blob: 5deeadda9083de848cc745d70bf9ef5429237093 [file] [log] [blame]
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090035#include <linux/slab.h>
Ryusuke Konishi9ff05122009-04-06 19:01:37 -070036#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"
Ryusuke Konishi9ff05122009-04-06 19:01:37 -070043#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,
Ryusuke Konishi9ff05122009-04-06 19:01:37 -070070 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 */
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +090082#define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
83#define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -070084
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 *);
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900107static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700108
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
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700118static 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,
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700129 * it is saved and will be restored on
130 * nilfs_transaction_commit().
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700131 */
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
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700161 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700162 * 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
Ryusuke Konishi7a650042010-03-14 03:32:40 +0900168 * instead; otherwise a new struct is assigned from a slab.
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700169 *
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 *
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700178 * %-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{
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700184 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
Ryusuke Konishi5beb6e02010-09-20 18:19:06 +0900192 vfs_check_frozen(sb, SB_FREEZE_WRITE);
193
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900194 nilfs = sb->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700195 down_read(&nilfs->ns_segctor_sem);
196 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
197 up_read(&nilfs->ns_segctor_sem);
198 ret = -ENOSPC;
199 goto failed;
200 }
201 return 0;
202
203 failed:
204 ti = current->journal_info;
205 current->journal_info = ti->ti_save;
206 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
207 kmem_cache_free(nilfs_transaction_cachep, ti);
208 return ret;
209}
210
211/**
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700212 * nilfs_transaction_commit - commit indivisible file operations.
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700213 * @sb: super block
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700214 *
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700215 * nilfs_transaction_commit() releases the read semaphore which is
216 * acquired by nilfs_transaction_begin(). This is only performed
217 * in outermost call of this function. If a commit flag is set,
218 * nilfs_transaction_commit() sets a timer to start the segment
219 * constructor. If a sync flag is set, it starts construction
220 * directly.
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700221 */
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700222int nilfs_transaction_commit(struct super_block *sb)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700223{
224 struct nilfs_transaction_info *ti = current->journal_info;
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900225 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700226 int err = 0;
227
228 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700229 ti->ti_flags |= NILFS_TI_COMMIT;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700230 if (ti->ti_count > 0) {
231 ti->ti_count--;
232 return 0;
233 }
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +0900234 if (nilfs->ns_writer) {
235 struct nilfs_sc_info *sci = nilfs->ns_writer;
236
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700237 if (ti->ti_flags & NILFS_TI_COMMIT)
238 nilfs_segctor_start_timer(sci);
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +0900239 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700240 nilfs_segctor_do_flush(sci, 0);
241 }
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +0900242 up_read(&nilfs->ns_segctor_sem);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700243 current->journal_info = ti->ti_save;
244
245 if (ti->ti_flags & NILFS_TI_SYNC)
246 err = nilfs_construct_segment(sb);
247 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
248 kmem_cache_free(nilfs_transaction_cachep, ti);
249 return err;
250}
251
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700252void nilfs_transaction_abort(struct super_block *sb)
253{
254 struct nilfs_transaction_info *ti = current->journal_info;
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900255 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700256
257 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
258 if (ti->ti_count > 0) {
259 ti->ti_count--;
260 return;
261 }
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900262 up_read(&nilfs->ns_segctor_sem);
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700263
264 current->journal_info = ti->ti_save;
265 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
266 kmem_cache_free(nilfs_transaction_cachep, ti);
267}
268
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700269void nilfs_relax_pressure_in_lock(struct super_block *sb)
270{
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900271 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +0900272 struct nilfs_sc_info *sci = nilfs->ns_writer;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700273
274 if (!sci || !sci->sc_flush_request)
275 return;
276
277 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
278 up_read(&nilfs->ns_segctor_sem);
279
280 down_write(&nilfs->ns_segctor_sem);
281 if (sci->sc_flush_request &&
282 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
283 struct nilfs_transaction_info *ti = current->journal_info;
284
285 ti->ti_flags |= NILFS_TI_WRITER;
286 nilfs_segctor_do_immediate_flush(sci);
287 ti->ti_flags &= ~NILFS_TI_WRITER;
288 }
289 downgrade_write(&nilfs->ns_segctor_sem);
290}
291
Ryusuke Konishif7545142011-03-09 11:05:08 +0900292static void nilfs_transaction_lock(struct super_block *sb,
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700293 struct nilfs_transaction_info *ti,
294 int gcflag)
295{
296 struct nilfs_transaction_info *cur_ti = current->journal_info;
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900297 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +0900298 struct nilfs_sc_info *sci = nilfs->ns_writer;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700299
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700300 WARN_ON(cur_ti);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700301 ti->ti_flags = NILFS_TI_WRITER;
302 ti->ti_count = 0;
303 ti->ti_save = cur_ti;
304 ti->ti_magic = NILFS_TI_MAGIC;
305 INIT_LIST_HEAD(&ti->ti_garbage);
306 current->journal_info = ti;
307
308 for (;;) {
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +0900309 down_write(&nilfs->ns_segctor_sem);
310 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700311 break;
312
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +0900313 nilfs_segctor_do_immediate_flush(sci);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700314
Ryusuke Konishif7545142011-03-09 11:05:08 +0900315 up_write(&nilfs->ns_segctor_sem);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700316 yield();
317 }
318 if (gcflag)
319 ti->ti_flags |= NILFS_TI_GC;
320}
321
Ryusuke Konishif7545142011-03-09 11:05:08 +0900322static void nilfs_transaction_unlock(struct super_block *sb)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700323{
324 struct nilfs_transaction_info *ti = current->journal_info;
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900325 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700326
327 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
328 BUG_ON(ti->ti_count > 0);
329
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900330 up_write(&nilfs->ns_segctor_sem);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700331 current->journal_info = ti->ti_save;
332 if (!list_empty(&ti->ti_garbage))
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900333 nilfs_dispose_list(nilfs, &ti->ti_garbage, 0);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700334}
335
336static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
337 struct nilfs_segsum_pointer *ssp,
338 unsigned bytes)
339{
340 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
341 unsigned blocksize = sci->sc_super->s_blocksize;
342 void *p;
343
344 if (unlikely(ssp->offset + bytes > blocksize)) {
345 ssp->offset = 0;
346 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
347 &segbuf->sb_segsum_buffers));
348 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
349 }
350 p = ssp->bh->b_data + ssp->offset;
351 ssp->offset += bytes;
352 return p;
353}
354
355/**
356 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
357 * @sci: nilfs_sc_info
358 */
359static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
360{
361 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
362 struct buffer_head *sumbh;
363 unsigned sumbytes;
364 unsigned flags = 0;
365 int err;
366
367 if (nilfs_doing_gc())
368 flags = NILFS_SS_GC;
Ryusuke Konishi6c43f412010-08-20 20:10:38 +0900369 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700370 if (unlikely(err))
371 return err;
372
373 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
374 sumbytes = segbuf->sb_sum.sumbytes;
375 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
376 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
377 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
378 return 0;
379}
380
381static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
382{
383 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
384 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
385 return -E2BIG; /* The current segment is filled up
386 (internal code) */
387 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
388 return nilfs_segctor_reset_segment_buffer(sci);
389}
390
391static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
392{
393 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
394 int err;
395
396 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
397 err = nilfs_segctor_feed_segment(sci);
398 if (err)
399 return err;
400 segbuf = sci->sc_curseg;
401 }
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +0900402 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700403 if (likely(!err))
404 segbuf->sb_sum.flags |= NILFS_SS_SR;
405 return err;
406}
407
408/*
409 * Functions for making segment summary and payloads
410 */
411static int nilfs_segctor_segsum_block_required(
412 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
413 unsigned binfo_size)
414{
415 unsigned blocksize = sci->sc_super->s_blocksize;
416 /* Size of finfo and binfo is enough small against blocksize */
417
418 return ssp->offset + binfo_size +
419 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
420 blocksize;
421}
422
423static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
424 struct inode *inode)
425{
426 sci->sc_curseg->sb_sum.nfinfo++;
427 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
428 nilfs_segctor_map_segsum_entry(
429 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
Ryusuke Konishic96fa462009-04-06 19:01:57 -0700430
Ryusuke Konishi72746ac2011-02-28 13:41:11 +0900431 if (NILFS_I(inode)->i_root &&
432 !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
Ryusuke Konishic96fa462009-04-06 19:01:57 -0700433 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700434 /* skip finfo */
435}
436
437static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
438 struct inode *inode)
439{
440 struct nilfs_finfo *finfo;
441 struct nilfs_inode_info *ii;
442 struct nilfs_segment_buffer *segbuf;
Ryusuke Konishi6c43f412010-08-20 20:10:38 +0900443 __u64 cno;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700444
445 if (sci->sc_blk_cnt == 0)
446 return;
447
448 ii = NILFS_I(inode);
Ryusuke Konishi6c43f412010-08-20 20:10:38 +0900449
450 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
451 cno = ii->i_cno;
452 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
453 cno = 0;
454 else
455 cno = sci->sc_cno;
456
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700457 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
458 sizeof(*finfo));
459 finfo->fi_ino = cpu_to_le64(inode->i_ino);
460 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
461 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
Ryusuke Konishi6c43f412010-08-20 20:10:38 +0900462 finfo->fi_cno = cpu_to_le64(cno);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700463
464 segbuf = sci->sc_curseg;
465 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
466 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
467 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
468 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
469}
470
471static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
472 struct buffer_head *bh,
473 struct inode *inode,
474 unsigned binfo_size)
475{
476 struct nilfs_segment_buffer *segbuf;
477 int required, err = 0;
478
479 retry:
480 segbuf = sci->sc_curseg;
481 required = nilfs_segctor_segsum_block_required(
482 sci, &sci->sc_binfo_ptr, binfo_size);
483 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
484 nilfs_segctor_end_finfo(sci, inode);
485 err = nilfs_segctor_feed_segment(sci);
486 if (err)
487 return err;
488 goto retry;
489 }
490 if (unlikely(required)) {
491 err = nilfs_segbuf_extend_segsum(segbuf);
492 if (unlikely(err))
493 goto failed;
494 }
495 if (sci->sc_blk_cnt == 0)
496 nilfs_segctor_begin_finfo(sci, inode);
497
498 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
499 /* Substitution to vblocknr is delayed until update_blocknr() */
500 nilfs_segbuf_add_file_buffer(segbuf, bh);
501 sci->sc_blk_cnt++;
502 failed:
503 return err;
504}
505
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700506/*
507 * Callback functions that enumerate, mark, and collect dirty blocks
508 */
509static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
510 struct buffer_head *bh, struct inode *inode)
511{
512 int err;
513
