1 // SPDX-License-Identifier: GPL-2.0+
3 * NILFS segment constructor.
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
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>
39 #define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
41 #define SC_MAX_SEGDELTA 64 /*
42 * Upper limit of the number of segments
43 * appended in collection retry loop
46 /* Construction mode */
48 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
50 * Flush data blocks of a given file and make
51 * a logical segment without a super root.
54 * Flush data files, leads to segment writes without
55 * creating a checkpoint.
58 * Flush DAT file. This also creates segments
59 * without a checkpoint.
63 /* Stage numbers of dirty block collection */
66 NILFS_ST_GC, /* Collecting dirty blocks for GC */
72 NILFS_ST_SR, /* Super root */
73 NILFS_ST_DSYNC, /* Data sync blocks */
77 #define CREATE_TRACE_POINTS
78 #include <trace/events/nilfs2.h>
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).
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
90 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
93 trace_nilfs2_collection_stage_transition(sci);
96 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
98 sci->sc_stage.scnt = next_scnt;
99 trace_nilfs2_collection_stage_transition(sci);
102 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
104 return sci->sc_stage.scnt;
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)
113 /* Operations depending on the construction mode and file type */
114 struct nilfs_sc_operations {
115 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
117 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
119 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
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 *);
132 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
133 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
134 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
135 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
137 #define nilfs_cnt32_ge(a, b) \
138 (typecheck(__u32, a) && typecheck(__u32, b) && \
139 ((__s32)(a) - (__s32)(b) >= 0))
141 static int nilfs_prepare_segment_lock(struct super_block *sb,
142 struct nilfs_transaction_info *ti)
144 struct nilfs_transaction_info *cur_ti = current->journal_info;
148 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
149 return ++cur_ti->ti_count;
152 * If journal_info field is occupied by other FS,
153 * it is saved and will be restored on
154 * nilfs_transaction_commit().
156 nilfs_warn(sb, "journal info from a different FS");
157 save = current->journal_info;
160 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
163 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
169 ti->ti_magic = NILFS_TI_MAGIC;
170 current->journal_info = ti;
175 * nilfs_transaction_begin - start indivisible file operations.
177 * @ti: nilfs_transaction_info
178 * @vacancy_check: flags for vacancy rate checks
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.
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.
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.
194 * Return Value: On success, 0 is returned. On error, one of the following
195 * negative error code is returned.
197 * %-ENOMEM - Insufficient memory available.
199 * %-ENOSPC - No space left on device
201 int nilfs_transaction_begin(struct super_block *sb,
202 struct nilfs_transaction_info *ti,
205 struct the_nilfs *nilfs;
206 int ret = nilfs_prepare_segment_lock(sb, ti);
207 struct nilfs_transaction_info *trace_ti;
209 if (unlikely(ret < 0))
212 trace_ti = current->journal_info;
214 trace_nilfs2_transaction_transition(sb, trace_ti,
215 trace_ti->ti_count, trace_ti->ti_flags,
216 TRACE_NILFS2_TRANSACTION_BEGIN);
220 sb_start_intwrite(sb);
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);
230 trace_ti = current->journal_info;
231 trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
233 TRACE_NILFS2_TRANSACTION_BEGIN);
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);
246 * nilfs_transaction_commit - commit indivisible file operations.
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
256 int nilfs_transaction_commit(struct super_block *sb)
258 struct nilfs_transaction_info *ti = current->journal_info;
259 struct the_nilfs *nilfs = sb->s_fs_info;
262 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
263 ti->ti_flags |= NILFS_TI_COMMIT;
264 if (ti->ti_count > 0) {
266 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
267 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
270 if (nilfs->ns_writer) {
271 struct nilfs_sc_info *sci = nilfs->ns_writer;
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);
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);
282 current->journal_info = ti->ti_save;
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);
292 void nilfs_transaction_abort(struct super_block *sb)
294 struct nilfs_transaction_info *ti = current->journal_info;
295 struct the_nilfs *nilfs = sb->s_fs_info;
297 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
298 if (ti->ti_count > 0) {
300 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
301 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
304 up_read(&nilfs->ns_segctor_sem);
306 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
307 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
309 current->journal_info = ti->ti_save;
310 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
311 kmem_cache_free(nilfs_transaction_cachep, ti);
315 void nilfs_relax_pressure_in_lock(struct super_block *sb)
317 struct the_nilfs *nilfs = sb->s_fs_info;
318 struct nilfs_sc_info *sci = nilfs->ns_writer;
320 if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request)
323 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
324 up_read(&nilfs->ns_segctor_sem);
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;
331 ti->ti_flags |= NILFS_TI_WRITER;
332 nilfs_segctor_do_immediate_flush(sci);
333 ti->ti_flags &= ~NILFS_TI_WRITER;
335 downgrade_write(&nilfs->ns_segctor_sem);
338 static void nilfs_transaction_lock(struct super_block *sb,
339 struct nilfs_transaction_info *ti,
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;
347 ti->ti_flags = NILFS_TI_WRITER;
349 ti->ti_save = cur_ti;
350 ti->ti_magic = NILFS_TI_MAGIC;
351 current->journal_info = ti;
354 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
355 ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
357 down_write(&nilfs->ns_segctor_sem);
358 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
361 nilfs_segctor_do_immediate_flush(sci);
363 up_write(&nilfs->ns_segctor_sem);
367 ti->ti_flags |= NILFS_TI_GC;
369 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
370 ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
373 static void nilfs_transaction_unlock(struct super_block *sb)
375 struct nilfs_transaction_info *ti = current->journal_info;
376 struct the_nilfs *nilfs = sb->s_fs_info;
378 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
379 BUG_ON(ti->ti_count > 0);
381 up_write(&nilfs->ns_segctor_sem);
382 current->journal_info = ti->ti_save;
384 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
385 ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
388 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
389 struct nilfs_segsum_pointer *ssp,
392 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
393 unsigned int blocksize = sci->sc_super->s_blocksize;
396 if (unlikely(ssp->offset + bytes > blocksize)) {
398 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
399 &segbuf->sb_segsum_buffers));
400 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
402 p = ssp->bh->b_data + ssp->offset;
403 ssp->offset += bytes;
408 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
409 * @sci: nilfs_sc_info
411 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
413 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
414 struct buffer_head *sumbh;
415 unsigned int sumbytes;
416 unsigned int flags = 0;
419 if (nilfs_doing_gc())
421 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
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;
434 * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
435 * @sci: segment constructor object
437 * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
438 * the current segment summary block.
