xref: /linux/fs/nilfs2/segment.c (revision 955abe0a1b41de5ba61fe4cd614ebc123084d499)
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) - (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 (sb_rdonly(sb) || unlikely(!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 /**
434  * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
435  * @sci: segment constructor object
436  *
437  * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
438  * the current segment summary block.
439  */
440 static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
441 {
442 	struct nilfs_segsum_pointer *ssp;
443 
444 	ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
445 	if (ssp->offset < ssp->bh->b_size)
446 		memset(ssp->bh->b_data + ssp->offset, 0,
447 		       ssp->bh->b_size - ssp->offset);
448 }
449 
450 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
451 {
452 	sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
453 	if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
454 		return -E2BIG; /*
455 				* The current segment is filled up
456 				* (internal code)
457 				*/
458 	nilfs_segctor_zeropad_segsum(sci);
459 	sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
460 	return nilfs_segctor_reset_segment_buffer(sci);
461 }
462 
463 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
464 {
465 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
466 	int err;
467 
468 	if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
469 		err = nilfs_segctor_feed_segment(sci);
470 		if (err)
471 			return err;
472 		segbuf = sci->sc_curseg;
473 	}
474 	err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
475 	if (likely(!err))
476 		segbuf->sb_sum.flags |= NILFS_SS_SR;
477 	return err;
478 }
479 
480 /*
481  * Functions for making segment summary and payloads
482  */
483 static int nilfs_segctor_segsum_block_required(
484 	struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
485 	unsigned int binfo_size)
486 {
487 	unsigned int blocksize = sci->sc_super->s_blocksize;
488 	/* Size of finfo and binfo is enough small against blocksize */
489 
490 	return ssp->offset + binfo_size +
491 		(!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
492 		blocksize;
493 }
494 
495 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
496 				      struct inode *inode)
497 {
498 	sci->sc_curseg->sb_sum.nfinfo++;
499 	sci->sc_binfo_ptr = sci->sc_finfo_ptr;
500 	nilfs_segctor_map_segsum_entry(
501 		sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
502 
503 	if (NILFS_I(inode)->i_root &&
504 	    !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
505 		set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
506 	/* skip finfo */
507 }
508 
509 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
510 				    struct inode *inode)
511 {
512 	struct nilfs_finfo *finfo;
513 	struct nilfs_inode_info *ii;
514 	struct nilfs_segment_buffer *segbuf;
515 	__u64 cno;
516 
517 	if (sci->sc_blk_cnt == 0)
518 		return;
519 
520 	ii = NILFS_I(inode);
521 
522 	if (test_bit(NILFS_I_GCINODE, &ii->i_state))
523 		cno = ii->i_cno;
524 	else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
525 		cno = 0;
526 	else
527 		cno = sci->sc_cno;
528 
529 	finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
530 						 sizeof(*finfo));
531 	finfo->fi_ino = cpu_to_le64(inode->i_ino);
532 	finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
533 	finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
534 	finfo->fi_cno = cpu_to_le64(cno);
535 
536 	segbuf = sci->sc_curseg;
537 	segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
538 		sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
539 	sci->sc_finfo_ptr = sci->sc_binfo_ptr;
540 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
541 }
542 
543 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
544 					struct buffer_head *bh,
545 					struct inode *inode,
546 					unsigned int binfo_size)
547 {
548 	struct nilfs_segment_buffer *segbuf;
549 	int required, err = 0;
550 
551  retry:
552 	segbuf = sci->sc_curseg;
553 	required = nilfs_segctor_segsum_block_required(
554 		sci, &sci->sc_binfo_ptr, binfo_size);
555 	if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
556 		nilfs_segctor_end_finfo(sci, inode);
557 		err = nilfs_segctor_feed_segment(sci);
558 		if (err)
559 			return err;
560 		goto retry;
561 	}
562 	if (unlikely(required)) {
563 		nilfs_segctor_zeropad_segsum(sci);
564 		err = nilfs_segbuf_extend_segsum(segbuf);
565 		if (unlikely(err))
566 			goto failed;
567 	}
568 	if (sci->sc_blk_cnt == 0)
569 		nilfs_segctor_begin_finfo(sci, inode);
570 
571 	nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
572 	/* Substitution to vblocknr is delayed until update_blocknr() */
573 	nilfs_segbuf_add_file_buffer(segbuf, bh);
574 	sci->sc_blk_cnt++;
575  failed:
576 	return err;
577 }
578 
579 /*
580  * Callback functions that enumerate, mark, and collect dirty blocks
581  */
582 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
583 				   struct buffer_head *bh, struct inode *inode)
584 {
585 	int err;
586 
587 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
588 	if (err < 0)
589 		return err;
590 
591 	err = nilfs_segctor_add_file_block(sci, bh, inode,
592 					   sizeof(struct nilfs_binfo_v));
593 	if (!err)
594 		sci->sc_datablk_cnt++;
595 	return err;
596 }
597 
598 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
599 				   struct buffer_head *bh,
600 				   struct inode *inode)
601 {
602 	return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
603 }
604 
605 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
606 				   struct buffer_head *bh,
607 				   struct inode *inode)
608 {
609 	WARN_ON(!buffer_dirty(bh));
610 	return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
611 }
612 
613 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
614 					struct nilfs_segsum_pointer *ssp,
615 					union nilfs_binfo *binfo)
616 {
617 	struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
618 		sci, ssp, sizeof(*binfo_v));
619 	*binfo_v = binfo->bi_v;
620 }
621 
622 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
623 					struct nilfs_segsum_pointer *ssp,
624 					union nilfs_binfo *binfo)
625 {
626 	__le64 *vblocknr = nilfs_segctor_map_segsum_entry(
627 		sci, ssp, sizeof(*vblocknr));
628 	*vblocknr = binfo->bi_v.bi_vblocknr;
629 }
630 
631 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
632 	.collect_data = nilfs_collect_file_data,
633 	.collect_node = nilfs_collect_file_node,
634 	.collect_bmap = nilfs_collect_file_bmap,
635 	.write_data_binfo = nilfs_write_file_data_binfo,
636 	.write_node_binfo = nilfs_write_file_node_binfo,
637 };
638 
639 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
640 				  struct buffer_head *bh, struct inode *inode)
641 {
642 	int err;
643 
644 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
645 	if (err < 0)
646 		return err;
647 
648 	err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
649 	if (!err)
650 		sci->sc_datablk_cnt++;
651 	return err;
652 }
653 
654 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
655 				  struct buffer_head *bh, struct inode *inode)
656 {
657 	WARN_ON(!buffer_dirty(bh));
658 	return nilfs_segctor_add_file_block(sci, bh, inode,
659 					    sizeof(struct nilfs_binfo_dat));
660 }
661 
662 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
663 				       struct nilfs_segsum_pointer *ssp,
664 				       union nilfs_binfo *binfo)
665 {
666 	__le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
667 							  sizeof(*blkoff));
668 	*blkoff = binfo->bi_dat.bi_blkoff;
669 }
670 
671 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
672 				       struct nilfs_segsum_pointer *ssp,
673 				       union nilfs_binfo *binfo)
674 {
675 	struct nilfs_binfo_dat *binfo_dat =
676 		nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
677 	*binfo_dat = binfo->bi_dat;
678 }
679 
680 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
681 	.collect_data = nilfs_collect_dat_data,
682 	.collect_node = nilfs_collect_file_node,
683 	.collect_bmap = nilfs_collect_dat_bmap,
684 	.write_data_binfo = nilfs_write_dat_data_binfo,
685 	.write_node_binfo = nilfs_write_dat_node_binfo,
686 };
687 
688 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
689 	.collect_data = nilfs_collect_file_data,
690 	.collect_node = NULL,
691 	.collect_bmap = NULL,
692 	.write_data_binfo = nilfs_write_file_data_binfo,
693 	.write_node_binfo = NULL,
694 };
695 
696 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
697 					      struct list_head *listp,
698 					      size_t nlimit,
699 					      loff_t start, loff_t end)
700 {
701 	struct address_space *mapping = inode->i_mapping;
702 	struct folio_batch fbatch;
703 	pgoff_t index = 0, last = ULONG_MAX;
704 	size_t ndirties = 0;
705 	int i;
706 
707 	if (unlikely(start != 0 || end != LLONG_MAX)) {
708 		/*
709 		 * A valid range is given for sync-ing data pages. The
710 		 * range is rounded to per-page; extra dirty buffers
711 		 * may be included if blocksize < pagesize.
