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>
35 #include <linux/slab.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 */
74 NILFS_ST_SR, /* Super root */
75 NILFS_ST_DSYNC, /* Data sync blocks */
79 /* State flags of collection */
80 #define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
81 #define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
82 #define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
83 #define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
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)
119 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
121 struct nilfs_transaction_info *cur_ti = current->journal_info;
125 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
126 return ++cur_ti->ti_count;
129 * If journal_info field is occupied by other FS,
130 * it is saved and will be restored on
131 * nilfs_transaction_commit().
134 "NILFS warning: journal info from a different "
136 save = current->journal_info;
140 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
143 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
149 ti->ti_magic = NILFS_TI_MAGIC;
150 current->journal_info = ti;
155 * nilfs_transaction_begin - start indivisible file operations.
157 * @ti: nilfs_transaction_info
158 * @vacancy_check: flags for vacancy rate checks
160 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
161 * the segment semaphore, to make a segment construction and write tasks
162 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
163 * The region enclosed by these two functions can be nested. To avoid a
164 * deadlock, the semaphore is only acquired or released in the outermost call.
166 * This function allocates a nilfs_transaction_info struct to keep context
167 * information on it. It is initialized and hooked onto the current task in
168 * the outermost call. If a pre-allocated struct is given to @ti, it is used
169 * instead; otherwise a new struct is assigned from a slab.
171 * When @vacancy_check flag is set, this function will check the amount of
172 * free space, and will wait for the GC to reclaim disk space if low capacity.
174 * Return Value: On success, 0 is returned. On error, one of the following
175 * negative error code is returned.
177 * %-ENOMEM - Insufficient memory available.
179 * %-ENOSPC - No space left on device
181 int nilfs_transaction_begin(struct super_block *sb,
182 struct nilfs_transaction_info *ti,
185 struct nilfs_sb_info *sbi;
186 struct the_nilfs *nilfs;
187 int ret = nilfs_prepare_segment_lock(ti);
189 if (unlikely(ret < 0))
195 nilfs = sbi->s_nilfs;
196 down_read(&nilfs->ns_segctor_sem);
197 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
198 up_read(&nilfs->ns_segctor_sem);
205 ti = current->journal_info;
206 current->journal_info = ti->ti_save;
207 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
208 kmem_cache_free(nilfs_transaction_cachep, ti);
213 * nilfs_transaction_commit - commit indivisible file operations.
216 * nilfs_transaction_commit() releases the read semaphore which is
217 * acquired by nilfs_transaction_begin(). This is only performed
218 * in outermost call of this function. If a commit flag is set,
219 * nilfs_transaction_commit() sets a timer to start the segment
220 * constructor. If a sync flag is set, it starts construction
223 int nilfs_transaction_commit(struct super_block *sb)
225 struct nilfs_transaction_info *ti = current->journal_info;
226 struct nilfs_sb_info *sbi;
227 struct nilfs_sc_info *sci;
230 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
231 ti->ti_flags |= NILFS_TI_COMMIT;
232 if (ti->ti_count > 0) {
239 if (ti->ti_flags & NILFS_TI_COMMIT)
240 nilfs_segctor_start_timer(sci);
241 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
243 nilfs_segctor_do_flush(sci, 0);
245 up_read(&sbi->s_nilfs->ns_segctor_sem);
246 current->journal_info = ti->ti_save;
248 if (ti->ti_flags & NILFS_TI_SYNC)
249 err = nilfs_construct_segment(sb);
250 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
251 kmem_cache_free(nilfs_transaction_cachep, ti);
255 void nilfs_transaction_abort(struct super_block *sb)
257 struct nilfs_transaction_info *ti = current->journal_info;
259 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
260 if (ti->ti_count > 0) {
264 up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
266 current->journal_info = ti->ti_save;
267 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
268 kmem_cache_free(nilfs_transaction_cachep, ti);
271 void nilfs_relax_pressure_in_lock(struct super_block *sb)
273 struct nilfs_sb_info *sbi = NILFS_SB(sb);
274 struct nilfs_sc_info *sci = NILFS_SC(sbi);
275 struct the_nilfs *nilfs = sbi->s_nilfs;
277 if (!sci || !sci->sc_flush_request)
280 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
281 up_read(&nilfs->ns_segctor_sem);
283 down_write(&nilfs->ns_segctor_sem);
284 if (sci->sc_flush_request &&
285 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
286 struct nilfs_transaction_info *ti = current->journal_info;
288 ti->ti_flags |= NILFS_TI_WRITER;
289 nilfs_segctor_do_immediate_flush(sci);
290 ti->ti_flags &= ~NILFS_TI_WRITER;
292 downgrade_write(&nilfs->ns_segctor_sem);
295 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
296 struct nilfs_transaction_info *ti,
299 struct nilfs_transaction_info *cur_ti = current->journal_info;
302 ti->ti_flags = NILFS_TI_WRITER;
304 ti->ti_save = cur_ti;
305 ti->ti_magic = NILFS_TI_MAGIC;
306 INIT_LIST_HEAD(&ti->ti_garbage);
307 current->journal_info = ti;
310 down_write(&sbi->s_nilfs->ns_segctor_sem);
311 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
314 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
316 up_write(&sbi->s_nilfs->ns_segctor_sem);
320 ti->ti_flags |= NILFS_TI_GC;
323 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
325 struct nilfs_transaction_info *ti = current->journal_info;
327 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
328 BUG_ON(ti->ti_count > 0);
330 up_write(&sbi->s_nilfs->ns_segctor_sem);
331 current->journal_info = ti->ti_save;
332 if (!list_empty(&ti->ti_garbage))
333 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
336 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
337 struct nilfs_segsum_pointer *ssp,
340 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
341 unsigned blocksize = sci->sc_super->s_blocksize;
344 if (unlikely(ssp->offset + bytes > blocksize)) {
346 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
347 &segbuf->sb_segsum_buffers));
348 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
350 p = ssp->bh->b_data + ssp->offset;
351 ssp->offset += bytes;
356 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
357 * @sci: nilfs_sc_info
359 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
361 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
362 struct buffer_head *sumbh;
367 if (nilfs_doing_gc())
369 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
373 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
374 sumbytes = segbuf->sb_sum.sumbytes;
375 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
376 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
377 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
381 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
383 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
384 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
385 return -E2BIG; /* The current segment is filled up
387 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
388 return nilfs_segctor_reset_segment_buffer(sci);
391 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
393 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
396 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
397 err = nilfs_segctor_feed_segment(sci);
400 segbuf = sci->sc_curseg;
402 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
404 segbuf->sb_sum.flags |= NILFS_SS_SR;
409 * Functions for making segment summary and payloads
411 static int nilfs_segctor_segsum_block_required(
412 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
415 unsigned blocksize = sci->sc_super->s_blocksize;
416 /* Size of finfo and binfo is enough small against blocksize */
418 return ssp->offset + binfo_size +
419 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
423 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
426 sci->sc_curseg->sb_sum.nfinfo++;
427 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
428 nilfs_segctor_map_segsum_entry(
429 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