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700514 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
Ryusuke Konishie8289492010-11-19 15:26:20 +0900515 if (err < 0)
516 return err;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700517
518 err = nilfs_segctor_add_file_block(sci, bh, inode,
519 sizeof(struct nilfs_binfo_v));
520 if (!err)
521 sci->sc_datablk_cnt++;
522 return err;
523}
524
525static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
526 struct buffer_head *bh,
527 struct inode *inode)
528{
Ryusuke Konishie8289492010-11-19 15:26:20 +0900529 return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700530}
531
532static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
533 struct buffer_head *bh,
534 struct inode *inode)
535{
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700536 WARN_ON(!buffer_dirty(bh));
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700537 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
538}
539
540static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
541 struct nilfs_segsum_pointer *ssp,
542 union nilfs_binfo *binfo)
543{
544 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
545 sci, ssp, sizeof(*binfo_v));
546 *binfo_v = binfo->bi_v;
547}
548
549static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
550 struct nilfs_segsum_pointer *ssp,
551 union nilfs_binfo *binfo)
552{
553 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
554 sci, ssp, sizeof(*vblocknr));
555 *vblocknr = binfo->bi_v.bi_vblocknr;
556}
557
Ryusuke Konishi4e819502010-04-23 17:35:23 +0900558static struct nilfs_sc_operations nilfs_sc_file_ops = {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700559 .collect_data = nilfs_collect_file_data,
560 .collect_node = nilfs_collect_file_node,
561 .collect_bmap = nilfs_collect_file_bmap,
562 .write_data_binfo = nilfs_write_file_data_binfo,
563 .write_node_binfo = nilfs_write_file_node_binfo,
564};
565
566static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
567 struct buffer_head *bh, struct inode *inode)
568{
569 int err;
570
571 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
Ryusuke Konishie8289492010-11-19 15:26:20 +0900572 if (err < 0)
573 return err;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700574
575 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
576 if (!err)
577 sci->sc_datablk_cnt++;
578 return err;
579}
580
581static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
582 struct buffer_head *bh, struct inode *inode)
583{
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700584 WARN_ON(!buffer_dirty(bh));
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700585 return nilfs_segctor_add_file_block(sci, bh, inode,
586 sizeof(struct nilfs_binfo_dat));
587}
588
589static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
590 struct nilfs_segsum_pointer *ssp,
591 union nilfs_binfo *binfo)
592{
593 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
594 sizeof(*blkoff));
595 *blkoff = binfo->bi_dat.bi_blkoff;
596}
597
598static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
599 struct nilfs_segsum_pointer *ssp,
600 union nilfs_binfo *binfo)
601{
602 struct nilfs_binfo_dat *binfo_dat =
603 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
604 *binfo_dat = binfo->bi_dat;
605}
606
Ryusuke Konishi4e819502010-04-23 17:35:23 +0900607static struct nilfs_sc_operations nilfs_sc_dat_ops = {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700608 .collect_data = nilfs_collect_dat_data,
609 .collect_node = nilfs_collect_file_node,
610 .collect_bmap = nilfs_collect_dat_bmap,
611 .write_data_binfo = nilfs_write_dat_data_binfo,
612 .write_node_binfo = nilfs_write_dat_node_binfo,
613};
614
Ryusuke Konishi4e819502010-04-23 17:35:23 +0900615static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700616 .collect_data = nilfs_collect_file_data,
617 .collect_node = NULL,
618 .collect_bmap = NULL,
619 .write_data_binfo = nilfs_write_file_data_binfo,
620 .write_node_binfo = NULL,
621};
622
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700623static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
624 struct list_head *listp,
625 size_t nlimit,
626 loff_t start, loff_t end)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700627{
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700628 struct address_space *mapping = inode->i_mapping;
629 struct pagevec pvec;
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700630 pgoff_t index = 0, last = ULONG_MAX;
631 size_t ndirties = 0;
632 int i;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700633
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700634 if (unlikely(start != 0 || end != LLONG_MAX)) {
635 /*
636 * A valid range is given for sync-ing data pages. The
637 * range is rounded to per-page; extra dirty buffers
638 * may be included if blocksize < pagesize.
639 */
640 index = start >> PAGE_SHIFT;
641 last = end >> PAGE_SHIFT;
642 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700643 pagevec_init(&pvec, 0);
644 repeat:
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700645 if (unlikely(index > last) ||
646 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
647 min_t(pgoff_t, last - index,
648 PAGEVEC_SIZE - 1) + 1))
649 return ndirties;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700650
651 for (i = 0; i < pagevec_count(&pvec); i++) {
652 struct buffer_head *bh, *head;
653 struct page *page = pvec.pages[i];
654
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700655 if (unlikely(page->index > last))
656 break;
657
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700658 if (mapping->host) {
659 lock_page(page);
660 if (!page_has_buffers(page))
661 create_empty_buffers(page,
662 1 << inode->i_blkbits, 0);
663 unlock_page(page);
664 }
665
666 bh = head = page_buffers(page);
667 do {
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700668 if (!buffer_dirty(bh))
669 continue;
670 get_bh(bh);
671 list_add_tail(&bh->b_assoc_buffers, listp);
672 ndirties++;
673 if (unlikely(ndirties >= nlimit)) {
674 pagevec_release(&pvec);
675 cond_resched();
676 return ndirties;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700677 }
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700678 } while (bh = bh->b_this_page, bh != head);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700679 }
680 pagevec_release(&pvec);
681 cond_resched();
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700682 goto repeat;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700683}
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 {
702 if (buffer_dirty(bh)) {
703 get_bh(bh);
704 list_add_tail(&bh->b_assoc_buffers,
705 listp);
706 }
707 bh = bh->b_this_page;
708 } while (bh != head);
709 }
710 pagevec_release(&pvec);
711 cond_resched();
712 }
713}
714
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900715static void nilfs_dispose_list(struct the_nilfs *nilfs,
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700716 struct list_head *head, int force)
717{
718 struct nilfs_inode_info *ii, *n;
719 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
720 unsigned nv = 0;
721
722 while (!list_empty(head)) {
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900723 spin_lock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700724 list_for_each_entry_safe(ii, n, head, i_dirty) {
725 list_del_init(&ii->i_dirty);
726 if (force) {
727 if (unlikely(ii->i_bh)) {
728 brelse(ii->i_bh);
729 ii->i_bh = NULL;
730 }
731 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
732 set_bit(NILFS_I_QUEUED, &ii->i_state);
733 list_add_tail(&ii->i_dirty,
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900734 &nilfs->ns_dirty_files);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700735 continue;
736 }
737 ivec[nv++] = ii;
738 if (nv == SC_N_INODEVEC)
739 break;
740 }
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900741 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700742
743 for (pii = ivec; nv > 0; pii++, nv--)
744 iput(&(*pii)->vfs_inode);
745 }
746}
747
Ryusuke Konishie912a5b2010-08-14 13:07:15 +0900748static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
749 struct nilfs_root *root)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700750{
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700751 int ret = 0;
752
Ryusuke Konishie912a5b2010-08-14 13:07:15 +0900753 if (nilfs_mdt_fetch_dirty(root->ifile))
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700754 ret++;
755 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
756 ret++;
757 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
758 ret++;
Ryusuke Konishi365e2152010-12-27 00:07:30 +0900759 if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
760 ret++;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700761 return ret;
762}
763
764static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
765{
766 return list_empty(&sci->sc_dirty_files) &&
767 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +0900768 sci->sc_nfreesegs == 0 &&
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700769 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
770}
771
772static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
773{
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900774 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700775 int ret = 0;
776
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900777 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700778 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
779
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900780 spin_lock(&nilfs->ns_inode_lock);
781 if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700782 ret++;
783
Ryusuke Konishi693dd322011-03-09 11:05:07 +0900784 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700785 return ret;
786}
787
788static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
789{
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900790 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700791
Ryusuke Konishie912a5b2010-08-14 13:07:15 +0900792 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700793 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
794 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
Ryusuke Konishi365e2152010-12-27 00:07:30 +0900795 nilfs_mdt_clear_dirty(nilfs->ns_dat);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700796}
797
798static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
799{
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900800 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700801 struct buffer_head *bh_cp;
802 struct nilfs_checkpoint *raw_cp;
803 int err;
804
805 /* XXX: this interface will be changed */
806 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
807 &raw_cp, &bh_cp);
808 if (likely(!err)) {
809 /* The following code is duplicated with cpfile. But, it is
810 needed to collect the checkpoint even if it was not newly
811 created */
812 nilfs_mdt_mark_buffer_dirty(bh_cp);
813 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
814 nilfs_cpfile_put_checkpoint(
815 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700816 } else
817 WARN_ON(err == -EINVAL || err == -ENOENT);
818
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700819 return err;
820}
821
822static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
823{
Ryusuke Konishie3154e92011-03-09 11:05:08 +0900824 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700825 struct buffer_head *bh_cp;
826 struct nilfs_checkpoint *raw_cp;
827 int err;
828
829 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
830 &raw_cp, &bh_cp);
831 if (unlikely(err)) {
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700832 WARN_ON(err == -EINVAL || err == -ENOENT);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700833 goto failed_ibh;
834 }
835 raw_cp->cp_snapshot_list.ssl_next = 0;
836 raw_cp->cp_snapshot_list.ssl_prev = 0;
837 raw_cp->cp_inodes_count =
Ryusuke Konishib7c06342010-08-14 14:48:32 +0900838 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700839 raw_cp->cp_blocks_count =
Ryusuke Konishib7c06342010-08-14 14:48:32 +0900840 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700841 raw_cp->cp_nblk_inc =
842 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
843 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
844 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
Ryusuke Konishi458c5b02009-04-06 19:01:56 -0700845
Ryusuke Konishic96fa462009-04-06 19:01:57 -0700846 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
847 nilfs_checkpoint_clear_minor(raw_cp);
848 else
849 nilfs_checkpoint_set_minor(raw_cp);
850
Ryusuke Konishie912a5b2010-08-14 13:07:15 +0900851 nilfs_write_inode_common(sci->sc_root->ifile,
852 &raw_cp->cp_ifile_inode, 1);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700853 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
854 return 0;
855
856 failed_ibh:
857 return err;
858}
859
860static void nilfs_fill_in_file_bmap(struct inode *ifile,
861 struct nilfs_inode_info *ii)
862
863{
864 struct buffer_head *ibh;
865 struct nilfs_inode *raw_inode;
866
867 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
868 ibh = ii->i_bh;
869 BUG_ON(!ibh);
870 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
871 ibh);
872 nilfs_bmap_write(ii->i_bmap, raw_inode);
873 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
874 }
875}
876
Ryusuke Konishie912a5b2010-08-14 13:07:15 +0900877static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700878{
879 struct nilfs_inode_info *ii;
880
881 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
Ryusuke Konishie912a5b2010-08-14 13:07:15 +0900882 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700883 set_bit(NILFS_I_COLLECTED, &ii->i_state);
884 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700885}
886
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700887static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
888 struct the_nilfs *nilfs)
889{
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +0900890 struct buffer_head *bh_sr;
891 struct nilfs_super_root *raw_sr;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700892 unsigned isz = nilfs->ns_inode_size;
893
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +0900894 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
895 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
896
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700897 raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
898 raw_sr->sr_nongc_ctime
899 = cpu_to_le64(nilfs_doing_gc() ?
