projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
nilfs2: call nilfs_error inside bmap routines
[linux-2.6-block.git] / fs / nilfs2 / segment.c
CommitLineData
9ff05123
RK		 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>
5a0e3ad6 35#include <linux/slab.h>
9ff05123
RK		 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"
9ff05123
RK		 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 */
55enum {
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 */
66enum {
67 NILFS_ST_INIT = 0,
68 NILFS_ST_GC, /* Collecting dirty blocks for GC */
69 NILFS_ST_FILE,
9ff05123
RK		 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 */
071cb4b8
RK		 82#define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
83#define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
9ff05123
RK		 84
85/* Operations depending on the construction mode and file type */
86struct 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 */
104static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
105static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
106static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
107static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
108 int);
109
110#define nilfs_cnt32_gt(a, b) \
111 (typecheck(__u32, a) && typecheck(__u32, b) && \
112 ((__s32)(b) - (__s32)(a) < 0))
113#define nilfs_cnt32_ge(a, b) \
114 (typecheck(__u32, a) && typecheck(__u32, b) && \
115 ((__s32)(a) - (__s32)(b) >= 0))
116#define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
117#define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
118
9ff05123
RK		 119static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
120{
121 struct nilfs_transaction_info *cur_ti = current->journal_info;
122 void *save = NULL;
123
124 if (cur_ti) {
125 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
126 return ++cur_ti->ti_count;
127 else {
128 /*
129 * If journal_info field is occupied by other FS,
47420c79
RK		 130 * it is saved and will be restored on
131 * nilfs_transaction_commit().
9ff05123
RK		 132 */
133 printk(KERN_WARNING
134 "NILFS warning: journal info from a different "
135 "FS\n");
136 save = current->journal_info;
137 }
138 }
139 if (!ti) {
140 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
141 if (!ti)
142 return -ENOMEM;
143 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
144 } else {
145 ti->ti_flags = 0;
146 }
147 ti->ti_count = 0;
148 ti->ti_save = save;
149 ti->ti_magic = NILFS_TI_MAGIC;
150 current->journal_info = ti;
151 return 0;
152}
153
154/**
155 * nilfs_transaction_begin - start indivisible file operations.
156 * @sb: super block
157 * @ti: nilfs_transaction_info
158 * @vacancy_check: flags for vacancy rate checks
159 *
160 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
161 * the segment semaphore, to make a segment construction and write tasks
47420c79 162 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
9ff05123
RK		 163 * The region enclosed by these two functions can be nested. To avoid a
164 * deadlock, the semaphore is only acquired or released in the outermost call.
165 *
166 * This function allocates a nilfs_transaction_info struct to keep context
167 * information on it. It is initialized and hooked onto the current task in
168 * the outermost call. If a pre-allocated struct is given to @ti, it is used
7a65004b 169 * instead; otherwise a new struct is assigned from a slab.
9ff05123
RK		 170 *
171 * When @vacancy_check flag is set, this function will check the amount of
172 * free space, and will wait for the GC to reclaim disk space if low capacity.
173 *
174 * Return Value: On success, 0 is returned. On error, one of the following
175 * negative error code is returned.
176 *
177 * %-ENOMEM - Insufficient memory available.
178 *
9ff05123
RK		 179 * %-ENOSPC - No space left on device
180 */
181int nilfs_transaction_begin(struct super_block *sb,
182 struct nilfs_transaction_info *ti,
183 int vacancy_check)
184{
185 struct nilfs_sb_info *sbi;
186 struct the_nilfs *nilfs;
187 int ret = nilfs_prepare_segment_lock(ti);
188
189 if (unlikely(ret < 0))
190 return ret;
191 if (ret > 0)
192 return 0;
193
5beb6e0b
RK		 194 vfs_check_frozen(sb, SB_FREEZE_WRITE);
195
9ff05123
RK		 196 sbi = NILFS_SB(sb);
197 nilfs = sbi->s_nilfs;
198 down_read(&nilfs->ns_segctor_sem);
199 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
200 up_read(&nilfs->ns_segctor_sem);
201 ret = -ENOSPC;
202 goto failed;
203 }
204 return 0;
205
206 failed:
207 ti = current->journal_info;
208 current->journal_info = ti->ti_save;
209 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
210 kmem_cache_free(nilfs_transaction_cachep, ti);
211 return ret;
212}
213
214/**
47420c79 215 * nilfs_transaction_commit - commit indivisible file operations.
9ff05123 216 * @sb: super block
9ff05123 217 *
47420c79
RK		 218 * nilfs_transaction_commit() releases the read semaphore which is
219 * acquired by nilfs_transaction_begin(). This is only performed
220 * in outermost call of this function. If a commit flag is set,
221 * nilfs_transaction_commit() sets a timer to start the segment
222 * constructor. If a sync flag is set, it starts construction
223 * directly.
9ff05123 224 */
47420c79 225int nilfs_transaction_commit(struct super_block *sb)
9ff05123
RK		 226{
227 struct nilfs_transaction_info *ti = current->journal_info;
228 struct nilfs_sb_info *sbi;
229 struct nilfs_sc_info *sci;
230 int err = 0;
231
232 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
47420c79 233 ti->ti_flags |= NILFS_TI_COMMIT;
9ff05123
RK		 234 if (ti->ti_count > 0) {
235 ti->ti_count--;
236 return 0;
237 }
238 sbi = NILFS_SB(sb);
239 sci = NILFS_SC(sbi);
240 if (sci != NULL) {
241 if (ti->ti_flags & NILFS_TI_COMMIT)
242 nilfs_segctor_start_timer(sci);
243 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
244 sci->sc_watermark)
245 nilfs_segctor_do_flush(sci, 0);
246 }
247 up_read(&sbi->s_nilfs->ns_segctor_sem);
248 current->journal_info = ti->ti_save;
249
250 if (ti->ti_flags & NILFS_TI_SYNC)
251 err = nilfs_construct_segment(sb);
252 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
253 kmem_cache_free(nilfs_transaction_cachep, ti);
254 return err;
255}
256
47420c79
RK		 257void nilfs_transaction_abort(struct super_block *sb)
258{
259 struct nilfs_transaction_info *ti = current->journal_info;
260
261 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
262 if (ti->ti_count > 0) {
263 ti->ti_count--;
264 return;
265 }
266 up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
267
268 current->journal_info = ti->ti_save;
269 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
270 kmem_cache_free(nilfs_transaction_cachep, ti);
271}
272
9ff05123
RK		 273void nilfs_relax_pressure_in_lock(struct super_block *sb)
274{
275 struct nilfs_sb_info *sbi = NILFS_SB(sb);
276 struct nilfs_sc_info *sci = NILFS_SC(sbi);
277 struct the_nilfs *nilfs = sbi->s_nilfs;
278
279 if (!sci || !sci->sc_flush_request)
280 return;
281
282 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
283 up_read(&nilfs->ns_segctor_sem);
284
285 down_write(&nilfs->ns_segctor_sem);
286 if (sci->sc_flush_request &&
287 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
288 struct nilfs_transaction_info *ti = current->journal_info;
289
290 ti->ti_flags |= NILFS_TI_WRITER;
291 nilfs_segctor_do_immediate_flush(sci);
292 ti->ti_flags &= ~NILFS_TI_WRITER;
293 }
294 downgrade_write(&nilfs->ns_segctor_sem);
295}
296
297static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
298 struct nilfs_transaction_info *ti,
299 int gcflag)
300{
301 struct nilfs_transaction_info *cur_ti = current->journal_info;
302
1f5abe7e 303 WARN_ON(cur_ti);
9ff05123
RK		 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(&sbi->s_nilfs->ns_segctor_sem);
313 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
314 break;
315
316 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
317
318 up_write(&sbi->s_nilfs->ns_segctor_sem);
319 yield();
320 }
321 if (gcflag)
322 ti->ti_flags |= NILFS_TI_GC;
323}
324
325static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
326{
327 struct nilfs_transaction_info *ti = current->journal_info;
328
329 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
330 BUG_ON(ti->ti_count > 0);
331
332 up_write(&sbi->s_nilfs->ns_segctor_sem);
333 current->journal_info = ti->ti_save;
334 if (!list_empty(&ti->ti_garbage))
335 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
336}
337
338static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
339 struct nilfs_segsum_pointer *ssp,
340 unsigned bytes)
341{
342 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
343 unsigned blocksize = sci->sc_super->s_blocksize;
344 void *p;
345
346 if (unlikely(ssp->offset + bytes > blocksize)) {
347 ssp->offset = 0;
348 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
349 &segbuf->sb_segsum_buffers));
350 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
351 }
352 p = ssp->bh->b_data + ssp->offset;
353 ssp->offset += bytes;
354 return p;
355}
356
357/**
358 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
359 * @sci: nilfs_sc_info
360 */
361static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
362{
363 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
364 struct buffer_head *sumbh;
365 unsigned sumbytes;
366 unsigned flags = 0;
367 int err;
368
369 if (nilfs_doing_gc())
370 flags = NILFS_SS_GC;
6c43f410 371 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
9ff05123
RK		 372 if (unlikely(err))
373 return err;
374
375 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
376 sumbytes = segbuf->sb_sum.sumbytes;
377 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
378 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
379 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
380 return 0;
381}
382
383static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
384{
385 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
386 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
387 return -E2BIG; /* The current segment is filled up
388 (internal code) */
389 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
390 return nilfs_segctor_reset_segment_buffer(sci);
391}
392
393static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
394{
395 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
396 int err;
397
398 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
399 err = nilfs_segctor_feed_segment(sci);
400 if (err)
401 return err;
402 segbuf = sci->sc_curseg;
403 }
1e2b68bf 404 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
9ff05123
RK		 405 if (likely(!err))
406 segbuf->sb_sum.flags |= NILFS_SS_SR;
407 return err;
408}
409
410/*
411 * Functions for making segment summary and payloads
412 */
413static int nilfs_segctor_segsum_block_required(
414 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
415 unsigned binfo_size)
416{
417 unsigned blocksize = sci->sc_super->s_blocksize;
418 /* Size of finfo and binfo is enough small against blocksize */
419
420 return ssp->offset + binfo_size +
421 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
422 blocksize;
423}
424
425static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
426 struct inode *inode)
427{
428 sci->sc_curseg->sb_sum.nfinfo++;
429 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
430 nilfs_segctor_map_segsum_entry(
431 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
c96fa464
RK		 432
433 if (inode->i_sb && !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
434 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
9ff05123
RK		 435 /* skip finfo */
436}
437
438static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
439 struct inode *inode)
440{
441 struct nilfs_finfo *finfo;
442 struct nilfs_inode_info *ii;
443 struct nilfs_segment_buffer *segbuf;
6c43f410 444 __u64 cno;
9ff05123
RK		 445
446 if (sci->sc_blk_cnt == 0)
447 return;
448
449 ii = NILFS_I(inode);
6c43f410
RK		 450
451 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
452 cno = ii->i_cno;
453 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
454 cno = 0;
455 else
456 cno = sci->sc_cno;
457
9ff05123
RK		 458 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
459 sizeof(*finfo));
460 finfo->fi_ino = cpu_to_le64(inode->i_ino);
461 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
462 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
6c43f410 463 finfo->fi_cno = cpu_to_le64(cno);
9ff05123
RK		 464
465 segbuf = sci->sc_curseg;
466 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
467 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
