projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
MAINTAINERS: remove SUPERH website
[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 *);
693dd321 107static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
9ff05123
RK		 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
9ff05123
RK		 118static 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,
47420c79
RK		 129 * it is saved and will be restored on
130 * nilfs_transaction_commit().
9ff05123
RK		 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
47420c79 161 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
9ff05123
RK		 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
7a65004b 168 * instead; otherwise a new struct is assigned from a slab.
9ff05123
RK		 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 *
9ff05123
RK		 178 * %-ENOSPC - No space left on device
179 */
180int nilfs_transaction_begin(struct super_block *sb,
181 struct nilfs_transaction_info *ti,
182 int vacancy_check)
183{
9ff05123
RK		 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
2c22b337 192 sb_start_intwrite(sb);
5beb6e0b 193
e3154e97 194 nilfs = sb->s_fs_info;
9ff05123
RK		 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);
2c22b337 208 sb_end_intwrite(sb);
9ff05123
RK		 209 return ret;
210}
211
212/**
47420c79 213 * nilfs_transaction_commit - commit indivisible file operations.
9ff05123 214 * @sb: super block
9ff05123 215 *
47420c79
RK		 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.
9ff05123 222 */
47420c79 223int nilfs_transaction_commit(struct super_block *sb)
9ff05123
RK		 224{
225 struct nilfs_transaction_info *ti = current->journal_info;
e3154e97 226 struct the_nilfs *nilfs = sb->s_fs_info;
9ff05123
RK		 227 int err = 0;
228
229 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
47420c79 230 ti->ti_flags |= NILFS_TI_COMMIT;
9ff05123
RK		 231 if (ti->ti_count > 0) {
232 ti->ti_count--;
233 return 0;
234 }
3fd3fe5a
RK		 235 if (nilfs->ns_writer) {
236 struct nilfs_sc_info *sci = nilfs->ns_writer;
237
9ff05123
RK		 238 if (ti->ti_flags & NILFS_TI_COMMIT)
239 nilfs_segctor_start_timer(sci);
3fd3fe5a 240 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
9ff05123
RK		 241 nilfs_segctor_do_flush(sci, 0);
242 }
3fd3fe5a 243 up_read(&nilfs->ns_segctor_sem);
9ff05123
RK		 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);
2c22b337 250 sb_end_intwrite(sb);
9ff05123
RK		 251 return err;
252}
253
47420c79
RK		 254void nilfs_transaction_abort(struct super_block *sb)
255{
256 struct nilfs_transaction_info *ti = current->journal_info;
e3154e97 257 struct the_nilfs *nilfs = sb->s_fs_info;
47420c79
RK		 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 }
e3154e97 264 up_read(&nilfs->ns_segctor_sem);
47420c79
RK		 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);
2c22b337 269 sb_end_intwrite(sb);
47420c79
RK		 270}
271
9ff05123
RK		 272void nilfs_relax_pressure_in_lock(struct super_block *sb)
273{
e3154e97 274 struct the_nilfs *nilfs = sb->s_fs_info;
3fd3fe5a 275 struct nilfs_sc_info *sci = nilfs->ns_writer;
9ff05123
RK		 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
f7545144 295static void nilfs_transaction_lock(struct super_block *sb,
9ff05123
RK		 296 struct nilfs_transaction_info *ti,
297 int gcflag)
298{
299 struct nilfs_transaction_info *cur_ti = current->journal_info;
e3154e97 300 struct the_nilfs *nilfs = sb->s_fs_info;
3fd3fe5a 301 struct nilfs_sc_info *sci = nilfs->ns_writer;
9ff05123 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 (;;) {
3fd3fe5a
RK		 312 down_write(&nilfs->ns_segctor_sem);
313 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
9ff05123
RK		 314 break;
315
3fd3fe5a 316 nilfs_segctor_do_immediate_flush(sci);
9ff05123 317
f7545144 318 up_write(&nilfs->ns_segctor_sem);
9ff05123
RK		 319 yield();
320 }
321 if (gcflag)
322 ti->ti_flags |= NILFS_TI_GC;
323}
324
f7545144 325static void nilfs_transaction_unlock(struct super_block *sb)
9ff05123
RK		 326{
327 struct nilfs_transaction_info *ti = current->journal_info;
e3154e97 328 struct the_nilfs *nilfs = sb->s_fs_info;
9ff05123
RK		 329
330 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
331 BUG_ON(ti->ti_count > 0);
332
693dd321 333 up_write(&nilfs->ns_segctor_sem);
9ff05123
RK		 334 current->journal_info = ti->ti_save;
335 if (!list_empty(&ti->ti_garbage))
693dd321 336 nilfs_dispose_list(nilfs, &ti->ti_garbage, 0);
9ff05123
RK		 337}
338
339static 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 */
362static 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;
6c43f410 372 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
9ff05123
RK		 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
384static 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
394static 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 }
1e2b68bf 405 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
9ff05123
RK		 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 */
414static 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
426static 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));
c96fa464 433
72746ac6
RK		 434 if (NILFS_I(inode)->i_root &&
435 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
c96fa464 436 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
9ff05123
RK		 437 /* skip finfo */
438}
439
440static 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;
6c43f410 446 __u64 cno;
9ff05123
RK		 447
448 if (sci->sc_blk_cnt == 0)
449 return;
450
451 ii = NILFS_I(inode);
6c43f410
RK		 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
9ff05123
RK		 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);
6c43f410 465 finfo->fi_cno = cpu_to_le64(cno);
9ff05123
RK		 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
474static 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
9ff05123
RK		 509/*
510 * Callback functions that enumerate, mark, and collect dirty blocks
511 */
512static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
513 struct buffer_head *bh, struct inode *inode)
514{
515 int err;
516
9ff05123 517 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
e828949e
RK		 518 if (err < 0)
519 return err;
9ff05123
RK		 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
528static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
529 struct buffer_head *bh,
530 struct inode *inode)
531{
e828949e 532 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
9ff05123
RK		 533}
534
535static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
536 struct buffer_head *bh,
537 struct inode *inode)
538{
1f5abe7e 539 WARN_ON(!buffer_dirty(bh));
9ff05123
RK		 540 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
541}
542
543static 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
552static 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
4e819509 561static struct nilfs_sc_operations nilfs_sc_file_ops = {
9ff05123
RK		 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
569static 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);
e828949e
RK		 575 if (err < 0)
576 return err;
9ff05123
RK		 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
584static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
585 struct buffer_head *bh, struct inode *inode)
586{
1f5abe7e 587 WARN_ON(!buffer_dirty(bh));
9ff05123
RK		 588 return nilfs_segctor_add_file_block(sci, bh, inode,
589 sizeof(struct nilfs_binfo_dat));
590}
591
592static 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
601static 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
4e819509 610static struct nilfs_sc_operations nilfs_sc_dat_ops = {
9ff05123
RK		 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
4e819509 618static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
9ff05123
RK		 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
