Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
[linux-2.6] / 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 "nilfs.h"
36 #include "btnode.h"
37 #include "page.h"
38 #include "segment.h"
39 #include "sufile.h"
40 #include "cpfile.h"
41 #include "ifile.h"
42 #include "seglist.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_HISTORY_MASK   (NILFS_CF_IFILE_STARTED)
83
84 /* Operations depending on the construction mode and file type */
85 struct nilfs_sc_operations {
86         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
87                             struct inode *);
88         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
89                             struct inode *);
90         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
91                             struct inode *);
92         void (*write_data_binfo)(struct nilfs_sc_info *,
93                                  struct nilfs_segsum_pointer *,
94                                  union nilfs_binfo *);
95         void (*write_node_binfo)(struct nilfs_sc_info *,
96                                  struct nilfs_segsum_pointer *,
97                                  union nilfs_binfo *);
98 };
99
100 /*
101  * Other definitions
102  */
103 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
104 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
105 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
106 static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
107                                int);
108
109 #define nilfs_cnt32_gt(a, b)   \
110         (typecheck(__u32, a) && typecheck(__u32, b) && \
111          ((__s32)(b) - (__s32)(a) < 0))
112 #define nilfs_cnt32_ge(a, b)   \
113         (typecheck(__u32, a) && typecheck(__u32, b) && \
114          ((__s32)(a) - (__s32)(b) >= 0))
115 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
116 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
117
118 /*
119  * Transaction
120  */
121 static struct kmem_cache *nilfs_transaction_cachep;
122
123 /**
124  * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
125  *
126  * nilfs_init_transaction_cache() creates a slab cache for the struct
127  * nilfs_transaction_info.
128  *
129  * Return Value: On success, it returns 0. On error, one of the following
130  * negative error code is returned.
131  *
132  * %-ENOMEM - Insufficient memory available.
133  */
134 int nilfs_init_transaction_cache(void)
135 {
136         nilfs_transaction_cachep =
137                 kmem_cache_create("nilfs2_transaction_cache",
138                                   sizeof(struct nilfs_transaction_info),
139                                   0, SLAB_RECLAIM_ACCOUNT, NULL);
140         return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
141 }
142
143 /**
144  * nilfs_detroy_transaction_cache - destroy the cache for transaction info
145  *
146  * nilfs_destroy_transaction_cache() frees the slab cache for the struct
147  * nilfs_transaction_info.
148  */
149 void nilfs_destroy_transaction_cache(void)
150 {
151         kmem_cache_destroy(nilfs_transaction_cachep);
152 }
153
154 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
155 {
156         struct nilfs_transaction_info *cur_ti = current->journal_info;
157         void *save = NULL;
158
159         if (cur_ti) {
160                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
161                         return ++cur_ti->ti_count;
162                 else {
163                         /*
164                          * If journal_info field is occupied by other FS,
165                          * it is saved and will be restored on
166                          * nilfs_transaction_commit().
167                          */
168                         printk(KERN_WARNING
169                                "NILFS warning: journal info from a different "
170                                "FS\n");
171                         save = current->journal_info;
172                 }
173         }
174         if (!ti) {
175                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
176                 if (!ti)
177                         return -ENOMEM;
178                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
179         } else {
180                 ti->ti_flags = 0;
181         }
182         ti->ti_count = 0;
183         ti->ti_save = save;
184         ti->ti_magic = NILFS_TI_MAGIC;
185         current->journal_info = ti;
186         return 0;
187 }
188
189 /**
190  * nilfs_transaction_begin - start indivisible file operations.
191  * @sb: super block
192  * @ti: nilfs_transaction_info
193  * @vacancy_check: flags for vacancy rate checks
194  *
195  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
196  * the segment semaphore, to make a segment construction and write tasks
197  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
198  * The region enclosed by these two functions can be nested.  To avoid a
199  * deadlock, the semaphore is only acquired or released in the outermost call.
200  *
201  * This function allocates a nilfs_transaction_info struct to keep context
202  * information on it.  It is initialized and hooked onto the current task in
203  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
204  * instead; othewise a new struct is assigned from a slab.
205  *
206  * When @vacancy_check flag is set, this function will check the amount of
207  * free space, and will wait for the GC to reclaim disk space if low capacity.
208  *
209  * Return Value: On success, 0 is returned. On error, one of the following
210  * negative error code is returned.
211  *
212  * %-ENOMEM - Insufficient memory available.
213  *
214  * %-ENOSPC - No space left on device
215  */
216 int nilfs_transaction_begin(struct super_block *sb,
217                             struct nilfs_transaction_info *ti,
218                             int vacancy_check)
219 {
220         struct nilfs_sb_info *sbi;
221         struct the_nilfs *nilfs;
222         int ret = nilfs_prepare_segment_lock(ti);
223
224         if (unlikely(ret < 0))
225                 return ret;
226         if (ret > 0)
227                 return 0;
228
229         sbi = NILFS_SB(sb);
230         nilfs = sbi->s_nilfs;
231         down_read(&nilfs->ns_segctor_sem);
232         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
233                 up_read(&nilfs->ns_segctor_sem);
234                 ret = -ENOSPC;
235                 goto failed;
236         }
237         return 0;
238
239  failed:
240         ti = current->journal_info;
241         current->journal_info = ti->ti_save;
242         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
243                 kmem_cache_free(nilfs_transaction_cachep, ti);
244         return ret;
245 }
246
247 /**
248  * nilfs_transaction_commit - commit indivisible file operations.
249  * @sb: super block
250  *
251  * nilfs_transaction_commit() releases the read semaphore which is
252  * acquired by nilfs_transaction_begin(). This is only performed
253  * in outermost call of this function.  If a commit flag is set,
254  * nilfs_transaction_commit() sets a timer to start the segment
255  * constructor.  If a sync flag is set, it starts construction
256  * directly.
257  */
258 int nilfs_transaction_commit(struct super_block *sb)
259 {
260         struct nilfs_transaction_info *ti = current->journal_info;
261         struct nilfs_sb_info *sbi;
262         struct nilfs_sc_info *sci;
263         int err = 0;
264
265         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
266         ti->ti_flags |= NILFS_TI_COMMIT;
267         if (ti->ti_count > 0) {
268                 ti->ti_count--;
269                 return 0;
270         }
271         sbi = NILFS_SB(sb);
272         sci = NILFS_SC(sbi);
273         if (sci != NULL) {
274                 if (ti->ti_flags & NILFS_TI_COMMIT)
275                         nilfs_segctor_start_timer(sci);
276                 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
277                     sci->sc_watermark)
278                         nilfs_segctor_do_flush(sci, 0);
279         }
280         up_read(&sbi->s_nilfs->ns_segctor_sem);
281         current->journal_info = ti->ti_save;
282
283         if (ti->ti_flags & NILFS_TI_SYNC)
284                 err = nilfs_construct_segment(sb);
285         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
286                 kmem_cache_free(nilfs_transaction_cachep, ti);
287         return err;
288 }
289
290 void nilfs_transaction_abort(struct super_block *sb)
291 {
292         struct nilfs_transaction_info *ti = current->journal_info;
293
294         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
295         if (ti->ti_count > 0) {
296                 ti->ti_count--;
297                 return;
298         }
299         up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
300
301         current->journal_info = ti->ti_save;
302         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
303                 kmem_cache_free(nilfs_transaction_cachep, ti);
304 }
305
306 void nilfs_relax_pressure_in_lock(struct super_block *sb)
307 {
308         struct nilfs_sb_info *sbi = NILFS_SB(sb);
309         struct nilfs_sc_info *sci = NILFS_SC(sbi);
310         struct the_nilfs *nilfs = sbi->s_nilfs;
311
312         if (!sci || !sci->sc_flush_request)
313                 return;
314
315         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
316         up_read(&nilfs->ns_segctor_sem);
317
318         down_write(&nilfs->ns_segctor_sem);
319         if (sci->sc_flush_request &&
320             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
321                 struct nilfs_transaction_info *ti = current->journal_info;
322
323                 ti->ti_flags |= NILFS_TI_WRITER;
324                 nilfs_segctor_do_immediate_flush(sci);
325                 ti->ti_flags &= ~NILFS_TI_WRITER;
326         }
327         downgrade_write(&nilfs->ns_segctor_sem);
328 }
329
330 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
331                                    struct nilfs_transaction_info *ti,
332                                    int gcflag)
333 {
334         struct nilfs_transaction_info *cur_ti = current->journal_info;
335
336         WARN_ON(cur_ti);
337         ti->ti_flags = NILFS_TI_WRITER;
338         ti->ti_count = 0;
339         ti->ti_save = cur_ti;
340         ti->ti_magic = NILFS_TI_MAGIC;
341         INIT_LIST_HEAD(&ti->ti_garbage);
342         current->journal_info = ti;
343
344         for (;;) {
345                 down_write(&sbi->s_nilfs->ns_segctor_sem);
346                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
347                         break;
348
349                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
350
351                 up_write(&sbi->s_nilfs->ns_segctor_sem);
352                 yield();
353         }
354         if (gcflag)
355                 ti->ti_flags |= NILFS_TI_GC;
356 }
357
358 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
359 {
360         struct nilfs_transaction_info *ti = current->journal_info;
361
362         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
363         BUG_ON(ti->ti_count > 0);
364
365         up_write(&sbi->s_nilfs->ns_segctor_sem);
366         current->journal_info = ti->ti_save;
367         if (!list_empty(&ti->ti_garbage))
368                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
369 }
370
371 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
372                                             struct nilfs_segsum_pointer *ssp,
373                                             unsigned bytes)
374 {
375         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
376         unsigned blocksize = sci->sc_super->s_blocksize;
377         void *p;
378
379         if (unlikely(ssp->offset + bytes > blocksize)) {
380                 ssp->offset = 0;
381                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
382                                                &segbuf->sb_segsum_buffers));
383                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
384         }
385         p = ssp->bh->b_data + ssp->offset;
386         ssp->offset += bytes;
387         return p;
388 }
389
390 /**
391  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
392  * @sci: nilfs_sc_info
393  */
394 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
395 {
396         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
397         struct buffer_head *sumbh;
398         unsigned sumbytes;
399         unsigned flags = 0;
400         int err;
401
402         if (nilfs_doing_gc())
403                 flags = NILFS_SS_GC;
404         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
405         if (unlikely(err))
406                 return err;
407
408         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
409         sumbytes = segbuf->sb_sum.sumbytes;
410         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
411         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
412         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
413         return 0;
414 }
415
416 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
417 {
418         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
419         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
420                 return -E2BIG; /* The current segment is filled up
421                                   (internal code) */
422         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
423         return nilfs_segctor_reset_segment_buffer(sci);
424 }
425
426 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
427 {
428         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
429         int err;
430
431         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
432                 err = nilfs_segctor_feed_segment(sci);
433                 if (err)
434                         return err;
435                 segbuf = sci->sc_curseg;
436         }
437         err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
438         if (likely(!err))
439                 segbuf->sb_sum.flags |= NILFS_SS_SR;
440         return err;
441 }
442
443 /*
444  * Functions for making segment summary and payloads
445  */
446 static int nilfs_segctor_segsum_block_required(
447         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
448         unsigned binfo_size)
449 {
450         unsigned blocksize = sci->sc_super->s_blocksize;
451         /* Size of finfo and binfo is enough small against blocksize */
452
453         return ssp->offset + binfo_size +
454                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
455                 blocksize;
456 }
457