431 if (inode->i_sb && !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
432 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
436 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
439 struct nilfs_finfo *finfo;
440 struct nilfs_inode_info *ii;
441 struct nilfs_segment_buffer *segbuf;
444 if (sci->sc_blk_cnt == 0)
449 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
451 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
456 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
458 finfo->fi_ino = cpu_to_le64(inode->i_ino);
459 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
460 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
461 finfo->fi_cno = cpu_to_le64(cno);
463 segbuf = sci->sc_curseg;
464 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
465 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
466 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
467 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
470 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
471 struct buffer_head *bh,
475 struct nilfs_segment_buffer *segbuf;
476 int required, err = 0;
479 segbuf = sci->sc_curseg;
480 required = nilfs_segctor_segsum_block_required(
481 sci, &sci->sc_binfo_ptr, binfo_size);
482 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
483 nilfs_segctor_end_finfo(sci, inode);
484 err = nilfs_segctor_feed_segment(sci);
489 if (unlikely(required)) {
490 err = nilfs_segbuf_extend_segsum(segbuf);
494 if (sci->sc_blk_cnt == 0)
495 nilfs_segctor_begin_finfo(sci, inode);
497 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
498 /* Substitution to vblocknr is delayed until update_blocknr() */
499 nilfs_segbuf_add_file_buffer(segbuf, bh);
505 static int nilfs_handle_bmap_error(int err, const char *fname,
506 struct inode *inode, struct super_block *sb)
508 if (err == -EINVAL) {
509 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
517 * Callback functions that enumerate, mark, and collect dirty blocks
519 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
520 struct buffer_head *bh, struct inode *inode)
524 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
525 if (unlikely(err < 0))
526 return nilfs_handle_bmap_error(err, __func__, inode,
529 err = nilfs_segctor_add_file_block(sci, bh, inode,
530 sizeof(struct nilfs_binfo_v));
532 sci->sc_datablk_cnt++;
536 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
537 struct buffer_head *bh,
542 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
543 if (unlikely(err < 0))
544 return nilfs_handle_bmap_error(err, __func__, inode,
549 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
550 struct buffer_head *bh,
553 WARN_ON(!buffer_dirty(bh));
554 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
557 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
558 struct nilfs_segsum_pointer *ssp,
559 union nilfs_binfo *binfo)
561 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
562 sci, ssp, sizeof(*binfo_v));
563 *binfo_v = binfo->bi_v;
566 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
567 struct nilfs_segsum_pointer *ssp,
568 union nilfs_binfo *binfo)
570 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
571 sci, ssp, sizeof(*vblocknr));
572 *vblocknr = binfo->bi_v.bi_vblocknr;
575 static struct nilfs_sc_operations nilfs_sc_file_ops = {
576 .collect_data = nilfs_collect_file_data,
577 .collect_node = nilfs_collect_file_node,
578 .collect_bmap = nilfs_collect_file_bmap,
579 .write_data_binfo = nilfs_write_file_data_binfo,
580 .write_node_binfo = nilfs_write_file_node_binfo,
583 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
584 struct buffer_head *bh, struct inode *inode)
588 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
589 if (unlikely(err < 0))
590 return nilfs_handle_bmap_error(err, __func__, inode,
593 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
595 sci->sc_datablk_cnt++;
599 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
600 struct buffer_head *bh, struct inode *inode)
602 WARN_ON(!buffer_dirty(bh));
603 return nilfs_segctor_add_file_block(sci, bh, inode,
604 sizeof(struct nilfs_binfo_dat));
607 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
608 struct nilfs_segsum_pointer *ssp,
609 union nilfs_binfo *binfo)
611 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
613 *blkoff = binfo->bi_dat.bi_blkoff;
616 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
617 struct nilfs_segsum_pointer *ssp,
618 union nilfs_binfo *binfo)
620 struct nilfs_binfo_dat *binfo_dat =
621 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
622 *binfo_dat = binfo->bi_dat;
625 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
626 .collect_data = nilfs_collect_dat_data,
627 .collect_node = nilfs_collect_file_node,
628 .collect_bmap = nilfs_collect_dat_bmap,
629 .write_data_binfo = nilfs_write_dat_data_binfo,
630 .write_node_binfo = nilfs_write_dat_node_binfo,
633 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
634 .collect_data = nilfs_collect_file_data,
635 .collect_node = NULL,
636 .collect_bmap = NULL,
637 .write_data_binfo = nilfs_write_file_data_binfo,
638 .write_node_binfo = NULL,
641 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
642 struct list_head *listp,
644 loff_t start, loff_t end)
646 struct address_space *mapping = inode->i_mapping;
648 pgoff_t index = 0, last = ULONG_MAX;
652 if (unlikely(start != 0 || end != LLONG_MAX)) {
654 * A valid range is given for sync-ing data pages. The
655 * range is rounded to per-page; extra dirty buffers
656 * may be included if blocksize < pagesize.
658 index = start >> PAGE_SHIFT;
659 last = end >> PAGE_SHIFT;
661 pagevec_init(&pvec, 0);
663 if (unlikely(index > last) ||
664 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
665 min_t(pgoff_t, last - index,
666 PAGEVEC_SIZE - 1) + 1))
669 for (i = 0; i < pagevec_count(&pvec); i++) {
670 struct buffer_head *bh, *head;
671 struct page *page = pvec.pages[i];
673 if (unlikely(page->index > last))
678 if (!page_has_buffers(page))
679 create_empty_buffers(page,
680 1 << inode->i_blkbits, 0);
684 bh = head = page_buffers(page);
686 if (!buffer_dirty(bh))
689 list_add_tail(&bh->b_assoc_buffers, listp);
691 if (unlikely(ndirties >= nlimit)) {
692 pagevec_release(&pvec);
696 } while (bh = bh->b_this_page, bh != head);
698 pagevec_release(&pvec);
703 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
704 struct list_head *listp)
706 struct nilfs_inode_info *ii = NILFS_I(inode);
707 struct address_space *mapping = &ii->i_btnode_cache;
709 struct buffer_head *bh, *head;
713 pagevec_init(&pvec, 0);
715 while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
717 for (i = 0; i < pagevec_count(&pvec); i++) {
718 bh = head = page_buffers(pvec.pages[i]);
720 if (buffer_dirty(bh)) {
722 list_add_tail(&bh->b_assoc_buffers,
725 bh = bh->b_this_page;
726 } while (bh != head);
728 pagevec_release(&pvec);
733 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
734 struct list_head *head, int force)
736 struct nilfs_inode_info *ii, *n;
737 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
740 while (!list_empty(head)) {
741 spin_lock(&sbi->s_inode_lock);
742 list_for_each_entry_safe(ii, n, head, i_dirty) {
743 list_del_init(&ii->i_dirty);
745 if (unlikely(ii->i_bh)) {
749 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
750 set_bit(NILFS_I_QUEUED, &ii->i_state);
751 list_add_tail(&ii->i_dirty,
752 &sbi->s_dirty_files);
756 if (nv == SC_N_INODEVEC)
759 spin_unlock(&sbi->s_inode_lock);
761 for (pii = ivec; nv > 0; pii++, nv--)
762 iput(&(*pii)->vfs_inode);
766 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
767 struct nilfs_root *root)
771 if (nilfs_mdt_fetch_dirty(root->ifile))
773 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
775 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
777 if (ret || nilfs_doing_gc())
778 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
783 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
785 return list_empty(&sci->sc_dirty_files) &&
786 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
787 sci->sc_nfreesegs == 0 &&
788 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
791 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
793 struct nilfs_sb_info *sbi = sci->sc_sbi;
796 if (nilfs_test_metadata_dirty(sbi->s_nilfs, sci->sc_root))
797 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
799 spin_lock(&sbi->s_inode_lock);
800 if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
803 spin_unlock(&sbi->s_inode_lock);
