xref: /openbmc/linux/fs/nilfs2/segment.c (revision e2c75e76)
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 <linux/sched/signal.h>
34 
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 "segbuf.h"
43 
44 
45 /*
46  * Segment constructor
47  */
48 #define SC_N_INODEVEC	16   /* Size of locally allocated inode vector */
49 
50 #define SC_MAX_SEGDELTA 64   /*
51 			      * Upper limit of the number of segments
52 			      * appended in collection retry loop
53 			      */
54 
55 /* Construction mode */
56 enum {
57 	SC_LSEG_SR = 1,	/* Make a logical segment having a super root */
58 	SC_LSEG_DSYNC,	/*
59 			 * Flush data blocks of a given file and make
60 			 * a logical segment without a super root.
61 			 */
62 	SC_FLUSH_FILE,	/*
63 			 * Flush data files, leads to segment writes without
64 			 * creating a checkpoint.
65 			 */
66 	SC_FLUSH_DAT,	/*
67 			 * Flush DAT file.  This also creates segments
68 			 * without a checkpoint.
69 			 */
70 };
71 
72 /* Stage numbers of dirty block collection */
73 enum {
74 	NILFS_ST_INIT = 0,
75 	NILFS_ST_GC,		/* Collecting dirty blocks for GC */
76 	NILFS_ST_FILE,
77 	NILFS_ST_IFILE,
78 	NILFS_ST_CPFILE,
79 	NILFS_ST_SUFILE,
80 	NILFS_ST_DAT,
81 	NILFS_ST_SR,		/* Super root */
82 	NILFS_ST_DSYNC,		/* Data sync blocks */
83 	NILFS_ST_DONE,
84 };
85 
86 #define CREATE_TRACE_POINTS
87 #include <trace/events/nilfs2.h>
88 
89 /*
90  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
91  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
92  * the variable must use them because transition of stage count must involve
93  * trace events (trace_nilfs2_collection_stage_transition).
94  *
95  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
96  * produce tracepoint events. It is provided just for making the intention
97  * clear.
98  */
99 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
100 {
101 	sci->sc_stage.scnt++;
102 	trace_nilfs2_collection_stage_transition(sci);
103 }
104 
105 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
106 {
107 	sci->sc_stage.scnt = next_scnt;
108 	trace_nilfs2_collection_stage_transition(sci);
109 }
110 
111 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
112 {
113 	return sci->sc_stage.scnt;
114 }
115 
116 /* State flags of collection */
117 #define NILFS_CF_NODE		0x0001	/* Collecting node blocks */
118 #define NILFS_CF_IFILE_STARTED	0x0002	/* IFILE stage has started */
119 #define NILFS_CF_SUFREED	0x0004	/* segment usages has been freed */
120 #define NILFS_CF_HISTORY_MASK	(NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
121 
122 /* Operations depending on the construction mode and file type */
123 struct nilfs_sc_operations {
124 	int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
125 			    struct inode *);
126 	int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
127 			    struct inode *);
128 	int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
129 			    struct inode *);
130 	void (*write_data_binfo)(struct nilfs_sc_info *,
131 				 struct nilfs_segsum_pointer *,
132 				 union nilfs_binfo *);
133 	void (*write_node_binfo)(struct nilfs_sc_info *,
134 				 struct nilfs_segsum_pointer *,
135 				 union nilfs_binfo *);
136 };
137 
138 /*
139  * Other definitions
140  */
141 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
142 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
143 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
144 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
145 
146 #define nilfs_cnt32_gt(a, b)   \
147 	(typecheck(__u32, a) && typecheck(__u32, b) && \
148 	 ((__s32)(b) - (__s32)(a) < 0))
149 #define nilfs_cnt32_ge(a, b)   \
150 	(typecheck(__u32, a) && typecheck(__u32, b) && \
151 	 ((__s32)(a) - (__s32)(b) >= 0))
152 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
153 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
154 
155 static int nilfs_prepare_segment_lock(struct super_block *sb,
156 				      struct nilfs_transaction_info *ti)
157 {
158 	struct nilfs_transaction_info *cur_ti = current->journal_info;
159 	void *save = NULL;
160 
161 	if (cur_ti) {
162 		if (cur_ti->ti_magic == NILFS_TI_MAGIC)
163 			return ++cur_ti->ti_count;
164 
165 		/*
166 		 * If journal_info field is occupied by other FS,
167 		 * it is saved and will be restored on
168 		 * nilfs_transaction_commit().
169 		 */
170 		nilfs_msg(sb, KERN_WARNING, "journal info from a different FS");
171 		save = current->journal_info;
172 	}
173 	if (!ti) {
174 		ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
175 		if (!ti)
176 			return -ENOMEM;
177 		ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
178 	} else {
179 		ti->ti_flags = 0;
180 	}
181 	ti->ti_count = 0;
182 	ti->ti_save = save;
183 	ti->ti_magic = NILFS_TI_MAGIC;
184 	current->journal_info = ti;
185 	return 0;
186 }
187 
188 /**
189  * nilfs_transaction_begin - start indivisible file operations.
190  * @sb: super block
191  * @ti: nilfs_transaction_info
192  * @vacancy_check: flags for vacancy rate checks
193  *
194  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
195  * the segment semaphore, to make a segment construction and write tasks
196  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
197  * The region enclosed by these two functions can be nested.  To avoid a
198  * deadlock, the semaphore is only acquired or released in the outermost call.
199  *
200  * This function allocates a nilfs_transaction_info struct to keep context
201  * information on it.  It is initialized and hooked onto the current task in
202  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
203  * instead; otherwise a new struct is assigned from a slab.
204  *
205  * When @vacancy_check flag is set, this function will check the amount of
206  * free space, and will wait for the GC to reclaim disk space if low capacity.
207  *
208  * Return Value: On success, 0 is returned. On error, one of the following
209  * negative error code is returned.
210  *
211  * %-ENOMEM - Insufficient memory available.
212  *
213  * %-ENOSPC - No space left on device
214  */
215 int nilfs_transaction_begin(struct super_block *sb,
216 			    struct nilfs_transaction_info *ti,
217 			    int vacancy_check)
218 {
219 	struct the_nilfs *nilfs;
220 	int ret = nilfs_prepare_segment_lock(sb, ti);
221 	struct nilfs_transaction_info *trace_ti;
222 
223 	if (unlikely(ret < 0))
224 		return ret;
225 	if (ret > 0) {
226 		trace_ti = current->journal_info;
227 
228 		trace_nilfs2_transaction_transition(sb, trace_ti,
229 				    trace_ti->ti_count, trace_ti->ti_flags,
230 				    TRACE_NILFS2_TRANSACTION_BEGIN);
231 		return 0;
232 	}
233 
234 	sb_start_intwrite(sb);
235 
236 	nilfs = sb->s_fs_info;
237 	down_read(&nilfs->ns_segctor_sem);
238 	if (vacancy_check && nilfs_near_disk_full(nilfs)) {
239 		up_read(&nilfs->ns_segctor_sem);
240 		ret = -ENOSPC;
241 		goto failed;
242 	}
243 
244 	trace_ti = current->journal_info;
245 	trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
246 					    trace_ti->ti_flags,
247 					    TRACE_NILFS2_TRANSACTION_BEGIN);
248 	return 0;
249 
250  failed:
251 	ti = current->journal_info;
252 	current->journal_info = ti->ti_save;
253 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
254 		kmem_cache_free(nilfs_transaction_cachep, ti);
255 	sb_end_intwrite(sb);
256 	return ret;
257 }
258 
259 /**
260  * nilfs_transaction_commit - commit indivisible file operations.
261  * @sb: super block
262  *
263  * nilfs_transaction_commit() releases the read semaphore which is
264  * acquired by nilfs_transaction_begin(). This is only performed
265  * in outermost call of this function.  If a commit flag is set,
266  * nilfs_transaction_commit() sets a timer to start the segment
267  * constructor.  If a sync flag is set, it starts construction
268  * directly.
