Loading...
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 int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
884{
885 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
886 struct buffer_head *bh_cp;
887 struct nilfs_checkpoint *raw_cp;
888 int err;
889
890 /* XXX: this interface will be changed */
891 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
892 &raw_cp, &bh_cp);
893 if (likely(!err)) {
894 /*
895 * The following code is duplicated with cpfile. But, it is
896 * needed to collect the checkpoint even if it was not newly
897 * created.
898 */
899 mark_buffer_dirty(bh_cp);
900 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
901 nilfs_cpfile_put_checkpoint(
902 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
903 } else if (err == -EINVAL || err == -ENOENT) {
904 nilfs_error(sci->sc_super,
905 "checkpoint creation failed due to metadata corruption.");
906 err = -EIO;
907 }
908 return err;
909}
910
911static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
912{
913 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
914 struct buffer_head *bh_cp;
915 struct nilfs_checkpoint *raw_cp;
916 int err;
917
918 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
919 &raw_cp, &bh_cp);
920 if (unlikely(err)) {
921 if (err == -EINVAL || err == -ENOENT) {
922 nilfs_error(sci->sc_super,
923 "checkpoint finalization failed due to metadata corruption.");
924 err = -EIO;
925 }
926 goto failed_ibh;
927 }
928 raw_cp->cp_snapshot_list.ssl_next = 0;
929 raw_cp->cp_snapshot_list.ssl_prev = 0;
930 raw_cp->cp_inodes_count =
931 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
932 raw_cp->cp_blocks_count =
933 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
934 raw_cp->cp_nblk_inc =
935 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
936 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
937 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
938
939 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
940 nilfs_checkpoint_clear_minor(raw_cp);
941 else
942 nilfs_checkpoint_set_minor(raw_cp);
943
944 nilfs_write_inode_common(sci->sc_root->ifile,
945 &raw_cp->cp_ifile_inode, 1);
946 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
947 return 0;
948
949 failed_ibh:
950 return err;
951}
952
953static void nilfs_fill_in_file_bmap(struct inode *ifile,
954 struct nilfs_inode_info *ii)
955
956{
957 struct buffer_head *ibh;
958 struct nilfs_inode *raw_inode;
959
960 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
961 ibh = ii->i_bh;
962 BUG_ON(!ibh);
963 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
964 ibh);
965 nilfs_bmap_write(ii->i_bmap, raw_inode);
966 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
967 }
968}
969
970static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
971{
972 struct nilfs_inode_info *ii;
973
974 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
975 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
976 set_bit(NILFS_I_COLLECTED, &ii->i_state);
977 }
978}
979
980static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
981 struct the_nilfs *nilfs)
982{
983 struct buffer_head *bh_sr;
984 struct nilfs_super_root *raw_sr;
985 unsigned int isz, srsz;
986
987 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
988
989 lock_buffer(bh_sr);
990 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
991 isz = nilfs->ns_inode_size;
992 srsz = NILFS_SR_BYTES(isz);
993
994 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
995 raw_sr->sr_bytes = cpu_to_le16(srsz);
996 raw_sr->sr_nongc_ctime
997 = cpu_to_le64(nilfs_doing_gc() ?
998 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
999 raw_sr->sr_flags = 0;
1000
1001 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
1002 NILFS_SR_DAT_OFFSET(isz), 1);
1003 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
1004 NILFS_SR_CPFILE_OFFSET(isz), 1);
1005 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
1006 NILFS_SR_SUFILE_OFFSET(isz), 1);
1007 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1008 set_buffer_uptodate(bh_sr);
1009 unlock_buffer(bh_sr);
1010}
1011
1012static void nilfs_redirty_inodes(struct list_head *head)
1013{
1014 struct nilfs_inode_info *ii;
1015
1016 list_for_each_entry(ii, head, i_dirty) {
1017 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1018 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1019 }
1020}
1021
1022static void nilfs_drop_collected_inodes(struct list_head *head)
1023{
1024 struct nilfs_inode_info *ii;
1025
1026 list_for_each_entry(ii, head, i_dirty) {
1027 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1028 continue;
1029
1030 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1031 set_bit(NILFS_I_UPDATED, &ii->i_state);
1032 }
1033}
1034
1035static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1036 struct inode *inode,
1037 struct list_head *listp,
1038 int (*collect)(struct nilfs_sc_info *,
1039 struct buffer_head *,
1040 struct inode *))
1041{
1042 struct buffer_head *bh, *n;
1043 int err = 0;
1044
1045 if (collect) {
1046 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1047 list_del_init(&bh->b_assoc_buffers);
1048 err = collect(sci, bh, inode);
1049 brelse(bh);
1050 if (unlikely(err))
1051 goto dispose_buffers;
1052 }
1053 return 0;
1054 }
1055
1056 dispose_buffers:
1057 while (!list_empty(listp)) {
1058 bh = list_first_entry(listp, struct buffer_head,
1059 b_assoc_buffers);
1060 list_del_init(&bh->b_assoc_buffers);
1061 brelse(bh);
1062 }
1063 return err;
1064}
1065
1066static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1067{
1068 /* Remaining number of blocks within segment buffer */
1069 return sci->sc_segbuf_nblocks -
1070 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1071}
1072
1073static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1074 struct inode *inode,
1075 const struct nilfs_sc_operations *sc_ops)
1076{
1077 LIST_HEAD(data_buffers);
1078 LIST_HEAD(node_buffers);
1079 int err;
1080
1081 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1082 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1083
1084 n = nilfs_lookup_dirty_data_buffers(
1085 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1086 if (n > rest) {
1087 err = nilfs_segctor_apply_buffers(
1088 sci, inode, &data_buffers,
1089 sc_ops->collect_data);
1090 BUG_ON(!err); /* always receive -E2BIG or true error */
1091 goto break_or_fail;
1092 }
1093 }
1094 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1095
1096 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1097 err = nilfs_segctor_apply_buffers(
1098 sci, inode, &data_buffers, sc_ops->collect_data);
1099 if (unlikely(err)) {
1100 /* dispose node list */
1101 nilfs_segctor_apply_buffers(
1102 sci, inode, &node_buffers, NULL);
1103 goto break_or_fail;
1104 }
1105 sci->sc_stage.flags |= NILFS_CF_NODE;
1106 }
1107 /* Collect node */
1108 err = nilfs_segctor_apply_buffers(
1109 sci, inode, &node_buffers, sc_ops->collect_node);
1110 if (unlikely(err))
1111 goto break_or_fail;
1112
1113 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1114 err = nilfs_segctor_apply_buffers(
1115 sci, inode, &node_buffers, sc_ops->collect_bmap);
1116 if (unlikely(err))
1117 goto break_or_fail;
1118
1119 nilfs_segctor_end_finfo(sci, inode);
1120 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1121
1122 break_or_fail:
1123 return err;
1124}
1125
1126static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1127 struct inode *inode)
1128{
1129 LIST_HEAD(data_buffers);
1130 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1131 int err;
1132
1133 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1134 sci->sc_dsync_start,
1135 sci->sc_dsync_end);
1136
1137 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1138 nilfs_collect_file_data);
1139 if (!err) {
1140 nilfs_segctor_end_finfo(sci, inode);
1141 BUG_ON(n > rest);
1142 /* always receive -E2BIG or true error if n > rest */
1143 }
1144 return err;
1145}
1146
1147static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1148{
1149 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1150 struct list_head *head;
1151 struct nilfs_inode_info *ii;
1152 size_t ndone;
1153 int err = 0;
1154
1155 switch (nilfs_sc_cstage_get(sci)) {
1156 case NILFS_ST_INIT:
1157 /* Pre-processes */
1158 sci->sc_stage.flags = 0;
1159
1160 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1161 sci->sc_nblk_inc = 0;
1162 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1163 if (mode == SC_LSEG_DSYNC) {
1164 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1165 goto dsync_mode;
1166 }
1167 }
1168
1169 sci->sc_stage.dirty_file_ptr = NULL;
1170 sci->sc_stage.gc_inode_ptr = NULL;
1171 if (mode == SC_FLUSH_DAT) {
1172 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1173 goto dat_stage;
1174 }
1175 nilfs_sc_cstage_inc(sci);
1176 fallthrough;
1177 case NILFS_ST_GC:
1178 if (nilfs_doing_gc()) {
1179 head = &sci->sc_gc_inodes;
1180 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1181 head, i_dirty);
1182 list_for_each_entry_continue(ii, head, i_dirty) {
1183 err = nilfs_segctor_scan_file(
1184 sci, &ii->vfs_inode,
1185 &nilfs_sc_file_ops);
1186 if (unlikely(err)) {
1187 sci->sc_stage.gc_inode_ptr = list_entry(
1188 ii->i_dirty.prev,
1189 struct nilfs_inode_info,
1190 i_dirty);
1191 goto break_or_fail;
1192 }
1193 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1194 }
1195 sci->sc_stage.gc_inode_ptr = NULL;
1196 }
1197 nilfs_sc_cstage_inc(sci);
1198 fallthrough;
1199 case NILFS_ST_FILE:
1200 head = &sci->sc_dirty_files;
1201 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1202 i_dirty);
1203 list_for_each_entry_continue(ii, head, i_dirty) {
1204 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1205
1206 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1207 &nilfs_sc_file_ops);
1208 if (unlikely(err)) {
1209 sci->sc_stage.dirty_file_ptr =
1210 list_entry(ii->i_dirty.prev,
1211 struct nilfs_inode_info,
1212 i_dirty);
1213 goto break_or_fail;
1214 }
1215 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1216 /* XXX: required ? */
1217 }
1218 sci->sc_stage.dirty_file_ptr = NULL;
1219 if (mode == SC_FLUSH_FILE) {
1220 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1221 return 0;
1222 }
1223 nilfs_sc_cstage_inc(sci);
1224 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1225 fallthrough;
1226 case NILFS_ST_IFILE:
1227 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1228 &nilfs_sc_file_ops);
1229 if (unlikely(err))
1230 break;
1231 nilfs_sc_cstage_inc(sci);
1232 /* Creating a checkpoint */
1233 err = nilfs_segctor_create_checkpoint(sci);
1234 if (unlikely(err))
1235 break;
1236 fallthrough;
1237 case NILFS_ST_CPFILE:
1238 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1239 &nilfs_sc_file_ops);
1240 if (unlikely(err))
1241 break;
1242 nilfs_sc_cstage_inc(sci);
1243 fallthrough;
1244 case NILFS_ST_SUFILE:
1245 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1246 sci->sc_nfreesegs, &ndone);
1247 if (unlikely(err)) {
1248 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1249 sci->sc_freesegs, ndone,
1250 NULL);
1251 break;
1252 }
1253 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1254
1255 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1256 &nilfs_sc_file_ops);
1257 if (unlikely(err))
1258 break;
1259 nilfs_sc_cstage_inc(sci);
1260 fallthrough;
1261 case NILFS_ST_DAT:
1262 dat_stage:
1263 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1264 &nilfs_sc_dat_ops);
1265 if (unlikely(err))
1266 break;
1267 if (mode == SC_FLUSH_DAT) {
1268 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1269 return 0;
1270 }
1271 nilfs_sc_cstage_inc(sci);
1272 fallthrough;
1273 case NILFS_ST_SR:
1274 if (mode == SC_LSEG_SR) {
1275 /* Appending a super root */
1276 err = nilfs_segctor_add_super_root(sci);
1277 if (unlikely(err))
1278 break;
1279 }
1280 /* End of a logical segment */
1281 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1282 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1283 return 0;
1284 case NILFS_ST_DSYNC:
1285 dsync_mode:
1286 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1287 ii = sci->sc_dsync_inode;
1288 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1289 break;
1290
1291 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1292 if (unlikely(err))
1293 break;
1294 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1295 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1296 return 0;
1297 case NILFS_ST_DONE:
1298 return 0;
1299 default:
1300 BUG();
1301 }
1302
1303 break_or_fail:
1304 return err;
1305}
1306
1307/**
1308 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1309 * @sci: nilfs_sc_info
1310 * @nilfs: nilfs object
1311 */
1312static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1313 struct the_nilfs *nilfs)
1314{
1315 struct nilfs_segment_buffer *segbuf, *prev;
1316 __u64 nextnum;
1317 int err, alloc = 0;
1318
1319 segbuf = nilfs_segbuf_new(sci->sc_super);
1320 if (unlikely(!segbuf))
1321 return -ENOMEM;
1322
1323 if (list_empty(&sci->sc_write_logs)) {
1324 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1325 nilfs->ns_pseg_offset, nilfs);
1326 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1327 nilfs_shift_to_next_segment(nilfs);
1328 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1329 }
1330
1331 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1332 nextnum = nilfs->ns_nextnum;
1333
1334 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1335 /* Start from the head of a new full segment */
1336 alloc++;
1337 } else {
1338 /* Continue logs */
1339 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1340 nilfs_segbuf_map_cont(segbuf, prev);
1341 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1342 nextnum = prev->sb_nextnum;
1343
1344 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1345 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1346 segbuf->sb_sum.seg_seq++;
1347 alloc++;
1348 }
1349 }
1350
1351 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1352 if (err)
1353 goto failed;
1354
1355 if (alloc) {
1356 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1357 if (err)
1358 goto failed;
1359 }
1360 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1361
1362 BUG_ON(!list_empty(&sci->sc_segbufs));
1363 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1364 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1365 return 0;
1366
1367 failed:
1368 nilfs_segbuf_free(segbuf);
1369 return err;
1370}
1371
1372static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1373 struct the_nilfs *nilfs, int nadd)
1374{
1375 struct nilfs_segment_buffer *segbuf, *prev;
1376 struct inode *sufile = nilfs->ns_sufile;
1377 __u64 nextnextnum;
1378 LIST_HEAD(list);
1379 int err, ret, i;
1380
1381 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1382 /*
1383 * Since the segment specified with nextnum might be allocated during
1384 * the previous construction, the buffer including its segusage may
1385 * not be dirty. The following call ensures that the buffer is dirty
1386 * and will pin the buffer on memory until the sufile is written.
