2 * segment.c - NILFS segment constructor.
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * Written by Ryusuke Konishi.
20 #include <linux/pagemap.h>
21 #include <linux/buffer_head.h>
22 #include <linux/writeback.h>
23 #include <linux/bitops.h>
24 #include <linux/bio.h>
25 #include <linux/completion.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/freezer.h>
29 #include <linux/kthread.h>
30 #include <linux/crc32.h>
31 #include <linux/pagevec.h>
32 #include <linux/slab.h>
33 #include <linux/sched/signal.h>
48 #define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
50 #define SC_MAX_SEGDELTA 64 /*
51 * Upper limit of the number of segments
52 * appended in collection retry loop
55 /* Construction mode */
57 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
59 * Flush data blocks of a given file and make
60 * a logical segment without a super root.
63 * Flush data files, leads to segment writes without
64 * creating a checkpoint.
67 * Flush DAT file. This also creates segments
68 * without a checkpoint.
72 /* Stage numbers of dirty block collection */
75 NILFS_ST_GC, /* Collecting dirty blocks for GC */
81 NILFS_ST_SR, /* Super root */
82 NILFS_ST_DSYNC, /* Data sync blocks */
86 #define CREATE_TRACE_POINTS
87 #include <trace/events/nilfs2.h>
90 * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
91 * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
92 * the variable must use them because transition of stage count must involve
93 * trace events (trace_nilfs2_collection_stage_transition).
95 * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
96 * produce tracepoint events. It is provided just for making the intention
99 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
101 sci->sc_stage.scnt++;
102 trace_nilfs2_collection_stage_transition(sci);
105 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
107 sci->sc_stage.scnt = next_scnt;
108 trace_nilfs2_collection_stage_transition(sci);
111 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
113 return sci->sc_stage.scnt;
116 /* State flags of collection */
117 #define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
118 #define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
119 #define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
120 #define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
122 /* Operations depending on the construction mode and file type */
123 struct nilfs_sc_operations {
124 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
126 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
128 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
130 void (*write_data_binfo)(struct nilfs_sc_info *,
131 struct nilfs_segsum_pointer *,
132 union nilfs_binfo *);
133 void (*write_node_binfo)(struct nilfs_sc_info *,
134 struct nilfs_segsum_pointer *,
135 union nilfs_binfo *);
141 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
142 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
143 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
144 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
146 #define nilfs_cnt32_gt(a, b) \
147 (typecheck(__u32, a) && typecheck(__u32, b) && \
148 ((__s32)(b) - (__s32)(a) < 0))
149 #define nilfs_cnt32_ge(a, b) \
150 (typecheck(__u32, a) && typecheck(__u32, b) && \
151 ((__s32)(a) - (__s32)(b) >= 0))
152 #define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
153 #define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
155 static int nilfs_prepare_segment_lock(struct super_block *sb,
156 struct nilfs_transaction_info *ti)
158 struct nilfs_transaction_info *cur_ti = current->journal_info;
162 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
163 return ++cur_ti->ti_count;
166 * If journal_info field is occupied by other FS,
167 * it is saved and will be restored on
168 * nilfs_transaction_commit().
170 nilfs_msg(sb, KERN_WARNING, "journal info from a different FS");
171 save = current->journal_info;
174 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
177 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
183 ti->ti_magic = NILFS_TI_MAGIC;
184 current->journal_info = ti;
189 * nilfs_transaction_begin - start indivisible file operations.
191 * @ti: nilfs_transaction_info
192 * @vacancy_check: flags for vacancy rate checks
194 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
195 * the segment semaphore, to make a segment construction and write tasks
196 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
197 * The region enclosed by these two functions can be nested. To avoid a
198 * deadlock, the semaphore is only acquired or released in the outermost call.
200 * This function allocates a nilfs_transaction_info struct to keep context
201 * information on it. It is initialized and hooked onto the current task in
202 * the outermost call. If a pre-allocated struct is given to @ti, it is used
203 * instead; otherwise a new struct is assigned from a slab.
205 * When @vacancy_check flag is set, this function will check the amount of
206 * free space, and will wait for the GC to reclaim disk space if low capacity.
208 * Return Value: On success, 0 is returned. On error, one of the following
209 * negative error code is returned.
211 * %-ENOMEM - Insufficient memory available.
213 * %-ENOSPC - No space left on device
215 int nilfs_transaction_begin(struct super_block *sb,
216 struct nilfs_transaction_info *ti,
219 struct the_nilfs *nilfs;
220 int ret = nilfs_prepare_segment_lock(sb, ti);
221 struct nilfs_transaction_info *trace_ti;
223 if (unlikely(ret < 0))
226 trace_ti = current->journal_info;
228 trace_nilfs2_transaction_transition(sb, trace_ti,
229 trace_ti->ti_count, trace_ti->ti_flags,
230 TRACE_NILFS2_TRANSACTION_BEGIN);
234 sb_start_intwrite(sb);
236 nilfs = sb->s_fs_info;
237 down_read(&nilfs->ns_segctor_sem);
238 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
239 up_read(&nilfs->ns_segctor_sem);
244 trace_ti = current->journal_info;
245 trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
247 TRACE_NILFS2_TRANSACTION_BEGIN);
251 ti = current->journal_info;
252 current->journal_info = ti->ti_save;
253 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
254 kmem_cache_free(nilfs_transaction_cachep, ti);
260 * nilfs_transaction_commit - commit indivisible file operations.
