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 * we save it and restore on nilfs_transaction_end().
167 * But this should never happen.
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_end() 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 * %-ERESTARTSYS - Interrupted
217 * %-ENOSPC - No space left on device
219 int nilfs_transaction_begin(struct super_block *sb,
220 struct nilfs_transaction_info *ti,
223 struct nilfs_sb_info *sbi;
224 struct the_nilfs *nilfs;
225 int ret = nilfs_prepare_segment_lock(ti);
227 if (unlikely(ret < 0))
233 nilfs = sbi->s_nilfs;
234 down_read(&nilfs->ns_segctor_sem);
235 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
236 up_read(&nilfs->ns_segctor_sem);
243 ti = current->journal_info;
244 current->journal_info = ti->ti_save;
245 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
246 kmem_cache_free(nilfs_transaction_cachep, ti);
251 * nilfs_transaction_end - end indivisible file operations.
253 * @commit: commit flag (0 for no change)
255 * nilfs_transaction_end() releases the read semaphore which is
256 * acquired by nilfs_transaction_begin(). Its releasing is only done
257 * in outermost call of this function. If the nilfs_transaction_info
258 * was allocated dynamically, it is given back to a slab cache.
260 int nilfs_transaction_end(struct super_block *sb, int commit)
262 struct nilfs_transaction_info *ti = current->journal_info;
263 struct nilfs_sb_info *sbi;
264 struct nilfs_sc_info *sci;
267 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
270 ti->ti_flags |= NILFS_TI_COMMIT;
271 if (ti->ti_count > 0) {
278 if (ti->ti_flags & NILFS_TI_COMMIT)
279 nilfs_segctor_start_timer(sci);
280 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
282 nilfs_segctor_do_flush(sci, 0);
284 up_read(&sbi->s_nilfs->ns_segctor_sem);
285 current->journal_info = ti->ti_save;
287 if (ti->ti_flags & NILFS_TI_SYNC)
288 err = nilfs_construct_segment(sb);
289 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
290 kmem_cache_free(nilfs_transaction_cachep, ti);
294 void nilfs_relax_pressure_in_lock(struct super_block *sb)
296 struct nilfs_sb_info *sbi = NILFS_SB(sb);
297 struct nilfs_sc_info *sci = NILFS_SC(sbi);
298 struct the_nilfs *nilfs = sbi->s_nilfs;
300 if (!sci || !sci->sc_flush_request)
303 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
304 up_read(&nilfs->ns_segctor_sem);
306 down_write(&nilfs->ns_segctor_sem);
307 if (sci->sc_flush_request &&
308 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
309 struct nilfs_transaction_info *ti = current->journal_info;
311 ti->ti_flags |= NILFS_TI_WRITER;
312 nilfs_segctor_do_immediate_flush(sci);
313 ti->ti_flags &= ~NILFS_TI_WRITER;
315 downgrade_write(&nilfs->ns_segctor_sem);
318 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
319 struct nilfs_transaction_info *ti,
322 struct nilfs_transaction_info *cur_ti = current->journal_info;
326 ti->ti_flags = NILFS_TI_WRITER;
328 ti->ti_save = cur_ti;
329 ti->ti_magic = NILFS_TI_MAGIC;
330 INIT_LIST_HEAD(&ti->ti_garbage);
331 current->journal_info = ti;
334 down_write(&sbi->s_nilfs->ns_segctor_sem);
335 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
338 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
340 up_write(&sbi->s_nilfs->ns_segctor_sem);
344 ti->ti_flags |= NILFS_TI_GC;
347 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
349 struct nilfs_transaction_info *ti = current->journal_info;
351 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
352 BUG_ON(ti->ti_count > 0);
354 up_write(&sbi->s_nilfs->ns_segctor_sem);
355 current->journal_info = ti->ti_save;
356 if (!list_empty(&ti->ti_garbage))
357 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
360 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
361 struct nilfs_segsum_pointer *ssp,
364 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
365 unsigned blocksize = sci->sc_super->s_blocksize;
368 if (unlikely(ssp->offset + bytes > blocksize)) {
370 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
371 &segbuf->sb_segsum_buffers));
372 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
374 p = ssp->bh->b_data + ssp->offset;
375 ssp->offset += bytes;
380 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
381 * @sci: nilfs_sc_info
383 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
385 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
386 struct buffer_head *sumbh;
391 if (nilfs_doing_gc())
393 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
397 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
398 sumbytes = segbuf->sb_sum.sumbytes;
399 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
400 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
401 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
405 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
407 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
408 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
409 return -E2BIG; /* The current segment is filled up
411 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
412 return nilfs_segctor_reset_segment_buffer(sci);
415 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
417 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
420 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
421 err = nilfs_segctor_feed_segment(sci);
424 segbuf = sci->sc_curseg;
426 err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
428 segbuf->sb_sum.flags |= NILFS_SS_SR;
433 * Functions for making segment summary and payloads
435 static int nilfs_segctor_segsum_block_required(
436 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
439 unsigned blocksize = sci->sc_super->s_blocksize;
440 /* Size of finfo and binfo is enough small against blocksize */
442 return ssp->offset + binfo_size +
443 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
447 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
450 sci->sc_curseg->sb_sum.nfinfo++;
451 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
452 nilfs_segctor_map_segsum_entry(
453 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
457 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
460 struct nilfs_finfo *finfo;
461 struct nilfs_inode_info *ii;
462 struct nilfs_segment_buffer *segbuf;
464 if (sci->sc_blk_cnt == 0)
468 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
470 finfo->fi_ino = cpu_to_le64(inode->i_ino);
471 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
472 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
473 finfo->fi_cno = cpu_to_le64(ii->i_cno);
475 segbuf = sci->sc_curseg;
476 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
477 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
478 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
479 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
482 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
483 struct buffer_head *bh,
487 struct nilfs_segment_buffer *segbuf;
488 int required, err = 0;
491 segbuf = sci->sc_curseg;
492 required = nilfs_segctor_segsum_block_required(
493 sci, &sci->sc_binfo_ptr, binfo_size);
494 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
495 nilfs_segctor_end_finfo(sci, inode);
496 err = nilfs_segctor_feed_segment(sci);
501 if (unlikely(required)) {
502 err = nilfs_segbuf_extend_segsum(segbuf);
506 if (sci->sc_blk_cnt == 0)
507 nilfs_segctor_begin_finfo(sci, inode);
509 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
510 /* Substitution to vblocknr is delayed until update_blocknr() */
511 nilfs_segbuf_add_file_buffer(segbuf, bh);
517 static int nilfs_handle_bmap_error(int err, const char *fname,
518 struct inode *inode, struct super_block *sb)
520 if (err == -EINVAL) {
521 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
529 * Callback functions that enumerate, mark, and collect dirty blocks
531 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
532 struct buffer_head *bh, struct inode *inode)
536 /* BUG_ON(!buffer_dirty(bh)); */
537 /* excluded by scan_dirty_data_buffers() */
538 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
539 if (unlikely(err < 0))
540 return nilfs_handle_bmap_error(err, __func__, inode,
543 err = nilfs_segctor_add_file_block(sci, bh, inode,
544 sizeof(struct nilfs_binfo_v));
546 sci->sc_datablk_cnt++;
550 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
551 struct buffer_head *bh,
556 /* BUG_ON(!buffer_dirty(bh)); */
557 /* excluded by scan_dirty_node_buffers() */
558 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
559 if (unlikely(err < 0))
560 return nilfs_handle_bmap_error(err, __func__, inode,
565 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
566 struct buffer_head *bh,
569 BUG_ON(!buffer_dirty(bh));
570 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
573 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
574 struct nilfs_segsum_pointer *ssp,
575 union nilfs_binfo *binfo)
577 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
578 sci, ssp, sizeof(*binfo_v));
579 *binfo_v = binfo->bi_v;
582 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
583 struct nilfs_segsum_pointer *ssp,
584 union nilfs_binfo *binfo)
586 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
587 sci, ssp, sizeof(*vblocknr));
588 *vblocknr = binfo->bi_v.bi_vblocknr;
591 struct nilfs_sc_operations nilfs_sc_file_ops = {
592 .collect_data = nilfs_collect_file_data,
593 .collect_node = nilfs_collect_file_node,
594 .collect_bmap = nilfs_collect_file_bmap,
595 .write_data_binfo = nilfs_write_file_data_binfo,
596 .write_node_binfo = nilfs_write_file_node_binfo,
599 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
600 struct buffer_head *bh, struct inode *inode)
604 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
605 if (unlikely(err < 0))
606 return nilfs_handle_bmap_error(err, __func__, inode,
609 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
611 sci->sc_datablk_cnt++;
615 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
616 struct buffer_head *bh, struct inode *inode)
618 BUG_ON(!buffer_dirty(bh));
619 return nilfs_segctor_add_file_block(sci, bh, inode,
620 sizeof(struct nilfs_binfo_dat));
623 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
624 struct nilfs_segsum_pointer *ssp,
625 union nilfs_binfo *binfo)
627 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
629 *blkoff = binfo->bi_dat.bi_blkoff;
632 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
633 struct nilfs_segsum_pointer *ssp,
634 union nilfs_binfo *binfo)
636 struct nilfs_binfo_dat *binfo_dat =
637 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
638 *binfo_dat = binfo->bi_dat;
641 struct nilfs_sc_operations nilfs_sc_dat_ops = {
642 .collect_data = nilfs_collect_dat_data,
643 .collect_node = nilfs_collect_file_node,
644 .collect_bmap = nilfs_collect_dat_bmap,
645 .write_data_binfo = nilfs_write_dat_data_binfo,
646 .write_node_binfo = nilfs_write_dat_node_binfo,
649 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
650 .collect_data = nilfs_collect_file_data,
651 .collect_node = NULL,
652 .collect_bmap = NULL,
653 .write_data_binfo = nilfs_write_file_data_binfo,
654 .write_node_binfo = NULL,
657 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
658 struct list_head *listp,
660 loff_t start, loff_t end)
662 struct address_space *mapping = inode->i_mapping;
664 pgoff_t index = 0, last = ULONG_MAX;
668 if (unlikely(start != 0 || end != LLONG_MAX)) {
670 * A valid range is given for sync-ing data pages. The
671 * range is rounded to per-page; extra dirty buffers
672 * may be included if blocksize < pagesize.
