xref: /openbmc/linux/fs/nilfs2/segment.c (revision e40573a4)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * NILFS segment constructor.
4  *
5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6  *
7  * Written by Ryusuke Konishi.
8  *
9  */
10 
11 #include <linux/pagemap.h>
12 #include <linux/buffer_head.h>
13 #include <linux/writeback.h>
14 #include <linux/bitops.h>
15 #include <linux/bio.h>
16 #include <linux/completion.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/freezer.h>
20 #include <linux/kthread.h>
21 #include <linux/crc32.h>
22 #include <linux/pagevec.h>
23 #include <linux/slab.h>
24 #include <linux/sched/signal.h>
25 
26 #include "nilfs.h"
27 #include "btnode.h"
28 #include "page.h"
29 #include "segment.h"
30 #include "sufile.h"
31 #include "cpfile.h"
32 #include "ifile.h"
33 #include "segbuf.h"
34 
35 
36 /*
37  * Segment constructor
38  */
39 #define SC_N_INODEVEC	16   /* Size of locally allocated inode vector */
40 
41 #define SC_MAX_SEGDELTA 64   /*
42 			      * Upper limit of the number of segments
43 			      * appended in collection retry loop
44 			      */
45 
46 /* Construction mode */
47 enum {
48 	SC_LSEG_SR = 1,	/* Make a logical segment having a super root */
49 	SC_LSEG_DSYNC,	/*
50 			 * Flush data blocks of a given file and make
51 			 * a logical segment without a super root.
52 			 */
53 	SC_FLUSH_FILE,	/*
54 			 * Flush data files, leads to segment writes without
55 			 * creating a checkpoint.
56 			 */
57 	SC_FLUSH_DAT,	/*
58 			 * Flush DAT file.  This also creates segments
59 			 * without a checkpoint.
60 			 */
61 };
62 
63 /* Stage numbers of dirty block collection */
64 enum {
65 	NILFS_ST_INIT = 0,
66 	NILFS_ST_GC,		/* Collecting dirty blocks for GC */
67 	NILFS_ST_FILE,
68 	NILFS_ST_IFILE,
69 	NILFS_ST_CPFILE,
70 	NILFS_ST_SUFILE,
71 	NILFS_ST_DAT,
72 	NILFS_ST_SR,		/* Super root */
73 	NILFS_ST_DSYNC,		/* Data sync blocks */
74 	NILFS_ST_DONE,
75 };
76 
77 #define CREATE_TRACE_POINTS
78 #include <trace/events/nilfs2.h>
79 
80 /*
81  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
82  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
83  * the variable must use them because transition of stage count must involve
84  * trace events (trace_nilfs2_collection_stage_transition).
85  *
86  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
87  * produce tracepoint events. It is provided just for making the intention
88  * clear.
89  */
90 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
91 {
92 	sci->sc_stage.scnt++;
93 	trace_nilfs2_collection_stage_transition(sci);
94 }
95 
96 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
97 {
98 	sci->sc_stage.scnt = next_scnt;
99 	trace_nilfs2_collection_stage_transition(sci);
100 }
101 
102 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
103 {
104 	return sci->sc_stage.scnt;
105 }
106 
107 /* State flags of collection */
108 #define NILFS_CF_NODE		0x0001	/* Collecting node blocks */
109 #define NILFS_CF_IFILE_STARTED	0x0002	/* IFILE stage has started */
110 #define NILFS_CF_SUFREED	0x0004	/* segment usages has been freed */
111 #define NILFS_CF_HISTORY_MASK	(NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
112 
113 /* Operations depending on the construction mode and file type */
114 struct nilfs_sc_operations {
115 	int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
116 			    struct inode *);
117 	int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
118 			    struct inode *);
119 	int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
120 			    struct inode *);
121 	void (*write_data_binfo)(struct nilfs_sc_info *,
122 				 struct nilfs_segsum_pointer *,
123 				 union nilfs_binfo *);
124 	void (*write_node_binfo)(struct nilfs_sc_info *,
125 				 struct nilfs_segsum_pointer *,
126 				 union nilfs_binfo *);
127 };
128 
129 /*
130  * Other definitions
131  */
132 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
133 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
134 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
135 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
136 
137 #define nilfs_cnt32_ge(a, b)   \
138 	(typecheck(__u32, a) && typecheck(__u32, b) && \
139 	 ((__s32)(a) - (__s32)(b) >= 0))
140 
141 static int nilfs_prepare_segment_lock(struct super_block *sb,
142 				      struct nilfs_transaction_info *ti)
143 {
144 	struct nilfs_transaction_info *cur_ti = current->journal_info;
145 	void *save = NULL;
146 
147 	if (cur_ti) {
148 		if (cur_ti->ti_magic == NILFS_TI_MAGIC)
149 			return ++cur_ti->ti_count;
150 
151 		/*
152 		 * If journal_info field is occupied by other FS,
153 		 * it is saved and will be restored on
154 		 * nilfs_transaction_commit().
155 		 */
156 		nilfs_warn(sb, "journal info from a different FS");
157 		save = current->journal_info;
158 	}
159 	if (!ti) {
160 		ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
161 		if (!ti)
162 			return -ENOMEM;
163 		ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
164 	} else {
165 		ti->ti_flags = 0;
166 	}
167 	ti->ti_count = 0;
168 	ti->ti_save = save;
169 	ti->ti_magic = NILFS_TI_MAGIC;
170 	current->journal_info = ti;
171 	return 0;
172 }
173 
174 /**
175  * nilfs_transaction_begin - start indivisible file operations.
176  * @sb: super block
177  * @ti: nilfs_transaction_info
178  * @vacancy_check: flags for vacancy rate checks
179  *
180  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
181  * the segment semaphore, to make a segment construction and write tasks
182  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
183  * The region enclosed by these two functions can be nested.  To avoid a
184  * deadlock, the semaphore is only acquired or released in the outermost call.
185  *
186  * This function allocates a nilfs_transaction_info struct to keep context
187  * information on it.  It is initialized and hooked onto the current task in
188  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
189  * instead; otherwise a new struct is assigned from a slab.
190  *
191  * When @vacancy_check flag is set, this function will check the amount of
192  * free space, and will wait for the GC to reclaim disk space if low capacity.
193  *
194  * Return Value: On success, 0 is returned. On error, one of the following
195  * negative error code is returned.
196  *
197  * %-ENOMEM - Insufficient memory available.
198  *
199  * %-ENOSPC - No space left on device
200  */
201 int nilfs_transaction_begin(struct super_block *sb,
202 			    struct nilfs_transaction_info *ti,
203 			    int vacancy_check)
204 {
205 	struct the_nilfs *nilfs;
206 	int ret = nilfs_prepare_segment_lock(sb, ti);
207 	struct nilfs_transaction_info *trace_ti;
208 
209 	if (unlikely(ret < 0))
210 		return ret;
211 	if (ret > 0) {
212 		trace_ti = current->journal_info;
213 
214 		trace_nilfs2_transaction_transition(sb, trace_ti,
215 				    trace_ti->ti_count, trace_ti->ti_flags,
216 				    TRACE_NILFS2_TRANSACTION_BEGIN);
217 		return 0;
218 	}
219 
220 	sb_start_intwrite(sb);
221 
222 	nilfs = sb->s_fs_info;
223 	down_read(&nilfs->ns_segctor_sem);
224 	if (vacancy_check && nilfs_near_disk_full(nilfs)) {
225 		up_read(&nilfs->ns_segctor_sem);
226 		ret = -ENOSPC;
227 		goto failed;
228 	}
229 
230 	trace_ti = current->journal_info;
231 	trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
232 					    trace_ti->ti_flags,
233 					    TRACE_NILFS2_TRANSACTION_BEGIN);
234 	return 0;
235 
236  failed:
237 	ti = current->journal_info;
238 	current->journal_info = ti->ti_save;
239 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
240 		kmem_cache_free(nilfs_transaction_cachep, ti);
241 	sb_end_intwrite(sb);
242 	return ret;
243 }
244 
245 /**
246  * nilfs_transaction_commit - commit indivisible file operations.
247  * @sb: super block
248  *
249  * nilfs_transaction_commit() releases the read semaphore which is
250  * acquired by nilfs_transaction_begin(). This is only performed
251  * in outermost call of this function.  If a commit flag is set,
252  * nilfs_transaction_commit() sets a timer to start the segment
253  * constructor.  If a sync flag is set, it starts construction
254  * directly.