440 static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
442 struct nilfs_segsum_pointer *ssp;
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);
450 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
452 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
453 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
455 * The current segment is filled up
458 nilfs_segctor_zeropad_segsum(sci);
459 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
460 return nilfs_segctor_reset_segment_buffer(sci);
463 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
465 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
468 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
469 err = nilfs_segctor_feed_segment(sci);
472 segbuf = sci->sc_curseg;
474 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
476 segbuf->sb_sum.flags |= NILFS_SS_SR;
481 * Functions for making segment summary and payloads
483 static int nilfs_segctor_segsum_block_required(
484 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
485 unsigned int binfo_size)
487 unsigned int blocksize = sci->sc_super->s_blocksize;
488 /* Size of finfo and binfo is enough small against blocksize */
490 return ssp->offset + binfo_size +
491 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
495 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
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));
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);
509 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
512 struct nilfs_finfo *finfo;
513 struct nilfs_inode_info *ii;
514 struct nilfs_segment_buffer *segbuf;
517 if (sci->sc_blk_cnt == 0)
522 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
524 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
529 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
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);
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;
543 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
544 struct buffer_head *bh,
546 unsigned int binfo_size)
548 struct nilfs_segment_buffer *segbuf;
549 int required, err = 0;
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);
562 if (unlikely(required)) {
563 nilfs_segctor_zeropad_segsum(sci);
564 err = nilfs_segbuf_extend_segsum(segbuf);
568 if (sci->sc_blk_cnt == 0)
569 nilfs_segctor_begin_finfo(sci, inode);
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);
580 * Callback functions that enumerate, mark, and collect dirty blocks
582 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
583 struct buffer_head *bh, struct inode *inode)
587 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
591 err = nilfs_segctor_add_file_block(sci, bh, inode,
592 sizeof(struct nilfs_binfo_v));
594 sci->sc_datablk_cnt++;
598 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
599 struct buffer_head *bh,
602 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
605 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
606 struct buffer_head *bh,
609 WARN_ON(!buffer_dirty(bh));
610 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
613 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
614 struct nilfs_segsum_pointer *ssp,
615 union nilfs_binfo *binfo)
617 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
618 sci, ssp, sizeof(*binfo_v));
619 *binfo_v = binfo->bi_v;
622 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
623 struct nilfs_segsum_pointer *ssp,
624 union nilfs_binfo *binfo)
626 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
627 sci, ssp, sizeof(*vblocknr));
628 *vblocknr = binfo->bi_v.bi_vblocknr;
631 static 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,
639 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
640 struct buffer_head *bh, struct inode *inode)
644 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
648 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
650 sci->sc_datablk_cnt++;
654 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
655 struct buffer_head *bh, struct inode *inode)
657 WARN_ON(!buffer_dirty(bh));
658 return nilfs_segctor_add_file_block(sci, bh, inode,
659 sizeof(struct nilfs_binfo_dat));
662 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
663 struct nilfs_segsum_pointer *ssp,
664 union nilfs_binfo *binfo)
666 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
668 *blkoff = binfo->bi_dat.bi_blkoff;
671 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
672 struct nilfs_segsum_pointer *ssp,
673 union nilfs_binfo *binfo)
675 struct nilfs_binfo_dat *binfo_dat =
676 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
677 *binfo_dat = binfo->bi_dat;
680 static 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,
688 static 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,
696 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
697 struct list_head *listp,
699 loff_t start, loff_t end)
701 struct address_space *mapping = inode->i_mapping;
702 struct folio_batch fbatch;
703 pgoff_t index = 0, last = ULONG_MAX;
707 if (unlikely(start != 0 || end != LLONG_MAX)) {
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.
713 index = start >> PAGE_SHIFT;
714 last = end >> PAGE_SHIFT;
716 folio_batch_init(&fbatch);
718 if (unlikely(index > last) ||
719 !filemap_get_folios_tag(mapping, &index, last,
720 PAGECACHE_TAG_DIRTY, &fbatch))
723 for (i = 0; i < folio_batch_count(&fbatch); i++) {
724 struct buffer_head *bh, *head;
725 struct folio *folio = fbatch.folios[i];
728 head = folio_buffers(folio);
730 create_empty_buffers(&folio->page, i_blocksize(inode), 0);
731 head = folio_buffers(folio);
737 if (!buffer_dirty(bh) || buffer_async_write(bh))
740 list_add_tail(&bh->b_assoc_buffers, listp);
742 if (unlikely(ndirties >= nlimit)) {
743 folio_batch_release(&fbatch);
747 } while (bh = bh->b_this_page, bh != head);
749 folio_batch_release(&fbatch);
754 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
755 struct list_head *listp)
757 struct nilfs_inode_info *ii = NILFS_I(inode);
758 struct inode *btnc_inode = ii->i_assoc_inode;
759 struct folio_batch fbatch;
760 struct buffer_head *bh, *head;
766 folio_batch_init(&fbatch);
768 while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
769 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
770 for (i = 0; i < folio_batch_count(&fbatch); i++) {
771 bh = head = folio_buffers(fbatch.folios[i]);
773 if (buffer_dirty(bh) &&
774 !buffer_async_write(bh)) {
776 list_add_tail(&bh->b_assoc_buffers,
779 bh = bh->b_this_page;
780 } while (bh != head);
782 folio_batch_release(&fbatch);
787 static void nilfs_dispose_list(struct the_nilfs *nilfs,
788 struct list_head *head, int force)
790 struct nilfs_inode_info *ii, *n;
791 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
794 while (!list_empty(head)) {
795 spin_lock(&nilfs->ns_inode_lock);
796 list_for_each_entry_safe(ii, n, head, i_dirty) {
797 list_del_init(&ii->i_dirty);
799 if (unlikely(ii->i_bh)) {
803 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
804 set_bit(NILFS_I_QUEUED, &ii->i_state);
805 list_add_tail(&ii->i_dirty,
806 &nilfs->ns_dirty_files);
810 if (nv == SC_N_INODEVEC)
813 spin_unlock(&nilfs->ns_inode_lock);
815 for (pii = ivec; nv > 0; pii++, nv--)
816 iput(&(*pii)->vfs_inode);
820 static void nilfs_iput_work_func(struct work_struct *work)
822 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
824 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
826 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
829 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
830 struct nilfs_root *root)
834 if (nilfs_mdt_fetch_dirty(root->ifile))
836 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
838 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
840 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
845 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
847 return list_empty(&sci->sc_dirty_files) &&
848 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
849 sci->sc_nfreesegs == 0 &&
850 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
853 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
855 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
858 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
859 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
861 spin_lock(&nilfs->ns_inode_lock);
862 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
865 spin_unlock(&nilfs->ns_inode_lock);
869 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
871 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
873 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
874 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
875 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
876 nilfs_mdt_clear_dirty(nilfs->ns_dat);
879 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
881 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
882 struct buffer_head *bh_cp;
883 struct nilfs_checkpoint *raw_cp;
886 /* XXX: this interface will be changed */
887 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
891 * The following code is duplicated with cpfile. But, it is
892 * needed to collect the checkpoint even if it was not newly
895 mark_buffer_dirty(bh_cp);
896 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
897 nilfs_cpfile_put_checkpoint(
898 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
899 } else if (err == -EINVAL || err == -ENOENT) {
900 nilfs_error(sci->sc_super,
901 "checkpoint creation failed due to metadata corruption.");
907 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
909 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
910 struct buffer_head *bh_cp;
911 struct nilfs_checkpoint *raw_cp;
914 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
917 if (err == -EINVAL || err == -ENOENT) {
918 nilfs_error(sci->sc_super,
919 "checkpoint finalization failed due to metadata corruption.");
924 raw_cp->cp_snapshot_list.ssl_next = 0;
925 raw_cp->cp_snapshot_list.ssl_prev = 0;
926 raw_cp->cp_inodes_count =
927 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
928 raw_cp->cp_blocks_count =
929 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
930 raw_cp->cp_nblk_inc =
931 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
932 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
933 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
935 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
936 nilfs_checkpoint_clear_minor(raw_cp);
938 nilfs_checkpoint_set_minor(raw_cp);
940 nilfs_write_inode_common(sci->sc_root->ifile,
941 &raw_cp->cp_ifile_inode, 1);
942 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
949 static void nilfs_fill_in_file_bmap(struct inode *ifile,
950 struct nilfs_inode_info *ii)
953 struct buffer_head *ibh;
954 struct nilfs_inode *raw_inode;
956 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
959 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
961 nilfs_bmap_write(ii->i_bmap, raw_inode);
962 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
966 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
968 struct nilfs_inode_info *ii;
970 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
971 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
972 set_bit(NILFS_I_COLLECTED, &ii->i_state);
976 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
977 struct the_nilfs *nilfs)
979 struct buffer_head *bh_sr;
980 struct nilfs_super_root *raw_sr;
981 unsigned int isz, srsz;
983 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
986 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
987 isz = nilfs->ns_inode_size;
988 srsz = NILFS_SR_BYTES(isz);
990 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
991 raw_sr->sr_bytes = cpu_to_le16(srsz);
992 raw_sr->sr_nongc_ctime
993 = cpu_to_le64(nilfs_doing_gc() ?
