xref: /linux/fs/nilfs2/segment.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
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  * Written by Ryusuke Konishi.
17  *
18  */
19 
20 #include <linux/pagemap.h>
21 #include <linux/buffer_head.h>
22 #include <linux/writeback.h>
23 #include <linux/bitops.h>
24 #include <linux/bio.h>
25 #include <linux/completion.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/freezer.h>
29 #include <linux/kthread.h>
30 #include <linux/crc32.h>
31 #include <linux/pagevec.h>
32 #include <linux/slab.h>
33 #include "nilfs.h"
34 #include "btnode.h"
35 #include "page.h"
36 #include "segment.h"
37 #include "sufile.h"
38 #include "cpfile.h"
39 #include "ifile.h"
40 #include "segbuf.h"
41 
42 
43 /*
44  * Segment constructor
45  */
46 #define SC_N_INODEVEC	16   /* Size of locally allocated inode vector */
47 
48 #define SC_MAX_SEGDELTA 64   /*
49 			      * Upper limit of the number of segments
50 			      * appended in collection retry loop
51 			      */
52 
53 /* Construction mode */
54 enum {
55 	SC_LSEG_SR = 1,	/* Make a logical segment having a super root */
56 	SC_LSEG_DSYNC,	/*
57 			 * Flush data blocks of a given file and make
58 			 * a logical segment without a super root.
59 			 */
60 	SC_FLUSH_FILE,	/*
61 			 * Flush data files, leads to segment writes without
62 			 * creating a checkpoint.
63 			 */
64 	SC_FLUSH_DAT,	/*
65 			 * Flush DAT file.  This also creates segments
66 			 * without a checkpoint.
67 			 */
68 };
69 
70 /* Stage numbers of dirty block collection */
71 enum {
72 	NILFS_ST_INIT = 0,
73 	NILFS_ST_GC,		/* Collecting dirty blocks for GC */
74 	NILFS_ST_FILE,
75 	NILFS_ST_IFILE,
76 	NILFS_ST_CPFILE,
77 	NILFS_ST_SUFILE,
78 	NILFS_ST_DAT,
79 	NILFS_ST_SR,		/* Super root */
80 	NILFS_ST_DSYNC,		/* Data sync blocks */
81 	NILFS_ST_DONE,
82 };
83 
84 #define CREATE_TRACE_POINTS
85 #include <trace/events/nilfs2.h>
86 
87 /*
88  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
89  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
90  * the variable must use them because transition of stage count must involve
91  * trace events (trace_nilfs2_collection_stage_transition).
92  *
93  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
94  * produce tracepoint events. It is provided just for making the intention
95  * clear.
96  */
97 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
98 {
99 	sci->sc_stage.scnt++;
100 	trace_nilfs2_collection_stage_transition(sci);
101 }
102 
103 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
104 {
105 	sci->sc_stage.scnt = next_scnt;
106 	trace_nilfs2_collection_stage_transition(sci);
107 }
108 
109 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
110 {
111 	return sci->sc_stage.scnt;
112 }
113 
114 /* State flags of collection */
115 #define NILFS_CF_NODE		0x0001	/* Collecting node blocks */
116 #define NILFS_CF_IFILE_STARTED	0x0002	/* IFILE stage has started */
117 #define NILFS_CF_SUFREED	0x0004	/* segment usages has been freed */
118 #define NILFS_CF_HISTORY_MASK	(NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
119 
120 /* Operations depending on the construction mode and file type */
121 struct nilfs_sc_operations {
122 	int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
123 			    struct inode *);
124 	int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
125 			    struct inode *);
126 	int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
127 			    struct inode *);
128 	void (*write_data_binfo)(struct nilfs_sc_info *,
129 				 struct nilfs_segsum_pointer *,
130 				 union nilfs_binfo *);
131 	void (*write_node_binfo)(struct nilfs_sc_info *,
132 				 struct nilfs_segsum_pointer *,
133 				 union nilfs_binfo *);
134 };
135 
136 /*
137  * Other definitions
138  */
139 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
140 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
141 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
142 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
143 
144 #define nilfs_cnt32_gt(a, b)   \
145 	(typecheck(__u32, a) && typecheck(__u32, b) && \
146 	 ((__s32)(b) - (__s32)(a) < 0))
147 #define nilfs_cnt32_ge(a, b)   \
148 	(typecheck(__u32, a) && typecheck(__u32, b) && \
149 	 ((__s32)(a) - (__s32)(b) >= 0))
150 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
151 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
152 
153 static int nilfs_prepare_segment_lock(struct super_block *sb,
154 				      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 
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 		nilfs_msg(sb, KERN_WARNING, "journal info from a different FS");
169 		save = current->journal_info;
170 	}
171 	if (!ti) {
172 		ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
173 		if (!ti)
174 			return -ENOMEM;
175 		ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
176 	} else {
177 		ti->ti_flags = 0;
178 	}
179 	ti->ti_count = 0;
180 	ti->ti_save = save;
181 	ti->ti_magic = NILFS_TI_MAGIC;
182 	current->journal_info = ti;
183 	return 0;
184 }
185 
186 /**
187  * nilfs_transaction_begin - start indivisible file operations.
188  * @sb: super block
189  * @ti: nilfs_transaction_info
190  * @vacancy_check: flags for vacancy rate checks
191  *
192  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
193  * the segment semaphore, to make a segment construction and write tasks
194  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
195  * The region enclosed by these two functions can be nested.  To avoid a
196  * deadlock, the semaphore is only acquired or released in the outermost call.
197  *
198  * This function allocates a nilfs_transaction_info struct to keep context
199  * information on it.  It is initialized and hooked onto the current task in
200  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
201  * instead; otherwise a new struct is assigned from a slab.
202  *
203  * When @vacancy_check flag is set, this function will check the amount of
204  * free space, and will wait for the GC to reclaim disk space if low capacity.
205  *
206  * Return Value: On success, 0 is returned. On error, one of the following
207  * negative error code is returned.
208  *
209  * %-ENOMEM - Insufficient memory available.
210  *
211  * %-ENOSPC - No space left on device
212  */
213 int nilfs_transaction_begin(struct super_block *sb,
214 			    struct nilfs_transaction_info *ti,
215 			    int vacancy_check)
216 {
217 	struct the_nilfs *nilfs;
218 	int ret = nilfs_prepare_segment_lock(sb, ti);
219 	struct nilfs_transaction_info *trace_ti;
220 
221 	if (unlikely(ret < 0))
222 		return ret;
223 	if (ret > 0) {
224 		trace_ti = current->journal_info;
225 
226 		trace_nilfs2_transaction_transition(sb, trace_ti,
227 				    trace_ti->ti_count, trace_ti->ti_flags,
228 				    TRACE_NILFS2_TRANSACTION_BEGIN);
229 		return 0;
230 	}
231 
232 	sb_start_intwrite(sb);
233 
234 	nilfs = sb->s_fs_info;
235 	down_read(&nilfs->ns_segctor_sem);
236 	if (vacancy_check && nilfs_near_disk_full(nilfs)) {
237 		up_read(&nilfs->ns_segctor_sem);
238 		ret = -ENOSPC;
239 		goto failed;
240 	}
241 
242 	trace_ti = current->journal_info;
243 	trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
244 					    trace_ti->ti_flags,
245 					    TRACE_NILFS2_TRANSACTION_BEGIN);
246 	return 0;
247 
248  failed:
249 	ti = current->journal_info;
250 	current->journal_info = ti->ti_save;
251 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
252 		kmem_cache_free(nilfs_transaction_cachep, ti);
253 	sb_end_intwrite(sb);
254 	return ret;
255 }
256 
257 /**
258  * nilfs_transaction_commit - commit indivisible file operations.
259  * @sb: super block
260  *
261  * nilfs_transaction_commit() releases the read semaphore which is
262  * acquired by nilfs_transaction_begin(). This is only performed
263  * in outermost call of this function.  If a commit flag is set,
264  * nilfs_transaction_commit() sets a timer to start the segment
265  * constructor.  If a sync flag is set, it starts construction
266  * directly.
