Loading...
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 * Written by Ryusuke Konishi.
17 *
18 */
19
20#include <linux/pagemap.h>
21#include <linux/buffer_head.h>
22#include <linux/writeback.h>
23#include <linux/bitops.h>
24#include <linux/bio.h>
25#include <linux/completion.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/freezer.h>
29#include <linux/kthread.h>
30#include <linux/crc32.h>
31#include <linux/pagevec.h>
32#include <linux/slab.h>
33#include <linux/sched/signal.h>
34
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"
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 /*
51 * Upper limit of the number of segments
52 * appended in collection retry loop
53 */
54
55/* Construction mode */
56enum {
57 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
58 SC_LSEG_DSYNC, /*
59 * Flush data blocks of a given file and make
60 * a logical segment without a super root.
61 */
62 SC_FLUSH_FILE, /*
63 * Flush data files, leads to segment writes without
64 * creating a checkpoint.
65 */
66 SC_FLUSH_DAT, /*
67 * Flush DAT file. This also creates segments
68 * without a checkpoint.
69 */
70};
71
72/* Stage numbers of dirty block collection */
73enum {
74 NILFS_ST_INIT = 0,
75 NILFS_ST_GC, /* Collecting dirty blocks for GC */
76 NILFS_ST_FILE,
77 NILFS_ST_IFILE,
78 NILFS_ST_CPFILE,
79 NILFS_ST_SUFILE,
80 NILFS_ST_DAT,
81 NILFS_ST_SR, /* Super root */
82 NILFS_ST_DSYNC, /* Data sync blocks */
83 NILFS_ST_DONE,
84};
85
86#define CREATE_TRACE_POINTS
87#include <trace/events/nilfs2.h>
88
89/*
90 * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
91 * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
92 * the variable must use them because transition of stage count must involve
93 * trace events (trace_nilfs2_collection_stage_transition).
94 *
95 * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
96 * produce tracepoint events. It is provided just for making the intention
97 * clear.
98 */
99static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
100{
101 sci->sc_stage.scnt++;
102 trace_nilfs2_collection_stage_transition(sci);
103}
104
105static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
106{
107 sci->sc_stage.scnt = next_scnt;
108 trace_nilfs2_collection_stage_transition(sci);
109}
110
111static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
112{
113 return sci->sc_stage.scnt;
114}
115
116/* State flags of collection */
117#define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
118#define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
119#define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
120#define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
121
122/* Operations depending on the construction mode and file type */
123struct nilfs_sc_operations {
124 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
125 struct inode *);
126 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
127 struct inode *);
128 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
129 struct inode *);
130 void (*write_data_binfo)(struct nilfs_sc_info *,
131 struct nilfs_segsum_pointer *,
132 union nilfs_binfo *);
133 void (*write_node_binfo)(struct nilfs_sc_info *,
134 struct nilfs_segsum_pointer *,
135 union nilfs_binfo *);
136};
137
138/*
139 * Other definitions
140 */
141static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
142static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
143static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
144static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
145
146#define nilfs_cnt32_gt(a, b) \
147 (typecheck(__u32, a) && typecheck(__u32, b) && \
148 ((__s32)(b) - (__s32)(a) < 0))
149#define nilfs_cnt32_ge(a, b) \
150 (typecheck(__u32, a) && typecheck(__u32, b) && \
151 ((__s32)(a) - (__s32)(b) >= 0))
152#define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
153#define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
154
155static int nilfs_prepare_segment_lock(struct super_block *sb,
156 struct nilfs_transaction_info *ti)
157{
158 struct nilfs_transaction_info *cur_ti = current->journal_info;
159 void *save = NULL;
160
161 if (cur_ti) {
162 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
163 return ++cur_ti->ti_count;
164
165 /*
166 * If journal_info field is occupied by other FS,
167 * it is saved and will be restored on
168 * nilfs_transaction_commit().
169 */
170 nilfs_msg(sb, KERN_WARNING, "journal info from a different FS");
171 save = current->journal_info;
172 }
173 if (!ti) {
174 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
175 if (!ti)
176 return -ENOMEM;
177 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
178 } else {
179 ti->ti_flags = 0;
180 }
181 ti->ti_count = 0;
182 ti->ti_save = save;
183 ti->ti_magic = NILFS_TI_MAGIC;
184 current->journal_info = ti;
185 return 0;
186}
187
188/**
189 * nilfs_transaction_begin - start indivisible file operations.
190 * @sb: super block
191 * @ti: nilfs_transaction_info
192 * @vacancy_check: flags for vacancy rate checks
193 *
194 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
195 * the segment semaphore, to make a segment construction and write tasks
196 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
197 * The region enclosed by these two functions can be nested. To avoid a
198 * deadlock, the semaphore is only acquired or released in the outermost call.
199 *
200 * This function allocates a nilfs_transaction_info struct to keep context
201 * information on it. It is initialized and hooked onto the current task in
202 * the outermost call. If a pre-allocated struct is given to @ti, it is used
203 * instead; otherwise a new struct is assigned from a slab.
204 *
205 * When @vacancy_check flag is set, this function will check the amount of
206 * free space, and will wait for the GC to reclaim disk space if low capacity.
207 *
208 * Return Value: On success, 0 is returned. On error, one of the following
209 * negative error code is returned.
210 *
211 * %-ENOMEM - Insufficient memory available.
212 *
213 * %-ENOSPC - No space left on device
214 */
215int nilfs_transaction_begin(struct super_block *sb,
216 struct nilfs_transaction_info *ti,
217 int vacancy_check)
218{
219 struct the_nilfs *nilfs;
220 int ret = nilfs_prepare_segment_lock(sb, ti);
221 struct nilfs_transaction_info *trace_ti;
222
223 if (unlikely(ret < 0))
224 return ret;
225 if (ret > 0) {
226 trace_ti = current->journal_info;
227
228 trace_nilfs2_transaction_transition(sb, trace_ti,
229 trace_ti->ti_count, trace_ti->ti_flags,
230 TRACE_NILFS2_TRANSACTION_BEGIN);
231 return 0;
232 }
233
234 sb_start_intwrite(sb);
235
236 nilfs = sb->s_fs_info;
237 down_read(&nilfs->ns_segctor_sem);
238 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
239 up_read(&nilfs->ns_segctor_sem);
240 ret = -ENOSPC;
241 goto failed;
242 }
243
244 trace_ti = current->journal_info;
245 trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
246 trace_ti->ti_flags,
247 TRACE_NILFS2_TRANSACTION_BEGIN);
248 return 0;
249
250 failed:
251 ti = current->journal_info;
252 current->journal_info = ti->ti_save;
253 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
254 kmem_cache_free(nilfs_transaction_cachep, ti);
255 sb_end_intwrite(sb);
256 return ret;
257}
258
259/**
260 * nilfs_transaction_commit - commit indivisible file operations.
261 * @sb: super block
262 *
263 * nilfs_transaction_commit() releases the read semaphore which is
264 * acquired by nilfs_transaction_begin(). This is only performed
265 * in outermost call of this function. If a commit flag is set,
266 * nilfs_transaction_commit() sets a timer to start the segment
267 * constructor. If a sync flag is set, it starts construction
268 * directly.
269 */
270int nilfs_transaction_commit(struct super_block *sb)
271{
272 struct nilfs_transaction_info *ti = current->journal_info;
273 struct the_nilfs *nilfs = sb->s_fs_info;
274 int err = 0;
275
276 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
277 ti->ti_flags |= NILFS_TI_COMMIT;
278 if (ti->ti_count > 0) {
279 ti->ti_count--;
280 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
281 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
282 return 0;
283 }
284 if (nilfs->ns_writer) {
285 struct nilfs_sc_info *sci = nilfs->ns_writer;
286
287 if (ti->ti_flags & NILFS_TI_COMMIT)
288 nilfs_segctor_start_timer(sci);
289 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
290 nilfs_segctor_do_flush(sci, 0);
291 }
292 up_read(&nilfs->ns_segctor_sem);
293 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
294 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
295
296 current->journal_info = ti->ti_save;
297
298 if (ti->ti_flags & NILFS_TI_SYNC)
299 err = nilfs_construct_segment(sb);
300 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
301 kmem_cache_free(nilfs_transaction_cachep, ti);
302 sb_end_intwrite(sb);
303 return err;
304}
305
306void nilfs_transaction_abort(struct super_block *sb)
307{
308 struct nilfs_transaction_info *ti = current->journal_info;
309 struct the_nilfs *nilfs = sb->s_fs_info;
310
311 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
312 if (ti->ti_count > 0) {
313 ti->ti_count--;
314 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
315 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
316 return;
317 }
318 up_read(&nilfs->ns_segctor_sem);
319
320 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
321 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
322
323 current->journal_info = ti->ti_save;
324 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
325 kmem_cache_free(nilfs_transaction_cachep, ti);
326 sb_end_intwrite(sb);
327}
328
329void nilfs_relax_pressure_in_lock(struct super_block *sb)
330{
331 struct the_nilfs *nilfs = sb->s_fs_info;
332 struct nilfs_sc_info *sci = nilfs->ns_writer;
333
334 if (!sci || !sci->sc_flush_request)
335 return;
336
337 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
338 up_read(&nilfs->ns_segctor_sem);
339
340 down_write(&nilfs->ns_segctor_sem);
341 if (sci->sc_flush_request &&
342 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
343 struct nilfs_transaction_info *ti = current->journal_info;
344
345 ti->ti_flags |= NILFS_TI_WRITER;
346 nilfs_segctor_do_immediate_flush(sci);
347 ti->ti_flags &= ~NILFS_TI_WRITER;
348 }
349 downgrade_write(&nilfs->ns_segctor_sem);
350}
351
352static void nilfs_transaction_lock(struct super_block *sb,
353 struct nilfs_transaction_info *ti,
354 int gcflag)
355{
356 struct nilfs_transaction_info *cur_ti = current->journal_info;
357 struct the_nilfs *nilfs = sb->s_fs_info;
358 struct nilfs_sc_info *sci = nilfs->ns_writer;
359
360 WARN_ON(cur_ti);
361 ti->ti_flags = NILFS_TI_WRITER;
362 ti->ti_count = 0;
363 ti->ti_save = cur_ti;
364 ti->ti_magic = NILFS_TI_MAGIC;
365 current->journal_info = ti;
366
367 for (;;) {
368 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
369 ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
370
371 down_write(&nilfs->ns_segctor_sem);
372 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
373 break;
374
375 nilfs_segctor_do_immediate_flush(sci);
376
377 up_write(&nilfs->ns_segctor_sem);
378 cond_resched();
379 }
380 if (gcflag)
381 ti->ti_flags |= NILFS_TI_GC;
382
383 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
384 ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
385}
386
387static void nilfs_transaction_unlock(struct super_block *sb)
388{
389 struct nilfs_transaction_info *ti = current->journal_info;
390 struct the_nilfs *nilfs = sb->s_fs_info;
391
392 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
393 BUG_ON(ti->ti_count > 0);
394
395 up_write(&nilfs->ns_segctor_sem);
396 current->journal_info = ti->ti_save;
397
398 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
399 ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
400}
401
402static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
403 struct nilfs_segsum_pointer *ssp,
404 unsigned int bytes)
405{
406 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
407 unsigned int blocksize = sci->sc_super->s_blocksize;
408 void *p;
409
410 if (unlikely(ssp->offset + bytes > blocksize)) {
411 ssp->offset = 0;
412 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
413 &segbuf->sb_segsum_buffers));
414 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
415 }
416 p = ssp->bh->b_data + ssp->offset;
417 ssp->offset += bytes;
418 return p;
419}
420
421/**
422 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
423 * @sci: nilfs_sc_info
424 */
425static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
426{
427 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
428 struct buffer_head *sumbh;
429 unsigned int sumbytes;
430 unsigned int flags = 0;
431 int err;
432
433 if (nilfs_doing_gc())
434 flags = NILFS_SS_GC;
435 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
436 if (unlikely(err))
437 return err;
438
439 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
440 sumbytes = segbuf->sb_sum.sumbytes;
441 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
442 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
443 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
444 return 0;
445}
446
447static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
448{
449 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
450 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
451 return -E2BIG; /*
452 * The current segment is filled up
453 * (internal code)
454 */
455 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
456 return nilfs_segctor_reset_segment_buffer(sci);
457}
458
459static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
460{
461 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
462 int err;
463
464 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
465 err = nilfs_segctor_feed_segment(sci);
466 if (err)
467 return err;
468 segbuf = sci->sc_curseg;
469 }
470 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
471 if (likely(!err))
472 segbuf->sb_sum.flags |= NILFS_SS_SR;
473 return err;
474}
475
476/*
477 * Functions for making segment summary and payloads
478 */
479static int nilfs_segctor_segsum_block_required(
480 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
481 unsigned int binfo_size)
482{
483 unsigned int blocksize = sci->sc_super->s_blocksize;
484 /* Size of finfo and binfo is enough small against blocksize */
485
486 return ssp->offset + binfo_size +
487 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
488 blocksize;
489}
490
491static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
492 struct inode *inode)
493{
494 sci->sc_curseg->sb_sum.nfinfo++;
495 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
496 nilfs_segctor_map_segsum_entry(
497 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
498
499 if (NILFS_I(inode)->i_root &&
500 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
501 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
502 /* skip finfo */
503}
504
505static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
506 struct inode *inode)
507{
508 struct nilfs_finfo *finfo;
509 struct nilfs_inode_info *ii;
510 struct nilfs_segment_buffer *segbuf;
511 __u64 cno;
512
513 if (sci->sc_blk_cnt == 0)
514 return;
515
516 ii = NILFS_I(inode);
517
518 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
519 cno = ii->i_cno;
520 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
521 cno = 0;
522 else
523 cno = sci->sc_cno;
524
525 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
526 sizeof(*finfo));
527 finfo->fi_ino = cpu_to_le64(inode->i_ino);
528 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
529 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
530 finfo->fi_cno = cpu_to_le64(cno);
531
532 segbuf = sci->sc_curseg;
533 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
534 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
535 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
536 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
537}
538
539static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
540 struct buffer_head *bh,
541 struct inode *inode,
542 unsigned int binfo_size)
543{
544 struct nilfs_segment_buffer *segbuf;
545 int required, err = 0;
546
547 retry:
548 segbuf = sci->sc_curseg;
549 required = nilfs_segctor_segsum_block_required(
550 sci, &sci->sc_binfo_ptr, binfo_size);
551 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
552 nilfs_segctor_end_finfo(sci, inode);
553 err = nilfs_segctor_feed_segment(sci);
554 if (err)
555 return err;
556 goto retry;
557 }
558 if (unlikely(required)) {
559 err = nilfs_segbuf_extend_segsum(segbuf);
560 if (unlikely(err))
561 goto failed;
562 }
563 if (sci->sc_blk_cnt == 0)
564 nilfs_segctor_begin_finfo(sci, inode);
565
566 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
567 /* Substitution to vblocknr is delayed until update_blocknr() */
568 nilfs_segbuf_add_file_buffer(segbuf, bh);
569 sci->sc_blk_cnt++;
570 failed:
571 return err;
572}
573
574/*
575 * Callback functions that enumerate, mark, and collect dirty blocks
576 */
577static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
578 struct buffer_head *bh, struct inode *inode)
579{
580 int err;
581
582 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
583 if (err < 0)
584 return err;
585
586 err = nilfs_segctor_add_file_block(sci, bh, inode,
587 sizeof(struct nilfs_binfo_v));
588 if (!err)
589 sci->sc_datablk_cnt++;
590 return err;
591}
592
593static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
594 struct buffer_head *bh,
595 struct inode *inode)
596{
597 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
598}
599
600static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
601 struct buffer_head *bh,
602 struct inode *inode)
603{
604 WARN_ON(!buffer_dirty(bh));
605 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
606}
607
608static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
609 struct nilfs_segsum_pointer *ssp,
610 union nilfs_binfo *binfo)
611{
612 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
613 sci, ssp, sizeof(*binfo_v));
614 *binfo_v = binfo->bi_v;
615}
616
617static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
618 struct nilfs_segsum_pointer *ssp,
619 union nilfs_binfo *binfo)
620{
621 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
622 sci, ssp, sizeof(*vblocknr));
623 *vblocknr = binfo->bi_v.bi_vblocknr;
624}
625
626static const struct nilfs_sc_operations nilfs_sc_file_ops = {
627 .collect_data = nilfs_collect_file_data,
628 .collect_node = nilfs_collect_file_node,
629 .collect_bmap = nilfs_collect_file_bmap,
630 .write_data_binfo = nilfs_write_file_data_binfo,
631 .write_node_binfo = nilfs_write_file_node_binfo,
632};
633
634static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
635 struct buffer_head *bh, struct inode *inode)
636{
637 int err;
638
639 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
640 if (err < 0)
641 return err;
642
643 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
644 if (!err)
645 sci->sc_datablk_cnt++;
646 return err;
647}
648
649static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
650 struct buffer_head *bh, struct inode *inode)
651{
652 WARN_ON(!buffer_dirty(bh));
653 return nilfs_segctor_add_file_block(sci, bh, inode,
654 sizeof(struct nilfs_binfo_dat));
655}
656
657static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
658 struct nilfs_segsum_pointer *ssp,
659 union nilfs_binfo *binfo)
660{
661 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
662 sizeof(*blkoff));
663 *blkoff = binfo->bi_dat.bi_blkoff;
664}
665
666static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
667 struct nilfs_segsum_pointer *ssp,
668 union nilfs_binfo *binfo)
669{
670 struct nilfs_binfo_dat *binfo_dat =
671 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
672 *binfo_dat = binfo->bi_dat;
673}
674
675static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
676 .collect_data = nilfs_collect_dat_data,
677 .collect_node = nilfs_collect_file_node,
678 .collect_bmap = nilfs_collect_dat_bmap,
679 .write_data_binfo = nilfs_write_dat_data_binfo,
680 .write_node_binfo = nilfs_write_dat_node_binfo,
681};
682
683static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
684 .collect_data = nilfs_collect_file_data,
685 .collect_node = NULL,
686 .collect_bmap = NULL,
687 .write_data_binfo = nilfs_write_file_data_binfo,
688 .write_node_binfo = NULL,
689};
690
691static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
692 struct list_head *listp,
693 size_t nlimit,
694 loff_t start, loff_t end)
695{
696 struct address_space *mapping = inode->i_mapping;
697 struct pagevec pvec;
698 pgoff_t index = 0, last = ULONG_MAX;
699 size_t ndirties = 0;
700 int i;
701
702 if (unlikely(start != 0 || end != LLONG_MAX)) {
703 /*
704 * A valid range is given for sync-ing data pages. The
705 * range is rounded to per-page; extra dirty buffers
706 * may be included if blocksize < pagesize.
