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 if (unlikely(folio->mapping != mapping)) {
729 /* Exclude folios removed from the address space */
733 head = folio_buffers(folio);
735 head = create_empty_buffers(folio,
736 i_blocksize(inode), 0);
741 if (!buffer_dirty(bh) || buffer_async_write(bh))
744 list_add_tail(&bh->b_assoc_buffers, listp);
746 if (unlikely(ndirties >= nlimit)) {
747 folio_batch_release(&fbatch);
751 } while (bh = bh->b_this_page, bh != head);
753 folio_batch_release(&fbatch);
758 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
759 struct list_head *listp)
761 struct nilfs_inode_info *ii = NILFS_I(inode);
762 struct inode *btnc_inode = ii->i_assoc_inode;
763 struct folio_batch fbatch;
764 struct buffer_head *bh, *head;
770 folio_batch_init(&fbatch);
772 while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
773 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
774 for (i = 0; i < folio_batch_count(&fbatch); i++) {
775 bh = head = folio_buffers(fbatch.folios[i]);
777 if (buffer_dirty(bh) &&
778 !buffer_async_write(bh)) {
780 list_add_tail(&bh->b_assoc_buffers,
783 bh = bh->b_this_page;
784 } while (bh != head);
786 folio_batch_release(&fbatch);
791 static void nilfs_dispose_list(struct the_nilfs *nilfs,
792 struct list_head *head, int force)
794 struct nilfs_inode_info *ii, *n;
795 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
798 while (!list_empty(head)) {
799 spin_lock(&nilfs->ns_inode_lock);
800 list_for_each_entry_safe(ii, n, head, i_dirty) {
801 list_del_init(&ii->i_dirty);
803 if (unlikely(ii->i_bh)) {
807 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
808 set_bit(NILFS_I_QUEUED, &ii->i_state);
809 list_add_tail(&ii->i_dirty,
810 &nilfs->ns_dirty_files);
814 if (nv == SC_N_INODEVEC)
817 spin_unlock(&nilfs->ns_inode_lock);
819 for (pii = ivec; nv > 0; pii++, nv--)
820 iput(&(*pii)->vfs_inode);
824 static void nilfs_iput_work_func(struct work_struct *work)
826 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
828 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
830 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
833 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
834 struct nilfs_root *root)
838 if (nilfs_mdt_fetch_dirty(root->ifile))
840 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
842 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
844 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
849 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
851 return list_empty(&sci->sc_dirty_files) &&
852 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
853 sci->sc_nfreesegs == 0 &&
854 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
857 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
859 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
862 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
863 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
865 spin_lock(&nilfs->ns_inode_lock);
866 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
869 spin_unlock(&nilfs->ns_inode_lock);
873 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
875 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
877 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
878 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
879 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
880 nilfs_mdt_clear_dirty(nilfs->ns_dat);
883 static void nilfs_fill_in_file_bmap(struct inode *ifile,
884 struct nilfs_inode_info *ii)
887 struct buffer_head *ibh;
888 struct nilfs_inode *raw_inode;
890 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
893 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
895 nilfs_bmap_write(ii->i_bmap, raw_inode);
896 nilfs_ifile_unmap_inode(raw_inode);
900 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
902 struct nilfs_inode_info *ii;
904 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
905 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
906 set_bit(NILFS_I_COLLECTED, &ii->i_state);
911 * nilfs_write_root_mdt_inode - export root metadata inode information to
913 * @inode: inode object of the root metadata file
914 * @raw_inode: on-disk inode
916 * nilfs_write_root_mdt_inode() writes inode information and bmap data of
917 * @inode to the inode area of the metadata file allocated on the super root
918 * block created to finalize the log. Since super root blocks are configured
919 * each time, this function zero-fills the unused area of @raw_inode.
921 static void nilfs_write_root_mdt_inode(struct inode *inode,
922 struct nilfs_inode *raw_inode)
924 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
926 nilfs_write_inode_common(inode, raw_inode);
928 /* zero-fill unused portion of raw_inode */
929 raw_inode->i_xattr = 0;
930 raw_inode->i_pad = 0;
931 memset((void *)raw_inode + sizeof(*raw_inode), 0,
932 nilfs->ns_inode_size - sizeof(*raw_inode));
934 nilfs_bmap_write(NILFS_I(inode)->i_bmap, raw_inode);
937 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
938 struct the_nilfs *nilfs)
940 struct buffer_head *bh_sr;
941 struct nilfs_super_root *raw_sr;
942 unsigned int isz, srsz;
944 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
947 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
948 isz = nilfs->ns_inode_size;
949 srsz = NILFS_SR_BYTES(isz);
951 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
952 raw_sr->sr_bytes = cpu_to_le16(srsz);
953 raw_sr->sr_nongc_ctime
954 = cpu_to_le64(nilfs_doing_gc() ?