267  */
268 int nilfs_transaction_commit(struct super_block *sb)
269 {
270 	struct nilfs_transaction_info *ti = current->journal_info;
271 	struct the_nilfs *nilfs = sb->s_fs_info;
272 	int err = 0;
273 
274 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
275 	ti->ti_flags |= NILFS_TI_COMMIT;
276 	if (ti->ti_count > 0) {
277 		ti->ti_count--;
278 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
279 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
280 		return 0;
281 	}
282 	if (nilfs->ns_writer) {
283 		struct nilfs_sc_info *sci = nilfs->ns_writer;
284 
285 		if (ti->ti_flags & NILFS_TI_COMMIT)
286 			nilfs_segctor_start_timer(sci);
287 		if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
288 			nilfs_segctor_do_flush(sci, 0);
289 	}
290 	up_read(&nilfs->ns_segctor_sem);
291 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
292 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
293 
294 	current->journal_info = ti->ti_save;
295 
296 	if (ti->ti_flags & NILFS_TI_SYNC)
297 		err = nilfs_construct_segment(sb);
298 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
299 		kmem_cache_free(nilfs_transaction_cachep, ti);
300 	sb_end_intwrite(sb);
301 	return err;
302 }
303 
304 void nilfs_transaction_abort(struct super_block *sb)
305 {
306 	struct nilfs_transaction_info *ti = current->journal_info;
307 	struct the_nilfs *nilfs = sb->s_fs_info;
308 
309 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
310 	if (ti->ti_count > 0) {
311 		ti->ti_count--;
312 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
313 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
314 		return;
315 	}
316 	up_read(&nilfs->ns_segctor_sem);
317 
318 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
319 		    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
320 
321 	current->journal_info = ti->ti_save;
322 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
323 		kmem_cache_free(nilfs_transaction_cachep, ti);
324 	sb_end_intwrite(sb);
325 }
326 
327 void nilfs_relax_pressure_in_lock(struct super_block *sb)
328 {
329 	struct the_nilfs *nilfs = sb->s_fs_info;
330 	struct nilfs_sc_info *sci = nilfs->ns_writer;
331 
332 	if (!sci || !sci->sc_flush_request)
333 		return;
334 
335 	set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
336 	up_read(&nilfs->ns_segctor_sem);
337 
338 	down_write(&nilfs->ns_segctor_sem);
339 	if (sci->sc_flush_request &&
340 	    test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
341 		struct nilfs_transaction_info *ti = current->journal_info;
342 
343 		ti->ti_flags |= NILFS_TI_WRITER;
344 		nilfs_segctor_do_immediate_flush(sci);
345 		ti->ti_flags &= ~NILFS_TI_WRITER;
346 	}
347 	downgrade_write(&nilfs->ns_segctor_sem);
348 }
349 
350 static void nilfs_transaction_lock(struct super_block *sb,
351 				   struct nilfs_transaction_info *ti,
352 				   int gcflag)
353 {
354 	struct nilfs_transaction_info *cur_ti = current->journal_info;
355 	struct the_nilfs *nilfs = sb->s_fs_info;
356 	struct nilfs_sc_info *sci = nilfs->ns_writer;
357 
358 	WARN_ON(cur_ti);
359 	ti->ti_flags = NILFS_TI_WRITER;
360 	ti->ti_count = 0;
361 	ti->ti_save = cur_ti;
362 	ti->ti_magic = NILFS_TI_MAGIC;
363 	current->journal_info = ti;
364 
365 	for (;;) {
366 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
367 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
368 
369 		down_write(&nilfs->ns_segctor_sem);
370 		if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
371 			break;
372 
373 		nilfs_segctor_do_immediate_flush(sci);
374 
375 		up_write(&nilfs->ns_segctor_sem);
376 		cond_resched();
377 	}
378 	if (gcflag)
379 		ti->ti_flags |= NILFS_TI_GC;
380 
381 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
382 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
383 }
384 
385 static void nilfs_transaction_unlock(struct super_block *sb)
386 {
387 	struct nilfs_transaction_info *ti = current->journal_info;
388 	struct the_nilfs *nilfs = sb->s_fs_info;
389 
390 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
391 	BUG_ON(ti->ti_count > 0);
392 
393 	up_write(&nilfs->ns_segctor_sem);
394 	current->journal_info = ti->ti_save;
395 
396 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
397 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
398 }
399 
400 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
401 					    struct nilfs_segsum_pointer *ssp,
402 					    unsigned int bytes)
403 {
404 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
405 	unsigned int blocksize = sci->sc_super->s_blocksize;
406 	void *p;
407 
408 	if (unlikely(ssp->offset + bytes > blocksize)) {
409 		ssp->offset = 0;
410 		BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
411 					       &segbuf->sb_segsum_buffers));
412 		ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
413 	}
414 	p = ssp->bh->b_data + ssp->offset;
415 	ssp->offset += bytes;
416 	return p;
417 }
418 
419 /**
420  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
421  * @sci: nilfs_sc_info
422  */
423 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
424 {
425 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
426 	struct buffer_head *sumbh;
427 	unsigned int sumbytes;
428 	unsigned int flags = 0;
429 	int err;
430 
431 	if (nilfs_doing_gc())
432 		flags = NILFS_SS_GC;
433 	err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
434 	if (unlikely(err))
435 		return err;
436 
437 	sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
438 	sumbytes = segbuf->sb_sum.sumbytes;
439 	sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
440 	sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
441 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
442 	return 0;
443 }
444 
445 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
446 {
447 	sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
448 	if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
449 		return -E2BIG; /*
450 				* The current segment is filled up
451 				* (internal code)
452 				*/
453 	sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
454 	return nilfs_segctor_reset_segment_buffer(sci);
455 }
456 
457 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
458 {
459 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
460 	int err;
461 
462 	if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
463 		err = nilfs_segctor_feed_segment(sci);
464 		if (err)
465 			return err;
466 		segbuf = sci->sc_curseg;
467 	}
468 	err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
469 	if (likely(!err))
470 		segbuf->sb_sum.flags |= NILFS_SS_SR;
471 	return err;
472 }
473 
474 /*
475  * Functions for making segment summary and payloads
476  */
477 static int nilfs_segctor_segsum_block_required(
478 	struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
479 	unsigned int binfo_size)
480 {
481 	unsigned int blocksize = sci->sc_super->s_blocksize;
482 	/* Size of finfo and binfo is enough small against blocksize */
483 
484 	return ssp->offset + binfo_size +
485 		(!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
486 		blocksize;
487 }
488 
489 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
490 				      struct inode *inode)
491 {
492 	sci->sc_curseg->sb_sum.nfinfo++;
493 	sci->sc_binfo_ptr = sci->sc_finfo_ptr;
494 	nilfs_segctor_map_segsum_entry(
495 		sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
496 
497 	if (NILFS_I(inode)->i_root &&
498 	    !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
499 		set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
500 	/* skip finfo */
501 }
502 
503 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
504 				    struct inode *inode)
505 {
506 	struct nilfs_finfo *finfo;
507 	struct nilfs_inode_info *ii;
508 	struct nilfs_segment_buffer *segbuf;
509 	__u64 cno;
510 
511 	if (sci->sc_blk_cnt == 0)
512 		return;
513 
514 	ii = NILFS_I(inode);
515 
516 	if (test_bit(NILFS_I_GCINODE, &ii->i_state))
517 		cno = ii->i_cno;
518 	else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
519 		cno = 0;
520 	else
521 		cno = sci->sc_cno;
522 
523 	finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
524 						 sizeof(*finfo));
525 	finfo->fi_ino = cpu_to_le64(inode->i_ino);
526 	finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
527 	finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
528 	finfo->fi_cno = cpu_to_le64(cno);
529 
530 	segbuf = sci->sc_curseg;
531 	segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
532 		sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
533 	sci->sc_finfo_ptr = sci->sc_binfo_ptr;
534 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
535 }
536 
537 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
538 					struct buffer_head *bh,
539 					struct inode *inode,
540 					unsigned int binfo_size)
541 {
542 	struct nilfs_segment_buffer *segbuf;
543 	int required, err = 0;
544 
545  retry:
546 	segbuf = sci->sc_curseg;
547 	required = nilfs_segctor_segsum_block_required(
548 		sci, &sci->sc_binfo_ptr, binfo_size);
549 	if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
550 		nilfs_segctor_end_finfo(sci, inode);
551 		err = nilfs_segctor_feed_segment(sci);
552 		if (err)
553 			return err;
554 		goto retry;
555 	}
556 	if (unlikely(required)) {
557 		err = nilfs_segbuf_extend_segsum(segbuf);
558 		if (unlikely(err))
559 			goto failed;
560 	}
561 	if (sci->sc_blk_cnt == 0)
562 		nilfs_segctor_begin_finfo(sci, inode);
563 
564 	nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
565 	/* Substitution to vblocknr is delayed until update_blocknr() */
566 	nilfs_segbuf_add_file_buffer(segbuf, bh);
567 	sci->sc_blk_cnt++;
568  failed:
569 	return err;
570 }
571 
572 /*
573  * Callback functions that enumerate, mark, and collect dirty blocks
574  */
575 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
576 				   struct buffer_head *bh, struct inode *inode)
577 {
578 	int err;
579 
580 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
581 	if (err < 0)
582 		return err;
583 
584 	err = nilfs_segctor_add_file_block(sci, bh, inode,
585 					   sizeof(struct nilfs_binfo_v));
586 	if (!err)
587 		sci->sc_datablk_cnt++;
588 	return err;
589 }
590 
591 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
592 				   struct buffer_head *bh,
593 				   struct inode *inode)
594 {
595 	return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
596 }
597 
598 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
599 				   struct buffer_head *bh,
600 				   struct inode *inode)
601 {
602 	WARN_ON(!buffer_dirty(bh));
603 	return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
604 }
605 
606 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
607 					struct nilfs_segsum_pointer *ssp,
608 					union nilfs_binfo *binfo)
609 {
610 	struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
611 		sci, ssp, sizeof(*binfo_v));
612 	*binfo_v = binfo->bi_v;
613 }
614 
615 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
616 					struct nilfs_segsum_pointer *ssp,
617 					union nilfs_binfo *binfo)
618 {
619 	__le64 *vblocknr = nilfs_segctor_map_segsum_entry(
620 		sci, ssp, sizeof(*vblocknr));
621 	*vblocknr = binfo->bi_v.bi_vblocknr;
622 }
623 
624 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
625 	.collect_data = nilfs_collect_file_data,
626 	.collect_node = nilfs_collect_file_node,
627 	.collect_bmap = nilfs_collect_file_bmap,
628 	.write_data_binfo = nilfs_write_file_data_binfo,
629 	.write_node_binfo = nilfs_write_file_node_binfo,
630 };
631 
632 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
633 				  struct buffer_head *bh, struct inode *inode)
634 {
635 	int err;
636 
637 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
638 	if (err < 0)
639 		return err;
640 
641 	err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
642 	if (!err)
643 		sci->sc_datablk_cnt++;
644 	return err;
645 }
646 
647 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
648 				  struct buffer_head *bh, struct inode *inode)
649 {
650 	WARN_ON(!buffer_dirty(bh));
651 	return nilfs_segctor_add_file_block(sci, bh, inode,
652 					    sizeof(struct nilfs_binfo_dat));
653 }
654 
655 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
656 				       struct nilfs_segsum_pointer *ssp,
657 				       union nilfs_binfo *binfo)
658 {
659 	__le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
660 							  sizeof(*blkoff));
661 	*blkoff = binfo->bi_dat.bi_blkoff;
662 }
663 
664 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
665 				       struct nilfs_segsum_pointer *ssp,
666 				       union nilfs_binfo *binfo)
667 {
668 	struct nilfs_binfo_dat *binfo_dat =
669 		nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
670 	*binfo_dat = binfo->bi_dat;
671 }
672 
673 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
674 	.collect_data = nilfs_collect_dat_data,
675 	.collect_node = nilfs_collect_file_node,
676 	.collect_bmap = nilfs_collect_dat_bmap,
677 	.write_data_binfo = nilfs_write_dat_data_binfo,
678 	.write_node_binfo = nilfs_write_dat_node_binfo,
679 };
680 
681 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
682 	.collect_data = nilfs_collect_file_data,
683 	.collect_node = NULL,
684 	.collect_bmap = NULL,
685 	.write_data_binfo = nilfs_write_file_data_binfo,
686 	.write_node_binfo = NULL,
687 };
688 
689 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
690 					      struct list_head *listp,
691 					      size_t nlimit,
692 					      loff_t start, loff_t end)
693 {
694 	struct address_space *mapping = inode->i_mapping;
695 	struct pagevec pvec;
696 	pgoff_t index = 0, last = ULONG_MAX;
697 	size_t ndirties = 0;
698 	int i;
699 
700 	if (unlikely(start != 0 || end != LLONG_MAX)) {
701 		/*
702 		 * A valid range is given for sync-ing data pages. The
703 		 * range is rounded to per-page; extra dirty buffers
704 		 * may be included if blocksize < pagesize.
