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 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
24 #include <linux/pagemap.h>
25 #include <linux/buffer_head.h>
26 #include <linux/writeback.h>
27 #include <linux/bio.h>
28 #include <linux/completion.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/freezer.h>
32 #include <linux/kthread.h>
33 #include <linux/crc32.h>
34 #include <linux/pagevec.h>
49 #define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
51 #define SC_MAX_SEGDELTA 64 /* Upper limit of the number of segments
52 appended in collection retry loop */
54 /* Construction mode */
56 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
57 SC_LSEG_DSYNC, /* Flush data blocks of a given file and make
58 a logical segment without a super root */
59 SC_FLUSH_FILE, /* Flush data files, leads to segment writes without
60 creating a checkpoint */
61 SC_FLUSH_DAT, /* Flush DAT file. This also creates segments without
65 /* Stage numbers of dirty block collection */
68 NILFS_ST_GC, /* Collecting dirty blocks for GC */
75 NILFS_ST_SR, /* Super root */
76 NILFS_ST_DSYNC, /* Data sync blocks */
80 /* State flags of collection */
81 #define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
82 #define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
83 #define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED)
85 /* Operations depending on the construction mode and file type */
86 struct nilfs_sc_operations {
87 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
89 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
91 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
93 void (*write_data_binfo)(struct nilfs_sc_info *,
94 struct nilfs_segsum_pointer *,
96 void (*write_node_binfo)(struct nilfs_sc_info *,
97 struct nilfs_segsum_pointer *,
104 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
105 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
107 static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
110 #define nilfs_cnt32_gt(a, b) \
111 (typecheck(__u32, a) && typecheck(__u32, b) && \
112 ((__s32)(b) - (__s32)(a) < 0))
113 #define nilfs_cnt32_ge(a, b) \
114 (typecheck(__u32, a) && typecheck(__u32, b) && \
115 ((__s32)(a) - (__s32)(b) >= 0))
116 #define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
117 #define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
122 static struct kmem_cache *nilfs_transaction_cachep;
125 * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
127 * nilfs_init_transaction_cache() creates a slab cache for the struct
128 * nilfs_transaction_info.
130 * Return Value: On success, it returns 0. On error, one of the following
131 * negative error code is returned.
133 * %-ENOMEM - Insufficient memory available.
135 int nilfs_init_transaction_cache(void)
137 nilfs_transaction_cachep =
138 kmem_cache_create("nilfs2_transaction_cache",
139 sizeof(struct nilfs_transaction_info),
140 0, SLAB_RECLAIM_ACCOUNT, NULL);
141 return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
145 * nilfs_detroy_transaction_cache - destroy the cache for transaction info
147 * nilfs_destroy_transaction_cache() frees the slab cache for the struct
148 * nilfs_transaction_info.
150 void nilfs_destroy_transaction_cache(void)
152 kmem_cache_destroy(nilfs_transaction_cachep);
155 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
157 struct nilfs_transaction_info *cur_ti = current->journal_info;
161 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
162 return ++cur_ti->ti_count;
165 * If journal_info field is occupied by other FS,
166 * it is saved and will be restored on
167 * nilfs_transaction_commit().
170 "NILFS warning: journal info from a different "
172 save = current->journal_info;
176 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
179 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
185 ti->ti_magic = NILFS_TI_MAGIC;
186 current->journal_info = ti;
191 * nilfs_transaction_begin - start indivisible file operations.
193 * @ti: nilfs_transaction_info
194 * @vacancy_check: flags for vacancy rate checks
196 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
197 * the segment semaphore, to make a segment construction and write tasks
198 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
199 * The region enclosed by these two functions can be nested. To avoid a
200 * deadlock, the semaphore is only acquired or released in the outermost call.
202 * This function allocates a nilfs_transaction_info struct to keep context
203 * information on it. It is initialized and hooked onto the current task in
204 * the outermost call. If a pre-allocated struct is given to @ti, it is used
205 * instead; othewise a new struct is assigned from a slab.
207 * When @vacancy_check flag is set, this function will check the amount of
208 * free space, and will wait for the GC to reclaim disk space if low capacity.
210 * Return Value: On success, 0 is returned. On error, one of the following
211 * negative error code is returned.
213 * %-ENOMEM - Insufficient memory available.
215 * %-ENOSPC - No space left on device
217 int nilfs_transaction_begin(struct super_block *sb,
218 struct nilfs_transaction_info *ti,
221 struct nilfs_sb_info *sbi;
222 struct the_nilfs *nilfs;
223 int ret = nilfs_prepare_segment_lock(ti);
225 if (unlikely(ret < 0))
231 nilfs = sbi->s_nilfs;
232 down_read(&nilfs->ns_segctor_sem);
233 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
234 up_read(&nilfs->ns_segctor_sem);
241 ti = current->journal_info;
242 current->journal_info = ti->ti_save;
243 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
244 kmem_cache_free(nilfs_transaction_cachep, ti);
249 * nilfs_transaction_commit - commit indivisible file operations.
252 * nilfs_transaction_commit() releases the read semaphore which is
253 * acquired by nilfs_transaction_begin(). This is only performed
254 * in outermost call of this function. If a commit flag is set,
255 * nilfs_transaction_commit() sets a timer to start the segment
256 * constructor. If a sync flag is set, it starts construction
259 int nilfs_transaction_commit(struct super_block *sb)
261 struct nilfs_transaction_info *ti = current->journal_info;
262 struct nilfs_sb_info *sbi;
263 struct nilfs_sc_info *sci;
266 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
267 ti->ti_flags |= NILFS_TI_COMMIT;
268 if (ti->ti_count > 0) {
275 if (ti->ti_flags & NILFS_TI_COMMIT)
276 nilfs_segctor_start_timer(sci);
277 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
279 nilfs_segctor_do_flush(sci, 0);
281 up_read(&sbi->s_nilfs->ns_segctor_sem);
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);
291 void nilfs_transaction_abort(struct super_block *sb)
293 struct nilfs_transaction_info *ti = current->journal_info;
295 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
296 if (ti->ti_count > 0) {
300 up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
302 current->journal_info = ti->ti_save;
303 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
304 kmem_cache_free(nilfs_transaction_cachep, ti);
307 void nilfs_relax_pressure_in_lock(struct super_block *sb)
309 struct nilfs_sb_info *sbi = NILFS_SB(sb);
310 struct nilfs_sc_info *sci = NILFS_SC(sbi);
311 struct the_nilfs *nilfs = sbi->s_nilfs;
313 if (!sci || !sci->sc_flush_request)
316 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
317 up_read(&nilfs->ns_segctor_sem);
319 down_write(&nilfs->ns_segctor_sem);
320 if (sci->sc_flush_request &&
321 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
322 struct nilfs_transaction_info *ti = current->journal_info;
324 ti->ti_flags |= NILFS_TI_WRITER;
325 nilfs_segctor_do_immediate_flush(sci);
326 ti->ti_flags &= ~NILFS_TI_WRITER;
328 downgrade_write(&nilfs->ns_segctor_sem);
331 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
332 struct nilfs_transaction_info *ti,
335 struct nilfs_transaction_info *cur_ti = current->journal_info;
339 ti->ti_flags = NILFS_TI_WRITER;
341 ti->ti_save = cur_ti;
342 ti->ti_magic = NILFS_TI_MAGIC;
343 INIT_LIST_HEAD(&ti->ti_garbage);
344 current->journal_info = ti;
347 down_write(&sbi->s_nilfs->ns_segctor_sem);
348 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
351 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
353 up_write(&sbi->s_nilfs->ns_segctor_sem);
357 ti->ti_flags |= NILFS_TI_GC;
360 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
362 struct nilfs_transaction_info *ti = current->journal_info;
364 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
365 BUG_ON(ti->ti_count > 0);
367 up_write(&sbi->s_nilfs->ns_segctor_sem);
368 current->journal_info = ti->ti_save;
369 if (!list_empty(&ti->ti_garbage))
370 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
373 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
374 struct nilfs_segsum_pointer *ssp,
377 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
378 unsigned blocksize = sci->sc_super->s_blocksize;
381 if (unlikely(ssp->offset + bytes > blocksize)) {
383 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
384 &segbuf->sb_segsum_buffers));
385 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
387 p = ssp->bh->b_data + ssp->offset;
388 ssp->offset += bytes;
393 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
394 * @sci: nilfs_sc_info
396 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
398 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
399 struct buffer_head *sumbh;
404 if (nilfs_doing_gc())
406 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
410 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
411 sumbytes = segbuf->sb_sum.sumbytes;
412 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
413 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
414 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
418 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
420 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
421 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
422 return -E2BIG; /* The current segment is filled up
424 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
425 return nilfs_segctor_reset_segment_buffer(sci);
428 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
430 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
433 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
434 err = nilfs_segctor_feed_segment(sci);
437 segbuf = sci->sc_curseg;
439 err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
441 segbuf->sb_sum.flags |= NILFS_SS_SR;
446 * Functions for making segment summary and payloads
448 static int nilfs_segctor_segsum_block_required(
449 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
452 unsigned blocksize = sci->sc_super->s_blocksize;
453 /* Size of finfo and binfo is enough small against blocksize */
455 return ssp->offset + binfo_size +
456 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
460 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
463 sci->sc_curseg->sb_sum.nfinfo++;
464 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
465 nilfs_segctor_map_segsum_entry(
466 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
470 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
473 struct nilfs_finfo *finfo;
474 struct nilfs_inode_info *ii;
475 struct nilfs_segment_buffer *segbuf;
477 if (sci->sc_blk_cnt == 0)
481 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
483 finfo->fi_ino = cpu_to_le64(inode->i_ino);
484 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
485 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
486 finfo->fi_cno = cpu_to_le64(ii->i_cno);
488 segbuf = sci->sc_curseg;
489 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
490 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
491 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