458 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
459                                       struct inode *inode)
460 {
461         sci->sc_curseg->sb_sum.nfinfo++;
462         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
463         nilfs_segctor_map_segsum_entry(
464                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
465
466         if (inode->i_sb && !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
467                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
468         /* skip finfo */
469 }
470
471 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
472                                     struct inode *inode)
473 {
474         struct nilfs_finfo *finfo;
475         struct nilfs_inode_info *ii;
476         struct nilfs_segment_buffer *segbuf;
477
478         if (sci->sc_blk_cnt == 0)
479                 return;
480
481         ii = NILFS_I(inode);
482         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
483                                                  sizeof(*finfo));
484         finfo->fi_ino = cpu_to_le64(inode->i_ino);
485         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
486         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
487         finfo->fi_cno = cpu_to_le64(ii->i_cno);
488
489         segbuf = sci->sc_curseg;
490         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
491                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
492         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
493         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
494 }
495
496 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
497                                         struct buffer_head *bh,
498                                         struct inode *inode,
499                                         unsigned binfo_size)
500 {
501         struct nilfs_segment_buffer *segbuf;
502         int required, err = 0;
503
504  retry:
505         segbuf = sci->sc_curseg;
506         required = nilfs_segctor_segsum_block_required(
507                 sci, &sci->sc_binfo_ptr, binfo_size);
508         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
509                 nilfs_segctor_end_finfo(sci, inode);
510                 err = nilfs_segctor_feed_segment(sci);
511                 if (err)
512                         return err;
513                 goto retry;
514         }
515         if (unlikely(required)) {
516                 err = nilfs_segbuf_extend_segsum(segbuf);
517                 if (unlikely(err))
518                         goto failed;
519         }
520         if (sci->sc_blk_cnt == 0)
521                 nilfs_segctor_begin_finfo(sci, inode);
522
523         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
524         /* Substitution to vblocknr is delayed until update_blocknr() */
525         nilfs_segbuf_add_file_buffer(segbuf, bh);
526         sci->sc_blk_cnt++;
527  failed:
528         return err;
529 }
530
531 static int nilfs_handle_bmap_error(int err, const char *fname,
532                                    struct inode *inode, struct super_block *sb)
533 {
534         if (err == -EINVAL) {
535                 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
536                             inode->i_ino);
537                 err = -EIO;
538         }
539         return err;
540 }
541
542 /*
543  * Callback functions that enumerate, mark, and collect dirty blocks
544  */
545 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
546                                    struct buffer_head *bh, struct inode *inode)
547 {
548         int err;
549
550         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
551         if (unlikely(err < 0))
552                 return nilfs_handle_bmap_error(err, __func__, inode,
553                                                sci->sc_super);
554
555         err = nilfs_segctor_add_file_block(sci, bh, inode,
556                                            sizeof(struct nilfs_binfo_v));
557         if (!err)
558                 sci->sc_datablk_cnt++;
559         return err;
560 }
561
562 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
563                                    struct buffer_head *bh,
564                                    struct inode *inode)
565 {
566         int err;
567
568         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
569         if (unlikely(err < 0))
570                 return nilfs_handle_bmap_error(err, __func__, inode,
571                                                sci->sc_super);
572         return 0;
573 }
574
575 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
576                                    struct buffer_head *bh,
577                                    struct inode *inode)
578 {
579         WARN_ON(!buffer_dirty(bh));
580         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
581 }
582
583 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
584                                         struct nilfs_segsum_pointer *ssp,
585                                         union nilfs_binfo *binfo)
586 {
587         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
588                 sci, ssp, sizeof(*binfo_v));
589         *binfo_v = binfo->bi_v;
590 }
591
592 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
593                                         struct nilfs_segsum_pointer *ssp,
594                                         union nilfs_binfo *binfo)
595 {
596         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
597                 sci, ssp, sizeof(*vblocknr));
598         *vblocknr = binfo->bi_v.bi_vblocknr;
599 }
600
601 struct nilfs_sc_operations nilfs_sc_file_ops = {
602         .collect_data = nilfs_collect_file_data,
603         .collect_node = nilfs_collect_file_node,
604         .collect_bmap = nilfs_collect_file_bmap,
605         .write_data_binfo = nilfs_write_file_data_binfo,
606         .write_node_binfo = nilfs_write_file_node_binfo,
607 };
608
609 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
610                                   struct buffer_head *bh, struct inode *inode)
611 {
612         int err;
613
614         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
615         if (unlikely(err < 0))
616                 return nilfs_handle_bmap_error(err, __func__, inode,
617                                                sci->sc_super);
618
619         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
620         if (!err)
621                 sci->sc_datablk_cnt++;
622         return err;
623 }
624
625 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
626                                   struct buffer_head *bh, struct inode *inode)
627 {
628         WARN_ON(!buffer_dirty(bh));
629         return nilfs_segctor_add_file_block(sci, bh, inode,
630                                             sizeof(struct nilfs_binfo_dat));
631 }
632
633 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
634                                        struct nilfs_segsum_pointer *ssp,
635                                        union nilfs_binfo *binfo)
636 {
637         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
638                                                           sizeof(*blkoff));
639         *blkoff = binfo->bi_dat.bi_blkoff;
640 }
641
642 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
643                                        struct nilfs_segsum_pointer *ssp,
644                                        union nilfs_binfo *binfo)
645 {
646         struct nilfs_binfo_dat *binfo_dat =
647                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
648         *binfo_dat = binfo->bi_dat;
649 }
650
651 struct nilfs_sc_operations nilfs_sc_dat_ops = {
652         .collect_data = nilfs_collect_dat_data,
653         .collect_node = nilfs_collect_file_node,
654         .collect_bmap = nilfs_collect_dat_bmap,
655         .write_data_binfo = nilfs_write_dat_data_binfo,
656         .write_node_binfo = nilfs_write_dat_node_binfo,
657 };
658
659 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
660         .collect_data = nilfs_collect_file_data,
661         .collect_node = NULL,
662         .collect_bmap = NULL,
663         .write_data_binfo = nilfs_write_file_data_binfo,
664         .write_node_binfo = NULL,
665 };
666
667 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
668                                               struct list_head *listp,
669                                               size_t nlimit,
670                                               loff_t start, loff_t end)
671 {
672         struct address_space *mapping = inode->i_mapping;
673         struct pagevec pvec;
674         pgoff_t index = 0, last = ULONG_MAX;
675         size_t ndirties = 0;
676         int i;
677
678         if (unlikely(start != 0 || end != LLONG_MAX)) {
679                 /*
680                  * A valid range is given for sync-ing data pages. The
681                  * range is rounded to per-page; extra dirty buffers
682                  * may be included if blocksize < pagesize.
683                  */
684                 index = start >> PAGE_SHIFT;
685                 last = end >> PAGE_SHIFT;
686         }
687         pagevec_init(&pvec, 0);
688  repeat:
689         if (unlikely(index > last) ||
690             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
691                                 min_t(pgoff_t, last - index,
692                                       PAGEVEC_SIZE - 1) + 1))
693                 return ndirties;
694
695         for (i = 0; i < pagevec_count(&pvec); i++) {
696                 struct buffer_head *bh, *head;
697                 struct page *page = pvec.pages[i];
698
699                 if (unlikely(page->index > last))
700                         break;
701
702                 if (mapping->host) {
703                         lock_page(page);
704                         if (!page_has_buffers(page))
705                                 create_empty_buffers(page,
706                                                      1 << inode->i_blkbits, 0);
707                         unlock_page(page);
708                 }
709
710                 bh = head = page_buffers(page);
711                 do {
712                         if (!buffer_dirty(bh))
713                                 continue;
714                         get_bh(bh);
715                         list_add_tail(&bh->b_assoc_buffers, listp);
716                         ndirties++;
717                         if (unlikely(ndirties >= nlimit)) {
718                                 pagevec_release(&pvec);
719                                 cond_resched();
720                                 return ndirties;
721                         }
722                 } while (bh = bh->b_this_page, bh != head);
723         }
724         pagevec_release(&pvec);
725         cond_resched();
726         goto repeat;
727 }
728
729 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
730                                             struct list_head *listp)
731 {
732         struct nilfs_inode_info *ii = NILFS_I(inode);
733         struct address_space *mapping = &ii->i_btnode_cache;
734         struct pagevec pvec;
735         struct buffer_head *bh, *head;
736         unsigned int i;
737         pgoff_t index = 0;
738
739         pagevec_init(&pvec, 0);
740
741         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
742                                   PAGEVEC_SIZE)) {
743                 for (i = 0; i < pagevec_count(&pvec); i++) {
744                         bh = head = page_buffers(pvec.pages[i]);
745                         do {
746                                 if (buffer_dirty(bh)) {
747                                         get_bh(bh);
748                                         list_add_tail(&bh->b_assoc_buffers,
749                                                       listp);
750                                 }
751                                 bh = bh->b_this_page;
752                         } while (bh != head);
753                 }
754                 pagevec_release(&pvec);
755                 cond_resched();
756         }
757 }
758
759 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
760                                struct list_head *head, int force)
761 {
762         struct nilfs_inode_info *ii, *n;
763         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
764         unsigned nv = 0;
765
766         while (!list_empty(head)) {
767                 spin_lock(&sbi->s_inode_lock);
768                 list_for_each_entry_safe(ii, n, head, i_dirty) {
769                         list_del_init(&ii->i_dirty);
770                         if (force) {
771                                 if (unlikely(ii->i_bh)) {
772                                         brelse(ii->i_bh);
773                                         ii->i_bh = NULL;
774                                 }
775                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
776                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
777                                 list_add_tail(&ii->i_dirty,
778                                               &sbi->s_dirty_files);
779                                 continue;
780                         }
781                         ivec[nv++] = ii;
782                         if (nv == SC_N_INODEVEC)
783                                 break;
784                 }
785                 spin_unlock(&sbi->s_inode_lock);
786
787                 for (pii = ivec; nv > 0; pii++, nv--)
788                         iput(&(*pii)->vfs_inode);
789         }
790 }
791
792 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
793 {
794         struct the_nilfs *nilfs = sbi->s_nilfs;
795         int ret = 0;
796
797         if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
798                 ret++;
799         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
800                 ret++;
801         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
802                 ret++;
803         if (ret || nilfs_doing_gc())
804                 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
805                         ret++;
806         return ret;
807 }
808
809 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
810 {
811         return list_empty(&sci->sc_dirty_files) &&
812                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
813                 list_empty(&sci->sc_cleaning_segments) &&
814                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
815 }
816
817 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
818 {
819         struct nilfs_sb_info *sbi = sci->sc_sbi;
820         int ret = 0;
