projects/linux·blob master · refs

2931 lines · 78.9 KiB

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/folio_batch.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: 0 on success, or one of the following negative error codes on
195 * failure:
196 * * %-ENOMEM - Insufficient memory available.
197 * * %-ENOSPC - No space left on device (if checking free space).
198 */
199int nilfs_transaction_begin(struct super_block *sb,
200 struct nilfs_transaction_info *ti,
201 int vacancy_check)
202{
203 struct the_nilfs *nilfs;
204 int ret = nilfs_prepare_segment_lock(sb, ti);
205 struct nilfs_transaction_info *trace_ti;
206
207 if (unlikely(ret < 0))
208 return ret;
209 if (ret > 0) {
210 trace_ti = current->journal_info;
211
212 trace_nilfs2_transaction_transition(sb, trace_ti,
213 trace_ti->ti_count, trace_ti->ti_flags,
214 TRACE_NILFS2_TRANSACTION_BEGIN);
215 return 0;
216 }
217
218 sb_start_intwrite(sb);
219
220 nilfs = sb->s_fs_info;
221 down_read(&nilfs->ns_segctor_sem);
222 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
223 up_read(&nilfs->ns_segctor_sem);
224 ret = -ENOSPC;
225 goto failed;
226 }
227
228 trace_ti = current->journal_info;
229 trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
230 trace_ti->ti_flags,
231 TRACE_NILFS2_TRANSACTION_BEGIN);
232 return 0;
233
234 failed:
235 ti = current->journal_info;
236 current->journal_info = ti->ti_save;
237 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
238 kmem_cache_free(nilfs_transaction_cachep, ti);
239 sb_end_intwrite(sb);
240 return ret;
241}
242
243/**
244 * nilfs_transaction_commit - commit indivisible file operations.
245 * @sb: super block
246 *
247 * nilfs_transaction_commit() releases the read semaphore which is
248 * acquired by nilfs_transaction_begin(). This is only performed
249 * in outermost call of this function. If a commit flag is set,
250 * nilfs_transaction_commit() sets a timer to start the segment
251 * constructor. If a sync flag is set, it starts construction
252 * directly.
253 *
254 * Return: 0 on success, or a negative error code on failure.
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 *
411 * Return: 0 on success, or a negative error code on failure.
412 */
413static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
414{
415 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
416 struct buffer_head *sumbh;
417 unsigned int sumbytes;
418 unsigned int flags = 0;
419 int err;
420
421 if (nilfs_doing_gc())
422 flags = NILFS_SS_GC;
423 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
424 if (unlikely(err))
425 return err;
426
427 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
428 sumbytes = segbuf->sb_sum.sumbytes;
429 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
430 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
431 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
432 return 0;
433}
434
435/**
436 * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
437 * @sci: segment constructor object
438 *
439 * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
440 * the current segment summary block.
441 */
442static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
443{
444 struct nilfs_segsum_pointer *ssp;
445
446 ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
447 if (ssp->offset < ssp->bh->b_size)
448 memset(ssp->bh->b_data + ssp->offset, 0,
449 ssp->bh->b_size - ssp->offset);
450}
451
452static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
453{
454 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
455 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
456 return -E2BIG; /*
457 * The current segment is filled up
458 * (internal code)
459 */
460 nilfs_segctor_zeropad_segsum(sci);
461 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
462 return nilfs_segctor_reset_segment_buffer(sci);
463}
464
465static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
466{
467 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
468 int err;
469
470 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
471 err = nilfs_segctor_feed_segment(sci);
472 if (err)
473 return err;
474 segbuf = sci->sc_curseg;
475 }
476 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
477 if (likely(!err))
478 segbuf->sb_sum.flags |= NILFS_SS_SR;
479 return err;
480}
481
482/*
483 * Functions for making segment summary and payloads
484 */
485static int nilfs_segctor_segsum_block_required(
486 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
487 unsigned int binfo_size)
488{
489 unsigned int blocksize = sci->sc_super->s_blocksize;
490 /* Size of finfo and binfo is enough small against blocksize */
491
492 return ssp->offset + binfo_size +
493 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
494 blocksize;
495}
496
497static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
498 struct inode *inode)
499{
500 sci->sc_curseg->sb_sum.nfinfo++;
501 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
502 nilfs_segctor_map_segsum_entry(
503 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
504
505 if (NILFS_I(inode)->i_root &&
506 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
507 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
508 /* skip finfo */
509}
510
511static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
512 struct inode *inode)
513{
514 struct nilfs_finfo *finfo;
515 struct nilfs_inode_info *ii;
516 struct nilfs_segment_buffer *segbuf;
517 __u64 cno;
518
519 if (sci->sc_blk_cnt == 0)
520 return;
521
522 ii = NILFS_I(inode);
523
524 if (ii->i_type & NILFS_I_TYPE_GC)
525 cno = ii->i_cno;
526 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
527 cno = 0;
528 else
529 cno = sci->sc_cno;
530
531 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
532 sizeof(*finfo));
533 finfo->fi_ino = cpu_to_le64(inode->i_ino);
534 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
535 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
536 finfo->fi_cno = cpu_to_le64(cno);
537
538 segbuf = sci->sc_curseg;
539 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
540 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
541 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
542 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
543}
544
545static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
546 struct buffer_head *bh,
547 struct inode *inode,
548 unsigned int binfo_size)
549{
550 struct nilfs_segment_buffer *segbuf;
551 int required, err = 0;
552
553 retry:
554 segbuf = sci->sc_curseg;
555 required = nilfs_segctor_segsum_block_required(
556 sci, &sci->sc_binfo_ptr, binfo_size);
557 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
558 nilfs_segctor_end_finfo(sci, inode);
559 err = nilfs_segctor_feed_segment(sci);
560 if (err)
561 return err;
562 goto retry;
563 }
564 if (unlikely(required)) {
565 nilfs_segctor_zeropad_segsum(sci);
566 err = nilfs_segbuf_extend_segsum(segbuf);
567 if (unlikely(err))
568 goto failed;
569 }
570 if (sci->sc_blk_cnt == 0)
571 nilfs_segctor_begin_finfo(sci, inode);
572
573 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
574 /* Substitution to vblocknr is delayed until update_blocknr() */
575 nilfs_segbuf_add_file_buffer(segbuf, bh);
576 sci->sc_blk_cnt++;
577 failed:
578 return err;
579}
580
581/*
582 * Callback functions that enumerate, mark, and collect dirty blocks
583 */
584static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
585 struct buffer_head *bh, struct inode *inode)
586{
587 int err;
588
589 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
590 if (err < 0)
591 return err;
592
593 err = nilfs_segctor_add_file_block(sci, bh, inode,
594 sizeof(struct nilfs_binfo_v));
595 if (!err)
596 sci->sc_datablk_cnt++;
597 return err;
598}
599
600static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
601 struct buffer_head *bh,
602 struct inode *inode)
603{
604 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
605}
606
607static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
608 struct buffer_head *bh,
609 struct inode *inode)
610{
611 WARN_ON(!buffer_dirty(bh));
612 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
613}
614
615static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
616 struct nilfs_segsum_pointer *ssp,
617 union nilfs_binfo *binfo)
618{
619 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
620 sci, ssp, sizeof(*binfo_v));
621 *binfo_v = binfo->bi_v;
622}
623
624static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
625 struct nilfs_segsum_pointer *ssp,
626 union nilfs_binfo *binfo)
627{
628 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
629 sci, ssp, sizeof(*vblocknr));
630 *vblocknr = binfo->bi_v.bi_vblocknr;
631}
632
633static const struct nilfs_sc_operations nilfs_sc_file_ops = {
634 .collect_data = nilfs_collect_file_data,
635 .collect_node = nilfs_collect_file_node,
636 .collect_bmap = nilfs_collect_file_bmap,
637 .write_data_binfo = nilfs_write_file_data_binfo,
638 .write_node_binfo = nilfs_write_file_node_binfo,
639};
640
641static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
642 struct buffer_head *bh, struct inode *inode)
643{
644 int err;
645
646 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
647 if (err < 0)
648 return err;
649
650 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
651 if (!err)
652 sci->sc_datablk_cnt++;
653 return err;
654}
655
656static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
657 struct buffer_head *bh, struct inode *inode)
658{
659 WARN_ON(!buffer_dirty(bh));
660 return nilfs_segctor_add_file_block(sci, bh, inode,
661 sizeof(struct nilfs_binfo_dat));
662}
663
664static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
665 struct nilfs_segsum_pointer *ssp,
666 union nilfs_binfo *binfo)
667{
668 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
669 sizeof(*blkoff));
670 *blkoff = binfo->bi_dat.bi_blkoff;
671}
672
673static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
674 struct nilfs_segsum_pointer *ssp,
675 union nilfs_binfo *binfo)
676{
677 struct nilfs_binfo_dat *binfo_dat =
678 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
679 *binfo_dat = binfo->bi_dat;
680}
681
682static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
683 .collect_data = nilfs_collect_dat_data,
684 .collect_node = nilfs_collect_file_node,
685 .collect_bmap = nilfs_collect_dat_bmap,
686 .write_data_binfo = nilfs_write_dat_data_binfo,
687 .write_node_binfo = nilfs_write_dat_node_binfo,
688};
689
690static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
691 .collect_data = nilfs_collect_file_data,
692 .collect_node = NULL,
693 .collect_bmap = NULL,
694 .write_data_binfo = nilfs_write_file_data_binfo,
695 .write_node_binfo = NULL,
696};
697
698static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
699 struct list_head *listp,
700 size_t nlimit,
701 loff_t start, loff_t end)
702{
703 struct address_space *mapping = inode->i_mapping;
704 struct folio_batch fbatch;
705 pgoff_t index = 0, last = ULONG_MAX;
706 size_t ndirties = 0;
707 int i;
708
709 if (unlikely(start != 0 || end != LLONG_MAX)) {
710 /*
711 * A valid range is given for sync-ing data pages. The
712 * range is rounded to per-page; extra dirty buffers
713 * may be included if blocksize < pagesize.