705 		 */
706 		index = start >> PAGE_SHIFT;
707 		last = end >> PAGE_SHIFT;
708 	}
709 	pagevec_init(&pvec, 0);
710  repeat:
711 	if (unlikely(index > last) ||
712 	    !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
713 				min_t(pgoff_t, last - index,
714 				      PAGEVEC_SIZE - 1) + 1))
715 		return ndirties;
716 
717 	for (i = 0; i < pagevec_count(&pvec); i++) {
718 		struct buffer_head *bh, *head;
719 		struct page *page = pvec.pages[i];
720 
721 		if (unlikely(page->index > last))
722 			break;
723 
724 		lock_page(page);
725 		if (!page_has_buffers(page))
726 			create_empty_buffers(page, 1 << inode->i_blkbits, 0);
727 		unlock_page(page);
728 
729 		bh = head = page_buffers(page);
730 		do {
731 			if (!buffer_dirty(bh) || buffer_async_write(bh))
732 				continue;
733 			get_bh(bh);
734 			list_add_tail(&bh->b_assoc_buffers, listp);
735 			ndirties++;
736 			if (unlikely(ndirties >= nlimit)) {
737 				pagevec_release(&pvec);
738 				cond_resched();
739 				return ndirties;
740 			}
741 		} while (bh = bh->b_this_page, bh != head);
742 	}
743 	pagevec_release(&pvec);
744 	cond_resched();
745 	goto repeat;
746 }
747 
748 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
749 					    struct list_head *listp)
750 {
751 	struct nilfs_inode_info *ii = NILFS_I(inode);
752 	struct address_space *mapping = &ii->i_btnode_cache;
753 	struct pagevec pvec;
754 	struct buffer_head *bh, *head;
755 	unsigned int i;
756 	pgoff_t index = 0;
757 
758 	pagevec_init(&pvec, 0);
759 
760 	while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
761 				  PAGEVEC_SIZE)) {
762 		for (i = 0; i < pagevec_count(&pvec); i++) {
763 			bh = head = page_buffers(pvec.pages[i]);
764 			do {
765 				if (buffer_dirty(bh) &&
766 						!buffer_async_write(bh)) {
767 					get_bh(bh);
768 					list_add_tail(&bh->b_assoc_buffers,
769 						      listp);
770 				}
771 				bh = bh->b_this_page;
772 			} while (bh != head);
773 		}
774 		pagevec_release(&pvec);
775 		cond_resched();
776 	}
777 }
778 
779 static void nilfs_dispose_list(struct the_nilfs *nilfs,
780 			       struct list_head *head, int force)
781 {
782 	struct nilfs_inode_info *ii, *n;
783 	struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
784 	unsigned int nv = 0;
785 
786 	while (!list_empty(head)) {
787 		spin_lock(&nilfs->ns_inode_lock);
788 		list_for_each_entry_safe(ii, n, head, i_dirty) {
789 			list_del_init(&ii->i_dirty);
790 			if (force) {
791 				if (unlikely(ii->i_bh)) {
792 					brelse(ii->i_bh);
793 					ii->i_bh = NULL;
794 				}
795 			} else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
796 				set_bit(NILFS_I_QUEUED, &ii->i_state);
797 				list_add_tail(&ii->i_dirty,
798 					      &nilfs->ns_dirty_files);
799 				continue;
800 			}
801 			ivec[nv++] = ii;
802 			if (nv == SC_N_INODEVEC)
803 				break;
804 		}
805 		spin_unlock(&nilfs->ns_inode_lock);
806 
807 		for (pii = ivec; nv > 0; pii++, nv--)
808 			iput(&(*pii)->vfs_inode);
809 	}
810 }
811 
812 static void nilfs_iput_work_func(struct work_struct *work)
813 {
814 	struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
815 						 sc_iput_work);
816 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
817 
818 	nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
819 }
820 
821 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
822 				     struct nilfs_root *root)
823 {
824 	int ret = 0;
825 
826 	if (nilfs_mdt_fetch_dirty(root->ifile))
827 		ret++;
828 	if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
829 		ret++;
830 	if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
831 		ret++;
832 	if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
833 		ret++;
834 	return ret;
835 }
836 
837 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
838 {
839 	return list_empty(&sci->sc_dirty_files) &&
840 		!test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
841 		sci->sc_nfreesegs == 0 &&
842 		(!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
843 }
844 
845 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
846 {
847 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
848 	int ret = 0;
849 
850 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
851 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
852 
853 	spin_lock(&nilfs->ns_inode_lock);
854 	if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
855 		ret++;
856 
857 	spin_unlock(&nilfs->ns_inode_lock);
858 	return ret;
859 }
860 
861 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
862 {
863 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
864 
865 	nilfs_mdt_clear_dirty(sci->sc_root->ifile);
866 	nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
867 	nilfs_mdt_clear_dirty(nilfs->ns_sufile);
868 	nilfs_mdt_clear_dirty(nilfs->ns_dat);
869 }
870 
871 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
872 {
873 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
874 	struct buffer_head *bh_cp;
875 	struct nilfs_checkpoint *raw_cp;
876 	int err;
877 
878 	/* XXX: this interface will be changed */
879 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
880 					  &raw_cp, &bh_cp);
881 	if (likely(!err)) {
882 		/*
883 		 * The following code is duplicated with cpfile.  But, it is
884 		 * needed to collect the checkpoint even if it was not newly
885 		 * created.
886 		 */
887 		mark_buffer_dirty(bh_cp);
888 		nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
889 		nilfs_cpfile_put_checkpoint(
890 			nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
891 	} else
892 		WARN_ON(err == -EINVAL || err == -ENOENT);
893 
894 	return err;
895 }
896 
897 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
898 {
899 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
900 	struct buffer_head *bh_cp;
901 	struct nilfs_checkpoint *raw_cp;
902 	int err;
903 
904 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
905 					  &raw_cp, &bh_cp);
906 	if (unlikely(err)) {
907 		WARN_ON(err == -EINVAL || err == -ENOENT);
908 		goto failed_ibh;
909 	}
910 	raw_cp->cp_snapshot_list.ssl_next = 0;
911 	raw_cp->cp_snapshot_list.ssl_prev = 0;
912 	raw_cp->cp_inodes_count =
913 		cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
914 	raw_cp->cp_blocks_count =
915 		cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
916 	raw_cp->cp_nblk_inc =
917 		cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
918 	raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
919 	raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
920 
921 	if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
922 		nilfs_checkpoint_clear_minor(raw_cp);
923 	else
924 		nilfs_checkpoint_set_minor(raw_cp);
925 
926 	nilfs_write_inode_common(sci->sc_root->ifile,
927 				 &raw_cp->cp_ifile_inode, 1);
928 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
929 	return 0;
930 
931  failed_ibh:
932 	return err;
933 }
934 
935 static void nilfs_fill_in_file_bmap(struct inode *ifile,
936 				    struct nilfs_inode_info *ii)
937 
938 {
939 	struct buffer_head *ibh;
940 	struct nilfs_inode *raw_inode;
941 
942 	if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
943 		ibh = ii->i_bh;
944 		BUG_ON(!ibh);
945 		raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
946 						  ibh);
947 		nilfs_bmap_write(ii->i_bmap, raw_inode);
948 		nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
949 	}
950 }
951 
952 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
953 {
954 	struct nilfs_inode_info *ii;
955 
956 	list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
957 		nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
958 		set_bit(NILFS_I_COLLECTED, &ii->i_state);
959 	}
960 }
961 
962 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
963 					     struct the_nilfs *nilfs)
964 {
965 	struct buffer_head *bh_sr;
966 	struct nilfs_super_root *raw_sr;
967 	unsigned int isz, srsz;
968 
969 	bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
970 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
971 	isz = nilfs->ns_inode_size;
972 	srsz = NILFS_SR_BYTES(isz);
973 
974 	raw_sr->sr_bytes = cpu_to_le16(srsz);
975 	raw_sr->sr_nongc_ctime
976 		= cpu_to_le64(nilfs_doing_gc() ?
977 			      nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
978 	raw_sr->sr_flags = 0;
979 
980 	nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
981 				 NILFS_SR_DAT_OFFSET(isz), 1);
982 	nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
983 				 NILFS_SR_CPFILE_OFFSET(isz), 1);
984 	nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
985 				 NILFS_SR_SUFILE_OFFSET(isz), 1);
986 	memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
987 }
988 
989 static void nilfs_redirty_inodes(struct list_head *head)
990 {
991 	struct nilfs_inode_info *ii;
992 
993 	list_for_each_entry(ii, head, i_dirty) {
994 		if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
995 			clear_bit(NILFS_I_COLLECTED, &ii->i_state);
996 	}
997 }
998 
999 static void nilfs_drop_collected_inodes(struct list_head *head)
1000 {
1001 	struct nilfs_inode_info *ii;
1002 
1003 	list_for_each_entry(ii, head, i_dirty) {
1004 		if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1005 			continue;
1006 
1007 		clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1008 		set_bit(NILFS_I_UPDATED, &ii->i_state);
1009 	}
1010 }
1011 
1012 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1013 				       struct inode *inode,
1014 				       struct list_head *listp,
1015 				       int (*collect)(struct nilfs_sc_info *,
1016 						      struct buffer_head *,
1017 						      struct inode *))
1018 {
1019 	struct buffer_head *bh, *n;
1020 	int err = 0;
1021 
1022 	if (collect) {
1023 		list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1024 			list_del_init(&bh->b_assoc_buffers);
1025 			err = collect(sci, bh, inode);
1026 			brelse(bh);
1027 			if (unlikely(err))
1028 				goto dispose_buffers;
1029 		}
1030 		return 0;
1031 	}
1032 
1033  dispose_buffers:
1034 	while (!list_empty(listp)) {
1035 		bh = list_first_entry(listp, struct buffer_head,
1036 				      b_assoc_buffers);
1037 		list_del_init(&bh->b_assoc_buffers);
1038 		brelse(bh);
1039 	}
1040 	return err;
1041 }
1042 
1043 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1044 {
