Commit | Line | Data |
---|---|---|
ae98043f | 1 | // SPDX-License-Identifier: GPL-2.0+ |
9ff05123 | 2 | /* |
94ee1d91 | 3 | * NILFS segment constructor. |
9ff05123 RK |
4 | * |
5 | * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation. | |
6 | * | |
4b420ab4 | 7 | * Written by Ryusuke Konishi. |
9ff05123 RK |
8 | * |
9 | */ | |
10 | ||
11 | #include <linux/pagemap.h> | |
12 | #include <linux/buffer_head.h> | |
13 | #include <linux/writeback.h> | |
ead8ecff | 14 | #include <linux/bitops.h> |
9ff05123 RK |
15 | #include <linux/bio.h> |
16 | #include <linux/completion.h> | |
17 | #include <linux/blkdev.h> | |
18 | #include <linux/backing-dev.h> | |
19 | #include <linux/freezer.h> | |
20 | #include <linux/kthread.h> | |
21 | #include <linux/crc32.h> | |
22 | #include <linux/pagevec.h> | |
5a0e3ad6 | 23 | #include <linux/slab.h> |
174cd4b1 IM |
24 | #include <linux/sched/signal.h> |
25 | ||
9ff05123 RK |
26 | #include "nilfs.h" |
27 | #include "btnode.h" | |
28 | #include "page.h" | |
29 | #include "segment.h" | |
30 | #include "sufile.h" | |
31 | #include "cpfile.h" | |
32 | #include "ifile.h" | |
9ff05123 RK |
33 | #include "segbuf.h" |
34 | ||
35 | ||
36 | /* | |
37 | * Segment constructor | |
38 | */ | |
39 | #define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */ | |
40 | ||
076a378b RK |
41 | #define SC_MAX_SEGDELTA 64 /* |
42 | * Upper limit of the number of segments | |
43 | * appended in collection retry loop | |
44 | */ | |
9ff05123 RK |
45 | |
46 | /* Construction mode */ | |
47 | enum { | |
48 | SC_LSEG_SR = 1, /* Make a logical segment having a super root */ | |
076a378b RK |
49 | SC_LSEG_DSYNC, /* |
50 | * Flush data blocks of a given file and make | |
51 | * a logical segment without a super root. | |
52 | */ | |
53 | SC_FLUSH_FILE, /* | |
54 | * Flush data files, leads to segment writes without | |
55 | * creating a checkpoint. | |
56 | */ | |
57 | SC_FLUSH_DAT, /* | |
58 | * Flush DAT file. This also creates segments | |
59 | * without a checkpoint. | |
60 | */ | |
9ff05123 RK |
61 | }; |
62 | ||
63 | /* Stage numbers of dirty block collection */ | |
64 | enum { | |
65 | NILFS_ST_INIT = 0, | |
66 | NILFS_ST_GC, /* Collecting dirty blocks for GC */ | |
67 | NILFS_ST_FILE, | |
9ff05123 RK |
68 | NILFS_ST_IFILE, |
69 | NILFS_ST_CPFILE, | |
70 | NILFS_ST_SUFILE, | |
71 | NILFS_ST_DAT, | |
72 | NILFS_ST_SR, /* Super root */ | |
73 | NILFS_ST_DSYNC, /* Data sync blocks */ | |
74 | NILFS_ST_DONE, | |
75 | }; | |
76 | ||
58497703 HM |
77 | #define CREATE_TRACE_POINTS |
78 | #include <trace/events/nilfs2.h> | |
79 | ||
80 | /* | |
81 | * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are | |
82 | * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of | |
83 | * the variable must use them because transition of stage count must involve | |
84 | * trace events (trace_nilfs2_collection_stage_transition). | |
85 | * | |
86 | * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't | |
87 | * produce tracepoint events. It is provided just for making the intention | |
88 | * clear. | |
89 | */ | |
90 | static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci) | |
91 | { | |
92 | sci->sc_stage.scnt++; | |
93 | trace_nilfs2_collection_stage_transition(sci); | |
94 | } | |
95 | ||
96 | static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt) | |
97 | { | |
98 | sci->sc_stage.scnt = next_scnt; | |
99 | trace_nilfs2_collection_stage_transition(sci); | |
100 | } | |
101 | ||
102 | static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci) | |
103 | { | |
104 | return sci->sc_stage.scnt; | |
105 | } | |
106 | ||
9ff05123 RK |
107 | /* State flags of collection */ |
108 | #define NILFS_CF_NODE 0x0001 /* Collecting node blocks */ | |
109 | #define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */ | |
071cb4b8 RK |
110 | #define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */ |
111 | #define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED) | |
9ff05123 RK |
112 | |
