nilfs2: move mount options to nilfs object
[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 nilfs_sb_info *, struct list_head *,
108                                int);
109
110 #define nilfs_cnt32_gt(a, b)   \
111         (typecheck(__u32, a) && typecheck(__u32, b) && \
112          ((__s32)(b) - (__s32)(a) < 0))
113 #define nilfs_cnt32_ge(a, b)   \
114         (typecheck(__u32, a) && typecheck(__u32, b) && \
115          ((__s32)(a) - (__s32)(b) >= 0))
116 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
117 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
118
119 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
120 {
121         struct nilfs_transaction_info *cur_ti = current->journal_info;
122         void *save = NULL;
123
124         if (cur_ti) {
125                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
126                         return ++cur_ti->ti_count;
127                 else {
128                         /*
129                          * If journal_info field is occupied by other FS,
130                          * it is saved and will be restored on
131                          * nilfs_transaction_commit().
132                          */
133                         printk(KERN_WARNING
134                                "NILFS warning: journal info from a different "
135                                "FS\n");
136                         save = current->journal_info;
137                 }
138         }
139         if (!ti) {
140                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
141                 if (!ti)
142                         return -ENOMEM;
143                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
144         } else {
145                 ti->ti_flags = 0;
146         }
147         ti->ti_count = 0;
148         ti->ti_save = save;
149         ti->ti_magic = NILFS_TI_MAGIC;
150         current->journal_info = ti;
151         return 0;
152 }
153
154 /**
155  * nilfs_transaction_begin - start indivisible file operations.
156  * @sb: super block
157  * @ti: nilfs_transaction_info
158  * @vacancy_check: flags for vacancy rate checks
159  *
160  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
161  * the segment semaphore, to make a segment construction and write tasks
162  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
163  * The region enclosed by these two functions can be nested.  To avoid a
164  * deadlock, the semaphore is only acquired or released in the outermost call.
165  *
166  * This function allocates a nilfs_transaction_info struct to keep context
167  * information on it.  It is initialized and hooked onto the current task in
168  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
169  * instead; otherwise a new struct is assigned from a slab.
170  *
171  * When @vacancy_check flag is set, this function will check the amount of
172  * free space, and will wait for the GC to reclaim disk space if low capacity.
173  *
174  * Return Value: On success, 0 is returned. On error, one of the following
175  * negative error code is returned.
176  *
177  * %-ENOMEM - Insufficient memory available.
178  *
179  * %-ENOSPC - No space left on device
180  */
181 int nilfs_transaction_begin(struct super_block *sb,
182                             struct nilfs_transaction_info *ti,
183                             int vacancy_check)
184 {
185         struct nilfs_sb_info *sbi;
186         struct the_nilfs *nilfs;
187         int ret = nilfs_prepare_segment_lock(ti);
188
189         if (unlikely(ret < 0))
190                 return ret;
191         if (ret > 0)
192                 return 0;
193
194         vfs_check_frozen(sb, SB_FREEZE_WRITE);
195
196         sbi = NILFS_SB(sb);
197         nilfs = sbi->s_nilfs;
198         down_read(&nilfs->ns_segctor_sem);
199         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
200                 up_read(&nilfs->ns_segctor_sem);
201                 ret = -ENOSPC;
202                 goto failed;
203         }
204         return 0;
205
206  failed:
207         ti = current->journal_info;
208         current->journal_info = ti->ti_save;
209         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
210                 kmem_cache_free(nilfs_transaction_cachep, ti);
211         return ret;
212 }
213
214 /**
215  * nilfs_transaction_commit - commit indivisible file operations.
216  * @sb: super block
217  *
218  * nilfs_transaction_commit() releases the read semaphore which is
219  * acquired by nilfs_transaction_begin(). This is only performed
220  * in outermost call of this function.  If a commit flag is set,
221  * nilfs_transaction_commit() sets a timer to start the segment
222  * constructor.  If a sync flag is set, it starts construction
223  * directly.
224  */
225 int nilfs_transaction_commit(struct super_block *sb)
226 {
227         struct nilfs_transaction_info *ti = current->journal_info;
228         struct nilfs_sb_info *sbi;
229         struct nilfs_sc_info *sci;
230         int err = 0;
231
232         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
233         ti->ti_flags |= NILFS_TI_COMMIT;
234         if (ti->ti_count > 0) {
235                 ti->ti_count--;
236                 return 0;
237         }
238         sbi = NILFS_SB(sb);
239         sci = NILFS_SC(sbi);
240         if (sci != NULL) {
241                 if (ti->ti_flags & NILFS_TI_COMMIT)
242                         nilfs_segctor_start_timer(sci);
243                 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
244                     sci->sc_watermark)
245                         nilfs_segctor_do_flush(sci, 0);
246         }
247         up_read(&sbi->s_nilfs->ns_segctor_sem);
248         current->journal_info = ti->ti_save;
249
250         if (ti->ti_flags & NILFS_TI_SYNC)
251                 err = nilfs_construct_segment(sb);
252         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
253                 kmem_cache_free(nilfs_transaction_cachep, ti);
254         return err;
255 }
256
257 void nilfs_transaction_abort(struct super_block *sb)
258 {
259         struct nilfs_transaction_info *ti = current->journal_info;
260
261         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
262         if (ti->ti_count > 0) {
263                 ti->ti_count--;
264                 return;
265         }
266         up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
267
268         current->journal_info = ti->ti_save;
269         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
270                 kmem_cache_free(nilfs_transaction_cachep, ti);
271 }
272
273 void nilfs_relax_pressure_in_lock(struct super_block *sb)
274 {
275         struct nilfs_sb_info *sbi = NILFS_SB(sb);
276         struct nilfs_sc_info *sci = NILFS_SC(sbi);
277         struct the_nilfs *nilfs = sbi->s_nilfs;
278
279         if (!sci || !sci->sc_flush_request)
280                 return;
281
282         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
283         up_read(&nilfs->ns_segctor_sem);
284
285         down_write(&nilfs->ns_segctor_sem);
286         if (sci->sc_flush_request &&
287             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
288                 struct nilfs_transaction_info *ti = current->journal_info;
289
290                 ti->ti_flags |= NILFS_TI_WRITER;
291                 nilfs_segctor_do_immediate_flush(sci);
292                 ti->ti_flags &= ~NILFS_TI_WRITER;
293         }
294         downgrade_write(&nilfs->ns_segctor_sem);
295 }
296
297 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
298                                    struct nilfs_transaction_info *ti,
299                                    int gcflag)
300 {
301         struct nilfs_transaction_info *cur_ti = current->journal_info;
302
303         WARN_ON(cur_ti);
304         ti->ti_flags = NILFS_TI_WRITER;
305         ti->ti_count = 0;
306         ti->ti_save = cur_ti;
307         ti->ti_magic = NILFS_TI_MAGIC;
308         INIT_LIST_HEAD(&ti->ti_garbage);
309         current->journal_info = ti;
310
311         for (;;) {
312                 down_write(&sbi->s_nilfs->ns_segctor_sem);
313                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
314                         break;
315
316                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
317
318                 up_write(&sbi->s_nilfs->ns_segctor_sem);
319                 yield();
320         }
321         if (gcflag)
322                 ti->ti_flags |= NILFS_TI_GC;
323 }
324
325 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
326 {
327         struct nilfs_transaction_info *ti = current->journal_info;
328
329         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
330         BUG_ON(ti->ti_count > 0);
331
332         up_write(&sbi->s_nilfs->ns_segctor_sem);
333         current->journal_info = ti->ti_save;
334         if (!list_empty(&ti->ti_garbage))
335                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
336 }
337
338 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
339                                             struct nilfs_segsum_pointer *ssp,
340                                             unsigned bytes)
341 {
342         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
343         unsigned blocksize = sci->sc_super->s_blocksize;
344         void *p;
345
346         if (unlikely(ssp->offset + bytes > blocksize)) {
347                 ssp->offset = 0;
348                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
349                                                &segbuf->sb_segsum_buffers));
350                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
351         }
352         p = ssp->bh->b_data + ssp->offset;
353         ssp->offset += bytes;
354         return p;
355 }
356
357 /**
358  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
359  * @sci: nilfs_sc_info
360  */
361 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
362 {
363         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
364         struct buffer_head *sumbh;
365         unsigned sumbytes;
366         unsigned flags = 0;
367         int err;
368
369         if (nilfs_doing_gc())
370                 flags = NILFS_SS_GC;
371         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
372         if (unlikely(err))
373                 return err;
374
375         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
376         sumbytes = segbuf->sb_sum.sumbytes;
377         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
378         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
379         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
380         return 0;
381 }
382
383 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
384 {
385         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
386         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
387                 return -E2BIG; /* The current segment is filled up
388                                   (internal code) */
389         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
390         return nilfs_segctor_reset_segment_buffer(sci);
391 }
392
393 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
394 {
395         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
396         int err;
397
398         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
399                 err = nilfs_segctor_feed_segment(sci);
400                 if (err)
401                         return err;
402                 segbuf = sci->sc_curseg;
403         }
404         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
405         if (likely(!err))
406                 segbuf->sb_sum.flags |= NILFS_SS_SR;
407         return err;
408 }
409
410 /*
411  * Functions for making segment summary and payloads
412  */
413 static int nilfs_segctor_segsum_block_required(
414         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
415         unsigned binfo_size)
416 {
417         unsigned blocksize = sci->sc_super->s_blocksize;
418         /* Size of finfo and binfo is enough small against blocksize */
419
420         return ssp->offset + binfo_size +
421                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
422                 blocksize;
423 }
424
425 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
426                                       struct inode *inode)
427 {
428         sci->sc_curseg->sb_sum.nfinfo++;
429         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
430         nilfs_segctor_map_segsum_entry(
431                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
432
433         if (NILFS_I(inode)->i_root &&
434             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
435                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
436         /* skip finfo */
437 }
438
439 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
440                                     struct inode *inode)
441 {
442         struct nilfs_finfo *finfo;
443         struct nilfs_inode_info *ii;
444         struct nilfs_segment_buffer *segbuf;
445         __u64 cno;
446
447         if (sci->sc_blk_cnt == 0)
448                 return;
449
450         ii = NILFS_I(inode);
451
452         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
453                 cno = ii->i_cno;
454         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
455                 cno = 0;
456         else
457                 cno = sci->sc_cno;
458
459         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
460                                                  sizeof(*finfo));
461         finfo->fi_ino = cpu_to_le64(inode->i_ino);
462         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
463         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
464         finfo->fi_cno = cpu_to_le64(cno);
465
466         segbuf = sci->sc_curseg;
467         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
468                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
469         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
470         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
471 }
472
473 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
474                                         struct buffer_head *bh,
475                                         struct inode *inode,
476                                         unsigned binfo_size)
477 {
478         struct nilfs_segment_buffer *segbuf;
479         int required, err = 0;
480
481  retry:
482         segbuf = sci->sc_curseg;
483         required = nilfs_segctor_segsum_block_required(
484                 sci, &sci->sc_binfo_ptr, binfo_size);
485         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
486                 nilfs_segctor_end_finfo(sci, inode);
487                 err = nilfs_segctor_feed_segment(sci);
488                 if (err)
489                         return err;
490                 goto retry;
491         }
492         if (unlikely(required)) {
493                 err = nilfs_segbuf_extend_segsum(segbuf);
494                 if (unlikely(err))
495                         goto failed;
496         }
497         if (sci->sc_blk_cnt == 0)
498                 nilfs_segctor_begin_finfo(sci, inode);
499
500         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
501         /* Substitution to vblocknr is delayed until update_blocknr() */
502         nilfs_segbuf_add_file_buffer(segbuf, bh);
503         sci->sc_blk_cnt++;
504  failed:
505         return err;
506 }
507
508 /*
509  * Callback functions that enumerate, mark, and collect dirty blocks
510  */
511 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
