xref: /linux/fs/nilfs2/segment.c (revision 0e685c3e7158d35626d6d76b9f859eae806d87fa)
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 (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 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 folio_batch fbatch;
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 	folio_batch_init(&fbatch);
698  repeat:
699 	if (unlikely(index > last) ||
700 	      !filemap_get_folios_tag(mapping, &index, last,
701 		      PAGECACHE_TAG_DIRTY, &fbatch))
702 		return ndirties;
703 
704 	for (i = 0; i < folio_batch_count(&fbatch); i++) {
705 		struct buffer_head *bh, *head;
706 		struct folio *folio = fbatch.folios[i];
707 
708 		folio_lock(folio);
709 		head = folio_buffers(folio);
710 		if (!head) {
711 			create_empty_buffers(&folio->page, i_blocksize(inode), 0);
712 			head = folio_buffers(folio);
713 		}
714 		folio_unlock(folio);
715 
716 		bh = head;
717 		do {
718 			if (!buffer_dirty(bh) || buffer_async_write(bh))
719 				continue;
720 			get_bh(bh);
721 			list_add_tail(&bh->b_assoc_buffers, listp);
722 			ndirties++;
723 			if (unlikely(ndirties >= nlimit)) {
724 				folio_batch_release(&fbatch);
725 				cond_resched();
726 				return ndirties;
727 			}
728 		} while (bh = bh->b_this_page, bh != head);
729 	}
730 	folio_batch_release(&fbatch);
731 	cond_resched();
732 	goto repeat;
733 }
734 
735 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
736 					    struct list_head *listp)
737 {
738 	struct nilfs_inode_info *ii = NILFS_I(inode);
739 	struct inode *btnc_inode = ii->i_assoc_inode;
740 	struct folio_batch fbatch;
741 	struct buffer_head *bh, *head;
742 	unsigned int i;
743 	pgoff_t index = 0;
744 
745 	if (!btnc_inode)
746 		return;
747 	folio_batch_init(&fbatch);
748 
749 	while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
750 				(pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
751 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
752 			bh = head = folio_buffers(fbatch.folios[i]);
753 			do {
754 				if (buffer_dirty(bh) &&
755 						!buffer_async_write(bh)) {
756 					get_bh(bh);
757 					list_add_tail(&bh->b_assoc_buffers,
758 						      listp);
759 				}
760 				bh = bh->b_this_page;
761 			} while (bh != head);
762 		}
763 		folio_batch_release(&fbatch);
764 		cond_resched();
765 	}
766 }
767 
768 static void nilfs_dispose_list(struct the_nilfs *nilfs,
769 			       struct list_head *head, int force)
770 {
771 	struct nilfs_inode_info *ii, *n;
772 	struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
773 	unsigned int nv = 0;
774 
775 	while (!list_empty(head)) {
776 		spin_lock(&nilfs->ns_inode_lock);
777 		list_for_each_entry_safe(ii, n, head, i_dirty) {
778 			list_del_init(&ii->i_dirty);
779 			if (force) {
780 				if (unlikely(ii->i_bh)) {
781 					brelse(ii->i_bh);
782 					ii->i_bh = NULL;
783 				}
784 			} else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
785 				set_bit(NILFS_I_QUEUED, &ii->i_state);
786 				list_add_tail(&ii->i_dirty,
787 					      &nilfs->ns_dirty_files);
788 				continue;
789 			}
790 			ivec[nv++] = ii;
791 			if (nv == SC_N_INODEVEC)
792 				break;
793 		}
794 		spin_unlock(&nilfs->ns_inode_lock);
795 
796 		for (pii = ivec; nv > 0; pii++, nv--)
797 			iput(&(*pii)->vfs_inode);
798 	}
799 }
800 
801 static void nilfs_iput_work_func(struct work_struct *work)
802 {
803 	struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
804 						 sc_iput_work);
805 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
806 
807 	nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
808 }
809 
810 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
811 				     struct nilfs_root *root)
812 {
813 	int ret = 0;
814 
815 	if (nilfs_mdt_fetch_dirty(root->ifile))
816 		ret++;
817 	if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
818 		ret++;
819 	if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
820 		ret++;
821 	if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
822 		ret++;
823 	return ret;
824 }
825 
826 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
827 {
828 	return list_empty(&sci->sc_dirty_files) &&
829 		!test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
830 		sci->sc_nfreesegs == 0 &&
831 		(!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
832 }
833 
834 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
835 {
836 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
837 	int ret = 0;
838 
839 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
840 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
841 
842 	spin_lock(&nilfs->ns_inode_lock);
843 	if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
844 		ret++;
845 
846 	spin_unlock(&nilfs->ns_inode_lock);
847 	return ret;
848 }
849 
850 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
851 {
852 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
853 
854 	nilfs_mdt_clear_dirty(sci->sc_root->ifile);
855 	nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
856 	nilfs_mdt_clear_dirty(nilfs->ns_sufile);
857 	nilfs_mdt_clear_dirty(nilfs->ns_dat);
858 }
859 
860 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
861 {
862 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
863 	struct buffer_head *bh_cp;
864 	struct nilfs_checkpoint *raw_cp;
865 	int err;
866 
867 	/* XXX: this interface will be changed */
868 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
869 					  &raw_cp, &bh_cp);
870 	if (likely(!err)) {
871 		/*
872 		 * The following code is duplicated with cpfile.  But, it is
873 		 * needed to collect the checkpoint even if it was not newly
874 		 * created.
875 		 */
876 		mark_buffer_dirty(bh_cp);
877 		nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
878 		nilfs_cpfile_put_checkpoint(
879 			nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
880 	} else if (err == -EINVAL || err == -ENOENT) {
881 		nilfs_error(sci->sc_super,
882 			    "checkpoint creation failed due to metadata corruption.");
883 		err = -EIO;
884 	}
885 	return err;
886 }
887 
888 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
889 {
890 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
891 	struct buffer_head *bh_cp;
892 	struct nilfs_checkpoint *raw_cp;
893 	int err;
894 
895 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
896 					  &raw_cp, &bh_cp);
897 	if (unlikely(err)) {
898 		if (err == -EINVAL || err == -ENOENT) {
899 			nilfs_error(sci->sc_super,
900 				    "checkpoint finalization failed due to metadata corruption.");
901 			err = -EIO;
902 		}
903 		goto failed_ibh;
904 	}
905 	raw_cp->cp_snapshot_list.ssl_next = 0;
906 	raw_cp->cp_snapshot_list.ssl_prev = 0;
907 	raw_cp->cp_inodes_count =
908 		cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
909 	raw_cp->cp_blocks_count =
910 		cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
911 	raw_cp->cp_nblk_inc =
912 		cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
913 	raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
914 	raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
915 
916 	if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
917 		nilfs_checkpoint_clear_minor(raw_cp);
918 	else
919 		nilfs_checkpoint_set_minor(raw_cp);
920 
921 	nilfs_write_inode_common(sci->sc_root->ifile,
922 				 &raw_cp->cp_ifile_inode, 1);
923 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
924 	return 0;
925 
926  failed_ibh:
927 	return err;
928 }
929 
930 static void nilfs_fill_in_file_bmap(struct inode *ifile,
931 				    struct nilfs_inode_info *ii)
932 
933 {
934 	struct buffer_head *ibh;
935 	struct nilfs_inode *raw_inode;
936 
937 	if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
938 		ibh = ii->i_bh;
939 		BUG_ON(!ibh);
940 		raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
941 						  ibh);
942 		nilfs_bmap_write(ii->i_bmap, raw_inode);
943 		nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
944 	}
945 }
946 
947 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
948 {
949 	struct nilfs_inode_info *ii;
950 
951 	list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
952 		nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
953 		set_bit(NILFS_I_COLLECTED, &ii->i_state);
954 	}
955 }
956 
957 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
958 					     struct the_nilfs *nilfs)
959 {
960 	struct buffer_head *bh_sr;
961 	struct nilfs_super_root *raw_sr;
962 	unsigned int isz, srsz;
963 
964 	bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
965 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
966 	isz = nilfs->ns_inode_size;
967 	srsz = NILFS_SR_BYTES(isz);
968 
969 	raw_sr->sr_bytes = cpu_to_le16(srsz);
970 	raw_sr->sr_nongc_ctime
971 		= cpu_to_le64(nilfs_doing_gc() ?