707 */
708 index = start >> PAGE_SHIFT;
709 last = end >> PAGE_SHIFT;
710 }
711 pagevec_init(&pvec);
712 repeat:
713 if (unlikely(index > last) ||
714 !pagevec_lookup_range_tag(&pvec, mapping, &index, last,
715 PAGECACHE_TAG_DIRTY))
716 return ndirties;
717
718 for (i = 0; i < pagevec_count(&pvec); i++) {
719 struct buffer_head *bh, *head;
720 struct page *page = pvec.pages[i];
721
722 lock_page(page);
723 if (!page_has_buffers(page))
724 create_empty_buffers(page, i_blocksize(inode), 0);
725 unlock_page(page);
726
727 bh = head = page_buffers(page);
728 do {
729 if (!buffer_dirty(bh) || buffer_async_write(bh))
730 continue;
731 get_bh(bh);
732 list_add_tail(&bh->b_assoc_buffers, listp);
733 ndirties++;
734 if (unlikely(ndirties >= nlimit)) {
735 pagevec_release(&pvec);
736 cond_resched();
737 return ndirties;
738 }
739 } while (bh = bh->b_this_page, bh != head);
740 }
741 pagevec_release(&pvec);
742 cond_resched();
743 goto repeat;
744}
745
746static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
747 struct list_head *listp)
748{
749 struct nilfs_inode_info *ii = NILFS_I(inode);
750 struct address_space *mapping = &ii->i_btnode_cache;
751 struct pagevec pvec;
752 struct buffer_head *bh, *head;
753 unsigned int i;
754 pgoff_t index = 0;
755
756 pagevec_init(&pvec);
757
758 while (pagevec_lookup_tag(&pvec, mapping, &index,
759 PAGECACHE_TAG_DIRTY)) {
760 for (i = 0; i < pagevec_count(&pvec); i++) {
761 bh = head = page_buffers(pvec.pages[i]);
762 do {
763 if (buffer_dirty(bh) &&
764 !buffer_async_write(bh)) {
765 get_bh(bh);
766 list_add_tail(&bh->b_assoc_buffers,
767 listp);
768 }
769 bh = bh->b_this_page;
770 } while (bh != head);
771 }
772 pagevec_release(&pvec);
773 cond_resched();
774 }
775}
776
777static void nilfs_dispose_list(struct the_nilfs *nilfs,
778 struct list_head *head, int force)
779{
780 struct nilfs_inode_info *ii, *n;
781 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
782 unsigned int nv = 0;
783
784 while (!list_empty(head)) {
785 spin_lock(&nilfs->ns_inode_lock);
786 list_for_each_entry_safe(ii, n, head, i_dirty) {
787 list_del_init(&ii->i_dirty);
788 if (force) {
789 if (unlikely(ii->i_bh)) {
790 brelse(ii->i_bh);
791 ii->i_bh = NULL;
792 }
793 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
794 set_bit(NILFS_I_QUEUED, &ii->i_state);
795 list_add_tail(&ii->i_dirty,
796 &nilfs->ns_dirty_files);
797 continue;
798 }
799 ivec[nv++] = ii;
800 if (nv == SC_N_INODEVEC)
801 break;
802 }
803 spin_unlock(&nilfs->ns_inode_lock);
804
805 for (pii = ivec; nv > 0; pii++, nv--)
806 iput(&(*pii)->vfs_inode);
807 }
808}
809
810static void nilfs_iput_work_func(struct work_struct *work)
811{
812 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
813 sc_iput_work);
814 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
815
816 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
817}
818
819static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
820 struct nilfs_root *root)
821{
822 int ret = 0;
823
824 if (nilfs_mdt_fetch_dirty(root->ifile))
825 ret++;
826 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
827 ret++;
828 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
829 ret++;
830 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
831 ret++;
832 return ret;
833}
834
835static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
836{
837 return list_empty(&sci->sc_dirty_files) &&
838 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
839 sci->sc_nfreesegs == 0 &&
840 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
841}
842
843static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
844{
845 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
846 int ret = 0;
847
848 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
849 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
850
851 spin_lock(&nilfs->ns_inode_lock);
852 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
853 ret++;
854
855 spin_unlock(&nilfs->ns_inode_lock);
856 return ret;
857}
858
859static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
860{
861 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
862
863 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
864 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
865 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
866 nilfs_mdt_clear_dirty(nilfs->ns_dat);
867}
868
869static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
870{
871 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
872 struct buffer_head *bh_cp;
873 struct nilfs_checkpoint *raw_cp;
874 int err;
875
876 /* XXX: this interface will be changed */
877 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
878 &raw_cp, &bh_cp);
879 if (likely(!err)) {
880 /*
881 * The following code is duplicated with cpfile. But, it is
882 * needed to collect the checkpoint even if it was not newly
883 * created.
884 */
885 mark_buffer_dirty(bh_cp);
886 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
887 nilfs_cpfile_put_checkpoint(
888 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
889 } else
890 WARN_ON(err == -EINVAL || err == -ENOENT);
891
892 return err;
893}
894
895static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
896{
897 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
898 struct buffer_head *bh_cp;
899 struct nilfs_checkpoint *raw_cp;
900 int err;
901
902 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
903 &raw_cp, &bh_cp);
904 if (unlikely(err)) {
905 WARN_ON(err == -EINVAL || err == -ENOENT);
906 goto failed_ibh;
907 }
908 raw_cp->cp_snapshot_list.ssl_next = 0;
909 raw_cp->cp_snapshot_list.ssl_prev = 0;
910 raw_cp->cp_inodes_count =
911 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
912 raw_cp->cp_blocks_count =
913 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
914 raw_cp->cp_nblk_inc =
915 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
916 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
917 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
918
919 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
920 nilfs_checkpoint_clear_minor(raw_cp);
921 else
922 nilfs_checkpoint_set_minor(raw_cp);
923
924 nilfs_write_inode_common(sci->sc_root->ifile,
925 &raw_cp->cp_ifile_inode, 1);
926 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
927 return 0;
928
929 failed_ibh:
930 return err;
931}
932
933static void nilfs_fill_in_file_bmap(struct inode *ifile,
934 struct nilfs_inode_info *ii)
935
936{
937 struct buffer_head *ibh;
938 struct nilfs_inode *raw_inode;
939
940 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
941 ibh = ii->i_bh;
942 BUG_ON(!ibh);
943 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
944 ibh);
945 nilfs_bmap_write(ii->i_bmap, raw_inode);
946 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
947 }
948}
949
950static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
951{
952 struct nilfs_inode_info *ii;
953
954 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
955 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
956 set_bit(NILFS_I_COLLECTED, &ii->i_state);
957 }
958}
959
960static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
961 struct the_nilfs *nilfs)
962{
963 struct buffer_head *bh_sr;
964 struct nilfs_super_root *raw_sr;
965 unsigned int isz, srsz;
966
967 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
968 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
969 isz = nilfs->ns_inode_size;
970 srsz = NILFS_SR_BYTES(isz);
971
972 raw_sr->sr_bytes = cpu_to_le16(srsz);
973 raw_sr->sr_nongc_ctime
974 = cpu_to_le64(nilfs_doing_gc() ?
975 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
976 raw_sr->sr_flags = 0;
977
978 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
979 NILFS_SR_DAT_OFFSET(isz), 1);
980 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
981 NILFS_SR_CPFILE_OFFSET(isz), 1);
982 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
983 NILFS_SR_SUFILE_OFFSET(isz), 1);
984 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
985}
986
987static void nilfs_redirty_inodes(struct list_head *head)
988{
989 struct nilfs_inode_info *ii;
990
991 list_for_each_entry(ii, head, i_dirty) {
992 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
993 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
994 }
995}
996
997static void nilfs_drop_collected_inodes(struct list_head *head)
998{
999 struct nilfs_inode_info *ii;
1000
1001 list_for_each_entry(ii, head, i_dirty) {
1002 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1003 continue;
1004
1005 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1006 set_bit(NILFS_I_UPDATED, &ii->i_state);
1007 }
1008}
1009
1010static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1011 struct inode *inode,
1012 struct list_head *listp,
1013 int (*collect)(struct nilfs_sc_info *,
1014 struct buffer_head *,
1015 struct inode *))
1016{
1017 struct buffer_head *bh, *n;
1018 int err = 0;
1019
1020 if (collect) {
1021 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1022 list_del_init(&bh->b_assoc_buffers);
1023 err = collect(sci, bh, inode);
1024 brelse(bh);
1025 if (unlikely(err))
1026 goto dispose_buffers;
1027 }
1028 return 0;
1029 }
1030
1031 dispose_buffers:
1032 while (!list_empty(listp)) {
1033 bh = list_first_entry(listp, struct buffer_head,
1034 b_assoc_buffers);
1035 list_del_init(&bh->b_assoc_buffers);
1036 brelse(bh);
1037 }
1038 return err;
1039}
1040
1041static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1042{
1043 /* Remaining number of blocks within segment buffer */
1044 return sci->sc_segbuf_nblocks -
1045 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1046}
1047
1048static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1049 struct inode *inode,
1050 const struct nilfs_sc_operations *sc_ops)
1051{
1052 LIST_HEAD(data_buffers);
1053 LIST_HEAD(node_buffers);
1054 int err;
1055
1056 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1057 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1058
1059 n = nilfs_lookup_dirty_data_buffers(
1060 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1061 if (n > rest) {
1062 err = nilfs_segctor_apply_buffers(
1063 sci, inode, &data_buffers,
1064 sc_ops->collect_data);
1065 BUG_ON(!err); /* always receive -E2BIG or true error */
1066 goto break_or_fail;
1067 }
1068 }
1069 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1070
1071 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1072 err = nilfs_segctor_apply_buffers(
1073 sci, inode, &data_buffers, sc_ops->collect_data);
1074 if (unlikely(err)) {
1075 /* dispose node list */
1076 nilfs_segctor_apply_buffers(
1077 sci, inode, &node_buffers, NULL);
1078 goto break_or_fail;
1079 }
1080 sci->sc_stage.flags |= NILFS_CF_NODE;
1081 }
1082 /* Collect node */
1083 err = nilfs_segctor_apply_buffers(
1084 sci, inode, &node_buffers, sc_ops->collect_node);
1085 if (unlikely(err))
1086 goto break_or_fail;
1087
1088 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1089 err = nilfs_segctor_apply_buffers(
1090 sci, inode, &node_buffers, sc_ops->collect_bmap);
1091 if (unlikely(err))
1092 goto break_or_fail;
1093
1094 nilfs_segctor_end_finfo(sci, inode);
1095 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1096
1097 break_or_fail:
1098 return err;
1099}
1100
1101static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1102 struct inode *inode)
1103{
1104 LIST_HEAD(data_buffers);
1105 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1106 int err;
1107
1108 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1109 sci->sc_dsync_start,
1110 sci->sc_dsync_end);
1111
1112 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1113 nilfs_collect_file_data);
1114 if (!err) {
1115 nilfs_segctor_end_finfo(sci, inode);
1116 BUG_ON(n > rest);
1117 /* always receive -E2BIG or true error if n > rest */
1118 }
1119 return err;
1120}
1121
1122static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1123{
1124 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1125 struct list_head *head;
1126 struct nilfs_inode_info *ii;
1127 size_t ndone;
1128 int err = 0;
1129
1130 switch (nilfs_sc_cstage_get(sci)) {
1131 case NILFS_ST_INIT:
1132 /* Pre-processes */
1133 sci->sc_stage.flags = 0;
1134
1135 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1136 sci->sc_nblk_inc = 0;
1137 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1138 if (mode == SC_LSEG_DSYNC) {
1139 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1140 goto dsync_mode;
1141 }
1142 }
1143
1144 sci->sc_stage.dirty_file_ptr = NULL;
1145 sci->sc_stage.gc_inode_ptr = NULL;
1146 if (mode == SC_FLUSH_DAT) {
1147 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1148 goto dat_stage;
1149 }
1150 nilfs_sc_cstage_inc(sci); /* Fall through */
1151 case NILFS_ST_GC:
1152 if (nilfs_doing_gc()) {
1153 head = &sci->sc_gc_inodes;
1154 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1155 head, i_dirty);
1156 list_for_each_entry_continue(ii, head, i_dirty) {
1157 err = nilfs_segctor_scan_file(
1158 sci, &ii->vfs_inode,
1159 &nilfs_sc_file_ops);
1160 if (unlikely(err)) {
1161 sci->sc_stage.gc_inode_ptr = list_entry(
1162 ii->i_dirty.prev,
1163 struct nilfs_inode_info,
1164 i_dirty);
1165 goto break_or_fail;
1166 }
1167 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1168 }
1169 sci->sc_stage.gc_inode_ptr = NULL;
1170 }
1171 nilfs_sc_cstage_inc(sci); /* Fall through */
1172 case NILFS_ST_FILE:
1173 head = &sci->sc_dirty_files;
1174 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1175 i_dirty);
1176 list_for_each_entry_continue(ii, head, i_dirty) {
1177 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1178
1179 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1180 &nilfs_sc_file_ops);
1181 if (unlikely(err)) {
1182 sci->sc_stage.dirty_file_ptr =
1183 list_entry(ii->i_dirty.prev,
1184 struct nilfs_inode_info,
1185 i_dirty);
1186 goto break_or_fail;
1187 }
1188 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1189 /* XXX: required ? */
1190 }
1191 sci->sc_stage.dirty_file_ptr = NULL;
1192 if (mode == SC_FLUSH_FILE) {
1193 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1194 return 0;
1195 }
1196 nilfs_sc_cstage_inc(sci);
1197 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1198 /* Fall through */
1199 case NILFS_ST_IFILE:
1200 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1201 &nilfs_sc_file_ops);
1202 if (unlikely(err))
1203 break;
1204 nilfs_sc_cstage_inc(sci);
1205 /* Creating a checkpoint */
1206 err = nilfs_segctor_create_checkpoint(sci);
1207 if (unlikely(err))
1208 break;
1209 /* Fall through */
1210 case NILFS_ST_CPFILE:
1211 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1212 &nilfs_sc_file_ops);
1213 if (unlikely(err))
1214 break;
1215 nilfs_sc_cstage_inc(sci); /* Fall through */
1216 case NILFS_ST_SUFILE:
1217 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1218 sci->sc_nfreesegs, &ndone);
1219 if (unlikely(err)) {
1220 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1221 sci->sc_freesegs, ndone,
1222 NULL);
1223 break;
1224 }
1225 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1226
1227 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1228 &nilfs_sc_file_ops);
1229 if (unlikely(err))
1230 break;
1231 nilfs_sc_cstage_inc(sci); /* Fall through */
1232 case NILFS_ST_DAT:
1233 dat_stage:
1234 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1235 &nilfs_sc_dat_ops);
1236 if (unlikely(err))
1237 break;
1238 if (mode == SC_FLUSH_DAT) {
1239 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1240 return 0;
1241 }
1242 nilfs_sc_cstage_inc(sci); /* Fall through */
1243 case NILFS_ST_SR:
1244 if (mode == SC_LSEG_SR) {
1245 /* Appending a super root */
1246 err = nilfs_segctor_add_super_root(sci);
1247 if (unlikely(err))
1248 break;
1249 }
1250 /* End of a logical segment */
1251 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1252 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1253 return 0;
1254 case NILFS_ST_DSYNC:
1255 dsync_mode:
1256 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1257 ii = sci->sc_dsync_inode;
1258 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1259 break;
1260
1261 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1262 if (unlikely(err))
1263 break;
1264 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1265 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1266 return 0;
1267 case NILFS_ST_DONE:
1268 return 0;
1269 default:
1270 BUG();
1271 }
1272
1273 break_or_fail:
1274 return err;
1275}
1276
1277/**
1278 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1279 * @sci: nilfs_sc_info
1280 * @nilfs: nilfs object
1281 */
1282static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1283 struct the_nilfs *nilfs)
1284{
1285 struct nilfs_segment_buffer *segbuf, *prev;
1286 __u64 nextnum;
1287 int err, alloc = 0;
1288
1289 segbuf = nilfs_segbuf_new(sci->sc_super);
1290 if (unlikely(!segbuf))
1291 return -ENOMEM;
1292
1293 if (list_empty(&sci->sc_write_logs)) {
1294 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1295 nilfs->ns_pseg_offset, nilfs);
1296 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1297 nilfs_shift_to_next_segment(nilfs);
1298 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1299 }
1300
1301 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1302 nextnum = nilfs->ns_nextnum;
1303
1304 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1305 /* Start from the head of a new full segment */
1306 alloc++;
1307 } else {
1308 /* Continue logs */
1309 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1310 nilfs_segbuf_map_cont(segbuf, prev);
1311 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1312 nextnum = prev->sb_nextnum;
1313
1314 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1315 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1316 segbuf->sb_sum.seg_seq++;
1317 alloc++;
1318 }
1319 }
1320
1321 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1322 if (err)
1323 goto failed;
1324
1325 if (alloc) {
1326 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1327 if (err)
1328 goto failed;
1329 }
1330 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1331
1332 BUG_ON(!list_empty(&sci->sc_segbufs));
1333 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1334 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1335 return 0;
1336
1337 failed:
1338 nilfs_segbuf_free(segbuf);
1339 return err;
1340}
1341
1342static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1343 struct the_nilfs *nilfs, int nadd)
1344{
1345 struct nilfs_segment_buffer *segbuf, *prev;
1346 struct inode *sufile = nilfs->ns_sufile;
1347 __u64 nextnextnum;
1348 LIST_HEAD(list);
1349 int err, ret, i;
1350
1351 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1352 /*
1353 * Since the segment specified with nextnum might be allocated during
1354 * the previous construction, the buffer including its segusage may
1355 * not be dirty. The following call ensures that the buffer is dirty
1356 * and will pin the buffer on memory until the sufile is written.