674 index = start >> PAGE_SHIFT;
675 last = end >> PAGE_SHIFT;
677 pagevec_init(&pvec, 0);
679 if (unlikely(index > last) ||
680 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
681 min_t(pgoff_t, last - index,
682 PAGEVEC_SIZE - 1) + 1))
685 for (i = 0; i < pagevec_count(&pvec); i++) {
686 struct buffer_head *bh, *head;
687 struct page *page = pvec.pages[i];
689 if (unlikely(page->index > last))
694 if (!page_has_buffers(page))
695 create_empty_buffers(page,
696 1 << inode->i_blkbits, 0);
700 bh = head = page_buffers(page);
702 if (!buffer_dirty(bh))
705 list_add_tail(&bh->b_assoc_buffers, listp);
707 if (unlikely(ndirties >= nlimit)) {
708 pagevec_release(&pvec);
712 } while (bh = bh->b_this_page, bh != head);
714 pagevec_release(&pvec);
719 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
720 struct list_head *listp)
722 struct nilfs_inode_info *ii = NILFS_I(inode);
723 struct address_space *mapping = &ii->i_btnode_cache;
725 struct buffer_head *bh, *head;
729 pagevec_init(&pvec, 0);
731 while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
733 for (i = 0; i < pagevec_count(&pvec); i++) {
734 bh = head = page_buffers(pvec.pages[i]);
736 if (buffer_dirty(bh)) {
738 list_add_tail(&bh->b_assoc_buffers,
741 bh = bh->b_this_page;
742 } while (bh != head);
744 pagevec_release(&pvec);
749 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
750 struct list_head *head, int force)
752 struct nilfs_inode_info *ii, *n;
753 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
756 while (!list_empty(head)) {
757 spin_lock(&sbi->s_inode_lock);
758 list_for_each_entry_safe(ii, n, head, i_dirty) {
759 list_del_init(&ii->i_dirty);
761 if (unlikely(ii->i_bh)) {
765 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
766 set_bit(NILFS_I_QUEUED, &ii->i_state);
767 list_add_tail(&ii->i_dirty,
768 &sbi->s_dirty_files);
772 if (nv == SC_N_INODEVEC)
775 spin_unlock(&sbi->s_inode_lock);
777 for (pii = ivec; nv > 0; pii++, nv--)
778 iput(&(*pii)->vfs_inode);
782 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
784 struct the_nilfs *nilfs = sbi->s_nilfs;
787 if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
789 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
791 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
793 if (ret || nilfs_doing_gc())
794 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
799 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
801 return list_empty(&sci->sc_dirty_files) &&
802 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
803 list_empty(&sci->sc_cleaning_segments) &&
804 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
807 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
809 struct nilfs_sb_info *sbi = sci->sc_sbi;
812 if (nilfs_test_metadata_dirty(sbi))
813 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
815 spin_lock(&sbi->s_inode_lock);
816 if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
819 spin_unlock(&sbi->s_inode_lock);
823 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
825 struct nilfs_sb_info *sbi = sci->sc_sbi;
826 struct the_nilfs *nilfs = sbi->s_nilfs;
828 nilfs_mdt_clear_dirty(sbi->s_ifile);
829 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
830 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
831 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
834 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
836 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
837 struct buffer_head *bh_cp;
838 struct nilfs_checkpoint *raw_cp;
841 /* XXX: this interface will be changed */
842 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
845 /* The following code is duplicated with cpfile. But, it is
846 needed to collect the checkpoint even if it was not newly
848 nilfs_mdt_mark_buffer_dirty(bh_cp);
849 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
850 nilfs_cpfile_put_checkpoint(
851 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
853 BUG_ON(err == -EINVAL || err == -ENOENT);
858 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
860 struct nilfs_sb_info *sbi = sci->sc_sbi;
861 struct the_nilfs *nilfs = sbi->s_nilfs;
862 struct buffer_head *bh_cp;
863 struct nilfs_checkpoint *raw_cp;
866 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
869 BUG_ON(err == -EINVAL || err == -ENOENT);
872 raw_cp->cp_snapshot_list.ssl_next = 0;
873 raw_cp->cp_snapshot_list.ssl_prev = 0;
874 raw_cp->cp_inodes_count =
875 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
876 raw_cp->cp_blocks_count =
877 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
878 raw_cp->cp_nblk_inc =
879 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
880 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
881 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
882 if (sci->sc_sketch_inode && i_size_read(sci->sc_sketch_inode) > 0)
883 nilfs_checkpoint_set_sketch(raw_cp);
884 nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
885 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
892 static void nilfs_fill_in_file_bmap(struct inode *ifile,
893 struct nilfs_inode_info *ii)
896 struct buffer_head *ibh;
897 struct nilfs_inode *raw_inode;
899 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
902 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
904 nilfs_bmap_write(ii->i_bmap, raw_inode);
905 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
909 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
912 struct nilfs_inode_info *ii;
914 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
915 nilfs_fill_in_file_bmap(ifile, ii);
916 set_bit(NILFS_I_COLLECTED, &ii->i_state);
918 if (sci->sc_sketch_inode) {
919 ii = NILFS_I(sci->sc_sketch_inode);
920 if (test_bit(NILFS_I_DIRTY, &ii->i_state))
921 nilfs_fill_in_file_bmap(ifile, ii);
926 * CRC calculation routines
928 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
930 struct nilfs_super_root *raw_sr =
931 (struct nilfs_super_root *)bh_sr->b_data;
934 BUG_ON(NILFS_SR_BYTES > bh_sr->b_size);
936 (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
937 NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
938 raw_sr->sr_sum = cpu_to_le32(crc);
941 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
944 struct nilfs_segment_buffer *segbuf;
946 if (sci->sc_super_root)
947 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
949 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
950 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
951 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
955 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
956 struct the_nilfs *nilfs)
958 struct buffer_head *bh_sr = sci->sc_super_root;
959 struct nilfs_super_root *raw_sr =
960 (struct nilfs_super_root *)bh_sr->b_data;
961 unsigned isz = nilfs->ns_inode_size;
963 raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
964 raw_sr->sr_nongc_ctime
965 = cpu_to_le64(nilfs_doing_gc() ?