712 		 */
713 		index = start >> PAGE_SHIFT;
714 		last = end >> PAGE_SHIFT;
715 	}
716 	folio_batch_init(&fbatch);
717  repeat:
718 	if (unlikely(index > last) ||
719 	      !filemap_get_folios_tag(mapping, &index, last,
720 		      PAGECACHE_TAG_DIRTY, &fbatch))
721 		return ndirties;
722 
723 	for (i = 0; i < folio_batch_count(&fbatch); i++) {
724 		struct buffer_head *bh, *head;
725 		struct folio *folio = fbatch.folios[i];
726 
727 		folio_lock(folio);
728 		if (unlikely(folio->mapping != mapping)) {
729 			/* Exclude folios removed from the address space */
730 			folio_unlock(folio);
731 			continue;
732 		}
733 		head = folio_buffers(folio);
734 		if (!head)
735 			head = create_empty_buffers(folio,
736 					i_blocksize(inode), 0);
737 		folio_unlock(folio);
738 
739 		bh = head;
740 		do {
741 			if (!buffer_dirty(bh) || buffer_async_write(bh))
742 				continue;
743 			get_bh(bh);
744 			list_add_tail(&bh->b_assoc_buffers, listp);
745 			ndirties++;
746 			if (unlikely(ndirties >= nlimit)) {
747 				folio_batch_release(&fbatch);
748 				cond_resched();
749 				return ndirties;
750 			}
751 		} while (bh = bh->b_this_page, bh != head);
752 	}
753 	folio_batch_release(&fbatch);
754 	cond_resched();
755 	goto repeat;
756 }
757 
758 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
759 					    struct list_head *listp)
760 {
761 	struct nilfs_inode_info *ii = NILFS_I(inode);
762 	struct inode *btnc_inode = ii->i_assoc_inode;
763 	struct folio_batch fbatch;
764 	struct buffer_head *bh, *head;
765 	unsigned int i;
766 	pgoff_t index = 0;
767 
768 	if (!btnc_inode)
769 		return;
770 	folio_batch_init(&fbatch);
771 
772 	while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
773 				(pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
774 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
775 			bh = head = folio_buffers(fbatch.folios[i]);
776 			do {
777 				if (buffer_dirty(bh) &&
778 						!buffer_async_write(bh)) {
779 					get_bh(bh);
780 					list_add_tail(&bh->b_assoc_buffers,
781 						      listp);
782 				}
783 				bh = bh->b_this_page;
784 			} while (bh != head);
785 		}
786 		folio_batch_release(&fbatch);
787 		cond_resched();
788 	}
789 }
790 
791 static void nilfs_dispose_list(struct the_nilfs *nilfs,
792 			       struct list_head *head, int force)
793 {
794 	struct nilfs_inode_info *ii, *n;
795 	struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
796 	unsigned int nv = 0;
797 
798 	while (!list_empty(head)) {
799 		spin_lock(&nilfs->ns_inode_lock);
800 		list_for_each_entry_safe(ii, n, head, i_dirty) {
801 			list_del_init(&ii->i_dirty);
802 			if (force) {
803 				if (unlikely(ii->i_bh)) {
804 					brelse(ii->i_bh);
805 					ii->i_bh = NULL;
806 				}
807 			} else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
808 				set_bit(NILFS_I_QUEUED, &ii->i_state);
809 				list_add_tail(&ii->i_dirty,
810 					      &nilfs->ns_dirty_files);
811 				continue;
812 			}
813 			ivec[nv++] = ii;
814 			if (nv == SC_N_INODEVEC)
815 				break;
816 		}
817 		spin_unlock(&nilfs->ns_inode_lock);
818 
819 		for (pii = ivec; nv > 0; pii++, nv--)
820 			iput(&(*pii)->vfs_inode);
821 	}
822 }
823 
824 static void nilfs_iput_work_func(struct work_struct *work)
825 {
826 	struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
827 						 sc_iput_work);
828 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
829 
830 	nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
831 }
832 
833 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
834 				     struct nilfs_root *root)
835 {
836 	int ret = 0;
837 
838 	if (nilfs_mdt_fetch_dirty(root->ifile))
839 		ret++;
840 	if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
841 		ret++;
842 	if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
843 		ret++;
844 	if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
845 		ret++;
846 	return ret;
847 }
848 
849 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
850 {
851 	return list_empty(&sci->sc_dirty_files) &&
852 		!test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
853 		sci->sc_nfreesegs == 0 &&
854 		(!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
855 }
856 
857 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
858 {
859 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
860 	int ret = 0;
861 
862 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
863 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
864 
865 	spin_lock(&nilfs->ns_inode_lock);
866 	if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
867 		ret++;
868 
869 	spin_unlock(&nilfs->ns_inode_lock);
870 	return ret;
871 }
872 
873 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
874 {
875 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
876 
877 	nilfs_mdt_clear_dirty(sci->sc_root->ifile);
878 	nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
879 	nilfs_mdt_clear_dirty(nilfs->ns_sufile);
880 	nilfs_mdt_clear_dirty(nilfs->ns_dat);
881 }
882 
883 static void nilfs_fill_in_file_bmap(struct inode *ifile,
884 				    struct nilfs_inode_info *ii)
885 
886 {
887 	struct buffer_head *ibh;
888 	struct nilfs_inode *raw_inode;
889 
890 	if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
891 		ibh = ii->i_bh;
892 		BUG_ON(!ibh);
893 		raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
894 						  ibh);
895 		nilfs_bmap_write(ii->i_bmap, raw_inode);
896 		nilfs_ifile_unmap_inode(raw_inode);
897 	}
898 }
899 
900 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
901 {
902 	struct nilfs_inode_info *ii;
903 
904 	list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
905 		nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
906 		set_bit(NILFS_I_COLLECTED, &ii->i_state);
907 	}
908 }
909 
910 /**
911  * nilfs_write_root_mdt_inode - export root metadata inode information to
912  *                              the on-disk inode
913  * @inode:     inode object of the root metadata file
914  * @raw_inode: on-disk inode
915  *
916  * nilfs_write_root_mdt_inode() writes inode information and bmap data of
917  * @inode to the inode area of the metadata file allocated on the super root
918  * block created to finalize the log.  Since super root blocks are configured
919  * each time, this function zero-fills the unused area of @raw_inode.
920  */
921 static void nilfs_write_root_mdt_inode(struct inode *inode,
922 				       struct nilfs_inode *raw_inode)
923 {
924 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
925 
926 	nilfs_write_inode_common(inode, raw_inode);
927 
928 	/* zero-fill unused portion of raw_inode */
929 	raw_inode->i_xattr = 0;
930 	raw_inode->i_pad = 0;
931 	memset((void *)raw_inode + sizeof(*raw_inode), 0,
932 	       nilfs->ns_inode_size - sizeof(*raw_inode));
933 
934 	nilfs_bmap_write(NILFS_I(inode)->i_bmap, raw_inode);
935 }
936 
937 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
938 					     struct the_nilfs *nilfs)
939 {
940 	struct buffer_head *bh_sr;
941 	struct nilfs_super_root *raw_sr;
942 	unsigned int isz, srsz;
943 
944 	bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
945 
946 	lock_buffer(bh_sr);
947 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
948 	isz = nilfs->ns_inode_size;
949 	srsz = NILFS_SR_BYTES(isz);
950 
951 	raw_sr->sr_sum = 0;  /* Ensure initialization within this update */
952 	raw_sr->sr_bytes = cpu_to_le16(srsz);
953 	raw_sr->sr_nongc_ctime
954 		= cpu_to_le64(nilfs_doing_gc() ?
955 			      nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
956 	raw_sr->sr_flags = 0;
957 
958 	nilfs_write_root_mdt_inode(nilfs->ns_dat, (void *)raw_sr +
959 				   NILFS_SR_DAT_OFFSET(isz));
960 	nilfs_write_root_mdt_inode(nilfs->ns_cpfile, (void *)raw_sr +
961 				   NILFS_SR_CPFILE_OFFSET(isz));
962 	nilfs_write_root_mdt_inode(nilfs->ns_sufile, (void *)raw_sr +
963 				   NILFS_SR_SUFILE_OFFSET(isz));
964 
965 	memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
966 	set_buffer_uptodate(bh_sr);
967 	unlock_buffer(bh_sr);
968 }
969 
970 static void nilfs_redirty_inodes(struct list_head *head)
971 {
972 	struct nilfs_inode_info *ii;
973 
974 	list_for_each_entry(ii, head, i_dirty) {
975 		if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
976 			clear_bit(NILFS_I_COLLECTED, &ii->i_state);
977 	}
978 }
979 
980 static void nilfs_drop_collected_inodes(struct list_head *head)
981 {
982 	struct nilfs_inode_info *ii;
983 
984 	list_for_each_entry(ii, head, i_dirty) {
985 		if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
986 			continue;
987 
988 		clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
989 		set_bit(NILFS_I_UPDATED, &ii->i_state);
990 	}
991 }
992 
993 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
994 				       struct inode *inode,
995 				       struct list_head *listp,
996 				       int (*collect)(struct nilfs_sc_info *,
997 						      struct buffer_head *,
998 						      struct inode *))
999 {
1000 	struct buffer_head *bh, *n;
1001 	int err = 0;
1002 
1003 	if (collect) {
1004 		list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1005 			list_del_init(&bh->b_assoc_buffers);
1006 			err = collect(sci, bh, inode);
1007 			brelse(bh);
1008 			if (unlikely(err))
1009 				goto dispose_buffers;
1010 		}
1011 		return 0;
1012 	}
1013 
1014  dispose_buffers:
1015 	while (!list_empty(listp)) {
1016 		bh = list_first_entry(listp, struct buffer_head,
1017 				      b_assoc_buffers);
1018 		list_del_init(&bh->b_assoc_buffers);
1019 		brelse(bh);
1020 	}
1021 	return err;
1022 }
1023 
1024 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1025 {
1026 	/* Remaining number of blocks within segment buffer */
1027 	return sci->sc_segbuf_nblocks -
1028 		(sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1029 }
1030 
1031 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1032 				   struct inode *inode,
1033 				   const struct nilfs_sc_operations *sc_ops)
1034 {