512                                    struct buffer_head *bh, struct inode *inode)
513 {
514         int err;
515
516         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
517         if (err < 0)
518                 return err;
519
520         err = nilfs_segctor_add_file_block(sci, bh, inode,
521                                            sizeof(struct nilfs_binfo_v));
522         if (!err)
523                 sci->sc_datablk_cnt++;
524         return err;
525 }
526
527 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
528                                    struct buffer_head *bh,
529                                    struct inode *inode)
530 {
531         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
532 }
533
534 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
535                                    struct buffer_head *bh,
536                                    struct inode *inode)
537 {
538         WARN_ON(!buffer_dirty(bh));
539         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
540 }
541
542 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
543                                         struct nilfs_segsum_pointer *ssp,
544                                         union nilfs_binfo *binfo)
545 {
546         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
547                 sci, ssp, sizeof(*binfo_v));
548         *binfo_v = binfo->bi_v;
549 }
550
551 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
552                                         struct nilfs_segsum_pointer *ssp,
553                                         union nilfs_binfo *binfo)
554 {
555         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
556                 sci, ssp, sizeof(*vblocknr));
557         *vblocknr = binfo->bi_v.bi_vblocknr;
558 }
559
560 static struct nilfs_sc_operations nilfs_sc_file_ops = {
561         .collect_data = nilfs_collect_file_data,
562         .collect_node = nilfs_collect_file_node,
563         .collect_bmap = nilfs_collect_file_bmap,
564         .write_data_binfo = nilfs_write_file_data_binfo,
565         .write_node_binfo = nilfs_write_file_node_binfo,
566 };
567
568 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
569                                   struct buffer_head *bh, struct inode *inode)
570 {
571         int err;
572
573         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
574         if (err < 0)
575                 return err;
576
577         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
578         if (!err)
579                 sci->sc_datablk_cnt++;
580         return err;
581 }
582
583 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
584                                   struct buffer_head *bh, struct inode *inode)
585 {
586         WARN_ON(!buffer_dirty(bh));
587         return nilfs_segctor_add_file_block(sci, bh, inode,
588                                             sizeof(struct nilfs_binfo_dat));
589 }
590
591 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
592                                        struct nilfs_segsum_pointer *ssp,
593                                        union nilfs_binfo *binfo)
594 {
595         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
596                                                           sizeof(*blkoff));
597         *blkoff = binfo->bi_dat.bi_blkoff;
598 }
599
600 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
601                                        struct nilfs_segsum_pointer *ssp,
602                                        union nilfs_binfo *binfo)
603 {
604         struct nilfs_binfo_dat *binfo_dat =
605                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
606         *binfo_dat = binfo->bi_dat;
607 }
608
609 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
610         .collect_data = nilfs_collect_dat_data,
611         .collect_node = nilfs_collect_file_node,
612         .collect_bmap = nilfs_collect_dat_bmap,
613         .write_data_binfo = nilfs_write_dat_data_binfo,
614         .write_node_binfo = nilfs_write_dat_node_binfo,
615 };
616
617 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
618         .collect_data = nilfs_collect_file_data,
619         .collect_node = NULL,
620         .collect_bmap = NULL,
621         .write_data_binfo = nilfs_write_file_data_binfo,
622         .write_node_binfo = NULL,
623 };
624
625 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
626                                               struct list_head *listp,
627                                               size_t nlimit,
628                                               loff_t start, loff_t end)
629 {
630         struct address_space *mapping = inode->i_mapping;
631         struct pagevec pvec;
632         pgoff_t index = 0, last = ULONG_MAX;
633         size_t ndirties = 0;
634         int i;
635
636         if (unlikely(start != 0 || end != LLONG_MAX)) {
637                 /*
638                  * A valid range is given for sync-ing data pages. The
639                  * range is rounded to per-page; extra dirty buffers
640                  * may be included if blocksize < pagesize.
641                  */
642                 index = start >> PAGE_SHIFT;
643                 last = end >> PAGE_SHIFT;
644         }
645         pagevec_init(&pvec, 0);
646  repeat:
647         if (unlikely(index > last) ||
648             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
649                                 min_t(pgoff_t, last - index,
650                                       PAGEVEC_SIZE - 1) + 1))
651                 return ndirties;
652
653         for (i = 0; i < pagevec_count(&pvec); i++) {
654                 struct buffer_head *bh, *head;
655                 struct page *page = pvec.pages[i];
656
657                 if (unlikely(page->index > last))
658                         break;
659
660                 if (mapping->host) {
661                         lock_page(page);
662                         if (!page_has_buffers(page))
663                                 create_empty_buffers(page,
664                                                      1 << inode->i_blkbits, 0);
665                         unlock_page(page);
666                 }
667
668                 bh = head = page_buffers(page);
669                 do {
670                         if (!buffer_dirty(bh))
671                                 continue;
672                         get_bh(bh);
673                         list_add_tail(&bh->b_assoc_buffers, listp);
674                         ndirties++;
675                         if (unlikely(ndirties >= nlimit)) {
676                                 pagevec_release(&pvec);
677                                 cond_resched();
678                                 return ndirties;
679                         }
680                 } while (bh = bh->b_this_page, bh != head);
681         }
682         pagevec_release(&pvec);
683         cond_resched();
684         goto repeat;
685 }
686
687 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
688                                             struct list_head *listp)
689 {
690         struct nilfs_inode_info *ii = NILFS_I(inode);
691         struct address_space *mapping = &ii->i_btnode_cache;
692         struct pagevec pvec;
693         struct buffer_head *bh, *head;
694         unsigned int i;
695         pgoff_t index = 0;
696
697         pagevec_init(&pvec, 0);
698
699         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
700                                   PAGEVEC_SIZE)) {
701                 for (i = 0; i < pagevec_count(&pvec); i++) {
702                         bh = head = page_buffers(pvec.pages[i]);
703                         do {
704                                 if (buffer_dirty(bh)) {
705                                         get_bh(bh);
706                                         list_add_tail(&bh->b_assoc_buffers,
707                                                       listp);
708                                 }
709                                 bh = bh->b_this_page;
710                         } while (bh != head);
711                 }
712                 pagevec_release(&pvec);
713                 cond_resched();
714         }
715 }
716
717 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
718                                struct list_head *head, int force)
719 {
720         struct nilfs_inode_info *ii, *n;
721         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
722         unsigned nv = 0;
723
724         while (!list_empty(head)) {
725                 spin_lock(&sbi->s_inode_lock);
726                 list_for_each_entry_safe(ii, n, head, i_dirty) {
727                         list_del_init(&ii->i_dirty);
728                         if (force) {
729                                 if (unlikely(ii->i_bh)) {
730                                         brelse(ii->i_bh);
731                                         ii->i_bh = NULL;
732                                 }
733                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
734                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
735                                 list_add_tail(&ii->i_dirty,
736                                               &sbi->s_dirty_files);
737                                 continue;
738                         }
739                         ivec[nv++] = ii;
740                         if (nv == SC_N_INODEVEC)
741                                 break;
742                 }
743                 spin_unlock(&sbi->s_inode_lock);
744
745                 for (pii = ivec; nv > 0; pii++, nv--)
746                         iput(&(*pii)->vfs_inode);
747         }
748 }
749
750 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
751                                      struct nilfs_root *root)
752 {
753         int ret = 0;
754
755         if (nilfs_mdt_fetch_dirty(root->ifile))
756                 ret++;
757         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
758                 ret++;
759         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
760                 ret++;
761         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
762                 ret++;
763         return ret;
764 }
765
766 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
767 {
768         return list_empty(&sci->sc_dirty_files) &&
769                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
770                 sci->sc_nfreesegs == 0 &&
771                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
772 }
773
774 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
775 {
776         struct nilfs_sb_info *sbi = sci->sc_sbi;
777         int ret = 0;
778
779         if (nilfs_test_metadata_dirty(sbi->s_nilfs, sci->sc_root))
780                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
781
782         spin_lock(&sbi->s_inode_lock);
783         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
784                 ret++;
785
786         spin_unlock(&sbi->s_inode_lock);
787         return ret;
788 }
789
790 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
791 {
792         struct nilfs_sb_info *sbi = sci->sc_sbi;
793         struct the_nilfs *nilfs = sbi->s_nilfs;
794
795         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
796         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
797         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
798         nilfs_mdt_clear_dirty(nilfs->ns_dat);
799 }
800
801 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
802 {
803         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
804         struct buffer_head *bh_cp;
805         struct nilfs_checkpoint *raw_cp;
806         int err;
807
808         /* XXX: this interface will be changed */
809         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
810                                           &raw_cp, &bh_cp);
811         if (likely(!err)) {
812                 /* The following code is duplicated with cpfile.  But, it is
813                    needed to collect the checkpoint even if it was not newly
814                    created */
815                 nilfs_mdt_mark_buffer_dirty(bh_cp);
816                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
817                 nilfs_cpfile_put_checkpoint(
818                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
819         } else
820                 WARN_ON(err == -EINVAL || err == -ENOENT);
821
822         return err;
823 }
824
825 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
826 {
827         struct nilfs_sb_info *sbi = sci->sc_sbi;
828         struct the_nilfs *nilfs = sbi->s_nilfs;
829         struct buffer_head *bh_cp;
830         struct nilfs_checkpoint *raw_cp;
831         int err;
832
833         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
834                                           &raw_cp, &bh_cp);
835         if (unlikely(err)) {
836                 WARN_ON(err == -EINVAL || err == -ENOENT);
837                 goto failed_ibh;
838         }
839         raw_cp->cp_snapshot_list.ssl_next = 0;
840         raw_cp->cp_snapshot_list.ssl_prev = 0;
841         raw_cp->cp_inodes_count =
842                 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
843         raw_cp->cp_blocks_count =
844                 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
845         raw_cp->cp_nblk_inc =
846                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
847         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
848         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
849
850         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
851                 nilfs_checkpoint_clear_minor(raw_cp);
852         else
853                 nilfs_checkpoint_set_minor(raw_cp);
854
855         nilfs_write_inode_common(sci->sc_root->ifile,
856                                  &raw_cp->cp_ifile_inode, 1);
857         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
858         return 0;
859
860  failed_ibh:
861         return err;
862 }
863
864 static void nilfs_fill_in_file_bmap(struct inode *ifile,
865                                     struct nilfs_inode_info *ii)
866
867 {
868         struct buffer_head *ibh;
869         struct nilfs_inode *raw_inode;
870
871         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
872                 ibh = ii->i_bh;
873                 BUG_ON(!ibh);
874                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
875                                                   ibh);
876                 nilfs_bmap_write(ii->i_bmap, raw_inode);
877                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
878         }
879 }
880
881 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
882 {
883         struct nilfs_inode_info *ii;
884
885         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
886                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
887                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
888         }
889 }
890
891 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
892                                              struct the_nilfs *nilfs)
893 {
894         struct buffer_head *bh_sr;
895         struct nilfs_super_root *raw_sr;
896         unsigned isz = nilfs->ns_inode_size;
897
898         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
899         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
900
901         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
902         raw_sr->sr_nongc_ctime
903                 = cpu_to_le64(nilfs_doing_gc() ?