994 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
995 raw_sr->sr_flags = 0;
997 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
998 NILFS_SR_DAT_OFFSET(isz), 1);
999 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
1000 NILFS_SR_CPFILE_OFFSET(isz), 1);
1001 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
1002 NILFS_SR_SUFILE_OFFSET(isz), 1);
1003 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1004 set_buffer_uptodate(bh_sr);
1005 unlock_buffer(bh_sr);
1008 static void nilfs_redirty_inodes(struct list_head *head)
1010 struct nilfs_inode_info *ii;
1012 list_for_each_entry(ii, head, i_dirty) {
1013 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1014 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1018 static void nilfs_drop_collected_inodes(struct list_head *head)
1020 struct nilfs_inode_info *ii;
1022 list_for_each_entry(ii, head, i_dirty) {
1023 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1026 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1027 set_bit(NILFS_I_UPDATED, &ii->i_state);
1031 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1032 struct inode *inode,
1033 struct list_head *listp,
1034 int (*collect)(struct nilfs_sc_info *,
1035 struct buffer_head *,
1038 struct buffer_head *bh, *n;
1042 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1043 list_del_init(&bh->b_assoc_buffers);
1044 err = collect(sci, bh, inode);
1047 goto dispose_buffers;
1053 while (!list_empty(listp)) {
1054 bh = list_first_entry(listp, struct buffer_head,
1056 list_del_init(&bh->b_assoc_buffers);
1062 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1064 /* Remaining number of blocks within segment buffer */
1065 return sci->sc_segbuf_nblocks -
1066 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1069 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1070 struct inode *inode,
1071 const struct nilfs_sc_operations *sc_ops)
1073 LIST_HEAD(data_buffers);
1074 LIST_HEAD(node_buffers);
1077 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1078 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1080 n = nilfs_lookup_dirty_data_buffers(
1081 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1083 err = nilfs_segctor_apply_buffers(
1084 sci, inode, &data_buffers,
1085 sc_ops->collect_data);
1086 BUG_ON(!err); /* always receive -E2BIG or true error */
1090 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1092 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1093 err = nilfs_segctor_apply_buffers(
1094 sci, inode, &data_buffers, sc_ops->collect_data);
1095 if (unlikely(err)) {
1096 /* dispose node list */
1097 nilfs_segctor_apply_buffers(
1098 sci, inode, &node_buffers, NULL);
1101 sci->sc_stage.flags |= NILFS_CF_NODE;
1104 err = nilfs_segctor_apply_buffers(
1105 sci, inode, &node_buffers, sc_ops->collect_node);
1109 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1110 err = nilfs_segctor_apply_buffers(
1111 sci, inode, &node_buffers, sc_ops->collect_bmap);
1115 nilfs_segctor_end_finfo(sci, inode);
1116 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1122 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1123 struct inode *inode)
1125 LIST_HEAD(data_buffers);
1126 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1129 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1130 sci->sc_dsync_start,
1133 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1134 nilfs_collect_file_data);
1136 nilfs_segctor_end_finfo(sci, inode);
1138 /* always receive -E2BIG or true error if n > rest */
1143 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1145 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1146 struct list_head *head;
1147 struct nilfs_inode_info *ii;
1151 switch (nilfs_sc_cstage_get(sci)) {
1154 sci->sc_stage.flags = 0;
1156 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1157 sci->sc_nblk_inc = 0;
1158 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1159 if (mode == SC_LSEG_DSYNC) {
1160 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1165 sci->sc_stage.dirty_file_ptr = NULL;
1166 sci->sc_stage.gc_inode_ptr = NULL;
1167 if (mode == SC_FLUSH_DAT) {
1168 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1171 nilfs_sc_cstage_inc(sci);
1174 if (nilfs_doing_gc()) {
1175 head = &sci->sc_gc_inodes;
1176 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1178 list_for_each_entry_continue(ii, head, i_dirty) {
1179 err = nilfs_segctor_scan_file(
1180 sci, &ii->vfs_inode,
1181 &nilfs_sc_file_ops);
1182 if (unlikely(err)) {
1183 sci->sc_stage.gc_inode_ptr = list_entry(
1185 struct nilfs_inode_info,
1189 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1191 sci->sc_stage.gc_inode_ptr = NULL;
1193 nilfs_sc_cstage_inc(sci);
1196 head = &sci->sc_dirty_files;
1197 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1199 list_for_each_entry_continue(ii, head, i_dirty) {
1200 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1202 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1203 &nilfs_sc_file_ops);
1204 if (unlikely(err)) {
1205 sci->sc_stage.dirty_file_ptr =
1206 list_entry(ii->i_dirty.prev,
1207 struct nilfs_inode_info,
1211 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1212 /* XXX: required ? */
1214 sci->sc_stage.dirty_file_ptr = NULL;
1215 if (mode == SC_FLUSH_FILE) {
1216 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1219 nilfs_sc_cstage_inc(sci);
1220 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1222 case NILFS_ST_IFILE:
1223 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1224 &nilfs_sc_file_ops);
1227 nilfs_sc_cstage_inc(sci);
1228 /* Creating a checkpoint */
1229 err = nilfs_segctor_create_checkpoint(sci);
1233 case NILFS_ST_CPFILE:
1234 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1235 &nilfs_sc_file_ops);
1238 nilfs_sc_cstage_inc(sci);
1240 case NILFS_ST_SUFILE:
1241 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1242 sci->sc_nfreesegs, &ndone);
1243 if (unlikely(err)) {
1244 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1245 sci->sc_freesegs, ndone,
1249 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1251 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1252 &nilfs_sc_file_ops);
1255 nilfs_sc_cstage_inc(sci);
1259 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1263 if (mode == SC_FLUSH_DAT) {
1264 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1267 nilfs_sc_cstage_inc(sci);
1270 if (mode == SC_LSEG_SR) {
1271 /* Appending a super root */
1272 err = nilfs_segctor_add_super_root(sci);
1276 /* End of a logical segment */
1277 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1278 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1280 case NILFS_ST_DSYNC:
1282 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1283 ii = sci->sc_dsync_inode;
1284 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1287 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1290 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1291 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1304 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1305 * @sci: nilfs_sc_info
1306 * @nilfs: nilfs object
1308 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1309 struct the_nilfs *nilfs)
1311 struct nilfs_segment_buffer *segbuf, *prev;
1315 segbuf = nilfs_segbuf_new(sci->sc_super);
1316 if (unlikely(!segbuf))
1319 if (list_empty(&sci->sc_write_logs)) {
1320 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1321 nilfs->ns_pseg_offset, nilfs);
1322 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1323 nilfs_shift_to_next_segment(nilfs);
1324 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1327 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1328 nextnum = nilfs->ns_nextnum;
1330 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1331 /* Start from the head of a new full segment */