1357 */
1358 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1359 if (unlikely(err))
1360 return err;
1361
1362 for (i = 0; i < nadd; i++) {
1363 /* extend segment info */
1364 err = -ENOMEM;
1365 segbuf = nilfs_segbuf_new(sci->sc_super);
1366 if (unlikely(!segbuf))
1367 goto failed;
1368
1369 /* map this buffer to region of segment on-disk */
1370 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1371 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1372
1373 /* allocate the next next full segment */
1374 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1375 if (unlikely(err))
1376 goto failed_segbuf;
1377
1378 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1379 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1380
1381 list_add_tail(&segbuf->sb_list, &list);
1382 prev = segbuf;
1383 }
1384 list_splice_tail(&list, &sci->sc_segbufs);
1385 return 0;
1386
1387 failed_segbuf:
1388 nilfs_segbuf_free(segbuf);
1389 failed:
1390 list_for_each_entry(segbuf, &list, sb_list) {
1391 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1392 WARN_ON(ret); /* never fails */
1393 }
1394 nilfs_destroy_logs(&list);
1395 return err;
1396}
1397
1398static void nilfs_free_incomplete_logs(struct list_head *logs,
1399 struct the_nilfs *nilfs)
1400{
1401 struct nilfs_segment_buffer *segbuf, *prev;
1402 struct inode *sufile = nilfs->ns_sufile;
1403 int ret;
1404
1405 segbuf = NILFS_FIRST_SEGBUF(logs);
1406 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1407 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1408 WARN_ON(ret); /* never fails */
1409 }
1410 if (atomic_read(&segbuf->sb_err)) {
1411 /* Case 1: The first segment failed */
1412 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1413 /*
1414 * Case 1a: Partial segment appended into an existing
1415 * segment
1416 */
1417 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1418 segbuf->sb_fseg_end);
1419 else /* Case 1b: New full segment */
1420 set_nilfs_discontinued(nilfs);
1421 }
1422
1423 prev = segbuf;
1424 list_for_each_entry_continue(segbuf, logs, sb_list) {
1425 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1426 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1427 WARN_ON(ret); /* never fails */
1428 }
1429 if (atomic_read(&segbuf->sb_err) &&
1430 segbuf->sb_segnum != nilfs->ns_nextnum)
1431 /* Case 2: extended segment (!= next) failed */
1432 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1433 prev = segbuf;
1434 }
1435}
1436
1437static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1438 struct inode *sufile)
1439{
1440 struct nilfs_segment_buffer *segbuf;
1441 unsigned long live_blocks;
1442 int ret;
1443
1444 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1445 live_blocks = segbuf->sb_sum.nblocks +
1446 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1447 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1448 live_blocks,
1449 sci->sc_seg_ctime);
1450 WARN_ON(ret); /* always succeed because the segusage is dirty */
1451 }
1452}
1453
1454static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1455{
1456 struct nilfs_segment_buffer *segbuf;
1457 int ret;
1458
1459 segbuf = NILFS_FIRST_SEGBUF(logs);
1460 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1461 segbuf->sb_pseg_start -
1462 segbuf->sb_fseg_start, 0);
1463 WARN_ON(ret); /* always succeed because the segusage is dirty */
1464
1465 list_for_each_entry_continue(segbuf, logs, sb_list) {
1466 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1467 0, 0);
1468 WARN_ON(ret); /* always succeed */
1469 }
1470}
1471
1472static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1473 struct nilfs_segment_buffer *last,
1474 struct inode *sufile)
1475{
1476 struct nilfs_segment_buffer *segbuf = last;
1477 int ret;
1478
1479 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1480 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1481 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1482 WARN_ON(ret);
1483 }
1484 nilfs_truncate_logs(&sci->sc_segbufs, last);
1485}
1486
1487
1488static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1489 struct the_nilfs *nilfs, int mode)
1490{
1491 struct nilfs_cstage prev_stage = sci->sc_stage;
1492 int err, nadd = 1;
1493
1494 /* Collection retry loop */
1495 for (;;) {
1496 sci->sc_nblk_this_inc = 0;
1497 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1498
1499 err = nilfs_segctor_reset_segment_buffer(sci);
1500 if (unlikely(err))
1501 goto failed;
1502
1503 err = nilfs_segctor_collect_blocks(sci, mode);
1504 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1505 if (!err)
1506 break;
1507
1508 if (unlikely(err != -E2BIG))
1509 goto failed;
1510
1511 /* The current segment is filled up */
1512 if (mode != SC_LSEG_SR ||
1513 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1514 break;
1515
1516 nilfs_clear_logs(&sci->sc_segbufs);
1517
1518 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1519 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1520 sci->sc_freesegs,
1521 sci->sc_nfreesegs,
1522 NULL);
1523 WARN_ON(err); /* do not happen */
1524 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1525 }
1526
1527 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1528 if (unlikely(err))
1529 return err;
1530
1531 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1532 sci->sc_stage = prev_stage;
1533 }
1534 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1535 return 0;
1536
1537 failed:
1538 return err;
1539}
1540
1541static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1542 struct buffer_head *new_bh)
1543{
1544 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1545
1546 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1547 /* The caller must release old_bh */
1548}
1549
1550static int
1551nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1552 struct nilfs_segment_buffer *segbuf,
1553 int mode)
1554{
1555 struct inode *inode = NULL;
1556 sector_t blocknr;
1557 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1558 unsigned long nblocks = 0, ndatablk = 0;
1559 const struct nilfs_sc_operations *sc_op = NULL;
1560 struct nilfs_segsum_pointer ssp;
1561 struct nilfs_finfo *finfo = NULL;
1562 union nilfs_binfo binfo;
1563 struct buffer_head *bh, *bh_org;
1564 ino_t ino = 0;
1565 int err = 0;
1566
1567 if (!nfinfo)
1568 goto out;
1569
1570 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1571 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1572 ssp.offset = sizeof(struct nilfs_segment_summary);
1573
1574 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1575 if (bh == segbuf->sb_super_root)
1576 break;
1577 if (!finfo) {
1578 finfo = nilfs_segctor_map_segsum_entry(
1579 sci, &ssp, sizeof(*finfo));
1580 ino = le64_to_cpu(finfo->fi_ino);
1581 nblocks = le32_to_cpu(finfo->fi_nblocks);
1582 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1583
1584 inode = bh->b_page->mapping->host;
1585
1586 if (mode == SC_LSEG_DSYNC)
1587 sc_op = &nilfs_sc_dsync_ops;
1588 else if (ino == NILFS_DAT_INO)
1589 sc_op = &nilfs_sc_dat_ops;
1590 else /* file blocks */
1591 sc_op = &nilfs_sc_file_ops;
1592 }
1593 bh_org = bh;
1594 get_bh(bh_org);
1595 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1596 &binfo);
1597 if (bh != bh_org)
1598 nilfs_list_replace_buffer(bh_org, bh);
1599 brelse(bh_org);
1600 if (unlikely(err))
1601 goto failed_bmap;
1602
1603 if (ndatablk > 0)
1604 sc_op->write_data_binfo(sci, &ssp, &binfo);
1605 else
1606 sc_op->write_node_binfo(sci, &ssp, &binfo);
1607
1608 blocknr++;
1609 if (--nblocks == 0) {
1610 finfo = NULL;
1611 if (--nfinfo == 0)
1612 break;
1613 } else if (ndatablk > 0)
1614 ndatablk--;
1615 }
1616 out:
1617 return 0;
1618
1619 failed_bmap:
1620 return err;
1621}
1622
1623static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1624{
1625 struct nilfs_segment_buffer *segbuf;
1626 int err;
1627
1628 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1629 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1630 if (unlikely(err))
1631 return err;
1632 nilfs_segbuf_fill_in_segsum(segbuf);
1633 }
1634 return 0;
1635}
1636
1637static void nilfs_begin_page_io(struct page *page)
1638{
1639 if (!page || PageWriteback(page))
1640 /*
1641 * For split b-tree node pages, this function may be called
1642 * twice. We ignore the 2nd or later calls by this check.
1643 */
1644 return;
1645
1646 lock_page(page);
1647 clear_page_dirty_for_io(page);
1648 set_page_writeback(page);
1649 unlock_page(page);
1650}
1651
1652static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1653{
1654 struct nilfs_segment_buffer *segbuf;
1655 struct page *bd_page = NULL, *fs_page = NULL;
1656
1657 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1658 struct buffer_head *bh;
1659
1660 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1661 b_assoc_buffers) {
1662 if (bh->b_page != bd_page) {
1663 if (bd_page) {
1664 lock_page(bd_page);
1665 clear_page_dirty_for_io(bd_page);
1666 set_page_writeback(bd_page);
1667 unlock_page(bd_page);
1668 }
1669 bd_page = bh->b_page;
1670 }
1671 }
1672
1673 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1674 b_assoc_buffers) {
1675 set_buffer_async_write(bh);
1676 if (bh == segbuf->sb_super_root) {
1677 if (bh->b_page != bd_page) {
1678 lock_page(bd_page);
1679 clear_page_dirty_for_io(bd_page);
1680 set_page_writeback(bd_page);
1681 unlock_page(bd_page);
1682 bd_page = bh->b_page;
1683 }
1684 break;
1685 }
1686 if (bh->b_page != fs_page) {
1687 nilfs_begin_page_io(fs_page);
1688 fs_page = bh->b_page;
1689 }
1690 }
1691 }
1692 if (bd_page) {
1693 lock_page(bd_page);
1694 clear_page_dirty_for_io(bd_page);
1695 set_page_writeback(bd_page);
1696 unlock_page(bd_page);
1697 }
1698 nilfs_begin_page_io(fs_page);
1699}
1700
1701static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1702 struct the_nilfs *nilfs)
1703{
1704 int ret;
1705
1706 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1707 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1708 return ret;
1709}
1710
1711static void nilfs_end_page_io(struct page *page, int err)
1712{
1713 if (!page)
1714 return;
1715
1716 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1717 /*
1718 * For b-tree node pages, this function may be called twice
1719 * or more because they might be split in a segment.
1720 */
1721 if (PageDirty(page)) {
1722 /*
1723 * For pages holding split b-tree node buffers, dirty
1724 * flag on the buffers may be cleared discretely.
1725 * In that case, the page is once redirtied for
1726 * remaining buffers, and it must be cancelled if
1727 * all the buffers get cleaned later.
1728 */
1729 lock_page(page);
1730 if (nilfs_page_buffers_clean(page))
1731 __nilfs_clear_page_dirty(page);
1732 unlock_page(page);
1733 }
1734 return;
1735 }
1736
1737 if (!err) {
1738 if (!nilfs_page_buffers_clean(page))
1739 __set_page_dirty_nobuffers(page);
1740 ClearPageError(page);
1741 } else {
1742 __set_page_dirty_nobuffers(page);
1743 SetPageError(page);
1744 }
1745
1746 end_page_writeback(page);
1747}
1748
1749static void nilfs_abort_logs(struct list_head *logs, int err)
1750{
1751 struct nilfs_segment_buffer *segbuf;
1752 struct page *bd_page = NULL, *fs_page = NULL;
1753 struct buffer_head *bh;
1754
1755 if (list_empty(logs))
1756 return;
1757
1758 list_for_each_entry(segbuf, logs, sb_list) {
1759 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1760 b_assoc_buffers) {
1761 if (bh->b_page != bd_page) {
1762 if (bd_page)
1763 end_page_writeback(bd_page);
1764 bd_page = bh->b_page;
1765 }
1766 }
1767
1768 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1769 b_assoc_buffers) {
1770 clear_buffer_async_write(bh);
1771 if (bh == segbuf->sb_super_root) {
1772 if (bh->b_page != bd_page) {
1773 end_page_writeback(bd_page);
1774 bd_page = bh->b_page;
1775 }
1776 break;
1777 }
1778 if (bh->b_page != fs_page) {
1779 nilfs_end_page_io(fs_page, err);
1780 fs_page = bh->b_page;
1781 }
1782 }
1783 }
1784 if (bd_page)
1785 end_page_writeback(bd_page);
1786
1787 nilfs_end_page_io(fs_page, err);
1788}
1789
1790static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1791 struct the_nilfs *nilfs, int err)
1792{
1793 LIST_HEAD(logs);
1794 int ret;
1795
1796 list_splice_tail_init(&sci->sc_write_logs, &logs);
1797 ret = nilfs_wait_on_logs(&logs);
1798 nilfs_abort_logs(&logs, ret ? : err);
1799
1800 list_splice_tail_init(&sci->sc_segbufs, &logs);
1801 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1802 nilfs_free_incomplete_logs(&logs, nilfs);
1803
1804 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1805 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1806 sci->sc_freesegs,
1807 sci->sc_nfreesegs,
1808 NULL);
1809 WARN_ON(ret); /* do not happen */
1810 }
1811
1812 nilfs_destroy_logs(&logs);
1813}
1814
1815static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1816 struct nilfs_segment_buffer *segbuf)
1817{
1818 nilfs->ns_segnum = segbuf->sb_segnum;
1819 nilfs->ns_nextnum = segbuf->sb_nextnum;
1820 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1821 + segbuf->sb_sum.nblocks;
1822 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1823 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1824}
1825
1826static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1827{
1828 struct nilfs_segment_buffer *segbuf;
1829 struct page *bd_page = NULL, *fs_page = NULL;
1830 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1831 int update_sr = false;
1832
1833 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1834 struct buffer_head *bh;
1835
1836 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1837 b_assoc_buffers) {
1838 set_buffer_uptodate(bh);
1839 clear_buffer_dirty(bh);
1840 if (bh->b_page != bd_page) {
1841 if (bd_page)
1842 end_page_writeback(bd_page);
1843 bd_page = bh->b_page;
1844 }
1845 }
1846 /*
1847 * We assume that the buffers which belong to the same page
1848 * continue over the buffer list.
1849 * Under this assumption, the last BHs of pages is
1850 * identifiable by the discontinuity of bh->b_page
1851 * (page != fs_page).
1852 *
1853 * For B-tree node blocks, however, this assumption is not
1854 * guaranteed. The cleanup code of B-tree node pages needs
1855 * special care.
1856 */
1857 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1858 b_assoc_buffers) {
1859 const unsigned long set_bits = BIT(BH_Uptodate);
1860 const unsigned long clear_bits =
1861 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1862 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1863 BIT(BH_NILFS_Redirected));
1864
1865 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1866 if (bh == segbuf->sb_super_root) {
1867 if (bh->b_page != bd_page) {
1868 end_page_writeback(bd_page);
1869 bd_page = bh->b_page;
1870 }
1871 update_sr = true;
1872 break;
1873 }
1874 if (bh->b_page != fs_page) {
1875 nilfs_end_page_io(fs_page, 0);
1876 fs_page = bh->b_page;
1877 }
1878 }
1879
1880 if (!nilfs_segbuf_simplex(segbuf)) {
1881 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1882 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1883 sci->sc_lseg_stime = jiffies;
1884 }
1885 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1886 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1887 }
1888 }
1889 /*
1890 * Since pages may continue over multiple segment buffers,
1891 * end of the last page must be checked outside of the loop.
1892 */
1893 if (bd_page)
1894 end_page_writeback(bd_page);
1895
1896 nilfs_end_page_io(fs_page, 0);
1897
1898 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1899
1900 if (nilfs_doing_gc())
1901 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1902 else
1903 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1904
1905 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1906
1907 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1908 nilfs_set_next_segment(nilfs, segbuf);
1909
1910 if (update_sr) {
1911 nilfs->ns_flushed_device = 0;
1912 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1913 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1914
1915 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1916 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1917 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1918 nilfs_segctor_clear_metadata_dirty(sci);
1919 } else
1920 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1921}
1922
1923static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1924{
1925 int ret;
1926
1927 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1928 if (!ret) {
1929 nilfs_segctor_complete_write(sci);
1930 nilfs_destroy_logs(&sci->sc_write_logs);
1931 }
1932 return ret;
1933}
1934
1935static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1936 struct the_nilfs *nilfs)
1937{
1938 struct nilfs_inode_info *ii, *n;
1939 struct inode *ifile = sci->sc_root->ifile;
1940
1941 spin_lock(&nilfs->ns_inode_lock);
1942 retry:
1943 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1944 if (!ii->i_bh) {
1945 struct buffer_head *ibh;
1946 int err;
1947
1948 spin_unlock(&nilfs->ns_inode_lock);
1949 err = nilfs_ifile_get_inode_block(
1950 ifile, ii->vfs_inode.i_ino, &ibh);
1951 if (unlikely(err)) {
1952 nilfs_msg(sci->sc_super, KERN_WARNING,
1953 "log writer: error %d getting inode block (ino=%lu)",
1954 err, ii->vfs_inode.i_ino);
1955 return err;
1956 }
1957 spin_lock(&nilfs->ns_inode_lock);
1958 if (likely(!ii->i_bh))
1959 ii->i_bh = ibh;
1960 else
1961 brelse(ibh);
1962 goto retry;
1963 }
1964
1965 // Always redirty the buffer to avoid race condition
1966 mark_buffer_dirty(ii->i_bh);
1967 nilfs_mdt_mark_dirty(ifile);
1968
1969 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1970 set_bit(NILFS_I_BUSY, &ii->i_state);
1971 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1972 }
1973 spin_unlock(&nilfs->ns_inode_lock);
1974
1975 return 0;
1976}
1977
1978static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1979 struct the_nilfs *nilfs)
1980{
1981 struct nilfs_inode_info *ii, *n;
1982 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1983 int defer_iput = false;
1984
1985 spin_lock(&nilfs->ns_inode_lock);
1986 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1987 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1988 test_bit(NILFS_I_DIRTY, &ii->i_state))
1989 continue;
1990
1991 clear_bit(NILFS_I_BUSY, &ii->i_state);
1992 brelse(ii->i_bh);
1993 ii->i_bh = NULL;
1994 list_del_init(&ii->i_dirty);
1995 if (!ii->vfs_inode.i_nlink || during_mount) {
1996 /*
1997 * Defer calling iput() to avoid deadlocks if
1998 * i_nlink == 0 or mount is not yet finished.
1999 */
2000 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2001 defer_iput = true;
2002 } else {
2003 spin_unlock(&nilfs->ns_inode_lock);
2004 iput(&ii->vfs_inode);
2005 spin_lock(&nilfs->ns_inode_lock);
2006 }
2007 }
2008 spin_unlock(&nilfs->ns_inode_lock);
2009
2010 if (defer_iput)
2011 schedule_work(&sci->sc_iput_work);
2012}
2013
2014/*
2015 * Main procedure of segment constructor
2016 */
2017static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2018{
2019 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2020 int err;
2021
2022 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2023 sci->sc_cno = nilfs->ns_cno;
2024
2025 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2026 if (unlikely(err))
2027 goto out;
2028
2029 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2030 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2031
2032 if (nilfs_segctor_clean(sci))
2033 goto out;
2034
2035 do {
2036 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2037
2038 err = nilfs_segctor_begin_construction(sci, nilfs);
2039 if (unlikely(err))
2040 goto out;
2041
2042 /* Update time stamp */
2043 sci->sc_seg_ctime = ktime_get_real_seconds();
2044
2045 err = nilfs_segctor_collect(sci, nilfs, mode);
2046 if (unlikely(err))
2047 goto failed;
2048
2049 /* Avoid empty segment */
2050 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2051 nilfs_segbuf_empty(sci->sc_curseg)) {
2052 nilfs_segctor_abort_construction(sci, nilfs, 1);
2053 goto out;
2054 }
2055
2056 err = nilfs_segctor_assign(sci, mode);
2057 if (unlikely(err))
2058 goto failed;
2059
2060 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2061 nilfs_segctor_fill_in_file_bmap(sci);
2062
2063 if (mode == SC_LSEG_SR &&
2064 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2065 err = nilfs_segctor_fill_in_checkpoint(sci);
2066 if (unlikely(err))
2067 goto failed_to_write;
2068
2069 nilfs_segctor_fill_in_super_root(sci, nilfs);
2070 }
2071 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2072
2073 /* Write partial segments */
2074 nilfs_segctor_prepare_write(sci);
2075
2076 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2077 nilfs->ns_crc_seed);
2078
2079 err = nilfs_segctor_write(sci, nilfs);
2080 if (unlikely(err))
2081 goto failed_to_write;
2082
2083 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2084 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2085 /*
2086 * At this point, we avoid double buffering
2087 * for blocksize < pagesize because page dirty
2088 * flag is turned off during write and dirty
2089 * buffers are not properly collected for
2090 * pages crossing over segments.
2091 */
2092 err = nilfs_segctor_wait(sci);
2093 if (err)
2094 goto failed_to_write;
2095 }
2096 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2097
2098 out:
2099 nilfs_segctor_drop_written_files(sci, nilfs);
2100 return err;
2101
2102 failed_to_write:
2103 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2104 nilfs_redirty_inodes(&sci->sc_dirty_files);
2105
2106 failed:
2107 if (nilfs_doing_gc())
2108 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2109 nilfs_segctor_abort_construction(sci, nilfs, err);
2110 goto out;
2111}
2112
2113/**
2114 * nilfs_segctor_start_timer - set timer of background write
2115 * @sci: nilfs_sc_info
2116 *
2117 * If the timer has already been set, it ignores the new request.
2118 * This function MUST be called within a section locking the segment
2119 * semaphore.
2120 */
2121static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2122{
2123 spin_lock(&sci->sc_state_lock);
2124 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2125 sci->sc_timer.expires = jiffies + sci->sc_interval;
2126 add_timer(&sci->sc_timer);
2127 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2128 }
2129 spin_unlock(&sci->sc_state_lock);
2130}
2131
2132static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2133{
2134 spin_lock(&sci->sc_state_lock);
2135 if (!(sci->sc_flush_request & BIT(bn))) {
2136 unsigned long prev_req = sci->sc_flush_request;
2137
2138 sci->sc_flush_request |= BIT(bn);
2139 if (!prev_req)
2140 wake_up(&sci->sc_wait_daemon);
2141 }
2142 spin_unlock(&sci->sc_state_lock);
2143}
2144
2145/**
2146 * nilfs_flush_segment - trigger a segment construction for resource control
2147 * @sb: super block
2148 * @ino: inode number of the file to be flushed out.