966 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
967 raw_sr->sr_flags = 0;
969 nilfs_mdt_write_inode_direct(
970 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
971 nilfs_mdt_write_inode_direct(
972 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
973 nilfs_mdt_write_inode_direct(
974 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
977 static void nilfs_redirty_inodes(struct list_head *head)
979 struct nilfs_inode_info *ii;
981 list_for_each_entry(ii, head, i_dirty) {
982 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
983 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
987 static void nilfs_drop_collected_inodes(struct list_head *head)
989 struct nilfs_inode_info *ii;
991 list_for_each_entry(ii, head, i_dirty) {
992 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
995 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
996 set_bit(NILFS_I_UPDATED, &ii->i_state);
1000 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
1001 struct inode *sufile)
1004 struct list_head *head = &sci->sc_cleaning_segments;
1005 struct nilfs_segment_entry *ent;
1008 list_for_each_entry(ent, head, list) {
1009 if (!(ent->flags & NILFS_SLH_FREED))
1011 err = nilfs_sufile_cancel_free(sufile, ent->segnum);
1014 ent->flags &= ~NILFS_SLH_FREED;
1018 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
1019 struct inode *sufile)
1021 struct list_head *head = &sci->sc_cleaning_segments;
1022 struct nilfs_segment_entry *ent;
1025 list_for_each_entry(ent, head, list) {
1026 err = nilfs_sufile_free(sufile, ent->segnum);
1029 ent->flags |= NILFS_SLH_FREED;
1034 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
1036 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
1039 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1040 struct inode *inode,
1041 struct list_head *listp,
1042 int (*collect)(struct nilfs_sc_info *,
1043 struct buffer_head *,
1046 struct buffer_head *bh, *n;
1050 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1051 list_del_init(&bh->b_assoc_buffers);
1052 err = collect(sci, bh, inode);
1055 goto dispose_buffers;
1061 while (!list_empty(listp)) {
1062 bh = list_entry(listp->next, struct buffer_head,
1064 list_del_init(&bh->b_assoc_buffers);
1070 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1072 /* Remaining number of blocks within segment buffer */
1073 return sci->sc_segbuf_nblocks -
1074 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1077 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1078 struct inode *inode,
1079 struct nilfs_sc_operations *sc_ops)
1081 LIST_HEAD(data_buffers);
1082 LIST_HEAD(node_buffers);
1085 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1086 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1088 n = nilfs_lookup_dirty_data_buffers(
1089 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1091 err = nilfs_segctor_apply_buffers(
1092 sci, inode, &data_buffers,
1093 sc_ops->collect_data);
1094 BUG_ON(!err); /* always receive -E2BIG or true error */
1098 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1100 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1101 err = nilfs_segctor_apply_buffers(
1102 sci, inode, &data_buffers, sc_ops->collect_data);
1103 if (unlikely(err)) {
1104 /* dispose node list */
1105 nilfs_segctor_apply_buffers(
1106 sci, inode, &node_buffers, NULL);
1109 sci->sc_stage.flags |= NILFS_CF_NODE;
1112 err = nilfs_segctor_apply_buffers(
1113 sci, inode, &node_buffers, sc_ops->collect_node);
1117 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1118 err = nilfs_segctor_apply_buffers(
1119 sci, inode, &node_buffers, sc_ops->collect_bmap);
1123 nilfs_segctor_end_finfo(sci, inode);
1124 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1130 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1131 struct inode *inode)
1133 LIST_HEAD(data_buffers);
1134 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1137 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1138 sci->sc_dsync_start,
1141 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1142 nilfs_collect_file_data);
1144 nilfs_segctor_end_finfo(sci, inode);
1146 /* always receive -E2BIG or true error if n > rest */
1151 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1153 struct nilfs_sb_info *sbi = sci->sc_sbi;
1154 struct the_nilfs *nilfs = sbi->s_nilfs;
1155 struct list_head *head;
1156 struct nilfs_inode_info *ii;
1159 switch (sci->sc_stage.scnt) {
1162 sci->sc_stage.flags = 0;
1164 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1165 sci->sc_nblk_inc = 0;
1166 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1167 if (mode == SC_LSEG_DSYNC) {
1168 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1173 sci->sc_stage.dirty_file_ptr = NULL;
1174 sci->sc_stage.gc_inode_ptr = NULL;
1175 if (mode == SC_FLUSH_DAT) {
1176 sci->sc_stage.scnt = NILFS_ST_DAT;
1179 sci->sc_stage.scnt++; /* Fall through */
1181 if (nilfs_doing_gc()) {
1182 head = &sci->sc_gc_inodes;
1183 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1185 list_for_each_entry_continue(ii, head, i_dirty) {
1186 err = nilfs_segctor_scan_file(
1187 sci, &ii->vfs_inode,
1188 &nilfs_sc_file_ops);
1189 if (unlikely(err)) {
1190 sci->sc_stage.gc_inode_ptr = list_entry(
1192 struct nilfs_inode_info,
1196 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1198 sci->sc_stage.gc_inode_ptr = NULL;
1200 sci->sc_stage.scnt++; /* Fall through */
1202 head = &sci->sc_dirty_files;
1203 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1205 list_for_each_entry_continue(ii, head, i_dirty) {
1206 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1208 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1209 &nilfs_sc_file_ops);
1210 if (unlikely(err)) {
1211 sci->sc_stage.dirty_file_ptr =
1212 list_entry(ii->i_dirty.prev,
1213 struct nilfs_inode_info,
1217 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1218 /* XXX: required ? */
1220 sci->sc_stage.dirty_file_ptr = NULL;
1221 if (mode == SC_FLUSH_FILE) {
1222 sci->sc_stage.scnt = NILFS_ST_DONE;
1225 sci->sc_stage.scnt++; /* Fall through */
1226 case NILFS_ST_SKETCH:
1227 if (mode == SC_LSEG_SR && sci->sc_sketch_inode) {
1228 ii = NILFS_I(sci->sc_sketch_inode);
1229 if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
1230 sci->sc_sketch_inode->i_ctime.tv_sec
1231 = sci->sc_seg_ctime;
1232 sci->sc_sketch_inode->i_mtime.tv_sec
1233 = sci->sc_seg_ctime;
1234 err = nilfs_mark_inode_dirty(
1235 sci->sc_sketch_inode);
1239 err = nilfs_segctor_scan_file(sci,
1240 sci->sc_sketch_inode,
1241 &nilfs_sc_file_ops);
1245 sci->sc_stage.scnt++;
1246 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1248 case NILFS_ST_IFILE:
1249 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1250 &nilfs_sc_file_ops);
1253 sci->sc_stage.scnt++;
1254 /* Creating a checkpoint */
1255 err = nilfs_segctor_create_checkpoint(sci);
1259 case NILFS_ST_CPFILE:
1260 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1261 &nilfs_sc_file_ops);
1264 sci->sc_stage.scnt++; /* Fall through */
1265 case NILFS_ST_SUFILE:
1266 err = nilfs_segctor_prepare_free_segments(sci,
1270 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1271 &nilfs_sc_file_ops);
1274 sci->sc_stage.scnt++; /* Fall through */
1277 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1281 if (mode == SC_FLUSH_DAT) {
1282 sci->sc_stage.scnt = NILFS_ST_DONE;
1285 sci->sc_stage.scnt++; /* Fall through */
1287 if (mode == SC_LSEG_SR) {
1288 /* Appending a super root */
1289 err = nilfs_segctor_add_super_root(sci);
1293 /* End of a logical segment */
1294 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1295 sci->sc_stage.scnt = NILFS_ST_DONE;
1297 case NILFS_ST_DSYNC:
1299 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1300 ii = sci->sc_dsync_inode;
1301 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1304 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1307 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1308 sci->sc_stage.scnt = NILFS_ST_DONE;
1320 static int nilfs_segctor_terminate_segment(struct nilfs_sc_info *sci,
1321 struct nilfs_segment_buffer *segbuf,
1322 struct inode *sufile)
1324 struct nilfs_segment_entry *ent = segbuf->sb_segent;
1327 err = nilfs_open_segment_entry(ent, sufile);
1330 nilfs_mdt_mark_buffer_dirty(ent->bh_su);
1331 nilfs_mdt_mark_dirty(sufile);
1332 nilfs_close_segment_entry(ent, sufile);
1334 list_add_tail(&ent->list, &sci->sc_active_segments);
1335 segbuf->sb_segent = NULL;
1339 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1341 struct buffer_head *bh_su;
1342 struct nilfs_segment_usage *raw_su;
1345 err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1348 nilfs_mdt_mark_buffer_dirty(bh_su);
1349 nilfs_mdt_mark_dirty(sufile);
1350 nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1354 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1355 struct the_nilfs *nilfs)
1357 struct nilfs_segment_buffer *segbuf, *n;
1358 struct inode *sufile = nilfs->ns_sufile;
1362 if (list_empty(&sci->sc_segbufs)) {
1363 segbuf = nilfs_segbuf_new(sci->sc_super);
1364 if (unlikely(!segbuf))
1366 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1368 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1370 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1371 nilfs->ns_pseg_offset, nilfs);
1375 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1376 err = nilfs_segctor_terminate_segment(sci, segbuf, sufile);
1380 nilfs_shift_to_next_segment(nilfs);
1381 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1383 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1385 err = nilfs_touch_segusage(sufile, segbuf->sb_segnum);
1389 if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1390 /* Start from the head of a new full segment */
1391 err = nilfs_sufile_alloc(sufile, &nextnum);
1395 nextnum = nilfs->ns_nextnum;
1397 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1398 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1400 /* truncating segment buffers */
1401 list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1403 list_del_init(&segbuf->sb_list);
1404 nilfs_segbuf_free(segbuf);
1409 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1410 struct the_nilfs *nilfs, int nadd)
1412 struct nilfs_segment_buffer *segbuf, *prev, *n;
1413 struct inode *sufile = nilfs->ns_sufile;
1418 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1420 * Since the segment specified with nextnum might be allocated during
1421 * the previous construction, the buffer including its segusage may
1422 * not be dirty. The following call ensures that the buffer is dirty
1423 * and will pin the buffer on memory until the sufile is written.
