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 int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
885 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
886 struct buffer_head *bh_cp;
887 struct nilfs_checkpoint *raw_cp;
890 /* XXX: this interface will be changed */
891 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
895 * The following code is duplicated with cpfile. But, it is
896 * needed to collect the checkpoint even if it was not newly
899 mark_buffer_dirty(bh_cp);
900 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
901 nilfs_cpfile_put_checkpoint(
902 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
903 } else if (err == -EINVAL || err == -ENOENT) {
904 nilfs_error(sci->sc_super,
905 "checkpoint creation failed due to metadata corruption.");
911 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
913 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
914 struct buffer_head *bh_cp;
915 struct nilfs_checkpoint *raw_cp;
918 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
921 if (err == -EINVAL || err == -ENOENT) {
922 nilfs_error(sci->sc_super,
923 "checkpoint finalization failed due to metadata corruption.");
928 raw_cp->cp_snapshot_list.ssl_next = 0;
929 raw_cp->cp_snapshot_list.ssl_prev = 0;
930 raw_cp->cp_inodes_count =
931 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
932 raw_cp->cp_blocks_count =
933 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
934 raw_cp->cp_nblk_inc =
935 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
936 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
937 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
939 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
940 nilfs_checkpoint_clear_minor(raw_cp);
942 nilfs_checkpoint_set_minor(raw_cp);
944 nilfs_write_inode_common(sci->sc_root->ifile,
945 &raw_cp->cp_ifile_inode, 1);
946 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
953 static void nilfs_fill_in_file_bmap(struct inode *ifile,
954 struct nilfs_inode_info *ii)
957 struct buffer_head *ibh;
958 struct nilfs_inode *raw_inode;
960 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
963 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
965 nilfs_bmap_write(ii->i_bmap, raw_inode);
966 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
970 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
972 struct nilfs_inode_info *ii;
974 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
975 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
976 set_bit(NILFS_I_COLLECTED, &ii->i_state);
980 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
981 struct the_nilfs *nilfs)
983 struct buffer_head *bh_sr;
984 struct nilfs_super_root *raw_sr;
985 unsigned int isz, srsz;
987 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
990 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
991 isz = nilfs->ns_inode_size;
992 srsz = NILFS_SR_BYTES(isz);
994 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
995 raw_sr->sr_bytes = cpu_to_le16(srsz);
996 raw_sr->sr_nongc_ctime
997 = cpu_to_le64(nilfs_doing_gc() ?
998 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
999 raw_sr->sr_flags = 0;
1001 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
1002 NILFS_SR_DAT_OFFSET(isz), 1);
1003 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
1004 NILFS_SR_CPFILE_OFFSET(isz), 1);
1005 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
1006 NILFS_SR_SUFILE_OFFSET(isz), 1);
1007 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1008 set_buffer_uptodate(bh_sr);
1009 unlock_buffer(bh_sr);
1012 static void nilfs_redirty_inodes(struct list_head *head)
1014 struct nilfs_inode_info *ii;
1016 list_for_each_entry(ii, head, i_dirty) {
1017 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1018 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1022 static void nilfs_drop_collected_inodes(struct list_head *head)
1024 struct nilfs_inode_info *ii;
1026 list_for_each_entry(ii, head, i_dirty) {
1027 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1030 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1031 set_bit(NILFS_I_UPDATED, &ii->i_state);
1035 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1036 struct inode *inode,
1037 struct list_head *listp,
1038 int (*collect)(struct nilfs_sc_info *,
1039 struct buffer_head *,
1042 struct buffer_head *bh, *n;
1046 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1047 list_del_init(&bh->b_assoc_buffers);
1048 err = collect(sci, bh, inode);
1051 goto dispose_buffers;
1057 while (!list_empty(listp)) {
1058 bh = list_first_entry(listp, struct buffer_head,
1060 list_del_init(&bh->b_assoc_buffers);
1066 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1068 /* Remaining number of blocks within segment buffer */
1069 return sci->sc_segbuf_nblocks -
1070 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1073 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1074 struct inode *inode,
1075 const struct nilfs_sc_operations *sc_ops)
1077 LIST_HEAD(data_buffers);
1078 LIST_HEAD(node_buffers);
1081 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1082 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1084 n = nilfs_lookup_dirty_data_buffers(
1085 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1087 err = nilfs_segctor_apply_buffers(
1088 sci, inode, &data_buffers,
1089 sc_ops->collect_data);
1090 BUG_ON(!err); /* always receive -E2BIG or true error */
1094 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1096 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1097 err = nilfs_segctor_apply_buffers(
1098 sci, inode, &data_buffers, sc_ops->collect_data);
1099 if (unlikely(err)) {
1100 /* dispose node list */
1101 nilfs_segctor_apply_buffers(
1102 sci, inode, &node_buffers, NULL);
1105 sci->sc_stage.flags |= NILFS_CF_NODE;
1108 err = nilfs_segctor_apply_buffers(
1109 sci, inode, &node_buffers, sc_ops->collect_node);
1113 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1114 err = nilfs_segctor_apply_buffers(
1115 sci, inode, &node_buffers, sc_ops->collect_bmap);