821
822         if (nilfs_test_metadata_dirty(sbi))
823                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
824
825         spin_lock(&sbi->s_inode_lock);
826         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
827                 ret++;
828
829         spin_unlock(&sbi->s_inode_lock);
830         return ret;
831 }
832
833 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
834 {
835         struct nilfs_sb_info *sbi = sci->sc_sbi;
836         struct the_nilfs *nilfs = sbi->s_nilfs;
837
838         nilfs_mdt_clear_dirty(sbi->s_ifile);
839         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
840         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
841         nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
842 }
843
844 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
845 {
846         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
847         struct buffer_head *bh_cp;
848         struct nilfs_checkpoint *raw_cp;
849         int err;
850
851         /* XXX: this interface will be changed */
852         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
853                                           &raw_cp, &bh_cp);
854         if (likely(!err)) {
855                 /* The following code is duplicated with cpfile.  But, it is
856                    needed to collect the checkpoint even if it was not newly
857                    created */
858                 nilfs_mdt_mark_buffer_dirty(bh_cp);
859                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
860                 nilfs_cpfile_put_checkpoint(
861                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
862         } else
863                 WARN_ON(err == -EINVAL || err == -ENOENT);
864
865         return err;
866 }
867
868 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
869 {
870         struct nilfs_sb_info *sbi = sci->sc_sbi;
871         struct the_nilfs *nilfs = sbi->s_nilfs;
872         struct buffer_head *bh_cp;
873         struct nilfs_checkpoint *raw_cp;
874         int err;
875
876         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
877                                           &raw_cp, &bh_cp);
878         if (unlikely(err)) {
879                 WARN_ON(err == -EINVAL || err == -ENOENT);
880                 goto failed_ibh;
881         }
882         raw_cp->cp_snapshot_list.ssl_next = 0;
883         raw_cp->cp_snapshot_list.ssl_prev = 0;
884         raw_cp->cp_inodes_count =
885                 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
886         raw_cp->cp_blocks_count =
887                 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
888         raw_cp->cp_nblk_inc =
889                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
890         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
891         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
892
893         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
894                 nilfs_checkpoint_clear_minor(raw_cp);
895         else
896                 nilfs_checkpoint_set_minor(raw_cp);
897
898         nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
899         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
900         return 0;
901
902  failed_ibh:
903         return err;
904 }
905
906 static void nilfs_fill_in_file_bmap(struct inode *ifile,
907                                     struct nilfs_inode_info *ii)
908
909 {
910         struct buffer_head *ibh;
911         struct nilfs_inode *raw_inode;
912
913         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
914                 ibh = ii->i_bh;
915                 BUG_ON(!ibh);
916                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
917                                                   ibh);
918                 nilfs_bmap_write(ii->i_bmap, raw_inode);
919                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
920         }
921 }
922
923 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
924                                             struct inode *ifile)
925 {
926         struct nilfs_inode_info *ii;
927
928         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
929                 nilfs_fill_in_file_bmap(ifile, ii);
930                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
931         }
932 }
933
934 /*
935  * CRC calculation routines
936  */
937 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
938 {
939         struct nilfs_super_root *raw_sr =
940                 (struct nilfs_super_root *)bh_sr->b_data;
941         u32 crc;
942
943         crc = crc32_le(seed,
944                        (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
945                        NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
946         raw_sr->sr_sum = cpu_to_le32(crc);
947 }
948
949 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
950                                             u32 seed)
951 {
952         struct nilfs_segment_buffer *segbuf;
953
954         if (sci->sc_super_root)
955                 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
956
957         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
958                 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
959                 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
960         }
961 }
962
963 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
964                                              struct the_nilfs *nilfs)
965 {
966         struct buffer_head *bh_sr = sci->sc_super_root;
967         struct nilfs_super_root *raw_sr =
968                 (struct nilfs_super_root *)bh_sr->b_data;
969         unsigned isz = nilfs->ns_inode_size;
970
971         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
972         raw_sr->sr_nongc_ctime
973                 = cpu_to_le64(nilfs_doing_gc() ?
974                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
975         raw_sr->sr_flags = 0;
976
977         nilfs_mdt_write_inode_direct(
978                 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
979         nilfs_mdt_write_inode_direct(
980                 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
981         nilfs_mdt_write_inode_direct(
982                 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
983 }
984
985 static void nilfs_redirty_inodes(struct list_head *head)
986 {
987         struct nilfs_inode_info *ii;
988
989         list_for_each_entry(ii, head, i_dirty) {
990                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
991                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
992         }
993 }
994
995 static void nilfs_drop_collected_inodes(struct list_head *head)
996 {
997         struct nilfs_inode_info *ii;
998
999         list_for_each_entry(ii, head, i_dirty) {
1000                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1001                         continue;
1002
1003                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
1004                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1005         }
1006 }
1007
1008 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
1009                                                struct inode *sufile)
1010
1011 {
1012         struct list_head *head = &sci->sc_cleaning_segments;
1013         struct nilfs_segment_entry *ent;
1014         int err;
1015
1016         list_for_each_entry(ent, head, list) {
1017                 if (!(ent->flags & NILFS_SLH_FREED))
1018                         break;
1019                 err = nilfs_sufile_cancel_free(sufile, ent->segnum);
1020                 WARN_ON(err); /* do not happen */
1021                 ent->flags &= ~NILFS_SLH_FREED;
1022         }
1023 }
1024
1025 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
1026                                                struct inode *sufile)
1027 {
1028         struct list_head *head = &sci->sc_cleaning_segments;
1029         struct nilfs_segment_entry *ent;
1030         int err;
1031
1032         list_for_each_entry(ent, head, list) {
1033                 err = nilfs_sufile_free(sufile, ent->segnum);
1034                 if (unlikely(err))
1035                         return err;
1036                 ent->flags |= NILFS_SLH_FREED;
1037         }
1038         return 0;
1039 }
1040
1041 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
1042 {
1043         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
1044 }
1045
1046 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1047                                        struct inode *inode,
1048                                        struct list_head *listp,
1049                                        int (*collect)(struct nilfs_sc_info *,
1050                                                       struct buffer_head *,
1051                                                       struct inode *))
1052 {
1053         struct buffer_head *bh, *n;
1054         int err = 0;
1055
1056         if (collect) {
1057                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1058                         list_del_init(&bh->b_assoc_buffers);
1059                         err = collect(sci, bh, inode);
1060                         brelse(bh);
1061                         if (unlikely(err))
1062                                 goto dispose_buffers;
1063                 }
1064                 return 0;
1065         }
1066
1067  dispose_buffers:
1068         while (!list_empty(listp)) {
1069                 bh = list_entry(listp->next, struct buffer_head,
1070                                 b_assoc_buffers);
1071                 list_del_init(&bh->b_assoc_buffers);
1072                 brelse(bh);
1073         }
1074         return err;
1075 }
1076
1077 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1078 {
1079         /* Remaining number of blocks within segment buffer */
1080         return sci->sc_segbuf_nblocks -
1081                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1082 }
1083
1084 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1085                                    struct inode *inode,
1086                                    struct nilfs_sc_operations *sc_ops)
1087 {
1088         LIST_HEAD(data_buffers);
1089         LIST_HEAD(node_buffers);
1090         int err;
1091
1092         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1093                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1094
1095                 n = nilfs_lookup_dirty_data_buffers(
1096                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1097                 if (n > rest) {
1098                         err = nilfs_segctor_apply_buffers(
1099                                 sci, inode, &data_buffers,
1100                                 sc_ops->collect_data);
1101                         BUG_ON(!err); /* always receive -E2BIG or true error */
1102                         goto break_or_fail;
1103                 }
1104         }
1105         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1106
1107         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1108                 err = nilfs_segctor_apply_buffers(
1109                         sci, inode, &data_buffers, sc_ops->collect_data);
1110                 if (unlikely(err)) {
1111                         /* dispose node list */
1112                         nilfs_segctor_apply_buffers(
1113                                 sci, inode, &node_buffers, NULL);
1114                         goto break_or_fail;
1115                 }
1116                 sci->sc_stage.flags |= NILFS_CF_NODE;
1117         }
1118         /* Collect node */
1119         err = nilfs_segctor_apply_buffers(
1120                 sci, inode, &node_buffers, sc_ops->collect_node);
1121         if (unlikely(err))
1122                 goto break_or_fail;
1123
1124         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1125         err = nilfs_segctor_apply_buffers(
1126                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1127         if (unlikely(err))
1128                 goto break_or_fail;
1129
1130         nilfs_segctor_end_finfo(sci, inode);
1131         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1132
1133  break_or_fail:
1134         return err;
1135 }
1136
1137 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1138                                          struct inode *inode)
1139 {
1140         LIST_HEAD(data_buffers);
1141         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1142         int err;
1143
1144         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1145                                             sci->sc_dsync_start,
1146                                             sci->sc_dsync_end);
1147
1148         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1149                                           nilfs_collect_file_data);
1150         if (!err) {
1151                 nilfs_segctor_end_finfo(sci, inode);
1152                 BUG_ON(n > rest);
1153                 /* always receive -E2BIG or true error if n > rest */
1154         }
1155         return err;
1156 }
1157
1158 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1159 {
1160         struct nilfs_sb_info *sbi = sci->sc_sbi;
1161         struct the_nilfs *nilfs = sbi->s_nilfs;
1162         struct list_head *head;
1163         struct nilfs_inode_info *ii;
1164         int err = 0;
1165
1166         switch (sci->sc_stage.scnt) {
1167         case NILFS_ST_INIT:
1168                 /* Pre-processes */
1169                 sci->sc_stage.flags = 0;
1170
1171                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1172                         sci->sc_nblk_inc = 0;
1173                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1174                         if (mode == SC_LSEG_DSYNC) {
1175                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1176                                 goto dsync_mode;
1177                         }
1178                 }
1179
1180                 sci->sc_stage.dirty_file_ptr = NULL;
1181                 sci->sc_stage.gc_inode_ptr = NULL;
1182                 if (mode == SC_FLUSH_DAT) {
1183                         sci->sc_stage.scnt = NILFS_ST_DAT;
1184                         goto dat_stage;
1185                 }
1186                 sci->sc_stage.scnt++;  /* Fall through */
1187         case NILFS_ST_GC:
1188                 if (nilfs_doing_gc()) {
1189                         head = &sci->sc_gc_inodes;
1190                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1191                                                 head, i_dirty);