1035 	LIST_HEAD(data_buffers);
1036 	LIST_HEAD(node_buffers);
1037 	int err;
1038 
1039 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1040 		size_t n, rest = nilfs_segctor_buffer_rest(sci);
1041 
1042 		n = nilfs_lookup_dirty_data_buffers(
1043 			inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1044 		if (n > rest) {
1045 			err = nilfs_segctor_apply_buffers(
1046 				sci, inode, &data_buffers,
1047 				sc_ops->collect_data);
1048 			BUG_ON(!err); /* always receive -E2BIG or true error */
1049 			goto break_or_fail;
1050 		}
1051 	}
1052 	nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1053 
1054 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1055 		err = nilfs_segctor_apply_buffers(
1056 			sci, inode, &data_buffers, sc_ops->collect_data);
1057 		if (unlikely(err)) {
1058 			/* dispose node list */
1059 			nilfs_segctor_apply_buffers(
1060 				sci, inode, &node_buffers, NULL);
1061 			goto break_or_fail;
1062 		}
1063 		sci->sc_stage.flags |= NILFS_CF_NODE;
1064 	}
1065 	/* Collect node */
1066 	err = nilfs_segctor_apply_buffers(
1067 		sci, inode, &node_buffers, sc_ops->collect_node);
1068 	if (unlikely(err))
1069 		goto break_or_fail;
1070 
1071 	nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1072 	err = nilfs_segctor_apply_buffers(
1073 		sci, inode, &node_buffers, sc_ops->collect_bmap);
1074 	if (unlikely(err))
1075 		goto break_or_fail;
1076 
1077 	nilfs_segctor_end_finfo(sci, inode);
1078 	sci->sc_stage.flags &= ~NILFS_CF_NODE;
1079 
1080  break_or_fail:
1081 	return err;
1082 }
1083 
1084 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1085 					 struct inode *inode)
1086 {
1087 	LIST_HEAD(data_buffers);
1088 	size_t n, rest = nilfs_segctor_buffer_rest(sci);
1089 	int err;
1090 
1091 	n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1092 					    sci->sc_dsync_start,
1093 					    sci->sc_dsync_end);
1094 
1095 	err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1096 					  nilfs_collect_file_data);
1097 	if (!err) {
1098 		nilfs_segctor_end_finfo(sci, inode);
1099 		BUG_ON(n > rest);
1100 		/* always receive -E2BIG or true error if n > rest */
1101 	}
1102 	return err;
1103 }
1104 
1105 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1106 {
1107 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1108 	struct list_head *head;
1109 	struct nilfs_inode_info *ii;
1110 	size_t ndone;
1111 	int err = 0;
1112 
1113 	switch (nilfs_sc_cstage_get(sci)) {
1114 	case NILFS_ST_INIT:
1115 		/* Pre-processes */
1116 		sci->sc_stage.flags = 0;
1117 
1118 		if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1119 			sci->sc_nblk_inc = 0;
1120 			sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1121 			if (mode == SC_LSEG_DSYNC) {
1122 				nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1123 				goto dsync_mode;
1124 			}
1125 		}
1126 
1127 		sci->sc_stage.dirty_file_ptr = NULL;
1128 		sci->sc_stage.gc_inode_ptr = NULL;
1129 		if (mode == SC_FLUSH_DAT) {
1130 			nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1131 			goto dat_stage;
1132 		}
1133 		nilfs_sc_cstage_inc(sci);
1134 		fallthrough;
1135 	case NILFS_ST_GC:
1136 		if (nilfs_doing_gc()) {
1137 			head = &sci->sc_gc_inodes;
1138 			ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1139 						head, i_dirty);
1140 			list_for_each_entry_continue(ii, head, i_dirty) {
1141 				err = nilfs_segctor_scan_file(
1142 					sci, &ii->vfs_inode,
1143 					&nilfs_sc_file_ops);
1144 				if (unlikely(err)) {
1145 					sci->sc_stage.gc_inode_ptr = list_entry(
1146 						ii->i_dirty.prev,
1147 						struct nilfs_inode_info,
1148 						i_dirty);
1149 					goto break_or_fail;
1150 				}
1151 				set_bit(NILFS_I_COLLECTED, &ii->i_state);
1152 			}
1153 			sci->sc_stage.gc_inode_ptr = NULL;
1154 		}
1155 		nilfs_sc_cstage_inc(sci);
1156 		fallthrough;
1157 	case NILFS_ST_FILE:
1158 		head = &sci->sc_dirty_files;
1159 		ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1160 					i_dirty);
1161 		list_for_each_entry_continue(ii, head, i_dirty) {
1162 			clear_bit(NILFS_I_DIRTY, &ii->i_state);
1163 
1164 			err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1165 						      &nilfs_sc_file_ops);
1166 			if (unlikely(err)) {
1167 				sci->sc_stage.dirty_file_ptr =
1168 					list_entry(ii->i_dirty.prev,
1169 						   struct nilfs_inode_info,
1170 						   i_dirty);
1171 				goto break_or_fail;
1172 			}
1173 			/* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1174 			/* XXX: required ? */
1175 		}
1176 		sci->sc_stage.dirty_file_ptr = NULL;
1177 		if (mode == SC_FLUSH_FILE) {
1178 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1179 			return 0;
1180 		}
1181 		nilfs_sc_cstage_inc(sci);
1182 		sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1183 		fallthrough;
1184 	case NILFS_ST_IFILE:
1185 		err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1186 					      &nilfs_sc_file_ops);
1187 		if (unlikely(err))
1188 			break;
1189 		nilfs_sc_cstage_inc(sci);
1190 		/* Creating a checkpoint */
1191 		err = nilfs_cpfile_create_checkpoint(nilfs->ns_cpfile,
1192 						     nilfs->ns_cno);
1193 		if (unlikely(err))
1194 			break;
1195 		fallthrough;
1196 	case NILFS_ST_CPFILE:
1197 		err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1198 					      &nilfs_sc_file_ops);
1199 		if (unlikely(err))
1200 			break;
1201 		nilfs_sc_cstage_inc(sci);
1202 		fallthrough;
1203 	case NILFS_ST_SUFILE:
1204 		err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1205 					 sci->sc_nfreesegs, &ndone);
1206 		if (unlikely(err)) {
1207 			nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1208 						  sci->sc_freesegs, ndone,
1209 						  NULL);
1210 			break;
1211 		}
1212 		sci->sc_stage.flags |= NILFS_CF_SUFREED;
1213 
1214 		err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1215 					      &nilfs_sc_file_ops);
1216 		if (unlikely(err))
1217 			break;
1218 		nilfs_sc_cstage_inc(sci);
1219 		fallthrough;
1220 	case NILFS_ST_DAT:
1221  dat_stage:
1222 		err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1223 					      &nilfs_sc_dat_ops);
1224 		if (unlikely(err))
1225 			break;
1226 		if (mode == SC_FLUSH_DAT) {
1227 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1228 			return 0;
1229 		}
1230 		nilfs_sc_cstage_inc(sci);
1231 		fallthrough;
1232 	case NILFS_ST_SR:
1233 		if (mode == SC_LSEG_SR) {
1234 			/* Appending a super root */
1235 			err = nilfs_segctor_add_super_root(sci);
1236 			if (unlikely(err))
1237 				break;
1238 		}
1239 		/* End of a logical segment */
1240 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1241 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1242 		return 0;
1243 	case NILFS_ST_DSYNC:
1244  dsync_mode:
1245 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1246 		ii = sci->sc_dsync_inode;
1247 		if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1248 			break;
1249 
1250 		err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1251 		if (unlikely(err))
1252 			break;
1253 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1254 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1255 		return 0;
1256 	case NILFS_ST_DONE:
1257 		return 0;
1258 	default:
1259 		BUG();
1260 	}
1261 
1262  break_or_fail:
1263 	return err;
1264 }
1265 
1266 /**
1267  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1268  * @sci: nilfs_sc_info
1269  * @nilfs: nilfs object
1270  */
1271 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1272 					    struct the_nilfs *nilfs)
1273 {
1274 	struct nilfs_segment_buffer *segbuf, *prev;
1275 	__u64 nextnum;
1276 	int err, alloc = 0;
1277 
1278 	segbuf = nilfs_segbuf_new(sci->sc_super);
1279 	if (unlikely(!segbuf))
1280 		return -ENOMEM;
1281 
1282 	if (list_empty(&sci->sc_write_logs)) {
1283 		nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1284 				 nilfs->ns_pseg_offset, nilfs);
1285 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1286 			nilfs_shift_to_next_segment(nilfs);
1287 			nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1288 		}
1289 
1290 		segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1291 		nextnum = nilfs->ns_nextnum;
1292 
1293 		if (nilfs->ns_segnum == nilfs->ns_nextnum)
1294 			/* Start from the head of a new full segment */
1295 			alloc++;
1296 	} else {
1297 		/* Continue logs */
1298 		prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1299 		nilfs_segbuf_map_cont(segbuf, prev);
1300 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1301 		nextnum = prev->sb_nextnum;
1302 
1303 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1304 			nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1305 			segbuf->sb_sum.seg_seq++;
1306 			alloc++;
1307 		}
1308 	}
1309 
1310 	err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1311 	if (err)
1312 		goto failed;
1313 
1314 	if (alloc) {
1315 		err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1316 		if (err)
1317 			goto failed;
1318 	}
1319 	nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1320 
1321 	BUG_ON(!list_empty(&sci->sc_segbufs));
1322 	list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1323 	sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1324 	return 0;
1325 
1326  failed:
1327 	nilfs_segbuf_free(segbuf);
1328 	return err;
1329 }
1330 
1331 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1332 					 struct the_nilfs *nilfs, int nadd)
1333 {
1334 	struct nilfs_segment_buffer *segbuf, *prev;
1335 	struct inode *sufile = nilfs->ns_sufile;
1336 	__u64 nextnextnum;
1337 	LIST_HEAD(list);
1338 	int err, ret, i;
1339 
1340 	prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1341 	/*
1342 	 * Since the segment specified with nextnum might be allocated during
1343 	 * the previous construction, the buffer including its segusage may
1344 	 * not be dirty.  The following call ensures that the buffer is dirty
1345 	 * and will pin the buffer on memory until the sufile is written.