255  */
256 int nilfs_transaction_commit(struct super_block *sb)
257 {
258 	struct nilfs_transaction_info *ti = current->journal_info;
259 	struct the_nilfs *nilfs = sb->s_fs_info;
260 	int err = 0;
261 
262 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
263 	ti->ti_flags |= NILFS_TI_COMMIT;
264 	if (ti->ti_count > 0) {
265 		ti->ti_count--;
266 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
267 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
268 		return 0;
269 	}
270 	if (nilfs->ns_writer) {
271 		struct nilfs_sc_info *sci = nilfs->ns_writer;
272 
273 		if (ti->ti_flags & NILFS_TI_COMMIT)
274 			nilfs_segctor_start_timer(sci);
275 		if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
276 			nilfs_segctor_do_flush(sci, 0);
277 	}
278 	up_read(&nilfs->ns_segctor_sem);
279 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
280 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
281 
282 	current->journal_info = ti->ti_save;
283 
284 	if (ti->ti_flags & NILFS_TI_SYNC)
285 		err = nilfs_construct_segment(sb);
286 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
287 		kmem_cache_free(nilfs_transaction_cachep, ti);
288 	sb_end_intwrite(sb);
289 	return err;
290 }
291 
292 void nilfs_transaction_abort(struct super_block *sb)
293 {
294 	struct nilfs_transaction_info *ti = current->journal_info;
295 	struct the_nilfs *nilfs = sb->s_fs_info;
296 
297 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
298 	if (ti->ti_count > 0) {
299 		ti->ti_count--;
300 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
301 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
302 		return;
303 	}
304 	up_read(&nilfs->ns_segctor_sem);
305 
306 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
307 		    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
308 
309 	current->journal_info = ti->ti_save;
310 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
311 		kmem_cache_free(nilfs_transaction_cachep, ti);
312 	sb_end_intwrite(sb);
313 }
314 
315 void nilfs_relax_pressure_in_lock(struct super_block *sb)
316 {
317 	struct the_nilfs *nilfs = sb->s_fs_info;
318 	struct nilfs_sc_info *sci = nilfs->ns_writer;
319 
320 	if (!sci || !sci->sc_flush_request)
321 		return;
322 
323 	set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
324 	up_read(&nilfs->ns_segctor_sem);
325 
326 	down_write(&nilfs->ns_segctor_sem);
327 	if (sci->sc_flush_request &&
328 	    test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
329 		struct nilfs_transaction_info *ti = current->journal_info;
330 
331 		ti->ti_flags |= NILFS_TI_WRITER;
332 		nilfs_segctor_do_immediate_flush(sci);
333 		ti->ti_flags &= ~NILFS_TI_WRITER;
334 	}
335 	downgrade_write(&nilfs->ns_segctor_sem);
336 }
337 
338 static void nilfs_transaction_lock(struct super_block *sb,
339 				   struct nilfs_transaction_info *ti,
340 				   int gcflag)
341 {
342 	struct nilfs_transaction_info *cur_ti = current->journal_info;
343 	struct the_nilfs *nilfs = sb->s_fs_info;
344 	struct nilfs_sc_info *sci = nilfs->ns_writer;
345 
346 	WARN_ON(cur_ti);
347 	ti->ti_flags = NILFS_TI_WRITER;
348 	ti->ti_count = 0;
349 	ti->ti_save = cur_ti;
350 	ti->ti_magic = NILFS_TI_MAGIC;
351 	current->journal_info = ti;
352 
353 	for (;;) {
354 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
355 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
356 
357 		down_write(&nilfs->ns_segctor_sem);
358 		if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
359 			break;
360 
361 		nilfs_segctor_do_immediate_flush(sci);
362 
363 		up_write(&nilfs->ns_segctor_sem);
364 		cond_resched();
365 	}
366 	if (gcflag)
367 		ti->ti_flags |= NILFS_TI_GC;
368 
369 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
370 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
371 }
372 
373 static void nilfs_transaction_unlock(struct super_block *sb)
374 {
375 	struct nilfs_transaction_info *ti = current->journal_info;
376 	struct the_nilfs *nilfs = sb->s_fs_info;
377 
378 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
379 	BUG_ON(ti->ti_count > 0);
380 
381 	up_write(&nilfs->ns_segctor_sem);
382 	current->journal_info = ti->ti_save;
383 
384 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
385 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
386 }
387 
388 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
389 					    struct nilfs_segsum_pointer *ssp,
390 					    unsigned int bytes)
391 {
392 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
393 	unsigned int blocksize = sci->sc_super->s_blocksize;
394 	void *p;
395 
396 	if (unlikely(ssp->offset + bytes > blocksize)) {
397 		ssp->offset = 0;
398 		BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
399 					       &segbuf->sb_segsum_buffers));
400 		ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
401 	}
402 	p = ssp->bh->b_data + ssp->offset;
403 	ssp->offset += bytes;
404 	return p;
405 }
406 
407 /**
408  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
409  * @sci: nilfs_sc_info
410  */
411 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
412 {
413 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
414 	struct buffer_head *sumbh;
415 	unsigned int sumbytes;
416 	unsigned int flags = 0;
417 	int err;
418 
419 	if (nilfs_doing_gc())
420 		flags = NILFS_SS_GC;
421 	err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
422 	if (unlikely(err))
423 		return err;
424 
425 	sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
426 	sumbytes = segbuf->sb_sum.sumbytes;
427 	sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
428 	sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
429 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
430 	return 0;
431 }
432 
433 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
434 {
435 	sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
436 	if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
437 		return -E2BIG; /*
438 				* The current segment is filled up
439 				* (internal code)
440 				*/
441 	sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
442 	return nilfs_segctor_reset_segment_buffer(sci);
443 }
444 
445 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
446 {
447 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
448 	int err;
449 
450 	if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
451 		err = nilfs_segctor_feed_segment(sci);
452 		if (err)
453 			return err;
454 		segbuf = sci->sc_curseg;
455 	}
456 	err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
457 	if (likely(!err))
458 		segbuf->sb_sum.flags |= NILFS_SS_SR;
459 	return err;
460 }
461 
462 /*
463  * Functions for making segment summary and payloads
464  */
465 static int nilfs_segctor_segsum_block_required(
466 	struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
467 	unsigned int binfo_size)
468 {
469 	unsigned int blocksize = sci->sc_super->s_blocksize;
470 	/* Size of finfo and binfo is enough small against blocksize */
471 
472 	return ssp->offset + binfo_size +
473 		(!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
474 		blocksize;
475 }
476 
477 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
478 				      struct inode *inode)
479 {
480 	sci->sc_curseg->sb_sum.nfinfo++;
481 	sci->sc_binfo_ptr = sci->sc_finfo_ptr;
482 	nilfs_segctor_map_segsum_entry(
483 		sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
484 
485 	if (NILFS_I(inode)->i_root &&
486 	    !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
487 		set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
488 	/* skip finfo */
489 }
490 
491 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
492 				    struct inode *inode)
493 {
494 	struct nilfs_finfo *finfo;
495 	struct nilfs_inode_info *ii;
496 	struct nilfs_segment_buffer *segbuf;
497 	__u64 cno;
498 
499 	if (sci->sc_blk_cnt == 0)
500 		return;
501 
502 	ii = NILFS_I(inode);
503 
504 	if (test_bit(NILFS_I_GCINODE, &ii->i_state))
505 		cno = ii->i_cno;
506 	else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
507 		cno = 0;
508 	else
509 		cno = sci->sc_cno;
510 
511 	finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
512 						 sizeof(*finfo));
513 	finfo->fi_ino = cpu_to_le64(inode->i_ino);
514 	finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
515 	finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
516 	finfo->fi_cno = cpu_to_le64(cno);
517 
518 	segbuf = sci->sc_curseg;
519 	segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
520 		sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
521 	sci->sc_finfo_ptr = sci->sc_binfo_ptr;
522 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
523 }
524 
525 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
526 					struct buffer_head *bh,
527 					struct inode *inode,
528 					unsigned int binfo_size)
529 {
530 	struct nilfs_segment_buffer *segbuf;
531 	int required, err = 0;
532 
533  retry:
534 	segbuf = sci->sc_curseg;
535 	required = nilfs_segctor_segsum_block_required(
536 		sci, &sci->sc_binfo_ptr, binfo_size);
537 	if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
538 		nilfs_segctor_end_finfo(sci, inode);
539 		err = nilfs_segctor_feed_segment(sci);
540 		if (err)
541 			return err;
542 		goto retry;
543 	}
544 	if (unlikely(required)) {
545 		err = nilfs_segbuf_extend_segsum(segbuf);
546 		if (unlikely(err))
547 			goto failed;
548 	}
549 	if (sci->sc_blk_cnt == 0)
550 		nilfs_segctor_begin_finfo(sci, inode);
551 
552 	nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
553 	/* Substitution to vblocknr is delayed until update_blocknr() */
554 	nilfs_segbuf_add_file_buffer(segbuf, bh);
555 	sci->sc_blk_cnt++;
556  failed:
557 	return err;
558 }
559 
560 /*
561  * Callback functions that enumerate, mark, and collect dirty blocks
562  */
563 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
564 				   struct buffer_head *bh, struct inode *inode)
565 {
566 	int err;
567 
568 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
569 	if (err < 0)
570 		return err;
571 
572 	err = nilfs_segctor_add_file_block(sci, bh, inode,
573 					   sizeof(struct nilfs_binfo_v));
574 	if (!err)
575 		sci->sc_datablk_cnt++;
576 	return err;
577 }
578 
579 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
580 				   struct buffer_head *bh,
581 				   struct inode *inode)
582 {
583 	return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
584 }
585 
586 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
587 				   struct buffer_head *bh,
588 				   struct inode *inode)
589 {
590 	WARN_ON(!buffer_dirty(bh));
591 	return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
592 }
593 
594 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
595 					struct nilfs_segsum_pointer *ssp,
596 					union nilfs_binfo *binfo)
597 {
598 	struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
599 		sci, ssp, sizeof(*binfo_v));
600 	*binfo_v = binfo->bi_v;
601 }
602 
603 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
604 					struct nilfs_segsum_pointer *ssp,
605 					union nilfs_binfo *binfo)
606 {
607 	__le64 *vblocknr = nilfs_segctor_map_segsum_entry(
608 		sci, ssp, sizeof(*vblocknr));
609 	*vblocknr = binfo->bi_v.bi_vblocknr;
610 }
611 
612 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
613 	.collect_data = nilfs_collect_file_data,
614 	.collect_node = nilfs_collect_file_node,
615 	.collect_bmap = nilfs_collect_file_bmap,
616 	.write_data_binfo = nilfs_write_file_data_binfo,
617 	.write_node_binfo = nilfs_write_file_node_binfo,
618 };
619 
620 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
621 				  struct buffer_head *bh, struct inode *inode)
622 {
623 	int err;
624 
625 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
626 	if (err < 0)
627 		return err;
628 
629 	err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
630 	if (!err)
631 		sci->sc_datablk_cnt++;
632 	return err;
633 }
634 
635 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
636 				  struct buffer_head *bh, struct inode *inode)
637 {
638 	WARN_ON(!buffer_dirty(bh));
639 	return nilfs_segctor_add_file_block(sci, bh, inode,
640 					    sizeof(struct nilfs_binfo_dat));
641 }
642 
643 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
644 				       struct nilfs_segsum_pointer *ssp,
645 				       union nilfs_binfo *binfo)
646 {
647 	__le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
648 							  sizeof(*blkoff));
649 	*blkoff = binfo->bi_dat.bi_blkoff;
650 }
651 
652 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
653 				       struct nilfs_segsum_pointer *ssp,
654 				       union nilfs_binfo *binfo)
655 {
656 	struct nilfs_binfo_dat *binfo_dat =
657 		nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
658 	*binfo_dat = binfo->bi_dat;
659 }
660 
661 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
662 	.collect_data = nilfs_collect_dat_data,
663 	.collect_node = nilfs_collect_file_node,
664 	.collect_bmap = nilfs_collect_dat_bmap,
665 	.write_data_binfo = nilfs_write_dat_data_binfo,
666 	.write_node_binfo = nilfs_write_dat_node_binfo,
667 };
668 
669 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
670 	.collect_data = nilfs_collect_file_data,
671 	.collect_node = NULL,
672 	.collect_bmap = NULL,
673 	.write_data_binfo = nilfs_write_file_data_binfo,
674 	.write_node_binfo = NULL,
675 };
676 
677 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
678 					      struct list_head *listp,
679 					      size_t nlimit,
680 					      loff_t start, loff_t end)
681 {
682 	struct address_space *mapping = inode->i_mapping;
683 	struct pagevec pvec;
684 	pgoff_t index = 0, last = ULONG_MAX;
685 	size_t ndirties = 0;
686 	int i;
687 
688 	if (unlikely(start != 0 || end != LLONG_MAX)) {
689 		/*
690 		 * A valid range is given for sync-ing data pages. The
691 		 * range is rounded to per-page; extra dirty buffers
692 		 * may be included if blocksize < pagesize.