1045 	/* Remaining number of blocks within segment buffer */
1046 	return sci->sc_segbuf_nblocks -
1047 		(sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1048 }
1049 
1050 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1051 				   struct inode *inode,
1052 				   const struct nilfs_sc_operations *sc_ops)
1053 {
1054 	LIST_HEAD(data_buffers);
1055 	LIST_HEAD(node_buffers);
1056 	int err;
1057 
1058 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1059 		size_t n, rest = nilfs_segctor_buffer_rest(sci);
1060 
1061 		n = nilfs_lookup_dirty_data_buffers(
1062 			inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1063 		if (n > rest) {
1064 			err = nilfs_segctor_apply_buffers(
1065 				sci, inode, &data_buffers,
1066 				sc_ops->collect_data);
1067 			BUG_ON(!err); /* always receive -E2BIG or true error */
1068 			goto break_or_fail;
1069 		}
1070 	}
1071 	nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1072 
1073 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1074 		err = nilfs_segctor_apply_buffers(
1075 			sci, inode, &data_buffers, sc_ops->collect_data);
1076 		if (unlikely(err)) {
1077 			/* dispose node list */
1078 			nilfs_segctor_apply_buffers(
1079 				sci, inode, &node_buffers, NULL);
1080 			goto break_or_fail;
1081 		}
1082 		sci->sc_stage.flags |= NILFS_CF_NODE;
1083 	}
1084 	/* Collect node */
1085 	err = nilfs_segctor_apply_buffers(
1086 		sci, inode, &node_buffers, sc_ops->collect_node);
1087 	if (unlikely(err))
1088 		goto break_or_fail;
1089 
1090 	nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1091 	err = nilfs_segctor_apply_buffers(
1092 		sci, inode, &node_buffers, sc_ops->collect_bmap);
1093 	if (unlikely(err))
1094 		goto break_or_fail;
1095 
1096 	nilfs_segctor_end_finfo(sci, inode);
1097 	sci->sc_stage.flags &= ~NILFS_CF_NODE;
1098 
1099  break_or_fail:
1100 	return err;
1101 }
1102 
1103 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1104 					 struct inode *inode)
1105 {
1106 	LIST_HEAD(data_buffers);
1107 	size_t n, rest = nilfs_segctor_buffer_rest(sci);
1108 	int err;
1109 
1110 	n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1111 					    sci->sc_dsync_start,
1112 					    sci->sc_dsync_end);
1113 
1114 	err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1115 					  nilfs_collect_file_data);
1116 	if (!err) {
1117 		nilfs_segctor_end_finfo(sci, inode);
1118 		BUG_ON(n > rest);
1119 		/* always receive -E2BIG or true error if n > rest */
1120 	}
1121 	return err;
1122 }
1123 
1124 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1125 {
1126 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1127 	struct list_head *head;
1128 	struct nilfs_inode_info *ii;
1129 	size_t ndone;
1130 	int err = 0;
1131 
1132 	switch (nilfs_sc_cstage_get(sci)) {
1133 	case NILFS_ST_INIT:
1134 		/* Pre-processes */
1135 		sci->sc_stage.flags = 0;
1136 
1137 		if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1138 			sci->sc_nblk_inc = 0;
1139 			sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1140 			if (mode == SC_LSEG_DSYNC) {
1141 				nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1142 				goto dsync_mode;
1143 			}
1144 		}
1145 
1146 		sci->sc_stage.dirty_file_ptr = NULL;
1147 		sci->sc_stage.gc_inode_ptr = NULL;
1148 		if (mode == SC_FLUSH_DAT) {
1149 			nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1150 			goto dat_stage;
1151 		}
1152 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1153 	case NILFS_ST_GC:
1154 		if (nilfs_doing_gc()) {
1155 			head = &sci->sc_gc_inodes;
1156 			ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1157 						head, i_dirty);
1158 			list_for_each_entry_continue(ii, head, i_dirty) {
1159 				err = nilfs_segctor_scan_file(
1160 					sci, &ii->vfs_inode,
1161 					&nilfs_sc_file_ops);
1162 				if (unlikely(err)) {
1163 					sci->sc_stage.gc_inode_ptr = list_entry(
1164 						ii->i_dirty.prev,
1165 						struct nilfs_inode_info,
1166 						i_dirty);
1167 					goto break_or_fail;
1168 				}
1169 				set_bit(NILFS_I_COLLECTED, &ii->i_state);
1170 			}
1171 			sci->sc_stage.gc_inode_ptr = NULL;
1172 		}
1173 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1174 	case NILFS_ST_FILE:
1175 		head = &sci->sc_dirty_files;
1176 		ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1177 					i_dirty);
1178 		list_for_each_entry_continue(ii, head, i_dirty) {
1179 			clear_bit(NILFS_I_DIRTY, &ii->i_state);
1180 
1181 			err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1182 						      &nilfs_sc_file_ops);
1183 			if (unlikely(err)) {
1184 				sci->sc_stage.dirty_file_ptr =
1185 					list_entry(ii->i_dirty.prev,
1186 						   struct nilfs_inode_info,
1187 						   i_dirty);
1188 				goto break_or_fail;
1189 			}
1190 			/* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1191 			/* XXX: required ? */
1192 		}
1193 		sci->sc_stage.dirty_file_ptr = NULL;
1194 		if (mode == SC_FLUSH_FILE) {
1195 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1196 			return 0;
1197 		}
1198 		nilfs_sc_cstage_inc(sci);
1199 		sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1200 		/* Fall through */
1201 	case NILFS_ST_IFILE:
1202 		err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1203 					      &nilfs_sc_file_ops);
1204 		if (unlikely(err))
1205 			break;
1206 		nilfs_sc_cstage_inc(sci);
1207 		/* Creating a checkpoint */
1208 		err = nilfs_segctor_create_checkpoint(sci);
1209 		if (unlikely(err))
1210 			break;
1211 		/* Fall through */
1212 	case NILFS_ST_CPFILE:
1213 		err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1214 					      &nilfs_sc_file_ops);
1215 		if (unlikely(err))
1216 			break;
1217 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1218 	case NILFS_ST_SUFILE:
1219 		err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1220 					 sci->sc_nfreesegs, &ndone);
1221 		if (unlikely(err)) {
1222 			nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1223 						  sci->sc_freesegs, ndone,
1224 						  NULL);
1225 			break;
1226 		}
1227 		sci->sc_stage.flags |= NILFS_CF_SUFREED;
1228 
1229 		err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1230 					      &nilfs_sc_file_ops);
1231 		if (unlikely(err))
1232 			break;
1233 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1234 	case NILFS_ST_DAT:
1235  dat_stage:
1236 		err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1237 					      &nilfs_sc_dat_ops);
1238 		if (unlikely(err))
1239 			break;
1240 		if (mode == SC_FLUSH_DAT) {
1241 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1242 			return 0;
1243 		}
1244 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1245 	case NILFS_ST_SR:
1246 		if (mode == SC_LSEG_SR) {
1247 			/* Appending a super root */
1248 			err = nilfs_segctor_add_super_root(sci);
1249 			if (unlikely(err))
1250 				break;
1251 		}
1252 		/* End of a logical segment */
1253 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1254 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1255 		return 0;
1256 	case NILFS_ST_DSYNC:
1257  dsync_mode:
1258 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1259 		ii = sci->sc_dsync_inode;
1260 		if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1261 			break;
1262 
1263 		err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1264 		if (unlikely(err))
1265 			break;
1266 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1267 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1268 		return 0;
1269 	case NILFS_ST_DONE:
1270 		return 0;
1271 	default:
1272 		BUG();
1273 	}
1274 
1275  break_or_fail:
1276 	return err;
1277 }
1278 
1279 /**
1280  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1281  * @sci: nilfs_sc_info
1282  * @nilfs: nilfs object
1283  */
1284 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1285 					    struct the_nilfs *nilfs)
1286 {
1287 	struct nilfs_segment_buffer *segbuf, *prev;
1288 	__u64 nextnum;
1289 	int err, alloc = 0;
1290 
1291 	segbuf = nilfs_segbuf_new(sci->sc_super);
1292 	if (unlikely(!segbuf))
1293 		return -ENOMEM;
1294 
1295 	if (list_empty(&sci->sc_write_logs)) {
1296 		nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1297 				 nilfs->ns_pseg_offset, nilfs);
1298 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1299 			nilfs_shift_to_next_segment(nilfs);
1300 			nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1301 		}
1302 
1303 		segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1304 		nextnum = nilfs->ns_nextnum;
1305 
1306 		if (nilfs->ns_segnum == nilfs->ns_nextnum)
1307 			/* Start from the head of a new full segment */
1308 			alloc++;
1309 	} else {
1310 		/* Continue logs */
1311 		prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1312 		nilfs_segbuf_map_cont(segbuf, prev);
1313 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1314 		nextnum = prev->sb_nextnum;
1315 
1316 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1317 			nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1318 			segbuf->sb_sum.seg_seq++;
1319 			alloc++;
1320 		}
1321 	}
1322 
1323 	err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1324 	if (err)
1325 		goto failed;
1326 
1327 	if (alloc) {
1328 		err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1329 		if (err)
1330 			goto failed;
1331 	}
1332 	nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1333 
1334 	BUG_ON(!list_empty(&sci->sc_segbufs));
1335 	list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1336 	sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1337 	return 0;
1338 
1339  failed:
1340 	nilfs_segbuf_free(segbuf);
1341 	return err;
1342 }
1343 
1344 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1345 					 struct the_nilfs *nilfs, int nadd)
1346 {
1347 	struct nilfs_segment_buffer *segbuf, *prev;
1348 	struct inode *sufile = nilfs->ns_sufile;
1349 	__u64 nextnextnum;
1350 	LIST_HEAD(list);
1351 	int err, ret, i;
1352 
1353 	prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1354 	/*
1355 	 * Since the segment specified with nextnum might be allocated during
1356 	 * the previous construction, the buffer including its segusage may
1357 	 * not be dirty.  The following call ensures that the buffer is dirty
1358 	 * and will pin the buffer on memory until the sufile is written.