1425 err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1429 for (i = 0; i < nadd; i++) {
1430 /* extend segment info */
1432 segbuf = nilfs_segbuf_new(sci->sc_super);
1433 if (unlikely(!segbuf))
1436 /* map this buffer to region of segment on-disk */
1437 err = nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1441 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1443 /* allocate the next next full segment */
1444 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1448 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1449 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1451 list_add_tail(&segbuf->sb_list, &list);
1454 list_splice(&list, sci->sc_segbufs.prev);
1458 nilfs_segbuf_free(segbuf);
1460 list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1461 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1463 list_del_init(&segbuf->sb_list);
1464 nilfs_segbuf_free(segbuf);
1469 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1470 struct the_nilfs *nilfs)
1472 struct nilfs_segment_buffer *segbuf;
1475 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1476 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1477 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1480 if (segbuf->sb_io_error) {
1481 /* Case 1: The first segment failed */
1482 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1483 /* Case 1a: Partial segment appended into an existing
1485 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1486 segbuf->sb_fseg_end);
1487 else /* Case 1b: New full segment */
1488 set_nilfs_discontinued(nilfs);
1492 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1493 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1495 if (!done && segbuf->sb_io_error) {
1496 if (segbuf->sb_segnum != nilfs->ns_nextnum)
1497 /* Case 2: extended segment (!= next) failed */
1498 nilfs_sufile_set_error(nilfs->ns_sufile,
1505 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1507 struct nilfs_segment_buffer *segbuf;
1509 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1510 nilfs_segbuf_clear(segbuf);
1511 sci->sc_super_root = NULL;
1514 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1516 struct nilfs_segment_buffer *segbuf;
1518 while (!list_empty(&sci->sc_segbufs)) {
1519 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1520 list_del_init(&segbuf->sb_list);
1521 nilfs_segbuf_free(segbuf);
1523 /* sci->sc_curseg = NULL; */
1526 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1527 struct the_nilfs *nilfs, int err)
1529 if (unlikely(err)) {
1530 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1531 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1533 nilfs_segctor_clear_segment_buffers(sci);
1536 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1537 struct inode *sufile)
1539 struct nilfs_segment_buffer *segbuf;
1540 struct buffer_head *bh_su;
1541 struct nilfs_segment_usage *raw_su;
1542 unsigned long live_blocks;
1545 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1546 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1548 BUG_ON(ret); /* always succeed because bh_su is dirty */
1549 live_blocks = segbuf->sb_sum.nblocks +
1550 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1551 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1552 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1553 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1558 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1559 struct inode *sufile)
1561 struct nilfs_segment_buffer *segbuf;
1562 struct buffer_head *bh_su;
1563 struct nilfs_segment_usage *raw_su;
1566 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1567 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1569 BUG_ON(ret); /* always succeed because bh_su is dirty */
1570 raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1571 segbuf->sb_fseg_start);
1572 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1574 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1575 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1577 BUG_ON(ret); /* always succeed */
1578 raw_su->su_nblocks = 0;
1579 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1584 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1585 struct nilfs_segment_buffer *last,
1586 struct inode *sufile)
1588 struct nilfs_segment_buffer *segbuf = last, *n;
1591 list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1593 list_del_init(&segbuf->sb_list);
1594 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1595 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1597 nilfs_segbuf_free(segbuf);
1602 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1603 struct the_nilfs *nilfs, int mode)
1605 struct nilfs_cstage prev_stage = sci->sc_stage;
1608 /* Collection retry loop */
1610 sci->sc_super_root = NULL;
1611 sci->sc_nblk_this_inc = 0;
1612 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1614 err = nilfs_segctor_reset_segment_buffer(sci);
1618 err = nilfs_segctor_collect_blocks(sci, mode);
1619 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1623 if (unlikely(err != -E2BIG))
1626 /* The current segment is filled up */
1627 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1630 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1631 nilfs_segctor_clear_segment_buffers(sci);
1633 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1637 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1638 sci->sc_stage = prev_stage;
1640 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1647 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1648 struct buffer_head *new_bh)
1650 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1652 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1653 /* The caller must release old_bh */
1657 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1658 struct nilfs_segment_buffer *segbuf,
1661 struct inode *inode = NULL;
1663 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1664 unsigned long nblocks = 0, ndatablk = 0;
1665 struct nilfs_sc_operations *sc_op = NULL;
1666 struct nilfs_segsum_pointer ssp;
1667 struct nilfs_finfo *finfo = NULL;
1668 union nilfs_binfo binfo;
1669 struct buffer_head *bh, *bh_org;
1676 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1677 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1678 ssp.offset = sizeof(struct nilfs_segment_summary);
1680 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1681 if (bh == sci->sc_super_root)
1684 finfo = nilfs_segctor_map_segsum_entry(
1685 sci, &ssp, sizeof(*finfo));
1686 ino = le64_to_cpu(finfo->fi_ino);
1687 nblocks = le32_to_cpu(finfo->fi_nblocks);
1688 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1690 if (buffer_nilfs_node(bh))
1691 inode = NILFS_BTNC_I(bh->b_page->mapping);
1693 inode = NILFS_AS_I(bh->b_page->mapping);
1695 if (mode == SC_LSEG_DSYNC)
1696 sc_op = &nilfs_sc_dsync_ops;
1697 else if (ino == NILFS_DAT_INO)
1698 sc_op = &nilfs_sc_dat_ops;
1699 else /* file blocks */
1700 sc_op = &nilfs_sc_file_ops;
1704 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1707 nilfs_list_replace_buffer(bh_org, bh);
1713 sc_op->write_data_binfo(sci, &ssp, &binfo);
1715 sc_op->write_node_binfo(sci, &ssp, &binfo);
1718 if (--nblocks == 0) {
1722 } else if (ndatablk > 0)
1729 err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1733 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1735 struct nilfs_segment_buffer *segbuf;
1738 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1739 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1742 nilfs_segbuf_fill_in_segsum(segbuf);
1748 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1750 struct page *clone_page;
1751 struct buffer_head *bh, *head, *bh2;
1754 bh = head = page_buffers(page);
1756 clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1757 if (unlikely(!clone_page))
1760 bh2 = page_buffers(clone_page);
1761 kaddr = kmap_atomic(page, KM_USER0);
1763 if (list_empty(&bh->b_assoc_buffers))
1766 page_cache_get(clone_page); /* for each bh */
1767 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1768 bh2->b_blocknr = bh->b_blocknr;
1769 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1770 list_add_tail(&bh->b_assoc_buffers, out);
1771 } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1772 kunmap_atomic(kaddr, KM_USER0);
1774 if (!TestSetPageWriteback(clone_page))
1775 inc_zone_page_state(clone_page, NR_WRITEBACK);
1776 unlock_page(clone_page);
1781 static int nilfs_test_page_to_be_frozen(struct page *page)
1783 struct address_space *mapping = page->mapping;
1785 if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1788 if (page_mapped(page)) {
1789 ClearPageChecked(page);
1792 return PageChecked(page);
1795 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1797 if (!page || PageWriteback(page))
1798 /* For split b-tree node pages, this function may be called
1799 twice. We ignore the 2nd or later calls by this check. */
1803 clear_page_dirty_for_io(page);
1804 set_page_writeback(page);
1807 if (nilfs_test_page_to_be_frozen(page)) {
1808 int err = nilfs_copy_replace_page_buffers(page, out);