1192                         list_for_each_entry_continue(ii, head, i_dirty) {
1193                                 err = nilfs_segctor_scan_file(
1194                                         sci, &ii->vfs_inode,
1195                                         &nilfs_sc_file_ops);
1196                                 if (unlikely(err)) {
1197                                         sci->sc_stage.gc_inode_ptr = list_entry(
1198                                                 ii->i_dirty.prev,
1199                                                 struct nilfs_inode_info,
1200                                                 i_dirty);
1201                                         goto break_or_fail;
1202                                 }
1203                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1204                         }
1205                         sci->sc_stage.gc_inode_ptr = NULL;
1206                 }
1207                 sci->sc_stage.scnt++;  /* Fall through */
1208         case NILFS_ST_FILE:
1209                 head = &sci->sc_dirty_files;
1210                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1211                                         i_dirty);
1212                 list_for_each_entry_continue(ii, head, i_dirty) {
1213                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1214
1215                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1216                                                       &nilfs_sc_file_ops);
1217                         if (unlikely(err)) {
1218                                 sci->sc_stage.dirty_file_ptr =
1219                                         list_entry(ii->i_dirty.prev,
1220                                                    struct nilfs_inode_info,
1221                                                    i_dirty);
1222                                 goto break_or_fail;
1223                         }
1224                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1225                         /* XXX: required ? */
1226                 }
1227                 sci->sc_stage.dirty_file_ptr = NULL;
1228                 if (mode == SC_FLUSH_FILE) {
1229                         sci->sc_stage.scnt = NILFS_ST_DONE;
1230                         return 0;
1231                 }
1232                 sci->sc_stage.scnt++;
1233                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1234                 /* Fall through */
1235         case NILFS_ST_IFILE:
1236                 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1237                                               &nilfs_sc_file_ops);
1238                 if (unlikely(err))
1239                         break;
1240                 sci->sc_stage.scnt++;
1241                 /* Creating a checkpoint */
1242                 err = nilfs_segctor_create_checkpoint(sci);
1243                 if (unlikely(err))
1244                         break;
1245                 /* Fall through */
1246         case NILFS_ST_CPFILE:
1247                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1248                                               &nilfs_sc_file_ops);
1249                 if (unlikely(err))
1250                         break;
1251                 sci->sc_stage.scnt++;  /* Fall through */
1252         case NILFS_ST_SUFILE:
1253                 err = nilfs_segctor_prepare_free_segments(sci,
1254                                                           nilfs->ns_sufile);
1255                 if (unlikely(err))
1256                         break;
1257                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1258                                               &nilfs_sc_file_ops);
1259                 if (unlikely(err))
1260                         break;
1261                 sci->sc_stage.scnt++;  /* Fall through */
1262         case NILFS_ST_DAT:
1263  dat_stage:
1264                 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1265                                               &nilfs_sc_dat_ops);
1266                 if (unlikely(err))
1267                         break;
1268                 if (mode == SC_FLUSH_DAT) {
1269                         sci->sc_stage.scnt = NILFS_ST_DONE;
1270                         return 0;
1271                 }
1272                 sci->sc_stage.scnt++;  /* Fall through */
1273         case NILFS_ST_SR:
1274                 if (mode == SC_LSEG_SR) {
1275                         /* Appending a super root */
1276                         err = nilfs_segctor_add_super_root(sci);
1277                         if (unlikely(err))
1278                                 break;
1279                 }
1280                 /* End of a logical segment */
1281                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1282                 sci->sc_stage.scnt = NILFS_ST_DONE;
1283                 return 0;
1284         case NILFS_ST_DSYNC:
1285  dsync_mode:
1286                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1287                 ii = sci->sc_dsync_inode;
1288                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1289                         break;
1290
1291                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1292                 if (unlikely(err))
1293                         break;
1294                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1295                 sci->sc_stage.scnt = NILFS_ST_DONE;
1296                 return 0;
1297         case NILFS_ST_DONE:
1298                 return 0;
1299         default:
1300                 BUG();
1301         }
1302
1303  break_or_fail:
1304         return err;
1305 }
1306
1307 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1308 {
1309         struct buffer_head *bh_su;
1310         struct nilfs_segment_usage *raw_su;
1311         int err;
1312
1313         err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1314         if (unlikely(err))
1315                 return err;
1316         nilfs_mdt_mark_buffer_dirty(bh_su);
1317         nilfs_mdt_mark_dirty(sufile);
1318         nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1319         return 0;
1320 }
1321
1322 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1323                                             struct the_nilfs *nilfs)
1324 {
1325         struct nilfs_segment_buffer *segbuf, *n;
1326         __u64 nextnum;
1327         int err;
1328
1329         if (list_empty(&sci->sc_segbufs)) {
1330                 segbuf = nilfs_segbuf_new(sci->sc_super);
1331                 if (unlikely(!segbuf))
1332                         return -ENOMEM;
1333                 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1334         } else
1335                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1336
1337         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, nilfs->ns_pseg_offset,
1338                          nilfs);
1339
1340         if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1341                 nilfs_shift_to_next_segment(nilfs);
1342                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1343         }
1344         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1345
1346         err = nilfs_touch_segusage(nilfs->ns_sufile, segbuf->sb_segnum);
1347         if (unlikely(err))
1348                 return err;
1349
1350         if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1351                 /* Start from the head of a new full segment */
1352                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1353                 if (unlikely(err))
1354                         return err;
1355         } else
1356                 nextnum = nilfs->ns_nextnum;
1357
1358         segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1359         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1360
1361         /* truncating segment buffers */
1362         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1363                                           sb_list) {
1364                 list_del_init(&segbuf->sb_list);
1365                 nilfs_segbuf_free(segbuf);
1366         }
1367         return 0;
1368 }
1369
1370 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1371                                          struct the_nilfs *nilfs, int nadd)
1372 {
1373         struct nilfs_segment_buffer *segbuf, *prev, *n;
1374         struct inode *sufile = nilfs->ns_sufile;
1375         __u64 nextnextnum;
1376         LIST_HEAD(list);
1377         int err, ret, i;
1378
1379         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1380         /*
1381          * Since the segment specified with nextnum might be allocated during
1382          * the previous construction, the buffer including its segusage may
1383          * not be dirty.  The following call ensures that the buffer is dirty
1384          * and will pin the buffer on memory until the sufile is written.
1385          */
1386         err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1387         if (unlikely(err))
1388                 return err;
1389
1390         for (i = 0; i < nadd; i++) {
1391                 /* extend segment info */
1392                 err = -ENOMEM;
1393                 segbuf = nilfs_segbuf_new(sci->sc_super);
1394                 if (unlikely(!segbuf))
1395                         goto failed;
1396
1397                 /* map this buffer to region of segment on-disk */
1398                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1399                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1400
1401                 /* allocate the next next full segment */
1402                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1403                 if (unlikely(err))
1404                         goto failed_segbuf;
1405
1406                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1407                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1408
1409                 list_add_tail(&segbuf->sb_list, &list);
1410                 prev = segbuf;
1411         }
1412         list_splice(&list, sci->sc_segbufs.prev);
1413         return 0;
1414
1415  failed_segbuf:
1416         nilfs_segbuf_free(segbuf);
1417  failed:
1418         list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1419                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1420                 WARN_ON(ret); /* never fails */
1421                 list_del_init(&segbuf->sb_list);
1422                 nilfs_segbuf_free(segbuf);
1423         }
1424         return err;
1425 }
1426
1427 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1428                                                    struct the_nilfs *nilfs)
1429 {
1430         struct nilfs_segment_buffer *segbuf;
1431         int ret, done = 0;
1432
1433         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1434         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1435                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1436                 WARN_ON(ret); /* never fails */
1437         }
1438         if (segbuf->sb_io_error) {
1439                 /* Case 1: The first segment failed */
1440                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1441                         /* Case 1a:  Partial segment appended into an existing
1442                            segment */
1443                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1444                                                 segbuf->sb_fseg_end);
1445                 else /* Case 1b:  New full segment */
1446                         set_nilfs_discontinued(nilfs);
1447                 done++;
1448         }
1449
1450         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1451                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1452                 WARN_ON(ret); /* never fails */
1453                 if (!done && segbuf->sb_io_error) {
1454                         if (segbuf->sb_segnum != nilfs->ns_nextnum)
1455                                 /* Case 2: extended segment (!= next) failed */
1456                                 nilfs_sufile_set_error(nilfs->ns_sufile,
1457                                                        segbuf->sb_segnum);
1458                         done++;
1459                 }
1460         }
1461 }
1462
1463 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1464 {
1465         struct nilfs_segment_buffer *segbuf;
1466
1467         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1468                 nilfs_segbuf_clear(segbuf);
1469         sci->sc_super_root = NULL;
1470 }
1471
1472 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1473 {
1474         struct nilfs_segment_buffer *segbuf;
1475
1476         while (!list_empty(&sci->sc_segbufs)) {
1477                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1478                 list_del_init(&segbuf->sb_list);
1479                 nilfs_segbuf_free(segbuf);
1480         }
1481         /* sci->sc_curseg = NULL; */
1482 }
1483
1484 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1485                                            struct the_nilfs *nilfs, int err)
1486 {
1487         if (unlikely(err)) {
1488                 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1489                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1490         }
1491         nilfs_segctor_clear_segment_buffers(sci);
1492 }
1493
1494 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1495                                           struct inode *sufile)
1496 {
1497         struct nilfs_segment_buffer *segbuf;
1498         struct buffer_head *bh_su;
1499         struct nilfs_segment_usage *raw_su;
1500         unsigned long live_blocks;
1501         int ret;
1502
1503         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1504                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1505                                                      &raw_su, &bh_su);
1506                 WARN_ON(ret); /* always succeed because bh_su is dirty */
1507                 live_blocks = segbuf->sb_sum.nblocks +
1508                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1509                 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1510                 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1511                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1512                                                bh_su);
1513         }
1514 }
1515
1516 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1517                                           struct inode *sufile)
1518 {
1519         struct nilfs_segment_buffer *segbuf;
1520         struct buffer_head *bh_su;