1119 nilfs_segctor_end_finfo(sci, inode);
1120 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1126 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1127 struct inode *inode)
1129 LIST_HEAD(data_buffers);
1130 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1133 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1134 sci->sc_dsync_start,
1137 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1138 nilfs_collect_file_data);
1140 nilfs_segctor_end_finfo(sci, inode);
1142 /* always receive -E2BIG or true error if n > rest */
1147 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1149 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1150 struct list_head *head;
1151 struct nilfs_inode_info *ii;
1155 switch (nilfs_sc_cstage_get(sci)) {
1158 sci->sc_stage.flags = 0;
1160 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1161 sci->sc_nblk_inc = 0;
1162 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1163 if (mode == SC_LSEG_DSYNC) {
1164 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1169 sci->sc_stage.dirty_file_ptr = NULL;
1170 sci->sc_stage.gc_inode_ptr = NULL;
1171 if (mode == SC_FLUSH_DAT) {
1172 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1175 nilfs_sc_cstage_inc(sci);
1178 if (nilfs_doing_gc()) {
1179 head = &sci->sc_gc_inodes;
1180 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1182 list_for_each_entry_continue(ii, head, i_dirty) {
1183 err = nilfs_segctor_scan_file(
1184 sci, &ii->vfs_inode,
1185 &nilfs_sc_file_ops);
1186 if (unlikely(err)) {
1187 sci->sc_stage.gc_inode_ptr = list_entry(
1189 struct nilfs_inode_info,
1193 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1195 sci->sc_stage.gc_inode_ptr = NULL;
1197 nilfs_sc_cstage_inc(sci);
1200 head = &sci->sc_dirty_files;
1201 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1203 list_for_each_entry_continue(ii, head, i_dirty) {
1204 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1206 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1207 &nilfs_sc_file_ops);
1208 if (unlikely(err)) {
1209 sci->sc_stage.dirty_file_ptr =
1210 list_entry(ii->i_dirty.prev,
1211 struct nilfs_inode_info,
1215 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1216 /* XXX: required ? */
1218 sci->sc_stage.dirty_file_ptr = NULL;
1219 if (mode == SC_FLUSH_FILE) {
1220 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1223 nilfs_sc_cstage_inc(sci);
1224 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1226 case NILFS_ST_IFILE:
1227 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1228 &nilfs_sc_file_ops);
1231 nilfs_sc_cstage_inc(sci);
1232 /* Creating a checkpoint */
1233 err = nilfs_segctor_create_checkpoint(sci);
1237 case NILFS_ST_CPFILE:
1238 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1239 &nilfs_sc_file_ops);
1242 nilfs_sc_cstage_inc(sci);
1244 case NILFS_ST_SUFILE:
1245 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1246 sci->sc_nfreesegs, &ndone);
1247 if (unlikely(err)) {
1248 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1249 sci->sc_freesegs, ndone,
1253 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1255 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1256 &nilfs_sc_file_ops);
1259 nilfs_sc_cstage_inc(sci);
1263 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1267 if (mode == SC_FLUSH_DAT) {
1268 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1271 nilfs_sc_cstage_inc(sci);
1274 if (mode == SC_LSEG_SR) {
1275 /* Appending a super root */
1276 err = nilfs_segctor_add_super_root(sci);
1280 /* End of a logical segment */
1281 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1282 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1284 case NILFS_ST_DSYNC:
1286 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1287 ii = sci->sc_dsync_inode;
1288 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1291 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1294 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1295 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1308 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1309 * @sci: nilfs_sc_info
1310 * @nilfs: nilfs object
1312 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1313 struct the_nilfs *nilfs)
1315 struct nilfs_segment_buffer *segbuf, *prev;
1319 segbuf = nilfs_segbuf_new(sci->sc_super);
1320 if (unlikely(!segbuf))
1323 if (list_empty(&sci->sc_write_logs)) {
1324 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1325 nilfs->ns_pseg_offset, nilfs);
1326 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1327 nilfs_shift_to_next_segment(nilfs);
1328 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1331 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1332 nextnum = nilfs->ns_nextnum;
1334 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1335 /* Start from the head of a new full segment */
1339 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1340 nilfs_segbuf_map_cont(segbuf, prev);
1341 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1342 nextnum = prev->sb_nextnum;
1344 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1345 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1346 segbuf->sb_sum.seg_seq++;
1351 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1356 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1360 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1362 BUG_ON(!list_empty(&sci->sc_segbufs));
1363 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1364 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1368 nilfs_segbuf_free(segbuf);
1372 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1373 struct the_nilfs *nilfs, int nadd)
1375 struct nilfs_segment_buffer *segbuf, *prev;
1376 struct inode *sufile = nilfs->ns_sufile;
1381 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1383 * Since the segment specified with nextnum might be allocated during
1384 * the previous construction, the buffer including its segusage may
1385 * not be dirty. The following call ensures that the buffer is dirty
1386 * and will pin the buffer on memory until the sufile is written.