904                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
905         raw_sr->sr_flags = 0;
906
907         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
908                                  NILFS_SR_DAT_OFFSET(isz), 1);
909         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
910                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
911         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
912                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
913 }
914
915 static void nilfs_redirty_inodes(struct list_head *head)
916 {
917         struct nilfs_inode_info *ii;
918
919         list_for_each_entry(ii, head, i_dirty) {
920                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
921                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
922         }
923 }
924
925 static void nilfs_drop_collected_inodes(struct list_head *head)
926 {
927         struct nilfs_inode_info *ii;
928
929         list_for_each_entry(ii, head, i_dirty) {
930                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
931                         continue;
932
933                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
934                 set_bit(NILFS_I_UPDATED, &ii->i_state);
935         }
936 }
937
938 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
939                                        struct inode *inode,
940                                        struct list_head *listp,
941                                        int (*collect)(struct nilfs_sc_info *,
942                                                       struct buffer_head *,
943                                                       struct inode *))
944 {
945         struct buffer_head *bh, *n;
946         int err = 0;
947
948         if (collect) {
949                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
950                         list_del_init(&bh->b_assoc_buffers);
951                         err = collect(sci, bh, inode);
952                         brelse(bh);
953                         if (unlikely(err))
954                                 goto dispose_buffers;
955                 }
956                 return 0;
957         }
958
959  dispose_buffers:
960         while (!list_empty(listp)) {
961                 bh = list_entry(listp->next, struct buffer_head,
962                                 b_assoc_buffers);
963                 list_del_init(&bh->b_assoc_buffers);
964                 brelse(bh);
965         }
966         return err;
967 }
968
969 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
970 {
971         /* Remaining number of blocks within segment buffer */
972         return sci->sc_segbuf_nblocks -
973                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
974 }
975
976 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
977                                    struct inode *inode,
978                                    struct nilfs_sc_operations *sc_ops)
979 {
980         LIST_HEAD(data_buffers);
981         LIST_HEAD(node_buffers);
982         int err;
983
984         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
985                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
986
987                 n = nilfs_lookup_dirty_data_buffers(
988                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
989                 if (n > rest) {
990                         err = nilfs_segctor_apply_buffers(
991                                 sci, inode, &data_buffers,
992                                 sc_ops->collect_data);
993                         BUG_ON(!err); /* always receive -E2BIG or true error */
994                         goto break_or_fail;
995                 }
996         }
997         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
998
999         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1000                 err = nilfs_segctor_apply_buffers(
1001                         sci, inode, &data_buffers, sc_ops->collect_data);
1002                 if (unlikely(err)) {
1003                         /* dispose node list */
1004                         nilfs_segctor_apply_buffers(
1005                                 sci, inode, &node_buffers, NULL);
1006                         goto break_or_fail;
1007                 }
1008                 sci->sc_stage.flags |= NILFS_CF_NODE;
1009         }
1010         /* Collect node */
1011         err = nilfs_segctor_apply_buffers(
1012                 sci, inode, &node_buffers, sc_ops->collect_node);
1013         if (unlikely(err))
1014                 goto break_or_fail;
1015
1016         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1017         err = nilfs_segctor_apply_buffers(
1018                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1019         if (unlikely(err))
1020                 goto break_or_fail;
1021
1022         nilfs_segctor_end_finfo(sci, inode);
1023         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1024
1025  break_or_fail:
1026         return err;
1027 }
1028
1029 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1030                                          struct inode *inode)
1031 {
1032         LIST_HEAD(data_buffers);
1033         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1034         int err;
1035
1036         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1037                                             sci->sc_dsync_start,
1038                                             sci->sc_dsync_end);
1039
1040         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1041                                           nilfs_collect_file_data);
1042         if (!err) {
1043                 nilfs_segctor_end_finfo(sci, inode);
1044                 BUG_ON(n > rest);
1045                 /* always receive -E2BIG or true error if n > rest */
1046         }
1047         return err;
1048 }
1049
1050 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1051 {
1052         struct nilfs_sb_info *sbi = sci->sc_sbi;
1053         struct the_nilfs *nilfs = sbi->s_nilfs;
1054         struct list_head *head;
1055         struct nilfs_inode_info *ii;
1056         size_t ndone;
1057         int err = 0;
1058
1059         switch (sci->sc_stage.scnt) {
1060         case NILFS_ST_INIT:
1061                 /* Pre-processes */
1062                 sci->sc_stage.flags = 0;
1063
1064                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1065                         sci->sc_nblk_inc = 0;
1066                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1067                         if (mode == SC_LSEG_DSYNC) {
1068                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1069                                 goto dsync_mode;
1070                         }
1071                 }
1072
1073                 sci->sc_stage.dirty_file_ptr = NULL;
1074                 sci->sc_stage.gc_inode_ptr = NULL;
1075                 if (mode == SC_FLUSH_DAT) {
1076                         sci->sc_stage.scnt = NILFS_ST_DAT;
1077                         goto dat_stage;
1078                 }
1079                 sci->sc_stage.scnt++;  /* Fall through */
1080         case NILFS_ST_GC:
1081                 if (nilfs_doing_gc()) {
1082                         head = &sci->sc_gc_inodes;
1083                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1084                                                 head, i_dirty);
1085                         list_for_each_entry_continue(ii, head, i_dirty) {
1086                                 err = nilfs_segctor_scan_file(
1087                                         sci, &ii->vfs_inode,
1088                                         &nilfs_sc_file_ops);
1089                                 if (unlikely(err)) {
1090                                         sci->sc_stage.gc_inode_ptr = list_entry(
1091                                                 ii->i_dirty.prev,
1092                                                 struct nilfs_inode_info,
1093                                                 i_dirty);
1094                                         goto break_or_fail;
1095                                 }
1096                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1097                         }
1098                         sci->sc_stage.gc_inode_ptr = NULL;
1099                 }
1100                 sci->sc_stage.scnt++;  /* Fall through */
1101         case NILFS_ST_FILE:
1102                 head = &sci->sc_dirty_files;
1103                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1104                                         i_dirty);
1105                 list_for_each_entry_continue(ii, head, i_dirty) {
1106                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1107
1108                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1109                                                       &nilfs_sc_file_ops);
1110                         if (unlikely(err)) {
1111                                 sci->sc_stage.dirty_file_ptr =
1112                                         list_entry(ii->i_dirty.prev,
1113                                                    struct nilfs_inode_info,
1114                                                    i_dirty);
1115                                 goto break_or_fail;
1116                         }
1117                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1118                         /* XXX: required ? */
1119                 }
1120                 sci->sc_stage.dirty_file_ptr = NULL;
1121                 if (mode == SC_FLUSH_FILE) {
1122                         sci->sc_stage.scnt = NILFS_ST_DONE;
1123                         return 0;
1124                 }
1125                 sci->sc_stage.scnt++;
1126                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1127                 /* Fall through */
1128         case NILFS_ST_IFILE:
1129                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1130                                               &nilfs_sc_file_ops);
1131                 if (unlikely(err))
1132                         break;
1133                 sci->sc_stage.scnt++;
1134                 /* Creating a checkpoint */
1135                 err = nilfs_segctor_create_checkpoint(sci);
1136                 if (unlikely(err))
1137                         break;
1138                 /* Fall through */
1139         case NILFS_ST_CPFILE:
1140                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1141                                               &nilfs_sc_file_ops);
1142                 if (unlikely(err))
1143                         break;
1144                 sci->sc_stage.scnt++;  /* Fall through */
1145         case NILFS_ST_SUFILE:
1146                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1147                                          sci->sc_nfreesegs, &ndone);
1148                 if (unlikely(err)) {
1149                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1150                                                   sci->sc_freesegs, ndone,
1151                                                   NULL);
1152                         break;
1153                 }
1154                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1155
1156                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1157                                               &nilfs_sc_file_ops);
1158                 if (unlikely(err))
1159                         break;
1160                 sci->sc_stage.scnt++;  /* Fall through */
1161         case NILFS_ST_DAT:
1162  dat_stage:
1163                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1164                                               &nilfs_sc_dat_ops);
1165                 if (unlikely(err))
1166                         break;
1167                 if (mode == SC_FLUSH_DAT) {
1168                         sci->sc_stage.scnt = NILFS_ST_DONE;
1169                         return 0;
1170                 }
1171                 sci->sc_stage.scnt++;  /* Fall through */
1172         case NILFS_ST_SR:
1173                 if (mode == SC_LSEG_SR) {
1174                         /* Appending a super root */
1175                         err = nilfs_segctor_add_super_root(sci);
1176                         if (unlikely(err))
1177                                 break;
1178                 }
1179                 /* End of a logical segment */
1180                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1181                 sci->sc_stage.scnt = NILFS_ST_DONE;
1182                 return 0;
1183         case NILFS_ST_DSYNC:
1184  dsync_mode:
1185                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1186                 ii = sci->sc_dsync_inode;
1187                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1188                         break;
1189
1190                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1191                 if (unlikely(err))
1192                         break;
1193                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1194                 sci->sc_stage.scnt = NILFS_ST_DONE;
1195                 return 0;
1196         case NILFS_ST_DONE:
1197                 return 0;
1198         default:
1199                 BUG();
1200         }
1201
1202  break_or_fail:
1203         return err;
1204 }
1205
1206 /**
1207  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1208  * @sci: nilfs_sc_info
1209  * @nilfs: nilfs object
1210  */
1211 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1212                                             struct the_nilfs *nilfs)
1213 {
1214         struct nilfs_segment_buffer *segbuf, *prev;
1215         __u64 nextnum;
1216         int err, alloc = 0;
1217