263 * nilfs_transaction_commit() releases the read semaphore which is
264 * acquired by nilfs_transaction_begin(). This is only performed
265 * in outermost call of this function. If a commit flag is set,
266 * nilfs_transaction_commit() sets a timer to start the segment
267 * constructor. If a sync flag is set, it starts construction
270 int nilfs_transaction_commit(struct super_block *sb)
272 struct nilfs_transaction_info *ti = current->journal_info;
273 struct the_nilfs *nilfs = sb->s_fs_info;
276 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
277 ti->ti_flags |= NILFS_TI_COMMIT;
278 if (ti->ti_count > 0) {
280 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
281 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
284 if (nilfs->ns_writer) {
285 struct nilfs_sc_info *sci = nilfs->ns_writer;
287 if (ti->ti_flags & NILFS_TI_COMMIT)
288 nilfs_segctor_start_timer(sci);
289 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
290 nilfs_segctor_do_flush(sci, 0);
292 up_read(&nilfs->ns_segctor_sem);
293 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
294 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
296 current->journal_info = ti->ti_save;
298 if (ti->ti_flags & NILFS_TI_SYNC)
299 err = nilfs_construct_segment(sb);
300 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
301 kmem_cache_free(nilfs_transaction_cachep, ti);
306 void nilfs_transaction_abort(struct super_block *sb)
308 struct nilfs_transaction_info *ti = current->journal_info;
309 struct the_nilfs *nilfs = sb->s_fs_info;
311 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
312 if (ti->ti_count > 0) {
314 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
315 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
318 up_read(&nilfs->ns_segctor_sem);
320 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
321 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
323 current->journal_info = ti->ti_save;
324 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
325 kmem_cache_free(nilfs_transaction_cachep, ti);
329 void nilfs_relax_pressure_in_lock(struct super_block *sb)
331 struct the_nilfs *nilfs = sb->s_fs_info;
332 struct nilfs_sc_info *sci = nilfs->ns_writer;
334 if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request)
337 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
338 up_read(&nilfs->ns_segctor_sem);
340 down_write(&nilfs->ns_segctor_sem);
341 if (sci->sc_flush_request &&
342 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
343 struct nilfs_transaction_info *ti = current->journal_info;
345 ti->ti_flags |= NILFS_TI_WRITER;
346 nilfs_segctor_do_immediate_flush(sci);
347 ti->ti_flags &= ~NILFS_TI_WRITER;
349 downgrade_write(&nilfs->ns_segctor_sem);
352 static void nilfs_transaction_lock(struct super_block *sb,
353 struct nilfs_transaction_info *ti,
356 struct nilfs_transaction_info *cur_ti = current->journal_info;
357 struct the_nilfs *nilfs = sb->s_fs_info;
358 struct nilfs_sc_info *sci = nilfs->ns_writer;
361 ti->ti_flags = NILFS_TI_WRITER;
363 ti->ti_save = cur_ti;
364 ti->ti_magic = NILFS_TI_MAGIC;
365 current->journal_info = ti;
368 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
369 ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
371 down_write(&nilfs->ns_segctor_sem);
372 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
375 nilfs_segctor_do_immediate_flush(sci);
377 up_write(&nilfs->ns_segctor_sem);
381 ti->ti_flags |= NILFS_TI_GC;
383 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
384 ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
387 static void nilfs_transaction_unlock(struct super_block *sb)
389 struct nilfs_transaction_info *ti = current->journal_info;
390 struct the_nilfs *nilfs = sb->s_fs_info;
392 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
393 BUG_ON(ti->ti_count > 0);
395 up_write(&nilfs->ns_segctor_sem);
396 current->journal_info = ti->ti_save;
398 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
399 ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
402 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
403 struct nilfs_segsum_pointer *ssp,
406 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
407 unsigned int blocksize = sci->sc_super->s_blocksize;
410 if (unlikely(ssp->offset + bytes > blocksize)) {
412 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
413 &segbuf->sb_segsum_buffers));
414 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
416 p = ssp->bh->b_data + ssp->offset;
417 ssp->offset += bytes;
422 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
423 * @sci: nilfs_sc_info
425 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
427 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
428 struct buffer_head *sumbh;
429 unsigned int sumbytes;
430 unsigned int flags = 0;
433 if (nilfs_doing_gc())
435 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
439 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
440 sumbytes = segbuf->sb_sum.sumbytes;
441 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
442 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
443 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
447 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
449 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
450 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
452 * The current segment is filled up
455 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
456 return nilfs_segctor_reset_segment_buffer(sci);
459 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
461 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
464 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
465 err = nilfs_segctor_feed_segment(sci);
468 segbuf = sci->sc_curseg;
470 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
472 segbuf->sb_sum.flags |= NILFS_SS_SR;
477 * Functions for making segment summary and payloads
479 static int nilfs_segctor_segsum_block_required(
480 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
481 unsigned int binfo_size)
483 unsigned int blocksize = sci->sc_super->s_blocksize;
484 /* Size of finfo and binfo is enough small against blocksize */
486 return ssp->offset + binfo_size +
487 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
491 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
494 sci->sc_curseg->sb_sum.nfinfo++;
495 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
496 nilfs_segctor_map_segsum_entry(
497 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
499 if (NILFS_I(inode)->i_root &&
500 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
501 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
505 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
508 struct nilfs_finfo *finfo;
509 struct nilfs_inode_info *ii;
510 struct nilfs_segment_buffer *segbuf;
513 if (sci->sc_blk_cnt == 0)
518 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
520 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
525 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
527 finfo->fi_ino = cpu_to_le64(inode->i_ino);
528 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
529 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
530 finfo->fi_cno = cpu_to_le64(cno);
532 segbuf = sci->sc_curseg;
533 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
534 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
535 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
536 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
539 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
540 struct buffer_head *bh,
542 unsigned int binfo_size)
544 struct nilfs_segment_buffer *segbuf;
545 int required, err = 0;
548 segbuf = sci->sc_curseg;
549 required = nilfs_segctor_segsum_block_required(
550 sci, &sci->sc_binfo_ptr, binfo_size);
551 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
552 nilfs_segctor_end_finfo(sci, inode);
553 err = nilfs_segctor_feed_segment(sci);
558 if (unlikely(required)) {
559 err = nilfs_segbuf_extend_segsum(segbuf);
563 if (sci->sc_blk_cnt == 0)
564 nilfs_segctor_begin_finfo(sci, inode);
566 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
567 /* Substitution to vblocknr is delayed until update_blocknr() */
568 nilfs_segbuf_add_file_buffer(segbuf, bh);
575 * Callback functions that enumerate, mark, and collect dirty blocks
577 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
578 struct buffer_head *bh, struct inode *inode)
582 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
586 err = nilfs_segctor_add_file_block(sci, bh, inode,
587 sizeof(struct nilfs_binfo_v));
589 sci->sc_datablk_cnt++;
593 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
594 struct buffer_head *bh,
597 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
600 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
601 struct buffer_head *bh,
604 WARN_ON(!buffer_dirty(bh));
605 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
608 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
609 struct nilfs_segsum_pointer *ssp,
610 union nilfs_binfo *binfo)
612 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
613 sci, ssp, sizeof(*binfo_v));
614 *binfo_v = binfo->bi_v;
617 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
618 struct nilfs_segsum_pointer *ssp,
619 union nilfs_binfo *binfo)
621 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
622 sci, ssp, sizeof(*vblocknr));
623 *vblocknr = binfo->bi_v.bi_vblocknr;
626 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
627 .collect_data = nilfs_collect_file_data,
628 .collect_node = nilfs_collect_file_node,
629 .collect_bmap = nilfs_collect_file_bmap,
630 .write_data_binfo = nilfs_write_file_data_binfo,
631 .write_node_binfo = nilfs_write_file_node_binfo,
634 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
635 struct buffer_head *bh, struct inode *inode)
639 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
643 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
645 sci->sc_datablk_cnt++;
649 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
650 struct buffer_head *bh, struct inode *inode)
652 WARN_ON(!buffer_dirty(bh));
653 return nilfs_segctor_add_file_block(sci, bh, inode,
654 sizeof(struct nilfs_binfo_dat));
657 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
658 struct nilfs_segsum_pointer *ssp,
659 union nilfs_binfo *binfo)
661 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
663 *blkoff = binfo->bi_dat.bi_blkoff;
666 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
667 struct nilfs_segsum_pointer *ssp,
668 union nilfs_binfo *binfo)
670 struct nilfs_binfo_dat *binfo_dat =
671 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
672 *binfo_dat = binfo->bi_dat;
675 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
676 .collect_data = nilfs_collect_dat_data,
677 .collect_node = nilfs_collect_file_node,
678 .collect_bmap = nilfs_collect_dat_bmap,
679 .write_data_binfo = nilfs_write_dat_data_binfo,
680 .write_node_binfo = nilfs_write_dat_node_binfo,
683 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
684 .collect_data = nilfs_collect_file_data,
685 .collect_node = NULL,
686 .collect_bmap = NULL,
687 .write_data_binfo = nilfs_write_file_data_binfo,
688 .write_node_binfo = NULL,
691 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
692 struct list_head *listp,
694 loff_t start, loff_t end)
696 struct address_space *mapping = inode->i_mapping;
698 pgoff_t index = 0, last = ULONG_MAX;
702 if (unlikely(start != 0 || end != LLONG_MAX)) {
704 * A valid range is given for sync-ing data pages. The
705 * range is rounded to per-page; extra dirty buffers
706 * may be included if blocksize < pagesize.