900 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
901 raw_sr->sr_flags = 0;
902
Ryusuke Konishi365e2152010-12-27 00:07:30 +0900903 nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
Ryusuke Konishi3961f0e2009-11-13 01:55:02 +0900904 NILFS_SR_DAT_OFFSET(isz), 1);
905 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
906 NILFS_SR_CPFILE_OFFSET(isz), 1);
907 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
908 NILFS_SR_SUFILE_OFFSET(isz), 1);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700909}
910
911static void nilfs_redirty_inodes(struct list_head *head)
912{
913 struct nilfs_inode_info *ii;
914
915 list_for_each_entry(ii, head, i_dirty) {
916 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
917 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
918 }
919}
920
921static void nilfs_drop_collected_inodes(struct list_head *head)
922{
923 struct nilfs_inode_info *ii;
924
925 list_for_each_entry(ii, head, i_dirty) {
926 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
927 continue;
928
929 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
930 set_bit(NILFS_I_UPDATED, &ii->i_state);
931 }
932}
933
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700934static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
935 struct inode *inode,
936 struct list_head *listp,
937 int (*collect)(struct nilfs_sc_info *,
938 struct buffer_head *,
939 struct inode *))
940{
941 struct buffer_head *bh, *n;
942 int err = 0;
943
944 if (collect) {
945 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
946 list_del_init(&bh->b_assoc_buffers);
947 err = collect(sci, bh, inode);
948 brelse(bh);
949 if (unlikely(err))
950 goto dispose_buffers;
951 }
952 return 0;
953 }
954
955 dispose_buffers:
956 while (!list_empty(listp)) {
957 bh = list_entry(listp->next, struct buffer_head,
958 b_assoc_buffers);
959 list_del_init(&bh->b_assoc_buffers);
960 brelse(bh);
961 }
962 return err;
963}
964
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700965static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
966{
967 /* Remaining number of blocks within segment buffer */
968 return sci->sc_segbuf_nblocks -
969 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
970}
971
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700972static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
973 struct inode *inode,
974 struct nilfs_sc_operations *sc_ops)
975{
976 LIST_HEAD(data_buffers);
977 LIST_HEAD(node_buffers);
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700978 int err;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700979
980 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700981 size_t n, rest = nilfs_segctor_buffer_rest(sci);
982
983 n = nilfs_lookup_dirty_data_buffers(
984 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
985 if (n > rest) {
986 err = nilfs_segctor_apply_buffers(
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700987 sci, inode, &data_buffers,
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -0700988 sc_ops->collect_data);
989 BUG_ON(!err); /* always receive -E2BIG or true error */
Ryusuke Konishi9ff05122009-04-06 19:01:37 -0700990 goto break_or_fail;
991 }
992 }
993 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
994
995 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
996 err = nilfs_segctor_apply_buffers(
997 sci, inode, &data_buffers, sc_ops->collect_data);
998 if (unlikely(err)) {
999 /* dispose node list */
1000 nilfs_segctor_apply_buffers(
1001 sci, inode, &node_buffers, NULL);
1002 goto break_or_fail;
1003 }
1004 sci->sc_stage.flags |= NILFS_CF_NODE;
1005 }
1006 /* Collect node */
1007 err = nilfs_segctor_apply_buffers(
1008 sci, inode, &node_buffers, sc_ops->collect_node);
1009 if (unlikely(err))
1010 goto break_or_fail;
1011
1012 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1013 err = nilfs_segctor_apply_buffers(
1014 sci, inode, &node_buffers, sc_ops->collect_bmap);
1015 if (unlikely(err))
1016 goto break_or_fail;
1017
1018 nilfs_segctor_end_finfo(sci, inode);
1019 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1020
1021 break_or_fail:
1022 return err;
1023}
1024
1025static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1026 struct inode *inode)
1027{
1028 LIST_HEAD(data_buffers);
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -07001029 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1030 int err;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001031
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -07001032 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1033 sci->sc_dsync_start,
1034 sci->sc_dsync_end);
1035
1036 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1037 nilfs_collect_file_data);
1038 if (!err) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001039 nilfs_segctor_end_finfo(sci, inode);
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -07001040 BUG_ON(n > rest);
1041 /* always receive -E2BIG or true error if n > rest */
1042 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001043 return err;
1044}
1045
1046static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1047{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09001048 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001049 struct list_head *head;
1050 struct nilfs_inode_info *ii;
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +09001051 size_t ndone;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001052 int err = 0;
1053
1054 switch (sci->sc_stage.scnt) {
1055 case NILFS_ST_INIT:
1056 /* Pre-processes */
1057 sci->sc_stage.flags = 0;
1058
1059 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1060 sci->sc_nblk_inc = 0;
1061 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1062 if (mode == SC_LSEG_DSYNC) {
1063 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1064 goto dsync_mode;
1065 }
1066 }
1067
1068 sci->sc_stage.dirty_file_ptr = NULL;
1069 sci->sc_stage.gc_inode_ptr = NULL;
1070 if (mode == SC_FLUSH_DAT) {
1071 sci->sc_stage.scnt = NILFS_ST_DAT;
1072 goto dat_stage;
1073 }
1074 sci->sc_stage.scnt++; /* Fall through */
1075 case NILFS_ST_GC:
1076 if (nilfs_doing_gc()) {
1077 head = &sci->sc_gc_inodes;
1078 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1079 head, i_dirty);
1080 list_for_each_entry_continue(ii, head, i_dirty) {
1081 err = nilfs_segctor_scan_file(
1082 sci, &ii->vfs_inode,
1083 &nilfs_sc_file_ops);
1084 if (unlikely(err)) {
1085 sci->sc_stage.gc_inode_ptr = list_entry(
1086 ii->i_dirty.prev,
1087 struct nilfs_inode_info,
1088 i_dirty);
1089 goto break_or_fail;
1090 }
1091 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1092 }
1093 sci->sc_stage.gc_inode_ptr = NULL;
1094 }
1095 sci->sc_stage.scnt++; /* Fall through */
1096 case NILFS_ST_FILE:
1097 head = &sci->sc_dirty_files;
1098 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1099 i_dirty);
1100 list_for_each_entry_continue(ii, head, i_dirty) {
1101 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1102
1103 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1104 &nilfs_sc_file_ops);
1105 if (unlikely(err)) {
1106 sci->sc_stage.dirty_file_ptr =
1107 list_entry(ii->i_dirty.prev,
1108 struct nilfs_inode_info,
1109 i_dirty);
1110 goto break_or_fail;
1111 }
1112 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1113 /* XXX: required ? */
1114 }
1115 sci->sc_stage.dirty_file_ptr = NULL;
1116 if (mode == SC_FLUSH_FILE) {
1117 sci->sc_stage.scnt = NILFS_ST_DONE;
1118 return 0;
1119 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001120 sci->sc_stage.scnt++;
1121 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1122 /* Fall through */
1123 case NILFS_ST_IFILE:
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09001124 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001125 &nilfs_sc_file_ops);
1126 if (unlikely(err))
1127 break;
1128 sci->sc_stage.scnt++;
1129 /* Creating a checkpoint */
1130 err = nilfs_segctor_create_checkpoint(sci);
1131 if (unlikely(err))
1132 break;
1133 /* Fall through */
1134 case NILFS_ST_CPFILE:
1135 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1136 &nilfs_sc_file_ops);
1137 if (unlikely(err))
1138 break;
1139 sci->sc_stage.scnt++; /* Fall through */
1140 case NILFS_ST_SUFILE:
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +09001141 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1142 sci->sc_nfreesegs, &ndone);
1143 if (unlikely(err)) {
1144 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1145 sci->sc_freesegs, ndone,
1146 NULL);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001147 break;
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +09001148 }
1149 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1150
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001151 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1152 &nilfs_sc_file_ops);
1153 if (unlikely(err))
1154 break;
1155 sci->sc_stage.scnt++; /* Fall through */
1156 case NILFS_ST_DAT:
1157 dat_stage:
Ryusuke Konishi365e2152010-12-27 00:07:30 +09001158 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001159 &nilfs_sc_dat_ops);
1160 if (unlikely(err))
1161 break;
1162 if (mode == SC_FLUSH_DAT) {
1163 sci->sc_stage.scnt = NILFS_ST_DONE;
1164 return 0;
1165 }
1166 sci->sc_stage.scnt++; /* Fall through */
1167 case NILFS_ST_SR:
1168 if (mode == SC_LSEG_SR) {
1169 /* Appending a super root */
1170 err = nilfs_segctor_add_super_root(sci);
1171 if (unlikely(err))
1172 break;
1173 }
1174 /* End of a logical segment */
1175 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1176 sci->sc_stage.scnt = NILFS_ST_DONE;
1177 return 0;
1178 case NILFS_ST_DSYNC:
1179 dsync_mode:
1180 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -07001181 ii = sci->sc_dsync_inode;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001182 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1183 break;
1184
1185 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1186 if (unlikely(err))
1187 break;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001188 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1189 sci->sc_stage.scnt = NILFS_ST_DONE;
1190 return 0;
1191 case NILFS_ST_DONE:
1192 return 0;
1193 default:
1194 BUG();
1195 }
1196
1197 break_or_fail:
1198 return err;
1199}
1200
Ryusuke Konishia6942912009-11-29 23:03:04 +09001201/**
1202 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1203 * @sci: nilfs_sc_info
1204 * @nilfs: nilfs object
1205 */
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001206static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1207 struct the_nilfs *nilfs)
1208{
Ryusuke Konishia6942912009-11-29 23:03:04 +09001209 struct nilfs_segment_buffer *segbuf, *prev;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001210 __u64 nextnum;
Ryusuke Konishia6942912009-11-29 23:03:04 +09001211 int err, alloc = 0;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001212
Ryusuke Konishia6942912009-11-29 23:03:04 +09001213 segbuf = nilfs_segbuf_new(sci->sc_super);
1214 if (unlikely(!segbuf))
1215 return -ENOMEM;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001216
Ryusuke Konishia6942912009-11-29 23:03:04 +09001217 if (list_empty(&sci->sc_write_logs)) {
1218 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1219 nilfs->ns_pseg_offset, nilfs);
1220 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1221 nilfs_shift_to_next_segment(nilfs);
1222 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1223 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001224
Ryusuke Konishia6942912009-11-29 23:03:04 +09001225 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001226 nextnum = nilfs->ns_nextnum;
1227
Ryusuke Konishia6942912009-11-29 23:03:04 +09001228 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1229 /* Start from the head of a new full segment */
1230 alloc++;
1231 } else {
1232 /* Continue logs */
1233 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1234 nilfs_segbuf_map_cont(segbuf, prev);
1235 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1236 nextnum = prev->sb_nextnum;
1237
1238 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1239 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1240 segbuf->sb_sum.seg_seq++;
1241 alloc++;
1242 }
1243 }
1244
1245 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1246 if (err)
1247 goto failed;
1248
1249 if (alloc) {
1250 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1251 if (err)
1252 goto failed;
1253 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001254 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1255
Ryusuke Konishia6942912009-11-29 23:03:04 +09001256 BUG_ON(!list_empty(&sci->sc_segbufs));
1257 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1258 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
Ryusuke Konishicece5522009-04-06 19:01:58 -07001259 return 0;
Ryusuke Konishia6942912009-11-29 23:03:04 +09001260
1261 failed:
1262 nilfs_segbuf_free(segbuf);
1263 return err;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001264}
1265
1266static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1267 struct the_nilfs *nilfs, int nadd)
1268{
Ryusuke Konishie29df392009-11-29 16:51:16 +09001269 struct nilfs_segment_buffer *segbuf, *prev;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001270 struct inode *sufile = nilfs->ns_sufile;
1271 __u64 nextnextnum;
1272 LIST_HEAD(list);
1273 int err, ret, i;
1274
1275 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1276 /*
1277 * Since the segment specified with nextnum might be allocated during
1278 * the previous construction, the buffer including its segusage may
1279 * not be dirty. The following call ensures that the buffer is dirty
1280 * and will pin the buffer on memory until the sufile is written.