1815 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1816 struct page **failed_page)
1818 struct nilfs_segment_buffer *segbuf;
1819 struct page *bd_page = NULL, *fs_page = NULL;
1820 struct list_head *list = &sci->sc_copied_buffers;
1823 *failed_page = NULL;
1824 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1825 struct buffer_head *bh;
1827 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1829 if (bh->b_page != bd_page) {
1832 clear_page_dirty_for_io(bd_page);
1833 set_page_writeback(bd_page);
1834 unlock_page(bd_page);
1836 bd_page = bh->b_page;
1840 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1842 if (bh == sci->sc_super_root) {
1843 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);
1848 bd_page = bh->b_page;
1852 if (bh->b_page != fs_page) {
1853 err = nilfs_begin_page_io(fs_page, list);
1854 if (unlikely(err)) {
1855 *failed_page = fs_page;
1858 fs_page = bh->b_page;
1864 clear_page_dirty_for_io(bd_page);
1865 set_page_writeback(bd_page);
1866 unlock_page(bd_page);
1868 err = nilfs_begin_page_io(fs_page, list);
1870 *failed_page = fs_page;
1875 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1876 struct backing_dev_info *bdi)
1878 struct nilfs_segment_buffer *segbuf;
1879 struct nilfs_write_info wi;
1882 wi.sb = sci->sc_super;
1883 wi.bh_sr = sci->sc_super_root;
1886 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1887 nilfs_segbuf_prepare_write(segbuf, &wi);
1888 err = nilfs_segbuf_write(segbuf, &wi);
1890 res = nilfs_segbuf_wait(segbuf, &wi);
1891 err = unlikely(err) ? : res;
1898 static int nilfs_page_has_uncleared_buffer(struct page *page)
1900 struct buffer_head *head, *bh;
1902 head = bh = page_buffers(page);
1904 if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
1906 bh = bh->b_this_page;
1907 } while (bh != head);
1911 static void __nilfs_end_page_io(struct page *page, int err)
1913 /* BUG_ON(err > 0); */
1915 if (!nilfs_page_buffers_clean(page))
1916 __set_page_dirty_nobuffers(page);
1917 ClearPageError(page);
1919 __set_page_dirty_nobuffers(page);
1923 if (buffer_nilfs_allocated(page_buffers(page))) {
1924 if (TestClearPageWriteback(page))
1925 dec_zone_page_state(page, NR_WRITEBACK);
1927 end_page_writeback(page);
1930 static void nilfs_end_page_io(struct page *page, int err)
1935 if (buffer_nilfs_node(page_buffers(page)) &&
1936 nilfs_page_has_uncleared_buffer(page))
1937 /* For b-tree node pages, this function may be called twice
1938 or more because they might be split in a segment.
1939 This check assures that cleanup has been done for all
1940 buffers in a split btnode page. */
1943 __nilfs_end_page_io(page, err);
1946 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1948 struct buffer_head *bh, *head;
1951 while (!list_empty(list)) {
1952 bh = list_entry(list->next, struct buffer_head,
1955 page_cache_get(page);
1956 head = bh = page_buffers(page);
1958 if (!list_empty(&bh->b_assoc_buffers)) {
1959 list_del_init(&bh->b_assoc_buffers);
1961 set_buffer_uptodate(bh);
1962 clear_buffer_dirty(bh);
1963 clear_buffer_nilfs_volatile(bh);
1965 brelse(bh); /* for b_assoc_buffers */
1967 } while ((bh = bh->b_this_page) != head);
1969 __nilfs_end_page_io(page, err);
1970 page_cache_release(page);
1974 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1975 struct page *failed_page, int err)
1977 struct nilfs_segment_buffer *segbuf;
1978 struct page *bd_page = NULL, *fs_page = NULL;
1980 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1981 struct buffer_head *bh;
1983 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1985 if (bh->b_page != bd_page) {
1987 end_page_writeback(bd_page);
1988 bd_page = bh->b_page;
1992 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1994 if (bh == sci->sc_super_root) {
1995 if (bh->b_page != bd_page) {
1996 end_page_writeback(bd_page);
1997 bd_page = bh->b_page;
2001 if (bh->b_page != fs_page) {
2002 nilfs_end_page_io(fs_page, err);
2003 if (unlikely(fs_page == failed_page))
2005 fs_page = bh->b_page;
2010 end_page_writeback(bd_page);
2012 nilfs_end_page_io(fs_page, err);
2014 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
2017 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
2018 struct nilfs_segment_buffer *segbuf)
2020 nilfs->ns_segnum = segbuf->sb_segnum;
2021 nilfs->ns_nextnum = segbuf->sb_nextnum;
2022 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
2023 + segbuf->sb_sum.nblocks;
2024 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
2025 nilfs->ns_ctime = segbuf->sb_sum.ctime;
2028 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
2030 struct nilfs_segment_buffer *segbuf;
2031 struct page *bd_page = NULL, *fs_page = NULL;
2032 struct nilfs_sb_info *sbi = sci->sc_sbi;
2033 struct the_nilfs *nilfs = sbi->s_nilfs;
2034 int update_sr = (sci->sc_super_root != NULL);
2036 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
2037 struct buffer_head *bh;
2039 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
2041 set_buffer_uptodate(bh);
2042 clear_buffer_dirty(bh);
2043 if (bh->b_page != bd_page) {
2045 end_page_writeback(bd_page);
2046 bd_page = bh->b_page;
2050 * We assume that the buffers which belong to the same page
2051 * continue over the buffer list.
2052 * Under this assumption, the last BHs of pages is
2053 * identifiable by the discontinuity of bh->b_page
2054 * (page != fs_page).
2056 * For B-tree node blocks, however, this assumption is not
2057 * guaranteed. The cleanup code of B-tree node pages needs
2060 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2062 set_buffer_uptodate(bh);
2063 clear_buffer_dirty(bh);
2064 clear_buffer_nilfs_volatile(bh);
2065 if (bh == sci->sc_super_root) {
2066 if (bh->b_page != bd_page) {
2067 end_page_writeback(bd_page);
2068 bd_page = bh->b_page;
2072 if (bh->b_page != fs_page) {
2073 nilfs_end_page_io(fs_page, 0);
2074 fs_page = bh->b_page;
2078 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2079 if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2080 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2081 sci->sc_lseg_stime = jiffies;
2083 if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2084 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2088 * Since pages may continue over multiple segment buffers,
2089 * end of the last page must be checked outside of the loop.
2092 end_page_writeback(bd_page);
2094 nilfs_end_page_io(fs_page, 0);
2096 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2098 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2100 if (nilfs_doing_gc()) {
2101 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2103 nilfs_commit_gcdat_inode(nilfs);
2105 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2106 set_nilfs_cond_nongc_write(nilfs);
2107 wake_up(&nilfs->ns_cleanerd_wq);
2110 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2112 segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2113 nilfs_set_next_segment(nilfs, segbuf);
2116 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2117 segbuf->sb_sum.seg_seq, nilfs->ns_cno);
2119 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2120 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2122 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2125 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2126 struct nilfs_sb_info *sbi)
2128 struct nilfs_inode_info *ii, *n;
2129 __u64 cno = sbi->s_nilfs->ns_cno;
2131 spin_lock(&sbi->s_inode_lock);
2133 list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2135 struct buffer_head *ibh;
2138 spin_unlock(&sbi->s_inode_lock);
2139 err = nilfs_ifile_get_inode_block(
2140 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2141 if (unlikely(err)) {
2142 nilfs_warning(sbi->s_super, __func__,
2143 "failed to get inode block.\n");
2146 nilfs_mdt_mark_buffer_dirty(ibh);
2147 nilfs_mdt_mark_dirty(sbi->s_ifile);
2148 spin_lock(&sbi->s_inode_lock);
2149 if (likely(!ii->i_bh))
2157 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2158 set_bit(NILFS_I_BUSY, &ii->i_state);
2159 list_del(&ii->i_dirty);
2160 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2162 spin_unlock(&sbi->s_inode_lock);
2164 NILFS_I(sbi->s_ifile)->i_cno = cno;
2169 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2170 struct nilfs_sb_info *sbi)
2172 struct nilfs_transaction_info *ti = current->journal_info;
2173 struct nilfs_inode_info *ii, *n;
2174 __u64 cno = sbi->s_nilfs->ns_cno;
2176 spin_lock(&sbi->s_inode_lock);
2177 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2178 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2179 test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2180 /* The current checkpoint number (=nilfs->ns_cno) is
2181 changed between check-in and check-out only if the
2182 super root is written out. So, we can update i_cno
2183 for the inodes that remain in the dirty list. */
2187 clear_bit(NILFS_I_BUSY, &ii->i_state);
2190 list_del(&ii->i_dirty);
2191 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2193 spin_unlock(&sbi->s_inode_lock);
2197 * Nasty routines to manipulate active flags on sufile.