708 index = start >> PAGE_SHIFT;
709 last = end >> PAGE_SHIFT;
711 pagevec_init(&pvec, 0);
713 if (unlikely(index > last) ||
714 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
715 min_t(pgoff_t, last - index,
716 PAGEVEC_SIZE - 1) + 1))
719 for (i = 0; i < pagevec_count(&pvec); i++) {
720 struct buffer_head *bh, *head;
721 struct page *page = pvec.pages[i];
723 if (unlikely(page->index > last))
727 if (!page_has_buffers(page))
728 create_empty_buffers(page, i_blocksize(inode), 0);
731 bh = head = page_buffers(page);
733 if (!buffer_dirty(bh) || buffer_async_write(bh))
736 list_add_tail(&bh->b_assoc_buffers, listp);
738 if (unlikely(ndirties >= nlimit)) {
739 pagevec_release(&pvec);
743 } while (bh = bh->b_this_page, bh != head);
745 pagevec_release(&pvec);
750 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
751 struct list_head *listp)
753 struct nilfs_inode_info *ii = NILFS_I(inode);
754 struct inode *btnc_inode = ii->i_assoc_inode;
756 struct buffer_head *bh, *head;
763 pagevec_init(&pvec, 0);
765 while (pagevec_lookup_tag(&pvec, btnc_inode->i_mapping, &index,
766 PAGECACHE_TAG_DIRTY, PAGEVEC_SIZE)) {
767 for (i = 0; i < pagevec_count(&pvec); i++) {
768 bh = head = page_buffers(pvec.pages[i]);
770 if (buffer_dirty(bh) &&
771 !buffer_async_write(bh)) {
773 list_add_tail(&bh->b_assoc_buffers,
776 bh = bh->b_this_page;
777 } while (bh != head);
779 pagevec_release(&pvec);
784 static void nilfs_dispose_list(struct the_nilfs *nilfs,
785 struct list_head *head, int force)
787 struct nilfs_inode_info *ii, *n;
788 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
791 while (!list_empty(head)) {
792 spin_lock(&nilfs->ns_inode_lock);
793 list_for_each_entry_safe(ii, n, head, i_dirty) {
794 list_del_init(&ii->i_dirty);
796 if (unlikely(ii->i_bh)) {
800 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
801 set_bit(NILFS_I_QUEUED, &ii->i_state);
802 list_add_tail(&ii->i_dirty,
803 &nilfs->ns_dirty_files);
807 if (nv == SC_N_INODEVEC)
810 spin_unlock(&nilfs->ns_inode_lock);
812 for (pii = ivec; nv > 0; pii++, nv--)
813 iput(&(*pii)->vfs_inode);
817 static void nilfs_iput_work_func(struct work_struct *work)
819 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
821 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
823 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
826 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
827 struct nilfs_root *root)
831 if (nilfs_mdt_fetch_dirty(root->ifile))
833 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
835 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
837 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
842 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
844 return list_empty(&sci->sc_dirty_files) &&
845 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
846 sci->sc_nfreesegs == 0 &&
847 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
850 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
852 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
855 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
856 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
858 spin_lock(&nilfs->ns_inode_lock);
859 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
862 spin_unlock(&nilfs->ns_inode_lock);
866 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
868 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
870 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
871 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
872 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
873 nilfs_mdt_clear_dirty(nilfs->ns_dat);
876 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
878 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
879 struct buffer_head *bh_cp;
880 struct nilfs_checkpoint *raw_cp;
883 /* XXX: this interface will be changed */
884 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
888 * The following code is duplicated with cpfile. But, it is
889 * needed to collect the checkpoint even if it was not newly
892 mark_buffer_dirty(bh_cp);
893 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
894 nilfs_cpfile_put_checkpoint(
895 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
896 } else if (err == -EINVAL || err == -ENOENT) {
897 nilfs_error(sci->sc_super,
898 "checkpoint creation failed due to metadata corruption.");
904 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
906 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
907 struct buffer_head *bh_cp;
908 struct nilfs_checkpoint *raw_cp;
911 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
914 if (err == -EINVAL || err == -ENOENT) {
915 nilfs_error(sci->sc_super,
916 "checkpoint finalization failed due to metadata corruption.");
921 raw_cp->cp_snapshot_list.ssl_next = 0;
922 raw_cp->cp_snapshot_list.ssl_prev = 0;
923 raw_cp->cp_inodes_count =
924 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
925 raw_cp->cp_blocks_count =
926 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
927 raw_cp->cp_nblk_inc =
928 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
929 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
930 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
932 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
933 nilfs_checkpoint_clear_minor(raw_cp);
935 nilfs_checkpoint_set_minor(raw_cp);
937 nilfs_write_inode_common(sci->sc_root->ifile,
938 &raw_cp->cp_ifile_inode, 1);
939 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
946 static void nilfs_fill_in_file_bmap(struct inode *ifile,
947 struct nilfs_inode_info *ii)
950 struct buffer_head *ibh;
951 struct nilfs_inode *raw_inode;
953 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
956 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
958 nilfs_bmap_write(ii->i_bmap, raw_inode);
959 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
963 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
965 struct nilfs_inode_info *ii;
967 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
968 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
969 set_bit(NILFS_I_COLLECTED, &ii->i_state);
973 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
974 struct the_nilfs *nilfs)
976 struct buffer_head *bh_sr;
977 struct nilfs_super_root *raw_sr;
978 unsigned int isz, srsz;
980 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
981 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
982 isz = nilfs->ns_inode_size;
983 srsz = NILFS_SR_BYTES(isz);
985 raw_sr->sr_bytes = cpu_to_le16(srsz);
986 raw_sr->sr_nongc_ctime
987 = cpu_to_le64(nilfs_doing_gc() ?
