2 * segment.c - NILFS segment constructor.
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
24 #include <linux/pagemap.h>
25 #include <linux/buffer_head.h>
26 #include <linux/writeback.h>
27 #include <linux/bio.h>
28 #include <linux/completion.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/freezer.h>
32 #include <linux/kthread.h>
33 #include <linux/crc32.h>
34 #include <linux/pagevec.h>
49 #define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
51 #define SC_MAX_SEGDELTA 64 /* Upper limit of the number of segments
52 appended in collection retry loop */
54 /* Construction mode */
56 SC_LSEG_SR
= 1, /* Make a logical segment having a super root */
57 SC_LSEG_DSYNC
, /* Flush data blocks of a given file and make
58 a logical segment without a super root */
59 SC_FLUSH_FILE
, /* Flush data files, leads to segment writes without
60 creating a checkpoint */
61 SC_FLUSH_DAT
, /* Flush DAT file. This also creates segments without
65 /* Stage numbers of dirty block collection */
68 NILFS_ST_GC
, /* Collecting dirty blocks for GC */
75 NILFS_ST_SR
, /* Super root */
76 NILFS_ST_DSYNC
, /* Data sync blocks */
80 /* State flags of collection */
81 #define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
82 #define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
83 #define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED)
85 /* Operations depending on the construction mode and file type */
86 struct nilfs_sc_operations
{
87 int (*collect_data
)(struct nilfs_sc_info
*, struct buffer_head
*,
89 int (*collect_node
)(struct nilfs_sc_info
*, struct buffer_head
*,
91 int (*collect_bmap
)(struct nilfs_sc_info
*, struct buffer_head
*,
93 void (*write_data_binfo
)(struct nilfs_sc_info
*,
94 struct nilfs_segsum_pointer
*,
96 void (*write_node_binfo
)(struct nilfs_sc_info
*,
97 struct nilfs_segsum_pointer
*,
104 static void nilfs_segctor_start_timer(struct nilfs_sc_info
*);
105 static void nilfs_segctor_do_flush(struct nilfs_sc_info
*, int);
106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info
*);
107 static void nilfs_dispose_list(struct nilfs_sb_info
*, struct list_head
*,
110 #define nilfs_cnt32_gt(a, b) \
111 (typecheck(__u32, a) && typecheck(__u32, b) && \
112 ((__s32)(b) - (__s32)(a) < 0))
113 #define nilfs_cnt32_ge(a, b) \
114 (typecheck(__u32, a) && typecheck(__u32, b) && \
115 ((__s32)(a) - (__s32)(b) >= 0))
116 #define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
117 #define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
122 static struct kmem_cache
*nilfs_transaction_cachep
;
125 * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
127 * nilfs_init_transaction_cache() creates a slab cache for the struct
128 * nilfs_transaction_info.
130 * Return Value: On success, it returns 0. On error, one of the following
131 * negative error code is returned.
133 * %-ENOMEM - Insufficient memory available.
135 int nilfs_init_transaction_cache(void)
137 nilfs_transaction_cachep
=
138 kmem_cache_create("nilfs2_transaction_cache",
139 sizeof(struct nilfs_transaction_info
),
140 0, SLAB_RECLAIM_ACCOUNT
, NULL
);
141 return (nilfs_transaction_cachep
== NULL
) ? -ENOMEM
: 0;
145 * nilfs_detroy_transaction_cache - destroy the cache for transaction info
147 * nilfs_destroy_transaction_cache() frees the slab cache for the struct
148 * nilfs_transaction_info.
150 void nilfs_destroy_transaction_cache(void)
152 kmem_cache_destroy(nilfs_transaction_cachep
);
155 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info
*ti
)
157 struct nilfs_transaction_info
*cur_ti
= current
->journal_info
;
161 if (cur_ti
->ti_magic
== NILFS_TI_MAGIC
)
162 return ++cur_ti
->ti_count
;
165 * If journal_info field is occupied by other FS,
166 * it is saved and will be restored on
167 * nilfs_transaction_commit().
170 "NILFS warning: journal info from a different "
172 save
= current
->journal_info
;
176 ti
= kmem_cache_alloc(nilfs_transaction_cachep
, GFP_NOFS
);
179 ti
->ti_flags
= NILFS_TI_DYNAMIC_ALLOC
;
185 ti
->ti_magic
= NILFS_TI_MAGIC
;
186 current
->journal_info
= ti
;
191 * nilfs_transaction_begin - start indivisible file operations.
193 * @ti: nilfs_transaction_info
194 * @vacancy_check: flags for vacancy rate checks
196 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
197 * the segment semaphore, to make a segment construction and write tasks
198 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
199 * The region enclosed by these two functions can be nested. To avoid a
200 * deadlock, the semaphore is only acquired or released in the outermost call.
202 * This function allocates a nilfs_transaction_info struct to keep context
203 * information on it. It is initialized and hooked onto the current task in
204 * the outermost call. If a pre-allocated struct is given to @ti, it is used
205 * instead; othewise a new struct is assigned from a slab.
207 * When @vacancy_check flag is set, this function will check the amount of
208 * free space, and will wait for the GC to reclaim disk space if low capacity.
210 * Return Value: On success, 0 is returned. On error, one of the following
211 * negative error code is returned.
213 * %-ENOMEM - Insufficient memory available.
215 * %-ENOSPC - No space left on device
217 int nilfs_transaction_begin(struct super_block
*sb
,
218 struct nilfs_transaction_info
*ti
,
221 struct nilfs_sb_info
*sbi
;
222 struct the_nilfs
*nilfs
;
223 int ret
= nilfs_prepare_segment_lock(ti
);
225 if (unlikely(ret
< 0))
231 nilfs
= sbi
->s_nilfs
;
232 down_read(&nilfs
->ns_segctor_sem
);
233 if (vacancy_check
&& nilfs_near_disk_full(nilfs
)) {
234 up_read(&nilfs
->ns_segctor_sem
);
241 ti
= current
->journal_info
;
242 current
->journal_info
= ti
->ti_save
;
243 if (ti
->ti_flags
& NILFS_TI_DYNAMIC_ALLOC
)
244 kmem_cache_free(nilfs_transaction_cachep
, ti
);
249 * nilfs_transaction_commit - commit indivisible file operations.
252 * nilfs_transaction_commit() releases the read semaphore which is
253 * acquired by nilfs_transaction_begin(). This is only performed
254 * in outermost call of this function. If a commit flag is set,
255 * nilfs_transaction_commit() sets a timer to start the segment
256 * constructor. If a sync flag is set, it starts construction
259 int nilfs_transaction_commit(struct super_block
*sb
)
261 struct nilfs_transaction_info
*ti
= current
->journal_info
;
262 struct nilfs_sb_info
*sbi
;
263 struct nilfs_sc_info
*sci
;
266 BUG_ON(ti
== NULL
|| ti
->ti_magic
!= NILFS_TI_MAGIC
);
267 ti
->ti_flags
|= NILFS_TI_COMMIT
;
268 if (ti
->ti_count
> 0) {
275 if (ti
->ti_flags
& NILFS_TI_COMMIT
)
276 nilfs_segctor_start_timer(sci
);
277 if (atomic_read(&sbi
->s_nilfs
->ns_ndirtyblks
) >
279 nilfs_segctor_do_flush(sci
, 0);
281 up_read(&sbi
->s_nilfs
->ns_segctor_sem
);
282 current
->journal_info
= ti
->ti_save
;
284 if (ti
->ti_flags
& NILFS_TI_SYNC
)
285 err
= nilfs_construct_segment(sb
);
286 if (ti
->ti_flags
& NILFS_TI_DYNAMIC_ALLOC
)
287 kmem_cache_free(nilfs_transaction_cachep
, ti
);
291 void nilfs_transaction_abort(struct super_block
*sb
)
293 struct nilfs_transaction_info
*ti
= current
->journal_info
;
295 BUG_ON(ti
== NULL
|| ti
->ti_magic
!= NILFS_TI_MAGIC
);
296 if (ti
->ti_count
> 0) {
300 up_read(&NILFS_SB(sb
)->s_nilfs
->ns_segctor_sem
);
302 current
->journal_info
= ti
->ti_save
;
303 if (ti
->ti_flags
& NILFS_TI_DYNAMIC_ALLOC
)
304 kmem_cache_free(nilfs_transaction_cachep
, ti
);
307 void nilfs_relax_pressure_in_lock(struct super_block
*sb
)
309 struct nilfs_sb_info
*sbi
= NILFS_SB(sb
);
310 struct nilfs_sc_info
*sci
= NILFS_SC(sbi
);
311 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
313 if (!sci
|| !sci
->sc_flush_request
)
316 set_bit(NILFS_SC_PRIOR_FLUSH
, &sci
->sc_flags
);
317 up_read(&nilfs
->ns_segctor_sem
);
319 down_write(&nilfs
->ns_segctor_sem
);
320 if (sci
->sc_flush_request
&&
321 test_bit(NILFS_SC_PRIOR_FLUSH
, &sci
->sc_flags
)) {
322 struct nilfs_transaction_info
*ti
= current
->journal_info
;
324 ti
->ti_flags
|= NILFS_TI_WRITER
;
325 nilfs_segctor_do_immediate_flush(sci
);
326 ti
->ti_flags
&= ~NILFS_TI_WRITER
;
328 downgrade_write(&nilfs
->ns_segctor_sem
);
331 static void nilfs_transaction_lock(struct nilfs_sb_info
*sbi
,
332 struct nilfs_transaction_info
*ti
,
335 struct nilfs_transaction_info
*cur_ti
= current
->journal_info
;
339 ti
->ti_flags
= NILFS_TI_WRITER
;
341 ti
->ti_save
= cur_ti
;
342 ti
->ti_magic
= NILFS_TI_MAGIC
;
343 INIT_LIST_HEAD(&ti
->ti_garbage
);
344 current
->journal_info
= ti
;
347 down_write(&sbi
->s_nilfs
->ns_segctor_sem
);
348 if (!test_bit(NILFS_SC_PRIOR_FLUSH
, &NILFS_SC(sbi
)->sc_flags
))
351 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi
));
353 up_write(&sbi
->s_nilfs
->ns_segctor_sem
);
357 ti
->ti_flags
|= NILFS_TI_GC
;
360 static void nilfs_transaction_unlock(struct nilfs_sb_info
*sbi
)
362 struct nilfs_transaction_info
*ti
= current
->journal_info
;
364 BUG_ON(ti
== NULL
|| ti
->ti_magic
!= NILFS_TI_MAGIC
);
365 BUG_ON(ti
->ti_count
> 0);
367 up_write(&sbi
->s_nilfs
->ns_segctor_sem
);
368 current
->journal_info
= ti
->ti_save
;
369 if (!list_empty(&ti
->ti_garbage
))
370 nilfs_dispose_list(sbi
, &ti
->ti_garbage
, 0);
373 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info
*sci
,
374 struct nilfs_segsum_pointer
*ssp
,
377 struct nilfs_segment_buffer
*segbuf
= sci
->sc_curseg
;
378 unsigned blocksize
= sci
->sc_super
->s_blocksize
;
381 if (unlikely(ssp
->offset
+ bytes
> blocksize
)) {
383 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp
->bh
,
384 &segbuf
->sb_segsum_buffers
));
385 ssp
->bh
= NILFS_SEGBUF_NEXT_BH(ssp
->bh
);
387 p
= ssp
->bh
->b_data
+ ssp
->offset
;
388 ssp
->offset
+= bytes
;
393 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
394 * @sci: nilfs_sc_info
396 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info
*sci
)
398 struct nilfs_segment_buffer
*segbuf
= sci
->sc_curseg
;
399 struct buffer_head
*sumbh
;
404 if (nilfs_doing_gc())
406 err
= nilfs_segbuf_reset(segbuf
, flags
, sci
->sc_seg_ctime
);
410 sumbh
= NILFS_SEGBUF_FIRST_BH(&segbuf
->sb_segsum_buffers
);
411 sumbytes
= segbuf
->sb_sum
.sumbytes
;
412 sci
->sc_finfo_ptr
.bh
= sumbh
; sci
->sc_finfo_ptr
.offset
= sumbytes
;
413 sci
->sc_binfo_ptr
.bh
= sumbh
; sci
->sc_binfo_ptr
.offset
= sumbytes
;
414 sci
->sc_blk_cnt
= sci
->sc_datablk_cnt
= 0;
