1 // SPDX-License-Identifier: GPL-2.0+
3 * segment.c - NILFS segment constructor.
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
11 #include <linux/pagemap.h>
12 #include <linux/buffer_head.h>
13 #include <linux/writeback.h>
14 #include <linux/bitops.h>
15 #include <linux/bio.h>
16 #include <linux/completion.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/freezer.h>
20 #include <linux/kthread.h>
21 #include <linux/crc32.h>
22 #include <linux/pagevec.h>
23 #include <linux/slab.h>
24 #include <linux/sched/signal.h>
39 #define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
41 #define SC_MAX_SEGDELTA 64 /*
42 * Upper limit of the number of segments
43 * appended in collection retry loop
46 /* Construction mode */
48 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
50 * Flush data blocks of a given file and make
51 * a logical segment without a super root.
54 * Flush data files, leads to segment writes without
55 * creating a checkpoint.
58 * Flush DAT file. This also creates segments
59 * without a checkpoint.
63 /* Stage numbers of dirty block collection */
66 NILFS_ST_GC, /* Collecting dirty blocks for GC */
72 NILFS_ST_SR, /* Super root */
73 NILFS_ST_DSYNC, /* Data sync blocks */
77 #define CREATE_TRACE_POINTS
78 #include <trace/events/nilfs2.h>
81 * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
82 * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
83 * the variable must use them because transition of stage count must involve
84 * trace events (trace_nilfs2_collection_stage_transition).
86 * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
87 * produce tracepoint events. It is provided just for making the intention
90 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
93 trace_nilfs2_collection_stage_transition(sci);
96 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
98 sci->sc_stage.scnt = next_scnt;
99 trace_nilfs2_collection_stage_transition(sci);
102 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
104 return sci->sc_stage.scnt;
107 /* State flags of collection */
108 #define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
109 #define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
110 #define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
111 #define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
113 /* Operations depending on the construction mode and file type */
114 struct nilfs_sc_operations {
115 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
117 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
119 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
121 void (*write_data_binfo)(struct nilfs_sc_info *,
122 struct nilfs_segsum_pointer *,
123 union nilfs_binfo *);
124 void (*write_node_binfo)(struct nilfs_sc_info *,
125 struct nilfs_segsum_pointer *,
126 union nilfs_binfo *);
132 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
133 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
134 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
135 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
137 #define nilfs_cnt32_gt(a, b) \
138 (typecheck(__u32, a) && typecheck(__u32, b) && \
139 ((__s32)(b) - (__s32)(a) < 0))
140 #define nilfs_cnt32_ge(a, b) \
141 (typecheck(__u32, a) && typecheck(__u32, b) && \
142 ((__s32)(a) - (__s32)(b) >= 0))
143 #define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
144 #define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
146 static int nilfs_prepare_segment_lock(struct super_block *sb,
147 struct nilfs_transaction_info *ti)
149 struct nilfs_transaction_info *cur_ti = current->journal_info;
153 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
154 return ++cur_ti->ti_count;
157 * If journal_info field is occupied by other FS,
158 * it is saved and will be restored on
159 * nilfs_transaction_commit().
161 nilfs_warn(sb, "journal info from a different FS");
162 save = current->journal_info;
165 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
168 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
174 ti->ti_magic = NILFS_TI_MAGIC;
175 current->journal_info = ti;
180 * nilfs_transaction_begin - start indivisible file operations.
182 * @ti: nilfs_transaction_info
183 * @vacancy_check: flags for vacancy rate checks
185 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
186 * the segment semaphore, to make a segment construction and write tasks
187 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
188 * The region enclosed by these two functions can be nested. To avoid a
189 * deadlock, the semaphore is only acquired or released in the outermost call.
191 * This function allocates a nilfs_transaction_info struct to keep context
192 * information on it. It is initialized and hooked onto the current task in
193 * the outermost call. If a pre-allocated struct is given to @ti, it is used
194 * instead; otherwise a new struct is assigned from a slab.
196 * When @vacancy_check flag is set, this function will check the amount of
197 * free space, and will wait for the GC to reclaim disk space if low capacity.
199 * Return Value: On success, 0 is returned. On error, one of the following
200 * negative error code is returned.
202 * %-ENOMEM - Insufficient memory available.
204 * %-ENOSPC - No space left on device
206 int nilfs_transaction_begin(struct super_block *sb,
207 struct nilfs_transaction_info *ti,
210 struct the_nilfs *nilfs;
211 int ret = nilfs_prepare_segment_lock(sb, ti);
212 struct nilfs_transaction_info *trace_ti;
214 if (unlikely(ret < 0))
217 trace_ti = current->journal_info;
219 trace_nilfs2_transaction_transition(sb, trace_ti,
220 trace_ti->ti_count, trace_ti->ti_flags,
221 TRACE_NILFS2_TRANSACTION_BEGIN);
225 sb_start_intwrite(sb);
227 nilfs = sb->s_fs_info;
228 down_read(&nilfs->ns_segctor_sem);
229 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
230 up_read(&nilfs->ns_segctor_sem);
235 trace_ti = current->journal_info;
236 trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
238 TRACE_NILFS2_TRANSACTION_BEGIN);
242 ti = current->journal_info;
243 current->journal_info = ti->ti_save;
244 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
245 kmem_cache_free(nilfs_transaction_cachep, ti);
251 * nilfs_transaction_commit - commit indivisible file operations.
254 * nilfs_transaction_commit() releases the read semaphore which is
255 * acquired by nilfs_transaction_begin(). This is only performed
256 * in outermost call of this function. If a commit flag is set,
257 * nilfs_transaction_commit() sets a timer to start the segment
258 * constructor. If a sync flag is set, it starts construction
261 int nilfs_transaction_commit(struct super_block *sb)
263 struct nilfs_transaction_info *ti = current->journal_info;
264 struct the_nilfs *nilfs = sb->s_fs_info;
267 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
268 ti->ti_flags |= NILFS_TI_COMMIT;
269 if (ti->ti_count > 0) {
271 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
272 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
275 if (nilfs->ns_writer) {
276 struct nilfs_sc_info *sci = nilfs->ns_writer;
278 if (ti->ti_flags & NILFS_TI_COMMIT)
279 nilfs_segctor_start_timer(sci);
280 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
281 nilfs_segctor_do_flush(sci, 0);
283 up_read(&nilfs->ns_segctor_sem);
284 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
285 ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
287 current->journal_info = ti->ti_save;
289 if (ti->ti_flags & NILFS_TI_SYNC)
290 err = nilfs_construct_segment(sb);
291 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
292 kmem_cache_free(nilfs_transaction_cachep, ti);
297 void nilfs_transaction_abort(struct super_block *sb)
299 struct nilfs_transaction_info *ti = current->journal_info;
300 struct the_nilfs *nilfs = sb->s_fs_info;
302 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
303 if (ti->ti_count > 0) {
305 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
306 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
309 up_read(&nilfs->ns_segctor_sem);
311 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
312 ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
314 current->journal_info = ti->ti_save;
315 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
316 kmem_cache_free(nilfs_transaction_cachep, ti);
320 void nilfs_relax_pressure_in_lock(struct super_block *sb)
322 struct the_nilfs *nilfs = sb->s_fs_info;
323 struct nilfs_sc_info *sci = nilfs->ns_writer;
325 if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request)
328 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
329 up_read(&nilfs->ns_segctor_sem);
331 down_write(&nilfs->ns_segctor_sem);
332 if (sci->sc_flush_request &&
333 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
334 struct nilfs_transaction_info *ti = current->journal_info;
336 ti->ti_flags |= NILFS_TI_WRITER;
337 nilfs_segctor_do_immediate_flush(sci);
338 ti->ti_flags &= ~NILFS_TI_WRITER;
340 downgrade_write(&nilfs->ns_segctor_sem);
343 static void nilfs_transaction_lock(struct super_block *sb,
344 struct nilfs_transaction_info *ti,
347 struct nilfs_transaction_info *cur_ti = current->journal_info;
348 struct the_nilfs *nilfs = sb->s_fs_info;
349 struct nilfs_sc_info *sci = nilfs->ns_writer;
352 ti->ti_flags = NILFS_TI_WRITER;
354 ti->ti_save = cur_ti;
355 ti->ti_magic = NILFS_TI_MAGIC;
356 current->journal_info = ti;
359 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
360 ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
362 down_write(&nilfs->ns_segctor_sem);
363 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
366 nilfs_segctor_do_immediate_flush(sci);
368 up_write(&nilfs->ns_segctor_sem);
372 ti->ti_flags |= NILFS_TI_GC;
374 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
375 ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
378 static void nilfs_transaction_unlock(struct super_block *sb)
380 struct nilfs_transaction_info *ti = current->journal_info;
381 struct the_nilfs *nilfs = sb->s_fs_info;
383 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
384 BUG_ON(ti->ti_count > 0);
386 up_write(&nilfs->ns_segctor_sem);
387 current->journal_info = ti->ti_save;
389 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
390 ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
393 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
394 struct nilfs_segsum_pointer *ssp,
397 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
398 unsigned int blocksize = sci->sc_super->s_blocksize;
401 if (unlikely(ssp->offset + bytes > blocksize)) {
403 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
404 &segbuf->sb_segsum_buffers));
405 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
407 p = ssp->bh->b_data + ssp->offset;
408 ssp->offset += bytes;
413 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
414 * @sci: nilfs_sc_info
416 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
418 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
419 struct buffer_head *sumbh;
420 unsigned int sumbytes;
421 unsigned int flags = 0;
424 if (nilfs_doing_gc())
426 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
430 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
431 sumbytes = segbuf->sb_sum.sumbytes;
432 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
433 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
434 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
439 * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
440 * @sci: segment constructor object
442 * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
443 * the current segment summary block.