f30bf3e4
RK		 626static 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)
9ff05123 630{
9ff05123
RK		 631 struct address_space *mapping = inode->i_mapping;
632 struct pagevec pvec;
f30bf3e4
RK		 633 pgoff_t index = 0, last = ULONG_MAX;
634 size_t ndirties = 0;
635 int i;
9ff05123 636
f30bf3e4
RK		 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 }
9ff05123
RK		 646 pagevec_init(&pvec, 0);
647 repeat:
f30bf3e4
RK		 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;
9ff05123
RK		 653
654 for (i = 0; i < pagevec_count(&pvec); i++) {
655 struct buffer_head *bh, *head;
656 struct page *page = pvec.pages[i];
657
f30bf3e4
RK		 658 if (unlikely(page->index > last))
659 break;
660
aa405b1f
RK		 661 lock_page(page);
662 if (!page_has_buffers(page))
663 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
664 unlock_page(page);
9ff05123
RK		 665
666 bh = head = page_buffers(page);
667 do {
7f42ec39 668 if (!buffer_dirty(bh) || buffer_async_write(bh))
f30bf3e4
RK		 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;
9ff05123 677 }
f30bf3e4 678 } while (bh = bh->b_this_page, bh != head);
9ff05123
RK		 679 }
680 pagevec_release(&pvec);
681 cond_resched();
f30bf3e4 682 goto repeat;
9ff05123
RK		 683}
684
685static 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 {
7f42ec39
VD		 702 if (buffer_dirty(bh) &&
703 !buffer_async_write(bh)) {
9ff05123
RK		 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
693dd321 716static void nilfs_dispose_list(struct the_nilfs *nilfs,
9ff05123
RK		 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)) {
693dd321 724 spin_lock(&nilfs->ns_inode_lock);
9ff05123
RK		 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,
693dd321 735 &nilfs->ns_dirty_files);
9ff05123
RK		 736 continue;
737 }
738 ivec[nv++] = ii;
739 if (nv == SC_N_INODEVEC)
740 break;
741 }
693dd321 742 spin_unlock(&nilfs->ns_inode_lock);
9ff05123
RK		 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++;
365e215c
RK		 760 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
761 ret++;
9ff05123
RK		 762 return ret;
763}
764
765static 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) &&
071cb4b8 769 sci->sc_nfreesegs == 0 &&
9ff05123
RK		 770 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
771}
772
773static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
774{
e3154e97 775 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
9ff05123
RK		 776 int ret = 0;
777
693dd321 778 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
9ff05123
RK		 779 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
780
693dd321
RK		 781 spin_lock(&nilfs->ns_inode_lock);
782 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
9ff05123
RK		 783 ret++;
784
693dd321 785 spin_unlock(&nilfs->ns_inode_lock);
9ff05123
RK		 786 return ret;
787}
788
789static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
790{
e3154e97 791 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
9ff05123 792
e912a5b6 793 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
9ff05123
RK		 794 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
795 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
365e215c 796 nilfs_mdt_clear_dirty(nilfs->ns_dat);
9ff05123
RK		 797}
798
799static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
800{
e3154e97 801 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
9ff05123
RK		 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 */
5fc7b141 813 mark_buffer_dirty(bh_cp);
9ff05123
RK		 814 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
815 nilfs_cpfile_put_checkpoint(
816 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
1f5abe7e
RK		 817 } else
818 WARN_ON(err == -EINVAL || err == -ENOENT);
819
9ff05123
RK		 820 return err;
821}
822
823static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
824{
e3154e97 825 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
9ff05123
RK		 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)) {
1f5abe7e 833 WARN_ON(err == -EINVAL || err == -ENOENT);
9ff05123
RK		 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 =
e5f7f848 839 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
9ff05123 840 raw_cp->cp_blocks_count =
e5f7f848 841 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
9ff05123
RK		 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);
458c5b08 846
c96fa464
RK		 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
e912a5b6
RK		 852 nilfs_write_inode_common(sci->sc_root->ifile,
853 &raw_cp->cp_ifile_inode, 1);
9ff05123
RK		 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
861static 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
e912a5b6 878static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
9ff05123
RK		 879{
880 struct nilfs_inode_info *ii;
881
882 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
e912a5b6 883 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
9ff05123
RK		 884 set_bit(NILFS_I_COLLECTED, &ii->i_state);
885 }
9ff05123
RK		 886}
887
9ff05123
RK		 888static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
889 struct the_nilfs *nilfs)
890{
1e2b68bf
RK		 891 struct buffer_head *bh_sr;
892 struct nilfs_super_root *raw_sr;
56eb5538 893 unsigned isz, srsz;
9ff05123 894
1e2b68bf
RK		 895 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
896 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
56eb5538
RK		 897 isz = nilfs->ns_inode_size;
898 srsz = NILFS_SR_BYTES(isz);
1e2b68bf 899
56eb5538 900 raw_sr->sr_bytes = cpu_to_le16(srsz);
9ff05123
RK		 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
365e215c 906 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
3961f0e2
RK		 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);
56eb5538 912 memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
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
b9f66140 933 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
9ff05123
RK		 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)) {
0cc12838
RK		 961 bh = list_first_entry(listp, struct buffer_head,
962 b_assoc_buffers);
9ff05123
RK		 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{
e3154e97 1052 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
9ff05123
RK		 1053 struct list_head *head;
1054 struct nilfs_inode_info *ii;
071cb4b8 1055 size_t ndone;
9ff05123
RK		 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 }
9ff05123
RK		 1124 sci->sc_stage.scnt++;
1125 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1126 /* Fall through */
1127 case NILFS_ST_IFILE:
e912a5b6 1128 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
9ff05123
RK		 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:
071cb4b8
RK		 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);
9ff05123 1151 break;
071cb4b8
RK		 1152 }
1153 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1154
9ff05123
RK		 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:
365e215c 1162 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
9ff05123
RK		 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;
f30bf3e4 1185 ii = sci->sc_dsync_inode;
9ff05123
RK		 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;
9ff05123
RK		 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
a694291a
RK		 1205/**
1206 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1207 * @sci: nilfs_sc_info
1208 * @nilfs: nilfs object
1209 */
9ff05123
RK		 1210static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1211 struct the_nilfs *nilfs)
1212{
a694291a 1213 struct nilfs_segment_buffer *segbuf, *prev;
9ff05123 1214 __u64 nextnum;
a694291a 1215 int err, alloc = 0;
9ff05123 1216
a694291a
RK		 1217 segbuf = nilfs_segbuf_new(sci->sc_super);
1218 if (unlikely(!segbuf))
1219 return -ENOMEM;
9ff05123 1220
a694291a
RK		 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 }