714 */
715 index = start >> PAGE_SHIFT;
716 last = end >> PAGE_SHIFT;
717 }
718 folio_batch_init(&fbatch);
719 repeat:
720 if (unlikely(index > last) ||
721 !filemap_get_folios_tag(mapping, &index, last,
722 PAGECACHE_TAG_DIRTY, &fbatch))
723 return ndirties;
724
725 for (i = 0; i < folio_batch_count(&fbatch); i++) {
726 struct buffer_head *bh, *head;
727 struct folio *folio = fbatch.folios[i];
728
729 folio_lock(folio);
730 if (unlikely(folio->mapping != mapping)) {
731 /* Exclude folios removed from the address space */
732 folio_unlock(folio);
733 continue;
734 }
735 head = folio_buffers(folio);
736 if (!head)
737 head = create_empty_buffers(folio,
738 i_blocksize(inode), 0);
739
740 bh = head;
741 do {
742 if (!buffer_dirty(bh) || buffer_async_write(bh))
743 continue;
744 get_bh(bh);
745 list_add_tail(&bh->b_assoc_buffers, listp);
746 ndirties++;
747 if (unlikely(ndirties >= nlimit)) {
748 folio_unlock(folio);
749 folio_batch_release(&fbatch);
750 cond_resched();
751 return ndirties;
752 }
753 } while (bh = bh->b_this_page, bh != head);
754
755 folio_unlock(folio);
756 }
757 folio_batch_release(&fbatch);
758 cond_resched();
759 goto repeat;
760}
761
762static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
763 struct list_head *listp)
764{
765 struct nilfs_inode_info *ii = NILFS_I(inode);
766 struct inode *btnc_inode = ii->i_assoc_inode;
767 struct folio_batch fbatch;
768 struct buffer_head *bh, *head;
769 unsigned int i;
770 pgoff_t index = 0;
771
772 if (!btnc_inode)
773 return;
774 folio_batch_init(&fbatch);
775
776 while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
777 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
778 for (i = 0; i < folio_batch_count(&fbatch); i++) {
779 bh = head = folio_buffers(fbatch.folios[i]);
780 do {
781 if (buffer_dirty(bh) &&
782 !buffer_async_write(bh)) {
783 get_bh(bh);
784 list_add_tail(&bh->b_assoc_buffers,
785 listp);
786 }
787 bh = bh->b_this_page;
788 } while (bh != head);
789 }
790 folio_batch_release(&fbatch);
791 cond_resched();
792 }
793}
794
795static void nilfs_dispose_list(struct the_nilfs *nilfs,
796 struct list_head *head, int force)
797{
798 struct nilfs_inode_info *ii, *n;
799 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
800 unsigned int nv = 0;
801
802 while (!list_empty(head)) {
803 spin_lock(&nilfs->ns_inode_lock);
804 list_for_each_entry_safe(ii, n, head, i_dirty) {
805 list_del_init(&ii->i_dirty);
806 if (force) {
807 if (unlikely(ii->i_bh)) {
808 brelse(ii->i_bh);
809 ii->i_bh = NULL;
810 }
811 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
812 set_bit(NILFS_I_QUEUED, &ii->i_state);
813 list_add_tail(&ii->i_dirty,
814 &nilfs->ns_dirty_files);
815 continue;
816 }
817 ivec[nv++] = ii;
818 if (nv == SC_N_INODEVEC)
819 break;
820 }
821 spin_unlock(&nilfs->ns_inode_lock);
822
823 for (pii = ivec; nv > 0; pii++, nv--)
824 iput(&(*pii)->vfs_inode);
825 }
826}
827
828static void nilfs_iput_work_func(struct work_struct *work)
829{
830 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
831 sc_iput_work);
832 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
833
834 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
835}
836
837static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
838 struct nilfs_root *root)
839{
840 int ret = 0;
841
842 if (nilfs_mdt_fetch_dirty(root->ifile))
843 ret++;
844 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
845 ret++;
846 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
847 ret++;
848 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
849 ret++;
850 return ret;
851}
852
853static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
854{
855 return list_empty(&sci->sc_dirty_files) &&
856 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
857 sci->sc_nfreesegs == 0 &&
858 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
859}
860
861static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
862{
863 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
864 int ret = 0;
865
866 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
867 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
868
869 spin_lock(&nilfs->ns_inode_lock);
870 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
871 ret++;
872
873 spin_unlock(&nilfs->ns_inode_lock);
874 return ret;
875}
876
877static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
878{
879 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
880
881 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
882 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
883 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
884 nilfs_mdt_clear_dirty(nilfs->ns_dat);
885}
886
887static void nilfs_fill_in_file_bmap(struct inode *ifile,
888 struct nilfs_inode_info *ii)
889
890{
891 struct buffer_head *ibh;
892 struct nilfs_inode *raw_inode;
893
894 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
895 ibh = ii->i_bh;
896 BUG_ON(!ibh);
897 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
898 ibh);
899 nilfs_bmap_write(ii->i_bmap, raw_inode);
900 nilfs_ifile_unmap_inode(raw_inode);
901 }
902}
903
904static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
905{
906 struct nilfs_inode_info *ii;
907
908 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
909 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
910 set_bit(NILFS_I_COLLECTED, &ii->i_state);
911 }
912}
913
914/**
915 * nilfs_write_root_mdt_inode - export root metadata inode information to
916 * the on-disk inode
917 * @inode: inode object of the root metadata file
918 * @raw_inode: on-disk inode
919 *
920 * nilfs_write_root_mdt_inode() writes inode information and bmap data of
921 * @inode to the inode area of the metadata file allocated on the super root
922 * block created to finalize the log. Since super root blocks are configured
923 * each time, this function zero-fills the unused area of @raw_inode.
924 */
925static void nilfs_write_root_mdt_inode(struct inode *inode,
926 struct nilfs_inode *raw_inode)
927{
928 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
929
930 nilfs_write_inode_common(inode, raw_inode);
931
932 /* zero-fill unused portion of raw_inode */
933 raw_inode->i_xattr = 0;
934 raw_inode->i_pad = 0;
935 memset((void *)raw_inode + sizeof(*raw_inode), 0,
936 nilfs->ns_inode_size - sizeof(*raw_inode));
937
938 nilfs_bmap_write(NILFS_I(inode)->i_bmap, raw_inode);
939}
940
941static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
942 struct the_nilfs *nilfs)
943{
944 struct buffer_head *bh_sr;
945 struct nilfs_super_root *raw_sr;
946 unsigned int isz, srsz;
947
948 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
949
950 lock_buffer(bh_sr);
951 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
952 isz = nilfs->ns_inode_size;
953 srsz = NILFS_SR_BYTES(isz);
954
955 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
956 raw_sr->sr_bytes = cpu_to_le16(srsz);
957 raw_sr->sr_nongc_ctime
958 = cpu_to_le64(nilfs_doing_gc() ?
959 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
960 raw_sr->sr_flags = 0;
961
962 nilfs_write_root_mdt_inode(nilfs->ns_dat, (void *)raw_sr +
963 NILFS_SR_DAT_OFFSET(isz));
964 nilfs_write_root_mdt_inode(nilfs->ns_cpfile, (void *)raw_sr +
965 NILFS_SR_CPFILE_OFFSET(isz));
966 nilfs_write_root_mdt_inode(nilfs->ns_sufile, (void *)raw_sr +
967 NILFS_SR_SUFILE_OFFSET(isz));
968
969 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
970 set_buffer_uptodate(bh_sr);
971 unlock_buffer(bh_sr);
972}
973
974static void nilfs_redirty_inodes(struct list_head *head)
975{
976 struct nilfs_inode_info *ii;
977
978 list_for_each_entry(ii, head, i_dirty) {
979 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
980 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
981 }
982}
983
984static void nilfs_drop_collected_inodes(struct list_head *head)
985{
986 struct nilfs_inode_info *ii;
987
988 list_for_each_entry(ii, head, i_dirty) {
989 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
990 continue;
991
992 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
993 set_bit(NILFS_I_UPDATED, &ii->i_state);
994 }
995}
996
997static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
998 struct inode *inode,
999 struct list_head *listp,
1000 int (*collect)(struct nilfs_sc_info *,
1001 struct buffer_head *,
1002 struct inode *))
1003{
1004 struct buffer_head *bh, *n;
1005 int err = 0;
1006
1007 if (collect) {
1008 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1009 list_del_init(&bh->b_assoc_buffers);
1010 err = collect(sci, bh, inode);
1011 brelse(bh);
1012 if (unlikely(err))
1013 goto dispose_buffers;
1014 }
1015 return 0;
1016 }
1017
1018 dispose_buffers:
1019 while (!list_empty(listp)) {
1020 bh = list_first_entry(listp, struct buffer_head,
1021 b_assoc_buffers);
1022 list_del_init(&bh->b_assoc_buffers);
1023 brelse(bh);
1024 }
1025 return err;
1026}
1027
1028static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1029{
1030 /* Remaining number of blocks within segment buffer */
1031 return sci->sc_segbuf_nblocks -
1032 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1033}
1034
1035static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1036 struct inode *inode,
1037 const struct nilfs_sc_operations *sc_ops)
1038{
1039 LIST_HEAD(data_buffers);
1040 LIST_HEAD(node_buffers);
1041 int err;
1042
1043 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1044 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1045
1046 n = nilfs_lookup_dirty_data_buffers(
1047 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1048 if (n > rest) {
1049 err = nilfs_segctor_apply_buffers(
1050 sci, inode, &data_buffers,
1051 sc_ops->collect_data);
1052 BUG_ON(!err); /* always receive -E2BIG or true error */
1053 goto break_or_fail;
1054 }
1055 }
1056 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1057
1058 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1059 err = nilfs_segctor_apply_buffers(
1060 sci, inode, &data_buffers, sc_ops->collect_data);
1061 if (unlikely(err)) {
1062 /* dispose node list */
1063 nilfs_segctor_apply_buffers(
1064 sci, inode, &node_buffers, NULL);
1065 goto break_or_fail;
1066 }
1067 sci->sc_stage.flags |= NILFS_CF_NODE;
1068 }
1069 /* Collect node */
1070 err = nilfs_segctor_apply_buffers(
1071 sci, inode, &node_buffers, sc_ops->collect_node);
1072 if (unlikely(err))
1073 goto break_or_fail;
1074
1075 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1076 err = nilfs_segctor_apply_buffers(
1077 sci, inode, &node_buffers, sc_ops->collect_bmap);
1078 if (unlikely(err))
1079 goto break_or_fail;
1080
1081 nilfs_segctor_end_finfo(sci, inode);
1082 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1083
1084 break_or_fail:
1085 return err;
1086}
1087
1088static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1089 struct inode *inode)
1090{
1091 LIST_HEAD(data_buffers);
1092 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1093 int err;
1094
1095 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1096 sci->sc_dsync_start,
1097 sci->sc_dsync_end);
1098
1099 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1100 nilfs_collect_file_data);
1101 if (!err) {
1102 nilfs_segctor_end_finfo(sci, inode);
1103 BUG_ON(n > rest);
1104 /* always receive -E2BIG or true error if n > rest */
1105 }
1106 return err;
1107}
1108
1109/**
1110 * nilfs_free_segments - free the segments given by an array of segment numbers
1111 * @nilfs: nilfs object
1112 * @segnumv: array of segment numbers to be freed
1113 * @nsegs: number of segments to be freed in @segnumv
1114 *
1115 * nilfs_free_segments() wraps nilfs_sufile_freev() and
1116 * nilfs_sufile_cancel_freev(), and edits the segment usage metadata file
1117 * (sufile) to free all segments given by @segnumv and @nsegs at once. If
1118 * it fails midway, it cancels the changes so that none of the segments are
1119 * freed. If @nsegs is 0, this function does nothing.
