nilfs2: use list_first_entry
[linux-2.6.git] / fs / nilfs2 / segment.c
1 /*
2  * segment.c - NILFS segment constructor.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
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.
10  *
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.
15  *
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
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  *
22  */
23
24 #include <linux/pagemap.h>
25 #include <linux/buffer_head.h>
26 #include <linux/writeback.h>
27 #include <linux/bio.h>
28 #include <linux/completion.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/freezer.h>
32 #include <linux/kthread.h>
33 #include <linux/crc32.h>
34 #include <linux/pagevec.h>
35 #include <linux/slab.h>
36 #include "nilfs.h"
37 #include "btnode.h"
38 #include "page.h"
39 #include "segment.h"
40 #include "sufile.h"
41 #include "cpfile.h"
42 #include "ifile.h"
43 #include "segbuf.h"
44
45
46 /*
47  * Segment constructor
48  */
49 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
50
51 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
52                                 appended in collection retry loop */
53
54 /* Construction mode */
55 enum {
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
62                            a checkpoint */
63 };
64
65 /* Stage numbers of dirty block collection */
66 enum {
67         NILFS_ST_INIT = 0,
68         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
69         NILFS_ST_FILE,
70         NILFS_ST_IFILE,
71         NILFS_ST_CPFILE,
72         NILFS_ST_SUFILE,
73         NILFS_ST_DAT,
74         NILFS_ST_SR,            /* Super root */
75         NILFS_ST_DSYNC,         /* Data sync blocks */
76         NILFS_ST_DONE,
77 };
78
79 /* State flags of collection */
80 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
81 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
82 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
83 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
84
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 *,
88                             struct inode *);
89         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
90                             struct inode *);
91         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
92                             struct inode *);
93         void (*write_data_binfo)(struct nilfs_sc_info *,
94                                  struct nilfs_segsum_pointer *,
95                                  union nilfs_binfo *);
96         void (*write_node_binfo)(struct nilfs_sc_info *,
97                                  struct nilfs_segsum_pointer *,
98                                  union nilfs_binfo *);
99 };
100
101 /*
102  * Other definitions
103  */
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 the_nilfs *, struct list_head *, int);
108
109 #define nilfs_cnt32_gt(a, b)   \
110         (typecheck(__u32, a) && typecheck(__u32, b) && \
111          ((__s32)(b) - (__s32)(a) < 0))
112 #define nilfs_cnt32_ge(a, b)   \
113         (typecheck(__u32, a) && typecheck(__u32, b) && \
114          ((__s32)(a) - (__s32)(b) >= 0))
115 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
116 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
117
118 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
119 {
120         struct nilfs_transaction_info *cur_ti = current->journal_info;
121         void *save = NULL;
122
123         if (cur_ti) {
124                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
125                         return ++cur_ti->ti_count;
126                 else {
127                         /*
128                          * If journal_info field is occupied by other FS,
129                          * it is saved and will be restored on
130                          * nilfs_transaction_commit().
131                          */
132                         printk(KERN_WARNING
133                                "NILFS warning: journal info from a different "
134                                "FS\n");
135                         save = current->journal_info;
136                 }
137         }
138         if (!ti) {
139                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
140                 if (!ti)
141                         return -ENOMEM;
142                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
143         } else {
144                 ti->ti_flags = 0;
145         }
146         ti->ti_count = 0;
147         ti->ti_save = save;
148         ti->ti_magic = NILFS_TI_MAGIC;
149         current->journal_info = ti;
150         return 0;
151 }
152
153 /**
154  * nilfs_transaction_begin - start indivisible file operations.
155  * @sb: super block
156  * @ti: nilfs_transaction_info
157  * @vacancy_check: flags for vacancy rate checks
158  *
159  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
160  * the segment semaphore, to make a segment construction and write tasks
161  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
162  * The region enclosed by these two functions can be nested.  To avoid a
163  * deadlock, the semaphore is only acquired or released in the outermost call.
164  *
165  * This function allocates a nilfs_transaction_info struct to keep context
166  * information on it.  It is initialized and hooked onto the current task in
167  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
168  * instead; otherwise a new struct is assigned from a slab.
169  *
170  * When @vacancy_check flag is set, this function will check the amount of
171  * free space, and will wait for the GC to reclaim disk space if low capacity.
172  *
173  * Return Value: On success, 0 is returned. On error, one of the following
174  * negative error code is returned.
175  *
176  * %-ENOMEM - Insufficient memory available.
177  *
178  * %-ENOSPC - No space left on device
179  */
180 int nilfs_transaction_begin(struct super_block *sb,
181                             struct nilfs_transaction_info *ti,
182                             int vacancy_check)
183 {
184         struct the_nilfs *nilfs;
185         int ret = nilfs_prepare_segment_lock(ti);
186
187         if (unlikely(ret < 0))
188                 return ret;
189         if (ret > 0)
190                 return 0;
191
192         vfs_check_frozen(sb, SB_FREEZE_WRITE);
193
194         nilfs = sb->s_fs_info;
195         down_read(&nilfs->ns_segctor_sem);
196         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
197                 up_read(&nilfs->ns_segctor_sem);
198                 ret = -ENOSPC;
199                 goto failed;
200         }
201         return 0;
202
203  failed:
204         ti = current->journal_info;
205         current->journal_info = ti->ti_save;
206         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
207                 kmem_cache_free(nilfs_transaction_cachep, ti);
208         return ret;
209 }
210
211 /**
212  * nilfs_transaction_commit - commit indivisible file operations.
213  * @sb: super block
214  *
215  * nilfs_transaction_commit() releases the read semaphore which is
216  * acquired by nilfs_transaction_begin(). This is only performed
217  * in outermost call of this function.  If a commit flag is set,
218  * nilfs_transaction_commit() sets a timer to start the segment
219  * constructor.  If a sync flag is set, it starts construction
220  * directly.
221  */
222 int nilfs_transaction_commit(struct super_block *sb)
223 {
224         struct nilfs_transaction_info *ti = current->journal_info;
225         struct the_nilfs *nilfs = sb->s_fs_info;
226         int err = 0;
227
228         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
229         ti->ti_flags |= NILFS_TI_COMMIT;
230         if (ti->ti_count > 0) {
231                 ti->ti_count--;
232                 return 0;
233         }
234         if (nilfs->ns_writer) {
235                 struct nilfs_sc_info *sci = nilfs->ns_writer;
236
237                 if (ti->ti_flags & NILFS_TI_COMMIT)
238                         nilfs_segctor_start_timer(sci);
239                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
240                         nilfs_segctor_do_flush(sci, 0);
241         }
242         up_read(&nilfs->ns_segctor_sem);
243         current->journal_info = ti->ti_save;
244
245         if (ti->ti_flags & NILFS_TI_SYNC)
246                 err = nilfs_construct_segment(sb);
247         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
248                 kmem_cache_free(nilfs_transaction_cachep, ti);
249         return err;
250 }
251
252 void nilfs_transaction_abort(struct super_block *sb)
253 {
254         struct nilfs_transaction_info *ti = current->journal_info;
255         struct the_nilfs *nilfs = sb->s_fs_info;
256
257         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
258         if (ti->ti_count > 0) {
259                 ti->ti_count--;
260                 return;
261         }
262         up_read(&nilfs->ns_segctor_sem);
263
264         current->journal_info = ti->ti_save;
265         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
266                 kmem_cache_free(nilfs_transaction_cachep, ti);
267 }
268
269 void nilfs_relax_pressure_in_lock(struct super_block *sb)
270 {
271         struct the_nilfs *nilfs = sb->s_fs_info;
272         struct nilfs_sc_info *sci = nilfs->ns_writer;
273
274         if (!sci || !sci->sc_flush_request)
275                 return;
276
277         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
278         up_read(&nilfs->ns_segctor_sem);
279
280         down_write(&nilfs->ns_segctor_sem);
281         if (sci->sc_flush_request &&
282             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
283                 struct nilfs_transaction_info *ti = current->journal_info;
284
285                 ti->ti_flags |= NILFS_TI_WRITER;
286                 nilfs_segctor_do_immediate_flush(sci);
287                 ti->ti_flags &= ~NILFS_TI_WRITER;
288         }
289         downgrade_write(&nilfs->ns_segctor_sem);
290 }
291
292 static void nilfs_transaction_lock(struct super_block *sb,
293                                    struct nilfs_transaction_info *ti,
294                                    int gcflag)
295 {
296         struct nilfs_transaction_info *cur_ti = current->journal_info;
297         struct the_nilfs *nilfs = sb->s_fs_info;
298         struct nilfs_sc_info *sci = nilfs->ns_writer;
299
300         WARN_ON(cur_ti);
301         ti->ti_flags = NILFS_TI_WRITER;
302         ti->ti_count = 0;
303         ti->ti_save = cur_ti;
304         ti->ti_magic = NILFS_TI_MAGIC;
305         INIT_LIST_HEAD(&ti->ti_garbage);
306         current->journal_info = ti;
307
308         for (;;) {
309                 down_write(&nilfs->ns_segctor_sem);
310                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
311                         break;
312
313                 nilfs_segctor_do_immediate_flush(sci);
314
315                 up_write(&nilfs->ns_segctor_sem);
316                 yield();
317         }
318         if (gcflag)
319                 ti->ti_flags |= NILFS_TI_GC;
320 }
321
322 static void nilfs_transaction_unlock(struct super_block *sb)
323 {
324         struct nilfs_transaction_info *ti = current->journal_info;
325         struct the_nilfs *nilfs = sb->s_fs_info;
326
327         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
328         BUG_ON(ti->ti_count > 0);
329
330         up_write(&nilfs->ns_segctor_sem);
331         current->journal_info = ti->ti_save;
332         if (!list_empty(&ti->ti_garbage))
333                 nilfs_dispose_list(nilfs, &ti->ti_garbage, 0);
334 }
335
336 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
337                                             struct nilfs_segsum_pointer *ssp,