1387 */
1388 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1389 if (unlikely(err))
1390 return err;
1391
1392 for (i = 0; i < nadd; i++) {
1393 /* extend segment info */
1394 err = -ENOMEM;
1395 segbuf = nilfs_segbuf_new(sci->sc_super);
1396 if (unlikely(!segbuf))
1397 goto failed;
1398
1399 /* map this buffer to region of segment on-disk */
1400 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1401 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1402
1403 /* allocate the next next full segment */
1404 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1405 if (unlikely(err))
1406 goto failed_segbuf;
1407
1408 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1409 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1410
1411 list_add_tail(&segbuf->sb_list, &list);
1412 prev = segbuf;
1413 }
1414 list_splice_tail(&list, &sci->sc_segbufs);
1415 return 0;
1416
1417 failed_segbuf:
1418 nilfs_segbuf_free(segbuf);
1419 failed:
1420 list_for_each_entry(segbuf, &list, sb_list) {
1421 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1422 WARN_ON(ret); /* never fails */
1423 }
1424 nilfs_destroy_logs(&list);
1425 return err;
1426}
1427
1428static void nilfs_free_incomplete_logs(struct list_head *logs,
1429 struct the_nilfs *nilfs)
1430{
1431 struct nilfs_segment_buffer *segbuf, *prev;
1432 struct inode *sufile = nilfs->ns_sufile;
1433 int ret;
1434
1435 segbuf = NILFS_FIRST_SEGBUF(logs);
1436 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1437 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1438 WARN_ON(ret); /* never fails */
1439 }
1440 if (atomic_read(&segbuf->sb_err)) {
1441 /* Case 1: The first segment failed */
1442 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1443 /*
1444 * Case 1a: Partial segment appended into an existing
1445 * segment
1446 */
1447 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1448 segbuf->sb_fseg_end);
1449 else /* Case 1b: New full segment */
1450 set_nilfs_discontinued(nilfs);
1451 }
1452
1453 prev = segbuf;
1454 list_for_each_entry_continue(segbuf, logs, sb_list) {
1455 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1456 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1457 WARN_ON(ret); /* never fails */
1458 }
1459 if (atomic_read(&segbuf->sb_err) &&
1460 segbuf->sb_segnum != nilfs->ns_nextnum)
1461 /* Case 2: extended segment (!= next) failed */
1462 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1463 prev = segbuf;
1464 }
1465}
1466
1467static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1468 struct inode *sufile)
1469{
1470 struct nilfs_segment_buffer *segbuf;
1471 unsigned long live_blocks;
1472 int ret;
1473
1474 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1475 live_blocks = segbuf->sb_sum.nblocks +
1476 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1477 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1478 live_blocks,
1479 sci->sc_seg_ctime);
1480 WARN_ON(ret); /* always succeed because the segusage is dirty */
1481 }
1482}
1483
1484static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1485{
1486 struct nilfs_segment_buffer *segbuf;
1487 int ret;
1488
1489 segbuf = NILFS_FIRST_SEGBUF(logs);
1490 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1491 segbuf->sb_pseg_start -
1492 segbuf->sb_fseg_start, 0);
1493 WARN_ON(ret); /* always succeed because the segusage is dirty */
1494
1495 list_for_each_entry_continue(segbuf, logs, sb_list) {
1496 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1497 0, 0);
1498 WARN_ON(ret); /* always succeed */
1499 }
1500}
1501
1502static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1503 struct nilfs_segment_buffer *last,
1504 struct inode *sufile)
1505{
1506 struct nilfs_segment_buffer *segbuf = last;
1507 int ret;
1508
1509 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1510 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1511 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1512 WARN_ON(ret);
1513 }
1514 nilfs_truncate_logs(&sci->sc_segbufs, last);
1515}
1516
1517
1518static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1519 struct the_nilfs *nilfs, int mode)
1520{
1521 struct nilfs_cstage prev_stage = sci->sc_stage;
1522 int err, nadd = 1;
1523
1524 /* Collection retry loop */
1525 for (;;) {
1526 sci->sc_nblk_this_inc = 0;
1527 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1528
1529 err = nilfs_segctor_reset_segment_buffer(sci);
1530 if (unlikely(err))
1531 goto failed;
1532
1533 err = nilfs_segctor_collect_blocks(sci, mode);
1534 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1535 if (!err)
1536 break;
1537
1538 if (unlikely(err != -E2BIG))
1539 goto failed;
1540
1541 /* The current segment is filled up */
1542 if (mode != SC_LSEG_SR ||
1543 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1544 break;
1545
1546 nilfs_clear_logs(&sci->sc_segbufs);
1547
1548 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1549 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1550 sci->sc_freesegs,
1551 sci->sc_nfreesegs,
1552 NULL);
1553 WARN_ON(err); /* do not happen */
1554 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1555 }
1556
1557 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1558 if (unlikely(err))
1559 return err;
1560
1561 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1562 sci->sc_stage = prev_stage;
1563 }
1564 nilfs_segctor_zeropad_segsum(sci);
1565 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1566 return 0;
1567
1568 failed:
1569 return err;
1570}
1571
1572static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1573 struct buffer_head *new_bh)
1574{
1575 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1576
1577 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1578 /* The caller must release old_bh */
1579}
1580
1581static int
1582nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1583 struct nilfs_segment_buffer *segbuf,
1584 int mode)
1585{
1586 struct inode *inode = NULL;
1587 sector_t blocknr;
1588 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1589 unsigned long nblocks = 0, ndatablk = 0;
1590 const struct nilfs_sc_operations *sc_op = NULL;
1591 struct nilfs_segsum_pointer ssp;
1592 struct nilfs_finfo *finfo = NULL;
1593 union nilfs_binfo binfo;
1594 struct buffer_head *bh, *bh_org;
1595 ino_t ino = 0;
1596 int err = 0;
1597
1598 if (!nfinfo)
1599 goto out;
1600
1601 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1602 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1603 ssp.offset = sizeof(struct nilfs_segment_summary);
1604
1605 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1606 if (bh == segbuf->sb_super_root)
1607 break;
1608 if (!finfo) {
1609 finfo = nilfs_segctor_map_segsum_entry(
1610 sci, &ssp, sizeof(*finfo));
1611 ino = le64_to_cpu(finfo->fi_ino);
1612 nblocks = le32_to_cpu(finfo->fi_nblocks);
1613 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1614
1615 inode = bh->b_folio->mapping->host;
1616
1617 if (mode == SC_LSEG_DSYNC)
1618 sc_op = &nilfs_sc_dsync_ops;
1619 else if (ino == NILFS_DAT_INO)
1620 sc_op = &nilfs_sc_dat_ops;
1621 else /* file blocks */
1622 sc_op = &nilfs_sc_file_ops;
1623 }
1624 bh_org = bh;
1625 get_bh(bh_org);
1626 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1627 &binfo);
1628 if (bh != bh_org)
1629 nilfs_list_replace_buffer(bh_org, bh);
1630 brelse(bh_org);
1631 if (unlikely(err))
1632 goto failed_bmap;
1633
1634 if (ndatablk > 0)
1635 sc_op->write_data_binfo(sci, &ssp, &binfo);
1636 else
1637 sc_op->write_node_binfo(sci, &ssp, &binfo);
1638
1639 blocknr++;
1640 if (--nblocks == 0) {
1641 finfo = NULL;
1642 if (--nfinfo == 0)
1643 break;
1644 } else if (ndatablk > 0)
1645 ndatablk--;
1646 }
1647 out:
1648 return 0;
1649
1650 failed_bmap:
1651 return err;
1652}
1653
1654static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1655{
1656 struct nilfs_segment_buffer *segbuf;
1657 int err;
1658
1659 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1660 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1661 if (unlikely(err))
1662 return err;
1663 nilfs_segbuf_fill_in_segsum(segbuf);
1664 }
1665 return 0;
1666}
1667
1668static void nilfs_begin_folio_io(struct folio *folio)
1669{
1670 if (!folio || folio_test_writeback(folio))
1671 /*
1672 * For split b-tree node pages, this function may be called
1673 * twice. We ignore the 2nd or later calls by this check.
1674 */
1675 return;
1676
1677 folio_lock(folio);
1678 folio_clear_dirty_for_io(folio);
1679 folio_start_writeback(folio);
1680 folio_unlock(folio);
1681}
1682
1683static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1684{
1685 struct nilfs_segment_buffer *segbuf;
1686 struct folio *bd_folio = NULL, *fs_folio = NULL;
1687
1688 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1689 struct buffer_head *bh;
1690
1691 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1692 b_assoc_buffers) {
1693 if (bh->b_folio != bd_folio) {
1694 if (bd_folio) {
1695 folio_lock(bd_folio);
1696 folio_clear_dirty_for_io(bd_folio);
1697 folio_start_writeback(bd_folio);
1698 folio_unlock(bd_folio);
1699 }
1700 bd_folio = bh->b_folio;
1701 }
1702 }
1703
1704 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1705 b_assoc_buffers) {
1706 if (bh == segbuf->sb_super_root) {
1707 if (bh->b_folio != bd_folio) {
1708 folio_lock(bd_folio);
1709 folio_clear_dirty_for_io(bd_folio);
1710 folio_start_writeback(bd_folio);
1711 folio_unlock(bd_folio);
1712 bd_folio = bh->b_folio;
1713 }
1714 break;
1715 }
1716 set_buffer_async_write(bh);
1717 if (bh->b_folio != fs_folio) {
1718 nilfs_begin_folio_io(fs_folio);
1719 fs_folio = bh->b_folio;
1720 }
1721 }
1722 }
1723 if (bd_folio) {
1724 folio_lock(bd_folio);
1725 folio_clear_dirty_for_io(bd_folio);
1726 folio_start_writeback(bd_folio);
1727 folio_unlock(bd_folio);
1728 }
1729 nilfs_begin_folio_io(fs_folio);
1730}
1731
1732static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1733 struct the_nilfs *nilfs)
1734{
1735 int ret;
1736
1737 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1738 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1739 return ret;
1740}
1741
1742static void nilfs_end_folio_io(struct folio *folio, int err)
1743{
1744 if (!folio)
1745 return;
1746
1747 if (buffer_nilfs_node(folio_buffers(folio)) &&
1748 !folio_test_writeback(folio)) {
1749 /*
1750 * For b-tree node pages, this function may be called twice
1751 * or more because they might be split in a segment.
1752 */
1753 if (folio_test_dirty(folio)) {
1754 /*
1755 * For pages holding split b-tree node buffers, dirty
1756 * flag on the buffers may be cleared discretely.
1757 * In that case, the page is once redirtied for
1758 * remaining buffers, and it must be cancelled if
1759 * all the buffers get cleaned later.
1760 */
1761 folio_lock(folio);
1762 if (nilfs_folio_buffers_clean(folio))
1763 __nilfs_clear_folio_dirty(folio);
1764 folio_unlock(folio);
1765 }
1766 return;
1767 }
1768
1769 if (!err) {
1770 if (!nilfs_folio_buffers_clean(folio))
1771 filemap_dirty_folio(folio->mapping, folio);
1772 folio_clear_error(folio);
1773 } else {
1774 filemap_dirty_folio(folio->mapping, folio);
1775 folio_set_error(folio);
1776 }
1777
1778 folio_end_writeback(folio);
1779}
1780
1781static void nilfs_abort_logs(struct list_head *logs, int err)
1782{
1783 struct nilfs_segment_buffer *segbuf;
1784 struct folio *bd_folio = NULL, *fs_folio = NULL;
1785 struct buffer_head *bh;
1786
1787 if (list_empty(logs))
1788 return;
1789
1790 list_for_each_entry(segbuf, logs, sb_list) {
1791 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1792 b_assoc_buffers) {
1793 clear_buffer_uptodate(bh);
1794 if (bh->b_folio != bd_folio) {
1795 if (bd_folio)
1796 folio_end_writeback(bd_folio);
1797 bd_folio = bh->b_folio;
1798 }
1799 }
1800
1801 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1802 b_assoc_buffers) {
1803 if (bh == segbuf->sb_super_root) {
1804 clear_buffer_uptodate(bh);
1805 if (bh->b_folio != bd_folio) {
1806 folio_end_writeback(bd_folio);
1807 bd_folio = bh->b_folio;
1808 }
1809 break;
1810 }
1811 clear_buffer_async_write(bh);
1812 if (bh->b_folio != fs_folio) {
1813 nilfs_end_folio_io(fs_folio, err);
1814 fs_folio = bh->b_folio;
1815 }
1816 }
1817 }
1818 if (bd_folio)
1819 folio_end_writeback(bd_folio);
1820
1821 nilfs_end_folio_io(fs_folio, err);
1822}
1823
1824static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1825 struct the_nilfs *nilfs, int err)
1826{
1827 LIST_HEAD(logs);
1828 int ret;
1829
1830 list_splice_tail_init(&sci->sc_write_logs, &logs);
1831 ret = nilfs_wait_on_logs(&logs);
1832 nilfs_abort_logs(&logs, ret ? : err);
1833
1834 list_splice_tail_init(&sci->sc_segbufs, &logs);
1835 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1836 nilfs_free_incomplete_logs(&logs, nilfs);
1837
1838 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1839 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1840 sci->sc_freesegs,
1841 sci->sc_nfreesegs,
1842 NULL);
1843 WARN_ON(ret); /* do not happen */
1844 }
1845
1846 nilfs_destroy_logs(&logs);
1847}
1848
1849static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1850 struct nilfs_segment_buffer *segbuf)
1851{
1852 nilfs->ns_segnum = segbuf->sb_segnum;
1853 nilfs->ns_nextnum = segbuf->sb_nextnum;
1854 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1855 + segbuf->sb_sum.nblocks;
1856 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1857 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1858}
1859
1860static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1861{
1862 struct nilfs_segment_buffer *segbuf;
1863 struct folio *bd_folio = NULL, *fs_folio = NULL;
1864 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1865 int update_sr = false;
1866
1867 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1868 struct buffer_head *bh;
1869
1870 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1871 b_assoc_buffers) {
1872 set_buffer_uptodate(bh);
1873 clear_buffer_dirty(bh);
1874 if (bh->b_folio != bd_folio) {
1875 if (bd_folio)
1876 folio_end_writeback(bd_folio);
1877 bd_folio = bh->b_folio;
1878 }
1879 }
1880 /*
1881 * We assume that the buffers which belong to the same folio
1882 * continue over the buffer list.
1883 * Under this assumption, the last BHs of folios is
1884 * identifiable by the discontinuity of bh->b_folio
1885 * (folio != fs_folio).
1886 *
1887 * For B-tree node blocks, however, this assumption is not
1888 * guaranteed. The cleanup code of B-tree node folios needs
1889 * special care.
1890 */
1891 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1892 b_assoc_buffers) {
1893 const unsigned long set_bits = BIT(BH_Uptodate);
1894 const unsigned long clear_bits =
1895 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1896 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1897 BIT(BH_NILFS_Redirected));
1898
1899 if (bh == segbuf->sb_super_root) {
1900 set_buffer_uptodate(bh);
1901 clear_buffer_dirty(bh);
1902 if (bh->b_folio != bd_folio) {
1903 folio_end_writeback(bd_folio);
1904 bd_folio = bh->b_folio;
1905 }
1906 update_sr = true;
1907 break;
1908 }
1909 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1910 if (bh->b_folio != fs_folio) {
1911 nilfs_end_folio_io(fs_folio, 0);
1912 fs_folio = bh->b_folio;
1913 }
1914 }
1915
1916 if (!nilfs_segbuf_simplex(segbuf)) {
1917 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1918 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1919 sci->sc_lseg_stime = jiffies;
1920 }
1921 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1922 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1923 }
1924 }
1925 /*
1926 * Since folios may continue over multiple segment buffers,
1927 * end of the last folio must be checked outside of the loop.