1281 */
Ryusuke Konishi61a189e2009-11-18 17:25:12 +09001282 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001283 if (unlikely(err))
1284 return err;
1285
1286 for (i = 0; i < nadd; i++) {
1287 /* extend segment info */
1288 err = -ENOMEM;
1289 segbuf = nilfs_segbuf_new(sci->sc_super);
1290 if (unlikely(!segbuf))
1291 goto failed;
1292
1293 /* map this buffer to region of segment on-disk */
Ryusuke Konishicece5522009-04-06 19:01:58 -07001294 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001295 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1296
1297 /* allocate the next next full segment */
1298 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1299 if (unlikely(err))
1300 goto failed_segbuf;
1301
1302 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1303 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1304
1305 list_add_tail(&segbuf->sb_list, &list);
1306 prev = segbuf;
1307 }
Ryusuke Konishi0935db72009-11-29 02:39:11 +09001308 list_splice_tail(&list, &sci->sc_segbufs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001309 return 0;
1310
1311 failed_segbuf:
1312 nilfs_segbuf_free(segbuf);
1313 failed:
Ryusuke Konishie29df392009-11-29 16:51:16 +09001314 list_for_each_entry(segbuf, &list, sb_list) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001315 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -07001316 WARN_ON(ret); /* never fails */
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001317 }
Ryusuke Konishie29df392009-11-29 16:51:16 +09001318 nilfs_destroy_logs(&list);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001319 return err;
1320}
1321
Ryusuke Konishia6942912009-11-29 23:03:04 +09001322static void nilfs_free_incomplete_logs(struct list_head *logs,
1323 struct the_nilfs *nilfs)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001324{
Ryusuke Konishia6942912009-11-29 23:03:04 +09001325 struct nilfs_segment_buffer *segbuf, *prev;
1326 struct inode *sufile = nilfs->ns_sufile;
Ryusuke Konishi9284ad22009-11-25 01:04:21 +09001327 int ret;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001328
Ryusuke Konishia6942912009-11-29 23:03:04 +09001329 segbuf = NILFS_FIRST_SEGBUF(logs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001330 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
Ryusuke Konishia6942912009-11-29 23:03:04 +09001331 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -07001332 WARN_ON(ret); /* never fails */
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001333 }
Ryusuke Konishi9284ad22009-11-25 01:04:21 +09001334 if (atomic_read(&segbuf->sb_err)) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001335 /* Case 1: The first segment failed */
1336 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1337 /* Case 1a: Partial segment appended into an existing
1338 segment */
1339 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1340 segbuf->sb_fseg_end);
1341 else /* Case 1b: New full segment */
1342 set_nilfs_discontinued(nilfs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001343 }
1344
Ryusuke Konishia6942912009-11-29 23:03:04 +09001345 prev = segbuf;
1346 list_for_each_entry_continue(segbuf, logs, sb_list) {
1347 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1348 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1349 WARN_ON(ret); /* never fails */
1350 }
Ryusuke Konishi9284ad22009-11-25 01:04:21 +09001351 if (atomic_read(&segbuf->sb_err) &&
1352 segbuf->sb_segnum != nilfs->ns_nextnum)
1353 /* Case 2: extended segment (!= next) failed */
Ryusuke Konishia6942912009-11-29 23:03:04 +09001354 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1355 prev = segbuf;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001356 }
1357}
1358
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001359static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1360 struct inode *sufile)
1361{
1362 struct nilfs_segment_buffer *segbuf;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001363 unsigned long live_blocks;
1364 int ret;
1365
1366 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001367 live_blocks = segbuf->sb_sum.nblocks +
1368 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
Ryusuke Konishi071ec542009-11-18 18:23:34 +09001369 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1370 live_blocks,
1371 sci->sc_seg_ctime);
1372 WARN_ON(ret); /* always succeed because the segusage is dirty */
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001373 }
1374}
1375
Ryusuke Konishia6942912009-11-29 23:03:04 +09001376static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001377{
1378 struct nilfs_segment_buffer *segbuf;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001379 int ret;
1380
Ryusuke Konishia6942912009-11-29 23:03:04 +09001381 segbuf = NILFS_FIRST_SEGBUF(logs);
Ryusuke Konishi071ec542009-11-18 18:23:34 +09001382 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1383 segbuf->sb_pseg_start -
1384 segbuf->sb_fseg_start, 0);
1385 WARN_ON(ret); /* always succeed because the segusage is dirty */
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001386
Ryusuke Konishia6942912009-11-29 23:03:04 +09001387 list_for_each_entry_continue(segbuf, logs, sb_list) {
Ryusuke Konishi071ec542009-11-18 18:23:34 +09001388 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1389 0, 0);
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -07001390 WARN_ON(ret); /* always succeed */
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001391 }
1392}
1393
1394static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1395 struct nilfs_segment_buffer *last,
1396 struct inode *sufile)
1397{
Ryusuke Konishie29df392009-11-29 16:51:16 +09001398 struct nilfs_segment_buffer *segbuf = last;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001399 int ret;
1400
Ryusuke Konishie29df392009-11-29 16:51:16 +09001401 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001402 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1403 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -07001404 WARN_ON(ret);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001405 }
Ryusuke Konishie29df392009-11-29 16:51:16 +09001406 nilfs_truncate_logs(&sci->sc_segbufs, last);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001407}
1408
1409
1410static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1411 struct the_nilfs *nilfs, int mode)
1412{
1413 struct nilfs_cstage prev_stage = sci->sc_stage;
1414 int err, nadd = 1;
1415
1416 /* Collection retry loop */
1417 for (;;) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001418 sci->sc_nblk_this_inc = 0;
1419 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1420
1421 err = nilfs_segctor_reset_segment_buffer(sci);
1422 if (unlikely(err))
1423 goto failed;
1424
1425 err = nilfs_segctor_collect_blocks(sci, mode);
1426 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1427 if (!err)
1428 break;
1429
1430 if (unlikely(err != -E2BIG))
1431 goto failed;
1432
1433 /* The current segment is filled up */
1434 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1435 break;
1436
Ryusuke Konishi2d8428a2010-03-22 14:01:24 +09001437 nilfs_clear_logs(&sci->sc_segbufs);
1438
1439 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1440 if (unlikely(err))
1441 return err;
1442
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +09001443 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1444 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1445 sci->sc_freesegs,
1446 sci->sc_nfreesegs,
1447 NULL);
1448 WARN_ON(err); /* do not happen */
1449 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001450 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1451 sci->sc_stage = prev_stage;
1452 }
1453 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1454 return 0;
1455
1456 failed:
1457 return err;
1458}
1459
1460static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1461 struct buffer_head *new_bh)
1462{
1463 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1464
1465 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1466 /* The caller must release old_bh */
1467}
1468
1469static int
1470nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1471 struct nilfs_segment_buffer *segbuf,
1472 int mode)
1473{
1474 struct inode *inode = NULL;
1475 sector_t blocknr;
1476 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1477 unsigned long nblocks = 0, ndatablk = 0;
1478 struct nilfs_sc_operations *sc_op = NULL;
1479 struct nilfs_segsum_pointer ssp;
1480 struct nilfs_finfo *finfo = NULL;
1481 union nilfs_binfo binfo;
1482 struct buffer_head *bh, *bh_org;
1483 ino_t ino = 0;
1484 int err = 0;
1485
1486 if (!nfinfo)
1487 goto out;
1488
1489 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1490 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1491 ssp.offset = sizeof(struct nilfs_segment_summary);
1492
1493 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001494 if (bh == segbuf->sb_super_root)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001495 break;
1496 if (!finfo) {
1497 finfo = nilfs_segctor_map_segsum_entry(
1498 sci, &ssp, sizeof(*finfo));
1499 ino = le64_to_cpu(finfo->fi_ino);
1500 nblocks = le32_to_cpu(finfo->fi_nblocks);
1501 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1502
1503 if (buffer_nilfs_node(bh))
1504 inode = NILFS_BTNC_I(bh->b_page->mapping);
1505 else
1506 inode = NILFS_AS_I(bh->b_page->mapping);
1507
1508 if (mode == SC_LSEG_DSYNC)
1509 sc_op = &nilfs_sc_dsync_ops;
1510 else if (ino == NILFS_DAT_INO)
1511 sc_op = &nilfs_sc_dat_ops;
1512 else /* file blocks */
1513 sc_op = &nilfs_sc_file_ops;
1514 }
1515 bh_org = bh;
1516 get_bh(bh_org);
1517 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1518 &binfo);
1519 if (bh != bh_org)
1520 nilfs_list_replace_buffer(bh_org, bh);
1521 brelse(bh_org);
1522 if (unlikely(err))
1523 goto failed_bmap;
1524
1525 if (ndatablk > 0)
1526 sc_op->write_data_binfo(sci, &ssp, &binfo);
1527 else
1528 sc_op->write_node_binfo(sci, &ssp, &binfo);
1529
1530 blocknr++;
1531 if (--nblocks == 0) {
1532 finfo = NULL;
1533 if (--nfinfo == 0)
1534 break;
1535 } else if (ndatablk > 0)
1536 ndatablk--;
1537 }
1538 out:
1539 return 0;
1540
1541 failed_bmap:
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001542 return err;
1543}
1544
1545static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1546{
1547 struct nilfs_segment_buffer *segbuf;
1548 int err;
1549
1550 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1551 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1552 if (unlikely(err))
1553 return err;
1554 nilfs_segbuf_fill_in_segsum(segbuf);
1555 }
1556 return 0;
1557}
1558
1559static int
1560nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1561{
1562 struct page *clone_page;
1563 struct buffer_head *bh, *head, *bh2;
1564 void *kaddr;
1565
1566 bh = head = page_buffers(page);
1567
1568 clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1569 if (unlikely(!clone_page))
1570 return -ENOMEM;
1571
1572 bh2 = page_buffers(clone_page);
1573 kaddr = kmap_atomic(page, KM_USER0);
1574 do {
1575 if (list_empty(&bh->b_assoc_buffers))
1576 continue;
1577 get_bh(bh2);
1578 page_cache_get(clone_page); /* for each bh */
1579 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1580 bh2->b_blocknr = bh->b_blocknr;
1581 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1582 list_add_tail(&bh->b_assoc_buffers, out);
1583 } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1584 kunmap_atomic(kaddr, KM_USER0);
1585
1586 if (!TestSetPageWriteback(clone_page))
Michael Rubinf629d1c2010-10-26 14:21:33 -07001587 account_page_writeback(clone_page);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001588 unlock_page(clone_page);