418 static int nilfs_segctor_feed_segment(struct nilfs_sc_info
*sci
)
420 sci
->sc_nblk_this_inc
+= sci
->sc_curseg
->sb_sum
.nblocks
;
421 if (NILFS_SEGBUF_IS_LAST(sci
->sc_curseg
, &sci
->sc_segbufs
))
422 return -E2BIG
; /* The current segment is filled up
424 sci
->sc_curseg
= NILFS_NEXT_SEGBUF(sci
->sc_curseg
);
425 return nilfs_segctor_reset_segment_buffer(sci
);
428 static int nilfs_segctor_add_super_root(struct nilfs_sc_info
*sci
)
430 struct nilfs_segment_buffer
*segbuf
= sci
->sc_curseg
;
433 if (segbuf
->sb_sum
.nblocks
>= segbuf
->sb_rest_blocks
) {
434 err
= nilfs_segctor_feed_segment(sci
);
437 segbuf
= sci
->sc_curseg
;
439 err
= nilfs_segbuf_extend_payload(segbuf
, &sci
->sc_super_root
);
441 segbuf
->sb_sum
.flags
|= NILFS_SS_SR
;
446 * Functions for making segment summary and payloads
448 static int nilfs_segctor_segsum_block_required(
449 struct nilfs_sc_info
*sci
, const struct nilfs_segsum_pointer
*ssp
,
452 unsigned blocksize
= sci
->sc_super
->s_blocksize
;
453 /* Size of finfo and binfo is enough small against blocksize */
455 return ssp
->offset
+ binfo_size
+
456 (!sci
->sc_blk_cnt
? sizeof(struct nilfs_finfo
) : 0) >
460 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info
*sci
,
463 sci
->sc_curseg
->sb_sum
.nfinfo
++;
464 sci
->sc_binfo_ptr
= sci
->sc_finfo_ptr
;
465 nilfs_segctor_map_segsum_entry(
466 sci
, &sci
->sc_binfo_ptr
, sizeof(struct nilfs_finfo
));
470 static void nilfs_segctor_end_finfo(struct nilfs_sc_info
*sci
,
473 struct nilfs_finfo
*finfo
;
474 struct nilfs_inode_info
*ii
;
475 struct nilfs_segment_buffer
*segbuf
;
477 if (sci
->sc_blk_cnt
== 0)
481 finfo
= nilfs_segctor_map_segsum_entry(sci
, &sci
->sc_finfo_ptr
,
483 finfo
->fi_ino
= cpu_to_le64(inode
->i_ino
);
484 finfo
->fi_nblocks
= cpu_to_le32(sci
->sc_blk_cnt
);
485 finfo
->fi_ndatablk
= cpu_to_le32(sci
->sc_datablk_cnt
);
486 finfo
->fi_cno
= cpu_to_le64(ii
->i_cno
);
488 segbuf
= sci
->sc_curseg
;
489 segbuf
->sb_sum
.sumbytes
= sci
->sc_binfo_ptr
.offset
+
490 sci
->sc_super
->s_blocksize
* (segbuf
->sb_sum
.nsumblk
- 1);
491 sci
->sc_finfo_ptr
= sci
->sc_binfo_ptr
;
492 sci
->sc_blk_cnt
= sci
->sc_datablk_cnt
= 0;
495 static int nilfs_segctor_add_file_block(struct nilfs_sc_info
*sci
,
496 struct buffer_head
*bh
,
500 struct nilfs_segment_buffer
*segbuf
;
501 int required
, err
= 0;
504 segbuf
= sci
->sc_curseg
;
505 required
= nilfs_segctor_segsum_block_required(
506 sci
, &sci
->sc_binfo_ptr
, binfo_size
);
507 if (segbuf
->sb_sum
.nblocks
+ required
+ 1 > segbuf
->sb_rest_blocks
) {
508 nilfs_segctor_end_finfo(sci
, inode
);
509 err
= nilfs_segctor_feed_segment(sci
);
514 if (unlikely(required
)) {
515 err
= nilfs_segbuf_extend_segsum(segbuf
);
519 if (sci
->sc_blk_cnt
== 0)
520 nilfs_segctor_begin_finfo(sci
, inode
);
522 nilfs_segctor_map_segsum_entry(sci
, &sci
->sc_binfo_ptr
, binfo_size
);
523 /* Substitution to vblocknr is delayed until update_blocknr() */
524 nilfs_segbuf_add_file_buffer(segbuf
, bh
);
530 static int nilfs_handle_bmap_error(int err
, const char *fname
,
531 struct inode
*inode
, struct super_block
*sb
)
533 if (err
== -EINVAL
) {
534 nilfs_error(sb
, fname
, "broken bmap (inode=%lu)\n",
542 * Callback functions that enumerate, mark, and collect dirty blocks
544 static int nilfs_collect_file_data(struct nilfs_sc_info
*sci
,
545 struct buffer_head
*bh
, struct inode
*inode
)
549 /* BUG_ON(!buffer_dirty(bh)); */
550 /* excluded by scan_dirty_data_buffers() */
551 err
= nilfs_bmap_propagate(NILFS_I(inode
)->i_bmap
, bh
);
552 if (unlikely(err
< 0))
553 return nilfs_handle_bmap_error(err
, __func__
, inode
,
556 err
= nilfs_segctor_add_file_block(sci
, bh
, inode
,
557 sizeof(struct nilfs_binfo_v
));
559 sci
->sc_datablk_cnt
++;
563 static int nilfs_collect_file_node(struct nilfs_sc_info
*sci
,
564 struct buffer_head
*bh
,
569 /* BUG_ON(!buffer_dirty(bh)); */
570 /* excluded by scan_dirty_node_buffers() */
571 err
= nilfs_bmap_propagate(NILFS_I(inode
)->i_bmap
, bh
);
572 if (unlikely(err
< 0))
573 return nilfs_handle_bmap_error(err
, __func__
, inode
,
578 static int nilfs_collect_file_bmap(struct nilfs_sc_info
*sci
,
579 struct buffer_head
*bh
,
582 BUG_ON(!buffer_dirty(bh
));
583 return nilfs_segctor_add_file_block(sci
, bh
, inode
, sizeof(__le64
));
586 static void nilfs_write_file_data_binfo(struct nilfs_sc_info
*sci
,
587 struct nilfs_segsum_pointer
*ssp
,
588 union nilfs_binfo
*binfo
)
590 struct nilfs_binfo_v
*binfo_v
= nilfs_segctor_map_segsum_entry(
591 sci
, ssp
, sizeof(*binfo_v
));
592 *binfo_v
= binfo
->bi_v
;
595 static void nilfs_write_file_node_binfo(struct nilfs_sc_info
*sci
,
596 struct nilfs_segsum_pointer
*ssp
,
597 union nilfs_binfo
*binfo
)
599 __le64
*vblocknr
= nilfs_segctor_map_segsum_entry(
600 sci
, ssp
, sizeof(*vblocknr
));
601 *vblocknr
= binfo
->bi_v
.bi_vblocknr
;
604 struct nilfs_sc_operations nilfs_sc_file_ops
= {
605 .collect_data
= nilfs_collect_file_data
,
606 .collect_node
= nilfs_collect_file_node
,
607 .collect_bmap
= nilfs_collect_file_bmap
,
608 .write_data_binfo
= nilfs_write_file_data_binfo
,
609 .write_node_binfo
= nilfs_write_file_node_binfo
,
612 static int nilfs_collect_dat_data(struct nilfs_sc_info
*sci
,
613 struct buffer_head
*bh
, struct inode
*inode
)
617 err
= nilfs_bmap_propagate(NILFS_I(inode
)->i_bmap
, bh
);
618 if (unlikely(err
< 0))
619 return nilfs_handle_bmap_error(err
, __func__
, inode
,
622 err
= nilfs_segctor_add_file_block(sci
, bh
, inode
, sizeof(__le64
));
624 sci
->sc_datablk_cnt
++;
628 static int nilfs_collect_dat_bmap(struct nilfs_sc_info
*sci
,
629 struct buffer_head
*bh
, struct inode
*inode
)
631 BUG_ON(!buffer_dirty(bh
));
632 return nilfs_segctor_add_file_block(sci
, bh
, inode
,
633 sizeof(struct nilfs_binfo_dat
));
636 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info
*sci
,
637 struct nilfs_segsum_pointer
*ssp
,
638 union nilfs_binfo
*binfo
)
640 __le64
*blkoff
= nilfs_segctor_map_segsum_entry(sci
, ssp
,
642 *blkoff
= binfo
->bi_dat
.bi_blkoff
;
645 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info
*sci
,
646 struct nilfs_segsum_pointer
*ssp
,
647 union nilfs_binfo
*binfo
)
649 struct nilfs_binfo_dat
*binfo_dat
=
650 nilfs_segctor_map_segsum_entry(sci
, ssp
, sizeof(*binfo_dat
));
651 *binfo_dat
= binfo
->bi_dat
;
654 struct nilfs_sc_operations nilfs_sc_dat_ops
= {
655 .collect_data
= nilfs_collect_dat_data
,
656 .collect_node
= nilfs_collect_file_node
,
657 .collect_bmap
= nilfs_collect_dat_bmap
,
658 .write_data_binfo
= nilfs_write_dat_data_binfo
,
659 .write_node_binfo
= nilfs_write_dat_node_binfo
,
662 struct nilfs_sc_operations nilfs_sc_dsync_ops
= {
663 .collect_data
= nilfs_collect_file_data
,
664 .collect_node
= NULL
,
665 .collect_bmap
= NULL
,
666 .write_data_binfo
= nilfs_write_file_data_binfo
,
667 .write_node_binfo
= NULL
,
670 static size_t nilfs_lookup_dirty_data_buffers(struct inode
*inode
,
671 struct list_head
*listp
,
673 loff_t start
, loff_t end
)
675 struct address_space
*mapping
= inode
->i_mapping
;
677 pgoff_t index
= 0, last
= ULONG_MAX
;
681 if (unlikely(start
!= 0 || end
!= LLONG_MAX
)) {
683 * A valid range is given for sync-ing data pages. The
684 * range is rounded to per-page; extra dirty buffers
685 * may be included if blocksize < pagesize.
687 index
= start
>> PAGE_SHIFT
;
688 last
= end
>> PAGE_SHIFT
;
690 pagevec_init(&pvec
, 0);
692 if (unlikely(index
> last
) ||
693 !pagevec_lookup_tag(&pvec
, mapping
, &index
, PAGECACHE_TAG_DIRTY
,
694 min_t(pgoff_t
, last
- index
,
695 PAGEVEC_SIZE
- 1) + 1))
698 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
699 struct buffer_head
*bh
, *head
;
700 struct page
*page
= pvec
.pages
[i
];
702 if (unlikely(page
->index
> last
))
707 if (!page_has_buffers(page
))
708 create_empty_buffers(page
,
709 1 << inode
->i_blkbits
, 0);
713 bh
= head
= page_buffers(page
);
715 if (!buffer_dirty(bh
))
718 list_add_tail(&bh
->b_assoc_buffers
, listp
);
720 if (unlikely(ndirties
>= nlimit
)) {
721 pagevec_release(&pvec
);
725 } while (bh
= bh
->b_this_page
, bh
!= head
);
727 pagevec_release(&pvec
);
732 static void nilfs_lookup_dirty_node_buffers(struct inode
*inode
,
733 struct list_head
*listp
)
735 struct nilfs_inode_info
*ii
= NILFS_I(inode
);
736 struct address_space
*mapping
= &ii
->i_btnode_cache
;
738 struct buffer_head
*bh
, *head
;
742 pagevec_init(&pvec
, 0);
744 while (pagevec_lookup_tag(&pvec
, mapping
, &index
, PAGECACHE_TAG_DIRTY
,
746 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
747 bh
= head
= page_buffers(pvec
.pages
[i
]);
749 if (buffer_dirty(bh
)) {
751 list_add_tail(&bh
->b_assoc_buffers
,
754 bh
= bh
->b_this_page
;
755 } while (bh
!= head
);
757 pagevec_release(&pvec
);
762 static void nilfs_dispose_list(struct nilfs_sb_info
*sbi
,
763 struct list_head
*head
, int force
)
765 struct nilfs_inode_info
*ii
, *n
;
766 struct nilfs_inode_info
*ivec
[SC_N_INODEVEC
], **pii
;
769 while (!list_empty(head
)) {
770 spin_lock(&sbi
->s_inode_lock
);
771 list_for_each_entry_safe(ii
, n
, head
, i_dirty
) {
772 list_del_init(&ii
->i_dirty
);
774 if (unlikely(ii
->i_bh
)) {
778 } else if (test_bit(NILFS_I_DIRTY
, &ii
->i_state
)) {
779 set_bit(NILFS_I_QUEUED
, &ii
->i_state
);
780 list_add_tail(&ii
->i_dirty
,
781 &sbi
->s_dirty_files
);
785 if (nv
== SC_N_INODEVEC
)
788 spin_unlock(&sbi
->s_inode_lock
);
790 for (pii
= ivec
; nv
> 0; pii
++, nv
--)
791 iput(&(*pii
)->vfs_inode
);
795 static int nilfs_test_metadata_dirty(struct nilfs_sb_info
*sbi
)
797 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
800 if (nilfs_mdt_fetch_dirty(sbi
->s_ifile
))
802 if (nilfs_mdt_fetch_dirty(nilfs
->ns_cpfile
))
804 if (nilfs_mdt_fetch_dirty(nilfs
->ns_sufile
))
806 if (ret
|| nilfs_doing_gc())
807 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs
)))
812 static int nilfs_segctor_clean(struct nilfs_sc_info
*sci
)
814 return list_empty(&sci
->sc_dirty_files
) &&
815 !test_bit(NILFS_SC_DIRTY
, &sci
->sc_flags
) &&
816 list_empty(&sci
->sc_cleaning_segments
) &&
817 (!nilfs_doing_gc() || list_empty(&sci
->sc_gc_inodes
));
820 static int nilfs_segctor_confirm(struct nilfs_sc_info
*sci
)
822 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
825 if (nilfs_test_metadata_dirty(sbi
))
826 set_bit(NILFS_SC_DIRTY
, &sci
->sc_flags
);
828 spin_lock(&sbi
->s_inode_lock
);