1928 */
1929 if (bd_folio)
1930 folio_end_writeback(bd_folio);
1931
1932 nilfs_end_folio_io(fs_folio, 0);
1933
1934 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1935
1936 if (nilfs_doing_gc())
1937 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1938 else
1939 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1940
1941 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1942
1943 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1944 nilfs_set_next_segment(nilfs, segbuf);
1945
1946 if (update_sr) {
1947 nilfs->ns_flushed_device = 0;
1948 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1949 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1950
1951 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1952 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1953 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1954 nilfs_segctor_clear_metadata_dirty(sci);
1955 } else
1956 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1957}
1958
1959static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1960{
1961 int ret;
1962
1963 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1964 if (!ret) {
1965 nilfs_segctor_complete_write(sci);
1966 nilfs_destroy_logs(&sci->sc_write_logs);
1967 }
1968 return ret;
1969}
1970
1971static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1972 struct the_nilfs *nilfs)
1973{
1974 struct nilfs_inode_info *ii, *n;
1975 struct inode *ifile = sci->sc_root->ifile;
1976
1977 spin_lock(&nilfs->ns_inode_lock);
1978 retry:
1979 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1980 if (!ii->i_bh) {
1981 struct buffer_head *ibh;
1982 int err;
1983
1984 spin_unlock(&nilfs->ns_inode_lock);
1985 err = nilfs_ifile_get_inode_block(
1986 ifile, ii->vfs_inode.i_ino, &ibh);
1987 if (unlikely(err)) {
1988 nilfs_warn(sci->sc_super,
1989 "log writer: error %d getting inode block (ino=%lu)",
1990 err, ii->vfs_inode.i_ino);
1991 return err;
1992 }
1993 spin_lock(&nilfs->ns_inode_lock);
1994 if (likely(!ii->i_bh))
1995 ii->i_bh = ibh;
1996 else
1997 brelse(ibh);
1998 goto retry;
1999 }
2000
2001 // Always redirty the buffer to avoid race condition
2002 mark_buffer_dirty(ii->i_bh);
2003 nilfs_mdt_mark_dirty(ifile);
2004
2005 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2006 set_bit(NILFS_I_BUSY, &ii->i_state);
2007 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2008 }
2009 spin_unlock(&nilfs->ns_inode_lock);
2010
2011 return 0;
2012}
2013
2014static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2015 struct the_nilfs *nilfs)
2016{
2017 struct nilfs_inode_info *ii, *n;
2018 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2019 int defer_iput = false;
2020
2021 spin_lock(&nilfs->ns_inode_lock);
2022 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2023 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2024 test_bit(NILFS_I_DIRTY, &ii->i_state))
2025 continue;
2026
2027 clear_bit(NILFS_I_BUSY, &ii->i_state);
2028 brelse(ii->i_bh);
2029 ii->i_bh = NULL;
2030 list_del_init(&ii->i_dirty);
2031 if (!ii->vfs_inode.i_nlink || during_mount) {
2032 /*
2033 * Defer calling iput() to avoid deadlocks if
2034 * i_nlink == 0 or mount is not yet finished.
2035 */
2036 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2037 defer_iput = true;
2038 } else {
2039 spin_unlock(&nilfs->ns_inode_lock);
2040 iput(&ii->vfs_inode);
2041 spin_lock(&nilfs->ns_inode_lock);
2042 }
2043 }
2044 spin_unlock(&nilfs->ns_inode_lock);
2045
2046 if (defer_iput)
2047 schedule_work(&sci->sc_iput_work);
2048}
2049
2050/*
2051 * Main procedure of segment constructor
2052 */
2053static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2054{
2055 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2056 int err;
2057
2058 if (sb_rdonly(sci->sc_super))
2059 return -EROFS;
2060
2061 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2062 sci->sc_cno = nilfs->ns_cno;
2063
2064 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2065 if (unlikely(err))
2066 goto out;
2067
2068 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2069 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2070
2071 if (nilfs_segctor_clean(sci))
2072 goto out;
2073
2074 do {
2075 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2076
2077 err = nilfs_segctor_begin_construction(sci, nilfs);
2078 if (unlikely(err))
2079 goto out;
2080
2081 /* Update time stamp */
2082 sci->sc_seg_ctime = ktime_get_real_seconds();
2083
2084 err = nilfs_segctor_collect(sci, nilfs, mode);
2085 if (unlikely(err))
2086 goto failed;
2087
2088 /* Avoid empty segment */
2089 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2090 nilfs_segbuf_empty(sci->sc_curseg)) {
2091 nilfs_segctor_abort_construction(sci, nilfs, 1);
2092 goto out;
2093 }
2094
2095 err = nilfs_segctor_assign(sci, mode);
2096 if (unlikely(err))
2097 goto failed;
2098
2099 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2100 nilfs_segctor_fill_in_file_bmap(sci);
2101
2102 if (mode == SC_LSEG_SR &&
2103 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2104 err = nilfs_segctor_fill_in_checkpoint(sci);
2105 if (unlikely(err))
2106 goto failed_to_write;
2107
2108 nilfs_segctor_fill_in_super_root(sci, nilfs);
2109 }
2110 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2111
2112 /* Write partial segments */
2113 nilfs_segctor_prepare_write(sci);
2114
2115 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2116 nilfs->ns_crc_seed);
2117
2118 err = nilfs_segctor_write(sci, nilfs);
2119 if (unlikely(err))
2120 goto failed_to_write;
2121
2122 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2123 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2124 /*
2125 * At this point, we avoid double buffering
2126 * for blocksize < pagesize because page dirty
2127 * flag is turned off during write and dirty
2128 * buffers are not properly collected for
2129 * pages crossing over segments.
2130 */
2131 err = nilfs_segctor_wait(sci);
2132 if (err)
2133 goto failed_to_write;
2134 }
2135 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2136
2137 out:
2138 nilfs_segctor_drop_written_files(sci, nilfs);
2139 return err;
2140
2141 failed_to_write:
2142 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2143 nilfs_redirty_inodes(&sci->sc_dirty_files);
2144
2145 failed:
2146 if (nilfs_doing_gc())
2147 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2148 nilfs_segctor_abort_construction(sci, nilfs, err);
2149 goto out;
2150}
2151
2152/**
2153 * nilfs_segctor_start_timer - set timer of background write
2154 * @sci: nilfs_sc_info
2155 *
2156 * If the timer has already been set, it ignores the new request.
2157 * This function MUST be called within a section locking the segment
2158 * semaphore.
2159 */
2160static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2161{
2162 spin_lock(&sci->sc_state_lock);
2163 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2164 sci->sc_timer.expires = jiffies + sci->sc_interval;
2165 add_timer(&sci->sc_timer);
2166 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2167 }
2168 spin_unlock(&sci->sc_state_lock);
2169}
2170
2171static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2172{
2173 spin_lock(&sci->sc_state_lock);
2174 if (!(sci->sc_flush_request & BIT(bn))) {
2175 unsigned long prev_req = sci->sc_flush_request;
2176
2177 sci->sc_flush_request |= BIT(bn);
2178 if (!prev_req)
2179 wake_up(&sci->sc_wait_daemon);
2180 }
2181 spin_unlock(&sci->sc_state_lock);
2182}
2183
2184/**
2185 * nilfs_flush_segment - trigger a segment construction for resource control
2186 * @sb: super block
2187 * @ino: inode number of the file to be flushed out.
2188 */
2189void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2190{
2191 struct the_nilfs *nilfs = sb->s_fs_info;
2192 struct nilfs_sc_info *sci = nilfs->ns_writer;
2193
2194 if (!sci || nilfs_doing_construction())
2195 return;
2196 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2197 /* assign bit 0 to data files */
2198}
2199
2200struct nilfs_segctor_wait_request {
2201 wait_queue_entry_t wq;
2202 __u32 seq;
2203 int err;
2204 atomic_t done;
2205};
2206
2207static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2208{
2209 struct nilfs_segctor_wait_request wait_req;
2210 int err = 0;
2211
2212 spin_lock(&sci->sc_state_lock);
2213 init_wait(&wait_req.wq);
2214 wait_req.err = 0;
2215 atomic_set(&wait_req.done, 0);
2216 wait_req.seq = ++sci->sc_seq_request;
2217 spin_unlock(&sci->sc_state_lock);
2218
2219 init_waitqueue_entry(&wait_req.wq, current);
2220 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2221 set_current_state(TASK_INTERRUPTIBLE);
2222 wake_up(&sci->sc_wait_daemon);
2223
2224 for (;;) {
2225 if (atomic_read(&wait_req.done)) {
2226 err = wait_req.err;
2227 break;
2228 }
2229 if (!signal_pending(current)) {
2230 schedule();
2231 continue;
2232 }
2233 err = -ERESTARTSYS;
2234 break;
2235 }
2236 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2237 return err;
2238}
2239
2240static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2241{
2242 struct nilfs_segctor_wait_request *wrq, *n;
2243 unsigned long flags;
2244
2245 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2246 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2247 if (!atomic_read(&wrq->done) &&
2248 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2249 wrq->err = err;
2250 atomic_set(&wrq->done, 1);
2251 }
2252 if (atomic_read(&wrq->done)) {
2253 wrq->wq.func(&wrq->wq,
2254 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2255 0, NULL);
2256 }
2257 }
2258 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2259}
2260
2261/**
2262 * nilfs_construct_segment - construct a logical segment
2263 * @sb: super block
2264 *
2265 * Return Value: On success, 0 is returned. On errors, one of the following
2266 * negative error code is returned.
2267 *
2268 * %-EROFS - Read only filesystem.
2269 *
2270 * %-EIO - I/O error
2271 *
2272 * %-ENOSPC - No space left on device (only in a panic state).
2273 *
2274 * %-ERESTARTSYS - Interrupted.
2275 *
2276 * %-ENOMEM - Insufficient memory available.
2277 */
2278int nilfs_construct_segment(struct super_block *sb)
2279{
2280 struct the_nilfs *nilfs = sb->s_fs_info;
2281 struct nilfs_sc_info *sci = nilfs->ns_writer;
2282 struct nilfs_transaction_info *ti;
2283
2284 if (sb_rdonly(sb) || unlikely(!sci))
2285 return -EROFS;
2286
2287 /* A call inside transactions causes a deadlock. */
2288 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2289
2290 return nilfs_segctor_sync(sci);
2291}
2292
2293/**
2294 * nilfs_construct_dsync_segment - construct a data-only logical segment
2295 * @sb: super block
2296 * @inode: inode whose data blocks should be written out
2297 * @start: start byte offset
2298 * @end: end byte offset (inclusive)
2299 *
2300 * Return Value: On success, 0 is returned. On errors, one of the following
2301 * negative error code is returned.
2302 *
2303 * %-EROFS - Read only filesystem.
2304 *
2305 * %-EIO - I/O error
2306 *
2307 * %-ENOSPC - No space left on device (only in a panic state).
2308 *
2309 * %-ERESTARTSYS - Interrupted.
2310 *
2311 * %-ENOMEM - Insufficient memory available.
2312 */
2313int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2314 loff_t start, loff_t end)
2315{
2316 struct the_nilfs *nilfs = sb->s_fs_info;
2317 struct nilfs_sc_info *sci = nilfs->ns_writer;
2318 struct nilfs_inode_info *ii;
2319 struct nilfs_transaction_info ti;
2320 int err = 0;
2321
2322 if (sb_rdonly(sb) || unlikely(!sci))
2323 return -EROFS;
2324
2325 nilfs_transaction_lock(sb, &ti, 0);
2326
2327 ii = NILFS_I(inode);
2328 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2329 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2330 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2331 nilfs_discontinued(nilfs)) {
2332 nilfs_transaction_unlock(sb);
2333 err = nilfs_segctor_sync(sci);
2334 return err;
2335 }
2336
2337 spin_lock(&nilfs->ns_inode_lock);
2338 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2339 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2340 spin_unlock(&nilfs->ns_inode_lock);
2341 nilfs_transaction_unlock(sb);
2342 return 0;
2343 }
2344 spin_unlock(&nilfs->ns_inode_lock);
2345 sci->sc_dsync_inode = ii;
2346 sci->sc_dsync_start = start;
2347 sci->sc_dsync_end = end;
2348
2349 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2350 if (!err)
2351 nilfs->ns_flushed_device = 0;
2352
2353 nilfs_transaction_unlock(sb);
2354 return err;
2355}
2356
2357#define FLUSH_FILE_BIT (0x1) /* data file only */
2358#define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2359
2360/**
2361 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2362 * @sci: segment constructor object
2363 */
2364static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2365{
2366 spin_lock(&sci->sc_state_lock);
2367 sci->sc_seq_accepted = sci->sc_seq_request;
2368 spin_unlock(&sci->sc_state_lock);
2369 del_timer_sync(&sci->sc_timer);
2370}
2371
2372/**
2373 * nilfs_segctor_notify - notify the result of request to caller threads
2374 * @sci: segment constructor object
2375 * @mode: mode of log forming
2376 * @err: error code to be notified
2377 */
2378static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2379{
2380 /* Clear requests (even when the construction failed) */
2381 spin_lock(&sci->sc_state_lock);
2382
2383 if (mode == SC_LSEG_SR) {
2384 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2385 sci->sc_seq_done = sci->sc_seq_accepted;
2386 nilfs_segctor_wakeup(sci, err);
2387 sci->sc_flush_request = 0;
2388 } else {
2389 if (mode == SC_FLUSH_FILE)
2390 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2391 else if (mode == SC_FLUSH_DAT)
2392 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2393
2394 /* re-enable timer if checkpoint creation was not done */
2395 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2396 time_before(jiffies, sci->sc_timer.expires))
2397 add_timer(&sci->sc_timer);
2398 }
2399 spin_unlock(&sci->sc_state_lock);
2400}
2401
2402/**
2403 * nilfs_segctor_construct - form logs and write them to disk
2404 * @sci: segment constructor object
2405 * @mode: mode of log forming
2406 */
2407static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2408{
2409 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2410 struct nilfs_super_block **sbp;
2411 int err = 0;
2412
2413 nilfs_segctor_accept(sci);
2414
2415 if (nilfs_discontinued(nilfs))
2416 mode = SC_LSEG_SR;
2417 if (!nilfs_segctor_confirm(sci))
2418 err = nilfs_segctor_do_construct(sci, mode);
2419
2420 if (likely(!err)) {
2421 if (mode != SC_FLUSH_DAT)
2422 atomic_set(&nilfs->ns_ndirtyblks, 0);
2423 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2424 nilfs_discontinued(nilfs)) {
2425 down_write(&nilfs->ns_sem);
2426 err = -EIO;
2427 sbp = nilfs_prepare_super(sci->sc_super,
2428 nilfs_sb_will_flip(nilfs));
2429 if (likely(sbp)) {
2430 nilfs_set_log_cursor(sbp[0], nilfs);
2431 err = nilfs_commit_super(sci->sc_super,
2432 NILFS_SB_COMMIT);
2433 }
2434 up_write(&nilfs->ns_sem);
2435 }
2436 }
2437
2438 nilfs_segctor_notify(sci, mode, err);
2439 return err;
2440}
2441
2442static void nilfs_construction_timeout(struct timer_list *t)
2443{
2444 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2445
2446 wake_up_process(sci->sc_timer_task);
2447}
2448
2449static void
2450nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2451{
2452 struct nilfs_inode_info *ii, *n;
2453
2454 list_for_each_entry_safe(ii, n, head, i_dirty) {
2455 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2456 continue;
2457 list_del_init(&ii->i_dirty);
2458 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2459 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2460 iput(&ii->vfs_inode);
2461 }
2462}
2463
2464int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2465 void **kbufs)
2466{
2467 struct the_nilfs *nilfs = sb->s_fs_info;
2468 struct nilfs_sc_info *sci = nilfs->ns_writer;
2469 struct nilfs_transaction_info ti;
2470 int err;
2471
2472 if (unlikely(!sci))
2473 return -EROFS;
2474
2475 nilfs_transaction_lock(sb, &ti, 1);
2476
2477 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2478 if (unlikely(err))
2479 goto out_unlock;
2480
2481 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2482 if (unlikely(err)) {
2483 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2484 goto out_unlock;
2485 }
2486
2487 sci->sc_freesegs = kbufs[4];
2488 sci->sc_nfreesegs = argv[4].v_nmembs;
2489 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2490
2491 for (;;) {
2492 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2493 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2494
2495 if (likely(!err))
2496 break;
2497
2498 nilfs_warn(sb, "error %d cleaning segments", err);
2499 set_current_state(TASK_INTERRUPTIBLE);
2500 schedule_timeout(sci->sc_interval);
2501 }
2502 if (nilfs_test_opt(nilfs, DISCARD)) {
2503 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2504 sci->sc_nfreesegs);
2505 if (ret) {
2506 nilfs_warn(sb,
2507 "error %d on discard request, turning discards off for the device",
2508 ret);
2509 nilfs_clear_opt(nilfs, DISCARD);
2510 }
2511 }
2512
2513 out_unlock:
2514 sci->sc_freesegs = NULL;
2515 sci->sc_nfreesegs = 0;
2516 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2517 nilfs_transaction_unlock(sb);
2518 return err;
2519}
2520
2521static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2522{
2523 struct nilfs_transaction_info ti;
2524
2525 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2526 nilfs_segctor_construct(sci, mode);
2527
2528 /*
2529 * Unclosed segment should be retried. We do this using sc_timer.