1388 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1392 for (i = 0; i < nadd; i++) {
1393 /* extend segment info */
1395 segbuf = nilfs_segbuf_new(sci->sc_super);
1396 if (unlikely(!segbuf))
1399 /* map this buffer to region of segment on-disk */
1400 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1401 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1403 /* allocate the next next full segment */
1404 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1408 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1409 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1411 list_add_tail(&segbuf->sb_list, &list);
1414 list_splice_tail(&list, &sci->sc_segbufs);
1418 nilfs_segbuf_free(segbuf);
1420 list_for_each_entry(segbuf, &list, sb_list) {
1421 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1422 WARN_ON(ret); /* never fails */
1424 nilfs_destroy_logs(&list);
1428 static void nilfs_free_incomplete_logs(struct list_head *logs,
1429 struct the_nilfs *nilfs)
1431 struct nilfs_segment_buffer *segbuf, *prev;
1432 struct inode *sufile = nilfs->ns_sufile;
1435 segbuf = NILFS_FIRST_SEGBUF(logs);
1436 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1437 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1438 WARN_ON(ret); /* never fails */
1440 if (atomic_read(&segbuf->sb_err)) {
1441 /* Case 1: The first segment failed */
1442 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1444 * Case 1a: Partial segment appended into an existing
1447 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1448 segbuf->sb_fseg_end);
1449 else /* Case 1b: New full segment */
1450 set_nilfs_discontinued(nilfs);
1454 list_for_each_entry_continue(segbuf, logs, sb_list) {
1455 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1456 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1457 WARN_ON(ret); /* never fails */
1459 if (atomic_read(&segbuf->sb_err) &&
1460 segbuf->sb_segnum != nilfs->ns_nextnum)
1461 /* Case 2: extended segment (!= next) failed */
1462 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1467 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1468 struct inode *sufile)
1470 struct nilfs_segment_buffer *segbuf;
1471 unsigned long live_blocks;
1474 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1475 live_blocks = segbuf->sb_sum.nblocks +
1476 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1477 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1480 WARN_ON(ret); /* always succeed because the segusage is dirty */
1484 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1486 struct nilfs_segment_buffer *segbuf;
1489 segbuf = NILFS_FIRST_SEGBUF(logs);
1490 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1491 segbuf->sb_pseg_start -
1492 segbuf->sb_fseg_start, 0);
1493 WARN_ON(ret); /* always succeed because the segusage is dirty */
1495 list_for_each_entry_continue(segbuf, logs, sb_list) {
1496 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1498 WARN_ON(ret); /* always succeed */
1502 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1503 struct nilfs_segment_buffer *last,
1504 struct inode *sufile)
1506 struct nilfs_segment_buffer *segbuf = last;
1509 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1510 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1511 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1514 nilfs_truncate_logs(&sci->sc_segbufs, last);
1518 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1519 struct the_nilfs *nilfs, int mode)
1521 struct nilfs_cstage prev_stage = sci->sc_stage;
1524 /* Collection retry loop */
1526 sci->sc_nblk_this_inc = 0;
1527 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1529 err = nilfs_segctor_reset_segment_buffer(sci);
1533 err = nilfs_segctor_collect_blocks(sci, mode);
1534 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1538 if (unlikely(err != -E2BIG))
1541 /* The current segment is filled up */
1542 if (mode != SC_LSEG_SR ||
1543 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1546 nilfs_clear_logs(&sci->sc_segbufs);
1548 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1549 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1553 WARN_ON(err); /* do not happen */
1554 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1557 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1561 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1562 sci->sc_stage = prev_stage;
1564 nilfs_segctor_zeropad_segsum(sci);
1565 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1572 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1573 struct buffer_head *new_bh)
1575 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1577 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1578 /* The caller must release old_bh */
1582 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1583 struct nilfs_segment_buffer *segbuf,
1586 struct inode *inode = NULL;
1588 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1589 unsigned long nblocks = 0, ndatablk = 0;
1590 const struct nilfs_sc_operations *sc_op = NULL;
1591 struct nilfs_segsum_pointer ssp;
1592 struct nilfs_finfo *finfo = NULL;
1593 union nilfs_binfo binfo;
1594 struct buffer_head *bh, *bh_org;
1601 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1602 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1603 ssp.offset = sizeof(struct nilfs_segment_summary);
1605 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1606 if (bh == segbuf->sb_super_root)
1609 finfo = nilfs_segctor_map_segsum_entry(
1610 sci, &ssp, sizeof(*finfo));
1611 ino = le64_to_cpu(finfo->fi_ino);
1612 nblocks = le32_to_cpu(finfo->fi_nblocks);
1613 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1615 inode = bh->b_folio->mapping->host;
1617 if (mode == SC_LSEG_DSYNC)
1618 sc_op = &nilfs_sc_dsync_ops;
1619 else if (ino == NILFS_DAT_INO)
1620 sc_op = &nilfs_sc_dat_ops;
1621 else /* file blocks */
1622 sc_op = &nilfs_sc_file_ops;
1626 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1629 nilfs_list_replace_buffer(bh_org, bh);
1635 sc_op->write_data_binfo(sci, &ssp, &binfo);
1637 sc_op->write_node_binfo(sci, &ssp, &binfo);
1640 if (--nblocks == 0) {
1644 } else if (ndatablk > 0)
1654 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1656 struct nilfs_segment_buffer *segbuf;
1659 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1660 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1663 nilfs_segbuf_fill_in_segsum(segbuf);
1668 static void nilfs_begin_page_io(struct page *page)
1670 if (!page || PageWriteback(page))
1672 * For split b-tree node pages, this function may be called
1673 * twice. We ignore the 2nd or later calls by this check.