269  */
270 int nilfs_transaction_commit(struct super_block *sb)
271 {
272 	struct nilfs_transaction_info *ti = current->journal_info;
273 	struct the_nilfs *nilfs = sb->s_fs_info;
274 	int err = 0;
275 
276 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
277 	ti->ti_flags |= NILFS_TI_COMMIT;
278 	if (ti->ti_count > 0) {
279 		ti->ti_count--;
280 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
281 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
282 		return 0;
283 	}
284 	if (nilfs->ns_writer) {
285 		struct nilfs_sc_info *sci = nilfs->ns_writer;
286 
287 		if (ti->ti_flags & NILFS_TI_COMMIT)
288 			nilfs_segctor_start_timer(sci);
289 		if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
290 			nilfs_segctor_do_flush(sci, 0);
291 	}
292 	up_read(&nilfs->ns_segctor_sem);
293 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
294 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
295 
296 	current->journal_info = ti->ti_save;
297 
298 	if (ti->ti_flags & NILFS_TI_SYNC)
299 		err = nilfs_construct_segment(sb);
300 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
301 		kmem_cache_free(nilfs_transaction_cachep, ti);
302 	sb_end_intwrite(sb);
303 	return err;
304 }
305 
306 void nilfs_transaction_abort(struct super_block *sb)
307 {
308 	struct nilfs_transaction_info *ti = current->journal_info;
309 	struct the_nilfs *nilfs = sb->s_fs_info;
310 
311 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
312 	if (ti->ti_count > 0) {
313 		ti->ti_count--;
314 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
315 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
316 		return;
317 	}
318 	up_read(&nilfs->ns_segctor_sem);
319 
320 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
321 		    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
322 
323 	current->journal_info = ti->ti_save;
324 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
325 		kmem_cache_free(nilfs_transaction_cachep, ti);
326 	sb_end_intwrite(sb);
327 }
328 
329 void nilfs_relax_pressure_in_lock(struct super_block *sb)
330 {
331 	struct the_nilfs *nilfs = sb->s_fs_info;
332 	struct nilfs_sc_info *sci = nilfs->ns_writer;
333 
334 	if (!sci || !sci->sc_flush_request)
335 		return;
336 
337 	set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
338 	up_read(&nilfs->ns_segctor_sem);
339 
340 	down_write(&nilfs->ns_segctor_sem);
341 	if (sci->sc_flush_request &&
342 	    test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
343 		struct nilfs_transaction_info *ti = current->journal_info;
344 
345 		ti->ti_flags |= NILFS_TI_WRITER;
346 		nilfs_segctor_do_immediate_flush(sci);
347 		ti->ti_flags &= ~NILFS_TI_WRITER;
348 	}
349 	downgrade_write(&nilfs->ns_segctor_sem);
350 }
351 
352 static void nilfs_transaction_lock(struct super_block *sb,
353 				   struct nilfs_transaction_info *ti,
354 				   int gcflag)
355 {
356 	struct nilfs_transaction_info *cur_ti = current->journal_info;
357 	struct the_nilfs *nilfs = sb->s_fs_info;
358 	struct nilfs_sc_info *sci = nilfs->ns_writer;
359 
360 	WARN_ON(cur_ti);
361 	ti->ti_flags = NILFS_TI_WRITER;
362 	ti->ti_count = 0;
363 	ti->ti_save = cur_ti;
364 	ti->ti_magic = NILFS_TI_MAGIC;
365 	current->journal_info = ti;
366 
367 	for (;;) {
368 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
369 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
370 
371 		down_write(&nilfs->ns_segctor_sem);
372 		if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
373 			break;
374 
375 		nilfs_segctor_do_immediate_flush(sci);
376 
377 		up_write(&nilfs->ns_segctor_sem);
378 		cond_resched();
379 	}
380 	if (gcflag)
381 		ti->ti_flags |= NILFS_TI_GC;
382 
383 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
384 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
385 }
386 
387 static void nilfs_transaction_unlock(struct super_block *sb)
388 {
389 	struct nilfs_transaction_info *ti = current->journal_info;
390 	struct the_nilfs *nilfs = sb->s_fs_info;
391 
392 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
393 	BUG_ON(ti->ti_count > 0);
394 
395 	up_write(&nilfs->ns_segctor_sem);
396 	current->journal_info = ti->ti_save;
397 
398 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
399 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
400 }
401 
402 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
403 					    struct nilfs_segsum_pointer *ssp,
404 					    unsigned int bytes)
405 {
406 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
407 	unsigned int blocksize = sci->sc_super->s_blocksize;
408 	void *p;
409 
410 	if (unlikely(ssp->offset + bytes > blocksize)) {
411 		ssp->offset = 0;
412 		BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
413 					       &segbuf->sb_segsum_buffers));
414 		ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
415 	}
416 	p = ssp->bh->b_data + ssp->offset;
417 	ssp->offset += bytes;
418 	return p;
419 }
420 
421 /**
422  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
423  * @sci: nilfs_sc_info
424  */
425 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
426 {
427 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
428 	struct buffer_head *sumbh;
429 	unsigned int sumbytes;
430 	unsigned int flags = 0;
431 	int err;
432 
433 	if (nilfs_doing_gc())
434 		flags = NILFS_SS_GC;
435 	err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
436 	if (unlikely(err))
437 		return err;
438 
439 	sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
440 	sumbytes = segbuf->sb_sum.sumbytes;
441 	sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
442 	sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
443 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
444 	return 0;
445 }
446 
447 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
448 {
449 	sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
450 	if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
451 		return -E2BIG; /*
452 				* The current segment is filled up
453 				* (internal code)
454 				*/
455 	sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
456 	return nilfs_segctor_reset_segment_buffer(sci);
457 }
458 
459 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
460 {
461 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
462 	int err;
463 
464 	if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
465 		err = nilfs_segctor_feed_segment(sci);
466 		if (err)
467 			return err;
468 		segbuf = sci->sc_curseg;
469 	}
470 	err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
471 	if (likely(!err))
472 		segbuf->sb_sum.flags |= NILFS_SS_SR;
473 	return err;
474 }
475 
476 /*
477  * Functions for making segment summary and payloads
478  */
479 static int nilfs_segctor_segsum_block_required(
480 	struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
481 	unsigned int binfo_size)
482 {
483 	unsigned int blocksize = sci->sc_super->s_blocksize;
484 	/* Size of finfo and binfo is enough small against blocksize */
485 
486 	return ssp->offset + binfo_size +
487 		(!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
488 		blocksize;
489 }
490 
491 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
492 				      struct inode *inode)
493 {
494 	sci->sc_curseg->sb_sum.nfinfo++;
495 	sci->sc_binfo_ptr = sci->sc_finfo_ptr;
496 	nilfs_segctor_map_segsum_entry(
497 		sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
498 
499 	if (NILFS_I(inode)->i_root &&
500 	    !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
501 		set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
502 	/* skip finfo */
503 }
504 
505 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
506 				    struct inode *inode)
507 {
508 	struct nilfs_finfo *finfo;
509 	struct nilfs_inode_info *ii;
510 	struct nilfs_segment_buffer *segbuf;
511 	__u64 cno;
512 
513 	if (sci->sc_blk_cnt == 0)
514 		return;
515 
516 	ii = NILFS_I(inode);
517 
518 	if (test_bit(NILFS_I_GCINODE, &ii->i_state))
519 		cno = ii->i_cno;
520 	else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
521 		cno = 0;
522 	else
523 		cno = sci->sc_cno;
524 
525 	finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
526 						 sizeof(*finfo));
527 	finfo->fi_ino = cpu_to_le64(inode->i_ino);
528 	finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
529 	finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
530 	finfo->fi_cno = cpu_to_le64(cno);
531 
532 	segbuf = sci->sc_curseg;
533 	segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
534 		sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
535 	sci->sc_finfo_ptr = sci->sc_binfo_ptr;
536 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
537 }
538 
539 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
540 					struct buffer_head *bh,
541 					struct inode *inode,
542 					unsigned int binfo_size)
543 {
544 	struct nilfs_segment_buffer *segbuf;
545 	int required, err = 0;
546 
547  retry:
548 	segbuf = sci->sc_curseg;
549 	required = nilfs_segctor_segsum_block_required(
550 		sci, &sci->sc_binfo_ptr, binfo_size);
551 	if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
552 		nilfs_segctor_end_finfo(sci, inode);
553 		err = nilfs_segctor_feed_segment(sci);
554 		if (err)
555 			return err;
556 		goto retry;
557 	}
558 	if (unlikely(required)) {
559 		err = nilfs_segbuf_extend_segsum(segbuf);
560 		if (unlikely(err))
561 			goto failed;
562 	}
563 	if (sci->sc_blk_cnt == 0)
564 		nilfs_segctor_begin_finfo(sci, inode);
565 
566 	nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
567 	/* Substitution to vblocknr is delayed until update_blocknr() */
568 	nilfs_segbuf_add_file_buffer(segbuf, bh);
569 	sci->sc_blk_cnt++;
570  failed:
571 	return err;
572 }
573 
574 /*
575  * Callback functions that enumerate, mark, and collect dirty blocks
576  */
577 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
578 				   struct buffer_head *bh, struct inode *inode)
579 {
580 	int err;
581 
582 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
583 	if (err < 0)
584 		return err;
585 
586 	err = nilfs_segctor_add_file_block(sci, bh, inode,
587 					   sizeof(struct nilfs_binfo_v));
588 	if (!err)
589 		sci->sc_datablk_cnt++;
590 	return err;
591 }
592 
593 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
594 				   struct buffer_head *bh,
595 				   struct inode *inode)
596 {
597 	return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
598 }
599 
600 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
601 				   struct buffer_head *bh,
602 				   struct inode *inode)
603 {
604 	WARN_ON(!buffer_dirty(bh));
605 	return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
606 }
607 
608 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
609 					struct nilfs_segsum_pointer *ssp,
610 					union nilfs_binfo *binfo)
611 {
612 	struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
613 		sci, ssp, sizeof(*binfo_v));
614 	*binfo_v = binfo->bi_v;
615 }
616 
617 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
618 					struct nilfs_segsum_pointer *ssp,
619 					union nilfs_binfo *binfo)
620 {
621 	__le64 *vblocknr = nilfs_segctor_map_segsum_entry(
622 		sci, ssp, sizeof(*vblocknr));
623 	*vblocknr = binfo->bi_v.bi_vblocknr;
624 }
625 
626 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
627 	.collect_data = nilfs_collect_file_data,
628 	.collect_node = nilfs_collect_file_node,
629 	.collect_bmap = nilfs_collect_file_bmap,
630 	.write_data_binfo = nilfs_write_file_data_binfo,
631 	.write_node_binfo = nilfs_write_file_node_binfo,
632 };
633 
634 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
635 				  struct buffer_head *bh, struct inode *inode)
636 {
637 	int err;
638 
639 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
640 	if (err < 0)
641 		return err;
642 
643 	err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
644 	if (!err)
645 		sci->sc_datablk_cnt++;
646 	return err;
647 }
648 
649 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
650 				  struct buffer_head *bh, struct inode *inode)
651 {
652 	WARN_ON(!buffer_dirty(bh));
653 	return nilfs_segctor_add_file_block(sci, bh, inode,
654 					    sizeof(struct nilfs_binfo_dat));
655 }
656 
657 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
658 				       struct nilfs_segsum_pointer *ssp,
659 				       union nilfs_binfo *binfo)
660 {
661 	__le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
662 							  sizeof(*blkoff));
663 	*blkoff = binfo->bi_dat.bi_blkoff;
664 }
665 
666 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
667 				       struct nilfs_segsum_pointer *ssp,
668 				       union nilfs_binfo *binfo)
669 {
670 	struct nilfs_binfo_dat *binfo_dat =
671 		nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
672 	*binfo_dat = binfo->bi_dat;
673 }
674 
675 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
676 	.collect_data = nilfs_collect_dat_data,
677 	.collect_node = nilfs_collect_file_node,
678 	.collect_bmap = nilfs_collect_dat_bmap,
679 	.write_data_binfo = nilfs_write_dat_data_binfo,
680 	.write_node_binfo = nilfs_write_dat_node_binfo,
681 };
682 
683 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
684 	.collect_data = nilfs_collect_file_data,
685 	.collect_node = NULL,
686 	.collect_bmap = NULL,
687 	.write_data_binfo = nilfs_write_file_data_binfo,
688 	.write_node_binfo = NULL,
689 };
690 
691 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
692 					      struct list_head *listp,
693 					      size_t nlimit,
694 					      loff_t start, loff_t end)
695 {
696 	struct address_space *mapping = inode->i_mapping;
697 	struct pagevec pvec;
698 	pgoff_t index = 0, last = ULONG_MAX;
699 	size_t ndirties = 0;
700 	int i;
701 
702 	if (unlikely(start != 0 || end != LLONG_MAX)) {
703 		/*
704 		 * A valid range is given for sync-ing data pages. The
705 		 * range is rounded to per-page; extra dirty buffers
706 		 * may be included if blocksize < pagesize.