807 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
809 struct nilfs_sb_info *sbi = sci->sc_sbi;
810 struct the_nilfs *nilfs = sbi->s_nilfs;
812 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
813 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
814 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
815 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
818 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
820 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
821 struct buffer_head *bh_cp;
822 struct nilfs_checkpoint *raw_cp;
825 /* XXX: this interface will be changed */
826 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
829 /* The following code is duplicated with cpfile. But, it is
830 needed to collect the checkpoint even if it was not newly
832 nilfs_mdt_mark_buffer_dirty(bh_cp);
833 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
834 nilfs_cpfile_put_checkpoint(
835 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
837 WARN_ON(err == -EINVAL || err == -ENOENT);
842 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
844 struct nilfs_sb_info *sbi = sci->sc_sbi;
845 struct the_nilfs *nilfs = sbi->s_nilfs;
846 struct buffer_head *bh_cp;
847 struct nilfs_checkpoint *raw_cp;
850 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
853 WARN_ON(err == -EINVAL || err == -ENOENT);
856 raw_cp->cp_snapshot_list.ssl_next = 0;
857 raw_cp->cp_snapshot_list.ssl_prev = 0;
858 raw_cp->cp_inodes_count =
859 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
860 raw_cp->cp_blocks_count =
861 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
862 raw_cp->cp_nblk_inc =
863 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
864 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
865 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
867 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
868 nilfs_checkpoint_clear_minor(raw_cp);
870 nilfs_checkpoint_set_minor(raw_cp);
872 nilfs_write_inode_common(sci->sc_root->ifile,
873 &raw_cp->cp_ifile_inode, 1);
874 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
881 static void nilfs_fill_in_file_bmap(struct inode *ifile,
882 struct nilfs_inode_info *ii)
885 struct buffer_head *ibh;
886 struct nilfs_inode *raw_inode;
888 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
891 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
893 nilfs_bmap_write(ii->i_bmap, raw_inode);
894 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
898 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
900 struct nilfs_inode_info *ii;
902 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
903 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
904 set_bit(NILFS_I_COLLECTED, &ii->i_state);
908 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
909 struct the_nilfs *nilfs)
911 struct buffer_head *bh_sr;
912 struct nilfs_super_root *raw_sr;
913 unsigned isz = nilfs->ns_inode_size;
915 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
916 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
918 raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
919 raw_sr->sr_nongc_ctime
920 = cpu_to_le64(nilfs_doing_gc() ?
921 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
922 raw_sr->sr_flags = 0;
924 nilfs_write_inode_common(nilfs_dat_inode(nilfs), (void *)raw_sr +
925 NILFS_SR_DAT_OFFSET(isz), 1);
926 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
927 NILFS_SR_CPFILE_OFFSET(isz), 1);
928 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
929 NILFS_SR_SUFILE_OFFSET(isz), 1);
932 static void nilfs_redirty_inodes(struct list_head *head)
934 struct nilfs_inode_info *ii;
936 list_for_each_entry(ii, head, i_dirty) {
937 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
938 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
942 static void nilfs_drop_collected_inodes(struct list_head *head)
944 struct nilfs_inode_info *ii;
946 list_for_each_entry(ii, head, i_dirty) {
947 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
950 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
951 set_bit(NILFS_I_UPDATED, &ii->i_state);
955 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
957 struct list_head *listp,
958 int (*collect)(struct nilfs_sc_info *,
959 struct buffer_head *,
962 struct buffer_head *bh, *n;
966 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
967 list_del_init(&bh->b_assoc_buffers);
968 err = collect(sci, bh, inode);
971 goto dispose_buffers;
977 while (!list_empty(listp)) {
978 bh = list_entry(listp->next, struct buffer_head,
980 list_del_init(&bh->b_assoc_buffers);
986 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
988 /* Remaining number of blocks within segment buffer */
989 return sci->sc_segbuf_nblocks -
990 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
993 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
995 struct nilfs_sc_operations *sc_ops)
997 LIST_HEAD(data_buffers);
998 LIST_HEAD(node_buffers);
1001 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1002 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1004 n = nilfs_lookup_dirty_data_buffers(
1005 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1007 err = nilfs_segctor_apply_buffers(
1008 sci, inode, &data_buffers,
1009 sc_ops->collect_data);
1010 BUG_ON(!err); /* always receive -E2BIG or true error */
1014 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1016 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1017 err = nilfs_segctor_apply_buffers(
1018 sci, inode, &data_buffers, sc_ops->collect_data);
1019 if (unlikely(err)) {
1020 /* dispose node list */
1021 nilfs_segctor_apply_buffers(
1022 sci, inode, &node_buffers, NULL);
1025 sci->sc_stage.flags |= NILFS_CF_NODE;
1028 err = nilfs_segctor_apply_buffers(
1029 sci, inode, &node_buffers, sc_ops->collect_node);
1033 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1034 err = nilfs_segctor_apply_buffers(
1035 sci, inode, &node_buffers, sc_ops->collect_bmap);
1039 nilfs_segctor_end_finfo(sci, inode);
1040 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1046 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1047 struct inode *inode)
1049 LIST_HEAD(data_buffers);
1050 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1053 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1054 sci->sc_dsync_start,
1057 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1058 nilfs_collect_file_data);
1060 nilfs_segctor_end_finfo(sci, inode);
1062 /* always receive -E2BIG or true error if n > rest */
1067 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1069 struct nilfs_sb_info *sbi = sci->sc_sbi;
1070 struct the_nilfs *nilfs = sbi->s_nilfs;
1071 struct list_head *head;
1072 struct nilfs_inode_info *ii;
1076 switch (sci->sc_stage.scnt) {
1079 sci->sc_stage.flags = 0;
1081 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1082 sci->sc_nblk_inc = 0;
1083 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1084 if (mode == SC_LSEG_DSYNC) {
1085 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1090 sci->sc_stage.dirty_file_ptr = NULL;
1091 sci->sc_stage.gc_inode_ptr = NULL;
1092 if (mode == SC_FLUSH_DAT) {
1093 sci->sc_stage.scnt = NILFS_ST_DAT;
1096 sci->sc_stage.scnt++; /* Fall through */
1098 if (nilfs_doing_gc()) {
1099 head = &sci->sc_gc_inodes;
1100 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1102 list_for_each_entry_continue(ii, head, i_dirty) {
1103 err = nilfs_segctor_scan_file(
1104 sci, &ii->vfs_inode,
1105 &nilfs_sc_file_ops);
1106 if (unlikely(err)) {
1107 sci->sc_stage.gc_inode_ptr = list_entry(
1109 struct nilfs_inode_info,
1113 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1115 sci->sc_stage.gc_inode_ptr = NULL;
1117 sci->sc_stage.scnt++; /* Fall through */
1119 head = &sci->sc_dirty_files;
1120 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1122 list_for_each_entry_continue(ii, head, i_dirty) {
1123 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1125 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1126 &nilfs_sc_file_ops);
1127 if (unlikely(err)) {
1128 sci->sc_stage.dirty_file_ptr =
1129 list_entry(ii->i_dirty.prev,
1130 struct nilfs_inode_info,
1134 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1135 /* XXX: required ? */
1137 sci->sc_stage.dirty_file_ptr = NULL;
1138 if (mode == SC_FLUSH_FILE) {
1139 sci->sc_stage.scnt = NILFS_ST_DONE;
1142 sci->sc_stage.scnt++;
1143 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1145 case NILFS_ST_IFILE:
1146 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1147 &nilfs_sc_file_ops);