2149 */
2150void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2151{
2152 struct the_nilfs *nilfs = sb->s_fs_info;
2153 struct nilfs_sc_info *sci = nilfs->ns_writer;
2154
2155 if (!sci || nilfs_doing_construction())
2156 return;
2157 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2158 /* assign bit 0 to data files */
2159}
2160
2161struct nilfs_segctor_wait_request {
2162 wait_queue_entry_t wq;
2163 __u32 seq;
2164 int err;
2165 atomic_t done;
2166};
2167
2168static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2169{
2170 struct nilfs_segctor_wait_request wait_req;
2171 int err = 0;
2172
2173 spin_lock(&sci->sc_state_lock);
2174 init_wait(&wait_req.wq);
2175 wait_req.err = 0;
2176 atomic_set(&wait_req.done, 0);
2177 wait_req.seq = ++sci->sc_seq_request;
2178 spin_unlock(&sci->sc_state_lock);
2179
2180 init_waitqueue_entry(&wait_req.wq, current);
2181 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2182 set_current_state(TASK_INTERRUPTIBLE);
2183 wake_up(&sci->sc_wait_daemon);
2184
2185 for (;;) {
2186 if (atomic_read(&wait_req.done)) {
2187 err = wait_req.err;
2188 break;
2189 }
2190 if (!signal_pending(current)) {
2191 schedule();
2192 continue;
2193 }
2194 err = -ERESTARTSYS;
2195 break;
2196 }
2197 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2198 return err;
2199}
2200
2201static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2202{
2203 struct nilfs_segctor_wait_request *wrq, *n;
2204 unsigned long flags;
2205
2206 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2207 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2208 if (!atomic_read(&wrq->done) &&
2209 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2210 wrq->err = err;
2211 atomic_set(&wrq->done, 1);
2212 }
2213 if (atomic_read(&wrq->done)) {
2214 wrq->wq.func(&wrq->wq,
2215 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2216 0, NULL);
2217 }
2218 }
2219 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2220}
2221
2222/**
2223 * nilfs_construct_segment - construct a logical segment
2224 * @sb: super block
2225 *
2226 * Return Value: On success, 0 is retured. On errors, one of the following
2227 * negative error code is returned.
2228 *
2229 * %-EROFS - Read only filesystem.
2230 *
2231 * %-EIO - I/O error
2232 *
2233 * %-ENOSPC - No space left on device (only in a panic state).
2234 *
2235 * %-ERESTARTSYS - Interrupted.
2236 *
2237 * %-ENOMEM - Insufficient memory available.
2238 */
2239int nilfs_construct_segment(struct super_block *sb)
2240{
2241 struct the_nilfs *nilfs = sb->s_fs_info;
2242 struct nilfs_sc_info *sci = nilfs->ns_writer;
2243 struct nilfs_transaction_info *ti;
2244 int err;
2245
2246 if (!sci)
2247 return -EROFS;
2248
2249 /* A call inside transactions causes a deadlock. */
2250 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2251
2252 err = nilfs_segctor_sync(sci);
2253 return err;
2254}
2255
2256/**
2257 * nilfs_construct_dsync_segment - construct a data-only logical segment
2258 * @sb: super block
2259 * @inode: inode whose data blocks should be written out
2260 * @start: start byte offset
2261 * @end: end byte offset (inclusive)
2262 *
2263 * Return Value: On success, 0 is retured. On errors, one of the following
2264 * negative error code is returned.
2265 *
2266 * %-EROFS - Read only filesystem.
2267 *
2268 * %-EIO - I/O error
2269 *
2270 * %-ENOSPC - No space left on device (only in a panic state).
2271 *
2272 * %-ERESTARTSYS - Interrupted.
2273 *
2274 * %-ENOMEM - Insufficient memory available.
2275 */
2276int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2277 loff_t start, loff_t end)
2278{
2279 struct the_nilfs *nilfs = sb->s_fs_info;
2280 struct nilfs_sc_info *sci = nilfs->ns_writer;
2281 struct nilfs_inode_info *ii;
2282 struct nilfs_transaction_info ti;
2283 int err = 0;
2284
2285 if (!sci)
2286 return -EROFS;
2287
2288 nilfs_transaction_lock(sb, &ti, 0);
2289
2290 ii = NILFS_I(inode);
2291 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2292 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2293 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2294 nilfs_discontinued(nilfs)) {
2295 nilfs_transaction_unlock(sb);
2296 err = nilfs_segctor_sync(sci);
2297 return err;
2298 }
2299
2300 spin_lock(&nilfs->ns_inode_lock);
2301 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2302 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2303 spin_unlock(&nilfs->ns_inode_lock);
2304 nilfs_transaction_unlock(sb);
2305 return 0;
2306 }
2307 spin_unlock(&nilfs->ns_inode_lock);
2308 sci->sc_dsync_inode = ii;
2309 sci->sc_dsync_start = start;
2310 sci->sc_dsync_end = end;
2311
2312 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2313 if (!err)
2314 nilfs->ns_flushed_device = 0;
2315
2316 nilfs_transaction_unlock(sb);
2317 return err;
2318}
2319
2320#define FLUSH_FILE_BIT (0x1) /* data file only */
2321#define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2322
2323/**
2324 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2325 * @sci: segment constructor object
2326 */
2327static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2328{
2329 spin_lock(&sci->sc_state_lock);
2330 sci->sc_seq_accepted = sci->sc_seq_request;
2331 spin_unlock(&sci->sc_state_lock);
2332 del_timer_sync(&sci->sc_timer);
2333}
2334
2335/**
2336 * nilfs_segctor_notify - notify the result of request to caller threads
2337 * @sci: segment constructor object
2338 * @mode: mode of log forming
2339 * @err: error code to be notified
2340 */
2341static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2342{
2343 /* Clear requests (even when the construction failed) */
2344 spin_lock(&sci->sc_state_lock);
2345
2346 if (mode == SC_LSEG_SR) {
2347 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2348 sci->sc_seq_done = sci->sc_seq_accepted;
2349 nilfs_segctor_wakeup(sci, err);
2350 sci->sc_flush_request = 0;
2351 } else {
2352 if (mode == SC_FLUSH_FILE)
2353 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2354 else if (mode == SC_FLUSH_DAT)
2355 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2356
2357 /* re-enable timer if checkpoint creation was not done */
2358 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2359 time_before(jiffies, sci->sc_timer.expires))
2360 add_timer(&sci->sc_timer);
2361 }
2362 spin_unlock(&sci->sc_state_lock);
2363}
2364
2365/**
2366 * nilfs_segctor_construct - form logs and write them to disk
2367 * @sci: segment constructor object
2368 * @mode: mode of log forming
2369 */
2370static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2371{
2372 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2373 struct nilfs_super_block **sbp;
2374 int err = 0;
2375
2376 nilfs_segctor_accept(sci);
2377
2378 if (nilfs_discontinued(nilfs))
2379 mode = SC_LSEG_SR;
2380 if (!nilfs_segctor_confirm(sci))
2381 err = nilfs_segctor_do_construct(sci, mode);
2382
2383 if (likely(!err)) {
2384 if (mode != SC_FLUSH_DAT)
2385 atomic_set(&nilfs->ns_ndirtyblks, 0);
2386 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2387 nilfs_discontinued(nilfs)) {
2388 down_write(&nilfs->ns_sem);
2389 err = -EIO;
2390 sbp = nilfs_prepare_super(sci->sc_super,
2391 nilfs_sb_will_flip(nilfs));
2392 if (likely(sbp)) {
2393 nilfs_set_log_cursor(sbp[0], nilfs);
2394 err = nilfs_commit_super(sci->sc_super,
2395 NILFS_SB_COMMIT);
2396 }
2397 up_write(&nilfs->ns_sem);
2398 }
2399 }
2400
2401 nilfs_segctor_notify(sci, mode, err);
2402 return err;
2403}
2404
2405static void nilfs_construction_timeout(struct timer_list *t)
2406{
2407 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2408
2409 wake_up_process(sci->sc_timer_task);
2410}
2411
2412static void
2413nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2414{
2415 struct nilfs_inode_info *ii, *n;
2416
2417 list_for_each_entry_safe(ii, n, head, i_dirty) {
2418 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2419 continue;
2420 list_del_init(&ii->i_dirty);
2421 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2422 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2423 iput(&ii->vfs_inode);
2424 }
2425}
2426
2427int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2428 void **kbufs)
2429{
2430 struct the_nilfs *nilfs = sb->s_fs_info;
2431 struct nilfs_sc_info *sci = nilfs->ns_writer;
2432 struct nilfs_transaction_info ti;
2433 int err;
2434
2435 if (unlikely(!sci))
2436 return -EROFS;
2437
2438 nilfs_transaction_lock(sb, &ti, 1);
2439
2440 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2441 if (unlikely(err))
2442 goto out_unlock;
2443
2444 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2445 if (unlikely(err)) {
2446 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2447 goto out_unlock;
2448 }
2449
2450 sci->sc_freesegs = kbufs[4];
2451 sci->sc_nfreesegs = argv[4].v_nmembs;
2452 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2453
2454 for (;;) {
2455 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2456 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2457
2458 if (likely(!err))
2459 break;
2460
2461 nilfs_msg(sb, KERN_WARNING, "error %d cleaning segments", err);
2462 set_current_state(TASK_INTERRUPTIBLE);
2463 schedule_timeout(sci->sc_interval);
2464 }
2465 if (nilfs_test_opt(nilfs, DISCARD)) {
2466 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2467 sci->sc_nfreesegs);
2468 if (ret) {
2469 nilfs_msg(sb, KERN_WARNING,
2470 "error %d on discard request, turning discards off for the device",
2471 ret);
2472 nilfs_clear_opt(nilfs, DISCARD);
2473 }
2474 }
2475
2476 out_unlock:
2477 sci->sc_freesegs = NULL;
2478 sci->sc_nfreesegs = 0;
2479 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2480 nilfs_transaction_unlock(sb);
2481 return err;
2482}
2483
2484static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2485{
2486 struct nilfs_transaction_info ti;
2487
2488 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2489 nilfs_segctor_construct(sci, mode);
2490
2491 /*
2492 * Unclosed segment should be retried. We do this using sc_timer.
2493 * Timeout of sc_timer will invoke complete construction which leads
2494 * to close the current logical segment.
2495 */
2496 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2497 nilfs_segctor_start_timer(sci);
2498
2499 nilfs_transaction_unlock(sci->sc_super);
2500}
2501
2502static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2503{
2504 int mode = 0;
2505
2506 spin_lock(&sci->sc_state_lock);
2507 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2508 SC_FLUSH_DAT : SC_FLUSH_FILE;
2509 spin_unlock(&sci->sc_state_lock);
2510
2511 if (mode) {
2512 nilfs_segctor_do_construct(sci, mode);
2513
2514 spin_lock(&sci->sc_state_lock);
2515 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2516 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2517 spin_unlock(&sci->sc_state_lock);
2518 }
2519 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2520}
2521
2522static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2523{
2524 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2525 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2526 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2527 return SC_FLUSH_FILE;
2528 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2529 return SC_FLUSH_DAT;
2530 }
2531 return SC_LSEG_SR;
2532}
2533
2534/**
2535 * nilfs_segctor_thread - main loop of the segment constructor thread.
2536 * @arg: pointer to a struct nilfs_sc_info.
2537 *
2538 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2539 * to execute segment constructions.
2540 */
2541static int nilfs_segctor_thread(void *arg)
2542{
2543 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2544 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2545 int timeout = 0;
2546
2547 sci->sc_timer_task = current;
2548
2549 /* start sync. */
2550 sci->sc_task = current;
2551 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2552 nilfs_msg(sci->sc_super, KERN_INFO,
2553 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2554 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2555
2556 spin_lock(&sci->sc_state_lock);
2557 loop:
2558 for (;;) {
2559 int mode;
2560
2561 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2562 goto end_thread;
2563
2564 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2565 mode = SC_LSEG_SR;
2566 else if (sci->sc_flush_request)
2567 mode = nilfs_segctor_flush_mode(sci);
2568 else
2569 break;
2570
2571 spin_unlock(&sci->sc_state_lock);
2572 nilfs_segctor_thread_construct(sci, mode);
2573 spin_lock(&sci->sc_state_lock);
2574 timeout = 0;
2575 }
2576
2577
2578 if (freezing(current)) {
2579 spin_unlock(&sci->sc_state_lock);
2580 try_to_freeze();
2581 spin_lock(&sci->sc_state_lock);
2582 } else {
2583 DEFINE_WAIT(wait);
2584 int should_sleep = 1;
2585
2586 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2587 TASK_INTERRUPTIBLE);
2588
2589 if (sci->sc_seq_request != sci->sc_seq_done)
2590 should_sleep = 0;
2591 else if (sci->sc_flush_request)
2592 should_sleep = 0;
2593 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2594 should_sleep = time_before(jiffies,
2595 sci->sc_timer.expires);
2596
2597 if (should_sleep) {
2598 spin_unlock(&sci->sc_state_lock);
2599 schedule();
2600 spin_lock(&sci->sc_state_lock);
2601 }
2602 finish_wait(&sci->sc_wait_daemon, &wait);
2603 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2604 time_after_eq(jiffies, sci->sc_timer.expires));
2605
2606 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2607 set_nilfs_discontinued(nilfs);
2608 }
2609 goto loop;
2610
2611 end_thread:
2612 spin_unlock(&sci->sc_state_lock);
2613
2614 /* end sync. */
2615 sci->sc_task = NULL;
2616 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2617 return 0;
2618}
2619
2620static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2621{
2622 struct task_struct *t;
2623
2624 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2625 if (IS_ERR(t)) {
2626 int err = PTR_ERR(t);
2627
2628 nilfs_msg(sci->sc_super, KERN_ERR,
2629 "error %d creating segctord thread", err);
2630 return err;
2631 }
2632 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2633 return 0;
2634}
2635
2636static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2637 __acquires(&sci->sc_state_lock)
2638 __releases(&sci->sc_state_lock)
2639{
2640 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2641
2642 while (sci->sc_task) {
2643 wake_up(&sci->sc_wait_daemon);
2644 spin_unlock(&sci->sc_state_lock);
2645 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2646 spin_lock(&sci->sc_state_lock);
2647 }
2648}
2649
2650/*
2651 * Setup & clean-up functions
2652 */
2653static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2654 struct nilfs_root *root)
2655{
2656 struct the_nilfs *nilfs = sb->s_fs_info;
2657 struct nilfs_sc_info *sci;
2658
2659 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2660 if (!sci)
2661 return NULL;
2662
2663 sci->sc_super = sb;
2664
2665 nilfs_get_root(root);
2666 sci->sc_root = root;
2667
2668 init_waitqueue_head(&sci->sc_wait_request);
2669 init_waitqueue_head(&sci->sc_wait_daemon);
2670 init_waitqueue_head(&sci->sc_wait_task);
2671 spin_lock_init(&sci->sc_state_lock);
2672 INIT_LIST_HEAD(&sci->sc_dirty_files);
2673 INIT_LIST_HEAD(&sci->sc_segbufs);
2674 INIT_LIST_HEAD(&sci->sc_write_logs);
2675 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2676 INIT_LIST_HEAD(&sci->sc_iput_queue);
2677 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2678 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2679
2680 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2681 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2682 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2683
2684 if (nilfs->ns_interval)
2685 sci->sc_interval = HZ * nilfs->ns_interval;
2686 if (nilfs->ns_watermark)
2687 sci->sc_watermark = nilfs->ns_watermark;
2688 return sci;
2689}
2690
2691static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2692{
2693 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2694
2695 /*
2696 * The segctord thread was stopped and its timer was removed.
2697 * But some tasks remain.
2698 */
2699 do {
2700 struct nilfs_transaction_info ti;
2701
2702 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2703 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2704 nilfs_transaction_unlock(sci->sc_super);
2705
2706 flush_work(&sci->sc_iput_work);
2707
2708 } while (ret && retrycount-- > 0);
2709}
2710
2711/**
2712 * nilfs_segctor_destroy - destroy the segment constructor.
2713 * @sci: nilfs_sc_info
2714 *
2715 * nilfs_segctor_destroy() kills the segctord thread and frees
2716 * the nilfs_sc_info struct.
2717 * Caller must hold the segment semaphore.
2718 */
2719static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2720{
2721 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2722 int flag;
2723
2724 up_write(&nilfs->ns_segctor_sem);
2725
2726 spin_lock(&sci->sc_state_lock);
2727 nilfs_segctor_kill_thread(sci);
2728 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2729 || sci->sc_seq_request != sci->sc_seq_done);
2730 spin_unlock(&sci->sc_state_lock);
2731
2732 if (flush_work(&sci->sc_iput_work))
2733 flag = true;
2734
2735 if (flag || !nilfs_segctor_confirm(sci))
2736 nilfs_segctor_write_out(sci);
2737
2738 if (!list_empty(&sci->sc_dirty_files)) {
2739 nilfs_msg(sci->sc_super, KERN_WARNING,
2740 "disposed unprocessed dirty file(s) when stopping log writer");
2741 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2742 }
2743
2744 if (!list_empty(&sci->sc_iput_queue)) {
2745 nilfs_msg(sci->sc_super, KERN_WARNING,
2746 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2747 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2748 }
2749
2750 WARN_ON(!list_empty(&sci->sc_segbufs));
2751 WARN_ON(!list_empty(&sci->sc_write_logs));
2752
2753 nilfs_put_root(sci->sc_root);
2754
2755 down_write(&nilfs->ns_segctor_sem);
2756
2757 del_timer_sync(&sci->sc_timer);
2758 kfree(sci);
2759}
2760
2761/**
2762 * nilfs_attach_log_writer - attach log writer
2763 * @sb: super block instance
2764 * @root: root object of the current filesystem tree
2765 *
2766 * This allocates a log writer object, initializes it, and starts the
2767 * log writer.
2768 *
2769 * Return Value: On success, 0 is returned. On error, one of the following
2770 * negative error code is returned.
2771 *
2772 * %-ENOMEM - Insufficient memory available.
2773 */
2774int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2775{
2776 struct the_nilfs *nilfs = sb->s_fs_info;
2777 int err;
2778
2779 if (nilfs->ns_writer) {
2780 /*
2781 * This happens if the filesystem was remounted
2782 * read/write after nilfs_error degenerated it into a
2783 * read-only mount.
2784 */
2785 nilfs_detach_log_writer(sb);
2786 }
2787
2788 nilfs->ns_writer = nilfs_segctor_new(sb, root);
2789 if (!nilfs->ns_writer)
2790 return -ENOMEM;
2791
2792 err = nilfs_segctor_start_thread(nilfs->ns_writer);
2793 if (err) {
2794 kfree(nilfs->ns_writer);
2795 nilfs->ns_writer = NULL;
2796 }
2797 return err;
2798}
2799
2800/**
2801 * nilfs_detach_log_writer - destroy log writer
2802 * @sb: super block instance
2803 *
2804 * This kills log writer daemon, frees the log writer object, and
2805 * destroys list of dirty files.
2806 */
2807void nilfs_detach_log_writer(struct super_block *sb)
2808{
2809 struct the_nilfs *nilfs = sb->s_fs_info;
2810 LIST_HEAD(garbage_list);
2811
2812 down_write(&nilfs->ns_segctor_sem);
2813 if (nilfs->ns_writer) {
2814 nilfs_segctor_destroy(nilfs->ns_writer);
2815 nilfs->ns_writer = NULL;
2816 }
2817
2818 /* Force to free the list of dirty files */
2819 spin_lock(&nilfs->ns_inode_lock);
2820 if (!list_empty(&nilfs->ns_dirty_files)) {
2821 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2822 nilfs_msg(sb, KERN_WARNING,
2823 "disposed unprocessed dirty file(s) when detaching log writer");
2824 }
2825 spin_unlock(&nilfs->ns_inode_lock);
2826 up_write(&nilfs->ns_segctor_sem);
2827
2828 nilfs_dispose_list(nilfs, &garbage_list, 1);
2829}
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * NILFS segment constructor.
4 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Written by Ryusuke Konishi.
8 *
9 */
10
11#include <linux/pagemap.h>
12#include <linux/buffer_head.h>
13#include <linux/writeback.h>
14#include <linux/bitops.h>
15#include <linux/bio.h>
16#include <linux/completion.h>
17#include <linux/blkdev.h>
18#include <linux/backing-dev.h>
19#include <linux/freezer.h>
20#include <linux/kthread.h>
21#include <linux/crc32.h>
22#include <linux/pagevec.h>
23#include <linux/slab.h>
24#include <linux/sched/signal.h>
25
26#include "nilfs.h"
27#include "btnode.h"
28#include "page.h"
29#include "segment.h"
30#include "sufile.h"
31#include "cpfile.h"
32#include "ifile.h"
33#include "segbuf.h"
34
35
36/*
37 * Segment constructor
38 */
39#define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
40
41#define SC_MAX_SEGDELTA 64 /*
42 * Upper limit of the number of segments
43 * appended in collection retry loop
44 */
45
46/* Construction mode */
47enum {
48 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
49 SC_LSEG_DSYNC, /*
50 * Flush data blocks of a given file and make
51 * a logical segment without a super root.