338                                             unsigned bytes)
339 {
340         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
341         unsigned blocksize = sci->sc_super->s_blocksize;
342         void *p;
343
344         if (unlikely(ssp->offset + bytes > blocksize)) {
345                 ssp->offset = 0;
346                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
347                                                &segbuf->sb_segsum_buffers));
348                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
349         }
350         p = ssp->bh->b_data + ssp->offset;
351         ssp->offset += bytes;
352         return p;
353 }
354
355 /**
356  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
357  * @sci: nilfs_sc_info
358  */
359 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
360 {
361         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
362         struct buffer_head *sumbh;
363         unsigned sumbytes;
364         unsigned flags = 0;
365         int err;
366
367         if (nilfs_doing_gc())
368                 flags = NILFS_SS_GC;
369         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
370         if (unlikely(err))
371                 return err;
372
373         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
374         sumbytes = segbuf->sb_sum.sumbytes;
375         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
376         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
377         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
378         return 0;
379 }
380
381 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
382 {
383         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
384         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
385                 return -E2BIG; /* The current segment is filled up
386                                   (internal code) */
387         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
388         return nilfs_segctor_reset_segment_buffer(sci);
389 }
390
391 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
392 {
393         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
394         int err;
395
396         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
397                 err = nilfs_segctor_feed_segment(sci);
398                 if (err)
399                         return err;
400                 segbuf = sci->sc_curseg;
401         }
402         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
403         if (likely(!err))
404                 segbuf->sb_sum.flags |= NILFS_SS_SR;
405         return err;
406 }
407
408 /*
409  * Functions for making segment summary and payloads
410  */
411 static int nilfs_segctor_segsum_block_required(
412         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
413         unsigned binfo_size)
414 {
415         unsigned blocksize = sci->sc_super->s_blocksize;
416         /* Size of finfo and binfo is enough small against blocksize */
417
418         return ssp->offset + binfo_size +
419                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
420                 blocksize;
421 }
422
423 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
424                                       struct inode *inode)
425 {
426         sci->sc_curseg->sb_sum.nfinfo++;
427         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
428         nilfs_segctor_map_segsum_entry(
429                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
430
431         if (NILFS_I(inode)->i_root &&
432             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
433                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
434         /* skip finfo */
435 }
436
437 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
438                                     struct inode *inode)
439 {
440         struct nilfs_finfo *finfo;
441         struct nilfs_inode_info *ii;
442         struct nilfs_segment_buffer *segbuf;
443         __u64 cno;
444
445         if (sci->sc_blk_cnt == 0)
446                 return;
447
448         ii = NILFS_I(inode);
449
450         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
451                 cno = ii->i_cno;
452         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
453                 cno = 0;
454         else
455                 cno = sci->sc_cno;
456
457         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
458                                                  sizeof(*finfo));
459         finfo->fi_ino = cpu_to_le64(inode->i_ino);
460         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
461         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
462         finfo->fi_cno = cpu_to_le64(cno);
463
464         segbuf = sci->sc_curseg;
465         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
466                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
467         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
468         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
469 }
470
471 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
472                                         struct buffer_head *bh,
473                                         struct inode *inode,
474                                         unsigned binfo_size)
475 {
476         struct nilfs_segment_buffer *segbuf;
477         int required, err = 0;
478
479  retry:
480         segbuf = sci->sc_curseg;
481         required = nilfs_segctor_segsum_block_required(
482                 sci, &sci->sc_binfo_ptr, binfo_size);
483         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
484                 nilfs_segctor_end_finfo(sci, inode);
485                 err = nilfs_segctor_feed_segment(sci);
486                 if (err)
487                         return err;
488                 goto retry;
489         }
490         if (unlikely(required)) {
491                 err = nilfs_segbuf_extend_segsum(segbuf);
492                 if (unlikely(err))
493                         goto failed;
494         }
495         if (sci->sc_blk_cnt == 0)
496                 nilfs_segctor_begin_finfo(sci, inode);
497
498         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
499         /* Substitution to vblocknr is delayed until update_blocknr() */
500         nilfs_segbuf_add_file_buffer(segbuf, bh);
501         sci->sc_blk_cnt++;
502  failed:
503         return err;
504 }
505
506 /*
507  * Callback functions that enumerate, mark, and collect dirty blocks
508  */
509 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
510                                    struct buffer_head *bh, struct inode *inode)
511 {
512         int err;
513
514         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
515         if (err < 0)
516                 return err;
517
518         err = nilfs_segctor_add_file_block(sci, bh, inode,
519                                            sizeof(struct nilfs_binfo_v));
520         if (!err)
521                 sci->sc_datablk_cnt++;
522         return err;
523 }
524
525 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
526                                    struct buffer_head *bh,
527                                    struct inode *inode)
528 {
529         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
530 }
531
532 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
533                                    struct buffer_head *bh,
534                                    struct inode *inode)
535 {
536         WARN_ON(!buffer_dirty(bh));
537         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
538 }
539
540 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
541                                         struct nilfs_segsum_pointer *ssp,
542                                         union nilfs_binfo *binfo)
543 {
544         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
545                 sci, ssp, sizeof(*binfo_v));
546         *binfo_v = binfo->bi_v;
547 }
548
549 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
550                                         struct nilfs_segsum_pointer *ssp,
551                                         union nilfs_binfo *binfo)
552 {
553         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
554                 sci, ssp, sizeof(*vblocknr));
555         *vblocknr = binfo->bi_v.bi_vblocknr;
556 }
557
558 static struct nilfs_sc_operations nilfs_sc_file_ops = {
559         .collect_data = nilfs_collect_file_data,
560         .collect_node = nilfs_collect_file_node,
561         .collect_bmap = nilfs_collect_file_bmap,
562         .write_data_binfo = nilfs_write_file_data_binfo,
563         .write_node_binfo = nilfs_write_file_node_binfo,
564 };
565
566 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
567                                   struct buffer_head *bh, struct inode *inode)
568 {
569         int err;
570
571         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
572         if (err < 0)
573                 return err;
574
575         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
576         if (!err)
577                 sci->sc_datablk_cnt++;
578         return err;
579 }
580
581 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
582                                   struct buffer_head *bh, struct inode *inode)
583 {
584         WARN_ON(!buffer_dirty(bh));
585         return nilfs_segctor_add_file_block(sci, bh, inode,
586                                             sizeof(struct nilfs_binfo_dat));
587 }
588
589 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
590                                        struct nilfs_segsum_pointer *ssp,
591                                        union nilfs_binfo *binfo)
592 {
593         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
594                                                           sizeof(*blkoff));
595         *blkoff = binfo->bi_dat.bi_blkoff;
596 }
597
598 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
599                                        struct nilfs_segsum_pointer *ssp,
600                                        union nilfs_binfo *binfo)
601 {
602         struct nilfs_binfo_dat *binfo_dat =
603                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
604         *binfo_dat = binfo->bi_dat;
605 }
606
607 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
608         .collect_data = nilfs_collect_dat_data,
609         .collect_node = nilfs_collect_file_node,
610         .collect_bmap = nilfs_collect_dat_bmap,
611         .write_data_binfo = nilfs_write_dat_data_binfo,
612         .write_node_binfo = nilfs_write_dat_node_binfo,
613 };
614
615 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
616         .collect_data = nilfs_collect_file_data,
617         .collect_node = NULL,
618         .collect_bmap = NULL,
619         .write_data_binfo = nilfs_write_file_data_binfo,
620         .write_node_binfo = NULL,
621 };
622
623 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
624                                               struct list_head *listp,
625                                               size_t nlimit,
626                                               loff_t start, loff_t end)
627 {
628         struct address_space *mapping = inode->i_mapping;
629         struct pagevec pvec;
630         pgoff_t index = 0, last = ULONG_MAX;
631         size_t ndirties = 0;
632         int i;
633
634         if (unlikely(start != 0 || end != LLONG_MAX)) {
635                 /*
636                  * A valid range is given for sync-ing data pages. The
637                  * range is rounded to per-page; extra dirty buffers
638                  * may be included if blocksize < pagesize.