445 static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
447 struct nilfs_segsum_pointer *ssp;
449 ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
450 if (ssp->offset < ssp->bh->b_size)
451 memset(ssp->bh->b_data + ssp->offset, 0,
452 ssp->bh->b_size - ssp->offset);
455 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
457 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
458 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
460 * The current segment is filled up
463 nilfs_segctor_zeropad_segsum(sci);
464 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
465 return nilfs_segctor_reset_segment_buffer(sci);
468 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
470 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
473 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
474 err = nilfs_segctor_feed_segment(sci);
477 segbuf = sci->sc_curseg;
479 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
481 segbuf->sb_sum.flags |= NILFS_SS_SR;
486 * Functions for making segment summary and payloads
488 static int nilfs_segctor_segsum_block_required(
489 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
490 unsigned int binfo_size)
492 unsigned int blocksize = sci->sc_super->s_blocksize;
493 /* Size of finfo and binfo is enough small against blocksize */
495 return ssp->offset + binfo_size +
496 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
500 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
503 sci->sc_curseg->sb_sum.nfinfo++;
504 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
505 nilfs_segctor_map_segsum_entry(
506 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
508 if (NILFS_I(inode)->i_root &&
509 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
510 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
514 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
517 struct nilfs_finfo *finfo;
518 struct nilfs_inode_info *ii;
519 struct nilfs_segment_buffer *segbuf;
522 if (sci->sc_blk_cnt == 0)
527 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
529 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
534 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
536 finfo->fi_ino = cpu_to_le64(inode->i_ino);
537 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
538 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
539 finfo->fi_cno = cpu_to_le64(cno);
541 segbuf = sci->sc_curseg;
542 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
543 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
544 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
545 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
548 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
549 struct buffer_head *bh,
551 unsigned int binfo_size)
553 struct nilfs_segment_buffer *segbuf;
554 int required, err = 0;
557 segbuf = sci->sc_curseg;
558 required = nilfs_segctor_segsum_block_required(
559 sci, &sci->sc_binfo_ptr, binfo_size);
560 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
561 nilfs_segctor_end_finfo(sci, inode);
562 err = nilfs_segctor_feed_segment(sci);
567 if (unlikely(required)) {
568 nilfs_segctor_zeropad_segsum(sci);
569 err = nilfs_segbuf_extend_segsum(segbuf);
573 if (sci->sc_blk_cnt == 0)
574 nilfs_segctor_begin_finfo(sci, inode);
576 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
577 /* Substitution to vblocknr is delayed until update_blocknr() */
578 nilfs_segbuf_add_file_buffer(segbuf, bh);
585 * Callback functions that enumerate, mark, and collect dirty blocks
587 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
588 struct buffer_head *bh, struct inode *inode)
592 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
596 err = nilfs_segctor_add_file_block(sci, bh, inode,
597 sizeof(struct nilfs_binfo_v));
599 sci->sc_datablk_cnt++;
603 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
604 struct buffer_head *bh,
607 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
610 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
611 struct buffer_head *bh,
614 WARN_ON(!buffer_dirty(bh));
615 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
618 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
619 struct nilfs_segsum_pointer *ssp,
620 union nilfs_binfo *binfo)
622 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
623 sci, ssp, sizeof(*binfo_v));
624 *binfo_v = binfo->bi_v;
627 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
628 struct nilfs_segsum_pointer *ssp,
629 union nilfs_binfo *binfo)
631 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
632 sci, ssp, sizeof(*vblocknr));
633 *vblocknr = binfo->bi_v.bi_vblocknr;
636 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
637 .collect_data = nilfs_collect_file_data,
638 .collect_node = nilfs_collect_file_node,
639 .collect_bmap = nilfs_collect_file_bmap,
640 .write_data_binfo = nilfs_write_file_data_binfo,
641 .write_node_binfo = nilfs_write_file_node_binfo,
644 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
645 struct buffer_head *bh, struct inode *inode)
649 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
653 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
655 sci->sc_datablk_cnt++;
659 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
660 struct buffer_head *bh, struct inode *inode)
662 WARN_ON(!buffer_dirty(bh));
663 return nilfs_segctor_add_file_block(sci, bh, inode,
664 sizeof(struct nilfs_binfo_dat));
667 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
668 struct nilfs_segsum_pointer *ssp,
669 union nilfs_binfo *binfo)
671 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
673 *blkoff = binfo->bi_dat.bi_blkoff;
676 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
677 struct nilfs_segsum_pointer *ssp,
678 union nilfs_binfo *binfo)
680 struct nilfs_binfo_dat *binfo_dat =
681 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
682 *binfo_dat = binfo->bi_dat;
685 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
686 .collect_data = nilfs_collect_dat_data,
687 .collect_node = nilfs_collect_file_node,
688 .collect_bmap = nilfs_collect_dat_bmap,
689 .write_data_binfo = nilfs_write_dat_data_binfo,
690 .write_node_binfo = nilfs_write_dat_node_binfo,
693 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
694 .collect_data = nilfs_collect_file_data,
695 .collect_node = NULL,
696 .collect_bmap = NULL,
697 .write_data_binfo = nilfs_write_file_data_binfo,
698 .write_node_binfo = NULL,
701 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
702 struct list_head *listp,
704 loff_t start, loff_t end)
706 struct address_space *mapping = inode->i_mapping;
708 pgoff_t index = 0, last = ULONG_MAX;
712 if (unlikely(start != 0 || end != LLONG_MAX)) {
714 * A valid range is given for sync-ing data pages. The
715 * range is rounded to per-page; extra dirty buffers
716 * may be included if blocksize < pagesize.