1120 *
1121 * The freeing of segments is not finalized until the writing of a log with
1122 * a super root block containing this sufile change is complete, and it can
1123 * be canceled with nilfs_sufile_cancel_freev() until then.
1124 *
1125 * Return: 0 on success, or one of the following negative error codes on
1126 * failure:
1127 * * %-EINVAL - Invalid segment number.
1128 * * %-EIO - I/O error (including metadata corruption).
1129 * * %-ENOMEM - Insufficient memory available.
1130 */
1131static int nilfs_free_segments(struct the_nilfs *nilfs, __u64 *segnumv,
1132 size_t nsegs)
1133{
1134 size_t ndone;
1135 int ret;
1136
1137 if (!nsegs)
1138 return 0;
1139
1140 ret = nilfs_sufile_freev(nilfs->ns_sufile, segnumv, nsegs, &ndone);
1141 if (unlikely(ret)) {
1142 nilfs_sufile_cancel_freev(nilfs->ns_sufile, segnumv, ndone,
1143 NULL);
1144 /*
1145 * If a segment usage of the segments to be freed is in a
1146 * hole block, nilfs_sufile_freev() will return -ENOENT.
1147 * In this case, -EINVAL should be returned to the caller
1148 * since there is something wrong with the given segment
1149 * number array. This error can only occur during GC, so
1150 * there is no need to worry about it propagating to other
1151 * callers (such as fsync).
1152 */
1153 if (ret == -ENOENT) {
1154 nilfs_err(nilfs->ns_sb,
1155 "The segment usage entry %llu to be freed is invalid (in a hole)",
1156 (unsigned long long)segnumv[ndone]);
1157 ret = -EINVAL;
1158 }
1159 }
1160 return ret;
1161}
1162
1163static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1164{
1165 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1166 struct list_head *head;
1167 struct nilfs_inode_info *ii;
1168 int err = 0;
1169
1170 switch (nilfs_sc_cstage_get(sci)) {
1171 case NILFS_ST_INIT:
1172 /* Pre-processes */
1173 sci->sc_stage.flags = 0;
1174
1175 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1176 sci->sc_nblk_inc = 0;
1177 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1178 if (mode == SC_LSEG_DSYNC) {
1179 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1180 goto dsync_mode;
1181 }
1182 }
1183
1184 sci->sc_stage.dirty_file_ptr = NULL;
1185 sci->sc_stage.gc_inode_ptr = NULL;
1186 if (mode == SC_FLUSH_DAT) {
1187 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1188 goto dat_stage;
1189 }
1190 nilfs_sc_cstage_inc(sci);
1191 fallthrough;
1192 case NILFS_ST_GC:
1193 if (nilfs_doing_gc()) {
1194 head = &sci->sc_gc_inodes;
1195 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1196 head, i_dirty);
1197 list_for_each_entry_continue(ii, head, i_dirty) {
1198 err = nilfs_segctor_scan_file(
1199 sci, &ii->vfs_inode,
1200 &nilfs_sc_file_ops);
1201 if (unlikely(err)) {
1202 sci->sc_stage.gc_inode_ptr = list_entry(
1203 ii->i_dirty.prev,
1204 struct nilfs_inode_info,
1205 i_dirty);
1206 goto break_or_fail;
1207 }
1208 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1209 }
1210 sci->sc_stage.gc_inode_ptr = NULL;
1211 }
1212 nilfs_sc_cstage_inc(sci);
1213 fallthrough;
1214 case NILFS_ST_FILE:
1215 head = &sci->sc_dirty_files;
1216 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1217 i_dirty);
1218 list_for_each_entry_continue(ii, head, i_dirty) {
1219 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1220
1221 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1222 &nilfs_sc_file_ops);
1223 if (unlikely(err)) {
1224 sci->sc_stage.dirty_file_ptr =
1225 list_entry(ii->i_dirty.prev,
1226 struct nilfs_inode_info,
1227 i_dirty);
1228 goto break_or_fail;
1229 }
1230 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1231 /* XXX: required ? */
1232 }
1233 sci->sc_stage.dirty_file_ptr = NULL;
1234 if (mode == SC_FLUSH_FILE) {
1235 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1236 return 0;
1237 }
1238 nilfs_sc_cstage_inc(sci);
1239 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1240 fallthrough;
1241 case NILFS_ST_IFILE:
1242 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1243 &nilfs_sc_file_ops);
1244 if (unlikely(err))
1245 break;
1246 nilfs_sc_cstage_inc(sci);
1247 /* Creating a checkpoint */
1248 err = nilfs_cpfile_create_checkpoint(nilfs->ns_cpfile,
1249 nilfs->ns_cno);
1250 if (unlikely(err))
1251 break;
1252 fallthrough;
1253 case NILFS_ST_CPFILE:
1254 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1255 &nilfs_sc_file_ops);
1256 if (unlikely(err))
1257 break;
1258 nilfs_sc_cstage_inc(sci);
1259 fallthrough;
1260 case NILFS_ST_SUFILE:
1261 err = nilfs_free_segments(nilfs, sci->sc_freesegs,
1262 sci->sc_nfreesegs);
1263 if (unlikely(err))
1264 break;
1265 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1266
1267 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1268 &nilfs_sc_file_ops);
1269 if (unlikely(err))
1270 break;
1271 nilfs_sc_cstage_inc(sci);
1272 fallthrough;
1273 case NILFS_ST_DAT:
1274 dat_stage:
1275 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1276 &nilfs_sc_dat_ops);
1277 if (unlikely(err))
1278 break;
1279 if (mode == SC_FLUSH_DAT) {
1280 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1281 return 0;
1282 }
1283 nilfs_sc_cstage_inc(sci);
1284 fallthrough;
1285 case NILFS_ST_SR:
1286 if (mode == SC_LSEG_SR) {
1287 /* Appending a super root */
1288 err = nilfs_segctor_add_super_root(sci);
1289 if (unlikely(err))
1290 break;
1291 }
1292 /* End of a logical segment */
1293 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1294 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1295 return 0;
1296 case NILFS_ST_DSYNC:
1297 dsync_mode:
1298 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1299 ii = sci->sc_dsync_inode;
1300 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1301 break;
1302
1303 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1304 if (unlikely(err))
1305 break;
1306 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1307 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1308 return 0;
1309 case NILFS_ST_DONE:
1310 return 0;
1311 default:
1312 BUG();
1313 }
1314
1315 break_or_fail:
1316 return err;
1317}
1318
1319/**
1320 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1321 * @sci: nilfs_sc_info
1322 * @nilfs: nilfs object
1323 *
1324 * Return: 0 on success, or a negative error code on failure.
