xref: /openbmc/linux/fs/nilfs2/segment.c (revision 8e9356c6)
1 /*
2  * segment.c - NILFS segment constructor.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  *
22  */
23 
24 #include <linux/pagemap.h>
25 #include <linux/buffer_head.h>
26 #include <linux/writeback.h>
27 #include <linux/bio.h>
28 #include <linux/completion.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/freezer.h>
32 #include <linux/kthread.h>
33 #include <linux/crc32.h>
34 #include <linux/pagevec.h>
35 #include <linux/slab.h>
36 #include "nilfs.h"
37 #include "btnode.h"
38 #include "page.h"
39 #include "segment.h"
40 #include "sufile.h"
41 #include "cpfile.h"
42 #include "ifile.h"
43 #include "segbuf.h"
44 
45 
46 /*
47  * Segment constructor
48  */
49 #define SC_N_INODEVEC	16   /* Size of locally allocated inode vector */
50 
51 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
52 				appended in collection retry loop */
53 
54 /* Construction mode */
55 enum {
56 	SC_LSEG_SR = 1,	/* Make a logical segment having a super root */
57 	SC_LSEG_DSYNC,	/* Flush data blocks of a given file and make
58 			   a logical segment without a super root */
59 	SC_FLUSH_FILE,	/* Flush data files, leads to segment writes without
60 			   creating a checkpoint */
61 	SC_FLUSH_DAT,	/* Flush DAT file. This also creates segments without
62 			   a checkpoint */
63 };
64 
65 /* Stage numbers of dirty block collection */
66 enum {
67 	NILFS_ST_INIT = 0,
68 	NILFS_ST_GC,		/* Collecting dirty blocks for GC */
69 	NILFS_ST_FILE,
70 	NILFS_ST_IFILE,
71 	NILFS_ST_CPFILE,
72 	NILFS_ST_SUFILE,
73 	NILFS_ST_DAT,
74 	NILFS_ST_SR,		/* Super root */
75 	NILFS_ST_DSYNC,		/* Data sync blocks */
76 	NILFS_ST_DONE,
77 };
78 
79 /* State flags of collection */
80 #define NILFS_CF_NODE		0x0001	/* Collecting node blocks */
81 #define NILFS_CF_IFILE_STARTED	0x0002	/* IFILE stage has started */
82 #define NILFS_CF_SUFREED	0x0004	/* segment usages has been freed */
83 #define NILFS_CF_HISTORY_MASK	(NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
84 
85 /* Operations depending on the construction mode and file type */
86 struct nilfs_sc_operations {
87 	int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
88 			    struct inode *);
89 	int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
90 			    struct inode *);
91 	int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
92 			    struct inode *);
93 	void (*write_data_binfo)(struct nilfs_sc_info *,
94 				 struct nilfs_segsum_pointer *,
95 				 union nilfs_binfo *);
96 	void (*write_node_binfo)(struct nilfs_sc_info *,
97 				 struct nilfs_segsum_pointer *,
98 				 union nilfs_binfo *);
99 };
100 
101 /*
102  * Other definitions
103  */
104 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
105 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
107 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
108 
109 #define nilfs_cnt32_gt(a, b)   \
110 	(typecheck(__u32, a) && typecheck(__u32, b) && \
111 	 ((__s32)(b) - (__s32)(a) < 0))
112 #define nilfs_cnt32_ge(a, b)   \
113 	(typecheck(__u32, a) && typecheck(__u32, b) && \
114 	 ((__s32)(a) - (__s32)(b) >= 0))
115 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
116 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
117 
118 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
119 {
120 	struct nilfs_transaction_info *cur_ti = current->journal_info;
121 	void *save = NULL;
122 
123 	if (cur_ti) {
124 		if (cur_ti->ti_magic == NILFS_TI_MAGIC)
125 			return ++cur_ti->ti_count;
126 		else {
127 			/*
128 			 * If journal_info field is occupied by other FS,
129 			 * it is saved and will be restored on
130 			 * nilfs_transaction_commit().
131 			 */
132 			printk(KERN_WARNING
133 			       "NILFS warning: journal info from a different "
134 			       "FS\n");
135 			save = current->journal_info;
136 		}
137 	}
138 	if (!ti) {
139 		ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
140 		if (!ti)
141 			return -ENOMEM;
142 		ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
143 	} else {
144 		ti->ti_flags = 0;
145 	}
146 	ti->ti_count = 0;
147 	ti->ti_save = save;
148 	ti->ti_magic = NILFS_TI_MAGIC;
149 	current->journal_info = ti;
150 	return 0;
151 }
152 
153 /**
154  * nilfs_transaction_begin - start indivisible file operations.
155  * @sb: super block
156  * @ti: nilfs_transaction_info
157  * @vacancy_check: flags for vacancy rate checks
158  *
159  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
160  * the segment semaphore, to make a segment construction and write tasks
161  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
162  * The region enclosed by these two functions can be nested.  To avoid a
163  * deadlock, the semaphore is only acquired or released in the outermost call.
164  *
165  * This function allocates a nilfs_transaction_info struct to keep context
166  * information on it.  It is initialized and hooked onto the current task in
167  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
168  * instead; otherwise a new struct is assigned from a slab.
169  *
170  * When @vacancy_check flag is set, this function will check the amount of
171  * free space, and will wait for the GC to reclaim disk space if low capacity.
172  *
173  * Return Value: On success, 0 is returned. On error, one of the following
174  * negative error code is returned.
175  *
176  * %-ENOMEM - Insufficient memory available.
177  *
178  * %-ENOSPC - No space left on device
179  */
180 int nilfs_transaction_begin(struct super_block *sb,
181 			    struct nilfs_transaction_info *ti,
182 			    int vacancy_check)
183 {
184 	struct the_nilfs *nilfs;
185 	int ret = nilfs_prepare_segment_lock(ti);
186 
187 	if (unlikely(ret < 0))
188 		return ret;
189 	if (ret > 0)
190 		return 0;
191 
192 	sb_start_intwrite(sb);
193 
194 	nilfs = sb->s_fs_info;
195 	down_read(&nilfs->ns_segctor_sem);
196 	if (vacancy_check && nilfs_near_disk_full(nilfs)) {
197 		up_read(&nilfs->ns_segctor_sem);
198 		ret = -ENOSPC;
199 		goto failed;
200 	}
201 	return 0;
202 
203  failed:
204 	ti = current->journal_info;
205 	current->journal_info = ti->ti_save;
206 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
207 		kmem_cache_free(nilfs_transaction_cachep, ti);
208 	sb_end_intwrite(sb);
209 	return ret;
210 }
211 
212 /**
213  * nilfs_transaction_commit - commit indivisible file operations.
214  * @sb: super block
215  *
216  * nilfs_transaction_commit() releases the read semaphore which is
217  * acquired by nilfs_transaction_begin(). This is only performed
218  * in outermost call of this function.  If a commit flag is set,
219  * nilfs_transaction_commit() sets a timer to start the segment
220  * constructor.  If a sync flag is set, it starts construction
221  * directly.
222  */
223 int nilfs_transaction_commit(struct super_block *sb)
224 {
225 	struct nilfs_transaction_info *ti = current->journal_info;
226 	struct the_nilfs *nilfs = sb->s_fs_info;
227 	int err = 0;
228 
229 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
230 	ti->ti_flags |= NILFS_TI_COMMIT;
231 	if (ti->ti_count > 0) {
232 		ti->ti_count--;
233 		return 0;
234 	}
235 	if (nilfs->ns_writer) {
236 		struct nilfs_sc_info *sci = nilfs->ns_writer;
237 
238 		if (ti->ti_flags & NILFS_TI_COMMIT)
239 			nilfs_segctor_start_timer(sci);
240 		if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
241 			nilfs_segctor_do_flush(sci, 0);
242 	}
243 	up_read(&nilfs->ns_segctor_sem);
244 	current->journal_info = ti->ti_save;
245 
246 	if (ti->ti_flags & NILFS_TI_SYNC)
247 		err = nilfs_construct_segment(sb);
248 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
249 		kmem_cache_free(nilfs_transaction_cachep, ti);
250 	sb_end_intwrite(sb);
251 	return err;
252 }
253 
254 void nilfs_transaction_abort(struct super_block *sb)
255 {
256 	struct nilfs_transaction_info *ti = current->journal_info;
257 	struct the_nilfs *nilfs = sb->s_fs_info;
258 
259 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
260 	if (ti->ti_count > 0) {
261 		ti->ti_count--;
262 		return;
263 	}
264 	up_read(&nilfs->ns_segctor_sem);
265 
266 	current->journal_info = ti->ti_save;
267 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
268 		kmem_cache_free(nilfs_transaction_cachep, ti);
269 	sb_end_intwrite(sb);
270 }
271 
272 void nilfs_relax_pressure_in_lock(struct super_block *sb)
273 {
274 	struct the_nilfs *nilfs = sb->s_fs_info;
275 	struct nilfs_sc_info *sci = nilfs->ns_writer;
276 
277 	if (!sci || !sci->sc_flush_request)
278 		return;
279 
280 	set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
281 	up_read(&nilfs->ns_segctor_sem);
282 
283 	down_write(&nilfs->ns_segctor_sem);
284 	if (sci->sc_flush_request &&
285 	    test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
286 		struct nilfs_transaction_info *ti = current->journal_info;
287 
288 		ti->ti_flags |= NILFS_TI_WRITER;
289 		nilfs_segctor_do_immediate_flush(sci);
290 		ti->ti_flags &= ~NILFS_TI_WRITER;
291 	}
292 	downgrade_write(&nilfs->ns_segctor_sem);
293 }
294 
295 static void nilfs_transaction_lock(struct super_block *sb,
296 				   struct nilfs_transaction_info *ti,
297 				   int gcflag)
298 {
299 	struct nilfs_transaction_info *cur_ti = current->journal_info;
300 	struct the_nilfs *nilfs = sb->s_fs_info;
301 	struct nilfs_sc_info *sci = nilfs->ns_writer;
302 
303 	WARN_ON(cur_ti);
304 	ti->ti_flags = NILFS_TI_WRITER;
305 	ti->ti_count = 0;
306 	ti->ti_save = cur_ti;
307 	ti->ti_magic = NILFS_TI_MAGIC;
308 	INIT_LIST_HEAD(&ti->ti_garbage);
309 	current->journal_info = ti;
310 
311 	for (;;) {
312 		down_write(&nilfs->ns_segctor_sem);
313 		if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
314 			break;
315 
316 		nilfs_segctor_do_immediate_flush(sci);
317 
318 		up_write(&nilfs->ns_segctor_sem);
319 		yield();
320 	}
321 	if (gcflag)
322 		ti->ti_flags |= NILFS_TI_GC;
323 }
324 
325 static void nilfs_transaction_unlock(struct super_block *sb)
326 {