718 index = start >> PAGE_SHIFT;
719 last = end >> PAGE_SHIFT;
723 if (unlikely(index > last) ||
724 !pagevec_lookup_range_tag(&pvec, mapping, &index, last,
725 PAGECACHE_TAG_DIRTY))
728 for (i = 0; i < pagevec_count(&pvec); i++) {
729 struct buffer_head *bh, *head;
730 struct page *page = pvec.pages[i];
733 if (unlikely(page->mapping != mapping)) {
734 /* Exclude pages removed from the address space */
738 if (!page_has_buffers(page))
739 create_empty_buffers(page, i_blocksize(inode), 0);
742 bh = head = page_buffers(page);
744 if (!buffer_dirty(bh) || buffer_async_write(bh))
747 list_add_tail(&bh->b_assoc_buffers, listp);
749 if (unlikely(ndirties >= nlimit)) {
750 pagevec_release(&pvec);
754 } while (bh = bh->b_this_page, bh != head);
756 pagevec_release(&pvec);
761 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
762 struct list_head *listp)
764 struct nilfs_inode_info *ii = NILFS_I(inode);
765 struct inode *btnc_inode = ii->i_assoc_inode;
767 struct buffer_head *bh, *head;
776 while (pagevec_lookup_tag(&pvec, btnc_inode->i_mapping, &index,
777 PAGECACHE_TAG_DIRTY)) {
778 for (i = 0; i < pagevec_count(&pvec); i++) {
779 bh = head = page_buffers(pvec.pages[i]);
781 if (buffer_dirty(bh) &&
782 !buffer_async_write(bh)) {
784 list_add_tail(&bh->b_assoc_buffers,
787 bh = bh->b_this_page;
788 } while (bh != head);
790 pagevec_release(&pvec);
795 static void nilfs_dispose_list(struct the_nilfs *nilfs,
796 struct list_head *head, int force)
798 struct nilfs_inode_info *ii, *n;
799 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
802 while (!list_empty(head)) {
803 spin_lock(&nilfs->ns_inode_lock);
804 list_for_each_entry_safe(ii, n, head, i_dirty) {
805 list_del_init(&ii->i_dirty);
807 if (unlikely(ii->i_bh)) {
811 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
812 set_bit(NILFS_I_QUEUED, &ii->i_state);
813 list_add_tail(&ii->i_dirty,
814 &nilfs->ns_dirty_files);
818 if (nv == SC_N_INODEVEC)
821 spin_unlock(&nilfs->ns_inode_lock);
823 for (pii = ivec; nv > 0; pii++, nv--)
824 iput(&(*pii)->vfs_inode);
828 static void nilfs_iput_work_func(struct work_struct *work)
830 struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
832 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
834 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
837 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
838 struct nilfs_root *root)
842 if (nilfs_mdt_fetch_dirty(root->ifile))
844 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
846 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
848 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
853 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
855 return list_empty(&sci->sc_dirty_files) &&
856 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
857 sci->sc_nfreesegs == 0 &&
858 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
861 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
863 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
866 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
867 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
869 spin_lock(&nilfs->ns_inode_lock);
870 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
873 spin_unlock(&nilfs->ns_inode_lock);
877 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
879 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
881 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
882 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
883 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
884 nilfs_mdt_clear_dirty(nilfs->ns_dat);
887 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
889 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
890 struct buffer_head *bh_cp;
891 struct nilfs_checkpoint *raw_cp;
894 /* XXX: this interface will be changed */
895 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
899 * The following code is duplicated with cpfile. But, it is
900 * needed to collect the checkpoint even if it was not newly
903 mark_buffer_dirty(bh_cp);
904 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
905 nilfs_cpfile_put_checkpoint(
906 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
907 } else if (err == -EINVAL || err == -ENOENT) {
908 nilfs_error(sci->sc_super,
909 "checkpoint creation failed due to metadata corruption.");
915 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
917 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
918 struct buffer_head *bh_cp;
919 struct nilfs_checkpoint *raw_cp;
922 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
925 if (err == -EINVAL || err == -ENOENT) {
926 nilfs_error(sci->sc_super,
927 "checkpoint finalization failed due to metadata corruption.");
932 raw_cp->cp_snapshot_list.ssl_next = 0;
933 raw_cp->cp_snapshot_list.ssl_prev = 0;
934 raw_cp->cp_inodes_count =
935 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
936 raw_cp->cp_blocks_count =
937 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
938 raw_cp->cp_nblk_inc =
939 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
940 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
941 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
943 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
944 nilfs_checkpoint_clear_minor(raw_cp);
946 nilfs_checkpoint_set_minor(raw_cp);
948 nilfs_write_inode_common(sci->sc_root->ifile,
949 &raw_cp->cp_ifile_inode, 1);
950 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
957 static void nilfs_fill_in_file_bmap(struct inode *ifile,
958 struct nilfs_inode_info *ii)
961 struct buffer_head *ibh;
962 struct nilfs_inode *raw_inode;
964 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
967 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
969 nilfs_bmap_write(ii->i_bmap, raw_inode);
970 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
974 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
976 struct nilfs_inode_info *ii;
978 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
979 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
980 set_bit(NILFS_I_COLLECTED, &ii->i_state);
984 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
985 struct the_nilfs *nilfs)
987 struct buffer_head *bh_sr;
988 struct nilfs_super_root *raw_sr;
989 unsigned int isz, srsz;
991 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
994 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
995 isz = nilfs->ns_inode_size;
996 srsz = NILFS_SR_BYTES(isz);
998 raw_sr->sr_sum = 0; /* Ensure initialization within this update */
999 raw_sr->sr_bytes = cpu_to_le16(srsz);
1000 raw_sr->sr_nongc_ctime
1001 = cpu_to_le64(nilfs_doing_gc() ?