468 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
469 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
470}
471
472static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
473 struct buffer_head *bh,
474 struct inode *inode,
475 unsigned binfo_size)
476{
477 struct nilfs_segment_buffer *segbuf;
478 int required, err = 0;
479
480 retry:
481 segbuf = sci->sc_curseg;
482 required = nilfs_segctor_segsum_block_required(
483 sci, &sci->sc_binfo_ptr, binfo_size);
484 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
485 nilfs_segctor_end_finfo(sci, inode);
486 err = nilfs_segctor_feed_segment(sci);
487 if (err)
488 return err;
489 goto retry;
490 }
491 if (unlikely(required)) {
492 err = nilfs_segbuf_extend_segsum(segbuf);
493 if (unlikely(err))
494 goto failed;
495 }
496 if (sci->sc_blk_cnt == 0)
497 nilfs_segctor_begin_finfo(sci, inode);
498
499 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
500 /* Substitution to vblocknr is delayed until update_blocknr() */
501 nilfs_segbuf_add_file_buffer(segbuf, bh);
502 sci->sc_blk_cnt++;
503 failed:
504 return err;
505}
506
9ff05123
RK		 507/*
508 * Callback functions that enumerate, mark, and collect dirty blocks
509 */
510static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
511 struct buffer_head *bh, struct inode *inode)
512{
513 int err;
514
9ff05123 515 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
e828949e
RK		 516 if (err < 0)
517 return err;
9ff05123
RK		 518
519 err = nilfs_segctor_add_file_block(sci, bh, inode,
520 sizeof(struct nilfs_binfo_v));
521 if (!err)
522 sci->sc_datablk_cnt++;
523 return err;
524}
525
526static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
527 struct buffer_head *bh,
528 struct inode *inode)
529{
e828949e 530 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
9ff05123
RK		 531}
532
533static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
534 struct buffer_head *bh,
535 struct inode *inode)
536{
1f5abe7e 537 WARN_ON(!buffer_dirty(bh));
9ff05123
RK		 538 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
539}
540
541static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
542 struct nilfs_segsum_pointer *ssp,
543 union nilfs_binfo *binfo)
544{
545 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
546 sci, ssp, sizeof(*binfo_v));
547 *binfo_v = binfo->bi_v;
548}
549
550static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
551 struct nilfs_segsum_pointer *ssp,
552 union nilfs_binfo *binfo)
553{
554 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
555 sci, ssp, sizeof(*vblocknr));
556 *vblocknr = binfo->bi_v.bi_vblocknr;
557}
558
4e819509 559static struct nilfs_sc_operations nilfs_sc_file_ops = {
9ff05123
RK		 560 .collect_data = nilfs_collect_file_data,
561 .collect_node = nilfs_collect_file_node,
562 .collect_bmap = nilfs_collect_file_bmap,
563 .write_data_binfo = nilfs_write_file_data_binfo,
564 .write_node_binfo = nilfs_write_file_node_binfo,
565};
566
567static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
568 struct buffer_head *bh, struct inode *inode)
569{
570 int err;
571
572 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
e828949e
RK		 573 if (err < 0)
574 return err;
9ff05123
RK		 575
576 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
577 if (!err)
578 sci->sc_datablk_cnt++;
579 return err;
580}
581
582static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
583 struct buffer_head *bh, struct inode *inode)
584{
1f5abe7e 585 WARN_ON(!buffer_dirty(bh));
9ff05123
RK		 586 return nilfs_segctor_add_file_block(sci, bh, inode,
587 sizeof(struct nilfs_binfo_dat));
588}
589
590static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
591 struct nilfs_segsum_pointer *ssp,
592 union nilfs_binfo *binfo)
593{
594 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
595 sizeof(*blkoff));
596 *blkoff = binfo->bi_dat.bi_blkoff;
597}
598
599static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
600 struct nilfs_segsum_pointer *ssp,
601 union nilfs_binfo *binfo)
602{
603 struct nilfs_binfo_dat *binfo_dat =
604 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
605 *binfo_dat = binfo->bi_dat;
606}
607
4e819509 608static struct nilfs_sc_operations nilfs_sc_dat_ops = {
9ff05123
RK		 609 .collect_data = nilfs_collect_dat_data,
610 .collect_node = nilfs_collect_file_node,
611 .collect_bmap = nilfs_collect_dat_bmap,
612 .write_data_binfo = nilfs_write_dat_data_binfo,
613 .write_node_binfo = nilfs_write_dat_node_binfo,
614};
615
4e819509 616static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
9ff05123
RK		 617 .collect_data = nilfs_collect_file_data,
618 .collect_node = NULL,
619 .collect_bmap = NULL,
620 .write_data_binfo = nilfs_write_file_data_binfo,
621 .write_node_binfo = NULL,
622};
623
f30bf3e4
RK		 624static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
625 struct list_head *listp,
626 size_t nlimit,
627 loff_t start, loff_t end)
9ff05123 628{
9ff05123
RK		 629 struct address_space *mapping = inode->i_mapping;
630 struct pagevec pvec;
f30bf3e4
RK		 631 pgoff_t index = 0, last = ULONG_MAX;
632 size_t ndirties = 0;
633 int i;
9ff05123 634
f30bf3e4
RK		 635 if (unlikely(start != 0 || end != LLONG_MAX)) {
636 /*
637 * A valid range is given for sync-ing data pages. The
638 * range is rounded to per-page; extra dirty buffers
639 * may be included if blocksize < pagesize.
640 */
641 index = start >> PAGE_SHIFT;
642 last = end >> PAGE_SHIFT;
643 }
9ff05123
RK		 644 pagevec_init(&pvec, 0);
645 repeat:
f30bf3e4
RK		 646 if (unlikely(index > last) ||
647 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
648 min_t(pgoff_t, last - index,
649 PAGEVEC_SIZE - 1) + 1))
650 return ndirties;
9ff05123
RK		 651
652 for (i = 0; i < pagevec_count(&pvec); i++) {
653 struct buffer_head *bh, *head;
654 struct page *page = pvec.pages[i];
655
f30bf3e4
RK		 656 if (unlikely(page->index > last))
657 break;
658
9ff05123
RK		 659 if (mapping->host) {
660 lock_page(page);
661 if (!page_has_buffers(page))
662 create_empty_buffers(page,
663 1 << inode->i_blkbits, 0);
664 unlock_page(page);
665 }
666
667 bh = head = page_buffers(page);
668 do {
f30bf3e4
RK		 669 if (!buffer_dirty(bh))
670 continue;
671 get_bh(bh);
672 list_add_tail(&bh->b_assoc_buffers, listp);
673 ndirties++;
674 if (unlikely(ndirties >= nlimit)) {
675 pagevec_release(&pvec);
676 cond_resched();
677 return ndirties;
9ff05123 678 }
f30bf3e4 679 } while (bh = bh->b_this_page, bh != head);
9ff05123
RK		 680 }
681 pagevec_release(&pvec);
682 cond_resched();
f30bf3e4 683 goto repeat;
9ff05123
RK		 684}
685
686static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
687 struct list_head *listp)
688{
689 struct nilfs_inode_info *ii = NILFS_I(inode);
690 struct address_space *mapping = &ii->i_btnode_cache;
691 struct pagevec pvec;
692 struct buffer_head *bh, *head;
693 unsigned int i;
694 pgoff_t index = 0;
695
696 pagevec_init(&pvec, 0);
697
698 while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
699 PAGEVEC_SIZE)) {
700 for (i = 0; i < pagevec_count(&pvec); i++) {
701 bh = head = page_buffers(pvec.pages[i]);
702 do {
703 if (buffer_dirty(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
716static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
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(&sbi->s_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 &sbi->s_dirty_files);
736 continue;
737 }
738 ivec[nv++] = ii;
739 if (nv == SC_N_INODEVEC)
740 break;
741 }
742 spin_unlock(&sbi->s_inode_lock);
743
744 for (pii = ivec; nv > 0; pii++, nv--)
745 iput(&(*pii)->vfs_inode);
746 }
747}
748
e912a5b6
RK		 749static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
750 struct nilfs_root *root)
9ff05123 751{
9ff05123
RK		 752 int ret = 0;
753
e912a5b6 754 if (nilfs_mdt_fetch_dirty(root->ifile))
9ff05123
RK		 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())
761 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
762 ret++;
763 return ret;
764}
765
766static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
767{
768 return list_empty(&sci->sc_dirty_files) &&
769 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
071cb4b8 770 sci->sc_nfreesegs == 0 &&
9ff05123
RK		 771 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
772}
773
774static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
775{
776 struct nilfs_sb_info *sbi = sci->sc_sbi;
777 int ret = 0;
778
e912a5b6 779 if (nilfs_test_metadata_dirty(sbi->s_nilfs, sci->sc_root))
9ff05123
RK		 780 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
781
782 spin_lock(&sbi->s_inode_lock);
783 if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
784 ret++;
785
786 spin_unlock(&sbi->s_inode_lock);
787 return ret;
788}
789
790static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
791{
792 struct nilfs_sb_info *sbi = sci->sc_sbi;
793 struct the_nilfs *nilfs = sbi->s_nilfs;
794
e912a5b6 795 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
9ff05123
RK		 796 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
797 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
798 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
799}
800
801static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
802{
803 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
804 struct buffer_head *bh_cp;
805 struct nilfs_checkpoint *raw_cp;
806 int err;
807
808 /* XXX: this interface will be changed */
809 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
810 &raw_cp, &bh_cp);
811 if (likely(!err)) {
812 /* The following code is duplicated with cpfile. But, it is
813 needed to collect the checkpoint even if it was not newly
814 created */
815 nilfs_mdt_mark_buffer_dirty(bh_cp);
816 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
817 nilfs_cpfile_put_checkpoint(
818 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
1f5abe7e
RK		 819 } else
820 WARN_ON(err == -EINVAL || err == -ENOENT);
821
9ff05123
RK		 822 return err;
823}
824
825static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
826{
827 struct nilfs_sb_info *sbi = sci->sc_sbi;
828 struct the_nilfs *nilfs = sbi->s_nilfs;
829 struct buffer_head *bh_cp;
830 struct nilfs_checkpoint *raw_cp;
831 int err;
832
833 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
834 &raw_cp, &bh_cp);
835 if (unlikely(err)) {
1f5abe7e 836 WARN_ON(err == -EINVAL || err == -ENOENT);
9ff05123
RK		 837 goto failed_ibh;
838 }
839 raw_cp->cp_snapshot_list.ssl_next = 0;
840 raw_cp->cp_snapshot_list.ssl_prev = 0;
841 raw_cp->cp_inodes_count =
b7c06342 842 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
9ff05123 843 raw_cp->cp_blocks_count =
b7c06342 844 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
9ff05123
RK		 845 raw_cp->cp_nblk_inc =
846 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
847 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
848 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
458c5b08 849
c96fa464
RK		 850 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
851 nilfs_checkpoint_clear_minor(raw_cp);
852 else
853 nilfs_checkpoint_set_minor(raw_cp);
854
e912a5b6
RK		 855 nilfs_write_inode_common(sci->sc_root->ifile,
856 &raw_cp->cp_ifile_inode, 1);
9ff05123
RK		 857 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
858 return 0;
859
860 failed_ibh:
861 return err;
862}
863
864static void nilfs_fill_in_file_bmap(struct inode *ifile,
865 struct nilfs_inode_info *ii)
866
867{
868 struct buffer_head *ibh;
869 struct nilfs_inode *raw_inode;
870
871 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
872 ibh = ii->i_bh;
873 BUG_ON(!ibh);
874 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
875 ibh);
876 nilfs_bmap_write(ii->i_bmap, raw_inode);
877 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
878 }
879}
880
e912a5b6 881static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
9ff05123
RK		 882{
883 struct nilfs_inode_info *ii;
884
885 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
e912a5b6 886 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
9ff05123
RK		 887 set_bit(NILFS_I_COLLECTED, &ii->i_state);
888 }
9ff05123
RK		 889}
890
9ff05123
RK		 891static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
892 struct the_nilfs *nilfs)
893{
1e2b68bf
RK		 894 struct buffer_head *bh_sr;
895 struct nilfs_super_root *raw_sr;
9ff05123
RK		 896 unsigned isz = nilfs->ns_inode_size;
897
1e2b68bf
RK		 898 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
899 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
900
9ff05123
RK		 901 raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
902 raw_sr->sr_nongc_ctime
903 = cpu_to_le64(nilfs_doing_gc() ?