1335 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1336 nilfs_segbuf_map_cont(segbuf, prev);
1337 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1338 nextnum = prev->sb_nextnum;
1340 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1341 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1342 segbuf->sb_sum.seg_seq++;
1347 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1352 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1356 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1358 BUG_ON(!list_empty(&sci->sc_segbufs));
1359 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1360 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1364 nilfs_segbuf_free(segbuf);
1368 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1369 struct the_nilfs *nilfs, int nadd)
1371 struct nilfs_segment_buffer *segbuf, *prev;
1372 struct inode *sufile = nilfs->ns_sufile;
1377 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1379 * Since the segment specified with nextnum might be allocated during
1380 * the previous construction, the buffer including its segusage may
1381 * not be dirty. The following call ensures that the buffer is dirty
1382 * and will pin the buffer on memory until the sufile is written.
1384 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1388 for (i = 0; i < nadd; i++) {
1389 /* extend segment info */
1391 segbuf = nilfs_segbuf_new(sci->sc_super);
1392 if (unlikely(!segbuf))
1395 /* map this buffer to region of segment on-disk */
1396 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1397 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1399 /* allocate the next next full segment */
1400 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1404 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1405 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1407 list_add_tail(&segbuf->sb_list, &list);
1410 list_splice_tail(&list, &sci->sc_segbufs);
1414 nilfs_segbuf_free(segbuf);
1416 list_for_each_entry(segbuf, &list, sb_list) {
1417 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1418 WARN_ON(ret); /* never fails */
1420 nilfs_destroy_logs(&list);
1424 static void nilfs_free_incomplete_logs(struct list_head *logs,
1425 struct the_nilfs *nilfs)
1427 struct nilfs_segment_buffer *segbuf, *prev;
1428 struct inode *sufile = nilfs->ns_sufile;
1431 segbuf = NILFS_FIRST_SEGBUF(logs);
1432 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1433 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1434 WARN_ON(ret); /* never fails */
1436 if (atomic_read(&segbuf->sb_err)) {
1437 /* Case 1: The first segment failed */
1438 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1440 * Case 1a: Partial segment appended into an existing
1443 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1444 segbuf->sb_fseg_end);
1445 else /* Case 1b: New full segment */
1446 set_nilfs_discontinued(nilfs);
1450 list_for_each_entry_continue(segbuf, logs, sb_list) {
1451 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1452 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1453 WARN_ON(ret); /* never fails */
1455 if (atomic_read(&segbuf->sb_err) &&
1456 segbuf->sb_segnum != nilfs->ns_nextnum)
1457 /* Case 2: extended segment (!= next) failed */
1458 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1463 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1464 struct inode *sufile)
1466 struct nilfs_segment_buffer *segbuf;
1467 unsigned long live_blocks;
1470 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1471 live_blocks = segbuf->sb_sum.nblocks +
1472 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1473 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1476 WARN_ON(ret); /* always succeed because the segusage is dirty */
1480 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1482 struct nilfs_segment_buffer *segbuf;
1485 segbuf = NILFS_FIRST_SEGBUF(logs);
1486 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1487 segbuf->sb_pseg_start -
1488 segbuf->sb_fseg_start, 0);
1489 WARN_ON(ret); /* always succeed because the segusage is dirty */
1491 list_for_each_entry_continue(segbuf, logs, sb_list) {
1492 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1494 WARN_ON(ret); /* always succeed */
1498 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1499 struct nilfs_segment_buffer *last,
1500 struct inode *sufile)
1502 struct nilfs_segment_buffer *segbuf = last;
1505 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1506 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1507 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1510 nilfs_truncate_logs(&sci->sc_segbufs, last);
1514 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1515 struct the_nilfs *nilfs, int mode)
1517 struct nilfs_cstage prev_stage = sci->sc_stage;
1520 /* Collection retry loop */
1522 sci->sc_nblk_this_inc = 0;
1523 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1525 err = nilfs_segctor_reset_segment_buffer(sci);
1529 err = nilfs_segctor_collect_blocks(sci, mode);
1530 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1534 if (unlikely(err != -E2BIG))
1537 /* The current segment is filled up */
1538 if (mode != SC_LSEG_SR ||
1539 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1542 nilfs_clear_logs(&sci->sc_segbufs);
1544 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1545 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1549 WARN_ON(err); /* do not happen */
1550 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1553 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1557 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1558 sci->sc_stage = prev_stage;
1560 nilfs_segctor_zeropad_segsum(sci);
1561 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1568 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1569 struct buffer_head *new_bh)
1571 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1573 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1574 /* The caller must release old_bh */
1578 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1579 struct nilfs_segment_buffer *segbuf,
1582 struct inode *inode = NULL;
1584 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1585 unsigned long nblocks = 0, ndatablk = 0;
1586 const struct nilfs_sc_operations *sc_op = NULL;
1587 struct nilfs_segsum_pointer ssp;
1588 struct nilfs_finfo *finfo = NULL;
1589 union nilfs_binfo binfo;
1590 struct buffer_head *bh, *bh_org;
1597 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1598 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1599 ssp.offset = sizeof(struct nilfs_segment_summary);
1601 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1602 if (bh == segbuf->sb_super_root)
1605 finfo = nilfs_segctor_map_segsum_entry(
1606 sci, &ssp, sizeof(*finfo));
1607 ino = le64_to_cpu(finfo->fi_ino);
1608 nblocks = le32_to_cpu(finfo->fi_nblocks);
1609 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1611 inode = bh->b_folio->mapping->host;
1613 if (mode == SC_LSEG_DSYNC)
1614 sc_op = &nilfs_sc_dsync_ops;
1615 else if (ino == NILFS_DAT_INO)
1616 sc_op = &nilfs_sc_dat_ops;
1617 else /* file blocks */
1618 sc_op = &nilfs_sc_file_ops;
1622 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1625 nilfs_list_replace_buffer(bh_org, bh);
1631 sc_op->write_data_binfo(sci, &ssp, &binfo);
1633 sc_op->write_node_binfo(sci, &ssp, &binfo);
1636 if (--nblocks == 0) {
1640 } else if (ndatablk > 0)
1650 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1652 struct nilfs_segment_buffer *segbuf;
1655 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1656 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1659 nilfs_segbuf_fill_in_segsum(segbuf);
1664 static void nilfs_begin_page_io(struct page *page)
1666 if (!page || PageWriteback(page))
1668 * For split b-tree node pages, this function may be called
1669 * twice. We ignore the 2nd or later calls by this check.