972 			      nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
973 	raw_sr->sr_flags = 0;
974 
975 	nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
976 				 NILFS_SR_DAT_OFFSET(isz), 1);
977 	nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
978 				 NILFS_SR_CPFILE_OFFSET(isz), 1);
979 	nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
980 				 NILFS_SR_SUFILE_OFFSET(isz), 1);
981 	memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
982 }
983 
984 static void nilfs_redirty_inodes(struct list_head *head)
985 {
986 	struct nilfs_inode_info *ii;
987 
988 	list_for_each_entry(ii, head, i_dirty) {
989 		if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
990 			clear_bit(NILFS_I_COLLECTED, &ii->i_state);
991 	}
992 }
993 
994 static void nilfs_drop_collected_inodes(struct list_head *head)
995 {
996 	struct nilfs_inode_info *ii;
997 
998 	list_for_each_entry(ii, head, i_dirty) {
999 		if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1000 			continue;
1001 
1002 		clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1003 		set_bit(NILFS_I_UPDATED, &ii->i_state);
1004 	}
1005 }
1006 
1007 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1008 				       struct inode *inode,
1009 				       struct list_head *listp,
1010 				       int (*collect)(struct nilfs_sc_info *,
1011 						      struct buffer_head *,
1012 						      struct inode *))
1013 {
1014 	struct buffer_head *bh, *n;
1015 	int err = 0;
1016 
1017 	if (collect) {
1018 		list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1019 			list_del_init(&bh->b_assoc_buffers);
1020 			err = collect(sci, bh, inode);
1021 			brelse(bh);
1022 			if (unlikely(err))
1023 				goto dispose_buffers;
1024 		}
1025 		return 0;
1026 	}
1027 
1028  dispose_buffers:
1029 	while (!list_empty(listp)) {
1030 		bh = list_first_entry(listp, struct buffer_head,
1031 				      b_assoc_buffers);
1032 		list_del_init(&bh->b_assoc_buffers);
1033 		brelse(bh);
1034 	}
1035 	return err;
1036 }
1037 
1038 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1039 {
1040 	/* Remaining number of blocks within segment buffer */
1041 	return sci->sc_segbuf_nblocks -
1042 		(sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1043 }
1044 
1045 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1046 				   struct inode *inode,
1047 				   const struct nilfs_sc_operations *sc_ops)
1048 {
1049 	LIST_HEAD(data_buffers);
1050 	LIST_HEAD(node_buffers);
1051 	int err;
1052 
1053 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1054 		size_t n, rest = nilfs_segctor_buffer_rest(sci);
1055 
1056 		n = nilfs_lookup_dirty_data_buffers(
1057 			inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1058 		if (n > rest) {
1059 			err = nilfs_segctor_apply_buffers(
1060 				sci, inode, &data_buffers,
1061 				sc_ops->collect_data);
1062 			BUG_ON(!err); /* always receive -E2BIG or true error */
1063 			goto break_or_fail;
1064 		}
1065 	}
1066 	nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1067 
1068 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1069 		err = nilfs_segctor_apply_buffers(
1070 			sci, inode, &data_buffers, sc_ops->collect_data);
1071 		if (unlikely(err)) {
1072 			/* dispose node list */
1073 			nilfs_segctor_apply_buffers(
1074 				sci, inode, &node_buffers, NULL);
1075 			goto break_or_fail;
1076 		}
1077 		sci->sc_stage.flags |= NILFS_CF_NODE;
1078 	}
1079 	/* Collect node */
1080 	err = nilfs_segctor_apply_buffers(
1081 		sci, inode, &node_buffers, sc_ops->collect_node);
1082 	if (unlikely(err))
1083 		goto break_or_fail;
1084 
1085 	nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1086 	err = nilfs_segctor_apply_buffers(
1087 		sci, inode, &node_buffers, sc_ops->collect_bmap);
1088 	if (unlikely(err))
1089 		goto break_or_fail;
1090 
1091 	nilfs_segctor_end_finfo(sci, inode);
1092 	sci->sc_stage.flags &= ~NILFS_CF_NODE;
1093 
1094  break_or_fail:
1095 	return err;
1096 }
1097 
1098 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1099 					 struct inode *inode)
1100 {
1101 	LIST_HEAD(data_buffers);
1102 	size_t n, rest = nilfs_segctor_buffer_rest(sci);
1103 	int err;
1104 
1105 	n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1106 					    sci->sc_dsync_start,
1107 					    sci->sc_dsync_end);
1108 
1109 	err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1110 					  nilfs_collect_file_data);
1111 	if (!err) {
1112 		nilfs_segctor_end_finfo(sci, inode);
1113 		BUG_ON(n > rest);
1114 		/* always receive -E2BIG or true error if n > rest */
1115 	}
1116 	return err;
1117 }
1118 
1119 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1120 {
1121 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1122 	struct list_head *head;
1123 	struct nilfs_inode_info *ii;
1124 	size_t ndone;
1125 	int err = 0;
1126 
1127 	switch (nilfs_sc_cstage_get(sci)) {
1128 	case NILFS_ST_INIT:
1129 		/* Pre-processes */
1130 		sci->sc_stage.flags = 0;
1131 
1132 		if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1133 			sci->sc_nblk_inc = 0;
1134 			sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1135 			if (mode == SC_LSEG_DSYNC) {
1136 				nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1137 				goto dsync_mode;
1138 			}
1139 		}
1140 
1141 		sci->sc_stage.dirty_file_ptr = NULL;
1142 		sci->sc_stage.gc_inode_ptr = NULL;
1143 		if (mode == SC_FLUSH_DAT) {
1144 			nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1145 			goto dat_stage;
1146 		}
1147 		nilfs_sc_cstage_inc(sci);
1148 		fallthrough;
1149 	case NILFS_ST_GC:
1150 		if (nilfs_doing_gc()) {
1151 			head = &sci->sc_gc_inodes;
1152 			ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1153 						head, i_dirty);
1154 			list_for_each_entry_continue(ii, head, i_dirty) {
1155 				err = nilfs_segctor_scan_file(
1156 					sci, &ii->vfs_inode,
1157 					&nilfs_sc_file_ops);
1158 				if (unlikely(err)) {
1159 					sci->sc_stage.gc_inode_ptr = list_entry(
1160 						ii->i_dirty.prev,
1161 						struct nilfs_inode_info,
1162 						i_dirty);
1163 					goto break_or_fail;
1164 				}
1165 				set_bit(NILFS_I_COLLECTED, &ii->i_state);
1166 			}
1167 			sci->sc_stage.gc_inode_ptr = NULL;
1168 		}
1169 		nilfs_sc_cstage_inc(sci);
1170 		fallthrough;
1171 	case NILFS_ST_FILE:
1172 		head = &sci->sc_dirty_files;
1173 		ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1174 					i_dirty);
1175 		list_for_each_entry_continue(ii, head, i_dirty) {
1176 			clear_bit(NILFS_I_DIRTY, &ii->i_state);
1177 
1178 			err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1179 						      &nilfs_sc_file_ops);
1180 			if (unlikely(err)) {
1181 				sci->sc_stage.dirty_file_ptr =
1182 					list_entry(ii->i_dirty.prev,
1183 						   struct nilfs_inode_info,
1184 						   i_dirty);
1185 				goto break_or_fail;
1186 			}
1187 			/* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1188 			/* XXX: required ? */
1189 		}
1190 		sci->sc_stage.dirty_file_ptr = NULL;
1191 		if (mode == SC_FLUSH_FILE) {
1192 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1193 			return 0;
1194 		}
1195 		nilfs_sc_cstage_inc(sci);
1196 		sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1197 		fallthrough;
1198 	case NILFS_ST_IFILE:
1199 		err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1200 					      &nilfs_sc_file_ops);
1201 		if (unlikely(err))
1202 			break;
1203 		nilfs_sc_cstage_inc(sci);
1204 		/* Creating a checkpoint */
1205 		err = nilfs_segctor_create_checkpoint(sci);
1206 		if (unlikely(err))
1207 			break;
1208 		fallthrough;
1209 	case NILFS_ST_CPFILE:
1210 		err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1211 					      &nilfs_sc_file_ops);
1212 		if (unlikely(err))
1213 			break;
1214 		nilfs_sc_cstage_inc(sci);
1215 		fallthrough;
1216 	case NILFS_ST_SUFILE:
1217 		err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1218 					 sci->sc_nfreesegs, &ndone);