327 	struct nilfs_transaction_info *ti = current->journal_info;
328 	struct the_nilfs *nilfs = sb->s_fs_info;
329 
330 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
331 	BUG_ON(ti->ti_count > 0);
332 
333 	up_write(&nilfs->ns_segctor_sem);
334 	current->journal_info = ti->ti_save;
335 	if (!list_empty(&ti->ti_garbage))
336 		nilfs_dispose_list(nilfs, &ti->ti_garbage, 0);
337 }
338 
339 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
340 					    struct nilfs_segsum_pointer *ssp,
341 					    unsigned bytes)
342 {
343 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
344 	unsigned blocksize = sci->sc_super->s_blocksize;
345 	void *p;
346 
347 	if (unlikely(ssp->offset + bytes > blocksize)) {
348 		ssp->offset = 0;
349 		BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
350 					       &segbuf->sb_segsum_buffers));
351 		ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
352 	}
353 	p = ssp->bh->b_data + ssp->offset;
354 	ssp->offset += bytes;
355 	return p;
356 }
357 
358 /**
359  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
360  * @sci: nilfs_sc_info
361  */
362 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
363 {
364 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
365 	struct buffer_head *sumbh;
366 	unsigned sumbytes;
367 	unsigned flags = 0;
368 	int err;
369 
370 	if (nilfs_doing_gc())
371 		flags = NILFS_SS_GC;
372 	err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
373 	if (unlikely(err))
374 		return err;
375 
376 	sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
377 	sumbytes = segbuf->sb_sum.sumbytes;
378 	sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
379 	sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
380 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
381 	return 0;
382 }
383 
384 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
385 {
386 	sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
387 	if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
388 		return -E2BIG; /* The current segment is filled up
389 				  (internal code) */
390 	sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
391 	return nilfs_segctor_reset_segment_buffer(sci);
392 }
393 
394 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
395 {
396 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
397 	int err;
398 
399 	if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
400 		err = nilfs_segctor_feed_segment(sci);
401 		if (err)
402 			return err;
403 		segbuf = sci->sc_curseg;
404 	}
405 	err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
406 	if (likely(!err))
407 		segbuf->sb_sum.flags |= NILFS_SS_SR;
408 	return err;
409 }
410 
411 /*
412  * Functions for making segment summary and payloads
413  */
414 static int nilfs_segctor_segsum_block_required(
415 	struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
416 	unsigned binfo_size)
417 {
418 	unsigned blocksize = sci->sc_super->s_blocksize;
419 	/* Size of finfo and binfo is enough small against blocksize */
420 
421 	return ssp->offset + binfo_size +
422 		(!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
423 		blocksize;
424 }
425 
426 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
427 				      struct inode *inode)
428 {
429 	sci->sc_curseg->sb_sum.nfinfo++;
430 	sci->sc_binfo_ptr = sci->sc_finfo_ptr;
431 	nilfs_segctor_map_segsum_entry(
432 		sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
433 
434 	if (NILFS_I(inode)->i_root &&
435 	    !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
436 		set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
437 	/* skip finfo */
438 }
439 
440 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
441 				    struct inode *inode)
442 {
443 	struct nilfs_finfo *finfo;
444 	struct nilfs_inode_info *ii;
445 	struct nilfs_segment_buffer *segbuf;
446 	__u64 cno;
447 
448 	if (sci->sc_blk_cnt == 0)
449 		return;
450 
451 	ii = NILFS_I(inode);
452 
453 	if (test_bit(NILFS_I_GCINODE, &ii->i_state))
454 		cno = ii->i_cno;
455 	else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
456 		cno = 0;
457 	else
458 		cno = sci->sc_cno;
459 
460 	finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
461 						 sizeof(*finfo));
462 	finfo->fi_ino = cpu_to_le64(inode->i_ino);
463 	finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
464 	finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
465 	finfo->fi_cno = cpu_to_le64(cno);
466 
467 	segbuf = sci->sc_curseg;
468 	segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
469 		sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
470 	sci->sc_finfo_ptr = sci->sc_binfo_ptr;
471 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
472 }
473 
474 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
475 					struct buffer_head *bh,
476 					struct inode *inode,
477 					unsigned binfo_size)
478 {
479 	struct nilfs_segment_buffer *segbuf;
480 	int required, err = 0;
481 
482  retry:
483 	segbuf = sci->sc_curseg;
484 	required = nilfs_segctor_segsum_block_required(
485 		sci, &sci->sc_binfo_ptr, binfo_size);
486 	if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
487 		nilfs_segctor_end_finfo(sci, inode);
488 		err = nilfs_segctor_feed_segment(sci);
489 		if (err)
490 			return err;
491 		goto retry;
492 	}
493 	if (unlikely(required)) {
494 		err = nilfs_segbuf_extend_segsum(segbuf);
495 		if (unlikely(err))
496 			goto failed;
497 	}
498 	if (sci->sc_blk_cnt == 0)
499 		nilfs_segctor_begin_finfo(sci, inode);
500 
501 	nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
502 	/* Substitution to vblocknr is delayed until update_blocknr() */
503 	nilfs_segbuf_add_file_buffer(segbuf, bh);
504 	sci->sc_blk_cnt++;
505  failed:
506 	return err;
507 }
508 
509 /*
510  * Callback functions that enumerate, mark, and collect dirty blocks
511  */
512 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
513 				   struct buffer_head *bh, struct inode *inode)
514 {
515 	int err;
516 
517 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
518 	if (err < 0)
519 		return err;
520 
521 	err = nilfs_segctor_add_file_block(sci, bh, inode,
522 					   sizeof(struct nilfs_binfo_v));
523 	if (!err)
524 		sci->sc_datablk_cnt++;
525 	return err;
526 }
527 
528 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
529 				   struct buffer_head *bh,
530 				   struct inode *inode)
531 {
532 	return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
533 }
534 
535 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
536 				   struct buffer_head *bh,
537 				   struct inode *inode)
538 {
539 	WARN_ON(!buffer_dirty(bh));
540 	return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
541 }
542 
543 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
544 					struct nilfs_segsum_pointer *ssp,
545 					union nilfs_binfo *binfo)
546 {
547 	struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
548 		sci, ssp, sizeof(*binfo_v));
549 	*binfo_v = binfo->bi_v;
550 }
551 
552 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
553 					struct nilfs_segsum_pointer *ssp,
554 					union nilfs_binfo *binfo)
555 {
556 	__le64 *vblocknr = nilfs_segctor_map_segsum_entry(
557 		sci, ssp, sizeof(*vblocknr));
558 	*vblocknr = binfo->bi_v.bi_vblocknr;
559 }
560 
561 static struct nilfs_sc_operations nilfs_sc_file_ops = {
562 	.collect_data = nilfs_collect_file_data,
563 	.collect_node = nilfs_collect_file_node,
564 	.collect_bmap = nilfs_collect_file_bmap,
565 	.write_data_binfo = nilfs_write_file_data_binfo,
566 	.write_node_binfo = nilfs_write_file_node_binfo,
567 };
568 
569 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
570 				  struct buffer_head *bh, struct inode *inode)
571 {
572 	int err;
573 
574 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
575 	if (err < 0)
576 		return err;
577 
578 	err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
579 	if (!err)
580 		sci->sc_datablk_cnt++;
581 	return err;
582 }
583 
584 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
585 				  struct buffer_head *bh, struct inode *inode)
586 {
587 	WARN_ON(!buffer_dirty(bh));
588 	return nilfs_segctor_add_file_block(sci, bh, inode,
589 					    sizeof(struct nilfs_binfo_dat));
590 }
591 
592 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
593 				       struct nilfs_segsum_pointer *ssp,
594 				       union nilfs_binfo *binfo)
595 {
596 	__le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
597 							  sizeof(*blkoff));
598 	*blkoff = binfo->bi_dat.bi_blkoff;
599 }
600 
601 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
602 				       struct nilfs_segsum_pointer *ssp,
603 				       union nilfs_binfo *binfo)
604 {
605 	struct nilfs_binfo_dat *binfo_dat =
606 		nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
607 	*binfo_dat = binfo->bi_dat;
608 }
609 
610 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
611 	.collect_data = nilfs_collect_dat_data,
612 	.collect_node = nilfs_collect_file_node,
613 	.collect_bmap = nilfs_collect_dat_bmap,
614 	.write_data_binfo = nilfs_write_dat_data_binfo,
615 	.write_node_binfo = nilfs_write_dat_node_binfo,
616 };
617 
618 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
619 	.collect_data = nilfs_collect_file_data,
620 	.collect_node = NULL,
621 	.collect_bmap = NULL,
622 	.write_data_binfo = nilfs_write_file_data_binfo,
623 	.write_node_binfo = NULL,
624 };
625 
626 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
627 					      struct list_head *listp,
628 					      size_t nlimit,
629 					      loff_t start, loff_t end)
630 {
631 	struct address_space *mapping = inode->i_mapping;
632 	struct pagevec pvec;
633 	pgoff_t index = 0, last = ULONG_MAX;
634 	size_t ndirties = 0;
635 	int i;
636 
637 	if (unlikely(start != 0 || end != LLONG_MAX)) {
638 		/*
639 		 * A valid range is given for sync-ing data pages. The
640 		 * range is rounded to per-page; extra dirty buffers
641 		 * may be included if blocksize < pagesize.