1674 clear_page_dirty_for_io(page);
1675 set_page_writeback(page);
1679 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1681 struct nilfs_segment_buffer *segbuf;
1682 struct page *bd_page = NULL, *fs_page = NULL;
1684 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1685 struct buffer_head *bh;
1687 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1689 if (bh->b_page != bd_page) {
1692 clear_page_dirty_for_io(bd_page);
1693 set_page_writeback(bd_page);
1694 unlock_page(bd_page);
1696 bd_page = bh->b_page;
1700 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1702 set_buffer_async_write(bh);
1703 if (bh == segbuf->sb_super_root) {
1704 if (bh->b_page != bd_page) {
1706 clear_page_dirty_for_io(bd_page);
1707 set_page_writeback(bd_page);
1708 unlock_page(bd_page);
1709 bd_page = bh->b_page;
1713 if (bh->b_page != fs_page) {
1714 nilfs_begin_page_io(fs_page);
1715 fs_page = bh->b_page;
1721 clear_page_dirty_for_io(bd_page);
1722 set_page_writeback(bd_page);
1723 unlock_page(bd_page);
1725 nilfs_begin_page_io(fs_page);
1728 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1729 struct the_nilfs *nilfs)
1733 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1734 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1738 static void nilfs_end_page_io(struct page *page, int err)
1743 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1745 * For b-tree node pages, this function may be called twice
1746 * or more because they might be split in a segment.
1748 if (PageDirty(page)) {
1750 * For pages holding split b-tree node buffers, dirty
1751 * flag on the buffers may be cleared discretely.
1752 * In that case, the page is once redirtied for
1753 * remaining buffers, and it must be cancelled if
1754 * all the buffers get cleaned later.
1757 if (nilfs_page_buffers_clean(page))
1758 __nilfs_clear_page_dirty(page);
1765 if (!nilfs_page_buffers_clean(page))
1766 __set_page_dirty_nobuffers(page);
1767 ClearPageError(page);
1769 __set_page_dirty_nobuffers(page);
1773 end_page_writeback(page);
1776 static void nilfs_abort_logs(struct list_head *logs, int err)
1778 struct nilfs_segment_buffer *segbuf;
1779 struct page *bd_page = NULL, *fs_page = NULL;
1780 struct buffer_head *bh;
1782 if (list_empty(logs))
1785 list_for_each_entry(segbuf, logs, sb_list) {
1786 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1788 clear_buffer_uptodate(bh);
1789 if (bh->b_page != bd_page) {
1791 end_page_writeback(bd_page);
1792 bd_page = bh->b_page;
1796 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1798 clear_buffer_async_write(bh);
1799 if (bh == segbuf->sb_super_root) {
1800 clear_buffer_uptodate(bh);
1801 if (bh->b_page != bd_page) {
1802 end_page_writeback(bd_page);
1803 bd_page = bh->b_page;
1807 if (bh->b_page != fs_page) {
1808 nilfs_end_page_io(fs_page, err);
1809 fs_page = bh->b_page;
1814 end_page_writeback(bd_page);
1816 nilfs_end_page_io(fs_page, err);
1819 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1820 struct the_nilfs *nilfs, int err)
1825 list_splice_tail_init(&sci->sc_write_logs, &logs);
1826 ret = nilfs_wait_on_logs(&logs);
1827 nilfs_abort_logs(&logs, ret ? : err);
1829 list_splice_tail_init(&sci->sc_segbufs, &logs);
1830 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1831 nilfs_free_incomplete_logs(&logs, nilfs);
1833 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1834 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1838 WARN_ON(ret); /* do not happen */
1841 nilfs_destroy_logs(&logs);
1844 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1845 struct nilfs_segment_buffer *segbuf)
1847 nilfs->ns_segnum = segbuf->sb_segnum;
1848 nilfs->ns_nextnum = segbuf->sb_nextnum;
1849 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1850 + segbuf->sb_sum.nblocks;
1851 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1852 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1855 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1857 struct nilfs_segment_buffer *segbuf;
1858 struct page *bd_page = NULL, *fs_page = NULL;
1859 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1860 int update_sr = false;
1862 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1863 struct buffer_head *bh;
1865 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1867 set_buffer_uptodate(bh);
1868 clear_buffer_dirty(bh);
1869 if (bh->b_page != bd_page) {
1871 end_page_writeback(bd_page);
1872 bd_page = bh->b_page;
1876 * We assume that the buffers which belong to the same page
1877 * continue over the buffer list.
1878 * Under this assumption, the last BHs of pages is
1879 * identifiable by the discontinuity of bh->b_page
1880 * (page != fs_page).
1882 * For B-tree node blocks, however, this assumption is not
1883 * guaranteed. The cleanup code of B-tree node pages needs
1886 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1888 const unsigned long set_bits = BIT(BH_Uptodate);
1889 const unsigned long clear_bits =
1890 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1891 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1892 BIT(BH_NILFS_Redirected));
1894 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1895 if (bh == segbuf->sb_super_root) {
1896 if (bh->b_page != bd_page) {
1897 end_page_writeback(bd_page);
1898 bd_page = bh->b_page;
1903 if (bh->b_page != fs_page) {
1904 nilfs_end_page_io(fs_page, 0);
1905 fs_page = bh->b_page;
1909 if (!nilfs_segbuf_simplex(segbuf)) {
1910 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1911 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1912 sci->sc_lseg_stime = jiffies;
1914 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1915 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1919 * Since pages may continue over multiple segment buffers,
1920 * end of the last page must be checked outside of the loop.