1219 		if (unlikely(err)) {
1220 			nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1221 						  sci->sc_freesegs, ndone,
1222 						  NULL);
1223 			break;
1224 		}
1225 		sci->sc_stage.flags |= NILFS_CF_SUFREED;
1226 
1227 		err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1228 					      &nilfs_sc_file_ops);
1229 		if (unlikely(err))
1230 			break;
1231 		nilfs_sc_cstage_inc(sci);
1232 		fallthrough;
1233 	case NILFS_ST_DAT:
1234  dat_stage:
1235 		err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1236 					      &nilfs_sc_dat_ops);
1237 		if (unlikely(err))
1238 			break;
1239 		if (mode == SC_FLUSH_DAT) {
1240 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1241 			return 0;
1242 		}
1243 		nilfs_sc_cstage_inc(sci);
1244 		fallthrough;
1245 	case NILFS_ST_SR:
1246 		if (mode == SC_LSEG_SR) {
1247 			/* Appending a super root */
1248 			err = nilfs_segctor_add_super_root(sci);
1249 			if (unlikely(err))
1250 				break;
1251 		}
1252 		/* End of a logical segment */
1253 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1254 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1255 		return 0;
1256 	case NILFS_ST_DSYNC:
1257  dsync_mode:
1258 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1259 		ii = sci->sc_dsync_inode;
1260 		if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1261 			break;
1262 
1263 		err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1264 		if (unlikely(err))
1265 			break;
1266 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1267 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1268 		return 0;
1269 	case NILFS_ST_DONE:
1270 		return 0;
1271 	default:
1272 		BUG();
1273 	}
1274 
1275  break_or_fail:
1276 	return err;
1277 }
1278 
1279 /**
1280  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1281  * @sci: nilfs_sc_info
1282  * @nilfs: nilfs object
1283  */
1284 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1285 					    struct the_nilfs *nilfs)
1286 {
1287 	struct nilfs_segment_buffer *segbuf, *prev;
1288 	__u64 nextnum;
1289 	int err, alloc = 0;
1290 
1291 	segbuf = nilfs_segbuf_new(sci->sc_super);
1292 	if (unlikely(!segbuf))
1293 		return -ENOMEM;
1294 
1295 	if (list_empty(&sci->sc_write_logs)) {
1296 		nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1297 				 nilfs->ns_pseg_offset, nilfs);
1298 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1299 			nilfs_shift_to_next_segment(nilfs);
1300 			nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1301 		}
1302 
1303 		segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1304 		nextnum = nilfs->ns_nextnum;
1305 
1306 		if (nilfs->ns_segnum == nilfs->ns_nextnum)
1307 			/* Start from the head of a new full segment */
1308 			alloc++;
1309 	} else {
1310 		/* Continue logs */
1311 		prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1312 		nilfs_segbuf_map_cont(segbuf, prev);
1313 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1314 		nextnum = prev->sb_nextnum;
1315 
1316 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1317 			nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1318 			segbuf->sb_sum.seg_seq++;
1319 			alloc++;
1320 		}
1321 	}
1322 
1323 	err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1324 	if (err)
1325 		goto failed;
1326 
1327 	if (alloc) {
1328 		err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1329 		if (err)
1330 			goto failed;
1331 	}
1332 	nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1333 
1334 	BUG_ON(!list_empty(&sci->sc_segbufs));
1335 	list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1336 	sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1337 	return 0;
1338 
1339  failed:
1340 	nilfs_segbuf_free(segbuf);
1341 	return err;
1342 }
1343 
1344 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1345 					 struct the_nilfs *nilfs, int nadd)
1346 {
1347 	struct nilfs_segment_buffer *segbuf, *prev;
1348 	struct inode *sufile = nilfs->ns_sufile;
1349 	__u64 nextnextnum;
1350 	LIST_HEAD(list);
1351 	int err, ret, i;
1352 
1353 	prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1354 	/*
1355 	 * Since the segment specified with nextnum might be allocated during
1356 	 * the previous construction, the buffer including its segusage may
1357 	 * not be dirty.  The following call ensures that the buffer is dirty
1358 	 * and will pin the buffer on memory until the sufile is written.
1359 	 */
1360 	err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1361 	if (unlikely(err))
1362 		return err;
1363 
1364 	for (i = 0; i < nadd; i++) {
1365 		/* extend segment info */
1366 		err = -ENOMEM;
1367 		segbuf = nilfs_segbuf_new(sci->sc_super);
1368 		if (unlikely(!segbuf))
1369 			goto failed;
1370 
1371 		/* map this buffer to region of segment on-disk */
1372 		nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1373 		sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1374 
1375 		/* allocate the next next full segment */
1376 		err = nilfs_sufile_alloc(sufile, &nextnextnum);
1377 		if (unlikely(err))
1378 			goto failed_segbuf;
1379 
1380 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1381 		nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1382 
1383 		list_add_tail(&segbuf->sb_list, &list);
1384 		prev = segbuf;
1385 	}
1386 	list_splice_tail(&list, &sci->sc_segbufs);
1387 	return 0;
1388 
1389  failed_segbuf:
1390 	nilfs_segbuf_free(segbuf);
1391  failed:
1392 	list_for_each_entry(segbuf, &list, sb_list) {
1393 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1394 		WARN_ON(ret); /* never fails */
1395 	}
1396 	nilfs_destroy_logs(&list);
1397 	return err;
1398 }
1399 
1400 static void nilfs_free_incomplete_logs(struct list_head *logs,
1401 				       struct the_nilfs *nilfs)
1402 {
1403 	struct nilfs_segment_buffer *segbuf, *prev;
1404 	struct inode *sufile = nilfs->ns_sufile;
1405 	int ret;
1406 
1407 	segbuf = NILFS_FIRST_SEGBUF(logs);
1408 	if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1409 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1410 		WARN_ON(ret); /* never fails */
1411 	}
1412 	if (atomic_read(&segbuf->sb_err)) {
1413 		/* Case 1: The first segment failed */
1414 		if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1415 			/*
1416 			 * Case 1a:  Partial segment appended into an existing
1417 			 * segment
1418 			 */
1419 			nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1420 						segbuf->sb_fseg_end);
1421 		else /* Case 1b:  New full segment */
1422 			set_nilfs_discontinued(nilfs);
1423 	}
1424 
1425 	prev = segbuf;
1426 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1427 		if (prev->sb_nextnum != segbuf->sb_nextnum) {
1428 			ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1429 			WARN_ON(ret); /* never fails */
1430 		}
1431 		if (atomic_read(&segbuf->sb_err) &&
1432 		    segbuf->sb_segnum != nilfs->ns_nextnum)
1433 			/* Case 2: extended segment (!= next) failed */
1434 			nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1435 		prev = segbuf;
1436 	}
1437 }
1438 
1439 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1440 					  struct inode *sufile)
1441 {
1442 	struct nilfs_segment_buffer *segbuf;
1443 	unsigned long live_blocks;
1444 	int ret;
1445 
1446 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1447 		live_blocks = segbuf->sb_sum.nblocks +
1448 			(segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1449 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1450 						     live_blocks,
1451 						     sci->sc_seg_ctime);
1452 		WARN_ON(ret); /* always succeed because the segusage is dirty */
1453 	}
1454 }
1455 
1456 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1457 {
1458 	struct nilfs_segment_buffer *segbuf;
1459 	int ret;
1460 
1461 	segbuf = NILFS_FIRST_SEGBUF(logs);
1462 	ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1463 					     segbuf->sb_pseg_start -
1464 					     segbuf->sb_fseg_start, 0);