9ff05123 1228
a694291a
RK		 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 }
9ff05123 1247 }
9ff05123 1248
61a189e9 1249 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
a694291a
RK		 1250 if (err)
1251 goto failed;
9ff05123 1252
a694291a 1253 if (alloc) {
cece5520 1254 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
a694291a
RK		 1255 if (err)
1256 goto failed;
1257 }
9ff05123
RK		 1258 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1259
a694291a
RK		 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;
cece5520 1263 return 0;
a694291a
RK		 1264
1265 failed:
1266 nilfs_segbuf_free(segbuf);
1267 return err;
9ff05123
RK		 1268}
1269
1270static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1271 struct the_nilfs *nilfs, int nadd)
1272{
e29df395 1273 struct nilfs_segment_buffer *segbuf, *prev;
9ff05123
RK		 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 */
61a189e9 1286 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
9ff05123
RK		 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 */
cece5520 1298 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
9ff05123
RK		 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 }
0935db74 1312 list_splice_tail(&list, &sci->sc_segbufs);
9ff05123
RK		 1313 return 0;
1314
1315 failed_segbuf:
1316 nilfs_segbuf_free(segbuf);
1317 failed:
e29df395 1318 list_for_each_entry(segbuf, &list, sb_list) {
9ff05123 1319 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1320 WARN_ON(ret); /* never fails */
9ff05123 1321 }
e29df395 1322 nilfs_destroy_logs(&list);
9ff05123
RK		 1323 return err;
1324}
1325
a694291a
RK		 1326static void nilfs_free_incomplete_logs(struct list_head *logs,
1327 struct the_nilfs *nilfs)
9ff05123 1328{
a694291a
RK		 1329 struct nilfs_segment_buffer *segbuf, *prev;
1330 struct inode *sufile = nilfs->ns_sufile;
9284ad2a 1331 int ret;
9ff05123 1332
a694291a 1333 segbuf = NILFS_FIRST_SEGBUF(logs);
9ff05123 1334 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
a694291a 1335 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1336 WARN_ON(ret); /* never fails */
9ff05123 1337 }
9284ad2a 1338 if (atomic_read(&segbuf->sb_err)) {
9ff05123
RK		 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);
9ff05123
RK		 1347 }
1348
a694291a
RK		 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 }
9284ad2a
RK		 1355 if (atomic_read(&segbuf->sb_err) &&
1356 segbuf->sb_segnum != nilfs->ns_nextnum)
1357 /* Case 2: extended segment (!= next) failed */
a694291a
RK		 1358 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1359 prev = segbuf;
9ff05123 1360 }
9ff05123
RK		 1361}
1362
1363static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1364 struct inode *sufile)
1365{
1366 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1367 unsigned long live_blocks;
1368 int ret;
1369
1370 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
9ff05123
RK		 1371 live_blocks = segbuf->sb_sum.nblocks +
1372 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
071ec54d
RK		 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 */
9ff05123
RK		 1377 }
1378}
1379
a694291a 1380static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
9ff05123
RK		 1381{
1382 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1383 int ret;
1384
a694291a 1385 segbuf = NILFS_FIRST_SEGBUF(logs);
071ec54d
RK		 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 */
9ff05123 1390
a694291a 1391 list_for_each_entry_continue(segbuf, logs, sb_list) {
071ec54d
RK		 1392 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1393 0, 0);
1f5abe7e 1394 WARN_ON(ret); /* always succeed */
9ff05123
RK		 1395 }
1396}
1397
1398static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1399 struct nilfs_segment_buffer *last,
1400 struct inode *sufile)
1401{
e29df395 1402 struct nilfs_segment_buffer *segbuf = last;
9ff05123
RK		 1403 int ret;
1404
e29df395 1405 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
9ff05123
RK		 1406 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1407 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1408 WARN_ON(ret);
9ff05123 1409 }
e29df395 1410 nilfs_truncate_logs(&sci->sc_segbufs, last);
9ff05123
RK		 1411}
1412
1413
1414static 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 (;;) {
9ff05123
RK		 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
2d8428ac
RK		 1441 nilfs_clear_logs(&sci->sc_segbufs);
1442
071cb4b8
RK		 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 */
70f2fe3a 1449 sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
071cb4b8 1450 }
70f2fe3a
AR		 1451
1452 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1453 if (unlikely(err))
1454 return err;
1455
9ff05123
RK		 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
1466static 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
1475static int
1476nilfs_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) {
1e2b68bf 1500 if (bh == segbuf->sb_super_root)
9ff05123
RK		 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
aa405b1f 1509 inode = bh->b_page->mapping->host;
9ff05123
RK		 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:
9ff05123
RK		 1545 return err;
1546}
1547
1548static 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
1cb2d38c 1562static void nilfs_begin_page_io(struct page *page)
9ff05123
RK		 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. */
1cb2d38c 1567 return;
9ff05123
RK		 1568
1569 lock_page(page);
1570 clear_page_dirty_for_io(page);
1571 set_page_writeback(page);
1572 unlock_page(page);
9ff05123
RK		 1573}
1574
1cb2d38c 1575static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
9ff05123
RK		 1576{
1577 struct nilfs_segment_buffer *segbuf;
1578 struct page *bd_page = NULL, *fs_page = NULL;
9ff05123 1579
9ff05123
RK		 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) {
7f42ec39 1585 set_buffer_async_write(bh);
9ff05123
RK		 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) {
7f42ec39 1599 set_buffer_async_write(bh);
1e2b68bf 1600 if (bh == segbuf->sb_super_root) {
9ff05123
RK		 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) {
1cb2d38c 1611 nilfs_begin_page_io(fs_page);
9ff05123
RK		 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 }
1cb2d38c 1622 nilfs_begin_page_io(fs_page);
9ff05123
RK		 1623}
1624
1625static int nilfs_segctor_write(struct nilfs_sc_info *sci,
9c965bac 1626 struct the_nilfs *nilfs)
9ff05123 1627{
d1c6b72a 1628 int ret;
9ff05123 1629
d1c6b72a 1630 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
a694291a
RK		 1631 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1632 return ret;
9ff05123
RK		 1633}
1634
9ff05123
RK		 1635static void nilfs_end_page_io(struct page *page, int err)
1636{
1637 if (!page)
1638 return;
1639
a9777845 1640 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
8227b297
RK		 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 */
a9777845
RK		 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 }
9ff05123 1658 return;
a9777845 1659 }
9ff05123 1660
1cb2d38c
RK		 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);
9ff05123 1668 }
1cb2d38c
RK		 1669
1670 end_page_writeback(page);
9ff05123
RK		 1671}
1672
1cb2d38c 1673static void nilfs_abort_logs(struct list_head *logs, int err)
9ff05123
RK		 1674{
1675 struct nilfs_segment_buffer *segbuf;
1676 struct page *bd_page = NULL, *fs_page = NULL;
a694291a 1677 struct buffer_head *bh;
9ff05123 1678
a694291a
RK		 1679 if (list_empty(logs))
1680 return;
9ff05123 1681
a694291a 1682 list_for_each_entry(segbuf, logs, sb_list) {
9ff05123
RK		 1683 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1684 b_assoc_buffers) {
7f42ec39 1685 clear_buffer_async_write(bh);
9ff05123
RK		 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) {
7f42ec39 1695 clear_buffer_async_write(bh);
1e2b68bf 1696 if (bh == segbuf->sb_super_root) {