1218         segbuf = nilfs_segbuf_new(sci->sc_super);
1219         if (unlikely(!segbuf))
1220                 return -ENOMEM;
1221
1222         if (list_empty(&sci->sc_write_logs)) {
1223                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1224                                  nilfs->ns_pseg_offset, nilfs);
1225                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1226                         nilfs_shift_to_next_segment(nilfs);
1227                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1228                 }
1229
1230                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1231                 nextnum = nilfs->ns_nextnum;
1232
1233                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1234                         /* Start from the head of a new full segment */
1235                         alloc++;
1236         } else {
1237                 /* Continue logs */
1238                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1239                 nilfs_segbuf_map_cont(segbuf, prev);
1240                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1241                 nextnum = prev->sb_nextnum;
1242
1243                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1244                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1245                         segbuf->sb_sum.seg_seq++;
1246                         alloc++;
1247                 }
1248         }
1249
1250         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1251         if (err)
1252                 goto failed;
1253
1254         if (alloc) {
1255                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1256                 if (err)
1257                         goto failed;
1258         }
1259         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1260
1261         BUG_ON(!list_empty(&sci->sc_segbufs));
1262         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1263         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1264         return 0;
1265
1266  failed:
1267         nilfs_segbuf_free(segbuf);
1268         return err;
1269 }
1270
1271 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1272                                          struct the_nilfs *nilfs, int nadd)
1273 {
1274         struct nilfs_segment_buffer *segbuf, *prev;
1275         struct inode *sufile = nilfs->ns_sufile;
1276         __u64 nextnextnum;
1277         LIST_HEAD(list);
1278         int err, ret, i;
1279
1280         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1281         /*
1282          * Since the segment specified with nextnum might be allocated during
1283          * the previous construction, the buffer including its segusage may
1284          * not be dirty.  The following call ensures that the buffer is dirty
1285          * and will pin the buffer on memory until the sufile is written.
1286          */
1287         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1288         if (unlikely(err))
1289                 return err;
1290
1291         for (i = 0; i < nadd; i++) {
1292                 /* extend segment info */
1293                 err = -ENOMEM;
1294                 segbuf = nilfs_segbuf_new(sci->sc_super);
1295                 if (unlikely(!segbuf))
1296                         goto failed;
1297
1298                 /* map this buffer to region of segment on-disk */
1299                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1300                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1301
1302                 /* allocate the next next full segment */
1303                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1304                 if (unlikely(err))
1305                         goto failed_segbuf;
1306
1307                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1308                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1309
1310                 list_add_tail(&segbuf->sb_list, &list);
1311                 prev = segbuf;
1312         }
1313         list_splice_tail(&list, &sci->sc_segbufs);
1314         return 0;
1315
1316  failed_segbuf:
1317         nilfs_segbuf_free(segbuf);
1318  failed:
1319         list_for_each_entry(segbuf, &list, sb_list) {
1320                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1321                 WARN_ON(ret); /* never fails */
1322         }
1323         nilfs_destroy_logs(&list);
1324         return err;
1325 }
1326
1327 static void nilfs_free_incomplete_logs(struct list_head *logs,
1328                                        struct the_nilfs *nilfs)
1329 {
1330         struct nilfs_segment_buffer *segbuf, *prev;
1331         struct inode *sufile = nilfs->ns_sufile;
1332         int ret;
1333
1334         segbuf = NILFS_FIRST_SEGBUF(logs);
1335         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1336                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1337                 WARN_ON(ret); /* never fails */
1338         }
1339         if (atomic_read(&segbuf->sb_err)) {
1340                 /* Case 1: The first segment failed */
1341                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1342                         /* Case 1a:  Partial segment appended into an existing
1343                            segment */
1344                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1345                                                 segbuf->sb_fseg_end);
1346                 else /* Case 1b:  New full segment */
1347                         set_nilfs_discontinued(nilfs);
1348         }
1349
1350         prev = segbuf;
1351         list_for_each_entry_continue(segbuf, logs, sb_list) {
1352                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1353                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1354                         WARN_ON(ret); /* never fails */
1355                 }
1356                 if (atomic_read(&segbuf->sb_err) &&
1357                     segbuf->sb_segnum != nilfs->ns_nextnum)
1358                         /* Case 2: extended segment (!= next) failed */
1359                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1360                 prev = segbuf;
1361         }
1362 }
1363
1364 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1365                                           struct inode *sufile)
1366 {
1367         struct nilfs_segment_buffer *segbuf;
1368         unsigned long live_blocks;
1369         int ret;
1370
1371         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1372                 live_blocks = segbuf->sb_sum.nblocks +
1373                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1374                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1375                                                      live_blocks,
1376                                                      sci->sc_seg_ctime);
1377                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1378         }
1379 }
1380
1381 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1382 {
1383         struct nilfs_segment_buffer *segbuf;
1384         int ret;
1385
1386         segbuf = NILFS_FIRST_SEGBUF(logs);
1387         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1388                                              segbuf->sb_pseg_start -
1389                                              segbuf->sb_fseg_start, 0);
1390         WARN_ON(ret); /* always succeed because the segusage is dirty */
1391
1392         list_for_each_entry_continue(segbuf, logs, sb_list) {
1393                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1394                                                      0, 0);
1395                 WARN_ON(ret); /* always succeed */
1396         }
1397 }
1398
1399 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1400                                             struct nilfs_segment_buffer *last,
1401                                             struct inode *sufile)
1402 {
1403         struct nilfs_segment_buffer *segbuf = last;
1404         int ret;
1405
1406         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1407                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1408                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1409                 WARN_ON(ret);
1410         }
1411         nilfs_truncate_logs(&sci->sc_segbufs, last);
1412 }
1413
1414
1415 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1416                                  struct the_nilfs *nilfs, int mode)
1417 {
1418         struct nilfs_cstage prev_stage = sci->sc_stage;
1419         int err, nadd = 1;
1420
1421         /* Collection retry loop */
1422         for (;;) {
1423                 sci->sc_nblk_this_inc = 0;
1424                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1425
1426                 err = nilfs_segctor_reset_segment_buffer(sci);
1427                 if (unlikely(err))
1428                         goto failed;
1429
1430                 err = nilfs_segctor_collect_blocks(sci, mode);
1431                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1432                 if (!err)
1433                         break;
1434
1435                 if (unlikely(err != -E2BIG))
1436                         goto failed;
1437
1438                 /* The current segment is filled up */
1439                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1440                         break;
1441
1442                 nilfs_clear_logs(&sci->sc_segbufs);
1443
1444                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1445                 if (unlikely(err))
1446                         return err;
1447
1448                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1449                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1450                                                         sci->sc_freesegs,
1451                                                         sci->sc_nfreesegs,
1452                                                         NULL);
1453                         WARN_ON(err); /* do not happen */
1454                 }
1455                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1456                 sci->sc_stage = prev_stage;
1457         }
1458         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1459         return 0;
1460
1461  failed:
1462         return err;
1463 }
1464
1465 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1466                                       struct buffer_head *new_bh)
1467 {
1468         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1469
1470         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1471         /* The caller must release old_bh */
1472 }
1473
1474 static int
1475 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1476                                      struct nilfs_segment_buffer *segbuf,
1477                                      int mode)
1478 {
1479         struct inode *inode = NULL;
1480         sector_t blocknr;
1481         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1482         unsigned long nblocks = 0, ndatablk = 0;
1483         struct nilfs_sc_operations *sc_op = NULL;
1484         struct nilfs_segsum_pointer ssp;
1485         struct nilfs_finfo *finfo = NULL;
1486         union nilfs_binfo binfo;
1487         struct buffer_head *bh, *bh_org;
1488         ino_t ino = 0;
1489         int err = 0;
1490
1491         if (!nfinfo)
1492                 goto out;
1493
1494         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1495         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1496         ssp.offset = sizeof(struct nilfs_segment_summary);
1497
1498         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1499                 if (bh == segbuf->sb_super_root)
1500                         break;
1501                 if (!finfo) {
1502                         finfo = nilfs_segctor_map_segsum_entry(
1503                                 sci, &ssp, sizeof(*finfo));
1504                         ino = le64_to_cpu(finfo->fi_ino);
1505                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1506                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1507
1508                         if (buffer_nilfs_node(bh))
1509                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1510                         else
1511                                 inode = NILFS_AS_I(bh->b_page->mapping);
1512
1513                         if (mode == SC_LSEG_DSYNC)
1514                                 sc_op = &nilfs_sc_dsync_ops;
1515                         else if (ino == NILFS_DAT_INO)
1516                                 sc_op = &nilfs_sc_dat_ops;
1517                         else /* file blocks */
1518                                 sc_op = &nilfs_sc_file_ops;
1519                 }
1520                 bh_org = bh;
1521                 get_bh(bh_org);
1522                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1523                                         &binfo);
1524                 if (bh != bh_org)
1525                         nilfs_list_replace_buffer(bh_org, bh);
1526                 brelse(bh_org);
1527                 if (unlikely(err))
1528                         goto failed_bmap;
1529
1530                 if (ndatablk > 0)
1531                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1532                 else
1533                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1534
1535                 blocknr++;
1536                 if (--nblocks == 0) {
1537                         finfo = NULL;
1538                         if (--nfinfo == 0)
1539                                 break;
1540                 } else if (ndatablk > 0)
1541                         ndatablk--;