1359 	 */
1360 	err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1361 	if (unlikely(err))
1362 		return err;
1363 
1364 	for (i = 0; i < nadd; i++) {
1365 		/* extend segment info */
1366 		err = -ENOMEM;
1367 		segbuf = nilfs_segbuf_new(sci->sc_super);
1368 		if (unlikely(!segbuf))
1369 			goto failed;
1370 
1371 		/* map this buffer to region of segment on-disk */
1372 		nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1373 		sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1374 
1375 		/* allocate the next next full segment */
1376 		err = nilfs_sufile_alloc(sufile, &nextnextnum);
1377 		if (unlikely(err))
1378 			goto failed_segbuf;
1379 
1380 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1381 		nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1382 
1383 		list_add_tail(&segbuf->sb_list, &list);
1384 		prev = segbuf;
1385 	}
1386 	list_splice_tail(&list, &sci->sc_segbufs);
1387 	return 0;
1388 
1389  failed_segbuf:
1390 	nilfs_segbuf_free(segbuf);
1391  failed:
1392 	list_for_each_entry(segbuf, &list, sb_list) {
1393 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1394 		WARN_ON(ret); /* never fails */
1395 	}
1396 	nilfs_destroy_logs(&list);
1397 	return err;
1398 }
1399 
1400 static void nilfs_free_incomplete_logs(struct list_head *logs,
1401 				       struct the_nilfs *nilfs)
1402 {
1403 	struct nilfs_segment_buffer *segbuf, *prev;
1404 	struct inode *sufile = nilfs->ns_sufile;
1405 	int ret;
1406 
1407 	segbuf = NILFS_FIRST_SEGBUF(logs);
1408 	if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1409 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1410 		WARN_ON(ret); /* never fails */
1411 	}
1412 	if (atomic_read(&segbuf->sb_err)) {
1413 		/* Case 1: The first segment failed */
1414 		if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1415 			/*
1416 			 * Case 1a:  Partial segment appended into an existing
1417 			 * segment
1418 			 */
1419 			nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1420 						segbuf->sb_fseg_end);
1421 		else /* Case 1b:  New full segment */
1422 			set_nilfs_discontinued(nilfs);
1423 	}
1424 
1425 	prev = segbuf;
1426 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1427 		if (prev->sb_nextnum != segbuf->sb_nextnum) {
1428 			ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1429 			WARN_ON(ret); /* never fails */
1430 		}
1431 		if (atomic_read(&segbuf->sb_err) &&
1432 		    segbuf->sb_segnum != nilfs->ns_nextnum)
1433 			/* Case 2: extended segment (!= next) failed */
1434 			nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1435 		prev = segbuf;
1436 	}
1437 }
1438 
1439 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1440 					  struct inode *sufile)
1441 {
1442 	struct nilfs_segment_buffer *segbuf;
1443 	unsigned long live_blocks;
1444 	int ret;
1445 
1446 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1447 		live_blocks = segbuf->sb_sum.nblocks +
1448 			(segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1449 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1450 						     live_blocks,
1451 						     sci->sc_seg_ctime);
1452 		WARN_ON(ret); /* always succeed because the segusage is dirty */
1453 	}
1454 }
1455 
1456 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1457 {
1458 	struct nilfs_segment_buffer *segbuf;
1459 	int ret;
1460 
1461 	segbuf = NILFS_FIRST_SEGBUF(logs);
1462 	ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1463 					     segbuf->sb_pseg_start -
1464 					     segbuf->sb_fseg_start, 0);
1465 	WARN_ON(ret); /* always succeed because the segusage is dirty */
1466 
1467 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1468 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1469 						     0, 0);
1470 		WARN_ON(ret); /* always succeed */
1471 	}
1472 }
1473 
1474 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1475 					    struct nilfs_segment_buffer *last,
1476 					    struct inode *sufile)
1477 {
1478 	struct nilfs_segment_buffer *segbuf = last;
1479 	int ret;
1480 
1481 	list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1482 		sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1483 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1484 		WARN_ON(ret);
1485 	}
1486 	nilfs_truncate_logs(&sci->sc_segbufs, last);
1487 }
1488 
1489 
1490 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1491 				 struct the_nilfs *nilfs, int mode)
1492 {
1493 	struct nilfs_cstage prev_stage = sci->sc_stage;
1494 	int err, nadd = 1;
1495 
1496 	/* Collection retry loop */
1497 	for (;;) {
1498 		sci->sc_nblk_this_inc = 0;
1499 		sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1500 
1501 		err = nilfs_segctor_reset_segment_buffer(sci);
1502 		if (unlikely(err))
1503 			goto failed;
1504 
1505 		err = nilfs_segctor_collect_blocks(sci, mode);
1506 		sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1507 		if (!err)
1508 			break;
1509 
1510 		if (unlikely(err != -E2BIG))
1511 			goto failed;
1512 
1513 		/* The current segment is filled up */
1514 		if (mode != SC_LSEG_SR ||
1515 		    nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1516 			break;
1517 
1518 		nilfs_clear_logs(&sci->sc_segbufs);
1519 
1520 		if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1521 			err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1522 							sci->sc_freesegs,
1523 							sci->sc_nfreesegs,
1524 							NULL);
1525 			WARN_ON(err); /* do not happen */
1526 			sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1527 		}
1528 
1529 		err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1530 		if (unlikely(err))
1531 			return err;
1532 
1533 		nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1534 		sci->sc_stage = prev_stage;
1535 	}
1536 	nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1537 	return 0;
1538 
1539  failed:
1540 	return err;
1541 }
1542 
1543 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1544 				      struct buffer_head *new_bh)
1545 {
1546 	BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1547 
1548 	list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1549 	/* The caller must release old_bh */
1550 }
1551 
1552 static int
1553 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1554 				     struct nilfs_segment_buffer *segbuf,
1555 				     int mode)
1556 {
1557 	struct inode *inode = NULL;
1558 	sector_t blocknr;
1559 	unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1560 	unsigned long nblocks = 0, ndatablk = 0;
1561 	const struct nilfs_sc_operations *sc_op = NULL;
1562 	struct nilfs_segsum_pointer ssp;
1563 	struct nilfs_finfo *finfo = NULL;
1564 	union nilfs_binfo binfo;
1565 	struct buffer_head *bh, *bh_org;
1566 	ino_t ino = 0;
1567 	int err = 0;
1568 
1569 	if (!nfinfo)
1570 		goto out;
1571 
1572 	blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1573 	ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1574 	ssp.offset = sizeof(struct nilfs_segment_summary);
1575 
1576 	list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1577 		if (bh == segbuf->sb_super_root)
1578 			break;
1579 		if (!finfo) {
1580 			finfo =	nilfs_segctor_map_segsum_entry(
1581 				sci, &ssp, sizeof(*finfo));
1582 			ino = le64_to_cpu(finfo->fi_ino);
1583 			nblocks = le32_to_cpu(finfo->fi_nblocks);
1584 			ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1585 
1586 			inode = bh->b_page->mapping->host;
1587 
1588 			if (mode == SC_LSEG_DSYNC)
1589 				sc_op = &nilfs_sc_dsync_ops;
1590 			else if (ino == NILFS_DAT_INO)
1591 				sc_op = &nilfs_sc_dat_ops;
1592 			else /* file blocks */
1593 				sc_op = &nilfs_sc_file_ops;
1594 		}
1595 		bh_org = bh;
1596 		get_bh(bh_org);
1597 		err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1598 					&binfo);
1599 		if (bh != bh_org)
1600 			nilfs_list_replace_buffer(bh_org, bh);
1601 		brelse(bh_org);
1602 		if (unlikely(err))
1603 			goto failed_bmap;
1604 
1605 		if (ndatablk > 0)
1606 			sc_op->write_data_binfo(sci, &ssp, &binfo);
1607 		else
1608 			sc_op->write_node_binfo(sci, &ssp, &binfo);
1609 
1610 		blocknr++;
1611 		if (--nblocks == 0) {
1612 			finfo = NULL;
1613 			if (--nfinfo == 0)
1614 				break;
1615 		} else if (ndatablk > 0)
1616 			ndatablk--;
1617 	}
1618  out:
1619 	return 0;
1620 
1621  failed_bmap:
1622 	return err;
1623 }
1624 
1625 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1626 {
1627 	struct nilfs_segment_buffer *segbuf;
1628 	int err;
1629 
1630 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1631 		err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1632 		if (unlikely(err))
1633 			return err;
1634 		nilfs_segbuf_fill_in_segsum(segbuf);
1635 	}
1636 	return 0;
1637 }
1638 
1639 static void nilfs_begin_page_io(struct page *page)
1640 {
1641 	if (!page || PageWriteback(page))
1642 		/*
1643 		 * For split b-tree node pages, this function may be called
1644 		 * twice.  We ignore the 2nd or later calls by this check.
1645 		 */
1646 		return;
1647 
1648 	lock_page(page);
1649 	clear_page_dirty_for_io(page);
1650 	set_page_writeback(page);
1651 	unlock_page(page);
1652 }
1653 
1654 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1655 {
1656 	struct nilfs_segment_buffer *segbuf;
1657 	struct page *bd_page = NULL, *fs_page = NULL;
1658 
1659 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1660 		struct buffer_head *bh;
1661 
1662 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1663 				    b_assoc_buffers) {
1664 			if (bh->b_page != bd_page) {
1665 				if (bd_page) {
1666 					lock_page(bd_page);
1667 					clear_page_dirty_for_io(bd_page);
1668 					set_page_writeback(bd_page);
1669 					unlock_page(bd_page);
1670 				}
1671 				bd_page = bh->b_page;
1672 			}
1673 		}
1674 
1675 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1676 				    b_assoc_buffers) {
1677 			set_buffer_async_write(bh);
1678 			if (bh == segbuf->sb_super_root) {
1679 				if (bh->b_page != bd_page) {
1680 					lock_page(bd_page);
1681 					clear_page_dirty_for_io(bd_page);
1682 					set_page_writeback(bd_page);
1683 					unlock_page(bd_page);
1684 					bd_page = bh->b_page;
1685 				}
1686 				break;
1687 			}
1688 			if (bh->b_page != fs_page) {
1689 				nilfs_begin_page_io(fs_page);
1690 				fs_page = bh->b_page;
1691 			}
1692 		}
1693 	}
1694 	if (bd_page) {
1695 		lock_page(bd_page);
1696 		clear_page_dirty_for_io(bd_page);
1697 		set_page_writeback(bd_page);
1698 		unlock_page(bd_page);
1699 	}
1700 	nilfs_begin_page_io(fs_page);
1701 }
1702 
1703 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1704 			       struct the_nilfs *nilfs)
1705 {
1706 	int ret;
1707 
1708 	ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1709 	list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1710 	return ret;
1711 }
1712 
1713 static void nilfs_end_page_io(struct page *page, int err)
1714 {
1715 	if (!page)
1716 		return;
1717 
1718 	if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1719 		/*
1720 		 * For b-tree node pages, this function may be called twice
1721 		 * or more because they might be split in a segment.
1722 		 */
1723 		if (PageDirty(page)) {
1724 			/*
1725 			 * For pages holding split b-tree node buffers, dirty
1726 			 * flag on the buffers may be cleared discretely.
1727 			 * In that case, the page is once redirtied for
1728 			 * remaining buffers, and it must be cancelled if
1729 			 * all the buffers get cleaned later.
1730 			 */
1731 			lock_page(page);
1732 			if (nilfs_page_buffers_clean(page))
1733 				__nilfs_clear_page_dirty(page);
1734 			unlock_page(page);
1735 		}
1736 		return;
1737 	}
1738 
1739 	if (!err) {
1740 		if (!nilfs_page_buffers_clean(page))
1741 			__set_page_dirty_nobuffers(page);
1742 		ClearPageError(page);
1743 	} else {
1744 		__set_page_dirty_nobuffers(page);
1745 		SetPageError(page);
1746 	}
1747 
1748 	end_page_writeback(page);
1749 }
1750 
1751 static void nilfs_abort_logs(struct list_head *logs, int err)
1752 {
1753 	struct nilfs_segment_buffer *segbuf;
1754 	struct page *bd_page = NULL, *fs_page = NULL;
1755 	struct buffer_head *bh;
1756 
1757 	if (list_empty(logs))
1758 		return;
1759 
1760 	list_for_each_entry(segbuf, logs, sb_list) {
1761 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1762 				    b_assoc_buffers) {
1763 			if (bh->b_page != bd_page) {
1764 				if (bd_page)
1765 					end_page_writeback(bd_page);
1766 				bd_page = bh->b_page;
1767 			}
1768 		}
1769 
1770 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1771 				    b_assoc_buffers) {
1772 			clear_buffer_async_write(bh);
1773 			if (bh == segbuf->sb_super_root) {
1774 				if (bh->b_page != bd_page) {
1775 					end_page_writeback(bd_page);
1776 					bd_page = bh->b_page;
1777 				}
1778 				break;
1779 			}
1780 			if (bh->b_page != fs_page) {
1781 				nilfs_end_page_io(fs_page, err);
1782 				fs_page = bh->b_page;
1783 			}
1784 		}
1785 	}
1786 	if (bd_page)
1787 		end_page_writeback(bd_page);
1788 
1789 	nilfs_end_page_io(fs_page, err);
1790 }
1791 
1792 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1793 					     struct the_nilfs *nilfs, int err)
1794 {
1795 	LIST_HEAD(logs);
1796 	int ret;
1797 
1798 	list_splice_tail_init(&sci->sc_write_logs, &logs);
1799 	ret = nilfs_wait_on_logs(&logs);
1800 	nilfs_abort_logs(&logs, ret ? : err);
1801 
1802 	list_splice_tail_init(&sci->sc_segbufs, &logs);
1803 	nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1804 	nilfs_free_incomplete_logs(&logs, nilfs);
1805 
1806 	if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1807 		ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1808 						sci->sc_freesegs,
1809 						sci->sc_nfreesegs,
1810 						NULL);
1811 		WARN_ON(ret); /* do not happen */
1812 	}
1813 
1814 	nilfs_destroy_logs(&logs);
1815 }
1816 
1817 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1818 				   struct nilfs_segment_buffer *segbuf)
1819 {
1820 	nilfs->ns_segnum = segbuf->sb_segnum;
1821 	nilfs->ns_nextnum = segbuf->sb_nextnum;
1822 	nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1823 		+ segbuf->sb_sum.nblocks;
1824 	nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1825 	nilfs->ns_ctime = segbuf->sb_sum.ctime;
1826 }
1827 
1828 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1829 {
1830 	struct nilfs_segment_buffer *segbuf;
1831 	struct page *bd_page = NULL, *fs_page = NULL;
1832 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1833 	int update_sr = false;
1834 
1835 	list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1836 		struct buffer_head *bh;
1837 
1838 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1839 				    b_assoc_buffers) {
1840 			set_buffer_uptodate(bh);
1841 			clear_buffer_dirty(bh);
1842 			if (bh->b_page != bd_page) {
1843 				if (bd_page)
1844 					end_page_writeback(bd_page);
1845 				bd_page = bh->b_page;
1846 			}
1847 		}
1848 		/*
1849 		 * We assume that the buffers which belong to the same page
1850 		 * continue over the buffer list.