113 | /* Operations depending on the construction mode and file type */ | |
114 | struct nilfs_sc_operations { | |
115 | int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *, | |
116 | struct inode *); | |
117 | int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *, | |
118 | struct inode *); | |
119 | int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *, | |
120 | struct inode *); | |
121 | void (*write_data_binfo)(struct nilfs_sc_info *, | |
122 | struct nilfs_segsum_pointer *, | |
123 | union nilfs_binfo *); | |
124 | void (*write_node_binfo)(struct nilfs_sc_info *, | |
125 | struct nilfs_segsum_pointer *, | |
126 | union nilfs_binfo *); | |
127 | }; | |
128 | ||
129 | /* | |
130 | * Other definitions | |
131 | */ | |
132 | static void nilfs_segctor_start_timer(struct nilfs_sc_info *); | |
133 | static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int); | |
134 | static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *); | |
693dd321 | 135 | static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int); |
9ff05123 | 136 | |
9ff05123 RK |
137 | #define nilfs_cnt32_ge(a, b) \ |
138 | (typecheck(__u32, a) && typecheck(__u32, b) && \ | |
139 | ((__s32)(a) - (__s32)(b) >= 0)) | |
9ff05123 | 140 | |
feee880f RK |
141 | static int nilfs_prepare_segment_lock(struct super_block *sb, |
142 | struct nilfs_transaction_info *ti) | |
9ff05123 RK |
143 | { |
144 | struct nilfs_transaction_info *cur_ti = current->journal_info; | |
145 | void *save = NULL; | |
146 | ||
147 | if (cur_ti) { | |
148 | if (cur_ti->ti_magic == NILFS_TI_MAGIC) | |
149 | return ++cur_ti->ti_count; | |
7f00184e RK |
150 | |
151 | /* | |
152 | * If journal_info field is occupied by other FS, | |
153 | * it is saved and will be restored on | |
154 | * nilfs_transaction_commit(). | |
155 | */ | |
a1d0747a | 156 | nilfs_warn(sb, "journal info from a different FS"); |
7f00184e | 157 | save = current->journal_info; |
9ff05123 RK |
158 | } |
159 | if (!ti) { | |
160 | ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS); | |
161 | if (!ti) | |
162 | return -ENOMEM; | |
163 | ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC; | |
164 | } else { | |
165 | ti->ti_flags = 0; | |
166 | } | |
167 | ti->ti_count = 0; | |
168 | ti->ti_save = save; | |
169 | ti->ti_magic = NILFS_TI_MAGIC; | |
170 | current->journal_info = ti; | |
171 | return 0; | |
172 | } | |
173 | ||
174 | /** | |
175 | * nilfs_transaction_begin - start indivisible file operations. | |
176 | * @sb: super block | |
177 | * @ti: nilfs_transaction_info | |
178 | * @vacancy_check: flags for vacancy rate checks | |
179 | * | |
180 | * nilfs_transaction_begin() acquires a reader/writer semaphore, called | |
181 | * the segment semaphore, to make a segment construction and write tasks | |
47420c79 | 182 | * exclusive. The function is used with nilfs_transaction_commit() in pairs. |
9ff05123 RK |
183 | * The region enclosed by these two functions can be nested. To avoid a |
184 | * deadlock, the semaphore is only acquired or released in the outermost call. | |
185 | * | |
186 | * This function allocates a nilfs_transaction_info struct to keep context | |
187 | * information on it. It is initialized and hooked onto the current task in | |
188 | * the outermost call. If a pre-allocated struct is given to @ti, it is used | |
7a65004b | 189 | * instead; otherwise a new struct is assigned from a slab. |
9ff05123 RK |
190 | * |
191 | * When @vacancy_check flag is set, this function will check the amount of | |
192 | * free space, and will wait for the GC to reclaim disk space if low capacity. | |
193 | * | |
194 | * Return Value: On success, 0 is returned. On error, one of the following | |
195 | * negative error code is returned. | |
196 | * | |
197 | * %-ENOMEM - Insufficient memory available. | |
198 | * | |
9ff05123 RK |
199 | * %-ENOSPC - No space left on device |
200 | */ | |
201 | int nilfs_transaction_begin(struct super_block *sb, | |
202 | struct nilfs_transaction_info *ti, | |
203 | int vacancy_check) | |
204 | { | |