1542         }
1543  out:
1544         return 0;
1545
1546  failed_bmap:
1547         return err;
1548 }
1549
1550 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1551 {
1552         struct nilfs_segment_buffer *segbuf;
1553         int err;
1554
1555         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1556                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1557                 if (unlikely(err))
1558                         return err;
1559                 nilfs_segbuf_fill_in_segsum(segbuf);
1560         }
1561         return 0;
1562 }
1563
1564 static int
1565 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1566 {
1567         struct page *clone_page;
1568         struct buffer_head *bh, *head, *bh2;
1569         void *kaddr;
1570
1571         bh = head = page_buffers(page);
1572
1573         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1574         if (unlikely(!clone_page))
1575                 return -ENOMEM;
1576
1577         bh2 = page_buffers(clone_page);
1578         kaddr = kmap_atomic(page, KM_USER0);
1579         do {
1580                 if (list_empty(&bh->b_assoc_buffers))
1581                         continue;
1582                 get_bh(bh2);
1583                 page_cache_get(clone_page); /* for each bh */
1584                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1585                 bh2->b_blocknr = bh->b_blocknr;
1586                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1587                 list_add_tail(&bh->b_assoc_buffers, out);
1588         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1589         kunmap_atomic(kaddr, KM_USER0);
1590
1591         if (!TestSetPageWriteback(clone_page))
1592                 account_page_writeback(clone_page);
1593         unlock_page(clone_page);
1594
1595         return 0;
1596 }
1597
1598 static int nilfs_test_page_to_be_frozen(struct page *page)
1599 {
1600         struct address_space *mapping = page->mapping;
1601
1602         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1603                 return 0;
1604
1605         if (page_mapped(page)) {
1606                 ClearPageChecked(page);
1607                 return 1;
1608         }
1609         return PageChecked(page);
1610 }
1611
1612 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1613 {
1614         if (!page || PageWriteback(page))
1615                 /* For split b-tree node pages, this function may be called
1616                    twice.  We ignore the 2nd or later calls by this check. */
1617                 return 0;
1618
1619         lock_page(page);
1620         clear_page_dirty_for_io(page);
1621         set_page_writeback(page);
1622         unlock_page(page);
1623
1624         if (nilfs_test_page_to_be_frozen(page)) {
1625                 int err = nilfs_copy_replace_page_buffers(page, out);
1626                 if (unlikely(err))
1627                         return err;
1628         }
1629         return 0;
1630 }
1631
1632 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1633                                        struct page **failed_page)
1634 {
1635         struct nilfs_segment_buffer *segbuf;
1636         struct page *bd_page = NULL, *fs_page = NULL;
1637         struct list_head *list = &sci->sc_copied_buffers;
1638         int err;
1639
1640         *failed_page = NULL;
1641         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1642                 struct buffer_head *bh;
1643
1644                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1645                                     b_assoc_buffers) {
1646                         if (bh->b_page != bd_page) {
1647                                 if (bd_page) {
1648                                         lock_page(bd_page);
1649                                         clear_page_dirty_for_io(bd_page);
1650                                         set_page_writeback(bd_page);
1651                                         unlock_page(bd_page);
1652                                 }
1653                                 bd_page = bh->b_page;
1654                         }
1655                 }
1656
1657                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1658                                     b_assoc_buffers) {
1659                         if (bh == segbuf->sb_super_root) {
1660                                 if (bh->b_page != bd_page) {
1661                                         lock_page(bd_page);
1662                                         clear_page_dirty_for_io(bd_page);
1663                                         set_page_writeback(bd_page);
1664                                         unlock_page(bd_page);
1665                                         bd_page = bh->b_page;
1666                                 }
1667                                 break;
1668                         }
1669                         if (bh->b_page != fs_page) {
1670                                 err = nilfs_begin_page_io(fs_page, list);
1671                                 if (unlikely(err)) {
1672                                         *failed_page = fs_page;
1673                                         goto out;
1674                                 }
1675                                 fs_page = bh->b_page;
1676                         }
1677                 }
1678         }
1679         if (bd_page) {
1680                 lock_page(bd_page);
1681                 clear_page_dirty_for_io(bd_page);
1682                 set_page_writeback(bd_page);
1683                 unlock_page(bd_page);
1684         }
1685         err = nilfs_begin_page_io(fs_page, list);
1686         if (unlikely(err))
1687                 *failed_page = fs_page;
1688  out:
1689         return err;
1690 }
1691
1692 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1693                                struct the_nilfs *nilfs)
1694 {
1695         int ret;
1696
1697         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1698         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1699         return ret;
1700 }
1701
1702 static void __nilfs_end_page_io(struct page *page, int err)
1703 {
1704         if (!err) {
1705                 if (!nilfs_page_buffers_clean(page))
1706                         __set_page_dirty_nobuffers(page);
1707                 ClearPageError(page);
1708         } else {
1709                 __set_page_dirty_nobuffers(page);
1710                 SetPageError(page);
1711         }
1712
1713         if (buffer_nilfs_allocated(page_buffers(page))) {
1714                 if (TestClearPageWriteback(page))
1715                         dec_zone_page_state(page, NR_WRITEBACK);
1716         } else
1717                 end_page_writeback(page);
1718 }
1719
1720 static void nilfs_end_page_io(struct page *page, int err)
1721 {
1722         if (!page)
1723                 return;
1724
1725         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1726                 /*
1727                  * For b-tree node pages, this function may be called twice
1728                  * or more because they might be split in a segment.
1729                  */
1730                 if (PageDirty(page)) {
1731                         /*
1732                          * For pages holding split b-tree node buffers, dirty
1733                          * flag on the buffers may be cleared discretely.
1734                          * In that case, the page is once redirtied for
1735                          * remaining buffers, and it must be cancelled if
1736                          * all the buffers get cleaned later.
1737                          */
1738                         lock_page(page);
1739                         if (nilfs_page_buffers_clean(page))
1740                                 __nilfs_clear_page_dirty(page);
1741                         unlock_page(page);
1742                 }
1743                 return;
1744         }
1745
1746         __nilfs_end_page_io(page, err);
1747 }
1748
1749 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1750 {
1751         struct buffer_head *bh, *head;
1752         struct page *page;
1753
1754         while (!list_empty(list)) {
1755                 bh = list_entry(list->next, struct buffer_head,
1756                                 b_assoc_buffers);
1757                 page = bh->b_page;
1758                 page_cache_get(page);
1759                 head = bh = page_buffers(page);
1760                 do {
1761                         if (!list_empty(&bh->b_assoc_buffers)) {
1762                                 list_del_init(&bh->b_assoc_buffers);
1763                                 if (!err) {
1764                                         set_buffer_uptodate(bh);
1765                                         clear_buffer_dirty(bh);
1766                                         clear_buffer_delay(bh);
1767                                         clear_buffer_nilfs_volatile(bh);
1768                                 }
1769                                 brelse(bh); /* for b_assoc_buffers */
1770                         }
1771                 } while ((bh = bh->b_this_page) != head);
1772
1773                 __nilfs_end_page_io(page, err);
1774                 page_cache_release(page);
1775         }
1776 }
1777
1778 static void nilfs_abort_logs(struct list_head *logs, struct page *failed_page,
1779                              int err)
1780 {
1781         struct nilfs_segment_buffer *segbuf;
1782         struct page *bd_page = NULL, *fs_page = NULL;
1783         struct buffer_head *bh;
1784
1785         if (list_empty(logs))
1786                 return;
1787
1788         list_for_each_entry(segbuf, logs, sb_list) {
1789                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1790                                     b_assoc_buffers) {
1791                         if (bh->b_page != bd_page) {
1792                                 if (bd_page)
1793                                         end_page_writeback(bd_page);
1794                                 bd_page = bh->b_page;
1795                         }
1796                 }
1797
1798                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1799                                     b_assoc_buffers) {
1800                         if (bh == segbuf->sb_super_root) {
1801                                 if (bh->b_page != bd_page) {
1802                                         end_page_writeback(bd_page);
1803                                         bd_page = bh->b_page;
1804                                 }
1805                                 break;
1806                         }
1807                         if (bh->b_page != fs_page) {
1808                                 nilfs_end_page_io(fs_page, err);
1809                                 if (fs_page && fs_page == failed_page)
1810                                         return;
1811                                 fs_page = bh->b_page;
1812                         }
1813                 }
1814         }
1815         if (bd_page)
1816                 end_page_writeback(bd_page);
1817
1818         nilfs_end_page_io(fs_page, err);
1819 }
1820
1821 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1822                                              struct the_nilfs *nilfs, int err)
1823 {
1824         LIST_HEAD(logs);
1825         int ret;
1826
1827         list_splice_tail_init(&sci->sc_write_logs, &logs);
1828         ret = nilfs_wait_on_logs(&logs);
1829         nilfs_abort_logs(&logs, NULL, ret ? : err);
1830
1831         list_splice_tail_init(&sci->sc_segbufs, &logs);
1832         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1833         nilfs_free_incomplete_logs(&logs, nilfs);
1834         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1835
1836         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1837                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1838                                                 sci->sc_freesegs,
1839                                                 sci->sc_nfreesegs,
1840                                                 NULL);
1841                 WARN_ON(ret); /* do not happen */
1842         }
1843
1844         nilfs_destroy_logs(&logs);
1845 }
1846
1847 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1848                                    struct nilfs_segment_buffer *segbuf)
1849 {
1850         nilfs->ns_segnum = segbuf->sb_segnum;
1851         nilfs->ns_nextnum = segbuf->sb_nextnum;
1852         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1853                 + segbuf->sb_sum.nblocks;
1854         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1855         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1856 }
1857
1858 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1859 {
1860         struct nilfs_segment_buffer *segbuf;
1861         struct page *bd_page = NULL, *fs_page = NULL;
1862         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
1863         int update_sr = false;
1864
1865         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1866                 struct buffer_head *bh;
1867
1868                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1869                                     b_assoc_buffers) {
1870                         set_buffer_uptodate(bh);
1871                         clear_buffer_dirty(bh);
1872                         if (bh->b_page != bd_page) {
1873                                 if (bd_page)
1874                                         end_page_writeback(bd_page);
1875                                 bd_page = bh->b_page;
1876                         }
1877                 }
1878                 /*
1879                  * We assume that the buffers which belong to the same page
1880                  * continue over the buffer list.