955 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
956 raw_sr->sr_flags = 0;
958 nilfs_write_root_mdt_inode(nilfs->ns_dat, (void *)raw_sr +
959 NILFS_SR_DAT_OFFSET(isz));
960 nilfs_write_root_mdt_inode(nilfs->ns_cpfile, (void *)raw_sr +
961 NILFS_SR_CPFILE_OFFSET(isz));
962 nilfs_write_root_mdt_inode(nilfs->ns_sufile, (void *)raw_sr +
963 NILFS_SR_SUFILE_OFFSET(isz));
965 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
966 set_buffer_uptodate(bh_sr);
967 unlock_buffer(bh_sr);
970 static void nilfs_redirty_inodes(struct list_head *head)
972 struct nilfs_inode_info *ii;
974 list_for_each_entry(ii, head, i_dirty) {
975 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
976 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
980 static void nilfs_drop_collected_inodes(struct list_head *head)
982 struct nilfs_inode_info *ii;
984 list_for_each_entry(ii, head, i_dirty) {
985 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
988 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
989 set_bit(NILFS_I_UPDATED, &ii->i_state);
993 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
995 struct list_head *listp,
996 int (*collect)(struct nilfs_sc_info *,
997 struct buffer_head *,
1000 struct buffer_head *bh, *n;
1004 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1005 list_del_init(&bh->b_assoc_buffers);
1006 err = collect(sci, bh, inode);
1009 goto dispose_buffers;
1015 while (!list_empty(listp)) {
1016 bh = list_first_entry(listp, struct buffer_head,
1018 list_del_init(&bh->b_assoc_buffers);
1024 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1026 /* Remaining number of blocks within segment buffer */
1027 return sci->sc_segbuf_nblocks -
1028 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1031 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1032 struct inode *inode,
1033 const struct nilfs_sc_operations *sc_ops)
1035 LIST_HEAD(data_buffers);
1036 LIST_HEAD(node_buffers);
1039 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1040 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1042 n = nilfs_lookup_dirty_data_buffers(
1043 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1045 err = nilfs_segctor_apply_buffers(
1046 sci, inode, &data_buffers,
1047 sc_ops->collect_data);
1048 BUG_ON(!err); /* always receive -E2BIG or true error */
1052 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1054 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1055 err = nilfs_segctor_apply_buffers(
1056 sci, inode, &data_buffers, sc_ops->collect_data);
1057 if (unlikely(err)) {
1058 /* dispose node list */
1059 nilfs_segctor_apply_buffers(
1060 sci, inode, &node_buffers, NULL);
1063 sci->sc_stage.flags |= NILFS_CF_NODE;
1066 err = nilfs_segctor_apply_buffers(
1067 sci, inode, &node_buffers, sc_ops->collect_node);
1071 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1072 err = nilfs_segctor_apply_buffers(
1073 sci, inode, &node_buffers, sc_ops->collect_bmap);
1077 nilfs_segctor_end_finfo(sci, inode);
1078 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1084 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1085 struct inode *inode)
1087 LIST_HEAD(data_buffers);
1088 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1091 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1092 sci->sc_dsync_start,
1095 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1096 nilfs_collect_file_data);
1098 nilfs_segctor_end_finfo(sci, inode);
1100 /* always receive -E2BIG or true error if n > rest */
1105 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1107 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1108 struct list_head *head;
1109 struct nilfs_inode_info *ii;
1113 switch (nilfs_sc_cstage_get(sci)) {
1116 sci->sc_stage.flags = 0;
1118 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1119 sci->sc_nblk_inc = 0;
1120 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1121 if (mode == SC_LSEG_DSYNC) {
1122 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1127 sci->sc_stage.dirty_file_ptr = NULL;
1128 sci->sc_stage.gc_inode_ptr = NULL;
1129 if (mode == SC_FLUSH_DAT) {
1130 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1133 nilfs_sc_cstage_inc(sci);
1136 if (nilfs_doing_gc()) {
1137 head = &sci->sc_gc_inodes;
1138 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1140 list_for_each_entry_continue(ii, head, i_dirty) {
1141 err = nilfs_segctor_scan_file(
1142 sci, &ii->vfs_inode,
1143 &nilfs_sc_file_ops);
1144 if (unlikely(err)) {
1145 sci->sc_stage.gc_inode_ptr = list_entry(
1147 struct nilfs_inode_info,
1151 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1153 sci->sc_stage.gc_inode_ptr = NULL;
1155 nilfs_sc_cstage_inc(sci);
1158 head = &sci->sc_dirty_files;
1159 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1161 list_for_each_entry_continue(ii, head, i_dirty) {
1162 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1164 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1165 &nilfs_sc_file_ops);
1166 if (unlikely(err)) {
1167 sci->sc_stage.dirty_file_ptr =
1168 list_entry(ii->i_dirty.prev,
1169 struct nilfs_inode_info,
1173 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1174 /* XXX: required ? */
1176 sci->sc_stage.dirty_file_ptr = NULL;
1177 if (mode == SC_FLUSH_FILE) {
1178 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1181 nilfs_sc_cstage_inc(sci);
1182 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1184 case NILFS_ST_IFILE:
1185 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1186 &nilfs_sc_file_ops);
1189 nilfs_sc_cstage_inc(sci);
1190 /* Creating a checkpoint */
1191 err = nilfs_cpfile_create_checkpoint(nilfs->ns_cpfile,
1196 case NILFS_ST_CPFILE:
1197 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1198 &nilfs_sc_file_ops);
1201 nilfs_sc_cstage_inc(sci);
1203 case NILFS_ST_SUFILE:
1204 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1205 sci->sc_nfreesegs, &ndone);
1206 if (unlikely(err)) {
1207 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1208 sci->sc_freesegs, ndone,
1212 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1214 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1215 &nilfs_sc_file_ops);
1218 nilfs_sc_cstage_inc(sci);
1222 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1226 if (mode == SC_FLUSH_DAT) {
1227 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1230 nilfs_sc_cstage_inc(sci);
1233 if (mode == SC_LSEG_SR) {
1234 /* Appending a super root */
1235 err = nilfs_segctor_add_super_root(sci);
1239 /* End of a logical segment */
1240 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1241 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1243 case NILFS_ST_DSYNC:
1245 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1246 ii = sci->sc_dsync_inode;
1247 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1250 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1253 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1254 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1267 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1268 * @sci: nilfs_sc_info
1269 * @nilfs: nilfs object
1271 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1272 struct the_nilfs *nilfs)
1274 struct nilfs_segment_buffer *segbuf, *prev;
1278 segbuf = nilfs_segbuf_new(sci->sc_super);
1279 if (unlikely(!segbuf))
1282 if (list_empty(&sci->sc_write_logs)) {
1283 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1284 nilfs->ns_pseg_offset, nilfs);
1285 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1286 nilfs_shift_to_next_segment(nilfs);
1287 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1290 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1291 nextnum = nilfs->ns_nextnum;
1293 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1294 /* Start from the head of a new full segment */
1298 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1299 nilfs_segbuf_map_cont(segbuf, prev);
1300 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1301 nextnum = prev->sb_nextnum;
1303 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1304 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1305 segbuf->sb_sum.seg_seq++;
1310 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1315 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1319 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1321 BUG_ON(!list_empty(&sci->sc_segbufs));
1322 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1323 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1327 nilfs_segbuf_free(segbuf);
1331 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1332 struct the_nilfs *nilfs, int nadd)
1334 struct nilfs_segment_buffer *segbuf, *prev;
1335 struct inode *sufile = nilfs->ns_sufile;
1340 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1342 * Since the segment specified with nextnum might be allocated during
1343 * the previous construction, the buffer including its segusage may
1344 * not be dirty. The following call ensures that the buffer is dirty
1345 * and will pin the buffer on memory until the sufile is written.