1851 		 * Under this assumption, the last BHs of pages is
1852 		 * identifiable by the discontinuity of bh->b_page
1853 		 * (page != fs_page).
1854 		 *
1855 		 * For B-tree node blocks, however, this assumption is not
1856 		 * guaranteed.  The cleanup code of B-tree node pages needs
1857 		 * special care.
1858 		 */
1859 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1860 				    b_assoc_buffers) {
1861 			const unsigned long set_bits = BIT(BH_Uptodate);
1862 			const unsigned long clear_bits =
1863 				(BIT(BH_Dirty) | BIT(BH_Async_Write) |
1864 				 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1865 				 BIT(BH_NILFS_Redirected));
1866 
1867 			set_mask_bits(&bh->b_state, clear_bits, set_bits);
1868 			if (bh == segbuf->sb_super_root) {
1869 				if (bh->b_page != bd_page) {
1870 					end_page_writeback(bd_page);
1871 					bd_page = bh->b_page;
1872 				}
1873 				update_sr = true;
1874 				break;
1875 			}
1876 			if (bh->b_page != fs_page) {
1877 				nilfs_end_page_io(fs_page, 0);
1878 				fs_page = bh->b_page;
1879 			}
1880 		}
1881 
1882 		if (!nilfs_segbuf_simplex(segbuf)) {
1883 			if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1884 				set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1885 				sci->sc_lseg_stime = jiffies;
1886 			}
1887 			if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1888 				clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1889 		}
1890 	}
1891 	/*
1892 	 * Since pages may continue over multiple segment buffers,
1893 	 * end of the last page must be checked outside of the loop.
1894 	 */
1895 	if (bd_page)
1896 		end_page_writeback(bd_page);
1897 
1898 	nilfs_end_page_io(fs_page, 0);
1899 
1900 	nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1901 
1902 	if (nilfs_doing_gc())
1903 		nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1904 	else
1905 		nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1906 
1907 	sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1908 
1909 	segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1910 	nilfs_set_next_segment(nilfs, segbuf);
1911 
1912 	if (update_sr) {
1913 		nilfs->ns_flushed_device = 0;
1914 		nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1915 				       segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1916 
1917 		clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1918 		clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1919 		set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1920 		nilfs_segctor_clear_metadata_dirty(sci);
1921 	} else
1922 		clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1923 }
1924 
1925 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1926 {
1927 	int ret;
1928 
1929 	ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1930 	if (!ret) {
1931 		nilfs_segctor_complete_write(sci);
1932 		nilfs_destroy_logs(&sci->sc_write_logs);
1933 	}
1934 	return ret;
1935 }
1936 
1937 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1938 					     struct the_nilfs *nilfs)
1939 {
1940 	struct nilfs_inode_info *ii, *n;
1941 	struct inode *ifile = sci->sc_root->ifile;
1942 
1943 	spin_lock(&nilfs->ns_inode_lock);
1944  retry:
1945 	list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1946 		if (!ii->i_bh) {
1947 			struct buffer_head *ibh;
1948 			int err;
1949 
1950 			spin_unlock(&nilfs->ns_inode_lock);
1951 			err = nilfs_ifile_get_inode_block(
1952 				ifile, ii->vfs_inode.i_ino, &ibh);
1953 			if (unlikely(err)) {
1954 				nilfs_msg(sci->sc_super, KERN_WARNING,
1955 					  "log writer: error %d getting inode block (ino=%lu)",
1956 					  err, ii->vfs_inode.i_ino);
1957 				return err;
1958 			}
1959 			mark_buffer_dirty(ibh);
1960 			nilfs_mdt_mark_dirty(ifile);
1961 			spin_lock(&nilfs->ns_inode_lock);
1962 			if (likely(!ii->i_bh))
1963 				ii->i_bh = ibh;
1964 			else
1965 				brelse(ibh);
1966 			goto retry;
1967 		}
1968 
1969 		clear_bit(NILFS_I_QUEUED, &ii->i_state);
1970 		set_bit(NILFS_I_BUSY, &ii->i_state);
1971 		list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1972 	}
1973 	spin_unlock(&nilfs->ns_inode_lock);
1974 
1975 	return 0;
1976 }
1977 
1978 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1979 					     struct the_nilfs *nilfs)
1980 {
1981 	struct nilfs_inode_info *ii, *n;
1982 	int during_mount = !(sci->sc_super->s_flags & MS_ACTIVE);
1983 	int defer_iput = false;
1984 
1985 	spin_lock(&nilfs->ns_inode_lock);
1986 	list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1987 		if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1988 		    test_bit(NILFS_I_DIRTY, &ii->i_state))
1989 			continue;
1990 
1991 		clear_bit(NILFS_I_BUSY, &ii->i_state);
1992 		brelse(ii->i_bh);
1993 		ii->i_bh = NULL;
1994 		list_del_init(&ii->i_dirty);
1995 		if (!ii->vfs_inode.i_nlink || during_mount) {
1996 			/*
1997 			 * Defer calling iput() to avoid deadlocks if
1998 			 * i_nlink == 0 or mount is not yet finished.
1999 			 */
2000 			list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2001 			defer_iput = true;
2002 		} else {
2003 			spin_unlock(&nilfs->ns_inode_lock);
2004 			iput(&ii->vfs_inode);
2005 			spin_lock(&nilfs->ns_inode_lock);
2006 		}
2007 	}
2008 	spin_unlock(&nilfs->ns_inode_lock);
2009 
2010 	if (defer_iput)
2011 		schedule_work(&sci->sc_iput_work);
2012 }
2013 
2014 /*
2015  * Main procedure of segment constructor
2016  */
2017 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2018 {
2019 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2020 	int err;
2021 
2022 	nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2023 	sci->sc_cno = nilfs->ns_cno;
2024 
2025 	err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2026 	if (unlikely(err))
2027 		goto out;
2028 
2029 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2030 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2031 
2032 	if (nilfs_segctor_clean(sci))
2033 		goto out;
2034 
2035 	do {
2036 		sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2037 
2038 		err = nilfs_segctor_begin_construction(sci, nilfs);
2039 		if (unlikely(err))
2040 			goto out;
2041 
2042 		/* Update time stamp */
2043 		sci->sc_seg_ctime = get_seconds();
2044 
2045 		err = nilfs_segctor_collect(sci, nilfs, mode);
2046 		if (unlikely(err))
2047 			goto failed;
2048 
2049 		/* Avoid empty segment */
2050 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2051 		    nilfs_segbuf_empty(sci->sc_curseg)) {
2052 			nilfs_segctor_abort_construction(sci, nilfs, 1);
2053 			goto out;
2054 		}
2055 
2056 		err = nilfs_segctor_assign(sci, mode);
2057 		if (unlikely(err))
2058 			goto failed;
2059 
2060 		if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2061 			nilfs_segctor_fill_in_file_bmap(sci);
2062 
2063 		if (mode == SC_LSEG_SR &&
2064 		    nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2065 			err = nilfs_segctor_fill_in_checkpoint(sci);
2066 			if (unlikely(err))
2067 				goto failed_to_write;
2068 
2069 			nilfs_segctor_fill_in_super_root(sci, nilfs);
2070 		}
2071 		nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2072 
2073 		/* Write partial segments */
2074 		nilfs_segctor_prepare_write(sci);
2075 
2076 		nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2077 					    nilfs->ns_crc_seed);
2078 
2079 		err = nilfs_segctor_write(sci, nilfs);
2080 		if (unlikely(err))
2081 			goto failed_to_write;
2082 
2083 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2084 		    nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2085 			/*
2086 			 * At this point, we avoid double buffering
2087 			 * for blocksize < pagesize because page dirty
2088 			 * flag is turned off during write and dirty
2089 			 * buffers are not properly collected for
2090 			 * pages crossing over segments.
2091 			 */
2092 			err = nilfs_segctor_wait(sci);
2093 			if (err)
2094 				goto failed_to_write;
2095 		}
2096 	} while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2097 
2098  out:
2099 	nilfs_segctor_drop_written_files(sci, nilfs);
2100 	return err;
2101 
2102  failed_to_write:
2103 	if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2104 		nilfs_redirty_inodes(&sci->sc_dirty_files);
2105 
2106  failed:
2107 	if (nilfs_doing_gc())
2108 		nilfs_redirty_inodes(&sci->sc_gc_inodes);
2109 	nilfs_segctor_abort_construction(sci, nilfs, err);
2110 	goto out;
2111 }
2112 
2113 /**
2114  * nilfs_segctor_start_timer - set timer of background write
2115  * @sci: nilfs_sc_info
2116  *
2117  * If the timer has already been set, it ignores the new request.
2118  * This function MUST be called within a section locking the segment
2119  * semaphore.
2120  */
2121 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2122 {
2123 	spin_lock(&sci->sc_state_lock);
2124 	if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2125 		sci->sc_timer.expires = jiffies + sci->sc_interval;
2126 		add_timer(&sci->sc_timer);
2127 		sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2128 	}
2129 	spin_unlock(&sci->sc_state_lock);
2130 }
2131 
2132 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2133 {
2134 	spin_lock(&sci->sc_state_lock);
2135 	if (!(sci->sc_flush_request & BIT(bn))) {
2136 		unsigned long prev_req = sci->sc_flush_request;
2137 
2138 		sci->sc_flush_request |= BIT(bn);
2139 		if (!prev_req)
2140 			wake_up(&sci->sc_wait_daemon);
2141 	}
2142 	spin_unlock(&sci->sc_state_lock);
2143 }
2144 
2145 /**
2146  * nilfs_flush_segment - trigger a segment construction for resource control
2147  * @sb: super block
2148  * @ino: inode number of the file to be flushed out.