639                  */
640                 index = start >> PAGE_SHIFT;
641                 last = end >> PAGE_SHIFT;
642         }
643         pagevec_init(&pvec, 0);
644  repeat:
645         if (unlikely(index > last) ||
646             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
647                                 min_t(pgoff_t, last - index,
648                                       PAGEVEC_SIZE - 1) + 1))
649                 return ndirties;
650
651         for (i = 0; i < pagevec_count(&pvec); i++) {
652                 struct buffer_head *bh, *head;
653                 struct page *page = pvec.pages[i];
654
655                 if (unlikely(page->index > last))
656                         break;
657
658                 if (mapping->host) {
659                         lock_page(page);
660                         if (!page_has_buffers(page))
661                                 create_empty_buffers(page,
662                                                      1 << inode->i_blkbits, 0);
663                         unlock_page(page);
664                 }
665
666                 bh = head = page_buffers(page);
667                 do {
668                         if (!buffer_dirty(bh))
669                                 continue;
670                         get_bh(bh);
671                         list_add_tail(&bh->b_assoc_buffers, listp);
672                         ndirties++;
673                         if (unlikely(ndirties >= nlimit)) {
674                                 pagevec_release(&pvec);
675                                 cond_resched();
676                                 return ndirties;
677                         }
678                 } while (bh = bh->b_this_page, bh != head);
679         }
680         pagevec_release(&pvec);
681         cond_resched();
682         goto repeat;
683 }
684
685 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
686                                             struct list_head *listp)
687 {
688         struct nilfs_inode_info *ii = NILFS_I(inode);
689         struct address_space *mapping = &ii->i_btnode_cache;
690         struct pagevec pvec;
691         struct buffer_head *bh, *head;
692         unsigned int i;
693         pgoff_t index = 0;
694
695         pagevec_init(&pvec, 0);
696
697         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
698                                   PAGEVEC_SIZE)) {
699                 for (i = 0; i < pagevec_count(&pvec); i++) {
700                         bh = head = page_buffers(pvec.pages[i]);
701                         do {
702                                 if (buffer_dirty(bh)) {
703                                         get_bh(bh);
704                                         list_add_tail(&bh->b_assoc_buffers,
705                                                       listp);
706                                 }
707                                 bh = bh->b_this_page;
708                         } while (bh != head);
709                 }
710                 pagevec_release(&pvec);
711                 cond_resched();
712         }
713 }
714
715 static void nilfs_dispose_list(struct the_nilfs *nilfs,
716                                struct list_head *head, int force)
717 {
718         struct nilfs_inode_info *ii, *n;
719         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
720         unsigned nv = 0;
721
722         while (!list_empty(head)) {
723                 spin_lock(&nilfs->ns_inode_lock);
724                 list_for_each_entry_safe(ii, n, head, i_dirty) {
725                         list_del_init(&ii->i_dirty);
726                         if (force) {
727                                 if (unlikely(ii->i_bh)) {
728                                         brelse(ii->i_bh);
729                                         ii->i_bh = NULL;
730                                 }
731                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
732                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
733                                 list_add_tail(&ii->i_dirty,
734                                               &nilfs->ns_dirty_files);
735                                 continue;
736                         }
737                         ivec[nv++] = ii;
738                         if (nv == SC_N_INODEVEC)
739                                 break;
740                 }
741                 spin_unlock(&nilfs->ns_inode_lock);
742
743                 for (pii = ivec; nv > 0; pii++, nv--)
744                         iput(&(*pii)->vfs_inode);
745         }
746 }
747
748 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
749                                      struct nilfs_root *root)
750 {
751         int ret = 0;
752
753         if (nilfs_mdt_fetch_dirty(root->ifile))
754                 ret++;
755         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
756                 ret++;
757         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
758                 ret++;
759         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
760                 ret++;
761         return ret;
762 }
763
764 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
765 {
766         return list_empty(&sci->sc_dirty_files) &&
767                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
768                 sci->sc_nfreesegs == 0 &&
769                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
770 }
771
772 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
773 {
774         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
775         int ret = 0;
776
777         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
778                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
779
780         spin_lock(&nilfs->ns_inode_lock);
781         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
782                 ret++;
783
784         spin_unlock(&nilfs->ns_inode_lock);
785         return ret;
786 }
787
788 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
789 {
790         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
791
792         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
793         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
794         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
795         nilfs_mdt_clear_dirty(nilfs->ns_dat);
796 }
797
798 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
799 {
800         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
801         struct buffer_head *bh_cp;
802         struct nilfs_checkpoint *raw_cp;
803         int err;
804
805         /* XXX: this interface will be changed */
806         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
807                                           &raw_cp, &bh_cp);
808         if (likely(!err)) {
809                 /* The following code is duplicated with cpfile.  But, it is
810                    needed to collect the checkpoint even if it was not newly
811                    created */
812                 nilfs_mdt_mark_buffer_dirty(bh_cp);
813                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
814                 nilfs_cpfile_put_checkpoint(
815                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
816         } else
817                 WARN_ON(err == -EINVAL || err == -ENOENT);
818
819         return err;
820 }
821
822 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
823 {
824         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
825         struct buffer_head *bh_cp;
826         struct nilfs_checkpoint *raw_cp;
827         int err;
828
829         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
830                                           &raw_cp, &bh_cp);
831         if (unlikely(err)) {
832                 WARN_ON(err == -EINVAL || err == -ENOENT);
833                 goto failed_ibh;
834         }
835         raw_cp->cp_snapshot_list.ssl_next = 0;
836         raw_cp->cp_snapshot_list.ssl_prev = 0;
837         raw_cp->cp_inodes_count =
838                 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
839         raw_cp->cp_blocks_count =
840                 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
841         raw_cp->cp_nblk_inc =
842                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
843         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
844         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
845
846         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
847                 nilfs_checkpoint_clear_minor(raw_cp);
848         else
849                 nilfs_checkpoint_set_minor(raw_cp);
850
851         nilfs_write_inode_common(sci->sc_root->ifile,
852                                  &raw_cp->cp_ifile_inode, 1);
853         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
854         return 0;
855
856  failed_ibh:
857         return err;
858 }
859
860 static void nilfs_fill_in_file_bmap(struct inode *ifile,
861                                     struct nilfs_inode_info *ii)
862
863 {
864         struct buffer_head *ibh;
865         struct nilfs_inode *raw_inode;
866
867         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
868                 ibh = ii->i_bh;
869                 BUG_ON(!ibh);
870                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
871                                                   ibh);
872                 nilfs_bmap_write(ii->i_bmap, raw_inode);
873                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
874         }
875 }
876
877 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
878 {
879         struct nilfs_inode_info *ii;
880
881         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
882                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
883                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
884         }
885 }
886
887 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
888                                              struct the_nilfs *nilfs)
889 {
890         struct buffer_head *bh_sr;
891         struct nilfs_super_root *raw_sr;
892         unsigned isz, srsz;
893
894         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
895         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
896         isz = nilfs->ns_inode_size;
897         srsz = NILFS_SR_BYTES(isz);
898
899         raw_sr->sr_bytes = cpu_to_le16(srsz);
900         raw_sr->sr_nongc_ctime
901                 = cpu_to_le64(nilfs_doing_gc() ?
902                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
903         raw_sr->sr_flags = 0;
904
905         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
906                                  NILFS_SR_DAT_OFFSET(isz), 1);
907         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
908                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
909         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
910                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
911         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
912 }
913
914 static void nilfs_redirty_inodes(struct list_head *head)
915 {
916         struct nilfs_inode_info *ii;
917
918         list_for_each_entry(ii, head, i_dirty) {
919                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
920                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
921         }
922 }
923
924 static void nilfs_drop_collected_inodes(struct list_head *head)
925 {
926         struct nilfs_inode_info *ii;
927
928         list_for_each_entry(ii, head, i_dirty) {
929                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
930                         continue;
931
932                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
933                 set_bit(NILFS_I_UPDATED, &ii->i_state);
934         }
935 }
936
937 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
938                                        struct inode *inode,
939                                        struct list_head *listp,
940                                        int (*collect)(struct nilfs_sc_info *,
941                                                       struct buffer_head *,
942                                                       struct inode *))
943 {
944         struct buffer_head *bh, *n;
945         int err = 0;
946
947         if (collect) {
948                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
949                         list_del_init(&bh->b_assoc_buffers);
950                         err = collect(sci, bh, inode);
951                         brelse(bh);
952                         if (unlikely(err))
953                                 goto dispose_buffers;
954                 }
955                 return 0;
956         }
957
958  dispose_buffers:
959         while (!list_empty(listp)) {
960                 bh = list_first_entry(listp, struct buffer_head,
961                                       b_assoc_buffers);
962                 list_del_init(&bh->b_assoc_buffers);
963                 brelse(bh);
964         }
965         return err;
966 }
967
968 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
969 {
970         /* Remaining number of blocks within segment buffer */
971         return sci->sc_segbuf_nblocks -
972                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
973 }
974
975 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
976                                    struct inode *inode,
977                                    struct nilfs_sc_operations *sc_ops)
978 {
979         LIST_HEAD(data_buffers);
980         LIST_HEAD(node_buffers);
981         int err;
982
983         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
984                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
985
986                 n = nilfs_lookup_dirty_data_buffers(
987                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
988                 if (n > rest) {
989                         err = nilfs_segctor_apply_buffers(
990                                 sci, inode, &data_buffers,
991                                 sc_ops->collect_data);
992                         BUG_ON(!err); /* always receive -E2BIG or true error */
993                         goto break_or_fail;
994                 }
995         }
996         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
997
998         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
999                 err = nilfs_segctor_apply_buffers(
1000                         sci, inode, &data_buffers, sc_ops->collect_data);
1001                 if (unlikely(err)) {
1002                         /* dispose node list */
1003                         nilfs_segctor_apply_buffers(
1004                                 sci, inode, &node_buffers, NULL);
1005                         goto break_or_fail;
1006                 }
1007                 sci->sc_stage.flags |= NILFS_CF_NODE;
1008         }
1009         /* Collect node */
1010         err = nilfs_segctor_apply_buffers(
1011                 sci, inode, &node_buffers, sc_ops->collect_node);
1012         if (unlikely(err))
1013                 goto break_or_fail;
1014
1015         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1016         err = nilfs_segctor_apply_buffers(
1017                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1018         if (unlikely(err))
1019                 goto break_or_fail;
1020
1021         nilfs_segctor_end_finfo(sci, inode);
1022         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1023
1024  break_or_fail:
1025         return err;
1026 }
1027
1028 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1029                                          struct inode *inode)
1030 {
1031         LIST_HEAD(data_buffers);
1032         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1033         int err;