642 		 */
643 		index = start >> PAGE_SHIFT;
644 		last = end >> PAGE_SHIFT;
645 	}
646 	pagevec_init(&pvec, 0);
647  repeat:
648 	if (unlikely(index > last) ||
649 	    !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
650 				min_t(pgoff_t, last - index,
651 				      PAGEVEC_SIZE - 1) + 1))
652 		return ndirties;
653 
654 	for (i = 0; i < pagevec_count(&pvec); i++) {
655 		struct buffer_head *bh, *head;
656 		struct page *page = pvec.pages[i];
657 
658 		if (unlikely(page->index > last))
659 			break;
660 
661 		lock_page(page);
662 		if (!page_has_buffers(page))
663 			create_empty_buffers(page, 1 << inode->i_blkbits, 0);
664 		unlock_page(page);
665 
666 		bh = head = page_buffers(page);
667 		do {
668 			if (!buffer_dirty(bh) || buffer_async_write(bh))
669 				continue;
670 			get_bh(bh);
671 			list_add_tail(&bh->b_assoc_buffers, listp);
672 			ndirties++;
673 			if (unlikely(ndirties >= nlimit)) {
674 				pagevec_release(&pvec);
675 				cond_resched();
676 				return ndirties;
677 			}
678 		} while (bh = bh->b_this_page, bh != head);
679 	}
680 	pagevec_release(&pvec);
681 	cond_resched();
682 	goto repeat;
683 }
684 
685 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
686 					    struct list_head *listp)
687 {
688 	struct nilfs_inode_info *ii = NILFS_I(inode);
689 	struct address_space *mapping = &ii->i_btnode_cache;
690 	struct pagevec pvec;
691 	struct buffer_head *bh, *head;
692 	unsigned int i;
693 	pgoff_t index = 0;
694 
695 	pagevec_init(&pvec, 0);
696 
697 	while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
698 				  PAGEVEC_SIZE)) {
699 		for (i = 0; i < pagevec_count(&pvec); i++) {
700 			bh = head = page_buffers(pvec.pages[i]);
701 			do {
702 				if (buffer_dirty(bh) &&
703 						!buffer_async_write(bh)) {
704 					get_bh(bh);
705 					list_add_tail(&bh->b_assoc_buffers,
706 						      listp);
707 				}
708 				bh = bh->b_this_page;
709 			} while (bh != head);
710 		}
711 		pagevec_release(&pvec);
712 		cond_resched();
713 	}
714 }
715 
716 static void nilfs_dispose_list(struct the_nilfs *nilfs,
717 			       struct list_head *head, int force)
718 {
719 	struct nilfs_inode_info *ii, *n;
720 	struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
721 	unsigned nv = 0;
722 
723 	while (!list_empty(head)) {
724 		spin_lock(&nilfs->ns_inode_lock);
725 		list_for_each_entry_safe(ii, n, head, i_dirty) {
726 			list_del_init(&ii->i_dirty);
727 			if (force) {
728 				if (unlikely(ii->i_bh)) {
729 					brelse(ii->i_bh);
730 					ii->i_bh = NULL;
731 				}
732 			} else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
733 				set_bit(NILFS_I_QUEUED, &ii->i_state);
734 				list_add_tail(&ii->i_dirty,
735 					      &nilfs->ns_dirty_files);
736 				continue;
737 			}
738 			ivec[nv++] = ii;
739 			if (nv == SC_N_INODEVEC)
740 				break;
741 		}
742 		spin_unlock(&nilfs->ns_inode_lock);
743 
744 		for (pii = ivec; nv > 0; pii++, nv--)
745 			iput(&(*pii)->vfs_inode);
746 	}
747 }
748 
749 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
750 				     struct nilfs_root *root)
751 {
752 	int ret = 0;
753 
754 	if (nilfs_mdt_fetch_dirty(root->ifile))
755 		ret++;
756 	if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
757 		ret++;
758 	if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
759 		ret++;
760 	if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
761 		ret++;
762 	return ret;
763 }
764 
765 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
766 {
767 	return list_empty(&sci->sc_dirty_files) &&
768 		!test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
769 		sci->sc_nfreesegs == 0 &&
770 		(!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
771 }
772 
773 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
774 {
775 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
776 	int ret = 0;
777 
778 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
779 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
780 
781 	spin_lock(&nilfs->ns_inode_lock);
782 	if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
783 		ret++;
784 
785 	spin_unlock(&nilfs->ns_inode_lock);
786 	return ret;
787 }
788 
789 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
790 {
791 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
792 
793 	nilfs_mdt_clear_dirty(sci->sc_root->ifile);
794 	nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
795 	nilfs_mdt_clear_dirty(nilfs->ns_sufile);
796 	nilfs_mdt_clear_dirty(nilfs->ns_dat);
797 }
798 
799 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
800 {
801 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
802 	struct buffer_head *bh_cp;
803 	struct nilfs_checkpoint *raw_cp;
804 	int err;
805 
806 	/* XXX: this interface will be changed */
807 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
808 					  &raw_cp, &bh_cp);
809 	if (likely(!err)) {
810 		/* The following code is duplicated with cpfile.  But, it is
811 		   needed to collect the checkpoint even if it was not newly
812 		   created */
813 		mark_buffer_dirty(bh_cp);
814 		nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
815 		nilfs_cpfile_put_checkpoint(
816 			nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
817 	} else
818 		WARN_ON(err == -EINVAL || err == -ENOENT);
819 
820 	return err;
821 }
822 
823 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
824 {
825 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
826 	struct buffer_head *bh_cp;
827 	struct nilfs_checkpoint *raw_cp;
828 	int err;
829 
830 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
831 					  &raw_cp, &bh_cp);
832 	if (unlikely(err)) {
833 		WARN_ON(err == -EINVAL || err == -ENOENT);
834 		goto failed_ibh;
835 	}
836 	raw_cp->cp_snapshot_list.ssl_next = 0;
837 	raw_cp->cp_snapshot_list.ssl_prev = 0;
838 	raw_cp->cp_inodes_count =
839 		cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
840 	raw_cp->cp_blocks_count =
841 		cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
842 	raw_cp->cp_nblk_inc =
843 		cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
844 	raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
845 	raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
846 
847 	if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
848 		nilfs_checkpoint_clear_minor(raw_cp);
849 	else
850 		nilfs_checkpoint_set_minor(raw_cp);
851 
852 	nilfs_write_inode_common(sci->sc_root->ifile,
853 				 &raw_cp->cp_ifile_inode, 1);
854 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
855 	return 0;
856 
857  failed_ibh:
858 	return err;
859 }
860 
861 static void nilfs_fill_in_file_bmap(struct inode *ifile,
862 				    struct nilfs_inode_info *ii)
863 
864 {
865 	struct buffer_head *ibh;
866 	struct nilfs_inode *raw_inode;
867 
868 	if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
869 		ibh = ii->i_bh;
870 		BUG_ON(!ibh);
871 		raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
872 						  ibh);
873 		nilfs_bmap_write(ii->i_bmap, raw_inode);
874 		nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
875 	}
876 }
877 
878 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
879 {
880 	struct nilfs_inode_info *ii;
881 
882 	list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
883 		nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
884 		set_bit(NILFS_I_COLLECTED, &ii->i_state);
885 	}
886 }
887 
888 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
889 					     struct the_nilfs *nilfs)
890 {
891 	struct buffer_head *bh_sr;
892 	struct nilfs_super_root *raw_sr;
893 	unsigned isz, srsz;
894 
895 	bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
896 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
897 	isz = nilfs->ns_inode_size;
898 	srsz = NILFS_SR_BYTES(isz);
899 
900 	raw_sr->sr_bytes = cpu_to_le16(srsz);
901 	raw_sr->sr_nongc_ctime
902 		= cpu_to_le64(nilfs_doing_gc() ?
903 			      nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
904 	raw_sr->sr_flags = 0;
905 
906 	nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
907 				 NILFS_SR_DAT_OFFSET(isz), 1);
908 	nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
909 				 NILFS_SR_CPFILE_OFFSET(isz), 1);
910 	nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
911 				 NILFS_SR_SUFILE_OFFSET(isz), 1);
912 	memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
913 }
914 
915 static void nilfs_redirty_inodes(struct list_head *head)
916 {
917 	struct nilfs_inode_info *ii;
918 
919 	list_for_each_entry(ii, head, i_dirty) {
920 		if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
921 			clear_bit(NILFS_I_COLLECTED, &ii->i_state);
922 	}
923 }
924 
925 static void nilfs_drop_collected_inodes(struct list_head *head)
926 {
927 	struct nilfs_inode_info *ii;
928 
929 	list_for_each_entry(ii, head, i_dirty) {
930 		if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
931 			continue;
932 
933 		clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
934 		set_bit(NILFS_I_UPDATED, &ii->i_state);
935 	}
936 }
937 
938 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
939 				       struct inode *inode,
940 				       struct list_head *listp,
941 				       int (*collect)(struct nilfs_sc_info *,
942 						      struct buffer_head *,
943 						      struct inode *))
944 {
945 	struct buffer_head *bh, *n;
946 	int err = 0;
947 
948 	if (collect) {
949 		list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
950 			list_del_init(&bh->b_assoc_buffers);
951 			err = collect(sci, bh, inode);
952 			brelse(bh);
953 			if (unlikely(err))
954 				goto dispose_buffers;
955 		}
956 		return 0;
957 	}
958 
959  dispose_buffers:
960 	while (!list_empty(listp)) {
961 		bh = list_first_entry(listp, struct buffer_head,
962 				      b_assoc_buffers);
963 		list_del_init(&bh->b_assoc_buffers);
964 		brelse(bh);
965 	}
966 	return err;
967 }
968 
969 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
970 {
971 	/* Remaining number of blocks within segment buffer */
972 	return sci->sc_segbuf_nblocks -