1150 sci->sc_stage.scnt++;
1151 /* Creating a checkpoint */
1152 err = nilfs_segctor_create_checkpoint(sci);
1156 case NILFS_ST_CPFILE:
1157 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1158 &nilfs_sc_file_ops);
1161 sci->sc_stage.scnt++; /* Fall through */
1162 case NILFS_ST_SUFILE:
1163 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1164 sci->sc_nfreesegs, &ndone);
1165 if (unlikely(err)) {
1166 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1167 sci->sc_freesegs, ndone,
1171 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1173 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1174 &nilfs_sc_file_ops);
1177 sci->sc_stage.scnt++; /* Fall through */
1180 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1184 if (mode == SC_FLUSH_DAT) {
1185 sci->sc_stage.scnt = NILFS_ST_DONE;
1188 sci->sc_stage.scnt++; /* Fall through */
1190 if (mode == SC_LSEG_SR) {
1191 /* Appending a super root */
1192 err = nilfs_segctor_add_super_root(sci);
1196 /* End of a logical segment */
1197 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1198 sci->sc_stage.scnt = NILFS_ST_DONE;
1200 case NILFS_ST_DSYNC:
1202 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1203 ii = sci->sc_dsync_inode;
1204 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1207 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1210 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1211 sci->sc_stage.scnt = NILFS_ST_DONE;
1224 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1225 * @sci: nilfs_sc_info
1226 * @nilfs: nilfs object
1228 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1229 struct the_nilfs *nilfs)
1231 struct nilfs_segment_buffer *segbuf, *prev;
1235 segbuf = nilfs_segbuf_new(sci->sc_super);
1236 if (unlikely(!segbuf))
1239 if (list_empty(&sci->sc_write_logs)) {
1240 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1241 nilfs->ns_pseg_offset, nilfs);
1242 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1243 nilfs_shift_to_next_segment(nilfs);
1244 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1247 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1248 nextnum = nilfs->ns_nextnum;
1250 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1251 /* Start from the head of a new full segment */
1255 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1256 nilfs_segbuf_map_cont(segbuf, prev);
1257 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1258 nextnum = prev->sb_nextnum;
1260 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1261 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1262 segbuf->sb_sum.seg_seq++;
1267 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1272 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1276 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1278 BUG_ON(!list_empty(&sci->sc_segbufs));
1279 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1280 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1284 nilfs_segbuf_free(segbuf);
1288 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1289 struct the_nilfs *nilfs, int nadd)
1291 struct nilfs_segment_buffer *segbuf, *prev;
1292 struct inode *sufile = nilfs->ns_sufile;
1297 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1299 * Since the segment specified with nextnum might be allocated during
1300 * the previous construction, the buffer including its segusage may
1301 * not be dirty. The following call ensures that the buffer is dirty
1302 * and will pin the buffer on memory until the sufile is written.
1304 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1308 for (i = 0; i < nadd; i++) {
1309 /* extend segment info */
1311 segbuf = nilfs_segbuf_new(sci->sc_super);
1312 if (unlikely(!segbuf))
1315 /* map this buffer to region of segment on-disk */
1316 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1317 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1319 /* allocate the next next full segment */
1320 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1324 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1325 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1327 list_add_tail(&segbuf->sb_list, &list);
1330 list_splice_tail(&list, &sci->sc_segbufs);
1334 nilfs_segbuf_free(segbuf);
1336 list_for_each_entry(segbuf, &list, sb_list) {
1337 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1338 WARN_ON(ret); /* never fails */
1340 nilfs_destroy_logs(&list);
1344 static void nilfs_free_incomplete_logs(struct list_head *logs,
1345 struct the_nilfs *nilfs)
1347 struct nilfs_segment_buffer *segbuf, *prev;
1348 struct inode *sufile = nilfs->ns_sufile;
1351 segbuf = NILFS_FIRST_SEGBUF(logs);
1352 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1353 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1354 WARN_ON(ret); /* never fails */
1356 if (atomic_read(&segbuf->sb_err)) {
1357 /* Case 1: The first segment failed */
1358 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1359 /* Case 1a: Partial segment appended into an existing
1361 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1362 segbuf->sb_fseg_end);
1363 else /* Case 1b: New full segment */
1364 set_nilfs_discontinued(nilfs);
1368 list_for_each_entry_continue(segbuf, logs, sb_list) {
1369 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1370 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1371 WARN_ON(ret); /* never fails */
1373 if (atomic_read(&segbuf->sb_err) &&
1374 segbuf->sb_segnum != nilfs->ns_nextnum)
1375 /* Case 2: extended segment (!= next) failed */
1376 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1381 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1382 struct inode *sufile)
1384 struct nilfs_segment_buffer *segbuf;
1385 unsigned long live_blocks;
1388 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1389 live_blocks = segbuf->sb_sum.nblocks +
1390 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1391 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1394 WARN_ON(ret); /* always succeed because the segusage is dirty */
1398 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1400 struct nilfs_segment_buffer *segbuf;
1403 segbuf = NILFS_FIRST_SEGBUF(logs);
1404 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1405 segbuf->sb_pseg_start -
1406 segbuf->sb_fseg_start, 0);
1407 WARN_ON(ret); /* always succeed because the segusage is dirty */
1409 list_for_each_entry_continue(segbuf, logs, sb_list) {
1410 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1412 WARN_ON(ret); /* always succeed */
1416 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1417 struct nilfs_segment_buffer *last,
1418 struct inode *sufile)
1420 struct nilfs_segment_buffer *segbuf = last;
1423 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1424 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1425 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1428 nilfs_truncate_logs(&sci->sc_segbufs, last);
1432 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1433 struct the_nilfs *nilfs, int mode)
1435 struct nilfs_cstage prev_stage = sci->sc_stage;
1438 /* Collection retry loop */
1440 sci->sc_nblk_this_inc = 0;
1441 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1443 err = nilfs_segctor_reset_segment_buffer(sci);
1447 err = nilfs_segctor_collect_blocks(sci, mode);
1448 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1452 if (unlikely(err != -E2BIG))
1455 /* The current segment is filled up */
1456 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1459 nilfs_clear_logs(&sci->sc_segbufs);
1461 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1465 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1466 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1470 WARN_ON(err); /* do not happen */
1472 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1473 sci->sc_stage = prev_stage;
1475 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1482 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1483 struct buffer_head *new_bh)
1485 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1487 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1488 /* The caller must release old_bh */
1492 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1493 struct nilfs_segment_buffer *segbuf,
1496 struct inode *inode = NULL;
1498 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1499 unsigned long nblocks = 0, ndatablk = 0;
1500 struct nilfs_sc_operations *sc_op = NULL;
1501 struct nilfs_segsum_pointer ssp;
1502 struct nilfs_finfo *finfo = NULL;