1034
1035         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1036                                             sci->sc_dsync_start,
1037                                             sci->sc_dsync_end);
1038
1039         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1040                                           nilfs_collect_file_data);
1041         if (!err) {
1042                 nilfs_segctor_end_finfo(sci, inode);
1043                 BUG_ON(n > rest);
1044                 /* always receive -E2BIG or true error if n > rest */
1045         }
1046         return err;
1047 }
1048
1049 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1050 {
1051         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1052         struct list_head *head;
1053         struct nilfs_inode_info *ii;
1054         size_t ndone;
1055         int err = 0;
1056
1057         switch (sci->sc_stage.scnt) {
1058         case NILFS_ST_INIT:
1059                 /* Pre-processes */
1060                 sci->sc_stage.flags = 0;
1061
1062                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1063                         sci->sc_nblk_inc = 0;
1064                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1065                         if (mode == SC_LSEG_DSYNC) {
1066                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1067                                 goto dsync_mode;
1068                         }
1069                 }
1070
1071                 sci->sc_stage.dirty_file_ptr = NULL;
1072                 sci->sc_stage.gc_inode_ptr = NULL;
1073                 if (mode == SC_FLUSH_DAT) {
1074                         sci->sc_stage.scnt = NILFS_ST_DAT;
1075                         goto dat_stage;
1076                 }
1077                 sci->sc_stage.scnt++;  /* Fall through */
1078         case NILFS_ST_GC:
1079                 if (nilfs_doing_gc()) {
1080                         head = &sci->sc_gc_inodes;
1081                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1082                                                 head, i_dirty);
1083                         list_for_each_entry_continue(ii, head, i_dirty) {
1084                                 err = nilfs_segctor_scan_file(
1085                                         sci, &ii->vfs_inode,
1086                                         &nilfs_sc_file_ops);
1087                                 if (unlikely(err)) {
1088                                         sci->sc_stage.gc_inode_ptr = list_entry(
1089                                                 ii->i_dirty.prev,
1090                                                 struct nilfs_inode_info,
1091                                                 i_dirty);
1092                                         goto break_or_fail;
1093                                 }
1094                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1095                         }
1096                         sci->sc_stage.gc_inode_ptr = NULL;
1097                 }
1098                 sci->sc_stage.scnt++;  /* Fall through */
1099         case NILFS_ST_FILE:
1100                 head = &sci->sc_dirty_files;
1101                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1102                                         i_dirty);
1103                 list_for_each_entry_continue(ii, head, i_dirty) {
1104                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1105
1106                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1107                                                       &nilfs_sc_file_ops);
1108                         if (unlikely(err)) {
1109                                 sci->sc_stage.dirty_file_ptr =
1110                                         list_entry(ii->i_dirty.prev,
1111                                                    struct nilfs_inode_info,
1112                                                    i_dirty);
1113                                 goto break_or_fail;
1114                         }
1115                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1116                         /* XXX: required ? */
1117                 }
1118                 sci->sc_stage.dirty_file_ptr = NULL;
1119                 if (mode == SC_FLUSH_FILE) {
1120                         sci->sc_stage.scnt = NILFS_ST_DONE;
1121                         return 0;
1122                 }
1123                 sci->sc_stage.scnt++;
1124                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1125                 /* Fall through */
1126         case NILFS_ST_IFILE:
1127                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1128                                               &nilfs_sc_file_ops);
1129                 if (unlikely(err))
1130                         break;
1131                 sci->sc_stage.scnt++;
1132                 /* Creating a checkpoint */
1133                 err = nilfs_segctor_create_checkpoint(sci);
1134                 if (unlikely(err))
1135                         break;
1136                 /* Fall through */
1137         case NILFS_ST_CPFILE:
1138                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1139                                               &nilfs_sc_file_ops);
1140                 if (unlikely(err))
1141                         break;
1142                 sci->sc_stage.scnt++;  /* Fall through */
1143         case NILFS_ST_SUFILE:
1144                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1145                                          sci->sc_nfreesegs, &ndone);
1146                 if (unlikely(err)) {
1147                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1148                                                   sci->sc_freesegs, ndone,
1149                                                   NULL);
1150                         break;
1151                 }
1152                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1153
1154                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1155                                               &nilfs_sc_file_ops);
1156                 if (unlikely(err))
1157                         break;
1158                 sci->sc_stage.scnt++;  /* Fall through */
1159         case NILFS_ST_DAT:
1160  dat_stage:
1161                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1162                                               &nilfs_sc_dat_ops);
1163                 if (unlikely(err))
1164                         break;
1165                 if (mode == SC_FLUSH_DAT) {
1166                         sci->sc_stage.scnt = NILFS_ST_DONE;
1167                         return 0;
1168                 }
1169                 sci->sc_stage.scnt++;  /* Fall through */
1170         case NILFS_ST_SR:
1171                 if (mode == SC_LSEG_SR) {
1172                         /* Appending a super root */
1173                         err = nilfs_segctor_add_super_root(sci);
1174                         if (unlikely(err))
1175                                 break;
1176                 }
1177                 /* End of a logical segment */
1178                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1179                 sci->sc_stage.scnt = NILFS_ST_DONE;
1180                 return 0;
1181         case NILFS_ST_DSYNC:
1182  dsync_mode:
1183                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1184                 ii = sci->sc_dsync_inode;
1185                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1186                         break;
1187
1188                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1189                 if (unlikely(err))
1190                         break;
1191                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1192                 sci->sc_stage.scnt = NILFS_ST_DONE;
1193                 return 0;
1194         case NILFS_ST_DONE:
1195                 return 0;
1196         default:
1197                 BUG();
1198         }
1199
1200  break_or_fail:
1201         return err;
1202 }
1203
1204 /**
1205  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1206  * @sci: nilfs_sc_info
1207  * @nilfs: nilfs object
1208  */
1209 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1210                                             struct the_nilfs *nilfs)
1211 {
1212         struct nilfs_segment_buffer *segbuf, *prev;
1213         __u64 nextnum;
1214         int err, alloc = 0;
1215
1216         segbuf = nilfs_segbuf_new(sci->sc_super);
1217         if (unlikely(!segbuf))
1218                 return -ENOMEM;
1219
1220         if (list_empty(&sci->sc_write_logs)) {
1221                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1222                                  nilfs->ns_pseg_offset, nilfs);
1223                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1224                         nilfs_shift_to_next_segment(nilfs);
1225                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1226                 }
1227
1228                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1229                 nextnum = nilfs->ns_nextnum;
1230
1231                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1232                         /* Start from the head of a new full segment */
1233                         alloc++;
1234         } else {
1235                 /* Continue logs */
1236                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1237                 nilfs_segbuf_map_cont(segbuf, prev);
1238                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1239                 nextnum = prev->sb_nextnum;
1240
1241                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1242                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1243                         segbuf->sb_sum.seg_seq++;
1244                         alloc++;
1245                 }
1246         }
1247
1248         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1249         if (err)
1250                 goto failed;
1251
1252         if (alloc) {
1253                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1254                 if (err)
1255                         goto failed;
1256         }
1257         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1258
1259         BUG_ON(!list_empty(&sci->sc_segbufs));
1260         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1261         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1262         return 0;
1263
1264  failed:
1265         nilfs_segbuf_free(segbuf);
1266         return err;
1267 }
1268
1269 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1270                                          struct the_nilfs *nilfs, int nadd)
1271 {
1272         struct nilfs_segment_buffer *segbuf, *prev;
1273         struct inode *sufile = nilfs->ns_sufile;
1274         __u64 nextnextnum;
1275         LIST_HEAD(list);
1276         int err, ret, i;
1277
1278         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1279         /*
1280          * Since the segment specified with nextnum might be allocated during
1281          * the previous construction, the buffer including its segusage may
1282          * not be dirty.  The following call ensures that the buffer is dirty
1283          * and will pin the buffer on memory until the sufile is written.
1284          */
1285         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1286         if (unlikely(err))
1287                 return err;
1288
1289         for (i = 0; i < nadd; i++) {
1290                 /* extend segment info */
1291                 err = -ENOMEM;
1292                 segbuf = nilfs_segbuf_new(sci->sc_super);
1293                 if (unlikely(!segbuf))
1294                         goto failed;
1295
1296                 /* map this buffer to region of segment on-disk */
1297                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1298                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1299
1300                 /* allocate the next next full segment */
1301                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1302                 if (unlikely(err))
1303                         goto failed_segbuf;
1304
1305                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1306                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1307
1308                 list_add_tail(&segbuf->sb_list, &list);
1309                 prev = segbuf;
1310         }
1311         list_splice_tail(&list, &sci->sc_segbufs);
1312         return 0;
1313
1314  failed_segbuf:
1315         nilfs_segbuf_free(segbuf);
1316  failed:
1317         list_for_each_entry(segbuf, &list, sb_list) {
1318                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1319                 WARN_ON(ret); /* never fails */
1320         }
1321         nilfs_destroy_logs(&list);
1322         return err;
1323 }
1324
1325 static void nilfs_free_incomplete_logs(struct list_head *logs,
1326                                        struct the_nilfs *nilfs)
1327 {
1328         struct nilfs_segment_buffer *segbuf, *prev;
1329         struct inode *sufile = nilfs->ns_sufile;
1330         int ret;
1331
1332         segbuf = NILFS_FIRST_SEGBUF(logs);
1333         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1334                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1335                 WARN_ON(ret); /* never fails */
1336         }
1337         if (atomic_read(&segbuf->sb_err)) {
1338                 /* Case 1: The first segment failed */
1339                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1340                         /* Case 1a:  Partial segment appended into an existing
1341                            segment */
1342                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1343                                                 segbuf->sb_fseg_end);
1344                 else /* Case 1b:  New full segment */
1345                         set_nilfs_discontinued(nilfs);
1346         }
1347
1348         prev = segbuf;
1349         list_for_each_entry_continue(segbuf, logs, sb_list) {
1350                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1351                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1352                         WARN_ON(ret); /* never fails */
1353                 }
1354                 if (atomic_read(&segbuf->sb_err) &&
1355                     segbuf->sb_segnum != nilfs->ns_nextnum)
1356                         /* Case 2: extended segment (!= next) failed */
1357                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1358                 prev = segbuf;
1359         }
1360 }
1361
1362 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1363                                           struct inode *sufile)
1364 {
1365         struct nilfs_segment_buffer *segbuf;
1366         unsigned long live_blocks;
1367         int ret;
1368
1369         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1370                 live_blocks = segbuf->sb_sum.nblocks +
1371                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1372                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1373                                                      live_blocks,
1374                                                      sci->sc_seg_ctime);
1375                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1376         }
1377 }
1378
1379 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1380 {
1381         struct nilfs_segment_buffer *segbuf;
1382         int ret;
1383
1384         segbuf = NILFS_FIRST_SEGBUF(logs);
1385         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1386                                              segbuf->sb_pseg_start -
1387                                              segbuf->sb_fseg_start, 0);