973 		(sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
974 }
975 
976 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
977 				   struct inode *inode,
978 				   struct nilfs_sc_operations *sc_ops)
979 {
980 	LIST_HEAD(data_buffers);
981 	LIST_HEAD(node_buffers);
982 	int err;
983 
984 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
985 		size_t n, rest = nilfs_segctor_buffer_rest(sci);
986 
987 		n = nilfs_lookup_dirty_data_buffers(
988 			inode, &data_buffers, rest + 1, 0, LLONG_MAX);
989 		if (n > rest) {
990 			err = nilfs_segctor_apply_buffers(
991 				sci, inode, &data_buffers,
992 				sc_ops->collect_data);
993 			BUG_ON(!err); /* always receive -E2BIG or true error */
994 			goto break_or_fail;
995 		}
996 	}
997 	nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
998 
999 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1000 		err = nilfs_segctor_apply_buffers(
1001 			sci, inode, &data_buffers, sc_ops->collect_data);
1002 		if (unlikely(err)) {
1003 			/* dispose node list */
1004 			nilfs_segctor_apply_buffers(
1005 				sci, inode, &node_buffers, NULL);
1006 			goto break_or_fail;
1007 		}
1008 		sci->sc_stage.flags |= NILFS_CF_NODE;
1009 	}
1010 	/* Collect node */
1011 	err = nilfs_segctor_apply_buffers(
1012 		sci, inode, &node_buffers, sc_ops->collect_node);
1013 	if (unlikely(err))
1014 		goto break_or_fail;
1015 
1016 	nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1017 	err = nilfs_segctor_apply_buffers(
1018 		sci, inode, &node_buffers, sc_ops->collect_bmap);
1019 	if (unlikely(err))
1020 		goto break_or_fail;
1021 
1022 	nilfs_segctor_end_finfo(sci, inode);
1023 	sci->sc_stage.flags &= ~NILFS_CF_NODE;
1024 
1025  break_or_fail:
1026 	return err;
1027 }
1028 
1029 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1030 					 struct inode *inode)
1031 {
1032 	LIST_HEAD(data_buffers);
1033 	size_t n, rest = nilfs_segctor_buffer_rest(sci);
1034 	int err;
1035 
1036 	n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1037 					    sci->sc_dsync_start,
1038 					    sci->sc_dsync_end);
1039 
1040 	err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1041 					  nilfs_collect_file_data);
1042 	if (!err) {
1043 		nilfs_segctor_end_finfo(sci, inode);
1044 		BUG_ON(n > rest);
1045 		/* always receive -E2BIG or true error if n > rest */
1046 	}
1047 	return err;
1048 }
1049 
1050 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1051 {
1052 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1053 	struct list_head *head;
1054 	struct nilfs_inode_info *ii;
1055 	size_t ndone;
1056 	int err = 0;
1057 
1058 	switch (sci->sc_stage.scnt) {
1059 	case NILFS_ST_INIT:
1060 		/* Pre-processes */
1061 		sci->sc_stage.flags = 0;
1062 
1063 		if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1064 			sci->sc_nblk_inc = 0;
1065 			sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1066 			if (mode == SC_LSEG_DSYNC) {
1067 				sci->sc_stage.scnt = NILFS_ST_DSYNC;
1068 				goto dsync_mode;
1069 			}
1070 		}
1071 
1072 		sci->sc_stage.dirty_file_ptr = NULL;
1073 		sci->sc_stage.gc_inode_ptr = NULL;
1074 		if (mode == SC_FLUSH_DAT) {
1075 			sci->sc_stage.scnt = NILFS_ST_DAT;
1076 			goto dat_stage;
1077 		}
1078 		sci->sc_stage.scnt++;  /* Fall through */
1079 	case NILFS_ST_GC:
1080 		if (nilfs_doing_gc()) {
1081 			head = &sci->sc_gc_inodes;
1082 			ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1083 						head, i_dirty);
1084 			list_for_each_entry_continue(ii, head, i_dirty) {
1085 				err = nilfs_segctor_scan_file(
1086 					sci, &ii->vfs_inode,
1087 					&nilfs_sc_file_ops);
1088 				if (unlikely(err)) {
1089 					sci->sc_stage.gc_inode_ptr = list_entry(
1090 						ii->i_dirty.prev,
1091 						struct nilfs_inode_info,
1092 						i_dirty);
1093 					goto break_or_fail;
1094 				}
1095 				set_bit(NILFS_I_COLLECTED, &ii->i_state);
1096 			}
1097 			sci->sc_stage.gc_inode_ptr = NULL;
1098 		}
1099 		sci->sc_stage.scnt++;  /* Fall through */
1100 	case NILFS_ST_FILE:
1101 		head = &sci->sc_dirty_files;
1102 		ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1103 					i_dirty);
1104 		list_for_each_entry_continue(ii, head, i_dirty) {
1105 			clear_bit(NILFS_I_DIRTY, &ii->i_state);
1106 
1107 			err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1108 						      &nilfs_sc_file_ops);
1109 			if (unlikely(err)) {
1110 				sci->sc_stage.dirty_file_ptr =
1111 					list_entry(ii->i_dirty.prev,
1112 						   struct nilfs_inode_info,
1113 						   i_dirty);
1114 				goto break_or_fail;
1115 			}
1116 			/* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1117 			/* XXX: required ? */
1118 		}
1119 		sci->sc_stage.dirty_file_ptr = NULL;
1120 		if (mode == SC_FLUSH_FILE) {
1121 			sci->sc_stage.scnt = NILFS_ST_DONE;
1122 			return 0;
1123 		}
1124 		sci->sc_stage.scnt++;
1125 		sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1126 		/* Fall through */
1127 	case NILFS_ST_IFILE:
1128 		err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1129 					      &nilfs_sc_file_ops);
1130 		if (unlikely(err))
1131 			break;
1132 		sci->sc_stage.scnt++;
1133 		/* Creating a checkpoint */
1134 		err = nilfs_segctor_create_checkpoint(sci);
1135 		if (unlikely(err))
1136 			break;
1137 		/* Fall through */
1138 	case NILFS_ST_CPFILE:
1139 		err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1140 					      &nilfs_sc_file_ops);
1141 		if (unlikely(err))
1142 			break;
1143 		sci->sc_stage.scnt++;  /* Fall through */
1144 	case NILFS_ST_SUFILE:
1145 		err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1146 					 sci->sc_nfreesegs, &ndone);
1147 		if (unlikely(err)) {
1148 			nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1149 						  sci->sc_freesegs, ndone,
1150 						  NULL);
1151 			break;
1152 		}
1153 		sci->sc_stage.flags |= NILFS_CF_SUFREED;
1154 
1155 		err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1156 					      &nilfs_sc_file_ops);
1157 		if (unlikely(err))
1158 			break;
1159 		sci->sc_stage.scnt++;  /* Fall through */
1160 	case NILFS_ST_DAT:
1161  dat_stage:
1162 		err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1163 					      &nilfs_sc_dat_ops);
1164 		if (unlikely(err))
1165 			break;
1166 		if (mode == SC_FLUSH_DAT) {
1167 			sci->sc_stage.scnt = NILFS_ST_DONE;
1168 			return 0;
1169 		}
1170 		sci->sc_stage.scnt++;  /* Fall through */
1171 	case NILFS_ST_SR:
1172 		if (mode == SC_LSEG_SR) {
1173 			/* Appending a super root */
1174 			err = nilfs_segctor_add_super_root(sci);
1175 			if (unlikely(err))
1176 				break;
1177 		}
1178 		/* End of a logical segment */
1179 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1180 		sci->sc_stage.scnt = NILFS_ST_DONE;
1181 		return 0;
1182 	case NILFS_ST_DSYNC:
1183  dsync_mode:
1184 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1185 		ii = sci->sc_dsync_inode;
1186 		if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1187 			break;
1188 
1189 		err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1190 		if (unlikely(err))
1191 			break;
1192 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1193 		sci->sc_stage.scnt = NILFS_ST_DONE;
1194 		return 0;
1195 	case NILFS_ST_DONE:
1196 		return 0;
1197 	default:
1198 		BUG();
1199 	}
1200 
1201  break_or_fail:
1202 	return err;
1203 }
1204 
1205 /**
1206  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1207  * @sci: nilfs_sc_info
1208  * @nilfs: nilfs object
1209  */
1210 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1211 					    struct the_nilfs *nilfs)
1212 {
1213 	struct nilfs_segment_buffer *segbuf, *prev;
1214 	__u64 nextnum;
1215 	int err, alloc = 0;
1216 
1217 	segbuf = nilfs_segbuf_new(sci->sc_super);
1218 	if (unlikely(!segbuf))
1219 		return -ENOMEM;
1220 
1221 	if (list_empty(&sci->sc_write_logs)) {
1222 		nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1223 				 nilfs->ns_pseg_offset, nilfs);
1224 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1225 			nilfs_shift_to_next_segment(nilfs);
1226 			nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1227 		}
1228 
1229 		segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1230 		nextnum = nilfs->ns_nextnum;
1231 
1232 		if (nilfs->ns_segnum == nilfs->ns_nextnum)
1233 			/* Start from the head of a new full segment */
1234 			alloc++;
1235 	} else {
1236 		/* Continue logs */
1237 		prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1238 		nilfs_segbuf_map_cont(segbuf, prev);
1239 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1240 		nextnum = prev->sb_nextnum;
1241 
1242 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1243 			nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1244 			segbuf->sb_sum.seg_seq++;
1245 			alloc++;
1246 		}
1247 	}
1248 
1249 	err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1250 	if (err)
1251 		goto failed;
1252 
1253 	if (alloc) {
1254 		err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1255 		if (err)
1256 			goto failed;
1257 	}
1258 	nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1259 
1260 	BUG_ON(!list_empty(&sci->sc_segbufs));
1261 	list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1262 	sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1263 	return 0;
1264 
1265  failed:
1266 	nilfs_segbuf_free(segbuf);
1267 	return err;
1268 }
1269 
1270 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1271 					 struct the_nilfs *nilfs, int nadd)
1272 {
1273 	struct nilfs_segment_buffer *segbuf, *prev;
1274 	struct inode *sufile = nilfs->ns_sufile;
1275 	__u64 nextnextnum;
1276 	LIST_HEAD(list);
1277 	int err, ret, i;
1278 
1279 	prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1280 	/*
1281 	 * Since the segment specified with nextnum might be allocated during
1282 	 * the previous construction, the buffer including its segusage may
1283 	 * not be dirty.  The following call ensures that the buffer is dirty
1284 	 * and will pin the buffer on memory until the sufile is written.