1881                  * Under this assumption, the last BHs of pages is
1882                  * identifiable by the discontinuity of bh->b_page
1883                  * (page != fs_page).
1884                  *
1885                  * For B-tree node blocks, however, this assumption is not
1886                  * guaranteed.  The cleanup code of B-tree node pages needs
1887                  * special care.
1888                  */
1889                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1890                                     b_assoc_buffers) {
1891                         set_buffer_uptodate(bh);
1892                         clear_buffer_dirty(bh);
1893                         clear_buffer_delay(bh);
1894                         clear_buffer_nilfs_volatile(bh);
1895                         clear_buffer_nilfs_redirected(bh);
1896                         if (bh == segbuf->sb_super_root) {
1897                                 if (bh->b_page != bd_page) {
1898                                         end_page_writeback(bd_page);
1899                                         bd_page = bh->b_page;
1900                                 }
1901                                 update_sr = true;
1902                                 break;
1903                         }
1904                         if (bh->b_page != fs_page) {
1905                                 nilfs_end_page_io(fs_page, 0);
1906                                 fs_page = bh->b_page;
1907                         }
1908                 }
1909
1910                 if (!nilfs_segbuf_simplex(segbuf)) {
1911                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1912                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1913                                 sci->sc_lseg_stime = jiffies;
1914                         }
1915                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1916                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1917                 }
1918         }
1919         /*
1920          * Since pages may continue over multiple segment buffers,
1921          * end of the last page must be checked outside of the loop.
1922          */
1923         if (bd_page)
1924                 end_page_writeback(bd_page);
1925
1926         nilfs_end_page_io(fs_page, 0);
1927
1928         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1929
1930         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1931
1932         if (nilfs_doing_gc())
1933                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1934         else
1935                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1936
1937         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1938
1939         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1940         nilfs_set_next_segment(nilfs, segbuf);
1941
1942         if (update_sr) {
1943                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1944                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1945
1946                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1947                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1948                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1949                 nilfs_segctor_clear_metadata_dirty(sci);
1950         } else
1951                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1952 }
1953
1954 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1955 {
1956         int ret;
1957
1958         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1959         if (!ret) {
1960                 nilfs_segctor_complete_write(sci);
1961                 nilfs_destroy_logs(&sci->sc_write_logs);
1962         }
1963         return ret;
1964 }
1965
1966 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
1967                                         struct nilfs_sb_info *sbi)
1968 {
1969         struct nilfs_inode_info *ii, *n;
1970         struct inode *ifile = sci->sc_root->ifile;
1971
1972         spin_lock(&sbi->s_inode_lock);
1973  retry:
1974         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
1975                 if (!ii->i_bh) {
1976                         struct buffer_head *ibh;
1977                         int err;
1978
1979                         spin_unlock(&sbi->s_inode_lock);
1980                         err = nilfs_ifile_get_inode_block(
1981                                 ifile, ii->vfs_inode.i_ino, &ibh);
1982                         if (unlikely(err)) {
1983                                 nilfs_warning(sbi->s_super, __func__,
1984                                               "failed to get inode block.\n");
1985                                 return err;
1986                         }
1987                         nilfs_mdt_mark_buffer_dirty(ibh);
1988                         nilfs_mdt_mark_dirty(ifile);
1989                         spin_lock(&sbi->s_inode_lock);
1990                         if (likely(!ii->i_bh))
1991                                 ii->i_bh = ibh;
1992                         else
1993                                 brelse(ibh);
1994                         goto retry;
1995                 }
1996
1997                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1998                 set_bit(NILFS_I_BUSY, &ii->i_state);
1999                 list_del(&ii->i_dirty);
2000                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2001         }
2002         spin_unlock(&sbi->s_inode_lock);
2003
2004         return 0;
2005 }
2006
2007 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2008                                           struct nilfs_sb_info *sbi)
2009 {
2010         struct nilfs_transaction_info *ti = current->journal_info;
2011         struct nilfs_inode_info *ii, *n;
2012
2013         spin_lock(&sbi->s_inode_lock);
2014         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2015                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2016                     test_bit(NILFS_I_DIRTY, &ii->i_state))
2017                         continue;
2018
2019                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2020                 brelse(ii->i_bh);
2021                 ii->i_bh = NULL;
2022                 list_del(&ii->i_dirty);
2023                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2024         }
2025         spin_unlock(&sbi->s_inode_lock);
2026 }
2027
2028 /*
2029  * Main procedure of segment constructor
2030  */
2031 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2032 {
2033         struct nilfs_sb_info *sbi = sci->sc_sbi;
2034         struct the_nilfs *nilfs = sbi->s_nilfs;
2035         struct page *failed_page;
2036         int err;
2037
2038         sci->sc_stage.scnt = NILFS_ST_INIT;
2039         sci->sc_cno = nilfs->ns_cno;
2040
2041         err = nilfs_segctor_check_in_files(sci, sbi);
2042         if (unlikely(err))
2043                 goto out;
2044
2045         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2046                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2047
2048         if (nilfs_segctor_clean(sci))
2049                 goto out;
2050
2051         do {
2052                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2053
2054                 err = nilfs_segctor_begin_construction(sci, nilfs);
2055                 if (unlikely(err))
2056                         goto out;
2057
2058                 /* Update time stamp */
2059                 sci->sc_seg_ctime = get_seconds();
2060
2061                 err = nilfs_segctor_collect(sci, nilfs, mode);
2062                 if (unlikely(err))
2063                         goto failed;
2064
2065                 /* Avoid empty segment */
2066                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2067                     nilfs_segbuf_empty(sci->sc_curseg)) {
2068                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2069                         goto out;
2070                 }
2071
2072                 err = nilfs_segctor_assign(sci, mode);
2073                 if (unlikely(err))
2074                         goto failed;
2075
2076                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2077                         nilfs_segctor_fill_in_file_bmap(sci);
2078
2079                 if (mode == SC_LSEG_SR &&
2080                     sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
2081                         err = nilfs_segctor_fill_in_checkpoint(sci);
2082                         if (unlikely(err))
2083                                 goto failed_to_write;
2084
2085                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2086                 }
2087                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2088
2089                 /* Write partial segments */
2090                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2091                 if (err) {
2092                         nilfs_abort_logs(&sci->sc_segbufs, failed_page, err);
2093                         goto failed_to_write;
2094                 }
2095
2096                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2097                                             nilfs->ns_crc_seed);
2098
2099                 err = nilfs_segctor_write(sci, nilfs);
2100                 if (unlikely(err))
2101                         goto failed_to_write;
2102
2103                 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2104                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2105                         /*
2106                          * At this point, we avoid double buffering
2107                          * for blocksize < pagesize because page dirty
2108                          * flag is turned off during write and dirty
2109                          * buffers are not properly collected for
2110                          * pages crossing over segments.
2111                          */
2112                         err = nilfs_segctor_wait(sci);
2113                         if (err)
2114                                 goto failed_to_write;
2115                 }
2116         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2117
2118  out:
2119         nilfs_segctor_check_out_files(sci, sbi);
2120         return err;
2121
2122  failed_to_write:
2123         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2124                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2125
2126  failed:
2127         if (nilfs_doing_gc())
2128                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2129         nilfs_segctor_abort_construction(sci, nilfs, err);
2130         goto out;
2131 }
2132
2133 /**
2134  * nilfs_segctor_start_timer - set timer of background write
2135  * @sci: nilfs_sc_info
2136  *
2137  * If the timer has already been set, it ignores the new request.
2138  * This function MUST be called within a section locking the segment
2139  * semaphore.
2140  */
2141 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2142 {
2143         spin_lock(&sci->sc_state_lock);
2144         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2145                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2146                 add_timer(&sci->sc_timer);
2147                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2148         }
2149         spin_unlock(&sci->sc_state_lock);
2150 }
2151
2152 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2153 {
2154         spin_lock(&sci->sc_state_lock);
2155         if (!(sci->sc_flush_request & (1 << bn))) {
2156                 unsigned long prev_req = sci->sc_flush_request;
2157
2158                 sci->sc_flush_request |= (1 << bn);
2159                 if (!prev_req)
2160                         wake_up(&sci->sc_wait_daemon);
2161         }
2162         spin_unlock(&sci->sc_state_lock);
2163 }
2164
2165 /**
2166  * nilfs_flush_segment - trigger a segment construction for resource control
2167  * @sb: super block
2168  * @ino: inode number of the file to be flushed out.