707 		 */
708 		index = start >> PAGE_SHIFT;
709 		last = end >> PAGE_SHIFT;
710 	}
711 	pagevec_init(&pvec);
712  repeat:
713 	if (unlikely(index > last) ||
714 	    !pagevec_lookup_range_tag(&pvec, mapping, &index, last,
715 				PAGECACHE_TAG_DIRTY))
716 		return ndirties;
717 
718 	for (i = 0; i < pagevec_count(&pvec); i++) {
719 		struct buffer_head *bh, *head;
720 		struct page *page = pvec.pages[i];
721 
722 		lock_page(page);
723 		if (!page_has_buffers(page))
724 			create_empty_buffers(page, i_blocksize(inode), 0);
725 		unlock_page(page);
726 
727 		bh = head = page_buffers(page);
728 		do {
729 			if (!buffer_dirty(bh) || buffer_async_write(bh))
730 				continue;
731 			get_bh(bh);
732 			list_add_tail(&bh->b_assoc_buffers, listp);
733 			ndirties++;
734 			if (unlikely(ndirties >= nlimit)) {
735 				pagevec_release(&pvec);
736 				cond_resched();
737 				return ndirties;
738 			}
739 		} while (bh = bh->b_this_page, bh != head);
740 	}
741 	pagevec_release(&pvec);
742 	cond_resched();
743 	goto repeat;
744 }
745 
746 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
747 					    struct list_head *listp)
748 {
749 	struct nilfs_inode_info *ii = NILFS_I(inode);
750 	struct address_space *mapping = &ii->i_btnode_cache;
751 	struct pagevec pvec;
752 	struct buffer_head *bh, *head;
753 	unsigned int i;
754 	pgoff_t index = 0;
755 
756 	pagevec_init(&pvec);
757 
758 	while (pagevec_lookup_tag(&pvec, mapping, &index,
759 					PAGECACHE_TAG_DIRTY)) {
760 		for (i = 0; i < pagevec_count(&pvec); i++) {
761 			bh = head = page_buffers(pvec.pages[i]);
762 			do {
763 				if (buffer_dirty(bh) &&
764 						!buffer_async_write(bh)) {
765 					get_bh(bh);
766 					list_add_tail(&bh->b_assoc_buffers,
767 						      listp);
768 				}
769 				bh = bh->b_this_page;
770 			} while (bh != head);
771 		}
772 		pagevec_release(&pvec);
773 		cond_resched();
774 	}
775 }
776 
777 static void nilfs_dispose_list(struct the_nilfs *nilfs,
778 			       struct list_head *head, int force)
779 {
780 	struct nilfs_inode_info *ii, *n;
781 	struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
782 	unsigned int nv = 0;
783 
784 	while (!list_empty(head)) {
785 		spin_lock(&nilfs->ns_inode_lock);
786 		list_for_each_entry_safe(ii, n, head, i_dirty) {
787 			list_del_init(&ii->i_dirty);
788 			if (force) {
789 				if (unlikely(ii->i_bh)) {
790 					brelse(ii->i_bh);
791 					ii->i_bh = NULL;
792 				}
793 			} else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
794 				set_bit(NILFS_I_QUEUED, &ii->i_state);
795 				list_add_tail(&ii->i_dirty,
796 					      &nilfs->ns_dirty_files);
797 				continue;
798 			}
799 			ivec[nv++] = ii;
800 			if (nv == SC_N_INODEVEC)
801 				break;
802 		}
803 		spin_unlock(&nilfs->ns_inode_lock);
804 
805 		for (pii = ivec; nv > 0; pii++, nv--)
806 			iput(&(*pii)->vfs_inode);
807 	}
808 }
809 
810 static void nilfs_iput_work_func(struct work_struct *work)
811 {
812 	struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
813 						 sc_iput_work);
814 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
815 
816 	nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
817 }
818 
819 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
820 				     struct nilfs_root *root)
821 {
822 	int ret = 0;
823 
824 	if (nilfs_mdt_fetch_dirty(root->ifile))
825 		ret++;
826 	if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
827 		ret++;
828 	if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
829 		ret++;
830 	if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
831 		ret++;
832 	return ret;
833 }
834 
835 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
836 {
837 	return list_empty(&sci->sc_dirty_files) &&
838 		!test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
839 		sci->sc_nfreesegs == 0 &&
840 		(!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
841 }
842 
843 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
844 {
845 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
846 	int ret = 0;
847 
848 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
849 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
850 
851 	spin_lock(&nilfs->ns_inode_lock);
852 	if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
853 		ret++;
854 
855 	spin_unlock(&nilfs->ns_inode_lock);
856 	return ret;
857 }
858 
859 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
860 {
861 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
862 
863 	nilfs_mdt_clear_dirty(sci->sc_root->ifile);
864 	nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
865 	nilfs_mdt_clear_dirty(nilfs->ns_sufile);
866 	nilfs_mdt_clear_dirty(nilfs->ns_dat);
867 }
868 
869 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
870 {
871 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
872 	struct buffer_head *bh_cp;
873 	struct nilfs_checkpoint *raw_cp;
874 	int err;
875 
876 	/* XXX: this interface will be changed */
877 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
878 					  &raw_cp, &bh_cp);
879 	if (likely(!err)) {
880 		/*
881 		 * The following code is duplicated with cpfile.  But, it is
882 		 * needed to collect the checkpoint even if it was not newly
883 		 * created.
884 		 */
885 		mark_buffer_dirty(bh_cp);
886 		nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
887 		nilfs_cpfile_put_checkpoint(
888 			nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
889 	} else
890 		WARN_ON(err == -EINVAL || err == -ENOENT);
891 
892 	return err;
893 }
894 
895 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
896 {
897 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
898 	struct buffer_head *bh_cp;
899 	struct nilfs_checkpoint *raw_cp;
900 	int err;
901 
902 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
903 					  &raw_cp, &bh_cp);
904 	if (unlikely(err)) {
905 		WARN_ON(err == -EINVAL || err == -ENOENT);
906 		goto failed_ibh;
907 	}
908 	raw_cp->cp_snapshot_list.ssl_next = 0;
909 	raw_cp->cp_snapshot_list.ssl_prev = 0;
910 	raw_cp->cp_inodes_count =
911 		cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
912 	raw_cp->cp_blocks_count =
913 		cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
914 	raw_cp->cp_nblk_inc =
915 		cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
916 	raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
917 	raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
918 
919 	if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
920 		nilfs_checkpoint_clear_minor(raw_cp);
921 	else
922 		nilfs_checkpoint_set_minor(raw_cp);
923 
924 	nilfs_write_inode_common(sci->sc_root->ifile,
925 				 &raw_cp->cp_ifile_inode, 1);
926 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
927 	return 0;
928 
929  failed_ibh:
930 	return err;
931 }
932 
933 static void nilfs_fill_in_file_bmap(struct inode *ifile,
934 				    struct nilfs_inode_info *ii)
935 
936 {
937 	struct buffer_head *ibh;
938 	struct nilfs_inode *raw_inode;
939 
940 	if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
941 		ibh = ii->i_bh;
942 		BUG_ON(!ibh);
943 		raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
944 						  ibh);
945 		nilfs_bmap_write(ii->i_bmap, raw_inode);
946 		nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
947 	}
948 }
949 
950 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
951 {
952 	struct nilfs_inode_info *ii;
953 
954 	list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
955 		nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
956 		set_bit(NILFS_I_COLLECTED, &ii->i_state);
957 	}
958 }
959 
960 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
961 					     struct the_nilfs *nilfs)
962 {
963 	struct buffer_head *bh_sr;
964 	struct nilfs_super_root *raw_sr;
965 	unsigned int isz, srsz;
966 
967 	bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
968 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
969 	isz = nilfs->ns_inode_size;
970 	srsz = NILFS_SR_BYTES(isz);
971 
972 	raw_sr->sr_bytes = cpu_to_le16(srsz);
973 	raw_sr->sr_nongc_ctime
974 		= cpu_to_le64(nilfs_doing_gc() ?