1346 	 */
1347 	err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1348 	if (unlikely(err))
1349 		return err;
1350 
1351 	for (i = 0; i < nadd; i++) {
1352 		/* extend segment info */
1353 		err = -ENOMEM;
1354 		segbuf = nilfs_segbuf_new(sci->sc_super);
1355 		if (unlikely(!segbuf))
1356 			goto failed;
1357 
1358 		/* map this buffer to region of segment on-disk */
1359 		nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1360 		sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1361 
1362 		/* allocate the next next full segment */
1363 		err = nilfs_sufile_alloc(sufile, &nextnextnum);
1364 		if (unlikely(err))
1365 			goto failed_segbuf;
1366 
1367 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1368 		nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1369 
1370 		list_add_tail(&segbuf->sb_list, &list);
1371 		prev = segbuf;
1372 	}
1373 	list_splice_tail(&list, &sci->sc_segbufs);
1374 	return 0;
1375 
1376  failed_segbuf:
1377 	nilfs_segbuf_free(segbuf);
1378  failed:
1379 	list_for_each_entry(segbuf, &list, sb_list) {
1380 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1381 		WARN_ON(ret); /* never fails */
1382 	}
1383 	nilfs_destroy_logs(&list);
1384 	return err;
1385 }
1386 
1387 static void nilfs_free_incomplete_logs(struct list_head *logs,
1388 				       struct the_nilfs *nilfs)
1389 {
1390 	struct nilfs_segment_buffer *segbuf, *prev;
1391 	struct inode *sufile = nilfs->ns_sufile;
1392 	int ret;
1393 
1394 	segbuf = NILFS_FIRST_SEGBUF(logs);
1395 	if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1396 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1397 		WARN_ON(ret); /* never fails */
1398 	}
1399 	if (atomic_read(&segbuf->sb_err)) {
1400 		/* Case 1: The first segment failed */
1401 		if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1402 			/*
1403 			 * Case 1a:  Partial segment appended into an existing
1404 			 * segment
1405 			 */
1406 			nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1407 						segbuf->sb_fseg_end);
1408 		else /* Case 1b:  New full segment */
1409 			set_nilfs_discontinued(nilfs);
1410 	}
1411 
1412 	prev = segbuf;
1413 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1414 		if (prev->sb_nextnum != segbuf->sb_nextnum) {
1415 			ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1416 			WARN_ON(ret); /* never fails */
1417 		}
1418 		if (atomic_read(&segbuf->sb_err) &&
1419 		    segbuf->sb_segnum != nilfs->ns_nextnum)
1420 			/* Case 2: extended segment (!= next) failed */
1421 			nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1422 		prev = segbuf;
1423 	}
1424 }
1425 
1426 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1427 					  struct inode *sufile)
1428 {
1429 	struct nilfs_segment_buffer *segbuf;
1430 	unsigned long live_blocks;
1431 	int ret;
1432 
1433 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1434 		live_blocks = segbuf->sb_sum.nblocks +
1435 			(segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1436 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1437 						     live_blocks,
1438 						     sci->sc_seg_ctime);
1439 		WARN_ON(ret); /* always succeed because the segusage is dirty */
1440 	}
1441 }
1442 
1443 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1444 {
1445 	struct nilfs_segment_buffer *segbuf;
1446 	int ret;
1447 
1448 	segbuf = NILFS_FIRST_SEGBUF(logs);
1449 	ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1450 					     segbuf->sb_pseg_start -
1451 					     segbuf->sb_fseg_start, 0);
1452 	WARN_ON(ret); /* always succeed because the segusage is dirty */
1453 
1454 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1455 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1456 						     0, 0);
1457 		WARN_ON(ret); /* always succeed */
1458 	}
1459 }
1460 
1461 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1462 					    struct nilfs_segment_buffer *last,
1463 					    struct inode *sufile)
1464 {
1465 	struct nilfs_segment_buffer *segbuf = last;
1466 	int ret;
1467 
1468 	list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1469 		sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1470 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1471 		WARN_ON(ret);
1472 	}
1473 	nilfs_truncate_logs(&sci->sc_segbufs, last);
1474 }
1475 
1476 
1477 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1478 				 struct the_nilfs *nilfs, int mode)
1479 {
1480 	struct nilfs_cstage prev_stage = sci->sc_stage;
1481 	int err, nadd = 1;
1482 
1483 	/* Collection retry loop */
1484 	for (;;) {
1485 		sci->sc_nblk_this_inc = 0;
1486 		sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1487 
1488 		err = nilfs_segctor_reset_segment_buffer(sci);
1489 		if (unlikely(err))
1490 			goto failed;
1491 
1492 		err = nilfs_segctor_collect_blocks(sci, mode);
1493 		sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1494 		if (!err)
1495 			break;
1496 
1497 		if (unlikely(err != -E2BIG))
1498 			goto failed;
1499 
1500 		/* The current segment is filled up */
1501 		if (mode != SC_LSEG_SR ||
1502 		    nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1503 			break;
1504 
1505 		nilfs_clear_logs(&sci->sc_segbufs);
1506 
1507 		if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1508 			err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1509 							sci->sc_freesegs,
1510 							sci->sc_nfreesegs,
1511 							NULL);
1512 			WARN_ON(err); /* do not happen */
1513 			sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1514 		}
1515 
1516 		err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1517 		if (unlikely(err))
1518 			return err;
1519 
1520 		nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1521 		sci->sc_stage = prev_stage;
1522 	}
1523 	nilfs_segctor_zeropad_segsum(sci);
1524 	nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1525 	return 0;
1526 
1527  failed:
1528 	return err;
1529 }
1530 
1531 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1532 				      struct buffer_head *new_bh)
1533 {
1534 	BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1535 
1536 	list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1537 	/* The caller must release old_bh */
1538 }
1539 
1540 static int
1541 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1542 				     struct nilfs_segment_buffer *segbuf,
1543 				     int mode)
1544 {
1545 	struct inode *inode = NULL;
1546 	sector_t blocknr;
1547 	unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1548 	unsigned long nblocks = 0, ndatablk = 0;
1549 	const struct nilfs_sc_operations *sc_op = NULL;
1550 	struct nilfs_segsum_pointer ssp;
1551 	struct nilfs_finfo *finfo = NULL;
1552 	union nilfs_binfo binfo;
1553 	struct buffer_head *bh, *bh_org;
1554 	ino_t ino = 0;
1555 	int err = 0;
1556 
1557 	if (!nfinfo)
1558 		goto out;
1559 
1560 	blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1561 	ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1562 	ssp.offset = sizeof(struct nilfs_segment_summary);
1563 
1564 	list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1565 		if (bh == segbuf->sb_super_root)
1566 			break;
1567 		if (!finfo) {
1568 			finfo =	nilfs_segctor_map_segsum_entry(
1569 				sci, &ssp, sizeof(*finfo));
1570 			ino = le64_to_cpu(finfo->fi_ino);
1571 			nblocks = le32_to_cpu(finfo->fi_nblocks);
1572 			ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1573 
1574 			inode = bh->b_folio->mapping->host;
1575 
1576 			if (mode == SC_LSEG_DSYNC)
1577 				sc_op = &nilfs_sc_dsync_ops;
1578 			else if (ino == NILFS_DAT_INO)
1579 				sc_op = &nilfs_sc_dat_ops;
1580 			else /* file blocks */
1581 				sc_op = &nilfs_sc_file_ops;
1582 		}
1583 		bh_org = bh;
1584 		get_bh(bh_org);
1585 		err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1586 					&binfo);
1587 		if (bh != bh_org)
1588 			nilfs_list_replace_buffer(bh_org, bh);
1589 		brelse(bh_org);
1590 		if (unlikely(err))
1591 			goto failed_bmap;
1592 
1593 		if (ndatablk > 0)
1594 			sc_op->write_data_binfo(sci, &ssp, &binfo);
1595 		else
1596 			sc_op->write_node_binfo(sci, &ssp, &binfo);
1597 
1598 		blocknr++;
1599 		if (--nblocks == 0) {
1600 			finfo = NULL;
1601 			if (--nfinfo == 0)
1602 				break;
1603 		} else if (ndatablk > 0)
1604 			ndatablk--;
1605 	}
1606  out:
1607 	return 0;
1608 
1609  failed_bmap:
1610 	return err;
1611 }
1612 
1613 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1614 {
1615 	struct nilfs_segment_buffer *segbuf;
1616 	int err;
1617 
1618 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1619 		err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1620 		if (unlikely(err))
1621 			return err;
1622 		nilfs_segbuf_fill_in_segsum(segbuf);
1623 	}
1624 	return 0;
1625 }
1626 
1627 static void nilfs_begin_folio_io(struct folio *folio)
1628 {
1629 	if (!folio || folio_test_writeback(folio))
1630 		/*
1631 		 * For split b-tree node pages, this function may be called
1632 		 * twice.  We ignore the 2nd or later calls by this check.
1633 		 */
1634 		return;
1635 
1636 	folio_lock(folio);
1637 	folio_clear_dirty_for_io(folio);
1638 	folio_start_writeback(folio);
1639 	folio_unlock(folio);
1640 }
1641 
1642 /**
1643  * nilfs_prepare_write_logs - prepare to write logs
1644  * @logs: logs to prepare for writing
1645  * @seed: checksum seed value
1646  *
1647  * nilfs_prepare_write_logs() adds checksums and prepares the block
1648  * buffers/folios for writing logs.  In order to stabilize folios of
1649  * memory-mapped file blocks by putting them in writeback state before
1650  * calculating the checksums, first prepare to write payload blocks other
1651  * than segment summary and super root blocks in which the checksums will
1652  * be embedded.