9ff05123
RK		 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);
9ff05123
RK		 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);
a694291a
RK		 1713}
1714
1715static 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);
1cb2d38c 1723 nilfs_abort_logs(&logs, ret ? : err);
a694291a
RK		 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);
a694291a
RK		 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);
9ff05123
RK		 1738}
1739
1740static 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
1751static 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;
e3154e97 1755 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1e2b68bf 1756 int update_sr = false;
9ff05123 1757
a694291a 1758 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
9ff05123
RK		 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);
7f42ec39 1765 clear_buffer_async_write(bh);
9ff05123
RK		 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);
7f42ec39 1787 clear_buffer_async_write(bh);
27e6c7a3 1788 clear_buffer_delay(bh);
9ff05123 1789 clear_buffer_nilfs_volatile(bh);
b1f6a4f2 1790 clear_buffer_nilfs_redirected(bh);
1e2b68bf 1791 if (bh == segbuf->sb_super_root) {
9ff05123
RK		 1792 if (bh->b_page != bd_page) {
1793 end_page_writeback(bd_page);
1794 bd_page = bh->b_page;
1795 }
1e2b68bf 1796 update_sr = true;
9ff05123
RK		 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
4762077c
RK		 1805 if (!nilfs_segbuf_simplex(segbuf)) {
1806 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
9ff05123
RK		 1807 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1808 sci->sc_lseg_stime = jiffies;
1809 }
4762077c 1810 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
9ff05123
RK		 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
9ff05123
RK		 1823 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1824
c1c1d709 1825 if (nilfs_doing_gc())
9ff05123 1826 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
c1c1d709 1827 else
9ff05123 1828 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
9ff05123
RK		 1829
1830 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1831
a694291a 1832 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
9ff05123
RK		 1833 nilfs_set_next_segment(nilfs, segbuf);
1834
1835 if (update_sr) {
e2c7617a 1836 nilfs->ns_flushed_device = 0;
9ff05123 1837 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
e339ad31 1838 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
9ff05123 1839
c96fa464 1840 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
9ff05123
RK		 1841 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1842 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
a694291a 1843 nilfs_segctor_clear_metadata_dirty(sci);
9ff05123
RK		 1844 } else
1845 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1846}
1847
a694291a
RK		 1848static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1849{
1850 int ret;
1851
1852 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1853 if (!ret) {
1854 nilfs_segctor_complete_write(sci);
1855 nilfs_destroy_logs(&sci->sc_write_logs);
1856 }
1857 return ret;
1858}
1859
693dd321
RK		 1860static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1861 struct the_nilfs *nilfs)
9ff05123
RK		 1862{
1863 struct nilfs_inode_info *ii, *n;
e912a5b6 1864 struct inode *ifile = sci->sc_root->ifile;
9ff05123 1865
693dd321 1866 spin_lock(&nilfs->ns_inode_lock);
9ff05123 1867 retry:
693dd321 1868 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
9ff05123
RK		 1869 if (!ii->i_bh) {
1870 struct buffer_head *ibh;
1871 int err;
1872
693dd321 1873 spin_unlock(&nilfs->ns_inode_lock);
9ff05123 1874 err = nilfs_ifile_get_inode_block(
e912a5b6 1875 ifile, ii->vfs_inode.i_ino, &ibh);
9ff05123 1876 if (unlikely(err)) {
693dd321 1877 nilfs_warning(sci->sc_super, __func__,
9ff05123
RK		 1878 "failed to get inode block.\n");
1879 return err;
1880 }
5fc7b141 1881 mark_buffer_dirty(ibh);
e912a5b6 1882 nilfs_mdt_mark_dirty(ifile);
693dd321 1883 spin_lock(&nilfs->ns_inode_lock);
9ff05123
RK		 1884 if (likely(!ii->i_bh))
1885 ii->i_bh = ibh;
1886 else
1887 brelse(ibh);
1888 goto retry;
1889 }
9ff05123
RK		 1890
1891 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1892 set_bit(NILFS_I_BUSY, &ii->i_state);
eaae0f37 1893 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
9ff05123 1894 }
693dd321 1895 spin_unlock(&nilfs->ns_inode_lock);
9ff05123 1896
9ff05123
RK		 1897 return 0;
1898}
1899
693dd321
RK		 1900static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1901 struct the_nilfs *nilfs)
9ff05123
RK		 1902{
1903 struct nilfs_transaction_info *ti = current->journal_info;
1904 struct nilfs_inode_info *ii, *n;
9ff05123 1905
693dd321 1906 spin_lock(&nilfs->ns_inode_lock);
9ff05123
RK		 1907 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1908 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
6c43f410 1909 test_bit(NILFS_I_DIRTY, &ii->i_state))
9ff05123 1910 continue;
6c43f410 1911
9ff05123
RK		 1912 clear_bit(NILFS_I_BUSY, &ii->i_state);
1913 brelse(ii->i_bh);
1914 ii->i_bh = NULL;
eaae0f37 1915 list_move_tail(&ii->i_dirty, &ti->ti_garbage);
9ff05123 1916 }
693dd321 1917 spin_unlock(&nilfs->ns_inode_lock);
9ff05123
RK		 1918}
1919
9ff05123
RK		 1920/*
1921 * Main procedure of segment constructor
1922 */
1923static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
1924{
e3154e97 1925 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1e2b68bf 1926 int err;
9ff05123
RK		 1927
1928 sci->sc_stage.scnt = NILFS_ST_INIT;
6c43f410 1929 sci->sc_cno = nilfs->ns_cno;
9ff05123 1930
693dd321 1931 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
9ff05123
RK		 1932 if (unlikely(err))
1933 goto out;
1934
e912a5b6 1935 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
9ff05123
RK		 1936 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1937
1938 if (nilfs_segctor_clean(sci))
1939 goto out;
1940
1941 do {
1942 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
1943
1944 err = nilfs_segctor_begin_construction(sci, nilfs);
1945 if (unlikely(err))
1946 goto out;
1947
1948 /* Update time stamp */
1949 sci->sc_seg_ctime = get_seconds();
1950
1951 err = nilfs_segctor_collect(sci, nilfs, mode);
1952 if (unlikely(err))
1953 goto failed;
1954
9ff05123
RK		 1955 /* Avoid empty segment */
1956 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
4762077c 1957 nilfs_segbuf_empty(sci->sc_curseg)) {
a694291a 1958 nilfs_segctor_abort_construction(sci, nilfs, 1);
9ff05123
RK		 1959 goto out;
1960 }
1961
1962 err = nilfs_segctor_assign(sci, mode);
1963 if (unlikely(err))
1964 goto failed;
1965
9ff05123 1966 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
e912a5b6 1967 nilfs_segctor_fill_in_file_bmap(sci);
9ff05123 1968
1e2b68bf
RK		 1969 if (mode == SC_LSEG_SR &&
1970 sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
9ff05123
RK		 1971 err = nilfs_segctor_fill_in_checkpoint(sci);
1972 if (unlikely(err))
a694291a 1973 goto failed_to_write;
9ff05123
RK		 1974
1975 nilfs_segctor_fill_in_super_root(sci, nilfs);
1976 }
1977 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
1978
1979 /* Write partial segments */
1cb2d38c 1980 nilfs_segctor_prepare_write(sci);
aaed1d5b
RK		 1981
1982 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
1983 nilfs->ns_crc_seed);
9ff05123 1984
9c965bac 1985 err = nilfs_segctor_write(sci, nilfs);
9ff05123
RK		 1986 if (unlikely(err))
1987 goto failed_to_write;
1988
a694291a
RK		 1989 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
1990 nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
1991 /*
1992 * At this point, we avoid double buffering
1993 * for blocksize < pagesize because page dirty
1994 * flag is turned off during write and dirty
1995 * buffers are not properly collected for
1996 * pages crossing over segments.