492 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
495 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
496 struct buffer_head *bh,
500 struct nilfs_segment_buffer *segbuf;
501 int required, err = 0;
504 segbuf = sci->sc_curseg;
505 required = nilfs_segctor_segsum_block_required(
506 sci, &sci->sc_binfo_ptr, binfo_size);
507 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
508 nilfs_segctor_end_finfo(sci, inode);
509 err = nilfs_segctor_feed_segment(sci);
514 if (unlikely(required)) {
515 err = nilfs_segbuf_extend_segsum(segbuf);
519 if (sci->sc_blk_cnt == 0)
520 nilfs_segctor_begin_finfo(sci, inode);
522 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
523 /* Substitution to vblocknr is delayed until update_blocknr() */
524 nilfs_segbuf_add_file_buffer(segbuf, bh);
530 static int nilfs_handle_bmap_error(int err, const char *fname,
531 struct inode *inode, struct super_block *sb)
533 if (err == -EINVAL) {
534 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
542 * Callback functions that enumerate, mark, and collect dirty blocks
544 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
545 struct buffer_head *bh, struct inode *inode)
549 /* BUG_ON(!buffer_dirty(bh)); */
550 /* excluded by scan_dirty_data_buffers() */
551 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
552 if (unlikely(err < 0))
553 return nilfs_handle_bmap_error(err, __func__, inode,
556 err = nilfs_segctor_add_file_block(sci, bh, inode,
557 sizeof(struct nilfs_binfo_v));
559 sci->sc_datablk_cnt++;
563 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
564 struct buffer_head *bh,
569 /* BUG_ON(!buffer_dirty(bh)); */
570 /* excluded by scan_dirty_node_buffers() */
571 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
572 if (unlikely(err < 0))
573 return nilfs_handle_bmap_error(err, __func__, inode,
578 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
579 struct buffer_head *bh,
582 BUG_ON(!buffer_dirty(bh));
583 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
586 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
587 struct nilfs_segsum_pointer *ssp,
588 union nilfs_binfo *binfo)
590 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
591 sci, ssp, sizeof(*binfo_v));
592 *binfo_v = binfo->bi_v;
595 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
596 struct nilfs_segsum_pointer *ssp,
597 union nilfs_binfo *binfo)
599 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
600 sci, ssp, sizeof(*vblocknr));
601 *vblocknr = binfo->bi_v.bi_vblocknr;
604 struct nilfs_sc_operations nilfs_sc_file_ops = {
605 .collect_data = nilfs_collect_file_data,
606 .collect_node = nilfs_collect_file_node,
607 .collect_bmap = nilfs_collect_file_bmap,
608 .write_data_binfo = nilfs_write_file_data_binfo,
609 .write_node_binfo = nilfs_write_file_node_binfo,
612 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
613 struct buffer_head *bh, struct inode *inode)
617 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
618 if (unlikely(err < 0))
619 return nilfs_handle_bmap_error(err, __func__, inode,
622 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
624 sci->sc_datablk_cnt++;
628 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
629 struct buffer_head *bh, struct inode *inode)
631 BUG_ON(!buffer_dirty(bh));
632 return nilfs_segctor_add_file_block(sci, bh, inode,
633 sizeof(struct nilfs_binfo_dat));
636 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
637 struct nilfs_segsum_pointer *ssp,
638 union nilfs_binfo *binfo)
640 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
642 *blkoff = binfo->bi_dat.bi_blkoff;
645 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
646 struct nilfs_segsum_pointer *ssp,
647 union nilfs_binfo *binfo)
649 struct nilfs_binfo_dat *binfo_dat =
650 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
651 *binfo_dat = binfo->bi_dat;
654 struct nilfs_sc_operations nilfs_sc_dat_ops = {
655 .collect_data = nilfs_collect_dat_data,
656 .collect_node = nilfs_collect_file_node,
657 .collect_bmap = nilfs_collect_dat_bmap,
658 .write_data_binfo = nilfs_write_dat_data_binfo,
659 .write_node_binfo = nilfs_write_dat_node_binfo,
662 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
663 .collect_data = nilfs_collect_file_data,
664 .collect_node = NULL,
665 .collect_bmap = NULL,
666 .write_data_binfo = nilfs_write_file_data_binfo,
667 .write_node_binfo = NULL,
670 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
671 struct list_head *listp,
673 loff_t start, loff_t end)
675 struct address_space *mapping = inode->i_mapping;
677 pgoff_t index = 0, last = ULONG_MAX;
681 if (unlikely(start != 0 || end != LLONG_MAX)) {
683 * A valid range is given for sync-ing data pages. The
684 * range is rounded to per-page; extra dirty buffers
685 * may be included if blocksize < pagesize.
687 index = start >> PAGE_SHIFT;
688 last = end >> PAGE_SHIFT;
690 pagevec_init(&pvec, 0);
692 if (unlikely(index > last) ||
693 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
694 min_t(pgoff_t, last - index,
695 PAGEVEC_SIZE - 1) + 1))
698 for (i = 0; i < pagevec_count(&pvec); i++) {
699 struct buffer_head *bh, *head;
700 struct page *page = pvec.pages[i];
702 if (unlikely(page->index > last))
707 if (!page_has_buffers(page))
708 create_empty_buffers(page,
709 1 << inode->i_blkbits, 0);
713 bh = head = page_buffers(page);
715 if (!buffer_dirty(bh))
718 list_add_tail(&bh->b_assoc_buffers, listp);
720 if (unlikely(ndirties >= nlimit)) {
721 pagevec_release(&pvec);
725 } while (bh = bh->b_this_page, bh != head);
727 pagevec_release(&pvec);
732 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
733 struct list_head *listp)
735 struct nilfs_inode_info *ii = NILFS_I(inode);
736 struct address_space *mapping = &ii->i_btnode_cache;
738 struct buffer_head *bh, *head;
742 pagevec_init(&pvec, 0);
744 while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
746 for (i = 0; i < pagevec_count(&pvec); i++) {
747 bh = head = page_buffers(pvec.pages[i]);
749 if (buffer_dirty(bh)) {
751 list_add_tail(&bh->b_assoc_buffers,
754 bh = bh->b_this_page;
755 } while (bh != head);
757 pagevec_release(&pvec);
762 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
763 struct list_head *head, int force)
765 struct nilfs_inode_info *ii, *n;
766 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
769 while (!list_empty(head)) {
770 spin_lock(&sbi->s_inode_lock);
771 list_for_each_entry_safe(ii, n, head, i_dirty) {
772 list_del_init(&ii->i_dirty);
774 if (unlikely(ii->i_bh)) {
778 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
779 set_bit(NILFS_I_QUEUED, &ii->i_state);
780 list_add_tail(&ii->i_dirty,
781 &sbi->s_dirty_files);
785 if (nv == SC_N_INODEVEC)
788 spin_unlock(&sbi->s_inode_lock);
790 for (pii = ivec; nv > 0; pii++, nv--)
791 iput(&(*pii)->vfs_inode);
795 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
797 struct the_nilfs *nilfs = sbi->s_nilfs;
800 if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
802 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
804 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
806 if (ret || nilfs_doing_gc())
807 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
812 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
814 return list_empty(&sci->sc_dirty_files) &&
815 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
816 list_empty(&sci->sc_cleaning_segments) &&
817 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
820 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
822 struct nilfs_sb_info *sbi = sci->sc_sbi;
825 if (nilfs_test_metadata_dirty(sbi))
826 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
828 spin_lock(&sbi->s_inode_lock);
829 if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
832 spin_unlock(&sbi->s_inode_lock);
836 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
838 struct nilfs_sb_info *sbi = sci->sc_sbi;
839 struct the_nilfs *nilfs = sbi->s_nilfs;
841 nilfs_mdt_clear_dirty(sbi->s_ifile);
842 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
843 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
844 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
847 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
849 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
850 struct buffer_head *bh_cp;
851 struct nilfs_checkpoint *raw_cp;
854 /* XXX: this interface will be changed */
855 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
858 /* The following code is duplicated with cpfile. But, it is
859 needed to collect the checkpoint even if it was not newly
861 nilfs_mdt_mark_buffer_dirty(bh_cp);
862 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
863 nilfs_cpfile_put_checkpoint(
864 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
866 BUG_ON(err == -EINVAL || err == -ENOENT);
871 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
873 struct nilfs_sb_info *sbi = sci->sc_sbi;
874 struct the_nilfs *nilfs = sbi->s_nilfs;
875 struct buffer_head *bh_cp;
876 struct nilfs_checkpoint *raw_cp;
879 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
882 BUG_ON(err == -EINVAL || err == -ENOENT);
885 raw_cp->cp_snapshot_list.ssl_next = 0;
886 raw_cp->cp_snapshot_list.ssl_prev = 0;
887 raw_cp->cp_inodes_count =
888 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
889 raw_cp->cp_blocks_count =
890 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
891 raw_cp->cp_nblk_inc =
892 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
893 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
894 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
895 if (sci->sc_sketch_inode && i_size_read(sci->sc_sketch_inode) > 0)
896 nilfs_checkpoint_set_sketch(raw_cp);
897 nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
898 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
905 static void nilfs_fill_in_file_bmap(struct inode *ifile,
906 struct nilfs_inode_info *ii)
909 struct buffer_head *ibh;
910 struct nilfs_inode *raw_inode;
912 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
915 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
917 nilfs_bmap_write(ii->i_bmap, raw_inode);
918 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
922 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
925 struct nilfs_inode_info *ii;
927 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
928 nilfs_fill_in_file_bmap(ifile, ii);
929 set_bit(NILFS_I_COLLECTED, &ii->i_state);
931 if (sci->sc_sketch_inode) {
932 ii = NILFS_I(sci->sc_sketch_inode);
933 if (test_bit(NILFS_I_DIRTY, &ii->i_state))
934 nilfs_fill_in_file_bmap(ifile, ii);
939 * CRC calculation routines
941 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
943 struct nilfs_super_root *raw_sr =
944 (struct nilfs_super_root *)bh_sr->b_data;
947 BUG_ON(NILFS_SR_BYTES > bh_sr->b_size);
949 (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
950 NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
951 raw_sr->sr_sum = cpu_to_le32(crc);
954 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
957 struct nilfs_segment_buffer *segbuf;
959 if (sci->sc_super_root)
960 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
962 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
963 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
964 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
968 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
969 struct the_nilfs *nilfs)
971 struct buffer_head *bh_sr = sci->sc_super_root;
972 struct nilfs_super_root *raw_sr =
973 (struct nilfs_super_root *)bh_sr->b_data;
974 unsigned isz = nilfs->ns_inode_size;
976 raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
977 raw_sr->sr_nongc_ctime
978 = cpu_to_le64(nilfs_doing_gc() ?