1503 union nilfs_binfo binfo;
1504 struct buffer_head *bh, *bh_org;
1511 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1512 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1513 ssp.offset = sizeof(struct nilfs_segment_summary);
1515 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1516 if (bh == segbuf->sb_super_root)
1519 finfo = nilfs_segctor_map_segsum_entry(
1520 sci, &ssp, sizeof(*finfo));
1521 ino = le64_to_cpu(finfo->fi_ino);
1522 nblocks = le32_to_cpu(finfo->fi_nblocks);
1523 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1525 if (buffer_nilfs_node(bh))
1526 inode = NILFS_BTNC_I(bh->b_page->mapping);
1528 inode = NILFS_AS_I(bh->b_page->mapping);
1530 if (mode == SC_LSEG_DSYNC)
1531 sc_op = &nilfs_sc_dsync_ops;
1532 else if (ino == NILFS_DAT_INO)
1533 sc_op = &nilfs_sc_dat_ops;
1534 else /* file blocks */
1535 sc_op = &nilfs_sc_file_ops;
1539 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1542 nilfs_list_replace_buffer(bh_org, bh);
1548 sc_op->write_data_binfo(sci, &ssp, &binfo);
1550 sc_op->write_node_binfo(sci, &ssp, &binfo);
1553 if (--nblocks == 0) {
1557 } else if (ndatablk > 0)
1564 err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1568 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1570 struct nilfs_segment_buffer *segbuf;
1573 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1574 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1577 nilfs_segbuf_fill_in_segsum(segbuf);
1583 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1585 struct page *clone_page;
1586 struct buffer_head *bh, *head, *bh2;
1589 bh = head = page_buffers(page);
1591 clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1592 if (unlikely(!clone_page))
1595 bh2 = page_buffers(clone_page);
1596 kaddr = kmap_atomic(page, KM_USER0);
1598 if (list_empty(&bh->b_assoc_buffers))
1601 page_cache_get(clone_page); /* for each bh */
1602 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1603 bh2->b_blocknr = bh->b_blocknr;
1604 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1605 list_add_tail(&bh->b_assoc_buffers, out);
1606 } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1607 kunmap_atomic(kaddr, KM_USER0);
1609 if (!TestSetPageWriteback(clone_page))
1610 inc_zone_page_state(clone_page, NR_WRITEBACK);
1611 unlock_page(clone_page);
1616 static int nilfs_test_page_to_be_frozen(struct page *page)
1618 struct address_space *mapping = page->mapping;
1620 if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1623 if (page_mapped(page)) {
1624 ClearPageChecked(page);
1627 return PageChecked(page);
1630 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1632 if (!page || PageWriteback(page))
1633 /* For split b-tree node pages, this function may be called
1634 twice. We ignore the 2nd or later calls by this check. */
1638 clear_page_dirty_for_io(page);
1639 set_page_writeback(page);
1642 if (nilfs_test_page_to_be_frozen(page)) {
1643 int err = nilfs_copy_replace_page_buffers(page, out);
1650 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1651 struct page **failed_page)
1653 struct nilfs_segment_buffer *segbuf;
1654 struct page *bd_page = NULL, *fs_page = NULL;
1655 struct list_head *list = &sci->sc_copied_buffers;
1658 *failed_page = NULL;
1659 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1660 struct buffer_head *bh;
1662 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1664 if (bh->b_page != bd_page) {
1667 clear_page_dirty_for_io(bd_page);
1668 set_page_writeback(bd_page);
1669 unlock_page(bd_page);
1671 bd_page = bh->b_page;
1675 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1677 if (bh == segbuf->sb_super_root) {
1678 if (bh->b_page != bd_page) {
1680 clear_page_dirty_for_io(bd_page);
1681 set_page_writeback(bd_page);
1682 unlock_page(bd_page);
1683 bd_page = bh->b_page;
1687 if (bh->b_page != fs_page) {
1688 err = nilfs_begin_page_io(fs_page, list);
1689 if (unlikely(err)) {
1690 *failed_page = fs_page;
1693 fs_page = bh->b_page;
1699 clear_page_dirty_for_io(bd_page);
1700 set_page_writeback(bd_page);
1701 unlock_page(bd_page);
1703 err = nilfs_begin_page_io(fs_page, list);
1705 *failed_page = fs_page;
1710 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1711 struct the_nilfs *nilfs)
1715 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1716 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1720 static void __nilfs_end_page_io(struct page *page, int err)
1723 if (!nilfs_page_buffers_clean(page))
1724 __set_page_dirty_nobuffers(page);
1725 ClearPageError(page);
1727 __set_page_dirty_nobuffers(page);
1731 if (buffer_nilfs_allocated(page_buffers(page))) {
1732 if (TestClearPageWriteback(page))
1733 dec_zone_page_state(page, NR_WRITEBACK);
1735 end_page_writeback(page);
1738 static void nilfs_end_page_io(struct page *page, int err)
1743 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1745 * For b-tree node pages, this function may be called twice
1746 * or more because they might be split in a segment.
1748 if (PageDirty(page)) {
1750 * For pages holding split b-tree node buffers, dirty
1751 * flag on the buffers may be cleared discretely.
1752 * In that case, the page is once redirtied for
1753 * remaining buffers, and it must be cancelled if
1754 * all the buffers get cleaned later.
1757 if (nilfs_page_buffers_clean(page))
1758 __nilfs_clear_page_dirty(page);
1764 __nilfs_end_page_io(page, err);
1767 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1769 struct buffer_head *bh, *head;
1772 while (!list_empty(list)) {
1773 bh = list_entry(list->next, struct buffer_head,
1776 page_cache_get(page);
1777 head = bh = page_buffers(page);
1779 if (!list_empty(&bh->b_assoc_buffers)) {
1780 list_del_init(&bh->b_assoc_buffers);
1782 set_buffer_uptodate(bh);
1783 clear_buffer_dirty(bh);
1784 clear_buffer_nilfs_volatile(bh);
1786 brelse(bh); /* for b_assoc_buffers */
1788 } while ((bh = bh->b_this_page) != head);
1790 __nilfs_end_page_io(page, err);
1791 page_cache_release(page);
1795 static void nilfs_abort_logs(struct list_head *logs, struct page *failed_page,
1798 struct nilfs_segment_buffer *segbuf;
1799 struct page *bd_page = NULL, *fs_page = NULL;
1800 struct buffer_head *bh;
1802 if (list_empty(logs))
1805 list_for_each_entry(segbuf, logs, sb_list) {
1806 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1808 if (bh->b_page != bd_page) {
1810 end_page_writeback(bd_page);
1811 bd_page = bh->b_page;
1815 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1817 if (bh == segbuf->sb_super_root) {
1818 if (bh->b_page != bd_page) {
1819 end_page_writeback(bd_page);
1820 bd_page = bh->b_page;
1824 if (bh->b_page != fs_page) {
1825 nilfs_end_page_io(fs_page, err);
1826 if (fs_page && fs_page == failed_page)
1828 fs_page = bh->b_page;
1833 end_page_writeback(bd_page);
1835 nilfs_end_page_io(fs_page, err);
1838 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1839 struct the_nilfs *nilfs, int err)
1844 list_splice_tail_init(&sci->sc_write_logs, &logs);
1845 ret = nilfs_wait_on_logs(&logs);
1846 nilfs_abort_logs(&logs, NULL, ret ? : err);
1848 list_splice_tail_init(&sci->sc_segbufs, &logs);
1849 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1850 nilfs_free_incomplete_logs(&logs, nilfs);
1851 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1853 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1854 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1858 WARN_ON(ret); /* do not happen */
1861 nilfs_destroy_logs(&logs);
1864 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1865 struct nilfs_segment_buffer *segbuf)
1867 nilfs->ns_segnum = segbuf->sb_segnum;
1868 nilfs->ns_nextnum = segbuf->sb_nextnum;
1869 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1870 + segbuf->sb_sum.nblocks;
1871 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1872 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1875 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1877 struct nilfs_segment_buffer *segbuf;
1878 struct page *bd_page = NULL, *fs_page = NULL;
1879 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
1880 int update_sr = false;
1882 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1883 struct buffer_head *bh;
1885 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1887 set_buffer_uptodate(bh);
1888 clear_buffer_dirty(bh);
1889 if (bh->b_page != bd_page) {
1891 end_page_writeback(bd_page);
1892 bd_page = bh->b_page;
1896 * We assume that the buffers which belong to the same page
1897 * continue over the buffer list.