2530 * Timeout of sc_timer will invoke complete construction which leads
2531 * to close the current logical segment.
2532 */
2533 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2534 nilfs_segctor_start_timer(sci);
2535
2536 nilfs_transaction_unlock(sci->sc_super);
2537}
2538
2539static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2540{
2541 int mode = 0;
2542
2543 spin_lock(&sci->sc_state_lock);
2544 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2545 SC_FLUSH_DAT : SC_FLUSH_FILE;
2546 spin_unlock(&sci->sc_state_lock);
2547
2548 if (mode) {
2549 nilfs_segctor_do_construct(sci, mode);
2550
2551 spin_lock(&sci->sc_state_lock);
2552 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2553 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2554 spin_unlock(&sci->sc_state_lock);
2555 }
2556 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2557}
2558
2559static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2560{
2561 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2562 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2563 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2564 return SC_FLUSH_FILE;
2565 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2566 return SC_FLUSH_DAT;
2567 }
2568 return SC_LSEG_SR;
2569}
2570
2571/**
2572 * nilfs_segctor_thread - main loop of the segment constructor thread.
2573 * @arg: pointer to a struct nilfs_sc_info.
2574 *
2575 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2576 * to execute segment constructions.
2577 */
2578static int nilfs_segctor_thread(void *arg)
2579{
2580 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2581 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2582 int timeout = 0;
2583
2584 sci->sc_timer_task = current;
2585
2586 /* start sync. */
2587 sci->sc_task = current;
2588 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2589 nilfs_info(sci->sc_super,
2590 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2591 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2592
2593 set_freezable();
2594 spin_lock(&sci->sc_state_lock);
2595 loop:
2596 for (;;) {
2597 int mode;
2598
2599 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2600 goto end_thread;
2601
2602 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2603 mode = SC_LSEG_SR;
2604 else if (sci->sc_flush_request)
2605 mode = nilfs_segctor_flush_mode(sci);
2606 else
2607 break;
2608
2609 spin_unlock(&sci->sc_state_lock);
2610 nilfs_segctor_thread_construct(sci, mode);
2611 spin_lock(&sci->sc_state_lock);
2612 timeout = 0;
2613 }
2614
2615
2616 if (freezing(current)) {
2617 spin_unlock(&sci->sc_state_lock);
2618 try_to_freeze();
2619 spin_lock(&sci->sc_state_lock);
2620 } else {
2621 DEFINE_WAIT(wait);
2622 int should_sleep = 1;
2623
2624 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2625 TASK_INTERRUPTIBLE);
2626
2627 if (sci->sc_seq_request != sci->sc_seq_done)
2628 should_sleep = 0;
2629 else if (sci->sc_flush_request)
2630 should_sleep = 0;
2631 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2632 should_sleep = time_before(jiffies,
2633 sci->sc_timer.expires);
2634
2635 if (should_sleep) {
2636 spin_unlock(&sci->sc_state_lock);
2637 schedule();
2638 spin_lock(&sci->sc_state_lock);
2639 }
2640 finish_wait(&sci->sc_wait_daemon, &wait);
2641 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2642 time_after_eq(jiffies, sci->sc_timer.expires));
2643
2644 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2645 set_nilfs_discontinued(nilfs);
2646 }
2647 goto loop;
2648
2649 end_thread:
2650 /* end sync. */
2651 sci->sc_task = NULL;
2652 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2653 spin_unlock(&sci->sc_state_lock);
2654 return 0;
2655}
2656
2657static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2658{
2659 struct task_struct *t;
2660
2661 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2662 if (IS_ERR(t)) {
2663 int err = PTR_ERR(t);
2664
2665 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2666 err);
2667 return err;
2668 }
2669 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2670 return 0;
2671}
2672
2673static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2674 __acquires(&sci->sc_state_lock)
2675 __releases(&sci->sc_state_lock)
2676{
2677 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2678
2679 while (sci->sc_task) {
2680 wake_up(&sci->sc_wait_daemon);
2681 spin_unlock(&sci->sc_state_lock);
2682 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2683 spin_lock(&sci->sc_state_lock);
2684 }
2685}
2686
2687/*
2688 * Setup & clean-up functions
2689 */
2690static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2691 struct nilfs_root *root)
2692{
2693 struct the_nilfs *nilfs = sb->s_fs_info;
2694 struct nilfs_sc_info *sci;
2695
2696 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2697 if (!sci)
2698 return NULL;
2699
2700 sci->sc_super = sb;
2701
2702 nilfs_get_root(root);
2703 sci->sc_root = root;
2704
2705 init_waitqueue_head(&sci->sc_wait_request);
2706 init_waitqueue_head(&sci->sc_wait_daemon);
2707 init_waitqueue_head(&sci->sc_wait_task);
2708 spin_lock_init(&sci->sc_state_lock);
2709 INIT_LIST_HEAD(&sci->sc_dirty_files);
2710 INIT_LIST_HEAD(&sci->sc_segbufs);
2711 INIT_LIST_HEAD(&sci->sc_write_logs);
2712 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2713 INIT_LIST_HEAD(&sci->sc_iput_queue);
2714 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2715 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2716
2717 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2718 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2719 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2720
2721 if (nilfs->ns_interval)
2722 sci->sc_interval = HZ * nilfs->ns_interval;
2723 if (nilfs->ns_watermark)
2724 sci->sc_watermark = nilfs->ns_watermark;
2725 return sci;
2726}
2727
2728static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2729{
2730 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2731
2732 /*
2733 * The segctord thread was stopped and its timer was removed.
2734 * But some tasks remain.
2735 */
2736 do {
2737 struct nilfs_transaction_info ti;
2738
2739 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2740 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2741 nilfs_transaction_unlock(sci->sc_super);
2742
2743 flush_work(&sci->sc_iput_work);
2744
2745 } while (ret && ret != -EROFS && retrycount-- > 0);
2746}
2747
2748/**
2749 * nilfs_segctor_destroy - destroy the segment constructor.
2750 * @sci: nilfs_sc_info
2751 *
2752 * nilfs_segctor_destroy() kills the segctord thread and frees
2753 * the nilfs_sc_info struct.
2754 * Caller must hold the segment semaphore.
2755 */
2756static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2757{
2758 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2759 int flag;
2760
2761 up_write(&nilfs->ns_segctor_sem);
2762
2763 spin_lock(&sci->sc_state_lock);
2764 nilfs_segctor_kill_thread(sci);
2765 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2766 || sci->sc_seq_request != sci->sc_seq_done);
2767 spin_unlock(&sci->sc_state_lock);
2768
2769 if (flush_work(&sci->sc_iput_work))
2770 flag = true;
2771
2772 if (flag || !nilfs_segctor_confirm(sci))
2773 nilfs_segctor_write_out(sci);
2774
2775 if (!list_empty(&sci->sc_dirty_files)) {
2776 nilfs_warn(sci->sc_super,
2777 "disposed unprocessed dirty file(s) when stopping log writer");
2778 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2779 }
2780
2781 if (!list_empty(&sci->sc_iput_queue)) {
2782 nilfs_warn(sci->sc_super,
2783 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2784 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2785 }
2786
2787 WARN_ON(!list_empty(&sci->sc_segbufs));
2788 WARN_ON(!list_empty(&sci->sc_write_logs));
2789
2790 nilfs_put_root(sci->sc_root);
2791
2792 down_write(&nilfs->ns_segctor_sem);
2793
2794 timer_shutdown_sync(&sci->sc_timer);
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}
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) - (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 (ii->i_type & NILFS_I_TYPE_GC)
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
738 bh = head;
739 do {
740 if (!buffer_dirty(bh) || buffer_async_write(bh))
741 continue;
742 get_bh(bh);
743 list_add_tail(&bh->b_assoc_buffers, listp);
744 ndirties++;
745 if (unlikely(ndirties >= nlimit)) {
746 folio_unlock(folio);
747 folio_batch_release(&fbatch);
748 cond_resched();
749 return ndirties;
750 }
751 } while (bh = bh->b_this_page, bh != head);
752
753 folio_unlock(folio);
754 }
755 folio_batch_release(&fbatch);
756 cond_resched();
757 goto repeat;
758}
759
760static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
761 struct list_head *listp)
762{
763 struct nilfs_inode_info *ii = NILFS_I(inode);
764 struct inode *btnc_inode = ii->i_assoc_inode;
765 struct folio_batch fbatch;
766 struct buffer_head *bh, *head;
767 unsigned int i;
768 pgoff_t index = 0;
769
770 if (!btnc_inode)
771 return;
772 folio_batch_init(&fbatch);
773
774 while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
775 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
776 for (i = 0; i < folio_batch_count(&fbatch); i++) {
777 bh = head = folio_buffers(fbatch.folios[i]);
778 do {
779 if (buffer_dirty(bh) &&
780 !buffer_async_write(bh)) {
781 get_bh(bh);
782 list_add_tail(&bh->b_assoc_buffers,
783 listp);
784 }
785 bh = bh->b_this_page;
786 } while (bh != head);
787 }
788 folio_batch_release(&fbatch);
789 cond_resched();
790 }
791}
792
793static void nilfs_dispose_list(struct the_nilfs *nilfs,
794 struct list_head *head, int force)
795{
796 struct nilfs_inode_info *ii, *n;
797 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
798 unsigned int nv = 0;
799
800 while (!list_empty(head)) {
801 spin_lock(&nilfs->ns_inode_lock);
802 list_for_each_entry_safe(ii, n, head, i_dirty) {
803 list_del_init(&ii->i_dirty);
804 if (force) {
805 if (unlikely(ii->i_bh)) {
806 brelse(ii->i_bh);
807 ii->i_bh = NULL;
808 }
809 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
810 set_bit(NILFS_I_QUEUED, &ii->i_state);
811 list_add_tail(&ii->i_dirty,
812 &nilfs->ns_dirty_files);
813 continue;
814 }
815 ivec[nv++] = ii;
816 if (nv == SC_N_INODEVEC)
817 break;
818 }
819 spin_unlock(&nilfs->ns_inode_lock);
820
821 for (pii = ivec; nv > 0; pii++, nv--)
822 iput(&(*pii)->vfs_inode);
823 }
824}
825
826static void nilfs_iput_work_func(struct work_struct *work)
827{
828 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
829 sc_iput_work);
830 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
831
832 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
833}
834
835static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
836 struct nilfs_root *root)
837{
838 int ret = 0;
839
840 if (nilfs_mdt_fetch_dirty(root->ifile))
841 ret++;
842 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
843 ret++;
844 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
845 ret++;
846 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
847 ret++;
848 return ret;
849}
850
851static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
852{
853 return list_empty(&sci->sc_dirty_files) &&
854 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
855 sci->sc_nfreesegs == 0 &&
856 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
857}
858
859static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
860{
861 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
862 int ret = 0;
863
864 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
865 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
866
867 spin_lock(&nilfs->ns_inode_lock);
868 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
869 ret++;
870
871 spin_unlock(&nilfs->ns_inode_lock);
872 return ret;
873}
874
875static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
876{
877 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
878
879 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
880 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
881 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
882 nilfs_mdt_clear_dirty(nilfs->ns_dat);
883}
884
885static void nilfs_fill_in_file_bmap(struct inode *ifile,
886 struct nilfs_inode_info *ii)
887
888{
889 struct buffer_head *ibh;
890 struct nilfs_inode *raw_inode;
891
892 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
893 ibh = ii->i_bh;
894 BUG_ON(!ibh);
895 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
896 ibh);
897 nilfs_bmap_write(ii->i_bmap, raw_inode);
898 nilfs_ifile_unmap_inode(raw_inode);
899 }
900}
901
902static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
903{
904 struct nilfs_inode_info *ii;
905
906 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
907 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
908 set_bit(NILFS_I_COLLECTED, &ii->i_state);
909 }
910}
911
912/**
913 * nilfs_write_root_mdt_inode - export root metadata inode information to
914 * the on-disk inode
915 * @inode: inode object of the root metadata file
916 * @raw_inode: on-disk inode
917 *
918 * nilfs_write_root_mdt_inode() writes inode information and bmap data of
919 * @inode to the inode area of the metadata file allocated on the super root
920 * block created to finalize the log. Since super root blocks are configured
921 * each time, this function zero-fills the unused area of @raw_inode.