1347 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1351 for (i = 0; i < nadd; i++) {
1352 /* extend segment info */
1354 segbuf = nilfs_segbuf_new(sci->sc_super);
1355 if (unlikely(!segbuf))
1358 /* map this buffer to region of segment on-disk */
1359 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1360 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1362 /* allocate the next next full segment */
1363 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1367 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1368 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1370 list_add_tail(&segbuf->sb_list, &list);
1373 list_splice_tail(&list, &sci->sc_segbufs);
1377 nilfs_segbuf_free(segbuf);
1379 list_for_each_entry(segbuf, &list, sb_list) {
1380 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1381 WARN_ON(ret); /* never fails */
1383 nilfs_destroy_logs(&list);
1387 static void nilfs_free_incomplete_logs(struct list_head *logs,
1388 struct the_nilfs *nilfs)
1390 struct nilfs_segment_buffer *segbuf, *prev;
1391 struct inode *sufile = nilfs->ns_sufile;
1394 segbuf = NILFS_FIRST_SEGBUF(logs);
1395 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1396 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1397 WARN_ON(ret); /* never fails */
1399 if (atomic_read(&segbuf->sb_err)) {
1400 /* Case 1: The first segment failed */
1401 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1403 * Case 1a: Partial segment appended into an existing
1406 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1407 segbuf->sb_fseg_end);
1408 else /* Case 1b: New full segment */
1409 set_nilfs_discontinued(nilfs);
1413 list_for_each_entry_continue(segbuf, logs, sb_list) {
1414 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1415 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1416 WARN_ON(ret); /* never fails */
1418 if (atomic_read(&segbuf->sb_err) &&
1419 segbuf->sb_segnum != nilfs->ns_nextnum)
1420 /* Case 2: extended segment (!= next) failed */
1421 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1426 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1427 struct inode *sufile)
1429 struct nilfs_segment_buffer *segbuf;
1430 unsigned long live_blocks;
1433 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1434 live_blocks = segbuf->sb_sum.nblocks +
1435 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1436 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1439 WARN_ON(ret); /* always succeed because the segusage is dirty */
1443 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1445 struct nilfs_segment_buffer *segbuf;
1448 segbuf = NILFS_FIRST_SEGBUF(logs);
1449 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1450 segbuf->sb_pseg_start -
1451 segbuf->sb_fseg_start, 0);
1452 WARN_ON(ret); /* always succeed because the segusage is dirty */
1454 list_for_each_entry_continue(segbuf, logs, sb_list) {
1455 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1457 WARN_ON(ret); /* always succeed */
1461 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1462 struct nilfs_segment_buffer *last,
1463 struct inode *sufile)
1465 struct nilfs_segment_buffer *segbuf = last;
1468 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1469 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1470 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1473 nilfs_truncate_logs(&sci->sc_segbufs, last);
1477 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1478 struct the_nilfs *nilfs, int mode)
1480 struct nilfs_cstage prev_stage = sci->sc_stage;
1483 /* Collection retry loop */
1485 sci->sc_nblk_this_inc = 0;
1486 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1488 err = nilfs_segctor_reset_segment_buffer(sci);
1492 err = nilfs_segctor_collect_blocks(sci, mode);
1493 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1497 if (unlikely(err != -E2BIG))
1500 /* The current segment is filled up */
1501 if (mode != SC_LSEG_SR ||
1502 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1505 nilfs_clear_logs(&sci->sc_segbufs);
1507 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1508 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1512 WARN_ON(err); /* do not happen */
1513 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1516 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1520 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1521 sci->sc_stage = prev_stage;
1523 nilfs_segctor_zeropad_segsum(sci);
1524 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1531 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1532 struct buffer_head *new_bh)
1534 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1536 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1537 /* The caller must release old_bh */
1541 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1542 struct nilfs_segment_buffer *segbuf,
1545 struct inode *inode = NULL;
1547 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1548 unsigned long nblocks = 0, ndatablk = 0;
1549 const struct nilfs_sc_operations *sc_op = NULL;
1550 struct nilfs_segsum_pointer ssp;
1551 struct nilfs_finfo *finfo = NULL;
1552 union nilfs_binfo binfo;
1553 struct buffer_head *bh, *bh_org;
1560 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1561 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1562 ssp.offset = sizeof(struct nilfs_segment_summary);
1564 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1565 if (bh == segbuf->sb_super_root)
1568 finfo = nilfs_segctor_map_segsum_entry(
1569 sci, &ssp, sizeof(*finfo));
1570 ino = le64_to_cpu(finfo->fi_ino);
1571 nblocks = le32_to_cpu(finfo->fi_nblocks);
1572 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1574 inode = bh->b_folio->mapping->host;
1576 if (mode == SC_LSEG_DSYNC)
1577 sc_op = &nilfs_sc_dsync_ops;
1578 else if (ino == NILFS_DAT_INO)
1579 sc_op = &nilfs_sc_dat_ops;
1580 else /* file blocks */
1581 sc_op = &nilfs_sc_file_ops;
1585 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1588 nilfs_list_replace_buffer(bh_org, bh);
1594 sc_op->write_data_binfo(sci, &ssp, &binfo);
1596 sc_op->write_node_binfo(sci, &ssp, &binfo);
1599 if (--nblocks == 0) {
1603 } else if (ndatablk > 0)
1613 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1615 struct nilfs_segment_buffer *segbuf;
1618 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1619 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1622 nilfs_segbuf_fill_in_segsum(segbuf);
1627 static void nilfs_begin_folio_io(struct folio *folio)
1629 if (!folio || folio_test_writeback(folio))
1631 * For split b-tree node pages, this function may be called
1632 * twice. We ignore the 2nd or later calls by this check.