979 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
980 raw_sr->sr_flags = 0;
982 nilfs_mdt_write_inode_direct(
983 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
984 nilfs_mdt_write_inode_direct(
985 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
986 nilfs_mdt_write_inode_direct(
987 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
990 static void nilfs_redirty_inodes(struct list_head *head)
992 struct nilfs_inode_info *ii;
994 list_for_each_entry(ii, head, i_dirty) {
995 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
996 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1000 static void nilfs_drop_collected_inodes(struct list_head *head)
1002 struct nilfs_inode_info *ii;
1004 list_for_each_entry(ii, head, i_dirty) {
1005 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1008 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
1009 set_bit(NILFS_I_UPDATED, &ii->i_state);
1013 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
1014 struct inode *sufile)
1017 struct list_head *head = &sci->sc_cleaning_segments;
1018 struct nilfs_segment_entry *ent;
1021 list_for_each_entry(ent, head, list) {
1022 if (!(ent->flags & NILFS_SLH_FREED))
1024 err = nilfs_sufile_cancel_free(sufile, ent->segnum);
1027 ent->flags &= ~NILFS_SLH_FREED;
1031 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
1032 struct inode *sufile)
1034 struct list_head *head = &sci->sc_cleaning_segments;
1035 struct nilfs_segment_entry *ent;
1038 list_for_each_entry(ent, head, list) {
1039 err = nilfs_sufile_free(sufile, ent->segnum);
1042 ent->flags |= NILFS_SLH_FREED;
1047 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
1049 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
1052 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1053 struct inode *inode,
1054 struct list_head *listp,
1055 int (*collect)(struct nilfs_sc_info *,
1056 struct buffer_head *,
1059 struct buffer_head *bh, *n;
1063 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1064 list_del_init(&bh->b_assoc_buffers);
1065 err = collect(sci, bh, inode);
1068 goto dispose_buffers;
1074 while (!list_empty(listp)) {
1075 bh = list_entry(listp->next, struct buffer_head,
1077 list_del_init(&bh->b_assoc_buffers);
1083 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1085 /* Remaining number of blocks within segment buffer */
1086 return sci->sc_segbuf_nblocks -
1087 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1090 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1091 struct inode *inode,
1092 struct nilfs_sc_operations *sc_ops)
1094 LIST_HEAD(data_buffers);
1095 LIST_HEAD(node_buffers);
1098 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1099 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1101 n = nilfs_lookup_dirty_data_buffers(
1102 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1104 err = nilfs_segctor_apply_buffers(
1105 sci, inode, &data_buffers,
1106 sc_ops->collect_data);
1107 BUG_ON(!err); /* always receive -E2BIG or true error */
1111 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1113 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1114 err = nilfs_segctor_apply_buffers(
1115 sci, inode, &data_buffers, sc_ops->collect_data);
1116 if (unlikely(err)) {
1117 /* dispose node list */
1118 nilfs_segctor_apply_buffers(
1119 sci, inode, &node_buffers, NULL);
1122 sci->sc_stage.flags |= NILFS_CF_NODE;
1125 err = nilfs_segctor_apply_buffers(
1126 sci, inode, &node_buffers, sc_ops->collect_node);
1130 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1131 err = nilfs_segctor_apply_buffers(
1132 sci, inode, &node_buffers, sc_ops->collect_bmap);
1136 nilfs_segctor_end_finfo(sci, inode);
1137 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1143 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1144 struct inode *inode)
1146 LIST_HEAD(data_buffers);
1147 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1150 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1151 sci->sc_dsync_start,
1154 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1155 nilfs_collect_file_data);
1157 nilfs_segctor_end_finfo(sci, inode);
1159 /* always receive -E2BIG or true error if n > rest */
1164 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1166 struct nilfs_sb_info *sbi = sci->sc_sbi;
1167 struct the_nilfs *nilfs = sbi->s_nilfs;
1168 struct list_head *head;
1169 struct nilfs_inode_info *ii;
1172 switch (sci->sc_stage.scnt) {
1175 sci->sc_stage.flags = 0;
1177 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1178 sci->sc_nblk_inc = 0;
1179 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1180 if (mode == SC_LSEG_DSYNC) {
1181 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1186 sci->sc_stage.dirty_file_ptr = NULL;
1187 sci->sc_stage.gc_inode_ptr = NULL;
1188 if (mode == SC_FLUSH_DAT) {
1189 sci->sc_stage.scnt = NILFS_ST_DAT;
1192 sci->sc_stage.scnt++; /* Fall through */
1194 if (nilfs_doing_gc()) {
1195 head = &sci->sc_gc_inodes;
1196 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1198 list_for_each_entry_continue(ii, head, i_dirty) {
1199 err = nilfs_segctor_scan_file(
1200 sci, &ii->vfs_inode,
1201 &nilfs_sc_file_ops);
1202 if (unlikely(err)) {
1203 sci->sc_stage.gc_inode_ptr = list_entry(
1205 struct nilfs_inode_info,
1209 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1211 sci->sc_stage.gc_inode_ptr = NULL;
1213 sci->sc_stage.scnt++; /* Fall through */
1215 head = &sci->sc_dirty_files;
1216 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1218 list_for_each_entry_continue(ii, head, i_dirty) {
1219 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1221 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1222 &nilfs_sc_file_ops);
1223 if (unlikely(err)) {
1224 sci->sc_stage.dirty_file_ptr =
1225 list_entry(ii->i_dirty.prev,
1226 struct nilfs_inode_info,
1230 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1231 /* XXX: required ? */
1233 sci->sc_stage.dirty_file_ptr = NULL;
1234 if (mode == SC_FLUSH_FILE) {
1235 sci->sc_stage.scnt = NILFS_ST_DONE;
1238 sci->sc_stage.scnt++; /* Fall through */
1239 case NILFS_ST_SKETCH:
1240 if (mode == SC_LSEG_SR && sci->sc_sketch_inode) {
1241 ii = NILFS_I(sci->sc_sketch_inode);
1242 if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
1243 sci->sc_sketch_inode->i_ctime.tv_sec
1244 = sci->sc_seg_ctime;
1245 sci->sc_sketch_inode->i_mtime.tv_sec
1246 = sci->sc_seg_ctime;
1247 err = nilfs_mark_inode_dirty(
1248 sci->sc_sketch_inode);
1252 err = nilfs_segctor_scan_file(sci,
1253 sci->sc_sketch_inode,
1254 &nilfs_sc_file_ops);
1258 sci->sc_stage.scnt++;
1259 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1261 case NILFS_ST_IFILE:
1262 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1263 &nilfs_sc_file_ops);
1266 sci->sc_stage.scnt++;
1267 /* Creating a checkpoint */
1268 err = nilfs_segctor_create_checkpoint(sci);
1272 case NILFS_ST_CPFILE:
1273 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1274 &nilfs_sc_file_ops);
1277 sci->sc_stage.scnt++; /* Fall through */
1278 case NILFS_ST_SUFILE:
1279 err = nilfs_segctor_prepare_free_segments(sci,
1283 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1284 &nilfs_sc_file_ops);
1287 sci->sc_stage.scnt++; /* Fall through */
1290 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1294 if (mode == SC_FLUSH_DAT) {
1295 sci->sc_stage.scnt = NILFS_ST_DONE;
1298 sci->sc_stage.scnt++; /* Fall through */
1300 if (mode == SC_LSEG_SR) {
1301 /* Appending a super root */
1302 err = nilfs_segctor_add_super_root(sci);
1306 /* End of a logical segment */
1307 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1308 sci->sc_stage.scnt = NILFS_ST_DONE;
1310 case NILFS_ST_DSYNC:
1312 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1313 ii = sci->sc_dsync_inode;
1314 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1317 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1320 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1321 sci->sc_stage.scnt = NILFS_ST_DONE;
1333 static int nilfs_segctor_terminate_segment(struct nilfs_sc_info *sci,
1334 struct nilfs_segment_buffer *segbuf,
1335 struct inode *sufile)
1337 struct nilfs_segment_entry *ent = segbuf->sb_segent;
1340 err = nilfs_open_segment_entry(ent, sufile);
1343 nilfs_mdt_mark_buffer_dirty(ent->bh_su);
1344 nilfs_mdt_mark_dirty(sufile);
1345 nilfs_close_segment_entry(ent, sufile);
1347 list_add_tail(&ent->list, &sci->sc_active_segments);
1348 segbuf->sb_segent = NULL;
1352 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1354 struct buffer_head *bh_su;
1355 struct nilfs_segment_usage *raw_su;
1358 err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1361 nilfs_mdt_mark_buffer_dirty(bh_su);
1362 nilfs_mdt_mark_dirty(sufile);