904 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
905 raw_sr->sr_flags = 0;
906
3961f0e2
RK		 907 nilfs_write_inode_common(nilfs_dat_inode(nilfs), (void *)raw_sr +
908 NILFS_SR_DAT_OFFSET(isz), 1);
909 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
910 NILFS_SR_CPFILE_OFFSET(isz), 1);
911 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
912 NILFS_SR_SUFILE_OFFSET(isz), 1);
9ff05123
RK		 913}
914
915static 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
925static 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
9ff05123
RK		 938static 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_entry(listp->next, struct buffer_head,
962 b_assoc_buffers);
963 list_del_init(&bh->b_assoc_buffers);
964 brelse(bh);
965 }
966 return err;
967}
968
f30bf3e4
RK		 969static 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
9ff05123
RK		 976static 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);
f30bf3e4 982 int err;
9ff05123
RK		 983
984 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
f30bf3e4
RK		 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(
9ff05123 991 sci, inode, &data_buffers,
f30bf3e4
RK		 992 sc_ops->collect_data);
993 BUG_ON(!err); /* always receive -E2BIG or true error */
9ff05123
RK		 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
1029static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1030 struct inode *inode)
1031{
1032 LIST_HEAD(data_buffers);
f30bf3e4
RK		 1033 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1034 int err;
9ff05123 1035
f30bf3e4
RK		 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) {
9ff05123 1043 nilfs_segctor_end_finfo(sci, inode);
f30bf3e4
RK		 1044 BUG_ON(n > rest);
1045 /* always receive -E2BIG or true error if n > rest */
1046 }
9ff05123
RK		 1047 return err;
1048}
1049
1050static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1051{
1052 struct nilfs_sb_info *sbi = sci->sc_sbi;
1053 struct the_nilfs *nilfs = sbi->s_nilfs;
1054 struct list_head *head;
1055 struct nilfs_inode_info *ii;
071cb4b8 1056 size_t ndone;
9ff05123
RK		 1057 int err = 0;
1058
1059 switch (sci->sc_stage.scnt) {
1060 case NILFS_ST_INIT:
1061 /* Pre-processes */
1062 sci->sc_stage.flags = 0;
1063
1064 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1065 sci->sc_nblk_inc = 0;
1066 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1067 if (mode == SC_LSEG_DSYNC) {
1068 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1069 goto dsync_mode;
1070 }
1071 }
1072
1073 sci->sc_stage.dirty_file_ptr = NULL;
1074 sci->sc_stage.gc_inode_ptr = NULL;
1075 if (mode == SC_FLUSH_DAT) {
1076 sci->sc_stage.scnt = NILFS_ST_DAT;
1077 goto dat_stage;
1078 }
1079 sci->sc_stage.scnt++; /* Fall through */
1080 case NILFS_ST_GC:
1081 if (nilfs_doing_gc()) {
1082 head = &sci->sc_gc_inodes;
1083 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1084 head, i_dirty);
1085 list_for_each_entry_continue(ii, head, i_dirty) {
1086 err = nilfs_segctor_scan_file(
1087 sci, &ii->vfs_inode,
1088 &nilfs_sc_file_ops);
1089 if (unlikely(err)) {
1090 sci->sc_stage.gc_inode_ptr = list_entry(
1091 ii->i_dirty.prev,
1092 struct nilfs_inode_info,
1093 i_dirty);
1094 goto break_or_fail;
1095 }
1096 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1097 }
1098 sci->sc_stage.gc_inode_ptr = NULL;
1099 }
1100 sci->sc_stage.scnt++; /* Fall through */
1101 case NILFS_ST_FILE:
1102 head = &sci->sc_dirty_files;
1103 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1104 i_dirty);
1105 list_for_each_entry_continue(ii, head, i_dirty) {
1106 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1107
1108 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1109 &nilfs_sc_file_ops);
1110 if (unlikely(err)) {
1111 sci->sc_stage.dirty_file_ptr =
1112 list_entry(ii->i_dirty.prev,
1113 struct nilfs_inode_info,
1114 i_dirty);
1115 goto break_or_fail;
1116 }
1117 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1118 /* XXX: required ? */
1119 }
1120 sci->sc_stage.dirty_file_ptr = NULL;
1121 if (mode == SC_FLUSH_FILE) {
1122 sci->sc_stage.scnt = NILFS_ST_DONE;
1123 return 0;
1124 }
9ff05123
RK		 1125 sci->sc_stage.scnt++;
1126 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1127 /* Fall through */
1128 case NILFS_ST_IFILE:
e912a5b6 1129 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
9ff05123
RK		 1130 &nilfs_sc_file_ops);
1131 if (unlikely(err))
1132 break;
1133 sci->sc_stage.scnt++;
1134 /* Creating a checkpoint */
1135 err = nilfs_segctor_create_checkpoint(sci);
1136 if (unlikely(err))
1137 break;
1138 /* Fall through */
1139 case NILFS_ST_CPFILE:
1140 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1141 &nilfs_sc_file_ops);
1142 if (unlikely(err))
1143 break;
1144 sci->sc_stage.scnt++; /* Fall through */
1145 case NILFS_ST_SUFILE:
071cb4b8
RK		 1146 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1147 sci->sc_nfreesegs, &ndone);
1148 if (unlikely(err)) {
1149 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1150 sci->sc_freesegs, ndone,
1151 NULL);
9ff05123 1152 break;
071cb4b8
RK		 1153 }
1154 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1155
9ff05123
RK		 1156 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1157 &nilfs_sc_file_ops);
1158 if (unlikely(err))
1159 break;
1160 sci->sc_stage.scnt++; /* Fall through */
1161 case NILFS_ST_DAT:
1162 dat_stage:
1163 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1164 &nilfs_sc_dat_ops);
1165 if (unlikely(err))
1166 break;
1167 if (mode == SC_FLUSH_DAT) {
1168 sci->sc_stage.scnt = NILFS_ST_DONE;
1169 return 0;
1170 }
1171 sci->sc_stage.scnt++; /* Fall through */
1172 case NILFS_ST_SR:
1173 if (mode == SC_LSEG_SR) {
1174 /* Appending a super root */
1175 err = nilfs_segctor_add_super_root(sci);
1176 if (unlikely(err))
1177 break;
1178 }
1179 /* End of a logical segment */
1180 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1181 sci->sc_stage.scnt = NILFS_ST_DONE;
1182 return 0;
1183 case NILFS_ST_DSYNC:
1184 dsync_mode:
1185 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
f30bf3e4 1186 ii = sci->sc_dsync_inode;
9ff05123
RK		 1187 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1188 break;
1189
1190 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1191 if (unlikely(err))
1192 break;
9ff05123
RK		 1193 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1194 sci->sc_stage.scnt = NILFS_ST_DONE;
1195 return 0;
1196 case NILFS_ST_DONE:
1197 return 0;
1198 default:
1199 BUG();
1200 }
1201
1202 break_or_fail:
1203 return err;
1204}
1205
a694291a
RK		 1206/**
1207 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1208 * @sci: nilfs_sc_info
1209 * @nilfs: nilfs object
1210 */
9ff05123
RK		 1211static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1212 struct the_nilfs *nilfs)
1213{
a694291a 1214 struct nilfs_segment_buffer *segbuf, *prev;
9ff05123 1215 __u64 nextnum;
a694291a 1216 int err, alloc = 0;
9ff05123 1217
a694291a
RK		 1218 segbuf = nilfs_segbuf_new(sci->sc_super);
1219 if (unlikely(!segbuf))
1220 return -ENOMEM;
9ff05123 1221
a694291a
RK		 1222 if (list_empty(&sci->sc_write_logs)) {
1223 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1224 nilfs->ns_pseg_offset, nilfs);
1225 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1226 nilfs_shift_to_next_segment(nilfs);
1227 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1228 }
9ff05123 1229
a694291a
RK		 1230 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1231 nextnum = nilfs->ns_nextnum;
1232
1233 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1234 /* Start from the head of a new full segment */
1235 alloc++;
1236 } else {
1237 /* Continue logs */
1238 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1239 nilfs_segbuf_map_cont(segbuf, prev);
1240 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1241 nextnum = prev->sb_nextnum;
1242
1243 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1244 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1245 segbuf->sb_sum.seg_seq++;
1246 alloc++;
1247 }
9ff05123 1248 }
9ff05123 1249
61a189e9 1250 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
a694291a
RK		 1251 if (err)
1252 goto failed;
9ff05123 1253
a694291a 1254 if (alloc) {
cece5520 1255 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
a694291a
RK		 1256 if (err)
1257 goto failed;
1258 }
9ff05123
RK		 1259 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1260
a694291a
RK		 1261 BUG_ON(!list_empty(&sci->sc_segbufs));
1262 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1263 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
cece5520 1264 return 0;
a694291a
RK		 1265
1266 failed:
1267 nilfs_segbuf_free(segbuf);
1268 return err;
9ff05123
RK		 1269}
1270
1271static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1272 struct the_nilfs *nilfs, int nadd)
1273{
e29df395 1274 struct nilfs_segment_buffer *segbuf, *prev;
9ff05123
RK		 1275 struct inode *sufile = nilfs->ns_sufile;
1276 __u64 nextnextnum;
1277 LIST_HEAD(list);
1278 int err, ret, i;
1279
1280 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1281 /*
1282 * Since the segment specified with nextnum might be allocated during
1283 * the previous construction, the buffer including its segusage may
1284 * not be dirty. The following call ensures that the buffer is dirty
1285 * and will pin the buffer on memory until the sufile is written.