1637 folio_clear_dirty_for_io(folio);
1638 folio_start_writeback(folio);
1639 folio_unlock(folio);
1642 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1644 struct nilfs_segment_buffer *segbuf;
1645 struct folio *bd_folio = NULL, *fs_folio = NULL;
1647 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1648 struct buffer_head *bh;
1650 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1652 if (bh->b_folio != bd_folio) {
1654 folio_lock(bd_folio);
1655 folio_clear_dirty_for_io(bd_folio);
1656 folio_start_writeback(bd_folio);
1657 folio_unlock(bd_folio);
1659 bd_folio = bh->b_folio;
1663 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1665 if (bh == segbuf->sb_super_root) {
1666 if (bh->b_folio != bd_folio) {
1667 folio_lock(bd_folio);
1668 folio_clear_dirty_for_io(bd_folio);
1669 folio_start_writeback(bd_folio);
1670 folio_unlock(bd_folio);
1671 bd_folio = bh->b_folio;
1675 set_buffer_async_write(bh);
1676 if (bh->b_folio != fs_folio) {
1677 nilfs_begin_folio_io(fs_folio);
1678 fs_folio = bh->b_folio;
1683 folio_lock(bd_folio);
1684 folio_clear_dirty_for_io(bd_folio);
1685 folio_start_writeback(bd_folio);
1686 folio_unlock(bd_folio);
1688 nilfs_begin_folio_io(fs_folio);
1691 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1692 struct the_nilfs *nilfs)
1696 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1697 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1701 static void nilfs_end_folio_io(struct folio *folio, int err)
1706 if (buffer_nilfs_node(folio_buffers(folio)) &&
1707 !folio_test_writeback(folio)) {
1709 * For b-tree node pages, this function may be called twice
1710 * or more because they might be split in a segment.
1712 if (folio_test_dirty(folio)) {
1714 * For pages holding split b-tree node buffers, dirty
1715 * flag on the buffers may be cleared discretely.
1716 * In that case, the page is once redirtied for
1717 * remaining buffers, and it must be cancelled if
1718 * all the buffers get cleaned later.
1721 if (nilfs_folio_buffers_clean(folio))
1722 __nilfs_clear_folio_dirty(folio);
1723 folio_unlock(folio);
1729 if (!nilfs_folio_buffers_clean(folio))
1730 filemap_dirty_folio(folio->mapping, folio);
1731 folio_clear_error(folio);
1733 filemap_dirty_folio(folio->mapping, folio);
1734 folio_set_error(folio);
1737 folio_end_writeback(folio);
1740 static void nilfs_abort_logs(struct list_head *logs, int err)
1742 struct nilfs_segment_buffer *segbuf;
1743 struct folio *bd_folio = NULL, *fs_folio = NULL;
1744 struct buffer_head *bh;
1746 if (list_empty(logs))
1749 list_for_each_entry(segbuf, logs, sb_list) {
1750 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1752 clear_buffer_uptodate(bh);
1753 if (bh->b_folio != bd_folio) {
1755 folio_end_writeback(bd_folio);
1756 bd_folio = bh->b_folio;
1760 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1762 if (bh == segbuf->sb_super_root) {
1763 clear_buffer_uptodate(bh);
1764 if (bh->b_folio != bd_folio) {
1765 folio_end_writeback(bd_folio);
1766 bd_folio = bh->b_folio;
1770 clear_buffer_async_write(bh);
1771 if (bh->b_folio != fs_folio) {
1772 nilfs_end_folio_io(fs_folio, err);
1773 fs_folio = bh->b_folio;
1778 folio_end_writeback(bd_folio);
1780 nilfs_end_folio_io(fs_folio, err);
1783 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1784 struct the_nilfs *nilfs, int err)
1789 list_splice_tail_init(&sci->sc_write_logs, &logs);
1790 ret = nilfs_wait_on_logs(&logs);
1791 nilfs_abort_logs(&logs, ret ? : err);
1793 list_splice_tail_init(&sci->sc_segbufs, &logs);
1794 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1795 nilfs_free_incomplete_logs(&logs, nilfs);
1797 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1798 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1802 WARN_ON(ret); /* do not happen */
1805 nilfs_destroy_logs(&logs);
1808 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1809 struct nilfs_segment_buffer *segbuf)
1811 nilfs->ns_segnum = segbuf->sb_segnum;
1812 nilfs->ns_nextnum = segbuf->sb_nextnum;
1813 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1814 + segbuf->sb_sum.nblocks;
1815 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1816 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1819 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1821 struct nilfs_segment_buffer *segbuf;
1822 struct folio *bd_folio = NULL, *fs_folio = NULL;
1823 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1824 int update_sr = false;
1826 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1827 struct buffer_head *bh;
1829 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1831 set_buffer_uptodate(bh);
1832 clear_buffer_dirty(bh);
1833 if (bh->b_folio != bd_folio) {
1835 folio_end_writeback(bd_folio);
1836 bd_folio = bh->b_folio;
1840 * We assume that the buffers which belong to the same folio
1841 * continue over the buffer list.