1521         struct nilfs_segment_usage *raw_su;
1522         int ret;
1523
1524         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1525         ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1526                                              &raw_su, &bh_su);
1527         WARN_ON(ret); /* always succeed because bh_su is dirty */
1528         raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1529                                          segbuf->sb_fseg_start);
1530         nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1531
1532         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1533                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1534                                                      &raw_su, &bh_su);
1535                 WARN_ON(ret); /* always succeed */
1536                 raw_su->su_nblocks = 0;
1537                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1538                                                bh_su);
1539         }
1540 }
1541
1542 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1543                                             struct nilfs_segment_buffer *last,
1544                                             struct inode *sufile)
1545 {
1546         struct nilfs_segment_buffer *segbuf = last, *n;
1547         int ret;
1548
1549         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1550                                           sb_list) {
1551                 list_del_init(&segbuf->sb_list);
1552                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1553                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1554                 WARN_ON(ret);
1555                 nilfs_segbuf_free(segbuf);
1556         }
1557 }
1558
1559
1560 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1561                                  struct the_nilfs *nilfs, int mode)
1562 {
1563         struct nilfs_cstage prev_stage = sci->sc_stage;
1564         int err, nadd = 1;
1565
1566         /* Collection retry loop */
1567         for (;;) {
1568                 sci->sc_super_root = NULL;
1569                 sci->sc_nblk_this_inc = 0;
1570                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1571
1572                 err = nilfs_segctor_reset_segment_buffer(sci);
1573                 if (unlikely(err))
1574                         goto failed;
1575
1576                 err = nilfs_segctor_collect_blocks(sci, mode);
1577                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1578                 if (!err)
1579                         break;
1580
1581                 if (unlikely(err != -E2BIG))
1582                         goto failed;
1583
1584                 /* The current segment is filled up */
1585                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1586                         break;
1587
1588                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1589                 nilfs_segctor_clear_segment_buffers(sci);
1590
1591                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1592                 if (unlikely(err))
1593                         return err;
1594
1595                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1596                 sci->sc_stage = prev_stage;
1597         }
1598         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1599         return 0;
1600
1601  failed:
1602         return err;
1603 }
1604
1605 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1606                                       struct buffer_head *new_bh)
1607 {
1608         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1609
1610         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1611         /* The caller must release old_bh */
1612 }
1613
1614 static int
1615 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1616                                      struct nilfs_segment_buffer *segbuf,
1617                                      int mode)
1618 {
1619         struct inode *inode = NULL;
1620         sector_t blocknr;
1621         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1622         unsigned long nblocks = 0, ndatablk = 0;
1623         struct nilfs_sc_operations *sc_op = NULL;
1624         struct nilfs_segsum_pointer ssp;
1625         struct nilfs_finfo *finfo = NULL;
1626         union nilfs_binfo binfo;
1627         struct buffer_head *bh, *bh_org;
1628         ino_t ino = 0;
1629         int err = 0;
1630
1631         if (!nfinfo)
1632                 goto out;
1633
1634         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1635         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1636         ssp.offset = sizeof(struct nilfs_segment_summary);
1637
1638         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1639                 if (bh == sci->sc_super_root)
1640                         break;
1641                 if (!finfo) {
1642                         finfo = nilfs_segctor_map_segsum_entry(
1643                                 sci, &ssp, sizeof(*finfo));
1644                         ino = le64_to_cpu(finfo->fi_ino);
1645                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1646                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1647
1648                         if (buffer_nilfs_node(bh))
1649                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1650                         else
1651                                 inode = NILFS_AS_I(bh->b_page->mapping);
1652
1653                         if (mode == SC_LSEG_DSYNC)
1654                                 sc_op = &nilfs_sc_dsync_ops;
1655                         else if (ino == NILFS_DAT_INO)
1656                                 sc_op = &nilfs_sc_dat_ops;
1657                         else /* file blocks */
1658                                 sc_op = &nilfs_sc_file_ops;
1659                 }
1660                 bh_org = bh;
1661                 get_bh(bh_org);
1662                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1663                                         &binfo);
1664                 if (bh != bh_org)
1665                         nilfs_list_replace_buffer(bh_org, bh);
1666                 brelse(bh_org);
1667                 if (unlikely(err))
1668                         goto failed_bmap;
1669
1670                 if (ndatablk > 0)
1671                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1672                 else
1673                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1674
1675                 blocknr++;
1676                 if (--nblocks == 0) {
1677                         finfo = NULL;
1678                         if (--nfinfo == 0)
1679                                 break;
1680                 } else if (ndatablk > 0)
1681                         ndatablk--;
1682         }
1683  out:
1684         return 0;
1685
1686  failed_bmap:
1687         err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1688         return err;
1689 }
1690
1691 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1692 {
1693         struct nilfs_segment_buffer *segbuf;
1694         int err;
1695
1696         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1697                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1698                 if (unlikely(err))
1699                         return err;
1700                 nilfs_segbuf_fill_in_segsum(segbuf);
1701         }
1702         return 0;
1703 }
1704
1705 static int
1706 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1707 {
1708         struct page *clone_page;
1709         struct buffer_head *bh, *head, *bh2;
1710         void *kaddr;
1711
1712         bh = head = page_buffers(page);
1713
1714         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1715         if (unlikely(!clone_page))
1716                 return -ENOMEM;
1717
1718         bh2 = page_buffers(clone_page);
1719         kaddr = kmap_atomic(page, KM_USER0);
1720         do {
1721                 if (list_empty(&bh->b_assoc_buffers))
1722                         continue;
1723                 get_bh(bh2);
1724                 page_cache_get(clone_page); /* for each bh */
1725                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1726                 bh2->b_blocknr = bh->b_blocknr;
1727                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1728                 list_add_tail(&bh->b_assoc_buffers, out);
1729         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1730         kunmap_atomic(kaddr, KM_USER0);
1731
1732         if (!TestSetPageWriteback(clone_page))
1733                 inc_zone_page_state(clone_page, NR_WRITEBACK);
1734         unlock_page(clone_page);
1735
1736         return 0;
1737 }
1738
1739 static int nilfs_test_page_to_be_frozen(struct page *page)
1740 {
1741         struct address_space *mapping = page->mapping;
1742
1743         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1744                 return 0;
1745
1746         if (page_mapped(page)) {
1747                 ClearPageChecked(page);
1748                 return 1;
1749         }
1750         return PageChecked(page);
1751 }
1752
1753 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1754 {
1755         if (!page || PageWriteback(page))
1756                 /* For split b-tree node pages, this function may be called
1757                    twice.  We ignore the 2nd or later calls by this check. */
1758                 return 0;
1759
1760         lock_page(page);
1761         clear_page_dirty_for_io(page);
1762         set_page_writeback(page);
1763         unlock_page(page);
1764
1765         if (nilfs_test_page_to_be_frozen(page)) {
1766                 int err = nilfs_copy_replace_page_buffers(page, out);
1767                 if (unlikely(err))
1768                         return err;
1769         }
1770         return 0;
1771 }
1772
1773 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1774                                        struct page **failed_page)
1775 {
1776         struct nilfs_segment_buffer *segbuf;
1777         struct page *bd_page = NULL, *fs_page = NULL;
1778         struct list_head *list = &sci->sc_copied_buffers;
1779         int err;
1780
1781         *failed_page = NULL;
1782         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1783                 struct buffer_head *bh;
1784
1785                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1786                                     b_assoc_buffers) {
1787                         if (bh->b_page != bd_page) {
1788                                 if (bd_page) {
1789                                         lock_page(bd_page);
1790                                         clear_page_dirty_for_io(bd_page);
1791                                         set_page_writeback(bd_page);
1792                                         unlock_page(bd_page);
1793                                 }
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 == sci->sc_super_root) {
1801                                 if (bh->b_page != bd_page) {
1802                                         lock_page(bd_page);
1803                                         clear_page_dirty_for_io(bd_page);
1804                                         set_page_writeback(bd_page);
1805                                         unlock_page(bd_page);
1806                                         bd_page = bh->b_page;
1807                                 }
1808                                 break;
1809                         }
1810                         if (bh->b_page != fs_page) {
1811                                 err = nilfs_begin_page_io(fs_page, list);
1812                                 if (unlikely(err)) {
1813                                         *failed_page = fs_page;
1814                                         goto out;
1815                                 }
1816                                 fs_page = bh->b_page;
1817                         }
1818                 }
1819         }
1820         if (bd_page) {
1821                 lock_page(bd_page);
1822                 clear_page_dirty_for_io(bd_page);
1823                 set_page_writeback(bd_page);
1824                 unlock_page(bd_page);
1825         }
1826         err = nilfs_begin_page_io(fs_page, list);
1827         if (unlikely(err))
1828                 *failed_page = fs_page;
1829  out:
1830         return err;
1831 }
1832
1833 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1834                                struct backing_dev_info *bdi)
1835 {
1836         struct nilfs_segment_buffer *segbuf;
1837         struct nilfs_write_info wi;
1838         int err, res;
1839
1840         wi.sb = sci->sc_super;
1841         wi.bh_sr = sci->sc_super_root;
1842         wi.bdi = bdi;
1843
1844         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1845                 nilfs_segbuf_prepare_write(segbuf, &wi);
1846                 err = nilfs_segbuf_write(segbuf, &wi);
1847
1848                 res = nilfs_segbuf_wait(segbuf, &wi);
1849                 err = unlikely(err) ? : res;
1850                 if (unlikely(err))
1851                         return err;
1852         }
1853         return 0;
1854 }
1855
1856 static int nilfs_page_has_uncleared_buffer(struct page *page)
1857 {
1858         struct buffer_head *head, *bh;
1859
1860         head = bh = page_buffers(page);
1861         do {
1862                 if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
1863                         return 1;
1864                 bh = bh->b_this_page;
1865         } while (bh != head);
1866         return 0;
1867 }
1868
1869 static void __nilfs_end_page_io(struct page *page, int err)
1870 {
1871         if (!err) {
1872                 if (!nilfs_page_buffers_clean(page))
1873                         __set_page_dirty_nobuffers(page);
1874                 ClearPageError(page);
1875         } else {
1876                 __set_page_dirty_nobuffers(page);
1877                 SetPageError(page);
1878         }
1879
1880         if (buffer_nilfs_allocated(page_buffers(page))) {
1881                 if (TestClearPageWriteback(page))
1882                         dec_zone_page_state(page, NR_WRITEBACK);
1883         } else
1884                 end_page_writeback(page);
1885 }
1886
1887 static void nilfs_end_page_io(struct page *page, int err)
1888 {
1889         if (!page)
1890                 return;
1891
1892         if (buffer_nilfs_node(page_buffers(page)) &&
1893             nilfs_page_has_uncleared_buffer(page))
1894                 /* For b-tree node pages, this function may be called twice
1895                    or more because they might be split in a segment.