975 			      nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
976 	raw_sr->sr_flags = 0;
977 
978 	nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
979 				 NILFS_SR_DAT_OFFSET(isz), 1);
980 	nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
981 				 NILFS_SR_CPFILE_OFFSET(isz), 1);
982 	nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
983 				 NILFS_SR_SUFILE_OFFSET(isz), 1);
984 	memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
985 }
986 
987 static void nilfs_redirty_inodes(struct list_head *head)
988 {
989 	struct nilfs_inode_info *ii;
990 
991 	list_for_each_entry(ii, head, i_dirty) {
992 		if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
993 			clear_bit(NILFS_I_COLLECTED, &ii->i_state);
994 	}
995 }
996 
997 static void nilfs_drop_collected_inodes(struct list_head *head)
998 {
999 	struct nilfs_inode_info *ii;
1000 
1001 	list_for_each_entry(ii, head, i_dirty) {
1002 		if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1003 			continue;
1004 
1005 		clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1006 		set_bit(NILFS_I_UPDATED, &ii->i_state);
1007 	}
1008 }
1009 
1010 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1011 				       struct inode *inode,
1012 				       struct list_head *listp,
1013 				       int (*collect)(struct nilfs_sc_info *,
1014 						      struct buffer_head *,
1015 						      struct inode *))
1016 {
1017 	struct buffer_head *bh, *n;
1018 	int err = 0;
1019 
1020 	if (collect) {
1021 		list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1022 			list_del_init(&bh->b_assoc_buffers);
1023 			err = collect(sci, bh, inode);
1024 			brelse(bh);
1025 			if (unlikely(err))
1026 				goto dispose_buffers;
1027 		}
1028 		return 0;
1029 	}
1030 
1031  dispose_buffers:
1032 	while (!list_empty(listp)) {
1033 		bh = list_first_entry(listp, struct buffer_head,
1034 				      b_assoc_buffers);
1035 		list_del_init(&bh->b_assoc_buffers);
1036 		brelse(bh);
1037 	}
1038 	return err;
1039 }
1040 
1041 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1042 {
1043 	/* Remaining number of blocks within segment buffer */
1044 	return sci->sc_segbuf_nblocks -
1045 		(sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1046 }
1047 
1048 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1049 				   struct inode *inode,
1050 				   const struct nilfs_sc_operations *sc_ops)
1051 {
1052 	LIST_HEAD(data_buffers);
1053 	LIST_HEAD(node_buffers);
1054 	int err;
1055 
1056 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1057 		size_t n, rest = nilfs_segctor_buffer_rest(sci);
1058 
1059 		n = nilfs_lookup_dirty_data_buffers(
1060 			inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1061 		if (n > rest) {
1062 			err = nilfs_segctor_apply_buffers(
1063 				sci, inode, &data_buffers,
1064 				sc_ops->collect_data);
1065 			BUG_ON(!err); /* always receive -E2BIG or true error */
1066 			goto break_or_fail;
1067 		}
1068 	}
1069 	nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1070 
1071 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1072 		err = nilfs_segctor_apply_buffers(
1073 			sci, inode, &data_buffers, sc_ops->collect_data);
1074 		if (unlikely(err)) {
1075 			/* dispose node list */
1076 			nilfs_segctor_apply_buffers(
1077 				sci, inode, &node_buffers, NULL);
1078 			goto break_or_fail;
1079 		}
1080 		sci->sc_stage.flags |= NILFS_CF_NODE;
1081 	}
1082 	/* Collect node */
1083 	err = nilfs_segctor_apply_buffers(
1084 		sci, inode, &node_buffers, sc_ops->collect_node);
1085 	if (unlikely(err))
1086 		goto break_or_fail;
1087 
1088 	nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1089 	err = nilfs_segctor_apply_buffers(
1090 		sci, inode, &node_buffers, sc_ops->collect_bmap);
1091 	if (unlikely(err))
1092 		goto break_or_fail;
1093 
1094 	nilfs_segctor_end_finfo(sci, inode);
1095 	sci->sc_stage.flags &= ~NILFS_CF_NODE;
1096 
1097  break_or_fail:
1098 	return err;
1099 }
1100 
1101 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1102 					 struct inode *inode)
1103 {
1104 	LIST_HEAD(data_buffers);
1105 	size_t n, rest = nilfs_segctor_buffer_rest(sci);
1106 	int err;
1107 
1108 	n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1109 					    sci->sc_dsync_start,
1110 					    sci->sc_dsync_end);
1111 
1112 	err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1113 					  nilfs_collect_file_data);
1114 	if (!err) {
1115 		nilfs_segctor_end_finfo(sci, inode);
1116 		BUG_ON(n > rest);
1117 		/* always receive -E2BIG or true error if n > rest */
1118 	}
1119 	return err;
1120 }
1121 
1122 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1123 {
1124 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1125 	struct list_head *head;
1126 	struct nilfs_inode_info *ii;
1127 	size_t ndone;
1128 	int err = 0;
1129 
1130 	switch (nilfs_sc_cstage_get(sci)) {
1131 	case NILFS_ST_INIT:
1132 		/* Pre-processes */
1133 		sci->sc_stage.flags = 0;
1134 
1135 		if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1136 			sci->sc_nblk_inc = 0;
1137 			sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1138 			if (mode == SC_LSEG_DSYNC) {
1139 				nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1140 				goto dsync_mode;
1141 			}
1142 		}
1143 
1144 		sci->sc_stage.dirty_file_ptr = NULL;
1145 		sci->sc_stage.gc_inode_ptr = NULL;
1146 		if (mode == SC_FLUSH_DAT) {
1147 			nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1148 			goto dat_stage;
1149 		}
1150 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1151 	case NILFS_ST_GC:
1152 		if (nilfs_doing_gc()) {
1153 			head = &sci->sc_gc_inodes;
1154 			ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1155 						head, i_dirty);
1156 			list_for_each_entry_continue(ii, head, i_dirty) {
1157 				err = nilfs_segctor_scan_file(
1158 					sci, &ii->vfs_inode,
1159 					&nilfs_sc_file_ops);
1160 				if (unlikely(err)) {
1161 					sci->sc_stage.gc_inode_ptr = list_entry(
1162 						ii->i_dirty.prev,
1163 						struct nilfs_inode_info,
1164 						i_dirty);
1165 					goto break_or_fail;
1166 				}
1167 				set_bit(NILFS_I_COLLECTED, &ii->i_state);
1168 			}
1169 			sci->sc_stage.gc_inode_ptr = NULL;
1170 		}
1171 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1172 	case NILFS_ST_FILE:
1173 		head = &sci->sc_dirty_files;
1174 		ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1175 					i_dirty);
1176 		list_for_each_entry_continue(ii, head, i_dirty) {
1177 			clear_bit(NILFS_I_DIRTY, &ii->i_state);
1178 
1179 			err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1180 						      &nilfs_sc_file_ops);
1181 			if (unlikely(err)) {
1182 				sci->sc_stage.dirty_file_ptr =
1183 					list_entry(ii->i_dirty.prev,
1184 						   struct nilfs_inode_info,
1185 						   i_dirty);
1186 				goto break_or_fail;
1187 			}
1188 			/* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1189 			/* XXX: required ? */
1190 		}
1191 		sci->sc_stage.dirty_file_ptr = NULL;
1192 		if (mode == SC_FLUSH_FILE) {
1193 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1194 			return 0;
1195 		}
1196 		nilfs_sc_cstage_inc(sci);
1197 		sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1198 		/* Fall through */
1199 	case NILFS_ST_IFILE:
1200 		err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1201 					      &nilfs_sc_file_ops);
1202 		if (unlikely(err))
1203 			break;
1204 		nilfs_sc_cstage_inc(sci);
1205 		/* Creating a checkpoint */
1206 		err = nilfs_segctor_create_checkpoint(sci);
1207 		if (unlikely(err))
1208 			break;
1209 		/* Fall through */
1210 	case NILFS_ST_CPFILE:
1211 		err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1212 					      &nilfs_sc_file_ops);
1213 		if (unlikely(err))
1214 			break;
1215 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1216 	case NILFS_ST_SUFILE:
1217 		err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1218 					 sci->sc_nfreesegs, &ndone);
1219 		if (unlikely(err)) {
1220 			nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1221 						  sci->sc_freesegs, ndone,
1222 						  NULL);
1223 			break;
1224 		}
1225 		sci->sc_stage.flags |= NILFS_CF_SUFREED;
1226 
1227 		err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1228 					      &nilfs_sc_file_ops);
1229 		if (unlikely(err))
1230 			break;
1231 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1232 	case NILFS_ST_DAT:
1233  dat_stage:
1234 		err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1235 					      &nilfs_sc_dat_ops);
1236 		if (unlikely(err))
1237 			break;
1238 		if (mode == SC_FLUSH_DAT) {
1239 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1240 			return 0;
1241 		}
1242 		nilfs_sc_cstage_inc(sci);  /* Fall through */
1243 	case NILFS_ST_SR:
1244 		if (mode == SC_LSEG_SR) {
1245 			/* Appending a super root */
1246 			err = nilfs_segctor_add_super_root(sci);
1247 			if (unlikely(err))
1248 				break;
1249 		}
1250 		/* End of a logical segment */
1251 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1252 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1253 		return 0;
1254 	case NILFS_ST_DSYNC:
1255  dsync_mode:
1256 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1257 		ii = sci->sc_dsync_inode;
1258 		if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1259 			break;
1260 
1261 		err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1262 		if (unlikely(err))
1263 			break;
1264 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1265 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1266 		return 0;
1267 	case NILFS_ST_DONE:
1268 		return 0;
1269 	default:
1270 		BUG();
1271 	}
1272 
1273  break_or_fail:
1274 	return err;
1275 }
1276 
1277 /**
1278  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1279  * @sci: nilfs_sc_info
1280  * @nilfs: nilfs object
1281  */
1282 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1283 					    struct the_nilfs *nilfs)
1284 {
1285 	struct nilfs_segment_buffer *segbuf, *prev;
1286 	__u64 nextnum;
1287 	int err, alloc = 0;
1288 
1289 	segbuf = nilfs_segbuf_new(sci->sc_super);
1290 	if (unlikely(!segbuf))
1291 		return -ENOMEM;
1292 
1293 	if (list_empty(&sci->sc_write_logs)) {
1294 		nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1295 				 nilfs->ns_pseg_offset, nilfs);
1296 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1297 			nilfs_shift_to_next_segment(nilfs);
1298 			nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1299 		}
1300 
1301 		segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1302 		nextnum = nilfs->ns_nextnum;
1303 
1304 		if (nilfs->ns_segnum == nilfs->ns_nextnum)
1305 			/* Start from the head of a new full segment */
1306 			alloc++;
1307 	} else {
1308 		/* Continue logs */
1309 		prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1310 		nilfs_segbuf_map_cont(segbuf, prev);
1311 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1312 		nextnum = prev->sb_nextnum;
1313 
1314 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1315 			nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1316 			segbuf->sb_sum.seg_seq++;
1317 			alloc++;
1318 		}
1319 	}
1320 
1321 	err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1322 	if (err)
1323 		goto failed;
1324 
1325 	if (alloc) {
1326 		err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1327 		if (err)
1328 			goto failed;
1329 	}
1330 	nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1331 
1332 	BUG_ON(!list_empty(&sci->sc_segbufs));
1333 	list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1334 	sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1335 	return 0;
1336 
1337  failed:
1338 	nilfs_segbuf_free(segbuf);
1339 	return err;
1340 }
1341 
1342 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1343 					 struct the_nilfs *nilfs, int nadd)
1344 {
1345 	struct nilfs_segment_buffer *segbuf, *prev;
1346 	struct inode *sufile = nilfs->ns_sufile;
1347 	__u64 nextnextnum;
1348 	LIST_HEAD(list);
1349 	int err, ret, i;
1350 
1351 	prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1352 	/*
1353 	 * Since the segment specified with nextnum might be allocated during
1354 	 * the previous construction, the buffer including its segusage may
1355 	 * not be dirty.  The following call ensures that the buffer is dirty
1356 	 * and will pin the buffer on memory until the sufile is written.