2198 * These would be removed in a future release.
2200 static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info *sci,
2201 struct the_nilfs *nilfs)
2203 struct nilfs_segment_buffer *segbuf, *last;
2204 struct nilfs_segment_entry *ent, *n;
2205 struct inode *sufile = nilfs->ns_sufile;
2206 struct list_head *head;
2208 last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2209 nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2210 ent = segbuf->sb_segent;
2212 break; /* ignore unmapped segments (should check it?)*/
2213 nilfs_segment_usage_set_active(ent->raw_su);
2214 nilfs_close_segment_entry(ent, sufile);
2217 head = &sci->sc_active_segments;
2218 list_for_each_entry_safe(ent, n, head, list) {
2219 nilfs_segment_usage_set_active(ent->raw_su);
2220 nilfs_close_segment_entry(ent, sufile);
2223 down_write(&nilfs->ns_sem);
2224 head = &nilfs->ns_used_segments;
2225 list_for_each_entry(ent, head, list) {
2226 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2228 up_write(&nilfs->ns_sem);
2231 static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info *sci,
2232 struct the_nilfs *nilfs)
2234 struct nilfs_segment_buffer *segbuf, *last;
2235 struct nilfs_segment_entry *ent;
2236 struct inode *sufile = nilfs->ns_sufile;
2237 struct list_head *head;
2240 last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2241 nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2243 * Deactivate ongoing full segments. The last segment is kept
2244 * active because it is a start point of recovery, and is not
2245 * relocatable until the super block points to a newer
2248 ent = segbuf->sb_segent;
2250 break; /* ignore unmapped segments (should check it?)*/
2251 err = nilfs_open_segment_entry(ent, sufile);
2254 nilfs_segment_usage_clear_active(ent->raw_su);
2255 BUG_ON(!buffer_dirty(ent->bh_su));
2258 head = &sci->sc_active_segments;
2259 list_for_each_entry(ent, head, list) {
2260 err = nilfs_open_segment_entry(ent, sufile);
2263 nilfs_segment_usage_clear_active(ent->raw_su);
2264 BUG_ON(!buffer_dirty(ent->bh_su));
2267 down_write(&nilfs->ns_sem);
2268 head = &nilfs->ns_used_segments;
2269 list_for_each_entry(ent, head, list) {
2270 /* clear volatile active for segments of older generations */
2271 nilfs_segment_usage_clear_volatile_active(ent->raw_su);
2273 up_write(&nilfs->ns_sem);
2277 nilfs_segctor_reactivate_segments(sci, nilfs);
2281 static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info *sci)
2283 struct nilfs_segment_buffer *segbuf, *last;
2284 struct nilfs_segment_entry *ent;
2286 /* move each segbuf->sb_segent to the list of used active segments */
2287 last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2288 nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2289 ent = segbuf->sb_segent;
2291 break; /* ignore unmapped segments (should check it?)*/
2292 list_add_tail(&ent->list, &sci->sc_active_segments);
2293 segbuf->sb_segent = NULL;
2298 __nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info *sci,
2299 struct the_nilfs *nilfs)
2302 struct nilfs_segment_entry *ent;
2304 list_splice_init(&sci->sc_active_segments,
2305 nilfs->ns_used_segments.prev);
2307 list_for_each_entry(ent, &nilfs->ns_used_segments, list) {
2308 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2309 /* These segments are kept open */
2314 * Main procedure of segment constructor
2316 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2318 struct nilfs_sb_info *sbi = sci->sc_sbi;
2319 struct the_nilfs *nilfs = sbi->s_nilfs;
2320 struct page *failed_page;
2321 int err, has_sr = 0;
2323 sci->sc_stage.scnt = NILFS_ST_INIT;
2325 err = nilfs_segctor_check_in_files(sci, sbi);
2329 if (nilfs_test_metadata_dirty(sbi))
2330 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2332 if (nilfs_segctor_clean(sci))
2336 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2338 err = nilfs_segctor_begin_construction(sci, nilfs);
2342 /* Update time stamp */
2343 sci->sc_seg_ctime = get_seconds();
2345 err = nilfs_segctor_collect(sci, nilfs, mode);
2349 has_sr = (sci->sc_super_root != NULL);
2351 /* Avoid empty segment */
2352 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2353 NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2354 BUG_ON(mode == SC_LSEG_SR);
2355 nilfs_segctor_end_construction(sci, nilfs, 1);
2359 err = nilfs_segctor_assign(sci, mode);
2364 err = nilfs_segctor_deactivate_segments(sci, nilfs);
2368 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2369 nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2372 err = nilfs_segctor_fill_in_checkpoint(sci);
2374 goto failed_to_make_up;
2376 nilfs_segctor_fill_in_super_root(sci, nilfs);
2378 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2380 /* Write partial segments */
2381 err = nilfs_segctor_prepare_write(sci, &failed_page);
2383 goto failed_to_write;
2385 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2387 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2389 goto failed_to_write;
2391 nilfs_segctor_complete_write(sci);
2393 /* Commit segments */
2394 nilfs_segctor_bead_completed_segments(sci);
2396 down_write(&nilfs->ns_sem);
2397 nilfs_update_last_segment(sbi, 1);
2398 __nilfs_segctor_commit_deactivate_segments(sci, nilfs);
2399 up_write(&nilfs->ns_sem);
2400 nilfs_segctor_commit_free_segments(sci);
2401 nilfs_segctor_clear_metadata_dirty(sci);
2404 nilfs_segctor_end_construction(sci, nilfs, 0);
2406 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2408 /* Clearing sketch data */
2409 if (has_sr && sci->sc_sketch_inode) {
2410 if (i_size_read(sci->sc_sketch_inode) == 0)
2411 clear_bit(NILFS_I_DIRTY,
2412 &NILFS_I(sci->sc_sketch_inode)->i_state);
2413 i_size_write(sci->sc_sketch_inode, 0);
2416 nilfs_segctor_destroy_segment_buffers(sci);
2417 nilfs_segctor_check_out_files(sci, sbi);
2421 nilfs_segctor_abort_write(sci, failed_page, err);
2422 nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2425 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2426 nilfs_redirty_inodes(&sci->sc_dirty_files);
2428 nilfs_segctor_reactivate_segments(sci, nilfs);
2431 if (nilfs_doing_gc())
2432 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2433 nilfs_segctor_end_construction(sci, nilfs, err);
2438 * nilfs_secgtor_start_timer - set timer of background write
2439 * @sci: nilfs_sc_info
2441 * If the timer has already been set, it ignores the new request.
2442 * This function MUST be called within a section locking the segment
2445 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2447 spin_lock(&sci->sc_state_lock);
2448 if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2449 sci->sc_timer->expires = jiffies + sci->sc_interval;
2450 add_timer(sci->sc_timer);
2451 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2453 spin_unlock(&sci->sc_state_lock);
2456 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2458 spin_lock(&sci->sc_state_lock);
2459 if (!(sci->sc_flush_request & (1 << bn))) {
2460 unsigned long prev_req = sci->sc_flush_request;
2462 sci->sc_flush_request |= (1 << bn);
2464 wake_up(&sci->sc_wait_daemon);
2466 spin_unlock(&sci->sc_state_lock);
2470 * nilfs_flush_segment - trigger a segment construction for resource control
2472 * @ino: inode number of the file to be flushed out.