1896                    This check assures that cleanup has been done for all
1897                    buffers in a split btnode page. */
1898                 return;
1899
1900         __nilfs_end_page_io(page, err);
1901 }
1902
1903 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1904 {
1905         struct buffer_head *bh, *head;
1906         struct page *page;
1907
1908         while (!list_empty(list)) {
1909                 bh = list_entry(list->next, struct buffer_head,
1910                                 b_assoc_buffers);
1911                 page = bh->b_page;
1912                 page_cache_get(page);
1913                 head = bh = page_buffers(page);
1914                 do {
1915                         if (!list_empty(&bh->b_assoc_buffers)) {
1916                                 list_del_init(&bh->b_assoc_buffers);
1917                                 if (!err) {
1918                                         set_buffer_uptodate(bh);
1919                                         clear_buffer_dirty(bh);
1920                                         clear_buffer_nilfs_volatile(bh);
1921                                 }
1922                                 brelse(bh); /* for b_assoc_buffers */
1923                         }
1924                 } while ((bh = bh->b_this_page) != head);
1925
1926                 __nilfs_end_page_io(page, err);
1927                 page_cache_release(page);
1928         }
1929 }
1930
1931 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1932                                       struct page *failed_page, int err)
1933 {
1934         struct nilfs_segment_buffer *segbuf;
1935         struct page *bd_page = NULL, *fs_page = NULL;
1936
1937         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1938                 struct buffer_head *bh;
1939
1940                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1941                                     b_assoc_buffers) {
1942                         if (bh->b_page != bd_page) {
1943                                 if (bd_page)
1944                                         end_page_writeback(bd_page);
1945                                 bd_page = bh->b_page;
1946                         }
1947                 }
1948
1949                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1950                                     b_assoc_buffers) {
1951                         if (bh == sci->sc_super_root) {
1952                                 if (bh->b_page != bd_page) {
1953                                         end_page_writeback(bd_page);
1954                                         bd_page = bh->b_page;
1955                                 }
1956                                 break;
1957                         }
1958                         if (bh->b_page != fs_page) {
1959                                 nilfs_end_page_io(fs_page, err);
1960                                 if (unlikely(fs_page == failed_page))
1961                                         goto done;
1962                                 fs_page = bh->b_page;
1963                         }
1964                 }
1965         }
1966         if (bd_page)
1967                 end_page_writeback(bd_page);
1968
1969         nilfs_end_page_io(fs_page, err);
1970  done:
1971         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1972 }
1973
1974 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1975                                    struct nilfs_segment_buffer *segbuf)
1976 {
1977         nilfs->ns_segnum = segbuf->sb_segnum;
1978         nilfs->ns_nextnum = segbuf->sb_nextnum;
1979         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1980                 + segbuf->sb_sum.nblocks;
1981         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1982         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1983 }
1984
1985 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1986 {
1987         struct nilfs_segment_buffer *segbuf;
1988         struct page *bd_page = NULL, *fs_page = NULL;
1989         struct nilfs_sb_info *sbi = sci->sc_sbi;
1990         struct the_nilfs *nilfs = sbi->s_nilfs;
1991         int update_sr = (sci->sc_super_root != NULL);
1992
1993         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1994                 struct buffer_head *bh;
1995
1996                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1997                                     b_assoc_buffers) {
1998                         set_buffer_uptodate(bh);
1999                         clear_buffer_dirty(bh);
2000                         if (bh->b_page != bd_page) {
2001                                 if (bd_page)
2002                                         end_page_writeback(bd_page);
2003                                 bd_page = bh->b_page;
2004                         }
2005                 }
2006                 /*
2007                  * We assume that the buffers which belong to the same page
2008                  * continue over the buffer list.
2009                  * Under this assumption, the last BHs of pages is
2010                  * identifiable by the discontinuity of bh->b_page
2011                  * (page != fs_page).
2012                  *
2013                  * For B-tree node blocks, however, this assumption is not
2014                  * guaranteed.  The cleanup code of B-tree node pages needs
2015                  * special care.
2016                  */
2017                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2018                                     b_assoc_buffers) {
2019                         set_buffer_uptodate(bh);
2020                         clear_buffer_dirty(bh);
2021                         clear_buffer_nilfs_volatile(bh);
2022                         if (bh == sci->sc_super_root) {
2023                                 if (bh->b_page != bd_page) {
2024                                         end_page_writeback(bd_page);
2025                                         bd_page = bh->b_page;
2026                                 }
2027                                 break;
2028                         }
2029                         if (bh->b_page != fs_page) {
2030                                 nilfs_end_page_io(fs_page, 0);
2031                                 fs_page = bh->b_page;
2032                         }
2033                 }
2034
2035                 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2036                         if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2037                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2038                                 sci->sc_lseg_stime = jiffies;
2039                         }
2040                         if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2041                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2042                 }
2043         }
2044         /*
2045          * Since pages may continue over multiple segment buffers,
2046          * end of the last page must be checked outside of the loop.
2047          */
2048         if (bd_page)
2049                 end_page_writeback(bd_page);
2050
2051         nilfs_end_page_io(fs_page, 0);
2052
2053         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2054
2055         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2056
2057         if (nilfs_doing_gc()) {
2058                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2059                 if (update_sr)
2060                         nilfs_commit_gcdat_inode(nilfs);
2061         } else
2062                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2063
2064         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2065
2066         segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2067         nilfs_set_next_segment(nilfs, segbuf);
2068
2069         if (update_sr) {
2070                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2071                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
2072                 sbi->s_super->s_dirt = 1;
2073
2074                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
2075                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2076                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2077         } else
2078                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2079 }
2080
2081 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2082                                         struct nilfs_sb_info *sbi)
2083 {
2084         struct nilfs_inode_info *ii, *n;
2085         __u64 cno = sbi->s_nilfs->ns_cno;
2086
2087         spin_lock(&sbi->s_inode_lock);
2088  retry:
2089         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2090                 if (!ii->i_bh) {
2091                         struct buffer_head *ibh;
2092                         int err;
2093
2094                         spin_unlock(&sbi->s_inode_lock);
2095                         err = nilfs_ifile_get_inode_block(
2096                                 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2097                         if (unlikely(err)) {
2098                                 nilfs_warning(sbi->s_super, __func__,
2099                                               "failed to get inode block.\n");
2100                                 return err;
2101                         }
2102                         nilfs_mdt_mark_buffer_dirty(ibh);
2103                         nilfs_mdt_mark_dirty(sbi->s_ifile);
2104                         spin_lock(&sbi->s_inode_lock);
2105                         if (likely(!ii->i_bh))
2106                                 ii->i_bh = ibh;
2107                         else
2108                                 brelse(ibh);
2109                         goto retry;
2110                 }
2111                 ii->i_cno = cno;
2112
2113                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2114                 set_bit(NILFS_I_BUSY, &ii->i_state);
2115                 list_del(&ii->i_dirty);
2116                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2117         }
2118         spin_unlock(&sbi->s_inode_lock);
2119
2120         NILFS_I(sbi->s_ifile)->i_cno = cno;
2121
2122         return 0;
2123 }
2124
2125 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2126                                           struct nilfs_sb_info *sbi)
2127 {
2128         struct nilfs_transaction_info *ti = current->journal_info;
2129         struct nilfs_inode_info *ii, *n;
2130         __u64 cno = sbi->s_nilfs->ns_cno;
2131
2132         spin_lock(&sbi->s_inode_lock);
2133         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2134                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2135                     test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2136                         /* The current checkpoint number (=nilfs->ns_cno) is
2137                            changed between check-in and check-out only if the
2138                            super root is written out.  So, we can update i_cno
2139                            for the inodes that remain in the dirty list. */
2140                         ii->i_cno = cno;
2141                         continue;
2142                 }
2143                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2144                 brelse(ii->i_bh);
2145                 ii->i_bh = NULL;
2146                 list_del(&ii->i_dirty);
2147                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2148         }
2149         spin_unlock(&sbi->s_inode_lock);
2150 }
2151
2152 /*
2153  * Main procedure of segment constructor
2154  */
2155 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2156 {
2157         struct nilfs_sb_info *sbi = sci->sc_sbi;
2158         struct the_nilfs *nilfs = sbi->s_nilfs;
2159         struct page *failed_page;
2160         int err, has_sr = 0;
2161
2162         sci->sc_stage.scnt = NILFS_ST_INIT;
2163
2164         err = nilfs_segctor_check_in_files(sci, sbi);
2165         if (unlikely(err))
2166                 goto out;
2167
2168         if (nilfs_test_metadata_dirty(sbi))
2169                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2170
2171         if (nilfs_segctor_clean(sci))
2172                 goto out;
2173
2174         do {
2175                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2176
2177                 err = nilfs_segctor_begin_construction(sci, nilfs);
2178                 if (unlikely(err))
2179                         goto out;
2180
2181                 /* Update time stamp */
2182                 sci->sc_seg_ctime = get_seconds();
2183
2184                 err = nilfs_segctor_collect(sci, nilfs, mode);
2185                 if (unlikely(err))
2186                         goto failed;
2187
2188                 has_sr = (sci->sc_super_root != NULL);
2189
2190                 /* Avoid empty segment */
2191                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2192                     NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2193                         nilfs_segctor_end_construction(sci, nilfs, 1);
2194                         goto out;
2195                 }
2196
2197                 err = nilfs_segctor_assign(sci, mode);
2198                 if (unlikely(err))
2199                         goto failed;
2200
2201                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2202                         nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2203
2204                 if (has_sr) {
2205                         err = nilfs_segctor_fill_in_checkpoint(sci);
2206                         if (unlikely(err))
2207                                 goto failed_to_make_up;
2208
2209                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2210                 }
2211                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2212
2213                 /* Write partial segments */
2214                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2215                 if (unlikely(err))
2216                         goto failed_to_write;
2217
2218                 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2219
2220                 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2221                 if (unlikely(err))
2222                         goto failed_to_write;
2223
2224                 nilfs_segctor_complete_write(sci);
2225
2226                 /* Commit segments */
2227                 if (has_sr) {
2228                         nilfs_segctor_commit_free_segments(sci);
2229                         nilfs_segctor_clear_metadata_dirty(sci);
2230                 }
2231
2232                 nilfs_segctor_end_construction(sci, nilfs, 0);
2233
2234         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2235
2236  out:
2237         nilfs_segctor_destroy_segment_buffers(sci);
2238         nilfs_segctor_check_out_files(sci, sbi);
2239         return err;
2240
2241  failed_to_write:
2242         nilfs_segctor_abort_write(sci, failed_page, err);
2243         nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2244
2245  failed_to_make_up:
2246         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2247                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2248
2249  failed:
2250         if (nilfs_doing_gc())
2251                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2252         nilfs_segctor_end_construction(sci, nilfs, err);
2253         goto out;
2254 }
2255
2256 /**
2257  * nilfs_secgtor_start_timer - set timer of background write
2258  * @sci: nilfs_sc_info
2259  *
2260  * If the timer has already been set, it ignores the new request.
2261  * This function MUST be called within a section locking the segment
2262  * semaphore.