1678 clear_page_dirty_for_io(page);
1679 set_page_writeback(page);
1683 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1685 struct nilfs_segment_buffer *segbuf;
1686 struct page *bd_page = NULL, *fs_page = NULL;
1688 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1689 struct buffer_head *bh;
1691 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1693 if (bh->b_page != bd_page) {
1696 clear_page_dirty_for_io(bd_page);
1697 set_page_writeback(bd_page);
1698 unlock_page(bd_page);
1700 bd_page = bh->b_page;
1704 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1706 set_buffer_async_write(bh);
1707 if (bh == segbuf->sb_super_root) {
1708 if (bh->b_page != bd_page) {
1710 clear_page_dirty_for_io(bd_page);
1711 set_page_writeback(bd_page);
1712 unlock_page(bd_page);
1713 bd_page = bh->b_page;
1717 if (bh->b_page != fs_page) {
1718 nilfs_begin_page_io(fs_page);
1719 fs_page = bh->b_page;
1725 clear_page_dirty_for_io(bd_page);
1726 set_page_writeback(bd_page);
1727 unlock_page(bd_page);
1729 nilfs_begin_page_io(fs_page);
1732 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1733 struct the_nilfs *nilfs)
1737 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1738 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1742 static void nilfs_end_page_io(struct page *page, int err)
1747 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1749 * For b-tree node pages, this function may be called twice
1750 * or more because they might be split in a segment.
1752 if (PageDirty(page)) {
1754 * For pages holding split b-tree node buffers, dirty
1755 * flag on the buffers may be cleared discretely.
1756 * In that case, the page is once redirtied for
1757 * remaining buffers, and it must be cancelled if
1758 * all the buffers get cleaned later.
1761 if (nilfs_page_buffers_clean(page))
1762 __nilfs_clear_page_dirty(page);
1769 if (!nilfs_page_buffers_clean(page))
1770 __set_page_dirty_nobuffers(page);
1771 ClearPageError(page);
1773 __set_page_dirty_nobuffers(page);
1777 end_page_writeback(page);
1780 static void nilfs_abort_logs(struct list_head *logs, int err)
1782 struct nilfs_segment_buffer *segbuf;
1783 struct page *bd_page = NULL, *fs_page = NULL;
1784 struct buffer_head *bh;
1786 if (list_empty(logs))
1789 list_for_each_entry(segbuf, logs, sb_list) {
1790 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1792 clear_buffer_uptodate(bh);
1793 if (bh->b_page != bd_page) {
1795 end_page_writeback(bd_page);
1796 bd_page = bh->b_page;
1800 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1802 clear_buffer_async_write(bh);
1803 if (bh == segbuf->sb_super_root) {
1804 clear_buffer_uptodate(bh);
1805 if (bh->b_page != bd_page) {
1806 end_page_writeback(bd_page);
1807 bd_page = bh->b_page;
1811 if (bh->b_page != fs_page) {
1812 nilfs_end_page_io(fs_page, err);
1813 fs_page = bh->b_page;
1818 end_page_writeback(bd_page);
1820 nilfs_end_page_io(fs_page, err);
1823 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1824 struct the_nilfs *nilfs, int err)
1829 list_splice_tail_init(&sci->sc_write_logs, &logs);
1830 ret = nilfs_wait_on_logs(&logs);
1831 nilfs_abort_logs(&logs, ret ? : err);
1833 list_splice_tail_init(&sci->sc_segbufs, &logs);
1834 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1835 nilfs_free_incomplete_logs(&logs, nilfs);
1837 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1838 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1842 WARN_ON(ret); /* do not happen */
1845 nilfs_destroy_logs(&logs);
1848 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1849 struct nilfs_segment_buffer *segbuf)
1851 nilfs->ns_segnum = segbuf->sb_segnum;
1852 nilfs->ns_nextnum = segbuf->sb_nextnum;
1853 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1854 + segbuf->sb_sum.nblocks;
1855 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1856 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1859 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1861 struct nilfs_segment_buffer *segbuf;
1862 struct page *bd_page = NULL, *fs_page = NULL;
1863 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1864 int update_sr = false;
1866 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1867 struct buffer_head *bh;
1869 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1871 set_buffer_uptodate(bh);
1872 clear_buffer_dirty(bh);
1873 if (bh->b_page != bd_page) {
1875 end_page_writeback(bd_page);
1876 bd_page = bh->b_page;
1880 * We assume that the buffers which belong to the same page
1881 * continue over the buffer list.
1882 * Under this assumption, the last BHs of pages is
1883 * identifiable by the discontinuity of bh->b_page
1884 * (page != fs_page).
1886 * For B-tree node blocks, however, this assumption is not
1887 * guaranteed. The cleanup code of B-tree node pages needs
1890 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1892 const unsigned long set_bits = BIT(BH_Uptodate);
1893 const unsigned long clear_bits =
1894 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1895 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1896 BIT(BH_NILFS_Redirected));
1898 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1899 if (bh == segbuf->sb_super_root) {
1900 if (bh->b_page != bd_page) {
1901 end_page_writeback(bd_page);
1902 bd_page = bh->b_page;
1907 if (bh->b_page != fs_page) {
1908 nilfs_end_page_io(fs_page, 0);
1909 fs_page = bh->b_page;
1913 if (!nilfs_segbuf_simplex(segbuf)) {
1914 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1915 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1916 sci->sc_lseg_stime = jiffies;
1918 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1919 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1923 * Since pages may continue over multiple segment buffers,
1924 * end of the last page must be checked outside of the loop.