1923 end_page_writeback(bd_page);
1925 nilfs_end_page_io(fs_page, 0);
1927 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1929 if (nilfs_doing_gc())
1930 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1932 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1934 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1936 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1937 nilfs_set_next_segment(nilfs, segbuf);
1940 nilfs->ns_flushed_device = 0;
1941 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1942 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1944 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1945 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1946 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1947 nilfs_segctor_clear_metadata_dirty(sci);
1949 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1952 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1956 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1958 nilfs_segctor_complete_write(sci);
1959 nilfs_destroy_logs(&sci->sc_write_logs);
1964 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1965 struct the_nilfs *nilfs)
1967 struct nilfs_inode_info *ii, *n;
1968 struct inode *ifile = sci->sc_root->ifile;
1970 spin_lock(&nilfs->ns_inode_lock);
1972 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1974 struct buffer_head *ibh;
1977 spin_unlock(&nilfs->ns_inode_lock);
1978 err = nilfs_ifile_get_inode_block(
1979 ifile, ii->vfs_inode.i_ino, &ibh);
1980 if (unlikely(err)) {
1981 nilfs_warn(sci->sc_super,
1982 "log writer: error %d getting inode block (ino=%lu)",
1983 err, ii->vfs_inode.i_ino);
1986 spin_lock(&nilfs->ns_inode_lock);
1987 if (likely(!ii->i_bh))
1994 // Always redirty the buffer to avoid race condition
1995 mark_buffer_dirty(ii->i_bh);
1996 nilfs_mdt_mark_dirty(ifile);
1998 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1999 set_bit(NILFS_I_BUSY, &ii->i_state);
2000 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2002 spin_unlock(&nilfs->ns_inode_lock);
2007 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2008 struct the_nilfs *nilfs)
2010 struct nilfs_inode_info *ii, *n;
2011 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2012 int defer_iput = false;
2014 spin_lock(&nilfs->ns_inode_lock);
2015 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2016 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2017 test_bit(NILFS_I_DIRTY, &ii->i_state))
2020 clear_bit(NILFS_I_BUSY, &ii->i_state);
2023 list_del_init(&ii->i_dirty);
2024 if (!ii->vfs_inode.i_nlink || during_mount) {
2026 * Defer calling iput() to avoid deadlocks if
2027 * i_nlink == 0 or mount is not yet finished.
2029 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2032 spin_unlock(&nilfs->ns_inode_lock);
2033 iput(&ii->vfs_inode);
2034 spin_lock(&nilfs->ns_inode_lock);
2037 spin_unlock(&nilfs->ns_inode_lock);
2040 schedule_work(&sci->sc_iput_work);
2044 * Main procedure of segment constructor
2046 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2048 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2051 if (sb_rdonly(sci->sc_super))
2054 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2055 sci->sc_cno = nilfs->ns_cno;
2057 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2061 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2062 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2064 if (nilfs_segctor_clean(sci))
2068 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2070 err = nilfs_segctor_begin_construction(sci, nilfs);
2074 /* Update time stamp */
2075 sci->sc_seg_ctime = ktime_get_real_seconds();
2077 err = nilfs_segctor_collect(sci, nilfs, mode);
2081 /* Avoid empty segment */
2082 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2083 nilfs_segbuf_empty(sci->sc_curseg)) {
2084 nilfs_segctor_abort_construction(sci, nilfs, 1);
2088 err = nilfs_segctor_assign(sci, mode);
2092 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2093 nilfs_segctor_fill_in_file_bmap(sci);
2095 if (mode == SC_LSEG_SR &&
2096 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2097 err = nilfs_segctor_fill_in_checkpoint(sci);
2099 goto failed_to_write;
2101 nilfs_segctor_fill_in_super_root(sci, nilfs);
2103 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2105 /* Write partial segments */
2106 nilfs_segctor_prepare_write(sci);
2108 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2109 nilfs->ns_crc_seed);
2111 err = nilfs_segctor_write(sci, nilfs);
2113 goto failed_to_write;
2115 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2116 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2118 * At this point, we avoid double buffering
2119 * for blocksize < pagesize because page dirty
2120 * flag is turned off during write and dirty
2121 * buffers are not properly collected for
2122 * pages crossing over segments.
2124 err = nilfs_segctor_wait(sci);
2126 goto failed_to_write;
2128 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2131 nilfs_segctor_drop_written_files(sci, nilfs);
2135 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2136 nilfs_redirty_inodes(&sci->sc_dirty_files);
2139 if (nilfs_doing_gc())
2140 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2141 nilfs_segctor_abort_construction(sci, nilfs, err);
2146 * nilfs_segctor_start_timer - set timer of background write
2147 * @sci: nilfs_sc_info
2149 * If the timer has already been set, it ignores the new request.
2150 * This function MUST be called within a section locking the segment
2153 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2155 spin_lock(&sci->sc_state_lock);
2156 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2157 sci->sc_timer.expires = jiffies + sci->sc_interval;
2158 add_timer(&sci->sc_timer);
2159 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2161 spin_unlock(&sci->sc_state_lock);
2164 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2166 spin_lock(&sci->sc_state_lock);
2167 if (!(sci->sc_flush_request & BIT(bn))) {
2168 unsigned long prev_req = sci->sc_flush_request;
2170 sci->sc_flush_request |= BIT(bn);
2172 wake_up(&sci->sc_wait_daemon);
2174 spin_unlock(&sci->sc_state_lock);
2178 * nilfs_flush_segment - trigger a segment construction for resource control
2180 * @ino: inode number of the file to be flushed out.
2182 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2184 struct the_nilfs *nilfs = sb->s_fs_info;
2185 struct nilfs_sc_info *sci = nilfs->ns_writer;
2187 if (!sci || nilfs_doing_construction())
2189 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2190 /* assign bit 0 to data files */
2193 struct nilfs_segctor_wait_request {
2194 wait_queue_entry_t wq;
2200 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2202 struct nilfs_segctor_wait_request wait_req;
2205 spin_lock(&sci->sc_state_lock);
2206 init_wait(&wait_req.wq);
2208 atomic_set(&wait_req.done, 0);
2209 wait_req.seq = ++sci->sc_seq_request;
2210 spin_unlock(&sci->sc_state_lock);
2212 init_waitqueue_entry(&wait_req.wq, current);
2213 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2214 set_current_state(TASK_INTERRUPTIBLE);
2215 wake_up(&sci->sc_wait_daemon);
2218 if (atomic_read(&wait_req.done)) {
2222 if (!signal_pending(current)) {
2229 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2233 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2235 struct nilfs_segctor_wait_request *wrq, *n;
2236 unsigned long flags;
2238 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2239 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2240 if (!atomic_read(&wrq->done) &&
2241 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2243 atomic_set(&wrq->done, 1);
2245 if (atomic_read(&wrq->done)) {
2246 wrq->wq.func(&wrq->wq,
2247 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2251 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2255 * nilfs_construct_segment - construct a logical segment
2258 * Return Value: On success, 0 is returned. On errors, one of the following
2259 * negative error code is returned.
2261 * %-EROFS - Read only filesystem.
2265 * %-ENOSPC - No space left on device (only in a panic state).
2267 * %-ERESTARTSYS - Interrupted.
