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