9ff05123 | 205 | struct the_nilfs *nilfs; |
feee880f | 206 | int ret = nilfs_prepare_segment_lock(sb, ti); |
44fda114 | 207 | struct nilfs_transaction_info *trace_ti; |
9ff05123 RK |
208 | |
209 | if (unlikely(ret < 0)) | |
210 | return ret; | |
44fda114 HM |
211 | if (ret > 0) { |
212 | trace_ti = current->journal_info; | |
213 | ||
214 | trace_nilfs2_transaction_transition(sb, trace_ti, | |
215 | trace_ti->ti_count, trace_ti->ti_flags, | |
216 | TRACE_NILFS2_TRANSACTION_BEGIN); | |
9ff05123 | 217 | return 0; |
44fda114 | 218 | } |
9ff05123 | 219 | |
2c22b337 | 220 | sb_start_intwrite(sb); |
5beb6e0b | 221 | |
e3154e97 | 222 | nilfs = sb->s_fs_info; |
9ff05123 RK |
223 | down_read(&nilfs->ns_segctor_sem); |
224 | if (vacancy_check && nilfs_near_disk_full(nilfs)) { | |
225 | up_read(&nilfs->ns_segctor_sem); | |
226 | ret = -ENOSPC; | |
227 | goto failed; | |
228 | } | |
44fda114 HM |
229 | |
230 | trace_ti = current->journal_info; | |
231 | trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count, | |
232 | trace_ti->ti_flags, | |
233 | TRACE_NILFS2_TRANSACTION_BEGIN); | |
9ff05123 RK |
234 | return 0; |
235 | ||
236 | failed: | |
237 | ti = current->journal_info; | |
238 | current->journal_info = ti->ti_save; | |
239 | if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC) | |
240 | kmem_cache_free(nilfs_transaction_cachep, ti); | |
2c22b337 | 241 | sb_end_intwrite(sb); |
9ff05123 RK |
242 | return ret; |
243 | } | |
244 | ||
245 | /** | |
47420c79 | 246 | * nilfs_transaction_commit - commit indivisible file operations. |
9ff05123 | 247 | * @sb: super block |
9ff05123 | 248 | * |
47420c79 RK |
249 | * nilfs_transaction_commit() releases the read semaphore which is |
250 | * acquired by nilfs_transaction_begin(). This is only performed | |
251 | * in outermost call of this function. If a commit flag is set, | |
252 | * nilfs_transaction_commit() sets a timer to start the segment | |
253 | * constructor. If a sync flag is set, it starts construction | |
254 | * directly. | |
9ff05123 | 255 | */ |
47420c79 | 256 | int nilfs_transaction_commit(struct super_block *sb) |
9ff05123 RK |
257 | { |
258 | struct nilfs_transaction_info *ti = current->journal_info; | |
e3154e97 | 259 | struct the_nilfs *nilfs = sb->s_fs_info; |
9ff05123 RK |
260 | int err = 0; |
261 | ||
262 | BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC); | |
47420c79 | 263 | ti->ti_flags |= NILFS_TI_COMMIT; |
9ff05123 RK |
264 | if (ti->ti_count > 0) { |
265 | ti->ti_count--; | |
44fda114 HM |
266 | trace_nilfs2_transaction_transition(sb, ti, ti->ti_count, |
267 | ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT); | |
9ff05123 RK |
268 | return 0; |
269 | } | |
3fd3fe5a RK |
270 | if (nilfs->ns_writer) { |
271 | struct nilfs_sc_info *sci = nilfs->ns_writer; | |
272 | ||
9ff05123 RK |
273 | if (ti->ti_flags & NILFS_TI_COMMIT) |
274 | nilfs_segctor_start_timer(sci); | |
3fd3fe5a | 275 | if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark) |
9ff05123 RK |
276 | nilfs_segctor_do_flush(sci, 0); |
277 | } | |
3fd3fe5a | 278 | up_read(&nilfs->ns_segctor_sem); |
44fda114 HM |
279 | trace_nilfs2_transaction_transition(sb, ti, ti->ti_count, |
280 | ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT); | |
281 | ||
9ff05123 RK |
282 | current->journal_info = ti->ti_save; |
283 | ||
284 | if (ti->ti_flags & NILFS_TI_SYNC) | |
285 | err = nilfs_construct_segment(sb); | |
286 | if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC) | |
287 | kmem_cache_free(nilfs_transaction_cachep, ti); | |
2c22b337 | 288 | sb_end_intwrite(sb); |
9ff05123 RK |
289 | return err; |
290 | } | |
291 | ||
47420c79 RK |
292 | void nilfs_transaction_abort(struct super_block *sb) |
293 | { | |
294 | struct nilfs_transaction_info *ti = current->journal_info; | |
e3154e97 | 295 | struct the_nilfs *nilfs = sb->s_fs_info; |
47420c79 RK |
296 | |
297 | BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC); | |
298 | if (ti->ti_count > 0) { | |
299 | ti->ti_count--; | |