1589
1590 return 0;
1591}
1592
1593static int nilfs_test_page_to_be_frozen(struct page *page)
1594{
1595 struct address_space *mapping = page->mapping;
1596
1597 if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1598 return 0;
1599
1600 if (page_mapped(page)) {
1601 ClearPageChecked(page);
1602 return 1;
1603 }
1604 return PageChecked(page);
1605}
1606
1607static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1608{
1609 if (!page || PageWriteback(page))
1610 /* For split b-tree node pages, this function may be called
1611 twice. We ignore the 2nd or later calls by this check. */
1612 return 0;
1613
1614 lock_page(page);
1615 clear_page_dirty_for_io(page);
1616 set_page_writeback(page);
1617 unlock_page(page);
1618
1619 if (nilfs_test_page_to_be_frozen(page)) {
1620 int err = nilfs_copy_replace_page_buffers(page, out);
1621 if (unlikely(err))
1622 return err;
1623 }
1624 return 0;
1625}
1626
1627static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1628 struct page **failed_page)
1629{
1630 struct nilfs_segment_buffer *segbuf;
1631 struct page *bd_page = NULL, *fs_page = NULL;
1632 struct list_head *list = &sci->sc_copied_buffers;
1633 int err;
1634
1635 *failed_page = NULL;
1636 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1637 struct buffer_head *bh;
1638
1639 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1640 b_assoc_buffers) {
1641 if (bh->b_page != bd_page) {
1642 if (bd_page) {
1643 lock_page(bd_page);
1644 clear_page_dirty_for_io(bd_page);
1645 set_page_writeback(bd_page);
1646 unlock_page(bd_page);
1647 }
1648 bd_page = bh->b_page;
1649 }
1650 }
1651
1652 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1653 b_assoc_buffers) {
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001654 if (bh == segbuf->sb_super_root) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001655 if (bh->b_page != bd_page) {
1656 lock_page(bd_page);
1657 clear_page_dirty_for_io(bd_page);
1658 set_page_writeback(bd_page);
1659 unlock_page(bd_page);
1660 bd_page = bh->b_page;
1661 }
1662 break;
1663 }
1664 if (bh->b_page != fs_page) {
1665 err = nilfs_begin_page_io(fs_page, list);
1666 if (unlikely(err)) {
1667 *failed_page = fs_page;
1668 goto out;
1669 }
1670 fs_page = bh->b_page;
1671 }
1672 }
1673 }
1674 if (bd_page) {
1675 lock_page(bd_page);
1676 clear_page_dirty_for_io(bd_page);
1677 set_page_writeback(bd_page);
1678 unlock_page(bd_page);
1679 }
1680 err = nilfs_begin_page_io(fs_page, list);
1681 if (unlikely(err))
1682 *failed_page = fs_page;
1683 out:
1684 return err;
1685}
1686
1687static int nilfs_segctor_write(struct nilfs_sc_info *sci,
Ryusuke Konishi9c965ba2009-11-29 01:17:31 +09001688 struct the_nilfs *nilfs)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001689{
Ryusuke Konishid1c6b722009-12-17 00:55:40 +09001690 int ret;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001691
Ryusuke Konishid1c6b722009-12-17 00:55:40 +09001692 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
Ryusuke Konishia6942912009-11-29 23:03:04 +09001693 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1694 return ret;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001695}
1696
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001697static void __nilfs_end_page_io(struct page *page, int err)
1698{
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001699 if (!err) {
1700 if (!nilfs_page_buffers_clean(page))
1701 __set_page_dirty_nobuffers(page);
1702 ClearPageError(page);
1703 } else {
1704 __set_page_dirty_nobuffers(page);
1705 SetPageError(page);
1706 }
1707
1708 if (buffer_nilfs_allocated(page_buffers(page))) {
1709 if (TestClearPageWriteback(page))
1710 dec_zone_page_state(page, NR_WRITEBACK);
1711 } else
1712 end_page_writeback(page);
1713}
1714
1715static void nilfs_end_page_io(struct page *page, int err)
1716{
1717 if (!page)
1718 return;
1719
Ryusuke Konishia9777842009-07-28 17:55:29 +09001720 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
Ryusuke Konishi8227b292009-06-18 23:52:23 +09001721 /*
1722 * For b-tree node pages, this function may be called twice
1723 * or more because they might be split in a segment.
1724 */
Ryusuke Konishia9777842009-07-28 17:55:29 +09001725 if (PageDirty(page)) {
1726 /*
1727 * For pages holding split b-tree node buffers, dirty
1728 * flag on the buffers may be cleared discretely.
1729 * In that case, the page is once redirtied for
1730 * remaining buffers, and it must be cancelled if
1731 * all the buffers get cleaned later.
1732 */
1733 lock_page(page);
1734 if (nilfs_page_buffers_clean(page))
1735 __nilfs_clear_page_dirty(page);
1736 unlock_page(page);
1737 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001738 return;
Ryusuke Konishia9777842009-07-28 17:55:29 +09001739 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001740
1741 __nilfs_end_page_io(page, err);
1742}
1743
1744static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1745{
1746 struct buffer_head *bh, *head;
1747 struct page *page;
1748
1749 while (!list_empty(list)) {
1750 bh = list_entry(list->next, struct buffer_head,
1751 b_assoc_buffers);
1752 page = bh->b_page;
1753 page_cache_get(page);
1754 head = bh = page_buffers(page);
1755 do {
1756 if (!list_empty(&bh->b_assoc_buffers)) {
1757 list_del_init(&bh->b_assoc_buffers);
1758 if (!err) {
1759 set_buffer_uptodate(bh);
1760 clear_buffer_dirty(bh);
Ryusuke Konishi27e6c7a2010-12-26 16:28:28 +09001761 clear_buffer_delay(bh);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001762 clear_buffer_nilfs_volatile(bh);
1763 }
1764 brelse(bh); /* for b_assoc_buffers */
1765 }
1766 } while ((bh = bh->b_this_page) != head);
1767
1768 __nilfs_end_page_io(page, err);
1769 page_cache_release(page);
1770 }
1771}
1772
Ryusuke Konishia6942912009-11-29 23:03:04 +09001773static void nilfs_abort_logs(struct list_head *logs, struct page *failed_page,
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001774 int err)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001775{
1776 struct nilfs_segment_buffer *segbuf;
1777 struct page *bd_page = NULL, *fs_page = NULL;
Ryusuke Konishia6942912009-11-29 23:03:04 +09001778 struct buffer_head *bh;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001779
Ryusuke Konishia6942912009-11-29 23:03:04 +09001780 if (list_empty(logs))
1781 return;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001782
Ryusuke Konishia6942912009-11-29 23:03:04 +09001783 list_for_each_entry(segbuf, logs, sb_list) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001784 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1785 b_assoc_buffers) {
1786 if (bh->b_page != bd_page) {
1787 if (bd_page)
1788 end_page_writeback(bd_page);
1789 bd_page = bh->b_page;
1790 }
1791 }
1792
1793 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1794 b_assoc_buffers) {
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001795 if (bh == segbuf->sb_super_root) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001796 if (bh->b_page != bd_page) {
1797 end_page_writeback(bd_page);
1798 bd_page = bh->b_page;
1799 }
1800 break;
1801 }
1802 if (bh->b_page != fs_page) {
1803 nilfs_end_page_io(fs_page, err);
Ryusuke Konishi8227b292009-06-18 23:52:23 +09001804 if (fs_page && fs_page == failed_page)
Ryusuke Konishia6942912009-11-29 23:03:04 +09001805 return;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001806 fs_page = bh->b_page;
1807 }
1808 }
1809 }
1810 if (bd_page)
1811 end_page_writeback(bd_page);
1812
1813 nilfs_end_page_io(fs_page, err);
Ryusuke Konishia6942912009-11-29 23:03:04 +09001814}
1815
1816static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1817 struct the_nilfs *nilfs, int err)
1818{
1819 LIST_HEAD(logs);
1820 int ret;
1821
1822 list_splice_tail_init(&sci->sc_write_logs, &logs);
1823 ret = nilfs_wait_on_logs(&logs);
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001824 nilfs_abort_logs(&logs, NULL, ret ? : err);
Ryusuke Konishia6942912009-11-29 23:03:04 +09001825
1826 list_splice_tail_init(&sci->sc_segbufs, &logs);
1827 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1828 nilfs_free_incomplete_logs(&logs, nilfs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001829 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
Ryusuke Konishia6942912009-11-29 23:03:04 +09001830
1831 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1832 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1833 sci->sc_freesegs,
1834 sci->sc_nfreesegs,
1835 NULL);
1836 WARN_ON(ret); /* do not happen */
1837 }
1838
1839 nilfs_destroy_logs(&logs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001840}
1841
1842static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1843 struct nilfs_segment_buffer *segbuf)
1844{
1845 nilfs->ns_segnum = segbuf->sb_segnum;
1846 nilfs->ns_nextnum = segbuf->sb_nextnum;
1847 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1848 + segbuf->sb_sum.nblocks;
1849 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1850 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1851}
1852
1853static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1854{
1855 struct nilfs_segment_buffer *segbuf;
1856 struct page *bd_page = NULL, *fs_page = NULL;
Ryusuke Konishie3154e92011-03-09 11:05:08 +09001857 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001858 int update_sr = false;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001859
Ryusuke Konishia6942912009-11-29 23:03:04 +09001860 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001861 struct buffer_head *bh;
1862
1863 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1864 b_assoc_buffers) {
1865 set_buffer_uptodate(bh);
1866 clear_buffer_dirty(bh);
1867 if (bh->b_page != bd_page) {
1868 if (bd_page)
1869 end_page_writeback(bd_page);
1870 bd_page = bh->b_page;
1871 }
1872 }
1873 /*
1874 * We assume that the buffers which belong to the same page
1875 * continue over the buffer list.
1876 * Under this assumption, the last BHs of pages is
1877 * identifiable by the discontinuity of bh->b_page
1878 * (page != fs_page).
1879 *
1880 * For B-tree node blocks, however, this assumption is not
1881 * guaranteed. The cleanup code of B-tree node pages needs
1882 * special care.
1883 */
1884 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1885 b_assoc_buffers) {
1886 set_buffer_uptodate(bh);
1887 clear_buffer_dirty(bh);
Ryusuke Konishi27e6c7a2010-12-26 16:28:28 +09001888 clear_buffer_delay(bh);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001889 clear_buffer_nilfs_volatile(bh);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +09001890 clear_buffer_nilfs_redirected(bh);
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001891 if (bh == segbuf->sb_super_root) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001892 if (bh->b_page != bd_page) {
1893 end_page_writeback(bd_page);
1894 bd_page = bh->b_page;
1895 }
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09001896 update_sr = true;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001897 break;
1898 }
1899 if (bh->b_page != fs_page) {
1900 nilfs_end_page_io(fs_page, 0);
1901 fs_page = bh->b_page;
1902 }
1903 }
1904
Ryusuke Konishi47620772010-05-23 21:48:36 +09001905 if (!nilfs_segbuf_simplex(segbuf)) {
1906 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001907 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1908 sci->sc_lseg_stime = jiffies;
1909 }
Ryusuke Konishi47620772010-05-23 21:48:36 +09001910 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001911 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1912 }
1913 }
1914 /*
1915 * Since pages may continue over multiple segment buffers,
1916 * end of the last page must be checked outside of the loop.