1325 */
1326static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1327 struct the_nilfs *nilfs)
1328{
1329 struct nilfs_segment_buffer *segbuf, *prev;
1330 __u64 nextnum;
1331 int err, alloc = 0;
1332
1333 segbuf = nilfs_segbuf_new(sci->sc_super);
1334 if (unlikely(!segbuf))
1335 return -ENOMEM;
1336
1337 if (list_empty(&sci->sc_write_logs)) {
1338 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1339 nilfs->ns_pseg_offset, nilfs);
1340 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1341 nilfs_shift_to_next_segment(nilfs);
1342 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1343 }
1344
1345 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1346 nextnum = nilfs->ns_nextnum;
1347
1348 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1349 /* Start from the head of a new full segment */
1350 alloc++;
1351 } else {
1352 /* Continue logs */
1353 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1354 nilfs_segbuf_map_cont(segbuf, prev);
1355 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1356 nextnum = prev->sb_nextnum;
1357
1358 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1359 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1360 segbuf->sb_sum.seg_seq++;
1361 alloc++;
1362 }
1363 }
1364
1365 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1366 if (err)
1367 goto failed;
1368
1369 if (alloc) {
1370 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1371 if (err)
1372 goto failed;
1373 }
1374 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1375
1376 BUG_ON(!list_empty(&sci->sc_segbufs));
1377 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1378 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1379 return 0;
1380
1381 failed:
1382 nilfs_segbuf_free(segbuf);
1383 return err;
1384}
1385
1386static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1387 struct the_nilfs *nilfs, int nadd)
1388{
1389 struct nilfs_segment_buffer *segbuf, *prev;
1390 struct inode *sufile = nilfs->ns_sufile;
1391 __u64 nextnextnum;
1392 LIST_HEAD(list);
1393 int err, ret, i;
1394
1395 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1396 /*
1397 * Since the segment specified with nextnum might be allocated during
1398 * the previous construction, the buffer including its segusage may
1399 * not be dirty. The following call ensures that the buffer is dirty
1400 * and will pin the buffer on memory until the sufile is written.
1401 */
1402 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1403 if (unlikely(err))
1404 return err;
1405
1406 for (i = 0; i < nadd; i++) {
1407 /* extend segment info */
1408 err = -ENOMEM;
1409 segbuf = nilfs_segbuf_new(sci->sc_super);
1410 if (unlikely(!segbuf))
1411 goto failed;
1412
1413 /* map this buffer to region of segment on-disk */
1414 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1415 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1416
1417 /* allocate the next next full segment */
1418 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1419 if (unlikely(err))
1420 goto failed_segbuf;
1421
1422 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1423 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1424
1425 list_add_tail(&segbuf->sb_list, &list);
1426 prev = segbuf;
1427 }
1428 list_splice_tail(&list, &sci->sc_segbufs);
1429 return 0;
1430
1431 failed_segbuf:
1432 nilfs_segbuf_free(segbuf);
1433 failed:
1434 list_for_each_entry(segbuf, &list, sb_list) {
1435 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1436 nilfs_sufile_warn_on_error(sufile, ret);
1437 }
1438 nilfs_destroy_logs(&list);
1439 return err;
1440}
1441
1442static void nilfs_free_incomplete_logs(struct list_head *logs,
1443 struct the_nilfs *nilfs)
1444{
1445 struct nilfs_segment_buffer *segbuf, *prev;
1446 struct inode *sufile = nilfs->ns_sufile;
1447 int ret;
1448
1449 segbuf = NILFS_FIRST_SEGBUF(logs);
1450 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1451 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1452 nilfs_sufile_warn_on_error(sufile, ret);
1453 }
1454 if (atomic_read(&segbuf->sb_err)) {
1455 /* Case 1: The first segment failed */
1456 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1457 /*
1458 * Case 1a: Partial segment appended into an existing
1459 * segment
1460 */
1461 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1462 segbuf->sb_fseg_end);
1463 else /* Case 1b: New full segment */
1464 set_nilfs_discontinued(nilfs);
1465 }
1466
1467 prev = segbuf;
1468 list_for_each_entry_continue(segbuf, logs, sb_list) {
1469 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1470 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1471 nilfs_sufile_warn_on_error(sufile, ret);
1472 }
1473 if (atomic_read(&segbuf->sb_err) &&
1474 segbuf->sb_segnum != nilfs->ns_nextnum)
1475 /* Case 2: extended segment (!= next) failed */
1476 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1477 prev = segbuf;
1478 }
1479}
1480
1481static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1482 struct inode *sufile)
1483{
1484 struct nilfs_segment_buffer *segbuf;
1485 unsigned long live_blocks;
1486 int ret;
1487
1488 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1489 live_blocks = segbuf->sb_sum.nblocks +
1490 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1491 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1492 live_blocks,
1493 sci->sc_seg_ctime);
1494 nilfs_sufile_warn_on_error(sufile, ret);
1495 }
1496}
1497
1498static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1499{
1500 struct nilfs_segment_buffer *segbuf;
1501 int ret;
1502
1503 segbuf = NILFS_FIRST_SEGBUF(logs);
1504 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1505 segbuf->sb_pseg_start -
1506 segbuf->sb_fseg_start, 0);
1507 nilfs_sufile_warn_on_error(sufile, ret);
1508
1509 list_for_each_entry_continue(segbuf, logs, sb_list) {
1510 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1511 0, 0);
1512 nilfs_sufile_warn_on_error(sufile, ret);
1513 }
1514}
1515
1516static int nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1517 struct nilfs_segment_buffer *last,
1518 struct inode *sufile)
1519{
1520 struct nilfs_segment_buffer *segbuf = last;
1521 int ret, err = 0;
1522
1523 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1524 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1525
1526 ret = nilfs_sufile_warn_on_error(
1527 sufile, nilfs_sufile_free(sufile, segbuf->sb_nextnum));
1528 if (unlikely(ret) && err != -EROFS)
1529 err = ret;
1530 }
1531 nilfs_truncate_logs(&sci->sc_segbufs, last);
1532 return err;
1533}
1534
1535
1536static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1537 struct the_nilfs *nilfs, int mode)
1538{
1539 struct nilfs_cstage prev_stage = sci->sc_stage;
1540 int err, nadd = 1;
1541
1542 /* Collection retry loop */
1543 for (;;) {
1544 sci->sc_nblk_this_inc = 0;
1545 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1546
1547 err = nilfs_segctor_reset_segment_buffer(sci);
1548 if (unlikely(err))
1549 goto failed;
1550
1551 err = nilfs_segctor_collect_blocks(sci, mode);
1552 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1553 if (!err)
1554 break;
1555
1556 if (unlikely(err != -E2BIG))
1557 goto failed;
1558
1559 /* The current segment is filled up */
1560 if (mode != SC_LSEG_SR ||
1561 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1562 break;
1563
1564 nilfs_clear_logs(&sci->sc_segbufs);
1565
1566 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1567 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1568 sci->sc_freesegs,
1569 sci->sc_nfreesegs,
1570 NULL);
1571 nilfs_sufile_warn_on_error(nilfs->ns_sufile, err);
1572 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1573 }
1574
1575 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1576 if (unlikely(err))
1577 return err;
1578
1579 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1580 sci->sc_stage = prev_stage;
1581 }
1582 nilfs_segctor_zeropad_segsum(sci);
1583 err = nilfs_segctor_truncate_segments(sci, sci->sc_curseg,
1584 nilfs->ns_sufile);
1585
1586 failed:
1587 return err;
1588}
1589
1590static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1591 struct buffer_head *new_bh)
1592{
1593 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1594
1595 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1596 /* The caller must release old_bh */
1597}
1598
1599static int
1600nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1601 struct nilfs_segment_buffer *segbuf,
1602 int mode)
1603{
1604 struct inode *inode = NULL;
1605 sector_t blocknr;
1606 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1607 unsigned long nblocks = 0, ndatablk = 0;
1608 const struct nilfs_sc_operations *sc_op = NULL;
1609 struct nilfs_segsum_pointer ssp;
1610 struct nilfs_finfo *finfo = NULL;
1611 union nilfs_binfo binfo;
1612 struct buffer_head *bh, *bh_org;
1613 u64 ino = 0;
1614 int err = 0;
1615
1616 if (!nfinfo)
1617 goto out;
1618
1619 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1620 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1621 ssp.offset = sizeof(struct nilfs_segment_summary);
1622
1623 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1624 if (bh == segbuf->sb_super_root)
1625 break;
1626 if (!finfo) {
1627 finfo = nilfs_segctor_map_segsum_entry(
1628 sci, &ssp, sizeof(*finfo));
1629 ino = le64_to_cpu(finfo->fi_ino);
1630 nblocks = le32_to_cpu(finfo->fi_nblocks);
1631 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1632
1633 inode = bh->b_folio->mapping->host;
1634
1635 if (mode == SC_LSEG_DSYNC)
1636 sc_op = &nilfs_sc_dsync_ops;
1637 else if (ino == NILFS_DAT_INO)
1638 sc_op = &nilfs_sc_dat_ops;
1639 else /* file blocks */
1640 sc_op = &nilfs_sc_file_ops;
1641 }
1642 bh_org = bh;
1643 get_bh(bh_org);
1644 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1645 &binfo);
1646 if (bh != bh_org)
1647 nilfs_list_replace_buffer(bh_org, bh);
1648 brelse(bh_org);
1649 if (unlikely(err))
1650 goto failed_bmap;
1651
1652 if (ndatablk > 0)
1653 sc_op->write_data_binfo(sci, &ssp, &binfo);
1654 else
1655 sc_op->write_node_binfo(sci, &ssp, &binfo);
1656
1657 blocknr++;
1658 if (--nblocks == 0) {
1659 finfo = NULL;
1660 if (--nfinfo == 0)
1661 break;
1662 } else if (ndatablk > 0)
1663 ndatablk--;
1664 }
1665 out:
1666 return 0;
1667
1668 failed_bmap:
1669 return err;
1670}
1671
1672static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1673{
1674 struct nilfs_segment_buffer *segbuf;
1675 int err;
1676
1677 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1678 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1679 if (unlikely(err))
1680 return err;
1681 nilfs_segbuf_fill_in_segsum(segbuf);
1682 }
1683 return 0;
1684}
1685
1686static void nilfs_begin_folio_io(struct folio *folio)
1687{
1688 if (!folio || folio_test_writeback(folio))
1689 /*
1690 * For split b-tree node pages, this function may be called
1691 * twice. We ignore the 2nd or later calls by this check.