693 		 */
694 		index = start >> PAGE_SHIFT;
695 		last = end >> PAGE_SHIFT;
696 	}
697 	pagevec_init(&pvec);
698  repeat:
699 	if (unlikely(index > last) ||
700 	    !pagevec_lookup_range_tag(&pvec, mapping, &index, last,
701 				PAGECACHE_TAG_DIRTY))
702 		return ndirties;
703 
704 	for (i = 0; i < pagevec_count(&pvec); i++) {
705 		struct buffer_head *bh, *head;
706 		struct page *page = pvec.pages[i];
707 
708 		lock_page(page);
709 		if (!page_has_buffers(page))
710 			create_empty_buffers(page, i_blocksize(inode), 0);
711 		unlock_page(page);
712 
713 		bh = head = page_buffers(page);
714 		do {
715 			if (!buffer_dirty(bh) || buffer_async_write(bh))
716 				continue;
717 			get_bh(bh);
718 			list_add_tail(&bh->b_assoc_buffers, listp);
719 			ndirties++;
720 			if (unlikely(ndirties >= nlimit)) {
721 				pagevec_release(&pvec);
722 				cond_resched();
723 				return ndirties;
724 			}
725 		} while (bh = bh->b_this_page, bh != head);
726 	}
727 	pagevec_release(&pvec);
728 	cond_resched();
729 	goto repeat;
730 }
731 
732 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
733 					    struct list_head *listp)
734 {
735 	struct nilfs_inode_info *ii = NILFS_I(inode);
736 	struct inode *btnc_inode = ii->i_assoc_inode;
737 	struct pagevec pvec;
738 	struct buffer_head *bh, *head;
739 	unsigned int i;
740 	pgoff_t index = 0;
741 
742 	if (!btnc_inode)
743 		return;
744 
745 	pagevec_init(&pvec);
746 
747 	while (pagevec_lookup_tag(&pvec, btnc_inode->i_mapping, &index,
748 					PAGECACHE_TAG_DIRTY)) {
749 		for (i = 0; i < pagevec_count(&pvec); i++) {
750 			bh = head = page_buffers(pvec.pages[i]);
751 			do {
752 				if (buffer_dirty(bh) &&
753 						!buffer_async_write(bh)) {
754 					get_bh(bh);
755 					list_add_tail(&bh->b_assoc_buffers,
756 						      listp);
757 				}
758 				bh = bh->b_this_page;
759 			} while (bh != head);
760 		}
761 		pagevec_release(&pvec);
762 		cond_resched();
763 	}
764 }
765 
766 static void nilfs_dispose_list(struct the_nilfs *nilfs,
767 			       struct list_head *head, int force)
768 {
769 	struct nilfs_inode_info *ii, *n;
770 	struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
771 	unsigned int nv = 0;
772 
773 	while (!list_empty(head)) {
774 		spin_lock(&nilfs->ns_inode_lock);
775 		list_for_each_entry_safe(ii, n, head, i_dirty) {
776 			list_del_init(&ii->i_dirty);
777 			if (force) {
778 				if (unlikely(ii->i_bh)) {
779 					brelse(ii->i_bh);
780 					ii->i_bh = NULL;
781 				}
782 			} else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
783 				set_bit(NILFS_I_QUEUED, &ii->i_state);
784 				list_add_tail(&ii->i_dirty,
785 					      &nilfs->ns_dirty_files);
786 				continue;
787 			}
788 			ivec[nv++] = ii;
789 			if (nv == SC_N_INODEVEC)
790 				break;
791 		}
792 		spin_unlock(&nilfs->ns_inode_lock);
793 
794 		for (pii = ivec; nv > 0; pii++, nv--)
795 			iput(&(*pii)->vfs_inode);
796 	}
797 }
798 
799 static void nilfs_iput_work_func(struct work_struct *work)
800 {
801 	struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
802 						 sc_iput_work);
803 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
804 
805 	nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
806 }
807 
808 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
809 				     struct nilfs_root *root)
810 {
811 	int ret = 0;
812 
813 	if (nilfs_mdt_fetch_dirty(root->ifile))
814 		ret++;
815 	if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
816 		ret++;
817 	if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
818 		ret++;
819 	if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
820 		ret++;
821 	return ret;
822 }
823 
824 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
825 {
826 	return list_empty(&sci->sc_dirty_files) &&
827 		!test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
828 		sci->sc_nfreesegs == 0 &&
829 		(!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
830 }
831 
832 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
833 {
834 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
835 	int ret = 0;
836 
837 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
838 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
839 
840 	spin_lock(&nilfs->ns_inode_lock);
841 	if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
842 		ret++;
843 
844 	spin_unlock(&nilfs->ns_inode_lock);
845 	return ret;
846 }
847 
848 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
849 {
850 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
851 
852 	nilfs_mdt_clear_dirty(sci->sc_root->ifile);
853 	nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
854 	nilfs_mdt_clear_dirty(nilfs->ns_sufile);
855 	nilfs_mdt_clear_dirty(nilfs->ns_dat);
856 }
857 
858 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
859 {
860 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
861 	struct buffer_head *bh_cp;
862 	struct nilfs_checkpoint *raw_cp;
863 	int err;
864 
865 	/* XXX: this interface will be changed */
866 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
867 					  &raw_cp, &bh_cp);
868 	if (likely(!err)) {
869 		/*
870 		 * The following code is duplicated with cpfile.  But, it is
871 		 * needed to collect the checkpoint even if it was not newly
872 		 * created.
873 		 */
874 		mark_buffer_dirty(bh_cp);
875 		nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
876 		nilfs_cpfile_put_checkpoint(
877 			nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
878 	} else
879 		WARN_ON(err == -EINVAL || err == -ENOENT);
880 
881 	return err;
882 }
883 
884 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
885 {
886 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
887 	struct buffer_head *bh_cp;
888 	struct nilfs_checkpoint *raw_cp;
889 	int err;
890 
891 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
892 					  &raw_cp, &bh_cp);
893 	if (unlikely(err)) {
894 		WARN_ON(err == -EINVAL || err == -ENOENT);
895 		goto failed_ibh;
896 	}
897 	raw_cp->cp_snapshot_list.ssl_next = 0;
898 	raw_cp->cp_snapshot_list.ssl_prev = 0;
899 	raw_cp->cp_inodes_count =
900 		cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
901 	raw_cp->cp_blocks_count =
902 		cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
903 	raw_cp->cp_nblk_inc =
904 		cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
905 	raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
906 	raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
907 
908 	if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
909 		nilfs_checkpoint_clear_minor(raw_cp);
910 	else
911 		nilfs_checkpoint_set_minor(raw_cp);
912 
913 	nilfs_write_inode_common(sci->sc_root->ifile,
914 				 &raw_cp->cp_ifile_inode, 1);
915 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
916 	return 0;
917 
918  failed_ibh:
919 	return err;
920 }
921 
922 static void nilfs_fill_in_file_bmap(struct inode *ifile,
923 				    struct nilfs_inode_info *ii)
924 
925 {
926 	struct buffer_head *ibh;
927 	struct nilfs_inode *raw_inode;
928 
929 	if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
930 		ibh = ii->i_bh;
931 		BUG_ON(!ibh);
932 		raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
933 						  ibh);
934 		nilfs_bmap_write(ii->i_bmap, raw_inode);
935 		nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
936 	}
937 }
938 
939 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
940 {
941 	struct nilfs_inode_info *ii;
942 
943 	list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
944 		nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
945 		set_bit(NILFS_I_COLLECTED, &ii->i_state);
946 	}
947 }
948 
949 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
950 					     struct the_nilfs *nilfs)
951 {
952 	struct buffer_head *bh_sr;
953 	struct nilfs_super_root *raw_sr;
954 	unsigned int isz, srsz;
955 
956 	bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
957 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
958 	isz = nilfs->ns_inode_size;
959 	srsz = NILFS_SR_BYTES(isz);
960 
961 	raw_sr->sr_bytes = cpu_to_le16(srsz);
962 	raw_sr->sr_nongc_ctime
963 		= cpu_to_le64(nilfs_doing_gc() ?
964 			      nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
965 	raw_sr->sr_flags = 0;
966 
967 	nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
968 				 NILFS_SR_DAT_OFFSET(isz), 1);
969 	nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
970 				 NILFS_SR_CPFILE_OFFSET(isz), 1);
971 	nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
972 				 NILFS_SR_SUFILE_OFFSET(isz), 1);
973 	memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
974 }
975 
976 static void nilfs_redirty_inodes(struct list_head *head)
977 {
978 	struct nilfs_inode_info *ii;
979 
980 	list_for_each_entry(ii, head, i_dirty) {
981 		if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
982 			clear_bit(NILFS_I_COLLECTED, &ii->i_state);
983 	}
984 }
985 
986 static void nilfs_drop_collected_inodes(struct list_head *head)
987 {
988 	struct nilfs_inode_info *ii;
989 
990 	list_for_each_entry(ii, head, i_dirty) {
991 		if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
992 			continue;
993 
994 		clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
995 		set_bit(NILFS_I_UPDATED, &ii->i_state);
996 	}
997 }
998 
999 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1000 				       struct inode *inode,
1001 				       struct list_head *listp,
1002 				       int (*collect)(struct nilfs_sc_info *,
1003 						      struct buffer_head *,
1004 						      struct inode *))
1005 {
1006 	struct buffer_head *bh, *n;
1007 	int err = 0;
1008 
1009 	if (collect) {
1010 		list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1011 			list_del_init(&bh->b_assoc_buffers);
1012 			err = collect(sci, bh, inode);
1013 			brelse(bh);
1014 			if (unlikely(err))
1015 				goto dispose_buffers;
1016 		}
1017 		return 0;
1018 	}
1019 
1020  dispose_buffers:
1021 	while (!list_empty(listp)) {
1022 		bh = list_first_entry(listp, struct buffer_head,
1023 				      b_assoc_buffers);
1024 		list_del_init(&bh->b_assoc_buffers);
1025 		brelse(bh);
1026 	}
1027 	return err;
1028 }
1029 
1030 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1031 {