1917 */
1918 if (bd_page)
1919 end_page_writeback(bd_page);
1920
1921 nilfs_end_page_io(fs_page, 0);
1922
1923 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1924
1925 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1926
Ryusuke Konishic1c1d70922010-08-29 12:44:56 +09001927 if (nilfs_doing_gc())
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001928 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
Ryusuke Konishic1c1d70922010-08-29 12:44:56 +09001929 else
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001930 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001931
1932 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1933
Ryusuke Konishia6942912009-11-29 23:03:04 +09001934 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001935 nilfs_set_next_segment(nilfs, segbuf);
1936
1937 if (update_sr) {
1938 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
Ryusuke Konishie339ad32009-04-06 19:01:59 -07001939 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001940
Ryusuke Konishic96fa462009-04-06 19:01:57 -07001941 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001942 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1943 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
Ryusuke Konishia6942912009-11-29 23:03:04 +09001944 nilfs_segctor_clear_metadata_dirty(sci);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001945 } else
1946 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1947}
1948
Ryusuke Konishia6942912009-11-29 23:03:04 +09001949static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1950{
1951 int ret;
1952
1953 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1954 if (!ret) {
1955 nilfs_segctor_complete_write(sci);
1956 nilfs_destroy_logs(&sci->sc_write_logs);
1957 }
1958 return ret;
1959}
1960
Ryusuke Konishi693dd322011-03-09 11:05:07 +09001961static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1962 struct the_nilfs *nilfs)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001963{
1964 struct nilfs_inode_info *ii, *n;
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09001965 struct inode *ifile = sci->sc_root->ifile;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001966
Ryusuke Konishi693dd322011-03-09 11:05:07 +09001967 spin_lock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001968 retry:
Ryusuke Konishi693dd322011-03-09 11:05:07 +09001969 list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001970 if (!ii->i_bh) {
1971 struct buffer_head *ibh;
1972 int err;
1973
Ryusuke Konishi693dd322011-03-09 11:05:07 +09001974 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001975 err = nilfs_ifile_get_inode_block(
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09001976 ifile, ii->vfs_inode.i_ino, &ibh);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001977 if (unlikely(err)) {
Ryusuke Konishi693dd322011-03-09 11:05:07 +09001978 nilfs_warning(sci->sc_super, __func__,
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001979 "failed to get inode block.\n");
1980 return err;
1981 }
1982 nilfs_mdt_mark_buffer_dirty(ibh);
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09001983 nilfs_mdt_mark_dirty(ifile);
Ryusuke Konishi693dd322011-03-09 11:05:07 +09001984 spin_lock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001985 if (likely(!ii->i_bh))
1986 ii->i_bh = ibh;
1987 else
1988 brelse(ibh);
1989 goto retry;
1990 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001991
1992 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1993 set_bit(NILFS_I_BUSY, &ii->i_state);
Nicolas Kaisereaae0f32011-03-19 16:45:30 +01001994 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001995 }
Ryusuke Konishi693dd322011-03-09 11:05:07 +09001996 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001997
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07001998 return 0;
1999}
2000
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002001static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2002 struct the_nilfs *nilfs)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002003{
2004 struct nilfs_transaction_info *ti = current->journal_info;
2005 struct nilfs_inode_info *ii, *n;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002006
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002007 spin_lock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002008 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2009 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
Ryusuke Konishi6c43f412010-08-20 20:10:38 +09002010 test_bit(NILFS_I_DIRTY, &ii->i_state))
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002011 continue;
Ryusuke Konishi6c43f412010-08-20 20:10:38 +09002012
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002013 clear_bit(NILFS_I_BUSY, &ii->i_state);
2014 brelse(ii->i_bh);
2015 ii->i_bh = NULL;
Nicolas Kaisereaae0f32011-03-19 16:45:30 +01002016 list_move_tail(&ii->i_dirty, &ti->ti_garbage);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002017 }
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002018 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002019}
2020
2021/*
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002022 * Main procedure of segment constructor
2023 */
2024static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2025{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002026 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002027 struct page *failed_page;
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09002028 int err;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002029
2030 sci->sc_stage.scnt = NILFS_ST_INIT;
Ryusuke Konishi6c43f412010-08-20 20:10:38 +09002031 sci->sc_cno = nilfs->ns_cno;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002032
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002033 err = nilfs_segctor_collect_dirty_files(sci, nilfs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002034 if (unlikely(err))
2035 goto out;
2036
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09002037 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002038 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2039
2040 if (nilfs_segctor_clean(sci))
2041 goto out;
2042
2043 do {
2044 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2045
2046 err = nilfs_segctor_begin_construction(sci, nilfs);
2047 if (unlikely(err))
2048 goto out;
2049
2050 /* Update time stamp */
2051 sci->sc_seg_ctime = get_seconds();
2052
2053 err = nilfs_segctor_collect(sci, nilfs, mode);
2054 if (unlikely(err))
2055 goto failed;
2056
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002057 /* Avoid empty segment */
2058 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
Ryusuke Konishi47620772010-05-23 21:48:36 +09002059 nilfs_segbuf_empty(sci->sc_curseg)) {
Ryusuke Konishia6942912009-11-29 23:03:04 +09002060 nilfs_segctor_abort_construction(sci, nilfs, 1);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002061 goto out;
2062 }
2063
2064 err = nilfs_segctor_assign(sci, mode);
2065 if (unlikely(err))
2066 goto failed;
2067
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002068 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09002069 nilfs_segctor_fill_in_file_bmap(sci);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002070
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09002071 if (mode == SC_LSEG_SR &&
2072 sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002073 err = nilfs_segctor_fill_in_checkpoint(sci);
2074 if (unlikely(err))
Ryusuke Konishia6942912009-11-29 23:03:04 +09002075 goto failed_to_write;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002076
2077 nilfs_segctor_fill_in_super_root(sci, nilfs);
2078 }
2079 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2080
2081 /* Write partial segments */
2082 err = nilfs_segctor_prepare_write(sci, &failed_page);
Ryusuke Konishia6942912009-11-29 23:03:04 +09002083 if (err) {
Ryusuke Konishi1e2b68b2010-03-23 01:15:31 +09002084 nilfs_abort_logs(&sci->sc_segbufs, failed_page, err);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002085 goto failed_to_write;
Ryusuke Konishia6942912009-11-29 23:03:04 +09002086 }
Ryusuke Konishiaaed1d52010-03-23 01:50:38 +09002087
2088 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2089 nilfs->ns_crc_seed);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002090
Ryusuke Konishi9c965ba2009-11-29 01:17:31 +09002091 err = nilfs_segctor_write(sci, nilfs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002092 if (unlikely(err))
2093 goto failed_to_write;
2094
Ryusuke Konishia6942912009-11-29 23:03:04 +09002095 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2096 nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2097 /*
2098 * At this point, we avoid double buffering
2099 * for blocksize < pagesize because page dirty
2100 * flag is turned off during write and dirty
2101 * buffers are not properly collected for
2102 * pages crossing over segments.
2103 */
2104 err = nilfs_segctor_wait(sci);
2105 if (err)
2106 goto failed_to_write;
2107 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002108 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2109
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002110 out:
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002111 nilfs_segctor_drop_written_files(sci, nilfs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002112 return err;
2113
2114 failed_to_write:
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002115 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2116 nilfs_redirty_inodes(&sci->sc_dirty_files);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002117
2118 failed:
2119 if (nilfs_doing_gc())
2120 nilfs_redirty_inodes(&sci->sc_gc_inodes);
Ryusuke Konishia6942912009-11-29 23:03:04 +09002121 nilfs_segctor_abort_construction(sci, nilfs, err);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002122 goto out;
2123}
2124
2125/**
Ryusuke Konishi9ccf56c2010-03-14 03:01:03 +09002126 * nilfs_segctor_start_timer - set timer of background write
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002127 * @sci: nilfs_sc_info
2128 *
2129 * If the timer has already been set, it ignores the new request.
2130 * This function MUST be called within a section locking the segment
2131 * semaphore.
2132 */
2133static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2134{
2135 spin_lock(&sci->sc_state_lock);
Li Hongfdce8952010-04-10 23:25:39 +08002136 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2137 sci->sc_timer.expires = jiffies + sci->sc_interval;
2138 add_timer(&sci->sc_timer);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002139 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2140 }
2141 spin_unlock(&sci->sc_state_lock);
2142}
2143
2144static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2145{
2146 spin_lock(&sci->sc_state_lock);
2147 if (!(sci->sc_flush_request & (1 << bn))) {
2148 unsigned long prev_req = sci->sc_flush_request;
2149
2150 sci->sc_flush_request |= (1 << bn);
2151 if (!prev_req)
2152 wake_up(&sci->sc_wait_daemon);
2153 }
2154 spin_unlock(&sci->sc_state_lock);
2155}
2156
2157/**
2158 * nilfs_flush_segment - trigger a segment construction for resource control
2159 * @sb: super block
2160 * @ino: inode number of the file to be flushed out.
2161 */
2162void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2163{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002164 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002165 struct nilfs_sc_info *sci = nilfs->ns_writer;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002166
2167 if (!sci || nilfs_doing_construction())
2168 return;
2169 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2170 /* assign bit 0 to data files */
2171}
2172
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002173struct nilfs_segctor_wait_request {
2174 wait_queue_t wq;
2175 __u32 seq;
2176 int err;
2177 atomic_t done;
2178};
2179
2180static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2181{
2182 struct nilfs_segctor_wait_request wait_req;
2183 int err = 0;
2184
2185 spin_lock(&sci->sc_state_lock);
2186 init_wait(&wait_req.wq);
2187 wait_req.err = 0;
2188 atomic_set(&wait_req.done, 0);
2189 wait_req.seq = ++sci->sc_seq_request;
2190 spin_unlock(&sci->sc_state_lock);
2191
2192 init_waitqueue_entry(&wait_req.wq, current);
2193 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2194 set_current_state(TASK_INTERRUPTIBLE);
2195 wake_up(&sci->sc_wait_daemon);
2196
2197 for (;;) {
2198 if (atomic_read(&wait_req.done)) {
2199 err = wait_req.err;
2200 break;
2201 }
2202 if (!signal_pending(current)) {
2203 schedule();
2204 continue;
2205 }
2206 err = -ERESTARTSYS;
2207 break;
2208 }
2209 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2210 return err;
2211}
2212
2213static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2214{
2215 struct nilfs_segctor_wait_request *wrq, *n;
2216 unsigned long flags;
2217
2218 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2219 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2220 wq.task_list) {
2221 if (!atomic_read(&wrq->done) &&
2222 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2223 wrq->err = err;
2224 atomic_set(&wrq->done, 1);
2225 }
2226 if (atomic_read(&wrq->done)) {
2227 wrq->wq.func(&wrq->wq,
2228 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2229 0, NULL);
2230 }
2231 }
2232 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2233}
2234
2235/**
2236 * nilfs_construct_segment - construct a logical segment
2237 * @sb: super block
2238 *
2239 * Return Value: On success, 0 is retured. On errors, one of the following
2240 * negative error code is returned.
2241 *
2242 * %-EROFS - Read only filesystem.
2243 *
2244 * %-EIO - I/O error
2245 *
2246 * %-ENOSPC - No space left on device (only in a panic state).