988 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
989 raw_sr->sr_flags = 0;
991 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
992 NILFS_SR_DAT_OFFSET(isz), 1);
993 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
994 NILFS_SR_CPFILE_OFFSET(isz), 1);
995 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
996 NILFS_SR_SUFILE_OFFSET(isz), 1);
997 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1000 static void nilfs_redirty_inodes(struct list_head *head)
1002 struct nilfs_inode_info *ii;
1004 list_for_each_entry(ii, head, i_dirty) {
1005 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1006 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1010 static void nilfs_drop_collected_inodes(struct list_head *head)
1012 struct nilfs_inode_info *ii;
1014 list_for_each_entry(ii, head, i_dirty) {
1015 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1018 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1019 set_bit(NILFS_I_UPDATED, &ii->i_state);
1023 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1024 struct inode *inode,
1025 struct list_head *listp,
1026 int (*collect)(struct nilfs_sc_info *,
1027 struct buffer_head *,
1030 struct buffer_head *bh, *n;
1034 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1035 list_del_init(&bh->b_assoc_buffers);
1036 err = collect(sci, bh, inode);
1039 goto dispose_buffers;
1045 while (!list_empty(listp)) {
1046 bh = list_first_entry(listp, struct buffer_head,
1048 list_del_init(&bh->b_assoc_buffers);
1054 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1056 /* Remaining number of blocks within segment buffer */
1057 return sci->sc_segbuf_nblocks -
1058 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1061 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1062 struct inode *inode,
1063 const struct nilfs_sc_operations *sc_ops)
1065 LIST_HEAD(data_buffers);
1066 LIST_HEAD(node_buffers);
1069 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1070 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1072 n = nilfs_lookup_dirty_data_buffers(
1073 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1075 err = nilfs_segctor_apply_buffers(
1076 sci, inode, &data_buffers,
1077 sc_ops->collect_data);
1078 BUG_ON(!err); /* always receive -E2BIG or true error */
1082 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1084 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1085 err = nilfs_segctor_apply_buffers(
1086 sci, inode, &data_buffers, sc_ops->collect_data);
1087 if (unlikely(err)) {
1088 /* dispose node list */
1089 nilfs_segctor_apply_buffers(
1090 sci, inode, &node_buffers, NULL);
1093 sci->sc_stage.flags |= NILFS_CF_NODE;
1096 err = nilfs_segctor_apply_buffers(
1097 sci, inode, &node_buffers, sc_ops->collect_node);
1101 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1102 err = nilfs_segctor_apply_buffers(
1103 sci, inode, &node_buffers, sc_ops->collect_bmap);
1107 nilfs_segctor_end_finfo(sci, inode);
1108 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1114 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1115 struct inode *inode)
1117 LIST_HEAD(data_buffers);
1118 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1121 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1122 sci->sc_dsync_start,
1125 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1126 nilfs_collect_file_data);
1128 nilfs_segctor_end_finfo(sci, inode);
1130 /* always receive -E2BIG or true error if n > rest */
1135 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1137 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1138 struct list_head *head;
1139 struct nilfs_inode_info *ii;
1143 switch (nilfs_sc_cstage_get(sci)) {
1146 sci->sc_stage.flags = 0;
1148 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1149 sci->sc_nblk_inc = 0;
1150 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1151 if (mode == SC_LSEG_DSYNC) {
1152 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1157 sci->sc_stage.dirty_file_ptr = NULL;
1158 sci->sc_stage.gc_inode_ptr = NULL;
1159 if (mode == SC_FLUSH_DAT) {
1160 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1163 nilfs_sc_cstage_inc(sci); /* Fall through */
1165 if (nilfs_doing_gc()) {
1166 head = &sci->sc_gc_inodes;
1167 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1169 list_for_each_entry_continue(ii, head, i_dirty) {
1170 err = nilfs_segctor_scan_file(
1171 sci, &ii->vfs_inode,
1172 &nilfs_sc_file_ops);
1173 if (unlikely(err)) {
1174 sci->sc_stage.gc_inode_ptr = list_entry(
1176 struct nilfs_inode_info,
1180 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1182 sci->sc_stage.gc_inode_ptr = NULL;
1184 nilfs_sc_cstage_inc(sci); /* Fall through */
1186 head = &sci->sc_dirty_files;
1187 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1189 list_for_each_entry_continue(ii, head, i_dirty) {
1190 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1192 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1193 &nilfs_sc_file_ops);
1194 if (unlikely(err)) {
1195 sci->sc_stage.dirty_file_ptr =
1196 list_entry(ii->i_dirty.prev,
1197 struct nilfs_inode_info,
1201 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1202 /* XXX: required ? */
1204 sci->sc_stage.dirty_file_ptr = NULL;
1205 if (mode == SC_FLUSH_FILE) {
1206 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1209 nilfs_sc_cstage_inc(sci);
1210 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1212 case NILFS_ST_IFILE:
1213 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1214 &nilfs_sc_file_ops);
1217 nilfs_sc_cstage_inc(sci);
1218 /* Creating a checkpoint */
1219 err = nilfs_segctor_create_checkpoint(sci);
1223 case NILFS_ST_CPFILE:
1224 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1225 &nilfs_sc_file_ops);
1228 nilfs_sc_cstage_inc(sci); /* Fall through */
1229 case NILFS_ST_SUFILE:
1230 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1231 sci->sc_nfreesegs, &ndone);
1232 if (unlikely(err)) {
1233 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1234 sci->sc_freesegs, ndone,
1238 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1240 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1241 &nilfs_sc_file_ops);
1244 nilfs_sc_cstage_inc(sci); /* Fall through */
1247 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1251 if (mode == SC_FLUSH_DAT) {
1252 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1255 nilfs_sc_cstage_inc(sci); /* Fall through */
1257 if (mode == SC_LSEG_SR) {
1258 /* Appending a super root */
1259 err = nilfs_segctor_add_super_root(sci);
1263 /* End of a logical segment */
1264 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1265 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1267 case NILFS_ST_DSYNC:
1269 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1270 ii = sci->sc_dsync_inode;
1271 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1274 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1277 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1278 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1291 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1292 * @sci: nilfs_sc_info
1293 * @nilfs: nilfs object
1295 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1296 struct the_nilfs *nilfs)
1298 struct nilfs_segment_buffer *segbuf, *prev;
1302 segbuf = nilfs_segbuf_new(sci->sc_super);
1303 if (unlikely(!segbuf))
1306 if (list_empty(&sci->sc_write_logs)) {
1307 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1308 nilfs->ns_pseg_offset, nilfs);
1309 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1310 nilfs_shift_to_next_segment(nilfs);
1311 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1314 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1315 nextnum = nilfs->ns_nextnum;
1317 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1318 /* Start from the head of a new full segment */
1322 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1323 nilfs_segbuf_map_cont(segbuf, prev);
1324 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1325 nextnum = prev->sb_nextnum;
1327 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1328 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1329 segbuf->sb_sum.seg_seq++;
1334 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1339 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1343 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1345 BUG_ON(!list_empty(&sci->sc_segbufs));
1346 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1347 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1351 nilfs_segbuf_free(segbuf);
1355 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1356 struct the_nilfs *nilfs, int nadd)
1358 struct nilfs_segment_buffer *segbuf, *prev;
1359 struct inode *sufile = nilfs->ns_sufile;
1364 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1366 * Since the segment specified with nextnum might be allocated during
1367 * the previous construction, the buffer including its segusage may
1368 * not be dirty. The following call ensures that the buffer is dirty
1369 * and will pin the buffer on memory until the sufile is written.