1465 	WARN_ON(ret); /* always succeed because the segusage is dirty */
1466 
1467 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1468 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1469 						     0, 0);
1470 		WARN_ON(ret); /* always succeed */
1471 	}
1472 }
1473 
1474 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1475 					    struct nilfs_segment_buffer *last,
1476 					    struct inode *sufile)
1477 {
1478 	struct nilfs_segment_buffer *segbuf = last;
1479 	int ret;
1480 
1481 	list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1482 		sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1483 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1484 		WARN_ON(ret);
1485 	}
1486 	nilfs_truncate_logs(&sci->sc_segbufs, last);
1487 }
1488 
1489 
1490 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1491 				 struct the_nilfs *nilfs, int mode)
1492 {
1493 	struct nilfs_cstage prev_stage = sci->sc_stage;
1494 	int err, nadd = 1;
1495 
1496 	/* Collection retry loop */
1497 	for (;;) {
1498 		sci->sc_nblk_this_inc = 0;
1499 		sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1500 
1501 		err = nilfs_segctor_reset_segment_buffer(sci);
1502 		if (unlikely(err))
1503 			goto failed;
1504 
1505 		err = nilfs_segctor_collect_blocks(sci, mode);
1506 		sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1507 		if (!err)
1508 			break;
1509 
1510 		if (unlikely(err != -E2BIG))
1511 			goto failed;
1512 
1513 		/* The current segment is filled up */
1514 		if (mode != SC_LSEG_SR ||
1515 		    nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1516 			break;
1517 
1518 		nilfs_clear_logs(&sci->sc_segbufs);
1519 
1520 		if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1521 			err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1522 							sci->sc_freesegs,
1523 							sci->sc_nfreesegs,
1524 							NULL);
1525 			WARN_ON(err); /* do not happen */
1526 			sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1527 		}
1528 
1529 		err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1530 		if (unlikely(err))
1531 			return err;
1532 
1533 		nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1534 		sci->sc_stage = prev_stage;
1535 	}
1536 	nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1537 	return 0;
1538 
1539  failed:
1540 	return err;
1541 }
1542 
1543 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1544 				      struct buffer_head *new_bh)
1545 {
1546 	BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1547 
1548 	list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1549 	/* The caller must release old_bh */
1550 }
1551 
1552 static int
1553 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1554 				     struct nilfs_segment_buffer *segbuf,
1555 				     int mode)
1556 {
1557 	struct inode *inode = NULL;
1558 	sector_t blocknr;
1559 	unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1560 	unsigned long nblocks = 0, ndatablk = 0;
1561 	const struct nilfs_sc_operations *sc_op = NULL;
1562 	struct nilfs_segsum_pointer ssp;
1563 	struct nilfs_finfo *finfo = NULL;
1564 	union nilfs_binfo binfo;
1565 	struct buffer_head *bh, *bh_org;
1566 	ino_t ino = 0;
1567 	int err = 0;
1568 
1569 	if (!nfinfo)
1570 		goto out;
1571 
1572 	blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1573 	ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1574 	ssp.offset = sizeof(struct nilfs_segment_summary);
1575 
1576 	list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1577 		if (bh == segbuf->sb_super_root)
1578 			break;
1579 		if (!finfo) {
1580 			finfo =	nilfs_segctor_map_segsum_entry(
1581 				sci, &ssp, sizeof(*finfo));
1582 			ino = le64_to_cpu(finfo->fi_ino);
1583 			nblocks = le32_to_cpu(finfo->fi_nblocks);
1584 			ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1585 
1586 			inode = bh->b_folio->mapping->host;
1587 
1588 			if (mode == SC_LSEG_DSYNC)
1589 				sc_op = &nilfs_sc_dsync_ops;
1590 			else if (ino == NILFS_DAT_INO)
1591 				sc_op = &nilfs_sc_dat_ops;
1592 			else /* file blocks */
1593 				sc_op = &nilfs_sc_file_ops;
1594 		}
1595 		bh_org = bh;
1596 		get_bh(bh_org);
1597 		err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1598 					&binfo);
1599 		if (bh != bh_org)
1600 			nilfs_list_replace_buffer(bh_org, bh);
1601 		brelse(bh_org);
1602 		if (unlikely(err))
1603 			goto failed_bmap;
1604 
1605 		if (ndatablk > 0)
1606 			sc_op->write_data_binfo(sci, &ssp, &binfo);
1607 		else
1608 			sc_op->write_node_binfo(sci, &ssp, &binfo);
1609 
1610 		blocknr++;
1611 		if (--nblocks == 0) {
1612 			finfo = NULL;
1613 			if (--nfinfo == 0)
1614 				break;
1615 		} else if (ndatablk > 0)
1616 			ndatablk--;
1617 	}
1618  out:
1619 	return 0;
1620 
1621  failed_bmap:
1622 	return err;
1623 }
1624 
1625 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1626 {
1627 	struct nilfs_segment_buffer *segbuf;
1628 	int err;
1629 
1630 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1631 		err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1632 		if (unlikely(err))
1633 			return err;
1634 		nilfs_segbuf_fill_in_segsum(segbuf);
1635 	}
1636 	return 0;
1637 }
1638 
1639 static void nilfs_begin_page_io(struct page *page)
1640 {
1641 	if (!page || PageWriteback(page))
1642 		/*
1643 		 * For split b-tree node pages, this function may be called
1644 		 * twice.  We ignore the 2nd or later calls by this check.
1645 		 */
1646 		return;
1647 
1648 	lock_page(page);
1649 	clear_page_dirty_for_io(page);
1650 	set_page_writeback(page);
1651 	unlock_page(page);
1652 }
1653 
1654 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1655 {
1656 	struct nilfs_segment_buffer *segbuf;
1657 	struct page *bd_page = NULL, *fs_page = NULL;
1658 
1659 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1660 		struct buffer_head *bh;
1661 
1662 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1663 				    b_assoc_buffers) {
1664 			if (bh->b_page != bd_page) {
1665 				if (bd_page) {
1666 					lock_page(bd_page);
1667 					clear_page_dirty_for_io(bd_page);
1668 					set_page_writeback(bd_page);
1669 					unlock_page(bd_page);
1670 				}
1671 				bd_page = bh->b_page;
1672 			}
1673 		}
1674 
1675 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1676 				    b_assoc_buffers) {
1677 			set_buffer_async_write(bh);
1678 			if (bh == segbuf->sb_super_root) {
1679 				if (bh->b_page != bd_page) {
1680 					lock_page(bd_page);
1681 					clear_page_dirty_for_io(bd_page);
1682 					set_page_writeback(bd_page);
1683 					unlock_page(bd_page);
1684 					bd_page = bh->b_page;
1685 				}
1686 				break;
1687 			}
1688 			if (bh->b_page != fs_page) {
1689 				nilfs_begin_page_io(fs_page);
1690 				fs_page = bh->b_page;
1691 			}
1692 		}
1693 	}
1694 	if (bd_page) {
1695 		lock_page(bd_page);
1696 		clear_page_dirty_for_io(bd_page);
1697 		set_page_writeback(bd_page);
1698 		unlock_page(bd_page);
1699 	}
1700 	nilfs_begin_page_io(fs_page);
1701 }
1702 
1703 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1704 			       struct the_nilfs *nilfs)
1705 {
1706 	int ret;
1707 
1708 	ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1709 	list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1710 	return ret;
1711 }
1712 
1713 static void nilfs_end_page_io(struct page *page, int err)
1714 {
1715 	if (!page)
1716 		return;
1717 
1718 	if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1719 		/*
1720 		 * For b-tree node pages, this function may be called twice
1721 		 * or more because they might be split in a segment.
1722 		 */
1723 		if (PageDirty(page)) {
1724 			/*
1725 			 * For pages holding split b-tree node buffers, dirty
1726 			 * flag on the buffers may be cleared discretely.
1727 			 * In that case, the page is once redirtied for
1728 			 * remaining buffers, and it must be cancelled if
1729 			 * all the buffers get cleaned later.