2269 * %-ENOMEM - Insufficient memory available.
2271 int nilfs_construct_segment(struct super_block *sb)
2273 struct the_nilfs *nilfs = sb->s_fs_info;
2274 struct nilfs_sc_info *sci = nilfs->ns_writer;
2275 struct nilfs_transaction_info *ti;
2277 if (sb_rdonly(sb) || unlikely(!sci))
2280 /* A call inside transactions causes a deadlock. */
2281 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2283 return nilfs_segctor_sync(sci);
2287 * nilfs_construct_dsync_segment - construct a data-only logical segment
2289 * @inode: inode whose data blocks should be written out
2290 * @start: start byte offset
2291 * @end: end byte offset (inclusive)
2293 * Return Value: On success, 0 is returned. On errors, one of the following
2294 * negative error code is returned.
2296 * %-EROFS - Read only filesystem.
2300 * %-ENOSPC - No space left on device (only in a panic state).
2302 * %-ERESTARTSYS - Interrupted.
2304 * %-ENOMEM - Insufficient memory available.
2306 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2307 loff_t start, loff_t end)
2309 struct the_nilfs *nilfs = sb->s_fs_info;
2310 struct nilfs_sc_info *sci = nilfs->ns_writer;
2311 struct nilfs_inode_info *ii;
2312 struct nilfs_transaction_info ti;
2315 if (sb_rdonly(sb) || unlikely(!sci))
2318 nilfs_transaction_lock(sb, &ti, 0);
2320 ii = NILFS_I(inode);
2321 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2322 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2323 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2324 nilfs_discontinued(nilfs)) {
2325 nilfs_transaction_unlock(sb);
2326 err = nilfs_segctor_sync(sci);
2330 spin_lock(&nilfs->ns_inode_lock);
2331 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2332 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2333 spin_unlock(&nilfs->ns_inode_lock);
2334 nilfs_transaction_unlock(sb);
2337 spin_unlock(&nilfs->ns_inode_lock);
2338 sci->sc_dsync_inode = ii;
2339 sci->sc_dsync_start = start;
2340 sci->sc_dsync_end = end;
2342 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2344 nilfs->ns_flushed_device = 0;
2346 nilfs_transaction_unlock(sb);
2350 #define FLUSH_FILE_BIT (0x1) /* data file only */
2351 #define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2354 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2355 * @sci: segment constructor object
2357 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2359 spin_lock(&sci->sc_state_lock);
2360 sci->sc_seq_accepted = sci->sc_seq_request;
2361 spin_unlock(&sci->sc_state_lock);
2362 del_timer_sync(&sci->sc_timer);
2366 * nilfs_segctor_notify - notify the result of request to caller threads
2367 * @sci: segment constructor object
2368 * @mode: mode of log forming
2369 * @err: error code to be notified
2371 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2373 /* Clear requests (even when the construction failed) */
2374 spin_lock(&sci->sc_state_lock);
2376 if (mode == SC_LSEG_SR) {
2377 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2378 sci->sc_seq_done = sci->sc_seq_accepted;
2379 nilfs_segctor_wakeup(sci, err);
2380 sci->sc_flush_request = 0;
2382 if (mode == SC_FLUSH_FILE)
2383 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2384 else if (mode == SC_FLUSH_DAT)
2385 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2387 /* re-enable timer if checkpoint creation was not done */
2388 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2389 time_before(jiffies, sci->sc_timer.expires))
2390 add_timer(&sci->sc_timer);
2392 spin_unlock(&sci->sc_state_lock);
2396 * nilfs_segctor_construct - form logs and write them to disk
2397 * @sci: segment constructor object
2398 * @mode: mode of log forming
2400 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2402 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2403 struct nilfs_super_block **sbp;
2406 nilfs_segctor_accept(sci);
2408 if (nilfs_discontinued(nilfs))
2410 if (!nilfs_segctor_confirm(sci))
2411 err = nilfs_segctor_do_construct(sci, mode);
2414 if (mode != SC_FLUSH_DAT)
2415 atomic_set(&nilfs->ns_ndirtyblks, 0);
2416 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2417 nilfs_discontinued(nilfs)) {
2418 down_write(&nilfs->ns_sem);
2420 sbp = nilfs_prepare_super(sci->sc_super,
2421 nilfs_sb_will_flip(nilfs));
2423 nilfs_set_log_cursor(sbp[0], nilfs);
2424 err = nilfs_commit_super(sci->sc_super,
2427 up_write(&nilfs->ns_sem);
2431 nilfs_segctor_notify(sci, mode, err);
2435 static void nilfs_construction_timeout(struct timer_list *t)
2437 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2439 wake_up_process(sci->sc_timer_task);
2443 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2445 struct nilfs_inode_info *ii, *n;
2447 list_for_each_entry_safe(ii, n, head, i_dirty) {
2448 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2450 list_del_init(&ii->i_dirty);
2451 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2452 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2453 iput(&ii->vfs_inode);
2457 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2460 struct the_nilfs *nilfs = sb->s_fs_info;
2461 struct nilfs_sc_info *sci = nilfs->ns_writer;
2462 struct nilfs_transaction_info ti;
2468 nilfs_transaction_lock(sb, &ti, 1);
2470 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2474 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2475 if (unlikely(err)) {
2476 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2480 sci->sc_freesegs = kbufs[4];
2481 sci->sc_nfreesegs = argv[4].v_nmembs;
2482 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2485 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2486 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2491 nilfs_warn(sb, "error %d cleaning segments", err);
2492 set_current_state(TASK_INTERRUPTIBLE);
2493 schedule_timeout(sci->sc_interval);
2495 if (nilfs_test_opt(nilfs, DISCARD)) {
2496 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2500 "error %d on discard request, turning discards off for the device",
2502 nilfs_clear_opt(nilfs, DISCARD);
2507 sci->sc_freesegs = NULL;
2508 sci->sc_nfreesegs = 0;
2509 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2510 nilfs_transaction_unlock(sb);
2514 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2516 struct nilfs_transaction_info ti;
2518 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2519 nilfs_segctor_construct(sci, mode);
2522 * Unclosed segment should be retried. We do this using sc_timer.
2523 * Timeout of sc_timer will invoke complete construction which leads
2524 * to close the current logical segment.
2526 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2527 nilfs_segctor_start_timer(sci);
2529 nilfs_transaction_unlock(sci->sc_super);
2532 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2536 spin_lock(&sci->sc_state_lock);
2537 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2538 SC_FLUSH_DAT : SC_FLUSH_FILE;
2539 spin_unlock(&sci->sc_state_lock);
2542 nilfs_segctor_do_construct(sci, mode);
2544 spin_lock(&sci->sc_state_lock);
2545 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2546 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2547 spin_unlock(&sci->sc_state_lock);
2549 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2552 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2554 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2555 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2556 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2557 return SC_FLUSH_FILE;
2558 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2559 return SC_FLUSH_DAT;
2565 * nilfs_segctor_thread - main loop of the segment constructor thread.