1692 */
1693 return;
1694
1695 folio_lock(folio);
1696 folio_clear_dirty_for_io(folio);
1697 folio_start_writeback(folio);
1698 folio_unlock(folio);
1699}
1700
1701/**
1702 * nilfs_prepare_write_logs - prepare to write logs
1703 * @logs: logs to prepare for writing
1704 * @seed: checksum seed value
1705 *
1706 * nilfs_prepare_write_logs() adds checksums and prepares the block
1707 * buffers/folios for writing logs. In order to stabilize folios of
1708 * memory-mapped file blocks by putting them in writeback state before
1709 * calculating the checksums, first prepare to write payload blocks other
1710 * than segment summary and super root blocks in which the checksums will
1711 * be embedded.
1712 */
1713static void nilfs_prepare_write_logs(struct list_head *logs, u32 seed)
1714{
1715 struct nilfs_segment_buffer *segbuf;
1716 struct folio *bd_folio = NULL, *fs_folio = NULL;
1717 struct buffer_head *bh;
1718
1719 /* Prepare to write payload blocks */
1720 list_for_each_entry(segbuf, logs, sb_list) {
1721 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1722 b_assoc_buffers) {
1723 if (bh == segbuf->sb_super_root)
1724 break;
1725 set_buffer_async_write(bh);
1726 if (bh->b_folio != fs_folio) {
1727 nilfs_begin_folio_io(fs_folio);
1728 fs_folio = bh->b_folio;
1729 }
1730 }
1731 }
1732 nilfs_begin_folio_io(fs_folio);
1733
1734 nilfs_add_checksums_on_logs(logs, seed);
1735
1736 /* Prepare to write segment summary blocks */
1737 list_for_each_entry(segbuf, logs, sb_list) {
1738 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1739 b_assoc_buffers) {
1740 mark_buffer_dirty(bh);
1741 if (bh->b_folio == bd_folio)
1742 continue;
1743 if (bd_folio) {
1744 folio_lock(bd_folio);
1745 folio_wait_writeback(bd_folio);
1746 folio_clear_dirty_for_io(bd_folio);
1747 folio_start_writeback(bd_folio);
1748 folio_unlock(bd_folio);
1749 }
1750 bd_folio = bh->b_folio;
1751 }
1752 }
1753
1754 /* Prepare to write super root block */
1755 bh = NILFS_LAST_SEGBUF(logs)->sb_super_root;
1756 if (bh) {
1757 mark_buffer_dirty(bh);
1758 if (bh->b_folio != bd_folio) {
1759 folio_lock(bd_folio);
1760 folio_wait_writeback(bd_folio);
1761 folio_clear_dirty_for_io(bd_folio);
1762 folio_start_writeback(bd_folio);
1763 folio_unlock(bd_folio);
1764 bd_folio = bh->b_folio;
1765 }
1766 }
1767
1768 if (bd_folio) {
1769 folio_lock(bd_folio);
1770 folio_wait_writeback(bd_folio);
1771 folio_clear_dirty_for_io(bd_folio);
1772 folio_start_writeback(bd_folio);
1773 folio_unlock(bd_folio);
1774 }
1775}
1776
1777static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1778 struct the_nilfs *nilfs)
1779{
1780 int ret;
1781
1782 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1783 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1784 return ret;
1785}
1786
1787static void nilfs_end_folio_io(struct folio *folio, int err)
1788{
1789 if (!folio)
1790 return;
1791
1792 if (buffer_nilfs_node(folio_buffers(folio)) &&
1793 !folio_test_writeback(folio)) {
1794 /*
1795 * For b-tree node pages, this function may be called twice
1796 * or more because they might be split in a segment.
1797 */
1798 if (folio_test_dirty(folio)) {
1799 /*
1800 * For pages holding split b-tree node buffers, dirty
1801 * flag on the buffers may be cleared discretely.
1802 * In that case, the page is once redirtied for
1803 * remaining buffers, and it must be cancelled if
1804 * all the buffers get cleaned later.
1805 */
1806 folio_lock(folio);
1807 if (nilfs_folio_buffers_clean(folio))
1808 __nilfs_clear_folio_dirty(folio);
1809 folio_unlock(folio);
1810 }
1811 return;
1812 }
1813
1814 if (err || !nilfs_folio_buffers_clean(folio))
1815 filemap_dirty_folio(folio->mapping, folio);
1816
1817 folio_end_writeback(folio);
1818}
1819
1820static void nilfs_abort_logs(struct list_head *logs, int err)
1821{
1822 struct nilfs_segment_buffer *segbuf;
1823 struct folio *bd_folio = NULL, *fs_folio = NULL;
1824 struct buffer_head *bh;
1825
1826 if (list_empty(logs))
1827 return;
1828
1829 list_for_each_entry(segbuf, logs, sb_list) {
1830 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1831 b_assoc_buffers) {
1832 clear_buffer_dirty(bh);
1833 clear_buffer_uptodate(bh);
1834 if (bh->b_folio != bd_folio) {
1835 if (bd_folio)
1836 folio_end_writeback(bd_folio);
1837 bd_folio = bh->b_folio;
1838 }
1839 }
1840
1841 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1842 b_assoc_buffers) {
1843 if (bh == segbuf->sb_super_root) {
1844 clear_buffer_dirty(bh);
1845 clear_buffer_uptodate(bh);
1846 if (bh->b_folio != bd_folio) {
1847 folio_end_writeback(bd_folio);
1848 bd_folio = bh->b_folio;
1849 }
1850 break;
1851 }
1852 clear_buffer_async_write(bh);
1853 if (bh->b_folio != fs_folio) {
1854 nilfs_end_folio_io(fs_folio, err);
1855 fs_folio = bh->b_folio;
1856 }
1857 }
1858 }
1859 if (bd_folio)
1860 folio_end_writeback(bd_folio);
1861
1862 nilfs_end_folio_io(fs_folio, err);
1863}
1864
1865static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1866 struct the_nilfs *nilfs, int err)
1867{
1868 LIST_HEAD(logs);
1869 int ret;
1870
1871 list_splice_tail_init(&sci->sc_write_logs, &logs);
1872 ret = nilfs_wait_on_logs(&logs);
1873 nilfs_abort_logs(&logs, ret ? : err);
1874
1875 list_splice_tail_init(&sci->sc_segbufs, &logs);
1876 if (list_empty(&logs))
1877 return; /* if the first segment buffer preparation failed */
1878
1879 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1880 nilfs_free_incomplete_logs(&logs, nilfs);
1881
1882 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1883 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1884 sci->sc_freesegs,
1885 sci->sc_nfreesegs,
1886 NULL);
1887 nilfs_sufile_warn_on_error(nilfs->ns_sufile, ret);
1888 }
1889
1890 nilfs_destroy_logs(&logs);
1891}
1892
1893static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1894 struct nilfs_segment_buffer *segbuf)
1895{
1896 nilfs->ns_segnum = segbuf->sb_segnum;
1897 nilfs->ns_nextnum = segbuf->sb_nextnum;
1898 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1899 + segbuf->sb_sum.nblocks;
1900 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1901 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1902}
1903
1904static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1905{
1906 struct nilfs_segment_buffer *segbuf;
1907 struct folio *bd_folio = NULL, *fs_folio = NULL;
1908 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1909 int update_sr = false;
1910
1911 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1912 struct buffer_head *bh;
1913
1914 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1915 b_assoc_buffers) {
1916 set_buffer_uptodate(bh);
1917 clear_buffer_dirty(bh);
1918 if (bh->b_folio != bd_folio) {
1919 if (bd_folio)
1920 folio_end_writeback(bd_folio);
1921 bd_folio = bh->b_folio;
1922 }
1923 }
1924 /*
1925 * We assume that the buffers which belong to the same folio
1926 * continue over the buffer list.
1927 * Under this assumption, the last BHs of folios is
1928 * identifiable by the discontinuity of bh->b_folio
1929 * (folio != fs_folio).
1930 *
1931 * For B-tree node blocks, however, this assumption is not
1932 * guaranteed. The cleanup code of B-tree node folios needs
1933 * special care.
1934 */
1935 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1936 b_assoc_buffers) {
1937 const unsigned long set_bits = BIT(BH_Uptodate);
1938 const unsigned long clear_bits =
1939 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1940 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1941 BIT(BH_NILFS_Redirected));
1942
1943 if (bh == segbuf->sb_super_root) {
1944 set_buffer_uptodate(bh);
1945 clear_buffer_dirty(bh);
1946 if (bh->b_folio != bd_folio) {
1947 folio_end_writeback(bd_folio);
1948 bd_folio = bh->b_folio;
1949 }
1950 update_sr = true;
1951 break;
1952 }
1953 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1954 if (bh->b_folio != fs_folio) {
1955 nilfs_end_folio_io(fs_folio, 0);
1956 fs_folio = bh->b_folio;
1957 }
1958 }
1959
1960 if (!nilfs_segbuf_simplex(segbuf)) {
1961 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1962 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1963 sci->sc_lseg_stime = jiffies;
1964 }
1965 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1966 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1967 }
1968 }
1969 /*
1970 * Since folios may continue over multiple segment buffers,
1971 * end of the last folio must be checked outside of the loop.