1653  */
1654 static void nilfs_prepare_write_logs(struct list_head *logs, u32 seed)
1655 {
1656 	struct nilfs_segment_buffer *segbuf;
1657 	struct folio *bd_folio = NULL, *fs_folio = NULL;
1658 	struct buffer_head *bh;
1659 
1660 	/* Prepare to write payload blocks */
1661 	list_for_each_entry(segbuf, logs, sb_list) {
1662 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1663 				    b_assoc_buffers) {
1664 			if (bh == segbuf->sb_super_root)
1665 				break;
1666 			set_buffer_async_write(bh);
1667 			if (bh->b_folio != fs_folio) {
1668 				nilfs_begin_folio_io(fs_folio);
1669 				fs_folio = bh->b_folio;
1670 			}
1671 		}
1672 	}
1673 	nilfs_begin_folio_io(fs_folio);
1674 
1675 	nilfs_add_checksums_on_logs(logs, seed);
1676 
1677 	/* Prepare to write segment summary blocks */
1678 	list_for_each_entry(segbuf, logs, sb_list) {
1679 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1680 				    b_assoc_buffers) {
1681 			mark_buffer_dirty(bh);
1682 			if (bh->b_folio == bd_folio)
1683 				continue;
1684 			if (bd_folio) {
1685 				folio_lock(bd_folio);
1686 				folio_wait_writeback(bd_folio);
1687 				folio_clear_dirty_for_io(bd_folio);
1688 				folio_start_writeback(bd_folio);
1689 				folio_unlock(bd_folio);
1690 			}
1691 			bd_folio = bh->b_folio;
1692 		}
1693 	}
1694 
1695 	/* Prepare to write super root block */
1696 	bh = NILFS_LAST_SEGBUF(logs)->sb_super_root;
1697 	if (bh) {
1698 		mark_buffer_dirty(bh);
1699 		if (bh->b_folio != bd_folio) {
1700 			folio_lock(bd_folio);
1701 			folio_wait_writeback(bd_folio);
1702 			folio_clear_dirty_for_io(bd_folio);
1703 			folio_start_writeback(bd_folio);
1704 			folio_unlock(bd_folio);
1705 			bd_folio = bh->b_folio;
1706 		}
1707 	}
1708 
1709 	if (bd_folio) {
1710 		folio_lock(bd_folio);
1711 		folio_wait_writeback(bd_folio);
1712 		folio_clear_dirty_for_io(bd_folio);
1713 		folio_start_writeback(bd_folio);
1714 		folio_unlock(bd_folio);
1715 	}
1716 }
1717 
1718 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1719 			       struct the_nilfs *nilfs)
1720 {
1721 	int ret;
1722 
1723 	ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1724 	list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1725 	return ret;
1726 }
1727 
1728 static void nilfs_end_folio_io(struct folio *folio, int err)
1729 {
1730 	if (!folio)
1731 		return;
1732 
1733 	if (buffer_nilfs_node(folio_buffers(folio)) &&
1734 			!folio_test_writeback(folio)) {
1735 		/*
1736 		 * For b-tree node pages, this function may be called twice
1737 		 * or more because they might be split in a segment.
1738 		 */
1739 		if (folio_test_dirty(folio)) {
1740 			/*
1741 			 * For pages holding split b-tree node buffers, dirty
1742 			 * flag on the buffers may be cleared discretely.
1743 			 * In that case, the page is once redirtied for
1744 			 * remaining buffers, and it must be cancelled if
1745 			 * all the buffers get cleaned later.
1746 			 */
1747 			folio_lock(folio);
1748 			if (nilfs_folio_buffers_clean(folio))
1749 				__nilfs_clear_folio_dirty(folio);
1750 			folio_unlock(folio);
1751 		}
1752 		return;
1753 	}
1754 
1755 	if (err || !nilfs_folio_buffers_clean(folio))
1756 		filemap_dirty_folio(folio->mapping, folio);
1757 
1758 	folio_end_writeback(folio);
1759 }
1760 
1761 static void nilfs_abort_logs(struct list_head *logs, int err)
1762 {
1763 	struct nilfs_segment_buffer *segbuf;
1764 	struct folio *bd_folio = NULL, *fs_folio = NULL;
1765 	struct buffer_head *bh;
1766 
1767 	if (list_empty(logs))
1768 		return;
1769 
1770 	list_for_each_entry(segbuf, logs, sb_list) {
1771 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1772 				    b_assoc_buffers) {
1773 			clear_buffer_uptodate(bh);
1774 			if (bh->b_folio != bd_folio) {
1775 				if (bd_folio)
1776 					folio_end_writeback(bd_folio);
1777 				bd_folio = bh->b_folio;
1778 			}
1779 		}
1780 
1781 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1782 				    b_assoc_buffers) {
1783 			if (bh == segbuf->sb_super_root) {
1784 				clear_buffer_uptodate(bh);
1785 				if (bh->b_folio != bd_folio) {
1786 					folio_end_writeback(bd_folio);
1787 					bd_folio = bh->b_folio;
1788 				}
1789 				break;
1790 			}
1791 			clear_buffer_async_write(bh);
1792 			if (bh->b_folio != fs_folio) {
1793 				nilfs_end_folio_io(fs_folio, err);
1794 				fs_folio = bh->b_folio;
1795 			}
1796 		}
1797 	}
1798 	if (bd_folio)
1799 		folio_end_writeback(bd_folio);
1800 
1801 	nilfs_end_folio_io(fs_folio, err);
1802 }
1803 
1804 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1805 					     struct the_nilfs *nilfs, int err)
1806 {
1807 	LIST_HEAD(logs);
1808 	int ret;
1809 
1810 	list_splice_tail_init(&sci->sc_write_logs, &logs);
1811 	ret = nilfs_wait_on_logs(&logs);
1812 	nilfs_abort_logs(&logs, ret ? : err);
1813 
1814 	list_splice_tail_init(&sci->sc_segbufs, &logs);
1815 	nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1816 	nilfs_free_incomplete_logs(&logs, nilfs);
1817 
1818 	if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1819 		ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1820 						sci->sc_freesegs,
1821 						sci->sc_nfreesegs,
1822 						NULL);
1823 		WARN_ON(ret); /* do not happen */
1824 	}
1825 
1826 	nilfs_destroy_logs(&logs);
1827 }
1828 
1829 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1830 				   struct nilfs_segment_buffer *segbuf)
1831 {
1832 	nilfs->ns_segnum = segbuf->sb_segnum;
1833 	nilfs->ns_nextnum = segbuf->sb_nextnum;
1834 	nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1835 		+ segbuf->sb_sum.nblocks;
1836 	nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1837 	nilfs->ns_ctime = segbuf->sb_sum.ctime;
1838 }
1839 
1840 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1841 {
1842 	struct nilfs_segment_buffer *segbuf;
1843 	struct folio *bd_folio = NULL, *fs_folio = NULL;
1844 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1845 	int update_sr = false;
1846 
1847 	list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1848 		struct buffer_head *bh;
1849 
1850 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1851 				    b_assoc_buffers) {
1852 			set_buffer_uptodate(bh);
1853 			clear_buffer_dirty(bh);
1854 			if (bh->b_folio != bd_folio) {
1855 				if (bd_folio)
1856 					folio_end_writeback(bd_folio);
1857 				bd_folio = bh->b_folio;
1858 			}
1859 		}
1860 		/*
1861 		 * We assume that the buffers which belong to the same folio
1862 		 * continue over the buffer list.
1863 		 * Under this assumption, the last BHs of folios is
1864 		 * identifiable by the discontinuity of bh->b_folio
1865 		 * (folio != fs_folio).
1866 		 *
1867 		 * For B-tree node blocks, however, this assumption is not
1868 		 * guaranteed.  The cleanup code of B-tree node folios needs
1869 		 * special care.
1870 		 */
1871 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1872 				    b_assoc_buffers) {
1873 			const unsigned long set_bits = BIT(BH_Uptodate);
1874 			const unsigned long clear_bits =
1875 				(BIT(BH_Dirty) | BIT(BH_Async_Write) |
1876 				 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1877 				 BIT(BH_NILFS_Redirected));
1878 
1879 			if (bh == segbuf->sb_super_root) {
1880 				set_buffer_uptodate(bh);
1881 				clear_buffer_dirty(bh);
1882 				if (bh->b_folio != bd_folio) {
1883 					folio_end_writeback(bd_folio);
1884 					bd_folio = bh->b_folio;
1885 				}
1886 				update_sr = true;
1887 				break;
1888 			}
1889 			set_mask_bits(&bh->b_state, clear_bits, set_bits);
1890 			if (bh->b_folio != fs_folio) {
1891 				nilfs_end_folio_io(fs_folio, 0);
1892 				fs_folio = bh->b_folio;
1893 			}
1894 		}
1895 
1896 		if (!nilfs_segbuf_simplex(segbuf)) {
1897 			if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1898 				set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1899 				sci->sc_lseg_stime = jiffies;
1900 			}
1901 			if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1902 				clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1903 		}
1904 	}
1905 	/*
1906 	 * Since folios may continue over multiple segment buffers,
1907 	 * end of the last folio must be checked outside of the loop.
1908 	 */
1909 	if (bd_folio)
1910 		folio_end_writeback(bd_folio);
1911 
1912 	nilfs_end_folio_io(fs_folio, 0);
1913 
1914 	nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1915 
1916 	if (nilfs_doing_gc())
1917 		nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1918 	else
1919 		nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1920 
1921 	sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1922 
1923 	segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1924 	nilfs_set_next_segment(nilfs, segbuf);
1925 
1926 	if (update_sr) {
1927 		nilfs->ns_flushed_device = 0;
1928 		nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1929 				       segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1930 
1931 		clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1932 		clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1933 		set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1934 		nilfs_segctor_clear_metadata_dirty(sci);
1935 	} else
1936 		clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1937 }
1938 
1939 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1940 {
1941 	int ret;
1942 
1943 	ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1944 	if (!ret) {
1945 		nilfs_segctor_complete_write(sci);
1946 		nilfs_destroy_logs(&sci->sc_write_logs);
1947 	}
1948 	return ret;
1949 }
1950 
1951 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1952 					     struct the_nilfs *nilfs)
1953 {
1954 	struct nilfs_inode_info *ii, *n;
1955 	struct inode *ifile = sci->sc_root->ifile;
1956 
1957 	spin_lock(&nilfs->ns_inode_lock);
1958  retry:
1959 	list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1960 		if (!ii->i_bh) {
1961 			struct buffer_head *ibh;
1962 			int err;
1963 
1964 			spin_unlock(&nilfs->ns_inode_lock);
1965 			err = nilfs_ifile_get_inode_block(
1966 				ifile, ii->vfs_inode.i_ino, &ibh);
1967 			if (unlikely(err)) {
1968 				nilfs_warn(sci->sc_super,
1969 					   "log writer: error %d getting inode block (ino=%lu)",
1970 					   err, ii->vfs_inode.i_ino);
1971 				return err;
1972 			}
1973 			spin_lock(&nilfs->ns_inode_lock);
1974 			if (likely(!ii->i_bh))
1975 				ii->i_bh = ibh;
1976 			else
1977 				brelse(ibh);
1978 			goto retry;
1979 		}
1980 
1981 		// Always redirty the buffer to avoid race condition
1982 		mark_buffer_dirty(ii->i_bh);
1983 		nilfs_mdt_mark_dirty(ifile);
1984 
1985 		clear_bit(NILFS_I_QUEUED, &ii->i_state);
1986 		set_bit(NILFS_I_BUSY, &ii->i_state);
1987 		list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1988 	}
1989 	spin_unlock(&nilfs->ns_inode_lock);
1990 
1991 	return 0;
1992 }
1993 
1994 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1995 					     struct the_nilfs *nilfs)
1996 {
1997 	struct nilfs_inode_info *ii, *n;
1998 	int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1999 	int defer_iput = false;
2000 
2001 	spin_lock(&nilfs->ns_inode_lock);
2002 	list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2003 		if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2004 		    test_bit(NILFS_I_DIRTY, &ii->i_state))
2005 			continue;
2006 
2007 		clear_bit(NILFS_I_BUSY, &ii->i_state);
2008 		brelse(ii->i_bh);
2009 		ii->i_bh = NULL;
2010 		list_del_init(&ii->i_dirty);
2011 		if (!ii->vfs_inode.i_nlink || during_mount) {
2012 			/*
2013 			 * Defer calling iput() to avoid deadlocks if
2014 			 * i_nlink == 0 or mount is not yet finished.