2566 * @arg: pointer to a struct nilfs_sc_info.
2568 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2569 * to execute segment constructions.
2571 static int nilfs_segctor_thread(void *arg)
2573 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2574 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2577 sci->sc_timer_task = current;
2580 sci->sc_task = current;
2581 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2582 nilfs_info(sci->sc_super,
2583 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2584 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2586 spin_lock(&sci->sc_state_lock);
2591 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2594 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2596 else if (sci->sc_flush_request)
2597 mode = nilfs_segctor_flush_mode(sci);
2601 spin_unlock(&sci->sc_state_lock);
2602 nilfs_segctor_thread_construct(sci, mode);
2603 spin_lock(&sci->sc_state_lock);
2608 if (freezing(current)) {
2609 spin_unlock(&sci->sc_state_lock);
2611 spin_lock(&sci->sc_state_lock);
2614 int should_sleep = 1;
2616 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2617 TASK_INTERRUPTIBLE);
2619 if (sci->sc_seq_request != sci->sc_seq_done)
2621 else if (sci->sc_flush_request)
2623 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2624 should_sleep = time_before(jiffies,
2625 sci->sc_timer.expires);
2628 spin_unlock(&sci->sc_state_lock);
2630 spin_lock(&sci->sc_state_lock);
2632 finish_wait(&sci->sc_wait_daemon, &wait);
2633 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2634 time_after_eq(jiffies, sci->sc_timer.expires));
2636 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2637 set_nilfs_discontinued(nilfs);
2643 sci->sc_task = NULL;
2644 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2645 spin_unlock(&sci->sc_state_lock);
2649 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2651 struct task_struct *t;
2653 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2655 int err = PTR_ERR(t);
2657 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2661 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2665 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2666 __acquires(&sci->sc_state_lock)
2667 __releases(&sci->sc_state_lock)
2669 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2671 while (sci->sc_task) {
2672 wake_up(&sci->sc_wait_daemon);
2673 spin_unlock(&sci->sc_state_lock);
2674 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2675 spin_lock(&sci->sc_state_lock);
2680 * Setup & clean-up functions
2682 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2683 struct nilfs_root *root)
2685 struct the_nilfs *nilfs = sb->s_fs_info;
2686 struct nilfs_sc_info *sci;
2688 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2694 nilfs_get_root(root);
2695 sci->sc_root = root;
2697 init_waitqueue_head(&sci->sc_wait_request);
2698 init_waitqueue_head(&sci->sc_wait_daemon);
2699 init_waitqueue_head(&sci->sc_wait_task);
2700 spin_lock_init(&sci->sc_state_lock);
2701 INIT_LIST_HEAD(&sci->sc_dirty_files);
2702 INIT_LIST_HEAD(&sci->sc_segbufs);
2703 INIT_LIST_HEAD(&sci->sc_write_logs);
2704 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2705 INIT_LIST_HEAD(&sci->sc_iput_queue);
2706 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2707 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2709 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2710 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2711 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2713 if (nilfs->ns_interval)
2714 sci->sc_interval = HZ * nilfs->ns_interval;
2715 if (nilfs->ns_watermark)
2716 sci->sc_watermark = nilfs->ns_watermark;
2720 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2722 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2725 * The segctord thread was stopped and its timer was removed.
2726 * But some tasks remain.
2729 struct nilfs_transaction_info ti;
2731 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2732 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2733 nilfs_transaction_unlock(sci->sc_super);
2735 flush_work(&sci->sc_iput_work);
2737 } while (ret && ret != -EROFS && retrycount-- > 0);
2741 * nilfs_segctor_destroy - destroy the segment constructor.
2742 * @sci: nilfs_sc_info
2744 * nilfs_segctor_destroy() kills the segctord thread and frees
2745 * the nilfs_sc_info struct.
2746 * Caller must hold the segment semaphore.
2748 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2750 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2753 up_write(&nilfs->ns_segctor_sem);
2755 spin_lock(&sci->sc_state_lock);
2756 nilfs_segctor_kill_thread(sci);
2757 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2758 || sci->sc_seq_request != sci->sc_seq_done);
2759 spin_unlock(&sci->sc_state_lock);
2761 if (flush_work(&sci->sc_iput_work))
2764 if (flag || !nilfs_segctor_confirm(sci))
2765 nilfs_segctor_write_out(sci);
2767 if (!list_empty(&sci->sc_dirty_files)) {
2768 nilfs_warn(sci->sc_super,
2769 "disposed unprocessed dirty file(s) when stopping log writer");
2770 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2773 if (!list_empty(&sci->sc_iput_queue)) {
2774 nilfs_warn(sci->sc_super,
2775 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2776 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2779 WARN_ON(!list_empty(&sci->sc_segbufs));
2780 WARN_ON(!list_empty(&sci->sc_write_logs));
2782 nilfs_put_root(sci->sc_root);
2784 down_write(&nilfs->ns_segctor_sem);
2786 timer_shutdown_sync(&sci->sc_timer);
2791 * nilfs_attach_log_writer - attach log writer
2792 * @sb: super block instance
2793 * @root: root object of the current filesystem tree
2795 * This allocates a log writer object, initializes it, and starts the
2798 * Return Value: On success, 0 is returned. On error, one of the following
2799 * negative error code is returned.
2801 * %-ENOMEM - Insufficient memory available.
2803 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2805 struct the_nilfs *nilfs = sb->s_fs_info;
2808 if (nilfs->ns_writer) {
2810 * This happens if the filesystem is made read-only by
2811 * __nilfs_error or nilfs_remount and then remounted
2812 * read/write. In these cases, reuse the existing
2818 nilfs->ns_writer = nilfs_segctor_new(sb, root);
2819 if (!nilfs->ns_writer)
2822 inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2824 err = nilfs_segctor_start_thread(nilfs->ns_writer);
2826 nilfs_detach_log_writer(sb);
2832 * nilfs_detach_log_writer - destroy log writer
2833 * @sb: super block instance
2835 * This kills log writer daemon, frees the log writer object, and
2836 * destroys list of dirty files.
2838 void nilfs_detach_log_writer(struct super_block *sb)
2840 struct the_nilfs *nilfs = sb->s_fs_info;
2841 LIST_HEAD(garbage_list);
2843 down_write(&nilfs->ns_segctor_sem);
2844 if (nilfs->ns_writer) {
2845 nilfs_segctor_destroy(nilfs->ns_writer);
2846 nilfs->ns_writer = NULL;
2849 /* Force to free the list of dirty files */
2850 spin_lock(&nilfs->ns_inode_lock);
2851 if (!list_empty(&nilfs->ns_dirty_files)) {
2852 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2854 "disposed unprocessed dirty file(s) when detaching log writer");
2856 spin_unlock(&nilfs->ns_inode_lock);
2857 up_write(&nilfs->ns_segctor_sem);
2859 nilfs_dispose_list(nilfs, &garbage_list, 1);