1002 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
1003 raw_sr->sr_flags = 0;
1005 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
1006 NILFS_SR_DAT_OFFSET(isz), 1);
1007 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
1008 NILFS_SR_CPFILE_OFFSET(isz), 1);
1009 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
1010 NILFS_SR_SUFILE_OFFSET(isz), 1);
1011 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1012 set_buffer_uptodate(bh_sr);
1013 unlock_buffer(bh_sr);
1016 static void nilfs_redirty_inodes(struct list_head *head)
1018 struct nilfs_inode_info *ii;
1020 list_for_each_entry(ii, head, i_dirty) {
1021 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1022 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1026 static void nilfs_drop_collected_inodes(struct list_head *head)
1028 struct nilfs_inode_info *ii;
1030 list_for_each_entry(ii, head, i_dirty) {
1031 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1034 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1035 set_bit(NILFS_I_UPDATED, &ii->i_state);
1039 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1040 struct inode *inode,
1041 struct list_head *listp,
1042 int (*collect)(struct nilfs_sc_info *,
1043 struct buffer_head *,
1046 struct buffer_head *bh, *n;
1050 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1051 list_del_init(&bh->b_assoc_buffers);
1052 err = collect(sci, bh, inode);
1055 goto dispose_buffers;
1061 while (!list_empty(listp)) {
1062 bh = list_first_entry(listp, struct buffer_head,
1064 list_del_init(&bh->b_assoc_buffers);
1070 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1072 /* Remaining number of blocks within segment buffer */
1073 return sci->sc_segbuf_nblocks -
1074 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1077 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1078 struct inode *inode,
1079 const struct nilfs_sc_operations *sc_ops)
1081 LIST_HEAD(data_buffers);
1082 LIST_HEAD(node_buffers);
1085 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1086 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1088 n = nilfs_lookup_dirty_data_buffers(
1089 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1091 err = nilfs_segctor_apply_buffers(
1092 sci, inode, &data_buffers,
1093 sc_ops->collect_data);
1094 BUG_ON(!err); /* always receive -E2BIG or true error */
1098 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1100 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1101 err = nilfs_segctor_apply_buffers(
1102 sci, inode, &data_buffers, sc_ops->collect_data);
1103 if (unlikely(err)) {
1104 /* dispose node list */
1105 nilfs_segctor_apply_buffers(
1106 sci, inode, &node_buffers, NULL);
1109 sci->sc_stage.flags |= NILFS_CF_NODE;
1112 err = nilfs_segctor_apply_buffers(
1113 sci, inode, &node_buffers, sc_ops->collect_node);
1117 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1118 err = nilfs_segctor_apply_buffers(
1119 sci, inode, &node_buffers, sc_ops->collect_bmap);
1123 nilfs_segctor_end_finfo(sci, inode);
1124 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1130 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1131 struct inode *inode)
1133 LIST_HEAD(data_buffers);
1134 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1137 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1138 sci->sc_dsync_start,
1141 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1142 nilfs_collect_file_data);
1144 nilfs_segctor_end_finfo(sci, inode);
1146 /* always receive -E2BIG or true error if n > rest */
1151 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1153 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1154 struct list_head *head;
1155 struct nilfs_inode_info *ii;
1159 switch (nilfs_sc_cstage_get(sci)) {
1162 sci->sc_stage.flags = 0;
1164 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1165 sci->sc_nblk_inc = 0;
1166 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1167 if (mode == SC_LSEG_DSYNC) {
1168 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1173 sci->sc_stage.dirty_file_ptr = NULL;
1174 sci->sc_stage.gc_inode_ptr = NULL;
1175 if (mode == SC_FLUSH_DAT) {
1176 nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1179 nilfs_sc_cstage_inc(sci);
1182 if (nilfs_doing_gc()) {
1183 head = &sci->sc_gc_inodes;
1184 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1186 list_for_each_entry_continue(ii, head, i_dirty) {
1187 err = nilfs_segctor_scan_file(
1188 sci, &ii->vfs_inode,
1189 &nilfs_sc_file_ops);
1190 if (unlikely(err)) {
1191 sci->sc_stage.gc_inode_ptr = list_entry(
1193 struct nilfs_inode_info,
1197 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1199 sci->sc_stage.gc_inode_ptr = NULL;
1201 nilfs_sc_cstage_inc(sci);
1204 head = &sci->sc_dirty_files;
1205 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1207 list_for_each_entry_continue(ii, head, i_dirty) {
1208 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1210 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1211 &nilfs_sc_file_ops);
1212 if (unlikely(err)) {
1213 sci->sc_stage.dirty_file_ptr =
1214 list_entry(ii->i_dirty.prev,
1215 struct nilfs_inode_info,
1219 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1220 /* XXX: required ? */
1222 sci->sc_stage.dirty_file_ptr = NULL;
1223 if (mode == SC_FLUSH_FILE) {
1224 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1227 nilfs_sc_cstage_inc(sci);
1228 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1230 case NILFS_ST_IFILE:
1231 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1232 &nilfs_sc_file_ops);
1235 nilfs_sc_cstage_inc(sci);
1236 /* Creating a checkpoint */
1237 err = nilfs_segctor_create_checkpoint(sci);
1241 case NILFS_ST_CPFILE:
1242 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1243 &nilfs_sc_file_ops);
1246 nilfs_sc_cstage_inc(sci);
1248 case NILFS_ST_SUFILE:
1249 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1250 sci->sc_nfreesegs, &ndone);
1251 if (unlikely(err)) {
1252 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1253 sci->sc_freesegs, ndone,
1257 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1259 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1260 &nilfs_sc_file_ops);
1263 nilfs_sc_cstage_inc(sci);
1267 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1271 if (mode == SC_FLUSH_DAT) {
1272 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1275 nilfs_sc_cstage_inc(sci);
1278 if (mode == SC_LSEG_SR) {
1279 /* Appending a super root */
1280 err = nilfs_segctor_add_super_root(sci);
1284 /* End of a logical segment */
1285 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1286 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1288 case NILFS_ST_DSYNC:
1290 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1291 ii = sci->sc_dsync_inode;
1292 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1295 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1298 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1299 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1312 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1313 * @sci: nilfs_sc_info
1314 * @nilfs: nilfs object
1316 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1317 struct the_nilfs *nilfs)
1319 struct nilfs_segment_buffer *segbuf, *prev;
1323 segbuf = nilfs_segbuf_new(sci->sc_super);
1324 if (unlikely(!segbuf))
1327 if (list_empty(&sci->sc_write_logs)) {
1328 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1329 nilfs->ns_pseg_offset, nilfs);
1330 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1331 nilfs_shift_to_next_segment(nilfs);
1332 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1335 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1336 nextnum = nilfs->ns_nextnum;
1338 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1339 /* Start from the head of a new full segment */
1343 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1344 nilfs_segbuf_map_cont(segbuf, prev);
1345 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1346 nextnum = prev->sb_nextnum;
1348 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1349 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1350 segbuf->sb_sum.seg_seq++;
1355 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1360 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1364 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1366 BUG_ON(!list_empty(&sci->sc_segbufs));
1367 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1368 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1372 nilfs_segbuf_free(segbuf);
1376 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1377 struct the_nilfs *nilfs, int nadd)
1379 struct nilfs_segment_buffer *segbuf, *prev;
1380 struct inode *sufile = nilfs->ns_sufile;
1385 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1387 * Since the segment specified with nextnum might be allocated during
1388 * the previous construction, the buffer including its segusage may
1389 * not be dirty. The following call ensures that the buffer is dirty
1390 * and will pin the buffer on memory until the sufile is written.