1286 */
61a189e9 1287 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
9ff05123
RK		 1288 if (unlikely(err))
1289 return err;
1290
1291 for (i = 0; i < nadd; i++) {
1292 /* extend segment info */
1293 err = -ENOMEM;
1294 segbuf = nilfs_segbuf_new(sci->sc_super);
1295 if (unlikely(!segbuf))
1296 goto failed;
1297
1298 /* map this buffer to region of segment on-disk */
cece5520 1299 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
9ff05123
RK		 1300 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1301
1302 /* allocate the next next full segment */
1303 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1304 if (unlikely(err))
1305 goto failed_segbuf;
1306
1307 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1308 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1309
1310 list_add_tail(&segbuf->sb_list, &list);
1311 prev = segbuf;
1312 }
0935db74 1313 list_splice_tail(&list, &sci->sc_segbufs);
9ff05123
RK		 1314 return 0;
1315
1316 failed_segbuf:
1317 nilfs_segbuf_free(segbuf);
1318 failed:
e29df395 1319 list_for_each_entry(segbuf, &list, sb_list) {
9ff05123 1320 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1321 WARN_ON(ret); /* never fails */
9ff05123 1322 }
e29df395 1323 nilfs_destroy_logs(&list);
9ff05123
RK		 1324 return err;
1325}
1326
a694291a
RK		 1327static void nilfs_free_incomplete_logs(struct list_head *logs,
1328 struct the_nilfs *nilfs)
9ff05123 1329{
a694291a
RK		 1330 struct nilfs_segment_buffer *segbuf, *prev;
1331 struct inode *sufile = nilfs->ns_sufile;
9284ad2a 1332 int ret;
9ff05123 1333
a694291a 1334 segbuf = NILFS_FIRST_SEGBUF(logs);
9ff05123 1335 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
a694291a 1336 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1337 WARN_ON(ret); /* never fails */
9ff05123 1338 }
9284ad2a 1339 if (atomic_read(&segbuf->sb_err)) {
9ff05123
RK		 1340 /* Case 1: The first segment failed */
1341 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1342 /* Case 1a: Partial segment appended into an existing
1343 segment */
1344 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1345 segbuf->sb_fseg_end);
1346 else /* Case 1b: New full segment */
1347 set_nilfs_discontinued(nilfs);
9ff05123
RK		 1348 }
1349
a694291a
RK		 1350 prev = segbuf;
1351 list_for_each_entry_continue(segbuf, logs, sb_list) {
1352 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1353 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1354 WARN_ON(ret); /* never fails */
1355 }
9284ad2a
RK		 1356 if (atomic_read(&segbuf->sb_err) &&
1357 segbuf->sb_segnum != nilfs->ns_nextnum)
1358 /* Case 2: extended segment (!= next) failed */
a694291a
RK		 1359 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1360 prev = segbuf;
9ff05123 1361 }
9ff05123
RK		 1362}
1363
1364static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1365 struct inode *sufile)
1366{
1367 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1368 unsigned long live_blocks;
1369 int ret;
1370
1371 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
9ff05123
RK		 1372 live_blocks = segbuf->sb_sum.nblocks +
1373 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
071ec54d
RK		 1374 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1375 live_blocks,
1376 sci->sc_seg_ctime);
1377 WARN_ON(ret); /* always succeed because the segusage is dirty */
9ff05123
RK		 1378 }
1379}
1380
a694291a 1381static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
9ff05123
RK		 1382{
1383 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1384 int ret;
1385
a694291a 1386 segbuf = NILFS_FIRST_SEGBUF(logs);
071ec54d
RK		 1387 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1388 segbuf->sb_pseg_start -
1389 segbuf->sb_fseg_start, 0);
1390 WARN_ON(ret); /* always succeed because the segusage is dirty */
9ff05123 1391
a694291a 1392 list_for_each_entry_continue(segbuf, logs, sb_list) {
071ec54d
RK		 1393 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1394 0, 0);
1f5abe7e 1395 WARN_ON(ret); /* always succeed */
9ff05123
RK		 1396 }
1397}
1398
1399static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1400 struct nilfs_segment_buffer *last,
1401 struct inode *sufile)
1402{
e29df395 1403 struct nilfs_segment_buffer *segbuf = last;
9ff05123
RK		 1404 int ret;
1405
e29df395 1406 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
9ff05123
RK		 1407 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1408 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1409 WARN_ON(ret);
9ff05123 1410 }
e29df395 1411 nilfs_truncate_logs(&sci->sc_segbufs, last);
9ff05123
RK		 1412}
1413
1414
1415static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1416 struct the_nilfs *nilfs, int mode)
1417{
1418 struct nilfs_cstage prev_stage = sci->sc_stage;
1419 int err, nadd = 1;
1420
1421 /* Collection retry loop */
1422 for (;;) {
9ff05123
RK		 1423 sci->sc_nblk_this_inc = 0;
1424 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1425
1426 err = nilfs_segctor_reset_segment_buffer(sci);
1427 if (unlikely(err))
1428 goto failed;
1429
1430 err = nilfs_segctor_collect_blocks(sci, mode);
1431 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1432 if (!err)
1433 break;
1434
1435 if (unlikely(err != -E2BIG))
1436 goto failed;
1437
1438 /* The current segment is filled up */
1439 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1440 break;
1441
2d8428ac
RK		 1442 nilfs_clear_logs(&sci->sc_segbufs);
1443
1444 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1445 if (unlikely(err))
1446 return err;
1447
071cb4b8
RK		 1448 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1449 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1450 sci->sc_freesegs,
1451 sci->sc_nfreesegs,
1452 NULL);
1453 WARN_ON(err); /* do not happen */
1454 }
9ff05123
RK		 1455 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1456 sci->sc_stage = prev_stage;
1457 }
1458 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1459 return 0;
1460
1461 failed:
1462 return err;
1463}
1464
1465static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1466 struct buffer_head *new_bh)
1467{
1468 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1469
1470 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1471 /* The caller must release old_bh */
1472}
1473
1474static int
1475nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1476 struct nilfs_segment_buffer *segbuf,
1477 int mode)
1478{
1479 struct inode *inode = NULL;
1480 sector_t blocknr;
1481 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1482 unsigned long nblocks = 0, ndatablk = 0;
1483 struct nilfs_sc_operations *sc_op = NULL;
1484 struct nilfs_segsum_pointer ssp;
1485 struct nilfs_finfo *finfo = NULL;
1486 union nilfs_binfo binfo;
1487 struct buffer_head *bh, *bh_org;
1488 ino_t ino = 0;
1489 int err = 0;
1490
1491 if (!nfinfo)
1492 goto out;
1493
1494 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1495 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1496 ssp.offset = sizeof(struct nilfs_segment_summary);
1497
1498 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1e2b68bf 1499 if (bh == segbuf->sb_super_root)
9ff05123
RK		 1500 break;
1501 if (!finfo) {
1502 finfo = nilfs_segctor_map_segsum_entry(
1503 sci, &ssp, sizeof(*finfo));
1504 ino = le64_to_cpu(finfo->fi_ino);
1505 nblocks = le32_to_cpu(finfo->fi_nblocks);
1506 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1507
1508 if (buffer_nilfs_node(bh))
1509 inode = NILFS_BTNC_I(bh->b_page->mapping);
1510 else
1511 inode = NILFS_AS_I(bh->b_page->mapping);
1512
1513 if (mode == SC_LSEG_DSYNC)
1514 sc_op = &nilfs_sc_dsync_ops;
1515 else if (ino == NILFS_DAT_INO)
1516 sc_op = &nilfs_sc_dat_ops;
1517 else /* file blocks */
1518 sc_op = &nilfs_sc_file_ops;
1519 }
1520 bh_org = bh;
1521 get_bh(bh_org);
1522 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1523 &binfo);
1524 if (bh != bh_org)
1525 nilfs_list_replace_buffer(bh_org, bh);
1526 brelse(bh_org);
1527 if (unlikely(err))
1528 goto failed_bmap;
1529
1530 if (ndatablk > 0)
1531 sc_op->write_data_binfo(sci, &ssp, &binfo);
1532 else
1533 sc_op->write_node_binfo(sci, &ssp, &binfo);
1534
1535 blocknr++;
1536 if (--nblocks == 0) {
1537 finfo = NULL;
1538 if (--nfinfo == 0)
1539 break;
1540 } else if (ndatablk > 0)
1541 ndatablk--;
1542 }
1543 out:
1544 return 0;
1545
1546 failed_bmap:
9ff05123
RK		 1547 return err;
1548}
1549
1550static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1551{
1552 struct nilfs_segment_buffer *segbuf;
1553 int err;
1554
1555 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1556 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1557 if (unlikely(err))
1558 return err;
1559 nilfs_segbuf_fill_in_segsum(segbuf);
1560 }
1561 return 0;
1562}
1563
1564static int
1565nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1566{
1567 struct page *clone_page;
1568 struct buffer_head *bh, *head, *bh2;
1569 void *kaddr;
1570
1571 bh = head = page_buffers(page);
1572
1573 clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1574 if (unlikely(!clone_page))
1575 return -ENOMEM;
1576