1842 * Under this assumption, the last BHs of folios is
1843 * identifiable by the discontinuity of bh->b_folio
1844 * (folio != fs_folio).
1846 * For B-tree node blocks, however, this assumption is not
1847 * guaranteed. The cleanup code of B-tree node folios needs
1850 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1852 const unsigned long set_bits = BIT(BH_Uptodate);
1853 const unsigned long clear_bits =
1854 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1855 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1856 BIT(BH_NILFS_Redirected));
1858 if (bh == segbuf->sb_super_root) {
1859 set_buffer_uptodate(bh);
1860 clear_buffer_dirty(bh);
1861 if (bh->b_folio != bd_folio) {
1862 folio_end_writeback(bd_folio);
1863 bd_folio = bh->b_folio;
1868 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1869 if (bh->b_folio != fs_folio) {
1870 nilfs_end_folio_io(fs_folio, 0);
1871 fs_folio = bh->b_folio;
1875 if (!nilfs_segbuf_simplex(segbuf)) {
1876 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1877 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1878 sci->sc_lseg_stime = jiffies;
1880 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1881 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1885 * Since folios may continue over multiple segment buffers,
1886 * end of the last folio must be checked outside of the loop.
1889 folio_end_writeback(bd_folio);
1891 nilfs_end_folio_io(fs_folio, 0);
1893 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1895 if (nilfs_doing_gc())
1896 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1898 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1900 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1902 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1903 nilfs_set_next_segment(nilfs, segbuf);
1906 nilfs->ns_flushed_device = 0;
1907 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1908 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1910 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1911 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1912 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1913 nilfs_segctor_clear_metadata_dirty(sci);
1915 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1918 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1922 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1924 nilfs_segctor_complete_write(sci);
1925 nilfs_destroy_logs(&sci->sc_write_logs);
1930 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1931 struct the_nilfs *nilfs)
1933 struct nilfs_inode_info *ii, *n;
1934 struct inode *ifile = sci->sc_root->ifile;
1936 spin_lock(&nilfs->ns_inode_lock);
1938 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1940 struct buffer_head *ibh;
1943 spin_unlock(&nilfs->ns_inode_lock);
1944 err = nilfs_ifile_get_inode_block(
1945 ifile, ii->vfs_inode.i_ino, &ibh);
1946 if (unlikely(err)) {
1947 nilfs_warn(sci->sc_super,
1948 "log writer: error %d getting inode block (ino=%lu)",
1949 err, ii->vfs_inode.i_ino);
1952 spin_lock(&nilfs->ns_inode_lock);
1953 if (likely(!ii->i_bh))
1960 // Always redirty the buffer to avoid race condition
1961 mark_buffer_dirty(ii->i_bh);
1962 nilfs_mdt_mark_dirty(ifile);
1964 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1965 set_bit(NILFS_I_BUSY, &ii->i_state);
1966 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1968 spin_unlock(&nilfs->ns_inode_lock);
1973 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1974 struct the_nilfs *nilfs)
1976 struct nilfs_inode_info *ii, *n;
1977 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1978 int defer_iput = false;
1980 spin_lock(&nilfs->ns_inode_lock);
1981 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1982 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1983 test_bit(NILFS_I_DIRTY, &ii->i_state))
1986 clear_bit(NILFS_I_BUSY, &ii->i_state);
1989 list_del_init(&ii->i_dirty);
1990 if (!ii->vfs_inode.i_nlink || during_mount) {
1992 * Defer calling iput() to avoid deadlocks if
1993 * i_nlink == 0 or mount is not yet finished.
1995 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
1998 spin_unlock(&nilfs->ns_inode_lock);
1999 iput(&ii->vfs_inode);
2000 spin_lock(&nilfs->ns_inode_lock);
2003 spin_unlock(&nilfs->ns_inode_lock);
2006 schedule_work(&sci->sc_iput_work);
2010 * Main procedure of segment constructor
2012 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2014 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2017 if (sb_rdonly(sci->sc_super))
2020 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2021 sci->sc_cno = nilfs->ns_cno;
2023 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2027 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2028 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2030 if (nilfs_segctor_clean(sci))
2034 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2036 err = nilfs_segctor_begin_construction(sci, nilfs);
2040 /* Update time stamp */
2041 sci->sc_seg_ctime = ktime_get_real_seconds();
2043 err = nilfs_segctor_collect(sci, nilfs, mode);
2047 /* Avoid empty segment */
2048 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2049 nilfs_segbuf_empty(sci->sc_curseg)) {
2050 nilfs_segctor_abort_construction(sci, nilfs, 1);
2054 err = nilfs_segctor_assign(sci, mode);
2058 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2059 nilfs_segctor_fill_in_file_bmap(sci);
2061 if (mode == SC_LSEG_SR &&
2062 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2063 err = nilfs_cpfile_finalize_checkpoint(
2064 nilfs->ns_cpfile, nilfs->ns_cno, sci->sc_root,
2065 sci->sc_nblk_inc + sci->sc_nblk_this_inc,
2067 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags));
2069 goto failed_to_write;
2071 nilfs_segctor_fill_in_super_root(sci, nilfs);
2073 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2075 /* Write partial segments */
2076 nilfs_segctor_prepare_write(sci);
2078 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2079 nilfs->ns_crc_seed);
2081 err = nilfs_segctor_write(sci, nilfs);
2083 goto failed_to_write;
2085 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2086 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2088 * At this point, we avoid double buffering
2089 * for blocksize < pagesize because page dirty
2090 * flag is turned off during write and dirty
2091 * buffers are not properly collected for
2092 * pages crossing over segments.