1392 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1396 for (i = 0; i < nadd; i++) {
1397 /* extend segment info */
1399 segbuf = nilfs_segbuf_new(sci->sc_super);
1400 if (unlikely(!segbuf))
1403 /* map this buffer to region of segment on-disk */
1404 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1405 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1407 /* allocate the next next full segment */
1408 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1412 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1413 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1415 list_add_tail(&segbuf->sb_list, &list);
1418 list_splice_tail(&list, &sci->sc_segbufs);
1422 nilfs_segbuf_free(segbuf);
1424 list_for_each_entry(segbuf, &list, sb_list) {
1425 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1426 WARN_ON(ret); /* never fails */
1428 nilfs_destroy_logs(&list);
1432 static void nilfs_free_incomplete_logs(struct list_head *logs,
1433 struct the_nilfs *nilfs)
1435 struct nilfs_segment_buffer *segbuf, *prev;
1436 struct inode *sufile = nilfs->ns_sufile;
1439 segbuf = NILFS_FIRST_SEGBUF(logs);
1440 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1441 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1442 WARN_ON(ret); /* never fails */
1444 if (atomic_read(&segbuf->sb_err)) {
1445 /* Case 1: The first segment failed */
1446 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1448 * Case 1a: Partial segment appended into an existing
1451 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1452 segbuf->sb_fseg_end);
1453 else /* Case 1b: New full segment */
1454 set_nilfs_discontinued(nilfs);
1458 list_for_each_entry_continue(segbuf, logs, sb_list) {
1459 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1460 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1461 WARN_ON(ret); /* never fails */
1463 if (atomic_read(&segbuf->sb_err) &&
1464 segbuf->sb_segnum != nilfs->ns_nextnum)
1465 /* Case 2: extended segment (!= next) failed */
1466 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1471 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1472 struct inode *sufile)
1474 struct nilfs_segment_buffer *segbuf;
1475 unsigned long live_blocks;
1478 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1479 live_blocks = segbuf->sb_sum.nblocks +
1480 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1481 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1484 WARN_ON(ret); /* always succeed because the segusage is dirty */
1488 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1490 struct nilfs_segment_buffer *segbuf;
1493 segbuf = NILFS_FIRST_SEGBUF(logs);
1494 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1495 segbuf->sb_pseg_start -
1496 segbuf->sb_fseg_start, 0);
1497 WARN_ON(ret); /* always succeed because the segusage is dirty */
1499 list_for_each_entry_continue(segbuf, logs, sb_list) {
1500 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1502 WARN_ON(ret); /* always succeed */
1506 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1507 struct nilfs_segment_buffer *last,
1508 struct inode *sufile)
1510 struct nilfs_segment_buffer *segbuf = last;
1513 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1514 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1515 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1518 nilfs_truncate_logs(&sci->sc_segbufs, last);
1522 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1523 struct the_nilfs *nilfs, int mode)
1525 struct nilfs_cstage prev_stage = sci->sc_stage;
1528 /* Collection retry loop */
1530 sci->sc_nblk_this_inc = 0;
1531 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1533 err = nilfs_segctor_reset_segment_buffer(sci);
1537 err = nilfs_segctor_collect_blocks(sci, mode);
1538 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1542 if (unlikely(err != -E2BIG))
1545 /* The current segment is filled up */
1546 if (mode != SC_LSEG_SR ||
1547 nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1550 nilfs_clear_logs(&sci->sc_segbufs);
1552 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1553 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1557 WARN_ON(err); /* do not happen */
1558 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1561 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1565 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1566 sci->sc_stage = prev_stage;
1568 nilfs_segctor_zeropad_segsum(sci);
1569 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1576 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1577 struct buffer_head *new_bh)
1579 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1581 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1582 /* The caller must release old_bh */
1586 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1587 struct nilfs_segment_buffer *segbuf,
1590 struct inode *inode = NULL;
1592 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1593 unsigned long nblocks = 0, ndatablk = 0;
1594 const struct nilfs_sc_operations *sc_op = NULL;
1595 struct nilfs_segsum_pointer ssp;
1596 struct nilfs_finfo *finfo = NULL;
1597 union nilfs_binfo binfo;
1598 struct buffer_head *bh, *bh_org;
1605 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1606 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1607 ssp.offset = sizeof(struct nilfs_segment_summary);
1609 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1610 if (bh == segbuf->sb_super_root)
1613 finfo = nilfs_segctor_map_segsum_entry(
1614 sci, &ssp, sizeof(*finfo));
1615 ino = le64_to_cpu(finfo->fi_ino);
1616 nblocks = le32_to_cpu(finfo->fi_nblocks);
1617 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1619 inode = bh->b_page->mapping->host;
1621 if (mode == SC_LSEG_DSYNC)
1622 sc_op = &nilfs_sc_dsync_ops;
1623 else if (ino == NILFS_DAT_INO)
1624 sc_op = &nilfs_sc_dat_ops;
1625 else /* file blocks */
1626 sc_op = &nilfs_sc_file_ops;
1630 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1633 nilfs_list_replace_buffer(bh_org, bh);
1639 sc_op->write_data_binfo(sci, &ssp, &binfo);
1641 sc_op->write_node_binfo(sci, &ssp, &binfo);
1644 if (--nblocks == 0) {
1648 } else if (ndatablk > 0)
1658 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1660 struct nilfs_segment_buffer *segbuf;
1663 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1664 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1667 nilfs_segbuf_fill_in_segsum(segbuf);
1672 static void nilfs_begin_page_io(struct page *page)
1674 if (!page || PageWriteback(page))
1676 * For split b-tree node pages, this function may be called
1677 * twice. We ignore the 2nd or later calls by this check.
1682 clear_page_dirty_for_io(page);
1683 set_page_writeback(page);
1687 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1689 struct nilfs_segment_buffer *segbuf;
1690 struct page *bd_page = NULL, *fs_page = NULL;
1692 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1693 struct buffer_head *bh;
1695 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1697 if (bh->b_page != bd_page) {
1700 clear_page_dirty_for_io(bd_page);
1701 set_page_writeback(bd_page);
1702 unlock_page(bd_page);
1704 bd_page = bh->b_page;
1708 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1710 if (bh == segbuf->sb_super_root) {
1711 if (bh->b_page != bd_page) {
1713 clear_page_dirty_for_io(bd_page);
1714 set_page_writeback(bd_page);
1715 unlock_page(bd_page);
1716 bd_page = bh->b_page;
1720 set_buffer_async_write(bh);
1721 if (bh->b_page != fs_page) {
1722 nilfs_begin_page_io(fs_page);
1723 fs_page = bh->b_page;
1729 clear_page_dirty_for_io(bd_page);
1730 set_page_writeback(bd_page);
1731 unlock_page(bd_page);
1733 nilfs_begin_page_io(fs_page);
1736 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1737 struct the_nilfs *nilfs)
1741 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1742 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1746 static void nilfs_end_page_io(struct page *page, int err)
1751 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1753 * For b-tree node pages, this function may be called twice
1754 * or more because they might be split in a segment.
1756 if (PageDirty(page)) {
1758 * For pages holding split b-tree node buffers, dirty
1759 * flag on the buffers may be cleared discretely.
1760 * In that case, the page is once redirtied for
1761 * remaining buffers, and it must be cancelled if
1762 * all the buffers get cleaned later.
1765 if (nilfs_page_buffers_clean(page))
1766 __nilfs_clear_page_dirty(page);
1773 if (!nilfs_page_buffers_clean(page))
1774 __set_page_dirty_nobuffers(page);
1775 ClearPageError(page);
1777 __set_page_dirty_nobuffers(page);
1781 end_page_writeback(page);
1784 static void nilfs_abort_logs(struct list_head *logs, int err)
1786 struct nilfs_segment_buffer *segbuf;
1787 struct page *bd_page = NULL, *fs_page = NULL;
1788 struct buffer_head *bh;
1790 if (list_empty(logs))
1793 list_for_each_entry(segbuf, logs, sb_list) {
1794 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1796 clear_buffer_uptodate(bh);
1797 if (bh->b_page != bd_page) {
1799 end_page_writeback(bd_page);
1800 bd_page = bh->b_page;
1804 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1806 if (bh == segbuf->sb_super_root) {
1807 clear_buffer_uptodate(bh);
1808 if (bh->b_page != bd_page) {
1809 end_page_writeback(bd_page);
1810 bd_page = bh->b_page;
1814 clear_buffer_async_write(bh);
1815 if (bh->b_page != fs_page) {
1816 nilfs_end_page_io(fs_page, err);
1817 fs_page = bh->b_page;
1822 end_page_writeback(bd_page);
1824 nilfs_end_page_io(fs_page, err);
1827 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1828 struct the_nilfs *nilfs, int err)
1833 list_splice_tail_init(&sci->sc_write_logs, &logs);
1834 ret = nilfs_wait_on_logs(&logs);
1835 nilfs_abort_logs(&logs, ret ? : err);
1837 list_splice_tail_init(&sci->sc_segbufs, &logs);
1838 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1839 nilfs_free_incomplete_logs(&logs, nilfs);
1841 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1842 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1846 WARN_ON(ret); /* do not happen */
1849 nilfs_destroy_logs(&logs);
1852 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1853 struct nilfs_segment_buffer *segbuf)
1855 nilfs->ns_segnum = segbuf->sb_segnum;
1856 nilfs->ns_nextnum = segbuf->sb_nextnum;
1857 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1858 + segbuf->sb_sum.nblocks;
1859 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1860 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1863 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1865 struct nilfs_segment_buffer *segbuf;
1866 struct page *bd_page = NULL, *fs_page = NULL;
1867 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1868 int update_sr = false;
1870 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1871 struct buffer_head *bh;
1873 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1875 set_buffer_uptodate(bh);
1876 clear_buffer_dirty(bh);
1877 if (bh->b_page != bd_page) {
1879 end_page_writeback(bd_page);
1880 bd_page = bh->b_page;
1884 * We assume that the buffers which belong to the same page
1885 * continue over the buffer list.