1363 nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1367 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1368 struct the_nilfs *nilfs)
1370 struct nilfs_segment_buffer *segbuf, *n;
1371 struct inode *sufile = nilfs->ns_sufile;
1375 if (list_empty(&sci->sc_segbufs)) {
1376 segbuf = nilfs_segbuf_new(sci->sc_super);
1377 if (unlikely(!segbuf))
1379 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1381 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1383 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1384 nilfs->ns_pseg_offset, nilfs);
1388 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1389 err = nilfs_segctor_terminate_segment(sci, segbuf, sufile);
1393 nilfs_shift_to_next_segment(nilfs);
1394 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1396 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1398 err = nilfs_touch_segusage(sufile, segbuf->sb_segnum);
1402 if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1403 /* Start from the head of a new full segment */
1404 err = nilfs_sufile_alloc(sufile, &nextnum);
1408 nextnum = nilfs->ns_nextnum;
1410 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1411 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1413 /* truncating segment buffers */
1414 list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1416 list_del_init(&segbuf->sb_list);
1417 nilfs_segbuf_free(segbuf);
1422 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1423 struct the_nilfs *nilfs, int nadd)
1425 struct nilfs_segment_buffer *segbuf, *prev, *n;
1426 struct inode *sufile = nilfs->ns_sufile;
1431 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1433 * Since the segment specified with nextnum might be allocated during
1434 * the previous construction, the buffer including its segusage may
1435 * not be dirty. The following call ensures that the buffer is dirty
1436 * and will pin the buffer on memory until the sufile is written.
1438 err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1442 for (i = 0; i < nadd; i++) {
1443 /* extend segment info */
1445 segbuf = nilfs_segbuf_new(sci->sc_super);
1446 if (unlikely(!segbuf))
1449 /* map this buffer to region of segment on-disk */
1450 err = nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1454 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1456 /* allocate the next next full segment */
1457 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1461 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1462 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1464 list_add_tail(&segbuf->sb_list, &list);
1467 list_splice(&list, sci->sc_segbufs.prev);
1471 nilfs_segbuf_free(segbuf);
1473 list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1474 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1476 list_del_init(&segbuf->sb_list);
1477 nilfs_segbuf_free(segbuf);
1482 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1483 struct the_nilfs *nilfs)
1485 struct nilfs_segment_buffer *segbuf;
1488 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1489 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1490 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1493 if (segbuf->sb_io_error) {
1494 /* Case 1: The first segment failed */
1495 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1496 /* Case 1a: Partial segment appended into an existing
1498 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1499 segbuf->sb_fseg_end);
1500 else /* Case 1b: New full segment */
1501 set_nilfs_discontinued(nilfs);
1505 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1506 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1508 if (!done && segbuf->sb_io_error) {
1509 if (segbuf->sb_segnum != nilfs->ns_nextnum)
1510 /* Case 2: extended segment (!= next) failed */
1511 nilfs_sufile_set_error(nilfs->ns_sufile,
1518 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1520 struct nilfs_segment_buffer *segbuf;
1522 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1523 nilfs_segbuf_clear(segbuf);
1524 sci->sc_super_root = NULL;
1527 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1529 struct nilfs_segment_buffer *segbuf;
1531 while (!list_empty(&sci->sc_segbufs)) {
1532 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1533 list_del_init(&segbuf->sb_list);
1534 nilfs_segbuf_free(segbuf);
1536 /* sci->sc_curseg = NULL; */
1539 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1540 struct the_nilfs *nilfs, int err)
1542 if (unlikely(err)) {
1543 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1544 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1546 nilfs_segctor_clear_segment_buffers(sci);
1549 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1550 struct inode *sufile)
1552 struct nilfs_segment_buffer *segbuf;
1553 struct buffer_head *bh_su;
1554 struct nilfs_segment_usage *raw_su;
1555 unsigned long live_blocks;
1558 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1559 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1561 BUG_ON(ret); /* always succeed because bh_su is dirty */
1562 live_blocks = segbuf->sb_sum.nblocks +
1563 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1564 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1565 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1566 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1571 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1572 struct inode *sufile)
1574 struct nilfs_segment_buffer *segbuf;
1575 struct buffer_head *bh_su;
1576 struct nilfs_segment_usage *raw_su;
1579 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1580 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1582 BUG_ON(ret); /* always succeed because bh_su is dirty */
1583 raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1584 segbuf->sb_fseg_start);
1585 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1587 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1588 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1590 BUG_ON(ret); /* always succeed */
1591 raw_su->su_nblocks = 0;
1592 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1597 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1598 struct nilfs_segment_buffer *last,
1599 struct inode *sufile)
1601 struct nilfs_segment_buffer *segbuf = last, *n;
1604 list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1606 list_del_init(&segbuf->sb_list);
1607 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1608 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1610 nilfs_segbuf_free(segbuf);
1615 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1616 struct the_nilfs *nilfs, int mode)
1618 struct nilfs_cstage prev_stage = sci->sc_stage;
1621 /* Collection retry loop */
1623 sci->sc_super_root = NULL;
1624 sci->sc_nblk_this_inc = 0;
1625 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1627 err = nilfs_segctor_reset_segment_buffer(sci);
1631 err = nilfs_segctor_collect_blocks(sci, mode);
1632 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1636 if (unlikely(err != -E2BIG))
1639 /* The current segment is filled up */
1640 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1643 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1644 nilfs_segctor_clear_segment_buffers(sci);
1646 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1650 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1651 sci->sc_stage = prev_stage;
1653 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1660 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1661 struct buffer_head *new_bh)
1663 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1665 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1666 /* The caller must release old_bh */
1670 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1671 struct nilfs_segment_buffer *segbuf,
1674 struct inode *inode = NULL;
1676 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1677 unsigned long nblocks = 0, ndatablk = 0;
1678 struct nilfs_sc_operations *sc_op = NULL;
1679 struct nilfs_segsum_pointer ssp;
1680 struct nilfs_finfo *finfo = NULL;
1681 union nilfs_binfo binfo;
1682 struct buffer_head *bh, *bh_org;
1689 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1690 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1691 ssp.offset = sizeof(struct nilfs_segment_summary);
1693 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1694 if (bh == sci->sc_super_root)
1697 finfo = nilfs_segctor_map_segsum_entry(
1698 sci, &ssp, sizeof(*finfo));
1699 ino = le64_to_cpu(finfo->fi_ino);
1700 nblocks = le32_to_cpu(finfo->fi_nblocks);
1701 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1703 if (buffer_nilfs_node(bh))
1704 inode = NILFS_BTNC_I(bh->b_page->mapping);