1285 	 */
1286 	err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1287 	if (unlikely(err))
1288 		return err;
1289 
1290 	for (i = 0; i < nadd; i++) {
1291 		/* extend segment info */
1292 		err = -ENOMEM;
1293 		segbuf = nilfs_segbuf_new(sci->sc_super);
1294 		if (unlikely(!segbuf))
1295 			goto failed;
1296 
1297 		/* map this buffer to region of segment on-disk */
1298 		nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1299 		sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1300 
1301 		/* allocate the next next full segment */
1302 		err = nilfs_sufile_alloc(sufile, &nextnextnum);
1303 		if (unlikely(err))
1304 			goto failed_segbuf;
1305 
1306 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1307 		nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1308 
1309 		list_add_tail(&segbuf->sb_list, &list);
1310 		prev = segbuf;
1311 	}
1312 	list_splice_tail(&list, &sci->sc_segbufs);
1313 	return 0;
1314 
1315  failed_segbuf:
1316 	nilfs_segbuf_free(segbuf);
1317  failed:
1318 	list_for_each_entry(segbuf, &list, sb_list) {
1319 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1320 		WARN_ON(ret); /* never fails */
1321 	}
1322 	nilfs_destroy_logs(&list);
1323 	return err;
1324 }
1325 
1326 static void nilfs_free_incomplete_logs(struct list_head *logs,
1327 				       struct the_nilfs *nilfs)
1328 {
1329 	struct nilfs_segment_buffer *segbuf, *prev;
1330 	struct inode *sufile = nilfs->ns_sufile;
1331 	int ret;
1332 
1333 	segbuf = NILFS_FIRST_SEGBUF(logs);
1334 	if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1335 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1336 		WARN_ON(ret); /* never fails */
1337 	}
1338 	if (atomic_read(&segbuf->sb_err)) {
1339 		/* Case 1: The first segment failed */
1340 		if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1341 			/* Case 1a:  Partial segment appended into an existing
1342 			   segment */
1343 			nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1344 						segbuf->sb_fseg_end);
1345 		else /* Case 1b:  New full segment */
1346 			set_nilfs_discontinued(nilfs);
1347 	}
1348 
1349 	prev = segbuf;
1350 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1351 		if (prev->sb_nextnum != segbuf->sb_nextnum) {
1352 			ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1353 			WARN_ON(ret); /* never fails */
1354 		}
1355 		if (atomic_read(&segbuf->sb_err) &&
1356 		    segbuf->sb_segnum != nilfs->ns_nextnum)
1357 			/* Case 2: extended segment (!= next) failed */
1358 			nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1359 		prev = segbuf;
1360 	}
1361 }
1362 
1363 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1364 					  struct inode *sufile)
1365 {
1366 	struct nilfs_segment_buffer *segbuf;
1367 	unsigned long live_blocks;
1368 	int ret;
1369 
1370 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1371 		live_blocks = segbuf->sb_sum.nblocks +
1372 			(segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1373 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1374 						     live_blocks,
1375 						     sci->sc_seg_ctime);
1376 		WARN_ON(ret); /* always succeed because the segusage is dirty */
1377 	}
1378 }
1379 
1380 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1381 {
1382 	struct nilfs_segment_buffer *segbuf;
1383 	int ret;
1384 
1385 	segbuf = NILFS_FIRST_SEGBUF(logs);
1386 	ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1387 					     segbuf->sb_pseg_start -
1388 					     segbuf->sb_fseg_start, 0);
1389 	WARN_ON(ret); /* always succeed because the segusage is dirty */
1390 
1391 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1392 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1393 						     0, 0);
1394 		WARN_ON(ret); /* always succeed */
1395 	}
1396 }
1397 
1398 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1399 					    struct nilfs_segment_buffer *last,
1400 					    struct inode *sufile)
1401 {
1402 	struct nilfs_segment_buffer *segbuf = last;
1403 	int ret;
1404 
1405 	list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1406 		sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1407 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1408 		WARN_ON(ret);
1409 	}
1410 	nilfs_truncate_logs(&sci->sc_segbufs, last);
1411 }
1412 
1413 
1414 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1415 				 struct the_nilfs *nilfs, int mode)
1416 {
1417 	struct nilfs_cstage prev_stage = sci->sc_stage;
1418 	int err, nadd = 1;
1419 
1420 	/* Collection retry loop */
1421 	for (;;) {
1422 		sci->sc_nblk_this_inc = 0;
1423 		sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1424 
1425 		err = nilfs_segctor_reset_segment_buffer(sci);
1426 		if (unlikely(err))
1427 			goto failed;
1428 
1429 		err = nilfs_segctor_collect_blocks(sci, mode);
1430 		sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1431 		if (!err)
1432 			break;
1433 
1434 		if (unlikely(err != -E2BIG))
1435 			goto failed;
1436 
1437 		/* The current segment is filled up */
1438 		if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1439 			break;
1440 
1441 		nilfs_clear_logs(&sci->sc_segbufs);
1442 
1443 		if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1444 			err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1445 							sci->sc_freesegs,
1446 							sci->sc_nfreesegs,
1447 							NULL);
1448 			WARN_ON(err); /* do not happen */
1449 			sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1450 		}
1451 
1452 		err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1453 		if (unlikely(err))
1454 			return err;
1455 
1456 		nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1457 		sci->sc_stage = prev_stage;
1458 	}
1459 	nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1460 	return 0;
1461 
1462  failed:
1463 	return err;
1464 }
1465 
1466 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1467 				      struct buffer_head *new_bh)
1468 {
1469 	BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1470 
1471 	list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1472 	/* The caller must release old_bh */
1473 }
1474 
1475 static int
1476 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1477 				     struct nilfs_segment_buffer *segbuf,
1478 				     int mode)
1479 {
1480 	struct inode *inode = NULL;
1481 	sector_t blocknr;
1482 	unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1483 	unsigned long nblocks = 0, ndatablk = 0;
1484 	struct nilfs_sc_operations *sc_op = NULL;
1485 	struct nilfs_segsum_pointer ssp;
1486 	struct nilfs_finfo *finfo = NULL;
1487 	union nilfs_binfo binfo;
1488 	struct buffer_head *bh, *bh_org;
1489 	ino_t ino = 0;
1490 	int err = 0;
1491 
1492 	if (!nfinfo)
1493 		goto out;
1494 
1495 	blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1496 	ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1497 	ssp.offset = sizeof(struct nilfs_segment_summary);
1498 
1499 	list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1500 		if (bh == segbuf->sb_super_root)
1501 			break;
1502 		if (!finfo) {
1503 			finfo =	nilfs_segctor_map_segsum_entry(
1504 				sci, &ssp, sizeof(*finfo));
1505 			ino = le64_to_cpu(finfo->fi_ino);
1506 			nblocks = le32_to_cpu(finfo->fi_nblocks);
1507 			ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1508 
1509 			inode = bh->b_page->mapping->host;
1510 
1511 			if (mode == SC_LSEG_DSYNC)
1512 				sc_op = &nilfs_sc_dsync_ops;
1513 			else if (ino == NILFS_DAT_INO)
1514 				sc_op = &nilfs_sc_dat_ops;
1515 			else /* file blocks */
1516 				sc_op = &nilfs_sc_file_ops;
1517 		}
1518 		bh_org = bh;
1519 		get_bh(bh_org);
1520 		err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1521 					&binfo);
1522 		if (bh != bh_org)
1523 			nilfs_list_replace_buffer(bh_org, bh);
1524 		brelse(bh_org);
1525 		if (unlikely(err))
1526 			goto failed_bmap;
1527 
1528 		if (ndatablk > 0)
1529 			sc_op->write_data_binfo(sci, &ssp, &binfo);
1530 		else
1531 			sc_op->write_node_binfo(sci, &ssp, &binfo);
1532 
1533 		blocknr++;
1534 		if (--nblocks == 0) {
1535 			finfo = NULL;
1536 			if (--nfinfo == 0)
1537 				break;
1538 		} else if (ndatablk > 0)
1539 			ndatablk--;
1540 	}
1541  out:
1542 	return 0;
1543 
1544  failed_bmap:
1545 	return err;
1546 }
1547 
1548 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1549 {
1550 	struct nilfs_segment_buffer *segbuf;
1551 	int err;
1552 
1553 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1554 		err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1555 		if (unlikely(err))
1556 			return err;
1557 		nilfs_segbuf_fill_in_segsum(segbuf);
1558 	}
1559 	return 0;
1560 }
1561 
1562 static void nilfs_begin_page_io(struct page *page)
1563 {
1564 	if (!page || PageWriteback(page))
1565 		/* For split b-tree node pages, this function may be called
1566 		   twice.  We ignore the 2nd or later calls by this check. */
1567 		return;
1568 
1569 	lock_page(page);
1570 	clear_page_dirty_for_io(page);
1571 	set_page_writeback(page);
1572 	unlock_page(page);
1573 }
1574 
1575 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1576 {
1577 	struct nilfs_segment_buffer *segbuf;
1578 	struct page *bd_page = NULL, *fs_page = NULL;
1579 
1580 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1581 		struct buffer_head *bh;
1582 
1583 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1584 				    b_assoc_buffers) {
1585 			set_buffer_async_write(bh);
1586 			if (bh->b_page != bd_page) {
1587 				if (bd_page) {
1588 					lock_page(bd_page);
1589 					clear_page_dirty_for_io(bd_page);
1590 					set_page_writeback(bd_page);
1591 					unlock_page(bd_page);
1592 				}
1593 				bd_page = bh->b_page;
1594 			}
1595 		}
1596 
1597 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1598 				    b_assoc_buffers) {
1599 			set_buffer_async_write(bh);
1600 			if (bh == segbuf->sb_super_root) {
1601 				if (bh->b_page != bd_page) {
1602 					lock_page(bd_page);
1603 					clear_page_dirty_for_io(bd_page);
1604 					set_page_writeback(bd_page);
1605 					unlock_page(bd_page);
1606 					bd_page = bh->b_page;
1607 				}
1608 				break;
1609 			}
1610 			if (bh->b_page != fs_page) {
1611 				nilfs_begin_page_io(fs_page);
1612 				fs_page = bh->b_page;
1613 			}
1614 		}
1615 	}
1616 	if (bd_page) {
1617 		lock_page(bd_page);
1618 		clear_page_dirty_for_io(bd_page);
1619 		set_page_writeback(bd_page);
1620 		unlock_page(bd_page);
1621 	}
1622 	nilfs_begin_page_io(fs_page);
1623 }
1624 
1625 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1626 			       struct the_nilfs *nilfs)
1627 {
1628 	int ret;
1629 
1630 	ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1631 	list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1632 	return ret;
1633 }
1634 
1635 static void nilfs_end_page_io(struct page *page, int err)
1636 {
1637 	if (!page)
1638 		return;
1639 
1640 	if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1641 		/*
1642 		 * For b-tree node pages, this function may be called twice
1643 		 * or more because they might be split in a segment.
1644 		 */
1645 		if (PageDirty(page)) {
1646 			/*
1647 			 * For pages holding split b-tree node buffers, dirty
1648 			 * flag on the buffers may be cleared discretely.
1649 			 * In that case, the page is once redirtied for
1650 			 * remaining buffers, and it must be cancelled if
1651 			 * all the buffers get cleaned later.
1652 			 */
1653 			lock_page(page);
1654 			if (nilfs_page_buffers_clean(page))
1655 				__nilfs_clear_page_dirty(page);
1656 			unlock_page(page);
1657 		}
1658 		return;
1659 	}
1660 
1661 	if (!err) {
1662 		if (!nilfs_page_buffers_clean(page))
1663 			__set_page_dirty_nobuffers(page);
1664 		ClearPageError(page);
1665 	} else {
1666 		__set_page_dirty_nobuffers(page);
1667 		SetPageError(page);
1668 	}
1669 
1670 	end_page_writeback(page);
1671 }
1672 
1673 static void nilfs_abort_logs(struct list_head *logs, int err)
1674 {
1675 	struct nilfs_segment_buffer *segbuf;
1676 	struct page *bd_page = NULL, *fs_page = NULL;
1677 	struct buffer_head *bh;
1678 
1679 	if (list_empty(logs))
1680 		return;
1681 
1682 	list_for_each_entry(segbuf, logs, sb_list) {
1683 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1684 				    b_assoc_buffers) {
1685 			clear_buffer_async_write(bh);
1686 			if (bh->b_page != bd_page) {
1687 				if (bd_page)
1688 					end_page_writeback(bd_page);
1689 				bd_page = bh->b_page;
1690 			}
1691 		}
1692 
1693 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1694 				    b_assoc_buffers) {
1695 			clear_buffer_async_write(bh);
1696 			if (bh == segbuf->sb_super_root) {
1697 				if (bh->b_page != bd_page) {
1698 					end_page_writeback(bd_page);
1699 					bd_page = bh->b_page;
1700 				}
1701 				break;
1702 			}
1703 			if (bh->b_page != fs_page) {
1704 				nilfs_end_page_io(fs_page, err);
1705 				fs_page = bh->b_page;
1706 			}
1707 		}
1708 	}
1709 	if (bd_page)
1710 		end_page_writeback(bd_page);
1711 
1712 	nilfs_end_page_io(fs_page, err);
1713 }
1714 
1715 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1716 					     struct the_nilfs *nilfs, int err)
1717 {
1718 	LIST_HEAD(logs);
1719 	int ret;
1720 
1721 	list_splice_tail_init(&sci->sc_write_logs, &logs);
1722 	ret = nilfs_wait_on_logs(&logs);
1723 	nilfs_abort_logs(&logs, ret ? : err);
1724 
1725 	list_splice_tail_init(&sci->sc_segbufs, &logs);
1726 	nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1727 	nilfs_free_incomplete_logs(&logs, nilfs);
1728 
1729 	if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1730 		ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1731 						sci->sc_freesegs,
1732 						sci->sc_nfreesegs,
1733 						NULL);
1734 		WARN_ON(ret); /* do not happen */
1735 	}
1736 
1737 	nilfs_destroy_logs(&logs);
1738 }
1739 
1740 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1741 				   struct nilfs_segment_buffer *segbuf)
1742 {
1743 	nilfs->ns_segnum = segbuf->sb_segnum;
1744 	nilfs->ns_nextnum = segbuf->sb_nextnum;
1745 	nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1746 		+ segbuf->sb_sum.nblocks;
1747 	nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1748 	nilfs->ns_ctime = segbuf->sb_sum.ctime;
1749 }
1750 
1751 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1752 {
1753 	struct nilfs_segment_buffer *segbuf;
1754 	struct page *bd_page = NULL, *fs_page = NULL;
1755 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1756 	int update_sr = false;
1757 
1758 	list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1759 		struct buffer_head *bh;
1760 
1761 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1762 				    b_assoc_buffers) {
1763 			set_buffer_uptodate(bh);
1764 			clear_buffer_dirty(bh);
1765 			clear_buffer_async_write(bh);
1766 			if (bh->b_page != bd_page) {
1767 				if (bd_page)
1768 					end_page_writeback(bd_page);
1769 				bd_page = bh->b_page;
1770 			}
1771 		}
1772 		/*
1773 		 * We assume that the buffers which belong to the same page
1774 		 * continue over the buffer list.