44fda114 HM |
300 | trace_nilfs2_transaction_transition(sb, ti, ti->ti_count, |
301 | ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT); | |
47420c79 RK |
302 | return; |
303 | } | |
e3154e97 | 304 | up_read(&nilfs->ns_segctor_sem); |
47420c79 | 305 | |
44fda114 HM |
306 | trace_nilfs2_transaction_transition(sb, ti, ti->ti_count, |
307 | ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT); | |
308 | ||
47420c79 RK |
309 | current->journal_info = ti->ti_save; |
310 | if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC) | |
311 | kmem_cache_free(nilfs_transaction_cachep, ti); | |
2c22b337 | 312 | sb_end_intwrite(sb); |
47420c79 RK |
313 | } |
314 | ||
9ff05123 RK |
315 | void nilfs_relax_pressure_in_lock(struct super_block *sb) |
316 | { | |
e3154e97 | 317 | struct the_nilfs *nilfs = sb->s_fs_info; |
3fd3fe5a | 318 | struct nilfs_sc_info *sci = nilfs->ns_writer; |
9ff05123 | 319 | |
8cccf05f | 320 | if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request) |
9ff05123 RK |
321 | return; |
322 | ||
323 | set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags); | |
324 | up_read(&nilfs->ns_segctor_sem); | |
325 | ||
326 | down_write(&nilfs->ns_segctor_sem); | |
327 | if (sci->sc_flush_request && | |
328 | test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) { | |
329 | struct nilfs_transaction_info *ti = current->journal_info; | |
330 | ||
331 | ti->ti_flags |= NILFS_TI_WRITER; | |
332 | nilfs_segctor_do_immediate_flush(sci); | |
333 | ti->ti_flags &= ~NILFS_TI_WRITER; | |
334 | } | |
335 | downgrade_write(&nilfs->ns_segctor_sem); | |
336 | } | |
337 | ||
f7545144 | 338 | static void nilfs_transaction_lock(struct super_block *sb, |
9ff05123 RK |
339 | struct nilfs_transaction_info *ti, |
340 | int gcflag) | |
341 | { | |
342 | struct nilfs_transaction_info *cur_ti = current->journal_info; | |
e3154e97 | 343 | struct the_nilfs *nilfs = sb->s_fs_info; |
3fd3fe5a | 344 | struct nilfs_sc_info *sci = nilfs->ns_writer; |
9ff05123 | 345 | |
1f5abe7e | 346 | WARN_ON(cur_ti); |
9ff05123 RK |
347 | ti->ti_flags = NILFS_TI_WRITER; |
348 | ti->ti_count = 0; | |
349 | ti->ti_save = cur_ti; | |
350 | ti->ti_magic = NILFS_TI_MAGIC; | |
9ff05123 RK |
351 | current->journal_info = ti; |
352 | ||
353 | for (;;) { | |
44fda114 HM |
354 | trace_nilfs2_transaction_transition(sb, ti, ti->ti_count, |
355 | ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK); | |
356 | ||
3fd3fe5a RK |
357 | down_write(&nilfs->ns_segctor_sem); |
358 | if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) | |
9ff05123 RK |
359 | break; |
360 | ||
3fd3fe5a | 361 | nilfs_segctor_do_immediate_flush(sci); |
9ff05123 | 362 | |
f7545144 | 363 | up_write(&nilfs->ns_segctor_sem); |
aceb4170 | 364 | cond_resched(); |
9ff05123 RK |
365 | } |
366 | if (gcflag) | |
367 | ti->ti_flags |= NILFS_TI_GC; | |
44fda114 HM |
368 | |
369 | trace_nilfs2_transaction_transition(sb, ti, ti->ti_count, | |
370 | ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK); | |
9ff05123 RK |
371 | } |
372 | ||
f7545144 | 373 | static void nilfs_transaction_unlock(struct super_block *sb) |
9ff05123 RK |
374 | { |
375 | struct nilfs_transaction_info *ti = current->journal_info; | |
e3154e97 | 376 | struct the_nilfs *nilfs = sb->s_fs_info; |
9ff05123 RK |
377 | |
378 | BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC); | |
379 | BUG_ON(ti->ti_count > 0); | |
380 | ||
693dd321 | 381 | up_write(&nilfs->ns_segctor_sem); |
9ff05123 | 382 | current->journal_info = ti->ti_save; |
44fda114 HM |
383 | |
384 | trace_nilfs2_transaction_transition(sb, ti, ti->ti_count, | |
385 | ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK); | |
9ff05123 RK |
386 | } |
387 | ||
388 | static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci, | |
389 | struct nilfs_segsum_pointer *ssp, | |
0c6c44cb | 390 | unsigned int bytes) |
9ff05123 RK |
391 | { |
392 | struct nilfs_segment_buffer *segbuf = sci->sc_curseg; | |
0c6c44cb | 393 | unsigned int blocksize = sci->sc_super->s_blocksize; |