52 */
53 SC_FLUSH_FILE, /*
54 * Flush data files, leads to segment writes without
55 * creating a checkpoint.
56 */
57 SC_FLUSH_DAT, /*
58 * Flush DAT file. This also creates segments
59 * without a checkpoint.
60 */
61};
62
63/* Stage numbers of dirty block collection */
64enum {
65 NILFS_ST_INIT = 0,
66 NILFS_ST_GC, /* Collecting dirty blocks for GC */
67 NILFS_ST_FILE,
68 NILFS_ST_IFILE,
69 NILFS_ST_CPFILE,
70 NILFS_ST_SUFILE,
71 NILFS_ST_DAT,
72 NILFS_ST_SR, /* Super root */
73 NILFS_ST_DSYNC, /* Data sync blocks */
74 NILFS_ST_DONE,
75};
76
77#define CREATE_TRACE_POINTS
78#include <trace/events/nilfs2.h>
79
80/*
81 * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
82 * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
83 * the variable must use them because transition of stage count must involve
84 * trace events (trace_nilfs2_collection_stage_transition).
85 *
86 * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
87 * produce tracepoint events. It is provided just for making the intention
88 * clear.
89 */
90static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
91{
92 sci->sc_stage.scnt++;
93 trace_nilfs2_collection_stage_transition(sci);
94}
95
96static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
97{
98 sci->sc_stage.scnt = next_scnt;
99 trace_nilfs2_collection_stage_transition(sci);
100}
101
102static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
103{
104 return sci->sc_stage.scnt;
105}
106
107/* State flags of collection */
108#define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
109#define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
110#define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
111#define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
112
113/* Operations depending on the construction mode and file type */
114struct nilfs_sc_operations {
115 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
116 struct inode *);
117 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
118 struct inode *);
119 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
120 struct inode *);
121 void (*write_data_binfo)(struct nilfs_sc_info *,
122 struct nilfs_segsum_pointer *,
123 union nilfs_binfo *);
124 void (*write_node_binfo)(struct nilfs_sc_info *,
125 struct nilfs_segsum_pointer *,
126 union nilfs_binfo *);
127};
128
129/*
130 * Other definitions
131 */
132static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
133static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
134static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
135static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
136
137#define nilfs_cnt32_ge(a, b) \
138 (typecheck(__u32, a) && typecheck(__u32, b) && \
139 ((__s32)(a) - (__s32)(b) >= 0))
140
141static int nilfs_prepare_segment_lock(struct super_block *sb,
142 struct nilfs_transaction_info *ti)
143{
144 struct nilfs_transaction_info *cur_ti = current->journal_info;
145 void *save = NULL;
146
147 if (cur_ti) {
148 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
149 return ++cur_ti->ti_count;
150
151 /*
152 * If journal_info field is occupied by other FS,
153 * it is saved and will be restored on
154 * nilfs_transaction_commit().
155 */
156 nilfs_warn(sb, "journal info from a different FS");
157 save = current->journal_info;
158 }
159 if (!ti) {
160 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
161 if (!ti)
162 return -ENOMEM;
163 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
164 } else {
165 ti->ti_flags = 0;
166 }
167 ti->ti_count = 0;
168 ti->ti_save = save;
169 ti->ti_magic = NILFS_TI_MAGIC;
170 current->journal_info = ti;
171 return 0;
172}
173
174/**
175 * nilfs_transaction_begin - start indivisible file operations.
176 * @sb: super block
177 * @ti: nilfs_transaction_info
178 * @vacancy_check: flags for vacancy rate checks
179 *
180 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
181 * the segment semaphore, to make a segment construction and write tasks
182 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
183 * The region enclosed by these two functions can be nested. To avoid a
184 * deadlock, the semaphore is only acquired or released in the outermost call.
185 *
186 * This function allocates a nilfs_transaction_info struct to keep context
187 * information on it. It is initialized and hooked onto the current task in
188 * the outermost call. If a pre-allocated struct is given to @ti, it is used
189 * instead; otherwise a new struct is assigned from a slab.
190 *
191 * When @vacancy_check flag is set, this function will check the amount of
192 * free space, and will wait for the GC to reclaim disk space if low capacity.
193 *
194 * Return Value: On success, 0 is returned. On error, one of the following
195 * negative error code is returned.
196 *
197 * %-ENOMEM - Insufficient memory available.
198 *
199 * %-ENOSPC - No space left on device
200 */
201int nilfs_transaction_begin(struct super_block *sb,
202 struct nilfs_transaction_info *ti,
203 int vacancy_check)
204{
205 struct the_nilfs *nilfs;
206 int ret = nilfs_prepare_segment_lock(sb, ti);
207 struct nilfs_transaction_info *trace_ti;
208
209 if (unlikely(ret < 0))
210 return ret;
211 if (ret > 0) {
212 trace_ti = current->journal_info;
213
214 trace_nilfs2_transaction_transition(sb, trace_ti,
215 trace_ti->ti_count, trace_ti->ti_flags,
216 TRACE_NILFS2_TRANSACTION_BEGIN);
217 return 0;
218 }
219
220 sb_start_intwrite(sb);
221
222 nilfs = sb->s_fs_info;
223 down_read(&nilfs->ns_segctor_sem);
224 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
225 up_read(&nilfs->ns_segctor_sem);
226 ret = -ENOSPC;
227 goto failed;
228 }
229
230 trace_ti = current->journal_info;
231 trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
232 trace_ti->ti_flags,
233 TRACE_NILFS2_TRANSACTION_BEGIN);
234 return 0;
235
236 failed:
237 ti = current->journal_info;
238 current->journal_info = ti->ti_save;
239 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
240 kmem_cache_free(nilfs_transaction_cachep, ti);
241 sb_end_intwrite(sb);
242 return ret;
243}
244
245/**
246 * nilfs_transaction_commit - commit indivisible file operations.
247 * @sb: super block
248 *
249 * nilfs_transaction_commit() releases the read semaphore which is
250 * acquired by nilfs_transaction_begin(). This is only performed
251 * in outermost call of this function. If a commit flag is set,
252 * nilfs_transaction_commit() sets a timer to start the segment
253 * constructor. If a sync flag is set, it starts construction
254 * directly.
255 */
256int nilfs_transaction_commit(struct super_block *sb)
257{
258 struct nilfs_transaction_info *ti = current->journal_info;
259 struct the_nilfs *nilfs = sb->s_fs_info;
260 int err = 0;
261
262 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
263 ti->ti_flags |= NILFS_TI_COMMIT;
264 if (ti->ti_count > 0) {
265 ti->ti_count--;
266 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
267 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
268 return 0;
269 }
270 if (nilfs->ns_writer) {
271 struct nilfs_sc_info *sci = nilfs->ns_writer;
272
273 if (ti->ti_flags & NILFS_TI_COMMIT)
274 nilfs_segctor_start_timer(sci);
275 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
276 nilfs_segctor_do_flush(sci, 0);
277 }
278 up_read(&nilfs->ns_segctor_sem);
279 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
280 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
281
282 current->journal_info = ti->ti_save;
283
284 if (ti->ti_flags & NILFS_TI_SYNC)
285 err = nilfs_construct_segment(sb);
286 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
287 kmem_cache_free(nilfs_transaction_cachep, ti);
288 sb_end_intwrite(sb);
289 return err;
290}
291
292void nilfs_transaction_abort(struct super_block *sb)
293{
294 struct nilfs_transaction_info *ti = current->journal_info;
295 struct the_nilfs *nilfs = sb->s_fs_info;
296
297 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
298 if (ti->ti_count > 0) {
299 ti->ti_count--;
300 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
301 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
302 return;
303 }
304 up_read(&nilfs->ns_segctor_sem);
305
306 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
307 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
308
309 current->journal_info = ti->ti_save;
310 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
311 kmem_cache_free(nilfs_transaction_cachep, ti);
312 sb_end_intwrite(sb);
313}
314
315void nilfs_relax_pressure_in_lock(struct super_block *sb)
316{
317 struct the_nilfs *nilfs = sb->s_fs_info;
318 struct nilfs_sc_info *sci = nilfs->ns_writer;
319
320 if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request)
321 return;
322
323 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
324 up_read(&nilfs->ns_segctor_sem);
325
326 down_write(&nilfs->ns_segctor_sem);
327 if (sci->sc_flush_request &&
328 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
329 struct nilfs_transaction_info *ti = current->journal_info;
330
331 ti->ti_flags |= NILFS_TI_WRITER;
332 nilfs_segctor_do_immediate_flush(sci);
333 ti->ti_flags &= ~NILFS_TI_WRITER;
334 }
335 downgrade_write(&nilfs->ns_segctor_sem);
336}
337
338static void nilfs_transaction_lock(struct super_block *sb,
339 struct nilfs_transaction_info *ti,
340 int gcflag)
341{
342 struct nilfs_transaction_info *cur_ti = current->journal_info;
343 struct the_nilfs *nilfs = sb->s_fs_info;
344 struct nilfs_sc_info *sci = nilfs->ns_writer;
345
346 WARN_ON(cur_ti);
347 ti->ti_flags = NILFS_TI_WRITER;
348 ti->ti_count = 0;
349 ti->ti_save = cur_ti;
350 ti->ti_magic = NILFS_TI_MAGIC;
351 current->journal_info = ti;
352
353 for (;;) {
354 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
355 ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
356
357 down_write(&nilfs->ns_segctor_sem);
358 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
359 break;
360
361 nilfs_segctor_do_immediate_flush(sci);
362
363 up_write(&nilfs->ns_segctor_sem);
364 cond_resched();
365 }
366 if (gcflag)
367 ti->ti_flags |= NILFS_TI_GC;
368
369 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
370 ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
371}
372
373static void nilfs_transaction_unlock(struct super_block *sb)
374{
375 struct nilfs_transaction_info *ti = current->journal_info;
376 struct the_nilfs *nilfs = sb->s_fs_info;
377
378 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
379 BUG_ON(ti->ti_count > 0);
380
381 up_write(&nilfs->ns_segctor_sem);
382 current->journal_info = ti->ti_save;
383
384 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
385 ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
386}
387
388static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
389 struct nilfs_segsum_pointer *ssp,
390 unsigned int bytes)
391{
392 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
393 unsigned int blocksize = sci->sc_super->s_blocksize;
394 void *p;
395
396 if (unlikely(ssp->offset + bytes > blocksize)) {
397 ssp->offset = 0;
398 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
399 &segbuf->sb_segsum_buffers));
400 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
401 }
402 p = ssp->bh->b_data + ssp->offset;
403 ssp->offset += bytes;
404 return p;
405}
406
407/**
408 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
409 * @sci: nilfs_sc_info
410 */
411static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
412{
413 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
414 struct buffer_head *sumbh;
415 unsigned int sumbytes;
416 unsigned int flags = 0;
417 int err;
418
419 if (nilfs_doing_gc())
420 flags = NILFS_SS_GC;
421 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
422 if (unlikely(err))
423 return err;
424
425 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
426 sumbytes = segbuf->sb_sum.sumbytes;
427 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
428 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
429 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
430 return 0;
431}
432
433/**
434 * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
435 * @sci: segment constructor object
436 *
437 * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
438 * the current segment summary block.
439 */
440static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
441{
442 struct nilfs_segsum_pointer *ssp;
443
444 ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
445 if (ssp->offset < ssp->bh->b_size)
446 memset(ssp->bh->b_data + ssp->offset, 0,
447 ssp->bh->b_size - ssp->offset);
448}
449
450static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
451{
452 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
453 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
454 return -E2BIG; /*
455 * The current segment is filled up
456 * (internal code)
457 */
458 nilfs_segctor_zeropad_segsum(sci);
459 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
460 return nilfs_segctor_reset_segment_buffer(sci);
461}
462
463static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
464{
465 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
466 int err;
467
468 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
469 err = nilfs_segctor_feed_segment(sci);
470 if (err)
471 return err;
472 segbuf = sci->sc_curseg;
473 }
474 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
475 if (likely(!err))
476 segbuf->sb_sum.flags |= NILFS_SS_SR;
477 return err;
478}
479
480/*
481 * Functions for making segment summary and payloads
482 */
483static int nilfs_segctor_segsum_block_required(
484 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
485 unsigned int binfo_size)
486{
487 unsigned int blocksize = sci->sc_super->s_blocksize;
488 /* Size of finfo and binfo is enough small against blocksize */
489
490 return ssp->offset + binfo_size +
491 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
492 blocksize;
493}
494
495static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
496 struct inode *inode)
497{
498 sci->sc_curseg->sb_sum.nfinfo++;
499 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
500 nilfs_segctor_map_segsum_entry(
501 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
502
503 if (NILFS_I(inode)->i_root &&
504 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
505 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
506 /* skip finfo */
507}
508
509static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
510 struct inode *inode)
511{
512 struct nilfs_finfo *finfo;
513 struct nilfs_inode_info *ii;
514 struct nilfs_segment_buffer *segbuf;
515 __u64 cno;
516
517 if (sci->sc_blk_cnt == 0)
518 return;
519
520 ii = NILFS_I(inode);
521
522 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
523 cno = ii->i_cno;
524 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
525 cno = 0;
526 else
527 cno = sci->sc_cno;
528
529 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
530 sizeof(*finfo));
531 finfo->fi_ino = cpu_to_le64(inode->i_ino);
532 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
533 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
534 finfo->fi_cno = cpu_to_le64(cno);
535
536 segbuf = sci->sc_curseg;
537 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
538 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
539 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
540 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
541}
542
543static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
544 struct buffer_head *bh,
545 struct inode *inode,
546 unsigned int binfo_size)
547{
548 struct nilfs_segment_buffer *segbuf;
549 int required, err = 0;
550
551 retry:
552 segbuf = sci->sc_curseg;
553 required = nilfs_segctor_segsum_block_required(
554 sci, &sci->sc_binfo_ptr, binfo_size);
555 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
556 nilfs_segctor_end_finfo(sci, inode);
557 err = nilfs_segctor_feed_segment(sci);
558 if (err)
559 return err;
560 goto retry;
561 }
562 if (unlikely(required)) {
563 nilfs_segctor_zeropad_segsum(sci);
564 err = nilfs_segbuf_extend_segsum(segbuf);
565 if (unlikely(err))
566 goto failed;
567 }
568 if (sci->sc_blk_cnt == 0)
569 nilfs_segctor_begin_finfo(sci, inode);
570
571 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
572 /* Substitution to vblocknr is delayed until update_blocknr() */
573 nilfs_segbuf_add_file_buffer(segbuf, bh);
574 sci->sc_blk_cnt++;
575 failed:
576 return err;
577}
578
579/*
580 * Callback functions that enumerate, mark, and collect dirty blocks
581 */
582static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
583 struct buffer_head *bh, struct inode *inode)
584{
585 int err;
586
587 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
588 if (err < 0)
589 return err;
590
591 err = nilfs_segctor_add_file_block(sci, bh, inode,
592 sizeof(struct nilfs_binfo_v));
593 if (!err)
594 sci->sc_datablk_cnt++;
595 return err;
596}
597
598static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
599 struct buffer_head *bh,
600 struct inode *inode)
601{
602 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
603}
604
605static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
606 struct buffer_head *bh,
607 struct inode *inode)
608{
609 WARN_ON(!buffer_dirty(bh));
610 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
611}
612
613static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
614 struct nilfs_segsum_pointer *ssp,
615 union nilfs_binfo *binfo)
616{
617 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
618 sci, ssp, sizeof(*binfo_v));
619 *binfo_v = binfo->bi_v;
620}
621
622static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
623 struct nilfs_segsum_pointer *ssp,
624 union nilfs_binfo *binfo)
625{
626 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
627 sci, ssp, sizeof(*vblocknr));
628 *vblocknr = binfo->bi_v.bi_vblocknr;
629}
630
631static const struct nilfs_sc_operations nilfs_sc_file_ops = {
632 .collect_data = nilfs_collect_file_data,
633 .collect_node = nilfs_collect_file_node,
634 .collect_bmap = nilfs_collect_file_bmap,
635 .write_data_binfo = nilfs_write_file_data_binfo,
636 .write_node_binfo = nilfs_write_file_node_binfo,
637};
638
639static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
640 struct buffer_head *bh, struct inode *inode)
641{
642 int err;
643
644 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
645 if (err < 0)
646 return err;
647
648 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
649 if (!err)
650 sci->sc_datablk_cnt++;
651 return err;
652}
653
654static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
655 struct buffer_head *bh, struct inode *inode)
656{
657 WARN_ON(!buffer_dirty(bh));
658 return nilfs_segctor_add_file_block(sci, bh, inode,
659 sizeof(struct nilfs_binfo_dat));
660}
661
662static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
663 struct nilfs_segsum_pointer *ssp,
664 union nilfs_binfo *binfo)
665{
666 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
667 sizeof(*blkoff));
668 *blkoff = binfo->bi_dat.bi_blkoff;
669}
670
671static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
672 struct nilfs_segsum_pointer *ssp,
673 union nilfs_binfo *binfo)
674{
675 struct nilfs_binfo_dat *binfo_dat =
676 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
677 *binfo_dat = binfo->bi_dat;
678}
679
680static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
681 .collect_data = nilfs_collect_dat_data,
682 .collect_node = nilfs_collect_file_node,
683 .collect_bmap = nilfs_collect_dat_bmap,
684 .write_data_binfo = nilfs_write_dat_data_binfo,
685 .write_node_binfo = nilfs_write_dat_node_binfo,
686};
687
688static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
689 .collect_data = nilfs_collect_file_data,
690 .collect_node = NULL,
691 .collect_bmap = NULL,
692 .write_data_binfo = nilfs_write_file_data_binfo,
693 .write_node_binfo = NULL,
694};
695
696static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
697 struct list_head *listp,
698 size_t nlimit,
699 loff_t start, loff_t end)
700{
701 struct address_space *mapping = inode->i_mapping;
702 struct folio_batch fbatch;
703 pgoff_t index = 0, last = ULONG_MAX;
704 size_t ndirties = 0;
705 int i;
706
707 if (unlikely(start != 0 || end != LLONG_MAX)) {
708 /*
709 * A valid range is given for sync-ing data pages. The
710 * range is rounded to per-page; extra dirty buffers
711 * may be included if blocksize < pagesize.