1927 end_page_writeback(bd_page);
1929 nilfs_end_page_io(fs_page, 0);
1931 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1933 if (nilfs_doing_gc())
1934 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1936 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1938 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1940 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1941 nilfs_set_next_segment(nilfs, segbuf);
1944 nilfs->ns_flushed_device = 0;
1945 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1946 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1948 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1949 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1950 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1951 nilfs_segctor_clear_metadata_dirty(sci);
1953 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1956 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1960 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1962 nilfs_segctor_complete_write(sci);
1963 nilfs_destroy_logs(&sci->sc_write_logs);
1968 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1969 struct the_nilfs *nilfs)
1971 struct nilfs_inode_info *ii, *n;
1972 struct inode *ifile = sci->sc_root->ifile;
1974 spin_lock(&nilfs->ns_inode_lock);
1976 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1978 struct buffer_head *ibh;
1981 spin_unlock(&nilfs->ns_inode_lock);
1982 err = nilfs_ifile_get_inode_block(
1983 ifile, ii->vfs_inode.i_ino, &ibh);
1984 if (unlikely(err)) {
1985 nilfs_warn(sci->sc_super,
1986 "log writer: error %d getting inode block (ino=%lu)",
1987 err, ii->vfs_inode.i_ino);
1990 spin_lock(&nilfs->ns_inode_lock);
1991 if (likely(!ii->i_bh))
1998 // Always redirty the buffer to avoid race condition
1999 mark_buffer_dirty(ii->i_bh);
2000 nilfs_mdt_mark_dirty(ifile);
2002 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2003 set_bit(NILFS_I_BUSY, &ii->i_state);
2004 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2006 spin_unlock(&nilfs->ns_inode_lock);
2011 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2012 struct the_nilfs *nilfs)
2014 struct nilfs_inode_info *ii, *n;
2015 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2016 int defer_iput = false;
2018 spin_lock(&nilfs->ns_inode_lock);
2019 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2020 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2021 test_bit(NILFS_I_DIRTY, &ii->i_state))
2024 clear_bit(NILFS_I_BUSY, &ii->i_state);
2027 list_del_init(&ii->i_dirty);
2028 if (!ii->vfs_inode.i_nlink || during_mount) {
2030 * Defer calling iput() to avoid deadlocks if
2031 * i_nlink == 0 or mount is not yet finished.
2033 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2036 spin_unlock(&nilfs->ns_inode_lock);
2037 iput(&ii->vfs_inode);
2038 spin_lock(&nilfs->ns_inode_lock);
2041 spin_unlock(&nilfs->ns_inode_lock);
2044 schedule_work(&sci->sc_iput_work);
2048 * Main procedure of segment constructor
2050 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2052 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2055 if (sb_rdonly(sci->sc_super))
2058 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2059 sci->sc_cno = nilfs->ns_cno;
2061 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2065 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2066 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2068 if (nilfs_segctor_clean(sci))
2072 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2074 err = nilfs_segctor_begin_construction(sci, nilfs);
2078 /* Update time stamp */
2079 sci->sc_seg_ctime = ktime_get_real_seconds();
2081 err = nilfs_segctor_collect(sci, nilfs, mode);
2085 /* Avoid empty segment */
2086 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2087 nilfs_segbuf_empty(sci->sc_curseg)) {
2088 nilfs_segctor_abort_construction(sci, nilfs, 1);
2092 err = nilfs_segctor_assign(sci, mode);
2096 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2097 nilfs_segctor_fill_in_file_bmap(sci);
2099 if (mode == SC_LSEG_SR &&
2100 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2101 err = nilfs_segctor_fill_in_checkpoint(sci);
2103 goto failed_to_write;
2105 nilfs_segctor_fill_in_super_root(sci, nilfs);
2107 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2109 /* Write partial segments */
2110 nilfs_segctor_prepare_write(sci);
2112 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2113 nilfs->ns_crc_seed);
2115 err = nilfs_segctor_write(sci, nilfs);
2117 goto failed_to_write;
2119 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2120 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2122 * At this point, we avoid double buffering
2123 * for blocksize < pagesize because page dirty
2124 * flag is turned off during write and dirty
2125 * buffers are not properly collected for
2126 * pages crossing over segments.
2128 err = nilfs_segctor_wait(sci);
2130 goto failed_to_write;
2132 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2135 nilfs_segctor_drop_written_files(sci, nilfs);
2139 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2140 nilfs_redirty_inodes(&sci->sc_dirty_files);
2143 if (nilfs_doing_gc())
2144 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2145 nilfs_segctor_abort_construction(sci, nilfs, err);
2150 * nilfs_segctor_start_timer - set timer of background write
2151 * @sci: nilfs_sc_info
2153 * If the timer has already been set, it ignores the new request.
2154 * This function MUST be called within a section locking the segment
2157 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2159 spin_lock(&sci->sc_state_lock);
2160 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2161 sci->sc_timer.expires = jiffies + sci->sc_interval;
2162 add_timer(&sci->sc_timer);
2163 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2165 spin_unlock(&sci->sc_state_lock);
2168 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2170 spin_lock(&sci->sc_state_lock);
2171 if (!(sci->sc_flush_request & BIT(bn))) {
2172 unsigned long prev_req = sci->sc_flush_request;
2174 sci->sc_flush_request |= BIT(bn);
2176 wake_up(&sci->sc_wait_daemon);
2178 spin_unlock(&sci->sc_state_lock);
2182 * nilfs_flush_segment - trigger a segment construction for resource control
2184 * @ino: inode number of the file to be flushed out.