2169  */
2170 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2171 {
2172         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2173         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2174
2175         if (!sci || nilfs_doing_construction())
2176                 return;
2177         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2178                                         /* assign bit 0 to data files */
2179 }
2180
2181 struct nilfs_segctor_wait_request {
2182         wait_queue_t    wq;
2183         __u32           seq;
2184         int             err;
2185         atomic_t        done;
2186 };
2187
2188 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2189 {
2190         struct nilfs_segctor_wait_request wait_req;
2191         int err = 0;
2192
2193         spin_lock(&sci->sc_state_lock);
2194         init_wait(&wait_req.wq);
2195         wait_req.err = 0;
2196         atomic_set(&wait_req.done, 0);
2197         wait_req.seq = ++sci->sc_seq_request;
2198         spin_unlock(&sci->sc_state_lock);
2199
2200         init_waitqueue_entry(&wait_req.wq, current);
2201         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2202         set_current_state(TASK_INTERRUPTIBLE);
2203         wake_up(&sci->sc_wait_daemon);
2204
2205         for (;;) {
2206                 if (atomic_read(&wait_req.done)) {
2207                         err = wait_req.err;
2208                         break;
2209                 }
2210                 if (!signal_pending(current)) {
2211                         schedule();
2212                         continue;
2213                 }
2214                 err = -ERESTARTSYS;
2215                 break;
2216         }
2217         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2218         return err;
2219 }
2220
2221 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2222 {
2223         struct nilfs_segctor_wait_request *wrq, *n;
2224         unsigned long flags;
2225
2226         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2227         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2228                                  wq.task_list) {
2229                 if (!atomic_read(&wrq->done) &&
2230                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2231                         wrq->err = err;
2232                         atomic_set(&wrq->done, 1);
2233                 }
2234                 if (atomic_read(&wrq->done)) {
2235                         wrq->wq.func(&wrq->wq,
2236                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2237                                      0, NULL);
2238                 }
2239         }
2240         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2241 }
2242
2243 /**
2244  * nilfs_construct_segment - construct a logical segment
2245  * @sb: super block
2246  *
2247  * Return Value: On success, 0 is retured. On errors, one of the following
2248  * negative error code is returned.
2249  *
2250  * %-EROFS - Read only filesystem.
2251  *
2252  * %-EIO - I/O error
2253  *
2254  * %-ENOSPC - No space left on device (only in a panic state).
2255  *
2256  * %-ERESTARTSYS - Interrupted.
2257  *
2258  * %-ENOMEM - Insufficient memory available.
2259  */
2260 int nilfs_construct_segment(struct super_block *sb)
2261 {
2262         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2263         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2264         struct nilfs_transaction_info *ti;
2265         int err;
2266
2267         if (!sci)
2268                 return -EROFS;
2269
2270         /* A call inside transactions causes a deadlock. */
2271         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2272
2273         err = nilfs_segctor_sync(sci);
2274         return err;
2275 }
2276
2277 /**
2278  * nilfs_construct_dsync_segment - construct a data-only logical segment
2279  * @sb: super block
2280  * @inode: inode whose data blocks should be written out
2281  * @start: start byte offset
2282  * @end: end byte offset (inclusive)
2283  *
2284  * Return Value: On success, 0 is retured. On errors, one of the following
2285  * negative error code is returned.
2286  *
2287  * %-EROFS - Read only filesystem.
2288  *
2289  * %-EIO - I/O error
2290  *
2291  * %-ENOSPC - No space left on device (only in a panic state).
2292  *
2293  * %-ERESTARTSYS - Interrupted.
2294  *
2295  * %-ENOMEM - Insufficient memory available.
2296  */
2297 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2298                                   loff_t start, loff_t end)
2299 {
2300         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2301         struct the_nilfs *nilfs = sbi->s_nilfs;
2302         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2303         struct nilfs_inode_info *ii;
2304         struct nilfs_transaction_info ti;
2305         int err = 0;
2306
2307         if (!sci)
2308                 return -EROFS;
2309
2310         nilfs_transaction_lock(sbi, &ti, 0);
2311
2312         ii = NILFS_I(inode);
2313         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2314             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2315             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2316             nilfs_discontinued(nilfs)) {
2317                 nilfs_transaction_unlock(sbi);
2318                 err = nilfs_segctor_sync(sci);
2319                 return err;
2320         }
2321
2322         spin_lock(&sbi->s_inode_lock);
2323         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2324             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2325                 spin_unlock(&sbi->s_inode_lock);
2326                 nilfs_transaction_unlock(sbi);
2327                 return 0;
2328         }
2329         spin_unlock(&sbi->s_inode_lock);
2330         sci->sc_dsync_inode = ii;
2331         sci->sc_dsync_start = start;
2332         sci->sc_dsync_end = end;
2333
2334         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2335
2336         nilfs_transaction_unlock(sbi);
2337         return err;
2338 }
2339
2340 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2341 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2342
2343 /**
2344  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2345  * @sci: segment constructor object
2346  */
2347 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2348 {
2349         spin_lock(&sci->sc_state_lock);
2350         sci->sc_seq_accepted = sci->sc_seq_request;
2351         spin_unlock(&sci->sc_state_lock);
2352         del_timer_sync(&sci->sc_timer);
2353 }
2354
2355 /**
2356  * nilfs_segctor_notify - notify the result of request to caller threads
2357  * @sci: segment constructor object
2358  * @mode: mode of log forming
2359  * @err: error code to be notified
2360  */
2361 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2362 {
2363         /* Clear requests (even when the construction failed) */
2364         spin_lock(&sci->sc_state_lock);
2365
2366         if (mode == SC_LSEG_SR) {
2367                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2368                 sci->sc_seq_done = sci->sc_seq_accepted;
2369                 nilfs_segctor_wakeup(sci, err);
2370                 sci->sc_flush_request = 0;
2371         } else {
2372                 if (mode == SC_FLUSH_FILE)
2373                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2374                 else if (mode == SC_FLUSH_DAT)
2375                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2376
2377                 /* re-enable timer if checkpoint creation was not done */
2378                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2379                     time_before(jiffies, sci->sc_timer.expires))
2380                         add_timer(&sci->sc_timer);
2381         }
2382         spin_unlock(&sci->sc_state_lock);
2383 }
2384
2385 /**
2386  * nilfs_segctor_construct - form logs and write them to disk
2387  * @sci: segment constructor object
2388  * @mode: mode of log forming
2389  */
2390 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2391 {
2392         struct nilfs_sb_info *sbi = sci->sc_sbi;
2393         struct the_nilfs *nilfs = sbi->s_nilfs;
2394         struct nilfs_super_block **sbp;
2395         int err = 0;
2396
2397         nilfs_segctor_accept(sci);
2398
2399         if (nilfs_discontinued(nilfs))
2400                 mode = SC_LSEG_SR;
2401         if (!nilfs_segctor_confirm(sci))
2402                 err = nilfs_segctor_do_construct(sci, mode);
2403
2404         if (likely(!err)) {
2405                 if (mode != SC_FLUSH_DAT)
2406                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2407                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2408                     nilfs_discontinued(nilfs)) {
2409                         down_write(&nilfs->ns_sem);
2410                         err = -EIO;
2411                         sbp = nilfs_prepare_super(sbi,
2412                                                   nilfs_sb_will_flip(nilfs));
2413                         if (likely(sbp)) {
2414                                 nilfs_set_log_cursor(sbp[0], nilfs);
2415                                 err = nilfs_commit_super(sbi, NILFS_SB_COMMIT);
2416                         }
2417                         up_write(&nilfs->ns_sem);
2418                 }
2419         }
2420
2421         nilfs_segctor_notify(sci, mode, err);
2422         return err;
2423 }
2424
2425 static void nilfs_construction_timeout(unsigned long data)
2426 {
2427         struct task_struct *p = (struct task_struct *)data;
2428         wake_up_process(p);
2429 }
2430
2431 static void
2432 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2433 {
2434         struct nilfs_inode_info *ii, *n;
2435
2436         list_for_each_entry_safe(ii, n, head, i_dirty) {
2437                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2438                         continue;
2439                 list_del_init(&ii->i_dirty);
2440                 iput(&ii->vfs_inode);
2441         }
2442 }
2443
2444 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2445                          void **kbufs)
2446 {
2447         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2448         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2449         struct the_nilfs *nilfs = sbi->s_nilfs;
2450         struct nilfs_transaction_info ti;
2451         int err;
2452
2453         if (unlikely(!sci))
2454                 return -EROFS;
2455
2456         nilfs_transaction_lock(sbi, &ti, 1);
2457
2458         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2459         if (unlikely(err))
2460                 goto out_unlock;
2461
2462         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2463         if (unlikely(err)) {
2464                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2465                 goto out_unlock;
2466         }
2467
2468         sci->sc_freesegs = kbufs[4];
2469         sci->sc_nfreesegs = argv[4].v_nmembs;
2470         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2471
2472         for (;;) {
2473                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2474                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2475
2476                 if (likely(!err))
2477                         break;
2478
2479                 nilfs_warning(sb, __func__,
2480                               "segment construction failed. (err=%d)", err);
2481                 set_current_state(TASK_INTERRUPTIBLE);
2482                 schedule_timeout(sci->sc_interval);
2483         }
2484         if (nilfs_test_opt(nilfs, DISCARD)) {
2485                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2486                                                  sci->sc_nfreesegs);
2487                 if (ret) {
2488                         printk(KERN_WARNING
2489                                "NILFS warning: error %d on discard request, "
2490                                "turning discards off for the device\n", ret);
2491                         nilfs_clear_opt(nilfs, DISCARD);
2492                 }
2493         }
2494
2495  out_unlock:
2496         sci->sc_freesegs = NULL;
2497         sci->sc_nfreesegs = 0;
2498         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2499         nilfs_transaction_unlock(sbi);
2500         return err;
2501 }
2502
2503 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2504 {
2505         struct nilfs_sb_info *sbi = sci->sc_sbi;
2506         struct nilfs_transaction_info ti;
2507
2508         nilfs_transaction_lock(sbi, &ti, 0);
2509         nilfs_segctor_construct(sci, mode);
2510
2511         /*
2512          * Unclosed segment should be retried.  We do this using sc_timer.