1032 	/* Remaining number of blocks within segment buffer */
1033 	return sci->sc_segbuf_nblocks -
1034 		(sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1035 }
1036 
1037 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1038 				   struct inode *inode,
1039 				   const struct nilfs_sc_operations *sc_ops)
1040 {
1041 	LIST_HEAD(data_buffers);
1042 	LIST_HEAD(node_buffers);
1043 	int err;
1044 
1045 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1046 		size_t n, rest = nilfs_segctor_buffer_rest(sci);
1047 
1048 		n = nilfs_lookup_dirty_data_buffers(
1049 			inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1050 		if (n > rest) {
1051 			err = nilfs_segctor_apply_buffers(
1052 				sci, inode, &data_buffers,
1053 				sc_ops->collect_data);
1054 			BUG_ON(!err); /* always receive -E2BIG or true error */
1055 			goto break_or_fail;
1056 		}
1057 	}
1058 	nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1059 
1060 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1061 		err = nilfs_segctor_apply_buffers(
1062 			sci, inode, &data_buffers, sc_ops->collect_data);
1063 		if (unlikely(err)) {
1064 			/* dispose node list */
1065 			nilfs_segctor_apply_buffers(
1066 				sci, inode, &node_buffers, NULL);
1067 			goto break_or_fail;
1068 		}
1069 		sci->sc_stage.flags |= NILFS_CF_NODE;
1070 	}
1071 	/* Collect node */
1072 	err = nilfs_segctor_apply_buffers(
1073 		sci, inode, &node_buffers, sc_ops->collect_node);
1074 	if (unlikely(err))
1075 		goto break_or_fail;
1076 
1077 	nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1078 	err = nilfs_segctor_apply_buffers(
1079 		sci, inode, &node_buffers, sc_ops->collect_bmap);
1080 	if (unlikely(err))
1081 		goto break_or_fail;
1082 
1083 	nilfs_segctor_end_finfo(sci, inode);
1084 	sci->sc_stage.flags &= ~NILFS_CF_NODE;
1085 
1086  break_or_fail:
1087 	return err;
1088 }
1089 
1090 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1091 					 struct inode *inode)
1092 {
1093 	LIST_HEAD(data_buffers);
1094 	size_t n, rest = nilfs_segctor_buffer_rest(sci);
1095 	int err;
1096 
1097 	n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1098 					    sci->sc_dsync_start,
1099 					    sci->sc_dsync_end);
1100 
1101 	err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1102 					  nilfs_collect_file_data);
1103 	if (!err) {
1104 		nilfs_segctor_end_finfo(sci, inode);
1105 		BUG_ON(n > rest);
1106 		/* always receive -E2BIG or true error if n > rest */
1107 	}
1108 	return err;
1109 }
1110 
1111 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1112 {
1113 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1114 	struct list_head *head;
1115 	struct nilfs_inode_info *ii;
1116 	size_t ndone;
1117 	int err = 0;
1118 
1119 	switch (nilfs_sc_cstage_get(sci)) {
1120 	case NILFS_ST_INIT:
1121 		/* Pre-processes */
1122 		sci->sc_stage.flags = 0;
1123 
1124 		if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1125 			sci->sc_nblk_inc = 0;
1126 			sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1127 			if (mode == SC_LSEG_DSYNC) {
1128 				nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1129 				goto dsync_mode;
1130 			}
1131 		}
1132 
1133 		sci->sc_stage.dirty_file_ptr = NULL;
1134 		sci->sc_stage.gc_inode_ptr = NULL;
1135 		if (mode == SC_FLUSH_DAT) {
1136 			nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1137 			goto dat_stage;
1138 		}
1139 		nilfs_sc_cstage_inc(sci);
1140 		fallthrough;
1141 	case NILFS_ST_GC:
1142 		if (nilfs_doing_gc()) {
1143 			head = &sci->sc_gc_inodes;
1144 			ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1145 						head, i_dirty);
1146 			list_for_each_entry_continue(ii, head, i_dirty) {
1147 				err = nilfs_segctor_scan_file(
1148 					sci, &ii->vfs_inode,
1149 					&nilfs_sc_file_ops);
1150 				if (unlikely(err)) {
1151 					sci->sc_stage.gc_inode_ptr = list_entry(
1152 						ii->i_dirty.prev,
1153 						struct nilfs_inode_info,
1154 						i_dirty);
1155 					goto break_or_fail;
1156 				}
1157 				set_bit(NILFS_I_COLLECTED, &ii->i_state);
1158 			}
1159 			sci->sc_stage.gc_inode_ptr = NULL;
1160 		}
1161 		nilfs_sc_cstage_inc(sci);
1162 		fallthrough;
1163 	case NILFS_ST_FILE:
1164 		head = &sci->sc_dirty_files;
1165 		ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1166 					i_dirty);
1167 		list_for_each_entry_continue(ii, head, i_dirty) {
1168 			clear_bit(NILFS_I_DIRTY, &ii->i_state);
1169 
1170 			err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1171 						      &nilfs_sc_file_ops);
1172 			if (unlikely(err)) {
1173 				sci->sc_stage.dirty_file_ptr =
1174 					list_entry(ii->i_dirty.prev,
1175 						   struct nilfs_inode_info,
1176 						   i_dirty);
1177 				goto break_or_fail;
1178 			}
1179 			/* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1180 			/* XXX: required ? */
1181 		}
1182 		sci->sc_stage.dirty_file_ptr = NULL;
1183 		if (mode == SC_FLUSH_FILE) {
1184 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1185 			return 0;
1186 		}
1187 		nilfs_sc_cstage_inc(sci);
1188 		sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1189 		fallthrough;
1190 	case NILFS_ST_IFILE:
1191 		err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1192 					      &nilfs_sc_file_ops);
1193 		if (unlikely(err))
1194 			break;
1195 		nilfs_sc_cstage_inc(sci);
1196 		/* Creating a checkpoint */
1197 		err = nilfs_segctor_create_checkpoint(sci);
1198 		if (unlikely(err))
1199 			break;
1200 		fallthrough;
1201 	case NILFS_ST_CPFILE:
1202 		err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1203 					      &nilfs_sc_file_ops);
1204 		if (unlikely(err))
1205 			break;
1206 		nilfs_sc_cstage_inc(sci);
1207 		fallthrough;
1208 	case NILFS_ST_SUFILE:
1209 		err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1210 					 sci->sc_nfreesegs, &ndone);
1211 		if (unlikely(err)) {
1212 			nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1213 						  sci->sc_freesegs, ndone,
1214 						  NULL);
1215 			break;
1216 		}
1217 		sci->sc_stage.flags |= NILFS_CF_SUFREED;
1218 
1219 		err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1220 					      &nilfs_sc_file_ops);
1221 		if (unlikely(err))
1222 			break;
1223 		nilfs_sc_cstage_inc(sci);
1224 		fallthrough;
1225 	case NILFS_ST_DAT:
1226  dat_stage:
1227 		err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1228 					      &nilfs_sc_dat_ops);
1229 		if (unlikely(err))
1230 			break;
1231 		if (mode == SC_FLUSH_DAT) {
1232 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1233 			return 0;
1234 		}
1235 		nilfs_sc_cstage_inc(sci);
1236 		fallthrough;
1237 	case NILFS_ST_SR:
1238 		if (mode == SC_LSEG_SR) {
1239 			/* Appending a super root */
1240 			err = nilfs_segctor_add_super_root(sci);
1241 			if (unlikely(err))
1242 				break;
1243 		}
1244 		/* End of a logical segment */
1245 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1246 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1247 		return 0;
1248 	case NILFS_ST_DSYNC:
1249  dsync_mode:
1250 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1251 		ii = sci->sc_dsync_inode;
1252 		if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1253 			break;
1254 
1255 		err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1256 		if (unlikely(err))
1257 			break;
1258 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1259 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1260 		return 0;
1261 	case NILFS_ST_DONE:
1262 		return 0;
1263 	default:
1264 		BUG();
1265 	}
1266 
1267  break_or_fail:
1268 	return err;
1269 }
1270 
1271 /**
1272  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1273  * @sci: nilfs_sc_info
1274  * @nilfs: nilfs object
1275  */
1276 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1277 					    struct the_nilfs *nilfs)
1278 {
1279 	struct nilfs_segment_buffer *segbuf, *prev;
1280 	__u64 nextnum;
1281 	int err, alloc = 0;
1282 
1283 	segbuf = nilfs_segbuf_new(sci->sc_super);
1284 	if (unlikely(!segbuf))
1285 		return -ENOMEM;
1286 
1287 	if (list_empty(&sci->sc_write_logs)) {
1288 		nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1289 				 nilfs->ns_pseg_offset, nilfs);
1290 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1291 			nilfs_shift_to_next_segment(nilfs);
1292 			nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1293 		}
1294 
1295 		segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1296 		nextnum = nilfs->ns_nextnum;
1297 
1298 		if (nilfs->ns_segnum == nilfs->ns_nextnum)
1299 			/* Start from the head of a new full segment */
1300 			alloc++;
1301 	} else {
1302 		/* Continue logs */
1303 		prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1304 		nilfs_segbuf_map_cont(segbuf, prev);
1305 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1306 		nextnum = prev->sb_nextnum;
1307 
1308 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1309 			nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1310 			segbuf->sb_sum.seg_seq++;
1311 			alloc++;
1312 		}
1313 	}
1314 
1315 	err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1316 	if (err)
1317 		goto failed;
1318 
1319 	if (alloc) {
1320 		err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1321 		if (err)
1322 			goto failed;
1323 	}
1324 	nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1325 
1326 	BUG_ON(!list_empty(&sci->sc_segbufs));
1327 	list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1328 	sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1329 	return 0;
1330 
1331  failed:
1332 	nilfs_segbuf_free(segbuf);
1333 	return err;
1334 }
1335 
1336 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1337 					 struct the_nilfs *nilfs, int nadd)
1338 {
1339 	struct nilfs_segment_buffer *segbuf, *prev;
1340 	struct inode *sufile = nilfs->ns_sufile;
1341 	__u64 nextnextnum;
1342 	LIST_HEAD(list);
1343 	int err, ret, i;
1344 
1345 	prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1346 	/*
1347 	 * Since the segment specified with nextnum might be allocated during
1348 	 * the previous construction, the buffer including its segusage may
1349 	 * not be dirty.  The following call ensures that the buffer is dirty
1350 	 * and will pin the buffer on memory until the sufile is written.