1706 inode = NILFS_AS_I(bh->b_page->mapping);
1708 if (mode == SC_LSEG_DSYNC)
1709 sc_op = &nilfs_sc_dsync_ops;
1710 else if (ino == NILFS_DAT_INO)
1711 sc_op = &nilfs_sc_dat_ops;
1712 else /* file blocks */
1713 sc_op = &nilfs_sc_file_ops;
1717 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1720 nilfs_list_replace_buffer(bh_org, bh);
1726 sc_op->write_data_binfo(sci, &ssp, &binfo);
1728 sc_op->write_node_binfo(sci, &ssp, &binfo);
1731 if (--nblocks == 0) {
1735 } else if (ndatablk > 0)
1742 err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1746 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1748 struct nilfs_segment_buffer *segbuf;
1751 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1752 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1755 nilfs_segbuf_fill_in_segsum(segbuf);
1761 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1763 struct page *clone_page;
1764 struct buffer_head *bh, *head, *bh2;
1767 bh = head = page_buffers(page);
1769 clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1770 if (unlikely(!clone_page))
1773 bh2 = page_buffers(clone_page);
1774 kaddr = kmap_atomic(page, KM_USER0);
1776 if (list_empty(&bh->b_assoc_buffers))
1779 page_cache_get(clone_page); /* for each bh */
1780 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1781 bh2->b_blocknr = bh->b_blocknr;
1782 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1783 list_add_tail(&bh->b_assoc_buffers, out);
1784 } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1785 kunmap_atomic(kaddr, KM_USER0);
1787 if (!TestSetPageWriteback(clone_page))
1788 inc_zone_page_state(clone_page, NR_WRITEBACK);
1789 unlock_page(clone_page);
1794 static int nilfs_test_page_to_be_frozen(struct page *page)
1796 struct address_space *mapping = page->mapping;
1798 if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1801 if (page_mapped(page)) {
1802 ClearPageChecked(page);
1805 return PageChecked(page);
1808 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1810 if (!page || PageWriteback(page))
1811 /* For split b-tree node pages, this function may be called
1812 twice. We ignore the 2nd or later calls by this check. */
1816 clear_page_dirty_for_io(page);
1817 set_page_writeback(page);
1820 if (nilfs_test_page_to_be_frozen(page)) {
1821 int err = nilfs_copy_replace_page_buffers(page, out);
1828 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1829 struct page **failed_page)
1831 struct nilfs_segment_buffer *segbuf;
1832 struct page *bd_page = NULL, *fs_page = NULL;
1833 struct list_head *list = &sci->sc_copied_buffers;
1836 *failed_page = NULL;
1837 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1838 struct buffer_head *bh;
1840 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1842 if (bh->b_page != bd_page) {
1845 clear_page_dirty_for_io(bd_page);
1846 set_page_writeback(bd_page);
1847 unlock_page(bd_page);
1849 bd_page = bh->b_page;
1853 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1855 if (bh == sci->sc_super_root) {
1856 if (bh->b_page != bd_page) {
1858 clear_page_dirty_for_io(bd_page);
1859 set_page_writeback(bd_page);
1860 unlock_page(bd_page);
1861 bd_page = bh->b_page;
1865 if (bh->b_page != fs_page) {
1866 err = nilfs_begin_page_io(fs_page, list);
1867 if (unlikely(err)) {
1868 *failed_page = fs_page;
1871 fs_page = bh->b_page;
1877 clear_page_dirty_for_io(bd_page);
1878 set_page_writeback(bd_page);
1879 unlock_page(bd_page);
1881 err = nilfs_begin_page_io(fs_page, list);
1883 *failed_page = fs_page;
1888 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1889 struct backing_dev_info *bdi)
1891 struct nilfs_segment_buffer *segbuf;
1892 struct nilfs_write_info wi;
1895 wi.sb = sci->sc_super;
1896 wi.bh_sr = sci->sc_super_root;
1899 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1900 nilfs_segbuf_prepare_write(segbuf, &wi);
1901 err = nilfs_segbuf_write(segbuf, &wi);
1903 res = nilfs_segbuf_wait(segbuf, &wi);
1904 err = unlikely(err) ? : res;
1911 static int nilfs_page_has_uncleared_buffer(struct page *page)
1913 struct buffer_head *head, *bh;
1915 head = bh = page_buffers(page);
1917 if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
1919 bh = bh->b_this_page;
1920 } while (bh != head);
1924 static void __nilfs_end_page_io(struct page *page, int err)
1926 /* BUG_ON(err > 0); */
1928 if (!nilfs_page_buffers_clean(page))
1929 __set_page_dirty_nobuffers(page);
1930 ClearPageError(page);
1932 __set_page_dirty_nobuffers(page);
1936 if (buffer_nilfs_allocated(page_buffers(page))) {
1937 if (TestClearPageWriteback(page))
1938 dec_zone_page_state(page, NR_WRITEBACK);
1940 end_page_writeback(page);
1943 static void nilfs_end_page_io(struct page *page, int err)
1948 if (buffer_nilfs_node(page_buffers(page)) &&
1949 nilfs_page_has_uncleared_buffer(page))
1950 /* For b-tree node pages, this function may be called twice
1951 or more because they might be split in a segment.
1952 This check assures that cleanup has been done for all
1953 buffers in a split btnode page. */
1956 __nilfs_end_page_io(page, err);
1959 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1961 struct buffer_head *bh, *head;
1964 while (!list_empty(list)) {
1965 bh = list_entry(list->next, struct buffer_head,
1968 page_cache_get(page);
1969 head = bh = page_buffers(page);
1971 if (!list_empty(&bh->b_assoc_buffers)) {
1972 list_del_init(&bh->b_assoc_buffers);
1974 set_buffer_uptodate(bh);
1975 clear_buffer_dirty(bh);
1976 clear_buffer_nilfs_volatile(bh);
1978 brelse(bh); /* for b_assoc_buffers */
1980 } while ((bh = bh->b_this_page) != head);
1982 __nilfs_end_page_io(page, err);
1983 page_cache_release(page);
1987 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1988 struct page *failed_page, int err)
1990 struct nilfs_segment_buffer *segbuf;
1991 struct page *bd_page = NULL, *fs_page = NULL;
1993 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1994 struct buffer_head *bh;
1996 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1998 if (bh->b_page != bd_page) {
2000 end_page_writeback(bd_page);
2001 bd_page = bh->b_page;
2005 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2007 if (bh == sci->sc_super_root) {
2008 if (bh->b_page != bd_page) {
2009 end_page_writeback(bd_page);
2010 bd_page = bh->b_page;
2014 if (bh->b_page != fs_page) {
2015 nilfs_end_page_io(fs_page, err);
2016 if (unlikely(fs_page == failed_page))
2018 fs_page = bh->b_page;
2023 end_page_writeback(bd_page);
2025 nilfs_end_page_io(fs_page, err);
2027 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
2030 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
2031 struct nilfs_segment_buffer *segbuf)
2033 nilfs->ns_segnum = segbuf->sb_segnum;
2034 nilfs->ns_nextnum = segbuf->sb_nextnum;
2035 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
2036 + segbuf->sb_sum.nblocks;
2037 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
2038 nilfs->ns_ctime = segbuf->sb_sum.ctime;
2041 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
2043 struct nilfs_segment_buffer *segbuf;
2044 struct page *bd_page = NULL, *fs_page = NULL;
2045 struct nilfs_sb_info *sbi = sci->sc_sbi;
2046 struct the_nilfs *nilfs = sbi->s_nilfs;
2047 int update_sr = (sci->sc_super_root != NULL);
2049 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
2050 struct buffer_head *bh;
2052 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
2054 set_buffer_uptodate(bh);
2055 clear_buffer_dirty(bh);
2056 if (bh->b_page != bd_page) {
2058 end_page_writeback(bd_page);
2059 bd_page = bh->b_page;
2063 * We assume that the buffers which belong to the same page
2064 * continue over the buffer list.
2065 * Under this assumption, the last BHs of pages is
2066 * identifiable by the discontinuity of bh->b_page
2067 * (page != fs_page).
2069 * For B-tree node blocks, however, this assumption is not
2070 * guaranteed. The cleanup code of B-tree node pages needs
2073 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2075 set_buffer_uptodate(bh);
2076 clear_buffer_dirty(bh);
2077 clear_buffer_nilfs_volatile(bh);
2078 if (bh == sci->sc_super_root) {
2079 if (bh->b_page != bd_page) {
2080 end_page_writeback(bd_page);
2081 bd_page = bh->b_page;
2085 if (bh->b_page != fs_page) {
2086 nilfs_end_page_io(fs_page, 0);
2087 fs_page = bh->b_page;
2091 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2092 if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2093 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2094 sci->sc_lseg_stime = jiffies;
2096 if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2097 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2101 * Since pages may continue over multiple segment buffers,
2102 * end of the last page must be checked outside of the loop.