1371 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1375 for (i = 0; i < nadd; i++) {
1376 /* extend segment info */
1378 segbuf = nilfs_segbuf_new(sci->sc_super);
1379 if (unlikely(!segbuf))
1382 /* map this buffer to region of segment on-disk */
1383 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1384 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1386 /* allocate the next next full segment */
1387 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1391 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1392 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1394 list_add_tail(&segbuf->sb_list, &list);
1397 list_splice_tail(&list, &sci->sc_segbufs);
1401 nilfs_segbuf_free(segbuf);
1403 list_for_each_entry(segbuf, &list, sb_list) {
1404 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1405 WARN_ON(ret); /* never fails */
1407 nilfs_destroy_logs(&list);
1411 static void nilfs_free_incomplete_logs(struct list_head *logs,
1412 struct the_nilfs *nilfs)
1414 struct nilfs_segment_buffer *segbuf, *prev;
1415 struct inode *sufile = nilfs->ns_sufile;
1418 segbuf = NILFS_FIRST_SEGBUF(logs);
1419 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1420 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1421 WARN_ON(ret); /* never fails */
1423 if (atomic_read(&segbuf->sb_err)) {
1424 /* Case 1: The first segment failed */
1425 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1427 * Case 1a: Partial segment appended into an existing
1430 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1431 segbuf->sb_fseg_end);
1432 else /* Case 1b: New full segment */
1433 set_nilfs_discontinued(nilfs);
1437 list_for_each_entry_continue(segbuf, logs, sb_list) {
1438 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1439 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1440 WARN_ON(ret); /* never fails */
1442 if (atomic_read(&segbuf->sb_err) &&
1443 segbuf->sb_segnum != nilfs->ns_nextnum)
1444 /* Case 2: extended segment (!= next) failed */
1445 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1450 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1451 struct inode *sufile)
1453 struct nilfs_segment_buffer *segbuf;
1454 unsigned long live_blocks;
1457 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1458 live_blocks = segbuf->sb_sum.nblocks +
1459 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1460 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1463 WARN_ON(ret); /* always succeed because the segusage is dirty */
1467 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1469 struct nilfs_segment_buffer *segbuf;
1472 segbuf = NILFS_FIRST_SEGBUF(logs);
1473 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1474 segbuf->sb_pseg_start -
1475 segbuf->sb_fseg_start, 0);
1476 WARN_ON(ret); /* always succeed because the segusage is dirty */
1478 list_for_each_entry_continue(segbuf, logs, sb_list) {
1479 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1481 WARN_ON(ret); /* always succeed */
1485 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1486 struct nilfs_segment_buffer *last,
1487 struct inode *sufile)
1489 struct nilfs_segment_buffer *segbuf = last;
1492 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1493 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1494 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1497 nilfs_truncate_logs(&sci->sc_segbufs, last);
1501 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1502 struct the_nilfs *nilfs, int mode)
1504 struct nilfs_cstage prev_stage = sci->sc_stage;
1507 /* Collection retry loop */
1509 sci->sc_nblk_this_inc = 0;
1510 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1512 err = nilfs_segctor_reset_segment_buffer(sci);
1516 err = nilfs_segctor_collect_blocks(sci, mode);
1517 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1521 if (unlikely(err != -E2BIG))
1524 /* The current segment is filled up */
1525 if (mode != SC_LSEG_SR ||
1526 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1529 nilfs_clear_logs(&sci->sc_segbufs);
1531 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1532 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1536 WARN_ON(err); /* do not happen */
1537 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1540 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1544 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1545 sci->sc_stage = prev_stage;
1547 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1554 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1555 struct buffer_head *new_bh)
1557 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1559 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1560 /* The caller must release old_bh */
1564 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1565 struct nilfs_segment_buffer *segbuf,
1568 struct inode *inode = NULL;
1570 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1571 unsigned long nblocks = 0, ndatablk = 0;
1572 const struct nilfs_sc_operations *sc_op = NULL;
1573 struct nilfs_segsum_pointer ssp;
1574 struct nilfs_finfo *finfo = NULL;
1575 union nilfs_binfo binfo;
1576 struct buffer_head *bh, *bh_org;
1583 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1584 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1585 ssp.offset = sizeof(struct nilfs_segment_summary);
1587 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1588 if (bh == segbuf->sb_super_root)
1591 finfo = nilfs_segctor_map_segsum_entry(
1592 sci, &ssp, sizeof(*finfo));
1593 ino = le64_to_cpu(finfo->fi_ino);
1594 nblocks = le32_to_cpu(finfo->fi_nblocks);
1595 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1597 inode = bh->b_page->mapping->host;
1599 if (mode == SC_LSEG_DSYNC)
1600 sc_op = &nilfs_sc_dsync_ops;
1601 else if (ino == NILFS_DAT_INO)
1602 sc_op = &nilfs_sc_dat_ops;
1603 else /* file blocks */
1604 sc_op = &nilfs_sc_file_ops;
1608 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1611 nilfs_list_replace_buffer(bh_org, bh);
1617 sc_op->write_data_binfo(sci, &ssp, &binfo);
1619 sc_op->write_node_binfo(sci, &ssp, &binfo);
1622 if (--nblocks == 0) {
1626 } else if (ndatablk > 0)
1636 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1638 struct nilfs_segment_buffer *segbuf;
1641 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1642 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1645 nilfs_segbuf_fill_in_segsum(segbuf);
1650 static void nilfs_begin_page_io(struct page *page)
1652 if (!page || PageWriteback(page))
1654 * For split b-tree node pages, this function may be called
1655 * twice. We ignore the 2nd or later calls by this check.
1660 clear_page_dirty_for_io(page);
1661 set_page_writeback(page);
1665 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1667 struct nilfs_segment_buffer *segbuf;
1668 struct page *bd_page = NULL, *fs_page = NULL;
1670 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1671 struct buffer_head *bh;
1673 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1675 if (bh->b_page != bd_page) {
1678 clear_page_dirty_for_io(bd_page);
1679 set_page_writeback(bd_page);
1680 unlock_page(bd_page);
1682 bd_page = bh->b_page;
1686 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1688 set_buffer_async_write(bh);
1689 if (bh == segbuf->sb_super_root) {
1690 if (bh->b_page != bd_page) {
1692 clear_page_dirty_for_io(bd_page);
1693 set_page_writeback(bd_page);
1694 unlock_page(bd_page);
1695 bd_page = bh->b_page;
1699 if (bh->b_page != fs_page) {
1700 nilfs_begin_page_io(fs_page);
1701 fs_page = bh->b_page;
1707 clear_page_dirty_for_io(bd_page);
1708 set_page_writeback(bd_page);
1709 unlock_page(bd_page);
1711 nilfs_begin_page_io(fs_page);
1714 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1715 struct the_nilfs *nilfs)
1719 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1720 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1724 static void nilfs_end_page_io(struct page *page, int err)
1729 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1731 * For b-tree node pages, this function may be called twice
1732 * or more because they might be split in a segment.
1734 if (PageDirty(page)) {
1736 * For pages holding split b-tree node buffers, dirty
1737 * flag on the buffers may be cleared discretely.
1738 * In that case, the page is once redirtied for
1739 * remaining buffers, and it must be cancelled if
1740 * all the buffers get cleaned later.
1743 if (nilfs_page_buffers_clean(page))
1744 __nilfs_clear_page_dirty(page);
1751 if (!nilfs_page_buffers_clean(page))
1752 __set_page_dirty_nobuffers(page);
1753 ClearPageError(page);
1755 __set_page_dirty_nobuffers(page);
1759 end_page_writeback(page);
1762 static void nilfs_abort_logs(struct list_head *logs, int err)
1764 struct nilfs_segment_buffer *segbuf;
1765 struct page *bd_page = NULL, *fs_page = NULL;
1766 struct buffer_head *bh;
1768 if (list_empty(logs))
1771 list_for_each_entry(segbuf, logs, sb_list) {
1772 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1774 if (bh->b_page != bd_page) {
1776 end_page_writeback(bd_page);
1777 bd_page = bh->b_page;
1781 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1783 clear_buffer_async_write(bh);
1784 if (bh == segbuf->sb_super_root) {
1785 if (bh->b_page != bd_page) {
1786 end_page_writeback(bd_page);
1787 bd_page = bh->b_page;
1791 if (bh->b_page != fs_page) {
1792 nilfs_end_page_io(fs_page, err);
1793 fs_page = bh->b_page;
1798 end_page_writeback(bd_page);
1800 nilfs_end_page_io(fs_page, err);
1803 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1804 struct the_nilfs *nilfs, int err)
1809 list_splice_tail_init(&sci->sc_write_logs, &logs);
1810 ret = nilfs_wait_on_logs(&logs);
1811 nilfs_abort_logs(&logs, ret ? : err);
1813 list_splice_tail_init(&sci->sc_segbufs, &logs);
1814 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1815 nilfs_free_incomplete_logs(&logs, nilfs);
1817 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1818 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1822 WARN_ON(ret); /* do not happen */
1825 nilfs_destroy_logs(&logs);
1828 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1829 struct nilfs_segment_buffer *segbuf)
1831 nilfs->ns_segnum = segbuf->sb_segnum;
1832 nilfs->ns_nextnum = segbuf->sb_nextnum;
1833 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1834 + segbuf->sb_sum.nblocks;
1835 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1836 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1839 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1841 struct nilfs_segment_buffer *segbuf;
1842 struct page *bd_page = NULL, *fs_page = NULL;
1843 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1844 int update_sr = false;
1846 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1847 struct buffer_head *bh;
1849 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1851 set_buffer_uptodate(bh);
1852 clear_buffer_dirty(bh);
1853 if (bh->b_page != bd_page) {
1855 end_page_writeback(bd_page);
1856 bd_page = bh->b_page;
1860 * We assume that the buffers which belong to the same page
1861 * continue over the buffer list.