1351 	 */
1352 	err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1353 	if (unlikely(err))
1354 		return err;
1355 
1356 	for (i = 0; i < nadd; i++) {
1357 		/* extend segment info */
1358 		err = -ENOMEM;
1359 		segbuf = nilfs_segbuf_new(sci->sc_super);
1360 		if (unlikely(!segbuf))
1361 			goto failed;
1362 
1363 		/* map this buffer to region of segment on-disk */
1364 		nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1365 		sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1366 
1367 		/* allocate the next next full segment */
1368 		err = nilfs_sufile_alloc(sufile, &nextnextnum);
1369 		if (unlikely(err))
1370 			goto failed_segbuf;
1371 
1372 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1373 		nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1374 
1375 		list_add_tail(&segbuf->sb_list, &list);
1376 		prev = segbuf;
1377 	}
1378 	list_splice_tail(&list, &sci->sc_segbufs);
1379 	return 0;
1380 
1381  failed_segbuf:
1382 	nilfs_segbuf_free(segbuf);
1383  failed:
1384 	list_for_each_entry(segbuf, &list, sb_list) {
1385 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1386 		WARN_ON(ret); /* never fails */
1387 	}
1388 	nilfs_destroy_logs(&list);
1389 	return err;
1390 }
1391 
1392 static void nilfs_free_incomplete_logs(struct list_head *logs,
1393 				       struct the_nilfs *nilfs)
1394 {
1395 	struct nilfs_segment_buffer *segbuf, *prev;
1396 	struct inode *sufile = nilfs->ns_sufile;
1397 	int ret;
1398 
1399 	segbuf = NILFS_FIRST_SEGBUF(logs);
1400 	if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1401 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1402 		WARN_ON(ret); /* never fails */
1403 	}
1404 	if (atomic_read(&segbuf->sb_err)) {
1405 		/* Case 1: The first segment failed */
1406 		if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1407 			/*
1408 			 * Case 1a:  Partial segment appended into an existing
1409 			 * segment
1410 			 */
1411 			nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1412 						segbuf->sb_fseg_end);
1413 		else /* Case 1b:  New full segment */
1414 			set_nilfs_discontinued(nilfs);
1415 	}
1416 
1417 	prev = segbuf;
1418 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1419 		if (prev->sb_nextnum != segbuf->sb_nextnum) {
1420 			ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1421 			WARN_ON(ret); /* never fails */
1422 		}
1423 		if (atomic_read(&segbuf->sb_err) &&
1424 		    segbuf->sb_segnum != nilfs->ns_nextnum)
1425 			/* Case 2: extended segment (!= next) failed */
1426 			nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1427 		prev = segbuf;
1428 	}
1429 }
1430 
1431 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1432 					  struct inode *sufile)
1433 {
1434 	struct nilfs_segment_buffer *segbuf;
1435 	unsigned long live_blocks;
1436 	int ret;
1437 
1438 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1439 		live_blocks = segbuf->sb_sum.nblocks +
1440 			(segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1441 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1442 						     live_blocks,
1443 						     sci->sc_seg_ctime);
1444 		WARN_ON(ret); /* always succeed because the segusage is dirty */
1445 	}
1446 }
1447 
1448 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1449 {
1450 	struct nilfs_segment_buffer *segbuf;
1451 	int ret;
1452 
1453 	segbuf = NILFS_FIRST_SEGBUF(logs);
1454 	ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1455 					     segbuf->sb_pseg_start -
1456 					     segbuf->sb_fseg_start, 0);
1457 	WARN_ON(ret); /* always succeed because the segusage is dirty */
1458 
1459 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1460 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1461 						     0, 0);
1462 		WARN_ON(ret); /* always succeed */
1463 	}
1464 }
1465 
1466 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1467 					    struct nilfs_segment_buffer *last,
1468 					    struct inode *sufile)
1469 {
1470 	struct nilfs_segment_buffer *segbuf = last;
1471 	int ret;
1472 
1473 	list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1474 		sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1475 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1476 		WARN_ON(ret);
1477 	}
1478 	nilfs_truncate_logs(&sci->sc_segbufs, last);
1479 }
1480 
1481 
1482 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1483 				 struct the_nilfs *nilfs, int mode)
1484 {
1485 	struct nilfs_cstage prev_stage = sci->sc_stage;
1486 	int err, nadd = 1;
1487 
1488 	/* Collection retry loop */
1489 	for (;;) {
1490 		sci->sc_nblk_this_inc = 0;
1491 		sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1492 
1493 		err = nilfs_segctor_reset_segment_buffer(sci);
1494 		if (unlikely(err))
1495 			goto failed;
1496 
1497 		err = nilfs_segctor_collect_blocks(sci, mode);
1498 		sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1499 		if (!err)
1500 			break;
1501 
1502 		if (unlikely(err != -E2BIG))
1503 			goto failed;
1504 
1505 		/* The current segment is filled up */
1506 		if (mode != SC_LSEG_SR ||
1507 		    nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1508 			break;
1509 
1510 		nilfs_clear_logs(&sci->sc_segbufs);
1511 
1512 		if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1513 			err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1514 							sci->sc_freesegs,
1515 							sci->sc_nfreesegs,
1516 							NULL);
1517 			WARN_ON(err); /* do not happen */
1518 			sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1519 		}
1520 
1521 		err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1522 		if (unlikely(err))
1523 			return err;
1524 
1525 		nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1526 		sci->sc_stage = prev_stage;
1527 	}
1528 	nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1529 	return 0;
1530 
1531  failed:
1532 	return err;
1533 }
1534 
1535 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1536 				      struct buffer_head *new_bh)
1537 {
1538 	BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1539 
1540 	list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1541 	/* The caller must release old_bh */
1542 }
1543 
1544 static int
1545 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1546 				     struct nilfs_segment_buffer *segbuf,
1547 				     int mode)
1548 {
1549 	struct inode *inode = NULL;
1550 	sector_t blocknr;
1551 	unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1552 	unsigned long nblocks = 0, ndatablk = 0;
1553 	const struct nilfs_sc_operations *sc_op = NULL;
1554 	struct nilfs_segsum_pointer ssp;
1555 	struct nilfs_finfo *finfo = NULL;
1556 	union nilfs_binfo binfo;
1557 	struct buffer_head *bh, *bh_org;
1558 	ino_t ino = 0;
1559 	int err = 0;
1560 
1561 	if (!nfinfo)
1562 		goto out;
1563 
1564 	blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1565 	ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1566 	ssp.offset = sizeof(struct nilfs_segment_summary);
1567 
1568 	list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1569 		if (bh == segbuf->sb_super_root)
1570 			break;
1571 		if (!finfo) {
1572 			finfo =	nilfs_segctor_map_segsum_entry(
1573 				sci, &ssp, sizeof(*finfo));
1574 			ino = le64_to_cpu(finfo->fi_ino);
1575 			nblocks = le32_to_cpu(finfo->fi_nblocks);
1576 			ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1577 
1578 			inode = bh->b_page->mapping->host;
1579 
1580 			if (mode == SC_LSEG_DSYNC)
1581 				sc_op = &nilfs_sc_dsync_ops;
1582 			else if (ino == NILFS_DAT_INO)
1583 				sc_op = &nilfs_sc_dat_ops;
1584 			else /* file blocks */
1585 				sc_op = &nilfs_sc_file_ops;
1586 		}
1587 		bh_org = bh;
1588 		get_bh(bh_org);
1589 		err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1590 					&binfo);
1591 		if (bh != bh_org)
1592 			nilfs_list_replace_buffer(bh_org, bh);
1593 		brelse(bh_org);
1594 		if (unlikely(err))
1595 			goto failed_bmap;
1596 
1597 		if (ndatablk > 0)
1598 			sc_op->write_data_binfo(sci, &ssp, &binfo);
1599 		else
1600 			sc_op->write_node_binfo(sci, &ssp, &binfo);
1601 
1602 		blocknr++;
1603 		if (--nblocks == 0) {
1604 			finfo = NULL;
1605 			if (--nfinfo == 0)
1606 				break;
1607 		} else if (ndatablk > 0)
1608 			ndatablk--;
1609 	}
1610  out:
1611 	return 0;
1612 
1613  failed_bmap:
1614 	return err;
1615 }
1616 
1617 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1618 {
1619 	struct nilfs_segment_buffer *segbuf;
1620 	int err;
1621 
1622 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1623 		err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1624 		if (unlikely(err))
1625 			return err;
1626 		nilfs_segbuf_fill_in_segsum(segbuf);
1627 	}
1628 	return 0;
1629 }
1630 
1631 static void nilfs_begin_page_io(struct page *page)
1632 {
1633 	if (!page || PageWriteback(page))
1634 		/*
1635 		 * For split b-tree node pages, this function may be called
1636 		 * twice.  We ignore the 2nd or later calls by this check.
1637 		 */
1638 		return;
1639 
1640 	lock_page(page);
1641 	clear_page_dirty_for_io(page);
1642 	set_page_writeback(page);
1643 	unlock_page(page);
1644 }
1645 
1646 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1647 {
1648 	struct nilfs_segment_buffer *segbuf;
1649 	struct page *bd_page = NULL, *fs_page = NULL;
1650 
1651 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1652 		struct buffer_head *bh;
1653 
1654 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1655 				    b_assoc_buffers) {
1656 			if (bh->b_page != bd_page) {
1657 				if (bd_page) {
1658 					lock_page(bd_page);
1659 					clear_page_dirty_for_io(bd_page);
1660 					set_page_writeback(bd_page);
1661 					unlock_page(bd_page);
1662 				}
1663 				bd_page = bh->b_page;
1664 			}
1665 		}
1666 
1667 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1668 				    b_assoc_buffers) {
1669 			set_buffer_async_write(bh);
1670 			if (bh == segbuf->sb_super_root) {
1671 				if (bh->b_page != bd_page) {
1672 					lock_page(bd_page);
1673 					clear_page_dirty_for_io(bd_page);
1674 					set_page_writeback(bd_page);
1675 					unlock_page(bd_page);
1676 					bd_page = bh->b_page;
1677 				}
1678 				break;
1679 			}
1680 			if (bh->b_page != fs_page) {
1681 				nilfs_begin_page_io(fs_page);
1682 				fs_page = bh->b_page;
1683 			}
1684 		}
1685 	}
1686 	if (bd_page) {
1687 		lock_page(bd_page);
1688 		clear_page_dirty_for_io(bd_page);
1689 		set_page_writeback(bd_page);
1690 		unlock_page(bd_page);
1691 	}
1692 	nilfs_begin_page_io(fs_page);
1693 }
1694 
1695 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1696 			       struct the_nilfs *nilfs)
1697 {
1698 	int ret;
1699 
1700 	ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1701 	list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1702 	return ret;
1703 }
1704 
1705 static void nilfs_end_page_io(struct page *page, int err)
1706 {
1707 	if (!page)
1708 		return;
1709 
1710 	if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1711 		/*
1712 		 * For b-tree node pages, this function may be called twice
1713 		 * or more because they might be split in a segment.
1714 		 */
1715 		if (PageDirty(page)) {
1716 			/*
1717 			 * For pages holding split b-tree node buffers, dirty
1718 			 * flag on the buffers may be cleared discretely.
1719 			 * In that case, the page is once redirtied for
1720 			 * remaining buffers, and it must be cancelled if
1721 			 * all the buffers get cleaned later.
1722 			 */
1723 			lock_page(page);
1724 			if (nilfs_page_buffers_clean(page))
1725 				__nilfs_clear_page_dirty(page);
1726 			unlock_page(page);
1727 		}
1728 		return;
1729 	}
1730 
1731 	if (!err) {
1732 		if (!nilfs_page_buffers_clean(page))
1733 			__set_page_dirty_nobuffers(page);
1734 		ClearPageError(page);
1735 	} else {
1736 		__set_page_dirty_nobuffers(page);
1737 		SetPageError(page);
1738 	}
1739 
1740 	end_page_writeback(page);
1741 }
1742 
1743 static void nilfs_abort_logs(struct list_head *logs, int err)
1744 {
1745 	struct nilfs_segment_buffer *segbuf;
1746 	struct page *bd_page = NULL, *fs_page = NULL;
1747 	struct buffer_head *bh;
1748 
1749 	if (list_empty(logs))
1750 		return;
1751 
1752 	list_for_each_entry(segbuf, logs, sb_list) {
1753 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1754 				    b_assoc_buffers) {
1755 			if (bh->b_page != bd_page) {
1756 				if (bd_page)
1757 					end_page_writeback(bd_page);
1758 				bd_page = bh->b_page;
1759 			}
1760 		}
1761 
1762 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1763 				    b_assoc_buffers) {
1764 			clear_buffer_async_write(bh);
1765 			if (bh == segbuf->sb_super_root) {
1766 				if (bh->b_page != bd_page) {
1767 					end_page_writeback(bd_page);
1768 					bd_page = bh->b_page;
1769 				}
1770 				break;
1771 			}
1772 			if (bh->b_page != fs_page) {
1773 				nilfs_end_page_io(fs_page, err);
1774 				fs_page = bh->b_page;
1775 			}
1776 		}
1777 	}
1778 	if (bd_page)
1779 		end_page_writeback(bd_page);
1780 
1781 	nilfs_end_page_io(fs_page, err);
1782 }
1783 
1784 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1785 					     struct the_nilfs *nilfs, int err)
1786 {
1787 	LIST_HEAD(logs);
1788 	int ret;
1789 
1790 	list_splice_tail_init(&sci->sc_write_logs, &logs);
1791 	ret = nilfs_wait_on_logs(&logs);
1792 	nilfs_abort_logs(&logs, ret ? : err);
1793 
1794 	list_splice_tail_init(&sci->sc_segbufs, &logs);
1795 	nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1796 	nilfs_free_incomplete_logs(&logs, nilfs);
1797 
1798 	if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1799 		ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1800 						sci->sc_freesegs,
1801 						sci->sc_nfreesegs,
1802 						NULL);
1803 		WARN_ON(ret); /* do not happen */
1804 	}
1805 
1806 	nilfs_destroy_logs(&logs);
1807 }
1808 
1809 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1810 				   struct nilfs_segment_buffer *segbuf)
1811 {
1812 	nilfs->ns_segnum = segbuf->sb_segnum;
1813 	nilfs->ns_nextnum = segbuf->sb_nextnum;
1814 	nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1815 		+ segbuf->sb_sum.nblocks;
1816 	nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1817 	nilfs->ns_ctime = segbuf->sb_sum.ctime;
1818 }
1819 
1820 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1821 {
1822 	struct nilfs_segment_buffer *segbuf;
1823 	struct page *bd_page = NULL, *fs_page = NULL;
1824 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1825 	int update_sr = false;
1826 
1827 	list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1828 		struct buffer_head *bh;
1829 
1830 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1831 				    b_assoc_buffers) {
1832 			set_buffer_uptodate(bh);
1833 			clear_buffer_dirty(bh);
1834 			if (bh->b_page != bd_page) {
1835 				if (bd_page)
1836 					end_page_writeback(bd_page);
1837 				bd_page = bh->b_page;
1838 			}
1839 		}
1840 		/*
1841 		 * We assume that the buffers which belong to the same page
1842 		 * continue over the buffer list.