1775 		 * Under this assumption, the last BHs of pages is
1776 		 * identifiable by the discontinuity of bh->b_page
1777 		 * (page != fs_page).
1778 		 *
1779 		 * For B-tree node blocks, however, this assumption is not
1780 		 * guaranteed.  The cleanup code of B-tree node pages needs
1781 		 * special care.
1782 		 */
1783 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1784 				    b_assoc_buffers) {
1785 			set_buffer_uptodate(bh);
1786 			clear_buffer_dirty(bh);
1787 			clear_buffer_async_write(bh);
1788 			clear_buffer_delay(bh);
1789 			clear_buffer_nilfs_volatile(bh);
1790 			clear_buffer_nilfs_redirected(bh);
1791 			if (bh == segbuf->sb_super_root) {
1792 				if (bh->b_page != bd_page) {
1793 					end_page_writeback(bd_page);
1794 					bd_page = bh->b_page;
1795 				}
1796 				update_sr = true;
1797 				break;
1798 			}
1799 			if (bh->b_page != fs_page) {
1800 				nilfs_end_page_io(fs_page, 0);
1801 				fs_page = bh->b_page;
1802 			}
1803 		}
1804 
1805 		if (!nilfs_segbuf_simplex(segbuf)) {
1806 			if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1807 				set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1808 				sci->sc_lseg_stime = jiffies;
1809 			}
1810 			if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1811 				clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1812 		}
1813 	}
1814 	/*
1815 	 * Since pages may continue over multiple segment buffers,
1816 	 * end of the last page must be checked outside of the loop.
1817 	 */
1818 	if (bd_page)
1819 		end_page_writeback(bd_page);
1820 
1821 	nilfs_end_page_io(fs_page, 0);
1822 
1823 	nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1824 
1825 	if (nilfs_doing_gc())
1826 		nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1827 	else
1828 		nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1829 
1830 	sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1831 
1832 	segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1833 	nilfs_set_next_segment(nilfs, segbuf);
1834 
1835 	if (update_sr) {
1836 		nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1837 				       segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1838 
1839 		clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1840 		clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1841 		set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1842 		nilfs_segctor_clear_metadata_dirty(sci);
1843 	} else
1844 		clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1845 }
1846 
1847 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1848 {
1849 	int ret;
1850 
1851 	ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1852 	if (!ret) {
1853 		nilfs_segctor_complete_write(sci);
1854 		nilfs_destroy_logs(&sci->sc_write_logs);
1855 	}
1856 	return ret;
1857 }
1858 
1859 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1860 					     struct the_nilfs *nilfs)
1861 {
1862 	struct nilfs_inode_info *ii, *n;
1863 	struct inode *ifile = sci->sc_root->ifile;
1864 
1865 	spin_lock(&nilfs->ns_inode_lock);
1866  retry:
1867 	list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1868 		if (!ii->i_bh) {
1869 			struct buffer_head *ibh;
1870 			int err;
1871 
1872 			spin_unlock(&nilfs->ns_inode_lock);
1873 			err = nilfs_ifile_get_inode_block(
1874 				ifile, ii->vfs_inode.i_ino, &ibh);
1875 			if (unlikely(err)) {
1876 				nilfs_warning(sci->sc_super, __func__,
1877 					      "failed to get inode block.\n");
1878 				return err;
1879 			}
1880 			mark_buffer_dirty(ibh);
1881 			nilfs_mdt_mark_dirty(ifile);
1882 			spin_lock(&nilfs->ns_inode_lock);
1883 			if (likely(!ii->i_bh))
1884 				ii->i_bh = ibh;
1885 			else
1886 				brelse(ibh);
1887 			goto retry;
1888 		}
1889 
1890 		clear_bit(NILFS_I_QUEUED, &ii->i_state);
1891 		set_bit(NILFS_I_BUSY, &ii->i_state);
1892 		list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1893 	}
1894 	spin_unlock(&nilfs->ns_inode_lock);
1895 
1896 	return 0;
1897 }
1898 
1899 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1900 					     struct the_nilfs *nilfs)
1901 {
1902 	struct nilfs_transaction_info *ti = current->journal_info;
1903 	struct nilfs_inode_info *ii, *n;
1904 
1905 	spin_lock(&nilfs->ns_inode_lock);
1906 	list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1907 		if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1908 		    test_bit(NILFS_I_DIRTY, &ii->i_state))
1909 			continue;
1910 
1911 		clear_bit(NILFS_I_BUSY, &ii->i_state);
1912 		brelse(ii->i_bh);
1913 		ii->i_bh = NULL;
1914 		list_move_tail(&ii->i_dirty, &ti->ti_garbage);
1915 	}
1916 	spin_unlock(&nilfs->ns_inode_lock);
1917 }
1918 
1919 /*
1920  * Main procedure of segment constructor
1921  */
1922 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
1923 {
1924 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1925 	int err;
1926 
1927 	sci->sc_stage.scnt = NILFS_ST_INIT;
1928 	sci->sc_cno = nilfs->ns_cno;
1929 
1930 	err = nilfs_segctor_collect_dirty_files(sci, nilfs);
1931 	if (unlikely(err))
1932 		goto out;
1933 
1934 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
1935 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1936 
1937 	if (nilfs_segctor_clean(sci))
1938 		goto out;
1939 
1940 	do {
1941 		sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
1942 
1943 		err = nilfs_segctor_begin_construction(sci, nilfs);
1944 		if (unlikely(err))
1945 			goto out;
1946 
1947 		/* Update time stamp */
1948 		sci->sc_seg_ctime = get_seconds();
1949 
1950 		err = nilfs_segctor_collect(sci, nilfs, mode);
1951 		if (unlikely(err))
1952 			goto failed;
1953 
1954 		/* Avoid empty segment */
1955 		if (sci->sc_stage.scnt == NILFS_ST_DONE &&
1956 		    nilfs_segbuf_empty(sci->sc_curseg)) {
1957 			nilfs_segctor_abort_construction(sci, nilfs, 1);
1958 			goto out;
1959 		}
1960 
1961 		err = nilfs_segctor_assign(sci, mode);
1962 		if (unlikely(err))
1963 			goto failed;
1964 
1965 		if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
1966 			nilfs_segctor_fill_in_file_bmap(sci);
1967 
1968 		if (mode == SC_LSEG_SR &&
1969 		    sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
1970 			err = nilfs_segctor_fill_in_checkpoint(sci);
1971 			if (unlikely(err))
1972 				goto failed_to_write;
1973 
1974 			nilfs_segctor_fill_in_super_root(sci, nilfs);
1975 		}
1976 		nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
1977 
1978 		/* Write partial segments */
1979 		nilfs_segctor_prepare_write(sci);
1980 
1981 		nilfs_add_checksums_on_logs(&sci->sc_segbufs,
1982 					    nilfs->ns_crc_seed);
1983 
1984 		err = nilfs_segctor_write(sci, nilfs);
1985 		if (unlikely(err))
1986 			goto failed_to_write;
1987 
1988 		if (sci->sc_stage.scnt == NILFS_ST_DONE ||
1989 		    nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
1990 			/*
1991 			 * At this point, we avoid double buffering
1992 			 * for blocksize < pagesize because page dirty
1993 			 * flag is turned off during write and dirty
1994 			 * buffers are not properly collected for
1995 			 * pages crossing over segments.
1996 			 */
1997 			err = nilfs_segctor_wait(sci);
1998 			if (err)
1999 				goto failed_to_write;
2000 		}
2001 	} while (sci->sc_stage.scnt != NILFS_ST_DONE);
2002 
2003  out:
2004 	nilfs_segctor_drop_written_files(sci, nilfs);
2005 	return err;
2006 
2007  failed_to_write:
2008 	if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2009 		nilfs_redirty_inodes(&sci->sc_dirty_files);
2010 
2011  failed:
2012 	if (nilfs_doing_gc())
2013 		nilfs_redirty_inodes(&sci->sc_gc_inodes);
2014 	nilfs_segctor_abort_construction(sci, nilfs, err);
2015 	goto out;
2016 }
2017 
2018 /**
2019  * nilfs_segctor_start_timer - set timer of background write
2020  * @sci: nilfs_sc_info
2021  *
2022  * If the timer has already been set, it ignores the new request.
2023  * This function MUST be called within a section locking the segment
2024  * semaphore.
2025  */
2026 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2027 {
2028 	spin_lock(&sci->sc_state_lock);
2029 	if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2030 		sci->sc_timer.expires = jiffies + sci->sc_interval;
2031 		add_timer(&sci->sc_timer);
2032 		sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2033 	}
2034 	spin_unlock(&sci->sc_state_lock);
2035 }
2036 
2037 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2038 {
2039 	spin_lock(&sci->sc_state_lock);
2040 	if (!(sci->sc_flush_request & (1 << bn))) {
2041 		unsigned long prev_req = sci->sc_flush_request;
2042 
2043 		sci->sc_flush_request |= (1 << bn);
2044 		if (!prev_req)
2045 			wake_up(&sci->sc_wait_daemon);
2046 	}
2047 	spin_unlock(&sci->sc_state_lock);
2048 }
2049 
2050 /**
2051  * nilfs_flush_segment - trigger a segment construction for resource control
2052  * @sb: super block
2053  * @ino: inode number of the file to be flushed out.