1730 			 */
1731 			lock_page(page);
1732 			if (nilfs_page_buffers_clean(page))
1733 				__nilfs_clear_page_dirty(page);
1734 			unlock_page(page);
1735 		}
1736 		return;
1737 	}
1738 
1739 	if (!err) {
1740 		if (!nilfs_page_buffers_clean(page))
1741 			__set_page_dirty_nobuffers(page);
1742 		ClearPageError(page);
1743 	} else {
1744 		__set_page_dirty_nobuffers(page);
1745 		SetPageError(page);
1746 	}
1747 
1748 	end_page_writeback(page);
1749 }
1750 
1751 static void nilfs_abort_logs(struct list_head *logs, int err)
1752 {
1753 	struct nilfs_segment_buffer *segbuf;
1754 	struct page *bd_page = NULL, *fs_page = NULL;
1755 	struct buffer_head *bh;
1756 
1757 	if (list_empty(logs))
1758 		return;
1759 
1760 	list_for_each_entry(segbuf, logs, sb_list) {
1761 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1762 				    b_assoc_buffers) {
1763 			if (bh->b_page != bd_page) {
1764 				if (bd_page)
1765 					end_page_writeback(bd_page);
1766 				bd_page = bh->b_page;
1767 			}
1768 		}
1769 
1770 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1771 				    b_assoc_buffers) {
1772 			clear_buffer_async_write(bh);
1773 			if (bh == segbuf->sb_super_root) {
1774 				if (bh->b_page != bd_page) {
1775 					end_page_writeback(bd_page);
1776 					bd_page = bh->b_page;
1777 				}
1778 				break;
1779 			}
1780 			if (bh->b_page != fs_page) {
1781 				nilfs_end_page_io(fs_page, err);
1782 				fs_page = bh->b_page;
1783 			}
1784 		}
1785 	}
1786 	if (bd_page)
1787 		end_page_writeback(bd_page);
1788 
1789 	nilfs_end_page_io(fs_page, err);
1790 }
1791 
1792 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1793 					     struct the_nilfs *nilfs, int err)
1794 {
1795 	LIST_HEAD(logs);
1796 	int ret;
1797 
1798 	list_splice_tail_init(&sci->sc_write_logs, &logs);
1799 	ret = nilfs_wait_on_logs(&logs);
1800 	nilfs_abort_logs(&logs, ret ? : err);
1801 
1802 	list_splice_tail_init(&sci->sc_segbufs, &logs);
1803 	nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1804 	nilfs_free_incomplete_logs(&logs, nilfs);
1805 
1806 	if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1807 		ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1808 						sci->sc_freesegs,
1809 						sci->sc_nfreesegs,
1810 						NULL);
1811 		WARN_ON(ret); /* do not happen */
1812 	}
1813 
1814 	nilfs_destroy_logs(&logs);
1815 }
1816 
1817 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1818 				   struct nilfs_segment_buffer *segbuf)
1819 {
1820 	nilfs->ns_segnum = segbuf->sb_segnum;
1821 	nilfs->ns_nextnum = segbuf->sb_nextnum;
1822 	nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1823 		+ segbuf->sb_sum.nblocks;
1824 	nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1825 	nilfs->ns_ctime = segbuf->sb_sum.ctime;
1826 }
1827 
1828 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1829 {
1830 	struct nilfs_segment_buffer *segbuf;
1831 	struct page *bd_page = NULL, *fs_page = NULL;
1832 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1833 	int update_sr = false;
1834 
1835 	list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1836 		struct buffer_head *bh;
1837 
1838 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1839 				    b_assoc_buffers) {
1840 			set_buffer_uptodate(bh);
1841 			clear_buffer_dirty(bh);
1842 			if (bh->b_page != bd_page) {
1843 				if (bd_page)
1844 					end_page_writeback(bd_page);
1845 				bd_page = bh->b_page;
1846 			}
1847 		}
1848 		/*
1849 		 * We assume that the buffers which belong to the same page
1850 		 * continue over the buffer list.
1851 		 * Under this assumption, the last BHs of pages is
1852 		 * identifiable by the discontinuity of bh->b_page
1853 		 * (page != fs_page).
1854 		 *
1855 		 * For B-tree node blocks, however, this assumption is not
1856 		 * guaranteed.  The cleanup code of B-tree node pages needs
1857 		 * special care.
1858 		 */
1859 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1860 				    b_assoc_buffers) {
1861 			const unsigned long set_bits = BIT(BH_Uptodate);
1862 			const unsigned long clear_bits =
1863 				(BIT(BH_Dirty) | BIT(BH_Async_Write) |
1864 				 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1865 				 BIT(BH_NILFS_Redirected));
1866 
1867 			set_mask_bits(&bh->b_state, clear_bits, set_bits);
1868 			if (bh == segbuf->sb_super_root) {
1869 				if (bh->b_page != bd_page) {
1870 					end_page_writeback(bd_page);
1871 					bd_page = bh->b_page;
1872 				}
1873 				update_sr = true;
1874 				break;
1875 			}
1876 			if (bh->b_page != fs_page) {
1877 				nilfs_end_page_io(fs_page, 0);
1878 				fs_page = bh->b_page;
1879 			}
1880 		}
1881 
1882 		if (!nilfs_segbuf_simplex(segbuf)) {
1883 			if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1884 				set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1885 				sci->sc_lseg_stime = jiffies;
1886 			}
1887 			if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1888 				clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1889 		}
1890 	}
1891 	/*
1892 	 * Since pages may continue over multiple segment buffers,
1893 	 * end of the last page must be checked outside of the loop.
1894 	 */
1895 	if (bd_page)
1896 		end_page_writeback(bd_page);
1897 
1898 	nilfs_end_page_io(fs_page, 0);
1899 
1900 	nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1901 
1902 	if (nilfs_doing_gc())
1903 		nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1904 	else
1905 		nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1906 
1907 	sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1908 
1909 	segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1910 	nilfs_set_next_segment(nilfs, segbuf);
1911 
1912 	if (update_sr) {
1913 		nilfs->ns_flushed_device = 0;
1914 		nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1915 				       segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1916 
1917 		clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1918 		clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1919 		set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1920 		nilfs_segctor_clear_metadata_dirty(sci);
1921 	} else
1922 		clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1923 }
1924 
1925 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1926 {
1927 	int ret;
1928 
1929 	ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1930 	if (!ret) {
1931 		nilfs_segctor_complete_write(sci);
1932 		nilfs_destroy_logs(&sci->sc_write_logs);
1933 	}
1934 	return ret;
1935 }
1936 
1937 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1938 					     struct the_nilfs *nilfs)
1939 {
1940 	struct nilfs_inode_info *ii, *n;
1941 	struct inode *ifile = sci->sc_root->ifile;
1942 
1943 	spin_lock(&nilfs->ns_inode_lock);
1944  retry:
1945 	list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1946 		if (!ii->i_bh) {
1947 			struct buffer_head *ibh;
1948 			int err;
1949 
1950 			spin_unlock(&nilfs->ns_inode_lock);
1951 			err = nilfs_ifile_get_inode_block(
1952 				ifile, ii->vfs_inode.i_ino, &ibh);
1953 			if (unlikely(err)) {
1954 				nilfs_warn(sci->sc_super,
1955 					   "log writer: error %d getting inode block (ino=%lu)",
1956 					   err, ii->vfs_inode.i_ino);
1957 				return err;
1958 			}
1959 			spin_lock(&nilfs->ns_inode_lock);
1960 			if (likely(!ii->i_bh))
1961 				ii->i_bh = ibh;
1962 			else
1963 				brelse(ibh);
1964 			goto retry;
1965 		}
1966 
1967 		// Always redirty the buffer to avoid race condition
1968 		mark_buffer_dirty(ii->i_bh);
1969 		nilfs_mdt_mark_dirty(ifile);
1970 
1971 		clear_bit(NILFS_I_QUEUED, &ii->i_state);
1972 		set_bit(NILFS_I_BUSY, &ii->i_state);
1973 		list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1974 	}
1975 	spin_unlock(&nilfs->ns_inode_lock);
1976 
1977 	return 0;
1978 }
1979 
1980 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1981 					     struct the_nilfs *nilfs)
1982 {
1983 	struct nilfs_inode_info *ii, *n;
1984 	int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1985 	int defer_iput = false;
1986 
1987 	spin_lock(&nilfs->ns_inode_lock);
1988 	list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1989 		if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1990 		    test_bit(NILFS_I_DIRTY, &ii->i_state))
1991 			continue;
1992 
1993 		clear_bit(NILFS_I_BUSY, &ii->i_state);
1994 		brelse(ii->i_bh);
1995 		ii->i_bh = NULL;
1996 		list_del_init(&ii->i_dirty);
1997 		if (!ii->vfs_inode.i_nlink || during_mount) {
1998 			/*
1999 			 * Defer calling iput() to avoid deadlocks if
2000 			 * i_nlink == 0 or mount is not yet finished.