829 if (list_empty(&sbi
->s_dirty_files
) && nilfs_segctor_clean(sci
))
832 spin_unlock(&sbi
->s_inode_lock
);
836 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info
*sci
)
838 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
839 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
841 nilfs_mdt_clear_dirty(sbi
->s_ifile
);
842 nilfs_mdt_clear_dirty(nilfs
->ns_cpfile
);
843 nilfs_mdt_clear_dirty(nilfs
->ns_sufile
);
844 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs
));
847 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info
*sci
)
849 struct the_nilfs
*nilfs
= sci
->sc_sbi
->s_nilfs
;
850 struct buffer_head
*bh_cp
;
851 struct nilfs_checkpoint
*raw_cp
;
854 /* XXX: this interface will be changed */
855 err
= nilfs_cpfile_get_checkpoint(nilfs
->ns_cpfile
, nilfs
->ns_cno
, 1,
858 /* The following code is duplicated with cpfile. But, it is
859 needed to collect the checkpoint even if it was not newly
861 nilfs_mdt_mark_buffer_dirty(bh_cp
);
862 nilfs_mdt_mark_dirty(nilfs
->ns_cpfile
);
863 nilfs_cpfile_put_checkpoint(
864 nilfs
->ns_cpfile
, nilfs
->ns_cno
, bh_cp
);
866 BUG_ON(err
== -EINVAL
|| err
== -ENOENT
);
871 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info
*sci
)
873 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
874 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
875 struct buffer_head
*bh_cp
;
876 struct nilfs_checkpoint
*raw_cp
;
879 err
= nilfs_cpfile_get_checkpoint(nilfs
->ns_cpfile
, nilfs
->ns_cno
, 0,
882 BUG_ON(err
== -EINVAL
|| err
== -ENOENT
);
885 raw_cp
->cp_snapshot_list
.ssl_next
= 0;
886 raw_cp
->cp_snapshot_list
.ssl_prev
= 0;
887 raw_cp
->cp_inodes_count
=
888 cpu_to_le64(atomic_read(&sbi
->s_inodes_count
));
889 raw_cp
->cp_blocks_count
=
890 cpu_to_le64(atomic_read(&sbi
->s_blocks_count
));
891 raw_cp
->cp_nblk_inc
=
892 cpu_to_le64(sci
->sc_nblk_inc
+ sci
->sc_nblk_this_inc
);
893 raw_cp
->cp_create
= cpu_to_le64(sci
->sc_seg_ctime
);
894 raw_cp
->cp_cno
= cpu_to_le64(nilfs
->ns_cno
);
895 if (sci
->sc_sketch_inode
&& i_size_read(sci
->sc_sketch_inode
) > 0)
896 nilfs_checkpoint_set_sketch(raw_cp
);
897 nilfs_write_inode_common(sbi
->s_ifile
, &raw_cp
->cp_ifile_inode
, 1);
898 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, nilfs
->ns_cno
, bh_cp
);
905 static void nilfs_fill_in_file_bmap(struct inode
*ifile
,
906 struct nilfs_inode_info
*ii
)
909 struct buffer_head
*ibh
;
910 struct nilfs_inode
*raw_inode
;
912 if (test_bit(NILFS_I_BMAP
, &ii
->i_state
)) {
915 raw_inode
= nilfs_ifile_map_inode(ifile
, ii
->vfs_inode
.i_ino
,
917 nilfs_bmap_write(ii
->i_bmap
, raw_inode
);
918 nilfs_ifile_unmap_inode(ifile
, ii
->vfs_inode
.i_ino
, ibh
);
922 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info
*sci
,
925 struct nilfs_inode_info
*ii
;
927 list_for_each_entry(ii
, &sci
->sc_dirty_files
, i_dirty
) {
928 nilfs_fill_in_file_bmap(ifile
, ii
);
929 set_bit(NILFS_I_COLLECTED
, &ii
->i_state
);
931 if (sci
->sc_sketch_inode
) {
932 ii
= NILFS_I(sci
->sc_sketch_inode
);
933 if (test_bit(NILFS_I_DIRTY
, &ii
->i_state
))
934 nilfs_fill_in_file_bmap(ifile
, ii
);
939 * CRC calculation routines
941 static void nilfs_fill_in_super_root_crc(struct buffer_head
*bh_sr
, u32 seed
)
943 struct nilfs_super_root
*raw_sr
=
944 (struct nilfs_super_root
*)bh_sr
->b_data
;
947 BUG_ON(NILFS_SR_BYTES
> bh_sr
->b_size
);
949 (unsigned char *)raw_sr
+ sizeof(raw_sr
->sr_sum
),
950 NILFS_SR_BYTES
- sizeof(raw_sr
->sr_sum
));
951 raw_sr
->sr_sum
= cpu_to_le32(crc
);
954 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info
*sci
,
957 struct nilfs_segment_buffer
*segbuf
;
959 if (sci
->sc_super_root
)
960 nilfs_fill_in_super_root_crc(sci
->sc_super_root
, seed
);
962 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
) {
963 nilfs_segbuf_fill_in_segsum_crc(segbuf
, seed
);
964 nilfs_segbuf_fill_in_data_crc(segbuf
, seed
);
968 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info
*sci
,
969 struct the_nilfs
*nilfs
)
971 struct buffer_head
*bh_sr
= sci
->sc_super_root
;
972 struct nilfs_super_root
*raw_sr
=
973 (struct nilfs_super_root
*)bh_sr
->b_data
;
974 unsigned isz
= nilfs
->ns_inode_size
;
976 raw_sr
->sr_bytes
= cpu_to_le16(NILFS_SR_BYTES
);
977 raw_sr
->sr_nongc_ctime
978 = cpu_to_le64(nilfs_doing_gc() ?
979 nilfs
->ns_nongc_ctime
: sci
->sc_seg_ctime
);
980 raw_sr
->sr_flags
= 0;
982 nilfs_mdt_write_inode_direct(
983 nilfs_dat_inode(nilfs
), bh_sr
, NILFS_SR_DAT_OFFSET(isz
));
984 nilfs_mdt_write_inode_direct(
985 nilfs
->ns_cpfile
, bh_sr
, NILFS_SR_CPFILE_OFFSET(isz
));
986 nilfs_mdt_write_inode_direct(
987 nilfs
->ns_sufile
, bh_sr
, NILFS_SR_SUFILE_OFFSET(isz
));
990 static void nilfs_redirty_inodes(struct list_head
*head
)
992 struct nilfs_inode_info
*ii
;
994 list_for_each_entry(ii
, head
, i_dirty
) {
995 if (test_bit(NILFS_I_COLLECTED
, &ii
->i_state
))
996 clear_bit(NILFS_I_COLLECTED
, &ii
->i_state
);
1000 static void nilfs_drop_collected_inodes(struct list_head
*head
)
1002 struct nilfs_inode_info
*ii
;
1004 list_for_each_entry(ii
, head
, i_dirty
) {
1005 if (!test_and_clear_bit(NILFS_I_COLLECTED
, &ii
->i_state
))
1008 clear_bit(NILFS_I_INODE_DIRTY
, &ii
->i_state
);
1009 set_bit(NILFS_I_UPDATED
, &ii
->i_state
);
1013 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info
*sci
,
1014 struct inode
*sufile
)
1017 struct list_head
*head
= &sci
->sc_cleaning_segments
;
1018 struct nilfs_segment_entry
*ent
;
1021 list_for_each_entry(ent
, head
, list
) {
1022 if (!(ent
->flags
& NILFS_SLH_FREED
))
1024 err
= nilfs_sufile_cancel_free(sufile
, ent
->segnum
);
1027 ent
->flags
&= ~NILFS_SLH_FREED
;
1031 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info
*sci
,
1032 struct inode
*sufile
)
1034 struct list_head
*head
= &sci
->sc_cleaning_segments
;
1035 struct nilfs_segment_entry
*ent
;
1038 list_for_each_entry(ent
, head
, list
) {
1039 err
= nilfs_sufile_free(sufile
, ent
->segnum
);
1042 ent
->flags
|= NILFS_SLH_FREED
;
1047 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info
*sci
)
1049 nilfs_dispose_segment_list(&sci
->sc_cleaning_segments
);
1052 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info
*sci
,
1053 struct inode
*inode
,
1054 struct list_head
*listp
,
1055 int (*collect
)(struct nilfs_sc_info
*,
1056 struct buffer_head
*,
1059 struct buffer_head
*bh
, *n
;
1063 list_for_each_entry_safe(bh
, n
, listp
, b_assoc_buffers
) {
1064 list_del_init(&bh
->b_assoc_buffers
);
1065 err
= collect(sci
, bh
, inode
);
1068 goto dispose_buffers
;
1074 while (!list_empty(listp
)) {
1075 bh
= list_entry(listp
->next
, struct buffer_head
,
1077 list_del_init(&bh
->b_assoc_buffers
);
1083 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info
*sci
)
1085 /* Remaining number of blocks within segment buffer */
1086 return sci
->sc_segbuf_nblocks
-
1087 (sci
->sc_nblk_this_inc
+ sci
->sc_curseg
->sb_sum
.nblocks
);
1090 static int nilfs_segctor_scan_file(struct nilfs_sc_info
*sci
,
1091 struct inode
*inode
,
1092 struct nilfs_sc_operations
*sc_ops
)
1094 LIST_HEAD(data_buffers
);
1095 LIST_HEAD(node_buffers
);
1098 if (!(sci
->sc_stage
.flags
& NILFS_CF_NODE
)) {
1099 size_t n
, rest
= nilfs_segctor_buffer_rest(sci
);
1101 n
= nilfs_lookup_dirty_data_buffers(
1102 inode
, &data_buffers
, rest
+ 1, 0, LLONG_MAX
);
1104 err
= nilfs_segctor_apply_buffers(
1105 sci
, inode
, &data_buffers
,
1106 sc_ops
->collect_data
);
1107 BUG_ON(!err
); /* always receive -E2BIG or true error */
1111 nilfs_lookup_dirty_node_buffers(inode
, &node_buffers
);
1113 if (!(sci
->sc_stage
.flags
& NILFS_CF_NODE
)) {
1114 err
= nilfs_segctor_apply_buffers(
1115 sci
, inode
, &data_buffers
, sc_ops
->collect_data
);
1116 if (unlikely(err
)) {
1117 /* dispose node list */
1118 nilfs_segctor_apply_buffers(
1119 sci
, inode
, &node_buffers
, NULL
);
1122 sci
->sc_stage
.flags
|= NILFS_CF_NODE
;
1125 err
= nilfs_segctor_apply_buffers(
1126 sci
, inode
, &node_buffers
, sc_ops
->collect_node
);
1130 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode
)->i_bmap
, &node_buffers
);
1131 err
= nilfs_segctor_apply_buffers(
1132 sci
, inode
, &node_buffers
, sc_ops
->collect_bmap
);
1136 nilfs_segctor_end_finfo(sci
, inode
);
1137 sci
->sc_stage
.flags
&= ~NILFS_CF_NODE
;
1143 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info
*sci
,
1144 struct inode
*inode
)
1146 LIST_HEAD(data_buffers
);
1147 size_t n
, rest
= nilfs_segctor_buffer_rest(sci
);
1150 n
= nilfs_lookup_dirty_data_buffers(inode
, &data_buffers
, rest
+ 1,
1151 sci
->sc_dsync_start
,
1154 err
= nilfs_segctor_apply_buffers(sci
, inode
, &data_buffers
,
1155 nilfs_collect_file_data
);
1157 nilfs_segctor_end_finfo(sci
, inode
);
1159 /* always receive -E2BIG or true error if n > rest */
1164 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info
*sci
, int mode
)
1166 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
1167 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
1168 struct list_head
*head
;
1169 struct nilfs_inode_info
*ii
;
1172 switch (sci
->sc_stage
.scnt
) {
1175 sci
->sc_stage
.flags
= 0;
1177 if (!test_bit(NILFS_SC_UNCLOSED
, &sci
->sc_flags
)) {
1178 sci
->sc_nblk_inc
= 0;
1179 sci
->sc_curseg
->sb_sum
.flags
= NILFS_SS_LOGBGN
;
1180 if (mode
== SC_LSEG_DSYNC
) {
1181 sci
->sc_stage
.scnt
= NILFS_ST_DSYNC
;
1186 sci
->sc_stage
.dirty_file_ptr
= NULL
;
1187 sci
->sc_stage
.gc_inode_ptr
= NULL
;
1188 if (mode
== SC_FLUSH_DAT
) {
1189 sci
->sc_stage
.scnt
= NILFS_ST_DAT
;
1192 sci
->sc_stage
.scnt
++; /* Fall through */
1194 if (nilfs_doing_gc()) {
1195 head
= &sci
->sc_gc_inodes
;
1196 ii
= list_prepare_entry(sci
->sc_stage
.gc_inode_ptr
,
1198 list_for_each_entry_continue(ii
, head
, i_dirty
) {
1199 err
= nilfs_segctor_scan_file(
1200 sci
, &ii
->vfs_inode
,
1201 &nilfs_sc_file_ops
);
1202 if (unlikely(err
)) {
1203 sci
->sc_stage
.gc_inode_ptr
= list_entry(
1205 struct nilfs_inode_info
,
1209 set_bit(NILFS_I_COLLECTED
, &ii
->i_state
);
1211 sci
->sc_stage
.gc_inode_ptr
= NULL
;
1213 sci
->sc_stage
.scnt
++; /* Fall through */
1215 head
= &sci
->sc_dirty_files
;
1216 ii
= list_prepare_entry(sci
->sc_stage
.dirty_file_ptr
, head
,
1218 list_for_each_entry_continue(ii
, head
, i_dirty
) {
1219 clear_bit(NILFS_I_DIRTY
, &ii
->i_state
);
1221 err
= nilfs_segctor_scan_file(sci
, &ii
->vfs_inode
,
1222 &nilfs_sc_file_ops
);