1972 */
1973 if (bd_folio)
1974 folio_end_writeback(bd_folio);
1975
1976 nilfs_end_folio_io(fs_folio, 0);
1977
1978 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1979
1980 if (nilfs_doing_gc())
1981 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1982 else
1983 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1984
1985 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1986
1987 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1988 nilfs_set_next_segment(nilfs, segbuf);
1989
1990 if (update_sr) {
1991 nilfs->ns_flushed_device = 0;
1992 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1993 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1994
1995 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1996 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1997 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1998 nilfs_segctor_clear_metadata_dirty(sci);
1999 } else
2000 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2001}
2002
2003static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
2004{
2005 int ret;
2006
2007 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
2008 if (!ret) {
2009 nilfs_segctor_complete_write(sci);
2010 nilfs_destroy_logs(&sci->sc_write_logs);
2011 }
2012 return ret;
2013}
2014
2015static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
2016 struct the_nilfs *nilfs)
2017{
2018 struct nilfs_inode_info *ii, *n;
2019 struct inode *ifile = sci->sc_root->ifile;
2020
2021 spin_lock(&nilfs->ns_inode_lock);
2022 retry:
2023 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
2024 if (!ii->i_bh) {
2025 struct buffer_head *ibh;
2026 int err;
2027
2028 spin_unlock(&nilfs->ns_inode_lock);
2029 err = nilfs_ifile_get_inode_block(
2030 ifile, ii->vfs_inode.i_ino, &ibh);
2031 if (unlikely(err)) {
2032 nilfs_warn(sci->sc_super,
2033 "log writer: error %d getting inode block (ino=%llu)",
2034 err, ii->vfs_inode.i_ino);
2035 return err;
2036 }
2037 spin_lock(&nilfs->ns_inode_lock);
2038 if (likely(!ii->i_bh))
2039 ii->i_bh = ibh;
2040 else
2041 brelse(ibh);
2042 goto retry;
2043 }
2044
2045 // Always redirty the buffer to avoid race condition
2046 mark_buffer_dirty(ii->i_bh);
2047 nilfs_mdt_mark_dirty(ifile);
2048
2049 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2050 set_bit(NILFS_I_BUSY, &ii->i_state);
2051 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2052 }
2053 spin_unlock(&nilfs->ns_inode_lock);
2054
2055 return 0;
2056}
2057
2058static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2059 struct the_nilfs *nilfs)
2060{
2061 struct nilfs_inode_info *ii, *n;
2062 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2063 int defer_iput = false;
2064
2065 spin_lock(&nilfs->ns_inode_lock);
2066 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2067 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2068 test_bit(NILFS_I_DIRTY, &ii->i_state))
2069 continue;
2070
2071 clear_bit(NILFS_I_BUSY, &ii->i_state);
2072 brelse(ii->i_bh);
2073 ii->i_bh = NULL;
2074 list_del_init(&ii->i_dirty);
2075 if (!ii->vfs_inode.i_nlink || during_mount) {
2076 /*
2077 * Defer calling iput() to avoid deadlocks if
2078 * i_nlink == 0 or mount is not yet finished.
2079 */
2080 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2081 defer_iput = true;
2082 } else {
2083 spin_unlock(&nilfs->ns_inode_lock);
2084 iput(&ii->vfs_inode);
2085 spin_lock(&nilfs->ns_inode_lock);
2086 }
2087 }
2088 spin_unlock(&nilfs->ns_inode_lock);
2089
2090 if (defer_iput)
2091 schedule_work(&sci->sc_iput_work);
2092}
2093
2094/*
2095 * Main procedure of segment constructor
2096 */
2097static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2098{
2099 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2100 int err;
2101
2102 if (sb_rdonly(sci->sc_super))
2103 return -EROFS;
2104
2105 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2106 sci->sc_cno = nilfs->ns_cno;
2107
2108 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2109 if (unlikely(err))
2110 goto out;
2111
2112 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2113 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2114
2115 if (nilfs_segctor_clean(sci))
2116 goto out;
2117
2118 do {
2119 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2120
2121 err = nilfs_segctor_begin_construction(sci, nilfs);
2122 if (unlikely(err))
2123 goto failed;
2124
2125 /* Update time stamp */
2126 sci->sc_seg_ctime = ktime_get_real_seconds();
2127
2128 err = nilfs_segctor_collect(sci, nilfs, mode);
2129 if (unlikely(err))
2130 goto failed;
2131
2132 /* Avoid empty segment */
2133 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2134 nilfs_segbuf_empty(sci->sc_curseg)) {
2135 nilfs_segctor_abort_construction(sci, nilfs, 1);
2136 goto out;
2137 }
2138
2139 err = nilfs_segctor_assign(sci, mode);
2140 if (unlikely(err))
2141 goto failed;
2142
2143 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2144 nilfs_segctor_fill_in_file_bmap(sci);
2145
2146 if (mode == SC_LSEG_SR &&
2147 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2148 err = nilfs_cpfile_finalize_checkpoint(
2149 nilfs->ns_cpfile, nilfs->ns_cno, sci->sc_root,
2150 sci->sc_nblk_inc + sci->sc_nblk_this_inc,
2151 sci->sc_seg_ctime,
2152 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags));
2153 if (unlikely(err))
2154 goto failed_to_write;
2155
2156 nilfs_segctor_fill_in_super_root(sci, nilfs);
2157 }
2158 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2159
2160 /* Write partial segments */
2161 nilfs_prepare_write_logs(&sci->sc_segbufs, nilfs->ns_crc_seed);
2162
2163 err = nilfs_segctor_write(sci, nilfs);
2164 if (unlikely(err))
2165 goto failed_to_write;
2166
2167 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2168 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2169 /*
2170 * At this point, we avoid double buffering
2171 * for blocksize < pagesize because page dirty
2172 * flag is turned off during write and dirty
2173 * buffers are not properly collected for
2174 * pages crossing over segments.
2175 */
2176 err = nilfs_segctor_wait(sci);
2177 if (err)
2178 goto failed_to_write;
2179 }
2180 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2181
2182 out:
2183 nilfs_segctor_drop_written_files(sci, nilfs);
2184 return err;
2185
2186 failed_to_write:
2187 failed:
2188 if (mode == SC_LSEG_SR && nilfs_sc_cstage_get(sci) >= NILFS_ST_IFILE)
2189 nilfs_redirty_inodes(&sci->sc_dirty_files);
2190 if (nilfs_doing_gc())
2191 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2192 nilfs_segctor_abort_construction(sci, nilfs, err);
2193 goto out;
2194}
2195
2196/**
2197 * nilfs_segctor_start_timer - set timer of background write
2198 * @sci: nilfs_sc_info
2199 *
2200 * If the timer has already been set, it ignores the new request.
2201 * This function MUST be called within a section locking the segment
2202 * semaphore.
2203 */
2204static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2205{
2206 spin_lock(&sci->sc_state_lock);
2207 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2208 if (sci->sc_task) {
2209 sci->sc_timer.expires = jiffies + sci->sc_interval;
2210 add_timer(&sci->sc_timer);
2211 }
2212 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2213 }
2214 spin_unlock(&sci->sc_state_lock);
2215}
2216
2217static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2218{
2219 spin_lock(&sci->sc_state_lock);
2220 if (!(sci->sc_flush_request & BIT(bn))) {
2221 unsigned long prev_req = sci->sc_flush_request;
2222
2223 sci->sc_flush_request |= BIT(bn);
2224 if (!prev_req)
2225 wake_up(&sci->sc_wait_daemon);
2226 }
2227 spin_unlock(&sci->sc_state_lock);
2228}
2229
2230struct nilfs_segctor_wait_request {
2231 wait_queue_entry_t wq;
2232 __u32 seq;
2233 int err;
2234 atomic_t done;
2235};
2236
2237static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2238{
2239 struct nilfs_segctor_wait_request wait_req;
2240 int err = 0;
2241
2242 init_wait(&wait_req.wq);
2243 wait_req.err = 0;
2244 atomic_set(&wait_req.done, 0);
2245 init_waitqueue_entry(&wait_req.wq, current);
2246
2247 /*
2248 * To prevent a race issue where completion notifications from the
2249 * log writer thread are missed, increment the request sequence count
2250 * "sc_seq_request" and insert a wait queue entry using the current
2251 * sequence number into the "sc_wait_request" queue at the same time
2252 * within the lock section of "sc_state_lock".
2253 */
2254 spin_lock(&sci->sc_state_lock);
2255 wait_req.seq = ++sci->sc_seq_request;
2256 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2257 spin_unlock(&sci->sc_state_lock);
2258
2259 wake_up(&sci->sc_wait_daemon);
2260
2261 for (;;) {
2262 set_current_state(TASK_KILLABLE);
2263
2264 /*
2265 * Synchronize only while the log writer thread is alive.
2266 * Leave flushing out after the log writer thread exits to
2267 * the cleanup work in nilfs_segctor_destroy().
2268 */
2269 if (!sci->sc_task)
2270 break;
2271
2272 if (atomic_read(&wait_req.done)) {
2273 err = wait_req.err;
2274 break;
2275 }
2276 if (!fatal_signal_pending(current)) {
2277 schedule();
2278 continue;
2279 }
2280 err = -EINTR;
2281 break;
2282 }
2283 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2284 return err;
2285}
2286
2287static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err, bool force)
2288{
2289 struct nilfs_segctor_wait_request *wrq, *n;
2290 unsigned long flags;
2291
2292 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2293 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2294 if (!atomic_read(&wrq->done) &&
2295 (force || nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq))) {
2296 wrq->err = err;
2297 atomic_set(&wrq->done, 1);
2298 }
2299 if (atomic_read(&wrq->done)) {
2300 wrq->wq.func(&wrq->wq,
2301 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2302 0, NULL);
2303 }
2304 }
2305 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2306}
2307
2308/**
2309 * nilfs_construct_segment - construct a logical segment
2310 * @sb: super block
2311 *
2312 * Return: 0 on success, or one of the following negative error codes on
2313 * failure:
2314 * * %-EINTR - Interrupted.
2315 * * %-EIO - I/O error (including metadata corruption).
2316 * * %-ENOMEM - Insufficient memory available.
2317 * * %-ENOSPC - No space left on device (only in a panic state).
2318 * * %-EROFS - Read only filesystem.