1843 		 * Under this assumption, the last BHs of pages is
1844 		 * identifiable by the discontinuity of bh->b_page
1845 		 * (page != fs_page).
1846 		 *
1847 		 * For B-tree node blocks, however, this assumption is not
1848 		 * guaranteed.  The cleanup code of B-tree node pages needs
1849 		 * special care.
1850 		 */
1851 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1852 				    b_assoc_buffers) {
1853 			const unsigned long set_bits = BIT(BH_Uptodate);
1854 			const unsigned long clear_bits =
1855 				(BIT(BH_Dirty) | BIT(BH_Async_Write) |
1856 				 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1857 				 BIT(BH_NILFS_Redirected));
1858 
1859 			set_mask_bits(&bh->b_state, clear_bits, set_bits);
1860 			if (bh == segbuf->sb_super_root) {
1861 				if (bh->b_page != bd_page) {
1862 					end_page_writeback(bd_page);
1863 					bd_page = bh->b_page;
1864 				}
1865 				update_sr = true;
1866 				break;
1867 			}
1868 			if (bh->b_page != fs_page) {
1869 				nilfs_end_page_io(fs_page, 0);
1870 				fs_page = bh->b_page;
1871 			}
1872 		}
1873 
1874 		if (!nilfs_segbuf_simplex(segbuf)) {
1875 			if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1876 				set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1877 				sci->sc_lseg_stime = jiffies;
1878 			}
1879 			if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1880 				clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1881 		}
1882 	}
1883 	/*
1884 	 * Since pages may continue over multiple segment buffers,
1885 	 * end of the last page must be checked outside of the loop.
1886 	 */
1887 	if (bd_page)
1888 		end_page_writeback(bd_page);
1889 
1890 	nilfs_end_page_io(fs_page, 0);
1891 
1892 	nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1893 
1894 	if (nilfs_doing_gc())
1895 		nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1896 	else
1897 		nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1898 
1899 	sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1900 
1901 	segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1902 	nilfs_set_next_segment(nilfs, segbuf);
1903 
1904 	if (update_sr) {
1905 		nilfs->ns_flushed_device = 0;
1906 		nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1907 				       segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1908 
1909 		clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1910 		clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1911 		set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1912 		nilfs_segctor_clear_metadata_dirty(sci);
1913 	} else
1914 		clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1915 }
1916 
1917 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1918 {
1919 	int ret;
1920 
1921 	ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1922 	if (!ret) {
1923 		nilfs_segctor_complete_write(sci);
1924 		nilfs_destroy_logs(&sci->sc_write_logs);
1925 	}
1926 	return ret;
1927 }
1928 
1929 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1930 					     struct the_nilfs *nilfs)
1931 {
1932 	struct nilfs_inode_info *ii, *n;
1933 	struct inode *ifile = sci->sc_root->ifile;
1934 
1935 	spin_lock(&nilfs->ns_inode_lock);
1936  retry:
1937 	list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1938 		if (!ii->i_bh) {
1939 			struct buffer_head *ibh;
1940 			int err;
1941 
1942 			spin_unlock(&nilfs->ns_inode_lock);
1943 			err = nilfs_ifile_get_inode_block(
1944 				ifile, ii->vfs_inode.i_ino, &ibh);
1945 			if (unlikely(err)) {
1946 				nilfs_warn(sci->sc_super,
1947 					   "log writer: error %d getting inode block (ino=%lu)",
1948 					   err, ii->vfs_inode.i_ino);
1949 				return err;
1950 			}
1951 			spin_lock(&nilfs->ns_inode_lock);
1952 			if (likely(!ii->i_bh))
1953 				ii->i_bh = ibh;
1954 			else
1955 				brelse(ibh);
1956 			goto retry;
1957 		}
1958 
1959 		// Always redirty the buffer to avoid race condition
1960 		mark_buffer_dirty(ii->i_bh);
1961 		nilfs_mdt_mark_dirty(ifile);
1962 
1963 		clear_bit(NILFS_I_QUEUED, &ii->i_state);
1964 		set_bit(NILFS_I_BUSY, &ii->i_state);
1965 		list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1966 	}
1967 	spin_unlock(&nilfs->ns_inode_lock);
1968 
1969 	return 0;
1970 }
1971 
1972 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1973 					     struct the_nilfs *nilfs)
1974 {
1975 	struct nilfs_inode_info *ii, *n;
1976 	int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1977 	int defer_iput = false;
1978 
1979 	spin_lock(&nilfs->ns_inode_lock);
1980 	list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1981 		if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1982 		    test_bit(NILFS_I_DIRTY, &ii->i_state))
1983 			continue;
1984 
1985 		clear_bit(NILFS_I_BUSY, &ii->i_state);
1986 		brelse(ii->i_bh);
1987 		ii->i_bh = NULL;
1988 		list_del_init(&ii->i_dirty);
1989 		if (!ii->vfs_inode.i_nlink || during_mount) {
1990 			/*
1991 			 * Defer calling iput() to avoid deadlocks if
1992 			 * i_nlink == 0 or mount is not yet finished.
1993 			 */
1994 			list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
1995 			defer_iput = true;
1996 		} else {
1997 			spin_unlock(&nilfs->ns_inode_lock);
1998 			iput(&ii->vfs_inode);
1999 			spin_lock(&nilfs->ns_inode_lock);
2000 		}
2001 	}
2002 	spin_unlock(&nilfs->ns_inode_lock);
2003 
2004 	if (defer_iput)
2005 		schedule_work(&sci->sc_iput_work);
2006 }
2007 
2008 /*
2009  * Main procedure of segment constructor
2010  */
2011 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2012 {
2013 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2014 	int err;
2015 
2016 	nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2017 	sci->sc_cno = nilfs->ns_cno;
2018 
2019 	err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2020 	if (unlikely(err))
2021 		goto out;
2022 
2023 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2024 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2025 
2026 	if (nilfs_segctor_clean(sci))
2027 		goto out;
2028 
2029 	do {
2030 		sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2031 
2032 		err = nilfs_segctor_begin_construction(sci, nilfs);
2033 		if (unlikely(err))
2034 			goto out;
2035 
2036 		/* Update time stamp */
2037 		sci->sc_seg_ctime = ktime_get_real_seconds();
2038 
2039 		err = nilfs_segctor_collect(sci, nilfs, mode);
2040 		if (unlikely(err))
2041 			goto failed;
2042 
2043 		/* Avoid empty segment */
2044 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2045 		    nilfs_segbuf_empty(sci->sc_curseg)) {
2046 			nilfs_segctor_abort_construction(sci, nilfs, 1);
2047 			goto out;
2048 		}
2049 
2050 		err = nilfs_segctor_assign(sci, mode);
2051 		if (unlikely(err))
2052 			goto failed;
2053 
2054 		if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2055 			nilfs_segctor_fill_in_file_bmap(sci);
2056 
2057 		if (mode == SC_LSEG_SR &&
2058 		    nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2059 			err = nilfs_segctor_fill_in_checkpoint(sci);
2060 			if (unlikely(err))
2061 				goto failed_to_write;
2062 
2063 			nilfs_segctor_fill_in_super_root(sci, nilfs);
2064 		}
2065 		nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2066 
2067 		/* Write partial segments */
2068 		nilfs_segctor_prepare_write(sci);
2069 
2070 		nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2071 					    nilfs->ns_crc_seed);
2072 
2073 		err = nilfs_segctor_write(sci, nilfs);
2074 		if (unlikely(err))
2075 			goto failed_to_write;
2076 
2077 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2078 		    nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2079 			/*
2080 			 * At this point, we avoid double buffering
2081 			 * for blocksize < pagesize because page dirty
2082 			 * flag is turned off during write and dirty
2083 			 * buffers are not properly collected for
2084 			 * pages crossing over segments.
2085 			 */
2086 			err = nilfs_segctor_wait(sci);
2087 			if (err)
2088 				goto failed_to_write;
2089 		}
2090 	} while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2091 
2092  out:
2093 	nilfs_segctor_drop_written_files(sci, nilfs);
2094 	return err;
2095 
2096  failed_to_write:
2097 	if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2098 		nilfs_redirty_inodes(&sci->sc_dirty_files);
2099 
2100  failed:
2101 	if (nilfs_doing_gc())
2102 		nilfs_redirty_inodes(&sci->sc_gc_inodes);
2103 	nilfs_segctor_abort_construction(sci, nilfs, err);
2104 	goto out;
2105 }
2106 
2107 /**
2108  * nilfs_segctor_start_timer - set timer of background write
2109  * @sci: nilfs_sc_info
2110  *
2111  * If the timer has already been set, it ignores the new request.
2112  * This function MUST be called within a section locking the segment
2113  * semaphore.
2114  */
2115 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2116 {
2117 	spin_lock(&sci->sc_state_lock);
2118 	if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2119 		sci->sc_timer.expires = jiffies + sci->sc_interval;
2120 		add_timer(&sci->sc_timer);
2121 		sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2122 	}
2123 	spin_unlock(&sci->sc_state_lock);
2124 }
2125 
2126 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2127 {
2128 	spin_lock(&sci->sc_state_lock);
2129 	if (!(sci->sc_flush_request & BIT(bn))) {
2130 		unsigned long prev_req = sci->sc_flush_request;
2131 
2132 		sci->sc_flush_request |= BIT(bn);
2133 		if (!prev_req)
2134 			wake_up(&sci->sc_wait_daemon);
2135 	}
2136 	spin_unlock(&sci->sc_state_lock);
2137 }
2138 
2139 /**
2140  * nilfs_flush_segment - trigger a segment construction for resource control
2141  * @sb: super block
2142  * @ino: inode number of the file to be flushed out.