1862 * Under this assumption, the last BHs of pages is
1863 * identifiable by the discontinuity of bh->b_page
1864 * (page != fs_page).
1866 * For B-tree node blocks, however, this assumption is not
1867 * guaranteed. The cleanup code of B-tree node pages needs
1870 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1872 const unsigned long set_bits = BIT(BH_Uptodate);
1873 const unsigned long clear_bits =
1874 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1875 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1876 BIT(BH_NILFS_Redirected));
1878 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1879 if (bh == segbuf->sb_super_root) {
1880 if (bh->b_page != bd_page) {
1881 end_page_writeback(bd_page);
1882 bd_page = bh->b_page;
1887 if (bh->b_page != fs_page) {
1888 nilfs_end_page_io(fs_page, 0);
1889 fs_page = bh->b_page;
1893 if (!nilfs_segbuf_simplex(segbuf)) {
1894 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1895 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1896 sci->sc_lseg_stime = jiffies;
1898 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1899 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1903 * Since pages may continue over multiple segment buffers,
1904 * end of the last page must be checked outside of the loop.
1907 end_page_writeback(bd_page);
1909 nilfs_end_page_io(fs_page, 0);
1911 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1913 if (nilfs_doing_gc())
1914 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1916 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1918 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1920 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1921 nilfs_set_next_segment(nilfs, segbuf);
1924 nilfs->ns_flushed_device = 0;
1925 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1926 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1928 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1929 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1930 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1931 nilfs_segctor_clear_metadata_dirty(sci);
1933 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1936 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1940 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1942 nilfs_segctor_complete_write(sci);
1943 nilfs_destroy_logs(&sci->sc_write_logs);
1948 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1949 struct the_nilfs *nilfs)
1951 struct nilfs_inode_info *ii, *n;
1952 struct inode *ifile = sci->sc_root->ifile;
1954 spin_lock(&nilfs->ns_inode_lock);
1956 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1958 struct buffer_head *ibh;
1961 spin_unlock(&nilfs->ns_inode_lock);
1962 err = nilfs_ifile_get_inode_block(
1963 ifile, ii->vfs_inode.i_ino, &ibh);
1964 if (unlikely(err)) {
1965 nilfs_msg(sci->sc_super, KERN_WARNING,
1966 "log writer: error %d getting inode block (ino=%lu)",
1967 err, ii->vfs_inode.i_ino);
1970 spin_lock(&nilfs->ns_inode_lock);
1971 if (likely(!ii->i_bh))
1978 // Always redirty the buffer to avoid race condition
1979 mark_buffer_dirty(ii->i_bh);
1980 nilfs_mdt_mark_dirty(ifile);
1982 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1983 set_bit(NILFS_I_BUSY, &ii->i_state);
1984 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1986 spin_unlock(&nilfs->ns_inode_lock);
1991 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1992 struct the_nilfs *nilfs)
1994 struct nilfs_inode_info *ii, *n;
1995 int during_mount = !(sci->sc_super->s_flags & MS_ACTIVE);
1996 int defer_iput = false;
1998 spin_lock(&nilfs->ns_inode_lock);
1999 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2000 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2001 test_bit(NILFS_I_DIRTY, &ii->i_state))
2004 clear_bit(NILFS_I_BUSY, &ii->i_state);
2007 list_del_init(&ii->i_dirty);
2008 if (!ii->vfs_inode.i_nlink || during_mount) {
2010 * Defer calling iput() to avoid deadlocks if
2011 * i_nlink == 0 or mount is not yet finished.
2013 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2016 spin_unlock(&nilfs->ns_inode_lock);
2017 iput(&ii->vfs_inode);
2018 spin_lock(&nilfs->ns_inode_lock);
2021 spin_unlock(&nilfs->ns_inode_lock);
2024 schedule_work(&sci->sc_iput_work);
2028 * Main procedure of segment constructor
2030 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2032 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2035 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2036 sci->sc_cno = nilfs->ns_cno;
2038 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2042 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2043 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2045 if (nilfs_segctor_clean(sci))
2049 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2051 err = nilfs_segctor_begin_construction(sci, nilfs);
2055 /* Update time stamp */
2056 sci->sc_seg_ctime = get_seconds();
2058 err = nilfs_segctor_collect(sci, nilfs, mode);
2062 /* Avoid empty segment */
2063 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2064 nilfs_segbuf_empty(sci->sc_curseg)) {
2065 nilfs_segctor_abort_construction(sci, nilfs, 1);
2069 err = nilfs_segctor_assign(sci, mode);
2073 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2074 nilfs_segctor_fill_in_file_bmap(sci);
2076 if (mode == SC_LSEG_SR &&
2077 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2078 err = nilfs_segctor_fill_in_checkpoint(sci);
2080 goto failed_to_write;
2082 nilfs_segctor_fill_in_super_root(sci, nilfs);
2084 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2086 /* Write partial segments */
2087 nilfs_segctor_prepare_write(sci);
2089 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2090 nilfs->ns_crc_seed);
2092 err = nilfs_segctor_write(sci, nilfs);
2094 goto failed_to_write;
2096 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2097 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2099 * At this point, we avoid double buffering
2100 * for blocksize < pagesize because page dirty
2101 * flag is turned off during write and dirty
2102 * buffers are not properly collected for
2103 * pages crossing over segments.
2105 err = nilfs_segctor_wait(sci);
2107 goto failed_to_write;
2109 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2112 nilfs_segctor_drop_written_files(sci, nilfs);
2116 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2117 nilfs_redirty_inodes(&sci->sc_dirty_files);
2120 if (nilfs_doing_gc())
2121 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2122 nilfs_segctor_abort_construction(sci, nilfs, err);
2127 * nilfs_segctor_start_timer - set timer of background write
2128 * @sci: nilfs_sc_info
2130 * If the timer has already been set, it ignores the new request.
2131 * This function MUST be called within a section locking the segment
2134 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2136 spin_lock(&sci->sc_state_lock);
2137 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2138 sci->sc_timer.expires = jiffies + sci->sc_interval;
2139 add_timer(&sci->sc_timer);
2140 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2142 spin_unlock(&sci->sc_state_lock);
2145 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2147 spin_lock(&sci->sc_state_lock);
2148 if (!(sci->sc_flush_request & BIT(bn))) {
2149 unsigned long prev_req = sci->sc_flush_request;
2151 sci->sc_flush_request |= BIT(bn);
2153 wake_up(&sci->sc_wait_daemon);
2155 spin_unlock(&sci->sc_state_lock);
2159 * nilfs_flush_segment - trigger a segment construction for resource control
2161 * @ino: inode number of the file to be flushed out.
2163 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2165 struct the_nilfs *nilfs = sb->s_fs_info;
2166 struct nilfs_sc_info *sci = nilfs->ns_writer;
2168 if (!sci || nilfs_doing_construction())
2170 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2171 /* assign bit 0 to data files */
2174 struct nilfs_segctor_wait_request {
2175 wait_queue_entry_t wq;
2181 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2183 struct nilfs_segctor_wait_request wait_req;
2186 spin_lock(&sci->sc_state_lock);
2187 init_wait(&wait_req.wq);
2189 atomic_set(&wait_req.done, 0);
2190 wait_req.seq = ++sci->sc_seq_request;
2191 spin_unlock(&sci->sc_state_lock);
2193 init_waitqueue_entry(&wait_req.wq, current);
2194 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2195 set_current_state(TASK_INTERRUPTIBLE);
2196 wake_up(&sci->sc_wait_daemon);
2199 if (atomic_read(&wait_req.done)) {
2203 if (!signal_pending(current)) {
2210 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2214 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2216 struct nilfs_segctor_wait_request *wrq, *n;
2217 unsigned long flags;
2219 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2220 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2221 if (!atomic_read(&wrq->done) &&
2222 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2224 atomic_set(&wrq->done, 1);
2226 if (atomic_read(&wrq->done)) {
2227 wrq->wq.func(&wrq->wq,
2228 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2232 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2236 * nilfs_construct_segment - construct a logical segment
2239 * Return Value: On success, 0 is retured. On errors, one of the following
2240 * negative error code is returned.