2186 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2188 struct the_nilfs *nilfs = sb->s_fs_info;
2189 struct nilfs_sc_info *sci = nilfs->ns_writer;
2191 if (!sci || nilfs_doing_construction())
2193 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2194 /* assign bit 0 to data files */
2197 struct nilfs_segctor_wait_request {
2198 wait_queue_entry_t wq;
2204 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2206 struct nilfs_segctor_wait_request wait_req;
2209 spin_lock(&sci->sc_state_lock);
2210 init_wait(&wait_req.wq);
2212 atomic_set(&wait_req.done, 0);
2213 wait_req.seq = ++sci->sc_seq_request;
2214 spin_unlock(&sci->sc_state_lock);
2216 init_waitqueue_entry(&wait_req.wq, current);
2217 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2218 set_current_state(TASK_INTERRUPTIBLE);
2219 wake_up(&sci->sc_wait_daemon);
2222 if (atomic_read(&wait_req.done)) {
2226 if (!signal_pending(current)) {
2233 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2237 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2239 struct nilfs_segctor_wait_request *wrq, *n;
2240 unsigned long flags;
2242 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2243 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2244 if (!atomic_read(&wrq->done) &&
2245 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2247 atomic_set(&wrq->done, 1);
2249 if (atomic_read(&wrq->done)) {
2250 wrq->wq.func(&wrq->wq,
2251 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2255 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2259 * nilfs_construct_segment - construct a logical segment
2262 * Return Value: On success, 0 is returned. On errors, one of the following
2263 * negative error code is returned.
2265 * %-EROFS - Read only filesystem.
2269 * %-ENOSPC - No space left on device (only in a panic state).
2271 * %-ERESTARTSYS - Interrupted.
2273 * %-ENOMEM - Insufficient memory available.
2275 int nilfs_construct_segment(struct super_block *sb)
2277 struct the_nilfs *nilfs = sb->s_fs_info;
2278 struct nilfs_sc_info *sci = nilfs->ns_writer;
2279 struct nilfs_transaction_info *ti;
2281 if (sb_rdonly(sb) || unlikely(!sci))
2284 /* A call inside transactions causes a deadlock. */
2285 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2287 return nilfs_segctor_sync(sci);
2291 * nilfs_construct_dsync_segment - construct a data-only logical segment
2293 * @inode: inode whose data blocks should be written out
2294 * @start: start byte offset
2295 * @end: end byte offset (inclusive)
2297 * Return Value: On success, 0 is returned. On errors, one of the following
2298 * negative error code is returned.
2300 * %-EROFS - Read only filesystem.
2304 * %-ENOSPC - No space left on device (only in a panic state).
2306 * %-ERESTARTSYS - Interrupted.
2308 * %-ENOMEM - Insufficient memory available.
2310 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2311 loff_t start, loff_t end)
2313 struct the_nilfs *nilfs = sb->s_fs_info;
2314 struct nilfs_sc_info *sci = nilfs->ns_writer;
2315 struct nilfs_inode_info *ii;
2316 struct nilfs_transaction_info ti;
2319 if (sb_rdonly(sb) || unlikely(!sci))
2322 nilfs_transaction_lock(sb, &ti, 0);
2324 ii = NILFS_I(inode);
2325 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2326 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2327 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2328 nilfs_discontinued(nilfs)) {
2329 nilfs_transaction_unlock(sb);
2330 err = nilfs_segctor_sync(sci);
2334 spin_lock(&nilfs->ns_inode_lock);
2335 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2336 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2337 spin_unlock(&nilfs->ns_inode_lock);
2338 nilfs_transaction_unlock(sb);
2341 spin_unlock(&nilfs->ns_inode_lock);
2342 sci->sc_dsync_inode = ii;
2343 sci->sc_dsync_start = start;
2344 sci->sc_dsync_end = end;
2346 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2348 nilfs->ns_flushed_device = 0;
2350 nilfs_transaction_unlock(sb);
2354 #define FLUSH_FILE_BIT (0x1) /* data file only */
2355 #define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2358 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2359 * @sci: segment constructor object
2361 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2363 spin_lock(&sci->sc_state_lock);
2364 sci->sc_seq_accepted = sci->sc_seq_request;
2365 spin_unlock(&sci->sc_state_lock);
2366 del_timer_sync(&sci->sc_timer);
2370 * nilfs_segctor_notify - notify the result of request to caller threads
2371 * @sci: segment constructor object
2372 * @mode: mode of log forming
2373 * @err: error code to be notified
2375 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2377 /* Clear requests (even when the construction failed) */
2378 spin_lock(&sci->sc_state_lock);
2380 if (mode == SC_LSEG_SR) {
2381 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2382 sci->sc_seq_done = sci->sc_seq_accepted;
2383 nilfs_segctor_wakeup(sci, err);
2384 sci->sc_flush_request = 0;
2386 if (mode == SC_FLUSH_FILE)
2387 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2388 else if (mode == SC_FLUSH_DAT)
2389 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2391 /* re-enable timer if checkpoint creation was not done */
2392 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2393 time_before(jiffies, sci->sc_timer.expires))
2394 add_timer(&sci->sc_timer);
2396 spin_unlock(&sci->sc_state_lock);
2400 * nilfs_segctor_construct - form logs and write them to disk
2401 * @sci: segment constructor object
2402 * @mode: mode of log forming
2404 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2406 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2407 struct nilfs_super_block **sbp;
2410 nilfs_segctor_accept(sci);
2412 if (nilfs_discontinued(nilfs))
2414 if (!nilfs_segctor_confirm(sci))
2415 err = nilfs_segctor_do_construct(sci, mode);
2418 if (mode != SC_FLUSH_DAT)
2419 atomic_set(&nilfs->ns_ndirtyblks, 0);
2420 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2421 nilfs_discontinued(nilfs)) {
2422 down_write(&nilfs->ns_sem);
2424 sbp = nilfs_prepare_super(sci->sc_super,
2425 nilfs_sb_will_flip(nilfs));
2427 nilfs_set_log_cursor(sbp[0], nilfs);
2428 err = nilfs_commit_super(sci->sc_super,
2431 up_write(&nilfs->ns_sem);
2435 nilfs_segctor_notify(sci, mode, err);
2439 static void nilfs_construction_timeout(struct timer_list *t)
2441 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2443 wake_up_process(sci->sc_timer_task);
2447 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2449 struct nilfs_inode_info *ii, *n;
2451 list_for_each_entry_safe(ii, n, head, i_dirty) {
2452 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2454 list_del_init(&ii->i_dirty);