2143  */
2144 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2145 {
2146 	struct the_nilfs *nilfs = sb->s_fs_info;
2147 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2148 
2149 	if (!sci || nilfs_doing_construction())
2150 		return;
2151 	nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2152 					/* assign bit 0 to data files */
2153 }
2154 
2155 struct nilfs_segctor_wait_request {
2156 	wait_queue_entry_t	wq;
2157 	__u32		seq;
2158 	int		err;
2159 	atomic_t	done;
2160 };
2161 
2162 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2163 {
2164 	struct nilfs_segctor_wait_request wait_req;
2165 	int err = 0;
2166 
2167 	spin_lock(&sci->sc_state_lock);
2168 	init_wait(&wait_req.wq);
2169 	wait_req.err = 0;
2170 	atomic_set(&wait_req.done, 0);
2171 	wait_req.seq = ++sci->sc_seq_request;
2172 	spin_unlock(&sci->sc_state_lock);
2173 
2174 	init_waitqueue_entry(&wait_req.wq, current);
2175 	add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2176 	set_current_state(TASK_INTERRUPTIBLE);
2177 	wake_up(&sci->sc_wait_daemon);
2178 
2179 	for (;;) {
2180 		if (atomic_read(&wait_req.done)) {
2181 			err = wait_req.err;
2182 			break;
2183 		}
2184 		if (!signal_pending(current)) {
2185 			schedule();
2186 			continue;
2187 		}
2188 		err = -ERESTARTSYS;
2189 		break;
2190 	}
2191 	finish_wait(&sci->sc_wait_request, &wait_req.wq);
2192 	return err;
2193 }
2194 
2195 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2196 {
2197 	struct nilfs_segctor_wait_request *wrq, *n;
2198 	unsigned long flags;
2199 
2200 	spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2201 	list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2202 		if (!atomic_read(&wrq->done) &&
2203 		    nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2204 			wrq->err = err;
2205 			atomic_set(&wrq->done, 1);
2206 		}
2207 		if (atomic_read(&wrq->done)) {
2208 			wrq->wq.func(&wrq->wq,
2209 				     TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2210 				     0, NULL);
2211 		}
2212 	}
2213 	spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2214 }
2215 
2216 /**
2217  * nilfs_construct_segment - construct a logical segment
2218  * @sb: super block
2219  *
2220  * Return Value: On success, 0 is returned. On errors, one of the following
2221  * negative error code is returned.
2222  *
2223  * %-EROFS - Read only filesystem.
2224  *
2225  * %-EIO - I/O error
2226  *
2227  * %-ENOSPC - No space left on device (only in a panic state).
2228  *
2229  * %-ERESTARTSYS - Interrupted.
2230  *
2231  * %-ENOMEM - Insufficient memory available.
2232  */
2233 int nilfs_construct_segment(struct super_block *sb)
2234 {
2235 	struct the_nilfs *nilfs = sb->s_fs_info;
2236 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2237 	struct nilfs_transaction_info *ti;
2238 	int err;
2239 
2240 	if (!sci)
2241 		return -EROFS;
2242 
2243 	/* A call inside transactions causes a deadlock. */
2244 	BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2245 
2246 	err = nilfs_segctor_sync(sci);
2247 	return err;
2248 }
2249 
2250 /**
2251  * nilfs_construct_dsync_segment - construct a data-only logical segment
2252  * @sb: super block
2253  * @inode: inode whose data blocks should be written out
2254  * @start: start byte offset
2255  * @end: end byte offset (inclusive)
2256  *
2257  * Return Value: On success, 0 is returned. On errors, one of the following
2258  * negative error code is returned.
2259  *
2260  * %-EROFS - Read only filesystem.
2261  *
2262  * %-EIO - I/O error
2263  *
2264  * %-ENOSPC - No space left on device (only in a panic state).
2265  *
2266  * %-ERESTARTSYS - Interrupted.
2267  *
2268  * %-ENOMEM - Insufficient memory available.
2269  */
2270 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2271 				  loff_t start, loff_t end)
2272 {
2273 	struct the_nilfs *nilfs = sb->s_fs_info;
2274 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2275 	struct nilfs_inode_info *ii;
2276 	struct nilfs_transaction_info ti;
2277 	int err = 0;
2278 
2279 	if (!sci)
2280 		return -EROFS;
2281 
2282 	nilfs_transaction_lock(sb, &ti, 0);
2283 
2284 	ii = NILFS_I(inode);
2285 	if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2286 	    nilfs_test_opt(nilfs, STRICT_ORDER) ||
2287 	    test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2288 	    nilfs_discontinued(nilfs)) {
2289 		nilfs_transaction_unlock(sb);
2290 		err = nilfs_segctor_sync(sci);
2291 		return err;
2292 	}
2293 
2294 	spin_lock(&nilfs->ns_inode_lock);
2295 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2296 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2297 		spin_unlock(&nilfs->ns_inode_lock);
2298 		nilfs_transaction_unlock(sb);
2299 		return 0;
2300 	}
2301 	spin_unlock(&nilfs->ns_inode_lock);
2302 	sci->sc_dsync_inode = ii;
2303 	sci->sc_dsync_start = start;
2304 	sci->sc_dsync_end = end;
2305 
2306 	err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2307 	if (!err)
2308 		nilfs->ns_flushed_device = 0;
2309 
2310 	nilfs_transaction_unlock(sb);
2311 	return err;
2312 }
2313 
2314 #define FLUSH_FILE_BIT	(0x1) /* data file only */
2315 #define FLUSH_DAT_BIT	BIT(NILFS_DAT_INO) /* DAT only */
2316 
2317 /**
2318  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2319  * @sci: segment constructor object
2320  */
2321 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2322 {
2323 	spin_lock(&sci->sc_state_lock);
2324 	sci->sc_seq_accepted = sci->sc_seq_request;
2325 	spin_unlock(&sci->sc_state_lock);
2326 	del_timer_sync(&sci->sc_timer);
2327 }
2328 
2329 /**
2330  * nilfs_segctor_notify - notify the result of request to caller threads
2331  * @sci: segment constructor object
2332  * @mode: mode of log forming
2333  * @err: error code to be notified
2334  */
2335 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2336 {
2337 	/* Clear requests (even when the construction failed) */
2338 	spin_lock(&sci->sc_state_lock);
2339 
2340 	if (mode == SC_LSEG_SR) {
2341 		sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2342 		sci->sc_seq_done = sci->sc_seq_accepted;
2343 		nilfs_segctor_wakeup(sci, err);
2344 		sci->sc_flush_request = 0;
2345 	} else {
2346 		if (mode == SC_FLUSH_FILE)
2347 			sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2348 		else if (mode == SC_FLUSH_DAT)
2349 			sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2350 
2351 		/* re-enable timer if checkpoint creation was not done */
2352 		if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2353 		    time_before(jiffies, sci->sc_timer.expires))
2354 			add_timer(&sci->sc_timer);
2355 	}
2356 	spin_unlock(&sci->sc_state_lock);
2357 }
2358 
2359 /**
2360  * nilfs_segctor_construct - form logs and write them to disk
2361  * @sci: segment constructor object
2362  * @mode: mode of log forming
2363  */
2364 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2365 {
2366 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2367 	struct nilfs_super_block **sbp;
2368 	int err = 0;
2369 
2370 	nilfs_segctor_accept(sci);
2371 
2372 	if (nilfs_discontinued(nilfs))
2373 		mode = SC_LSEG_SR;
2374 	if (!nilfs_segctor_confirm(sci))
2375 		err = nilfs_segctor_do_construct(sci, mode);
2376 
2377 	if (likely(!err)) {
2378 		if (mode != SC_FLUSH_DAT)
2379 			atomic_set(&nilfs->ns_ndirtyblks, 0);
2380 		if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2381 		    nilfs_discontinued(nilfs)) {
2382 			down_write(&nilfs->ns_sem);
2383 			err = -EIO;
2384 			sbp = nilfs_prepare_super(sci->sc_super,
2385 						  nilfs_sb_will_flip(nilfs));
2386 			if (likely(sbp)) {
2387 				nilfs_set_log_cursor(sbp[0], nilfs);
2388 				err = nilfs_commit_super(sci->sc_super,
2389 							 NILFS_SB_COMMIT);
2390 			}
2391 			up_write(&nilfs->ns_sem);
2392 		}
2393 	}
2394 
2395 	nilfs_segctor_notify(sci, mode, err);
2396 	return err;
2397 }
2398 
2399 static void nilfs_construction_timeout(struct timer_list *t)
2400 {
2401 	struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2402 
2403 	wake_up_process(sci->sc_timer_task);
2404 }
2405 
2406 static void
2407 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2408 {
2409 	struct nilfs_inode_info *ii, *n;
2410 
2411 	list_for_each_entry_safe(ii, n, head, i_dirty) {
2412 		if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2413 			continue;
2414 		list_del_init(&ii->i_dirty);
2415 		truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2416 		nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2417 		iput(&ii->vfs_inode);
2418 	}
2419 }
2420 
2421 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2422 			 void **kbufs)
2423 {
2424 	struct the_nilfs *nilfs = sb->s_fs_info;
2425 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2426 	struct nilfs_transaction_info ti;
2427 	int err;
2428 
2429 	if (unlikely(!sci))
2430 		return -EROFS;
2431 
2432 	nilfs_transaction_lock(sb, &ti, 1);
2433 
2434 	err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2435 	if (unlikely(err))
2436 		goto out_unlock;
2437 
2438 	err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2439 	if (unlikely(err)) {
2440 		nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2441 		goto out_unlock;
2442 	}
2443 
2444 	sci->sc_freesegs = kbufs[4];
2445 	sci->sc_nfreesegs = argv[4].v_nmembs;
2446 	list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2447 
2448 	for (;;) {
2449 		err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2450 		nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2451 
2452 		if (likely(!err))
2453 			break;
2454 
2455 		nilfs_warn(sb, "error %d cleaning segments", err);
2456 		set_current_state(TASK_INTERRUPTIBLE);
2457 		schedule_timeout(sci->sc_interval);
2458 	}
2459 	if (nilfs_test_opt(nilfs, DISCARD)) {
2460 		int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2461 						 sci->sc_nfreesegs);
2462 		if (ret) {
2463 			nilfs_warn(sb,
2464 				   "error %d on discard request, turning discards off for the device",
2465 				   ret);
2466 			nilfs_clear_opt(nilfs, DISCARD);
2467 		}
2468 	}
2469 
2470  out_unlock:
2471 	sci->sc_freesegs = NULL;
2472 	sci->sc_nfreesegs = 0;
2473 	nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2474 	nilfs_transaction_unlock(sb);
2475 	return err;
2476 }
2477 
2478 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2479 {
2480 	struct nilfs_transaction_info ti;
2481 
2482 	nilfs_transaction_lock(sci->sc_super, &ti, 0);
2483 	nilfs_segctor_construct(sci, mode);
2484 
2485 	/*
2486 	 * Unclosed segment should be retried.  We do this using sc_timer.