1577 bh2 = page_buffers(clone_page);
1578 kaddr = kmap_atomic(page, KM_USER0);
1579 do {
1580 if (list_empty(&bh->b_assoc_buffers))
1581 continue;
1582 get_bh(bh2);
1583 page_cache_get(clone_page); /* for each bh */
1584 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1585 bh2->b_blocknr = bh->b_blocknr;
1586 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1587 list_add_tail(&bh->b_assoc_buffers, out);
1588 } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1589 kunmap_atomic(kaddr, KM_USER0);
1590
1591 if (!TestSetPageWriteback(clone_page))
f629d1c9 1592 account_page_writeback(clone_page);
9ff05123
RK		 1593 unlock_page(clone_page);
1594
1595 return 0;
1596}
1597
1598static int nilfs_test_page_to_be_frozen(struct page *page)
1599{
1600 struct address_space *mapping = page->mapping;
1601
1602 if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1603 return 0;
1604
1605 if (page_mapped(page)) {
1606 ClearPageChecked(page);
1607 return 1;
1608 }
1609 return PageChecked(page);
1610}
1611
1612static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1613{
1614 if (!page || PageWriteback(page))
1615 /* For split b-tree node pages, this function may be called
1616 twice. We ignore the 2nd or later calls by this check. */
1617 return 0;
1618
1619 lock_page(page);
1620 clear_page_dirty_for_io(page);
1621 set_page_writeback(page);
1622 unlock_page(page);
1623
1624 if (nilfs_test_page_to_be_frozen(page)) {
1625 int err = nilfs_copy_replace_page_buffers(page, out);
1626 if (unlikely(err))
1627 return err;
1628 }
1629 return 0;
1630}
1631
1632static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1633 struct page **failed_page)
1634{
1635 struct nilfs_segment_buffer *segbuf;
1636 struct page *bd_page = NULL, *fs_page = NULL;
1637 struct list_head *list = &sci->sc_copied_buffers;
1638 int err;
1639
1640 *failed_page = NULL;
1641 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1642 struct buffer_head *bh;
1643
1644 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1645 b_assoc_buffers) {
1646 if (bh->b_page != bd_page) {
1647 if (bd_page) {
1648 lock_page(bd_page);
1649 clear_page_dirty_for_io(bd_page);
1650 set_page_writeback(bd_page);
1651 unlock_page(bd_page);
1652 }
1653 bd_page = bh->b_page;
1654 }
1655 }
1656
1657 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1658 b_assoc_buffers) {
1e2b68bf 1659 if (bh == segbuf->sb_super_root) {
9ff05123
RK		 1660 if (bh->b_page != bd_page) {
1661 lock_page(bd_page);
1662 clear_page_dirty_for_io(bd_page);
1663 set_page_writeback(bd_page);
1664 unlock_page(bd_page);
1665 bd_page = bh->b_page;
1666 }
1667 break;
1668 }
1669 if (bh->b_page != fs_page) {
1670 err = nilfs_begin_page_io(fs_page, list);
1671 if (unlikely(err)) {
1672 *failed_page = fs_page;
1673 goto out;
1674 }
1675 fs_page = bh->b_page;
1676 }
1677 }
1678 }
1679 if (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 }
1685 err = nilfs_begin_page_io(fs_page, list);
1686 if (unlikely(err))
1687 *failed_page = fs_page;
1688 out:
1689 return err;
1690}
1691
1692static int nilfs_segctor_write(struct nilfs_sc_info *sci,
9c965bac 1693 struct the_nilfs *nilfs)
9ff05123 1694{
d1c6b72a 1695 int ret;
9ff05123 1696
d1c6b72a 1697 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
a694291a
RK		 1698 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1699 return ret;
9ff05123
RK		 1700}
1701
9ff05123
RK		 1702static void __nilfs_end_page_io(struct page *page, int err)
1703{
9ff05123
RK		 1704 if (!err) {
1705 if (!nilfs_page_buffers_clean(page))
1706 __set_page_dirty_nobuffers(page);
1707 ClearPageError(page);
1708 } else {
1709 __set_page_dirty_nobuffers(page);
1710 SetPageError(page);
1711 }
1712
1713 if (buffer_nilfs_allocated(page_buffers(page))) {
1714 if (TestClearPageWriteback(page))
1715 dec_zone_page_state(page, NR_WRITEBACK);
1716 } else
1717 end_page_writeback(page);
1718}
1719
1720static void nilfs_end_page_io(struct page *page, int err)
1721{
1722 if (!page)
1723 return;
1724
a9777845 1725 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
8227b297
RK		 1726 /*
1727 * For b-tree node pages, this function may be called twice
1728 * or more because they might be split in a segment.
1729 */
a9777845
RK		 1730 if (PageDirty(page)) {
1731 /*
1732 * For pages holding split b-tree node buffers, dirty
1733 * flag on the buffers may be cleared discretely.
1734 * In that case, the page is once redirtied for
1735 * remaining buffers, and it must be cancelled if
1736 * all the buffers get cleaned later.
1737 */
1738 lock_page(page);
1739 if (nilfs_page_buffers_clean(page))
1740 __nilfs_clear_page_dirty(page);
1741 unlock_page(page);
1742 }
9ff05123 1743 return;
a9777845 1744 }
9ff05123
RK		 1745
1746 __nilfs_end_page_io(page, err);
1747}
1748
1749static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1750{
1751 struct buffer_head *bh, *head;
1752 struct page *page;
1753
1754 while (!list_empty(list)) {
1755 bh = list_entry(list->next, struct buffer_head,
1756 b_assoc_buffers);
1757 page = bh->b_page;
1758 page_cache_get(page);
1759 head = bh = page_buffers(page);
1760 do {
1761 if (!list_empty(&bh->b_assoc_buffers)) {
1762 list_del_init(&bh->b_assoc_buffers);
1763 if (!err) {
1764 set_buffer_uptodate(bh);
1765 clear_buffer_dirty(bh);
1766 clear_buffer_nilfs_volatile(bh);
1767 }
1768 brelse(bh); /* for b_assoc_buffers */
1769 }
1770 } while ((bh = bh->b_this_page) != head);
1771
1772 __nilfs_end_page_io(page, err);
1773 page_cache_release(page);
1774 }
1775}
1776
a694291a 1777static void nilfs_abort_logs(struct list_head *logs, struct page *failed_page,
1e2b68bf 1778 int err)
9ff05123
RK		 1779{
1780 struct nilfs_segment_buffer *segbuf;
1781 struct page *bd_page = NULL, *fs_page = NULL;
a694291a 1782 struct buffer_head *bh;
9ff05123 1783
a694291a
RK		 1784 if (list_empty(logs))
1785 return;
9ff05123 1786
a694291a 1787 list_for_each_entry(segbuf, logs, sb_list) {
9ff05123
RK		 1788 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1789 b_assoc_buffers) {
1790 if (bh->b_page != bd_page) {
1791 if (bd_page)
1792 end_page_writeback(bd_page);
1793 bd_page = bh->b_page;
1794 }
1795 }
1796
1797 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1798 b_assoc_buffers) {
1e2b68bf 1799 if (bh == segbuf->sb_super_root) {
9ff05123
RK		 1800 if (bh->b_page != bd_page) {
1801 end_page_writeback(bd_page);
1802 bd_page = bh->b_page;
1803 }
1804 break;
1805 }
1806 if (bh->b_page != fs_page) {
1807 nilfs_end_page_io(fs_page, err);
8227b297 1808 if (fs_page && fs_page == failed_page)
a694291a 1809 return;
9ff05123
RK		 1810 fs_page = bh->b_page;
1811 }
1812 }
1813 }
1814 if (bd_page)
1815 end_page_writeback(bd_page);
1816
1817 nilfs_end_page_io(fs_page, err);
a694291a
RK		 1818}
1819
1820static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1821 struct the_nilfs *nilfs, int err)
1822{
1823 LIST_HEAD(logs);
1824 int ret;
1825
1826 list_splice_tail_init(&sci->sc_write_logs, &logs);
1827 ret = nilfs_wait_on_logs(&logs);
1e2b68bf 1828 nilfs_abort_logs(&logs, NULL, ret ? : err);
a694291a
RK		 1829
1830 list_splice_tail_init(&sci->sc_segbufs, &logs);
1831 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1832 nilfs_free_incomplete_logs(&logs, nilfs);
9ff05123 1833 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
a694291a
RK		 1834
1835 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1836 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1837 sci->sc_freesegs,
1838 sci->sc_nfreesegs,
1839 NULL);
1840 WARN_ON(ret); /* do not happen */
1841 }
1842
1843 nilfs_destroy_logs(&logs);
9ff05123
RK		 1844}
1845
1846static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1847 struct nilfs_segment_buffer *segbuf)
1848{
1849 nilfs->ns_segnum = segbuf->sb_segnum;
1850 nilfs->ns_nextnum = segbuf->sb_nextnum;
1851 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1852 + segbuf->sb_sum.nblocks;
1853 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1854 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1855}
1856
1857static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1858{
1859 struct nilfs_segment_buffer *segbuf;
1860 struct page *bd_page = NULL, *fs_page = NULL;
e605f0a7 1861 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
1e2b68bf 1862 int update_sr = false;
9ff05123 1863
a694291a 1864 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
9ff05123
RK		 1865 struct buffer_head *bh;
1866
1867 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1868 b_assoc_buffers) {
1869 set_buffer_uptodate(bh);
1870 clear_buffer_dirty(bh);
1871 if (bh->b_page != bd_page) {
1872 if (bd_page)
1873 end_page_writeback(bd_page);
1874 bd_page = bh->b_page;
1875 }
1876 }
1877 /*
1878 * We assume that the buffers which belong to the same page
1879 * continue over the buffer list.
1880 * Under this assumption, the last BHs of pages is
1881 * identifiable by the discontinuity of bh->b_page
1882 * (page != fs_page).
1883 *
1884 * For B-tree node blocks, however, this assumption is not
1885 * guaranteed. The cleanup code of B-tree node pages needs
1886 * special care.