2455 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2456 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2457 iput(&ii->vfs_inode);
2461 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2464 struct the_nilfs *nilfs = sb->s_fs_info;
2465 struct nilfs_sc_info *sci = nilfs->ns_writer;
2466 struct nilfs_transaction_info ti;
2472 nilfs_transaction_lock(sb, &ti, 1);
2474 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2478 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2479 if (unlikely(err)) {
2480 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2484 sci->sc_freesegs = kbufs[4];
2485 sci->sc_nfreesegs = argv[4].v_nmembs;
2486 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2489 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2490 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2495 nilfs_warn(sb, "error %d cleaning segments", err);
2496 set_current_state(TASK_INTERRUPTIBLE);
2497 schedule_timeout(sci->sc_interval);
2499 if (nilfs_test_opt(nilfs, DISCARD)) {
2500 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2504 "error %d on discard request, turning discards off for the device",
2506 nilfs_clear_opt(nilfs, DISCARD);
2511 sci->sc_freesegs = NULL;
2512 sci->sc_nfreesegs = 0;
2513 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2514 nilfs_transaction_unlock(sb);
2518 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2520 struct nilfs_transaction_info ti;
2522 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2523 nilfs_segctor_construct(sci, mode);
2526 * Unclosed segment should be retried. We do this using sc_timer.
2527 * Timeout of sc_timer will invoke complete construction which leads
2528 * to close the current logical segment.
2530 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2531 nilfs_segctor_start_timer(sci);
2533 nilfs_transaction_unlock(sci->sc_super);
2536 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2540 spin_lock(&sci->sc_state_lock);
2541 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2542 SC_FLUSH_DAT : SC_FLUSH_FILE;
2543 spin_unlock(&sci->sc_state_lock);
2546 nilfs_segctor_do_construct(sci, mode);
2548 spin_lock(&sci->sc_state_lock);
2549 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2550 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2551 spin_unlock(&sci->sc_state_lock);
2553 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2556 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2558 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2559 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2560 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2561 return SC_FLUSH_FILE;
2562 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2563 return SC_FLUSH_DAT;
2569 * nilfs_segctor_thread - main loop of the segment constructor thread.
2570 * @arg: pointer to a struct nilfs_sc_info.
2572 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2573 * to execute segment constructions.
2575 static int nilfs_segctor_thread(void *arg)
2577 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2578 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2581 sci->sc_timer_task = current;
2584 sci->sc_task = current;
2585 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2586 nilfs_info(sci->sc_super,
2587 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2588 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2590 spin_lock(&sci->sc_state_lock);
2595 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2598 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2600 else if (sci->sc_flush_request)
2601 mode = nilfs_segctor_flush_mode(sci);
2605 spin_unlock(&sci->sc_state_lock);
2606 nilfs_segctor_thread_construct(sci, mode);
2607 spin_lock(&sci->sc_state_lock);
2612 if (freezing(current)) {
2613 spin_unlock(&sci->sc_state_lock);
2615 spin_lock(&sci->sc_state_lock);
2618 int should_sleep = 1;
2620 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2621 TASK_INTERRUPTIBLE);
2623 if (sci->sc_seq_request != sci->sc_seq_done)
2625 else if (sci->sc_flush_request)
2627 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2628 should_sleep = time_before(jiffies,
2629 sci->sc_timer.expires);
2632 spin_unlock(&sci->sc_state_lock);
2634 spin_lock(&sci->sc_state_lock);
2636 finish_wait(&sci->sc_wait_daemon, &wait);
2637 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2638 time_after_eq(jiffies, sci->sc_timer.expires));
2640 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2641 set_nilfs_discontinued(nilfs);
2647 sci->sc_task = NULL;
2648 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2649 spin_unlock(&sci->sc_state_lock);
2653 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2655 struct task_struct *t;
2657 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2659 int err = PTR_ERR(t);
2661 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2665 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2669 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2670 __acquires(&sci->sc_state_lock)
2671 __releases(&sci->sc_state_lock)
2673 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2675 while (sci->sc_task) {
2676 wake_up(&sci->sc_wait_daemon);
2677 spin_unlock(&sci->sc_state_lock);
2678 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2679 spin_lock(&sci->sc_state_lock);
2684 * Setup & clean-up functions
2686 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2687 struct nilfs_root *root)
2689 struct the_nilfs *nilfs = sb->s_fs_info;
2690 struct nilfs_sc_info *sci;
2692 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2698 nilfs_get_root(root);
2699 sci->sc_root = root;
2701 init_waitqueue_head(&sci->sc_wait_request);
2702 init_waitqueue_head(&sci->sc_wait_daemon);
2703 init_waitqueue_head(&sci->sc_wait_task);
2704 spin_lock_init(&sci->sc_state_lock);
2705 INIT_LIST_HEAD(&sci->sc_dirty_files);
2706 INIT_LIST_HEAD(&sci->sc_segbufs);
2707 INIT_LIST_HEAD(&sci->sc_write_logs);
2708 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2709 INIT_LIST_HEAD(&sci->sc_iput_queue);
2710 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2711 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2713 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2714 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2715 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2717 if (nilfs->ns_interval)
2718 sci->sc_interval = HZ * nilfs->ns_interval;
2719 if (nilfs->ns_watermark)
2720 sci->sc_watermark = nilfs->ns_watermark;
2724 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2726 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2729 * The segctord thread was stopped and its timer was removed.