1357 	 */
1358 	err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1359 	if (unlikely(err))
1360 		return err;
1361 
1362 	for (i = 0; i < nadd; i++) {
1363 		/* extend segment info */
1364 		err = -ENOMEM;
1365 		segbuf = nilfs_segbuf_new(sci->sc_super);
1366 		if (unlikely(!segbuf))
1367 			goto failed;
1368 
1369 		/* map this buffer to region of segment on-disk */
1370 		nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1371 		sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1372 
1373 		/* allocate the next next full segment */
1374 		err = nilfs_sufile_alloc(sufile, &nextnextnum);
1375 		if (unlikely(err))
1376 			goto failed_segbuf;
1377 
1378 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1379 		nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1380 
1381 		list_add_tail(&segbuf->sb_list, &list);
1382 		prev = segbuf;
1383 	}
1384 	list_splice_tail(&list, &sci->sc_segbufs);
1385 	return 0;
1386 
1387  failed_segbuf:
1388 	nilfs_segbuf_free(segbuf);
1389  failed:
1390 	list_for_each_entry(segbuf, &list, sb_list) {
1391 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1392 		WARN_ON(ret); /* never fails */
1393 	}
1394 	nilfs_destroy_logs(&list);
1395 	return err;
1396 }
1397 
1398 static void nilfs_free_incomplete_logs(struct list_head *logs,
1399 				       struct the_nilfs *nilfs)
1400 {
1401 	struct nilfs_segment_buffer *segbuf, *prev;
1402 	struct inode *sufile = nilfs->ns_sufile;
1403 	int ret;
1404 
1405 	segbuf = NILFS_FIRST_SEGBUF(logs);
1406 	if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1407 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1408 		WARN_ON(ret); /* never fails */
1409 	}
1410 	if (atomic_read(&segbuf->sb_err)) {
1411 		/* Case 1: The first segment failed */
1412 		if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1413 			/*
1414 			 * Case 1a:  Partial segment appended into an existing
1415 			 * segment
1416 			 */
1417 			nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1418 						segbuf->sb_fseg_end);
1419 		else /* Case 1b:  New full segment */
1420 			set_nilfs_discontinued(nilfs);
1421 	}
1422 
1423 	prev = segbuf;
1424 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1425 		if (prev->sb_nextnum != segbuf->sb_nextnum) {
1426 			ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1427 			WARN_ON(ret); /* never fails */
1428 		}
1429 		if (atomic_read(&segbuf->sb_err) &&
1430 		    segbuf->sb_segnum != nilfs->ns_nextnum)
1431 			/* Case 2: extended segment (!= next) failed */
1432 			nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1433 		prev = segbuf;
1434 	}
1435 }
1436 
1437 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1438 					  struct inode *sufile)
1439 {
1440 	struct nilfs_segment_buffer *segbuf;
1441 	unsigned long live_blocks;
1442 	int ret;
1443 
1444 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1445 		live_blocks = segbuf->sb_sum.nblocks +
1446 			(segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1447 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1448 						     live_blocks,
1449 						     sci->sc_seg_ctime);
1450 		WARN_ON(ret); /* always succeed because the segusage is dirty */
1451 	}
1452 }
1453 
1454 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1455 {
1456 	struct nilfs_segment_buffer *segbuf;
1457 	int ret;
1458 
1459 	segbuf = NILFS_FIRST_SEGBUF(logs);
1460 	ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1461 					     segbuf->sb_pseg_start -
1462 					     segbuf->sb_fseg_start, 0);
1463 	WARN_ON(ret); /* always succeed because the segusage is dirty */
1464 
1465 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1466 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1467 						     0, 0);
1468 		WARN_ON(ret); /* always succeed */
1469 	}
1470 }
1471 
1472 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1473 					    struct nilfs_segment_buffer *last,
1474 					    struct inode *sufile)
1475 {
1476 	struct nilfs_segment_buffer *segbuf = last;
1477 	int ret;
1478 
1479 	list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1480 		sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1481 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1482 		WARN_ON(ret);
1483 	}
1484 	nilfs_truncate_logs(&sci->sc_segbufs, last);
1485 }
1486 
1487 
1488 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1489 				 struct the_nilfs *nilfs, int mode)
1490 {
1491 	struct nilfs_cstage prev_stage = sci->sc_stage;
1492 	int err, nadd = 1;
1493 
1494 	/* Collection retry loop */
1495 	for (;;) {
1496 		sci->sc_nblk_this_inc = 0;
1497 		sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1498 
1499 		err = nilfs_segctor_reset_segment_buffer(sci);
1500 		if (unlikely(err))
1501 			goto failed;
1502 
1503 		err = nilfs_segctor_collect_blocks(sci, mode);
1504 		sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1505 		if (!err)
1506 			break;
1507 
1508 		if (unlikely(err != -E2BIG))
1509 			goto failed;
1510 
1511 		/* The current segment is filled up */
1512 		if (mode != SC_LSEG_SR ||
1513 		    nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1514 			break;
1515 
1516 		nilfs_clear_logs(&sci->sc_segbufs);
1517 
1518 		if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1519 			err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1520 							sci->sc_freesegs,
1521 							sci->sc_nfreesegs,
1522 							NULL);
1523 			WARN_ON(err); /* do not happen */
1524 			sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1525 		}
1526 
1527 		err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1528 		if (unlikely(err))
1529 			return err;
1530 
1531 		nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1532 		sci->sc_stage = prev_stage;
1533 	}
1534 	nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1535 	return 0;
1536 
1537  failed:
1538 	return err;
1539 }
1540 
1541 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1542 				      struct buffer_head *new_bh)
1543 {
1544 	BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1545 
1546 	list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1547 	/* The caller must release old_bh */
1548 }
1549 
1550 static int
1551 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1552 				     struct nilfs_segment_buffer *segbuf,
1553 				     int mode)
1554 {
1555 	struct inode *inode = NULL;
1556 	sector_t blocknr;
1557 	unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1558 	unsigned long nblocks = 0, ndatablk = 0;
1559 	const struct nilfs_sc_operations *sc_op = NULL;
1560 	struct nilfs_segsum_pointer ssp;
1561 	struct nilfs_finfo *finfo = NULL;
1562 	union nilfs_binfo binfo;
1563 	struct buffer_head *bh, *bh_org;
1564 	ino_t ino = 0;
1565 	int err = 0;
1566 
1567 	if (!nfinfo)
1568 		goto out;
1569 
1570 	blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1571 	ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1572 	ssp.offset = sizeof(struct nilfs_segment_summary);
1573 
1574 	list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1575 		if (bh == segbuf->sb_super_root)
1576 			break;
1577 		if (!finfo) {
1578 			finfo =	nilfs_segctor_map_segsum_entry(
1579 				sci, &ssp, sizeof(*finfo));
1580 			ino = le64_to_cpu(finfo->fi_ino);
1581 			nblocks = le32_to_cpu(finfo->fi_nblocks);
1582 			ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1583 
1584 			inode = bh->b_page->mapping->host;
1585 
1586 			if (mode == SC_LSEG_DSYNC)
1587 				sc_op = &nilfs_sc_dsync_ops;
1588 			else if (ino == NILFS_DAT_INO)
1589 				sc_op = &nilfs_sc_dat_ops;
1590 			else /* file blocks */
1591 				sc_op = &nilfs_sc_file_ops;
1592 		}
1593 		bh_org = bh;
1594 		get_bh(bh_org);
1595 		err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1596 					&binfo);
1597 		if (bh != bh_org)
1598 			nilfs_list_replace_buffer(bh_org, bh);
1599 		brelse(bh_org);
1600 		if (unlikely(err))
1601 			goto failed_bmap;
1602 
1603 		if (ndatablk > 0)
1604 			sc_op->write_data_binfo(sci, &ssp, &binfo);
1605 		else
1606 			sc_op->write_node_binfo(sci, &ssp, &binfo);
1607 
1608 		blocknr++;
1609 		if (--nblocks == 0) {
1610 			finfo = NULL;
1611 			if (--nfinfo == 0)
1612 				break;
1613 		} else if (ndatablk > 0)
1614 			ndatablk--;
1615 	}
1616  out:
1617 	return 0;
1618 
1619  failed_bmap:
1620 	return err;
1621 }
1622 
1623 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1624 {
1625 	struct nilfs_segment_buffer *segbuf;
1626 	int err;
1627 
1628 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1629 		err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1630 		if (unlikely(err))
1631 			return err;
1632 		nilfs_segbuf_fill_in_segsum(segbuf);
1633 	}
1634 	return 0;
1635 }
1636 
1637 static void nilfs_begin_page_io(struct page *page)
1638 {
1639 	if (!page || PageWriteback(page))
1640 		/*
1641 		 * For split b-tree node pages, this function may be called
1642 		 * twice.  We ignore the 2nd or later calls by this check.