2094 err = nilfs_segctor_wait(sci);
2096 goto failed_to_write;
2098 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2101 nilfs_segctor_drop_written_files(sci, nilfs);
2105 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2106 nilfs_redirty_inodes(&sci->sc_dirty_files);
2109 if (nilfs_doing_gc())
2110 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2111 nilfs_segctor_abort_construction(sci, nilfs, err);
2116 * nilfs_segctor_start_timer - set timer of background write
2117 * @sci: nilfs_sc_info
2119 * If the timer has already been set, it ignores the new request.
2120 * This function MUST be called within a section locking the segment
2123 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2125 spin_lock(&sci->sc_state_lock);
2126 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2127 sci->sc_timer.expires = jiffies + sci->sc_interval;
2128 add_timer(&sci->sc_timer);
2129 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2131 spin_unlock(&sci->sc_state_lock);
2134 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2136 spin_lock(&sci->sc_state_lock);
2137 if (!(sci->sc_flush_request & BIT(bn))) {
2138 unsigned long prev_req = sci->sc_flush_request;
2140 sci->sc_flush_request |= BIT(bn);
2142 wake_up(&sci->sc_wait_daemon);
2144 spin_unlock(&sci->sc_state_lock);
2148 * nilfs_flush_segment - trigger a segment construction for resource control
2150 * @ino: inode number of the file to be flushed out.
2152 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2154 struct the_nilfs *nilfs = sb->s_fs_info;
2155 struct nilfs_sc_info *sci = nilfs->ns_writer;
2157 if (!sci || nilfs_doing_construction())
2159 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2160 /* assign bit 0 to data files */
2163 struct nilfs_segctor_wait_request {
2164 wait_queue_entry_t wq;
2170 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2172 struct nilfs_segctor_wait_request wait_req;
2175 spin_lock(&sci->sc_state_lock);
2176 init_wait(&wait_req.wq);
2178 atomic_set(&wait_req.done, 0);
2179 wait_req.seq = ++sci->sc_seq_request;
2180 spin_unlock(&sci->sc_state_lock);
2182 init_waitqueue_entry(&wait_req.wq, current);
2183 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2184 set_current_state(TASK_INTERRUPTIBLE);
2185 wake_up(&sci->sc_wait_daemon);
2188 if (atomic_read(&wait_req.done)) {
2192 if (!signal_pending(current)) {
2199 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2203 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2205 struct nilfs_segctor_wait_request *wrq, *n;
2206 unsigned long flags;
2208 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2209 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2210 if (!atomic_read(&wrq->done) &&
2211 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2213 atomic_set(&wrq->done, 1);
2215 if (atomic_read(&wrq->done)) {
2216 wrq->wq.func(&wrq->wq,
2217 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2221 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2225 * nilfs_construct_segment - construct a logical segment
2228 * Return Value: On success, 0 is returned. On errors, one of the following
2229 * negative error code is returned.
2231 * %-EROFS - Read only filesystem.
2235 * %-ENOSPC - No space left on device (only in a panic state).
2237 * %-ERESTARTSYS - Interrupted.
2239 * %-ENOMEM - Insufficient memory available.
2241 int nilfs_construct_segment(struct super_block *sb)
2243 struct the_nilfs *nilfs = sb->s_fs_info;
2244 struct nilfs_sc_info *sci = nilfs->ns_writer;
2245 struct nilfs_transaction_info *ti;
2247 if (sb_rdonly(sb) || unlikely(!sci))
2250 /* A call inside transactions causes a deadlock. */
2251 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2253 return nilfs_segctor_sync(sci);
2257 * nilfs_construct_dsync_segment - construct a data-only logical segment
2259 * @inode: inode whose data blocks should be written out
2260 * @start: start byte offset
2261 * @end: end byte offset (inclusive)
2263 * Return Value: On success, 0 is returned. On errors, one of the following
2264 * negative error code is returned.
2266 * %-EROFS - Read only filesystem.
2270 * %-ENOSPC - No space left on device (only in a panic state).
2272 * %-ERESTARTSYS - Interrupted.
2274 * %-ENOMEM - Insufficient memory available.