1898 * Under this assumption, the last BHs of pages is
1899 * identifiable by the discontinuity of bh->b_page
1900 * (page != fs_page).
1902 * For B-tree node blocks, however, this assumption is not
1903 * guaranteed. The cleanup code of B-tree node pages needs
1906 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1908 set_buffer_uptodate(bh);
1909 clear_buffer_dirty(bh);
1910 clear_buffer_nilfs_volatile(bh);
1911 clear_buffer_nilfs_redirected(bh);
1912 if (bh == segbuf->sb_super_root) {
1913 if (bh->b_page != bd_page) {
1914 end_page_writeback(bd_page);
1915 bd_page = bh->b_page;
1920 if (bh->b_page != fs_page) {
1921 nilfs_end_page_io(fs_page, 0);
1922 fs_page = bh->b_page;
1926 if (!nilfs_segbuf_simplex(segbuf)) {
1927 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1928 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1929 sci->sc_lseg_stime = jiffies;
1931 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1932 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1936 * Since pages may continue over multiple segment buffers,
1937 * end of the last page must be checked outside of the loop.
1940 end_page_writeback(bd_page);
1942 nilfs_end_page_io(fs_page, 0);
1944 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1946 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1948 if (nilfs_doing_gc())
1949 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1951 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1953 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1955 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1956 nilfs_set_next_segment(nilfs, segbuf);
1959 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1960 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1962 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1963 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1964 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1965 nilfs_segctor_clear_metadata_dirty(sci);
1967 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1970 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1974 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1976 nilfs_segctor_complete_write(sci);
1977 nilfs_destroy_logs(&sci->sc_write_logs);
1982 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
1983 struct nilfs_sb_info *sbi)
1985 struct nilfs_inode_info *ii, *n;
1986 struct inode *ifile = sci->sc_root->ifile;
1988 spin_lock(&sbi->s_inode_lock);
1990 list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
1992 struct buffer_head *ibh;
1995 spin_unlock(&sbi->s_inode_lock);
1996 err = nilfs_ifile_get_inode_block(
1997 ifile, ii->vfs_inode.i_ino, &ibh);
1998 if (unlikely(err)) {
1999 nilfs_warning(sbi->s_super, __func__,
2000 "failed to get inode block.\n");
2003 nilfs_mdt_mark_buffer_dirty(ibh);
2004 nilfs_mdt_mark_dirty(ifile);
2005 spin_lock(&sbi->s_inode_lock);
2006 if (likely(!ii->i_bh))
2013 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2014 set_bit(NILFS_I_BUSY, &ii->i_state);
2015 list_del(&ii->i_dirty);
2016 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2018 spin_unlock(&sbi->s_inode_lock);
2023 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2024 struct nilfs_sb_info *sbi)
2026 struct nilfs_transaction_info *ti = current->journal_info;
2027 struct nilfs_inode_info *ii, *n;
2029 spin_lock(&sbi->s_inode_lock);
2030 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2031 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2032 test_bit(NILFS_I_DIRTY, &ii->i_state))
2035 clear_bit(NILFS_I_BUSY, &ii->i_state);
2038 list_del(&ii->i_dirty);
2039 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2041 spin_unlock(&sbi->s_inode_lock);
2045 * Main procedure of segment constructor
2047 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2049 struct nilfs_sb_info *sbi = sci->sc_sbi;
2050 struct the_nilfs *nilfs = sbi->s_nilfs;
2051 struct page *failed_page;
2054 sci->sc_stage.scnt = NILFS_ST_INIT;
2055 sci->sc_cno = nilfs->ns_cno;
2057 err = nilfs_segctor_check_in_files(sci, sbi);
2061 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2062 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2064 if (nilfs_segctor_clean(sci))
2068 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2070 err = nilfs_segctor_begin_construction(sci, nilfs);
2074 /* Update time stamp */
2075 sci->sc_seg_ctime = get_seconds();
2077 err = nilfs_segctor_collect(sci, nilfs, mode);
2081 /* Avoid empty segment */
2082 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2083 nilfs_segbuf_empty(sci->sc_curseg)) {
2084 nilfs_segctor_abort_construction(sci, nilfs, 1);
2088 err = nilfs_segctor_assign(sci, mode);
2092 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2093 nilfs_segctor_fill_in_file_bmap(sci);
2095 if (mode == SC_LSEG_SR &&
2096 sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
2097 err = nilfs_segctor_fill_in_checkpoint(sci);
2099 goto failed_to_write;
2101 nilfs_segctor_fill_in_super_root(sci, nilfs);
2103 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2105 /* Write partial segments */
2106 err = nilfs_segctor_prepare_write(sci, &failed_page);
2108 nilfs_abort_logs(&sci->sc_segbufs, failed_page, err);
2109 goto failed_to_write;
2112 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2113 nilfs->ns_crc_seed);
2115 err = nilfs_segctor_write(sci, nilfs);
2117 goto failed_to_write;
2119 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2120 nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2122 * At this point, we avoid double buffering
2123 * for blocksize < pagesize because page dirty
2124 * flag is turned off during write and dirty
2125 * buffers are not properly collected for
2126 * pages crossing over segments.
2128 err = nilfs_segctor_wait(sci);
2130 goto failed_to_write;
2132 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2135 nilfs_segctor_check_out_files(sci, sbi);
2139 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2140 nilfs_redirty_inodes(&sci->sc_dirty_files);
2143 if (nilfs_doing_gc())
2144 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2145 nilfs_segctor_abort_construction(sci, nilfs, err);
2150 * nilfs_segctor_start_timer - set timer of background write
2151 * @sci: nilfs_sc_info
2153 * If the timer has already been set, it ignores the new request.
2154 * This function MUST be called within a section locking the segment
2157 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2159 spin_lock(&sci->sc_state_lock);
2160 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2161 sci->sc_timer.expires = jiffies + sci->sc_interval;
2162 add_timer(&sci->sc_timer);
2163 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2165 spin_unlock(&sci->sc_state_lock);
2168 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2170 spin_lock(&sci->sc_state_lock);
2171 if (!(sci->sc_flush_request & (1 << bn))) {
2172 unsigned long prev_req = sci->sc_flush_request;
2174 sci->sc_flush_request |= (1 << bn);
2176 wake_up(&sci->sc_wait_daemon);
2178 spin_unlock(&sci->sc_state_lock);
2182 * nilfs_flush_segment - trigger a segment construction for resource control
2184 * @ino: inode number of the file to be flushed out.