2319 */
2320int nilfs_construct_segment(struct super_block *sb)
2321{
2322 struct the_nilfs *nilfs = sb->s_fs_info;
2323 struct nilfs_sc_info *sci = nilfs->ns_writer;
2324 struct nilfs_transaction_info *ti;
2325
2326 if (sb_rdonly(sb) || unlikely(!sci))
2327 return -EROFS;
2328
2329 /* A call inside transactions causes a deadlock. */
2330 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2331
2332 return nilfs_segctor_sync(sci);
2333}
2334
2335/**
2336 * nilfs_construct_dsync_segment - construct a data-only logical segment
2337 * @sb: super block
2338 * @inode: inode whose data blocks should be written out
2339 * @start: start byte offset
2340 * @end: end byte offset (inclusive)
2341 *
2342 * Return: 0 on success, or one of the following negative error codes on
2343 * failure:
2344 * * %-EINTR - Interrupted.
2345 * * %-EIO - I/O error (including metadata corruption).
2346 * * %-ENOMEM - Insufficient memory available.
2347 * * %-ENOSPC - No space left on device (only in a panic state).
2348 * * %-EROFS - Read only filesystem.
2349 */
2350int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2351 loff_t start, loff_t end)
2352{
2353 struct the_nilfs *nilfs = sb->s_fs_info;
2354 struct nilfs_sc_info *sci = nilfs->ns_writer;
2355 struct nilfs_inode_info *ii;
2356 struct nilfs_transaction_info ti;
2357 int err = 0;
2358
2359 if (sb_rdonly(sb) || unlikely(!sci))
2360 return -EROFS;
2361
2362 nilfs_transaction_lock(sb, &ti, 0);
2363
2364 ii = NILFS_I(inode);
2365 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2366 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2367 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2368 nilfs_discontinued(nilfs)) {
2369 nilfs_transaction_unlock(sb);
2370 err = nilfs_segctor_sync(sci);
2371 return err;
2372 }
2373
2374 spin_lock(&nilfs->ns_inode_lock);
2375 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2376 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2377 spin_unlock(&nilfs->ns_inode_lock);
2378 nilfs_transaction_unlock(sb);
2379 return 0;
2380 }
2381 spin_unlock(&nilfs->ns_inode_lock);
2382 sci->sc_dsync_inode = ii;
2383 sci->sc_dsync_start = start;
2384 sci->sc_dsync_end = end;
2385
2386 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2387 if (!err)
2388 nilfs->ns_flushed_device = 0;
2389
2390 nilfs_transaction_unlock(sb);
2391 return err;
2392}
2393
2394#define FLUSH_FILE_BIT (0x1) /* data file only */
2395#define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2396
2397/**
2398 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2399 * @sci: segment constructor object
2400 */
2401static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2402{
2403 bool thread_is_alive;
2404
2405 spin_lock(&sci->sc_state_lock);
2406 sci->sc_seq_accepted = sci->sc_seq_request;
2407 thread_is_alive = (bool)sci->sc_task;
2408 spin_unlock(&sci->sc_state_lock);
2409
2410 /*
2411 * This function does not race with the log writer thread's
2412 * termination. Therefore, deleting sc_timer, which should not be
2413 * done after the log writer thread exits, can be done safely outside
2414 * the area protected by sc_state_lock.
2415 */
2416 if (thread_is_alive)
2417 timer_delete_sync(&sci->sc_timer);
2418}
2419
2420/**
2421 * nilfs_segctor_notify - notify the result of request to caller threads
2422 * @sci: segment constructor object
2423 * @mode: mode of log forming
2424 * @err: error code to be notified
2425 */
2426static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2427{
2428 /* Clear requests (even when the construction failed) */
2429 spin_lock(&sci->sc_state_lock);
2430
2431 if (mode == SC_LSEG_SR) {
2432 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2433 sci->sc_seq_done = sci->sc_seq_accepted;
2434 nilfs_segctor_wakeup(sci, err, false);
2435 sci->sc_flush_request = 0;
2436 } else {
2437 if (mode == SC_FLUSH_FILE)
2438 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2439 else if (mode == SC_FLUSH_DAT)
2440 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2441
2442 /* re-enable timer if checkpoint creation was not done */
2443 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) && sci->sc_task &&
2444 time_before(jiffies, sci->sc_timer.expires))
2445 add_timer(&sci->sc_timer);
2446 }
2447 spin_unlock(&sci->sc_state_lock);
2448}
2449
2450/**
2451 * nilfs_segctor_construct - form logs and write them to disk
2452 * @sci: segment constructor object
2453 * @mode: mode of log forming
2454 *
2455 * Return: 0 on success, or a negative error code on failure.
2456 */
2457static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2458{
2459 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2460 struct nilfs_super_block **sbp;
2461 int err = 0;
2462
2463 nilfs_segctor_accept(sci);
2464
2465 if (nilfs_discontinued(nilfs))
2466 mode = SC_LSEG_SR;
2467 if (!nilfs_segctor_confirm(sci))
2468 err = nilfs_segctor_do_construct(sci, mode);
2469
2470 if (likely(!err)) {
2471 if (mode != SC_FLUSH_DAT)
2472 atomic_set(&nilfs->ns_ndirtyblks, 0);
2473 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2474 nilfs_discontinued(nilfs)) {
2475 down_write(&nilfs->ns_sem);
2476 err = -EIO;
2477 sbp = nilfs_prepare_super(sci->sc_super,
2478 nilfs_sb_will_flip(nilfs));
2479 if (likely(sbp)) {
2480 nilfs_set_log_cursor(sbp[0], nilfs);
2481 err = nilfs_commit_super(sci->sc_super,
2482 NILFS_SB_COMMIT);
2483 }
2484 up_write(&nilfs->ns_sem);
2485 }
2486 }
2487
2488 nilfs_segctor_notify(sci, mode, err);
2489 return err;
2490}
2491
2492static void nilfs_construction_timeout(struct timer_list *t)
2493{
2494 struct nilfs_sc_info *sci = timer_container_of(sci, t, sc_timer);
2495
2496 wake_up_process(sci->sc_task);
2497}
2498
2499static void
2500nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2501{
2502 struct nilfs_inode_info *ii, *n;
2503
2504 list_for_each_entry_safe(ii, n, head, i_dirty) {
2505 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2506 continue;
2507 list_del_init(&ii->i_dirty);
2508 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2509 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2510 iput(&ii->vfs_inode);
2511 }
2512}
2513
2514int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2515 void **kbufs)
2516{
2517 struct the_nilfs *nilfs = sb->s_fs_info;
2518 struct nilfs_sc_info *sci = nilfs->ns_writer;
2519 struct nilfs_transaction_info ti;
2520 int err;
2521 size_t i, nfreesegs = argv[4].v_nmembs;
2522 __u64 *segnumv = kbufs[4];
2523
2524 if (unlikely(!sci))
2525 return -EROFS;
2526
2527 nilfs_transaction_lock(sb, &ti, 1);
2528
2529 /*
2530 * Validate segment numbers under ns_segctor_sem (held for write
2531 * by nilfs_transaction_lock above) so the check is serialized
2532 * against nilfs_ioctl_resize(), which can modify ns_nsegments.
2533 * Rejecting bad input here, before any segment-cleaning work
2534 * begins, avoids the per-element diagnostic path inside
2535 * nilfs_sufile_updatev() that would otherwise run under this
2536 * same lock and stall concurrent readers.
2537 */
2538 for (i = 0; i < nfreesegs; i++) {
2539 if (segnumv[i] >= nilfs->ns_nsegments) {
2540 nilfs_err(sb,
2541 "Segment number %llu to be freed is out of range",
2542 (unsigned long long)segnumv[i]);
2543 err = -EINVAL;
2544 goto bail_unlock;
2545 }
2546 }
2547
2548 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2549 if (unlikely(err))
2550 goto out_unlock;
2551
2552 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2553 if (unlikely(err)) {
2554 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2555 goto out_unlock;
2556 }
2557
2558 sci->sc_freesegs = kbufs[4];
2559 sci->sc_nfreesegs = argv[4].v_nmembs;
2560 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2561
2562 for (;;) {
2563 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2564 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2565
2566 if (likely(!err))
2567 break;
2568
2569 nilfs_warn(sb, "error %d cleaning segments", err);
2570
2571 if (unlikely(err == -EROFS))
2572 goto out_unlock;
2573
2574 set_current_state(TASK_INTERRUPTIBLE);
2575 schedule_timeout(sci->sc_interval);
2576 }
2577 if (nilfs_test_opt(nilfs, DISCARD)) {
2578 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2579 sci->sc_nfreesegs);
2580 if (ret) {
2581 nilfs_warn(sb,
2582 "error %d on discard request, turning discards off for the device",
2583 ret);
2584 nilfs_clear_opt(nilfs, DISCARD);
2585 }
2586 }
2587
2588 out_unlock:
2589 sci->sc_freesegs = NULL;
2590 sci->sc_nfreesegs = 0;
2591 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2592 bail_unlock:
2593 nilfs_transaction_unlock(sb);
2594 return err;
2595}
2596
2597static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2598{
2599 struct nilfs_transaction_info ti;
2600
2601 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2602 nilfs_segctor_construct(sci, mode);
2603
2604 /*
2605 * Unclosed segment should be retried. We do this using sc_timer.
2606 * Timeout of sc_timer will invoke complete construction which leads
2607 * to close the current logical segment.