2276 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2277 loff_t start, loff_t end)
2279 struct the_nilfs *nilfs = sb->s_fs_info;
2280 struct nilfs_sc_info *sci = nilfs->ns_writer;
2281 struct nilfs_inode_info *ii;
2282 struct nilfs_transaction_info ti;
2285 if (sb_rdonly(sb) || unlikely(!sci))
2288 nilfs_transaction_lock(sb, &ti, 0);
2290 ii = NILFS_I(inode);
2291 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2292 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2293 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2294 nilfs_discontinued(nilfs)) {
2295 nilfs_transaction_unlock(sb);
2296 err = nilfs_segctor_sync(sci);
2300 spin_lock(&nilfs->ns_inode_lock);
2301 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2302 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2303 spin_unlock(&nilfs->ns_inode_lock);
2304 nilfs_transaction_unlock(sb);
2307 spin_unlock(&nilfs->ns_inode_lock);
2308 sci->sc_dsync_inode = ii;
2309 sci->sc_dsync_start = start;
2310 sci->sc_dsync_end = end;
2312 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2314 nilfs->ns_flushed_device = 0;
2316 nilfs_transaction_unlock(sb);
2320 #define FLUSH_FILE_BIT (0x1) /* data file only */
2321 #define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2324 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2325 * @sci: segment constructor object
2327 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2329 spin_lock(&sci->sc_state_lock);
2330 sci->sc_seq_accepted = sci->sc_seq_request;
2331 spin_unlock(&sci->sc_state_lock);
2332 del_timer_sync(&sci->sc_timer);
2336 * nilfs_segctor_notify - notify the result of request to caller threads
2337 * @sci: segment constructor object
2338 * @mode: mode of log forming
2339 * @err: error code to be notified
2341 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2343 /* Clear requests (even when the construction failed) */
2344 spin_lock(&sci->sc_state_lock);
2346 if (mode == SC_LSEG_SR) {
2347 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2348 sci->sc_seq_done = sci->sc_seq_accepted;
2349 nilfs_segctor_wakeup(sci, err);
2350 sci->sc_flush_request = 0;
2352 if (mode == SC_FLUSH_FILE)
2353 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2354 else if (mode == SC_FLUSH_DAT)
2355 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2357 /* re-enable timer if checkpoint creation was not done */
2358 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2359 time_before(jiffies, sci->sc_timer.expires))
2360 add_timer(&sci->sc_timer);
2362 spin_unlock(&sci->sc_state_lock);
2366 * nilfs_segctor_construct - form logs and write them to disk
2367 * @sci: segment constructor object
2368 * @mode: mode of log forming
2370 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2372 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2373 struct nilfs_super_block **sbp;
2376 nilfs_segctor_accept(sci);
2378 if (nilfs_discontinued(nilfs))
2380 if (!nilfs_segctor_confirm(sci))
2381 err = nilfs_segctor_do_construct(sci, mode);
2384 if (mode != SC_FLUSH_DAT)
2385 atomic_set(&nilfs->ns_ndirtyblks, 0);
2386 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2387 nilfs_discontinued(nilfs)) {
2388 down_write(&nilfs->ns_sem);
2390 sbp = nilfs_prepare_super(sci->sc_super,
2391 nilfs_sb_will_flip(nilfs));
2393 nilfs_set_log_cursor(sbp[0], nilfs);
2394 err = nilfs_commit_super(sci->sc_super,
2397 up_write(&nilfs->ns_sem);
2401 nilfs_segctor_notify(sci, mode, err);
2405 static void nilfs_construction_timeout(struct timer_list *t)
2407 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2409 wake_up_process(sci->sc_timer_task);
2413 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2415 struct nilfs_inode_info *ii, *n;
2417 list_for_each_entry_safe(ii, n, head, i_dirty) {
2418 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2420 list_del_init(&ii->i_dirty);
2421 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2422 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2423 iput(&ii->vfs_inode);
2427 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2430 struct the_nilfs *nilfs = sb->s_fs_info;
2431 struct nilfs_sc_info *sci = nilfs->ns_writer;
2432 struct nilfs_transaction_info ti;
2438 nilfs_transaction_lock(sb, &ti, 1);
2440 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2444 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2445 if (unlikely(err)) {
2446 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2450 sci->sc_freesegs = kbufs[4];
2451 sci->sc_nfreesegs = argv[4].v_nmembs;
2452 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2455 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2456 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2461 nilfs_warn(sb, "error %d cleaning segments", err);
2462 set_current_state(TASK_INTERRUPTIBLE);
2463 schedule_timeout(sci->sc_interval);
2465 if (nilfs_test_opt(nilfs, DISCARD)) {
2466 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2470 "error %d on discard request, turning discards off for the device",
2472 nilfs_clear_opt(nilfs, DISCARD);
2477 sci->sc_freesegs = NULL;
2478 sci->sc_nfreesegs = 0;
2479 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2480 nilfs_transaction_unlock(sb);
2484 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2486 struct nilfs_transaction_info ti;
2488 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2489 nilfs_segctor_construct(sci, mode);
2492 * Unclosed segment should be retried. We do this using sc_timer.
2493 * Timeout of sc_timer will invoke complete construction which leads
2494 * to close the current logical segment.
2496 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2497 nilfs_segctor_start_timer(sci);
2499 nilfs_transaction_unlock(sci->sc_super);
2502 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2506 spin_lock(&sci->sc_state_lock);
2507 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2508 SC_FLUSH_DAT : SC_FLUSH_FILE;
2509 spin_unlock(&sci->sc_state_lock);
2512 nilfs_segctor_do_construct(sci, mode);
2514 spin_lock(&sci->sc_state_lock);
2515 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2516 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2517 spin_unlock(&sci->sc_state_lock);
2519 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2522 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2524 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2525 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2526 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2527 return SC_FLUSH_FILE;
2528 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2529 return SC_FLUSH_DAT;
2535 * nilfs_segctor_thread - main loop of the segment constructor thread.
2536 * @arg: pointer to a struct nilfs_sc_info.
2538 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2539 * to execute segment constructions.