1388         WARN_ON(ret); /* always succeed because the segusage is dirty */
1389
1390         list_for_each_entry_continue(segbuf, logs, sb_list) {
1391                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1392                                                      0, 0);
1393                 WARN_ON(ret); /* always succeed */
1394         }
1395 }
1396
1397 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1398                                             struct nilfs_segment_buffer *last,
1399                                             struct inode *sufile)
1400 {
1401         struct nilfs_segment_buffer *segbuf = last;
1402         int ret;
1403
1404         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1405                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1406                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1407                 WARN_ON(ret);
1408         }
1409         nilfs_truncate_logs(&sci->sc_segbufs, last);
1410 }
1411
1412
1413 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1414                                  struct the_nilfs *nilfs, int mode)
1415 {
1416         struct nilfs_cstage prev_stage = sci->sc_stage;
1417         int err, nadd = 1;
1418
1419         /* Collection retry loop */
1420         for (;;) {
1421                 sci->sc_nblk_this_inc = 0;
1422                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1423
1424                 err = nilfs_segctor_reset_segment_buffer(sci);
1425                 if (unlikely(err))
1426                         goto failed;
1427
1428                 err = nilfs_segctor_collect_blocks(sci, mode);
1429                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1430                 if (!err)
1431                         break;
1432
1433                 if (unlikely(err != -E2BIG))
1434                         goto failed;
1435
1436                 /* The current segment is filled up */
1437                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1438                         break;
1439
1440                 nilfs_clear_logs(&sci->sc_segbufs);
1441
1442                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1443                 if (unlikely(err))
1444                         return err;
1445
1446                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1447                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1448                                                         sci->sc_freesegs,
1449                                                         sci->sc_nfreesegs,
1450                                                         NULL);
1451                         WARN_ON(err); /* do not happen */
1452                 }
1453                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1454                 sci->sc_stage = prev_stage;
1455         }
1456         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1457         return 0;
1458
1459  failed:
1460         return err;
1461 }
1462
1463 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1464                                       struct buffer_head *new_bh)
1465 {
1466         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1467
1468         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1469         /* The caller must release old_bh */
1470 }
1471
1472 static int
1473 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1474                                      struct nilfs_segment_buffer *segbuf,
1475                                      int mode)
1476 {
1477         struct inode *inode = NULL;
1478         sector_t blocknr;
1479         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1480         unsigned long nblocks = 0, ndatablk = 0;
1481         struct nilfs_sc_operations *sc_op = NULL;
1482         struct nilfs_segsum_pointer ssp;
1483         struct nilfs_finfo *finfo = NULL;
1484         union nilfs_binfo binfo;
1485         struct buffer_head *bh, *bh_org;
1486         ino_t ino = 0;
1487         int err = 0;
1488
1489         if (!nfinfo)
1490                 goto out;
1491
1492         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1493         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1494         ssp.offset = sizeof(struct nilfs_segment_summary);
1495
1496         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1497                 if (bh == segbuf->sb_super_root)
1498                         break;
1499                 if (!finfo) {
1500                         finfo = nilfs_segctor_map_segsum_entry(
1501                                 sci, &ssp, sizeof(*finfo));
1502                         ino = le64_to_cpu(finfo->fi_ino);
1503                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1504                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1505
1506                         if (buffer_nilfs_node(bh))
1507                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1508                         else
1509                                 inode = NILFS_AS_I(bh->b_page->mapping);
1510
1511                         if (mode == SC_LSEG_DSYNC)
1512                                 sc_op = &nilfs_sc_dsync_ops;
1513                         else if (ino == NILFS_DAT_INO)
1514                                 sc_op = &nilfs_sc_dat_ops;
1515                         else /* file blocks */
1516                                 sc_op = &nilfs_sc_file_ops;
1517                 }
1518                 bh_org = bh;
1519                 get_bh(bh_org);
1520                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1521                                         &binfo);
1522                 if (bh != bh_org)
1523                         nilfs_list_replace_buffer(bh_org, bh);
1524                 brelse(bh_org);
1525                 if (unlikely(err))
1526                         goto failed_bmap;
1527
1528                 if (ndatablk > 0)
1529                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1530                 else
1531                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1532
1533                 blocknr++;
1534                 if (--nblocks == 0) {
1535                         finfo = NULL;
1536                         if (--nfinfo == 0)
1537                                 break;
1538                 } else if (ndatablk > 0)
1539                         ndatablk--;
1540         }
1541  out:
1542         return 0;
1543
1544  failed_bmap:
1545         return err;
1546 }
1547
1548 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1549 {
1550         struct nilfs_segment_buffer *segbuf;
1551         int err;
1552
1553         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1554                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1555                 if (unlikely(err))
1556                         return err;
1557                 nilfs_segbuf_fill_in_segsum(segbuf);
1558         }
1559         return 0;
1560 }
1561
1562 static void nilfs_begin_page_io(struct page *page)
1563 {
1564         if (!page || PageWriteback(page))
1565                 /* For split b-tree node pages, this function may be called
1566                    twice.  We ignore the 2nd or later calls by this check. */
1567                 return;
1568
1569         lock_page(page);
1570         clear_page_dirty_for_io(page);
1571         set_page_writeback(page);
1572         unlock_page(page);
1573 }
1574
1575 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1576 {
1577         struct nilfs_segment_buffer *segbuf;
1578         struct page *bd_page = NULL, *fs_page = NULL;
1579
1580         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1581                 struct buffer_head *bh;
1582
1583                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1584                                     b_assoc_buffers) {
1585                         if (bh->b_page != bd_page) {
1586                                 if (bd_page) {
1587                                         lock_page(bd_page);
1588                                         clear_page_dirty_for_io(bd_page);
1589                                         set_page_writeback(bd_page);
1590                                         unlock_page(bd_page);
1591                                 }
1592                                 bd_page = bh->b_page;
1593                         }
1594                 }
1595
1596                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1597                                     b_assoc_buffers) {
1598                         if (bh == segbuf->sb_super_root) {
1599                                 if (bh->b_page != bd_page) {
1600                                         lock_page(bd_page);
1601                                         clear_page_dirty_for_io(bd_page);
1602                                         set_page_writeback(bd_page);
1603                                         unlock_page(bd_page);
1604                                         bd_page = bh->b_page;
1605                                 }
1606                                 break;
1607                         }
1608                         if (bh->b_page != fs_page) {
1609                                 nilfs_begin_page_io(fs_page);
1610                                 fs_page = bh->b_page;
1611                         }
1612                 }
1613         }
1614         if (bd_page) {
1615                 lock_page(bd_page);
1616                 clear_page_dirty_for_io(bd_page);
1617                 set_page_writeback(bd_page);
1618                 unlock_page(bd_page);
1619         }
1620         nilfs_begin_page_io(fs_page);
1621 }
1622
1623 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1624                                struct the_nilfs *nilfs)
1625 {
1626         int ret;
1627
1628         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1629         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1630         return ret;
1631 }
1632
1633 static void nilfs_end_page_io(struct page *page, int err)
1634 {
1635         if (!page)
1636                 return;
1637
1638         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1639                 /*
1640                  * For b-tree node pages, this function may be called twice
1641                  * or more because they might be split in a segment.
1642                  */
1643                 if (PageDirty(page)) {
1644                         /*
1645                          * For pages holding split b-tree node buffers, dirty
1646                          * flag on the buffers may be cleared discretely.
1647                          * In that case, the page is once redirtied for
1648                          * remaining buffers, and it must be cancelled if
1649                          * all the buffers get cleaned later.
1650                          */
1651                         lock_page(page);
1652                         if (nilfs_page_buffers_clean(page))
1653                                 __nilfs_clear_page_dirty(page);
1654                         unlock_page(page);
1655                 }
1656                 return;
1657         }
1658
1659         if (!err) {
1660                 if (!nilfs_page_buffers_clean(page))
1661                         __set_page_dirty_nobuffers(page);
1662                 ClearPageError(page);
1663         } else {
1664                 __set_page_dirty_nobuffers(page);
1665                 SetPageError(page);
1666         }
1667
1668         end_page_writeback(page);
1669 }
1670
1671 static void nilfs_abort_logs(struct list_head *logs, int err)
1672 {
1673         struct nilfs_segment_buffer *segbuf;
1674         struct page *bd_page = NULL, *fs_page = NULL;
1675         struct buffer_head *bh;
1676
1677         if (list_empty(logs))
1678                 return;
1679
1680         list_for_each_entry(segbuf, logs, sb_list) {
1681                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1682                                     b_assoc_buffers) {
1683                         if (bh->b_page != bd_page) {
1684                                 if (bd_page)
1685                                         end_page_writeback(bd_page);
1686                                 bd_page = bh->b_page;
1687                         }
1688                 }
1689
1690                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1691                                     b_assoc_buffers) {
1692                         if (bh == segbuf->sb_super_root) {
1693                                 if (bh->b_page != bd_page) {
1694                                         end_page_writeback(bd_page);
1695                                         bd_page = bh->b_page;
1696                                 }
1697                                 break;
1698                         }
1699                         if (bh->b_page != fs_page) {
1700                                 nilfs_end_page_io(fs_page, err);
1701                                 fs_page = bh->b_page;
1702                         }
1703                 }
1704         }
1705         if (bd_page)
1706                 end_page_writeback(bd_page);
1707
1708         nilfs_end_page_io(fs_page, err);
1709 }
1710
1711 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1712                                              struct the_nilfs *nilfs, int err)
1713 {
1714         LIST_HEAD(logs);
1715         int ret;
1716
1717         list_splice_tail_init(&sci->sc_write_logs, &logs);
1718         ret = nilfs_wait_on_logs(&logs);
1719         nilfs_abort_logs(&logs, ret ? : err);
1720
1721         list_splice_tail_init(&sci->sc_segbufs, &logs);
1722         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1723         nilfs_free_incomplete_logs(&logs, nilfs);
1724
1725         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1726                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1727                                                 sci->sc_freesegs,
1728                                                 sci->sc_nfreesegs,
1729                                                 NULL);
1730                 WARN_ON(ret); /* do not happen */
1731         }
1732
1733         nilfs_destroy_logs(&logs);
1734 }
1735
1736 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1737                                    struct nilfs_segment_buffer *segbuf)
1738 {
1739         nilfs->ns_segnum = segbuf->sb_segnum;
1740         nilfs->ns_nextnum = segbuf->sb_nextnum;
1741         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1742                 + segbuf->sb_sum.nblocks;
1743         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1744         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1745 }
1746
1747 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1748 {
1749         struct nilfs_segment_buffer *segbuf;
1750         struct page *bd_page = NULL, *fs_page = NULL;
1751         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1752         int update_sr = false;
1753
1754         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1755                 struct buffer_head *bh;
1756
1757                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1758                                     b_assoc_buffers) {
1759                         set_buffer_uptodate(bh);
1760                         clear_buffer_dirty(bh);
1761                         if (bh->b_page != bd_page) {
1762                                 if (bd_page)
1763                                         end_page_writeback(bd_page);
1764                                 bd_page = bh->b_page;
1765                         }
1766                 }
1767                 /*
1768                  * We assume that the buffers which belong to the same page
1769                  * continue over the buffer list.