922 */
923static void nilfs_write_root_mdt_inode(struct inode *inode,
924 struct nilfs_inode *raw_inode)
925{
926 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
927
928 nilfs_write_inode_common(inode, raw_inode);
929
930 /* zero-fill unused portion of raw_inode */
931 raw_inode->i_xattr = 0;
932 raw_inode->i_pad = 0;
933 memset((void *)raw_inode + sizeof(*raw_inode), 0,
934 nilfs->ns_inode_size - sizeof(*raw_inode));
935
936 nilfs_bmap_write(NILFS_I(inode)->i_bmap, raw_inode);
937}
938
939static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
940 struct the_nilfs *nilfs)
941{
942 struct buffer_head *bh_sr;
943 struct nilfs_super_root *raw_sr;
944 unsigned int isz, srsz;
945
946 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
947
948 lock_buffer(bh_sr);
949 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
950 isz = nilfs->ns_inode_size;
951 srsz = NILFS_SR_BYTES(isz);
952
953 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
954 raw_sr->sr_bytes = cpu_to_le16(srsz);
955 raw_sr->sr_nongc_ctime
956 = cpu_to_le64(nilfs_doing_gc() ?
957 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
958 raw_sr->sr_flags = 0;
959
960 nilfs_write_root_mdt_inode(nilfs->ns_dat, (void *)raw_sr +
961 NILFS_SR_DAT_OFFSET(isz));
962 nilfs_write_root_mdt_inode(nilfs->ns_cpfile, (void *)raw_sr +
963 NILFS_SR_CPFILE_OFFSET(isz));
964 nilfs_write_root_mdt_inode(nilfs->ns_sufile, (void *)raw_sr +
965 NILFS_SR_SUFILE_OFFSET(isz));
966
967 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
968 set_buffer_uptodate(bh_sr);
969 unlock_buffer(bh_sr);
970}
971
972static void nilfs_redirty_inodes(struct list_head *head)
973{
974 struct nilfs_inode_info *ii;
975
976 list_for_each_entry(ii, head, i_dirty) {
977 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
978 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
979 }
980}
981
982static void nilfs_drop_collected_inodes(struct list_head *head)
983{
984 struct nilfs_inode_info *ii;
985
986 list_for_each_entry(ii, head, i_dirty) {
987 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
988 continue;
989
990 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
991 set_bit(NILFS_I_UPDATED, &ii->i_state);
992 }
993}
994
995static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
996 struct inode *inode,
997 struct list_head *listp,
998 int (*collect)(struct nilfs_sc_info *,
999 struct buffer_head *,
1000 struct inode *))
1001{
1002 struct buffer_head *bh, *n;
1003 int err = 0;
1004
1005 if (collect) {
1006 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1007 list_del_init(&bh->b_assoc_buffers);
1008 err = collect(sci, bh, inode);
1009 brelse(bh);
1010 if (unlikely(err))
1011 goto dispose_buffers;
1012 }
1013 return 0;
1014 }
1015
1016 dispose_buffers:
1017 while (!list_empty(listp)) {
1018 bh = list_first_entry(listp, struct buffer_head,
1019 b_assoc_buffers);
1020 list_del_init(&bh->b_assoc_buffers);
1021 brelse(bh);
1022 }
1023 return err;
1024}
1025
1026static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1027{
1028 /* Remaining number of blocks within segment buffer */
1029 return sci->sc_segbuf_nblocks -
1030 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1031}
1032
1033static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1034 struct inode *inode,
1035 const struct nilfs_sc_operations *sc_ops)
1036{
1037 LIST_HEAD(data_buffers);
1038 LIST_HEAD(node_buffers);
1039 int err;
1040
1041 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1042 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1043
1044 n = nilfs_lookup_dirty_data_buffers(
1045 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1046 if (n > rest) {
1047 err = nilfs_segctor_apply_buffers(
1048 sci, inode, &data_buffers,
1049 sc_ops->collect_data);
1050 BUG_ON(!err); /* always receive -E2BIG or true error */
1051 goto break_or_fail;
1052 }
1053 }
1054 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1055
1056 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1057 err = nilfs_segctor_apply_buffers(
1058 sci, inode, &data_buffers, sc_ops->collect_data);
1059 if (unlikely(err)) {
1060 /* dispose node list */
1061 nilfs_segctor_apply_buffers(
1062 sci, inode, &node_buffers, NULL);
1063 goto break_or_fail;
1064 }
1065 sci->sc_stage.flags |= NILFS_CF_NODE;
1066 }
1067 /* Collect node */
1068 err = nilfs_segctor_apply_buffers(
1069 sci, inode, &node_buffers, sc_ops->collect_node);
1070 if (unlikely(err))
1071 goto break_or_fail;
1072
1073 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1074 err = nilfs_segctor_apply_buffers(
1075 sci, inode, &node_buffers, sc_ops->collect_bmap);
1076 if (unlikely(err))
1077 goto break_or_fail;
1078
1079 nilfs_segctor_end_finfo(sci, inode);
1080 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1081
1082 break_or_fail:
1083 return err;
1084}
1085
1086static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1087 struct inode *inode)
1088{
1089 LIST_HEAD(data_buffers);
1090 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1091 int err;
1092
1093 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1094 sci->sc_dsync_start,
1095 sci->sc_dsync_end);
1096
1097 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1098 nilfs_collect_file_data);
1099 if (!err) {
1100 nilfs_segctor_end_finfo(sci, inode);
1101 BUG_ON(n > rest);
1102 /* always receive -E2BIG or true error if n > rest */
1103 }
1104 return err;
1105}
1106
1107/**
1108 * nilfs_free_segments - free the segments given by an array of segment numbers
1109 * @nilfs: nilfs object
1110 * @segnumv: array of segment numbers to be freed
1111 * @nsegs: number of segments to be freed in @segnumv
1112 *
1113 * nilfs_free_segments() wraps nilfs_sufile_freev() and
1114 * nilfs_sufile_cancel_freev(), and edits the segment usage metadata file
1115 * (sufile) to free all segments given by @segnumv and @nsegs at once. If
1116 * it fails midway, it cancels the changes so that none of the segments are
1117 * freed. If @nsegs is 0, this function does nothing.
1118 *
1119 * The freeing of segments is not finalized until the writing of a log with
1120 * a super root block containing this sufile change is complete, and it can
1121 * be canceled with nilfs_sufile_cancel_freev() until then.
1122 *
1123 * Return: 0 on success, or the following negative error code on failure.
1124 * * %-EINVAL - Invalid segment number.
1125 * * %-EIO - I/O error (including metadata corruption).
1126 * * %-ENOMEM - Insufficient memory available.
1127 */
1128static int nilfs_free_segments(struct the_nilfs *nilfs, __u64 *segnumv,
1129 size_t nsegs)
1130{
1131 size_t ndone;
1132 int ret;
1133
1134 if (!nsegs)
1135 return 0;
1136
1137 ret = nilfs_sufile_freev(nilfs->ns_sufile, segnumv, nsegs, &ndone);
1138 if (unlikely(ret)) {
1139 nilfs_sufile_cancel_freev(nilfs->ns_sufile, segnumv, ndone,
1140 NULL);
1141 /*
1142 * If a segment usage of the segments to be freed is in a
1143 * hole block, nilfs_sufile_freev() will return -ENOENT.
1144 * In this case, -EINVAL should be returned to the caller
1145 * since there is something wrong with the given segment
1146 * number array. This error can only occur during GC, so
1147 * there is no need to worry about it propagating to other
1148 * callers (such as fsync).
1149 */
1150 if (ret == -ENOENT) {
1151 nilfs_err(nilfs->ns_sb,
1152 "The segment usage entry %llu to be freed is invalid (in a hole)",
1153 (unsigned long long)segnumv[ndone]);
1154 ret = -EINVAL;
1155 }
1156 }
1157 return ret;
1158}
1159
1160static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1161{
1162 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1163 struct list_head *head;
1164 struct nilfs_inode_info *ii;
1165 int err = 0;
1166
1167 switch (nilfs_sc_cstage_get(sci)) {
1168 case NILFS_ST_INIT:
1169 /* Pre-processes */
1170 sci->sc_stage.flags = 0;
1171
1172 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1173 sci->sc_nblk_inc = 0;
1174 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1175 if (mode == SC_LSEG_DSYNC) {
1176 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1177 goto dsync_mode;
1178 }
1179 }
1180
1181 sci->sc_stage.dirty_file_ptr = NULL;
1182 sci->sc_stage.gc_inode_ptr = NULL;
1183 if (mode == SC_FLUSH_DAT) {
1184 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1185 goto dat_stage;
1186 }
1187 nilfs_sc_cstage_inc(sci);
1188 fallthrough;
1189 case NILFS_ST_GC:
1190 if (nilfs_doing_gc()) {
1191 head = &sci->sc_gc_inodes;
1192 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1193 head, i_dirty);
1194 list_for_each_entry_continue(ii, head, i_dirty) {
1195 err = nilfs_segctor_scan_file(
1196 sci, &ii->vfs_inode,
1197 &nilfs_sc_file_ops);
1198 if (unlikely(err)) {
1199 sci->sc_stage.gc_inode_ptr = list_entry(
1200 ii->i_dirty.prev,
1201 struct nilfs_inode_info,
1202 i_dirty);
1203 goto break_or_fail;
1204 }
1205 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1206 }
1207 sci->sc_stage.gc_inode_ptr = NULL;
1208 }
1209 nilfs_sc_cstage_inc(sci);
1210 fallthrough;
1211 case NILFS_ST_FILE:
1212 head = &sci->sc_dirty_files;
1213 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1214 i_dirty);
1215 list_for_each_entry_continue(ii, head, i_dirty) {
1216 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1217
1218 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1219 &nilfs_sc_file_ops);
1220 if (unlikely(err)) {
1221 sci->sc_stage.dirty_file_ptr =
1222 list_entry(ii->i_dirty.prev,
1223 struct nilfs_inode_info,
1224 i_dirty);
1225 goto break_or_fail;
1226 }
1227 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1228 /* XXX: required ? */
1229 }
1230 sci->sc_stage.dirty_file_ptr = NULL;
1231 if (mode == SC_FLUSH_FILE) {
1232 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1233 return 0;
1234 }
1235 nilfs_sc_cstage_inc(sci);
1236 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1237 fallthrough;
1238 case NILFS_ST_IFILE:
1239 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1240 &nilfs_sc_file_ops);
1241 if (unlikely(err))
1242 break;
1243 nilfs_sc_cstage_inc(sci);
1244 /* Creating a checkpoint */
1245 err = nilfs_cpfile_create_checkpoint(nilfs->ns_cpfile,
1246 nilfs->ns_cno);
1247 if (unlikely(err))
1248 break;
1249 fallthrough;
1250 case NILFS_ST_CPFILE:
1251 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1252 &nilfs_sc_file_ops);
1253 if (unlikely(err))
1254 break;
1255 nilfs_sc_cstage_inc(sci);
1256 fallthrough;
1257 case NILFS_ST_SUFILE:
1258 err = nilfs_free_segments(nilfs, sci->sc_freesegs,
1259 sci->sc_nfreesegs);
1260 if (unlikely(err))
1261 break;
1262 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1263
1264 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1265 &nilfs_sc_file_ops);
1266 if (unlikely(err))
1267 break;
1268 nilfs_sc_cstage_inc(sci);
1269 fallthrough;
1270 case NILFS_ST_DAT:
1271 dat_stage:
1272 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1273 &nilfs_sc_dat_ops);
1274 if (unlikely(err))
1275 break;
1276 if (mode == SC_FLUSH_DAT) {
1277 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1278 return 0;
1279 }
1280 nilfs_sc_cstage_inc(sci);
1281 fallthrough;
1282 case NILFS_ST_SR:
1283 if (mode == SC_LSEG_SR) {
1284 /* Appending a super root */
1285 err = nilfs_segctor_add_super_root(sci);
1286 if (unlikely(err))
1287 break;
1288 }
1289 /* End of a logical segment */
1290 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1291 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1292 return 0;
1293 case NILFS_ST_DSYNC:
1294 dsync_mode:
1295 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1296 ii = sci->sc_dsync_inode;
1297 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1298 break;
1299
1300 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1301 if (unlikely(err))
1302 break;
1303 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1304 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1305 return 0;
1306 case NILFS_ST_DONE:
1307 return 0;
1308 default:
1309 BUG();
1310 }
1311
1312 break_or_fail:
1313 return err;
1314}
1315
1316/**
1317 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1318 * @sci: nilfs_sc_info
1319 * @nilfs: nilfs object
1320 */
1321static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1322 struct the_nilfs *nilfs)
1323{
1324 struct nilfs_segment_buffer *segbuf, *prev;
1325 __u64 nextnum;
1326 int err, alloc = 0;
1327
1328 segbuf = nilfs_segbuf_new(sci->sc_super);
1329 if (unlikely(!segbuf))
1330 return -ENOMEM;
1331
1332 if (list_empty(&sci->sc_write_logs)) {
1333 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1334 nilfs->ns_pseg_offset, nilfs);
1335 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1336 nilfs_shift_to_next_segment(nilfs);
1337 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1338 }
1339
1340 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1341 nextnum = nilfs->ns_nextnum;
1342
1343 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1344 /* Start from the head of a new full segment */
1345 alloc++;
1346 } else {
1347 /* Continue logs */
1348 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1349 nilfs_segbuf_map_cont(segbuf, prev);
1350 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1351 nextnum = prev->sb_nextnum;
1352
1353 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1354 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1355 segbuf->sb_sum.seg_seq++;
1356 alloc++;
1357 }
1358 }
1359
1360 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1361 if (err)
1362 goto failed;
1363
1364 if (alloc) {
1365 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1366 if (err)
1367 goto failed;
1368 }
1369 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1370
1371 BUG_ON(!list_empty(&sci->sc_segbufs));
1372 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1373 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1374 return 0;
1375
1376 failed:
1377 nilfs_segbuf_free(segbuf);
1378 return err;
1379}
1380
1381static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1382 struct the_nilfs *nilfs, int nadd)
1383{
1384 struct nilfs_segment_buffer *segbuf, *prev;
1385 struct inode *sufile = nilfs->ns_sufile;
1386 __u64 nextnextnum;
1387 LIST_HEAD(list);
1388 int err, ret, i;
1389
1390 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1391 /*
1392 * Since the segment specified with nextnum might be allocated during
1393 * the previous construction, the buffer including its segusage may
1394 * not be dirty. The following call ensures that the buffer is dirty
1395 * and will pin the buffer on memory until the sufile is written.