2149  */
2150 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2151 {
2152 	struct the_nilfs *nilfs = sb->s_fs_info;
2153 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2154 
2155 	if (!sci || nilfs_doing_construction())
2156 		return;
2157 	nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2158 					/* assign bit 0 to data files */
2159 }
2160 
2161 struct nilfs_segctor_wait_request {
2162 	wait_queue_t	wq;
2163 	__u32		seq;
2164 	int		err;
2165 	atomic_t	done;
2166 };
2167 
2168 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2169 {
2170 	struct nilfs_segctor_wait_request wait_req;
2171 	int err = 0;
2172 
2173 	spin_lock(&sci->sc_state_lock);
2174 	init_wait(&wait_req.wq);
2175 	wait_req.err = 0;
2176 	atomic_set(&wait_req.done, 0);
2177 	wait_req.seq = ++sci->sc_seq_request;
2178 	spin_unlock(&sci->sc_state_lock);
2179 
2180 	init_waitqueue_entry(&wait_req.wq, current);
2181 	add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2182 	set_current_state(TASK_INTERRUPTIBLE);
2183 	wake_up(&sci->sc_wait_daemon);
2184 
2185 	for (;;) {
2186 		if (atomic_read(&wait_req.done)) {
2187 			err = wait_req.err;
2188 			break;
2189 		}
2190 		if (!signal_pending(current)) {
2191 			schedule();
2192 			continue;
2193 		}
2194 		err = -ERESTARTSYS;
2195 		break;
2196 	}
2197 	finish_wait(&sci->sc_wait_request, &wait_req.wq);
2198 	return err;
2199 }
2200 
2201 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2202 {
2203 	struct nilfs_segctor_wait_request *wrq, *n;
2204 	unsigned long flags;
2205 
2206 	spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2207 	list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2208 				 wq.task_list) {
2209 		if (!atomic_read(&wrq->done) &&
2210 		    nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2211 			wrq->err = err;
2212 			atomic_set(&wrq->done, 1);
2213 		}
2214 		if (atomic_read(&wrq->done)) {
2215 			wrq->wq.func(&wrq->wq,
2216 				     TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2217 				     0, NULL);
2218 		}
2219 	}
2220 	spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2221 }
2222 
2223 /**
2224  * nilfs_construct_segment - construct a logical segment
2225  * @sb: super block
2226  *
2227  * Return Value: On success, 0 is retured. On errors, one of the following
2228  * negative error code is returned.
2229  *
2230  * %-EROFS - Read only filesystem.
2231  *
2232  * %-EIO - I/O error
2233  *
2234  * %-ENOSPC - No space left on device (only in a panic state).
2235  *
2236  * %-ERESTARTSYS - Interrupted.
2237  *
2238  * %-ENOMEM - Insufficient memory available.
2239  */
2240 int nilfs_construct_segment(struct super_block *sb)
2241 {
2242 	struct the_nilfs *nilfs = sb->s_fs_info;
2243 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2244 	struct nilfs_transaction_info *ti;
2245 	int err;
2246 
2247 	if (!sci)
2248 		return -EROFS;
2249 
2250 	/* A call inside transactions causes a deadlock. */
2251 	BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2252 
2253 	err = nilfs_segctor_sync(sci);
2254 	return err;
2255 }
2256 
2257 /**
2258  * nilfs_construct_dsync_segment - construct a data-only logical segment
2259  * @sb: super block
2260  * @inode: inode whose data blocks should be written out
2261  * @start: start byte offset
2262  * @end: end byte offset (inclusive)
2263  *
2264  * Return Value: On success, 0 is retured. On errors, one of the following
2265  * negative error code is returned.
2266  *
2267  * %-EROFS - Read only filesystem.
2268  *
2269  * %-EIO - I/O error
2270  *
2271  * %-ENOSPC - No space left on device (only in a panic state).
2272  *
2273  * %-ERESTARTSYS - Interrupted.
2274  *
2275  * %-ENOMEM - Insufficient memory available.
2276  */
2277 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2278 				  loff_t start, loff_t end)
2279 {
2280 	struct the_nilfs *nilfs = sb->s_fs_info;
2281 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2282 	struct nilfs_inode_info *ii;
2283 	struct nilfs_transaction_info ti;
2284 	int err = 0;
2285 
2286 	if (!sci)
2287 		return -EROFS;
2288 
2289 	nilfs_transaction_lock(sb, &ti, 0);
2290 
2291 	ii = NILFS_I(inode);
2292 	if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2293 	    nilfs_test_opt(nilfs, STRICT_ORDER) ||
2294 	    test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2295 	    nilfs_discontinued(nilfs)) {
2296 		nilfs_transaction_unlock(sb);
2297 		err = nilfs_segctor_sync(sci);
2298 		return err;
2299 	}
2300 
2301 	spin_lock(&nilfs->ns_inode_lock);
2302 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2303 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2304 		spin_unlock(&nilfs->ns_inode_lock);
2305 		nilfs_transaction_unlock(sb);
2306 		return 0;
2307 	}
2308 	spin_unlock(&nilfs->ns_inode_lock);
2309 	sci->sc_dsync_inode = ii;
2310 	sci->sc_dsync_start = start;
2311 	sci->sc_dsync_end = end;
2312 
2313 	err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2314 	if (!err)
2315 		nilfs->ns_flushed_device = 0;
2316 
2317 	nilfs_transaction_unlock(sb);
2318 	return err;
2319 }
2320 
2321 #define FLUSH_FILE_BIT	(0x1) /* data file only */
2322 #define FLUSH_DAT_BIT	BIT(NILFS_DAT_INO) /* DAT only */
2323 
2324 /**
2325  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2326  * @sci: segment constructor object
2327  */
2328 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2329 {
2330 	spin_lock(&sci->sc_state_lock);
2331 	sci->sc_seq_accepted = sci->sc_seq_request;
2332 	spin_unlock(&sci->sc_state_lock);
2333 	del_timer_sync(&sci->sc_timer);
2334 }
2335 
2336 /**
2337  * nilfs_segctor_notify - notify the result of request to caller threads
2338  * @sci: segment constructor object
2339  * @mode: mode of log forming
2340  * @err: error code to be notified
2341  */
2342 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2343 {
2344 	/* Clear requests (even when the construction failed) */
2345 	spin_lock(&sci->sc_state_lock);
2346 
2347 	if (mode == SC_LSEG_SR) {
2348 		sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2349 		sci->sc_seq_done = sci->sc_seq_accepted;
2350 		nilfs_segctor_wakeup(sci, err);
2351 		sci->sc_flush_request = 0;
2352 	} else {
2353 		if (mode == SC_FLUSH_FILE)
2354 			sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2355 		else if (mode == SC_FLUSH_DAT)
2356 			sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2357 
2358 		/* re-enable timer if checkpoint creation was not done */
2359 		if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2360 		    time_before(jiffies, sci->sc_timer.expires))
2361 			add_timer(&sci->sc_timer);
2362 	}
2363 	spin_unlock(&sci->sc_state_lock);
2364 }
2365 
2366 /**
2367  * nilfs_segctor_construct - form logs and write them to disk
2368  * @sci: segment constructor object
2369  * @mode: mode of log forming
2370  */
2371 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2372 {
2373 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2374 	struct nilfs_super_block **sbp;
2375 	int err = 0;
2376 
2377 	nilfs_segctor_accept(sci);
2378 
2379 	if (nilfs_discontinued(nilfs))
2380 		mode = SC_LSEG_SR;
2381 	if (!nilfs_segctor_confirm(sci))
2382 		err = nilfs_segctor_do_construct(sci, mode);
2383 
2384 	if (likely(!err)) {
2385 		if (mode != SC_FLUSH_DAT)
2386 			atomic_set(&nilfs->ns_ndirtyblks, 0);
2387 		if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2388 		    nilfs_discontinued(nilfs)) {
2389 			down_write(&nilfs->ns_sem);
2390 			err = -EIO;
2391 			sbp = nilfs_prepare_super(sci->sc_super,
2392 						  nilfs_sb_will_flip(nilfs));
2393 			if (likely(sbp)) {
2394 				nilfs_set_log_cursor(sbp[0], nilfs);
2395 				err = nilfs_commit_super(sci->sc_super,
2396 							 NILFS_SB_COMMIT);
2397 			}
2398 			up_write(&nilfs->ns_sem);
2399 		}
2400 	}
2401 
2402 	nilfs_segctor_notify(sci, mode, err);
2403 	return err;
2404 }
2405 
2406 static void nilfs_construction_timeout(unsigned long data)
2407 {
2408 	struct task_struct *p = (struct task_struct *)data;
2409 
2410 	wake_up_process(p);
2411 }
2412 
2413 static void
2414 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2415 {
2416 	struct nilfs_inode_info *ii, *n;
2417 
2418 	list_for_each_entry_safe(ii, n, head, i_dirty) {
2419 		if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2420 			continue;
2421 		list_del_init(&ii->i_dirty);
2422 		truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2423 		nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2424 		iput(&ii->vfs_inode);
2425 	}
2426 }
2427 
2428 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2429 			 void **kbufs)
2430 {
2431 	struct the_nilfs *nilfs = sb->s_fs_info;
2432 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2433 	struct nilfs_transaction_info ti;
2434 	int err;
2435 
2436 	if (unlikely(!sci))
2437 		return -EROFS;
2438 
2439 	nilfs_transaction_lock(sb, &ti, 1);
2440 
2441 	err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2442 	if (unlikely(err))
2443 		goto out_unlock;
2444 
2445 	err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2446 	if (unlikely(err)) {
2447 		nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2448 		goto out_unlock;
2449 	}
2450 
2451 	sci->sc_freesegs = kbufs[4];
2452 	sci->sc_nfreesegs = argv[4].v_nmembs;
2453 	list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2454 
2455 	for (;;) {
2456 		err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2457 		nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2458 
2459 		if (likely(!err))
2460 			break;
2461 
2462 		nilfs_msg(sb, KERN_WARNING, "error %d cleaning segments", err);
2463 		set_current_state(TASK_INTERRUPTIBLE);
2464 		schedule_timeout(sci->sc_interval);
2465 	}
2466 	if (nilfs_test_opt(nilfs, DISCARD)) {
2467 		int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2468 						 sci->sc_nfreesegs);
2469 		if (ret) {
2470 			nilfs_msg(sb, KERN_WARNING,
2471 				  "error %d on discard request, turning discards off for the device",
2472 				  ret);
2473 			nilfs_clear_opt(nilfs, DISCARD);
2474 		}
2475 	}
2476 
2477  out_unlock:
2478 	sci->sc_freesegs = NULL;
2479 	sci->sc_nfreesegs = 0;
2480 	nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2481 	nilfs_transaction_unlock(sb);
2482 	return err;
2483 }
2484 
2485 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2486 {
2487 	struct nilfs_transaction_info ti;
2488 
2489 	nilfs_transaction_lock(sci->sc_super, &ti, 0);
2490 	nilfs_segctor_construct(sci, mode);
2491 
2492 	/*
2493 	 * Unclosed segment should be retried.  We do this using sc_timer.
2494 	 * Timeout of sc_timer will invoke complete construction which leads
2495 	 * to close the current logical segment.