2242 * %-EROFS - Read only filesystem.
2246 * %-ENOSPC - No space left on device (only in a panic state).
2248 * %-ERESTARTSYS - Interrupted.
2250 * %-ENOMEM - Insufficient memory available.
2252 int nilfs_construct_segment(struct super_block *sb)
2254 struct the_nilfs *nilfs = sb->s_fs_info;
2255 struct nilfs_sc_info *sci = nilfs->ns_writer;
2256 struct nilfs_transaction_info *ti;
2259 if (sb_rdonly(sb) || unlikely(!sci))
2262 /* A call inside transactions causes a deadlock. */
2263 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2265 err = nilfs_segctor_sync(sci);
2270 * nilfs_construct_dsync_segment - construct a data-only logical segment
2272 * @inode: inode whose data blocks should be written out
2273 * @start: start byte offset
2274 * @end: end byte offset (inclusive)
2276 * Return Value: On success, 0 is retured. On errors, one of the following
2277 * negative error code is returned.
2279 * %-EROFS - Read only filesystem.
2283 * %-ENOSPC - No space left on device (only in a panic state).
2285 * %-ERESTARTSYS - Interrupted.
2287 * %-ENOMEM - Insufficient memory available.
2289 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2290 loff_t start, loff_t end)
2292 struct the_nilfs *nilfs = sb->s_fs_info;
2293 struct nilfs_sc_info *sci = nilfs->ns_writer;
2294 struct nilfs_inode_info *ii;
2295 struct nilfs_transaction_info ti;
2298 if (sb_rdonly(sb) || unlikely(!sci))
2301 nilfs_transaction_lock(sb, &ti, 0);
2303 ii = NILFS_I(inode);
2304 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2305 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2306 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2307 nilfs_discontinued(nilfs)) {
2308 nilfs_transaction_unlock(sb);
2309 err = nilfs_segctor_sync(sci);
2313 spin_lock(&nilfs->ns_inode_lock);
2314 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2315 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2316 spin_unlock(&nilfs->ns_inode_lock);
2317 nilfs_transaction_unlock(sb);
2320 spin_unlock(&nilfs->ns_inode_lock);
2321 sci->sc_dsync_inode = ii;
2322 sci->sc_dsync_start = start;
2323 sci->sc_dsync_end = end;
2325 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2327 nilfs->ns_flushed_device = 0;
2329 nilfs_transaction_unlock(sb);
2333 #define FLUSH_FILE_BIT (0x1) /* data file only */
2334 #define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2337 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2338 * @sci: segment constructor object
2340 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2342 spin_lock(&sci->sc_state_lock);
2343 sci->sc_seq_accepted = sci->sc_seq_request;
2344 spin_unlock(&sci->sc_state_lock);
2345 del_timer_sync(&sci->sc_timer);
2349 * nilfs_segctor_notify - notify the result of request to caller threads
2350 * @sci: segment constructor object
2351 * @mode: mode of log forming
2352 * @err: error code to be notified
2354 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2356 /* Clear requests (even when the construction failed) */
2357 spin_lock(&sci->sc_state_lock);
2359 if (mode == SC_LSEG_SR) {
2360 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2361 sci->sc_seq_done = sci->sc_seq_accepted;
2362 nilfs_segctor_wakeup(sci, err);
2363 sci->sc_flush_request = 0;
2365 if (mode == SC_FLUSH_FILE)
2366 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2367 else if (mode == SC_FLUSH_DAT)
2368 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2370 /* re-enable timer if checkpoint creation was not done */
2371 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2372 time_before(jiffies, sci->sc_timer.expires))
2373 add_timer(&sci->sc_timer);
2375 spin_unlock(&sci->sc_state_lock);
2379 * nilfs_segctor_construct - form logs and write them to disk
2380 * @sci: segment constructor object
2381 * @mode: mode of log forming
2383 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2385 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2386 struct nilfs_super_block **sbp;
2389 nilfs_segctor_accept(sci);
2391 if (nilfs_discontinued(nilfs))
2393 if (!nilfs_segctor_confirm(sci))
2394 err = nilfs_segctor_do_construct(sci, mode);
2397 if (mode != SC_FLUSH_DAT)
2398 atomic_set(&nilfs->ns_ndirtyblks, 0);
2399 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2400 nilfs_discontinued(nilfs)) {
2401 down_write(&nilfs->ns_sem);
2403 sbp = nilfs_prepare_super(sci->sc_super,
2404 nilfs_sb_will_flip(nilfs));
2406 nilfs_set_log_cursor(sbp[0], nilfs);
2407 err = nilfs_commit_super(sci->sc_super,
2410 up_write(&nilfs->ns_sem);
2414 nilfs_segctor_notify(sci, mode, err);
2418 static void nilfs_construction_timeout(unsigned long data)
2420 struct task_struct *p = (struct task_struct *)data;
2426 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2428 struct nilfs_inode_info *ii, *n;
2430 list_for_each_entry_safe(ii, n, head, i_dirty) {
2431 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2433 list_del_init(&ii->i_dirty);
2434 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2435 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2436 iput(&ii->vfs_inode);
2440 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2443 struct the_nilfs *nilfs = sb->s_fs_info;
2444 struct nilfs_sc_info *sci = nilfs->ns_writer;
2445 struct nilfs_transaction_info ti;
2451 nilfs_transaction_lock(sb, &ti, 1);
2453 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2457 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2458 if (unlikely(err)) {
2459 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2463 sci->sc_freesegs = kbufs[4];
2464 sci->sc_nfreesegs = argv[4].v_nmembs;
2465 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2468 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2469 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2474 nilfs_msg(sb, KERN_WARNING, "error %d cleaning segments", err);
2475 set_current_state(TASK_INTERRUPTIBLE);
2476 schedule_timeout(sci->sc_interval);
2478 if (nilfs_test_opt(nilfs, DISCARD)) {
2479 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2482 nilfs_msg(sb, KERN_WARNING,
2483 "error %d on discard request, turning discards off for the device",
2485 nilfs_clear_opt(nilfs, DISCARD);
2490 sci->sc_freesegs = NULL;
2491 sci->sc_nfreesegs = 0;
2492 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2493 nilfs_transaction_unlock(sb);
2497 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2499 struct nilfs_transaction_info ti;
2501 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2502 nilfs_segctor_construct(sci, mode);
2505 * Unclosed segment should be retried. We do this using sc_timer.
2506 * Timeout of sc_timer will invoke complete construction which leads
2507 * to close the current logical segment.