1643 		 */
1644 		return;
1645 
1646 	lock_page(page);
1647 	clear_page_dirty_for_io(page);
1648 	set_page_writeback(page);
1649 	unlock_page(page);
1650 }
1651 
1652 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1653 {
1654 	struct nilfs_segment_buffer *segbuf;
1655 	struct page *bd_page = NULL, *fs_page = NULL;
1656 
1657 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1658 		struct buffer_head *bh;
1659 
1660 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1661 				    b_assoc_buffers) {
1662 			if (bh->b_page != bd_page) {
1663 				if (bd_page) {
1664 					lock_page(bd_page);
1665 					clear_page_dirty_for_io(bd_page);
1666 					set_page_writeback(bd_page);
1667 					unlock_page(bd_page);
1668 				}
1669 				bd_page = bh->b_page;
1670 			}
1671 		}
1672 
1673 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1674 				    b_assoc_buffers) {
1675 			set_buffer_async_write(bh);
1676 			if (bh == segbuf->sb_super_root) {
1677 				if (bh->b_page != bd_page) {
1678 					lock_page(bd_page);
1679 					clear_page_dirty_for_io(bd_page);
1680 					set_page_writeback(bd_page);
1681 					unlock_page(bd_page);
1682 					bd_page = bh->b_page;
1683 				}
1684 				break;
1685 			}
1686 			if (bh->b_page != fs_page) {
1687 				nilfs_begin_page_io(fs_page);
1688 				fs_page = bh->b_page;
1689 			}
1690 		}
1691 	}
1692 	if (bd_page) {
1693 		lock_page(bd_page);
1694 		clear_page_dirty_for_io(bd_page);
1695 		set_page_writeback(bd_page);
1696 		unlock_page(bd_page);
1697 	}
1698 	nilfs_begin_page_io(fs_page);
1699 }
1700 
1701 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1702 			       struct the_nilfs *nilfs)
1703 {
1704 	int ret;
1705 
1706 	ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1707 	list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1708 	return ret;
1709 }
1710 
1711 static void nilfs_end_page_io(struct page *page, int err)
1712 {
1713 	if (!page)
1714 		return;
1715 
1716 	if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1717 		/*
1718 		 * For b-tree node pages, this function may be called twice
1719 		 * or more because they might be split in a segment.
1720 		 */
1721 		if (PageDirty(page)) {
1722 			/*
1723 			 * For pages holding split b-tree node buffers, dirty
1724 			 * flag on the buffers may be cleared discretely.
1725 			 * In that case, the page is once redirtied for
1726 			 * remaining buffers, and it must be cancelled if
1727 			 * all the buffers get cleaned later.
1728 			 */
1729 			lock_page(page);
1730 			if (nilfs_page_buffers_clean(page))
1731 				__nilfs_clear_page_dirty(page);
1732 			unlock_page(page);
1733 		}
1734 		return;
1735 	}
1736 
1737 	if (!err) {
1738 		if (!nilfs_page_buffers_clean(page))
1739 			__set_page_dirty_nobuffers(page);
1740 		ClearPageError(page);
1741 	} else {
1742 		__set_page_dirty_nobuffers(page);
1743 		SetPageError(page);
1744 	}
1745 
1746 	end_page_writeback(page);
1747 }
1748 
1749 static void nilfs_abort_logs(struct list_head *logs, int err)
1750 {
1751 	struct nilfs_segment_buffer *segbuf;
1752 	struct page *bd_page = NULL, *fs_page = NULL;
1753 	struct buffer_head *bh;
1754 
1755 	if (list_empty(logs))
1756 		return;
1757 
1758 	list_for_each_entry(segbuf, logs, sb_list) {
1759 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1760 				    b_assoc_buffers) {
1761 			if (bh->b_page != bd_page) {
1762 				if (bd_page)
1763 					end_page_writeback(bd_page);
1764 				bd_page = bh->b_page;
1765 			}
1766 		}
1767 
1768 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1769 				    b_assoc_buffers) {
1770 			clear_buffer_async_write(bh);
1771 			if (bh == segbuf->sb_super_root) {
1772 				if (bh->b_page != bd_page) {
1773 					end_page_writeback(bd_page);
1774 					bd_page = bh->b_page;
1775 				}
1776 				break;
1777 			}
1778 			if (bh->b_page != fs_page) {
1779 				nilfs_end_page_io(fs_page, err);
1780 				fs_page = bh->b_page;
1781 			}
1782 		}
1783 	}
1784 	if (bd_page)
1785 		end_page_writeback(bd_page);
1786 
1787 	nilfs_end_page_io(fs_page, err);
1788 }
1789 
1790 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1791 					     struct the_nilfs *nilfs, int err)
1792 {
1793 	LIST_HEAD(logs);
1794 	int ret;
1795 
1796 	list_splice_tail_init(&sci->sc_write_logs, &logs);
1797 	ret = nilfs_wait_on_logs(&logs);
1798 	nilfs_abort_logs(&logs, ret ? : err);
1799 
1800 	list_splice_tail_init(&sci->sc_segbufs, &logs);
1801 	nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1802 	nilfs_free_incomplete_logs(&logs, nilfs);
1803 
1804 	if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1805 		ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1806 						sci->sc_freesegs,
1807 						sci->sc_nfreesegs,
1808 						NULL);
1809 		WARN_ON(ret); /* do not happen */
1810 	}
1811 
1812 	nilfs_destroy_logs(&logs);
1813 }
1814 
1815 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1816 				   struct nilfs_segment_buffer *segbuf)
1817 {
1818 	nilfs->ns_segnum = segbuf->sb_segnum;
1819 	nilfs->ns_nextnum = segbuf->sb_nextnum;
1820 	nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1821 		+ segbuf->sb_sum.nblocks;
1822 	nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1823 	nilfs->ns_ctime = segbuf->sb_sum.ctime;
1824 }
1825 
1826 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1827 {
1828 	struct nilfs_segment_buffer *segbuf;
1829 	struct page *bd_page = NULL, *fs_page = NULL;
1830 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1831 	int update_sr = false;
1832 
1833 	list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1834 		struct buffer_head *bh;
1835 
1836 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1837 				    b_assoc_buffers) {
1838 			set_buffer_uptodate(bh);
1839 			clear_buffer_dirty(bh);
1840 			if (bh->b_page != bd_page) {
1841 				if (bd_page)
1842 					end_page_writeback(bd_page);
1843 				bd_page = bh->b_page;
1844 			}
1845 		}
1846 		/*
1847 		 * We assume that the buffers which belong to the same page
1848 		 * continue over the buffer list.
1849 		 * Under this assumption, the last BHs of pages is
1850 		 * identifiable by the discontinuity of bh->b_page
1851 		 * (page != fs_page).
1852 		 *
1853 		 * For B-tree node blocks, however, this assumption is not
1854 		 * guaranteed.  The cleanup code of B-tree node pages needs
1855 		 * special care.
1856 		 */
1857 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1858 				    b_assoc_buffers) {
1859 			const unsigned long set_bits = BIT(BH_Uptodate);
1860 			const unsigned long clear_bits =
1861 				(BIT(BH_Dirty) | BIT(BH_Async_Write) |
1862 				 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1863 				 BIT(BH_NILFS_Redirected));
1864 
1865 			set_mask_bits(&bh->b_state, clear_bits, set_bits);
1866 			if (bh == segbuf->sb_super_root) {
1867 				if (bh->b_page != bd_page) {
1868 					end_page_writeback(bd_page);
1869 					bd_page = bh->b_page;
1870 				}
1871 				update_sr = true;
1872 				break;
1873 			}
1874 			if (bh->b_page != fs_page) {
1875 				nilfs_end_page_io(fs_page, 0);
1876 				fs_page = bh->b_page;
1877 			}
1878 		}
1879 
1880 		if (!nilfs_segbuf_simplex(segbuf)) {
1881 			if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1882 				set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1883 				sci->sc_lseg_stime = jiffies;
1884 			}
1885 			if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1886 				clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1887 		}
1888 	}
1889 	/*
1890 	 * Since pages may continue over multiple segment buffers,
1891 	 * end of the last page must be checked outside of the loop.
1892 	 */
1893 	if (bd_page)
1894 		end_page_writeback(bd_page);
1895 
1896 	nilfs_end_page_io(fs_page, 0);
1897 
1898 	nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1899 
1900 	if (nilfs_doing_gc())
1901 		nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1902 	else
1903 		nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1904 
1905 	sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1906 
1907 	segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1908 	nilfs_set_next_segment(nilfs, segbuf);
1909 
1910 	if (update_sr) {
1911 		nilfs->ns_flushed_device = 0;
1912 		nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1913 				       segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1914 
1915 		clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1916 		clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1917 		set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1918 		nilfs_segctor_clear_metadata_dirty(sci);
1919 	} else
1920 		clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1921 }
1922 
1923 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1924 {
1925 	int ret;
1926 
1927 	ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1928 	if (!ret) {
1929 		nilfs_segctor_complete_write(sci);
1930 		nilfs_destroy_logs(&sci->sc_write_logs);
1931 	}
1932 	return ret;
1933 }
1934 
1935 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1936 					     struct the_nilfs *nilfs)
1937 {
1938 	struct nilfs_inode_info *ii, *n;
1939 	struct inode *ifile = sci->sc_root->ifile;
1940 
1941 	spin_lock(&nilfs->ns_inode_lock);
1942  retry:
1943 	list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1944 		if (!ii->i_bh) {
1945 			struct buffer_head *ibh;
1946 			int err;
1947 
1948 			spin_unlock(&nilfs->ns_inode_lock);
1949 			err = nilfs_ifile_get_inode_block(
1950 				ifile, ii->vfs_inode.i_ino, &ibh);
1951 			if (unlikely(err)) {
1952 				nilfs_msg(sci->sc_super, KERN_WARNING,
1953 					  "log writer: error %d getting inode block (ino=%lu)",
1954 					  err, ii->vfs_inode.i_ino);
1955 				return err;
1956 			}
1957 			spin_lock(&nilfs->ns_inode_lock);
1958 			if (likely(!ii->i_bh))
1959 				ii->i_bh = ibh;
1960 			else
1961 				brelse(ibh);
1962 			goto retry;
1963 		}
1964 
1965 		// Always redirty the buffer to avoid race condition
1966 		mark_buffer_dirty(ii->i_bh);
1967 		nilfs_mdt_mark_dirty(ifile);
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 & SB_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 = ktime_get_real_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_entry_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.head, wq.entry) {
2208 		if (!atomic_read(&wrq->done) &&
2209 		    nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2210 			wrq->err = err;
2211 			atomic_set(&wrq->done, 1);
2212 		}
2213 		if (atomic_read(&wrq->done)) {
2214 			wrq->wq.func(&wrq->wq,
2215 				     TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2216 				     0, NULL);
2217 		}
2218 	}
2219 	spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2220 }
2221 
2222 /**
2223  * nilfs_construct_segment - construct a logical segment
2224  * @sb: super block
2225  *
2226  * Return Value: On success, 0 is retured. On errors, one of the following
2227  * negative error code is returned.