2186 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2188 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2189 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2191 if (!sci || nilfs_doing_construction())
2193 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2194 /* assign bit 0 to data files */
2197 struct nilfs_segctor_wait_request {
2204 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2206 struct nilfs_segctor_wait_request wait_req;
2209 spin_lock(&sci->sc_state_lock);
2210 init_wait(&wait_req.wq);
2212 atomic_set(&wait_req.done, 0);
2213 wait_req.seq = ++sci->sc_seq_request;
2214 spin_unlock(&sci->sc_state_lock);
2216 init_waitqueue_entry(&wait_req.wq, current);
2217 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2218 set_current_state(TASK_INTERRUPTIBLE);
2219 wake_up(&sci->sc_wait_daemon);
2222 if (atomic_read(&wait_req.done)) {
2226 if (!signal_pending(current)) {
2233 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2237 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2239 struct nilfs_segctor_wait_request *wrq, *n;
2240 unsigned long flags;
2242 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2243 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2245 if (!atomic_read(&wrq->done) &&
2246 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2248 atomic_set(&wrq->done, 1);
2250 if (atomic_read(&wrq->done)) {
2251 wrq->wq.func(&wrq->wq,
2252 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2256 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2260 * nilfs_construct_segment - construct a logical segment
2263 * Return Value: On success, 0 is retured. On errors, one of the following
2264 * negative error code is returned.
2266 * %-EROFS - Read only filesystem.
2270 * %-ENOSPC - No space left on device (only in a panic state).
2272 * %-ERESTARTSYS - Interrupted.
2274 * %-ENOMEM - Insufficient memory available.
2276 int nilfs_construct_segment(struct super_block *sb)
2278 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2279 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2280 struct nilfs_transaction_info *ti;
2286 /* A call inside transactions causes a deadlock. */
2287 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2289 err = nilfs_segctor_sync(sci);
2294 * nilfs_construct_dsync_segment - construct a data-only logical segment
2296 * @inode: inode whose data blocks should be written out
2297 * @start: start byte offset
2298 * @end: end byte offset (inclusive)
2300 * Return Value: On success, 0 is retured. On errors, one of the following
2301 * negative error code is returned.
2303 * %-EROFS - Read only filesystem.
2307 * %-ENOSPC - No space left on device (only in a panic state).
2309 * %-ERESTARTSYS - Interrupted.
2311 * %-ENOMEM - Insufficient memory available.
2313 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2314 loff_t start, loff_t end)
2316 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2317 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2318 struct nilfs_inode_info *ii;
2319 struct nilfs_transaction_info ti;
2325 nilfs_transaction_lock(sbi, &ti, 0);
2327 ii = NILFS_I(inode);
2328 if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2329 nilfs_test_opt(sbi, STRICT_ORDER) ||
2330 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2331 nilfs_discontinued(sbi->s_nilfs)) {
2332 nilfs_transaction_unlock(sbi);
2333 err = nilfs_segctor_sync(sci);
2337 spin_lock(&sbi->s_inode_lock);
2338 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2339 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2340 spin_unlock(&sbi->s_inode_lock);
2341 nilfs_transaction_unlock(sbi);
2344 spin_unlock(&sbi->s_inode_lock);
2345 sci->sc_dsync_inode = ii;
2346 sci->sc_dsync_start = start;
2347 sci->sc_dsync_end = end;
2349 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2351 nilfs_transaction_unlock(sbi);
2355 #define FLUSH_FILE_BIT (0x1) /* data file only */
2356 #define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2359 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2360 * @sci: segment constructor object
2362 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2364 spin_lock(&sci->sc_state_lock);
2365 sci->sc_seq_accepted = sci->sc_seq_request;
2366 spin_unlock(&sci->sc_state_lock);
2367 del_timer_sync(&sci->sc_timer);
2371 * nilfs_segctor_notify - notify the result of request to caller threads
2372 * @sci: segment constructor object
2373 * @mode: mode of log forming
2374 * @err: error code to be notified
2376 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2378 /* Clear requests (even when the construction failed) */
2379 spin_lock(&sci->sc_state_lock);
2381 if (mode == SC_LSEG_SR) {
2382 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2383 sci->sc_seq_done = sci->sc_seq_accepted;
2384 nilfs_segctor_wakeup(sci, err);
2385 sci->sc_flush_request = 0;
2387 if (mode == SC_FLUSH_FILE)
2388 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2389 else if (mode == SC_FLUSH_DAT)
2390 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2392 /* re-enable timer if checkpoint creation was not done */
2393 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2394 time_before(jiffies, sci->sc_timer.expires))
2395 add_timer(&sci->sc_timer);
2397 spin_unlock(&sci->sc_state_lock);
2401 * nilfs_segctor_construct - form logs and write them to disk
2402 * @sci: segment constructor object
2403 * @mode: mode of log forming
2405 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2407 struct nilfs_sb_info *sbi = sci->sc_sbi;
2408 struct the_nilfs *nilfs = sbi->s_nilfs;
2409 struct nilfs_super_block **sbp;
2412 nilfs_segctor_accept(sci);
2414 if (nilfs_discontinued(nilfs))
2416 if (!nilfs_segctor_confirm(sci))
2417 err = nilfs_segctor_do_construct(sci, mode);
2420 if (mode != SC_FLUSH_DAT)
2421 atomic_set(&nilfs->ns_ndirtyblks, 0);
2422 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2423 nilfs_discontinued(nilfs)) {
2424 down_write(&nilfs->ns_sem);
2426 sbp = nilfs_prepare_super(sbi,
2427 nilfs_sb_will_flip(nilfs));
2429 nilfs_set_log_cursor(sbp[0], nilfs);
2430 err = nilfs_commit_super(sbi, NILFS_SB_COMMIT);
2432 up_write(&nilfs->ns_sem);
2436 nilfs_segctor_notify(sci, mode, err);
2440 static void nilfs_construction_timeout(unsigned long data)
2442 struct task_struct *p = (struct task_struct *)data;
2447 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2449 struct nilfs_inode_info *ii, *n;
2451 list_for_each_entry_safe(ii, n, head, i_dirty) {
2452 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2454 list_del_init(&ii->i_dirty);
2455 iput(&ii->vfs_inode);
2459 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2462 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2463 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2464 struct the_nilfs *nilfs = sbi->s_nilfs;
2465 struct nilfs_transaction_info ti;
2471 nilfs_transaction_lock(sbi, &ti, 1);
2473 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2477 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2478 if (unlikely(err)) {
2479 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2483 sci->sc_freesegs = kbufs[4];
2484 sci->sc_nfreesegs = argv[4].v_nmembs;
2485 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2488 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2489 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2494 nilfs_warning(sb, __func__,
2495 "segment construction failed. (err=%d)", err);
2496 set_current_state(TASK_INTERRUPTIBLE);
2497 schedule_timeout(sci->sc_interval);
2499 if (nilfs_test_opt(sbi, DISCARD)) {
2500 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2504 "NILFS warning: error %d on discard request, "
2505 "turning discards off for the device\n", ret);
2506 nilfs_clear_opt(sbi, DISCARD);
2511 sci->sc_freesegs = NULL;
2512 sci->sc_nfreesegs = 0;
2513 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2514 nilfs_transaction_unlock(sbi);
2518 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2520 struct nilfs_sb_info *sbi = sci->sc_sbi;
2521 struct nilfs_transaction_info ti;
2523 nilfs_transaction_lock(sbi, &ti, 0);
2524 nilfs_segctor_construct(sci, mode);
2527 * Unclosed segment should be retried. We do this using sc_timer.
2528 * Timeout of sc_timer will invoke complete construction which leads
2529 * to close the current logical segment.