1997 */
1998 err = nilfs_segctor_wait(sci);
1999 if (err)
2000 goto failed_to_write;
2001 }
9ff05123
RK		 2002 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2003
9ff05123 2004 out:
693dd321 2005 nilfs_segctor_drop_written_files(sci, nilfs);
9ff05123
RK		 2006 return err;
2007
2008 failed_to_write:
9ff05123
RK		 2009 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2010 nilfs_redirty_inodes(&sci->sc_dirty_files);
9ff05123
RK		 2011
2012 failed:
2013 if (nilfs_doing_gc())
2014 nilfs_redirty_inodes(&sci->sc_gc_inodes);
a694291a 2015 nilfs_segctor_abort_construction(sci, nilfs, err);
9ff05123
RK		 2016 goto out;
2017}
2018
2019/**
9ccf56c1 2020 * nilfs_segctor_start_timer - set timer of background write
9ff05123
RK		 2021 * @sci: nilfs_sc_info
2022 *
2023 * If the timer has already been set, it ignores the new request.
2024 * This function MUST be called within a section locking the segment
2025 * semaphore.
2026 */
2027static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2028{
2029 spin_lock(&sci->sc_state_lock);
fdce895e
LH		 2030 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2031 sci->sc_timer.expires = jiffies + sci->sc_interval;
2032 add_timer(&sci->sc_timer);
9ff05123
RK		 2033 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2034 }
2035 spin_unlock(&sci->sc_state_lock);
2036}
2037
2038static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2039{
2040 spin_lock(&sci->sc_state_lock);
2041 if (!(sci->sc_flush_request & (1 << bn))) {
2042 unsigned long prev_req = sci->sc_flush_request;
2043
2044 sci->sc_flush_request |= (1 << bn);
2045 if (!prev_req)
2046 wake_up(&sci->sc_wait_daemon);
2047 }
2048 spin_unlock(&sci->sc_state_lock);
2049}
2050
2051/**
2052 * nilfs_flush_segment - trigger a segment construction for resource control
2053 * @sb: super block
2054 * @ino: inode number of the file to be flushed out.
2055 */
2056void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2057{
e3154e97 2058 struct the_nilfs *nilfs = sb->s_fs_info;
3fd3fe5a 2059 struct nilfs_sc_info *sci = nilfs->ns_writer;
9ff05123
RK		 2060
2061 if (!sci || nilfs_doing_construction())
2062 return;
2063 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2064 /* assign bit 0 to data files */
2065}
2066
9ff05123
RK		 2067struct nilfs_segctor_wait_request {
2068 wait_queue_t wq;
2069 __u32 seq;
2070 int err;
2071 atomic_t done;
2072};
2073
2074static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2075{
2076 struct nilfs_segctor_wait_request wait_req;
2077 int err = 0;
2078
2079 spin_lock(&sci->sc_state_lock);
2080 init_wait(&wait_req.wq);
2081 wait_req.err = 0;
2082 atomic_set(&wait_req.done, 0);
2083 wait_req.seq = ++sci->sc_seq_request;
2084 spin_unlock(&sci->sc_state_lock);
2085
2086 init_waitqueue_entry(&wait_req.wq, current);
2087 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2088 set_current_state(TASK_INTERRUPTIBLE);
2089 wake_up(&sci->sc_wait_daemon);
2090
2091 for (;;) {
2092 if (atomic_read(&wait_req.done)) {
2093 err = wait_req.err;
2094 break;
2095 }
2096 if (!signal_pending(current)) {
2097 schedule();
2098 continue;
2099 }
2100 err = -ERESTARTSYS;
2101 break;
2102 }
2103 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2104 return err;
2105}
2106
2107static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2108{
2109 struct nilfs_segctor_wait_request *wrq, *n;
2110 unsigned long flags;
2111
2112 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2113 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2114 wq.task_list) {
2115 if (!atomic_read(&wrq->done) &&
2116 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2117 wrq->err = err;
2118 atomic_set(&wrq->done, 1);
2119 }
2120 if (atomic_read(&wrq->done)) {
2121 wrq->wq.func(&wrq->wq,
2122 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2123 0, NULL);
2124 }
2125 }
2126 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2127}
2128
2129/**
2130 * nilfs_construct_segment - construct a logical segment
2131 * @sb: super block
2132 *
2133 * Return Value: On success, 0 is retured. On errors, one of the following
2134 * negative error code is returned.
2135 *
2136 * %-EROFS - Read only filesystem.
2137 *
2138 * %-EIO - I/O error
2139 *
2140 * %-ENOSPC - No space left on device (only in a panic state).
2141 *
2142 * %-ERESTARTSYS - Interrupted.
2143 *
2144 * %-ENOMEM - Insufficient memory available.
2145 */
2146int nilfs_construct_segment(struct super_block *sb)
2147{
e3154e97 2148 struct the_nilfs *nilfs = sb->s_fs_info;
3fd3fe5a 2149 struct nilfs_sc_info *sci = nilfs->ns_writer;
9ff05123
RK		 2150 struct nilfs_transaction_info *ti;
2151 int err;
2152
2153 if (!sci)
2154 return -EROFS;
2155
2156 /* A call inside transactions causes a deadlock. */
2157 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2158
2159 err = nilfs_segctor_sync(sci);
2160 return err;
2161}
2162
2163/**
2164 * nilfs_construct_dsync_segment - construct a data-only logical segment
2165 * @sb: super block
f30bf3e4
RK		 2166 * @inode: inode whose data blocks should be written out
2167 * @start: start byte offset
2168 * @end: end byte offset (inclusive)
9ff05123
RK		 2169 *
2170 * Return Value: On success, 0 is retured. On errors, one of the following
2171 * negative error code is returned.
2172 *
2173 * %-EROFS - Read only filesystem.
2174 *
2175 * %-EIO - I/O error
2176 *
2177 * %-ENOSPC - No space left on device (only in a panic state).