1886 * Under this assumption, the last BHs of pages is
1887 * identifiable by the discontinuity of bh->b_page
1888 * (page != fs_page).
1890 * For B-tree node blocks, however, this assumption is not
1891 * guaranteed. The cleanup code of B-tree node pages needs
1894 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1896 const unsigned long set_bits = BIT(BH_Uptodate);
1897 const unsigned long clear_bits =
1898 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1899 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1900 BIT(BH_NILFS_Redirected));
1902 if (bh == segbuf->sb_super_root) {
1903 set_buffer_uptodate(bh);
1904 clear_buffer_dirty(bh);
1905 if (bh->b_page != bd_page) {
1906 end_page_writeback(bd_page);
1907 bd_page = bh->b_page;
1912 set_mask_bits(&bh->b_state, clear_bits, set_bits);
1913 if (bh->b_page != fs_page) {
1914 nilfs_end_page_io(fs_page, 0);
1915 fs_page = bh->b_page;
1919 if (!nilfs_segbuf_simplex(segbuf)) {
1920 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1921 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1922 sci->sc_lseg_stime = jiffies;
1924 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1925 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1929 * Since pages may continue over multiple segment buffers,
1930 * end of the last page must be checked outside of the loop.
1933 end_page_writeback(bd_page);
1935 nilfs_end_page_io(fs_page, 0);
1937 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1939 if (nilfs_doing_gc())
1940 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1942 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1944 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1946 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1947 nilfs_set_next_segment(nilfs, segbuf);
1950 nilfs->ns_flushed_device = 0;
1951 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1952 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1954 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1955 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1956 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1957 nilfs_segctor_clear_metadata_dirty(sci);
1959 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1962 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1966 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1968 nilfs_segctor_complete_write(sci);
1969 nilfs_destroy_logs(&sci->sc_write_logs);
1974 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1975 struct the_nilfs *nilfs)
1977 struct nilfs_inode_info *ii, *n;
1978 struct inode *ifile = sci->sc_root->ifile;
1980 spin_lock(&nilfs->ns_inode_lock);
1982 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1984 struct buffer_head *ibh;
1987 spin_unlock(&nilfs->ns_inode_lock);
1988 err = nilfs_ifile_get_inode_block(
1989 ifile, ii->vfs_inode.i_ino, &ibh);
1990 if (unlikely(err)) {
1991 nilfs_warn(sci->sc_super,
1992 "log writer: error %d getting inode block (ino=%lu)",
1993 err, ii->vfs_inode.i_ino);
1996 spin_lock(&nilfs->ns_inode_lock);
1997 if (likely(!ii->i_bh))
2004 // Always redirty the buffer to avoid race condition
2005 mark_buffer_dirty(ii->i_bh);
2006 nilfs_mdt_mark_dirty(ifile);
2008 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2009 set_bit(NILFS_I_BUSY, &ii->i_state);
2010 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2012 spin_unlock(&nilfs->ns_inode_lock);
2017 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2018 struct the_nilfs *nilfs)
2020 struct nilfs_inode_info *ii, *n;
2021 int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2022 int defer_iput = false;
2024 spin_lock(&nilfs->ns_inode_lock);
2025 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2026 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2027 test_bit(NILFS_I_DIRTY, &ii->i_state))
2030 clear_bit(NILFS_I_BUSY, &ii->i_state);
2033 list_del_init(&ii->i_dirty);
2034 if (!ii->vfs_inode.i_nlink || during_mount) {
2036 * Defer calling iput() to avoid deadlocks if
2037 * i_nlink == 0 or mount is not yet finished.
2039 list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2042 spin_unlock(&nilfs->ns_inode_lock);
2043 iput(&ii->vfs_inode);
2044 spin_lock(&nilfs->ns_inode_lock);
2047 spin_unlock(&nilfs->ns_inode_lock);
2050 schedule_work(&sci->sc_iput_work);
2054 * Main procedure of segment constructor
2056 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2058 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2061 if (sb_rdonly(sci->sc_super))
2064 nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2065 sci->sc_cno = nilfs->ns_cno;
2067 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2071 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2072 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2074 if (nilfs_segctor_clean(sci))
2078 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2080 err = nilfs_segctor_begin_construction(sci, nilfs);
2084 /* Update time stamp */
2085 sci->sc_seg_ctime = ktime_get_real_seconds();
2087 err = nilfs_segctor_collect(sci, nilfs, mode);
2091 /* Avoid empty segment */
2092 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2093 nilfs_segbuf_empty(sci->sc_curseg)) {
2094 nilfs_segctor_abort_construction(sci, nilfs, 1);
2098 err = nilfs_segctor_assign(sci, mode);
2102 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2103 nilfs_segctor_fill_in_file_bmap(sci);
2105 if (mode == SC_LSEG_SR &&
2106 nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2107 err = nilfs_segctor_fill_in_checkpoint(sci);
2109 goto failed_to_write;
2111 nilfs_segctor_fill_in_super_root(sci, nilfs);
2113 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2115 /* Write partial segments */
2116 nilfs_segctor_prepare_write(sci);
2118 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2119 nilfs->ns_crc_seed);
2121 err = nilfs_segctor_write(sci, nilfs);
2123 goto failed_to_write;
2125 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2126 nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2128 * At this point, we avoid double buffering
2129 * for blocksize < pagesize because page dirty
2130 * flag is turned off during write and dirty
2131 * buffers are not properly collected for
2132 * pages crossing over segments.
2134 err = nilfs_segctor_wait(sci);
2136 goto failed_to_write;
2138 } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2141 nilfs_segctor_drop_written_files(sci, nilfs);
2145 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2146 nilfs_redirty_inodes(&sci->sc_dirty_files);
2149 if (nilfs_doing_gc())
2150 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2151 nilfs_segctor_abort_construction(sci, nilfs, err);
2156 * nilfs_segctor_start_timer - set timer of background write
2157 * @sci: nilfs_sc_info
2159 * If the timer has already been set, it ignores the new request.
2160 * This function MUST be called within a section locking the segment
2163 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2165 spin_lock(&sci->sc_state_lock);
2166 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2167 sci->sc_timer.expires = jiffies + sci->sc_interval;
2168 add_timer(&sci->sc_timer);
2169 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2171 spin_unlock(&sci->sc_state_lock);
2174 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2176 spin_lock(&sci->sc_state_lock);
2177 if (!(sci->sc_flush_request & BIT(bn))) {
2178 unsigned long prev_req = sci->sc_flush_request;
2180 sci->sc_flush_request |= BIT(bn);
2182 wake_up(&sci->sc_wait_daemon);
2184 spin_unlock(&sci->sc_state_lock);
2188 * nilfs_flush_segment - trigger a segment construction for resource control
2190 * @ino: inode number of the file to be flushed out.