2496 	 */
2497 	if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2498 		nilfs_segctor_start_timer(sci);
2499 
2500 	nilfs_transaction_unlock(sci->sc_super);
2501 }
2502 
2503 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2504 {
2505 	int mode = 0;
2506 
2507 	spin_lock(&sci->sc_state_lock);
2508 	mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2509 		SC_FLUSH_DAT : SC_FLUSH_FILE;
2510 	spin_unlock(&sci->sc_state_lock);
2511 
2512 	if (mode) {
2513 		nilfs_segctor_do_construct(sci, mode);
2514 
2515 		spin_lock(&sci->sc_state_lock);
2516 		sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2517 			~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2518 		spin_unlock(&sci->sc_state_lock);
2519 	}
2520 	clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2521 }
2522 
2523 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2524 {
2525 	if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2526 	    time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2527 		if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2528 			return SC_FLUSH_FILE;
2529 		else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2530 			return SC_FLUSH_DAT;
2531 	}
2532 	return SC_LSEG_SR;
2533 }
2534 
2535 /**
2536  * nilfs_segctor_thread - main loop of the segment constructor thread.
2537  * @arg: pointer to a struct nilfs_sc_info.
2538  *
2539  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2540  * to execute segment constructions.
2541  */
2542 static int nilfs_segctor_thread(void *arg)
2543 {
2544 	struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2545 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2546 	int timeout = 0;
2547 
2548 	sci->sc_timer.data = (unsigned long)current;
2549 	sci->sc_timer.function = nilfs_construction_timeout;
2550 
2551 	/* start sync. */
2552 	sci->sc_task = current;
2553 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2554 	nilfs_msg(sci->sc_super, KERN_INFO,
2555 		  "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2556 		  sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2557 
2558 	spin_lock(&sci->sc_state_lock);
2559  loop:
2560 	for (;;) {
2561 		int mode;
2562 
2563 		if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2564 			goto end_thread;
2565 
2566 		if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2567 			mode = SC_LSEG_SR;
2568 		else if (sci->sc_flush_request)
2569 			mode = nilfs_segctor_flush_mode(sci);
2570 		else
2571 			break;
2572 
2573 		spin_unlock(&sci->sc_state_lock);
2574 		nilfs_segctor_thread_construct(sci, mode);
2575 		spin_lock(&sci->sc_state_lock);
2576 		timeout = 0;
2577 	}
2578 
2579 
2580 	if (freezing(current)) {
2581 		spin_unlock(&sci->sc_state_lock);
2582 		try_to_freeze();
2583 		spin_lock(&sci->sc_state_lock);
2584 	} else {
2585 		DEFINE_WAIT(wait);
2586 		int should_sleep = 1;
2587 
2588 		prepare_to_wait(&sci->sc_wait_daemon, &wait,
2589 				TASK_INTERRUPTIBLE);
2590 
2591 		if (sci->sc_seq_request != sci->sc_seq_done)
2592 			should_sleep = 0;
2593 		else if (sci->sc_flush_request)
2594 			should_sleep = 0;
2595 		else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2596 			should_sleep = time_before(jiffies,
2597 					sci->sc_timer.expires);
2598 
2599 		if (should_sleep) {
2600 			spin_unlock(&sci->sc_state_lock);
2601 			schedule();
2602 			spin_lock(&sci->sc_state_lock);
2603 		}
2604 		finish_wait(&sci->sc_wait_daemon, &wait);
2605 		timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2606 			   time_after_eq(jiffies, sci->sc_timer.expires));
2607 
2608 		if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2609 			set_nilfs_discontinued(nilfs);
2610 	}
2611 	goto loop;
2612 
2613  end_thread:
2614 	spin_unlock(&sci->sc_state_lock);
2615 
2616 	/* end sync. */
2617 	sci->sc_task = NULL;
2618 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2619 	return 0;
2620 }
2621 
2622 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2623 {
2624 	struct task_struct *t;
2625 
2626 	t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2627 	if (IS_ERR(t)) {
2628 		int err = PTR_ERR(t);
2629 
2630 		nilfs_msg(sci->sc_super, KERN_ERR,
2631 			  "error %d creating segctord thread", err);
2632 		return err;
2633 	}
2634 	wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2635 	return 0;
2636 }
2637 
2638 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2639 	__acquires(&sci->sc_state_lock)
2640 	__releases(&sci->sc_state_lock)
2641 {
2642 	sci->sc_state |= NILFS_SEGCTOR_QUIT;
2643 
2644 	while (sci->sc_task) {
2645 		wake_up(&sci->sc_wait_daemon);
2646 		spin_unlock(&sci->sc_state_lock);
2647 		wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2648 		spin_lock(&sci->sc_state_lock);
2649 	}
2650 }
2651 
2652 /*
2653  * Setup & clean-up functions
2654  */
2655 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2656 					       struct nilfs_root *root)
2657 {
2658 	struct the_nilfs *nilfs = sb->s_fs_info;
2659 	struct nilfs_sc_info *sci;
2660 
2661 	sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2662 	if (!sci)
2663 		return NULL;
2664 
2665 	sci->sc_super = sb;
2666 
2667 	nilfs_get_root(root);
2668 	sci->sc_root = root;
2669 
2670 	init_waitqueue_head(&sci->sc_wait_request);
2671 	init_waitqueue_head(&sci->sc_wait_daemon);
2672 	init_waitqueue_head(&sci->sc_wait_task);
2673 	spin_lock_init(&sci->sc_state_lock);
2674 	INIT_LIST_HEAD(&sci->sc_dirty_files);
2675 	INIT_LIST_HEAD(&sci->sc_segbufs);
2676 	INIT_LIST_HEAD(&sci->sc_write_logs);
2677 	INIT_LIST_HEAD(&sci->sc_gc_inodes);
2678 	INIT_LIST_HEAD(&sci->sc_iput_queue);
2679 	INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2680 	init_timer(&sci->sc_timer);
2681 
2682 	sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2683 	sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2684 	sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2685 
2686 	if (nilfs->ns_interval)
2687 		sci->sc_interval = HZ * nilfs->ns_interval;
2688 	if (nilfs->ns_watermark)
2689 		sci->sc_watermark = nilfs->ns_watermark;
2690 	return sci;
2691 }
2692 
2693 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2694 {
2695 	int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2696 
2697 	/*
2698 	 * The segctord thread was stopped and its timer was removed.
2699 	 * But some tasks remain.
2700 	 */
2701 	do {
2702 		struct nilfs_transaction_info ti;
2703 
2704 		nilfs_transaction_lock(sci->sc_super, &ti, 0);
2705 		ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2706 		nilfs_transaction_unlock(sci->sc_super);
2707 
2708 		flush_work(&sci->sc_iput_work);
2709 
2710 	} while (ret && retrycount-- > 0);
2711 }
2712 
2713 /**
2714  * nilfs_segctor_destroy - destroy the segment constructor.
2715  * @sci: nilfs_sc_info
2716  *
2717  * nilfs_segctor_destroy() kills the segctord thread and frees
2718  * the nilfs_sc_info struct.
2719  * Caller must hold the segment semaphore.
2720  */
2721 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2722 {
2723 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2724 	int flag;
2725 
2726 	up_write(&nilfs->ns_segctor_sem);
2727 
2728 	spin_lock(&sci->sc_state_lock);
2729 	nilfs_segctor_kill_thread(sci);
2730 	flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2731 		|| sci->sc_seq_request != sci->sc_seq_done);
2732 	spin_unlock(&sci->sc_state_lock);
2733 
2734 	if (flush_work(&sci->sc_iput_work))
2735 		flag = true;
2736 
2737 	if (flag || !nilfs_segctor_confirm(sci))
2738 		nilfs_segctor_write_out(sci);
2739 
2740 	if (!list_empty(&sci->sc_dirty_files)) {
2741 		nilfs_msg(sci->sc_super, KERN_WARNING,
2742 			  "disposed unprocessed dirty file(s) when stopping log writer");
2743 		nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2744 	}
2745 
2746 	if (!list_empty(&sci->sc_iput_queue)) {
2747 		nilfs_msg(sci->sc_super, KERN_WARNING,
2748 			  "disposed unprocessed inode(s) in iput queue when stopping log writer");
2749 		nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2750 	}
2751 
2752 	WARN_ON(!list_empty(&sci->sc_segbufs));
2753 	WARN_ON(!list_empty(&sci->sc_write_logs));
2754 
2755 	nilfs_put_root(sci->sc_root);
2756 
2757 	down_write(&nilfs->ns_segctor_sem);
2758 
2759 	del_timer_sync(&sci->sc_timer);
2760 	kfree(sci);
2761 }
2762 
2763 /**
2764  * nilfs_attach_log_writer - attach log writer
2765  * @sb: super block instance
2766  * @root: root object of the current filesystem tree
2767  *
2768  * This allocates a log writer object, initializes it, and starts the
2769  * log writer.
2770  *
2771  * Return Value: On success, 0 is returned. On error, one of the following
2772  * negative error code is returned.
2773  *
2774  * %-ENOMEM - Insufficient memory available.
2775  */
2776 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2777 {
2778 	struct the_nilfs *nilfs = sb->s_fs_info;
2779 	int err;
2780 
2781 	if (nilfs->ns_writer) {
2782 		/*
2783 		 * This happens if the filesystem was remounted
2784 		 * read/write after nilfs_error degenerated it into a
2785 		 * read-only mount.
2786 		 */
2787 		nilfs_detach_log_writer(sb);
2788 	}
2789 
2790 	nilfs->ns_writer = nilfs_segctor_new(sb, root);
2791 	if (!nilfs->ns_writer)
2792 		return -ENOMEM;
2793 
2794 	err = nilfs_segctor_start_thread(nilfs->ns_writer);
2795 	if (err) {
2796 		kfree(nilfs->ns_writer);
2797 		nilfs->ns_writer = NULL;
2798 	}
2799 	return err;
2800 }
2801 
2802 /**
2803  * nilfs_detach_log_writer - destroy log writer
2804  * @sb: super block instance
2805  *
2806  * This kills log writer daemon, frees the log writer object, and
2807  * destroys list of dirty files.
2808  */
2809 void nilfs_detach_log_writer(struct super_block *sb)
2810 {
2811 	struct the_nilfs *nilfs = sb->s_fs_info;
2812 	LIST_HEAD(garbage_list);
2813 
2814 	down_write(&nilfs->ns_segctor_sem);
2815 	if (nilfs->ns_writer) {
2816 		nilfs_segctor_destroy(nilfs->ns_writer);
2817 		nilfs->ns_writer = NULL;
2818 	}
2819 
2820 	/* Force to free the list of dirty files */
2821 	spin_lock(&nilfs->ns_inode_lock);
2822 	if (!list_empty(&nilfs->ns_dirty_files)) {
2823 		list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2824 		nilfs_msg(sb, KERN_WARNING,
2825 			  "disposed unprocessed dirty file(s) when detaching log writer");
2826 	}
2827 	spin_unlock(&nilfs->ns_inode_lock);
2828 	up_write(&nilfs->ns_segctor_sem);
2829 
2830 	nilfs_dispose_list(nilfs, &garbage_list, 1);
2831 }
2832