2263  */
2264 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2265 {
2266         spin_lock(&sci->sc_state_lock);
2267         if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2268                 sci->sc_timer->expires = jiffies + sci->sc_interval;
2269                 add_timer(sci->sc_timer);
2270                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2271         }
2272         spin_unlock(&sci->sc_state_lock);
2273 }
2274
2275 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2276 {
2277         spin_lock(&sci->sc_state_lock);
2278         if (!(sci->sc_flush_request & (1 << bn))) {
2279                 unsigned long prev_req = sci->sc_flush_request;
2280
2281                 sci->sc_flush_request |= (1 << bn);
2282                 if (!prev_req)
2283                         wake_up(&sci->sc_wait_daemon);
2284         }
2285         spin_unlock(&sci->sc_state_lock);
2286 }
2287
2288 /**
2289  * nilfs_flush_segment - trigger a segment construction for resource control
2290  * @sb: super block
2291  * @ino: inode number of the file to be flushed out.
2292  */
2293 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2294 {
2295         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2296         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2297
2298         if (!sci || nilfs_doing_construction())
2299                 return;
2300         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2301                                         /* assign bit 0 to data files */
2302 }
2303
2304 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
2305                                            __u64 *segnum, size_t nsegs)
2306 {
2307         struct nilfs_segment_entry *ent;
2308         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2309         struct inode *sufile = nilfs->ns_sufile;
2310         LIST_HEAD(list);
2311         __u64 *pnum;
2312         size_t i;
2313         int err;
2314
2315         for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
2316                 ent = nilfs_alloc_segment_entry(*pnum);
2317                 if (unlikely(!ent)) {
2318                         err = -ENOMEM;
2319                         goto failed;
2320                 }
2321                 list_add_tail(&ent->list, &list);
2322
2323                 err = nilfs_open_segment_entry(ent, sufile);
2324                 if (unlikely(err))
2325                         goto failed;
2326
2327                 if (unlikely(!nilfs_segment_usage_dirty(ent->raw_su)))
2328                         printk(KERN_WARNING "NILFS: unused segment is "
2329                                "requested to be cleaned (segnum=%llu)\n",
2330                                (unsigned long long)ent->segnum);
2331                 nilfs_close_segment_entry(ent, sufile);
2332         }
2333         list_splice(&list, sci->sc_cleaning_segments.prev);
2334         return 0;
2335
2336  failed:
2337         nilfs_dispose_segment_list(&list);
2338         return err;
2339 }
2340
2341 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
2342 {
2343         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2344 }
2345
2346 struct nilfs_segctor_wait_request {
2347         wait_queue_t    wq;
2348         __u32           seq;
2349         int             err;
2350         atomic_t        done;
2351 };
2352
2353 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2354 {
2355         struct nilfs_segctor_wait_request wait_req;
2356         int err = 0;
2357
2358         spin_lock(&sci->sc_state_lock);
2359         init_wait(&wait_req.wq);
2360         wait_req.err = 0;
2361         atomic_set(&wait_req.done, 0);
2362         wait_req.seq = ++sci->sc_seq_request;
2363         spin_unlock(&sci->sc_state_lock);
2364
2365         init_waitqueue_entry(&wait_req.wq, current);
2366         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2367         set_current_state(TASK_INTERRUPTIBLE);
2368         wake_up(&sci->sc_wait_daemon);
2369
2370         for (;;) {
2371                 if (atomic_read(&wait_req.done)) {
2372                         err = wait_req.err;
2373                         break;
2374                 }
2375                 if (!signal_pending(current)) {
2376                         schedule();
2377                         continue;
2378                 }
2379                 err = -ERESTARTSYS;
2380                 break;
2381         }
2382         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2383         return err;
2384 }
2385
2386 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2387 {
2388         struct nilfs_segctor_wait_request *wrq, *n;
2389         unsigned long flags;
2390
2391         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2392         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2393                                  wq.task_list) {
2394                 if (!atomic_read(&wrq->done) &&
2395                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2396                         wrq->err = err;
2397                         atomic_set(&wrq->done, 1);
2398                 }
2399                 if (atomic_read(&wrq->done)) {
2400                         wrq->wq.func(&wrq->wq,
2401                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2402                                      0, NULL);
2403                 }
2404         }
2405         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2406 }
2407
2408 /**
2409  * nilfs_construct_segment - construct a logical segment
2410  * @sb: super block
2411  *
2412  * Return Value: On success, 0 is retured. On errors, one of the following
2413  * negative error code is returned.
2414  *
2415  * %-EROFS - Read only filesystem.
2416  *
2417  * %-EIO - I/O error
2418  *
2419  * %-ENOSPC - No space left on device (only in a panic state).
2420  *
2421  * %-ERESTARTSYS - Interrupted.
2422  *
2423  * %-ENOMEM - Insufficient memory available.
2424  */
2425 int nilfs_construct_segment(struct super_block *sb)
2426 {
2427         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2428         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2429         struct nilfs_transaction_info *ti;
2430         int err;
2431
2432         if (!sci)
2433                 return -EROFS;
2434
2435         /* A call inside transactions causes a deadlock. */
2436         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2437
2438         err = nilfs_segctor_sync(sci);
2439         return err;
2440 }
2441
2442 /**
2443  * nilfs_construct_dsync_segment - construct a data-only logical segment
2444  * @sb: super block
2445  * @inode: inode whose data blocks should be written out
2446  * @start: start byte offset
2447  * @end: end byte offset (inclusive)
2448  *
2449  * Return Value: On success, 0 is retured. On errors, one of the following
2450  * negative error code is returned.
2451  *
2452  * %-EROFS - Read only filesystem.
2453  *
2454  * %-EIO - I/O error
2455  *
2456  * %-ENOSPC - No space left on device (only in a panic state).
2457  *
2458  * %-ERESTARTSYS - Interrupted.
2459  *
2460  * %-ENOMEM - Insufficient memory available.
2461  */
2462 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2463                                   loff_t start, loff_t end)
2464 {
2465         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2466         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2467         struct nilfs_inode_info *ii;
2468         struct nilfs_transaction_info ti;
2469         int err = 0;
2470
2471         if (!sci)
2472                 return -EROFS;
2473
2474         nilfs_transaction_lock(sbi, &ti, 0);
2475
2476         ii = NILFS_I(inode);
2477         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2478             nilfs_test_opt(sbi, STRICT_ORDER) ||
2479             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2480             nilfs_discontinued(sbi->s_nilfs)) {
2481                 nilfs_transaction_unlock(sbi);
2482                 err = nilfs_segctor_sync(sci);
2483                 return err;
2484         }
2485
2486         spin_lock(&sbi->s_inode_lock);
2487         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2488             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2489                 spin_unlock(&sbi->s_inode_lock);
2490                 nilfs_transaction_unlock(sbi);
2491                 return 0;
2492         }
2493         spin_unlock(&sbi->s_inode_lock);
2494         sci->sc_dsync_inode = ii;
2495         sci->sc_dsync_start = start;
2496         sci->sc_dsync_end = end;
2497
2498         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2499
2500         nilfs_transaction_unlock(sbi);
2501         return err;
2502 }
2503
2504 struct nilfs_segctor_req {
2505         int mode;
2506         __u32 seq_accepted;
2507         int sc_err;  /* construction failure */
2508         int sb_err;  /* super block writeback failure */
2509 };
2510
2511 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2512 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2513
2514 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2515                                  struct nilfs_segctor_req *req)
2516 {
2517         req->sc_err = req->sb_err = 0;
2518         spin_lock(&sci->sc_state_lock);
2519         req->seq_accepted = sci->sc_seq_request;
2520         spin_unlock(&sci->sc_state_lock);
2521
2522         if (sci->sc_timer)
2523                 del_timer_sync(sci->sc_timer);
2524 }
2525
2526 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2527                                  struct nilfs_segctor_req *req)
2528 {
2529         /* Clear requests (even when the construction failed) */
2530         spin_lock(&sci->sc_state_lock);
2531
2532         sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2533
2534         if (req->mode == SC_LSEG_SR) {
2535                 sci->sc_seq_done = req->seq_accepted;
2536                 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2537                 sci->sc_flush_request = 0;
2538         } else if (req->mode == SC_FLUSH_FILE)
2539                 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2540         else if (req->mode == SC_FLUSH_DAT)
2541                 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2542
2543         spin_unlock(&sci->sc_state_lock);
2544 }
2545
2546 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2547                                    struct nilfs_segctor_req *req)
2548 {
2549         struct nilfs_sb_info *sbi = sci->sc_sbi;
2550         struct the_nilfs *nilfs = sbi->s_nilfs;
2551         int err = 0;
2552
2553         if (nilfs_discontinued(nilfs))
2554                 req->mode = SC_LSEG_SR;
2555         if (!nilfs_segctor_confirm(sci)) {
2556                 err = nilfs_segctor_do_construct(sci, req->mode);
2557                 req->sc_err = err;
2558         }
2559         if (likely(!err)) {
2560                 if (req->mode != SC_FLUSH_DAT)
2561                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2562                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2563                     nilfs_discontinued(nilfs)) {
2564                         down_write(&nilfs->ns_sem);
2565                         req->sb_err = nilfs_commit_super(sbi, 0);
2566                         up_write(&nilfs->ns_sem);
2567                 }
2568         }
2569         return err;
2570 }
2571
2572 static void nilfs_construction_timeout(unsigned long data)
2573 {
2574         struct task_struct *p = (struct task_struct *)data;
2575         wake_up_process(p);
2576 }
2577
2578 static void
2579 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2580 {
2581         struct nilfs_inode_info *ii, *n;
2582
2583         list_for_each_entry_safe(ii, n, head, i_dirty) {
2584                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2585                         continue;
2586                 hlist_del_init(&ii->vfs_inode.i_hash);
2587                 list_del_init(&ii->i_dirty);
2588                 nilfs_clear_gcinode(&ii->vfs_inode);
2589         }
2590 }
2591
2592 int nilfs_clean_segments(struct super_block *sb, void __user *argp)
2593 {
2594         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2595         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2596         struct the_nilfs *nilfs = sbi->s_nilfs;
2597         struct nilfs_transaction_info ti;
2598         struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2599         int err;
2600
2601         if (unlikely(!sci))
2602                 return -EROFS;
2603
2604         nilfs_transaction_lock(sbi, &ti, 1);
2605
2606         err = nilfs_init_gcdat_inode(nilfs);
2607         if (unlikely(err))
2608                 goto out_unlock;
2609         err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
2610         if (unlikely(err))
2611                 goto out_unlock;
2612
2613         list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2614
2615         for (;;) {
2616                 nilfs_segctor_accept(sci, &req);
2617                 err = nilfs_segctor_construct(sci, &req);
2618                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2619                 nilfs_segctor_notify(sci, &req);
2620
2621                 if (likely(!err))
2622                         break;
2623
2624                 nilfs_warning(sb, __func__,
2625                               "segment construction failed. (err=%d)", err);
2626                 set_current_state(TASK_INTERRUPTIBLE);
2627                 schedule_timeout(sci->sc_interval);
2628         }
2629
2630  out_unlock:
2631         nilfs_clear_gcdat_inode(nilfs);
2632         nilfs_transaction_unlock(sbi);
2633         return err;
2634 }
2635
2636 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2637 {
2638         struct nilfs_sb_info *sbi = sci->sc_sbi;
2639         struct nilfs_transaction_info ti;
2640         struct nilfs_segctor_req req = { .mode = mode };
2641
2642         nilfs_transaction_lock(sbi, &ti, 0);
2643
2644         nilfs_segctor_accept(sci, &req);
2645         nilfs_segctor_construct(sci, &req);
2646         nilfs_segctor_notify(sci, &req);
2647
2648         /*
2649          * Unclosed segment should be retried.  We do this using sc_timer.