1223 if (unlikely(err
)) {
1224 sci
->sc_stage
.dirty_file_ptr
=
1225 list_entry(ii
->i_dirty
.prev
,
1226 struct nilfs_inode_info
,
1230 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1231 /* XXX: required ? */
1233 sci
->sc_stage
.dirty_file_ptr
= NULL
;
1234 if (mode
== SC_FLUSH_FILE
) {
1235 sci
->sc_stage
.scnt
= NILFS_ST_DONE
;
1238 sci
->sc_stage
.scnt
++; /* Fall through */
1239 case NILFS_ST_SKETCH
:
1240 if (mode
== SC_LSEG_SR
&& sci
->sc_sketch_inode
) {
1241 ii
= NILFS_I(sci
->sc_sketch_inode
);
1242 if (test_bit(NILFS_I_DIRTY
, &ii
->i_state
)) {
1243 sci
->sc_sketch_inode
->i_ctime
.tv_sec
1244 = sci
->sc_seg_ctime
;
1245 sci
->sc_sketch_inode
->i_mtime
.tv_sec
1246 = sci
->sc_seg_ctime
;
1247 err
= nilfs_mark_inode_dirty(
1248 sci
->sc_sketch_inode
);
1252 err
= nilfs_segctor_scan_file(sci
,
1253 sci
->sc_sketch_inode
,
1254 &nilfs_sc_file_ops
);
1258 sci
->sc_stage
.scnt
++;
1259 sci
->sc_stage
.flags
|= NILFS_CF_IFILE_STARTED
;
1261 case NILFS_ST_IFILE
:
1262 err
= nilfs_segctor_scan_file(sci
, sbi
->s_ifile
,
1263 &nilfs_sc_file_ops
);
1266 sci
->sc_stage
.scnt
++;
1267 /* Creating a checkpoint */
1268 err
= nilfs_segctor_create_checkpoint(sci
);
1272 case NILFS_ST_CPFILE
:
1273 err
= nilfs_segctor_scan_file(sci
, nilfs
->ns_cpfile
,
1274 &nilfs_sc_file_ops
);
1277 sci
->sc_stage
.scnt
++; /* Fall through */
1278 case NILFS_ST_SUFILE
:
1279 err
= nilfs_segctor_prepare_free_segments(sci
,
1283 err
= nilfs_segctor_scan_file(sci
, nilfs
->ns_sufile
,
1284 &nilfs_sc_file_ops
);
1287 sci
->sc_stage
.scnt
++; /* Fall through */
1290 err
= nilfs_segctor_scan_file(sci
, nilfs_dat_inode(nilfs
),
1294 if (mode
== SC_FLUSH_DAT
) {
1295 sci
->sc_stage
.scnt
= NILFS_ST_DONE
;
1298 sci
->sc_stage
.scnt
++; /* Fall through */
1300 if (mode
== SC_LSEG_SR
) {
1301 /* Appending a super root */
1302 err
= nilfs_segctor_add_super_root(sci
);
1306 /* End of a logical segment */
1307 sci
->sc_curseg
->sb_sum
.flags
|= NILFS_SS_LOGEND
;
1308 sci
->sc_stage
.scnt
= NILFS_ST_DONE
;
1310 case NILFS_ST_DSYNC
:
1312 sci
->sc_curseg
->sb_sum
.flags
|= NILFS_SS_SYNDT
;
1313 ii
= sci
->sc_dsync_inode
;
1314 if (!test_bit(NILFS_I_BUSY
, &ii
->i_state
))
1317 err
= nilfs_segctor_scan_file_dsync(sci
, &ii
->vfs_inode
);
1320 sci
->sc_curseg
->sb_sum
.flags
|= NILFS_SS_LOGEND
;
1321 sci
->sc_stage
.scnt
= NILFS_ST_DONE
;
1333 static int nilfs_segctor_terminate_segment(struct nilfs_sc_info
*sci
,
1334 struct nilfs_segment_buffer
*segbuf
,
1335 struct inode
*sufile
)
1337 struct nilfs_segment_entry
*ent
= segbuf
->sb_segent
;
1340 err
= nilfs_open_segment_entry(ent
, sufile
);
1343 nilfs_mdt_mark_buffer_dirty(ent
->bh_su
);
1344 nilfs_mdt_mark_dirty(sufile
);
1345 nilfs_close_segment_entry(ent
, sufile
);
1347 list_add_tail(&ent
->list
, &sci
->sc_active_segments
);
1348 segbuf
->sb_segent
= NULL
;
1352 static int nilfs_touch_segusage(struct inode
*sufile
, __u64 segnum
)
1354 struct buffer_head
*bh_su
;
1355 struct nilfs_segment_usage
*raw_su
;
1358 err
= nilfs_sufile_get_segment_usage(sufile
, segnum
, &raw_su
, &bh_su
);
1361 nilfs_mdt_mark_buffer_dirty(bh_su
);
1362 nilfs_mdt_mark_dirty(sufile
);
1363 nilfs_sufile_put_segment_usage(sufile
, segnum
, bh_su
);
1367 static int nilfs_segctor_begin_construction(struct nilfs_sc_info
*sci
,
1368 struct the_nilfs
*nilfs
)
1370 struct nilfs_segment_buffer
*segbuf
, *n
;
1371 struct inode
*sufile
= nilfs
->ns_sufile
;
1375 if (list_empty(&sci
->sc_segbufs
)) {
1376 segbuf
= nilfs_segbuf_new(sci
->sc_super
);
1377 if (unlikely(!segbuf
))
1379 list_add(&segbuf
->sb_list
, &sci
->sc_segbufs
);
1381 segbuf
= NILFS_FIRST_SEGBUF(&sci
->sc_segbufs
);
1383 err
= nilfs_segbuf_map(segbuf
, nilfs
->ns_segnum
,
1384 nilfs
->ns_pseg_offset
, nilfs
);
1388 if (segbuf
->sb_rest_blocks
< NILFS_PSEG_MIN_BLOCKS
) {
1389 err
= nilfs_segctor_terminate_segment(sci
, segbuf
, sufile
);
1393 nilfs_shift_to_next_segment(nilfs
);
1394 err
= nilfs_segbuf_map(segbuf
, nilfs
->ns_segnum
, 0, nilfs
);
1396 sci
->sc_segbuf_nblocks
= segbuf
->sb_rest_blocks
;
1398 err
= nilfs_touch_segusage(sufile
, segbuf
->sb_segnum
);
1402 if (nilfs
->ns_segnum
== nilfs
->ns_nextnum
) {
1403 /* Start from the head of a new full segment */
1404 err
= nilfs_sufile_alloc(sufile
, &nextnum
);
1408 nextnum
= nilfs
->ns_nextnum
;
1410 segbuf
->sb_sum
.seg_seq
= nilfs
->ns_seg_seq
;
1411 nilfs_segbuf_set_next_segnum(segbuf
, nextnum
, nilfs
);
1413 /* truncating segment buffers */
1414 list_for_each_entry_safe_continue(segbuf
, n
, &sci
->sc_segbufs
,
1416 list_del_init(&segbuf
->sb_list
);
1417 nilfs_segbuf_free(segbuf
);
1422 static int nilfs_segctor_extend_segments(struct nilfs_sc_info
*sci
,
1423 struct the_nilfs
*nilfs
, int nadd
)
1425 struct nilfs_segment_buffer
*segbuf
, *prev
, *n
;
1426 struct inode
*sufile
= nilfs
->ns_sufile
;
1431 prev
= NILFS_LAST_SEGBUF(&sci
->sc_segbufs
);
1433 * Since the segment specified with nextnum might be allocated during
1434 * the previous construction, the buffer including its segusage may
1435 * not be dirty. The following call ensures that the buffer is dirty
1436 * and will pin the buffer on memory until the sufile is written.
1438 err
= nilfs_touch_segusage(sufile
, prev
->sb_nextnum
);
1442 for (i
= 0; i
< nadd
; i
++) {
1443 /* extend segment info */
1445 segbuf
= nilfs_segbuf_new(sci
->sc_super
);
1446 if (unlikely(!segbuf
))
1449 /* map this buffer to region of segment on-disk */
1450 err
= nilfs_segbuf_map(segbuf
, prev
->sb_nextnum
, 0, nilfs
);
1454 sci
->sc_segbuf_nblocks
+= segbuf
->sb_rest_blocks
;
1456 /* allocate the next next full segment */
1457 err
= nilfs_sufile_alloc(sufile
, &nextnextnum
);
1461 segbuf
->sb_sum
.seg_seq
= prev
->sb_sum
.seg_seq
+ 1;
1462 nilfs_segbuf_set_next_segnum(segbuf
, nextnextnum
, nilfs
);
1464 list_add_tail(&segbuf
->sb_list
, &list
);
1467 list_splice(&list
, sci
->sc_segbufs
.prev
);
1471 nilfs_segbuf_free(segbuf
);
1473 list_for_each_entry_safe(segbuf
, n
, &list
, sb_list
) {
1474 ret
= nilfs_sufile_free(sufile
, segbuf
->sb_nextnum
);
1476 list_del_init(&segbuf
->sb_list
);
1477 nilfs_segbuf_free(segbuf
);
1482 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info
*sci
,
1483 struct the_nilfs
*nilfs
)
1485 struct nilfs_segment_buffer
*segbuf
;
1488 segbuf
= NILFS_FIRST_SEGBUF(&sci
->sc_segbufs
);
1489 if (nilfs
->ns_nextnum
!= segbuf
->sb_nextnum
) {
1490 ret
= nilfs_sufile_free(nilfs
->ns_sufile
, segbuf
->sb_nextnum
);
1493 if (segbuf
->sb_io_error
) {
1494 /* Case 1: The first segment failed */
1495 if (segbuf
->sb_pseg_start
!= segbuf
->sb_fseg_start
)
1496 /* Case 1a: Partial segment appended into an existing
1498 nilfs_terminate_segment(nilfs
, segbuf
->sb_fseg_start
,
1499 segbuf
->sb_fseg_end
);
1500 else /* Case 1b: New full segment */
1501 set_nilfs_discontinued(nilfs
);
1505 list_for_each_entry_continue(segbuf
, &sci
->sc_segbufs
, sb_list
) {
1506 ret
= nilfs_sufile_free(nilfs
->ns_sufile
, segbuf
->sb_nextnum
);
1508 if (!done
&& segbuf
->sb_io_error
) {
1509 if (segbuf
->sb_segnum
!= nilfs
->ns_nextnum
)
1510 /* Case 2: extended segment (!= next) failed */
1511 nilfs_sufile_set_error(nilfs
->ns_sufile
,
1518 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info
*sci
)
1520 struct nilfs_segment_buffer
*segbuf
;
1522 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
)
1523 nilfs_segbuf_clear(segbuf
);
1524 sci
->sc_super_root
= NULL
;
1527 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info
*sci
)
1529 struct nilfs_segment_buffer
*segbuf
;
1531 while (!list_empty(&sci
->sc_segbufs
)) {
1532 segbuf
= NILFS_FIRST_SEGBUF(&sci
->sc_segbufs
);
1533 list_del_init(&segbuf
->sb_list
);
1534 nilfs_segbuf_free(segbuf
);
1536 /* sci->sc_curseg = NULL; */
1539 static void nilfs_segctor_end_construction(struct nilfs_sc_info
*sci
,
1540 struct the_nilfs
*nilfs
, int err
)
1542 if (unlikely(err
)) {
1543 nilfs_segctor_free_incomplete_segments(sci
, nilfs
);
1544 nilfs_segctor_cancel_free_segments(sci
, nilfs
->ns_sufile
);
1546 nilfs_segctor_clear_segment_buffers(sci
);
1549 static void nilfs_segctor_update_segusage(struct nilfs_sc_info
*sci
,
1550 struct inode
*sufile
)
1552 struct nilfs_segment_buffer
*segbuf
;
1553 struct buffer_head
*bh_su
;
1554 struct nilfs_segment_usage
*raw_su
;
1555 unsigned long live_blocks
;
1558 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
) {
1559 ret
= nilfs_sufile_get_segment_usage(sufile
, segbuf
->sb_segnum
,
1561 BUG_ON(ret
); /* always succeed because bh_su is dirty */
1562 live_blocks
= segbuf
->sb_sum
.nblocks
+
1563 (segbuf
->sb_pseg_start
- segbuf
->sb_fseg_start
);
1564 raw_su
->su_lastmod
= cpu_to_le64(sci
->sc_seg_ctime
);
1565 raw_su
->su_nblocks
= cpu_to_le32(live_blocks
);
1566 nilfs_sufile_put_segment_usage(sufile
, segbuf
->sb_segnum
,
1571 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info
*sci
,
1572 struct inode
*sufile
)
1574 struct nilfs_segment_buffer
*segbuf
;
1575 struct buffer_head
*bh_su
;
1576 struct nilfs_segment_usage
*raw_su
;
1579 segbuf
= NILFS_FIRST_SEGBUF(&sci
->sc_segbufs
);
1580 ret
= nilfs_sufile_get_segment_usage(sufile
, segbuf
->sb_segnum
,
1582 BUG_ON(ret
); /* always succeed because bh_su is dirty */
1583 raw_su
->su_nblocks
= cpu_to_le32(segbuf
->sb_pseg_start
-
1584 segbuf
->sb_fseg_start
);
1585 nilfs_sufile_put_segment_usage(sufile
, segbuf
->sb_segnum
, bh_su
);
1587 list_for_each_entry_continue(segbuf
, &sci
->sc_segbufs
, sb_list
) {
1588 ret
= nilfs_sufile_get_segment_usage(sufile
, segbuf
->sb_segnum
,
1590 BUG_ON(ret
); /* always succeed */
1591 raw_su
->su_nblocks
= 0;
1592 nilfs_sufile_put_segment_usage(sufile
, segbuf
->sb_segnum
,
1597 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info
*sci
,
1598 struct nilfs_segment_buffer
*last
,
1599 struct inode
*sufile
)
1601 struct nilfs_segment_buffer
*segbuf
= last
, *n
;
1604 list_for_each_entry_safe_continue(segbuf
, n
, &sci
->sc_segbufs
,
1606 list_del_init(&segbuf
->sb_list
);
1607 sci
->sc_segbuf_nblocks
-= segbuf
->sb_rest_blocks
;
1608 ret
= nilfs_sufile_free(sufile
, segbuf
->sb_nextnum
);
1610 nilfs_segbuf_free(segbuf
);
1615 static int nilfs_segctor_collect(struct nilfs_sc_info
*sci
,
1616 struct the_nilfs
*nilfs
, int mode
)
1618 struct nilfs_cstage prev_stage
= sci
->sc_stage
;
1621 /* Collection retry loop */
1623 sci
->sc_super_root