2178 *
2179 * %-ERESTARTSYS - Interrupted.
2180 *
2181 * %-ENOMEM - Insufficient memory available.
2182 */
f30bf3e4
RK		 2183int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2184 loff_t start, loff_t end)
9ff05123 2185{
e3154e97 2186 struct the_nilfs *nilfs = sb->s_fs_info;
3fd3fe5a 2187 struct nilfs_sc_info *sci = nilfs->ns_writer;
9ff05123
RK		 2188 struct nilfs_inode_info *ii;
2189 struct nilfs_transaction_info ti;
2190 int err = 0;
2191
2192 if (!sci)
2193 return -EROFS;
2194
f7545144 2195 nilfs_transaction_lock(sb, &ti, 0);
9ff05123
RK		 2196
2197 ii = NILFS_I(inode);
b9f66140 2198 if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
3b2ce58b 2199 nilfs_test_opt(nilfs, STRICT_ORDER) ||
9ff05123 2200 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
3b2ce58b 2201 nilfs_discontinued(nilfs)) {
f7545144 2202 nilfs_transaction_unlock(sb);
9ff05123
RK		 2203 err = nilfs_segctor_sync(sci);
2204 return err;
2205 }
2206
693dd321 2207 spin_lock(&nilfs->ns_inode_lock);
9ff05123
RK		 2208 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2209 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
693dd321 2210 spin_unlock(&nilfs->ns_inode_lock);
f7545144 2211 nilfs_transaction_unlock(sb);
9ff05123
RK		 2212 return 0;
2213 }
693dd321 2214 spin_unlock(&nilfs->ns_inode_lock);
f30bf3e4
RK		 2215 sci->sc_dsync_inode = ii;
2216 sci->sc_dsync_start = start;
2217 sci->sc_dsync_end = end;
9ff05123
RK		 2218
2219 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
e2c7617a
AR		 2220 if (!err)
2221 nilfs->ns_flushed_device = 0;
9ff05123 2222
f7545144 2223 nilfs_transaction_unlock(sb);
9ff05123
RK		 2224 return err;
2225}
2226
9ff05123
RK		 2227#define FLUSH_FILE_BIT (0x1) /* data file only */
2228#define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2229
dcd76186
RK		 2230/**
2231 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2232 * @sci: segment constructor object
2233 */
2234static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
9ff05123 2235{
9ff05123 2236 spin_lock(&sci->sc_state_lock);
dcd76186 2237 sci->sc_seq_accepted = sci->sc_seq_request;
9ff05123 2238 spin_unlock(&sci->sc_state_lock);
fdce895e 2239 del_timer_sync(&sci->sc_timer);
9ff05123
RK		 2240}
2241
dcd76186
RK		 2242/**
2243 * nilfs_segctor_notify - notify the result of request to caller threads
2244 * @sci: segment constructor object
2245 * @mode: mode of log forming
2246 * @err: error code to be notified
2247 */
2248static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
9ff05123
RK		 2249{
2250 /* Clear requests (even when the construction failed) */
2251 spin_lock(&sci->sc_state_lock);
2252
dcd76186 2253 if (mode == SC_LSEG_SR) {
aeda7f63 2254 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
dcd76186
RK		 2255 sci->sc_seq_done = sci->sc_seq_accepted;
2256 nilfs_segctor_wakeup(sci, err);
9ff05123 2257 sci->sc_flush_request = 0;
aeda7f63 2258 } else {
dcd76186 2259 if (mode == SC_FLUSH_FILE)
aeda7f63 2260 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
dcd76186 2261 else if (mode == SC_FLUSH_DAT)
aeda7f63
RK		 2262 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2263
2264 /* re-enable timer if checkpoint creation was not done */
fdce895e
LH		 2265 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2266 time_before(jiffies, sci->sc_timer.expires))
2267 add_timer(&sci->sc_timer);
aeda7f63 2268 }
9ff05123
RK		 2269 spin_unlock(&sci->sc_state_lock);
2270}
2271
dcd76186
RK		 2272/**
2273 * nilfs_segctor_construct - form logs and write them to disk
2274 * @sci: segment constructor object
2275 * @mode: mode of log forming
2276 */
2277static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
9ff05123 2278{
e3154e97 2279 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
d26493b6 2280 struct nilfs_super_block **sbp;
9ff05123
RK		 2281 int err = 0;
2282
dcd76186
RK		 2283 nilfs_segctor_accept(sci);
2284
9ff05123 2285 if (nilfs_discontinued(nilfs))
dcd76186
RK		 2286 mode = SC_LSEG_SR;
2287 if (!nilfs_segctor_confirm(sci))
2288 err = nilfs_segctor_do_construct(sci, mode);
2289
9ff05123 2290 if (likely(!err)) {
dcd76186 2291 if (mode != SC_FLUSH_DAT)
9ff05123
RK		 2292 atomic_set(&nilfs->ns_ndirtyblks, 0);
2293 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2294 nilfs_discontinued(nilfs)) {
2295 down_write(&nilfs->ns_sem);
d26493b6 2296 err = -EIO;
f7545144 2297 sbp = nilfs_prepare_super(sci->sc_super,
b2ac86e1
JS		 2298 nilfs_sb_will_flip(nilfs));
2299 if (likely(sbp)) {
2300 nilfs_set_log_cursor(sbp[0], nilfs);
f7545144
RK		 2301 err = nilfs_commit_super(sci->sc_super,
2302 NILFS_SB_COMMIT);
b2ac86e1 2303 }
9ff05123
RK		 2304 up_write(&nilfs->ns_sem);
2305 }
2306 }
dcd76186
RK		 2307
2308 nilfs_segctor_notify(sci, mode, err);
9ff05123
RK		 2309 return err;
2310}
2311
2312static void nilfs_construction_timeout(unsigned long data)
2313{
2314 struct task_struct *p = (struct task_struct *)data;
2315 wake_up_process(p);
2316}
2317
2318static void
2319nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2320{
2321 struct nilfs_inode_info *ii, *n;
2322
2323 list_for_each_entry_safe(ii, n, head, i_dirty) {
2324 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2325 continue;
9ff05123 2326 list_del_init(&ii->i_dirty);
fbb24a3a
RK		 2327 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2328 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
263d90ce 2329 iput(&ii->vfs_inode);
9ff05123
RK		 2330 }
2331}
2332
4f6b8288
RK		 2333int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2334 void **kbufs)
9ff05123 2335{
e3154e97 2336 struct the_nilfs *nilfs = sb->s_fs_info;
3fd3fe5a 2337 struct nilfs_sc_info *sci = nilfs->ns_writer;
9ff05123 2338 struct nilfs_transaction_info ti;
9ff05123
RK		 2339 int err;
2340
2341 if (unlikely(!sci))
2342 return -EROFS;
2343
f7545144 2344 nilfs_transaction_lock(sb, &ti, 1);
9ff05123 2345
c1c1d709 2346 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
9ff05123
RK		 2347 if (unlikely(err))
2348 goto out_unlock;
071cb4b8 2349
4f6b8288 2350 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
c1c1d709
RK		 2351 if (unlikely(err)) {
2352 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
9ff05123 2353 goto out_unlock;
c1c1d709 2354 }
9ff05123 2355
071cb4b8
RK		 2356 sci->sc_freesegs = kbufs[4];
2357 sci->sc_nfreesegs = argv[4].v_nmembs;
0935db74 2358 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
9ff05123
RK		 2359
2360 for (;;) {
dcd76186 2361 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
9ff05123 2362 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
9ff05123
RK		 2363
2364 if (likely(!err))
2365 break;
2366
2367 nilfs_warning(sb, __func__,
2368 "segment construction failed. (err=%d)", err);
2369 set_current_state(TASK_INTERRUPTIBLE);
2370 schedule_timeout(sci->sc_interval);
2371 }
3b2ce58b 2372 if (nilfs_test_opt(nilfs, DISCARD)) {
e902ec99
JS		 2373 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2374 sci->sc_nfreesegs);
2375 if (ret) {
2376 printk(KERN_WARNING
2377 "NILFS warning: error %d on discard request, "
2378 "turning discards off for the device\n", ret);
3b2ce58b 2379 nilfs_clear_opt(nilfs, DISCARD);
e902ec99
JS		 2380 }
2381 }
9ff05123
RK		 2382
2383 out_unlock:
071cb4b8
RK		 2384 sci->sc_freesegs = NULL;
2385 sci->sc_nfreesegs = 0;
c1c1d709 2386 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
f7545144 2387 nilfs_transaction_unlock(sb);
9ff05123
RK		 2388 return err;
2389}
2390
2391static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2392{
9ff05123 2393 struct nilfs_transaction_info ti;
9ff05123 2394
f7545144 2395 nilfs_transaction_lock(sci->sc_super, &ti, 0);
dcd76186 2396 nilfs_segctor_construct(sci, mode);
9ff05123
RK		 2397
2398 /*
2399 * Unclosed segment should be retried. We do this using sc_timer.