2054  */
2055 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2056 {
2057 	struct the_nilfs *nilfs = sb->s_fs_info;
2058 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2059 
2060 	if (!sci || nilfs_doing_construction())
2061 		return;
2062 	nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2063 					/* assign bit 0 to data files */
2064 }
2065 
2066 struct nilfs_segctor_wait_request {
2067 	wait_queue_t	wq;
2068 	__u32		seq;
2069 	int		err;
2070 	atomic_t	done;
2071 };
2072 
2073 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2074 {
2075 	struct nilfs_segctor_wait_request wait_req;
2076 	int err = 0;
2077 
2078 	spin_lock(&sci->sc_state_lock);
2079 	init_wait(&wait_req.wq);
2080 	wait_req.err = 0;
2081 	atomic_set(&wait_req.done, 0);
2082 	wait_req.seq = ++sci->sc_seq_request;
2083 	spin_unlock(&sci->sc_state_lock);
2084 
2085 	init_waitqueue_entry(&wait_req.wq, current);
2086 	add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2087 	set_current_state(TASK_INTERRUPTIBLE);
2088 	wake_up(&sci->sc_wait_daemon);
2089 
2090 	for (;;) {
2091 		if (atomic_read(&wait_req.done)) {
2092 			err = wait_req.err;
2093 			break;
2094 		}
2095 		if (!signal_pending(current)) {
2096 			schedule();
2097 			continue;
2098 		}
2099 		err = -ERESTARTSYS;
2100 		break;
2101 	}
2102 	finish_wait(&sci->sc_wait_request, &wait_req.wq);
2103 	return err;
2104 }
2105 
2106 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2107 {
2108 	struct nilfs_segctor_wait_request *wrq, *n;
2109 	unsigned long flags;
2110 
2111 	spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2112 	list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2113 				 wq.task_list) {
2114 		if (!atomic_read(&wrq->done) &&
2115 		    nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2116 			wrq->err = err;
2117 			atomic_set(&wrq->done, 1);
2118 		}
2119 		if (atomic_read(&wrq->done)) {
2120 			wrq->wq.func(&wrq->wq,
2121 				     TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2122 				     0, NULL);
2123 		}
2124 	}
2125 	spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2126 }
2127 
2128 /**
2129  * nilfs_construct_segment - construct a logical segment
2130  * @sb: super block
2131  *
2132  * Return Value: On success, 0 is retured. On errors, one of the following
2133  * negative error code is returned.
2134  *
2135  * %-EROFS - Read only filesystem.
2136  *
2137  * %-EIO - I/O error
2138  *
2139  * %-ENOSPC - No space left on device (only in a panic state).
2140  *
2141  * %-ERESTARTSYS - Interrupted.
2142  *
2143  * %-ENOMEM - Insufficient memory available.
2144  */
2145 int nilfs_construct_segment(struct super_block *sb)
2146 {
2147 	struct the_nilfs *nilfs = sb->s_fs_info;
2148 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2149 	struct nilfs_transaction_info *ti;
2150 	int err;
2151 
2152 	if (!sci)
2153 		return -EROFS;
2154 
2155 	/* A call inside transactions causes a deadlock. */
2156 	BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2157 
2158 	err = nilfs_segctor_sync(sci);
2159 	return err;
2160 }
2161 
2162 /**
2163  * nilfs_construct_dsync_segment - construct a data-only logical segment
2164  * @sb: super block
2165  * @inode: inode whose data blocks should be written out
2166  * @start: start byte offset
2167  * @end: end byte offset (inclusive)
2168  *
2169  * Return Value: On success, 0 is retured. On errors, one of the following
2170  * negative error code is returned.
2171  *
2172  * %-EROFS - Read only filesystem.
2173  *
2174  * %-EIO - I/O error
2175  *
2176  * %-ENOSPC - No space left on device (only in a panic state).
2177  *
2178  * %-ERESTARTSYS - Interrupted.
2179  *
2180  * %-ENOMEM - Insufficient memory available.
2181  */
2182 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2183 				  loff_t start, loff_t end)
2184 {
2185 	struct the_nilfs *nilfs = sb->s_fs_info;
2186 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2187 	struct nilfs_inode_info *ii;
2188 	struct nilfs_transaction_info ti;
2189 	int err = 0;
2190 
2191 	if (!sci)
2192 		return -EROFS;
2193 
2194 	nilfs_transaction_lock(sb, &ti, 0);
2195 
2196 	ii = NILFS_I(inode);
2197 	if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2198 	    nilfs_test_opt(nilfs, STRICT_ORDER) ||
2199 	    test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2200 	    nilfs_discontinued(nilfs)) {
2201 		nilfs_transaction_unlock(sb);
2202 		err = nilfs_segctor_sync(sci);
2203 		return err;
2204 	}
2205 
2206 	spin_lock(&nilfs->ns_inode_lock);
2207 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2208 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2209 		spin_unlock(&nilfs->ns_inode_lock);
2210 		nilfs_transaction_unlock(sb);
2211 		return 0;
2212 	}
2213 	spin_unlock(&nilfs->ns_inode_lock);
2214 	sci->sc_dsync_inode = ii;
2215 	sci->sc_dsync_start = start;
2216 	sci->sc_dsync_end = end;
2217 
2218 	err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2219 
2220 	nilfs_transaction_unlock(sb);
2221 	return err;
2222 }
2223 
2224 #define FLUSH_FILE_BIT	(0x1) /* data file only */
2225 #define FLUSH_DAT_BIT	(1 << NILFS_DAT_INO) /* DAT only */
2226 
2227 /**
2228  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2229  * @sci: segment constructor object
2230  */
2231 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2232 {
2233 	spin_lock(&sci->sc_state_lock);
2234 	sci->sc_seq_accepted = sci->sc_seq_request;
2235 	spin_unlock(&sci->sc_state_lock);
2236 	del_timer_sync(&sci->sc_timer);
2237 }
2238 
2239 /**
2240  * nilfs_segctor_notify - notify the result of request to caller threads
2241  * @sci: segment constructor object
2242  * @mode: mode of log forming
2243  * @err: error code to be notified
2244  */
2245 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2246 {
2247 	/* Clear requests (even when the construction failed) */
2248 	spin_lock(&sci->sc_state_lock);
2249 
2250 	if (mode == SC_LSEG_SR) {
2251 		sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2252 		sci->sc_seq_done = sci->sc_seq_accepted;
2253 		nilfs_segctor_wakeup(sci, err);
2254 		sci->sc_flush_request = 0;
2255 	} else {
2256 		if (mode == SC_FLUSH_FILE)
2257 			sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2258 		else if (mode == SC_FLUSH_DAT)
2259 			sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2260 
2261 		/* re-enable timer if checkpoint creation was not done */
2262 		if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2263 		    time_before(jiffies, sci->sc_timer.expires))
2264 			add_timer(&sci->sc_timer);
2265 	}
2266 	spin_unlock(&sci->sc_state_lock);
2267 }
2268 
2269 /**
2270  * nilfs_segctor_construct - form logs and write them to disk
2271  * @sci: segment constructor object
2272  * @mode: mode of log forming
2273  */
2274 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2275 {
2276 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2277 	struct nilfs_super_block **sbp;
2278 	int err = 0;
2279 
2280 	nilfs_segctor_accept(sci);
2281 
2282 	if (nilfs_discontinued(nilfs))
2283 		mode = SC_LSEG_SR;
2284 	if (!nilfs_segctor_confirm(sci))
2285 		err = nilfs_segctor_do_construct(sci, mode);
2286 
2287 	if (likely(!err)) {
2288 		if (mode != SC_FLUSH_DAT)
2289 			atomic_set(&nilfs->ns_ndirtyblks, 0);
2290 		if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2291 		    nilfs_discontinued(nilfs)) {
2292 			down_write(&nilfs->ns_sem);
2293 			err = -EIO;
2294 			sbp = nilfs_prepare_super(sci->sc_super,
2295 						  nilfs_sb_will_flip(nilfs));
2296 			if (likely(sbp)) {
2297 				nilfs_set_log_cursor(sbp[0], nilfs);
2298 				err = nilfs_commit_super(sci->sc_super,
2299 							 NILFS_SB_COMMIT);
2300 			}
2301 			up_write(&nilfs->ns_sem);
2302 		}
2303 	}
2304 
2305 	nilfs_segctor_notify(sci, mode, err);
2306 	return err;
2307 }
2308 
2309 static void nilfs_construction_timeout(unsigned long data)
2310 {
2311 	struct task_struct *p = (struct task_struct *)data;
2312 	wake_up_process(p);
2313 }
2314 
2315 static void
2316 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2317 {
2318 	struct nilfs_inode_info *ii, *n;
2319 
2320 	list_for_each_entry_safe(ii, n, head, i_dirty) {
2321 		if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2322 			continue;
2323 		list_del_init(&ii->i_dirty);
2324 		truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2325 		nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2326 		iput(&ii->vfs_inode);
2327 	}
2328 }
2329 
2330 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2331 			 void **kbufs)
2332 {
2333 	struct the_nilfs *nilfs = sb->s_fs_info;
2334 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2335 	struct nilfs_transaction_info ti;
2336 	int err;
2337 
2338 	if (unlikely(!sci))
2339 		return -EROFS;
2340 
2341 	nilfs_transaction_lock(sb, &ti, 1);
2342 
2343 	err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2344 	if (unlikely(err))
2345 		goto out_unlock;
2346 
2347 	err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2348 	if (unlikely(err)) {
2349 		nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2350 		goto out_unlock;
2351 	}
2352 
2353 	sci->sc_freesegs = kbufs[4];
2354 	sci->sc_nfreesegs = argv[4].v_nmembs;
2355 	list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2356 
2357 	for (;;) {
2358 		err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2359 		nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2360 
2361 		if (likely(!err))
2362 			break;
2363 
2364 		nilfs_warning(sb, __func__,
2365 			      "segment construction failed. (err=%d)", err);
2366 		set_current_state(TASK_INTERRUPTIBLE);
2367 		schedule_timeout(sci->sc_interval);
2368 	}
2369 	if (nilfs_test_opt(nilfs, DISCARD)) {
2370 		int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2371 						 sci->sc_nfreesegs);
2372 		if (ret) {
2373 			printk(KERN_WARNING
2374 			       "NILFS warning: error %d on discard request, "
2375 			       "turning discards off for the device\n", ret);
2376 			nilfs_clear_opt(nilfs, DISCARD);
2377 		}
2378 	}
2379 
2380  out_unlock:
2381 	sci->sc_freesegs = NULL;
2382 	sci->sc_nfreesegs = 0;
2383 	nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2384 	nilfs_transaction_unlock(sb);
2385 	return err;
2386 }
2387 
2388 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2389 {
2390 	struct nilfs_transaction_info ti;
2391 
2392 	nilfs_transaction_lock(sci->sc_super, &ti, 0);
2393 	nilfs_segctor_construct(sci, mode);
2394 
2395 	/*
2396 	 * Unclosed segment should be retried.  We do this using sc_timer.