2513          * Timeout of sc_timer will invoke complete construction which leads
2514          * to close the current logical segment.
2515          */
2516         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2517                 nilfs_segctor_start_timer(sci);
2518
2519         nilfs_transaction_unlock(sbi);
2520 }
2521
2522 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2523 {
2524         int mode = 0;
2525         int err;
2526
2527         spin_lock(&sci->sc_state_lock);
2528         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2529                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2530         spin_unlock(&sci->sc_state_lock);
2531
2532         if (mode) {
2533                 err = nilfs_segctor_do_construct(sci, mode);
2534
2535                 spin_lock(&sci->sc_state_lock);
2536                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2537                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2538                 spin_unlock(&sci->sc_state_lock);
2539         }
2540         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2541 }
2542
2543 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2544 {
2545         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2546             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2547                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2548                         return SC_FLUSH_FILE;
2549                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2550                         return SC_FLUSH_DAT;
2551         }
2552         return SC_LSEG_SR;
2553 }
2554
2555 /**
2556  * nilfs_segctor_thread - main loop of the segment constructor thread.
2557  * @arg: pointer to a struct nilfs_sc_info.
2558  *
2559  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2560  * to execute segment constructions.
2561  */
2562 static int nilfs_segctor_thread(void *arg)
2563 {
2564         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2565         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2566         int timeout = 0;
2567
2568         sci->sc_timer.data = (unsigned long)current;
2569         sci->sc_timer.function = nilfs_construction_timeout;
2570
2571         /* start sync. */
2572         sci->sc_task = current;
2573         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2574         printk(KERN_INFO
2575                "segctord starting. Construction interval = %lu seconds, "
2576                "CP frequency < %lu seconds\n",
2577                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2578
2579         spin_lock(&sci->sc_state_lock);
2580  loop:
2581         for (;;) {
2582                 int mode;
2583
2584                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2585                         goto end_thread;
2586
2587                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2588                         mode = SC_LSEG_SR;
2589                 else if (!sci->sc_flush_request)
2590                         break;
2591                 else
2592                         mode = nilfs_segctor_flush_mode(sci);
2593
2594                 spin_unlock(&sci->sc_state_lock);
2595                 nilfs_segctor_thread_construct(sci, mode);
2596                 spin_lock(&sci->sc_state_lock);
2597                 timeout = 0;
2598         }
2599
2600
2601         if (freezing(current)) {
2602                 spin_unlock(&sci->sc_state_lock);
2603                 refrigerator();
2604                 spin_lock(&sci->sc_state_lock);
2605         } else {
2606                 DEFINE_WAIT(wait);
2607                 int should_sleep = 1;
2608
2609                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2610                                 TASK_INTERRUPTIBLE);
2611
2612                 if (sci->sc_seq_request != sci->sc_seq_done)
2613                         should_sleep = 0;
2614                 else if (sci->sc_flush_request)
2615                         should_sleep = 0;
2616                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2617                         should_sleep = time_before(jiffies,
2618                                         sci->sc_timer.expires);
2619
2620                 if (should_sleep) {
2621                         spin_unlock(&sci->sc_state_lock);
2622                         schedule();
2623                         spin_lock(&sci->sc_state_lock);
2624                 }
2625                 finish_wait(&sci->sc_wait_daemon, &wait);
2626                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2627                            time_after_eq(jiffies, sci->sc_timer.expires));
2628
2629                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2630                         set_nilfs_discontinued(nilfs);
2631         }
2632         goto loop;
2633
2634  end_thread:
2635         spin_unlock(&sci->sc_state_lock);
2636
2637         /* end sync. */
2638         sci->sc_task = NULL;
2639         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2640         return 0;
2641 }
2642
2643 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2644 {
2645         struct task_struct *t;
2646
2647         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2648         if (IS_ERR(t)) {
2649                 int err = PTR_ERR(t);
2650
2651                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2652                        err);
2653                 return err;
2654         }
2655         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2656         return 0;
2657 }
2658
2659 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2660         __acquires(&sci->sc_state_lock)
2661         __releases(&sci->sc_state_lock)
2662 {
2663         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2664
2665         while (sci->sc_task) {
2666                 wake_up(&sci->sc_wait_daemon);
2667                 spin_unlock(&sci->sc_state_lock);
2668                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2669                 spin_lock(&sci->sc_state_lock);
2670         }
2671 }
2672
2673 /*
2674  * Setup & clean-up functions
2675  */
2676 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi,
2677                                                struct nilfs_root *root)
2678 {
2679         struct nilfs_sc_info *sci;
2680
2681         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2682         if (!sci)
2683                 return NULL;
2684
2685         sci->sc_sbi = sbi;
2686         sci->sc_super = sbi->s_super;
2687
2688         nilfs_get_root(root);
2689         sci->sc_root = root;
2690
2691         init_waitqueue_head(&sci->sc_wait_request);
2692         init_waitqueue_head(&sci->sc_wait_daemon);
2693         init_waitqueue_head(&sci->sc_wait_task);
2694         spin_lock_init(&sci->sc_state_lock);
2695         INIT_LIST_HEAD(&sci->sc_dirty_files);
2696         INIT_LIST_HEAD(&sci->sc_segbufs);
2697         INIT_LIST_HEAD(&sci->sc_write_logs);
2698         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2699         INIT_LIST_HEAD(&sci->sc_copied_buffers);
2700         init_timer(&sci->sc_timer);
2701
2702         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2703         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2704         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2705
2706         if (sbi->s_interval)
2707                 sci->sc_interval = sbi->s_interval;
2708         if (sbi->s_watermark)
2709                 sci->sc_watermark = sbi->s_watermark;
2710         return sci;
2711 }
2712
2713 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2714 {
2715         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2716
2717         /* The segctord thread was stopped and its timer was removed.
2718            But some tasks remain. */
2719         do {
2720                 struct nilfs_sb_info *sbi = sci->sc_sbi;
2721                 struct nilfs_transaction_info ti;
2722
2723                 nilfs_transaction_lock(sbi, &ti, 0);
2724                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2725                 nilfs_transaction_unlock(sbi);
2726
2727         } while (ret && retrycount-- > 0);
2728 }
2729
2730 /**
2731  * nilfs_segctor_destroy - destroy the segment constructor.
2732  * @sci: nilfs_sc_info
2733  *
2734  * nilfs_segctor_destroy() kills the segctord thread and frees
2735  * the nilfs_sc_info struct.
2736  * Caller must hold the segment semaphore.
2737  */
2738 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2739 {
2740         struct nilfs_sb_info *sbi = sci->sc_sbi;
2741         int flag;
2742
2743         up_write(&sbi->s_nilfs->ns_segctor_sem);
2744
2745         spin_lock(&sci->sc_state_lock);
2746         nilfs_segctor_kill_thread(sci);
2747         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2748                 || sci->sc_seq_request != sci->sc_seq_done);
2749         spin_unlock(&sci->sc_state_lock);
2750
2751         if (flag || !nilfs_segctor_confirm(sci))
2752                 nilfs_segctor_write_out(sci);
2753
2754         WARN_ON(!list_empty(&sci->sc_copied_buffers));
2755
2756         if (!list_empty(&sci->sc_dirty_files)) {
2757                 nilfs_warning(sbi->s_super, __func__,
2758                               "dirty file(s) after the final construction\n");
2759                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2760         }
2761
2762         WARN_ON(!list_empty(&sci->sc_segbufs));
2763         WARN_ON(!list_empty(&sci->sc_write_logs));
2764
2765         nilfs_put_root(sci->sc_root);
2766
2767         down_write(&sbi->s_nilfs->ns_segctor_sem);
2768
2769         del_timer_sync(&sci->sc_timer);
2770         kfree(sci);
2771 }
2772
2773 /**
2774  * nilfs_attach_segment_constructor - attach a segment constructor
2775  * @sbi: nilfs_sb_info
2776  * @root: root object of the current filesystem tree
2777  *
2778  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2779  * initializes it, and starts the segment constructor.
2780  *
2781  * Return Value: On success, 0 is returned. On error, one of the following
2782  * negative error code is returned.
2783  *
2784  * %-ENOMEM - Insufficient memory available.
2785  */
2786 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
2787                                      struct nilfs_root *root)
2788 {
2789         int err;
2790
2791         if (NILFS_SC(sbi)) {
2792                 /*
2793                  * This happens if the filesystem was remounted
2794                  * read/write after nilfs_error degenerated it into a
2795                  * read-only mount.
2796                  */
2797                 nilfs_detach_segment_constructor(sbi);
2798         }
2799
2800         sbi->s_sc_info = nilfs_segctor_new(sbi, root);
2801         if (!sbi->s_sc_info)
2802                 return -ENOMEM;
2803
2804         err = nilfs_segctor_start_thread(NILFS_SC(sbi));
2805         if (err) {
2806                 kfree(sbi->s_sc_info);
2807                 sbi->s_sc_info = NULL;
2808         }
2809         return err;
2810 }
2811
2812 /**
2813  * nilfs_detach_segment_constructor - destroy the segment constructor
2814  * @sbi: nilfs_sb_info
2815  *
2816  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2817  * frees the struct nilfs_sc_info, and destroy the dirty file list.
2818  */
2819 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2820 {
2821         struct the_nilfs *nilfs = sbi->s_nilfs;
2822         LIST_HEAD(garbage_list);
2823
2824         down_write(&nilfs->ns_segctor_sem);
2825         if (NILFS_SC(sbi)) {
2826                 nilfs_segctor_destroy(NILFS_SC(sbi));
2827                 sbi->s_sc_info = NULL;
2828         }
2829
2830         /* Force to free the list of dirty files */
2831         spin_lock(&sbi->s_inode_lock);
2832         if (!list_empty(&sbi->s_dirty_files)) {
2833                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2834                 nilfs_warning(sbi->s_super, __func__,
2835                               "Non empty dirty list after the last "
2836                               "segment construction\n");
2837         }
2838         spin_unlock(&sbi->s_inode_lock);
2839         up_write(&nilfs->ns_segctor_sem);
2840
2841         nilfs_dispose_list(sbi, &garbage_list, 1);
2842 }