2400 * Timeout of sc_timer will invoke complete construction which leads
2401 * to close the current logical segment.
2402 */
2403 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2404 nilfs_segctor_start_timer(sci);
2405
f7545144 2406 nilfs_transaction_unlock(sci->sc_super);
9ff05123
RK		 2407}
2408
2409static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2410{
2411 int mode = 0;
2412 int err;
2413
2414 spin_lock(&sci->sc_state_lock);
2415 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2416 SC_FLUSH_DAT : SC_FLUSH_FILE;
2417 spin_unlock(&sci->sc_state_lock);
2418
2419 if (mode) {
2420 err = nilfs_segctor_do_construct(sci, mode);
2421
2422 spin_lock(&sci->sc_state_lock);
2423 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2424 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2425 spin_unlock(&sci->sc_state_lock);
2426 }
2427 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2428}
2429
2430static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2431{
2432 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2433 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2434 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2435 return SC_FLUSH_FILE;
2436 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2437 return SC_FLUSH_DAT;
2438 }
2439 return SC_LSEG_SR;
2440}
2441
2442/**
2443 * nilfs_segctor_thread - main loop of the segment constructor thread.
2444 * @arg: pointer to a struct nilfs_sc_info.
2445 *
2446 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2447 * to execute segment constructions.
2448 */
2449static int nilfs_segctor_thread(void *arg)
2450{
2451 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
e3154e97 2452 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
9ff05123
RK		 2453 int timeout = 0;
2454
fdce895e
LH		 2455 sci->sc_timer.data = (unsigned long)current;
2456 sci->sc_timer.function = nilfs_construction_timeout;
9ff05123
RK		 2457
2458 /* start sync. */
2459 sci->sc_task = current;
2460 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2461 printk(KERN_INFO
2462 "segctord starting. Construction interval = %lu seconds, "
2463 "CP frequency < %lu seconds\n",
2464 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2465
2466 spin_lock(&sci->sc_state_lock);
2467 loop:
2468 for (;;) {
2469 int mode;
2470
2471 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2472 goto end_thread;
2473
2474 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2475 mode = SC_LSEG_SR;
2476 else if (!sci->sc_flush_request)
2477 break;
2478 else
2479 mode = nilfs_segctor_flush_mode(sci);
2480
2481 spin_unlock(&sci->sc_state_lock);
2482 nilfs_segctor_thread_construct(sci, mode);
2483 spin_lock(&sci->sc_state_lock);
2484 timeout = 0;
2485 }
2486
2487
2488 if (freezing(current)) {
2489 spin_unlock(&sci->sc_state_lock);
a0acae0e 2490 try_to_freeze();
9ff05123
RK		 2491 spin_lock(&sci->sc_state_lock);
2492 } else {
2493 DEFINE_WAIT(wait);
2494 int should_sleep = 1;
2495
2496 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2497 TASK_INTERRUPTIBLE);
2498
2499 if (sci->sc_seq_request != sci->sc_seq_done)
2500 should_sleep = 0;
2501 else if (sci->sc_flush_request)
2502 should_sleep = 0;
2503 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2504 should_sleep = time_before(jiffies,
fdce895e 2505 sci->sc_timer.expires);
9ff05123
RK		 2506
2507 if (should_sleep) {
2508 spin_unlock(&sci->sc_state_lock);
2509 schedule();
2510 spin_lock(&sci->sc_state_lock);
2511 }
2512 finish_wait(&sci->sc_wait_daemon, &wait);
2513 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
fdce895e 2514 time_after_eq(jiffies, sci->sc_timer.expires));
e605f0a7
RK		 2515
2516 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
1dfa2710 2517 set_nilfs_discontinued(nilfs);
9ff05123
RK		 2518 }
2519 goto loop;
2520
2521 end_thread:
2522 spin_unlock(&sci->sc_state_lock);
9ff05123
RK		 2523
2524 /* end sync. */
2525 sci->sc_task = NULL;
2526 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2527 return 0;
2528}
2529
2530static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2531{
2532 struct task_struct *t;
2533
2534 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2535 if (IS_ERR(t)) {
2536 int err = PTR_ERR(t);
2537
2538 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2539 err);
2540 return err;
2541 }
2542 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2543 return 0;
2544}
2545
2546static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
6b81e14e
JS		 2547 __acquires(&sci->sc_state_lock)
2548 __releases(&sci->sc_state_lock)
9ff05123
RK		 2549{
2550 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2551
2552 while (sci->sc_task) {
2553 wake_up(&sci->sc_wait_daemon);
2554 spin_unlock(&sci->sc_state_lock);
2555 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2556 spin_lock(&sci->sc_state_lock);
2557 }
2558}
2559
9ff05123
RK		 2560/*
2561 * Setup & clean-up functions
2562 */
f7545144 2563static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
e912a5b6 2564 struct nilfs_root *root)
9ff05123 2565{
e3154e97 2566 struct the_nilfs *nilfs = sb->s_fs_info;
9ff05123
RK		 2567 struct nilfs_sc_info *sci;
2568
2569 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2570 if (!sci)
2571 return NULL;
2572
f7545144 2573 sci->sc_super = sb;
9ff05123 2574
e912a5b6
RK		 2575 nilfs_get_root(root);
2576 sci->sc_root = root;
2577
9ff05123
RK		 2578 init_waitqueue_head(&sci->sc_wait_request);
2579 init_waitqueue_head(&sci->sc_wait_daemon);
2580 init_waitqueue_head(&sci->sc_wait_task);
2581 spin_lock_init(&sci->sc_state_lock);
2582 INIT_LIST_HEAD(&sci->sc_dirty_files);
2583 INIT_LIST_HEAD(&sci->sc_segbufs);
a694291a 2584 INIT_LIST_HEAD(&sci->sc_write_logs);
9ff05123 2585 INIT_LIST_HEAD(&sci->sc_gc_inodes);
fdce895e 2586 init_timer(&sci->sc_timer);
9ff05123
RK		 2587
2588 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2589 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2590 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2591
574e6c31 2592 if (nilfs->ns_interval)
071d73cf 2593 sci->sc_interval = HZ * nilfs->ns_interval;
574e6c31
RK		 2594 if (nilfs->ns_watermark)
2595 sci->sc_watermark = nilfs->ns_watermark;
9ff05123
RK		 2596 return sci;
2597}
2598
2599static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2600{
2601 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2602
2603 /* The segctord thread was stopped and its timer was removed.