2487 	 * Timeout of sc_timer will invoke complete construction which leads
2488 	 * to close the current logical segment.
2489 	 */
2490 	if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2491 		nilfs_segctor_start_timer(sci);
2492 
2493 	nilfs_transaction_unlock(sci->sc_super);
2494 }
2495 
2496 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2497 {
2498 	int mode = 0;
2499 
2500 	spin_lock(&sci->sc_state_lock);
2501 	mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2502 		SC_FLUSH_DAT : SC_FLUSH_FILE;
2503 	spin_unlock(&sci->sc_state_lock);
2504 
2505 	if (mode) {
2506 		nilfs_segctor_do_construct(sci, mode);
2507 
2508 		spin_lock(&sci->sc_state_lock);
2509 		sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2510 			~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2511 		spin_unlock(&sci->sc_state_lock);
2512 	}
2513 	clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2514 }
2515 
2516 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2517 {
2518 	if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2519 	    time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2520 		if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2521 			return SC_FLUSH_FILE;
2522 		else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2523 			return SC_FLUSH_DAT;
2524 	}
2525 	return SC_LSEG_SR;
2526 }
2527 
2528 /**
2529  * nilfs_segctor_thread - main loop of the segment constructor thread.
2530  * @arg: pointer to a struct nilfs_sc_info.
2531  *
2532  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2533  * to execute segment constructions.
2534  */
2535 static int nilfs_segctor_thread(void *arg)
2536 {
2537 	struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2538 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2539 	int timeout = 0;
2540 
2541 	sci->sc_timer_task = current;
2542 
2543 	/* start sync. */
2544 	sci->sc_task = current;
2545 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2546 	nilfs_info(sci->sc_super,
2547 		   "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2548 		   sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2549 
2550 	spin_lock(&sci->sc_state_lock);
2551  loop:
2552 	for (;;) {
2553 		int mode;
2554 
2555 		if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2556 			goto end_thread;
2557 
2558 		if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2559 			mode = SC_LSEG_SR;
2560 		else if (sci->sc_flush_request)
2561 			mode = nilfs_segctor_flush_mode(sci);
2562 		else
2563 			break;
2564 
2565 		spin_unlock(&sci->sc_state_lock);
2566 		nilfs_segctor_thread_construct(sci, mode);
2567 		spin_lock(&sci->sc_state_lock);
2568 		timeout = 0;
2569 	}
2570 
2571 
2572 	if (freezing(current)) {
2573 		spin_unlock(&sci->sc_state_lock);
2574 		try_to_freeze();
2575 		spin_lock(&sci->sc_state_lock);
2576 	} else {
2577 		DEFINE_WAIT(wait);
2578 		int should_sleep = 1;
2579 
2580 		prepare_to_wait(&sci->sc_wait_daemon, &wait,
2581 				TASK_INTERRUPTIBLE);
2582 
2583 		if (sci->sc_seq_request != sci->sc_seq_done)
2584 			should_sleep = 0;
2585 		else if (sci->sc_flush_request)
2586 			should_sleep = 0;
2587 		else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2588 			should_sleep = time_before(jiffies,
2589 					sci->sc_timer.expires);
2590 
2591 		if (should_sleep) {
2592 			spin_unlock(&sci->sc_state_lock);
2593 			schedule();
2594 			spin_lock(&sci->sc_state_lock);
2595 		}
2596 		finish_wait(&sci->sc_wait_daemon, &wait);
2597 		timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2598 			   time_after_eq(jiffies, sci->sc_timer.expires));
2599 
2600 		if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2601 			set_nilfs_discontinued(nilfs);
2602 	}
2603 	goto loop;
2604 
2605  end_thread:
2606 	spin_unlock(&sci->sc_state_lock);
2607 
2608 	/* end sync. */
2609 	sci->sc_task = NULL;
2610 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2611 	return 0;
2612 }
2613 
2614 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2615 {
2616 	struct task_struct *t;
2617 
2618 	t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2619 	if (IS_ERR(t)) {
2620 		int err = PTR_ERR(t);
2621 
2622 		nilfs_err(sci->sc_super, "error %d creating segctord thread",
2623 			  err);
2624 		return err;
2625 	}
2626 	wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2627 	return 0;
2628 }
2629 
2630 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2631 	__acquires(&sci->sc_state_lock)
2632 	__releases(&sci->sc_state_lock)
2633 {
2634 	sci->sc_state |= NILFS_SEGCTOR_QUIT;
2635 
2636 	while (sci->sc_task) {
2637 		wake_up(&sci->sc_wait_daemon);
2638 		spin_unlock(&sci->sc_state_lock);
2639 		wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2640 		spin_lock(&sci->sc_state_lock);
2641 	}
2642 }
2643 
2644 /*
2645  * Setup & clean-up functions
2646  */
2647 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2648 					       struct nilfs_root *root)
2649 {
2650 	struct the_nilfs *nilfs = sb->s_fs_info;
2651 	struct nilfs_sc_info *sci;
2652 
2653 	sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2654 	if (!sci)
2655 		return NULL;
2656 
2657 	sci->sc_super = sb;
2658 
2659 	nilfs_get_root(root);
2660 	sci->sc_root = root;
2661 
2662 	init_waitqueue_head(&sci->sc_wait_request);
2663 	init_waitqueue_head(&sci->sc_wait_daemon);
2664 	init_waitqueue_head(&sci->sc_wait_task);
2665 	spin_lock_init(&sci->sc_state_lock);
2666 	INIT_LIST_HEAD(&sci->sc_dirty_files);
2667 	INIT_LIST_HEAD(&sci->sc_segbufs);
2668 	INIT_LIST_HEAD(&sci->sc_write_logs);
2669 	INIT_LIST_HEAD(&sci->sc_gc_inodes);
2670 	INIT_LIST_HEAD(&sci->sc_iput_queue);
2671 	INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2672 	timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2673 
2674 	sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2675 	sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2676 	sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2677 
2678 	if (nilfs->ns_interval)
2679 		sci->sc_interval = HZ * nilfs->ns_interval;
2680 	if (nilfs->ns_watermark)
2681 		sci->sc_watermark = nilfs->ns_watermark;
2682 	return sci;
2683 }
2684 
2685 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2686 {
2687 	int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2688 
2689 	/*
2690 	 * The segctord thread was stopped and its timer was removed.
2691 	 * But some tasks remain.
2692 	 */
2693 	do {
2694 		struct nilfs_transaction_info ti;
2695 
2696 		nilfs_transaction_lock(sci->sc_super, &ti, 0);
2697 		ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2698 		nilfs_transaction_unlock(sci->sc_super);
2699 
2700 		flush_work(&sci->sc_iput_work);
2701 
2702 	} while (ret && retrycount-- > 0);
2703 }
2704 
2705 /**
2706  * nilfs_segctor_destroy - destroy the segment constructor.
2707  * @sci: nilfs_sc_info
2708  *
2709  * nilfs_segctor_destroy() kills the segctord thread and frees
2710  * the nilfs_sc_info struct.
2711  * Caller must hold the segment semaphore.
2712  */
2713 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2714 {
2715 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2716 	int flag;
2717 
2718 	up_write(&nilfs->ns_segctor_sem);
2719 
2720 	spin_lock(&sci->sc_state_lock);
2721 	nilfs_segctor_kill_thread(sci);
2722 	flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2723 		|| sci->sc_seq_request != sci->sc_seq_done);
2724 	spin_unlock(&sci->sc_state_lock);
2725 
2726 	if (flush_work(&sci->sc_iput_work))
2727 		flag = true;
2728 
2729 	if (flag || !nilfs_segctor_confirm(sci))
2730 		nilfs_segctor_write_out(sci);
2731 
2732 	if (!list_empty(&sci->sc_dirty_files)) {
2733 		nilfs_warn(sci->sc_super,
2734 			   "disposed unprocessed dirty file(s) when stopping log writer");
2735 		nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2736 	}
2737 
2738 	if (!list_empty(&sci->sc_iput_queue)) {
2739 		nilfs_warn(sci->sc_super,
2740 			   "disposed unprocessed inode(s) in iput queue when stopping log writer");
2741 		nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2742 	}
2743 
2744 	WARN_ON(!list_empty(&sci->sc_segbufs));
2745 	WARN_ON(!list_empty(&sci->sc_write_logs));
2746 
2747 	nilfs_put_root(sci->sc_root);
2748 
2749 	down_write(&nilfs->ns_segctor_sem);
2750 
2751 	del_timer_sync(&sci->sc_timer);
2752 	kfree(sci);
2753 }
2754 
2755 /**
2756  * nilfs_attach_log_writer - attach log writer
2757  * @sb: super block instance
2758  * @root: root object of the current filesystem tree
2759  *
2760  * This allocates a log writer object, initializes it, and starts the
2761  * log writer.
2762  *
2763  * Return Value: On success, 0 is returned. On error, one of the following
2764  * negative error code is returned.
2765  *
2766  * %-ENOMEM - Insufficient memory available.
2767  */
2768 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2769 {
2770 	struct the_nilfs *nilfs = sb->s_fs_info;
2771 	int err;
2772 
2773 	if (nilfs->ns_writer) {
2774 		/*
2775 		 * This happens if the filesystem was remounted
2776 		 * read/write after nilfs_error degenerated it into a
2777 		 * read-only mount.
2778 		 */
2779 		nilfs_detach_log_writer(sb);
2780 	}
2781 
2782 	nilfs->ns_writer = nilfs_segctor_new(sb, root);
2783 	if (!nilfs->ns_writer)
2784 		return -ENOMEM;
2785 
2786 	inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2787 
2788 	err = nilfs_segctor_start_thread(nilfs->ns_writer);
2789 	if (err) {
2790 		kfree(nilfs->ns_writer);
2791 		nilfs->ns_writer = NULL;
2792 	}
2793 	return err;
2794 }
2795 
2796 /**
2797  * nilfs_detach_log_writer - destroy log writer
2798  * @sb: super block instance
2799  *
2800  * This kills log writer daemon, frees the log writer object, and
2801  * destroys list of dirty files.
2802  */
2803 void nilfs_detach_log_writer(struct super_block *sb)
2804 {
2805 	struct the_nilfs *nilfs = sb->s_fs_info;
2806 	LIST_HEAD(garbage_list);
2807 
2808 	down_write(&nilfs->ns_segctor_sem);
2809 	if (nilfs->ns_writer) {
2810 		nilfs_segctor_destroy(nilfs->ns_writer);
2811 		nilfs->ns_writer = NULL;
2812 	}
2813 
2814 	/* Force to free the list of dirty files */
2815 	spin_lock(&nilfs->ns_inode_lock);
2816 	if (!list_empty(&nilfs->ns_dirty_files)) {
2817 		list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2818 		nilfs_warn(sb,
2819 			   "disposed unprocessed dirty file(s) when detaching log writer");
2820 	}
2821 	spin_unlock(&nilfs->ns_inode_lock);
2822 	up_write(&nilfs->ns_segctor_sem);
2823 
2824 	nilfs_dispose_list(nilfs, &garbage_list, 1);
2825 }
2826