2608 */
2609 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2610 nilfs_segctor_start_timer(sci);
2611
2612 nilfs_transaction_unlock(sci->sc_super);
2613}
2614
2615static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2616{
2617 int mode = 0;
2618
2619 spin_lock(&sci->sc_state_lock);
2620 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2621 SC_FLUSH_DAT : SC_FLUSH_FILE;
2622 spin_unlock(&sci->sc_state_lock);
2623
2624 if (mode) {
2625 nilfs_segctor_do_construct(sci, mode);
2626
2627 spin_lock(&sci->sc_state_lock);
2628 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2629 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2630 spin_unlock(&sci->sc_state_lock);
2631 }
2632 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2633}
2634
2635static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2636{
2637 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2638 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2639 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2640 return SC_FLUSH_FILE;
2641 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2642 return SC_FLUSH_DAT;
2643 }
2644 return SC_LSEG_SR;
2645}
2646
2647/**
2648 * nilfs_log_write_required - determine whether log writing is required
2649 * @sci: nilfs_sc_info struct
2650 * @modep: location for storing log writing mode
2651 *
2652 * Return: true if log writing is required, false otherwise. If log writing
2653 * is required, the mode is stored in the location pointed to by @modep.
2654 */
2655static bool nilfs_log_write_required(struct nilfs_sc_info *sci, int *modep)
2656{
2657 bool timedout, ret = true;
2658
2659 spin_lock(&sci->sc_state_lock);
2660 timedout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2661 time_after_eq(jiffies, sci->sc_timer.expires));
2662 if (timedout || sci->sc_seq_request != sci->sc_seq_done)
2663 *modep = SC_LSEG_SR;
2664 else if (sci->sc_flush_request)
2665 *modep = nilfs_segctor_flush_mode(sci);
2666 else
2667 ret = false;
2668
2669 spin_unlock(&sci->sc_state_lock);
2670 return ret;
2671}
2672
2673/**
2674 * nilfs_segctor_thread - main loop of the log writer thread
2675 * @arg: pointer to a struct nilfs_sc_info.
2676 *
2677 * nilfs_segctor_thread() is the main loop function of the log writer kernel
2678 * thread, which determines whether log writing is necessary, and if so,
2679 * performs the log write in the background, or waits if not. It is also
2680 * used to decide the background writeback of the superblock.
2681 *
2682 * Return: Always 0.
2683 */
2684static int nilfs_segctor_thread(void *arg)
2685{
2686 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2687 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2688
2689 nilfs_info(sci->sc_super,
2690 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2691 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2692
2693 set_freezable();
2694
2695 while (!kthread_should_stop()) {
2696 DEFINE_WAIT(wait);
2697 bool should_write;
2698 int mode;
2699
2700 if (freezing(current)) {
2701 try_to_freeze();
2702 continue;
2703 }
2704
2705 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2706 TASK_INTERRUPTIBLE);
2707 should_write = nilfs_log_write_required(sci, &mode);
2708 if (!should_write)
2709 schedule();
2710 finish_wait(&sci->sc_wait_daemon, &wait);
2711
2712 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2713 set_nilfs_discontinued(nilfs);
2714
2715 if (should_write)
2716 nilfs_segctor_thread_construct(sci, mode);
2717 }
2718
2719 /* end sync. */
2720 spin_lock(&sci->sc_state_lock);
2721 sci->sc_task = NULL;
2722 timer_shutdown_sync(&sci->sc_timer);
2723 spin_unlock(&sci->sc_state_lock);
2724 return 0;
2725}
2726
2727/*
2728 * Setup & clean-up functions
2729 */
2730static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2731 struct nilfs_root *root)
2732{
2733 struct the_nilfs *nilfs = sb->s_fs_info;
2734 struct nilfs_sc_info *sci;
2735
2736 sci = kzalloc_obj(*sci);
2737 if (!sci)
2738 return NULL;
2739
2740 sci->sc_super = sb;
2741
2742 nilfs_get_root(root);
2743 sci->sc_root = root;
2744
2745 init_waitqueue_head(&sci->sc_wait_request);
2746 init_waitqueue_head(&sci->sc_wait_daemon);
2747 spin_lock_init(&sci->sc_state_lock);
2748 INIT_LIST_HEAD(&sci->sc_dirty_files);
2749 INIT_LIST_HEAD(&sci->sc_segbufs);
2750 INIT_LIST_HEAD(&sci->sc_write_logs);
2751 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2752 INIT_LIST_HEAD(&sci->sc_iput_queue);
2753 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2754
2755 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2756 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2757 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2758
2759 if (nilfs->ns_interval)
2760 sci->sc_interval = HZ * nilfs->ns_interval;
2761 if (nilfs->ns_watermark)
2762 sci->sc_watermark = nilfs->ns_watermark;
2763 return sci;
2764}
2765
2766static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2767{
2768 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2769
2770 /*
2771 * The segctord thread was stopped and its timer was removed.
2772 * But some tasks remain.
2773 */
2774 do {
2775 struct nilfs_transaction_info ti;
2776
2777 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2778 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2779 nilfs_transaction_unlock(sci->sc_super);
2780
2781 flush_work(&sci->sc_iput_work);
2782
2783 } while (ret && ret != -EROFS && retrycount-- > 0);
2784}
2785
2786/**
2787 * nilfs_segctor_destroy - destroy the segment constructor.
2788 * @sci: nilfs_sc_info
2789 *
2790 * nilfs_segctor_destroy() kills the segctord thread and frees
2791 * the nilfs_sc_info struct.
2792 * Caller must hold the segment semaphore.
2793 */
2794static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2795{
2796 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2797 int flag;
2798
2799 up_write(&nilfs->ns_segctor_sem);
2800
2801 if (sci->sc_task) {
2802 wake_up(&sci->sc_wait_daemon);
2803 if (kthread_stop(sci->sc_task)) {
2804 spin_lock(&sci->sc_state_lock);
2805 sci->sc_task = NULL;
2806 timer_shutdown_sync(&sci->sc_timer);
2807 spin_unlock(&sci->sc_state_lock);
2808 }
2809 }
2810
2811 spin_lock(&sci->sc_state_lock);
2812 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2813 || sci->sc_seq_request != sci->sc_seq_done);
2814 spin_unlock(&sci->sc_state_lock);
2815
2816 /*
2817 * Forcibly wake up tasks waiting in nilfs_segctor_sync(), which can
2818 * be called from delayed iput() via nilfs_evict_inode() and can race
2819 * with the above log writer thread termination.
2820 */
2821 nilfs_segctor_wakeup(sci, 0, true);
2822
2823 if (flush_work(&sci->sc_iput_work))
2824 flag = true;
2825
2826 if (flag || !nilfs_segctor_confirm(sci))
2827 nilfs_segctor_write_out(sci);
2828
2829 if (!list_empty(&sci->sc_dirty_files)) {
2830 nilfs_warn(sci->sc_super,
2831 "disposed unprocessed dirty file(s) when stopping log writer");
2832 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2833 }
2834
2835 if (!list_empty(&sci->sc_iput_queue)) {
2836 nilfs_warn(sci->sc_super,
2837 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2838 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2839 }
2840
2841 WARN_ON(!list_empty(&sci->sc_segbufs));
2842 WARN_ON(!list_empty(&sci->sc_write_logs));
2843
2844 nilfs_put_root(sci->sc_root);
2845
2846 down_write(&nilfs->ns_segctor_sem);
2847
2848 kfree(sci);
2849}
2850
2851/**
2852 * nilfs_attach_log_writer - attach log writer
2853 * @sb: super block instance
2854 * @root: root object of the current filesystem tree
2855 *
2856 * This allocates a log writer object, initializes it, and starts the
2857 * log writer.
2858 *
2859 * Return: 0 on success, or one of the following negative error codes on
2860 * failure:
2861 * * %-EINTR - Log writer thread creation failed due to interruption.
2862 * * %-ENOMEM - Insufficient memory available.
2863 */
2864int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2865{
2866 struct the_nilfs *nilfs = sb->s_fs_info;
2867 struct nilfs_sc_info *sci;
2868 struct task_struct *t;
2869 int err;
2870
2871 if (nilfs->ns_writer) {
2872 /*
2873 * This happens if the filesystem is made read-only by
2874 * __nilfs_error or nilfs_remount and then remounted
2875 * read/write. In these cases, reuse the existing
2876 * writer.
2877 */
2878 return 0;
2879 }
2880
2881 sci = nilfs_segctor_new(sb, root);
2882 if (unlikely(!sci))
2883 return -ENOMEM;
2884
2885 nilfs->ns_writer = sci;
2886 t = kthread_create(nilfs_segctor_thread, sci, "segctord");
2887 if (IS_ERR(t)) {
2888 err = PTR_ERR(t);
2889 nilfs_err(sb, "error %d creating segctord thread", err);
2890 nilfs_detach_log_writer(sb);
2891 return err;
2892 }
2893 sci->sc_task = t;
2894 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2895
2896 wake_up_process(sci->sc_task);
2897 return 0;
2898}
2899
2900/**
2901 * nilfs_detach_log_writer - destroy log writer
2902 * @sb: super block instance
2903 *
2904 * This kills log writer daemon, frees the log writer object, and
2905 * destroys list of dirty files.
2906 */
2907void nilfs_detach_log_writer(struct super_block *sb)
2908{
2909 struct the_nilfs *nilfs = sb->s_fs_info;
2910 LIST_HEAD(garbage_list);
2911
2912 down_write(&nilfs->ns_segctor_sem);
2913 if (nilfs->ns_writer) {
2914 nilfs_segctor_destroy(nilfs->ns_writer);
2915 nilfs->ns_writer = NULL;
2916 }
2917 set_nilfs_purging(nilfs);
2918
2919 /* Force to free the list of dirty files */
2920 spin_lock(&nilfs->ns_inode_lock);
2921 if (!list_empty(&nilfs->ns_dirty_files)) {
2922 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2923 nilfs_warn(sb,
2924 "disposed unprocessed dirty file(s) when detaching log writer");
2925 }
2926 spin_unlock(&nilfs->ns_inode_lock);
2927 up_write(&nilfs->ns_segctor_sem);
2928
2929 nilfs_dispose_list(nilfs, &garbage_list, 1);
2930 clear_nilfs_purging(nilfs);
2931}