1887 */
1888 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1889 b_assoc_buffers) {
1890 set_buffer_uptodate(bh);
1891 clear_buffer_dirty(bh);
1892 clear_buffer_nilfs_volatile(bh);
b1f6a4f2 1893 clear_buffer_nilfs_redirected(bh);
1e2b68bf 1894 if (bh == segbuf->sb_super_root) {
9ff05123
RK		 1895 if (bh->b_page != bd_page) {
1896 end_page_writeback(bd_page);
1897 bd_page = bh->b_page;
1898 }
1e2b68bf 1899 update_sr = true;
9ff05123
RK		 1900 break;
1901 }
1902 if (bh->b_page != fs_page) {
1903 nilfs_end_page_io(fs_page, 0);
1904 fs_page = bh->b_page;
1905 }
1906 }
1907
4762077c
RK		 1908 if (!nilfs_segbuf_simplex(segbuf)) {
1909 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
9ff05123
RK		 1910 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1911 sci->sc_lseg_stime = jiffies;
1912 }
4762077c 1913 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
9ff05123
RK		 1914 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1915 }
1916 }
1917 /*
1918 * Since pages may continue over multiple segment buffers,
1919 * end of the last page must be checked outside of the loop.
1920 */
1921 if (bd_page)
1922 end_page_writeback(bd_page);
1923
1924 nilfs_end_page_io(fs_page, 0);
1925
1926 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1927
1928 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1929
c1c1d709 1930 if (nilfs_doing_gc())
9ff05123 1931 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
c1c1d709 1932 else
9ff05123 1933 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
9ff05123
RK		 1934
1935 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1936
a694291a 1937 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
9ff05123
RK		 1938 nilfs_set_next_segment(nilfs, segbuf);
1939
1940 if (update_sr) {
1941 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
e339ad31 1942 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
9ff05123 1943
c96fa464 1944 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
9ff05123
RK		 1945 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1946 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
a694291a 1947 nilfs_segctor_clear_metadata_dirty(sci);
9ff05123
RK		 1948 } else
1949 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1950}
1951
a694291a
RK		 1952static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1953{
1954 int ret;
1955
1956 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1957 if (!ret) {
1958 nilfs_segctor_complete_write(sci);
1959 nilfs_destroy_logs(&sci->sc_write_logs);
1960 }
1961 return ret;
1962}
1963
9ff05123
RK		 1964static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
1965 struct nilfs_sb_info *sbi)
1966{
1967 struct nilfs_inode_info *ii, *n;
e912a5b6 1968 struct inode *ifile = sci->sc_root->ifile;
9ff05123
RK		 1969
1970 spin_lock(&sbi->s_inode_lock);
1971 retry:
1972 list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
1973 if (!ii->i_bh) {
1974 struct buffer_head *ibh;
1975 int err;
1976
1977 spin_unlock(&sbi->s_inode_lock);
1978 err = nilfs_ifile_get_inode_block(
e912a5b6 1979 ifile, ii->vfs_inode.i_ino, &ibh);
9ff05123
RK		 1980 if (unlikely(err)) {
1981 nilfs_warning(sbi->s_super, __func__,
1982 "failed to get inode block.\n");
1983 return err;
1984 }
1985 nilfs_mdt_mark_buffer_dirty(ibh);
e912a5b6 1986 nilfs_mdt_mark_dirty(ifile);
9ff05123
RK		 1987 spin_lock(&sbi->s_inode_lock);
1988 if (likely(!ii->i_bh))
1989 ii->i_bh = ibh;
1990 else
1991 brelse(ibh);
1992 goto retry;
1993 }
9ff05123
RK		 1994
1995 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1996 set_bit(NILFS_I_BUSY, &ii->i_state);
1997 list_del(&ii->i_dirty);
1998 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
1999 }
2000 spin_unlock(&sbi->s_inode_lock);
2001
9ff05123
RK		 2002 return 0;
2003}
2004
2005static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2006 struct nilfs_sb_info *sbi)
2007{
2008 struct nilfs_transaction_info *ti = current->journal_info;
2009 struct nilfs_inode_info *ii, *n;
9ff05123
RK		 2010
2011 spin_lock(&sbi->s_inode_lock);
2012 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2013 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
6c43f410 2014 test_bit(NILFS_I_DIRTY, &ii->i_state))
9ff05123 2015 continue;
6c43f410 2016
9ff05123
RK		 2017 clear_bit(NILFS_I_BUSY, &ii->i_state);
2018 brelse(ii->i_bh);
2019 ii->i_bh = NULL;
2020 list_del(&ii->i_dirty);
2021 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2022 }
2023 spin_unlock(&sbi->s_inode_lock);
2024}
2025
9ff05123
RK		 2026/*
2027 * Main procedure of segment constructor
2028 */
2029static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2030{
2031 struct nilfs_sb_info *sbi = sci->sc_sbi;
2032 struct the_nilfs *nilfs = sbi->s_nilfs;
2033 struct page *failed_page;
1e2b68bf 2034 int err;
9ff05123
RK		 2035
2036 sci->sc_stage.scnt = NILFS_ST_INIT;
6c43f410 2037 sci->sc_cno = nilfs->ns_cno;
9ff05123
RK		 2038
2039 err = nilfs_segctor_check_in_files(sci, sbi);
2040 if (unlikely(err))
2041 goto out;
2042
e912a5b6 2043 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
9ff05123
RK		 2044 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2045
2046 if (nilfs_segctor_clean(sci))
2047 goto out;
2048
2049 do {
2050 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2051
2052 err = nilfs_segctor_begin_construction(sci, nilfs);
2053 if (unlikely(err))
2054 goto out;
2055
2056 /* Update time stamp */
2057 sci->sc_seg_ctime = get_seconds();
2058
2059 err = nilfs_segctor_collect(sci, nilfs, mode);
2060 if (unlikely(err))
2061 goto failed;
2062
9ff05123
RK		 2063 /* Avoid empty segment */
2064 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
4762077c 2065 nilfs_segbuf_empty(sci->sc_curseg)) {
a694291a 2066 nilfs_segctor_abort_construction(sci, nilfs, 1);
9ff05123
RK		 2067 goto out;
2068 }
2069
2070 err = nilfs_segctor_assign(sci, mode);
2071 if (unlikely(err))
2072 goto failed;
2073
9ff05123 2074 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
e912a5b6 2075 nilfs_segctor_fill_in_file_bmap(sci);
9ff05123 2076
1e2b68bf
RK		 2077 if (mode == SC_LSEG_SR &&
2078 sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
9ff05123
RK		 2079 err = nilfs_segctor_fill_in_checkpoint(sci);
2080 if (unlikely(err))
a694291a 2081 goto failed_to_write;
9ff05123
RK		 2082
2083 nilfs_segctor_fill_in_super_root(sci, nilfs);
2084 }
2085 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2086
2087 /* Write partial segments */
2088 err = nilfs_segctor_prepare_write(sci, &failed_page);
a694291a 2089 if (err) {
1e2b68bf 2090 nilfs_abort_logs(&sci->sc_segbufs, failed_page, err);
9ff05123 2091 goto failed_to_write;
a694291a 2092 }
aaed1d5b
RK		 2093
2094 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2095 nilfs->ns_crc_seed);
9ff05123 2096
9c965bac 2097 err = nilfs_segctor_write(sci, nilfs);
9ff05123
RK		 2098 if (unlikely(err))
2099 goto failed_to_write;
2100
a694291a
RK		 2101 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2102 nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2103 /*
2104 * At this point, we avoid double buffering
2105 * for blocksize < pagesize because page dirty
2106 * flag is turned off during write and dirty
2107 * buffers are not properly collected for
2108 * pages crossing over segments.
2109 */
2110 err = nilfs_segctor_wait(sci);
2111 if (err)
2112 goto failed_to_write;
2113 }
9ff05123
RK		 2114 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2115
9ff05123 2116 out:
9ff05123
RK		 2117 nilfs_segctor_check_out_files(sci, sbi);
2118 return err;
2119
2120 failed_to_write:
9ff05123
RK		 2121 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2122 nilfs_redirty_inodes(&sci->sc_dirty_files);
9ff05123
RK		 2123
2124 failed:
2125 if (nilfs_doing_gc())
2126 nilfs_redirty_inodes(&sci->sc_gc_inodes);
a694291a 2127 nilfs_segctor_abort_construction(sci, nilfs, err);
9ff05123
RK		 2128 goto out;
2129}
2130
2131/**
9ccf56c1 2132 * nilfs_segctor_start_timer - set timer of background write
9ff05123
RK		 2133 * @sci: nilfs_sc_info
2134 *
2135 * If the timer has already been set, it ignores the new request.
2136 * This function MUST be called within a section locking the segment
2137 * semaphore.
2138 */
2139static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2140{
2141 spin_lock(&sci->sc_state_lock);
fdce895e
LH		 2142 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2143 sci->sc_timer.expires = jiffies + sci->sc_interval;
2144 add_timer(&sci->sc_timer);
9ff05123
RK		 2145 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2146 }
2147 spin_unlock(&sci->sc_state_lock);
2148}
2149
2150static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2151{
2152 spin_lock(&sci->sc_state_lock);
2153 if (!(sci->sc_flush_request & (1 << bn))) {
2154 unsigned long prev_req = sci->sc_flush_request;
2155
2156 sci->sc_flush_request |= (1 << bn);
2157 if (!prev_req)
2158 wake_up(&sci->sc_wait_daemon);
2159 }
2160 spin_unlock(&sci->sc_state_lock);
2161}
2162
2163/**
2164 * nilfs_flush_segment - trigger a segment construction for resource control
2165 * @sb: super block
2166 * @ino: inode number of the file to be flushed out.
2167 */
2168void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2169{
2170 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2171 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2172
2173 if (!sci || nilfs_doing_construction())
2174 return;
2175 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2176 /* assign bit 0 to data files */
2177}
2178
9ff05123
RK		 2179struct nilfs_segctor_wait_request {
2180 wait_queue_t wq;
2181 __u32 seq;
2182 int err;
2183 atomic_t done;
2184};
2185
2186static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2187{
2188 struct nilfs_segctor_wait_request wait_req;
2189 int err = 0;
2190
2191 spin_lock(&sci->sc_state_lock);
2192 init_wait(&wait_req.wq);
2193 wait_req.err = 0;
2194 atomic_set(&wait_req.done, 0);
2195 wait_req.seq = ++sci->sc_seq_request;
2196 spin_unlock(&sci->sc_state_lock);
2197
2198 init_waitqueue_entry(&wait_req.wq, current);
2199 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2200 set_current_state(TASK_INTERRUPTIBLE);
2201 wake_up(&sci->sc_wait_daemon);
2202
2203 for (;;) {
2204 if (atomic_read(&wait_req.done)) {
2205 err = wait_req.err;
2206 break;
2207 }
2208 if (!signal_pending(current)) {
2209 schedule();
2210 continue;
2211 }
2212 err = -ERESTARTSYS;
2213 break;
2214 }
2215 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2216 return err;
2217}
2218
2219static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2220{
2221 struct nilfs_segctor_wait_request *wrq, *n;
2222 unsigned long flags;
2223
2224 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2225 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2226 wq.task_list) {
2227 if (!atomic_read(&wrq->done) &&
2228 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2229 wrq->err = err;
2230 atomic_set(&wrq->done, 1);
2231 }
2232 if (atomic_read(&wrq->done)) {
2233 wrq->wq.func(&wrq->wq,
2234 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2235 0, NULL);
2236 }
2237 }
2238 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2239}
2240
2241/**
2242 * nilfs_construct_segment - construct a logical segment
2243 * @sb: super block
2244 *
2245 * Return Value: On success, 0 is retured. On errors, one of the following
2246 * negative error code is returned.