712 */
713 index = start >> PAGE_SHIFT;
714 last = end >> PAGE_SHIFT;
715 }
716 folio_batch_init(&fbatch);
717 repeat:
718 if (unlikely(index > last) ||
719 !filemap_get_folios_tag(mapping, &index, last,
720 PAGECACHE_TAG_DIRTY, &fbatch))
721 return ndirties;
722
723 for (i = 0; i < folio_batch_count(&fbatch); i++) {
724 struct buffer_head *bh, *head;
725 struct folio *folio = fbatch.folios[i];
726
727 folio_lock(folio);
728 if (unlikely(folio->mapping != mapping)) {
729 /* Exclude folios removed from the address space */
730 folio_unlock(folio);
731 continue;
732 }
733 head = folio_buffers(folio);
734 if (!head)
735 head = create_empty_buffers(folio,
736 i_blocksize(inode), 0);
737 folio_unlock(folio);
738
739 bh = head;
740 do {
741 if (!buffer_dirty(bh) || buffer_async_write(bh))
742 continue;
743 get_bh(bh);
744 list_add_tail(&bh->b_assoc_buffers, listp);
745 ndirties++;
746 if (unlikely(ndirties >= nlimit)) {
747 folio_batch_release(&fbatch);
748 cond_resched();
749 return ndirties;
750 }
751 } while (bh = bh->b_this_page, bh != head);
752 }
753 folio_batch_release(&fbatch);
754 cond_resched();
755 goto repeat;
756}
757
758static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
759 struct list_head *listp)
760{
761 struct nilfs_inode_info *ii = NILFS_I(inode);
762 struct inode *btnc_inode = ii->i_assoc_inode;
763 struct folio_batch fbatch;
764 struct buffer_head *bh, *head;
765 unsigned int i;
766 pgoff_t index = 0;
767
768 if (!btnc_inode)
769 return;
770 folio_batch_init(&fbatch);
771
772 while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
773 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
774 for (i = 0; i < folio_batch_count(&fbatch); i++) {
775 bh = head = folio_buffers(fbatch.folios[i]);
776 do {
777 if (buffer_dirty(bh) &&
778 !buffer_async_write(bh)) {
779 get_bh(bh);
780 list_add_tail(&bh->b_assoc_buffers,
781 listp);
782 }
783 bh = bh->b_this_page;
784 } while (bh != head);
785 }
786 folio_batch_release(&fbatch);
787 cond_resched();
788 }
789}
790
791static void nilfs_dispose_list(struct the_nilfs *nilfs,
792 struct list_head *head, int force)
793{
794 struct nilfs_inode_info *ii, *n;
795 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
796 unsigned int nv = 0;
797
798 while (!list_empty(head)) {
799 spin_lock(&nilfs->ns_inode_lock);
800 list_for_each_entry_safe(ii, n, head, i_dirty) {
801 list_del_init(&ii->i_dirty);
802 if (force) {
803 if (unlikely(ii->i_bh)) {
804 brelse(ii->i_bh);
805 ii->i_bh = NULL;
806 }
807 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
808 set_bit(NILFS_I_QUEUED, &ii->i_state);
809 list_add_tail(&ii->i_dirty,
810 &nilfs->ns_dirty_files);
811 continue;
812 }
813 ivec[nv++] = ii;
814 if (nv == SC_N_INODEVEC)
815 break;
816 }
817 spin_unlock(&nilfs->ns_inode_lock);
818
819 for (pii = ivec; nv > 0; pii++, nv--)
820 iput(&(*pii)->vfs_inode);
821 }
822}
823
824static void nilfs_iput_work_func(struct work_struct *work)
825{
826 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
827 sc_iput_work);
828 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
829
830 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
831}
832
833static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
834 struct nilfs_root *root)
835{
836 int ret = 0;
837
838 if (nilfs_mdt_fetch_dirty(root->ifile))
839 ret++;
840 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
841 ret++;
842 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
843 ret++;
844 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
845 ret++;
846 return ret;
847}
848
849static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
850{
851 return list_empty(&sci->sc_dirty_files) &&
852 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
853 sci->sc_nfreesegs == 0 &&
854 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
855}
856
857static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
858{
859 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
860 int ret = 0;
861
862 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
863 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
864
865 spin_lock(&nilfs->ns_inode_lock);
866 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
867 ret++;
868
869 spin_unlock(&nilfs->ns_inode_lock);
870 return ret;
871}
872
873static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
874{
875 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
876
877 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
878 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
879 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
880 nilfs_mdt_clear_dirty(nilfs->ns_dat);
881}
882
883static void nilfs_fill_in_file_bmap(struct inode *ifile,
884 struct nilfs_inode_info *ii)
885
886{
887 struct buffer_head *ibh;
888 struct nilfs_inode *raw_inode;
889
890 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
891 ibh = ii->i_bh;
892 BUG_ON(!ibh);
893 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
894 ibh);
895 nilfs_bmap_write(ii->i_bmap, raw_inode);
896 nilfs_ifile_unmap_inode(raw_inode);
897 }
898}
899
900static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
901{
902 struct nilfs_inode_info *ii;
903
904 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
905 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
906 set_bit(NILFS_I_COLLECTED, &ii->i_state);
907 }
908}
909
910/**
911 * nilfs_write_root_mdt_inode - export root metadata inode information to
912 * the on-disk inode
913 * @inode: inode object of the root metadata file
914 * @raw_inode: on-disk inode
915 *
916 * nilfs_write_root_mdt_inode() writes inode information and bmap data of
917 * @inode to the inode area of the metadata file allocated on the super root
918 * block created to finalize the log. Since super root blocks are configured
919 * each time, this function zero-fills the unused area of @raw_inode.
920 */
921static void nilfs_write_root_mdt_inode(struct inode *inode,
922 struct nilfs_inode *raw_inode)
923{
924 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
925
926 nilfs_write_inode_common(inode, raw_inode);
927
928 /* zero-fill unused portion of raw_inode */
929 raw_inode->i_xattr = 0;
930 raw_inode->i_pad = 0;
931 memset((void *)raw_inode + sizeof(*raw_inode), 0,
932 nilfs->ns_inode_size - sizeof(*raw_inode));
933
934 nilfs_bmap_write(NILFS_I(inode)->i_bmap, raw_inode);
935}
936
937static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
938 struct the_nilfs *nilfs)
939{
940 struct buffer_head *bh_sr;
941 struct nilfs_super_root *raw_sr;
942 unsigned int isz, srsz;
943
944 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
945
946 lock_buffer(bh_sr);
947 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
948 isz = nilfs->ns_inode_size;
949 srsz = NILFS_SR_BYTES(isz);
950
951 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
952 raw_sr->sr_bytes = cpu_to_le16(srsz);
953 raw_sr->sr_nongc_ctime
954 = cpu_to_le64(nilfs_doing_gc() ?
955 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
956 raw_sr->sr_flags = 0;
957
958 nilfs_write_root_mdt_inode(nilfs->ns_dat, (void *)raw_sr +
959 NILFS_SR_DAT_OFFSET(isz));
960 nilfs_write_root_mdt_inode(nilfs->ns_cpfile, (void *)raw_sr +
961 NILFS_SR_CPFILE_OFFSET(isz));
962 nilfs_write_root_mdt_inode(nilfs->ns_sufile, (void *)raw_sr +
963 NILFS_SR_SUFILE_OFFSET(isz));
964
965 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
966 set_buffer_uptodate(bh_sr);
967 unlock_buffer(bh_sr);
968}
969
970static void nilfs_redirty_inodes(struct list_head *head)
971{
972 struct nilfs_inode_info *ii;
973
974 list_for_each_entry(ii, head, i_dirty) {
975 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
976 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
977 }
978}
979
980static void nilfs_drop_collected_inodes(struct list_head *head)
981{
982 struct nilfs_inode_info *ii;
983
984 list_for_each_entry(ii, head, i_dirty) {
985 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
986 continue;
987
988 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
989 set_bit(NILFS_I_UPDATED, &ii->i_state);
990 }
991}
992
993static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
994 struct inode *inode,
995 struct list_head *listp,
996 int (*collect)(struct nilfs_sc_info *,
997 struct buffer_head *,
998 struct inode *))
999{
1000 struct buffer_head *bh, *n;
1001 int err = 0;
1002
1003 if (collect) {
1004 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1005 list_del_init(&bh->b_assoc_buffers);
1006 err = collect(sci, bh, inode);
1007 brelse(bh);
1008 if (unlikely(err))
1009 goto dispose_buffers;
1010 }
1011 return 0;
1012 }
1013
1014 dispose_buffers:
1015 while (!list_empty(listp)) {
1016 bh = list_first_entry(listp, struct buffer_head,
1017 b_assoc_buffers);
1018 list_del_init(&bh->b_assoc_buffers);
1019 brelse(bh);
1020 }
1021 return err;
1022}
1023
1024static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1025{
1026 /* Remaining number of blocks within segment buffer */
1027 return sci->sc_segbuf_nblocks -
1028 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1029}
1030
1031static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1032 struct inode *inode,
1033 const struct nilfs_sc_operations *sc_ops)
1034{
1035 LIST_HEAD(data_buffers);
1036 LIST_HEAD(node_buffers);
1037 int err;
1038
1039 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1040 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1041
1042 n = nilfs_lookup_dirty_data_buffers(
1043 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1044 if (n > rest) {
1045 err = nilfs_segctor_apply_buffers(
1046 sci, inode, &data_buffers,
1047 sc_ops->collect_data);
1048 BUG_ON(!err); /* always receive -E2BIG or true error */
1049 goto break_or_fail;
1050 }
1051 }
1052 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1053
1054 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1055 err = nilfs_segctor_apply_buffers(
1056 sci, inode, &data_buffers, sc_ops->collect_data);
1057 if (unlikely(err)) {
1058 /* dispose node list */
1059 nilfs_segctor_apply_buffers(
1060 sci, inode, &node_buffers, NULL);
1061 goto break_or_fail;
1062 }
1063 sci->sc_stage.flags |= NILFS_CF_NODE;
1064 }
1065 /* Collect node */
1066 err = nilfs_segctor_apply_buffers(
1067 sci, inode, &node_buffers, sc_ops->collect_node);
1068 if (unlikely(err))
1069 goto break_or_fail;
1070
1071 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1072 err = nilfs_segctor_apply_buffers(
1073 sci, inode, &node_buffers, sc_ops->collect_bmap);
1074 if (unlikely(err))
1075 goto break_or_fail;
1076
1077 nilfs_segctor_end_finfo(sci, inode);
1078 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1079
1080 break_or_fail:
1081 return err;
1082}
1083
1084static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1085 struct inode *inode)
1086{
1087 LIST_HEAD(data_buffers);
1088 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1089 int err;
1090
1091 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1092 sci->sc_dsync_start,
1093 sci->sc_dsync_end);
1094
1095 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1096 nilfs_collect_file_data);
1097 if (!err) {
1098 nilfs_segctor_end_finfo(sci, inode);
1099 BUG_ON(n > rest);
1100 /* always receive -E2BIG or true error if n > rest */
1101 }
1102 return err;
1103}
1104
1105static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1106{
1107 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1108 struct list_head *head;
1109 struct nilfs_inode_info *ii;
1110 size_t ndone;
1111 int err = 0;
1112
1113 switch (nilfs_sc_cstage_get(sci)) {
1114 case NILFS_ST_INIT:
1115 /* Pre-processes */
1116 sci->sc_stage.flags = 0;
1117
1118 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1119 sci->sc_nblk_inc = 0;
1120 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1121 if (mode == SC_LSEG_DSYNC) {
1122 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1123 goto dsync_mode;
1124 }
1125 }
1126
1127 sci->sc_stage.dirty_file_ptr = NULL;
1128 sci->sc_stage.gc_inode_ptr = NULL;
1129 if (mode == SC_FLUSH_DAT) {
1130 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1131 goto dat_stage;
1132 }
1133 nilfs_sc_cstage_inc(sci);
1134 fallthrough;
1135 case NILFS_ST_GC:
1136 if (nilfs_doing_gc()) {
1137 head = &sci->sc_gc_inodes;
1138 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1139 head, i_dirty);
1140 list_for_each_entry_continue(ii, head, i_dirty) {
1141 err = nilfs_segctor_scan_file(
1142 sci, &ii->vfs_inode,
1143 &nilfs_sc_file_ops);
1144 if (unlikely(err)) {
1145 sci->sc_stage.gc_inode_ptr = list_entry(
1146 ii->i_dirty.prev,
1147 struct nilfs_inode_info,
1148 i_dirty);
1149 goto break_or_fail;
1150 }
1151 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1152 }
1153 sci->sc_stage.gc_inode_ptr = NULL;
1154 }
1155 nilfs_sc_cstage_inc(sci);
1156 fallthrough;
1157 case NILFS_ST_FILE:
1158 head = &sci->sc_dirty_files;
1159 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1160 i_dirty);
1161 list_for_each_entry_continue(ii, head, i_dirty) {
1162 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1163
1164 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1165 &nilfs_sc_file_ops);
1166 if (unlikely(err)) {
1167 sci->sc_stage.dirty_file_ptr =
1168 list_entry(ii->i_dirty.prev,
1169 struct nilfs_inode_info,
1170 i_dirty);
1171 goto break_or_fail;
1172 }
1173 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1174 /* XXX: required ? */
1175 }
1176 sci->sc_stage.dirty_file_ptr = NULL;
1177 if (mode == SC_FLUSH_FILE) {
1178 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1179 return 0;
1180 }
1181 nilfs_sc_cstage_inc(sci);
1182 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1183 fallthrough;
1184 case NILFS_ST_IFILE:
1185 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1186 &nilfs_sc_file_ops);
1187 if (unlikely(err))
1188 break;
1189 nilfs_sc_cstage_inc(sci);
1190 /* Creating a checkpoint */
1191 err = nilfs_cpfile_create_checkpoint(nilfs->ns_cpfile,
1192 nilfs->ns_cno);
1193 if (unlikely(err))
1194 break;
1195 fallthrough;
1196 case NILFS_ST_CPFILE:
1197 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1198 &nilfs_sc_file_ops);
1199 if (unlikely(err))
1200 break;
1201 nilfs_sc_cstage_inc(sci);
1202 fallthrough;
1203 case NILFS_ST_SUFILE:
1204 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1205 sci->sc_nfreesegs, &ndone);
1206 if (unlikely(err)) {
1207 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1208 sci->sc_freesegs, ndone,
1209 NULL);
1210 break;
1211 }
1212 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1213
1214 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1215 &nilfs_sc_file_ops);
1216 if (unlikely(err))
1217 break;
1218 nilfs_sc_cstage_inc(sci);
1219 fallthrough;
1220 case NILFS_ST_DAT:
1221 dat_stage:
1222 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1223 &nilfs_sc_dat_ops);
1224 if (unlikely(err))
1225 break;
1226 if (mode == SC_FLUSH_DAT) {
1227 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1228 return 0;
1229 }
1230 nilfs_sc_cstage_inc(sci);
1231 fallthrough;
1232 case NILFS_ST_SR:
1233 if (mode == SC_LSEG_SR) {
1234 /* Appending a super root */
1235 err = nilfs_segctor_add_super_root(sci);
1236 if (unlikely(err))
1237 break;
1238 }
1239 /* End of a logical segment */
1240 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1241 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1242 return 0;
1243 case NILFS_ST_DSYNC:
1244 dsync_mode:
1245 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1246 ii = sci->sc_dsync_inode;
1247 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1248 break;
1249
1250 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1251 if (unlikely(err))
1252 break;
1253 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1254 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1255 return 0;
1256 case NILFS_ST_DONE:
1257 return 0;
1258 default:
1259 BUG();
1260 }
1261
1262 break_or_fail:
1263 return err;
1264}
1265
1266/**
1267 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1268 * @sci: nilfs_sc_info
1269 * @nilfs: nilfs object
1270 */
1271static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1272 struct the_nilfs *nilfs)
1273{
1274 struct nilfs_segment_buffer *segbuf, *prev;
1275 __u64 nextnum;
1276 int err, alloc = 0;
1277
1278 segbuf = nilfs_segbuf_new(sci->sc_super);
1279 if (unlikely(!segbuf))
1280 return -ENOMEM;
1281
1282 if (list_empty(&sci->sc_write_logs)) {
1283 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1284 nilfs->ns_pseg_offset, nilfs);
1285 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1286 nilfs_shift_to_next_segment(nilfs);
1287 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1288 }
1289
1290 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1291 nextnum = nilfs->ns_nextnum;
1292
1293 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1294 /* Start from the head of a new full segment */
1295 alloc++;
1296 } else {
1297 /* Continue logs */
1298 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1299 nilfs_segbuf_map_cont(segbuf, prev);
1300 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1301 nextnum = prev->sb_nextnum;
1302
1303 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1304 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1305 segbuf->sb_sum.seg_seq++;
1306 alloc++;
1307 }
1308 }
1309
1310 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1311 if (err)
1312 goto failed;
1313
1314 if (alloc) {
1315 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1316 if (err)
1317 goto failed;
1318 }
1319 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1320
1321 BUG_ON(!list_empty(&sci->sc_segbufs));
1322 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1323 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1324 return 0;
1325
1326 failed:
1327 nilfs_segbuf_free(segbuf);
1328 return err;
1329}
1330
1331static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1332 struct the_nilfs *nilfs, int nadd)
1333{
1334 struct nilfs_segment_buffer *segbuf, *prev;
1335 struct inode *sufile = nilfs->ns_sufile;
1336 __u64 nextnextnum;
1337 LIST_HEAD(list);
1338 int err, ret, i;
1339
1340 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1341 /*
1342 * Since the segment specified with nextnum might be allocated during
1343 * the previous construction, the buffer including its segusage may
1344 * not be dirty. The following call ensures that the buffer is dirty
1345 * and will pin the buffer on memory until the sufile is written.
1346 */
1347 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1348 if (unlikely(err))
1349 return err;
1350
1351 for (i = 0; i < nadd; i++) {
1352 /* extend segment info */
1353 err = -ENOMEM;
1354 segbuf = nilfs_segbuf_new(sci->sc_super);
1355 if (unlikely(!segbuf))
1356 goto failed;
1357
1358 /* map this buffer to region of segment on-disk */
1359 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1360 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1361
1362 /* allocate the next next full segment */
1363 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1364 if (unlikely(err))
1365 goto failed_segbuf;
1366
1367 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1368 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1369
1370 list_add_tail(&segbuf->sb_list, &list);
1371 prev = segbuf;
1372 }
1373 list_splice_tail(&list, &sci->sc_segbufs);
1374 return 0;
1375
1376 failed_segbuf:
1377 nilfs_segbuf_free(segbuf);
1378 failed:
1379 list_for_each_entry(segbuf, &list, sb_list) {
1380 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1381 WARN_ON(ret); /* never fails */
1382 }
1383 nilfs_destroy_logs(&list);
1384 return err;
1385}
1386
1387static void nilfs_free_incomplete_logs(struct list_head *logs,
1388 struct the_nilfs *nilfs)
1389{
1390 struct nilfs_segment_buffer *segbuf, *prev;
1391 struct inode *sufile = nilfs->ns_sufile;
1392 int ret;
1393
1394 segbuf = NILFS_FIRST_SEGBUF(logs);
1395 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1396 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1397 WARN_ON(ret); /* never fails */
1398 }
1399 if (atomic_read(&segbuf->sb_err)) {
1400 /* Case 1: The first segment failed */
1401 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1402 /*
1403 * Case 1a: Partial segment appended into an existing
1404 * segment
1405 */
1406 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1407 segbuf->sb_fseg_end);
1408 else /* Case 1b: New full segment */
1409 set_nilfs_discontinued(nilfs);
1410 }
1411
1412 prev = segbuf;
1413 list_for_each_entry_continue(segbuf, logs, sb_list) {
1414 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1415 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1416 WARN_ON(ret); /* never fails */
1417 }
1418 if (atomic_read(&segbuf->sb_err) &&
1419 segbuf->sb_segnum != nilfs->ns_nextnum)
1420 /* Case 2: extended segment (!= next) failed */
1421 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1422 prev = segbuf;
1423 }
1424}
1425
1426static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1427 struct inode *sufile)
1428{
1429 struct nilfs_segment_buffer *segbuf;
1430 unsigned long live_blocks;
1431 int ret;
1432
1433 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1434 live_blocks = segbuf->sb_sum.nblocks +
1435 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1436 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1437 live_blocks,
1438 sci->sc_seg_ctime);
1439 WARN_ON(ret); /* always succeed because the segusage is dirty */
1440 }
1441}
1442
1443static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1444{
1445 struct nilfs_segment_buffer *segbuf;
1446 int ret;
1447
1448 segbuf = NILFS_FIRST_SEGBUF(logs);
1449 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1450 segbuf->sb_pseg_start -
1451 segbuf->sb_fseg_start, 0);
1452 WARN_ON(ret); /* always succeed because the segusage is dirty */
1453
1454 list_for_each_entry_continue(segbuf, logs, sb_list) {
1455 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1456 0, 0);
1457 WARN_ON(ret); /* always succeed */
1458 }
1459}
1460
1461static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1462 struct nilfs_segment_buffer *last,
1463 struct inode *sufile)
1464{
1465 struct nilfs_segment_buffer *segbuf = last;
1466 int ret;
1467
1468 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1469 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1470 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1471 WARN_ON(ret);
1472 }
1473 nilfs_truncate_logs(&sci->sc_segbufs, last);
1474}
1475
1476
1477static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1478 struct the_nilfs *nilfs, int mode)
1479{
1480 struct nilfs_cstage prev_stage = sci->sc_stage;
1481 int err, nadd = 1;
1482
1483 /* Collection retry loop */
1484 for (;;) {
1485 sci->sc_nblk_this_inc = 0;
1486 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1487
1488 err = nilfs_segctor_reset_segment_buffer(sci);
1489 if (unlikely(err))
1490 goto failed;
1491
1492 err = nilfs_segctor_collect_blocks(sci, mode);
1493 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1494 if (!err)
1495 break;
1496
1497 if (unlikely(err != -E2BIG))
1498 goto failed;
1499
1500 /* The current segment is filled up */
1501 if (mode != SC_LSEG_SR ||
1502 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1503 break;
1504
1505 nilfs_clear_logs(&sci->sc_segbufs);
1506
1507 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1508 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1509 sci->sc_freesegs,
1510 sci->sc_nfreesegs,
1511 NULL);
1512 WARN_ON(err); /* do not happen */
1513 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1514 }
1515
1516 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1517 if (unlikely(err))
1518 return err;
1519
1520 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1521 sci->sc_stage = prev_stage;
1522 }
1523 nilfs_segctor_zeropad_segsum(sci);
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 const 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 == segbuf->sb_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 inode = bh->b_folio->mapping->host;
1575
1576 if (mode == SC_LSEG_DSYNC)
1577 sc_op = &nilfs_sc_dsync_ops;
1578 else if (ino == NILFS_DAT_INO)
1579 sc_op = &nilfs_sc_dat_ops;
1580 else /* file blocks */
1581 sc_op = &nilfs_sc_file_ops;
1582 }
1583 bh_org = bh;
1584 get_bh(bh_org);
1585 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1586 &binfo);
1587 if (bh != bh_org)
1588 nilfs_list_replace_buffer(bh_org, bh);
1589 brelse(bh_org);
1590 if (unlikely(err))
1591 goto failed_bmap;
1592
1593 if (ndatablk > 0)
1594 sc_op->write_data_binfo(sci, &ssp, &binfo);
1595 else
1596 sc_op->write_node_binfo(sci, &ssp, &binfo);
1597
1598 blocknr++;
1599 if (--nblocks == 0) {
1600 finfo = NULL;
1601 if (--nfinfo == 0)
1602 break;
1603 } else if (ndatablk > 0)
1604 ndatablk--;
1605 }
1606 out:
1607 return 0;
1608
1609 failed_bmap:
1610 return err;
1611}
1612
1613static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1614{
1615 struct nilfs_segment_buffer *segbuf;
1616 int err;
1617
1618 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1619 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1620 if (unlikely(err))
1621 return err;
1622 nilfs_segbuf_fill_in_segsum(segbuf);
1623 }
1624 return 0;
1625}
1626
1627static void nilfs_begin_folio_io(struct folio *folio)
1628{
1629 if (!folio || folio_test_writeback(folio))
1630 /*
1631 * For split b-tree node pages, this function may be called
1632 * twice. We ignore the 2nd or later calls by this check.