2397 	 * Timeout of sc_timer will invoke complete construction which leads
2398 	 * to close the current logical segment.
2399 	 */
2400 	if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2401 		nilfs_segctor_start_timer(sci);
2402 
2403 	nilfs_transaction_unlock(sci->sc_super);
2404 }
2405 
2406 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2407 {
2408 	int mode = 0;
2409 	int err;
2410 
2411 	spin_lock(&sci->sc_state_lock);
2412 	mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2413 		SC_FLUSH_DAT : SC_FLUSH_FILE;
2414 	spin_unlock(&sci->sc_state_lock);
2415 
2416 	if (mode) {
2417 		err = nilfs_segctor_do_construct(sci, mode);
2418 
2419 		spin_lock(&sci->sc_state_lock);
2420 		sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2421 			~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2422 		spin_unlock(&sci->sc_state_lock);
2423 	}
2424 	clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2425 }
2426 
2427 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2428 {
2429 	if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2430 	    time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2431 		if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2432 			return SC_FLUSH_FILE;
2433 		else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2434 			return SC_FLUSH_DAT;
2435 	}
2436 	return SC_LSEG_SR;
2437 }
2438 
2439 /**
2440  * nilfs_segctor_thread - main loop of the segment constructor thread.
2441  * @arg: pointer to a struct nilfs_sc_info.
2442  *
2443  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2444  * to execute segment constructions.
2445  */
2446 static int nilfs_segctor_thread(void *arg)
2447 {
2448 	struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2449 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2450 	int timeout = 0;
2451 
2452 	sci->sc_timer.data = (unsigned long)current;
2453 	sci->sc_timer.function = nilfs_construction_timeout;
2454 
2455 	/* start sync. */
2456 	sci->sc_task = current;
2457 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2458 	printk(KERN_INFO
2459 	       "segctord starting. Construction interval = %lu seconds, "
2460 	       "CP frequency < %lu seconds\n",
2461 	       sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2462 
2463 	spin_lock(&sci->sc_state_lock);
2464  loop:
2465 	for (;;) {
2466 		int mode;
2467 
2468 		if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2469 			goto end_thread;
2470 
2471 		if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2472 			mode = SC_LSEG_SR;
2473 		else if (!sci->sc_flush_request)
2474 			break;
2475 		else
2476 			mode = nilfs_segctor_flush_mode(sci);
2477 
2478 		spin_unlock(&sci->sc_state_lock);
2479 		nilfs_segctor_thread_construct(sci, mode);
2480 		spin_lock(&sci->sc_state_lock);
2481 		timeout = 0;
2482 	}
2483 
2484 
2485 	if (freezing(current)) {
2486 		spin_unlock(&sci->sc_state_lock);
2487 		try_to_freeze();
2488 		spin_lock(&sci->sc_state_lock);
2489 	} else {
2490 		DEFINE_WAIT(wait);
2491 		int should_sleep = 1;
2492 
2493 		prepare_to_wait(&sci->sc_wait_daemon, &wait,
2494 				TASK_INTERRUPTIBLE);
2495 
2496 		if (sci->sc_seq_request != sci->sc_seq_done)
2497 			should_sleep = 0;
2498 		else if (sci->sc_flush_request)
2499 			should_sleep = 0;
2500 		else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2501 			should_sleep = time_before(jiffies,
2502 					sci->sc_timer.expires);
2503 
2504 		if (should_sleep) {
2505 			spin_unlock(&sci->sc_state_lock);
2506 			schedule();
2507 			spin_lock(&sci->sc_state_lock);
2508 		}
2509 		finish_wait(&sci->sc_wait_daemon, &wait);
2510 		timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2511 			   time_after_eq(jiffies, sci->sc_timer.expires));
2512 
2513 		if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2514 			set_nilfs_discontinued(nilfs);
2515 	}
2516 	goto loop;
2517 
2518  end_thread:
2519 	spin_unlock(&sci->sc_state_lock);
2520 
2521 	/* end sync. */
2522 	sci->sc_task = NULL;
2523 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2524 	return 0;
2525 }
2526 
2527 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2528 {
2529 	struct task_struct *t;
2530 
2531 	t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2532 	if (IS_ERR(t)) {
2533 		int err = PTR_ERR(t);
2534 
2535 		printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2536 		       err);
2537 		return err;
2538 	}
2539 	wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2540 	return 0;
2541 }
2542 
2543 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2544 	__acquires(&sci->sc_state_lock)
2545 	__releases(&sci->sc_state_lock)
2546 {
2547 	sci->sc_state |= NILFS_SEGCTOR_QUIT;
2548 
2549 	while (sci->sc_task) {
2550 		wake_up(&sci->sc_wait_daemon);
2551 		spin_unlock(&sci->sc_state_lock);
2552 		wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2553 		spin_lock(&sci->sc_state_lock);
2554 	}
2555 }
2556 
2557 /*
2558  * Setup & clean-up functions
2559  */
2560 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2561 					       struct nilfs_root *root)
2562 {
2563 	struct the_nilfs *nilfs = sb->s_fs_info;
2564 	struct nilfs_sc_info *sci;
2565 
2566 	sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2567 	if (!sci)
2568 		return NULL;
2569 
2570 	sci->sc_super = sb;
2571 
2572 	nilfs_get_root(root);
2573 	sci->sc_root = root;
2574 
2575 	init_waitqueue_head(&sci->sc_wait_request);
2576 	init_waitqueue_head(&sci->sc_wait_daemon);
2577 	init_waitqueue_head(&sci->sc_wait_task);
2578 	spin_lock_init(&sci->sc_state_lock);
2579 	INIT_LIST_HEAD(&sci->sc_dirty_files);
2580 	INIT_LIST_HEAD(&sci->sc_segbufs);
2581 	INIT_LIST_HEAD(&sci->sc_write_logs);
2582 	INIT_LIST_HEAD(&sci->sc_gc_inodes);
2583 	init_timer(&sci->sc_timer);
2584 
2585 	sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2586 	sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2587 	sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2588 
2589 	if (nilfs->ns_interval)
2590 		sci->sc_interval = HZ * nilfs->ns_interval;
2591 	if (nilfs->ns_watermark)
2592 		sci->sc_watermark = nilfs->ns_watermark;
2593 	return sci;
2594 }
2595 
2596 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2597 {
2598 	int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2599 
2600 	/* The segctord thread was stopped and its timer was removed.
2601 	   But some tasks remain. */
2602 	do {
2603 		struct nilfs_transaction_info ti;
2604 
2605 		nilfs_transaction_lock(sci->sc_super, &ti, 0);
2606 		ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2607 		nilfs_transaction_unlock(sci->sc_super);
2608 
2609 	} while (ret && retrycount-- > 0);
2610 }
2611 
2612 /**
2613  * nilfs_segctor_destroy - destroy the segment constructor.
2614  * @sci: nilfs_sc_info
2615  *
2616  * nilfs_segctor_destroy() kills the segctord thread and frees
2617  * the nilfs_sc_info struct.
2618  * Caller must hold the segment semaphore.
2619  */
2620 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2621 {
2622 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2623 	int flag;
2624 
2625 	up_write(&nilfs->ns_segctor_sem);
2626 
2627 	spin_lock(&sci->sc_state_lock);
2628 	nilfs_segctor_kill_thread(sci);
2629 	flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2630 		|| sci->sc_seq_request != sci->sc_seq_done);
2631 	spin_unlock(&sci->sc_state_lock);
2632 
2633 	if (flag || !nilfs_segctor_confirm(sci))
2634 		nilfs_segctor_write_out(sci);
2635 
2636 	if (!list_empty(&sci->sc_dirty_files)) {
2637 		nilfs_warning(sci->sc_super, __func__,
2638 			      "dirty file(s) after the final construction\n");
2639 		nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2640 	}
2641 
2642 	WARN_ON(!list_empty(&sci->sc_segbufs));
2643 	WARN_ON(!list_empty(&sci->sc_write_logs));
2644 
2645 	nilfs_put_root(sci->sc_root);
2646 
2647 	down_write(&nilfs->ns_segctor_sem);
2648 
2649 	del_timer_sync(&sci->sc_timer);
2650 	kfree(sci);
2651 }
2652 
2653 /**
2654  * nilfs_attach_log_writer - attach log writer
2655  * @sb: super block instance
2656  * @root: root object of the current filesystem tree
2657  *
2658  * This allocates a log writer object, initializes it, and starts the
2659  * log writer.
2660  *
2661  * Return Value: On success, 0 is returned. On error, one of the following
2662  * negative error code is returned.
2663  *
2664  * %-ENOMEM - Insufficient memory available.
2665  */
2666 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2667 {
2668 	struct the_nilfs *nilfs = sb->s_fs_info;
2669 	int err;
2670 
2671 	if (nilfs->ns_writer) {
2672 		/*
2673 		 * This happens if the filesystem was remounted
2674 		 * read/write after nilfs_error degenerated it into a
2675 		 * read-only mount.
2676 		 */
2677 		nilfs_detach_log_writer(sb);
2678 	}
2679 
2680 	nilfs->ns_writer = nilfs_segctor_new(sb, root);
2681 	if (!nilfs->ns_writer)
2682 		return -ENOMEM;
2683 
2684 	err = nilfs_segctor_start_thread(nilfs->ns_writer);
2685 	if (err) {
2686 		kfree(nilfs->ns_writer);
2687 		nilfs->ns_writer = NULL;
2688 	}
2689 	return err;
2690 }
2691 
2692 /**
2693  * nilfs_detach_log_writer - destroy log writer
2694  * @sb: super block instance
2695  *
2696  * This kills log writer daemon, frees the log writer object, and
2697  * destroys list of dirty files.
2698  */
2699 void nilfs_detach_log_writer(struct super_block *sb)
2700 {
2701 	struct the_nilfs *nilfs = sb->s_fs_info;
2702 	LIST_HEAD(garbage_list);
2703 
2704 	down_write(&nilfs->ns_segctor_sem);
2705 	if (nilfs->ns_writer) {
2706 		nilfs_segctor_destroy(nilfs->ns_writer);
2707 		nilfs->ns_writer = NULL;
2708 	}
2709 
2710 	/* Force to free the list of dirty files */
2711 	spin_lock(&nilfs->ns_inode_lock);
2712 	if (!list_empty(&nilfs->ns_dirty_files)) {
2713 		list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2714 		nilfs_warning(sb, __func__,
2715 			      "Hit dirty file after stopped log writer\n");
2716 	}
2717 	spin_unlock(&nilfs->ns_inode_lock);
2718 	up_write(&nilfs->ns_segctor_sem);
2719 
2720 	nilfs_dispose_list(nilfs, &garbage_list, 1);
2721 }
2722