2650          * Timeout of sc_timer will invoke complete construction which leads
2651          * to close the current logical segment.
2652          */
2653         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2654                 nilfs_segctor_start_timer(sci);
2655
2656         nilfs_transaction_unlock(sbi);
2657 }
2658
2659 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2660 {
2661         int mode = 0;
2662         int err;
2663
2664         spin_lock(&sci->sc_state_lock);
2665         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2666                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2667         spin_unlock(&sci->sc_state_lock);
2668
2669         if (mode) {
2670                 err = nilfs_segctor_do_construct(sci, mode);
2671
2672                 spin_lock(&sci->sc_state_lock);
2673                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2674                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2675                 spin_unlock(&sci->sc_state_lock);
2676         }
2677         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2678 }
2679
2680 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2681 {
2682         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2683             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2684                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2685                         return SC_FLUSH_FILE;
2686                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2687                         return SC_FLUSH_DAT;
2688         }
2689         return SC_LSEG_SR;
2690 }
2691
2692 /**
2693  * nilfs_segctor_thread - main loop of the segment constructor thread.
2694  * @arg: pointer to a struct nilfs_sc_info.
2695  *
2696  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2697  * to execute segment constructions.
2698  */
2699 static int nilfs_segctor_thread(void *arg)
2700 {
2701         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2702         struct timer_list timer;
2703         int timeout = 0;
2704
2705         init_timer(&timer);
2706         timer.data = (unsigned long)current;
2707         timer.function = nilfs_construction_timeout;
2708         sci->sc_timer = &timer;
2709
2710         /* start sync. */
2711         sci->sc_task = current;
2712         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2713         printk(KERN_INFO
2714                "segctord starting. Construction interval = %lu seconds, "
2715                "CP frequency < %lu seconds\n",
2716                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2717
2718         spin_lock(&sci->sc_state_lock);
2719  loop:
2720         for (;;) {
2721                 int mode;
2722
2723                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2724                         goto end_thread;
2725
2726                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2727                         mode = SC_LSEG_SR;
2728                 else if (!sci->sc_flush_request)
2729                         break;
2730                 else
2731                         mode = nilfs_segctor_flush_mode(sci);
2732
2733                 spin_unlock(&sci->sc_state_lock);
2734                 nilfs_segctor_thread_construct(sci, mode);
2735                 spin_lock(&sci->sc_state_lock);
2736                 timeout = 0;
2737         }
2738
2739
2740         if (freezing(current)) {
2741                 spin_unlock(&sci->sc_state_lock);
2742                 refrigerator();
2743                 spin_lock(&sci->sc_state_lock);
2744         } else {
2745                 DEFINE_WAIT(wait);
2746                 int should_sleep = 1;
2747
2748                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2749                                 TASK_INTERRUPTIBLE);
2750
2751                 if (sci->sc_seq_request != sci->sc_seq_done)
2752                         should_sleep = 0;
2753                 else if (sci->sc_flush_request)
2754                         should_sleep = 0;
2755                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2756                         should_sleep = time_before(jiffies,
2757                                                    sci->sc_timer->expires);
2758
2759                 if (should_sleep) {
2760                         spin_unlock(&sci->sc_state_lock);
2761                         schedule();
2762                         spin_lock(&sci->sc_state_lock);
2763                 }
2764                 finish_wait(&sci->sc_wait_daemon, &wait);
2765                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2766                            time_after_eq(jiffies, sci->sc_timer->expires));
2767         }
2768         goto loop;
2769
2770  end_thread:
2771         spin_unlock(&sci->sc_state_lock);
2772         del_timer_sync(sci->sc_timer);
2773         sci->sc_timer = NULL;
2774
2775         /* end sync. */
2776         sci->sc_task = NULL;
2777         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2778         return 0;
2779 }
2780
2781 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2782 {
2783         struct task_struct *t;
2784
2785         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2786         if (IS_ERR(t)) {
2787                 int err = PTR_ERR(t);
2788
2789                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2790                        err);
2791                 return err;
2792         }
2793         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2794         return 0;
2795 }
2796
2797 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2798 {
2799         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2800
2801         while (sci->sc_task) {
2802                 wake_up(&sci->sc_wait_daemon);
2803                 spin_unlock(&sci->sc_state_lock);
2804                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2805                 spin_lock(&sci->sc_state_lock);
2806         }
2807 }
2808
2809 static int nilfs_segctor_init(struct nilfs_sc_info *sci)
2810 {
2811         sci->sc_seq_done = sci->sc_seq_request;
2812
2813         return nilfs_segctor_start_thread(sci);
2814 }
2815
2816 /*
2817  * Setup & clean-up functions
2818  */
2819 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
2820 {
2821         struct nilfs_sc_info *sci;
2822
2823         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2824         if (!sci)
2825                 return NULL;
2826
2827         sci->sc_sbi = sbi;
2828         sci->sc_super = sbi->s_super;
2829
2830         init_waitqueue_head(&sci->sc_wait_request);
2831         init_waitqueue_head(&sci->sc_wait_daemon);
2832         init_waitqueue_head(&sci->sc_wait_task);
2833         spin_lock_init(&sci->sc_state_lock);
2834         INIT_LIST_HEAD(&sci->sc_dirty_files);
2835         INIT_LIST_HEAD(&sci->sc_segbufs);
2836         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2837         INIT_LIST_HEAD(&sci->sc_cleaning_segments);
2838         INIT_LIST_HEAD(&sci->sc_copied_buffers);
2839
2840         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2841         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2842         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2843
2844         if (sbi->s_interval)
2845                 sci->sc_interval = sbi->s_interval;
2846         if (sbi->s_watermark)
2847                 sci->sc_watermark = sbi->s_watermark;
2848         return sci;
2849 }
2850
2851 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2852 {
2853         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2854
2855         /* The segctord thread was stopped and its timer was removed.
2856            But some tasks remain. */
2857         do {
2858                 struct nilfs_sb_info *sbi = sci->sc_sbi;
2859                 struct nilfs_transaction_info ti;
2860                 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2861
2862                 nilfs_transaction_lock(sbi, &ti, 0);
2863                 nilfs_segctor_accept(sci, &req);
2864                 ret = nilfs_segctor_construct(sci, &req);
2865                 nilfs_segctor_notify(sci, &req);
2866                 nilfs_transaction_unlock(sbi);
2867
2868         } while (ret && retrycount-- > 0);
2869 }
2870
2871 /**
2872  * nilfs_segctor_destroy - destroy the segment constructor.
2873  * @sci: nilfs_sc_info
2874  *
2875  * nilfs_segctor_destroy() kills the segctord thread and frees
2876  * the nilfs_sc_info struct.
2877  * Caller must hold the segment semaphore.
2878  */
2879 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2880 {
2881         struct nilfs_sb_info *sbi = sci->sc_sbi;
2882         int flag;
2883
2884         up_write(&sbi->s_nilfs->ns_segctor_sem);
2885
2886         spin_lock(&sci->sc_state_lock);
2887         nilfs_segctor_kill_thread(sci);
2888         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2889                 || sci->sc_seq_request != sci->sc_seq_done);
2890         spin_unlock(&sci->sc_state_lock);
2891
2892         if (flag || nilfs_segctor_confirm(sci))
2893                 nilfs_segctor_write_out(sci);
2894
2895         WARN_ON(!list_empty(&sci->sc_copied_buffers));
2896
2897         if (!list_empty(&sci->sc_dirty_files)) {
2898                 nilfs_warning(sbi->s_super, __func__,
2899                               "dirty file(s) after the final construction\n");
2900                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2901         }
2902
2903         if (!list_empty(&sci->sc_cleaning_segments))
2904                 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2905
2906         WARN_ON(!list_empty(&sci->sc_segbufs));
2907
2908         down_write(&sbi->s_nilfs->ns_segctor_sem);
2909
2910         kfree(sci);
2911 }
2912
2913 /**
2914  * nilfs_attach_segment_constructor - attach a segment constructor
2915  * @sbi: nilfs_sb_info
2916  *
2917  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2918  * initilizes it, and starts the segment constructor.
2919  *
2920  * Return Value: On success, 0 is returned. On error, one of the following
2921  * negative error code is returned.
2922  *
2923  * %-ENOMEM - Insufficient memory available.
2924  */
2925 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi)
2926 {
2927         struct the_nilfs *nilfs = sbi->s_nilfs;
2928         int err;
2929
2930         /* Each field of nilfs_segctor is cleared through the initialization
2931            of super-block info */
2932         sbi->s_sc_info = nilfs_segctor_new(sbi);
2933         if (!sbi->s_sc_info)
2934                 return -ENOMEM;
2935
2936         nilfs_attach_writer(nilfs, sbi);
2937         err = nilfs_segctor_init(NILFS_SC(sbi));
2938         if (err) {
2939                 nilfs_detach_writer(nilfs, sbi);
2940                 kfree(sbi->s_sc_info);
2941                 sbi->s_sc_info = NULL;
2942         }
2943         return err;
2944 }
2945
2946 /**
2947  * nilfs_detach_segment_constructor - destroy the segment constructor
2948  * @sbi: nilfs_sb_info
2949  *
2950  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2951  * frees the struct nilfs_sc_info, and destroy the dirty file list.
2952  */
2953 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2954 {
2955         struct the_nilfs *nilfs = sbi->s_nilfs;
2956         LIST_HEAD(garbage_list);
2957
2958         down_write(&nilfs->ns_segctor_sem);
2959         if (NILFS_SC(sbi)) {
2960                 nilfs_segctor_destroy(NILFS_SC(sbi));
2961                 sbi->s_sc_info = NULL;
2962         }
2963
2964         /* Force to free the list of dirty files */
2965         spin_lock(&sbi->s_inode_lock);
2966         if (!list_empty(&sbi->s_dirty_files)) {
2967                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2968                 nilfs_warning(sbi->s_super, __func__,
2969                               "Non empty dirty list after the last "
2970                               "segment construction\n");
2971         }
2972         spin_unlock(&sbi->s_inode_lock);
2973         up_write(&nilfs->ns_segctor_sem);
2974
2975         nilfs_dispose_list(sbi, &garbage_list, 1);
2976         nilfs_detach_writer(nilfs, sbi);
2977 }