1770                  * Under this assumption, the last BHs of pages is
1771                  * identifiable by the discontinuity of bh->b_page
1772                  * (page != fs_page).
1773                  *
1774                  * For B-tree node blocks, however, this assumption is not
1775                  * guaranteed.  The cleanup code of B-tree node pages needs
1776                  * special care.
1777                  */
1778                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1779                                     b_assoc_buffers) {
1780                         set_buffer_uptodate(bh);
1781                         clear_buffer_dirty(bh);
1782                         clear_buffer_delay(bh);
1783                         clear_buffer_nilfs_volatile(bh);
1784                         clear_buffer_nilfs_redirected(bh);
1785                         if (bh == segbuf->sb_super_root) {
1786                                 if (bh->b_page != bd_page) {
1787                                         end_page_writeback(bd_page);
1788                                         bd_page = bh->b_page;
1789                                 }
1790                                 update_sr = true;
1791                                 break;
1792                         }
1793                         if (bh->b_page != fs_page) {
1794                                 nilfs_end_page_io(fs_page, 0);
1795                                 fs_page = bh->b_page;
1796                         }
1797                 }
1798
1799                 if (!nilfs_segbuf_simplex(segbuf)) {
1800                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1801                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1802                                 sci->sc_lseg_stime = jiffies;
1803                         }
1804                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1805                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1806                 }
1807         }
1808         /*
1809          * Since pages may continue over multiple segment buffers,
1810          * end of the last page must be checked outside of the loop.
1811          */
1812         if (bd_page)
1813                 end_page_writeback(bd_page);
1814
1815         nilfs_end_page_io(fs_page, 0);
1816
1817         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1818
1819         if (nilfs_doing_gc())
1820                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1821         else
1822                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1823
1824         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1825
1826         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1827         nilfs_set_next_segment(nilfs, segbuf);
1828
1829         if (update_sr) {
1830                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1831                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1832
1833                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1834                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1835                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1836                 nilfs_segctor_clear_metadata_dirty(sci);
1837         } else
1838                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1839 }
1840
1841 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1842 {
1843         int ret;
1844
1845         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1846         if (!ret) {
1847                 nilfs_segctor_complete_write(sci);
1848                 nilfs_destroy_logs(&sci->sc_write_logs);
1849         }
1850         return ret;
1851 }
1852
1853 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1854                                              struct the_nilfs *nilfs)
1855 {
1856         struct nilfs_inode_info *ii, *n;
1857         struct inode *ifile = sci->sc_root->ifile;
1858
1859         spin_lock(&nilfs->ns_inode_lock);
1860  retry:
1861         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1862                 if (!ii->i_bh) {
1863                         struct buffer_head *ibh;
1864                         int err;
1865
1866                         spin_unlock(&nilfs->ns_inode_lock);
1867                         err = nilfs_ifile_get_inode_block(
1868                                 ifile, ii->vfs_inode.i_ino, &ibh);
1869                         if (unlikely(err)) {
1870                                 nilfs_warning(sci->sc_super, __func__,
1871                                               "failed to get inode block.\n");
1872                                 return err;
1873                         }
1874                         nilfs_mdt_mark_buffer_dirty(ibh);
1875                         nilfs_mdt_mark_dirty(ifile);
1876                         spin_lock(&nilfs->ns_inode_lock);
1877                         if (likely(!ii->i_bh))
1878                                 ii->i_bh = ibh;
1879                         else
1880                                 brelse(ibh);
1881                         goto retry;
1882                 }
1883
1884                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1885                 set_bit(NILFS_I_BUSY, &ii->i_state);
1886                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1887         }
1888         spin_unlock(&nilfs->ns_inode_lock);
1889
1890         return 0;
1891 }
1892
1893 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1894                                              struct the_nilfs *nilfs)
1895 {
1896         struct nilfs_transaction_info *ti = current->journal_info;
1897         struct nilfs_inode_info *ii, *n;
1898
1899         spin_lock(&nilfs->ns_inode_lock);
1900         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1901                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1902                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1903                         continue;
1904
1905                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1906                 brelse(ii->i_bh);
1907                 ii->i_bh = NULL;
1908                 list_move_tail(&ii->i_dirty, &ti->ti_garbage);
1909         }
1910         spin_unlock(&nilfs->ns_inode_lock);
1911 }
1912
1913 /*
1914  * Main procedure of segment constructor
1915  */
1916 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
1917 {
1918         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1919         int err;
1920
1921         sci->sc_stage.scnt = NILFS_ST_INIT;
1922         sci->sc_cno = nilfs->ns_cno;
1923
1924         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
1925         if (unlikely(err))
1926                 goto out;
1927
1928         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
1929                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1930
1931         if (nilfs_segctor_clean(sci))
1932                 goto out;
1933
1934         do {
1935                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
1936
1937                 err = nilfs_segctor_begin_construction(sci, nilfs);
1938                 if (unlikely(err))
1939                         goto out;
1940
1941                 /* Update time stamp */
1942                 sci->sc_seg_ctime = get_seconds();
1943
1944                 err = nilfs_segctor_collect(sci, nilfs, mode);
1945                 if (unlikely(err))
1946                         goto failed;
1947
1948                 /* Avoid empty segment */
1949                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
1950                     nilfs_segbuf_empty(sci->sc_curseg)) {
1951                         nilfs_segctor_abort_construction(sci, nilfs, 1);
1952                         goto out;
1953                 }
1954
1955                 err = nilfs_segctor_assign(sci, mode);
1956                 if (unlikely(err))
1957                         goto failed;
1958
1959                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
1960                         nilfs_segctor_fill_in_file_bmap(sci);
1961
1962                 if (mode == SC_LSEG_SR &&
1963                     sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
1964                         err = nilfs_segctor_fill_in_checkpoint(sci);
1965                         if (unlikely(err))
1966                                 goto failed_to_write;
1967
1968                         nilfs_segctor_fill_in_super_root(sci, nilfs);
1969                 }
1970                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
1971
1972                 /* Write partial segments */
1973                 nilfs_segctor_prepare_write(sci);
1974
1975                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
1976                                             nilfs->ns_crc_seed);
1977
1978                 err = nilfs_segctor_write(sci, nilfs);
1979                 if (unlikely(err))
1980                         goto failed_to_write;
1981
1982                 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
1983                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
1984                         /*
1985                          * At this point, we avoid double buffering
1986                          * for blocksize < pagesize because page dirty
1987                          * flag is turned off during write and dirty
1988                          * buffers are not properly collected for
1989                          * pages crossing over segments.
1990                          */
1991                         err = nilfs_segctor_wait(sci);
1992                         if (err)
1993                                 goto failed_to_write;
1994                 }
1995         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
1996
1997  out:
1998         nilfs_segctor_drop_written_files(sci, nilfs);
1999         return err;
2000
2001  failed_to_write:
2002         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2003                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2004
2005  failed:
2006         if (nilfs_doing_gc())
2007                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2008         nilfs_segctor_abort_construction(sci, nilfs, err);
2009         goto out;
2010 }
2011
2012 /**
2013  * nilfs_segctor_start_timer - set timer of background write
2014  * @sci: nilfs_sc_info
2015  *
2016  * If the timer has already been set, it ignores the new request.
2017  * This function MUST be called within a section locking the segment
2018  * semaphore.
2019  */
2020 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2021 {
2022         spin_lock(&sci->sc_state_lock);
2023         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2024                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2025                 add_timer(&sci->sc_timer);
2026                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2027         }
2028         spin_unlock(&sci->sc_state_lock);
2029 }
2030
2031 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2032 {
2033         spin_lock(&sci->sc_state_lock);
2034         if (!(sci->sc_flush_request & (1 << bn))) {
2035                 unsigned long prev_req = sci->sc_flush_request;
2036
2037                 sci->sc_flush_request |= (1 << bn);
2038                 if (!prev_req)
2039                         wake_up(&sci->sc_wait_daemon);
2040         }
2041         spin_unlock(&sci->sc_state_lock);
2042 }
2043
2044 /**
2045  * nilfs_flush_segment - trigger a segment construction for resource control
2046  * @sb: super block
2047  * @ino: inode number of the file to be flushed out.