2001 			 */
2002 			list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2003 			defer_iput = true;
2004 		} else {
2005 			spin_unlock(&nilfs->ns_inode_lock);
2006 			iput(&ii->vfs_inode);
2007 			spin_lock(&nilfs->ns_inode_lock);
2008 		}
2009 	}
2010 	spin_unlock(&nilfs->ns_inode_lock);
2011 
2012 	if (defer_iput)
2013 		schedule_work(&sci->sc_iput_work);
2014 }
2015 
2016 /*
2017  * Main procedure of segment constructor
2018  */
2019 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2020 {
2021 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2022 	int err;
2023 
2024 	nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2025 	sci->sc_cno = nilfs->ns_cno;
2026 
2027 	err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2028 	if (unlikely(err))
2029 		goto out;
2030 
2031 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2032 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2033 
2034 	if (nilfs_segctor_clean(sci))
2035 		goto out;
2036 
2037 	do {
2038 		sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2039 
2040 		err = nilfs_segctor_begin_construction(sci, nilfs);
2041 		if (unlikely(err))
2042 			goto out;
2043 
2044 		/* Update time stamp */
2045 		sci->sc_seg_ctime = ktime_get_real_seconds();
2046 
2047 		err = nilfs_segctor_collect(sci, nilfs, mode);
2048 		if (unlikely(err))
2049 			goto failed;
2050 
2051 		/* Avoid empty segment */
2052 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2053 		    nilfs_segbuf_empty(sci->sc_curseg)) {
2054 			nilfs_segctor_abort_construction(sci, nilfs, 1);
2055 			goto out;
2056 		}
2057 
2058 		err = nilfs_segctor_assign(sci, mode);
2059 		if (unlikely(err))
2060 			goto failed;
2061 
2062 		if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2063 			nilfs_segctor_fill_in_file_bmap(sci);
2064 
2065 		if (mode == SC_LSEG_SR &&
2066 		    nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2067 			err = nilfs_segctor_fill_in_checkpoint(sci);
2068 			if (unlikely(err))
2069 				goto failed_to_write;
2070 
2071 			nilfs_segctor_fill_in_super_root(sci, nilfs);
2072 		}
2073 		nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2074 
2075 		/* Write partial segments */
2076 		nilfs_segctor_prepare_write(sci);
2077 
2078 		nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2079 					    nilfs->ns_crc_seed);
2080 
2081 		err = nilfs_segctor_write(sci, nilfs);
2082 		if (unlikely(err))
2083 			goto failed_to_write;
2084 
2085 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2086 		    nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2087 			/*
2088 			 * At this point, we avoid double buffering
2089 			 * for blocksize < pagesize because page dirty
2090 			 * flag is turned off during write and dirty
2091 			 * buffers are not properly collected for
2092 			 * pages crossing over segments.
2093 			 */
2094 			err = nilfs_segctor_wait(sci);
2095 			if (err)
2096 				goto failed_to_write;
2097 		}
2098 	} while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2099 
2100  out:
2101 	nilfs_segctor_drop_written_files(sci, nilfs);
2102 	return err;
2103 
2104  failed_to_write:
2105 	if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2106 		nilfs_redirty_inodes(&sci->sc_dirty_files);
2107 
2108  failed:
2109 	if (nilfs_doing_gc())
2110 		nilfs_redirty_inodes(&sci->sc_gc_inodes);
2111 	nilfs_segctor_abort_construction(sci, nilfs, err);
2112 	goto out;
2113 }
2114 
2115 /**
2116  * nilfs_segctor_start_timer - set timer of background write
2117  * @sci: nilfs_sc_info
2118  *
2119  * If the timer has already been set, it ignores the new request.
2120  * This function MUST be called within a section locking the segment
2121  * semaphore.
2122  */
2123 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2124 {
2125 	spin_lock(&sci->sc_state_lock);
2126 	if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2127 		sci->sc_timer.expires = jiffies + sci->sc_interval;
2128 		add_timer(&sci->sc_timer);
2129 		sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2130 	}
2131 	spin_unlock(&sci->sc_state_lock);
2132 }
2133 
2134 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2135 {
2136 	spin_lock(&sci->sc_state_lock);
2137 	if (!(sci->sc_flush_request & BIT(bn))) {
2138 		unsigned long prev_req = sci->sc_flush_request;
2139 
2140 		sci->sc_flush_request |= BIT(bn);
2141 		if (!prev_req)
2142 			wake_up(&sci->sc_wait_daemon);
2143 	}
2144 	spin_unlock(&sci->sc_state_lock);
2145 }
2146 
2147 /**
2148  * nilfs_flush_segment - trigger a segment construction for resource control
2149  * @sb: super block
2150  * @ino: inode number of the file to be flushed out.
2151  */
2152 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2153 {
2154 	struct the_nilfs *nilfs = sb->s_fs_info;
2155 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2156 
2157 	if (!sci || nilfs_doing_construction())
2158 		return;
2159 	nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2160 					/* assign bit 0 to data files */
2161 }
2162 
2163 struct nilfs_segctor_wait_request {
2164 	wait_queue_entry_t	wq;
2165 	__u32		seq;
2166 	int		err;
2167 	atomic_t	done;
2168 };
2169 
2170 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2171 {
2172 	struct nilfs_segctor_wait_request wait_req;
2173 	int err = 0;
2174 
2175 	spin_lock(&sci->sc_state_lock);
2176 	init_wait(&wait_req.wq);
2177 	wait_req.err = 0;
2178 	atomic_set(&wait_req.done, 0);
2179 	wait_req.seq = ++sci->sc_seq_request;
2180 	spin_unlock(&sci->sc_state_lock);
2181 
2182 	init_waitqueue_entry(&wait_req.wq, current);
2183 	add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2184 	set_current_state(TASK_INTERRUPTIBLE);
2185 	wake_up(&sci->sc_wait_daemon);
2186 
2187 	for (;;) {
2188 		if (atomic_read(&wait_req.done)) {
2189 			err = wait_req.err;
2190 			break;
2191 		}
2192 		if (!signal_pending(current)) {
2193 			schedule();
2194 			continue;
2195 		}
2196 		err = -ERESTARTSYS;
2197 		break;
2198 	}
2199 	finish_wait(&sci->sc_wait_request, &wait_req.wq);
2200 	return err;
2201 }
2202 
2203 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2204 {
2205 	struct nilfs_segctor_wait_request *wrq, *n;
2206 	unsigned long flags;
2207 
2208 	spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2209 	list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2210 		if (!atomic_read(&wrq->done) &&
2211 		    nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2212 			wrq->err = err;
2213 			atomic_set(&wrq->done, 1);
2214 		}
2215 		if (atomic_read(&wrq->done)) {
2216 			wrq->wq.func(&wrq->wq,
2217 				     TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2218 				     0, NULL);
2219 		}
2220 	}
2221 	spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2222 }
2223 
2224 /**
2225  * nilfs_construct_segment - construct a logical segment
2226  * @sb: super block
2227  *
2228  * Return Value: On success, 0 is returned. On errors, one of the following
2229  * negative error code is returned.
2230  *
2231  * %-EROFS - Read only filesystem.
2232  *
2233  * %-EIO - I/O error
2234  *
2235  * %-ENOSPC - No space left on device (only in a panic state).
2236  *
2237  * %-ERESTARTSYS - Interrupted.
2238  *
2239  * %-ENOMEM - Insufficient memory available.
2240  */
2241 int nilfs_construct_segment(struct super_block *sb)
2242 {
2243 	struct the_nilfs *nilfs = sb->s_fs_info;
2244 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2245 	struct nilfs_transaction_info *ti;
2246 
2247 	if (sb_rdonly(sb) || unlikely(!sci))
2248 		return -EROFS;
2249 
2250 	/* A call inside transactions causes a deadlock. */
2251 	BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2252 
2253 	return nilfs_segctor_sync(sci);
2254 }
2255 
2256 /**
2257  * nilfs_construct_dsync_segment - construct a data-only logical segment
2258  * @sb: super block
2259  * @inode: inode whose data blocks should be written out
2260  * @start: start byte offset
2261  * @end: end byte offset (inclusive)
2262  *
2263  * Return Value: On success, 0 is returned. On errors, one of the following
2264  * negative error code is returned.
2265  *
2266  * %-EROFS - Read only filesystem.
2267  *
2268  * %-EIO - I/O error
2269  *
2270  * %-ENOSPC - No space left on device (only in a panic state).
2271  *
2272  * %-ERESTARTSYS - Interrupted.
2273  *
2274  * %-ENOMEM - Insufficient memory available.