2247 *
2248 * %-EROFS - Read only filesystem.
2249 *
2250 * %-EIO - I/O error
2251 *
2252 * %-ENOSPC - No space left on device (only in a panic state).
2253 *
2254 * %-ERESTARTSYS - Interrupted.
2255 *
2256 * %-ENOMEM - Insufficient memory available.
2257 */
2258int nilfs_construct_segment(struct super_block *sb)
2259{
2260 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2261 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2262 struct nilfs_transaction_info *ti;
2263 int err;
2264
2265 if (!sci)
2266 return -EROFS;
2267
2268 /* A call inside transactions causes a deadlock. */
2269 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2270
2271 err = nilfs_segctor_sync(sci);
2272 return err;
2273}
2274
2275/**
2276 * nilfs_construct_dsync_segment - construct a data-only logical segment
2277 * @sb: super block
f30bf3e4
RK		 2278 * @inode: inode whose data blocks should be written out
2279 * @start: start byte offset
2280 * @end: end byte offset (inclusive)
9ff05123
RK		 2281 *
2282 * Return Value: On success, 0 is retured. On errors, one of the following
2283 * negative error code is returned.
2284 *
2285 * %-EROFS - Read only filesystem.
2286 *
2287 * %-EIO - I/O error
2288 *
2289 * %-ENOSPC - No space left on device (only in a panic state).
2290 *
2291 * %-ERESTARTSYS - Interrupted.
2292 *
2293 * %-ENOMEM - Insufficient memory available.
2294 */
f30bf3e4
RK		 2295int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2296 loff_t start, loff_t end)
9ff05123
RK		 2297{
2298 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2299 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2300 struct nilfs_inode_info *ii;
2301 struct nilfs_transaction_info ti;
2302 int err = 0;
2303
2304 if (!sci)
2305 return -EROFS;
2306
2307 nilfs_transaction_lock(sbi, &ti, 0);
2308
2309 ii = NILFS_I(inode);
2310 if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2311 nilfs_test_opt(sbi, STRICT_ORDER) ||
2312 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2313 nilfs_discontinued(sbi->s_nilfs)) {
2314 nilfs_transaction_unlock(sbi);
2315 err = nilfs_segctor_sync(sci);
2316 return err;
2317 }
2318
2319 spin_lock(&sbi->s_inode_lock);
2320 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2321 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2322 spin_unlock(&sbi->s_inode_lock);
2323 nilfs_transaction_unlock(sbi);
2324 return 0;
2325 }
2326 spin_unlock(&sbi->s_inode_lock);
f30bf3e4
RK		 2327 sci->sc_dsync_inode = ii;
2328 sci->sc_dsync_start = start;
2329 sci->sc_dsync_end = end;
9ff05123
RK		 2330
2331 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2332
2333 nilfs_transaction_unlock(sbi);
2334 return err;
2335}
2336
9ff05123
RK		 2337#define FLUSH_FILE_BIT (0x1) /* data file only */
2338#define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2339
dcd76186
RK		 2340/**
2341 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2342 * @sci: segment constructor object
2343 */
2344static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
9ff05123 2345{
9ff05123 2346 spin_lock(&sci->sc_state_lock);
dcd76186 2347 sci->sc_seq_accepted = sci->sc_seq_request;
9ff05123 2348 spin_unlock(&sci->sc_state_lock);
fdce895e 2349 del_timer_sync(&sci->sc_timer);
9ff05123
RK		 2350}
2351
dcd76186
RK		 2352/**
2353 * nilfs_segctor_notify - notify the result of request to caller threads
2354 * @sci: segment constructor object
2355 * @mode: mode of log forming
2356 * @err: error code to be notified
2357 */
2358static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
9ff05123
RK		 2359{
2360 /* Clear requests (even when the construction failed) */
2361 spin_lock(&sci->sc_state_lock);
2362
dcd76186 2363 if (mode == SC_LSEG_SR) {
aeda7f63 2364 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
dcd76186
RK		 2365 sci->sc_seq_done = sci->sc_seq_accepted;
2366 nilfs_segctor_wakeup(sci, err);
9ff05123 2367 sci->sc_flush_request = 0;
aeda7f63 2368 } else {
dcd76186 2369 if (mode == SC_FLUSH_FILE)
aeda7f63 2370 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
dcd76186 2371 else if (mode == SC_FLUSH_DAT)
aeda7f63
RK		 2372 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2373
2374 /* re-enable timer if checkpoint creation was not done */
fdce895e
LH		 2375 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2376 time_before(jiffies, sci->sc_timer.expires))
2377 add_timer(&sci->sc_timer);
aeda7f63 2378 }
9ff05123
RK		 2379 spin_unlock(&sci->sc_state_lock);
2380}
2381
dcd76186
RK		 2382/**
2383 * nilfs_segctor_construct - form logs and write them to disk
2384 * @sci: segment constructor object
2385 * @mode: mode of log forming
2386 */
2387static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
9ff05123
RK		 2388{
2389 struct nilfs_sb_info *sbi = sci->sc_sbi;
2390 struct the_nilfs *nilfs = sbi->s_nilfs;
d26493b6 2391 struct nilfs_super_block **sbp;
9ff05123
RK		 2392 int err = 0;
2393
dcd76186
RK		 2394 nilfs_segctor_accept(sci);
2395
9ff05123 2396 if (nilfs_discontinued(nilfs))
dcd76186
RK		 2397 mode = SC_LSEG_SR;
2398 if (!nilfs_segctor_confirm(sci))
2399 err = nilfs_segctor_do_construct(sci, mode);
2400
9ff05123 2401 if (likely(!err)) {
dcd76186 2402 if (mode != SC_FLUSH_DAT)
9ff05123
RK		 2403 atomic_set(&nilfs->ns_ndirtyblks, 0);
2404 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2405 nilfs_discontinued(nilfs)) {
2406 down_write(&nilfs->ns_sem);
d26493b6 2407 err = -EIO;
b2ac86e1
JS		 2408 sbp = nilfs_prepare_super(sbi,
2409 nilfs_sb_will_flip(nilfs));
2410 if (likely(sbp)) {
2411 nilfs_set_log_cursor(sbp[0], nilfs);
2412 err = nilfs_commit_super(sbi, NILFS_SB_COMMIT);
2413 }
9ff05123
RK		 2414 up_write(&nilfs->ns_sem);
2415 }
2416 }
dcd76186
RK		 2417
2418 nilfs_segctor_notify(sci, mode, err);
9ff05123
RK		 2419 return err;
2420}
2421
2422static void nilfs_construction_timeout(unsigned long data)
2423{
2424 struct task_struct *p = (struct task_struct *)data;
2425 wake_up_process(p);
2426}
2427
2428static void
2429nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2430{
2431 struct nilfs_inode_info *ii, *n;
2432
2433 list_for_each_entry_safe(ii, n, head, i_dirty) {
2434 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2435 continue;
9ff05123 2436 list_del_init(&ii->i_dirty);
263d90ce 2437 iput(&ii->vfs_inode);
9ff05123
RK		 2438 }
2439}
2440
4f6b8288
RK		 2441int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2442 void **kbufs)
9ff05123
RK		 2443{
2444 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2445 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2446 struct the_nilfs *nilfs = sbi->s_nilfs;
2447 struct nilfs_transaction_info ti;
9ff05123
RK		 2448 int err;
2449
2450 if (unlikely(!sci))
2451 return -EROFS;
2452
2453 nilfs_transaction_lock(sbi, &ti, 1);
2454
c1c1d709 2455 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
9ff05123
RK		 2456 if (unlikely(err))
2457 goto out_unlock;
071cb4b8 2458
4f6b8288 2459 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
c1c1d709
RK		 2460 if (unlikely(err)) {
2461 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
9ff05123 2462 goto out_unlock;
c1c1d709 2463 }
9ff05123 2464
071cb4b8
RK		 2465 sci->sc_freesegs = kbufs[4];
2466 sci->sc_nfreesegs = argv[4].v_nmembs;
0935db74 2467 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
9ff05123
RK		 2468
2469 for (;;) {
dcd76186 2470 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
9ff05123 2471 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
9ff05123
RK		 2472
2473 if (likely(!err))
2474 break;
2475
2476 nilfs_warning(sb, __func__,
2477 "segment construction failed. (err=%d)", err);
2478 set_current_state(TASK_INTERRUPTIBLE);
2479 schedule_timeout(sci->sc_interval);
2480 }
e902ec99
JS		 2481 if (nilfs_test_opt(sbi, DISCARD)) {
2482 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2483 sci->sc_nfreesegs);
2484 if (ret) {
2485 printk(KERN_WARNING
2486 "NILFS warning: error %d on discard request, "
2487 "turning discards off for the device\n", ret);
2488 nilfs_clear_opt(sbi, DISCARD);
2489 }
2490 }
9ff05123
RK		 2491
2492 out_unlock:
071cb4b8
RK		 2493 sci->sc_freesegs = NULL;
2494 sci->sc_nfreesegs = 0;
c1c1d709 2495 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
9ff05123
RK		 2496 nilfs_transaction_unlock(sbi);
2497 return err;
2498}
2499
2500static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2501{
2502 struct nilfs_sb_info *sbi = sci->sc_sbi;
2503 struct nilfs_transaction_info ti;
9ff05123
RK		 2504
2505 nilfs_transaction_lock(sbi, &ti, 0);
dcd76186 2506 nilfs_segctor_construct(sci, mode);
9ff05123
RK		 2507
2508 /*
2509 * Unclosed segment should be retried. We do this using sc_timer.
2510 * Timeout of sc_timer will invoke complete construction which leads
2511 * to close the current logical segment.