= NULL
;
1624 sci
->sc_nblk_this_inc
= 0;
1625 sci
->sc_curseg
= NILFS_FIRST_SEGBUF(&sci
->sc_segbufs
);
1627 err
= nilfs_segctor_reset_segment_buffer(sci
);
1631 err
= nilfs_segctor_collect_blocks(sci
, mode
);
1632 sci
->sc_nblk_this_inc
+= sci
->sc_curseg
->sb_sum
.nblocks
;
1636 if (unlikely(err
!= -E2BIG
))
1639 /* The current segment is filled up */
1640 if (mode
!= SC_LSEG_SR
|| sci
->sc_stage
.scnt
< NILFS_ST_CPFILE
)
1643 nilfs_segctor_cancel_free_segments(sci
, nilfs
->ns_sufile
);
1644 nilfs_segctor_clear_segment_buffers(sci
);
1646 err
= nilfs_segctor_extend_segments(sci
, nilfs
, nadd
);
1650 nadd
= min_t(int, nadd
<< 1, SC_MAX_SEGDELTA
);
1651 sci
->sc_stage
= prev_stage
;
1653 nilfs_segctor_truncate_segments(sci
, sci
->sc_curseg
, nilfs
->ns_sufile
);
1660 static void nilfs_list_replace_buffer(struct buffer_head
*old_bh
,
1661 struct buffer_head
*new_bh
)
1663 BUG_ON(!list_empty(&new_bh
->b_assoc_buffers
));
1665 list_replace_init(&old_bh
->b_assoc_buffers
, &new_bh
->b_assoc_buffers
);
1666 /* The caller must release old_bh */
1670 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info
*sci
,
1671 struct nilfs_segment_buffer
*segbuf
,
1674 struct inode
*inode
= NULL
;
1676 unsigned long nfinfo
= segbuf
->sb_sum
.nfinfo
;
1677 unsigned long nblocks
= 0, ndatablk
= 0;
1678 struct nilfs_sc_operations
*sc_op
= NULL
;
1679 struct nilfs_segsum_pointer ssp
;
1680 struct nilfs_finfo
*finfo
= NULL
;
1681 union nilfs_binfo binfo
;
1682 struct buffer_head
*bh
, *bh_org
;
1689 blocknr
= segbuf
->sb_pseg_start
+ segbuf
->sb_sum
.nsumblk
;
1690 ssp
.bh
= NILFS_SEGBUF_FIRST_BH(&segbuf
->sb_segsum_buffers
);
1691 ssp
.offset
= sizeof(struct nilfs_segment_summary
);
1693 list_for_each_entry(bh
, &segbuf
->sb_payload_buffers
, b_assoc_buffers
) {
1694 if (bh
== sci
->sc_super_root
)
1697 finfo
= nilfs_segctor_map_segsum_entry(
1698 sci
, &ssp
, sizeof(*finfo
));
1699 ino
= le64_to_cpu(finfo
->fi_ino
);
1700 nblocks
= le32_to_cpu(finfo
->fi_nblocks
);
1701 ndatablk
= le32_to_cpu(finfo
->fi_ndatablk
);
1703 if (buffer_nilfs_node(bh
))
1704 inode
= NILFS_BTNC_I(bh
->b_page
->mapping
);
1706 inode
= NILFS_AS_I(bh
->b_page
->mapping
);
1708 if (mode
== SC_LSEG_DSYNC
)
1709 sc_op
= &nilfs_sc_dsync_ops
;
1710 else if (ino
== NILFS_DAT_INO
)
1711 sc_op
= &nilfs_sc_dat_ops
;
1712 else /* file blocks */
1713 sc_op
= &nilfs_sc_file_ops
;
1717 err
= nilfs_bmap_assign(NILFS_I(inode
)->i_bmap
, &bh
, blocknr
,
1720 nilfs_list_replace_buffer(bh_org
, bh
);
1726 sc_op
->write_data_binfo(sci
, &ssp
, &binfo
);
1728 sc_op
->write_node_binfo(sci
, &ssp
, &binfo
);
1731 if (--nblocks
== 0) {
1735 } else if (ndatablk
> 0)
1742 err
= nilfs_handle_bmap_error(err
, __func__
, inode
, sci
->sc_super
);
1746 static int nilfs_segctor_assign(struct nilfs_sc_info
*sci
, int mode
)
1748 struct nilfs_segment_buffer
*segbuf
;
1751 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
) {
1752 err
= nilfs_segctor_update_payload_blocknr(sci
, segbuf
, mode
);
1755 nilfs_segbuf_fill_in_segsum(segbuf
);
1761 nilfs_copy_replace_page_buffers(struct page
*page
, struct list_head
*out
)
1763 struct page
*clone_page
;
1764 struct buffer_head
*bh
, *head
, *bh2
;
1767 bh
= head
= page_buffers(page
);
1769 clone_page
= nilfs_alloc_private_page(bh
->b_bdev
, bh
->b_size
, 0);
1770 if (unlikely(!clone_page
))
1773 bh2
= page_buffers(clone_page
);
1774 kaddr
= kmap_atomic(page
, KM_USER0
);
1776 if (list_empty(&bh
->b_assoc_buffers
))
1779 page_cache_get(clone_page
); /* for each bh */
1780 memcpy(bh2
->b_data
, kaddr
+ bh_offset(bh
), bh2
->b_size
);
1781 bh2
->b_blocknr
= bh
->b_blocknr
;
1782 list_replace(&bh
->b_assoc_buffers
, &bh2
->b_assoc_buffers
);
1783 list_add_tail(&bh
->b_assoc_buffers
, out
);
1784 } while (bh
= bh
->b_this_page
, bh2
= bh2
->b_this_page
, bh
!= head
);
1785 kunmap_atomic(kaddr
, KM_USER0
);
1787 if (!TestSetPageWriteback(clone_page
))
1788 inc_zone_page_state(clone_page
, NR_WRITEBACK
);
1789 unlock_page(clone_page
);
1794 static int nilfs_test_page_to_be_frozen(struct page
*page
)
1796 struct address_space
*mapping
= page
->mapping
;
1798 if (!mapping
|| !mapping
->host
|| S_ISDIR(mapping
->host
->i_mode
))
1801 if (page_mapped(page
)) {
1802 ClearPageChecked(page
);
1805 return PageChecked(page
);
1808 static int nilfs_begin_page_io(struct page
*page
, struct list_head
*out
)
1810 if (!page
|| PageWriteback(page
))
1811 /* For split b-tree node pages, this function may be called
1812 twice. We ignore the 2nd or later calls by this check. */
1816 clear_page_dirty_for_io(page
);
1817 set_page_writeback(page
);
1820 if (nilfs_test_page_to_be_frozen(page
)) {
1821 int err
= nilfs_copy_replace_page_buffers(page
, out
);
1828 static int nilfs_segctor_prepare_write(struct nilfs_sc_info
*sci
,
1829 struct page
**failed_page
)
1831 struct nilfs_segment_buffer
*segbuf
;
1832 struct page
*bd_page
= NULL
, *fs_page
= NULL
;
1833 struct list_head
*list
= &sci
->sc_copied_buffers
;
1836 *failed_page
= NULL
;
1837 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
) {
1838 struct buffer_head
*bh
;
1840 list_for_each_entry(bh
, &segbuf
->sb_segsum_buffers
,
1842 if (bh
->b_page
!= bd_page
) {
1845 clear_page_dirty_for_io(bd_page
);
1846 set_page_writeback(bd_page
);
1847 unlock_page(bd_page
);
1849 bd_page
= bh
->b_page
;
1853 list_for_each_entry(bh
, &segbuf
->sb_payload_buffers
,
1855 if (bh
== sci
->sc_super_root
) {
1856 if (bh
->b_page
!= bd_page
) {
1858 clear_page_dirty_for_io(bd_page
);
1859 set_page_writeback(bd_page
);
1860 unlock_page(bd_page
);
1861 bd_page
= bh
->b_page
;
1865 if (bh
->b_page
!= fs_page
) {
1866 err
= nilfs_begin_page_io(fs_page
, list
);
1867 if (unlikely(err
)) {
1868 *failed_page
= fs_page
;
1871 fs_page
= bh
->b_page
;
1877 clear_page_dirty_for_io(bd_page
);
1878 set_page_writeback(bd_page
);
1879 unlock_page(bd_page
);
1881 err
= nilfs_begin_page_io(fs_page
, list
);
1883 *failed_page
= fs_page
;
1888 static int nilfs_segctor_write(struct nilfs_sc_info
*sci
,
1889 struct backing_dev_info
*bdi
)
1891 struct nilfs_segment_buffer
*segbuf
;
1892 struct nilfs_write_info wi
;
1895 wi
.sb
= sci
->sc_super
;
1896 wi
.bh_sr
= sci
->sc_super_root
;
1899 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
) {
1900 nilfs_segbuf_prepare_write(segbuf
, &wi
);
1901 err
= nilfs_segbuf_write(segbuf
, &wi
);
1903 res
= nilfs_segbuf_wait(segbuf
, &wi
);
1904 err
= unlikely(err
) ? : res
;
1911 static int nilfs_page_has_uncleared_buffer(struct page
*page
)
1913 struct buffer_head
*head
, *bh
;
1915 head
= bh
= page_buffers(page
);
1917 if (buffer_dirty(bh
) && !list_empty(&bh
->b_assoc_buffers
))
1919 bh
= bh
->b_this_page
;
1920 } while (bh
!= head
);
1924 static void __nilfs_end_page_io(struct page
*page
, int err
)
1926 /* BUG_ON(err > 0); */
1928 if (!nilfs_page_buffers_clean(page
))
1929 __set_page_dirty_nobuffers(page
);
1930 ClearPageError(page
);
1932 __set_page_dirty_nobuffers(page
);
1936 if (buffer_nilfs_allocated(page_buffers(page
))) {
1937 if (TestClearPageWriteback(page
))
1938 dec_zone_page_state(page
, NR_WRITEBACK
);
1940 end_page_writeback(page
);
1943 static void nilfs_end_page_io(struct page
*page
, int err
)
1948 if (buffer_nilfs_node(page_buffers(page
)) &&
1949 nilfs_page_has_uncleared_buffer(page
))
1950 /* For b-tree node pages, this function may be called twice
1951 or more because they might be split in a segment.
1952 This check assures that cleanup has been done for all
1953 buffers in a split btnode page. */
1956 __nilfs_end_page_io(page
, err
);
1959 static void nilfs_clear_copied_buffers(struct list_head
*list
, int err
)
1961 struct buffer_head
*bh
, *head
;
1964 while (!list_empty(list
)) {
1965 bh
= list_entry(list
->next
, struct buffer_head
,
1968 page_cache_get(page
);
1969 head
= bh
= page_buffers(page
);
1971 if (!list_empty(&bh
->b_assoc_buffers
)) {
1972 list_del_init(&bh
->b_assoc_buffers
);
1974 set_buffer_uptodate(bh
);
1975 clear_buffer_dirty(bh
);
1976 clear_buffer_nilfs_volatile(bh
);
1978 brelse(bh
); /* for b_assoc_buffers */
1980 } while ((bh
= bh
->b_this_page
) != head
);
1982 __nilfs_end_page_io(page
, err
);
1983 page_cache_release(page
);
1987 static void nilfs_segctor_abort_write(struct nilfs_sc_info
*sci
,
1988 struct page
*failed_page
, int err
)
1990 struct nilfs_segment_buffer
*segbuf
;
1991 struct page
*bd_page
= NULL
, *fs_page
= NULL
;
1993 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
) {
1994 struct buffer_head
*bh
;
1996 list_for_each_entry(bh
, &segbuf
->sb_segsum_buffers
,
1998 if (bh
->b_page
!= bd_page
) {
2000 end_page_writeback(bd_page
);
2001 bd_page
= bh
->b_page
;
2005 list_for_each_entry(bh
, &segbuf
->sb_payload_buffers
,
2007 if (bh
== sci
->sc_super_root
) {
2008 if (bh
->b_page
!= bd_page
) {
2009 end_page_writeback(bd_page
);
2010 bd_page
= bh
->b_page
;
2014 if (bh
->b_page
!= fs_page
) {
2015 nilfs_end_page_io(fs_page
, err
);
2016 if (unlikely(fs_page
== failed_page
))
2018 fs_page
= bh
->b_page
;
2023 end_page_writeback(bd_page
);
2025 nilfs_end_page_io(fs_page
, err
);
2027 nilfs_clear_copied_buffers(&sci
->sc_copied_buffers
, err
);
2030 static void nilfs_set_next_segment(struct the_nilfs
*nilfs
,
2031 struct nilfs_segment_buffer
*segbuf
)
2033 nilfs
->ns_segnum
= segbuf
->sb_segnum
;
2034 nilfs
->ns_nextnum
= segbuf
->sb_nextnum
;
2035 nilfs
->ns_pseg_offset
= segbuf
->sb_pseg_start
- segbuf
->sb_fseg_start
2036 + segbuf
->sb_sum
.nblocks
;
2037 nilfs
->ns_seg_seq
= segbuf
->sb_sum
.seg_seq
;
2038 nilfs
->ns_ctime
= segbuf
->sb_sum
.ctime
;
2041 static void nilfs_segctor_complete_write(struct nilfs_sc_info
*sci
)
2043 struct nilfs_segment_buffer
*segbuf
;
2044 struct page
*bd_page
= NULL
, *fs_page
= NULL
;
2045 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
2046 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
2047 int update_sr
= (sci
->sc_super_root
!= NULL
);
2049 list_for_each_entry(segbuf
, &sci
->sc_segbufs
, sb_list
) {
2050 struct buffer_head
*bh
;
2052 list_for_each_entry(bh
, &segbuf
->sb_segsum_buffers
,
2054 set_buffer_uptodate(bh
);
2055 clear_buffer_dirty(bh
);
2056 if (bh
->b_page
!= bd_page
) {
2058 end_page_writeback(bd_page
);
2059 bd_page
= bh
->b_page
;
2063 * We assume that the buffers which belong to the same page
2064 * continue over the buffer list.