2048  */
2049 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2050 {
2051         struct the_nilfs *nilfs = sb->s_fs_info;
2052         struct nilfs_sc_info *sci = nilfs->ns_writer;
2053
2054         if (!sci || nilfs_doing_construction())
2055                 return;
2056         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2057                                         /* assign bit 0 to data files */
2058 }
2059
2060 struct nilfs_segctor_wait_request {
2061         wait_queue_t    wq;
2062         __u32           seq;
2063         int             err;
2064         atomic_t        done;
2065 };
2066
2067 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2068 {
2069         struct nilfs_segctor_wait_request wait_req;
2070         int err = 0;
2071
2072         spin_lock(&sci->sc_state_lock);
2073         init_wait(&wait_req.wq);
2074         wait_req.err = 0;
2075         atomic_set(&wait_req.done, 0);
2076         wait_req.seq = ++sci->sc_seq_request;
2077         spin_unlock(&sci->sc_state_lock);
2078
2079         init_waitqueue_entry(&wait_req.wq, current);
2080         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2081         set_current_state(TASK_INTERRUPTIBLE);
2082         wake_up(&sci->sc_wait_daemon);
2083
2084         for (;;) {
2085                 if (atomic_read(&wait_req.done)) {
2086                         err = wait_req.err;
2087                         break;
2088                 }
2089                 if (!signal_pending(current)) {
2090                         schedule();
2091                         continue;
2092                 }
2093                 err = -ERESTARTSYS;
2094                 break;
2095         }
2096         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2097         return err;
2098 }
2099
2100 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2101 {
2102         struct nilfs_segctor_wait_request *wrq, *n;
2103         unsigned long flags;
2104
2105         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2106         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2107                                  wq.task_list) {
2108                 if (!atomic_read(&wrq->done) &&
2109                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2110                         wrq->err = err;
2111                         atomic_set(&wrq->done, 1);
2112                 }
2113                 if (atomic_read(&wrq->done)) {
2114                         wrq->wq.func(&wrq->wq,
2115                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2116                                      0, NULL);
2117                 }
2118         }
2119         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2120 }
2121
2122 /**
2123  * nilfs_construct_segment - construct a logical segment
2124  * @sb: super block
2125  *
2126  * Return Value: On success, 0 is retured. On errors, one of the following
2127  * negative error code is returned.
2128  *
2129  * %-EROFS - Read only filesystem.
2130  *
2131  * %-EIO - I/O error
2132  *
2133  * %-ENOSPC - No space left on device (only in a panic state).
2134  *
2135  * %-ERESTARTSYS - Interrupted.
2136  *
2137  * %-ENOMEM - Insufficient memory available.
2138  */
2139 int nilfs_construct_segment(struct super_block *sb)
2140 {
2141         struct the_nilfs *nilfs = sb->s_fs_info;
2142         struct nilfs_sc_info *sci = nilfs->ns_writer;
2143         struct nilfs_transaction_info *ti;
2144         int err;
2145
2146         if (!sci)
2147                 return -EROFS;
2148
2149         /* A call inside transactions causes a deadlock. */
2150         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2151
2152         err = nilfs_segctor_sync(sci);
2153         return err;
2154 }
2155
2156 /**
2157  * nilfs_construct_dsync_segment - construct a data-only logical segment
2158  * @sb: super block
2159  * @inode: inode whose data blocks should be written out
2160  * @start: start byte offset
2161  * @end: end byte offset (inclusive)
2162  *
2163  * Return Value: On success, 0 is retured. On errors, one of the following
2164  * negative error code is returned.
2165  *
2166  * %-EROFS - Read only filesystem.
2167  *
2168  * %-EIO - I/O error
2169  *
2170  * %-ENOSPC - No space left on device (only in a panic state).
2171  *
2172  * %-ERESTARTSYS - Interrupted.
2173  *
2174  * %-ENOMEM - Insufficient memory available.
2175  */
2176 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2177                                   loff_t start, loff_t end)
2178 {
2179         struct the_nilfs *nilfs = sb->s_fs_info;
2180         struct nilfs_sc_info *sci = nilfs->ns_writer;
2181         struct nilfs_inode_info *ii;
2182         struct nilfs_transaction_info ti;
2183         int err = 0;
2184
2185         if (!sci)
2186                 return -EROFS;
2187
2188         nilfs_transaction_lock(sb, &ti, 0);
2189
2190         ii = NILFS_I(inode);
2191         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2192             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2193             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2194             nilfs_discontinued(nilfs)) {
2195                 nilfs_transaction_unlock(sb);
2196                 err = nilfs_segctor_sync(sci);
2197                 return err;
2198         }
2199
2200         spin_lock(&nilfs->ns_inode_lock);
2201         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2202             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2203                 spin_unlock(&nilfs->ns_inode_lock);
2204                 nilfs_transaction_unlock(sb);
2205                 return 0;
2206         }
2207         spin_unlock(&nilfs->ns_inode_lock);
2208         sci->sc_dsync_inode = ii;
2209         sci->sc_dsync_start = start;
2210         sci->sc_dsync_end = end;
2211
2212         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2213
2214         nilfs_transaction_unlock(sb);
2215         return err;
2216 }
2217
2218 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2219 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2220
2221 /**
2222  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2223  * @sci: segment constructor object
2224  */
2225 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2226 {
2227         spin_lock(&sci->sc_state_lock);
2228         sci->sc_seq_accepted = sci->sc_seq_request;
2229         spin_unlock(&sci->sc_state_lock);
2230         del_timer_sync(&sci->sc_timer);
2231 }
2232
2233 /**
2234  * nilfs_segctor_notify - notify the result of request to caller threads
2235  * @sci: segment constructor object
2236  * @mode: mode of log forming
2237  * @err: error code to be notified
2238  */
2239 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2240 {
2241         /* Clear requests (even when the construction failed) */
2242         spin_lock(&sci->sc_state_lock);
2243
2244         if (mode == SC_LSEG_SR) {
2245                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2246                 sci->sc_seq_done = sci->sc_seq_accepted;
2247                 nilfs_segctor_wakeup(sci, err);
2248                 sci->sc_flush_request = 0;
2249         } else {
2250                 if (mode == SC_FLUSH_FILE)
2251                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2252                 else if (mode == SC_FLUSH_DAT)
2253                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2254
2255                 /* re-enable timer if checkpoint creation was not done */
2256                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2257                     time_before(jiffies, sci->sc_timer.expires))
2258                         add_timer(&sci->sc_timer);
2259         }
2260         spin_unlock(&sci->sc_state_lock);
2261 }
2262
2263 /**
2264  * nilfs_segctor_construct - form logs and write them to disk
2265  * @sci: segment constructor object
2266  * @mode: mode of log forming
2267  */
2268 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2269 {
2270         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2271         struct nilfs_super_block **sbp;
2272         int err = 0;
2273
2274         nilfs_segctor_accept(sci);
2275
2276         if (nilfs_discontinued(nilfs))
2277                 mode = SC_LSEG_SR;
2278         if (!nilfs_segctor_confirm(sci))
2279                 err = nilfs_segctor_do_construct(sci, mode);
2280
2281         if (likely(!err)) {
2282                 if (mode != SC_FLUSH_DAT)
2283                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2284                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2285                     nilfs_discontinued(nilfs)) {
2286                         down_write(&nilfs->ns_sem);
2287                         err = -EIO;
2288                         sbp = nilfs_prepare_super(sci->sc_super,
2289                                                   nilfs_sb_will_flip(nilfs));
2290                         if (likely(sbp)) {
2291                                 nilfs_set_log_cursor(sbp[0], nilfs);
2292                                 err = nilfs_commit_super(sci->sc_super,
2293                                                          NILFS_SB_COMMIT);
2294                         }
2295                         up_write(&nilfs->ns_sem);
2296                 }
2297         }
2298
2299         nilfs_segctor_notify(sci, mode, err);
2300         return err;
2301 }
2302
2303 static void nilfs_construction_timeout(unsigned long data)
2304 {
2305         struct task_struct *p = (struct task_struct *)data;
2306         wake_up_process(p);
2307 }
2308
2309 static void
2310 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2311 {
2312         struct nilfs_inode_info *ii, *n;
2313
2314         list_for_each_entry_safe(ii, n, head, i_dirty) {
2315                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2316                         continue;
2317                 list_del_init(&ii->i_dirty);
2318                 iput(&ii->vfs_inode);
2319         }
2320 }
2321
2322 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2323                          void **kbufs)
2324 {
2325         struct the_nilfs *nilfs = sb->s_fs_info;
2326         struct nilfs_sc_info *sci = nilfs->ns_writer;
2327         struct nilfs_transaction_info ti;
2328         int err;
2329
2330         if (unlikely(!sci))
2331                 return -EROFS;
2332
2333         nilfs_transaction_lock(sb, &ti, 1);
2334
2335         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2336         if (unlikely(err))
2337                 goto out_unlock;
2338
2339         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2340         if (unlikely(err)) {
2341                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2342                 goto out_unlock;
2343         }
2344
2345         sci->sc_freesegs = kbufs[4];
2346         sci->sc_nfreesegs = argv[4].v_nmembs;
2347         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2348
2349         for (;;) {
2350                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2351                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2352
2353                 if (likely(!err))
2354                         break;
2355
2356                 nilfs_warning(sb, __func__,
2357                               "segment construction failed. (err=%d)", err);
2358                 set_current_state(TASK_INTERRUPTIBLE);
2359                 schedule_timeout(sci->sc_interval);
2360         }
2361         if (nilfs_test_opt(nilfs, DISCARD)) {
2362                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2363                                                  sci->sc_nfreesegs);
2364                 if (ret) {
2365                         printk(KERN_WARNING
2366                                "NILFS warning: error %d on discard request, "
2367                                "turning discards off for the device\n", ret);
2368                         nilfs_clear_opt(nilfs, DISCARD);
2369                 }
2370         }
2371
2372  out_unlock:
2373         sci->sc_freesegs = NULL;
2374         sci->sc_nfreesegs = 0;
2375         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2376         nilfs_transaction_unlock(sb);
2377         return err;
2378 }
2379
2380 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2381 {
2382         struct nilfs_transaction_info ti;
2383
2384         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2385         nilfs_segctor_construct(sci, mode);
2386
2387         /*
2388          * Unclosed segment should be retried.  We do this using sc_timer.