2247 *
2248 * %-ERESTARTSYS - Interrupted.
2249 *
2250 * %-ENOMEM - Insufficient memory available.
2251 */
2252int nilfs_construct_segment(struct super_block *sb)
2253{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002254 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002255 struct nilfs_sc_info *sci = nilfs->ns_writer;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002256 struct nilfs_transaction_info *ti;
2257 int err;
2258
2259 if (!sci)
2260 return -EROFS;
2261
2262 /* A call inside transactions causes a deadlock. */
2263 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2264
2265 err = nilfs_segctor_sync(sci);
2266 return err;
2267}
2268
2269/**
2270 * nilfs_construct_dsync_segment - construct a data-only logical segment
2271 * @sb: super block
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -07002272 * @inode: inode whose data blocks should be written out
2273 * @start: start byte offset
2274 * @end: end byte offset (inclusive)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002275 *
2276 * Return Value: On success, 0 is retured. On errors, one of the following
2277 * negative error code is returned.
2278 *
2279 * %-EROFS - Read only filesystem.
2280 *
2281 * %-EIO - I/O error
2282 *
2283 * %-ENOSPC - No space left on device (only in a panic state).
2284 *
2285 * %-ERESTARTSYS - Interrupted.
2286 *
2287 * %-ENOMEM - Insufficient memory available.
2288 */
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -07002289int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2290 loff_t start, loff_t end)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002291{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002292 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002293 struct nilfs_sc_info *sci = nilfs->ns_writer;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002294 struct nilfs_inode_info *ii;
2295 struct nilfs_transaction_info ti;
2296 int err = 0;
2297
2298 if (!sci)
2299 return -EROFS;
2300
Ryusuke Konishif7545142011-03-09 11:05:08 +09002301 nilfs_transaction_lock(sb, &ti, 0);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002302
2303 ii = NILFS_I(inode);
2304 if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
Ryusuke Konishi3b2ce582011-03-09 11:05:07 +09002305 nilfs_test_opt(nilfs, STRICT_ORDER) ||
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002306 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
Ryusuke Konishi3b2ce582011-03-09 11:05:07 +09002307 nilfs_discontinued(nilfs)) {
Ryusuke Konishif7545142011-03-09 11:05:08 +09002308 nilfs_transaction_unlock(sb);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002309 err = nilfs_segctor_sync(sci);
2310 return err;
2311 }
2312
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002313 spin_lock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002314 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2315 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002316 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishif7545142011-03-09 11:05:08 +09002317 nilfs_transaction_unlock(sb);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002318 return 0;
2319 }
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002320 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishif30bf3e2009-04-06 19:01:38 -07002321 sci->sc_dsync_inode = ii;
2322 sci->sc_dsync_start = start;
2323 sci->sc_dsync_end = end;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002324
2325 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2326
Ryusuke Konishif7545142011-03-09 11:05:08 +09002327 nilfs_transaction_unlock(sb);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002328 return err;
2329}
2330
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002331#define FLUSH_FILE_BIT (0x1) /* data file only */
2332#define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2333
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002334/**
2335 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2336 * @sci: segment constructor object
2337 */
2338static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002339{
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002340 spin_lock(&sci->sc_state_lock);
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002341 sci->sc_seq_accepted = sci->sc_seq_request;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002342 spin_unlock(&sci->sc_state_lock);
Li Hongfdce8952010-04-10 23:25:39 +08002343 del_timer_sync(&sci->sc_timer);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002344}
2345
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002346/**
2347 * nilfs_segctor_notify - notify the result of request to caller threads
2348 * @sci: segment constructor object
2349 * @mode: mode of log forming
2350 * @err: error code to be notified
2351 */
2352static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002353{
2354 /* Clear requests (even when the construction failed) */
2355 spin_lock(&sci->sc_state_lock);
2356
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002357 if (mode == SC_LSEG_SR) {
Ryusuke Konishiaeda7f62009-11-02 15:08:13 +09002358 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002359 sci->sc_seq_done = sci->sc_seq_accepted;
2360 nilfs_segctor_wakeup(sci, err);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002361 sci->sc_flush_request = 0;
Ryusuke Konishiaeda7f62009-11-02 15:08:13 +09002362 } else {
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002363 if (mode == SC_FLUSH_FILE)
Ryusuke Konishiaeda7f62009-11-02 15:08:13 +09002364 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002365 else if (mode == SC_FLUSH_DAT)
Ryusuke Konishiaeda7f62009-11-02 15:08:13 +09002366 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002367
Ryusuke Konishiaeda7f62009-11-02 15:08:13 +09002368 /* re-enable timer if checkpoint creation was not done */
Li Hongfdce8952010-04-10 23:25:39 +08002369 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2370 time_before(jiffies, sci->sc_timer.expires))
2371 add_timer(&sci->sc_timer);
Ryusuke Konishiaeda7f62009-11-02 15:08:13 +09002372 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002373 spin_unlock(&sci->sc_state_lock);
2374}
2375
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002376/**
2377 * nilfs_segctor_construct - form logs and write them to disk
2378 * @sci: segment constructor object
2379 * @mode: mode of log forming
2380 */
2381static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002382{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002383 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Jiro SEKIBAd26493b2010-06-28 17:49:32 +09002384 struct nilfs_super_block **sbp;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002385 int err = 0;
2386
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002387 nilfs_segctor_accept(sci);
2388
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002389 if (nilfs_discontinued(nilfs))
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002390 mode = SC_LSEG_SR;
2391 if (!nilfs_segctor_confirm(sci))
2392 err = nilfs_segctor_do_construct(sci, mode);
2393
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002394 if (likely(!err)) {
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002395 if (mode != SC_FLUSH_DAT)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002396 atomic_set(&nilfs->ns_ndirtyblks, 0);
2397 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2398 nilfs_discontinued(nilfs)) {
2399 down_write(&nilfs->ns_sem);
Jiro SEKIBAd26493b2010-06-28 17:49:32 +09002400 err = -EIO;
Ryusuke Konishif7545142011-03-09 11:05:08 +09002401 sbp = nilfs_prepare_super(sci->sc_super,
Jiro SEKIBAb2ac86e2010-06-28 17:49:33 +09002402 nilfs_sb_will_flip(nilfs));
2403 if (likely(sbp)) {
2404 nilfs_set_log_cursor(sbp[0], nilfs);
Ryusuke Konishif7545142011-03-09 11:05:08 +09002405 err = nilfs_commit_super(sci->sc_super,
2406 NILFS_SB_COMMIT);
Jiro SEKIBAb2ac86e2010-06-28 17:49:33 +09002407 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002408 up_write(&nilfs->ns_sem);
2409 }
2410 }
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002411
2412 nilfs_segctor_notify(sci, mode, err);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002413 return err;
2414}
2415
2416static void nilfs_construction_timeout(unsigned long data)
2417{
2418 struct task_struct *p = (struct task_struct *)data;
2419 wake_up_process(p);
2420}
2421
2422static void
2423nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2424{
2425 struct nilfs_inode_info *ii, *n;
2426
2427 list_for_each_entry_safe(ii, n, head, i_dirty) {
2428 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2429 continue;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002430 list_del_init(&ii->i_dirty);
Ryusuke Konishi263d90c2010-08-20 19:06:11 +09002431 iput(&ii->vfs_inode);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002432 }
2433}
2434
Ryusuke Konishi4f6b8282009-05-10 22:41:43 +09002435int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2436 void **kbufs)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002437{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002438 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002439 struct nilfs_sc_info *sci = nilfs->ns_writer;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002440 struct nilfs_transaction_info ti;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002441 int err;
2442
2443 if (unlikely(!sci))
2444 return -EROFS;
2445
Ryusuke Konishif7545142011-03-09 11:05:08 +09002446 nilfs_transaction_lock(sb, &ti, 1);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002447
Ryusuke Konishic1c1d70922010-08-29 12:44:56 +09002448 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002449 if (unlikely(err))
2450 goto out_unlock;
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +09002451
Ryusuke Konishi4f6b8282009-05-10 22:41:43 +09002452 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
Ryusuke Konishic1c1d70922010-08-29 12:44:56 +09002453 if (unlikely(err)) {
2454 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002455 goto out_unlock;
Ryusuke Konishic1c1d70922010-08-29 12:44:56 +09002456 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002457
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +09002458 sci->sc_freesegs = kbufs[4];
2459 sci->sc_nfreesegs = argv[4].v_nmembs;
Ryusuke Konishi0935db72009-11-29 02:39:11 +09002460 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002461
2462 for (;;) {
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002463 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002464 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002465
2466 if (likely(!err))
2467 break;
2468
2469 nilfs_warning(sb, __func__,
2470 "segment construction failed. (err=%d)", err);
2471 set_current_state(TASK_INTERRUPTIBLE);
2472 schedule_timeout(sci->sc_interval);
2473 }
Ryusuke Konishi3b2ce582011-03-09 11:05:07 +09002474 if (nilfs_test_opt(nilfs, DISCARD)) {
Jiro SEKIBAe902ec92010-01-30 18:06:35 +09002475 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2476 sci->sc_nfreesegs);
2477 if (ret) {
2478 printk(KERN_WARNING
2479 "NILFS warning: error %d on discard request, "
2480 "turning discards off for the device\n", ret);
Ryusuke Konishi3b2ce582011-03-09 11:05:07 +09002481 nilfs_clear_opt(nilfs, DISCARD);
Jiro SEKIBAe902ec92010-01-30 18:06:35 +09002482 }
2483 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002484
2485 out_unlock:
Ryusuke Konishi071cb4b2009-05-16 23:44:55 +09002486 sci->sc_freesegs = NULL;
2487 sci->sc_nfreesegs = 0;
Ryusuke Konishic1c1d70922010-08-29 12:44:56 +09002488 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
Ryusuke Konishif7545142011-03-09 11:05:08 +09002489 nilfs_transaction_unlock(sb);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002490 return err;
2491}
2492
2493static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2494{
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002495 struct nilfs_transaction_info ti;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002496
Ryusuke Konishif7545142011-03-09 11:05:08 +09002497 nilfs_transaction_lock(sci->sc_super, &ti, 0);
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002498 nilfs_segctor_construct(sci, mode);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002499
2500 /*
2501 * Unclosed segment should be retried. We do this using sc_timer.
2502 * Timeout of sc_timer will invoke complete construction which leads
2503 * to close the current logical segment.
2504 */
2505 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2506 nilfs_segctor_start_timer(sci);
2507
Ryusuke Konishif7545142011-03-09 11:05:08 +09002508 nilfs_transaction_unlock(sci->sc_super);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002509}
2510
2511static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2512{
2513 int mode = 0;
2514 int err;
2515
2516 spin_lock(&sci->sc_state_lock);
2517 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2518 SC_FLUSH_DAT : SC_FLUSH_FILE;
2519 spin_unlock(&sci->sc_state_lock);
2520
2521 if (mode) {
2522 err = nilfs_segctor_do_construct(sci, mode);
2523
2524 spin_lock(&sci->sc_state_lock);
2525 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2526 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2527 spin_unlock(&sci->sc_state_lock);
2528 }
2529 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2530}
2531
2532static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2533{
2534 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2535 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2536 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2537 return SC_FLUSH_FILE;
2538 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2539 return SC_FLUSH_DAT;
2540 }
2541 return SC_LSEG_SR;
2542}
2543
2544/**
2545 * nilfs_segctor_thread - main loop of the segment constructor thread.