2541 static int nilfs_segctor_thread(void *arg)
2543 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2544 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2547 sci->sc_timer_task = current;
2550 sci->sc_task = current;
2551 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2552 nilfs_info(sci->sc_super,
2553 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2554 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2557 spin_lock(&sci->sc_state_lock);
2562 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2565 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2567 else if (sci->sc_flush_request)
2568 mode = nilfs_segctor_flush_mode(sci);
2572 spin_unlock(&sci->sc_state_lock);
2573 nilfs_segctor_thread_construct(sci, mode);
2574 spin_lock(&sci->sc_state_lock);
2579 if (freezing(current)) {
2580 spin_unlock(&sci->sc_state_lock);
2582 spin_lock(&sci->sc_state_lock);
2585 int should_sleep = 1;
2587 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2588 TASK_INTERRUPTIBLE);
2590 if (sci->sc_seq_request != sci->sc_seq_done)
2592 else if (sci->sc_flush_request)
2594 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2595 should_sleep = time_before(jiffies,
2596 sci->sc_timer.expires);
2599 spin_unlock(&sci->sc_state_lock);
2601 spin_lock(&sci->sc_state_lock);
2603 finish_wait(&sci->sc_wait_daemon, &wait);
2604 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2605 time_after_eq(jiffies, sci->sc_timer.expires));
2607 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2608 set_nilfs_discontinued(nilfs);
2614 sci->sc_task = NULL;
2615 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2616 spin_unlock(&sci->sc_state_lock);
2620 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2622 struct task_struct *t;
2624 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2626 int err = PTR_ERR(t);
2628 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2632 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2636 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2637 __acquires(&sci->sc_state_lock)
2638 __releases(&sci->sc_state_lock)
2640 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2642 while (sci->sc_task) {
2643 wake_up(&sci->sc_wait_daemon);
2644 spin_unlock(&sci->sc_state_lock);
2645 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2646 spin_lock(&sci->sc_state_lock);
2651 * Setup & clean-up functions
2653 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2654 struct nilfs_root *root)
2656 struct the_nilfs *nilfs = sb->s_fs_info;
2657 struct nilfs_sc_info *sci;
2659 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2665 nilfs_get_root(root);
2666 sci->sc_root = root;
2668 init_waitqueue_head(&sci->sc_wait_request);
2669 init_waitqueue_head(&sci->sc_wait_daemon);
2670 init_waitqueue_head(&sci->sc_wait_task);
2671 spin_lock_init(&sci->sc_state_lock);
2672 INIT_LIST_HEAD(&sci->sc_dirty_files);
2673 INIT_LIST_HEAD(&sci->sc_segbufs);
2674 INIT_LIST_HEAD(&sci->sc_write_logs);
2675 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2676 INIT_LIST_HEAD(&sci->sc_iput_queue);
2677 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2678 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2680 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2681 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2682 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2684 if (nilfs->ns_interval)
2685 sci->sc_interval = HZ * nilfs->ns_interval;
2686 if (nilfs->ns_watermark)
2687 sci->sc_watermark = nilfs->ns_watermark;
2691 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2693 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2696 * The segctord thread was stopped and its timer was removed.
2697 * But some tasks remain.
2700 struct nilfs_transaction_info ti;
2702 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2703 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2704 nilfs_transaction_unlock(sci->sc_super);
2706 flush_work(&sci->sc_iput_work);
2708 } while (ret && ret != -EROFS && retrycount-- > 0);
2712 * nilfs_segctor_destroy - destroy the segment constructor.
2713 * @sci: nilfs_sc_info
2715 * nilfs_segctor_destroy() kills the segctord thread and frees
2716 * the nilfs_sc_info struct.
2717 * Caller must hold the segment semaphore.
2719 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2721 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2724 up_write(&nilfs->ns_segctor_sem);
2726 spin_lock(&sci->sc_state_lock);
2727 nilfs_segctor_kill_thread(sci);
2728 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2729 || sci->sc_seq_request != sci->sc_seq_done);
2730 spin_unlock(&sci->sc_state_lock);
2732 if (flush_work(&sci->sc_iput_work))
2735 if (flag || !nilfs_segctor_confirm(sci))
2736 nilfs_segctor_write_out(sci);
2738 if (!list_empty(&sci->sc_dirty_files)) {
2739 nilfs_warn(sci->sc_super,
2740 "disposed unprocessed dirty file(s) when stopping log writer");
2741 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2744 if (!list_empty(&sci->sc_iput_queue)) {
2745 nilfs_warn(sci->sc_super,
2746 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2747 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2750 WARN_ON(!list_empty(&sci->sc_segbufs));
2751 WARN_ON(!list_empty(&sci->sc_write_logs));
2753 nilfs_put_root(sci->sc_root);
2755 down_write(&nilfs->ns_segctor_sem);
2757 timer_shutdown_sync(&sci->sc_timer);
2762 * nilfs_attach_log_writer - attach log writer
2763 * @sb: super block instance
2764 * @root: root object of the current filesystem tree
2766 * This allocates a log writer object, initializes it, and starts the
2769 * Return Value: On success, 0 is returned. On error, one of the following
2770 * negative error code is returned.
2772 * %-ENOMEM - Insufficient memory available.
2774 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2776 struct the_nilfs *nilfs = sb->s_fs_info;
2779 if (nilfs->ns_writer) {
2781 * This happens if the filesystem is made read-only by
2782 * __nilfs_error or nilfs_remount and then remounted
2783 * read/write. In these cases, reuse the existing
2789 nilfs->ns_writer = nilfs_segctor_new(sb, root);
2790 if (!nilfs->ns_writer)
2793 inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2795 err = nilfs_segctor_start_thread(nilfs->ns_writer);
2797 nilfs_detach_log_writer(sb);
2803 * nilfs_detach_log_writer - destroy log writer
2804 * @sb: super block instance
2806 * This kills log writer daemon, frees the log writer object, and
2807 * destroys list of dirty files.
2809 void nilfs_detach_log_writer(struct super_block *sb)
2811 struct the_nilfs *nilfs = sb->s_fs_info;
2812 LIST_HEAD(garbage_list);
2814 down_write(&nilfs->ns_segctor_sem);
2815 if (nilfs->ns_writer) {
2816 nilfs_segctor_destroy(nilfs->ns_writer);
2817 nilfs->ns_writer = NULL;
2819 set_nilfs_purging(nilfs);
2821 /* Force to free the list of dirty files */
2822 spin_lock(&nilfs->ns_inode_lock);
2823 if (!list_empty(&nilfs->ns_dirty_files)) {
2824 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2826 "disposed unprocessed dirty file(s) when detaching log writer");
2828 spin_unlock(&nilfs->ns_inode_lock);
2829 up_write(&nilfs->ns_segctor_sem);
2831 nilfs_dispose_list(nilfs, &garbage_list, 1);
2832 clear_nilfs_purging(nilfs);