1396 */
1397 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1398 if (unlikely(err))
1399 return err;
1400
1401 for (i = 0; i < nadd; i++) {
1402 /* extend segment info */
1403 err = -ENOMEM;
1404 segbuf = nilfs_segbuf_new(sci->sc_super);
1405 if (unlikely(!segbuf))
1406 goto failed;
1407
1408 /* map this buffer to region of segment on-disk */
1409 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1410 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1411
1412 /* allocate the next next full segment */
1413 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1414 if (unlikely(err))
1415 goto failed_segbuf;
1416
1417 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1418 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1419
1420 list_add_tail(&segbuf->sb_list, &list);
1421 prev = segbuf;
1422 }
1423 list_splice_tail(&list, &sci->sc_segbufs);
1424 return 0;
1425
1426 failed_segbuf:
1427 nilfs_segbuf_free(segbuf);
1428 failed:
1429 list_for_each_entry(segbuf, &list, sb_list) {
1430 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1431 WARN_ON(ret); /* never fails */
1432 }
1433 nilfs_destroy_logs(&list);
1434 return err;
1435}
1436
1437static void nilfs_free_incomplete_logs(struct list_head *logs,
1438 struct the_nilfs *nilfs)
1439{
1440 struct nilfs_segment_buffer *segbuf, *prev;
1441 struct inode *sufile = nilfs->ns_sufile;
1442 int ret;
1443
1444 segbuf = NILFS_FIRST_SEGBUF(logs);
1445 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1446 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1447 WARN_ON(ret); /* never fails */
1448 }
1449 if (atomic_read(&segbuf->sb_err)) {
1450 /* Case 1: The first segment failed */
1451 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1452 /*
1453 * Case 1a: Partial segment appended into an existing
1454 * segment
1455 */
1456 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1457 segbuf->sb_fseg_end);
1458 else /* Case 1b: New full segment */
1459 set_nilfs_discontinued(nilfs);
1460 }
1461
1462 prev = segbuf;
1463 list_for_each_entry_continue(segbuf, logs, sb_list) {
1464 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1465 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1466 WARN_ON(ret); /* never fails */
1467 }
1468 if (atomic_read(&segbuf->sb_err) &&
1469 segbuf->sb_segnum != nilfs->ns_nextnum)
1470 /* Case 2: extended segment (!= next) failed */
1471 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1472 prev = segbuf;
1473 }
1474}
1475
1476static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1477 struct inode *sufile)
1478{
1479 struct nilfs_segment_buffer *segbuf;
1480 unsigned long live_blocks;
1481 int ret;
1482
1483 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1484 live_blocks = segbuf->sb_sum.nblocks +
1485 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1486 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1487 live_blocks,
1488 sci->sc_seg_ctime);
1489 WARN_ON(ret); /* always succeed because the segusage is dirty */
1490 }
1491}
1492
1493static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1494{
1495 struct nilfs_segment_buffer *segbuf;
1496 int ret;
1497
1498 segbuf = NILFS_FIRST_SEGBUF(logs);
1499 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1500 segbuf->sb_pseg_start -
1501 segbuf->sb_fseg_start, 0);
1502 WARN_ON(ret); /* always succeed because the segusage is dirty */
1503
1504 list_for_each_entry_continue(segbuf, logs, sb_list) {
1505 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1506 0, 0);
1507 WARN_ON(ret); /* always succeed */
1508 }
1509}
1510
1511static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1512 struct nilfs_segment_buffer *last,
1513 struct inode *sufile)
1514{
1515 struct nilfs_segment_buffer *segbuf = last;
1516 int ret;
1517
1518 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1519 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1520 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1521 WARN_ON(ret);
1522 }
1523 nilfs_truncate_logs(&sci->sc_segbufs, last);
1524}
1525
1526
1527static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1528 struct the_nilfs *nilfs, int mode)
1529{
1530 struct nilfs_cstage prev_stage = sci->sc_stage;
1531 int err, nadd = 1;
1532
1533 /* Collection retry loop */
1534 for (;;) {
1535 sci->sc_nblk_this_inc = 0;
1536 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1537
1538 err = nilfs_segctor_reset_segment_buffer(sci);
1539 if (unlikely(err))
1540 goto failed;
1541
1542 err = nilfs_segctor_collect_blocks(sci, mode);
1543 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1544 if (!err)
1545 break;
1546
1547 if (unlikely(err != -E2BIG))
1548 goto failed;
1549
1550 /* The current segment is filled up */
1551 if (mode != SC_LSEG_SR ||
1552 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1553 break;
1554
1555 nilfs_clear_logs(&sci->sc_segbufs);
1556
1557 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1558 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1559 sci->sc_freesegs,
1560 sci->sc_nfreesegs,
1561 NULL);
1562 WARN_ON(err); /* do not happen */
1563 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1564 }
1565
1566 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1567 if (unlikely(err))
1568 return err;
1569
1570 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1571 sci->sc_stage = prev_stage;
1572 }
1573 nilfs_segctor_zeropad_segsum(sci);
1574 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1575 return 0;
1576
1577 failed:
1578 return err;
1579}
1580
1581static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1582 struct buffer_head *new_bh)
1583{
1584 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1585
1586 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1587 /* The caller must release old_bh */
1588}
1589
1590static int
1591nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1592 struct nilfs_segment_buffer *segbuf,
1593 int mode)
1594{
1595 struct inode *inode = NULL;
1596 sector_t blocknr;
1597 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1598 unsigned long nblocks = 0, ndatablk = 0;
1599 const struct nilfs_sc_operations *sc_op = NULL;
1600 struct nilfs_segsum_pointer ssp;
1601 struct nilfs_finfo *finfo = NULL;
1602 union nilfs_binfo binfo;
1603 struct buffer_head *bh, *bh_org;
1604 ino_t ino = 0;
1605 int err = 0;
1606
1607 if (!nfinfo)
1608 goto out;
1609
1610 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1611 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1612 ssp.offset = sizeof(struct nilfs_segment_summary);
1613
1614 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1615 if (bh == segbuf->sb_super_root)
1616 break;
1617 if (!finfo) {
1618 finfo = nilfs_segctor_map_segsum_entry(
1619 sci, &ssp, sizeof(*finfo));
1620 ino = le64_to_cpu(finfo->fi_ino);
1621 nblocks = le32_to_cpu(finfo->fi_nblocks);
1622 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1623
1624 inode = bh->b_folio->mapping->host;
1625
1626 if (mode == SC_LSEG_DSYNC)
1627 sc_op = &nilfs_sc_dsync_ops;
1628 else if (ino == NILFS_DAT_INO)
1629 sc_op = &nilfs_sc_dat_ops;
1630 else /* file blocks */
1631 sc_op = &nilfs_sc_file_ops;
1632 }
1633 bh_org = bh;
1634 get_bh(bh_org);
1635 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1636 &binfo);
1637 if (bh != bh_org)
1638 nilfs_list_replace_buffer(bh_org, bh);
1639 brelse(bh_org);
1640 if (unlikely(err))
1641 goto failed_bmap;
1642
1643 if (ndatablk > 0)
1644 sc_op->write_data_binfo(sci, &ssp, &binfo);
1645 else
1646 sc_op->write_node_binfo(sci, &ssp, &binfo);
1647
1648 blocknr++;
1649 if (--nblocks == 0) {
1650 finfo = NULL;
1651 if (--nfinfo == 0)
1652 break;
1653 } else if (ndatablk > 0)
1654 ndatablk--;
1655 }
1656 out:
1657 return 0;
1658
1659 failed_bmap:
1660 return err;
1661}
1662
1663static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1664{
1665 struct nilfs_segment_buffer *segbuf;
1666 int err;
1667
1668 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1669 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1670 if (unlikely(err))
1671 return err;
1672 nilfs_segbuf_fill_in_segsum(segbuf);
1673 }
1674 return 0;
1675}
1676
1677static void nilfs_begin_folio_io(struct folio *folio)
1678{
1679 if (!folio || folio_test_writeback(folio))
1680 /*
1681 * For split b-tree node pages, this function may be called
1682 * twice. We ignore the 2nd or later calls by this check.
1683 */
1684 return;
1685
1686 folio_lock(folio);
1687 folio_clear_dirty_for_io(folio);
1688 folio_start_writeback(folio);
1689 folio_unlock(folio);
1690}
1691
1692/**
1693 * nilfs_prepare_write_logs - prepare to write logs
1694 * @logs: logs to prepare for writing
1695 * @seed: checksum seed value
1696 *
1697 * nilfs_prepare_write_logs() adds checksums and prepares the block
1698 * buffers/folios for writing logs. In order to stabilize folios of
1699 * memory-mapped file blocks by putting them in writeback state before
1700 * calculating the checksums, first prepare to write payload blocks other
1701 * than segment summary and super root blocks in which the checksums will
1702 * be embedded.
1703 */
1704static void nilfs_prepare_write_logs(struct list_head *logs, u32 seed)
1705{
1706 struct nilfs_segment_buffer *segbuf;
1707 struct folio *bd_folio = NULL, *fs_folio = NULL;
1708 struct buffer_head *bh;
1709
1710 /* Prepare to write payload blocks */
1711 list_for_each_entry(segbuf, logs, sb_list) {
1712 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1713 b_assoc_buffers) {
1714 if (bh == segbuf->sb_super_root)
1715 break;
1716 set_buffer_async_write(bh);
1717 if (bh->b_folio != fs_folio) {
1718 nilfs_begin_folio_io(fs_folio);
1719 fs_folio = bh->b_folio;
1720 }
1721 }
1722 }
1723 nilfs_begin_folio_io(fs_folio);
1724
1725 nilfs_add_checksums_on_logs(logs, seed);
1726
1727 /* Prepare to write segment summary blocks */
1728 list_for_each_entry(segbuf, logs, sb_list) {
1729 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1730 b_assoc_buffers) {
1731 mark_buffer_dirty(bh);
1732 if (bh->b_folio == bd_folio)
1733 continue;
1734 if (bd_folio) {
1735 folio_lock(bd_folio);
1736 folio_wait_writeback(bd_folio);
1737 folio_clear_dirty_for_io(bd_folio);
1738 folio_start_writeback(bd_folio);
1739 folio_unlock(bd_folio);
1740 }
1741 bd_folio = bh->b_folio;
1742 }
1743 }
1744
1745 /* Prepare to write super root block */
1746 bh = NILFS_LAST_SEGBUF(logs)->sb_super_root;
1747 if (bh) {
1748 mark_buffer_dirty(bh);
1749 if (bh->b_folio != bd_folio) {
1750 folio_lock(bd_folio);
1751 folio_wait_writeback(bd_folio);
1752 folio_clear_dirty_for_io(bd_folio);
1753 folio_start_writeback(bd_folio);
1754 folio_unlock(bd_folio);
1755 bd_folio = bh->b_folio;
1756 }
1757 }
1758
1759 if (bd_folio) {
1760 folio_lock(bd_folio);
1761 folio_wait_writeback(bd_folio);
1762 folio_clear_dirty_for_io(bd_folio);
1763 folio_start_writeback(bd_folio);
1764 folio_unlock(bd_folio);
1765 }
1766}
1767
1768static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1769 struct the_nilfs *nilfs)
1770{
1771 int ret;
1772
1773 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1774 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1775 return ret;
1776}
1777
1778static void nilfs_end_folio_io(struct folio *folio, int err)
1779{
1780 if (!folio)
1781 return;
1782
1783 if (buffer_nilfs_node(folio_buffers(folio)) &&
1784 !folio_test_writeback(folio)) {
1785 /*
1786 * For b-tree node pages, this function may be called twice
1787 * or more because they might be split in a segment.
1788 */
1789 if (folio_test_dirty(folio)) {
1790 /*
1791 * For pages holding split b-tree node buffers, dirty
1792 * flag on the buffers may be cleared discretely.
1793 * In that case, the page is once redirtied for
1794 * remaining buffers, and it must be cancelled if
1795 * all the buffers get cleaned later.
1796 */
1797 folio_lock(folio);
1798 if (nilfs_folio_buffers_clean(folio))
1799 __nilfs_clear_folio_dirty(folio);
1800 folio_unlock(folio);
1801 }
1802 return;
1803 }
1804
1805 if (err || !nilfs_folio_buffers_clean(folio))
1806 filemap_dirty_folio(folio->mapping, folio);
1807
1808 folio_end_writeback(folio);
1809}
1810
1811static void nilfs_abort_logs(struct list_head *logs, int err)
1812{
1813 struct nilfs_segment_buffer *segbuf;
1814 struct folio *bd_folio = NULL, *fs_folio = NULL;
1815 struct buffer_head *bh;
1816
1817 if (list_empty(logs))
1818 return;
1819
1820 list_for_each_entry(segbuf, logs, sb_list) {
1821 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1822 b_assoc_buffers) {
1823 clear_buffer_uptodate(bh);
1824 if (bh->b_folio != bd_folio) {
1825 if (bd_folio)
1826 folio_end_writeback(bd_folio);
1827 bd_folio = bh->b_folio;
1828 }
1829 }
1830
1831 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1832 b_assoc_buffers) {
1833 if (bh == segbuf->sb_super_root) {
1834 clear_buffer_uptodate(bh);
1835 if (bh->b_folio != bd_folio) {
1836 folio_end_writeback(bd_folio);
1837 bd_folio = bh->b_folio;
1838 }
1839 break;
1840 }
1841 clear_buffer_async_write(bh);
1842 if (bh->b_folio != fs_folio) {
1843 nilfs_end_folio_io(fs_folio, err);
1844 fs_folio = bh->b_folio;
1845 }
1846 }
1847 }
1848 if (bd_folio)
1849 folio_end_writeback(bd_folio);
1850
1851 nilfs_end_folio_io(fs_folio, err);
1852}
1853
1854static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1855 struct the_nilfs *nilfs, int err)
1856{
1857 LIST_HEAD(logs);
1858 int ret;
1859
1860 list_splice_tail_init(&sci->sc_write_logs, &logs);
1861 ret = nilfs_wait_on_logs(&logs);
1862 nilfs_abort_logs(&logs, ret ? : err);
1863
1864 list_splice_tail_init(&sci->sc_segbufs, &logs);
1865 if (list_empty(&logs))
1866 return; /* if the first segment buffer preparation failed */
1867
1868 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1869 nilfs_free_incomplete_logs(&logs, nilfs);
1870
1871 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1872 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1873 sci->sc_freesegs,
1874 sci->sc_nfreesegs,
1875 NULL);
1876 WARN_ON(ret); /* do not happen */
1877 }
1878
1879 nilfs_destroy_logs(&logs);
1880}
1881
1882static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1883 struct nilfs_segment_buffer *segbuf)
1884{
1885 nilfs->ns_segnum = segbuf->sb_segnum;
1886 nilfs->ns_nextnum = segbuf->sb_nextnum;
1887 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1888 + segbuf->sb_sum.nblocks;
1889 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1890 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1891}
1892
1893static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1894{
1895 struct nilfs_segment_buffer *segbuf;
1896 struct folio *bd_folio = NULL, *fs_folio = NULL;
1897 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1898 int update_sr = false;
1899
1900 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1901 struct buffer_head *bh;
1902
1903 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1904 b_assoc_buffers) {
1905 set_buffer_uptodate(bh);
1906 clear_buffer_dirty(bh);
1907 if (bh->b_folio != bd_folio) {
1908 if (bd_folio)
1909 folio_end_writeback(bd_folio);
1910 bd_folio = bh->b_folio;
1911 }
1912 }
1913 /*
1914 * We assume that the buffers which belong to the same folio
1915 * continue over the buffer list.
1916 * Under this assumption, the last BHs of folios is
1917 * identifiable by the discontinuity of bh->b_folio
1918 * (folio != fs_folio).
1919 *
1920 * For B-tree node blocks, however, this assumption is not
1921 * guaranteed. The cleanup code of B-tree node folios needs
1922 * special care.