2389          * Timeout of sc_timer will invoke complete construction which leads
2390          * to close the current logical segment.
2391          */
2392         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2393                 nilfs_segctor_start_timer(sci);
2394
2395         nilfs_transaction_unlock(sci->sc_super);
2396 }
2397
2398 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2399 {
2400         int mode = 0;
2401         int err;
2402
2403         spin_lock(&sci->sc_state_lock);
2404         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2405                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2406         spin_unlock(&sci->sc_state_lock);
2407
2408         if (mode) {
2409                 err = nilfs_segctor_do_construct(sci, mode);
2410
2411                 spin_lock(&sci->sc_state_lock);
2412                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2413                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2414                 spin_unlock(&sci->sc_state_lock);
2415         }
2416         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2417 }
2418
2419 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2420 {
2421         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2422             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2423                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2424                         return SC_FLUSH_FILE;
2425                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2426                         return SC_FLUSH_DAT;
2427         }
2428         return SC_LSEG_SR;
2429 }
2430
2431 /**
2432  * nilfs_segctor_thread - main loop of the segment constructor thread.
2433  * @arg: pointer to a struct nilfs_sc_info.
2434  *
2435  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2436  * to execute segment constructions.
2437  */
2438 static int nilfs_segctor_thread(void *arg)
2439 {
2440         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2441         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2442         int timeout = 0;
2443
2444         sci->sc_timer.data = (unsigned long)current;
2445         sci->sc_timer.function = nilfs_construction_timeout;
2446
2447         /* start sync. */
2448         sci->sc_task = current;
2449         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2450         printk(KERN_INFO
2451                "segctord starting. Construction interval = %lu seconds, "
2452                "CP frequency < %lu seconds\n",
2453                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2454
2455         spin_lock(&sci->sc_state_lock);
2456  loop:
2457         for (;;) {
2458                 int mode;
2459
2460                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2461                         goto end_thread;
2462
2463                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2464                         mode = SC_LSEG_SR;
2465                 else if (!sci->sc_flush_request)
2466                         break;
2467                 else
2468                         mode = nilfs_segctor_flush_mode(sci);
2469
2470                 spin_unlock(&sci->sc_state_lock);
2471                 nilfs_segctor_thread_construct(sci, mode);
2472                 spin_lock(&sci->sc_state_lock);
2473                 timeout = 0;
2474         }
2475
2476
2477         if (freezing(current)) {
2478                 spin_unlock(&sci->sc_state_lock);
2479                 refrigerator();
2480                 spin_lock(&sci->sc_state_lock);
2481         } else {
2482                 DEFINE_WAIT(wait);
2483                 int should_sleep = 1;
2484
2485                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2486                                 TASK_INTERRUPTIBLE);
2487
2488                 if (sci->sc_seq_request != sci->sc_seq_done)
2489                         should_sleep = 0;
2490                 else if (sci->sc_flush_request)
2491                         should_sleep = 0;
2492                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2493                         should_sleep = time_before(jiffies,
2494                                         sci->sc_timer.expires);
2495
2496                 if (should_sleep) {
2497                         spin_unlock(&sci->sc_state_lock);
2498                         schedule();
2499                         spin_lock(&sci->sc_state_lock);
2500                 }
2501                 finish_wait(&sci->sc_wait_daemon, &wait);
2502                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2503                            time_after_eq(jiffies, sci->sc_timer.expires));
2504
2505                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2506                         set_nilfs_discontinued(nilfs);
2507         }
2508         goto loop;
2509
2510  end_thread:
2511         spin_unlock(&sci->sc_state_lock);
2512
2513         /* end sync. */
2514         sci->sc_task = NULL;
2515         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2516         return 0;
2517 }
2518
2519 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2520 {
2521         struct task_struct *t;
2522
2523         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2524         if (IS_ERR(t)) {
2525                 int err = PTR_ERR(t);
2526
2527                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2528                        err);
2529                 return err;
2530         }
2531         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2532         return 0;
2533 }
2534
2535 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2536         __acquires(&sci->sc_state_lock)
2537         __releases(&sci->sc_state_lock)
2538 {
2539         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2540
2541         while (sci->sc_task) {
2542                 wake_up(&sci->sc_wait_daemon);
2543                 spin_unlock(&sci->sc_state_lock);
2544                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2545                 spin_lock(&sci->sc_state_lock);
2546         }
2547 }
2548
2549 /*
2550  * Setup & clean-up functions
2551  */
2552 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2553                                                struct nilfs_root *root)
2554 {
2555         struct the_nilfs *nilfs = sb->s_fs_info;
2556         struct nilfs_sc_info *sci;
2557
2558         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2559         if (!sci)
2560                 return NULL;
2561
2562         sci->sc_super = sb;
2563
2564         nilfs_get_root(root);
2565         sci->sc_root = root;
2566
2567         init_waitqueue_head(&sci->sc_wait_request);
2568         init_waitqueue_head(&sci->sc_wait_daemon);
2569         init_waitqueue_head(&sci->sc_wait_task);
2570         spin_lock_init(&sci->sc_state_lock);
2571         INIT_LIST_HEAD(&sci->sc_dirty_files);
2572         INIT_LIST_HEAD(&sci->sc_segbufs);
2573         INIT_LIST_HEAD(&sci->sc_write_logs);
2574         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2575         init_timer(&sci->sc_timer);
2576
2577         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2578         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2579         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2580
2581         if (nilfs->ns_interval)
2582                 sci->sc_interval = nilfs->ns_interval;
2583         if (nilfs->ns_watermark)
2584                 sci->sc_watermark = nilfs->ns_watermark;
2585         return sci;
2586 }
2587
2588 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2589 {
2590         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2591
2592         /* The segctord thread was stopped and its timer was removed.
2593            But some tasks remain. */
2594         do {
2595                 struct nilfs_transaction_info ti;
2596
2597                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2598                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2599                 nilfs_transaction_unlock(sci->sc_super);
2600
2601         } while (ret && retrycount-- > 0);
2602 }
2603
2604 /**
2605  * nilfs_segctor_destroy - destroy the segment constructor.
2606  * @sci: nilfs_sc_info
2607  *
2608  * nilfs_segctor_destroy() kills the segctord thread and frees
2609  * the nilfs_sc_info struct.
2610  * Caller must hold the segment semaphore.
2611  */
2612 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2613 {
2614         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2615         int flag;
2616
2617         up_write(&nilfs->ns_segctor_sem);
2618
2619         spin_lock(&sci->sc_state_lock);
2620         nilfs_segctor_kill_thread(sci);
2621         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2622                 || sci->sc_seq_request != sci->sc_seq_done);
2623         spin_unlock(&sci->sc_state_lock);
2624
2625         if (flag || !nilfs_segctor_confirm(sci))
2626                 nilfs_segctor_write_out(sci);
2627
2628         if (!list_empty(&sci->sc_dirty_files)) {
2629                 nilfs_warning(sci->sc_super, __func__,
2630                               "dirty file(s) after the final construction\n");
2631                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2632         }
2633
2634         WARN_ON(!list_empty(&sci->sc_segbufs));
2635         WARN_ON(!list_empty(&sci->sc_write_logs));
2636
2637         nilfs_put_root(sci->sc_root);
2638
2639         down_write(&nilfs->ns_segctor_sem);
2640
2641         del_timer_sync(&sci->sc_timer);
2642         kfree(sci);
2643 }
2644
2645 /**
2646  * nilfs_attach_log_writer - attach log writer
2647  * @sb: super block instance
2648  * @root: root object of the current filesystem tree
2649  *
2650  * This allocates a log writer object, initializes it, and starts the
2651  * log writer.
2652  *
2653  * Return Value: On success, 0 is returned. On error, one of the following
2654  * negative error code is returned.
2655  *
2656  * %-ENOMEM - Insufficient memory available.
2657  */
2658 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2659 {
2660         struct the_nilfs *nilfs = sb->s_fs_info;
2661         int err;
2662
2663         if (nilfs->ns_writer) {
2664                 /*
2665                  * This happens if the filesystem was remounted
2666                  * read/write after nilfs_error degenerated it into a
2667                  * read-only mount.
2668                  */
2669                 nilfs_detach_log_writer(sb);
2670         }
2671
2672         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2673         if (!nilfs->ns_writer)
2674                 return -ENOMEM;
2675
2676         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2677         if (err) {
2678                 kfree(nilfs->ns_writer);
2679                 nilfs->ns_writer = NULL;
2680         }
2681         return err;
2682 }
2683
2684 /**
2685  * nilfs_detach_log_writer - destroy log writer
2686  * @sb: super block instance
2687  *
2688  * This kills log writer daemon, frees the log writer object, and
2689  * destroys list of dirty files.
2690  */
2691 void nilfs_detach_log_writer(struct super_block *sb)
2692 {
2693         struct the_nilfs *nilfs = sb->s_fs_info;
2694         LIST_HEAD(garbage_list);
2695
2696         down_write(&nilfs->ns_segctor_sem);
2697         if (nilfs->ns_writer) {
2698                 nilfs_segctor_destroy(nilfs->ns_writer);
2699                 nilfs->ns_writer = NULL;
2700         }
2701
2702         /* Force to free the list of dirty files */
2703         spin_lock(&nilfs->ns_inode_lock);
2704         if (!list_empty(&nilfs->ns_dirty_files)) {
2705                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2706                 nilfs_warning(sb, __func__,
2707                               "Hit dirty file after stopped log writer\n");
2708         }
2709         spin_unlock(&nilfs->ns_inode_lock);
2710         up_write(&nilfs->ns_segctor_sem);
2711
2712         nilfs_dispose_list(nilfs, &garbage_list, 1);
2713 }