2228  *
2229  * %-EROFS - Read only filesystem.
2230  *
2231  * %-EIO - I/O error
2232  *
2233  * %-ENOSPC - No space left on device (only in a panic state).
2234  *
2235  * %-ERESTARTSYS - Interrupted.
2236  *
2237  * %-ENOMEM - Insufficient memory available.
2238  */
2239 int nilfs_construct_segment(struct super_block *sb)
2240 {
2241 	struct the_nilfs *nilfs = sb->s_fs_info;
2242 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2243 	struct nilfs_transaction_info *ti;
2244 	int err;
2245 
2246 	if (!sci)
2247 		return -EROFS;
2248 
2249 	/* A call inside transactions causes a deadlock. */
2250 	BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2251 
2252 	err = nilfs_segctor_sync(sci);
2253 	return err;
2254 }
2255 
2256 /**
2257  * nilfs_construct_dsync_segment - construct a data-only logical segment
2258  * @sb: super block
2259  * @inode: inode whose data blocks should be written out
2260  * @start: start byte offset
2261  * @end: end byte offset (inclusive)
2262  *
2263  * Return Value: On success, 0 is retured. On errors, one of the following
2264  * negative error code is returned.
2265  *
2266  * %-EROFS - Read only filesystem.
2267  *
2268  * %-EIO - I/O error
2269  *
2270  * %-ENOSPC - No space left on device (only in a panic state).
2271  *
2272  * %-ERESTARTSYS - Interrupted.
2273  *
2274  * %-ENOMEM - Insufficient memory available.
2275  */
2276 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2277 				  loff_t start, loff_t end)
2278 {
2279 	struct the_nilfs *nilfs = sb->s_fs_info;
2280 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2281 	struct nilfs_inode_info *ii;
2282 	struct nilfs_transaction_info ti;
2283 	int err = 0;
2284 
2285 	if (!sci)
2286 		return -EROFS;
2287 
2288 	nilfs_transaction_lock(sb, &ti, 0);
2289 
2290 	ii = NILFS_I(inode);
2291 	if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2292 	    nilfs_test_opt(nilfs, STRICT_ORDER) ||
2293 	    test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2294 	    nilfs_discontinued(nilfs)) {
2295 		nilfs_transaction_unlock(sb);
2296 		err = nilfs_segctor_sync(sci);
2297 		return err;
2298 	}
2299 
2300 	spin_lock(&nilfs->ns_inode_lock);
2301 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2302 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2303 		spin_unlock(&nilfs->ns_inode_lock);
2304 		nilfs_transaction_unlock(sb);
2305 		return 0;
2306 	}
2307 	spin_unlock(&nilfs->ns_inode_lock);
2308 	sci->sc_dsync_inode = ii;
2309 	sci->sc_dsync_start = start;
2310 	sci->sc_dsync_end = end;
2311 
2312 	err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2313 	if (!err)
2314 		nilfs->ns_flushed_device = 0;
2315 
2316 	nilfs_transaction_unlock(sb);
2317 	return err;
2318 }
2319 
2320 #define FLUSH_FILE_BIT	(0x1) /* data file only */
2321 #define FLUSH_DAT_BIT	BIT(NILFS_DAT_INO) /* DAT only */
2322 
2323 /**
2324  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2325  * @sci: segment constructor object
2326  */
2327 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2328 {
2329 	spin_lock(&sci->sc_state_lock);
2330 	sci->sc_seq_accepted = sci->sc_seq_request;
2331 	spin_unlock(&sci->sc_state_lock);
2332 	del_timer_sync(&sci->sc_timer);
2333 }
2334 
2335 /**
2336  * nilfs_segctor_notify - notify the result of request to caller threads
2337  * @sci: segment constructor object
2338  * @mode: mode of log forming
2339  * @err: error code to be notified
2340  */
2341 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2342 {
2343 	/* Clear requests (even when the construction failed) */
2344 	spin_lock(&sci->sc_state_lock);
2345 
2346 	if (mode == SC_LSEG_SR) {
2347 		sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2348 		sci->sc_seq_done = sci->sc_seq_accepted;
2349 		nilfs_segctor_wakeup(sci, err);
2350 		sci->sc_flush_request = 0;
2351 	} else {
2352 		if (mode == SC_FLUSH_FILE)
2353 			sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2354 		else if (mode == SC_FLUSH_DAT)
2355 			sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2356 
2357 		/* re-enable timer if checkpoint creation was not done */
2358 		if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2359 		    time_before(jiffies, sci->sc_timer.expires))
2360 			add_timer(&sci->sc_timer);
2361 	}
2362 	spin_unlock(&sci->sc_state_lock);
2363 }
2364 
2365 /**
2366  * nilfs_segctor_construct - form logs and write them to disk
2367  * @sci: segment constructor object
2368  * @mode: mode of log forming
2369  */
2370 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2371 {
2372 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2373 	struct nilfs_super_block **sbp;
2374 	int err = 0;
2375 
2376 	nilfs_segctor_accept(sci);
2377 
2378 	if (nilfs_discontinued(nilfs))
2379 		mode = SC_LSEG_SR;
2380 	if (!nilfs_segctor_confirm(sci))
2381 		err = nilfs_segctor_do_construct(sci, mode);
2382 
2383 	if (likely(!err)) {
2384 		if (mode != SC_FLUSH_DAT)
2385 			atomic_set(&nilfs->ns_ndirtyblks, 0);
2386 		if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2387 		    nilfs_discontinued(nilfs)) {
2388 			down_write(&nilfs->ns_sem);
2389 			err = -EIO;
2390 			sbp = nilfs_prepare_super(sci->sc_super,
2391 						  nilfs_sb_will_flip(nilfs));
2392 			if (likely(sbp)) {
2393 				nilfs_set_log_cursor(sbp[0], nilfs);
2394 				err = nilfs_commit_super(sci->sc_super,
2395 							 NILFS_SB_COMMIT);
2396 			}
2397 			up_write(&nilfs->ns_sem);
2398 		}
2399 	}
2400 
2401 	nilfs_segctor_notify(sci, mode, err);
2402 	return err;
2403 }
2404 
2405 static void nilfs_construction_timeout(struct timer_list *t)
2406 {
2407 	struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2408 
2409 	wake_up_process(sci->sc_timer_task);
2410 }
2411 
2412 static void
2413 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2414 {
2415 	struct nilfs_inode_info *ii, *n;
2416 
2417 	list_for_each_entry_safe(ii, n, head, i_dirty) {
2418 		if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2419 			continue;
2420 		list_del_init(&ii->i_dirty);
2421 		truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2422 		nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2423 		iput(&ii->vfs_inode);
2424 	}
2425 }
2426 
2427 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2428 			 void **kbufs)
2429 {
2430 	struct the_nilfs *nilfs = sb->s_fs_info;
2431 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2432 	struct nilfs_transaction_info ti;
2433 	int err;
2434 
2435 	if (unlikely(!sci))
2436 		return -EROFS;
2437 
2438 	nilfs_transaction_lock(sb, &ti, 1);
2439 
2440 	err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2441 	if (unlikely(err))
2442 		goto out_unlock;
2443 
2444 	err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2445 	if (unlikely(err)) {
2446 		nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2447 		goto out_unlock;
2448 	}
2449 
2450 	sci->sc_freesegs = kbufs[4];
2451 	sci->sc_nfreesegs = argv[4].v_nmembs;
2452 	list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2453 
2454 	for (;;) {
2455 		err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2456 		nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2457 
2458 		if (likely(!err))
2459 			break;
2460 
2461 		nilfs_msg(sb, KERN_WARNING, "error %d cleaning segments", err);
2462 		set_current_state(TASK_INTERRUPTIBLE);
2463 		schedule_timeout(sci->sc_interval);
2464 	}
2465 	if (nilfs_test_opt(nilfs, DISCARD)) {
2466 		int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2467 						 sci->sc_nfreesegs);
2468 		if (ret) {
2469 			nilfs_msg(sb, KERN_WARNING,
2470 				  "error %d on discard request, turning discards off for the device",
2471 				  ret);
2472 			nilfs_clear_opt(nilfs, DISCARD);
2473 		}
2474 	}
2475 
2476  out_unlock:
2477 	sci->sc_freesegs = NULL;
2478 	sci->sc_nfreesegs = 0;
2479 	nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2480 	nilfs_transaction_unlock(sb);
2481 	return err;
2482 }
2483 
2484 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2485 {
2486 	struct nilfs_transaction_info ti;
2487 
2488 	nilfs_transaction_lock(sci->sc_super, &ti, 0);
2489 	nilfs_segctor_construct(sci, mode);
2490 
2491 	/*
2492 	 * Unclosed segment should be retried.  We do this using sc_timer.
2493 	 * Timeout of sc_timer will invoke complete construction which leads
2494 	 * to close the current logical segment.
2495 	 */
2496 	if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2497 		nilfs_segctor_start_timer(sci);
2498 
2499 	nilfs_transaction_unlock(sci->sc_super);
2500 }
2501 
2502 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2503 {
2504 	int mode = 0;
2505 
2506 	spin_lock(&sci->sc_state_lock);
2507 	mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2508 		SC_FLUSH_DAT : SC_FLUSH_FILE;
2509 	spin_unlock(&sci->sc_state_lock);
2510 
2511 	if (mode) {
2512 		nilfs_segctor_do_construct(sci, mode);
2513 
2514 		spin_lock(&sci->sc_state_lock);
2515 		sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2516 			~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2517 		spin_unlock(&sci->sc_state_lock);
2518 	}
2519 	clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2520 }
2521 
2522 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2523 {
2524 	if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2525 	    time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2526 		if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2527 			return SC_FLUSH_FILE;
2528 		else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2529 			return SC_FLUSH_DAT;
2530 	}
2531 	return SC_LSEG_SR;
2532 }
2533 
2534 /**
2535  * nilfs_segctor_thread - main loop of the segment constructor thread.