2065 * Under this assumption, the last BHs of pages is
2066 * identifiable by the discontinuity of bh->b_page
2067 * (page != fs_page).
2069 * For B-tree node blocks, however, this assumption is not
2070 * guaranteed. The cleanup code of B-tree node pages needs
2073 list_for_each_entry(bh
, &segbuf
->sb_payload_buffers
,
2075 set_buffer_uptodate(bh
);
2076 clear_buffer_dirty(bh
);
2077 clear_buffer_nilfs_volatile(bh
);
2078 if (bh
== sci
->sc_super_root
) {
2079 if (bh
->b_page
!= bd_page
) {
2080 end_page_writeback(bd_page
);
2081 bd_page
= bh
->b_page
;
2085 if (bh
->b_page
!= fs_page
) {
2086 nilfs_end_page_io(fs_page
, 0);
2087 fs_page
= bh
->b_page
;
2091 if (!NILFS_SEG_SIMPLEX(&segbuf
->sb_sum
)) {
2092 if (NILFS_SEG_LOGBGN(&segbuf
->sb_sum
)) {
2093 set_bit(NILFS_SC_UNCLOSED
, &sci
->sc_flags
);
2094 sci
->sc_lseg_stime
= jiffies
;
2096 if (NILFS_SEG_LOGEND(&segbuf
->sb_sum
))
2097 clear_bit(NILFS_SC_UNCLOSED
, &sci
->sc_flags
);
2101 * Since pages may continue over multiple segment buffers,
2102 * end of the last page must be checked outside of the loop.
2105 end_page_writeback(bd_page
);
2107 nilfs_end_page_io(fs_page
, 0);
2109 nilfs_clear_copied_buffers(&sci
->sc_copied_buffers
, 0);
2111 nilfs_drop_collected_inodes(&sci
->sc_dirty_files
);
2113 if (nilfs_doing_gc()) {
2114 nilfs_drop_collected_inodes(&sci
->sc_gc_inodes
);
2116 nilfs_commit_gcdat_inode(nilfs
);
2118 nilfs
->ns_nongc_ctime
= sci
->sc_seg_ctime
;
2120 sci
->sc_nblk_inc
+= sci
->sc_nblk_this_inc
;
2122 segbuf
= NILFS_LAST_SEGBUF(&sci
->sc_segbufs
);
2123 nilfs_set_next_segment(nilfs
, segbuf
);
2126 nilfs_set_last_segment(nilfs
, segbuf
->sb_pseg_start
,
2127 segbuf
->sb_sum
.seg_seq
, nilfs
->ns_cno
);
2129 clear_bit(NILFS_SC_DIRTY
, &sci
->sc_flags
);
2130 set_bit(NILFS_SC_SUPER_ROOT
, &sci
->sc_flags
);
2132 clear_bit(NILFS_SC_SUPER_ROOT
, &sci
->sc_flags
);
2135 static int nilfs_segctor_check_in_files(struct nilfs_sc_info
*sci
,
2136 struct nilfs_sb_info
*sbi
)
2138 struct nilfs_inode_info
*ii
, *n
;
2139 __u64 cno
= sbi
->s_nilfs
->ns_cno
;
2141 spin_lock(&sbi
->s_inode_lock
);
2143 list_for_each_entry_safe(ii
, n
, &sbi
->s_dirty_files
, i_dirty
) {
2145 struct buffer_head
*ibh
;
2148 spin_unlock(&sbi
->s_inode_lock
);
2149 err
= nilfs_ifile_get_inode_block(
2150 sbi
->s_ifile
, ii
->vfs_inode
.i_ino
, &ibh
);
2151 if (unlikely(err
)) {
2152 nilfs_warning(sbi
->s_super
, __func__
,
2153 "failed to get inode block.\n");
2156 nilfs_mdt_mark_buffer_dirty(ibh
);
2157 nilfs_mdt_mark_dirty(sbi
->s_ifile
);
2158 spin_lock(&sbi
->s_inode_lock
);
2159 if (likely(!ii
->i_bh
))
2167 clear_bit(NILFS_I_QUEUED
, &ii
->i_state
);
2168 set_bit(NILFS_I_BUSY
, &ii
->i_state
);
2169 list_del(&ii
->i_dirty
);
2170 list_add_tail(&ii
->i_dirty
, &sci
->sc_dirty_files
);
2172 spin_unlock(&sbi
->s_inode_lock
);
2174 NILFS_I(sbi
->s_ifile
)->i_cno
= cno
;
2179 static void nilfs_segctor_check_out_files(struct nilfs_sc_info
*sci
,
2180 struct nilfs_sb_info
*sbi
)
2182 struct nilfs_transaction_info
*ti
= current
->journal_info
;
2183 struct nilfs_inode_info
*ii
, *n
;
2184 __u64 cno
= sbi
->s_nilfs
->ns_cno
;
2186 spin_lock(&sbi
->s_inode_lock
);
2187 list_for_each_entry_safe(ii
, n
, &sci
->sc_dirty_files
, i_dirty
) {
2188 if (!test_and_clear_bit(NILFS_I_UPDATED
, &ii
->i_state
) ||
2189 test_bit(NILFS_I_DIRTY
, &ii
->i_state
)) {
2190 /* The current checkpoint number (=nilfs->ns_cno) is
2191 changed between check-in and check-out only if the
2192 super root is written out. So, we can update i_cno
2193 for the inodes that remain in the dirty list. */
2197 clear_bit(NILFS_I_BUSY
, &ii
->i_state
);
2200 list_del(&ii
->i_dirty
);
2201 list_add_tail(&ii
->i_dirty
, &ti
->ti_garbage
);
2203 spin_unlock(&sbi
->s_inode_lock
);
2207 * Nasty routines to manipulate active flags on sufile.
2208 * These would be removed in a future release.
2210 static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info
*sci
,
2211 struct the_nilfs
*nilfs
)
2213 struct nilfs_segment_buffer
*segbuf
, *last
;
2214 struct nilfs_segment_entry
*ent
, *n
;
2215 struct inode
*sufile
= nilfs
->ns_sufile
;
2216 struct list_head
*head
;
2218 last
= NILFS_LAST_SEGBUF(&sci
->sc_segbufs
);
2219 nilfs_for_each_segbuf_before(segbuf
, last
, &sci
->sc_segbufs
) {
2220 ent
= segbuf
->sb_segent
;
2222 break; /* ignore unmapped segments (should check it?)*/
2223 nilfs_segment_usage_set_active(ent
->raw_su
);
2224 nilfs_close_segment_entry(ent
, sufile
);
2227 head
= &sci
->sc_active_segments
;
2228 list_for_each_entry_safe(ent
, n
, head
, list
) {
2229 nilfs_segment_usage_set_active(ent
->raw_su
);
2230 nilfs_close_segment_entry(ent
, sufile
);
2233 down_write(&nilfs
->ns_sem
);
2234 head
= &nilfs
->ns_used_segments
;
2235 list_for_each_entry(ent
, head
, list
) {
2236 nilfs_segment_usage_set_volatile_active(ent
->raw_su
);
2238 up_write(&nilfs
->ns_sem
);
2241 static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info
*sci
,
2242 struct the_nilfs
*nilfs
)
2244 struct nilfs_segment_buffer
*segbuf
, *last
;
2245 struct nilfs_segment_entry
*ent
;
2246 struct inode
*sufile
= nilfs
->ns_sufile
;
2247 struct list_head
*head
;
2250 last
= NILFS_LAST_SEGBUF(&sci
->sc_segbufs
);
2251 nilfs_for_each_segbuf_before(segbuf
, last
, &sci
->sc_segbufs
) {
2253 * Deactivate ongoing full segments. The last segment is kept
2254 * active because it is a start point of recovery, and is not
2255 * relocatable until the super block points to a newer
2258 ent
= segbuf
->sb_segent
;
2260 break; /* ignore unmapped segments (should check it?)*/
2261 err
= nilfs_open_segment_entry(ent
, sufile
);
2264 nilfs_segment_usage_clear_active(ent
->raw_su
);
2265 BUG_ON(!buffer_dirty(ent
->bh_su
));
2268 head
= &sci
->sc_active_segments
;
2269 list_for_each_entry(ent
, head
, list
) {
2270 err
= nilfs_open_segment_entry(ent
, sufile
);
2273 nilfs_segment_usage_clear_active(ent
->raw_su
);
2274 BUG_ON(!buffer_dirty(ent
->bh_su
));
2277 down_write(&nilfs
->ns_sem
);
2278 head
= &nilfs
->ns_used_segments
;
2279 list_for_each_entry(ent
, head
, list
) {
2280 /* clear volatile active for segments of older generations */
2281 nilfs_segment_usage_clear_volatile_active(ent
->raw_su
);
2283 up_write(&nilfs
->ns_sem
);
2287 nilfs_segctor_reactivate_segments(sci
, nilfs
);
2291 static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info
*sci
)
2293 struct nilfs_segment_buffer
*segbuf
, *last
;
2294 struct nilfs_segment_entry
*ent
;
2296 /* move each segbuf->sb_segent to the list of used active segments */
2297 last
= NILFS_LAST_SEGBUF(&sci
->sc_segbufs
);
2298 nilfs_for_each_segbuf_before(segbuf
, last
, &sci
->sc_segbufs
) {
2299 ent
= segbuf
->sb_segent
;
2301 break; /* ignore unmapped segments (should check it?)*/
2302 list_add_tail(&ent
->list
, &sci
->sc_active_segments
);
2303 segbuf
->sb_segent
= NULL
;
2308 __nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info
*sci
,
2309 struct the_nilfs
*nilfs
)
2312 struct nilfs_segment_entry
*ent
;
2314 list_splice_init(&sci
->sc_active_segments
,
2315 nilfs
->ns_used_segments
.prev
);
2317 list_for_each_entry(ent
, &nilfs
->ns_used_segments
, list
) {
2318 nilfs_segment_usage_set_volatile_active(ent
->raw_su
);
2319 /* These segments are kept open */
2324 * Main procedure of segment constructor
2326 static int nilfs_segctor_do_construct(struct nilfs_sc_info
*sci
, int mode
)
2328 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
2329 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
2330 struct page
*failed_page
;
2331 int err
, has_sr
= 0;
2333 sci
->sc_stage
.scnt
= NILFS_ST_INIT
;
2335 err
= nilfs_segctor_check_in_files(sci
, sbi
);
2339 if (nilfs_test_metadata_dirty(sbi
))
2340 set_bit(NILFS_SC_DIRTY
, &sci
->sc_flags
);
2342 if (nilfs_segctor_clean(sci
))
2346 sci
->sc_stage
.flags
&= ~NILFS_CF_HISTORY_MASK
;
2348 err
= nilfs_segctor_begin_construction(sci
, nilfs
);
2352 /* Update time stamp */
2353 sci
->sc_seg_ctime
= get_seconds();
2355 err
= nilfs_segctor_collect(sci
, nilfs
, mode
);
2359 has_sr
= (sci
->sc_super_root
!= NULL
);
2361 /* Avoid empty segment */
2362 if (sci
->sc_stage
.scnt
== NILFS_ST_DONE
&&
2363 NILFS_SEG_EMPTY(&sci
->sc_curseg
->sb_sum
)) {
2364 BUG_ON(mode
== SC_LSEG_SR
);
2365 nilfs_segctor_end_construction(sci
, nilfs
, 1);
2369 err
= nilfs_segctor_assign(sci
, mode
);
2374 err
= nilfs_segctor_deactivate_segments(sci
, nilfs
);
2378 if (sci
->sc_stage
.flags
& NILFS_CF_IFILE_STARTED
)
2379 nilfs_segctor_fill_in_file_bmap(sci
, sbi
->s_ifile
);
2382 err
= nilfs_segctor_fill_in_checkpoint(sci
);
2384 goto failed_to_make_up
;
2386 nilfs_segctor_fill_in_super_root(sci
, nilfs
);
2388 nilfs_segctor_update_segusage(sci
, nilfs
->ns_sufile
);
2390 /* Write partial segments */
2391 err
= nilfs_segctor_prepare_write(sci
, &failed_page
);
2393 goto failed_to_write
;
2395 nilfs_segctor_fill_in_checksums(sci
, nilfs
->ns_crc_seed
);
2397 err
= nilfs_segctor_write(sci
, nilfs
->ns_bdi
);
2399 goto failed_to_write
;
2401 nilfs_segctor_complete_write(sci
);
2403 /* Commit segments */
2404 nilfs_segctor_bead_completed_segments(sci
);
2406 down_write(&nilfs
->ns_sem
);
2407 nilfs_update_last_segment(sbi
, 1);
2408 __nilfs_segctor_commit_deactivate_segments(sci
, nilfs
);
2409 up_write(&nilfs
->ns_sem
);
2410 nilfs_segctor_commit_free_segments(sci
);
2411 nilfs_segctor_clear_metadata_dirty(sci
);
2414 nilfs_segctor_end_construction(sci
, nilfs
, 0);
2416 } while (sci
->sc_stage
.scnt
!= NILFS_ST_DONE
);
2418 /* Clearing sketch data */
2419 if (has_sr
&& sci
->sc_sketch_inode
) {
2420 if (i_size_read(sci
->sc_sketch_inode
) == 0)
2421 clear_bit(NILFS_I_DIRTY
,
2422 &NILFS_I(sci
->sc_sketch_inode
)->i_state
);
2423 i_size_write(sci
->sc_sketch_inode
, 0);
2426 nilfs_segctor_destroy_segment_buffers(sci
);
2427 nilfs_segctor_check_out_files(sci
, sbi
);
2431 nilfs_segctor_abort_write(sci
, failed_page
, err
);
2432 nilfs_segctor_cancel_segusage(sci
, nilfs
->ns_sufile
);
2435 if (sci
->sc_stage
.flags
& NILFS_CF_IFILE_STARTED
)
2436 nilfs_redirty_inodes(&sci
->sc_dirty_files
);
2438 nilfs_segctor_reactivate_segments(sci
, nilfs
);
2441 if (nilfs_doing_gc())
2442 nilfs_redirty_inodes(&sci
->sc_gc_inodes
);
2443 nilfs_segctor_end_construction(sci
, nilfs
, err
);
2448 * nilfs_secgtor_start_timer - set timer of background write
2449 * @sci: nilfs_sc_info
2451 * If the timer has already been set, it ignores the new request.