2105 end_page_writeback(bd_page);
2107 nilfs_end_page_io(fs_page, 0);
2109 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2111 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2113 if (nilfs_doing_gc()) {
2114 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2116 nilfs_commit_gcdat_inode(nilfs);
2118 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2119 set_nilfs_cond_nongc_write(nilfs);
2120 wake_up(&nilfs->ns_cleanerd_wq);
2123 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2125 segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2126 nilfs_set_next_segment(nilfs, segbuf);
2129 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2130 segbuf->sb_sum.seg_seq, nilfs->ns_cno);
2132 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2133 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2135 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2138 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2139 struct nilfs_sb_info *sbi)
2141 struct nilfs_inode_info *ii, *n;
2142 __u64 cno = sbi->s_nilfs->ns_cno;
2144 spin_lock(&sbi->s_inode_lock);
2146 list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2148 struct buffer_head *ibh;
2151 spin_unlock(&sbi->s_inode_lock);
2152 err = nilfs_ifile_get_inode_block(
2153 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2154 if (unlikely(err)) {
2155 nilfs_warning(sbi->s_super, __func__,
2156 "failed to get inode block.\n");
2159 nilfs_mdt_mark_buffer_dirty(ibh);
2160 nilfs_mdt_mark_dirty(sbi->s_ifile);
2161 spin_lock(&sbi->s_inode_lock);
2162 if (likely(!ii->i_bh))
2170 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2171 set_bit(NILFS_I_BUSY, &ii->i_state);
2172 list_del(&ii->i_dirty);
2173 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2175 spin_unlock(&sbi->s_inode_lock);
2177 NILFS_I(sbi->s_ifile)->i_cno = cno;
2182 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2183 struct nilfs_sb_info *sbi)
2185 struct nilfs_transaction_info *ti = current->journal_info;
2186 struct nilfs_inode_info *ii, *n;
2187 __u64 cno = sbi->s_nilfs->ns_cno;
2189 spin_lock(&sbi->s_inode_lock);
2190 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2191 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2192 test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2193 /* The current checkpoint number (=nilfs->ns_cno) is
2194 changed between check-in and check-out only if the
2195 super root is written out. So, we can update i_cno
2196 for the inodes that remain in the dirty list. */
2200 clear_bit(NILFS_I_BUSY, &ii->i_state);
2203 list_del(&ii->i_dirty);
2204 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2206 spin_unlock(&sbi->s_inode_lock);
2210 * Nasty routines to manipulate active flags on sufile.
2211 * These would be removed in a future release.
2213 static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info *sci,
2214 struct the_nilfs *nilfs)
2216 struct nilfs_segment_buffer *segbuf, *last;
2217 struct nilfs_segment_entry *ent, *n;
2218 struct inode *sufile = nilfs->ns_sufile;
2219 struct list_head *head;
2221 last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2222 nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2223 ent = segbuf->sb_segent;
2225 break; /* ignore unmapped segments (should check it?)*/
2226 nilfs_segment_usage_set_active(ent->raw_su);
2227 nilfs_close_segment_entry(ent, sufile);
2230 head = &sci->sc_active_segments;
2231 list_for_each_entry_safe(ent, n, head, list) {
2232 nilfs_segment_usage_set_active(ent->raw_su);
2233 nilfs_close_segment_entry(ent, sufile);
2236 down_write(&nilfs->ns_sem);
2237 head = &nilfs->ns_used_segments;
2238 list_for_each_entry(ent, head, list) {
2239 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2241 up_write(&nilfs->ns_sem);
2244 static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info *sci,
2245 struct the_nilfs *nilfs)
2247 struct nilfs_segment_buffer *segbuf, *last;
2248 struct nilfs_segment_entry *ent;
2249 struct inode *sufile = nilfs->ns_sufile;
2250 struct list_head *head;
2253 last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2254 nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2256 * Deactivate ongoing full segments. The last segment is kept
2257 * active because it is a start point of recovery, and is not
2258 * relocatable until the super block points to a newer
2261 ent = segbuf->sb_segent;
2263 break; /* ignore unmapped segments (should check it?)*/
2264 err = nilfs_open_segment_entry(ent, sufile);
2267 nilfs_segment_usage_clear_active(ent->raw_su);
2268 BUG_ON(!buffer_dirty(ent->bh_su));
2271 head = &sci->sc_active_segments;
2272 list_for_each_entry(ent, head, list) {
2273 err = nilfs_open_segment_entry(ent, sufile);
2276 nilfs_segment_usage_clear_active(ent->raw_su);
2277 BUG_ON(!buffer_dirty(ent->bh_su));
2280 down_write(&nilfs->ns_sem);
2281 head = &nilfs->ns_used_segments;
2282 list_for_each_entry(ent, head, list) {
2283 /* clear volatile active for segments of older generations */
2284 nilfs_segment_usage_clear_volatile_active(ent->raw_su);
2286 up_write(&nilfs->ns_sem);
2290 nilfs_segctor_reactivate_segments(sci, nilfs);
2294 static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info *sci)
2296 struct nilfs_segment_buffer *segbuf, *last;
2297 struct nilfs_segment_entry *ent;
2299 /* move each segbuf->sb_segent to the list of used active segments */
2300 last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2301 nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2302 ent = segbuf->sb_segent;
2304 break; /* ignore unmapped segments (should check it?)*/
2305 list_add_tail(&ent->list, &sci->sc_active_segments);
2306 segbuf->sb_segent = NULL;
2311 __nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info *sci,
2312 struct the_nilfs *nilfs)
2315 struct nilfs_segment_entry *ent;
2317 list_splice_init(&sci->sc_active_segments,
2318 nilfs->ns_used_segments.prev);
2320 list_for_each_entry(ent, &nilfs->ns_used_segments, list) {
2321 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2322 /* These segments are kept open */
2327 * Main procedure of segment constructor
2329 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2331 struct nilfs_sb_info *sbi = sci->sc_sbi;
2332 struct the_nilfs *nilfs = sbi->s_nilfs;
2333 struct page *failed_page;
2334 int err, has_sr = 0;
2336 sci->sc_stage.scnt = NILFS_ST_INIT;
2338 err = nilfs_segctor_check_in_files(sci, sbi);
2342 if (nilfs_test_metadata_dirty(sbi))
2343 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2345 if (nilfs_segctor_clean(sci))
2349 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2351 err = nilfs_segctor_begin_construction(sci, nilfs);
2355 /* Update time stamp */
2356 sci->sc_seg_ctime = get_seconds();
2358 err = nilfs_segctor_collect(sci, nilfs, mode);
2362 has_sr = (sci->sc_super_root != NULL);
2364 /* Avoid empty segment */
2365 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2366 NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2367 BUG_ON(mode == SC_LSEG_SR);
2368 nilfs_segctor_end_construction(sci, nilfs, 1);
2372 err = nilfs_segctor_assign(sci, mode);
2377 err = nilfs_segctor_deactivate_segments(sci, nilfs);
2381 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2382 nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2385 err = nilfs_segctor_fill_in_checkpoint(sci);
2387 goto failed_to_make_up;
2389 nilfs_segctor_fill_in_super_root(sci, nilfs);
2391 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2393 /* Write partial segments */
2394 err = nilfs_segctor_prepare_write(sci, &failed_page);
2396 goto failed_to_write;
2398 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2400 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2402 goto failed_to_write;
2404 nilfs_segctor_complete_write(sci);
2406 /* Commit segments */
2407 nilfs_segctor_bead_completed_segments(sci);
2409 down_write(&nilfs->ns_sem);
2410 nilfs_update_last_segment(sbi, 1);
2411 __nilfs_segctor_commit_deactivate_segments(sci, nilfs);
2412 up_write(&nilfs->ns_sem);
2413 nilfs_segctor_commit_free_segments(sci);
2414 nilfs_segctor_clear_metadata_dirty(sci);
2417 nilfs_segctor_end_construction(sci, nilfs, 0);
2419 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2421 /* Clearing sketch data */
2422 if (has_sr && sci->sc_sketch_inode) {
2423 if (i_size_read(sci->sc_sketch_inode) == 0)
2424 clear_bit(NILFS_I_DIRTY,
2425 &NILFS_I(sci->sc_sketch_inode)->i_state);
2426 i_size_write(sci->sc_sketch_inode, 0);
2429 nilfs_segctor_destroy_segment_buffers(sci);
2430 nilfs_segctor_check_out_files(sci, sbi);
2434 nilfs_segctor_abort_write(sci, failed_page, err);
2435 nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2438 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2439 nilfs_redirty_inodes(&sci->sc_dirty_files);
2441 nilfs_segctor_reactivate_segments(sci, nilfs);
2444 if (nilfs_doing_gc())
2445 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2446 nilfs_segctor_end_construction(sci, nilfs, err);
2451 * nilfs_secgtor_start_timer - set timer of background write
2452 * @sci: nilfs_sc_info
2454 * If the timer has already been set, it ignores the new request.
2455 * This function MUST be called within a section locking the segment
2458 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2460 spin_lock(&sci->sc_state_lock);
2461 if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2462 sci->sc_timer->expires = jiffies + sci->sc_interval;
2463 add_timer(sci->sc_timer);
2464 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2466 spin_unlock(&sci->sc_state_lock);
2469 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2471 spin_lock(&sci->sc_state_lock);
2472 if (!(sci->sc_flush_request & (1 << bn))) {
2473 unsigned long prev_req = sci->sc_flush_request;
2475 sci->sc_flush_request |= (1 << bn);
2477 wake_up(&sci->sc_wait_daemon);
2479 spin_unlock(&sci->sc_state_lock);
2483 * nilfs_flush_segment - trigger a segment construction for resource control
2485 * @ino: inode number of the file to be flushed out.