2536  * @arg: pointer to a struct nilfs_sc_info.
2537  *
2538  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2539  * to execute segment constructions.
2540  */
2541 static int nilfs_segctor_thread(void *arg)
2542 {
2543 	struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2544 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2545 	int timeout = 0;
2546 
2547 	sci->sc_timer_task = current;
2548 
2549 	/* start sync. */
2550 	sci->sc_task = current;
2551 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2552 	nilfs_msg(sci->sc_super, KERN_INFO,
2553 		  "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2554 		  sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2555 
2556 	spin_lock(&sci->sc_state_lock);
2557  loop:
2558 	for (;;) {
2559 		int mode;
2560 
2561 		if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2562 			goto end_thread;
2563 
2564 		if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2565 			mode = SC_LSEG_SR;
2566 		else if (sci->sc_flush_request)
2567 			mode = nilfs_segctor_flush_mode(sci);
2568 		else
2569 			break;
2570 
2571 		spin_unlock(&sci->sc_state_lock);
2572 		nilfs_segctor_thread_construct(sci, mode);
2573 		spin_lock(&sci->sc_state_lock);
2574 		timeout = 0;
2575 	}
2576 
2577 
2578 	if (freezing(current)) {
2579 		spin_unlock(&sci->sc_state_lock);
2580 		try_to_freeze();
2581 		spin_lock(&sci->sc_state_lock);
2582 	} else {
2583 		DEFINE_WAIT(wait);
2584 		int should_sleep = 1;
2585 
2586 		prepare_to_wait(&sci->sc_wait_daemon, &wait,
2587 				TASK_INTERRUPTIBLE);
2588 
2589 		if (sci->sc_seq_request != sci->sc_seq_done)
2590 			should_sleep = 0;
2591 		else if (sci->sc_flush_request)
2592 			should_sleep = 0;
2593 		else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2594 			should_sleep = time_before(jiffies,
2595 					sci->sc_timer.expires);
2596 
2597 		if (should_sleep) {
2598 			spin_unlock(&sci->sc_state_lock);
2599 			schedule();
2600 			spin_lock(&sci->sc_state_lock);
2601 		}
2602 		finish_wait(&sci->sc_wait_daemon, &wait);
2603 		timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2604 			   time_after_eq(jiffies, sci->sc_timer.expires));
2605 
2606 		if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2607 			set_nilfs_discontinued(nilfs);
2608 	}
2609 	goto loop;
2610 
2611  end_thread:
2612 	spin_unlock(&sci->sc_state_lock);
2613 
2614 	/* end sync. */
2615 	sci->sc_task = NULL;
2616 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2617 	return 0;
2618 }
2619 
2620 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2621 {
2622 	struct task_struct *t;
2623 
2624 	t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2625 	if (IS_ERR(t)) {
2626 		int err = PTR_ERR(t);
2627 
2628 		nilfs_msg(sci->sc_super, KERN_ERR,
2629 			  "error %d creating segctord thread", err);
2630 		return err;
2631 	}
2632 	wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2633 	return 0;
2634 }
2635 
2636 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2637 	__acquires(&sci->sc_state_lock)
2638 	__releases(&sci->sc_state_lock)
2639 {
2640 	sci->sc_state |= NILFS_SEGCTOR_QUIT;
2641 
2642 	while (sci->sc_task) {
2643 		wake_up(&sci->sc_wait_daemon);
2644 		spin_unlock(&sci->sc_state_lock);
2645 		wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2646 		spin_lock(&sci->sc_state_lock);
2647 	}
2648 }
2649 
2650 /*
2651  * Setup & clean-up functions
2652  */
2653 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2654 					       struct nilfs_root *root)
2655 {
2656 	struct the_nilfs *nilfs = sb->s_fs_info;
2657 	struct nilfs_sc_info *sci;
2658 
2659 	sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2660 	if (!sci)
2661 		return NULL;
2662 
2663 	sci->sc_super = sb;
2664 
2665 	nilfs_get_root(root);
2666 	sci->sc_root = root;
2667 
2668 	init_waitqueue_head(&sci->sc_wait_request);
2669 	init_waitqueue_head(&sci->sc_wait_daemon);
2670 	init_waitqueue_head(&sci->sc_wait_task);
2671 	spin_lock_init(&sci->sc_state_lock);
2672 	INIT_LIST_HEAD(&sci->sc_dirty_files);
2673 	INIT_LIST_HEAD(&sci->sc_segbufs);
2674 	INIT_LIST_HEAD(&sci->sc_write_logs);
2675 	INIT_LIST_HEAD(&sci->sc_gc_inodes);
2676 	INIT_LIST_HEAD(&sci->sc_iput_queue);
2677 	INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2678 	timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2679 
2680 	sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2681 	sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2682 	sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2683 
2684 	if (nilfs->ns_interval)
2685 		sci->sc_interval = HZ * nilfs->ns_interval;
2686 	if (nilfs->ns_watermark)
2687 		sci->sc_watermark = nilfs->ns_watermark;
2688 	return sci;
2689 }
2690 
2691 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2692 {
2693 	int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2694 
2695 	/*
2696 	 * The segctord thread was stopped and its timer was removed.
2697 	 * But some tasks remain.
2698 	 */
2699 	do {
2700 		struct nilfs_transaction_info ti;
2701 
2702 		nilfs_transaction_lock(sci->sc_super, &ti, 0);
2703 		ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2704 		nilfs_transaction_unlock(sci->sc_super);
2705 
2706 		flush_work(&sci->sc_iput_work);
2707 
2708 	} while (ret && retrycount-- > 0);
2709 }
2710 
2711 /**
2712  * nilfs_segctor_destroy - destroy the segment constructor.
2713  * @sci: nilfs_sc_info
2714  *
2715  * nilfs_segctor_destroy() kills the segctord thread and frees
2716  * the nilfs_sc_info struct.
2717  * Caller must hold the segment semaphore.
2718  */
2719 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2720 {
2721 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2722 	int flag;
2723 
2724 	up_write(&nilfs->ns_segctor_sem);
2725 
2726 	spin_lock(&sci->sc_state_lock);
2727 	nilfs_segctor_kill_thread(sci);
2728 	flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2729 		|| sci->sc_seq_request != sci->sc_seq_done);
2730 	spin_unlock(&sci->sc_state_lock);
2731 
2732 	if (flush_work(&sci->sc_iput_work))
2733 		flag = true;
2734 
2735 	if (flag || !nilfs_segctor_confirm(sci))
2736 		nilfs_segctor_write_out(sci);
2737 
2738 	if (!list_empty(&sci->sc_dirty_files)) {
2739 		nilfs_msg(sci->sc_super, KERN_WARNING,
2740 			  "disposed unprocessed dirty file(s) when stopping log writer");
2741 		nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2742 	}
2743 
2744 	if (!list_empty(&sci->sc_iput_queue)) {
2745 		nilfs_msg(sci->sc_super, KERN_WARNING,
2746 			  "disposed unprocessed inode(s) in iput queue when stopping log writer");
2747 		nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2748 	}
2749 
2750 	WARN_ON(!list_empty(&sci->sc_segbufs));
2751 	WARN_ON(!list_empty(&sci->sc_write_logs));
2752 
2753 	nilfs_put_root(sci->sc_root);
2754 
2755 	down_write(&nilfs->ns_segctor_sem);
2756 
2757 	del_timer_sync(&sci->sc_timer);
2758 	kfree(sci);
2759 }
2760 
2761 /**
2762  * nilfs_attach_log_writer - attach log writer
2763  * @sb: super block instance
2764  * @root: root object of the current filesystem tree
2765  *
2766  * This allocates a log writer object, initializes it, and starts the
2767  * log writer.
2768  *
2769  * Return Value: On success, 0 is returned. On error, one of the following
2770  * negative error code is returned.
2771  *
2772  * %-ENOMEM - Insufficient memory available.
2773  */
2774 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2775 {
2776 	struct the_nilfs *nilfs = sb->s_fs_info;
2777 	int err;
2778 
2779 	if (nilfs->ns_writer) {
2780 		/*
2781 		 * This happens if the filesystem was remounted
2782 		 * read/write after nilfs_error degenerated it into a
2783 		 * read-only mount.
2784 		 */
2785 		nilfs_detach_log_writer(sb);
2786 	}
2787 
2788 	nilfs->ns_writer = nilfs_segctor_new(sb, root);
2789 	if (!nilfs->ns_writer)
2790 		return -ENOMEM;
2791 
2792 	err = nilfs_segctor_start_thread(nilfs->ns_writer);
2793 	if (err) {
2794 		kfree(nilfs->ns_writer);
2795 		nilfs->ns_writer = NULL;
2796 	}
2797 	return err;
2798 }
2799 
2800 /**
2801  * nilfs_detach_log_writer - destroy log writer
2802  * @sb: super block instance
2803  *
2804  * This kills log writer daemon, frees the log writer object, and
2805  * destroys list of dirty files.
2806  */
2807 void nilfs_detach_log_writer(struct super_block *sb)
2808 {
2809 	struct the_nilfs *nilfs = sb->s_fs_info;
2810 	LIST_HEAD(garbage_list);
2811 
2812 	down_write(&nilfs->ns_segctor_sem);
2813 	if (nilfs->ns_writer) {
2814 		nilfs_segctor_destroy(nilfs->ns_writer);
2815 		nilfs->ns_writer = NULL;
2816 	}
2817 
2818 	/* Force to free the list of dirty files */
2819 	spin_lock(&nilfs->ns_inode_lock);
2820 	if (!list_empty(&nilfs->ns_dirty_files)) {
2821 		list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2822 		nilfs_msg(sb, KERN_WARNING,
2823 			  "disposed unprocessed dirty file(s) when detaching log writer");
2824 	}
2825 	spin_unlock(&nilfs->ns_inode_lock);
2826 	up_write(&nilfs->ns_segctor_sem);
2827 
2828 	nilfs_dispose_list(nilfs, &garbage_list, 1);
2829 }
2830