2275  */
2276 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2277 				  loff_t start, loff_t end)
2278 {
2279 	struct the_nilfs *nilfs = sb->s_fs_info;
2280 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2281 	struct nilfs_inode_info *ii;
2282 	struct nilfs_transaction_info ti;
2283 	int err = 0;
2284 
2285 	if (sb_rdonly(sb) || unlikely(!sci))
2286 		return -EROFS;
2287 
2288 	nilfs_transaction_lock(sb, &ti, 0);
2289 
2290 	ii = NILFS_I(inode);
2291 	if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2292 	    nilfs_test_opt(nilfs, STRICT_ORDER) ||
2293 	    test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2294 	    nilfs_discontinued(nilfs)) {
2295 		nilfs_transaction_unlock(sb);
2296 		err = nilfs_segctor_sync(sci);
2297 		return err;
2298 	}
2299 
2300 	spin_lock(&nilfs->ns_inode_lock);
2301 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2302 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2303 		spin_unlock(&nilfs->ns_inode_lock);
2304 		nilfs_transaction_unlock(sb);
2305 		return 0;
2306 	}
2307 	spin_unlock(&nilfs->ns_inode_lock);
2308 	sci->sc_dsync_inode = ii;
2309 	sci->sc_dsync_start = start;
2310 	sci->sc_dsync_end = end;
2311 
2312 	err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2313 	if (!err)
2314 		nilfs->ns_flushed_device = 0;
2315 
2316 	nilfs_transaction_unlock(sb);
2317 	return err;
2318 }
2319 
2320 #define FLUSH_FILE_BIT	(0x1) /* data file only */
2321 #define FLUSH_DAT_BIT	BIT(NILFS_DAT_INO) /* DAT only */
2322 
2323 /**
2324  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2325  * @sci: segment constructor object
2326  */
2327 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2328 {
2329 	spin_lock(&sci->sc_state_lock);
2330 	sci->sc_seq_accepted = sci->sc_seq_request;
2331 	spin_unlock(&sci->sc_state_lock);
2332 	del_timer_sync(&sci->sc_timer);
2333 }
2334 
2335 /**
2336  * nilfs_segctor_notify - notify the result of request to caller threads
2337  * @sci: segment constructor object
2338  * @mode: mode of log forming
2339  * @err: error code to be notified
2340  */
2341 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2342 {
2343 	/* Clear requests (even when the construction failed) */
2344 	spin_lock(&sci->sc_state_lock);
2345 
2346 	if (mode == SC_LSEG_SR) {
2347 		sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2348 		sci->sc_seq_done = sci->sc_seq_accepted;
2349 		nilfs_segctor_wakeup(sci, err);
2350 		sci->sc_flush_request = 0;
2351 	} else {
2352 		if (mode == SC_FLUSH_FILE)
2353 			sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2354 		else if (mode == SC_FLUSH_DAT)
2355 			sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2356 
2357 		/* re-enable timer if checkpoint creation was not done */
2358 		if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2359 		    time_before(jiffies, sci->sc_timer.expires))
2360 			add_timer(&sci->sc_timer);
2361 	}
2362 	spin_unlock(&sci->sc_state_lock);
2363 }
2364 
2365 /**
2366  * nilfs_segctor_construct - form logs and write them to disk
2367  * @sci: segment constructor object
2368  * @mode: mode of log forming
2369  */
2370 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2371 {
2372 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2373 	struct nilfs_super_block **sbp;
2374 	int err = 0;
2375 
2376 	nilfs_segctor_accept(sci);
2377 
2378 	if (nilfs_discontinued(nilfs))
2379 		mode = SC_LSEG_SR;
2380 	if (!nilfs_segctor_confirm(sci))
2381 		err = nilfs_segctor_do_construct(sci, mode);
2382 
2383 	if (likely(!err)) {
2384 		if (mode != SC_FLUSH_DAT)
2385 			atomic_set(&nilfs->ns_ndirtyblks, 0);
2386 		if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2387 		    nilfs_discontinued(nilfs)) {
2388 			down_write(&nilfs->ns_sem);
2389 			err = -EIO;
2390 			sbp = nilfs_prepare_super(sci->sc_super,
2391 						  nilfs_sb_will_flip(nilfs));
2392 			if (likely(sbp)) {
2393 				nilfs_set_log_cursor(sbp[0], nilfs);
2394 				err = nilfs_commit_super(sci->sc_super,
2395 							 NILFS_SB_COMMIT);
2396 			}
2397 			up_write(&nilfs->ns_sem);
2398 		}
2399 	}
2400 
2401 	nilfs_segctor_notify(sci, mode, err);
2402 	return err;
2403 }
2404 
2405 static void nilfs_construction_timeout(struct timer_list *t)
2406 {
2407 	struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2408 
2409 	wake_up_process(sci->sc_timer_task);
2410 }
2411 
2412 static void
2413 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2414 {
2415 	struct nilfs_inode_info *ii, *n;
2416 
2417 	list_for_each_entry_safe(ii, n, head, i_dirty) {
2418 		if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2419 			continue;
2420 		list_del_init(&ii->i_dirty);
2421 		truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2422 		nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2423 		iput(&ii->vfs_inode);
2424 	}
2425 }
2426 
2427 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2428 			 void **kbufs)
2429 {
2430 	struct the_nilfs *nilfs = sb->s_fs_info;
2431 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2432 	struct nilfs_transaction_info ti;
2433 	int err;
2434 
2435 	if (unlikely(!sci))
2436 		return -EROFS;
2437 
2438 	nilfs_transaction_lock(sb, &ti, 1);
2439 
2440 	err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2441 	if (unlikely(err))
2442 		goto out_unlock;
2443 
2444 	err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2445 	if (unlikely(err)) {
2446 		nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2447 		goto out_unlock;
2448 	}
2449 
2450 	sci->sc_freesegs = kbufs[4];
2451 	sci->sc_nfreesegs = argv[4].v_nmembs;
2452 	list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2453 
2454 	for (;;) {
2455 		err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2456 		nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2457 
2458 		if (likely(!err))
2459 			break;
2460 
2461 		nilfs_warn(sb, "error %d cleaning segments", err);
2462 		set_current_state(TASK_INTERRUPTIBLE);
2463 		schedule_timeout(sci->sc_interval);
2464 	}
2465 	if (nilfs_test_opt(nilfs, DISCARD)) {
2466 		int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2467 						 sci->sc_nfreesegs);
2468 		if (ret) {
2469 			nilfs_warn(sb,
2470 				   "error %d on discard request, turning discards off for the device",
2471 				   ret);
2472 			nilfs_clear_opt(nilfs, DISCARD);
2473 		}
2474 	}
2475 
2476  out_unlock:
2477 	sci->sc_freesegs = NULL;
2478 	sci->sc_nfreesegs = 0;
2479 	nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2480 	nilfs_transaction_unlock(sb);
2481 	return err;
2482 }
2483 
2484 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2485 {
2486 	struct nilfs_transaction_info ti;
2487 
2488 	nilfs_transaction_lock(sci->sc_super, &ti, 0);
2489 	nilfs_segctor_construct(sci, mode);
2490 
2491 	/*
2492 	 * Unclosed segment should be retried.  We do this using sc_timer.
2493 	 * Timeout of sc_timer will invoke complete construction which leads
2494 	 * to close the current logical segment.
2495 	 */
2496 	if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2497 		nilfs_segctor_start_timer(sci);
2498 
2499 	nilfs_transaction_unlock(sci->sc_super);
2500 }
2501 
2502 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2503 {
2504 	int mode = 0;
2505 
2506 	spin_lock(&sci->sc_state_lock);
2507 	mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2508 		SC_FLUSH_DAT : SC_FLUSH_FILE;
2509 	spin_unlock(&sci->sc_state_lock);
2510 
2511 	if (mode) {
2512 		nilfs_segctor_do_construct(sci, mode);
2513 
2514 		spin_lock(&sci->sc_state_lock);
2515 		sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2516 			~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2517 		spin_unlock(&sci->sc_state_lock);
2518 	}
2519 	clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2520 }
2521 
2522 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2523 {
2524 	if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2525 	    time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2526 		if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2527 			return SC_FLUSH_FILE;
2528 		else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2529 			return SC_FLUSH_DAT;
2530 	}
2531 	return SC_LSEG_SR;
2532 }
2533 
2534 /**
2535  * nilfs_segctor_thread - main loop of the segment constructor thread.
2536  * @arg: pointer to a struct nilfs_sc_info.
2537  *
2538  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2539  * to execute segment constructions.