2487 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2489 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2490 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2492 if (!sci || nilfs_doing_construction())
2494 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2495 /* assign bit 0 to data files */
2498 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
2499 __u64 *segnum, size_t nsegs)
2501 struct nilfs_segment_entry *ent;
2502 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2503 struct inode *sufile = nilfs->ns_sufile;
2506 const char *flag_name;
2510 for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
2511 ent = nilfs_alloc_segment_entry(*pnum);
2512 if (unlikely(!ent)) {
2516 list_add_tail(&ent->list, &list);
2518 err = nilfs_open_segment_entry(ent, sufile);
2522 if (unlikely(le32_to_cpu(ent->raw_su->su_flags) !=
2523 (1UL << NILFS_SEGMENT_USAGE_DIRTY))) {
2524 if (nilfs_segment_usage_clean(ent->raw_su))
2525 flag_name = "clean";
2526 else if (nilfs_segment_usage_active(ent->raw_su))
2527 flag_name = "active";
2528 else if (nilfs_segment_usage_volatile_active(
2530 flag_name = "volatile active";
2531 else if (!nilfs_segment_usage_dirty(ent->raw_su))
2532 flag_name = "non-dirty";
2534 flag_name = "erroneous";
2537 "NILFS: %s segment is requested to be cleaned "
2539 flag_name, (unsigned long long)ent->segnum);
2542 nilfs_close_segment_entry(ent, sufile);
2544 if (unlikely(err2)) {
2548 list_splice(&list, sci->sc_cleaning_segments.prev);
2552 nilfs_dispose_segment_list(&list);
2556 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
2558 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2561 struct nilfs_segctor_wait_request {
2568 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2570 struct nilfs_segctor_wait_request wait_req;
2573 spin_lock(&sci->sc_state_lock);
2574 init_wait(&wait_req.wq);
2576 atomic_set(&wait_req.done, 0);
2577 wait_req.seq = ++sci->sc_seq_request;
2578 spin_unlock(&sci->sc_state_lock);
2580 init_waitqueue_entry(&wait_req.wq, current);
2581 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2582 set_current_state(TASK_INTERRUPTIBLE);
2583 wake_up(&sci->sc_wait_daemon);
2586 if (atomic_read(&wait_req.done)) {
2590 if (!signal_pending(current)) {
2597 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2601 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2603 struct nilfs_segctor_wait_request *wrq, *n;
2604 unsigned long flags;
2606 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2607 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2609 if (!atomic_read(&wrq->done) &&
2610 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2612 atomic_set(&wrq->done, 1);
2614 if (atomic_read(&wrq->done)) {
2615 wrq->wq.func(&wrq->wq,
2616 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2620 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2624 * nilfs_construct_segment - construct a logical segment
2627 * Return Value: On success, 0 is retured. On errors, one of the following
2628 * negative error code is returned.
2630 * %-EROFS - Read only filesystem.
2634 * %-ENOSPC - No space left on device (only in a panic state).
2636 * %-ERESTARTSYS - Interrupted.
2638 * %-ENOMEM - Insufficient memory available.
2640 int nilfs_construct_segment(struct super_block *sb)
2642 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2643 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2644 struct nilfs_transaction_info *ti;
2650 /* A call inside transactions causes a deadlock. */
2651 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2653 err = nilfs_segctor_sync(sci);
2658 * nilfs_construct_dsync_segment - construct a data-only logical segment
2660 * @inode: inode whose data blocks should be written out
2661 * @start: start byte offset
2662 * @end: end byte offset (inclusive)
2664 * Return Value: On success, 0 is retured. On errors, one of the following
2665 * negative error code is returned.
2667 * %-EROFS - Read only filesystem.
2671 * %-ENOSPC - No space left on device (only in a panic state).
2673 * %-ERESTARTSYS - Interrupted.
2675 * %-ENOMEM - Insufficient memory available.
2677 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2678 loff_t start, loff_t end)
2680 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2681 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2682 struct nilfs_inode_info *ii;
2683 struct nilfs_transaction_info ti;
2689 nilfs_transaction_lock(sbi, &ti, 0);
2691 ii = NILFS_I(inode);
2692 if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2693 nilfs_test_opt(sbi, STRICT_ORDER) ||
2694 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2695 nilfs_discontinued(sbi->s_nilfs)) {
2696 nilfs_transaction_unlock(sbi);
2697 err = nilfs_segctor_sync(sci);
2701 spin_lock(&sbi->s_inode_lock);
2702 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2703 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2704 spin_unlock(&sbi->s_inode_lock);
2705 nilfs_transaction_unlock(sbi);
2708 spin_unlock(&sbi->s_inode_lock);
2709 sci->sc_dsync_inode = ii;
2710 sci->sc_dsync_start = start;
2711 sci->sc_dsync_end = end;
2713 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2715 nilfs_transaction_unlock(sbi);
2719 struct nilfs_segctor_req {
2722 int sc_err; /* construction failure */
2723 int sb_err; /* super block writeback failure */
2726 #define FLUSH_FILE_BIT (0x1) /* data file only */
2727 #define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2729 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2730 struct nilfs_segctor_req *req)
2734 req->sc_err = req->sb_err = 0;
2735 spin_lock(&sci->sc_state_lock);
2736 req->seq_accepted = sci->sc_seq_request;
2737 spin_unlock(&sci->sc_state_lock);
2740 del_timer_sync(sci->sc_timer);
2743 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2744 struct nilfs_segctor_req *req)
2746 /* Clear requests (even when the construction failed) */
2747 spin_lock(&sci->sc_state_lock);
2749 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2751 if (req->mode == SC_LSEG_SR) {
2752 sci->sc_seq_done = req->seq_accepted;
2753 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2754 sci->sc_flush_request = 0;
2755 } else if (req->mode == SC_FLUSH_FILE)
2756 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2757 else if (req->mode == SC_FLUSH_DAT)
2758 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2760 spin_unlock(&sci->sc_state_lock);
2763 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2764 struct nilfs_segctor_req *req)
2766 struct nilfs_sb_info *sbi = sci->sc_sbi;
2767 struct the_nilfs *nilfs = sbi->s_nilfs;
2770 if (nilfs_discontinued(nilfs))
2771 req->mode = SC_LSEG_SR;
2772 if (!nilfs_segctor_confirm(sci)) {
2773 err = nilfs_segctor_do_construct(sci, req->mode);
2777 if (req->mode != SC_FLUSH_DAT)
2778 atomic_set(&nilfs->ns_ndirtyblks, 0);
2779 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2780 nilfs_discontinued(nilfs)) {
2781 down_write(&nilfs->ns_sem);
2782 req->sb_err = nilfs_commit_super(sbi);
2783 up_write(&nilfs->ns_sem);
2789 static void nilfs_construction_timeout(unsigned long data)
2791 struct task_struct *p = (struct task_struct *)data;
2796 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2798 struct nilfs_inode_info *ii, *n;
2800 list_for_each_entry_safe(ii, n, head, i_dirty) {
2801 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2803 hlist_del_init(&ii->vfs_inode.i_hash);
2804 list_del_init(&ii->i_dirty);
2805 nilfs_clear_gcinode(&ii->vfs_inode);
2809 int nilfs_clean_segments(struct super_block *sb, void __user *argp)
2811 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2812 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2813 struct the_nilfs *nilfs = sbi->s_nilfs;
2814 struct nilfs_transaction_info ti;
2815 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2821 nilfs_transaction_lock(sbi, &ti, 1);
2823 err = nilfs_init_gcdat_inode(nilfs);
2826 err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
2830 list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2833 nilfs_segctor_accept(sci, &req);
2834 err = nilfs_segctor_construct(sci, &req);
2835 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2836 nilfs_segctor_notify(sci, &req);
2841 nilfs_warning(sb, __func__,
2842 "segment construction failed. (err=%d)", err);
2843 set_current_state(TASK_INTERRUPTIBLE);
2844 schedule_timeout(sci->sc_interval);
2848 nilfs_clear_gcdat_inode(nilfs);
2849 nilfs_transaction_unlock(sbi);
2853 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2855 struct nilfs_sb_info *sbi = sci->sc_sbi;
2856 struct nilfs_transaction_info ti;
2857 struct nilfs_segctor_req req = { .mode = mode };
2859 nilfs_transaction_lock(sbi, &ti, 0);
2861 nilfs_segctor_accept(sci, &req);
2862 nilfs_segctor_construct(sci, &req);
2863 nilfs_segctor_notify(sci, &req);
2866 * Unclosed segment should be retried. We do this using sc_timer.
2867 * Timeout of sc_timer will invoke complete construction which leads
2868 * to close the current logical segment.