1923 */
1924 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1925 b_assoc_buffers) {
1926 const unsigned long set_bits = BIT(BH_Uptodate);
1927 const unsigned long clear_bits =
1928 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1929 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1930 BIT(BH_NILFS_Redirected));
1931
1932 if (bh == segbuf->sb_super_root) {
1933 set_buffer_uptodate(bh);
1934 clear_buffer_dirty(bh);
1935 if (bh->b_folio != bd_folio) {
1936 folio_end_writeback(bd_folio);
1937 bd_folio = bh->b_folio;
1938 }
1939 update_sr = true;
1940 break;
1941 }
1942 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1943 if (bh->b_folio != fs_folio) {
1944 nilfs_end_folio_io(fs_folio, 0);
1945 fs_folio = bh->b_folio;
1946 }
1947 }
1948
1949 if (!nilfs_segbuf_simplex(segbuf)) {
1950 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1951 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1952 sci->sc_lseg_stime = jiffies;
1953 }
1954 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1955 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1956 }
1957 }
1958 /*
1959 * Since folios may continue over multiple segment buffers,
1960 * end of the last folio must be checked outside of the loop.
1961 */
1962 if (bd_folio)
1963 folio_end_writeback(bd_folio);
1964
1965 nilfs_end_folio_io(fs_folio, 0);
1966
1967 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1968
1969 if (nilfs_doing_gc())
1970 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1971 else
1972 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1973
1974 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1975
1976 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1977 nilfs_set_next_segment(nilfs, segbuf);
1978
1979 if (update_sr) {
1980 nilfs->ns_flushed_device = 0;
1981 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1982 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1983
1984 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1985 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1986 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1987 nilfs_segctor_clear_metadata_dirty(sci);
1988 } else
1989 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1990}
1991
1992static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1993{
1994 int ret;
1995
1996 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1997 if (!ret) {
1998 nilfs_segctor_complete_write(sci);
1999 nilfs_destroy_logs(&sci->sc_write_logs);
2000 }
2001 return ret;
2002}
2003
2004static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
2005 struct the_nilfs *nilfs)
2006{
2007 struct nilfs_inode_info *ii, *n;
2008 struct inode *ifile = sci->sc_root->ifile;
2009
2010 spin_lock(&nilfs->ns_inode_lock);
2011 retry:
2012 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
2013 if (!ii->i_bh) {
2014 struct buffer_head *ibh;
2015 int err;
2016
2017 spin_unlock(&nilfs->ns_inode_lock);
2018 err = nilfs_ifile_get_inode_block(
2019 ifile, ii->vfs_inode.i_ino, &ibh);
2020 if (unlikely(err)) {
2021 nilfs_warn(sci->sc_super,
2022 "log writer: error %d getting inode block (ino=%lu)",
2023 err, ii->vfs_inode.i_ino);
2024 return err;
2025 }
2026 spin_lock(&nilfs->ns_inode_lock);
2027 if (likely(!ii->i_bh))
2028 ii->i_bh = ibh;
2029 else
2030 brelse(ibh);
2031 goto retry;
2032 }
2033
2034 // Always redirty the buffer to avoid race condition
2035 mark_buffer_dirty(ii->i_bh);
2036 nilfs_mdt_mark_dirty(ifile);
2037
2038 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2039 set_bit(NILFS_I_BUSY, &ii->i_state);
2040 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2041 }
2042 spin_unlock(&nilfs->ns_inode_lock);
2043
2044 return 0;
2045}
2046
2047static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2048 struct the_nilfs *nilfs)
2049{
2050 struct nilfs_inode_info *ii, *n;
2051 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2052 int defer_iput = false;
2053
2054 spin_lock(&nilfs->ns_inode_lock);
2055 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2056 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2057 test_bit(NILFS_I_DIRTY, &ii->i_state))
2058 continue;
2059
2060 clear_bit(NILFS_I_BUSY, &ii->i_state);
2061 brelse(ii->i_bh);
2062 ii->i_bh = NULL;
2063 list_del_init(&ii->i_dirty);
2064 if (!ii->vfs_inode.i_nlink || during_mount) {
2065 /*
2066 * Defer calling iput() to avoid deadlocks if
2067 * i_nlink == 0 or mount is not yet finished.
2068 */
2069 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2070 defer_iput = true;
2071 } else {
2072 spin_unlock(&nilfs->ns_inode_lock);
2073 iput(&ii->vfs_inode);
2074 spin_lock(&nilfs->ns_inode_lock);
2075 }
2076 }
2077 spin_unlock(&nilfs->ns_inode_lock);
2078
2079 if (defer_iput)
2080 schedule_work(&sci->sc_iput_work);
2081}
2082
2083/*
2084 * Main procedure of segment constructor
2085 */
2086static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2087{
2088 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2089 int err;
2090
2091 if (sb_rdonly(sci->sc_super))
2092 return -EROFS;
2093
2094 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2095 sci->sc_cno = nilfs->ns_cno;
2096
2097 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2098 if (unlikely(err))
2099 goto out;
2100
2101 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2102 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2103
2104 if (nilfs_segctor_clean(sci))
2105 goto out;
2106
2107 do {
2108 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2109
2110 err = nilfs_segctor_begin_construction(sci, nilfs);
2111 if (unlikely(err))
2112 goto failed;
2113
2114 /* Update time stamp */
2115 sci->sc_seg_ctime = ktime_get_real_seconds();
2116
2117 err = nilfs_segctor_collect(sci, nilfs, mode);
2118 if (unlikely(err))
2119 goto failed;
2120
2121 /* Avoid empty segment */
2122 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2123 nilfs_segbuf_empty(sci->sc_curseg)) {
2124 nilfs_segctor_abort_construction(sci, nilfs, 1);
2125 goto out;
2126 }
2127
2128 err = nilfs_segctor_assign(sci, mode);
2129 if (unlikely(err))
2130 goto failed;
2131
2132 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2133 nilfs_segctor_fill_in_file_bmap(sci);
2134
2135 if (mode == SC_LSEG_SR &&
2136 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2137 err = nilfs_cpfile_finalize_checkpoint(
2138 nilfs->ns_cpfile, nilfs->ns_cno, sci->sc_root,
2139 sci->sc_nblk_inc + sci->sc_nblk_this_inc,
2140 sci->sc_seg_ctime,
2141 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags));
2142 if (unlikely(err))
2143 goto failed_to_write;
2144
2145 nilfs_segctor_fill_in_super_root(sci, nilfs);
2146 }
2147 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2148
2149 /* Write partial segments */
2150 nilfs_prepare_write_logs(&sci->sc_segbufs, nilfs->ns_crc_seed);
2151
2152 err = nilfs_segctor_write(sci, nilfs);
2153 if (unlikely(err))
2154 goto failed_to_write;
2155
2156 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2157 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2158 /*
2159 * At this point, we avoid double buffering
2160 * for blocksize < pagesize because page dirty
2161 * flag is turned off during write and dirty
2162 * buffers are not properly collected for
2163 * pages crossing over segments.
2164 */
2165 err = nilfs_segctor_wait(sci);
2166 if (err)
2167 goto failed_to_write;
2168 }
2169 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2170
2171 out:
2172 nilfs_segctor_drop_written_files(sci, nilfs);
2173 return err;
2174
2175 failed_to_write:
2176 failed:
2177 if (mode == SC_LSEG_SR && nilfs_sc_cstage_get(sci) >= NILFS_ST_IFILE)
2178 nilfs_redirty_inodes(&sci->sc_dirty_files);
2179 if (nilfs_doing_gc())
2180 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2181 nilfs_segctor_abort_construction(sci, nilfs, err);
2182 goto out;
2183}
2184
2185/**
2186 * nilfs_segctor_start_timer - set timer of background write
2187 * @sci: nilfs_sc_info
2188 *
2189 * If the timer has already been set, it ignores the new request.
2190 * This function MUST be called within a section locking the segment
2191 * semaphore.
2192 */
2193static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2194{
2195 spin_lock(&sci->sc_state_lock);
2196 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2197 if (sci->sc_task) {
2198 sci->sc_timer.expires = jiffies + sci->sc_interval;
2199 add_timer(&sci->sc_timer);
2200 }
2201 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2202 }
2203 spin_unlock(&sci->sc_state_lock);
2204}
2205
2206static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2207{
2208 spin_lock(&sci->sc_state_lock);
2209 if (!(sci->sc_flush_request & BIT(bn))) {
2210 unsigned long prev_req = sci->sc_flush_request;
2211
2212 sci->sc_flush_request |= BIT(bn);
2213 if (!prev_req)
2214 wake_up(&sci->sc_wait_daemon);
2215 }
2216 spin_unlock(&sci->sc_state_lock);
2217}
2218
2219/**
2220 * nilfs_flush_segment - trigger a segment construction for resource control
2221 * @sb: super block
2222 * @ino: inode number of the file to be flushed out.
2223 */
2224void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2225{
2226 struct the_nilfs *nilfs = sb->s_fs_info;
2227 struct nilfs_sc_info *sci = nilfs->ns_writer;
2228
2229 if (!sci || nilfs_doing_construction())
2230 return;
2231 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2232 /* assign bit 0 to data files */
2233}
2234
2235struct nilfs_segctor_wait_request {
2236 wait_queue_entry_t wq;
2237 __u32 seq;
2238 int err;
2239 atomic_t done;
2240};
2241
2242static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2243{
2244 struct nilfs_segctor_wait_request wait_req;
2245 int err = 0;
2246
2247 init_wait(&wait_req.wq);
2248 wait_req.err = 0;
2249 atomic_set(&wait_req.done, 0);
2250 init_waitqueue_entry(&wait_req.wq, current);
2251
2252 /*
2253 * To prevent a race issue where completion notifications from the
2254 * log writer thread are missed, increment the request sequence count
2255 * "sc_seq_request" and insert a wait queue entry using the current
2256 * sequence number into the "sc_wait_request" queue at the same time
2257 * within the lock section of "sc_state_lock".
2258 */
2259 spin_lock(&sci->sc_state_lock);
2260 wait_req.seq = ++sci->sc_seq_request;
2261 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2262 spin_unlock(&sci->sc_state_lock);
2263
2264 wake_up(&sci->sc_wait_daemon);
2265
2266 for (;;) {
2267 set_current_state(TASK_INTERRUPTIBLE);
2268
2269 /*
2270 * Synchronize only while the log writer thread is alive.
2271 * Leave flushing out after the log writer thread exits to
2272 * the cleanup work in nilfs_segctor_destroy().
2273 */
2274 if (!sci->sc_task)
2275 break;
2276
2277 if (atomic_read(&wait_req.done)) {
2278 err = wait_req.err;
2279 break;
2280 }
2281 if (!signal_pending(current)) {
2282 schedule();
2283 continue;
2284 }
2285 err = -ERESTARTSYS;
2286 break;
2287 }
2288 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2289 return err;
2290}
2291
2292static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err, bool force)
2293{
2294 struct nilfs_segctor_wait_request *wrq, *n;
2295 unsigned long flags;
2296
2297 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2298 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2299 if (!atomic_read(&wrq->done) &&
2300 (force || nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq))) {
2301 wrq->err = err;
2302 atomic_set(&wrq->done, 1);
2303 }
2304 if (atomic_read(&wrq->done)) {
2305 wrq->wq.func(&wrq->wq,
2306 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2307 0, NULL);
2308 }
2309 }
2310 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2311}
2312
2313/**
2314 * nilfs_construct_segment - construct a logical segment
2315 * @sb: super block
2316 *
2317 * Return Value: On success, 0 is returned. On errors, one of the following
2318 * negative error code is returned.
2319 *
2320 * %-EROFS - Read only filesystem.
2321 *
2322 * %-EIO - I/O error
2323 *
2324 * %-ENOSPC - No space left on device (only in a panic state).
2325 *
2326 * %-ERESTARTSYS - Interrupted.
2327 *
2328 * %-ENOMEM - Insufficient memory available.
2329 */
2330int nilfs_construct_segment(struct super_block *sb)
2331{
2332 struct the_nilfs *nilfs = sb->s_fs_info;
2333 struct nilfs_sc_info *sci = nilfs->ns_writer;
2334 struct nilfs_transaction_info *ti;
2335
2336 if (sb_rdonly(sb) || unlikely(!sci))
2337 return -EROFS;
2338
2339 /* A call inside transactions causes a deadlock. */
2340 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2341
2342 return nilfs_segctor_sync(sci);
2343}
2344
2345/**
2346 * nilfs_construct_dsync_segment - construct a data-only logical segment
2347 * @sb: super block
2348 * @inode: inode whose data blocks should be written out
2349 * @start: start byte offset
2350 * @end: end byte offset (inclusive)
2351 *
2352 * Return Value: On success, 0 is returned. On errors, one of the following
2353 * negative error code is returned.
2354 *
2355 * %-EROFS - Read only filesystem.
2356 *
2357 * %-EIO - I/O error
2358 *
2359 * %-ENOSPC - No space left on device (only in a panic state).
2360 *
2361 * %-ERESTARTSYS - Interrupted.
2362 *
2363 * %-ENOMEM - Insufficient memory available.
2364 */
2365int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2366 loff_t start, loff_t end)
2367{
2368 struct the_nilfs *nilfs = sb->s_fs_info;
2369 struct nilfs_sc_info *sci = nilfs->ns_writer;
2370 struct nilfs_inode_info *ii;
2371 struct nilfs_transaction_info ti;
2372 int err = 0;
2373
2374 if (sb_rdonly(sb) || unlikely(!sci))
2375 return -EROFS;
2376
2377 nilfs_transaction_lock(sb, &ti, 0);
2378
2379 ii = NILFS_I(inode);
2380 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2381 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2382 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2383 nilfs_discontinued(nilfs)) {
2384 nilfs_transaction_unlock(sb);
2385 err = nilfs_segctor_sync(sci);
2386 return err;
2387 }
2388
2389 spin_lock(&nilfs->ns_inode_lock);
2390 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2391 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2392 spin_unlock(&nilfs->ns_inode_lock);
2393 nilfs_transaction_unlock(sb);
2394 return 0;
2395 }
2396 spin_unlock(&nilfs->ns_inode_lock);
2397 sci->sc_dsync_inode = ii;
2398 sci->sc_dsync_start = start;
2399 sci->sc_dsync_end = end;
2400
2401 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2402 if (!err)
2403 nilfs->ns_flushed_device = 0;
2404
2405 nilfs_transaction_unlock(sb);
2406 return err;
2407}
2408
2409#define FLUSH_FILE_BIT (0x1) /* data file only */
2410#define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2411
2412/**
2413 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2414 * @sci: segment constructor object
2415 */
2416static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2417{
2418 bool thread_is_alive;
2419
2420 spin_lock(&sci->sc_state_lock);
2421 sci->sc_seq_accepted = sci->sc_seq_request;
2422 thread_is_alive = (bool)sci->sc_task;
2423 spin_unlock(&sci->sc_state_lock);
2424
2425 /*
2426 * This function does not race with the log writer thread's
2427 * termination. Therefore, deleting sc_timer, which should not be
2428 * done after the log writer thread exits, can be done safely outside
2429 * the area protected by sc_state_lock.