2192 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2194 struct the_nilfs *nilfs = sb->s_fs_info;
2195 struct nilfs_sc_info *sci = nilfs->ns_writer;
2197 if (!sci || nilfs_doing_construction())
2199 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2200 /* assign bit 0 to data files */
2203 struct nilfs_segctor_wait_request {
2204 wait_queue_entry_t wq;
2210 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2212 struct nilfs_segctor_wait_request wait_req;
2215 spin_lock(&sci->sc_state_lock);
2216 init_wait(&wait_req.wq);
2218 atomic_set(&wait_req.done, 0);
2219 wait_req.seq = ++sci->sc_seq_request;
2220 spin_unlock(&sci->sc_state_lock);
2222 init_waitqueue_entry(&wait_req.wq, current);
2223 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2224 set_current_state(TASK_INTERRUPTIBLE);
2225 wake_up(&sci->sc_wait_daemon);
2228 if (atomic_read(&wait_req.done)) {
2232 if (!signal_pending(current)) {
2239 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2243 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2245 struct nilfs_segctor_wait_request *wrq, *n;
2246 unsigned long flags;
2248 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2249 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2250 if (!atomic_read(&wrq->done) &&
2251 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2253 atomic_set(&wrq->done, 1);
2255 if (atomic_read(&wrq->done)) {
2256 wrq->wq.func(&wrq->wq,
2257 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2261 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2265 * nilfs_construct_segment - construct a logical segment
2268 * Return Value: On success, 0 is retured. On errors, one of the following
2269 * negative error code is returned.
2271 * %-EROFS - Read only filesystem.
2275 * %-ENOSPC - No space left on device (only in a panic state).
2277 * %-ERESTARTSYS - Interrupted.
2279 * %-ENOMEM - Insufficient memory available.
2281 int nilfs_construct_segment(struct super_block *sb)
2283 struct the_nilfs *nilfs = sb->s_fs_info;
2284 struct nilfs_sc_info *sci = nilfs->ns_writer;
2285 struct nilfs_transaction_info *ti;
2288 if (sb_rdonly(sb) || unlikely(!sci))
2291 /* A call inside transactions causes a deadlock. */
2292 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2294 err = nilfs_segctor_sync(sci);
2299 * nilfs_construct_dsync_segment - construct a data-only logical segment
2301 * @inode: inode whose data blocks should be written out
2302 * @start: start byte offset
2303 * @end: end byte offset (inclusive)
2305 * Return Value: On success, 0 is retured. On errors, one of the following
2306 * negative error code is returned.
2308 * %-EROFS - Read only filesystem.
2312 * %-ENOSPC - No space left on device (only in a panic state).
2314 * %-ERESTARTSYS - Interrupted.
2316 * %-ENOMEM - Insufficient memory available.
2318 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2319 loff_t start, loff_t end)
2321 struct the_nilfs *nilfs = sb->s_fs_info;
2322 struct nilfs_sc_info *sci = nilfs->ns_writer;
2323 struct nilfs_inode_info *ii;
2324 struct nilfs_transaction_info ti;
2327 if (sb_rdonly(sb) || unlikely(!sci))
2330 nilfs_transaction_lock(sb, &ti, 0);
2332 ii = NILFS_I(inode);
2333 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2334 nilfs_test_opt(nilfs, STRICT_ORDER) ||
2335 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2336 nilfs_discontinued(nilfs)) {
2337 nilfs_transaction_unlock(sb);
2338 err = nilfs_segctor_sync(sci);
2342 spin_lock(&nilfs->ns_inode_lock);
2343 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2344 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2345 spin_unlock(&nilfs->ns_inode_lock);
2346 nilfs_transaction_unlock(sb);
2349 spin_unlock(&nilfs->ns_inode_lock);
2350 sci->sc_dsync_inode = ii;
2351 sci->sc_dsync_start = start;
2352 sci->sc_dsync_end = end;
2354 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2356 nilfs->ns_flushed_device = 0;
2358 nilfs_transaction_unlock(sb);
2362 #define FLUSH_FILE_BIT (0x1) /* data file only */
2363 #define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */
2366 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2367 * @sci: segment constructor object
2369 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2371 spin_lock(&sci->sc_state_lock);
2372 sci->sc_seq_accepted = sci->sc_seq_request;
2373 spin_unlock(&sci->sc_state_lock);
2374 del_timer_sync(&sci->sc_timer);
2378 * nilfs_segctor_notify - notify the result of request to caller threads
2379 * @sci: segment constructor object
2380 * @mode: mode of log forming
2381 * @err: error code to be notified
2383 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2385 /* Clear requests (even when the construction failed) */
2386 spin_lock(&sci->sc_state_lock);
2388 if (mode == SC_LSEG_SR) {
2389 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2390 sci->sc_seq_done = sci->sc_seq_accepted;
2391 nilfs_segctor_wakeup(sci, err);
2392 sci->sc_flush_request = 0;
2394 if (mode == SC_FLUSH_FILE)
2395 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2396 else if (mode == SC_FLUSH_DAT)
2397 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2399 /* re-enable timer if checkpoint creation was not done */
2400 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2401 time_before(jiffies, sci->sc_timer.expires))
2402 add_timer(&sci->sc_timer);
2404 spin_unlock(&sci->sc_state_lock);
2408 * nilfs_segctor_construct - form logs and write them to disk
2409 * @sci: segment constructor object
2410 * @mode: mode of log forming
2412 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2414 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2415 struct nilfs_super_block **sbp;
2418 nilfs_segctor_accept(sci);
2420 if (nilfs_discontinued(nilfs))
2422 if (!nilfs_segctor_confirm(sci))
2423 err = nilfs_segctor_do_construct(sci, mode);
2426 if (mode != SC_FLUSH_DAT)
2427 atomic_set(&nilfs->ns_ndirtyblks, 0);
2428 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2429 nilfs_discontinued(nilfs)) {
2430 down_write(&nilfs->ns_sem);
2432 sbp = nilfs_prepare_super(sci->sc_super,
2433 nilfs_sb_will_flip(nilfs));
2435 nilfs_set_log_cursor(sbp[0], nilfs);
2436 err = nilfs_commit_super(sci->sc_super,
2439 up_write(&nilfs->ns_sem);
2443 nilfs_segctor_notify(sci, mode, err);
2447 static void nilfs_construction_timeout(struct timer_list *t)
2449 struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2451 wake_up_process(sci->sc_timer_task);
2455 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2457 struct nilfs_inode_info *ii, *n;
2459 list_for_each_entry_safe(ii, n, head, i_dirty) {
2460 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2462 list_del_init(&ii->i_dirty);
2463 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2464 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2465 iput(&ii->vfs_inode);
2469 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2472 struct the_nilfs *nilfs = sb->s_fs_info;
2473 struct nilfs_sc_info *sci = nilfs->ns_writer;
2474 struct nilfs_transaction_info ti;
2480 nilfs_transaction_lock(sb, &ti, 1);
2482 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2486 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2487 if (unlikely(err)) {
2488 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2492 sci->sc_freesegs = kbufs[4];
2493 sci->sc_nfreesegs = argv[4].v_nmembs;
2494 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2497 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2498 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2503 nilfs_warn(sb, "error %d cleaning segments", err);
2504 set_current_state(TASK_INTERRUPTIBLE);
2505 schedule_timeout(sci->sc_interval);
2507 if (nilfs_test_opt(nilfs, DISCARD)) {
2508 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2512 "error %d on discard request, turning discards off for the device",
2514 nilfs_clear_opt(nilfs, DISCARD);
2519 sci->sc_freesegs = NULL;
2520 sci->sc_nfreesegs = 0;
2521 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2522 nilfs_transaction_unlock(sb);
2526 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2528 struct nilfs_transaction_info ti;
2530 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2531 nilfs_segctor_construct(sci, mode);
2534 * Unclosed segment should be retried. We do this using sc_timer.
2535 * Timeout of sc_timer will invoke complete construction which leads
2536 * to close the current logical segment.