2604 But some tasks remain. */
2605 do {
9ff05123 2606 struct nilfs_transaction_info ti;
9ff05123 2607
f7545144 2608 nilfs_transaction_lock(sci->sc_super, &ti, 0);
dcd76186 2609 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
f7545144 2610 nilfs_transaction_unlock(sci->sc_super);
9ff05123
RK		 2611
2612 } while (ret && retrycount-- > 0);
2613}
2614
2615/**
2616 * nilfs_segctor_destroy - destroy the segment constructor.
2617 * @sci: nilfs_sc_info
2618 *
2619 * nilfs_segctor_destroy() kills the segctord thread and frees
2620 * the nilfs_sc_info struct.
2621 * Caller must hold the segment semaphore.
2622 */
2623static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2624{
e3154e97 2625 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
9ff05123
RK		 2626 int flag;
2627
693dd321 2628 up_write(&nilfs->ns_segctor_sem);
9ff05123
RK		 2629
2630 spin_lock(&sci->sc_state_lock);
2631 nilfs_segctor_kill_thread(sci);
2632 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2633 || sci->sc_seq_request != sci->sc_seq_done);
2634 spin_unlock(&sci->sc_state_lock);
2635
3256a055 2636 if (flag || !nilfs_segctor_confirm(sci))
9ff05123
RK		 2637 nilfs_segctor_write_out(sci);
2638
9ff05123 2639 if (!list_empty(&sci->sc_dirty_files)) {
693dd321 2640 nilfs_warning(sci->sc_super, __func__,
9ff05123 2641 "dirty file(s) after the final construction\n");
693dd321 2642 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
9ff05123 2643 }
9ff05123 2644
1f5abe7e 2645 WARN_ON(!list_empty(&sci->sc_segbufs));
a694291a 2646 WARN_ON(!list_empty(&sci->sc_write_logs));
9ff05123 2647
e912a5b6
RK		 2648 nilfs_put_root(sci->sc_root);
2649
693dd321 2650 down_write(&nilfs->ns_segctor_sem);
9ff05123 2651
fdce895e 2652 del_timer_sync(&sci->sc_timer);
9ff05123
RK		 2653 kfree(sci);
2654}
2655
2656/**
f7545144
RK		 2657 * nilfs_attach_log_writer - attach log writer
2658 * @sb: super block instance
e912a5b6 2659 * @root: root object of the current filesystem tree
9ff05123 2660 *
f7545144
RK		 2661 * This allocates a log writer object, initializes it, and starts the
2662 * log writer.
9ff05123
RK		 2663 *
2664 * Return Value: On success, 0 is returned. On error, one of the following
2665 * negative error code is returned.
2666 *
2667 * %-ENOMEM - Insufficient memory available.
2668 */
f7545144 2669int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
9ff05123 2670{
e3154e97 2671 struct the_nilfs *nilfs = sb->s_fs_info;
9ff05123
RK		 2672 int err;
2673
3fd3fe5a 2674 if (nilfs->ns_writer) {
fe5f171b
RK		 2675 /*
2676 * This happens if the filesystem was remounted
2677 * read/write after nilfs_error degenerated it into a
2678 * read-only mount.
2679 */
f7545144 2680 nilfs_detach_log_writer(sb);
fe5f171b
RK		 2681 }
2682
f7545144 2683 nilfs->ns_writer = nilfs_segctor_new(sb, root);
3fd3fe5a 2684 if (!nilfs->ns_writer)
9ff05123
RK		 2685 return -ENOMEM;
2686
3fd3fe5a 2687 err = nilfs_segctor_start_thread(nilfs->ns_writer);
9ff05123 2688 if (err) {
3fd3fe5a
RK		 2689 kfree(nilfs->ns_writer);
2690 nilfs->ns_writer = NULL;
9ff05123
RK		 2691 }
2692 return err;
2693}
2694
2695/**
f7545144
RK		 2696 * nilfs_detach_log_writer - destroy log writer
2697 * @sb: super block instance
9ff05123 2698 *
f7545144
RK		 2699 * This kills log writer daemon, frees the log writer object, and
2700 * destroys list of dirty files.
9ff05123 2701 */
f7545144 2702void nilfs_detach_log_writer(struct super_block *sb)
9ff05123 2703{
e3154e97 2704 struct the_nilfs *nilfs = sb->s_fs_info;
9ff05123
RK		 2705 LIST_HEAD(garbage_list);
2706
2707 down_write(&nilfs->ns_segctor_sem);
3fd3fe5a
RK		 2708 if (nilfs->ns_writer) {
2709 nilfs_segctor_destroy(nilfs->ns_writer);
2710 nilfs->ns_writer = NULL;
9ff05123
RK		 2711 }
2712
2713 /* Force to free the list of dirty files */
693dd321
RK		 2714 spin_lock(&nilfs->ns_inode_lock);
2715 if (!list_empty(&nilfs->ns_dirty_files)) {
2716 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
f7545144
RK		 2717 nilfs_warning(sb, __func__,
2718 "Hit dirty file after stopped log writer\n");
9ff05123 2719 }
693dd321 2720 spin_unlock(&nilfs->ns_inode_lock);
9ff05123
RK		 2721 up_write(&nilfs->ns_segctor_sem);
2722
693dd321 2723 nilfs_dispose_list(nilfs, &garbage_list, 1);
9ff05123 2724}
Linux 6.x block layer and io_uring tree(s)