2540  */
2541 static int nilfs_segctor_thread(void *arg)
2542 {
2543 	struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2544 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2545 	int timeout = 0;
2546 
2547 	sci->sc_timer_task = current;
2548 
2549 	/* start sync. */
2550 	sci->sc_task = current;
2551 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2552 	nilfs_info(sci->sc_super,
2553 		   "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2554 		   sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2555 
2556 	spin_lock(&sci->sc_state_lock);
2557  loop:
2558 	for (;;) {
2559 		int mode;
2560 
2561 		if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2562 			goto end_thread;
2563 
2564 		if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2565 			mode = SC_LSEG_SR;
2566 		else if (sci->sc_flush_request)
2567 			mode = nilfs_segctor_flush_mode(sci);
2568 		else
2569 			break;
2570 
2571 		spin_unlock(&sci->sc_state_lock);
2572 		nilfs_segctor_thread_construct(sci, mode);
2573 		spin_lock(&sci->sc_state_lock);
2574 		timeout = 0;
2575 	}
2576 
2577 
2578 	if (freezing(current)) {
2579 		spin_unlock(&sci->sc_state_lock);
2580 		try_to_freeze();
2581 		spin_lock(&sci->sc_state_lock);
2582 	} else {
2583 		DEFINE_WAIT(wait);
2584 		int should_sleep = 1;
2585 
2586 		prepare_to_wait(&sci->sc_wait_daemon, &wait,
2587 				TASK_INTERRUPTIBLE);
2588 
2589 		if (sci->sc_seq_request != sci->sc_seq_done)
2590 			should_sleep = 0;
2591 		else if (sci->sc_flush_request)
2592 			should_sleep = 0;
2593 		else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2594 			should_sleep = time_before(jiffies,
2595 					sci->sc_timer.expires);
2596 
2597 		if (should_sleep) {
2598 			spin_unlock(&sci->sc_state_lock);
2599 			schedule();
2600 			spin_lock(&sci->sc_state_lock);
2601 		}
2602 		finish_wait(&sci->sc_wait_daemon, &wait);
2603 		timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2604 			   time_after_eq(jiffies, sci->sc_timer.expires));
2605 
2606 		if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2607 			set_nilfs_discontinued(nilfs);
2608 	}
2609 	goto loop;
2610 
2611  end_thread:
2612 	spin_unlock(&sci->sc_state_lock);
2613 
2614 	/* end sync. */
2615 	sci->sc_task = NULL;
2616 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2617 	return 0;
2618 }
2619 
2620 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2621 {
2622 	struct task_struct *t;
2623 
2624 	t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2625 	if (IS_ERR(t)) {
2626 		int err = PTR_ERR(t);
2627 
2628 		nilfs_err(sci->sc_super, "error %d creating segctord thread",
2629 			  err);
2630 		return err;
2631 	}
2632 	wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2633 	return 0;
2634 }
2635 
2636 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2637 	__acquires(&sci->sc_state_lock)
2638 	__releases(&sci->sc_state_lock)
2639 {
2640 	sci->sc_state |= NILFS_SEGCTOR_QUIT;
2641 
2642 	while (sci->sc_task) {
2643 		wake_up(&sci->sc_wait_daemon);
2644 		spin_unlock(&sci->sc_state_lock);
2645 		wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2646 		spin_lock(&sci->sc_state_lock);
2647 	}
2648 }
2649 
2650 /*
2651  * Setup & clean-up functions
2652  */
2653 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2654 					       struct nilfs_root *root)
2655 {
2656 	struct the_nilfs *nilfs = sb->s_fs_info;
2657 	struct nilfs_sc_info *sci;
2658 
2659 	sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2660 	if (!sci)
2661 		return NULL;
2662 
2663 	sci->sc_super = sb;
2664 
2665 	nilfs_get_root(root);
2666 	sci->sc_root = root;
2667 
2668 	init_waitqueue_head(&sci->sc_wait_request);
2669 	init_waitqueue_head(&sci->sc_wait_daemon);
2670 	init_waitqueue_head(&sci->sc_wait_task);
2671 	spin_lock_init(&sci->sc_state_lock);
2672 	INIT_LIST_HEAD(&sci->sc_dirty_files);
2673 	INIT_LIST_HEAD(&sci->sc_segbufs);
2674 	INIT_LIST_HEAD(&sci->sc_write_logs);
2675 	INIT_LIST_HEAD(&sci->sc_gc_inodes);
2676 	INIT_LIST_HEAD(&sci->sc_iput_queue);
2677 	INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2678 	timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2679 
2680 	sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2681 	sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2682 	sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2683 
2684 	if (nilfs->ns_interval)
2685 		sci->sc_interval = HZ * nilfs->ns_interval;
2686 	if (nilfs->ns_watermark)
2687 		sci->sc_watermark = nilfs->ns_watermark;
2688 	return sci;
2689 }
2690 
2691 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2692 {
2693 	int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2694 
2695 	/*
2696 	 * The segctord thread was stopped and its timer was removed.
2697 	 * But some tasks remain.
2698 	 */
2699 	do {
2700 		struct nilfs_transaction_info ti;
2701 
2702 		nilfs_transaction_lock(sci->sc_super, &ti, 0);
2703 		ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2704 		nilfs_transaction_unlock(sci->sc_super);
2705 
2706 		flush_work(&sci->sc_iput_work);
2707 
2708 	} while (ret && retrycount-- > 0);
2709 }
2710 
2711 /**
2712  * nilfs_segctor_destroy - destroy the segment constructor.
2713  * @sci: nilfs_sc_info
2714  *
2715  * nilfs_segctor_destroy() kills the segctord thread and frees
2716  * the nilfs_sc_info struct.
2717  * Caller must hold the segment semaphore.
2718  */
2719 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2720 {
2721 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2722 	int flag;
2723 
2724 	up_write(&nilfs->ns_segctor_sem);
2725 
2726 	spin_lock(&sci->sc_state_lock);
2727 	nilfs_segctor_kill_thread(sci);
2728 	flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2729 		|| sci->sc_seq_request != sci->sc_seq_done);
2730 	spin_unlock(&sci->sc_state_lock);
2731 
2732 	if (flush_work(&sci->sc_iput_work))
2733 		flag = true;
2734 
2735 	if (flag || !nilfs_segctor_confirm(sci))
2736 		nilfs_segctor_write_out(sci);
2737 
2738 	if (!list_empty(&sci->sc_dirty_files)) {
2739 		nilfs_warn(sci->sc_super,
2740 			   "disposed unprocessed dirty file(s) when stopping log writer");
2741 		nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2742 	}
2743 
2744 	if (!list_empty(&sci->sc_iput_queue)) {
2745 		nilfs_warn(sci->sc_super,
2746 			   "disposed unprocessed inode(s) in iput queue when stopping log writer");
2747 		nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2748 	}
2749 
2750 	WARN_ON(!list_empty(&sci->sc_segbufs));
2751 	WARN_ON(!list_empty(&sci->sc_write_logs));
2752 
2753 	nilfs_put_root(sci->sc_root);
2754 
2755 	down_write(&nilfs->ns_segctor_sem);
2756 
2757 	timer_shutdown_sync(&sci->sc_timer);
2758 	kfree(sci);
2759 }
2760 
2761 /**
2762  * nilfs_attach_log_writer - attach log writer
2763  * @sb: super block instance
2764  * @root: root object of the current filesystem tree
2765  *
2766  * This allocates a log writer object, initializes it, and starts the
2767  * log writer.
2768  *
2769  * Return Value: On success, 0 is returned. On error, one of the following
2770  * negative error code is returned.
2771  *
2772  * %-ENOMEM - Insufficient memory available.
2773  */
2774 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2775 {
2776 	struct the_nilfs *nilfs = sb->s_fs_info;
2777 	int err;
2778 
2779 	if (nilfs->ns_writer) {
2780 		/*
2781 		 * This happens if the filesystem is made read-only by
2782 		 * __nilfs_error or nilfs_remount and then remounted
2783 		 * read/write.  In these cases, reuse the existing
2784 		 * writer.
2785 		 */
2786 		return 0;
2787 	}
2788 
2789 	nilfs->ns_writer = nilfs_segctor_new(sb, root);
2790 	if (!nilfs->ns_writer)
2791 		return -ENOMEM;
2792 
2793 	inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2794 
2795 	err = nilfs_segctor_start_thread(nilfs->ns_writer);
2796 	if (unlikely(err))
2797 		nilfs_detach_log_writer(sb);
2798 
2799 	return err;
2800 }
2801 
2802 /**
2803  * nilfs_detach_log_writer - destroy log writer
2804  * @sb: super block instance
2805  *
2806  * This kills log writer daemon, frees the log writer object, and
2807  * destroys list of dirty files.
2808  */
2809 void nilfs_detach_log_writer(struct super_block *sb)
2810 {
2811 	struct the_nilfs *nilfs = sb->s_fs_info;
2812 	LIST_HEAD(garbage_list);
2813 
2814 	down_write(&nilfs->ns_segctor_sem);
2815 	if (nilfs->ns_writer) {
2816 		nilfs_segctor_destroy(nilfs->ns_writer);
2817 		nilfs->ns_writer = NULL;
2818 	}
2819 
2820 	/* Force to free the list of dirty files */
2821 	spin_lock(&nilfs->ns_inode_lock);
2822 	if (!list_empty(&nilfs->ns_dirty_files)) {
2823 		list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2824 		nilfs_warn(sb,
2825 			   "disposed unprocessed dirty file(s) when detaching log writer");
2826 	}
2827 	spin_unlock(&nilfs->ns_inode_lock);
2828 	up_write(&nilfs->ns_segctor_sem);
2829 
2830 	nilfs_dispose_list(nilfs, &garbage_list, 1);
2831 }
2832