2509 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2510 nilfs_segctor_start_timer(sci);
2512 nilfs_transaction_unlock(sci->sc_super);
2515 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2519 spin_lock(&sci->sc_state_lock);
2520 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2521 SC_FLUSH_DAT : SC_FLUSH_FILE;
2522 spin_unlock(&sci->sc_state_lock);
2525 nilfs_segctor_do_construct(sci, mode);
2527 spin_lock(&sci->sc_state_lock);
2528 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2529 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2530 spin_unlock(&sci->sc_state_lock);
2532 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2535 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2537 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2538 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2539 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2540 return SC_FLUSH_FILE;
2541 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2542 return SC_FLUSH_DAT;
2548 * nilfs_segctor_thread - main loop of the segment constructor thread.
2549 * @arg: pointer to a struct nilfs_sc_info.
2551 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2552 * to execute segment constructions.
2554 static int nilfs_segctor_thread(void *arg)
2556 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2557 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2560 sci->sc_timer.data = (unsigned long)current;
2561 sci->sc_timer.function = nilfs_construction_timeout;
2564 sci->sc_task = current;
2565 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2566 nilfs_msg(sci->sc_super, KERN_INFO,
2567 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2568 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2570 spin_lock(&sci->sc_state_lock);
2575 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2578 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2580 else if (sci->sc_flush_request)
2581 mode = nilfs_segctor_flush_mode(sci);
2585 spin_unlock(&sci->sc_state_lock);
2586 nilfs_segctor_thread_construct(sci, mode);
2587 spin_lock(&sci->sc_state_lock);
2592 if (freezing(current)) {
2593 spin_unlock(&sci->sc_state_lock);
2595 spin_lock(&sci->sc_state_lock);
2598 int should_sleep = 1;
2600 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2601 TASK_INTERRUPTIBLE);
2603 if (sci->sc_seq_request != sci->sc_seq_done)
2605 else if (sci->sc_flush_request)
2607 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2608 should_sleep = time_before(jiffies,
2609 sci->sc_timer.expires);
2612 spin_unlock(&sci->sc_state_lock);
2614 spin_lock(&sci->sc_state_lock);
2616 finish_wait(&sci->sc_wait_daemon, &wait);
2617 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2618 time_after_eq(jiffies, sci->sc_timer.expires));
2620 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2621 set_nilfs_discontinued(nilfs);
2626 spin_unlock(&sci->sc_state_lock);
2629 sci->sc_task = NULL;
2630 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2634 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2636 struct task_struct *t;
2638 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2640 int err = PTR_ERR(t);
2642 nilfs_msg(sci->sc_super, KERN_ERR,
2643 "error %d creating segctord thread", err);
2646 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2650 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2651 __acquires(&sci->sc_state_lock)
2652 __releases(&sci->sc_state_lock)
2654 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2656 while (sci->sc_task) {
2657 wake_up(&sci->sc_wait_daemon);
2658 spin_unlock(&sci->sc_state_lock);
2659 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2660 spin_lock(&sci->sc_state_lock);
2665 * Setup & clean-up functions
2667 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2668 struct nilfs_root *root)
2670 struct the_nilfs *nilfs = sb->s_fs_info;
2671 struct nilfs_sc_info *sci;
2673 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2679 nilfs_get_root(root);
2680 sci->sc_root = root;
2682 init_waitqueue_head(&sci->sc_wait_request);
2683 init_waitqueue_head(&sci->sc_wait_daemon);
2684 init_waitqueue_head(&sci->sc_wait_task);
2685 spin_lock_init(&sci->sc_state_lock);
2686 INIT_LIST_HEAD(&sci->sc_dirty_files);
2687 INIT_LIST_HEAD(&sci->sc_segbufs);
2688 INIT_LIST_HEAD(&sci->sc_write_logs);
2689 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2690 INIT_LIST_HEAD(&sci->sc_iput_queue);
2691 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2692 init_timer(&sci->sc_timer);
2694 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2695 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2696 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2698 if (nilfs->ns_interval)
2699 sci->sc_interval = HZ * nilfs->ns_interval;
2700 if (nilfs->ns_watermark)
2701 sci->sc_watermark = nilfs->ns_watermark;
2705 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2707 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2710 * The segctord thread was stopped and its timer was removed.
2711 * But some tasks remain.
2714 struct nilfs_transaction_info ti;
2716 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2717 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2718 nilfs_transaction_unlock(sci->sc_super);
2720 flush_work(&sci->sc_iput_work);
2722 } while (ret && retrycount-- > 0);
2726 * nilfs_segctor_destroy - destroy the segment constructor.
2727 * @sci: nilfs_sc_info
2729 * nilfs_segctor_destroy() kills the segctord thread and frees
2730 * the nilfs_sc_info struct.
2731 * Caller must hold the segment semaphore.
2733 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2735 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2738 up_write(&nilfs->ns_segctor_sem);
2740 spin_lock(&sci->sc_state_lock);
2741 nilfs_segctor_kill_thread(sci);
2742 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2743 || sci->sc_seq_request != sci->sc_seq_done);
2744 spin_unlock(&sci->sc_state_lock);
2746 if (flush_work(&sci->sc_iput_work))
2749 if (flag || !nilfs_segctor_confirm(sci))
2750 nilfs_segctor_write_out(sci);
2752 if (!list_empty(&sci->sc_dirty_files)) {
2753 nilfs_msg(sci->sc_super, KERN_WARNING,
2754 "disposed unprocessed dirty file(s) when stopping log writer");
2755 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2758 if (!list_empty(&sci->sc_iput_queue)) {
2759 nilfs_msg(sci->sc_super, KERN_WARNING,
2760 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2761 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2764 WARN_ON(!list_empty(&sci->sc_segbufs));
2765 WARN_ON(!list_empty(&sci->sc_write_logs));
2767 nilfs_put_root(sci->sc_root);
2769 down_write(&nilfs->ns_segctor_sem);
2771 del_timer_sync(&sci->sc_timer);
2776 * nilfs_attach_log_writer - attach log writer
2777 * @sb: super block instance
2778 * @root: root object of the current filesystem tree
2780 * This allocates a log writer object, initializes it, and starts the
2783 * Return Value: On success, 0 is returned. On error, one of the following
2784 * negative error code is returned.
2786 * %-ENOMEM - Insufficient memory available.
2788 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2790 struct the_nilfs *nilfs = sb->s_fs_info;
2793 if (nilfs->ns_writer) {
2795 * This happens if the filesystem is made read-only by
2796 * __nilfs_error or nilfs_remount and then remounted
2797 * read/write. In these cases, reuse the existing
2803 nilfs->ns_writer = nilfs_segctor_new(sb, root);
2804 if (!nilfs->ns_writer)
2807 inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2809 err = nilfs_segctor_start_thread(nilfs->ns_writer);
2811 nilfs_detach_log_writer(sb);
2817 * nilfs_detach_log_writer - destroy log writer
2818 * @sb: super block instance
2820 * This kills log writer daemon, frees the log writer object, and
2821 * destroys list of dirty files.
2823 void nilfs_detach_log_writer(struct super_block *sb)
2825 struct the_nilfs *nilfs = sb->s_fs_info;
2826 LIST_HEAD(garbage_list);
2828 down_write(&nilfs->ns_segctor_sem);
2829 if (nilfs->ns_writer) {
2830 nilfs_segctor_destroy(nilfs->ns_writer);
2831 nilfs->ns_writer = NULL;
2834 /* Force to free the list of dirty files */
2835 spin_lock(&nilfs->ns_inode_lock);
2836 if (!list_empty(&nilfs->ns_dirty_files)) {
2837 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2838 nilfs_msg(sb, KERN_WARNING,
2839 "disposed unprocessed dirty file(s) when detaching log writer");
2841 spin_unlock(&nilfs->ns_inode_lock);
2842 up_write(&nilfs->ns_segctor_sem);
2844 nilfs_dispose_list(nilfs, &garbage_list, 1);