2452 * This function MUST be called within a section locking the segment
2455 static void nilfs_segctor_start_timer(struct nilfs_sc_info
*sci
)
2457 spin_lock(&sci
->sc_state_lock
);
2458 if (sci
->sc_timer
&& !(sci
->sc_state
& NILFS_SEGCTOR_COMMIT
)) {
2459 sci
->sc_timer
->expires
= jiffies
+ sci
->sc_interval
;
2460 add_timer(sci
->sc_timer
);
2461 sci
->sc_state
|= NILFS_SEGCTOR_COMMIT
;
2463 spin_unlock(&sci
->sc_state_lock
);
2466 static void nilfs_segctor_do_flush(struct nilfs_sc_info
*sci
, int bn
)
2468 spin_lock(&sci
->sc_state_lock
);
2469 if (!(sci
->sc_flush_request
& (1 << bn
))) {
2470 unsigned long prev_req
= sci
->sc_flush_request
;
2472 sci
->sc_flush_request
|= (1 << bn
);
2474 wake_up(&sci
->sc_wait_daemon
);
2476 spin_unlock(&sci
->sc_state_lock
);
2480 * nilfs_flush_segment - trigger a segment construction for resource control
2482 * @ino: inode number of the file to be flushed out.
2484 void nilfs_flush_segment(struct super_block
*sb
, ino_t ino
)
2486 struct nilfs_sb_info
*sbi
= NILFS_SB(sb
);
2487 struct nilfs_sc_info
*sci
= NILFS_SC(sbi
);
2489 if (!sci
|| nilfs_doing_construction())
2491 nilfs_segctor_do_flush(sci
, NILFS_MDT_INODE(sb
, ino
) ? ino
: 0);
2492 /* assign bit 0 to data files */
2495 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info
*sci
,
2496 __u64
*segnum
, size_t nsegs
)
2498 struct nilfs_segment_entry
*ent
;
2499 struct the_nilfs
*nilfs
= sci
->sc_sbi
->s_nilfs
;
2500 struct inode
*sufile
= nilfs
->ns_sufile
;
2503 const char *flag_name
;
2507 for (pnum
= segnum
, i
= 0; i
< nsegs
; pnum
++, i
++) {
2508 ent
= nilfs_alloc_segment_entry(*pnum
);
2509 if (unlikely(!ent
)) {
2513 list_add_tail(&ent
->list
, &list
);
2515 err
= nilfs_open_segment_entry(ent
, sufile
);
2519 if (unlikely(le32_to_cpu(ent
->raw_su
->su_flags
) !=
2520 (1UL << NILFS_SEGMENT_USAGE_DIRTY
))) {
2521 if (nilfs_segment_usage_clean(ent
->raw_su
))
2522 flag_name
= "clean";
2523 else if (nilfs_segment_usage_active(ent
->raw_su
))
2524 flag_name
= "active";
2525 else if (nilfs_segment_usage_volatile_active(
2527 flag_name
= "volatile active";
2528 else if (!nilfs_segment_usage_dirty(ent
->raw_su
))
2529 flag_name
= "non-dirty";
2531 flag_name
= "erroneous";
2534 "NILFS: %s segment is requested to be cleaned "
2536 flag_name
, (unsigned long long)ent
->segnum
);
2539 nilfs_close_segment_entry(ent
, sufile
);
2541 if (unlikely(err2
)) {
2545 list_splice(&list
, sci
->sc_cleaning_segments
.prev
);
2549 nilfs_dispose_segment_list(&list
);
2553 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info
*sci
)
2555 nilfs_dispose_segment_list(&sci
->sc_cleaning_segments
);
2558 struct nilfs_segctor_wait_request
{
2565 static int nilfs_segctor_sync(struct nilfs_sc_info
*sci
)
2567 struct nilfs_segctor_wait_request wait_req
;
2570 spin_lock(&sci
->sc_state_lock
);
2571 init_wait(&wait_req
.wq
);
2573 atomic_set(&wait_req
.done
, 0);
2574 wait_req
.seq
= ++sci
->sc_seq_request
;
2575 spin_unlock(&sci
->sc_state_lock
);
2577 init_waitqueue_entry(&wait_req
.wq
, current
);
2578 add_wait_queue(&sci
->sc_wait_request
, &wait_req
.wq
);
2579 set_current_state(TASK_INTERRUPTIBLE
);
2580 wake_up(&sci
->sc_wait_daemon
);
2583 if (atomic_read(&wait_req
.done
)) {
2587 if (!signal_pending(current
)) {
2594 finish_wait(&sci
->sc_wait_request
, &wait_req
.wq
);
2598 static void nilfs_segctor_wakeup(struct nilfs_sc_info
*sci
, int err
)
2600 struct nilfs_segctor_wait_request
*wrq
, *n
;
2601 unsigned long flags
;
2603 spin_lock_irqsave(&sci
->sc_wait_request
.lock
, flags
);
2604 list_for_each_entry_safe(wrq
, n
, &sci
->sc_wait_request
.task_list
,
2606 if (!atomic_read(&wrq
->done
) &&
2607 nilfs_cnt32_ge(sci
->sc_seq_done
, wrq
->seq
)) {
2609 atomic_set(&wrq
->done
, 1);
2611 if (atomic_read(&wrq
->done
)) {
2612 wrq
->wq
.func(&wrq
->wq
,
2613 TASK_UNINTERRUPTIBLE
| TASK_INTERRUPTIBLE
,
2617 spin_unlock_irqrestore(&sci
->sc_wait_request
.lock
, flags
);
2621 * nilfs_construct_segment - construct a logical segment
2624 * Return Value: On success, 0 is retured. On errors, one of the following
2625 * negative error code is returned.
2627 * %-EROFS - Read only filesystem.
2631 * %-ENOSPC - No space left on device (only in a panic state).
2633 * %-ERESTARTSYS - Interrupted.
2635 * %-ENOMEM - Insufficient memory available.
2637 int nilfs_construct_segment(struct super_block
*sb
)
2639 struct nilfs_sb_info
*sbi
= NILFS_SB(sb
);
2640 struct nilfs_sc_info
*sci
= NILFS_SC(sbi
);
2641 struct nilfs_transaction_info
*ti
;
2647 /* A call inside transactions causes a deadlock. */
2648 BUG_ON((ti
= current
->journal_info
) && ti
->ti_magic
== NILFS_TI_MAGIC
);
2650 err
= nilfs_segctor_sync(sci
);
2655 * nilfs_construct_dsync_segment - construct a data-only logical segment
2657 * @inode: inode whose data blocks should be written out
2658 * @start: start byte offset
2659 * @end: end byte offset (inclusive)
2661 * Return Value: On success, 0 is retured. On errors, one of the following
2662 * negative error code is returned.
2664 * %-EROFS - Read only filesystem.
2668 * %-ENOSPC - No space left on device (only in a panic state).
2670 * %-ERESTARTSYS - Interrupted.
2672 * %-ENOMEM - Insufficient memory available.
2674 int nilfs_construct_dsync_segment(struct super_block
*sb
, struct inode
*inode
,
2675 loff_t start
, loff_t end
)
2677 struct nilfs_sb_info
*sbi
= NILFS_SB(sb
);
2678 struct nilfs_sc_info
*sci
= NILFS_SC(sbi
);
2679 struct nilfs_inode_info
*ii
;
2680 struct nilfs_transaction_info ti
;
2686 nilfs_transaction_lock(sbi
, &ti
, 0);
2688 ii
= NILFS_I(inode
);
2689 if (test_bit(NILFS_I_INODE_DIRTY
, &ii
->i_state
) ||
2690 nilfs_test_opt(sbi
, STRICT_ORDER
) ||
2691 test_bit(NILFS_SC_UNCLOSED
, &sci
->sc_flags
) ||
2692 nilfs_discontinued(sbi
->s_nilfs
)) {
2693 nilfs_transaction_unlock(sbi
);
2694 err
= nilfs_segctor_sync(sci
);
2698 spin_lock(&sbi
->s_inode_lock
);
2699 if (!test_bit(NILFS_I_QUEUED
, &ii
->i_state
) &&
2700 !test_bit(NILFS_I_BUSY
, &ii
->i_state
)) {
2701 spin_unlock(&sbi
->s_inode_lock
);
2702 nilfs_transaction_unlock(sbi
);
2705 spin_unlock(&sbi
->s_inode_lock
);
2706 sci
->sc_dsync_inode
= ii
;
2707 sci
->sc_dsync_start
= start
;
2708 sci
->sc_dsync_end
= end
;
2710 err
= nilfs_segctor_do_construct(sci
, SC_LSEG_DSYNC
);
2712 nilfs_transaction_unlock(sbi
);
2716 struct nilfs_segctor_req
{
2719 int sc_err
; /* construction failure */
2720 int sb_err
; /* super block writeback failure */
2723 #define FLUSH_FILE_BIT (0x1) /* data file only */
2724 #define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2726 static void nilfs_segctor_accept(struct nilfs_sc_info
*sci
,
2727 struct nilfs_segctor_req
*req
)
2731 req
->sc_err
= req
->sb_err
= 0;
2732 spin_lock(&sci
->sc_state_lock
);
2733 req
->seq_accepted
= sci
->sc_seq_request
;
2734 spin_unlock(&sci
->sc_state_lock
);
2737 del_timer_sync(sci
->sc_timer
);
2740 static void nilfs_segctor_notify(struct nilfs_sc_info
*sci
,
2741 struct nilfs_segctor_req
*req
)
2743 /* Clear requests (even when the construction failed) */
2744 spin_lock(&sci
->sc_state_lock
);
2746 sci
->sc_state
&= ~NILFS_SEGCTOR_COMMIT
;
2748 if (req
->mode
== SC_LSEG_SR
) {
2749 sci
->sc_seq_done
= req
->seq_accepted
;
2750 nilfs_segctor_wakeup(sci
, req
->sc_err
? : req
->sb_err
);
2751 sci
->sc_flush_request
= 0;
2752 } else if (req
->mode
== SC_FLUSH_FILE
)
2753 sci
->sc_flush_request
&= ~FLUSH_FILE_BIT
;
2754 else if (req
->mode
== SC_FLUSH_DAT
)
2755 sci
->sc_flush_request
&= ~FLUSH_DAT_BIT
;
2757 spin_unlock(&sci
->sc_state_lock
);
2760 static int nilfs_segctor_construct(struct nilfs_sc_info
*sci
,
2761 struct nilfs_segctor_req
*req
)
2763 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
2764 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
2767 if (nilfs_discontinued(nilfs
))
2768 req
->mode
= SC_LSEG_SR
;
2769 if (!nilfs_segctor_confirm(sci
)) {
2770 err
= nilfs_segctor_do_construct(sci
, req
->mode
);
2774 if (req
->mode
!= SC_FLUSH_DAT
)
2775 atomic_set(&nilfs
->ns_ndirtyblks
, 0);
2776 if (test_bit(NILFS_SC_SUPER_ROOT
, &sci
->sc_flags
) &&
2777 nilfs_discontinued(nilfs
)) {
2778 down_write(&nilfs
->ns_sem
);
2779 req
->sb_err
= nilfs_commit_super(sbi
);
2780 up_write(&nilfs
->ns_sem
);
2786 static void nilfs_construction_timeout(unsigned long data
)
2788 struct task_struct
*p
= (struct task_struct
*)data
;
2793 nilfs_remove_written_gcinodes(struct the_nilfs
*nilfs
, struct list_head
*head
)
2795 struct nilfs_inode_info
*ii
, *n
;
2797 list_for_each_entry_safe(ii
, n
, head
, i_dirty
) {
2798 if (!test_bit(NILFS_I_UPDATED
, &ii
->i_state
))
2800 hlist_del_init(&ii
->vfs_inode
.i_hash
);
2801 list_del_init(&ii
->i_dirty
);
2802 nilfs_clear_gcinode(&ii
->vfs_inode
);
2806 int nilfs_clean_segments(struct super_block
*sb
, void __user
*argp
)
2808 struct nilfs_sb_info
*sbi
= NILFS_SB(sb
);
2809 struct nilfs_sc_info
*sci
= NILFS_SC(sbi
);
2810 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
2811 struct nilfs_transaction_info ti
;
2812 struct nilfs_segctor_req req
= { .mode
= SC_LSEG_SR
};
2818 nilfs_transaction_lock(sbi
, &ti
, 1);
2820 err
= nilfs_init_gcdat_inode(nilfs
);
2823 err
= nilfs_ioctl_prepare_clean_segments(nilfs
, argp
);
2827 list_splice_init(&nilfs
->ns_gc_inodes
, sci
->sc_gc_inodes
.prev
);
2830 nilfs_segctor_accept(sci
, &req
);
2831 err
= nilfs_segctor_construct(sci
, &req
);
2832 nilfs_remove_written_gcinodes(nilfs
, &sci
->sc_gc_inodes
);
2833 nilfs_segctor_notify(sci
, &req
);
2838 nilfs_warning(sb
, __func__
,
2839 "segment construction failed. (err=%d)", err
);
2840 set_current_state(TASK_INTERRUPTIBLE
);
2841 schedule_timeout(sci
->sc_interval
);
2845 nilfs_clear_gcdat_inode(nilfs
);
2846 nilfs_transaction_unlock(sbi
);
2850 static void nilfs_segctor_thread_construct(struct nilfs_sc_info
*sci
, int mode
)
2852 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
2853 struct nilfs_transaction_info ti
;
2854 struct nilfs_segctor_req req
= { .mode
= mode
};
2856 nilfs_transaction_lock(sbi
, &ti
, 0);
2858 nilfs_segctor_accept(sci
, &req
);
2859 nilfs_segctor_construct(sci
, &req
);
2860 nilfs_segctor_notify(sci
, &req
);
2863 * Unclosed segment should be retried. We do this using sc_timer.