2474 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2476 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2477 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2479 if (!sci || nilfs_doing_construction())
2481 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2482 /* assign bit 0 to data files */
2485 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
2486 __u64 *segnum, size_t nsegs)
2488 struct nilfs_segment_entry *ent;
2489 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2490 struct inode *sufile = nilfs->ns_sufile;
2493 const char *flag_name;
2497 for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
2498 ent = nilfs_alloc_segment_entry(*pnum);
2499 if (unlikely(!ent)) {
2503 list_add_tail(&ent->list, &list);
2505 err = nilfs_open_segment_entry(ent, sufile);
2509 if (unlikely(le32_to_cpu(ent->raw_su->su_flags) !=
2510 (1UL << NILFS_SEGMENT_USAGE_DIRTY))) {
2511 if (nilfs_segment_usage_clean(ent->raw_su))
2512 flag_name = "clean";
2513 else if (nilfs_segment_usage_active(ent->raw_su))
2514 flag_name = "active";
2515 else if (nilfs_segment_usage_volatile_active(
2517 flag_name = "volatile active";
2518 else if (!nilfs_segment_usage_dirty(ent->raw_su))
2519 flag_name = "non-dirty";
2521 flag_name = "erroneous";
2524 "NILFS: %s segment is requested to be cleaned "
2526 flag_name, (unsigned long long)ent->segnum);
2529 nilfs_close_segment_entry(ent, sufile);
2531 if (unlikely(err2)) {
2535 list_splice(&list, sci->sc_cleaning_segments.prev);
2539 nilfs_dispose_segment_list(&list);
2543 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
2545 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2548 struct nilfs_segctor_wait_request {
2555 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2557 struct nilfs_segctor_wait_request wait_req;
2560 spin_lock(&sci->sc_state_lock);
2561 init_wait(&wait_req.wq);
2563 atomic_set(&wait_req.done, 0);
2564 wait_req.seq = ++sci->sc_seq_request;
2565 spin_unlock(&sci->sc_state_lock);
2567 init_waitqueue_entry(&wait_req.wq, current);
2568 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2569 set_current_state(TASK_INTERRUPTIBLE);
2570 wake_up(&sci->sc_wait_daemon);
2573 if (atomic_read(&wait_req.done)) {
2577 if (!signal_pending(current)) {
2584 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2588 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2590 struct nilfs_segctor_wait_request *wrq, *n;
2591 unsigned long flags;
2593 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2594 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2596 if (!atomic_read(&wrq->done) &&
2597 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2599 atomic_set(&wrq->done, 1);
2601 if (atomic_read(&wrq->done)) {
2602 wrq->wq.func(&wrq->wq,
2603 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2607 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2611 * nilfs_construct_segment - construct a logical segment
2614 * Return Value: On success, 0 is retured. On errors, one of the following
2615 * negative error code is returned.
2617 * %-EROFS - Read only filesystem.
2621 * %-ENOSPC - No space left on device (only in a panic state).
2623 * %-ERESTARTSYS - Interrupted.
2625 * %-ENOMEM - Insufficient memory available.
2627 int nilfs_construct_segment(struct super_block *sb)
2629 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2630 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2631 struct nilfs_transaction_info *ti;
2637 /* A call inside transactions causes a deadlock. */
2638 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2640 err = nilfs_segctor_sync(sci);
2645 * nilfs_construct_dsync_segment - construct a data-only logical segment
2647 * @inode: inode whose data blocks should be written out
2648 * @start: start byte offset
2649 * @end: end byte offset (inclusive)
2651 * Return Value: On success, 0 is retured. On errors, one of the following
2652 * negative error code is returned.
2654 * %-EROFS - Read only filesystem.
2658 * %-ENOSPC - No space left on device (only in a panic state).
2660 * %-ERESTARTSYS - Interrupted.
2662 * %-ENOMEM - Insufficient memory available.
2664 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2665 loff_t start, loff_t end)
2667 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2668 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2669 struct nilfs_inode_info *ii;
2670 struct nilfs_transaction_info ti;
2676 nilfs_transaction_lock(sbi, &ti, 0);
2678 ii = NILFS_I(inode);
2679 if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2680 nilfs_test_opt(sbi, STRICT_ORDER) ||
2681 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2682 nilfs_discontinued(sbi->s_nilfs)) {
2683 nilfs_transaction_unlock(sbi);
2684 err = nilfs_segctor_sync(sci);
2688 spin_lock(&sbi->s_inode_lock);
2689 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2690 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2691 spin_unlock(&sbi->s_inode_lock);
2692 nilfs_transaction_unlock(sbi);
2695 spin_unlock(&sbi->s_inode_lock);
2696 sci->sc_dsync_inode = ii;
2697 sci->sc_dsync_start = start;
2698 sci->sc_dsync_end = end;
2700 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2702 nilfs_transaction_unlock(sbi);
2706 struct nilfs_segctor_req {
2709 int sc_err; /* construction failure */
2710 int sb_err; /* super block writeback failure */
2713 #define FLUSH_FILE_BIT (0x1) /* data file only */
2714 #define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2716 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2717 struct nilfs_segctor_req *req)
2721 req->sc_err = req->sb_err = 0;
2722 spin_lock(&sci->sc_state_lock);
2723 req->seq_accepted = sci->sc_seq_request;
2724 spin_unlock(&sci->sc_state_lock);
2727 del_timer_sync(sci->sc_timer);
2730 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2731 struct nilfs_segctor_req *req)
2733 /* Clear requests (even when the construction failed) */
2734 spin_lock(&sci->sc_state_lock);
2736 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2738 if (req->mode == SC_LSEG_SR) {
2739 sci->sc_seq_done = req->seq_accepted;
2740 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2741 sci->sc_flush_request = 0;
2742 } else if (req->mode == SC_FLUSH_FILE)
2743 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2744 else if (req->mode == SC_FLUSH_DAT)
2745 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2747 spin_unlock(&sci->sc_state_lock);
2750 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2751 struct nilfs_segctor_req *req)
2753 struct nilfs_sb_info *sbi = sci->sc_sbi;
2754 struct the_nilfs *nilfs = sbi->s_nilfs;
2757 if (nilfs_discontinued(nilfs))
2758 req->mode = SC_LSEG_SR;
2759 if (!nilfs_segctor_confirm(sci)) {
2760 err = nilfs_segctor_do_construct(sci, req->mode);
2764 if (req->mode != SC_FLUSH_DAT)
2765 atomic_set(&nilfs->ns_ndirtyblks, 0);
2766 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2767 nilfs_discontinued(nilfs)) {
2768 down_write(&nilfs->ns_sem);
2769 req->sb_err = nilfs_commit_super(sbi);
2770 up_write(&nilfs->ns_sem);
2776 static void nilfs_construction_timeout(unsigned long data)
2778 struct task_struct *p = (struct task_struct *)data;
2783 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2785 struct nilfs_inode_info *ii, *n;
2787 list_for_each_entry_safe(ii, n, head, i_dirty) {
2788 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2790 hlist_del_init(&ii->vfs_inode.i_hash);
2791 list_del_init(&ii->i_dirty);
2792 nilfs_clear_gcinode(&ii->vfs_inode);
2796 int nilfs_clean_segments(struct super_block *sb, void __user *argp)
2798 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2799 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2800 struct the_nilfs *nilfs = sbi->s_nilfs;
2801 struct nilfs_transaction_info ti;
2802 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2808 nilfs_transaction_lock(sbi, &ti, 1);
2810 err = nilfs_init_gcdat_inode(nilfs);
2813 err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
2817 list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2820 nilfs_segctor_accept(sci, &req);
2821 err = nilfs_segctor_construct(sci, &req);
2822 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2823 nilfs_segctor_notify(sci, &req);
2828 nilfs_warning(sb, __func__,
2829 "segment construction failed. (err=%d)", err);
2830 set_current_state(TASK_INTERRUPTIBLE);
2831 schedule_timeout(sci->sc_interval);
2835 nilfs_clear_gcdat_inode(nilfs);
2836 nilfs_transaction_unlock(sbi);
2840 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2842 struct nilfs_sb_info *sbi = sci->sc_sbi;
2843 struct nilfs_transaction_info ti;
2844 struct nilfs_segctor_req req = { .mode = mode };
2846 nilfs_transaction_lock(sbi, &ti, 0);
2848 nilfs_segctor_accept(sci, &req);
2849 nilfs_segctor_construct(sci, &req);
2850 nilfs_segctor_notify(sci, &req);
2853 * Unclosed segment should be retried. We do this using sc_timer.
2854 * Timeout of sc_timer will invoke complete construction which leads
2855 * to close the current logical segment.
2857 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2858 nilfs_segctor_start_timer(sci);
2860 nilfs_transaction_unlock(sbi);
2863 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2868 spin_lock(&sci->sc_state_lock);
2869 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2870 SC_FLUSH_DAT : SC_FLUSH_FILE;
2871 spin_unlock(&sci->sc_state_lock);
2874 err = nilfs_segctor_do_construct(sci, mode);
2876 spin_lock(&sci->sc_state_lock);
2877 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2878 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2879 spin_unlock(&sci->sc_state_lock);
2881 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2884 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2886 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2887 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2888 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2889 return SC_FLUSH_FILE;
2890 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2891 return SC_FLUSH_DAT;
2897 * nilfs_segctor_thread - main loop of the segment constructor thread.
2898 * @arg: pointer to a struct nilfs_sc_info.
2900 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2901 * to execute segment constructions.