2538 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2539 nilfs_segctor_start_timer(sci);
2541 nilfs_transaction_unlock(sci->sc_super);
2544 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2548 spin_lock(&sci->sc_state_lock);
2549 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2550 SC_FLUSH_DAT : SC_FLUSH_FILE;
2551 spin_unlock(&sci->sc_state_lock);
2554 nilfs_segctor_do_construct(sci, mode);
2556 spin_lock(&sci->sc_state_lock);
2557 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2558 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2559 spin_unlock(&sci->sc_state_lock);
2561 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2564 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2566 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2567 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2568 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2569 return SC_FLUSH_FILE;
2570 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2571 return SC_FLUSH_DAT;
2577 * nilfs_segctor_thread - main loop of the segment constructor thread.
2578 * @arg: pointer to a struct nilfs_sc_info.
2580 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2581 * to execute segment constructions.
2583 static int nilfs_segctor_thread(void *arg)
2585 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2586 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2589 sci->sc_timer_task = current;
2592 sci->sc_task = current;
2593 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2594 nilfs_info(sci->sc_super,
2595 "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2596 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2598 spin_lock(&sci->sc_state_lock);
2603 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2606 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2608 else if (sci->sc_flush_request)
2609 mode = nilfs_segctor_flush_mode(sci);
2613 spin_unlock(&sci->sc_state_lock);
2614 nilfs_segctor_thread_construct(sci, mode);
2615 spin_lock(&sci->sc_state_lock);
2620 if (freezing(current)) {
2621 spin_unlock(&sci->sc_state_lock);
2623 spin_lock(&sci->sc_state_lock);
2626 int should_sleep = 1;
2628 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2629 TASK_INTERRUPTIBLE);
2631 if (sci->sc_seq_request != sci->sc_seq_done)
2633 else if (sci->sc_flush_request)
2635 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2636 should_sleep = time_before(jiffies,
2637 sci->sc_timer.expires);
2640 spin_unlock(&sci->sc_state_lock);
2642 spin_lock(&sci->sc_state_lock);
2644 finish_wait(&sci->sc_wait_daemon, &wait);
2645 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2646 time_after_eq(jiffies, sci->sc_timer.expires));
2648 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2649 set_nilfs_discontinued(nilfs);
2655 sci->sc_task = NULL;
2656 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2657 spin_unlock(&sci->sc_state_lock);
2661 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2663 struct task_struct *t;
2665 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2667 int err = PTR_ERR(t);
2669 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2673 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2677 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2678 __acquires(&sci->sc_state_lock)
2679 __releases(&sci->sc_state_lock)
2681 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2683 while (sci->sc_task) {
2684 wake_up(&sci->sc_wait_daemon);
2685 spin_unlock(&sci->sc_state_lock);
2686 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2687 spin_lock(&sci->sc_state_lock);
2692 * Setup & clean-up functions
2694 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2695 struct nilfs_root *root)
2697 struct the_nilfs *nilfs = sb->s_fs_info;
2698 struct nilfs_sc_info *sci;
2700 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2706 nilfs_get_root(root);
2707 sci->sc_root = root;
2709 init_waitqueue_head(&sci->sc_wait_request);
2710 init_waitqueue_head(&sci->sc_wait_daemon);
2711 init_waitqueue_head(&sci->sc_wait_task);
2712 spin_lock_init(&sci->sc_state_lock);
2713 INIT_LIST_HEAD(&sci->sc_dirty_files);
2714 INIT_LIST_HEAD(&sci->sc_segbufs);
2715 INIT_LIST_HEAD(&sci->sc_write_logs);
2716 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2717 INIT_LIST_HEAD(&sci->sc_iput_queue);
2718 INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2719 timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2721 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2722 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2723 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2725 if (nilfs->ns_interval)
2726 sci->sc_interval = HZ * nilfs->ns_interval;
2727 if (nilfs->ns_watermark)
2728 sci->sc_watermark = nilfs->ns_watermark;
2732 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2734 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2737 * The segctord thread was stopped and its timer was removed.
2738 * But some tasks remain.
2741 struct nilfs_transaction_info ti;
2743 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2744 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2745 nilfs_transaction_unlock(sci->sc_super);
2747 flush_work(&sci->sc_iput_work);
2749 } while (ret && ret != -EROFS && retrycount-- > 0);
2753 * nilfs_segctor_destroy - destroy the segment constructor.
2754 * @sci: nilfs_sc_info
2756 * nilfs_segctor_destroy() kills the segctord thread and frees
2757 * the nilfs_sc_info struct.
2758 * Caller must hold the segment semaphore.
2760 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2762 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2765 up_write(&nilfs->ns_segctor_sem);
2767 spin_lock(&sci->sc_state_lock);
2768 nilfs_segctor_kill_thread(sci);
2769 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2770 || sci->sc_seq_request != sci->sc_seq_done);
2771 spin_unlock(&sci->sc_state_lock);
2773 if (flush_work(&sci->sc_iput_work))
2776 if (flag || !nilfs_segctor_confirm(sci))
2777 nilfs_segctor_write_out(sci);
2779 if (!list_empty(&sci->sc_dirty_files)) {
2780 nilfs_warn(sci->sc_super,
2781 "disposed unprocessed dirty file(s) when stopping log writer");
2782 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2785 if (!list_empty(&sci->sc_iput_queue)) {
2786 nilfs_warn(sci->sc_super,
2787 "disposed unprocessed inode(s) in iput queue when stopping log writer");
2788 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2791 WARN_ON(!list_empty(&sci->sc_segbufs));
2792 WARN_ON(!list_empty(&sci->sc_write_logs));
2794 nilfs_put_root(sci->sc_root);
2796 down_write(&nilfs->ns_segctor_sem);
2798 del_timer_sync(&sci->sc_timer);
2803 * nilfs_attach_log_writer - attach log writer
2804 * @sb: super block instance
2805 * @root: root object of the current filesystem tree
2807 * This allocates a log writer object, initializes it, and starts the
2810 * Return Value: On success, 0 is returned. On error, one of the following
2811 * negative error code is returned.
2813 * %-ENOMEM - Insufficient memory available.
2815 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2817 struct the_nilfs *nilfs = sb->s_fs_info;
2820 if (nilfs->ns_writer) {
2822 * This happens if the filesystem is made read-only by
2823 * __nilfs_error or nilfs_remount and then remounted
2824 * read/write. In these cases, reuse the existing
2830 nilfs->ns_writer = nilfs_segctor_new(sb, root);
2831 if (!nilfs->ns_writer)
2834 inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2836 err = nilfs_segctor_start_thread(nilfs->ns_writer);
2838 nilfs_detach_log_writer(sb);
2844 * nilfs_detach_log_writer - destroy log writer
2845 * @sb: super block instance
2847 * This kills log writer daemon, frees the log writer object, and
2848 * destroys list of dirty files.
2850 void nilfs_detach_log_writer(struct super_block *sb)
2852 struct the_nilfs *nilfs = sb->s_fs_info;
2853 LIST_HEAD(garbage_list);
2855 down_write(&nilfs->ns_segctor_sem);
2856 if (nilfs->ns_writer) {
2857 nilfs_segctor_destroy(nilfs->ns_writer);
2858 nilfs->ns_writer = NULL;
2860 set_nilfs_purging(nilfs);
2862 /* Force to free the list of dirty files */
2863 spin_lock(&nilfs->ns_inode_lock);
2864 if (!list_empty(&nilfs->ns_dirty_files)) {
2865 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2867 "disposed unprocessed dirty file(s) when detaching log writer");
2869 spin_unlock(&nilfs->ns_inode_lock);
2870 up_write(&nilfs->ns_segctor_sem);
2872 nilfs_dispose_list(nilfs, &garbage_list, 1);
2873 clear_nilfs_purging(nilfs);