2864 * Timeout of sc_timer will invoke complete construction which leads
2865 * to close the current logical segment.
2867 if (test_bit(NILFS_SC_UNCLOSED
, &sci
->sc_flags
))
2868 nilfs_segctor_start_timer(sci
);
2870 nilfs_transaction_unlock(sbi
);
2873 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info
*sci
)
2878 spin_lock(&sci
->sc_state_lock
);
2879 mode
= (sci
->sc_flush_request
& FLUSH_DAT_BIT
) ?
2880 SC_FLUSH_DAT
: SC_FLUSH_FILE
;
2881 spin_unlock(&sci
->sc_state_lock
);
2884 err
= nilfs_segctor_do_construct(sci
, mode
);
2886 spin_lock(&sci
->sc_state_lock
);
2887 sci
->sc_flush_request
&= (mode
== SC_FLUSH_FILE
) ?
2888 ~FLUSH_FILE_BIT
: ~FLUSH_DAT_BIT
;
2889 spin_unlock(&sci
->sc_state_lock
);
2891 clear_bit(NILFS_SC_PRIOR_FLUSH
, &sci
->sc_flags
);
2894 static int nilfs_segctor_flush_mode(struct nilfs_sc_info
*sci
)
2896 if (!test_bit(NILFS_SC_UNCLOSED
, &sci
->sc_flags
) ||
2897 time_before(jiffies
, sci
->sc_lseg_stime
+ sci
->sc_mjcp_freq
)) {
2898 if (!(sci
->sc_flush_request
& ~FLUSH_FILE_BIT
))
2899 return SC_FLUSH_FILE
;
2900 else if (!(sci
->sc_flush_request
& ~FLUSH_DAT_BIT
))
2901 return SC_FLUSH_DAT
;
2907 * nilfs_segctor_thread - main loop of the segment constructor thread.
2908 * @arg: pointer to a struct nilfs_sc_info.
2910 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2911 * to execute segment constructions.
2913 static int nilfs_segctor_thread(void *arg
)
2915 struct nilfs_sc_info
*sci
= (struct nilfs_sc_info
*)arg
;
2916 struct timer_list timer
;
2920 timer
.data
= (unsigned long)current
;
2921 timer
.function
= nilfs_construction_timeout
;
2922 sci
->sc_timer
= &timer
;
2925 sci
->sc_task
= current
;
2926 wake_up(&sci
->sc_wait_task
); /* for nilfs_segctor_start_thread() */
2928 "segctord starting. Construction interval = %lu seconds, "
2929 "CP frequency < %lu seconds\n",
2930 sci
->sc_interval
/ HZ
, sci
->sc_mjcp_freq
/ HZ
);
2932 spin_lock(&sci
->sc_state_lock
);
2937 if (sci
->sc_state
& NILFS_SEGCTOR_QUIT
)
2940 if (timeout
|| sci
->sc_seq_request
!= sci
->sc_seq_done
)
2942 else if (!sci
->sc_flush_request
)
2945 mode
= nilfs_segctor_flush_mode(sci
);
2947 spin_unlock(&sci
->sc_state_lock
);
2948 nilfs_segctor_thread_construct(sci
, mode
);
2949 spin_lock(&sci
->sc_state_lock
);
2954 if (freezing(current
)) {
2955 spin_unlock(&sci
->sc_state_lock
);
2957 spin_lock(&sci
->sc_state_lock
);
2960 int should_sleep
= 1;
2962 prepare_to_wait(&sci
->sc_wait_daemon
, &wait
,
2963 TASK_INTERRUPTIBLE
);
2965 if (sci
->sc_seq_request
!= sci
->sc_seq_done
)
2967 else if (sci
->sc_flush_request
)
2969 else if (sci
->sc_state
& NILFS_SEGCTOR_COMMIT
)
2970 should_sleep
= time_before(jiffies
,
2971 sci
->sc_timer
->expires
);
2974 spin_unlock(&sci
->sc_state_lock
);
2976 spin_lock(&sci
->sc_state_lock
);
2978 finish_wait(&sci
->sc_wait_daemon
, &wait
);
2979 timeout
= ((sci
->sc_state
& NILFS_SEGCTOR_COMMIT
) &&
2980 time_after_eq(jiffies
, sci
->sc_timer
->expires
));
2985 spin_unlock(&sci
->sc_state_lock
);
2986 del_timer_sync(sci
->sc_timer
);
2987 sci
->sc_timer
= NULL
;
2990 sci
->sc_task
= NULL
;
2991 wake_up(&sci
->sc_wait_task
); /* for nilfs_segctor_kill_thread() */
2995 static int nilfs_segctor_start_thread(struct nilfs_sc_info
*sci
)
2997 struct task_struct
*t
;
2999 t
= kthread_run(nilfs_segctor_thread
, sci
, "segctord");
3001 int err
= PTR_ERR(t
);
3003 printk(KERN_ERR
"NILFS: error %d creating segctord thread\n",
3007 wait_event(sci
->sc_wait_task
, sci
->sc_task
!= NULL
);
3011 static void nilfs_segctor_kill_thread(struct nilfs_sc_info
*sci
)
3013 sci
->sc_state
|= NILFS_SEGCTOR_QUIT
;
3015 while (sci
->sc_task
) {
3016 wake_up(&sci
->sc_wait_daemon
);
3017 spin_unlock(&sci
->sc_state_lock
);
3018 wait_event(sci
->sc_wait_task
, sci
->sc_task
== NULL
);
3019 spin_lock(&sci
->sc_state_lock
);
3023 static int nilfs_segctor_init(struct nilfs_sc_info
*sci
,
3024 struct nilfs_recovery_info
*ri
)
3027 struct inode
*inode
= nilfs_iget(sci
->sc_super
, NILFS_SKETCH_INO
);
3029 sci
->sc_sketch_inode
= IS_ERR(inode
) ? NULL
: inode
;
3030 if (sci
->sc_sketch_inode
)
3031 i_size_write(sci
->sc_sketch_inode
, 0);
3033 sci
->sc_seq_done
= sci
->sc_seq_request
;
3035 list_splice_init(&ri
->ri_used_segments
,
3036 sci
->sc_active_segments
.prev
);
3038 err
= nilfs_segctor_start_thread(sci
);
3041 list_splice_init(&sci
->sc_active_segments
,
3042 ri
->ri_used_segments
.prev
);
3043 if (sci
->sc_sketch_inode
) {
3044 iput(sci
->sc_sketch_inode
);
3045 sci
->sc_sketch_inode
= NULL
;
3052 * Setup & clean-up functions
3054 static struct nilfs_sc_info
*nilfs_segctor_new(struct nilfs_sb_info
*sbi
)
3056 struct nilfs_sc_info
*sci
;
3058 sci
= kzalloc(sizeof(*sci
), GFP_KERNEL
);
3063 sci
->sc_super
= sbi
->s_super
;
3065 init_waitqueue_head(&sci
->sc_wait_request
);
3066 init_waitqueue_head(&sci
->sc_wait_daemon
);
3067 init_waitqueue_head(&sci
->sc_wait_task
);
3068 spin_lock_init(&sci
->sc_state_lock
);
3069 INIT_LIST_HEAD(&sci
->sc_dirty_files
);
3070 INIT_LIST_HEAD(&sci
->sc_segbufs
);
3071 INIT_LIST_HEAD(&sci
->sc_gc_inodes
);
3072 INIT_LIST_HEAD(&sci
->sc_active_segments
);
3073 INIT_LIST_HEAD(&sci
->sc_cleaning_segments
);
3074 INIT_LIST_HEAD(&sci
->sc_copied_buffers
);
3076 sci
->sc_interval
= HZ
* NILFS_SC_DEFAULT_TIMEOUT
;
3077 sci
->sc_mjcp_freq
= HZ
* NILFS_SC_DEFAULT_SR_FREQ
;
3078 sci
->sc_watermark
= NILFS_SC_DEFAULT_WATERMARK
;
3080 if (sbi
->s_interval
)
3081 sci
->sc_interval
= sbi
->s_interval
;
3082 if (sbi
->s_watermark
)
3083 sci
->sc_watermark
= sbi
->s_watermark
;
3087 static void nilfs_segctor_write_out(struct nilfs_sc_info
*sci
)
3089 int ret
, retrycount
= NILFS_SC_CLEANUP_RETRY
;
3091 /* The segctord thread was stopped and its timer was removed.
3092 But some tasks remain. */
3094 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
3095 struct nilfs_transaction_info ti
;
3096 struct nilfs_segctor_req req
= { .mode
= SC_LSEG_SR
};
3098 nilfs_transaction_lock(sbi
, &ti
, 0);
3099 nilfs_segctor_accept(sci
, &req
);
3100 ret
= nilfs_segctor_construct(sci
, &req
);
3101 nilfs_segctor_notify(sci
, &req
);
3102 nilfs_transaction_unlock(sbi
);
3104 } while (ret
&& retrycount
-- > 0);
3108 * nilfs_segctor_destroy - destroy the segment constructor.
3109 * @sci: nilfs_sc_info
3111 * nilfs_segctor_destroy() kills the segctord thread and frees
3112 * the nilfs_sc_info struct.
3113 * Caller must hold the segment semaphore.
3115 static void nilfs_segctor_destroy(struct nilfs_sc_info
*sci
)
3117 struct nilfs_sb_info
*sbi
= sci
->sc_sbi
;
3120 up_write(&sbi
->s_nilfs
->ns_segctor_sem
);
3122 spin_lock(&sci
->sc_state_lock
);
3123 nilfs_segctor_kill_thread(sci
);
3124 flag
= ((sci
->sc_state
& NILFS_SEGCTOR_COMMIT
) || sci
->sc_flush_request
3125 || sci
->sc_seq_request
!= sci
->sc_seq_done
);
3126 spin_unlock(&sci
->sc_state_lock
);
3128 if (flag
|| nilfs_segctor_confirm(sci
))
3129 nilfs_segctor_write_out(sci
);
3131 BUG_ON(!list_empty(&sci
->sc_copied_buffers
));
3133 if (!list_empty(&sci
->sc_dirty_files
)) {
3134 nilfs_warning(sbi
->s_super
, __func__
,
3135 "dirty file(s) after the final construction\n");
3136 nilfs_dispose_list(sbi
, &sci
->sc_dirty_files
, 1);
3138 if (!list_empty(&sci
->sc_active_segments
))
3139 nilfs_dispose_segment_list(&sci
->sc_active_segments
);
3141 if (!list_empty(&sci
->sc_cleaning_segments
))
3142 nilfs_dispose_segment_list(&sci
->sc_cleaning_segments
);
3144 BUG_ON(!list_empty(&sci
->sc_segbufs
));
3146 if (sci
->sc_sketch_inode
) {
3147 iput(sci
->sc_sketch_inode
);
3148 sci
->sc_sketch_inode
= NULL
;
3150 down_write(&sbi
->s_nilfs
->ns_segctor_sem
);
3156 * nilfs_attach_segment_constructor - attach a segment constructor
3157 * @sbi: nilfs_sb_info
3158 * @ri: nilfs_recovery_info
3160 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
3161 * initilizes it, and starts the segment constructor.
3163 * Return Value: On success, 0 is returned. On error, one of the following
3164 * negative error code is returned.
3166 * %-ENOMEM - Insufficient memory available.
3168 int nilfs_attach_segment_constructor(struct nilfs_sb_info
*sbi
,
3169 struct nilfs_recovery_info
*ri
)
3171 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
3174 /* Each field of nilfs_segctor is cleared through the initialization
3175 of super-block info */
3176 sbi
->s_sc_info
= nilfs_segctor_new(sbi
);
3177 if (!sbi
->s_sc_info
)
3180 nilfs_attach_writer(nilfs
, sbi
);
3181 err
= nilfs_segctor_init(NILFS_SC(sbi
), ri
);
3183 nilfs_detach_writer(nilfs
, sbi
);
3184 kfree(sbi
->s_sc_info
);
3185 sbi
->s_sc_info
= NULL
;
3191 * nilfs_detach_segment_constructor - destroy the segment constructor
3192 * @sbi: nilfs_sb_info
3194 * nilfs_detach_segment_constructor() kills the segment constructor daemon,
3195 * frees the struct nilfs_sc_info, and destroy the dirty file list.
3197 void nilfs_detach_segment_constructor(struct nilfs_sb_info
*sbi
)
3199 struct the_nilfs
*nilfs
= sbi
->s_nilfs
;
3200 LIST_HEAD(garbage_list
);
3202 down_write(&nilfs
->ns_segctor_sem
);
3203 if (NILFS_SC(sbi
)) {
3204 nilfs_segctor_destroy(NILFS_SC(sbi
));
3205 sbi
->s_sc_info
= NULL
;
3208 /* Force to free the list of dirty files */
3209 spin_lock(&sbi
->s_inode_lock
);
3210 if (!list_empty(&sbi
->s_dirty_files
)) {
3211 list_splice_init(&sbi
->s_dirty_files
, &garbage_list
);
3212 nilfs_warning(sbi
->s_super
, __func__
,
3213 "Non empty dirty list after the last "
3214 "segment construction\n");
3216 spin_unlock(&sbi
->s_inode_lock
);
3217 up_write(&nilfs
->ns_segctor_sem
);
3219 nilfs_dispose_list(sbi
, &garbage_list
, 1);
3220 nilfs_detach_writer(nilfs
, sbi
);