2903 static int nilfs_segctor_thread(void *arg)
2905 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2906 struct timer_list timer;
2910 timer.data = (unsigned long)current;
2911 timer.function = nilfs_construction_timeout;
2912 sci->sc_timer = &timer;
2915 sci->sc_task = current;
2916 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2918 "segctord starting. Construction interval = %lu seconds, "
2919 "CP frequency < %lu seconds\n",
2920 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2922 spin_lock(&sci->sc_state_lock);
2927 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2930 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2932 else if (!sci->sc_flush_request)
2935 mode = nilfs_segctor_flush_mode(sci);
2937 spin_unlock(&sci->sc_state_lock);
2938 nilfs_segctor_thread_construct(sci, mode);
2939 spin_lock(&sci->sc_state_lock);
2944 if (freezing(current)) {
2945 spin_unlock(&sci->sc_state_lock);
2947 spin_lock(&sci->sc_state_lock);
2950 int should_sleep = 1;
2952 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2953 TASK_INTERRUPTIBLE);
2955 if (sci->sc_seq_request != sci->sc_seq_done)
2957 else if (sci->sc_flush_request)
2959 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2960 should_sleep = time_before(jiffies,
2961 sci->sc_timer->expires);
2964 spin_unlock(&sci->sc_state_lock);
2966 spin_lock(&sci->sc_state_lock);
2968 finish_wait(&sci->sc_wait_daemon, &wait);
2969 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2970 time_after_eq(jiffies, sci->sc_timer->expires));
2975 spin_unlock(&sci->sc_state_lock);
2976 del_timer_sync(sci->sc_timer);
2977 sci->sc_timer = NULL;
2980 sci->sc_task = NULL;
2981 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2985 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2987 struct task_struct *t;
2989 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2991 int err = PTR_ERR(t);
2993 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2997 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
3001 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
3003 sci->sc_state |= NILFS_SEGCTOR_QUIT;
3005 while (sci->sc_task) {
3006 wake_up(&sci->sc_wait_daemon);
3007 spin_unlock(&sci->sc_state_lock);
3008 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
3009 spin_lock(&sci->sc_state_lock);
3013 static int nilfs_segctor_init(struct nilfs_sc_info *sci,
3014 struct nilfs_recovery_info *ri)
3017 struct inode *inode = nilfs_iget(sci->sc_super, NILFS_SKETCH_INO);
3019 sci->sc_sketch_inode = IS_ERR(inode) ? NULL : inode;
3020 if (sci->sc_sketch_inode)
3021 i_size_write(sci->sc_sketch_inode, 0);
3023 sci->sc_seq_done = sci->sc_seq_request;
3025 list_splice_init(&ri->ri_used_segments,
3026 sci->sc_active_segments.prev);
3028 err = nilfs_segctor_start_thread(sci);
3031 list_splice_init(&sci->sc_active_segments,
3032 ri->ri_used_segments.prev);
3033 if (sci->sc_sketch_inode) {
3034 iput(sci->sc_sketch_inode);
3035 sci->sc_sketch_inode = NULL;
3042 * Setup & clean-up functions
3044 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
3046 struct nilfs_sc_info *sci;
3048 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
3053 sci->sc_super = sbi->s_super;
3055 init_waitqueue_head(&sci->sc_wait_request);
3056 init_waitqueue_head(&sci->sc_wait_daemon);
3057 init_waitqueue_head(&sci->sc_wait_task);
3058 spin_lock_init(&sci->sc_state_lock);
3059 INIT_LIST_HEAD(&sci->sc_dirty_files);
3060 INIT_LIST_HEAD(&sci->sc_segbufs);
3061 INIT_LIST_HEAD(&sci->sc_gc_inodes);
3062 INIT_LIST_HEAD(&sci->sc_active_segments);
3063 INIT_LIST_HEAD(&sci->sc_cleaning_segments);
3064 INIT_LIST_HEAD(&sci->sc_copied_buffers);
3066 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
3067 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
3068 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
3070 if (sbi->s_interval)
3071 sci->sc_interval = sbi->s_interval;
3072 if (sbi->s_watermark)
3073 sci->sc_watermark = sbi->s_watermark;
3077 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
3079 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
3081 /* The segctord thread was stopped and its timer was removed.
3082 But some tasks remain. */
3084 struct nilfs_sb_info *sbi = sci->sc_sbi;
3085 struct nilfs_transaction_info ti;
3086 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
3088 nilfs_transaction_lock(sbi, &ti, 0);
3089 nilfs_segctor_accept(sci, &req);
3090 ret = nilfs_segctor_construct(sci, &req);
3091 nilfs_segctor_notify(sci, &req);
3092 nilfs_transaction_unlock(sbi);
3094 } while (ret && retrycount-- > 0);
3098 * nilfs_segctor_destroy - destroy the segment constructor.
3099 * @sci: nilfs_sc_info
3101 * nilfs_segctor_destroy() kills the segctord thread and frees
3102 * the nilfs_sc_info struct.
3103 * Caller must hold the segment semaphore.
3105 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
3107 struct nilfs_sb_info *sbi = sci->sc_sbi;
3110 up_write(&sbi->s_nilfs->ns_segctor_sem);
3112 spin_lock(&sci->sc_state_lock);
3113 nilfs_segctor_kill_thread(sci);
3114 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
3115 || sci->sc_seq_request != sci->sc_seq_done);
3116 spin_unlock(&sci->sc_state_lock);
3118 if (flag || nilfs_segctor_confirm(sci))
3119 nilfs_segctor_write_out(sci);
3121 BUG_ON(!list_empty(&sci->sc_copied_buffers));
3123 if (!list_empty(&sci->sc_dirty_files)) {
3124 nilfs_warning(sbi->s_super, __func__,
3125 "dirty file(s) after the final construction\n");
3126 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
3128 if (!list_empty(&sci->sc_active_segments))
3129 nilfs_dispose_segment_list(&sci->sc_active_segments);
3131 if (!list_empty(&sci->sc_cleaning_segments))
3132 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
3134 BUG_ON(!list_empty(&sci->sc_segbufs));
3136 if (sci->sc_sketch_inode) {
3137 iput(sci->sc_sketch_inode);
3138 sci->sc_sketch_inode = NULL;
3140 down_write(&sbi->s_nilfs->ns_segctor_sem);
3146 * nilfs_attach_segment_constructor - attach a segment constructor
3147 * @sbi: nilfs_sb_info
3148 * @ri: nilfs_recovery_info
3150 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
3151 * initilizes it, and starts the segment constructor.
3153 * Return Value: On success, 0 is returned. On error, one of the following
3154 * negative error code is returned.
3156 * %-ENOMEM - Insufficient memory available.
3158 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
3159 struct nilfs_recovery_info *ri)
3161 struct the_nilfs *nilfs = sbi->s_nilfs;
3164 /* Each field of nilfs_segctor is cleared through the initialization
3165 of super-block info */
3166 sbi->s_sc_info = nilfs_segctor_new(sbi);
3167 if (!sbi->s_sc_info)
3170 nilfs_attach_writer(nilfs, sbi);
3171 err = nilfs_segctor_init(NILFS_SC(sbi), ri);
3173 nilfs_detach_writer(nilfs, sbi);
3174 kfree(sbi->s_sc_info);
3175 sbi->s_sc_info = NULL;
3181 * nilfs_detach_segment_constructor - destroy the segment constructor
3182 * @sbi: nilfs_sb_info
3184 * nilfs_detach_segment_constructor() kills the segment constructor daemon,
3185 * frees the struct nilfs_sc_info, and destroy the dirty file list.
3187 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
3189 struct the_nilfs *nilfs = sbi->s_nilfs;
3190 LIST_HEAD(garbage_list);
3192 down_write(&nilfs->ns_segctor_sem);
3193 if (NILFS_SC(sbi)) {
3194 nilfs_segctor_destroy(NILFS_SC(sbi));
3195 sbi->s_sc_info = NULL;
3198 /* Force to free the list of dirty files */
3199 spin_lock(&sbi->s_inode_lock);
3200 if (!list_empty(&sbi->s_dirty_files)) {
3201 list_splice_init(&sbi->s_dirty_files, &garbage_list);
3202 nilfs_warning(sbi->s_super, __func__,
3203 "Non empty dirty list after the last "
3204 "segment construction\n");
3206 spin_unlock(&sbi->s_inode_lock);
3207 up_write(&nilfs->ns_segctor_sem);
3209 nilfs_dispose_list(sbi, &garbage_list, 1);
3210 nilfs_detach_writer(nilfs, sbi);