2015 			 */
2016 			list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2017 			defer_iput = true;
2018 		} else {
2019 			spin_unlock(&nilfs->ns_inode_lock);
2020 			iput(&ii->vfs_inode);
2021 			spin_lock(&nilfs->ns_inode_lock);
2022 		}
2023 	}
2024 	spin_unlock(&nilfs->ns_inode_lock);
2025 
2026 	if (defer_iput)
2027 		schedule_work(&sci->sc_iput_work);
2028 }
2029 
2030 /*
2031  * Main procedure of segment constructor
2032  */
2033 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2034 {
2035 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2036 	int err;
2037 
2038 	if (sb_rdonly(sci->sc_super))
2039 		return -EROFS;
2040 
2041 	nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2042 	sci->sc_cno = nilfs->ns_cno;
2043 
2044 	err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2045 	if (unlikely(err))
2046 		goto out;
2047 
2048 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2049 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2050 
2051 	if (nilfs_segctor_clean(sci))
2052 		goto out;
2053 
2054 	do {
2055 		sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2056 
2057 		err = nilfs_segctor_begin_construction(sci, nilfs);
2058 		if (unlikely(err))
2059 			goto out;
2060 
2061 		/* Update time stamp */
2062 		sci->sc_seg_ctime = ktime_get_real_seconds();
2063 
2064 		err = nilfs_segctor_collect(sci, nilfs, mode);
2065 		if (unlikely(err))
2066 			goto failed;
2067 
2068 		/* Avoid empty segment */
2069 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2070 		    nilfs_segbuf_empty(sci->sc_curseg)) {
2071 			nilfs_segctor_abort_construction(sci, nilfs, 1);
2072 			goto out;
2073 		}
2074 
2075 		err = nilfs_segctor_assign(sci, mode);
2076 		if (unlikely(err))
2077 			goto failed;
2078 
2079 		if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2080 			nilfs_segctor_fill_in_file_bmap(sci);
2081 
2082 		if (mode == SC_LSEG_SR &&
2083 		    nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2084 			err = nilfs_cpfile_finalize_checkpoint(
2085 				nilfs->ns_cpfile, nilfs->ns_cno, sci->sc_root,
2086 				sci->sc_nblk_inc + sci->sc_nblk_this_inc,
2087 				sci->sc_seg_ctime,
2088 				!test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags));
2089 			if (unlikely(err))
2090 				goto failed_to_write;
2091 
2092 			nilfs_segctor_fill_in_super_root(sci, nilfs);
2093 		}
2094 		nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2095 
2096 		/* Write partial segments */
2097 		nilfs_prepare_write_logs(&sci->sc_segbufs, nilfs->ns_crc_seed);
2098 
2099 		err = nilfs_segctor_write(sci, nilfs);
2100 		if (unlikely(err))
2101 			goto failed_to_write;
2102 
2103 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2104 		    nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2105 			/*
2106 			 * At this point, we avoid double buffering
2107 			 * for blocksize < pagesize because page dirty
2108 			 * flag is turned off during write and dirty
2109 			 * buffers are not properly collected for
2110 			 * pages crossing over segments.
2111 			 */
2112 			err = nilfs_segctor_wait(sci);
2113 			if (err)
2114 				goto failed_to_write;
2115 		}
2116 	} while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2117 
2118  out:
2119 	nilfs_segctor_drop_written_files(sci, nilfs);
2120 	return err;
2121 
2122  failed_to_write:
2123 	if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2124 		nilfs_redirty_inodes(&sci->sc_dirty_files);
2125 
2126  failed:
2127 	if (nilfs_doing_gc())
2128 		nilfs_redirty_inodes(&sci->sc_gc_inodes);
2129 	nilfs_segctor_abort_construction(sci, nilfs, err);
2130 	goto out;
2131 }
2132 
2133 /**
2134  * nilfs_segctor_start_timer - set timer of background write
2135  * @sci: nilfs_sc_info
2136  *
2137  * If the timer has already been set, it ignores the new request.
2138  * This function MUST be called within a section locking the segment
2139  * semaphore.
2140  */
2141 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2142 {
2143 	spin_lock(&sci->sc_state_lock);
2144 	if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2145 		if (sci->sc_task) {
2146 			sci->sc_timer.expires = jiffies + sci->sc_interval;
2147 			add_timer(&sci->sc_timer);
2148 		}
2149 		sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2150 	}
2151 	spin_unlock(&sci->sc_state_lock);
2152 }
2153 
2154 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2155 {
2156 	spin_lock(&sci->sc_state_lock);
2157 	if (!(sci->sc_flush_request & BIT(bn))) {
2158 		unsigned long prev_req = sci->sc_flush_request;
2159 
2160 		sci->sc_flush_request |= BIT(bn);
2161 		if (!prev_req)
2162 			wake_up(&sci->sc_wait_daemon);
2163 	}
2164 	spin_unlock(&sci->sc_state_lock);
2165 }
2166 
2167 /**
2168  * nilfs_flush_segment - trigger a segment construction for resource control
2169  * @sb: super block
2170  * @ino: inode number of the file to be flushed out.
2171  */
2172 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2173 {
2174 	struct the_nilfs *nilfs = sb->s_fs_info;
2175 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2176 
2177 	if (!sci || nilfs_doing_construction())
2178 		return;
2179 	nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2180 					/* assign bit 0 to data files */
2181 }
2182 
2183 struct nilfs_segctor_wait_request {
2184 	wait_queue_entry_t	wq;
2185 	__u32		seq;
2186 	int		err;
2187 	atomic_t	done;
2188 };
2189 
2190 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2191 {
2192 	struct nilfs_segctor_wait_request wait_req;
2193 	int err = 0;
2194 
2195 	init_wait(&wait_req.wq);
2196 	wait_req.err = 0;
2197 	atomic_set(&wait_req.done, 0);
2198 	init_waitqueue_entry(&wait_req.wq, current);
2199 
2200 	/*
2201 	 * To prevent a race issue where completion notifications from the
2202 	 * log writer thread are missed, increment the request sequence count
2203 	 * "sc_seq_request" and insert a wait queue entry using the current
2204 	 * sequence number into the "sc_wait_request" queue at the same time
2205 	 * within the lock section of "sc_state_lock".
2206 	 */
2207 	spin_lock(&sci->sc_state_lock);
2208 	wait_req.seq = ++sci->sc_seq_request;
2209 	add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2210 	spin_unlock(&sci->sc_state_lock);
2211 
2212 	wake_up(&sci->sc_wait_daemon);
2213 
2214 	for (;;) {
2215 		set_current_state(TASK_INTERRUPTIBLE);
2216 
2217 		/*
2218 		 * Synchronize only while the log writer thread is alive.
2219 		 * Leave flushing out after the log writer thread exits to
2220 		 * the cleanup work in nilfs_segctor_destroy().
2221 		 */
2222 		if (!sci->sc_task)
2223 			break;
2224 
2225 		if (atomic_read(&wait_req.done)) {
2226 			err = wait_req.err;
2227 			break;
2228 		}
2229 		if (!signal_pending(current)) {
2230 			schedule();
2231 			continue;
2232 		}
2233 		err = -ERESTARTSYS;
2234 		break;
2235 	}
2236 	finish_wait(&sci->sc_wait_request, &wait_req.wq);
2237 	return err;
2238 }
2239 
2240 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err, bool force)
2241 {
2242 	struct nilfs_segctor_wait_request *wrq, *n;
2243 	unsigned long flags;
2244 
2245 	spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2246 	list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2247 		if (!atomic_read(&wrq->done) &&
2248 		    (force || nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq))) {
2249 			wrq->err = err;
2250 			atomic_set(&wrq->done, 1);
2251 		}
2252 		if (atomic_read(&wrq->done)) {
2253 			wrq->wq.func(&wrq->wq,
2254 				     TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2255 				     0, NULL);
2256 		}
2257 	}
2258 	spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2259 }
2260 
2261 /**
2262  * nilfs_construct_segment - construct a logical segment
2263  * @sb: super block
2264  *
2265  * Return Value: On success, 0 is returned. On errors, one of the following
2266  * negative error code is returned.
2267  *
2268  * %-EROFS - Read only filesystem.
2269  *
2270  * %-EIO - I/O error
2271  *
2272  * %-ENOSPC - No space left on device (only in a panic state).
2273  *
2274  * %-ERESTARTSYS - Interrupted.
2275  *
2276  * %-ENOMEM - Insufficient memory available.