2512 */
2513 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2514 nilfs_segctor_start_timer(sci);
2515
2516 nilfs_transaction_unlock(sbi);
2517}
2518
2519static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2520{
2521 int mode = 0;
2522 int err;
2523
2524 spin_lock(&sci->sc_state_lock);
2525 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2526 SC_FLUSH_DAT : SC_FLUSH_FILE;
2527 spin_unlock(&sci->sc_state_lock);
2528
2529 if (mode) {
2530 err = nilfs_segctor_do_construct(sci, mode);
2531
2532 spin_lock(&sci->sc_state_lock);
2533 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2534 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2535 spin_unlock(&sci->sc_state_lock);
2536 }
2537 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2538}
2539
2540static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2541{
2542 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2543 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2544 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2545 return SC_FLUSH_FILE;
2546 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2547 return SC_FLUSH_DAT;
2548 }
2549 return SC_LSEG_SR;
2550}
2551
2552/**
2553 * nilfs_segctor_thread - main loop of the segment constructor thread.
2554 * @arg: pointer to a struct nilfs_sc_info.
2555 *
2556 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2557 * to execute segment constructions.
2558 */
2559static int nilfs_segctor_thread(void *arg)
2560{
2561 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
e605f0a7 2562 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
9ff05123
RK		 2563 int timeout = 0;
2564
fdce895e
LH		 2565 sci->sc_timer.data = (unsigned long)current;
2566 sci->sc_timer.function = nilfs_construction_timeout;
9ff05123
RK		 2567
2568 /* start sync. */
2569 sci->sc_task = current;
2570 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2571 printk(KERN_INFO
2572 "segctord starting. Construction interval = %lu seconds, "
2573 "CP frequency < %lu seconds\n",
2574 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2575
2576 spin_lock(&sci->sc_state_lock);
2577 loop:
2578 for (;;) {
2579 int mode;
2580
2581 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2582 goto end_thread;
2583
2584 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2585 mode = SC_LSEG_SR;
2586 else if (!sci->sc_flush_request)
2587 break;
2588 else
2589 mode = nilfs_segctor_flush_mode(sci);
2590
2591 spin_unlock(&sci->sc_state_lock);
2592 nilfs_segctor_thread_construct(sci, mode);
2593 spin_lock(&sci->sc_state_lock);
2594 timeout = 0;
2595 }
2596
2597
2598 if (freezing(current)) {
2599 spin_unlock(&sci->sc_state_lock);
2600 refrigerator();
2601 spin_lock(&sci->sc_state_lock);
2602 } else {
2603 DEFINE_WAIT(wait);
2604 int should_sleep = 1;
2605
2606 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2607 TASK_INTERRUPTIBLE);
2608
2609 if (sci->sc_seq_request != sci->sc_seq_done)
2610 should_sleep = 0;
2611 else if (sci->sc_flush_request)
2612 should_sleep = 0;
2613 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2614 should_sleep = time_before(jiffies,
fdce895e 2615 sci->sc_timer.expires);
9ff05123
RK		 2616
2617 if (should_sleep) {
2618 spin_unlock(&sci->sc_state_lock);
2619 schedule();
2620 spin_lock(&sci->sc_state_lock);
2621 }
2622 finish_wait(&sci->sc_wait_daemon, &wait);
2623 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
fdce895e 2624 time_after_eq(jiffies, sci->sc_timer.expires));
e605f0a7
RK		 2625
2626 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
1dfa2710 2627 set_nilfs_discontinued(nilfs);
9ff05123
RK		 2628 }
2629 goto loop;
2630
2631 end_thread:
2632 spin_unlock(&sci->sc_state_lock);
9ff05123
RK		 2633
2634 /* end sync. */
2635 sci->sc_task = NULL;
2636 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2637 return 0;
2638}
2639
2640static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2641{
2642 struct task_struct *t;
2643
2644 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2645 if (IS_ERR(t)) {
2646 int err = PTR_ERR(t);
2647
2648 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2649 err);
2650 return err;
2651 }
2652 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2653 return 0;
2654}
2655
2656static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
6b81e14e
JS		 2657 __acquires(&sci->sc_state_lock)
2658 __releases(&sci->sc_state_lock)
9ff05123
RK		 2659{
2660 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2661
2662 while (sci->sc_task) {
2663 wake_up(&sci->sc_wait_daemon);
2664 spin_unlock(&sci->sc_state_lock);
2665 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2666 spin_lock(&sci->sc_state_lock);
2667 }
2668}
2669
9ff05123
RK		 2670/*
2671 * Setup & clean-up functions
2672 */
e912a5b6
RK		 2673static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi,
2674 struct nilfs_root *root)
9ff05123
RK		 2675{
2676 struct nilfs_sc_info *sci;
2677
2678 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2679 if (!sci)
2680 return NULL;
2681
2682 sci->sc_sbi = sbi;
2683 sci->sc_super = sbi->s_super;
2684
e912a5b6
RK		 2685 nilfs_get_root(root);
2686 sci->sc_root = root;
2687
9ff05123
RK		 2688 init_waitqueue_head(&sci->sc_wait_request);
2689 init_waitqueue_head(&sci->sc_wait_daemon);
2690 init_waitqueue_head(&sci->sc_wait_task);
2691 spin_lock_init(&sci->sc_state_lock);
2692 INIT_LIST_HEAD(&sci->sc_dirty_files);
2693 INIT_LIST_HEAD(&sci->sc_segbufs);
a694291a 2694 INIT_LIST_HEAD(&sci->sc_write_logs);
9ff05123 2695 INIT_LIST_HEAD(&sci->sc_gc_inodes);
9ff05123 2696 INIT_LIST_HEAD(&sci->sc_copied_buffers);
fdce895e 2697 init_timer(&sci->sc_timer);
9ff05123
RK		 2698
2699 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2700 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2701 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2702
2703 if (sbi->s_interval)
2704 sci->sc_interval = sbi->s_interval;
2705 if (sbi->s_watermark)
2706 sci->sc_watermark = sbi->s_watermark;
2707 return sci;
2708}
2709
2710static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2711{
2712 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2713
2714 /* The segctord thread was stopped and its timer was removed.
2715 But some tasks remain. */
2716 do {
2717 struct nilfs_sb_info *sbi = sci->sc_sbi;
2718 struct nilfs_transaction_info ti;
9ff05123
RK		 2719
2720 nilfs_transaction_lock(sbi, &ti, 0);
dcd76186 2721 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
9ff05123
RK		 2722 nilfs_transaction_unlock(sbi);
2723
2724 } while (ret && retrycount-- > 0);
2725}
2726
2727/**
2728 * nilfs_segctor_destroy - destroy the segment constructor.
2729 * @sci: nilfs_sc_info
2730 *
2731 * nilfs_segctor_destroy() kills the segctord thread and frees
2732 * the nilfs_sc_info struct.
2733 * Caller must hold the segment semaphore.
2734 */
2735static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2736{
2737 struct nilfs_sb_info *sbi = sci->sc_sbi;
2738 int flag;
2739
2740 up_write(&sbi->s_nilfs->ns_segctor_sem);
2741
2742 spin_lock(&sci->sc_state_lock);
2743 nilfs_segctor_kill_thread(sci);
2744 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2745 || sci->sc_seq_request != sci->sc_seq_done);
2746 spin_unlock(&sci->sc_state_lock);
2747
3256a055 2748 if (flag || !nilfs_segctor_confirm(sci))
9ff05123
RK		 2749 nilfs_segctor_write_out(sci);
2750
1f5abe7e 2751 WARN_ON(!list_empty(&sci->sc_copied_buffers));
9ff05123
RK		 2752
2753 if (!list_empty(&sci->sc_dirty_files)) {
2754 nilfs_warning(sbi->s_super, __func__,
2755 "dirty file(s) after the final construction\n");
2756 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2757 }
9ff05123 2758
1f5abe7e 2759 WARN_ON(!list_empty(&sci->sc_segbufs));
a694291a 2760 WARN_ON(!list_empty(&sci->sc_write_logs));
9ff05123 2761
e912a5b6
RK		 2762 nilfs_put_root(sci->sc_root);
2763
9ff05123
RK		 2764 down_write(&sbi->s_nilfs->ns_segctor_sem);
2765
fdce895e 2766 del_timer_sync(&sci->sc_timer);
9ff05123
RK		 2767 kfree(sci);
2768}
2769
2770/**
2771 * nilfs_attach_segment_constructor - attach a segment constructor
2772 * @sbi: nilfs_sb_info
e912a5b6 2773 * @root: root object of the current filesystem tree
9ff05123
RK		 2774 *
2775 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
7a65004b 2776 * initializes it, and starts the segment constructor.
9ff05123
RK		 2777 *
2778 * Return Value: On success, 0 is returned. On error, one of the following
2779 * negative error code is returned.
2780 *
2781 * %-ENOMEM - Insufficient memory available.
2782 */
e912a5b6
RK		 2783int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
2784 struct nilfs_root *root)
9ff05123 2785{
9ff05123
RK		 2786 int err;
2787
fe5f171b
RK		 2788 if (NILFS_SC(sbi)) {
2789 /*
2790 * This happens if the filesystem was remounted
2791 * read/write after nilfs_error degenerated it into a
2792 * read-only mount.
2793 */
2794 nilfs_detach_segment_constructor(sbi);
2795 }
2796
e912a5b6 2797 sbi->s_sc_info = nilfs_segctor_new(sbi, root);
9ff05123
RK		 2798 if (!sbi->s_sc_info)
2799 return -ENOMEM;
2800
154ac5a8 2801 err = nilfs_segctor_start_thread(NILFS_SC(sbi));
9ff05123 2802 if (err) {
9ff05123
RK		 2803 kfree(sbi->s_sc_info);
2804 sbi->s_sc_info = NULL;
2805 }
2806 return err;
2807}
2808
2809/**
2810 * nilfs_detach_segment_constructor - destroy the segment constructor
2811 * @sbi: nilfs_sb_info
2812 *
2813 * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2814 * frees the struct nilfs_sc_info, and destroy the dirty file list.
2815 */
2816void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2817{
2818 struct the_nilfs *nilfs = sbi->s_nilfs;
2819 LIST_HEAD(garbage_list);
2820
2821 down_write(&nilfs->ns_segctor_sem);
2822 if (NILFS_SC(sbi)) {
2823 nilfs_segctor_destroy(NILFS_SC(sbi));
2824 sbi->s_sc_info = NULL;
2825 }
2826
2827 /* Force to free the list of dirty files */
2828 spin_lock(&sbi->s_inode_lock);
2829 if (!list_empty(&sbi->s_dirty_files)) {
2830 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2831 nilfs_warning(sbi->s_super, __func__,
2832 "Non empty dirty list after the last "
2833 "segment construction\n");
2834 }
2835 spin_unlock(&sbi->s_inode_lock);
2836 up_write(&nilfs->ns_segctor_sem);
2837
2838 nilfs_dispose_list(sbi, &garbage_list, 1);
9ff05123 2839}
Linux 6.x block layer and io_uring tree(s)