2546 * @arg: pointer to a struct nilfs_sc_info.
2547 *
2548 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2549 * to execute segment constructions.
2550 */
2551static int nilfs_segctor_thread(void *arg)
2552{
2553 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002554 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002555 int timeout = 0;
2556
Li Hongfdce8952010-04-10 23:25:39 +08002557 sci->sc_timer.data = (unsigned long)current;
2558 sci->sc_timer.function = nilfs_construction_timeout;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002559
2560 /* start sync. */
2561 sci->sc_task = current;
2562 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2563 printk(KERN_INFO
2564 "segctord starting. Construction interval = %lu seconds, "
2565 "CP frequency < %lu seconds\n",
2566 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2567
2568 spin_lock(&sci->sc_state_lock);
2569 loop:
2570 for (;;) {
2571 int mode;
2572
2573 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2574 goto end_thread;
2575
2576 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2577 mode = SC_LSEG_SR;
2578 else if (!sci->sc_flush_request)
2579 break;
2580 else
2581 mode = nilfs_segctor_flush_mode(sci);
2582
2583 spin_unlock(&sci->sc_state_lock);
2584 nilfs_segctor_thread_construct(sci, mode);
2585 spin_lock(&sci->sc_state_lock);
2586 timeout = 0;
2587 }
2588
2589
2590 if (freezing(current)) {
2591 spin_unlock(&sci->sc_state_lock);
2592 refrigerator();
2593 spin_lock(&sci->sc_state_lock);
2594 } else {
2595 DEFINE_WAIT(wait);
2596 int should_sleep = 1;
2597
2598 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2599 TASK_INTERRUPTIBLE);
2600
2601 if (sci->sc_seq_request != sci->sc_seq_done)
2602 should_sleep = 0;
2603 else if (sci->sc_flush_request)
2604 should_sleep = 0;
2605 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2606 should_sleep = time_before(jiffies,
Li Hongfdce8952010-04-10 23:25:39 +08002607 sci->sc_timer.expires);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002608
2609 if (should_sleep) {
2610 spin_unlock(&sci->sc_state_lock);
2611 schedule();
2612 spin_lock(&sci->sc_state_lock);
2613 }
2614 finish_wait(&sci->sc_wait_daemon, &wait);
2615 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
Li Hongfdce8952010-04-10 23:25:39 +08002616 time_after_eq(jiffies, sci->sc_timer.expires));
Ryusuke Konishie605f0a2009-12-09 00:57:52 +09002617
2618 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
Jiro SEKIBA1dfa2712009-07-23 01:33:49 +09002619 set_nilfs_discontinued(nilfs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002620 }
2621 goto loop;
2622
2623 end_thread:
2624 spin_unlock(&sci->sc_state_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002625
2626 /* end sync. */
2627 sci->sc_task = NULL;
2628 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2629 return 0;
2630}
2631
2632static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2633{
2634 struct task_struct *t;
2635
2636 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2637 if (IS_ERR(t)) {
2638 int err = PTR_ERR(t);
2639
2640 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2641 err);
2642 return err;
2643 }
2644 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2645 return 0;
2646}
2647
2648static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
Jiro SEKIBA6b81e142010-10-14 13:52:00 +09002649 __acquires(&sci->sc_state_lock)
2650 __releases(&sci->sc_state_lock)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002651{
2652 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2653
2654 while (sci->sc_task) {
2655 wake_up(&sci->sc_wait_daemon);
2656 spin_unlock(&sci->sc_state_lock);
2657 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2658 spin_lock(&sci->sc_state_lock);
2659 }
2660}
2661
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002662/*
2663 * Setup & clean-up functions
2664 */
Ryusuke Konishif7545142011-03-09 11:05:08 +09002665static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09002666 struct nilfs_root *root)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002667{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002668 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002669 struct nilfs_sc_info *sci;
2670
2671 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2672 if (!sci)
2673 return NULL;
2674
Ryusuke Konishif7545142011-03-09 11:05:08 +09002675 sci->sc_super = sb;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002676
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09002677 nilfs_get_root(root);
2678 sci->sc_root = root;
2679
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002680 init_waitqueue_head(&sci->sc_wait_request);
2681 init_waitqueue_head(&sci->sc_wait_daemon);
2682 init_waitqueue_head(&sci->sc_wait_task);
2683 spin_lock_init(&sci->sc_state_lock);
2684 INIT_LIST_HEAD(&sci->sc_dirty_files);
2685 INIT_LIST_HEAD(&sci->sc_segbufs);
Ryusuke Konishia6942912009-11-29 23:03:04 +09002686 INIT_LIST_HEAD(&sci->sc_write_logs);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002687 INIT_LIST_HEAD(&sci->sc_gc_inodes);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002688 INIT_LIST_HEAD(&sci->sc_copied_buffers);
Li Hongfdce8952010-04-10 23:25:39 +08002689 init_timer(&sci->sc_timer);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002690
2691 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2692 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2693 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2694
Ryusuke Konishi574e6c32011-03-09 11:05:07 +09002695 if (nilfs->ns_interval)
2696 sci->sc_interval = nilfs->ns_interval;
2697 if (nilfs->ns_watermark)
2698 sci->sc_watermark = nilfs->ns_watermark;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002699 return sci;
2700}
2701
2702static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2703{
2704 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2705
2706 /* The segctord thread was stopped and its timer was removed.
2707 But some tasks remain. */
2708 do {
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002709 struct nilfs_transaction_info ti;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002710
Ryusuke Konishif7545142011-03-09 11:05:08 +09002711 nilfs_transaction_lock(sci->sc_super, &ti, 0);
Ryusuke Konishidcd76182010-01-26 15:20:15 +09002712 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
Ryusuke Konishif7545142011-03-09 11:05:08 +09002713 nilfs_transaction_unlock(sci->sc_super);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002714
2715 } while (ret && retrycount-- > 0);
2716}
2717
2718/**
2719 * nilfs_segctor_destroy - destroy the segment constructor.
2720 * @sci: nilfs_sc_info
2721 *
2722 * nilfs_segctor_destroy() kills the segctord thread and frees
2723 * the nilfs_sc_info struct.
2724 * Caller must hold the segment semaphore.
2725 */
2726static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2727{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002728 struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002729 int flag;
2730
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002731 up_write(&nilfs->ns_segctor_sem);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002732
2733 spin_lock(&sci->sc_state_lock);
2734 nilfs_segctor_kill_thread(sci);
2735 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2736 || sci->sc_seq_request != sci->sc_seq_done);
2737 spin_unlock(&sci->sc_state_lock);
2738
Ryusuke Konishi3256a052010-01-31 12:39:50 +09002739 if (flag || !nilfs_segctor_confirm(sci))
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002740 nilfs_segctor_write_out(sci);
2741
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -07002742 WARN_ON(!list_empty(&sci->sc_copied_buffers));
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002743
2744 if (!list_empty(&sci->sc_dirty_files)) {
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002745 nilfs_warning(sci->sc_super, __func__,
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002746 "dirty file(s) after the final construction\n");
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002747 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002748 }
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002749
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -07002750 WARN_ON(!list_empty(&sci->sc_segbufs));
Ryusuke Konishia6942912009-11-29 23:03:04 +09002751 WARN_ON(!list_empty(&sci->sc_write_logs));
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002752
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09002753 nilfs_put_root(sci->sc_root);
2754
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002755 down_write(&nilfs->ns_segctor_sem);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002756
Li Hongfdce8952010-04-10 23:25:39 +08002757 del_timer_sync(&sci->sc_timer);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002758 kfree(sci);
2759}
2760
2761/**
Ryusuke Konishif7545142011-03-09 11:05:08 +09002762 * nilfs_attach_log_writer - attach log writer
2763 * @sb: super block instance
Ryusuke Konishie912a5b2010-08-14 13:07:15 +09002764 * @root: root object of the current filesystem tree
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002765 *
Ryusuke Konishif7545142011-03-09 11:05:08 +09002766 * This allocates a log writer object, initializes it, and starts the
2767 * log writer.
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002768 *
2769 * Return Value: On success, 0 is returned. On error, one of the following
2770 * negative error code is returned.
2771 *
2772 * %-ENOMEM - Insufficient memory available.
2773 */
Ryusuke Konishif7545142011-03-09 11:05:08 +09002774int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002775{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002776 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002777 int err;
2778
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002779 if (nilfs->ns_writer) {
Ryusuke Konishife5f1712010-01-31 19:46:40 +09002780 /*
2781 * This happens if the filesystem was remounted
2782 * read/write after nilfs_error degenerated it into a
2783 * read-only mount.
2784 */
Ryusuke Konishif7545142011-03-09 11:05:08 +09002785 nilfs_detach_log_writer(sb);
Ryusuke Konishife5f1712010-01-31 19:46:40 +09002786 }
2787
Ryusuke Konishif7545142011-03-09 11:05:08 +09002788 nilfs->ns_writer = nilfs_segctor_new(sb, root);
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002789 if (!nilfs->ns_writer)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002790 return -ENOMEM;
2791
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002792 err = nilfs_segctor_start_thread(nilfs->ns_writer);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002793 if (err) {
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002794 kfree(nilfs->ns_writer);
2795 nilfs->ns_writer = NULL;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002796 }
2797 return err;
2798}
2799
2800/**
Ryusuke Konishif7545142011-03-09 11:05:08 +09002801 * nilfs_detach_log_writer - destroy log writer
2802 * @sb: super block instance
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002803 *
Ryusuke Konishif7545142011-03-09 11:05:08 +09002804 * This kills log writer daemon, frees the log writer object, and
2805 * destroys list of dirty files.
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002806 */
Ryusuke Konishif7545142011-03-09 11:05:08 +09002807void nilfs_detach_log_writer(struct super_block *sb)
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002808{
Ryusuke Konishie3154e92011-03-09 11:05:08 +09002809 struct the_nilfs *nilfs = sb->s_fs_info;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002810 LIST_HEAD(garbage_list);
2811
2812 down_write(&nilfs->ns_segctor_sem);
Ryusuke Konishi3fd3fe52011-03-09 11:05:08 +09002813 if (nilfs->ns_writer) {
2814 nilfs_segctor_destroy(nilfs->ns_writer);
2815 nilfs->ns_writer = NULL;
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002816 }
2817
2818 /* Force to free the list of dirty files */
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002819 spin_lock(&nilfs->ns_inode_lock);
2820 if (!list_empty(&nilfs->ns_dirty_files)) {
2821 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
Ryusuke Konishif7545142011-03-09 11:05:08 +09002822 nilfs_warning(sb, __func__,
2823 "Hit dirty file after stopped log writer\n");
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002824 }
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002825 spin_unlock(&nilfs->ns_inode_lock);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002826 up_write(&nilfs->ns_segctor_sem);
2827
Ryusuke Konishi693dd322011-03-09 11:05:07 +09002828 nilfs_dispose_list(nilfs, &garbage_list, 1);
Ryusuke Konishi9ff05122009-04-06 19:01:37 -07002829}