2277  */
2278 int nilfs_construct_segment(struct super_block *sb)
2279 {
2280 	struct the_nilfs *nilfs = sb->s_fs_info;
2281 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2282 	struct nilfs_transaction_info *ti;
2283 
2284 	if (sb_rdonly(sb) || unlikely(!sci))
2285 		return -EROFS;
2286 
2287 	/* A call inside transactions causes a deadlock. */
2288 	BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2289 
2290 	return nilfs_segctor_sync(sci);
2291 }
2292 
2293 /**
2294  * nilfs_construct_dsync_segment - construct a data-only logical segment
2295  * @sb: super block
2296  * @inode: inode whose data blocks should be written out
2297  * @start: start byte offset
2298  * @end: end byte offset (inclusive)
2299  *
2300  * Return Value: On success, 0 is returned. On errors, one of the following
2301  * negative error code is returned.
2302  *
2303  * %-EROFS - Read only filesystem.
2304  *
2305  * %-EIO - I/O error
2306  *
2307  * %-ENOSPC - No space left on device (only in a panic state).
2308  *
2309  * %-ERESTARTSYS - Interrupted.
2310  *
2311  * %-ENOMEM - Insufficient memory available.
2312  */
2313 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2314 				  loff_t start, loff_t end)
2315 {
2316 	struct the_nilfs *nilfs = sb->s_fs_info;
2317 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2318 	struct nilfs_inode_info *ii;
2319 	struct nilfs_transaction_info ti;
2320 	int err = 0;
2321 
2322 	if (sb_rdonly(sb) || unlikely(!sci))
2323 		return -EROFS;
2324 
2325 	nilfs_transaction_lock(sb, &ti, 0);
2326 
2327 	ii = NILFS_I(inode);
2328 	if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2329 	    nilfs_test_opt(nilfs, STRICT_ORDER) ||
2330 	    test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2331 	    nilfs_discontinued(nilfs)) {
2332 		nilfs_transaction_unlock(sb);
2333 		err = nilfs_segctor_sync(sci);
2334 		return err;
2335 	}
2336 
2337 	spin_lock(&nilfs->ns_inode_lock);
2338 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2339 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2340 		spin_unlock(&nilfs->ns_inode_lock);
2341 		nilfs_transaction_unlock(sb);
2342 		return 0;
2343 	}
2344 	spin_unlock(&nilfs->ns_inode_lock);
2345 	sci->sc_dsync_inode = ii;
2346 	sci->sc_dsync_start = start;
2347 	sci->sc_dsync_end = end;
2348 
2349 	err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2350 	if (!err)
2351 		nilfs->ns_flushed_device = 0;
2352 
2353 	nilfs_transaction_unlock(sb);
2354 	return err;
2355 }
2356 
2357 #define FLUSH_FILE_BIT	(0x1) /* data file only */
2358 #define FLUSH_DAT_BIT	BIT(NILFS_DAT_INO) /* DAT only */
2359 
2360 /**
2361  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2362  * @sci: segment constructor object
2363  */
2364 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2365 {
2366 	bool thread_is_alive;
2367 
2368 	spin_lock(&sci->sc_state_lock);
2369 	sci->sc_seq_accepted = sci->sc_seq_request;
2370 	thread_is_alive = (bool)sci->sc_task;
2371 	spin_unlock(&sci->sc_state_lock);
2372 
2373 	/*
2374 	 * This function does not race with the log writer thread's
2375 	 * termination.  Therefore, deleting sc_timer, which should not be
2376 	 * done after the log writer thread exits, can be done safely outside
2377 	 * the area protected by sc_state_lock.
2378 	 */
2379 	if (thread_is_alive)
2380 		del_timer_sync(&sci->sc_timer);
2381 }
2382 
2383 /**
2384  * nilfs_segctor_notify - notify the result of request to caller threads
2385  * @sci: segment constructor object
2386  * @mode: mode of log forming
2387  * @err: error code to be notified
2388  */
2389 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2390 {
2391 	/* Clear requests (even when the construction failed) */
2392 	spin_lock(&sci->sc_state_lock);
2393 
2394 	if (mode == SC_LSEG_SR) {
2395 		sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2396 		sci->sc_seq_done = sci->sc_seq_accepted;
2397 		nilfs_segctor_wakeup(sci, err, false);
2398 		sci->sc_flush_request = 0;
2399 	} else {
2400 		if (mode == SC_FLUSH_FILE)
2401 			sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2402 		else if (mode == SC_FLUSH_DAT)
2403 			sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2404 
2405 		/* re-enable timer if checkpoint creation was not done */
2406 		if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) && sci->sc_task &&
2407 		    time_before(jiffies, sci->sc_timer.expires))
2408 			add_timer(&sci->sc_timer);
2409 	}
2410 	spin_unlock(&sci->sc_state_lock);
2411 }
2412 
2413 /**
2414  * nilfs_segctor_construct - form logs and write them to disk
2415  * @sci: segment constructor object
2416  * @mode: mode of log forming
2417  */
2418 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2419 {
2420 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2421 	struct nilfs_super_block **sbp;
2422 	int err = 0;
2423 
2424 	nilfs_segctor_accept(sci);
2425 
2426 	if (nilfs_discontinued(nilfs))
2427 		mode = SC_LSEG_SR;
2428 	if (!nilfs_segctor_confirm(sci))
2429 		err = nilfs_segctor_do_construct(sci, mode);
2430 
2431 	if (likely(!err)) {
2432 		if (mode != SC_FLUSH_DAT)
2433 			atomic_set(&nilfs->ns_ndirtyblks, 0);
2434 		if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2435 		    nilfs_discontinued(nilfs)) {
2436 			down_write(&nilfs->ns_sem);
2437 			err = -EIO;
2438 			sbp = nilfs_prepare_super(sci->sc_super,
2439 						  nilfs_sb_will_flip(nilfs));
2440 			if (likely(sbp)) {
2441 				nilfs_set_log_cursor(sbp[0], nilfs);
2442 				err = nilfs_commit_super(sci->sc_super,
2443 							 NILFS_SB_COMMIT);
2444 			}
2445 			up_write(&nilfs->ns_sem);
2446 		}
2447 	}
2448 
2449 	nilfs_segctor_notify(sci, mode, err);
2450 	return err;
2451 }
2452 
2453 static void nilfs_construction_timeout(struct timer_list *t)
2454 {
2455 	struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2456 
2457 	wake_up_process(sci->sc_timer_task);
2458 }
2459 
2460 static void
2461 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2462 {
2463 	struct nilfs_inode_info *ii, *n;
2464 
2465 	list_for_each_entry_safe(ii, n, head, i_dirty) {
2466 		if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2467 			continue;
2468 		list_del_init(&ii->i_dirty);
2469 		truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2470 		nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2471 		iput(&ii->vfs_inode);
2472 	}
2473 }
2474 
2475 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2476 			 void **kbufs)
2477 {
2478 	struct the_nilfs *nilfs = sb->s_fs_info;
2479 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2480 	struct nilfs_transaction_info ti;
2481 	int err;
2482 
2483 	if (unlikely(!sci))
2484 		return -EROFS;
2485 
2486 	nilfs_transaction_lock(sb, &ti, 1);
2487 
2488 	err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2489 	if (unlikely(err))
2490 		goto out_unlock;
2491 
2492 	err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2493 	if (unlikely(err)) {
2494 		nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2495 		goto out_unlock;
2496 	}
2497 
2498 	sci->sc_freesegs = kbufs[4];
2499 	sci->sc_nfreesegs = argv[4].v_nmembs;
2500 	list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2501 
2502 	for (;;) {
2503 		err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2504 		nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2505 
2506 		if (likely(!err))
2507 			break;
2508 
2509 		nilfs_warn(sb, "error %d cleaning segments", err);
2510 		set_current_state(TASK_INTERRUPTIBLE);
2511 		schedule_timeout(sci->sc_interval);
2512 	}
2513 	if (nilfs_test_opt(nilfs, DISCARD)) {
2514 		int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2515 						 sci->sc_nfreesegs);
2516 		if (ret) {
2517 			nilfs_warn(sb,
2518 				   "error %d on discard request, turning discards off for the device",
2519 				   ret);
2520 			nilfs_clear_opt(nilfs, DISCARD);
2521 		}
2522 	}
2523 
2524  out_unlock:
2525 	sci->sc_freesegs = NULL;
2526 	sci->sc_nfreesegs = 0;
2527 	nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2528 	nilfs_transaction_unlock(sb);
2529 	return err;
2530 }
2531 
2532 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2533 {
2534 	struct nilfs_transaction_info ti;
2535 
2536 	nilfs_transaction_lock(sci->sc_super, &ti, 0);
2537 	nilfs_segctor_construct(sci, mode);
2538 
2539 	/*
2540 	 * Unclosed segment should be retried.  We do this using sc_timer.
2541 	 * Timeout of sc_timer will invoke complete construction which leads
2542 	 * to close the current logical segment.
2543 	 */
2544 	if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2545 		nilfs_segctor_start_timer(sci);
2546 
2547 	nilfs_transaction_unlock(sci->sc_super);
2548 }
2549 
2550 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2551 {
2552 	int mode = 0;
2553 
2554 	spin_lock(&sci->sc_state_lock);
2555 	mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2556 		SC_FLUSH_DAT : SC_FLUSH_FILE;
2557 	spin_unlock(&sci->sc_state_lock);
2558 
2559 	if (mode) {
2560 		nilfs_segctor_do_construct(sci, mode);
2561 
2562 		spin_lock(&sci->sc_state_lock);
2563 		sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2564 			~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2565 		spin_unlock(&sci->sc_state_lock);
2566 	}
2567 	clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2568 }
2569 
2570 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2571 {
2572 	if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2573 	    time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2574 		if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2575 			return SC_FLUSH_FILE;
2576 		else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2577 			return SC_FLUSH_DAT;
2578 	}
2579 	return SC_LSEG_SR;
2580 }
2581 
2582 /**
2583  * nilfs_segctor_thread - main loop of the segment constructor thread.
2584  * @arg: pointer to a struct nilfs_sc_info.
2585  *
2586  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2587  * to execute segment constructions.