1633 */
1634 return;
1635
1636 folio_lock(folio);
1637 folio_clear_dirty_for_io(folio);
1638 folio_start_writeback(folio);
1639 folio_unlock(folio);
1640}
1641
1642static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1643{
1644 struct nilfs_segment_buffer *segbuf;
1645 struct folio *bd_folio = NULL, *fs_folio = NULL;
1646
1647 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1648 struct buffer_head *bh;
1649
1650 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1651 b_assoc_buffers) {
1652 if (bh->b_folio != bd_folio) {
1653 if (bd_folio) {
1654 folio_lock(bd_folio);
1655 folio_clear_dirty_for_io(bd_folio);
1656 folio_start_writeback(bd_folio);
1657 folio_unlock(bd_folio);
1658 }
1659 bd_folio = bh->b_folio;
1660 }
1661 }
1662
1663 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1664 b_assoc_buffers) {
1665 if (bh == segbuf->sb_super_root) {
1666 if (bh->b_folio != bd_folio) {
1667 folio_lock(bd_folio);
1668 folio_clear_dirty_for_io(bd_folio);
1669 folio_start_writeback(bd_folio);
1670 folio_unlock(bd_folio);
1671 bd_folio = bh->b_folio;
1672 }
1673 break;
1674 }
1675 set_buffer_async_write(bh);
1676 if (bh->b_folio != fs_folio) {
1677 nilfs_begin_folio_io(fs_folio);
1678 fs_folio = bh->b_folio;
1679 }
1680 }
1681 }
1682 if (bd_folio) {
1683 folio_lock(bd_folio);
1684 folio_clear_dirty_for_io(bd_folio);
1685 folio_start_writeback(bd_folio);
1686 folio_unlock(bd_folio);
1687 }
1688 nilfs_begin_folio_io(fs_folio);
1689}
1690
1691static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1692 struct the_nilfs *nilfs)
1693{
1694 int ret;
1695
1696 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1697 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1698 return ret;
1699}
1700
1701static void nilfs_end_folio_io(struct folio *folio, int err)
1702{
1703 if (!folio)
1704 return;
1705
1706 if (buffer_nilfs_node(folio_buffers(folio)) &&
1707 !folio_test_writeback(folio)) {
1708 /*
1709 * For b-tree node pages, this function may be called twice
1710 * or more because they might be split in a segment.
1711 */
1712 if (folio_test_dirty(folio)) {
1713 /*
1714 * For pages holding split b-tree node buffers, dirty
1715 * flag on the buffers may be cleared discretely.
1716 * In that case, the page is once redirtied for
1717 * remaining buffers, and it must be cancelled if
1718 * all the buffers get cleaned later.
1719 */
1720 folio_lock(folio);
1721 if (nilfs_folio_buffers_clean(folio))
1722 __nilfs_clear_folio_dirty(folio);
1723 folio_unlock(folio);
1724 }
1725 return;
1726 }
1727
1728 if (!err) {
1729 if (!nilfs_folio_buffers_clean(folio))
1730 filemap_dirty_folio(folio->mapping, folio);
1731 folio_clear_error(folio);
1732 } else {
1733 filemap_dirty_folio(folio->mapping, folio);
1734 folio_set_error(folio);
1735 }
1736
1737 folio_end_writeback(folio);
1738}
1739
1740static void nilfs_abort_logs(struct list_head *logs, int err)
1741{
1742 struct nilfs_segment_buffer *segbuf;
1743 struct folio *bd_folio = NULL, *fs_folio = NULL;
1744 struct buffer_head *bh;
1745
1746 if (list_empty(logs))
1747 return;
1748
1749 list_for_each_entry(segbuf, logs, sb_list) {
1750 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1751 b_assoc_buffers) {
1752 clear_buffer_uptodate(bh);
1753 if (bh->b_folio != bd_folio) {
1754 if (bd_folio)
1755 folio_end_writeback(bd_folio);
1756 bd_folio = bh->b_folio;
1757 }
1758 }
1759
1760 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1761 b_assoc_buffers) {
1762 if (bh == segbuf->sb_super_root) {
1763 clear_buffer_uptodate(bh);
1764 if (bh->b_folio != bd_folio) {
1765 folio_end_writeback(bd_folio);
1766 bd_folio = bh->b_folio;
1767 }
1768 break;
1769 }
1770 clear_buffer_async_write(bh);
1771 if (bh->b_folio != fs_folio) {
1772 nilfs_end_folio_io(fs_folio, err);
1773 fs_folio = bh->b_folio;
1774 }
1775 }
1776 }
1777 if (bd_folio)
1778 folio_end_writeback(bd_folio);
1779
1780 nilfs_end_folio_io(fs_folio, err);
1781}
1782
1783static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1784 struct the_nilfs *nilfs, int err)
1785{
1786 LIST_HEAD(logs);
1787 int ret;
1788
1789 list_splice_tail_init(&sci->sc_write_logs, &logs);
1790 ret = nilfs_wait_on_logs(&logs);
1791 nilfs_abort_logs(&logs, ret ? : err);
1792
1793 list_splice_tail_init(&sci->sc_segbufs, &logs);
1794 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1795 nilfs_free_incomplete_logs(&logs, nilfs);
1796
1797 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1798 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1799 sci->sc_freesegs,
1800 sci->sc_nfreesegs,
1801 NULL);
1802 WARN_ON(ret); /* do not happen */
1803 }
1804
1805 nilfs_destroy_logs(&logs);
1806}
1807
1808static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1809 struct nilfs_segment_buffer *segbuf)
1810{
1811 nilfs->ns_segnum = segbuf->sb_segnum;
1812 nilfs->ns_nextnum = segbuf->sb_nextnum;
1813 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1814 + segbuf->sb_sum.nblocks;
1815 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1816 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1817}
1818
1819static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1820{
1821 struct nilfs_segment_buffer *segbuf;
1822 struct folio *bd_folio = NULL, *fs_folio = NULL;
1823 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1824 int update_sr = false;
1825
1826 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1827 struct buffer_head *bh;
1828
1829 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1830 b_assoc_buffers) {
1831 set_buffer_uptodate(bh);
1832 clear_buffer_dirty(bh);
1833 if (bh->b_folio != bd_folio) {
1834 if (bd_folio)
1835 folio_end_writeback(bd_folio);
1836 bd_folio = bh->b_folio;
1837 }
1838 }
1839 /*
1840 * We assume that the buffers which belong to the same folio
1841 * continue over the buffer list.
1842 * Under this assumption, the last BHs of folios is
1843 * identifiable by the discontinuity of bh->b_folio
1844 * (folio != fs_folio).
1845 *
1846 * For B-tree node blocks, however, this assumption is not
1847 * guaranteed. The cleanup code of B-tree node folios needs
1848 * special care.
1849 */
1850 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1851 b_assoc_buffers) {
1852 const unsigned long set_bits = BIT(BH_Uptodate);
1853 const unsigned long clear_bits =
1854 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1855 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1856 BIT(BH_NILFS_Redirected));
1857
1858 if (bh == segbuf->sb_super_root) {
1859 set_buffer_uptodate(bh);
1860 clear_buffer_dirty(bh);
1861 if (bh->b_folio != bd_folio) {
1862 folio_end_writeback(bd_folio);
1863 bd_folio = bh->b_folio;
1864 }
1865 update_sr = true;
1866 break;
1867 }
1868 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1869 if (bh->b_folio != fs_folio) {
1870 nilfs_end_folio_io(fs_folio, 0);
1871 fs_folio = bh->b_folio;
1872 }
1873 }
1874
1875 if (!nilfs_segbuf_simplex(segbuf)) {
1876 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1877 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1878 sci->sc_lseg_stime = jiffies;
1879 }
1880 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1881 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1882 }
1883 }
1884 /*
1885 * Since folios may continue over multiple segment buffers,
1886 * end of the last folio must be checked outside of the loop.
1887 */
1888 if (bd_folio)
1889 folio_end_writeback(bd_folio);
1890
1891 nilfs_end_folio_io(fs_folio, 0);
1892
1893 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1894
1895 if (nilfs_doing_gc())
1896 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1897 else
1898 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1899
1900 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1901
1902 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1903 nilfs_set_next_segment(nilfs, segbuf);
1904
1905 if (update_sr) {
1906 nilfs->ns_flushed_device = 0;
1907 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1908 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1909
1910 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1911 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1912 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1913 nilfs_segctor_clear_metadata_dirty(sci);
1914 } else
1915 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1916}
1917
1918static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1919{
1920 int ret;
1921
1922 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1923 if (!ret) {
1924 nilfs_segctor_complete_write(sci);
1925 nilfs_destroy_logs(&sci->sc_write_logs);
1926 }
1927 return ret;
1928}
1929
1930static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1931 struct the_nilfs *nilfs)
1932{
1933 struct nilfs_inode_info *ii, *n;
1934 struct inode *ifile = sci->sc_root->ifile;
1935
1936 spin_lock(&nilfs->ns_inode_lock);
1937 retry:
1938 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1939 if (!ii->i_bh) {
1940 struct buffer_head *ibh;
1941 int err;
1942
1943 spin_unlock(&nilfs->ns_inode_lock);
1944 err = nilfs_ifile_get_inode_block(
1945 ifile, ii->vfs_inode.i_ino, &ibh);
1946 if (unlikely(err)) {
1947 nilfs_warn(sci->sc_super,
1948 "log writer: error %d getting inode block (ino=%lu)",
1949 err, ii->vfs_inode.i_ino);
1950 return err;
1951 }
1952 spin_lock(&nilfs->ns_inode_lock);
1953 if (likely(!ii->i_bh))
1954 ii->i_bh = ibh;
1955 else
1956 brelse(ibh);
1957 goto retry;
1958 }
1959
1960 // Always redirty the buffer to avoid race condition
1961 mark_buffer_dirty(ii->i_bh);
1962 nilfs_mdt_mark_dirty(ifile);
1963
1964 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1965 set_bit(NILFS_I_BUSY, &ii->i_state);
1966 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1967 }
1968 spin_unlock(&nilfs->ns_inode_lock);
1969
1970 return 0;
1971}
1972
1973static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1974 struct the_nilfs *nilfs)
1975{
1976 struct nilfs_inode_info *ii, *n;
1977 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1978 int defer_iput = false;
1979
1980 spin_lock(&nilfs->ns_inode_lock);
1981 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1982 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1983 test_bit(NILFS_I_DIRTY, &ii->i_state))
1984 continue;
1985
1986 clear_bit(NILFS_I_BUSY, &ii->i_state);
1987 brelse(ii->i_bh);
1988 ii->i_bh = NULL;
1989 list_del_init(&ii->i_dirty);
1990 if (!ii->vfs_inode.i_nlink || during_mount) {
1991 /*
1992 * Defer calling iput() to avoid deadlocks if
1993 * i_nlink == 0 or mount is not yet finished.
1994 */
1995 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
1996 defer_iput = true;
1997 } else {
1998 spin_unlock(&nilfs->ns_inode_lock);
1999 iput(&ii->vfs_inode);
2000 spin_lock(&nilfs->ns_inode_lock);
2001 }
2002 }
2003 spin_unlock(&nilfs->ns_inode_lock);
2004
2005 if (defer_iput)
2006 schedule_work(&sci->sc_iput_work);
2007}
2008
2009/*
2010 * Main procedure of segment constructor
2011 */
2012static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2013{
2014 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2015 int err;
2016
2017 if (sb_rdonly(sci->sc_super))
2018 return -EROFS;
2019
2020 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2021 sci->sc_cno = nilfs->ns_cno;
2022
2023 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2024 if (unlikely(err))
2025 goto out;
2026
2027 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2028 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2029
2030 if (nilfs_segctor_clean(sci))
2031 goto out;
2032
2033 do {
2034 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2035
2036 err = nilfs_segctor_begin_construction(sci, nilfs);
2037 if (unlikely(err))
2038 goto out;
2039
2040 /* Update time stamp */
2041 sci->sc_seg_ctime = ktime_get_real_seconds();
2042
2043 err = nilfs_segctor_collect(sci, nilfs, mode);
2044 if (unlikely(err))
2045 goto failed;
2046
2047 /* Avoid empty segment */
2048 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2049 nilfs_segbuf_empty(sci->sc_curseg)) {
2050 nilfs_segctor_abort_construction(sci, nilfs, 1);
2051 goto out;
2052 }
2053
2054 err = nilfs_segctor_assign(sci, mode);
2055 if (unlikely(err))
2056 goto failed;
2057
2058 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2059 nilfs_segctor_fill_in_file_bmap(sci);
2060
2061 if (mode == SC_LSEG_SR &&
2062 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2063 err = nilfs_cpfile_finalize_checkpoint(
2064 nilfs->ns_cpfile, nilfs->ns_cno, sci->sc_root,
2065 sci->sc_nblk_inc + sci->sc_nblk_this_inc,
2066 sci->sc_seg_ctime,
2067 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags));
2068 if (unlikely(err))
2069 goto failed_to_write;
2070
2071 nilfs_segctor_fill_in_super_root(sci, nilfs);
2072 }
2073 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2074
2075 /* Write partial segments */
2076 nilfs_segctor_prepare_write(sci);
2077
2078 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2079 nilfs->ns_crc_seed);
2080
2081 err = nilfs_segctor_write(sci, nilfs);
2082 if (unlikely(err))
2083 goto failed_to_write;
2084
2085 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2086 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2087 /*
2088 * At this point, we avoid double buffering
2089 * for blocksize < pagesize because page dirty
2090 * flag is turned off during write and dirty
2091 * buffers are not properly collected for
2092 * pages crossing over segments.
2093 */
2094 err = nilfs_segctor_wait(sci);
2095 if (err)
2096 goto failed_to_write;
2097 }
2098 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2099
2100 out:
2101 nilfs_segctor_drop_written_files(sci, nilfs);
2102 return err;
2103
2104 failed_to_write:
2105 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2106 nilfs_redirty_inodes(&sci->sc_dirty_files);
2107
2108 failed:
2109 if (nilfs_doing_gc())
2110 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2111 nilfs_segctor_abort_construction(sci, nilfs, err);
2112 goto out;
2113}
2114
2115/**
2116 * nilfs_segctor_start_timer - set timer of background write
2117 * @sci: nilfs_sc_info
2118 *
2119 * If the timer has already been set, it ignores the new request.
2120 * This function MUST be called within a section locking the segment
2121 * semaphore.
2122 */
2123static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2124{
2125 spin_lock(&sci->sc_state_lock);
2126 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2127 if (sci->sc_task) {
2128 sci->sc_timer.expires = jiffies + sci->sc_interval;
2129 add_timer(&sci->sc_timer);
2130 }
2131 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2132 }
2133 spin_unlock(&sci->sc_state_lock);
2134}
2135
2136static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2137{
2138 spin_lock(&sci->sc_state_lock);
2139 if (!(sci->sc_flush_request & BIT(bn))) {
2140 unsigned long prev_req = sci->sc_flush_request;
2141
2142 sci->sc_flush_request |= BIT(bn);
2143 if (!prev_req)
2144 wake_up(&sci->sc_wait_daemon);
2145 }
2146 spin_unlock(&sci->sc_state_lock);
2147}
2148
2149/**
2150 * nilfs_flush_segment - trigger a segment construction for resource control
2151 * @sb: super block
2152 * @ino: inode number of the file to be flushed out.
2153 */
2154void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2155{
2156 struct the_nilfs *nilfs = sb->s_fs_info;
2157 struct nilfs_sc_info *sci = nilfs->ns_writer;
2158
2159 if (!sci || nilfs_doing_construction())
2160 return;
2161 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2162 /* assign bit 0 to data files */
2163}
2164
2165struct nilfs_segctor_wait_request {
2166 wait_queue_entry_t wq;
2167 __u32 seq;
2168 int err;
2169 atomic_t done;
2170};
2171
2172static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2173{
2174 struct nilfs_segctor_wait_request wait_req;
2175 int err = 0;
2176
2177 init_wait(&wait_req.wq);
2178 wait_req.err = 0;
2179 atomic_set(&wait_req.done, 0);
2180 init_waitqueue_entry(&wait_req.wq, current);
2181
2182 /*
2183 * To prevent a race issue where completion notifications from the
2184 * log writer thread are missed, increment the request sequence count
2185 * "sc_seq_request" and insert a wait queue entry using the current
2186 * sequence number into the "sc_wait_request" queue at the same time
2187 * within the lock section of "sc_state_lock".
2188 */
2189 spin_lock(&sci->sc_state_lock);
2190 wait_req.seq = ++sci->sc_seq_request;
2191 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2192 spin_unlock(&sci->sc_state_lock);
2193
2194 wake_up(&sci->sc_wait_daemon);
2195
2196 for (;;) {
2197 set_current_state(TASK_INTERRUPTIBLE);
2198
2199 /*
2200 * Synchronize only while the log writer thread is alive.