2430 */
2431 if (thread_is_alive)
2432 del_timer_sync(&sci->sc_timer);
2433}
2434
2435/**
2436 * nilfs_segctor_notify - notify the result of request to caller threads
2437 * @sci: segment constructor object
2438 * @mode: mode of log forming
2439 * @err: error code to be notified
2440 */
2441static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2442{
2443 /* Clear requests (even when the construction failed) */
2444 spin_lock(&sci->sc_state_lock);
2445
2446 if (mode == SC_LSEG_SR) {
2447 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2448 sci->sc_seq_done = sci->sc_seq_accepted;
2449 nilfs_segctor_wakeup(sci, err, false);
2450 sci->sc_flush_request = 0;
2451 } else {
2452 if (mode == SC_FLUSH_FILE)
2453 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2454 else if (mode == SC_FLUSH_DAT)
2455 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2456
2457 /* re-enable timer if checkpoint creation was not done */
2458 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) && sci->sc_task &&
2459 time_before(jiffies, sci->sc_timer.expires))
2460 add_timer(&sci->sc_timer);
2461 }
2462 spin_unlock(&sci->sc_state_lock);
2463}
2464
2465/**
2466 * nilfs_segctor_construct - form logs and write them to disk
2467 * @sci: segment constructor object
2468 * @mode: mode of log forming
2469 */
2470static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2471{
2472 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2473 struct nilfs_super_block **sbp;
2474 int err = 0;
2475
2476 nilfs_segctor_accept(sci);
2477
2478 if (nilfs_discontinued(nilfs))
2479 mode = SC_LSEG_SR;
2480 if (!nilfs_segctor_confirm(sci))
2481 err = nilfs_segctor_do_construct(sci, mode);
2482
2483 if (likely(!err)) {
2484 if (mode != SC_FLUSH_DAT)
2485 atomic_set(&nilfs->ns_ndirtyblks, 0);
2486 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2487 nilfs_discontinued(nilfs)) {
2488 down_write(&nilfs->ns_sem);
2489 err = -EIO;
2490 sbp = nilfs_prepare_super(sci->sc_super,
2491 nilfs_sb_will_flip(nilfs));
2492 if (likely(sbp)) {
2493 nilfs_set_log_cursor(sbp[0], nilfs);
2494 err = nilfs_commit_super(sci->sc_super,
2495 NILFS_SB_COMMIT);
2496 }
2497 up_write(&nilfs->ns_sem);
2498 }
2499 }
2500
2501 nilfs_segctor_notify(sci, mode, err);
2502 return err;
2503}
2504
2505static void nilfs_construction_timeout(struct timer_list *t)
2506{
2507 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2508
2509 wake_up_process(sci->sc_task);
2510}
2511
2512static void
2513nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2514{
2515 struct nilfs_inode_info *ii, *n;
2516
2517 list_for_each_entry_safe(ii, n, head, i_dirty) {
2518 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2519 continue;
2520 list_del_init(&ii->i_dirty);
2521 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2522 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2523 iput(&ii->vfs_inode);
2524 }
2525}
2526
2527int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2528 void **kbufs)
2529{
2530 struct the_nilfs *nilfs = sb->s_fs_info;
2531 struct nilfs_sc_info *sci = nilfs->ns_writer;
2532 struct nilfs_transaction_info ti;
2533 int err;
2534
2535 if (unlikely(!sci))
2536 return -EROFS;
2537
2538 nilfs_transaction_lock(sb, &ti, 1);
2539
2540 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2541 if (unlikely(err))
2542 goto out_unlock;
2543
2544 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2545 if (unlikely(err)) {
2546 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2547 goto out_unlock;
2548 }
2549
2550 sci->sc_freesegs = kbufs[4];
2551 sci->sc_nfreesegs = argv[4].v_nmembs;
2552 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2553
2554 for (;;) {
2555 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2556 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2557
2558 if (likely(!err))
2559 break;
2560
2561 nilfs_warn(sb, "error %d cleaning segments", err);
2562 set_current_state(TASK_INTERRUPTIBLE);
2563 schedule_timeout(sci->sc_interval);
2564 }
2565 if (nilfs_test_opt(nilfs, DISCARD)) {
2566 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2567 sci->sc_nfreesegs);
2568 if (ret) {
2569 nilfs_warn(sb,
2570 "error %d on discard request, turning discards off for the device",
2571 ret);
2572 nilfs_clear_opt(nilfs, DISCARD);
2573 }
2574 }
2575
2576 out_unlock:
2577 sci->sc_freesegs = NULL;
2578 sci->sc_nfreesegs = 0;
2579 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2580 nilfs_transaction_unlock(sb);
2581 return err;
2582}
2583
2584static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2585{
2586 struct nilfs_transaction_info ti;
2587
2588 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2589 nilfs_segctor_construct(sci, mode);
2590
2591 /*
2592 * Unclosed segment should be retried. We do this using sc_timer.
2593 * Timeout of sc_timer will invoke complete construction which leads
2594 * to close the current logical segment.
2595 */
2596 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2597 nilfs_segctor_start_timer(sci);
2598
2599 nilfs_transaction_unlock(sci->sc_super);
2600}
2601
2602static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2603{
2604 int mode = 0;
2605
2606 spin_lock(&sci->sc_state_lock);
2607 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2608 SC_FLUSH_DAT : SC_FLUSH_FILE;
2609 spin_unlock(&sci->sc_state_lock);
2610
2611 if (mode) {
2612 nilfs_segctor_do_construct(sci, mode);
2613
2614 spin_lock(&sci->sc_state_lock);
2615 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2616 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2617 spin_unlock(&sci->sc_state_lock);
2618 }
2619 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2620}
2621
2622static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2623{
2624 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2625 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2626 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2627 return SC_FLUSH_FILE;
2628 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2629 return SC_FLUSH_DAT;
2630 }
2631 return SC_LSEG_SR;
2632}
2633
2634/**
2635 * nilfs_log_write_required - determine whether log writing is required
2636 * @sci: nilfs_sc_info struct
2637 * @modep: location for storing log writing mode
2638 *
2639 * Return: true if log writing is required, false otherwise. If log writing
2640 * is required, the mode is stored in the location pointed to by @modep.
2641 */
2642static bool nilfs_log_write_required(struct nilfs_sc_info *sci, int *modep)
2643{
2644 bool timedout, ret = true;
2645
2646 spin_lock(&sci->sc_state_lock);
2647 timedout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2648 time_after_eq(jiffies, sci->sc_timer.expires));
2649 if (timedout || sci->sc_seq_request != sci->sc_seq_done)
2650 *modep = SC_LSEG_SR;
2651 else if (sci->sc_flush_request)
2652 *modep = nilfs_segctor_flush_mode(sci);
2653 else
2654 ret = false;
2655
2656 spin_unlock(&sci->sc_state_lock);
2657 return ret;
2658}
2659
2660/**
2661 * nilfs_segctor_thread - main loop of the log writer thread
2662 * @arg: pointer to a struct nilfs_sc_info.
2663 *
2664 * nilfs_segctor_thread() is the main loop function of the log writer kernel
2665 * thread, which determines whether log writing is necessary, and if so,
2666 * performs the log write in the background, or waits if not. It is also
2667 * used to decide the background writeback of the superblock.
2668 *
2669 * Return: Always 0.
2670 */
2671static int nilfs_segctor_thread(void *arg)
2672{
2673 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2674 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2675
2676 nilfs_info(sci->sc_super,
2677 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2678 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2679
2680 set_freezable();
2681
2682 while (!kthread_should_stop()) {
2683 DEFINE_WAIT(wait);
2684 bool should_write;
2685 int mode;
2686
2687 if (freezing(current)) {
2688 try_to_freeze();
2689 continue;
2690 }
2691
2692 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2693 TASK_INTERRUPTIBLE);
2694 should_write = nilfs_log_write_required(sci, &mode);
2695 if (!should_write)
2696 schedule();
2697 finish_wait(&sci->sc_wait_daemon, &wait);
2698
2699 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2700 set_nilfs_discontinued(nilfs);
2701
2702 if (should_write)
2703 nilfs_segctor_thread_construct(sci, mode);
2704 }
2705
2706 /* end sync. */
2707 spin_lock(&sci->sc_state_lock);
2708 sci->sc_task = NULL;
2709 timer_shutdown_sync(&sci->sc_timer);
2710 spin_unlock(&sci->sc_state_lock);
2711 return 0;
2712}
2713
2714/*
2715 * Setup & clean-up functions
2716 */
2717static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2718 struct nilfs_root *root)
2719{
2720 struct the_nilfs *nilfs = sb->s_fs_info;
2721 struct nilfs_sc_info *sci;
2722
2723 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2724 if (!sci)
2725 return NULL;
2726
2727 sci->sc_super = sb;
2728
2729 nilfs_get_root(root);
2730 sci->sc_root = root;
2731
2732 init_waitqueue_head(&sci->sc_wait_request);
2733 init_waitqueue_head(&sci->sc_wait_daemon);
2734 spin_lock_init(&sci->sc_state_lock);
2735 INIT_LIST_HEAD(&sci->sc_dirty_files);
2736 INIT_LIST_HEAD(&sci->sc_segbufs);
2737 INIT_LIST_HEAD(&sci->sc_write_logs);
2738 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2739 INIT_LIST_HEAD(&sci->sc_iput_queue);
2740 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2741
2742 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2743 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2744 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2745
2746 if (nilfs->ns_interval)
2747 sci->sc_interval = HZ * nilfs->ns_interval;
2748 if (nilfs->ns_watermark)
2749 sci->sc_watermark = nilfs->ns_watermark;
2750 return sci;
2751}
2752
2753static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2754{
2755 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2756
2757 /*
2758 * The segctord thread was stopped and its timer was removed.
2759 * But some tasks remain.
2760 */
2761 do {
2762 struct nilfs_transaction_info ti;
2763
2764 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2765 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2766 nilfs_transaction_unlock(sci->sc_super);
2767
2768 flush_work(&sci->sc_iput_work);
2769
2770 } while (ret && ret != -EROFS && retrycount-- > 0);
2771}
2772
2773/**
2774 * nilfs_segctor_destroy - destroy the segment constructor.
2775 * @sci: nilfs_sc_info
2776 *
2777 * nilfs_segctor_destroy() kills the segctord thread and frees
2778 * the nilfs_sc_info struct.
2779 * Caller must hold the segment semaphore.
2780 */
2781static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2782{
2783 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2784 int flag;
2785
2786 up_write(&nilfs->ns_segctor_sem);
2787
2788 if (sci->sc_task) {
2789 wake_up(&sci->sc_wait_daemon);
2790 kthread_stop(sci->sc_task);
2791 }
2792
2793 spin_lock(&sci->sc_state_lock);
2794 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2795 || sci->sc_seq_request != sci->sc_seq_done);
2796 spin_unlock(&sci->sc_state_lock);
2797
2798 /*
2799 * Forcibly wake up tasks waiting in nilfs_segctor_sync(), which can
2800 * be called from delayed iput() via nilfs_evict_inode() and can race
2801 * with the above log writer thread termination.
2802 */
2803 nilfs_segctor_wakeup(sci, 0, true);
2804
2805 if (flush_work(&sci->sc_iput_work))
2806 flag = true;
2807
2808 if (flag || !nilfs_segctor_confirm(sci))
2809 nilfs_segctor_write_out(sci);
2810
2811 if (!list_empty(&sci->sc_dirty_files)) {
2812 nilfs_warn(sci->sc_super,
2813 "disposed unprocessed dirty file(s) when stopping log writer");
2814 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2815 }
2816
2817 if (!list_empty(&sci->sc_iput_queue)) {
2818 nilfs_warn(sci->sc_super,
2819 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2820 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2821 }
2822
2823 WARN_ON(!list_empty(&sci->sc_segbufs));
2824 WARN_ON(!list_empty(&sci->sc_write_logs));
2825
2826 nilfs_put_root(sci->sc_root);
2827
2828 down_write(&nilfs->ns_segctor_sem);
2829
2830 kfree(sci);
2831}
2832
2833/**
2834 * nilfs_attach_log_writer - attach log writer
2835 * @sb: super block instance
2836 * @root: root object of the current filesystem tree
2837 *
2838 * This allocates a log writer object, initializes it, and starts the
2839 * log writer.
2840 *
2841 * Return: 0 on success, or the following negative error code on failure.
2842 * * %-EINTR - Log writer thread creation failed due to interruption.
2843 * * %-ENOMEM - Insufficient memory available.
2844 */
2845int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2846{
2847 struct the_nilfs *nilfs = sb->s_fs_info;
2848 struct nilfs_sc_info *sci;
2849 struct task_struct *t;
2850 int err;
2851
2852 if (nilfs->ns_writer) {
2853 /*
2854 * This happens if the filesystem is made read-only by
2855 * __nilfs_error or nilfs_remount and then remounted
2856 * read/write. In these cases, reuse the existing
2857 * writer.
2858 */
2859 return 0;
2860 }
2861
2862 sci = nilfs_segctor_new(sb, root);
2863 if (unlikely(!sci))
2864 return -ENOMEM;
2865
2866 nilfs->ns_writer = sci;
2867 t = kthread_create(nilfs_segctor_thread, sci, "segctord");
2868 if (IS_ERR(t)) {
2869 err = PTR_ERR(t);
2870 nilfs_err(sb, "error %d creating segctord thread", err);
2871 nilfs_detach_log_writer(sb);
2872 return err;
2873 }
2874 sci->sc_task = t;
2875 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2876
2877 wake_up_process(sci->sc_task);
2878 return 0;
2879}
2880
2881/**
2882 * nilfs_detach_log_writer - destroy log writer
2883 * @sb: super block instance
2884 *
2885 * This kills log writer daemon, frees the log writer object, and
2886 * destroys list of dirty files.
2887 */
2888void nilfs_detach_log_writer(struct super_block *sb)
2889{
2890 struct the_nilfs *nilfs = sb->s_fs_info;
2891 LIST_HEAD(garbage_list);
2892
2893 down_write(&nilfs->ns_segctor_sem);
2894 if (nilfs->ns_writer) {
2895 nilfs_segctor_destroy(nilfs->ns_writer);
2896 nilfs->ns_writer = NULL;
2897 }
2898 set_nilfs_purging(nilfs);
2899
2900 /* Force to free the list of dirty files */
2901 spin_lock(&nilfs->ns_inode_lock);
2902 if (!list_empty(&nilfs->ns_dirty_files)) {
2903 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2904 nilfs_warn(sb,
2905 "disposed unprocessed dirty file(s) when detaching log writer");
2906 }
2907 spin_unlock(&nilfs->ns_inode_lock);
2908 up_write(&nilfs->ns_segctor_sem);
2909
2910 nilfs_dispose_list(nilfs, &garbage_list, 1);
2911 clear_nilfs_purging(nilfs);
2912}