2588  */
2589 static int nilfs_segctor_thread(void *arg)
2590 {
2591 	struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2592 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2593 	int timeout = 0;
2594 
2595 	sci->sc_timer_task = current;
2596 	timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2597 
2598 	/* start sync. */
2599 	sci->sc_task = current;
2600 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2601 	nilfs_info(sci->sc_super,
2602 		   "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2603 		   sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2604 
2605 	set_freezable();
2606 	spin_lock(&sci->sc_state_lock);
2607  loop:
2608 	for (;;) {
2609 		int mode;
2610 
2611 		if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2612 			goto end_thread;
2613 
2614 		if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2615 			mode = SC_LSEG_SR;
2616 		else if (sci->sc_flush_request)
2617 			mode = nilfs_segctor_flush_mode(sci);
2618 		else
2619 			break;
2620 
2621 		spin_unlock(&sci->sc_state_lock);
2622 		nilfs_segctor_thread_construct(sci, mode);
2623 		spin_lock(&sci->sc_state_lock);
2624 		timeout = 0;
2625 	}
2626 
2627 
2628 	if (freezing(current)) {
2629 		spin_unlock(&sci->sc_state_lock);
2630 		try_to_freeze();
2631 		spin_lock(&sci->sc_state_lock);
2632 	} else {
2633 		DEFINE_WAIT(wait);
2634 		int should_sleep = 1;
2635 
2636 		prepare_to_wait(&sci->sc_wait_daemon, &wait,
2637 				TASK_INTERRUPTIBLE);
2638 
2639 		if (sci->sc_seq_request != sci->sc_seq_done)
2640 			should_sleep = 0;
2641 		else if (sci->sc_flush_request)
2642 			should_sleep = 0;
2643 		else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2644 			should_sleep = time_before(jiffies,
2645 					sci->sc_timer.expires);
2646 
2647 		if (should_sleep) {
2648 			spin_unlock(&sci->sc_state_lock);
2649 			schedule();
2650 			spin_lock(&sci->sc_state_lock);
2651 		}
2652 		finish_wait(&sci->sc_wait_daemon, &wait);
2653 		timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2654 			   time_after_eq(jiffies, sci->sc_timer.expires));
2655 
2656 		if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2657 			set_nilfs_discontinued(nilfs);
2658 	}
2659 	goto loop;
2660 
2661  end_thread:
2662 	/* end sync. */
2663 	sci->sc_task = NULL;
2664 	timer_shutdown_sync(&sci->sc_timer);
2665 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2666 	spin_unlock(&sci->sc_state_lock);
2667 	return 0;
2668 }
2669 
2670 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2671 {
2672 	struct task_struct *t;
2673 
2674 	t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2675 	if (IS_ERR(t)) {
2676 		int err = PTR_ERR(t);
2677 
2678 		nilfs_err(sci->sc_super, "error %d creating segctord thread",
2679 			  err);
2680 		return err;
2681 	}
2682 	wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2683 	return 0;
2684 }
2685 
2686 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2687 	__acquires(&sci->sc_state_lock)
2688 	__releases(&sci->sc_state_lock)
2689 {
2690 	sci->sc_state |= NILFS_SEGCTOR_QUIT;
2691 
2692 	while (sci->sc_task) {
2693 		wake_up(&sci->sc_wait_daemon);
2694 		spin_unlock(&sci->sc_state_lock);
2695 		wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2696 		spin_lock(&sci->sc_state_lock);
2697 	}
2698 }
2699 
2700 /*
2701  * Setup & clean-up functions
2702  */
2703 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2704 					       struct nilfs_root *root)
2705 {
2706 	struct the_nilfs *nilfs = sb->s_fs_info;
2707 	struct nilfs_sc_info *sci;
2708 
2709 	sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2710 	if (!sci)
2711 		return NULL;
2712 
2713 	sci->sc_super = sb;
2714 
2715 	nilfs_get_root(root);
2716 	sci->sc_root = root;
2717 
2718 	init_waitqueue_head(&sci->sc_wait_request);
2719 	init_waitqueue_head(&sci->sc_wait_daemon);
2720 	init_waitqueue_head(&sci->sc_wait_task);
2721 	spin_lock_init(&sci->sc_state_lock);
2722 	INIT_LIST_HEAD(&sci->sc_dirty_files);
2723 	INIT_LIST_HEAD(&sci->sc_segbufs);
2724 	INIT_LIST_HEAD(&sci->sc_write_logs);
2725 	INIT_LIST_HEAD(&sci->sc_gc_inodes);
2726 	INIT_LIST_HEAD(&sci->sc_iput_queue);
2727 	INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2728 
2729 	sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2730 	sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2731 	sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2732 
2733 	if (nilfs->ns_interval)
2734 		sci->sc_interval = HZ * nilfs->ns_interval;
2735 	if (nilfs->ns_watermark)
2736 		sci->sc_watermark = nilfs->ns_watermark;
2737 	return sci;
2738 }
2739 
2740 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2741 {
2742 	int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2743 
2744 	/*
2745 	 * The segctord thread was stopped and its timer was removed.
2746 	 * But some tasks remain.
2747 	 */
2748 	do {
2749 		struct nilfs_transaction_info ti;
2750 
2751 		nilfs_transaction_lock(sci->sc_super, &ti, 0);
2752 		ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2753 		nilfs_transaction_unlock(sci->sc_super);
2754 
2755 		flush_work(&sci->sc_iput_work);
2756 
2757 	} while (ret && ret != -EROFS && retrycount-- > 0);
2758 }
2759 
2760 /**
2761  * nilfs_segctor_destroy - destroy the segment constructor.
2762  * @sci: nilfs_sc_info
2763  *
2764  * nilfs_segctor_destroy() kills the segctord thread and frees
2765  * the nilfs_sc_info struct.
2766  * Caller must hold the segment semaphore.
2767  */
2768 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2769 {
2770 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2771 	int flag;
2772 
2773 	up_write(&nilfs->ns_segctor_sem);
2774 
2775 	spin_lock(&sci->sc_state_lock);
2776 	nilfs_segctor_kill_thread(sci);
2777 	flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2778 		|| sci->sc_seq_request != sci->sc_seq_done);
2779 	spin_unlock(&sci->sc_state_lock);
2780 
2781 	/*
2782 	 * Forcibly wake up tasks waiting in nilfs_segctor_sync(), which can
2783 	 * be called from delayed iput() via nilfs_evict_inode() and can race
2784 	 * with the above log writer thread termination.
2785 	 */
2786 	nilfs_segctor_wakeup(sci, 0, true);
2787 
2788 	if (flush_work(&sci->sc_iput_work))
2789 		flag = true;
2790 
2791 	if (flag || !nilfs_segctor_confirm(sci))
2792 		nilfs_segctor_write_out(sci);
2793 
2794 	if (!list_empty(&sci->sc_dirty_files)) {
2795 		nilfs_warn(sci->sc_super,
2796 			   "disposed unprocessed dirty file(s) when stopping log writer");
2797 		nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2798 	}
2799 
2800 	if (!list_empty(&sci->sc_iput_queue)) {
2801 		nilfs_warn(sci->sc_super,
2802 			   "disposed unprocessed inode(s) in iput queue when stopping log writer");
2803 		nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2804 	}
2805 
2806 	WARN_ON(!list_empty(&sci->sc_segbufs));
2807 	WARN_ON(!list_empty(&sci->sc_write_logs));
2808 
2809 	nilfs_put_root(sci->sc_root);
2810 
2811 	down_write(&nilfs->ns_segctor_sem);
2812 
2813 	kfree(sci);
2814 }
2815 
2816 /**
2817  * nilfs_attach_log_writer - attach log writer
2818  * @sb: super block instance
2819  * @root: root object of the current filesystem tree
2820  *
2821  * This allocates a log writer object, initializes it, and starts the
2822  * log writer.
2823  *
2824  * Return Value: On success, 0 is returned. On error, one of the following
2825  * negative error code is returned.
2826  *
2827  * %-ENOMEM - Insufficient memory available.
2828  */
2829 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2830 {
2831 	struct the_nilfs *nilfs = sb->s_fs_info;
2832 	int err;
2833 
2834 	if (nilfs->ns_writer) {
2835 		/*
2836 		 * This happens if the filesystem is made read-only by
2837 		 * __nilfs_error or nilfs_remount and then remounted
2838 		 * read/write.  In these cases, reuse the existing
2839 		 * writer.
2840 		 */
2841 		return 0;
2842 	}
2843 
2844 	nilfs->ns_writer = nilfs_segctor_new(sb, root);
2845 	if (!nilfs->ns_writer)
2846 		return -ENOMEM;
2847 
2848 	err = nilfs_segctor_start_thread(nilfs->ns_writer);
2849 	if (unlikely(err))
2850 		nilfs_detach_log_writer(sb);
2851 
2852 	return err;
2853 }
2854 
2855 /**
2856  * nilfs_detach_log_writer - destroy log writer
2857  * @sb: super block instance
2858  *
2859  * This kills log writer daemon, frees the log writer object, and
2860  * destroys list of dirty files.
2861  */
2862 void nilfs_detach_log_writer(struct super_block *sb)
2863 {
2864 	struct the_nilfs *nilfs = sb->s_fs_info;
2865 	LIST_HEAD(garbage_list);
2866 
2867 	down_write(&nilfs->ns_segctor_sem);
2868 	if (nilfs->ns_writer) {
2869 		nilfs_segctor_destroy(nilfs->ns_writer);
2870 		nilfs->ns_writer = NULL;
2871 	}
2872 	set_nilfs_purging(nilfs);
2873 
2874 	/* Force to free the list of dirty files */
2875 	spin_lock(&nilfs->ns_inode_lock);
2876 	if (!list_empty(&nilfs->ns_dirty_files)) {
2877 		list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2878 		nilfs_warn(sb,
2879 			   "disposed unprocessed dirty file(s) when detaching log writer");
2880 	}
2881 	spin_unlock(&nilfs->ns_inode_lock);
2882 	up_write(&nilfs->ns_segctor_sem);
2883 
2884 	nilfs_dispose_list(nilfs, &garbage_list, 1);
2885 	clear_nilfs_purging(nilfs);
2886 }
2887