2201 * Leave flushing out after the log writer thread exits to
2202 * the cleanup work in nilfs_segctor_destroy().
2203 */
2204 if (!sci->sc_task)
2205 break;
2206
2207 if (atomic_read(&wait_req.done)) {
2208 err = wait_req.err;
2209 break;
2210 }
2211 if (!signal_pending(current)) {
2212 schedule();
2213 continue;
2214 }
2215 err = -ERESTARTSYS;
2216 break;
2217 }
2218 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2219 return err;
2220}
2221
2222static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err, bool force)
2223{
2224 struct nilfs_segctor_wait_request *wrq, *n;
2225 unsigned long flags;
2226
2227 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2228 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2229 if (!atomic_read(&wrq->done) &&
2230 (force || nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq))) {
2231 wrq->err = err;
2232 atomic_set(&wrq->done, 1);
2233 }
2234 if (atomic_read(&wrq->done)) {
2235 wrq->wq.func(&wrq->wq,
2236 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2237 0, NULL);
2238 }
2239 }
2240 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2241}
2242
2243/**
2244 * nilfs_construct_segment - construct a logical segment
2245 * @sb: super block
2246 *
2247 * Return Value: On success, 0 is returned. On errors, one of the following
2248 * negative error code is returned.
2249 *
2250 * %-EROFS - Read only filesystem.
2251 *
2252 * %-EIO - I/O error
2253 *
2254 * %-ENOSPC - No space left on device (only in a panic state).
2255 *
2256 * %-ERESTARTSYS - Interrupted.
2257 *
2258 * %-ENOMEM - Insufficient memory available.
2259 */
2260int nilfs_construct_segment(struct super_block *sb)
2261{
2262 struct the_nilfs *nilfs = sb->s_fs_info;
2263 struct nilfs_sc_info *sci = nilfs->ns_writer;
2264 struct nilfs_transaction_info *ti;
2265
2266 if (sb_rdonly(sb) || unlikely(!sci))
2267 return -EROFS;
2268
2269 /* A call inside transactions causes a deadlock. */
2270 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2271
2272 return nilfs_segctor_sync(sci);
2273}
2274
2275/**
2276 * nilfs_construct_dsync_segment - construct a data-only logical segment
2277 * @sb: super block
2278 * @inode: inode whose data blocks should be written out
2279 * @start: start byte offset
2280 * @end: end byte offset (inclusive)
2281 *
2282 * Return Value: On success, 0 is returned. On errors, one of the following
2283 * negative error code is returned.
2284 *
2285 * %-EROFS - Read only filesystem.
2286 *
2287 * %-EIO - I/O error
2288 *
2289 * %-ENOSPC - No space left on device (only in a panic state).
2290 *
2291 * %-ERESTARTSYS - Interrupted.
2292 *
2293 * %-ENOMEM - Insufficient memory available.
2294 */
2295int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2296 loff_t start, loff_t end)
2297{
2298 struct the_nilfs *nilfs = sb->s_fs_info;
2299 struct nilfs_sc_info *sci = nilfs->ns_writer;
2300 struct nilfs_inode_info *ii;
2301 struct nilfs_transaction_info ti;
2302 int err = 0;
2303
2304 if (sb_rdonly(sb) || unlikely(!sci))
2305 return -EROFS;
2306
2307 nilfs_transaction_lock(sb, &ti, 0);
2308
2309 ii = NILFS_I(inode);
2310 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2311 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2312 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2313 nilfs_discontinued(nilfs)) {
2314 nilfs_transaction_unlock(sb);
2315 err = nilfs_segctor_sync(sci);
2316 return err;
2317 }
2318
2319 spin_lock(&nilfs->ns_inode_lock);
2320 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2321 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2322 spin_unlock(&nilfs->ns_inode_lock);
2323 nilfs_transaction_unlock(sb);
2324 return 0;
2325 }
2326 spin_unlock(&nilfs->ns_inode_lock);
2327 sci->sc_dsync_inode = ii;
2328 sci->sc_dsync_start = start;
2329 sci->sc_dsync_end = end;
2330
2331 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2332 if (!err)
2333 nilfs->ns_flushed_device = 0;
2334
2335 nilfs_transaction_unlock(sb);
2336 return err;
2337}
2338
2339#define FLUSH_FILE_BIT (0x1) /* data file only */
2340#define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2341
2342/**
2343 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2344 * @sci: segment constructor object
2345 */
2346static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2347{
2348 bool thread_is_alive;
2349
2350 spin_lock(&sci->sc_state_lock);
2351 sci->sc_seq_accepted = sci->sc_seq_request;
2352 thread_is_alive = (bool)sci->sc_task;
2353 spin_unlock(&sci->sc_state_lock);
2354
2355 /*
2356 * This function does not race with the log writer thread's
2357 * termination. Therefore, deleting sc_timer, which should not be
2358 * done after the log writer thread exits, can be done safely outside
2359 * the area protected by sc_state_lock.
2360 */
2361 if (thread_is_alive)
2362 del_timer_sync(&sci->sc_timer);
2363}
2364
2365/**
2366 * nilfs_segctor_notify - notify the result of request to caller threads
2367 * @sci: segment constructor object
2368 * @mode: mode of log forming
2369 * @err: error code to be notified
2370 */
2371static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2372{
2373 /* Clear requests (even when the construction failed) */
2374 spin_lock(&sci->sc_state_lock);
2375
2376 if (mode == SC_LSEG_SR) {
2377 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2378 sci->sc_seq_done = sci->sc_seq_accepted;
2379 nilfs_segctor_wakeup(sci, err, false);
2380 sci->sc_flush_request = 0;
2381 } else {
2382 if (mode == SC_FLUSH_FILE)
2383 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2384 else if (mode == SC_FLUSH_DAT)
2385 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2386
2387 /* re-enable timer if checkpoint creation was not done */
2388 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) && sci->sc_task &&
2389 time_before(jiffies, sci->sc_timer.expires))
2390 add_timer(&sci->sc_timer);
2391 }
2392 spin_unlock(&sci->sc_state_lock);
2393}
2394
2395/**
2396 * nilfs_segctor_construct - form logs and write them to disk
2397 * @sci: segment constructor object
2398 * @mode: mode of log forming
2399 */
2400static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2401{
2402 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2403 struct nilfs_super_block **sbp;
2404 int err = 0;
2405
2406 nilfs_segctor_accept(sci);
2407
2408 if (nilfs_discontinued(nilfs))
2409 mode = SC_LSEG_SR;
2410 if (!nilfs_segctor_confirm(sci))
2411 err = nilfs_segctor_do_construct(sci, mode);
2412
2413 if (likely(!err)) {
2414 if (mode != SC_FLUSH_DAT)
2415 atomic_set(&nilfs->ns_ndirtyblks, 0);
2416 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2417 nilfs_discontinued(nilfs)) {
2418 down_write(&nilfs->ns_sem);
2419 err = -EIO;
2420 sbp = nilfs_prepare_super(sci->sc_super,
2421 nilfs_sb_will_flip(nilfs));
2422 if (likely(sbp)) {
2423 nilfs_set_log_cursor(sbp[0], nilfs);
2424 err = nilfs_commit_super(sci->sc_super,
2425 NILFS_SB_COMMIT);
2426 }
2427 up_write(&nilfs->ns_sem);
2428 }
2429 }
2430
2431 nilfs_segctor_notify(sci, mode, err);
2432 return err;
2433}
2434
2435static void nilfs_construction_timeout(struct timer_list *t)
2436{
2437 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2438
2439 wake_up_process(sci->sc_timer_task);
2440}
2441
2442static void
2443nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2444{
2445 struct nilfs_inode_info *ii, *n;
2446
2447 list_for_each_entry_safe(ii, n, head, i_dirty) {
2448 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2449 continue;
2450 list_del_init(&ii->i_dirty);
2451 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2452 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2453 iput(&ii->vfs_inode);
2454 }
2455}
2456
2457int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2458 void **kbufs)
2459{
2460 struct the_nilfs *nilfs = sb->s_fs_info;
2461 struct nilfs_sc_info *sci = nilfs->ns_writer;
2462 struct nilfs_transaction_info ti;
2463 int err;
2464
2465 if (unlikely(!sci))
2466 return -EROFS;
2467
2468 nilfs_transaction_lock(sb, &ti, 1);
2469
2470 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2471 if (unlikely(err))
2472 goto out_unlock;
2473
2474 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2475 if (unlikely(err)) {
2476 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2477 goto out_unlock;
2478 }
2479
2480 sci->sc_freesegs = kbufs[4];
2481 sci->sc_nfreesegs = argv[4].v_nmembs;
2482 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2483
2484 for (;;) {
2485 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2486 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2487
2488 if (likely(!err))
2489 break;
2490
2491 nilfs_warn(sb, "error %d cleaning segments", err);
2492 set_current_state(TASK_INTERRUPTIBLE);
2493 schedule_timeout(sci->sc_interval);
2494 }
2495 if (nilfs_test_opt(nilfs, DISCARD)) {
2496 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2497 sci->sc_nfreesegs);
2498 if (ret) {
2499 nilfs_warn(sb,
2500 "error %d on discard request, turning discards off for the device",
2501 ret);
2502 nilfs_clear_opt(nilfs, DISCARD);
2503 }
2504 }
2505
2506 out_unlock:
2507 sci->sc_freesegs = NULL;
2508 sci->sc_nfreesegs = 0;
2509 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2510 nilfs_transaction_unlock(sb);
2511 return err;
2512}
2513
2514static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2515{
2516 struct nilfs_transaction_info ti;
2517
2518 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2519 nilfs_segctor_construct(sci, mode);
2520
2521 /*
2522 * Unclosed segment should be retried. We do this using sc_timer.
2523 * Timeout of sc_timer will invoke complete construction which leads
2524 * to close the current logical segment.
2525 */
2526 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2527 nilfs_segctor_start_timer(sci);
2528
2529 nilfs_transaction_unlock(sci->sc_super);
2530}
2531
2532static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2533{
2534 int mode = 0;
2535
2536 spin_lock(&sci->sc_state_lock);
2537 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2538 SC_FLUSH_DAT : SC_FLUSH_FILE;
2539 spin_unlock(&sci->sc_state_lock);
2540
2541 if (mode) {
2542 nilfs_segctor_do_construct(sci, mode);
2543
2544 spin_lock(&sci->sc_state_lock);
2545 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2546 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2547 spin_unlock(&sci->sc_state_lock);
2548 }
2549 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2550}
2551
2552static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2553{
2554 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2555 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2556 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2557 return SC_FLUSH_FILE;
2558 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2559 return SC_FLUSH_DAT;
2560 }
2561 return SC_LSEG_SR;
2562}
2563
2564/**
2565 * nilfs_segctor_thread - main loop of the segment constructor thread.
2566 * @arg: pointer to a struct nilfs_sc_info.
2567 *
2568 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2569 * to execute segment constructions.
2570 */
2571static int nilfs_segctor_thread(void *arg)
2572{
2573 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2574 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2575 int timeout = 0;
2576
2577 sci->sc_timer_task = current;
2578 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2579
2580 /* start sync. */
2581 sci->sc_task = current;
2582 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2583 nilfs_info(sci->sc_super,
2584 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2585 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2586
2587 set_freezable();
2588 spin_lock(&sci->sc_state_lock);
2589 loop:
2590 for (;;) {
2591 int mode;
2592
2593 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2594 goto end_thread;
2595
2596 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2597 mode = SC_LSEG_SR;
2598 else if (sci->sc_flush_request)
2599 mode = nilfs_segctor_flush_mode(sci);
2600 else
2601 break;
2602
2603 spin_unlock(&sci->sc_state_lock);
2604 nilfs_segctor_thread_construct(sci, mode);
2605 spin_lock(&sci->sc_state_lock);
2606 timeout = 0;
2607 }
2608
2609
2610 if (freezing(current)) {
2611 spin_unlock(&sci->sc_state_lock);
2612 try_to_freeze();
2613 spin_lock(&sci->sc_state_lock);
2614 } else {
2615 DEFINE_WAIT(wait);
2616 int should_sleep = 1;
2617
2618 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2619 TASK_INTERRUPTIBLE);
2620
2621 if (sci->sc_seq_request != sci->sc_seq_done)
2622 should_sleep = 0;
2623 else if (sci->sc_flush_request)
2624 should_sleep = 0;
2625 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2626 should_sleep = time_before(jiffies,
2627 sci->sc_timer.expires);
2628
2629 if (should_sleep) {
2630 spin_unlock(&sci->sc_state_lock);
2631 schedule();
2632 spin_lock(&sci->sc_state_lock);
2633 }
2634 finish_wait(&sci->sc_wait_daemon, &wait);
2635 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2636 time_after_eq(jiffies, sci->sc_timer.expires));
2637
2638 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2639 set_nilfs_discontinued(nilfs);
2640 }
2641 goto loop;
2642
2643 end_thread:
2644 /* end sync. */
2645 sci->sc_task = NULL;
2646 timer_shutdown_sync(&sci->sc_timer);
2647 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2648 spin_unlock(&sci->sc_state_lock);
2649 return 0;
2650}
2651
2652static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2653{
2654 struct task_struct *t;
2655
2656 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2657 if (IS_ERR(t)) {
2658 int err = PTR_ERR(t);
2659
2660 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2661 err);
2662 return err;
2663 }
2664 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2665 return 0;
2666}
2667
2668static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2669 __acquires(&sci->sc_state_lock)
2670 __releases(&sci->sc_state_lock)
2671{
2672 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2673
2674 while (sci->sc_task) {
2675 wake_up(&sci->sc_wait_daemon);
2676 spin_unlock(&sci->sc_state_lock);
2677 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2678 spin_lock(&sci->sc_state_lock);
2679 }
2680}
2681
2682/*
2683 * Setup & clean-up functions
2684 */
2685static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2686 struct nilfs_root *root)
2687{
2688 struct the_nilfs *nilfs = sb->s_fs_info;
2689 struct nilfs_sc_info *sci;
2690
2691 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2692 if (!sci)
2693 return NULL;
2694
2695 sci->sc_super = sb;
2696
2697 nilfs_get_root(root);
2698 sci->sc_root = root;
2699
2700 init_waitqueue_head(&sci->sc_wait_request);
2701 init_waitqueue_head(&sci->sc_wait_daemon);
2702 init_waitqueue_head(&sci->sc_wait_task);
2703 spin_lock_init(&sci->sc_state_lock);
2704 INIT_LIST_HEAD(&sci->sc_dirty_files);
2705 INIT_LIST_HEAD(&sci->sc_segbufs);
2706 INIT_LIST_HEAD(&sci->sc_write_logs);
2707 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2708 INIT_LIST_HEAD(&sci->sc_iput_queue);
2709 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2710
2711 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2712 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2713 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2714
2715 if (nilfs->ns_interval)
2716 sci->sc_interval = HZ * nilfs->ns_interval;
2717 if (nilfs->ns_watermark)
2718 sci->sc_watermark = nilfs->ns_watermark;
2719 return sci;
2720}
2721
2722static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2723{
2724 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2725
2726 /*
2727 * The segctord thread was stopped and its timer was removed.
2728 * But some tasks remain.
2729 */
2730 do {
2731 struct nilfs_transaction_info ti;
2732
2733 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2734 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2735 nilfs_transaction_unlock(sci->sc_super);
2736
2737 flush_work(&sci->sc_iput_work);
2738
2739 } while (ret && ret != -EROFS && retrycount-- > 0);
2740}
2741
2742/**
2743 * nilfs_segctor_destroy - destroy the segment constructor.
2744 * @sci: nilfs_sc_info
2745 *
2746 * nilfs_segctor_destroy() kills the segctord thread and frees
2747 * the nilfs_sc_info struct.
2748 * Caller must hold the segment semaphore.
2749 */
2750static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2751{
2752 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2753 int flag;
2754
2755 up_write(&nilfs->ns_segctor_sem);
2756
2757 spin_lock(&sci->sc_state_lock);
2758 nilfs_segctor_kill_thread(sci);
2759 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2760 || sci->sc_seq_request != sci->sc_seq_done);
2761 spin_unlock(&sci->sc_state_lock);
2762
2763 /*
2764 * Forcibly wake up tasks waiting in nilfs_segctor_sync(), which can
2765 * be called from delayed iput() via nilfs_evict_inode() and can race
2766 * with the above log writer thread termination.
2767 */
2768 nilfs_segctor_wakeup(sci, 0, true);
2769
2770 if (flush_work(&sci->sc_iput_work))
2771 flag = true;
2772
2773 if (flag || !nilfs_segctor_confirm(sci))
2774 nilfs_segctor_write_out(sci);
2775
2776 if (!list_empty(&sci->sc_dirty_files)) {
2777 nilfs_warn(sci->sc_super,
2778 "disposed unprocessed dirty file(s) when stopping log writer");
2779 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2780 }
2781
2782 if (!list_empty(&sci->sc_iput_queue)) {
2783 nilfs_warn(sci->sc_super,
2784 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2785 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2786 }
2787
2788 WARN_ON(!list_empty(&sci->sc_segbufs));
2789 WARN_ON(!list_empty(&sci->sc_write_logs));
2790
2791 nilfs_put_root(sci->sc_root);
2792
2793 down_write(&nilfs->ns_segctor_sem);
2794
2795 kfree(sci);
2796}
2797
2798/**
2799 * nilfs_attach_log_writer - attach log writer
2800 * @sb: super block instance
2801 * @root: root object of the current filesystem tree
2802 *
2803 * This allocates a log writer object, initializes it, and starts the
2804 * log writer.
2805 *
2806 * Return Value: On success, 0 is returned. On error, one of the following
2807 * negative error code is returned.
2808 *
2809 * %-ENOMEM - Insufficient memory available.
2810 */
2811int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2812{
2813 struct the_nilfs *nilfs = sb->s_fs_info;
2814 int err;
2815
2816 if (nilfs->ns_writer) {
2817 /*
2818 * This happens if the filesystem is made read-only by
2819 * __nilfs_error or nilfs_remount and then remounted
2820 * read/write. In these cases, reuse the existing
2821 * writer.
2822 */
2823 return 0;
2824 }
2825
2826 nilfs->ns_writer = nilfs_segctor_new(sb, root);
2827 if (!nilfs->ns_writer)
2828 return -ENOMEM;
2829
2830 inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2831
2832 err = nilfs_segctor_start_thread(nilfs->ns_writer);
2833 if (unlikely(err))
2834 nilfs_detach_log_writer(sb);
2835
2836 return err;
2837}
2838
2839/**
2840 * nilfs_detach_log_writer - destroy log writer
2841 * @sb: super block instance
2842 *
2843 * This kills log writer daemon, frees the log writer object, and
2844 * destroys list of dirty files.
2845 */
2846void nilfs_detach_log_writer(struct super_block *sb)
2847{
2848 struct the_nilfs *nilfs = sb->s_fs_info;
2849 LIST_HEAD(garbage_list);
2850
2851 down_write(&nilfs->ns_segctor_sem);
2852 if (nilfs->ns_writer) {
2853 nilfs_segctor_destroy(nilfs->ns_writer);
2854 nilfs->ns_writer = NULL;
2855 }
2856 set_nilfs_purging(nilfs);
2857
2858 /* Force to free the list of dirty files */
2859 spin_lock(&nilfs->ns_inode_lock);
2860 if (!list_empty(&nilfs->ns_dirty_files)) {
2861 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2862 nilfs_warn(sb,
2863 "disposed unprocessed dirty file(s) when detaching log writer");
2864 }
2865 spin_unlock(&nilfs->ns_inode_lock);
2866 up_write(&nilfs->ns_segctor_sem);
2867
2868 nilfs_dispose_list(nilfs, &garbage_list, 1);
2869 clear_nilfs_purging(nilfs);
2870}