2730 * But some tasks remain.
2733 struct nilfs_transaction_info ti;
2735 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2736 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2737 nilfs_transaction_unlock(sci->sc_super);
2739 flush_work(&sci->sc_iput_work);
2741 } while (ret && ret != -EROFS && retrycount-- > 0);
2745 * nilfs_segctor_destroy - destroy the segment constructor.
2746 * @sci: nilfs_sc_info
2748 * nilfs_segctor_destroy() kills the segctord thread and frees
2749 * the nilfs_sc_info struct.
2750 * Caller must hold the segment semaphore.
2752 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2754 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2757 up_write(&nilfs->ns_segctor_sem);
2759 spin_lock(&sci->sc_state_lock);
2760 nilfs_segctor_kill_thread(sci);
2761 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2762 || sci->sc_seq_request != sci->sc_seq_done);
2763 spin_unlock(&sci->sc_state_lock);
2765 if (flush_work(&sci->sc_iput_work))
2768 if (flag || !nilfs_segctor_confirm(sci))
2769 nilfs_segctor_write_out(sci);
2771 if (!list_empty(&sci->sc_dirty_files)) {
2772 nilfs_warn(sci->sc_super,
2773 "disposed unprocessed dirty file(s) when stopping log writer");
2774 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2777 if (!list_empty(&sci->sc_iput_queue)) {
2778 nilfs_warn(sci->sc_super,
2779 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2780 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2783 WARN_ON(!list_empty(&sci->sc_segbufs));
2784 WARN_ON(!list_empty(&sci->sc_write_logs));
2786 nilfs_put_root(sci->sc_root);
2788 down_write(&nilfs->ns_segctor_sem);
2790 timer_shutdown_sync(&sci->sc_timer);
2795 * nilfs_attach_log_writer - attach log writer
2796 * @sb: super block instance
2797 * @root: root object of the current filesystem tree
2799 * This allocates a log writer object, initializes it, and starts the
2802 * Return Value: On success, 0 is returned. On error, one of the following
2803 * negative error code is returned.
2805 * %-ENOMEM - Insufficient memory available.
2807 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2809 struct the_nilfs *nilfs = sb->s_fs_info;
2812 if (nilfs->ns_writer) {
2814 * This happens if the filesystem is made read-only by
2815 * __nilfs_error or nilfs_remount and then remounted
2816 * read/write. In these cases, reuse the existing
2822 nilfs->ns_writer = nilfs_segctor_new(sb, root);
2823 if (!nilfs->ns_writer)
2826 inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2828 err = nilfs_segctor_start_thread(nilfs->ns_writer);
2830 nilfs_detach_log_writer(sb);
2836 * nilfs_detach_log_writer - destroy log writer
2837 * @sb: super block instance
2839 * This kills log writer daemon, frees the log writer object, and
2840 * destroys list of dirty files.
2842 void nilfs_detach_log_writer(struct super_block *sb)
2844 struct the_nilfs *nilfs = sb->s_fs_info;
2845 LIST_HEAD(garbage_list);
2847 down_write(&nilfs->ns_segctor_sem);
2848 if (nilfs->ns_writer) {
2849 nilfs_segctor_destroy(nilfs->ns_writer);
2850 nilfs->ns_writer = NULL;
2852 set_nilfs_purging(nilfs);
2854 /* Force to free the list of dirty files */
2855 spin_lock(&nilfs->ns_inode_lock);
2856 if (!list_empty(&nilfs->ns_dirty_files)) {
2857 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2859 "disposed unprocessed dirty file(s) when detaching log writer");
2861 spin_unlock(&nilfs->ns_inode_lock);
2862 up_write(&nilfs->ns_segctor_sem);
2864 nilfs_dispose_list(nilfs, &garbage_list, 1);
2865 clear_nilfs_purging(nilfs);