2870 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2871 nilfs_segctor_start_timer(sci);
2873 nilfs_transaction_unlock(sbi);
2876 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2881 spin_lock(&sci->sc_state_lock);
2882 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2883 SC_FLUSH_DAT : SC_FLUSH_FILE;
2884 spin_unlock(&sci->sc_state_lock);
2887 err = nilfs_segctor_do_construct(sci, mode);
2889 spin_lock(&sci->sc_state_lock);
2890 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2891 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2892 spin_unlock(&sci->sc_state_lock);
2894 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2897 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2899 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2900 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2901 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2902 return SC_FLUSH_FILE;
2903 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2904 return SC_FLUSH_DAT;
2910 * nilfs_segctor_thread - main loop of the segment constructor thread.
2911 * @arg: pointer to a struct nilfs_sc_info.
2913 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2914 * to execute segment constructions.
2916 static int nilfs_segctor_thread(void *arg)
2918 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2919 struct timer_list timer;
2923 timer.data = (unsigned long)current;
2924 timer.function = nilfs_construction_timeout;
2925 sci->sc_timer = &timer;
2928 sci->sc_task = current;
2929 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2931 "segctord starting. Construction interval = %lu seconds, "
2932 "CP frequency < %lu seconds\n",
2933 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2935 spin_lock(&sci->sc_state_lock);
2940 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2943 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2945 else if (!sci->sc_flush_request)
2948 mode = nilfs_segctor_flush_mode(sci);
2950 spin_unlock(&sci->sc_state_lock);
2951 nilfs_segctor_thread_construct(sci, mode);
2952 spin_lock(&sci->sc_state_lock);
2957 if (freezing(current)) {
2958 spin_unlock(&sci->sc_state_lock);
2960 spin_lock(&sci->sc_state_lock);
2963 int should_sleep = 1;
2965 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2966 TASK_INTERRUPTIBLE);
2968 if (sci->sc_seq_request != sci->sc_seq_done)
2970 else if (sci->sc_flush_request)
2972 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2973 should_sleep = time_before(jiffies,
2974 sci->sc_timer->expires);
2977 spin_unlock(&sci->sc_state_lock);
2979 spin_lock(&sci->sc_state_lock);
2981 finish_wait(&sci->sc_wait_daemon, &wait);
2982 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2983 time_after_eq(jiffies, sci->sc_timer->expires));
2988 spin_unlock(&sci->sc_state_lock);
2989 del_timer_sync(sci->sc_timer);
2990 sci->sc_timer = NULL;
2993 sci->sc_task = NULL;
2994 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2998 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
3000 struct task_struct *t;
3002 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
3004 int err = PTR_ERR(t);
3006 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
3010 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
3014 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
3016 sci->sc_state |= NILFS_SEGCTOR_QUIT;
3018 while (sci->sc_task) {
3019 wake_up(&sci->sc_wait_daemon);
3020 spin_unlock(&sci->sc_state_lock);
3021 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
3022 spin_lock(&sci->sc_state_lock);
3026 static int nilfs_segctor_init(struct nilfs_sc_info *sci,
3027 struct nilfs_recovery_info *ri)
3030 struct inode *inode = nilfs_iget(sci->sc_super, NILFS_SKETCH_INO);
3032 sci->sc_sketch_inode = IS_ERR(inode) ? NULL : inode;
3033 if (sci->sc_sketch_inode)
3034 i_size_write(sci->sc_sketch_inode, 0);
3036 sci->sc_seq_done = sci->sc_seq_request;
3038 list_splice_init(&ri->ri_used_segments,
3039 sci->sc_active_segments.prev);
3041 err = nilfs_segctor_start_thread(sci);
3044 list_splice_init(&sci->sc_active_segments,
3045 ri->ri_used_segments.prev);
3046 if (sci->sc_sketch_inode) {
3047 iput(sci->sc_sketch_inode);
3048 sci->sc_sketch_inode = NULL;
3055 * Setup & clean-up functions
3057 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
3059 struct nilfs_sc_info *sci;
3061 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
3066 sci->sc_super = sbi->s_super;
3068 init_waitqueue_head(&sci->sc_wait_request);
3069 init_waitqueue_head(&sci->sc_wait_daemon);
3070 init_waitqueue_head(&sci->sc_wait_task);
3071 spin_lock_init(&sci->sc_state_lock);
3072 INIT_LIST_HEAD(&sci->sc_dirty_files);
3073 INIT_LIST_HEAD(&sci->sc_segbufs);
3074 INIT_LIST_HEAD(&sci->sc_gc_inodes);
3075 INIT_LIST_HEAD(&sci->sc_active_segments);
3076 INIT_LIST_HEAD(&sci->sc_cleaning_segments);
3077 INIT_LIST_HEAD(&sci->sc_copied_buffers);
3079 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
3080 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
3081 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
3083 if (sbi->s_interval)
3084 sci->sc_interval = sbi->s_interval;
3085 if (sbi->s_watermark)
3086 sci->sc_watermark = sbi->s_watermark;
3090 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
3092 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
3094 /* The segctord thread was stopped and its timer was removed.
3095 But some tasks remain. */
3097 struct nilfs_sb_info *sbi = sci->sc_sbi;
3098 struct nilfs_transaction_info ti;
3099 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
3101 nilfs_transaction_lock(sbi, &ti, 0);
3102 nilfs_segctor_accept(sci, &req);
3103 ret = nilfs_segctor_construct(sci, &req);
3104 nilfs_segctor_notify(sci, &req);
3105 nilfs_transaction_unlock(sbi);
3107 } while (ret && retrycount-- > 0);
3111 * nilfs_segctor_destroy - destroy the segment constructor.
3112 * @sci: nilfs_sc_info
3114 * nilfs_segctor_destroy() kills the segctord thread and frees
3115 * the nilfs_sc_info struct.
3116 * Caller must hold the segment semaphore.
3118 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
3120 struct nilfs_sb_info *sbi = sci->sc_sbi;
3123 up_write(&sbi->s_nilfs->ns_segctor_sem);
3125 spin_lock(&sci->sc_state_lock);
3126 nilfs_segctor_kill_thread(sci);
3127 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
3128 || sci->sc_seq_request != sci->sc_seq_done);
3129 spin_unlock(&sci->sc_state_lock);
3131 if (flag || nilfs_segctor_confirm(sci))
3132 nilfs_segctor_write_out(sci);
3134 BUG_ON(!list_empty(&sci->sc_copied_buffers));
3136 if (!list_empty(&sci->sc_dirty_files)) {
3137 nilfs_warning(sbi->s_super, __func__,
3138 "dirty file(s) after the final construction\n");
3139 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
3141 if (!list_empty(&sci->sc_active_segments))
3142 nilfs_dispose_segment_list(&sci->sc_active_segments);
3144 if (!list_empty(&sci->sc_cleaning_segments))
3145 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
3147 BUG_ON(!list_empty(&sci->sc_segbufs));
3149 if (sci->sc_sketch_inode) {
3150 iput(sci->sc_sketch_inode);
3151 sci->sc_sketch_inode = NULL;
3153 down_write(&sbi->s_nilfs->ns_segctor_sem);
3159 * nilfs_attach_segment_constructor - attach a segment constructor
3160 * @sbi: nilfs_sb_info
3161 * @ri: nilfs_recovery_info
3163 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
3164 * initilizes it, and starts the segment constructor.
3166 * Return Value: On success, 0 is returned. On error, one of the following
3167 * negative error code is returned.
3169 * %-ENOMEM - Insufficient memory available.
3171 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
3172 struct nilfs_recovery_info *ri)
3174 struct the_nilfs *nilfs = sbi->s_nilfs;
3177 /* Each field of nilfs_segctor is cleared through the initialization
3178 of super-block info */
3179 sbi->s_sc_info = nilfs_segctor_new(sbi);
3180 if (!sbi->s_sc_info)
3183 nilfs_attach_writer(nilfs, sbi);
3184 err = nilfs_segctor_init(NILFS_SC(sbi), ri);
3186 nilfs_detach_writer(nilfs, sbi);
3187 kfree(sbi->s_sc_info);
3188 sbi->s_sc_info = NULL;
3194 * nilfs_detach_segment_constructor - destroy the segment constructor
3195 * @sbi: nilfs_sb_info
3197 * nilfs_detach_segment_constructor() kills the segment constructor daemon,
3198 * frees the struct nilfs_sc_info, and destroy the dirty file list.
3200 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
3202 struct the_nilfs *nilfs = sbi->s_nilfs;
3203 LIST_HEAD(garbage_list);
3205 down_write(&nilfs->ns_segctor_sem);
3206 if (NILFS_SC(sbi)) {
3207 nilfs_segctor_destroy(NILFS_SC(sbi));
3208 sbi->s_sc_info = NULL;
3211 /* Force to free the list of dirty files */
3212 spin_lock(&sbi->s_inode_lock);
3213 if (!list_empty(&sbi->s_dirty_files)) {
3214 list_splice_init(&sbi->s_dirty_files, &garbage_list);
3215 nilfs_warning(sbi->s_super, __func__,
3216 "Non empty dirty list after the last "
3217 "segment construction\n");
3219 spin_unlock(&sbi->s_inode_lock);
3220 up_write(&nilfs->ns_segctor_sem);
3222 nilfs_dispose_list(sbi, &garbage_list, 1);
3223 nilfs_detach_writer(nilfs, sbi);