2531 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2532 nilfs_segctor_start_timer(sci);
2534 nilfs_transaction_unlock(sbi);
2537 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2542 spin_lock(&sci->sc_state_lock);
2543 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2544 SC_FLUSH_DAT : SC_FLUSH_FILE;
2545 spin_unlock(&sci->sc_state_lock);
2548 err = nilfs_segctor_do_construct(sci, mode);
2550 spin_lock(&sci->sc_state_lock);
2551 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2552 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2553 spin_unlock(&sci->sc_state_lock);
2555 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2558 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2560 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2561 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2562 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2563 return SC_FLUSH_FILE;
2564 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2565 return SC_FLUSH_DAT;
2571 * nilfs_segctor_thread - main loop of the segment constructor thread.
2572 * @arg: pointer to a struct nilfs_sc_info.
2574 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2575 * to execute segment constructions.
2577 static int nilfs_segctor_thread(void *arg)
2579 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2580 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2583 sci->sc_timer.data = (unsigned long)current;
2584 sci->sc_timer.function = nilfs_construction_timeout;
2587 sci->sc_task = current;
2588 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2590 "segctord starting. Construction interval = %lu seconds, "
2591 "CP frequency < %lu seconds\n",
2592 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2594 spin_lock(&sci->sc_state_lock);
2599 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2602 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2604 else if (!sci->sc_flush_request)
2607 mode = nilfs_segctor_flush_mode(sci);
2609 spin_unlock(&sci->sc_state_lock);
2610 nilfs_segctor_thread_construct(sci, mode);
2611 spin_lock(&sci->sc_state_lock);
2616 if (freezing(current)) {
2617 spin_unlock(&sci->sc_state_lock);
2619 spin_lock(&sci->sc_state_lock);
2622 int should_sleep = 1;
2624 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2625 TASK_INTERRUPTIBLE);
2627 if (sci->sc_seq_request != sci->sc_seq_done)
2629 else if (sci->sc_flush_request)
2631 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2632 should_sleep = time_before(jiffies,
2633 sci->sc_timer.expires);
2636 spin_unlock(&sci->sc_state_lock);
2638 spin_lock(&sci->sc_state_lock);
2640 finish_wait(&sci->sc_wait_daemon, &wait);
2641 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2642 time_after_eq(jiffies, sci->sc_timer.expires));
2644 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2645 set_nilfs_discontinued(nilfs);
2650 spin_unlock(&sci->sc_state_lock);
2653 sci->sc_task = NULL;
2654 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2658 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2660 struct task_struct *t;
2662 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2664 int err = PTR_ERR(t);
2666 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2670 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2674 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2676 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2678 while (sci->sc_task) {
2679 wake_up(&sci->sc_wait_daemon);
2680 spin_unlock(&sci->sc_state_lock);
2681 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2682 spin_lock(&sci->sc_state_lock);
2687 * Setup & clean-up functions
2689 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi,
2690 struct nilfs_root *root)
2692 struct nilfs_sc_info *sci;
2694 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2699 sci->sc_super = sbi->s_super;
2701 nilfs_get_root(root);
2702 sci->sc_root = root;
2704 init_waitqueue_head(&sci->sc_wait_request);
2705 init_waitqueue_head(&sci->sc_wait_daemon);
2706 init_waitqueue_head(&sci->sc_wait_task);
2707 spin_lock_init(&sci->sc_state_lock);
2708 INIT_LIST_HEAD(&sci->sc_dirty_files);
2709 INIT_LIST_HEAD(&sci->sc_segbufs);
2710 INIT_LIST_HEAD(&sci->sc_write_logs);
2711 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2712 INIT_LIST_HEAD(&sci->sc_copied_buffers);
2713 init_timer(&sci->sc_timer);
2715 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2716 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2717 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2719 if (sbi->s_interval)
2720 sci->sc_interval = sbi->s_interval;
2721 if (sbi->s_watermark)
2722 sci->sc_watermark = sbi->s_watermark;
2726 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2728 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2730 /* The segctord thread was stopped and its timer was removed.
2731 But some tasks remain. */
2733 struct nilfs_sb_info *sbi = sci->sc_sbi;
2734 struct nilfs_transaction_info ti;
2736 nilfs_transaction_lock(sbi, &ti, 0);
2737 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2738 nilfs_transaction_unlock(sbi);
2740 } while (ret && retrycount-- > 0);
2744 * nilfs_segctor_destroy - destroy the segment constructor.
2745 * @sci: nilfs_sc_info
2747 * nilfs_segctor_destroy() kills the segctord thread and frees
2748 * the nilfs_sc_info struct.
2749 * Caller must hold the segment semaphore.
2751 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2753 struct nilfs_sb_info *sbi = sci->sc_sbi;
2756 up_write(&sbi->s_nilfs->ns_segctor_sem);
2758 spin_lock(&sci->sc_state_lock);
2759 nilfs_segctor_kill_thread(sci);
2760 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2761 || sci->sc_seq_request != sci->sc_seq_done);
2762 spin_unlock(&sci->sc_state_lock);
2764 if (flag || !nilfs_segctor_confirm(sci))
2765 nilfs_segctor_write_out(sci);
2767 WARN_ON(!list_empty(&sci->sc_copied_buffers));
2769 if (!list_empty(&sci->sc_dirty_files)) {
2770 nilfs_warning(sbi->s_super, __func__,
2771 "dirty file(s) after the final construction\n");
2772 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2775 WARN_ON(!list_empty(&sci->sc_segbufs));
2776 WARN_ON(!list_empty(&sci->sc_write_logs));
2778 nilfs_put_root(sci->sc_root);
2780 down_write(&sbi->s_nilfs->ns_segctor_sem);
2782 del_timer_sync(&sci->sc_timer);
2787 * nilfs_attach_segment_constructor - attach a segment constructor
2788 * @sbi: nilfs_sb_info
2789 * @root: root object of the current filesystem tree
2791 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2792 * initializes it, and starts the segment constructor.
2794 * Return Value: On success, 0 is returned. On error, one of the following
2795 * negative error code is returned.
2797 * %-ENOMEM - Insufficient memory available.
2799 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
2800 struct nilfs_root *root)
2804 if (NILFS_SC(sbi)) {
2806 * This happens if the filesystem was remounted
2807 * read/write after nilfs_error degenerated it into a
2810 nilfs_detach_segment_constructor(sbi);
2813 sbi->s_sc_info = nilfs_segctor_new(sbi, root);
2814 if (!sbi->s_sc_info)
2817 err = nilfs_segctor_start_thread(NILFS_SC(sbi));
2819 kfree(sbi->s_sc_info);
2820 sbi->s_sc_info = NULL;
2826 * nilfs_detach_segment_constructor - destroy the segment constructor
2827 * @sbi: nilfs_sb_info
2829 * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2830 * frees the struct nilfs_sc_info, and destroy the dirty file list.
2832 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2834 struct the_nilfs *nilfs = sbi->s_nilfs;
2835 LIST_HEAD(garbage_list);
2837 down_write(&nilfs->ns_segctor_sem);
2838 if (NILFS_SC(sbi)) {
2839 nilfs_segctor_destroy(NILFS_SC(sbi));
2840 sbi->s_sc_info = NULL;
2843 /* Force to free the list of dirty files */
2844 spin_lock(&sbi->s_inode_lock);
2845 if (!list_empty(&sbi->s_dirty_files)) {
2846 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2847 nilfs_warning(sbi->s_super, __func__,
2848 "Non empty dirty list after the last "
2849 "segment construction\n");
2851 spin_unlock(&sbi->s_inode_lock);
2852 up_write(&nilfs->ns_segctor_sem);
2854 nilfs_dispose_list(sbi, &garbage_list, 1);