9ff05123 RK |
394 | void *p; |
395 | ||
396 | if (unlikely(ssp->offset + bytes > blocksize)) { | |
397 | ssp->offset = 0; | |
398 | BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh, | |
399 | &segbuf->sb_segsum_buffers)); | |
400 | ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh); | |
401 | } | |
402 | p = ssp->bh->b_data + ssp->offset; | |
403 | ssp->offset += bytes; | |
404 | return p; | |
405 | } | |
406 | ||
407 | /** | |
408 | * nilfs_segctor_reset_segment_buffer - reset the current segment buffer | |
409 | * @sci: nilfs_sc_info | |
410 | */ | |
411 | static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci) | |
412 | { | |
413 | struct nilfs_segment_buffer *segbuf = sci->sc_curseg; | |
414 | struct buffer_head *sumbh; | |
0c6c44cb RK |
415 | unsigned int sumbytes; |
416 | unsigned int flags = 0; | |
9ff05123 RK |
417 | int err; |
418 | ||
419 | if (nilfs_doing_gc()) | |
420 | flags = NILFS_SS_GC; | |
6c43f410 | 421 | err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno); |
9ff05123 RK |
422 | if (unlikely(err)) |
423 | return err; | |
424 | ||
425 | sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers); | |
426 | sumbytes = segbuf->sb_sum.sumbytes; | |
427 | sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes; | |
428 | sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes; | |
429 | sci->sc_blk_cnt = sci->sc_datablk_cnt = 0; | |
430 | return 0; | |
431 | } | |
432 | ||
433 | static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci) | |
434 | { | |
435 | sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks; | |
436 | if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs)) | |
076a378b RK |
437 | return -E2BIG; /* |
438 | * The current segment is filled up | |
439 | * (internal code) | |
440 | */ | |
9ff05123 RK |
441 | sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg); |
442 | return nilfs_segctor_reset_segment_buffer(sci); | |
443 | } | |
444 | ||
445 | static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci) | |
446 | { | |
447 | struct nilfs_segment_buffer *segbuf = sci->sc_curseg; | |
448 | int err; | |
449 | ||
450 | if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) { | |
451 | err = nilfs_segctor_feed_segment(sci); | |
452 | if (err) | |
453 | return err; | |
454 | segbuf = sci->sc_curseg; | |
455 | } | |
1e2b68bf | 456 | err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root); |
9ff05123 RK |
457 | if (likely(!err)) |
458 | segbuf->sb_sum.flags |= NILFS_SS_SR; | |
459 | return err; | |
460 | } | |
461 | ||
462 | /* | |
463 | * Functions for making segment summary and payloads | |
464 | */ | |
465 | static int nilfs_segctor_segsum_block_required( | |
466 | struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp, | |
0c6c44cb | 467 | unsigned int binfo_size) |
9ff05123 | 468 | { |
0c6c44cb | 469 | unsigned int blocksize = sci->sc_super->s_blocksize; |
9ff05123 RK |
470 | /* Size of finfo and binfo is enough small against blocksize */ |
471 | ||
472 | return ssp->offset + binfo_size + | |
473 | (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) > | |
474 | blocksize; | |
475 | } | |
476 | ||
477 | static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci, | |
478 | struct inode *inode) | |
479 | { | |
480 | sci->sc_curseg->sb_sum.nfinfo++; | |
481 | sci->sc_binfo_ptr = sci->sc_finfo_ptr; | |
482 | nilfs_segctor_map_segsum_entry( | |
483 | sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo)); | |
c96fa464 | 484 | |
72746ac6 RK |
485 | if (NILFS_I(inode)->i_root && |
486 | !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags)) | |
c96fa464 | 487 | set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags); |
9ff05123 RK |
488 | /* skip finfo */ |
489 | } | |
490 | ||
491 | static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci, | |
492 | struct inode *inode) | |
493 | { | |
494 | struct nilfs_finfo *finfo; | |
495 | struct nilfs_inode_info *ii; | |
496 | struct nilfs_segment_buffer *segbuf; | |
6c43f410 | 497 | __u64 cno; |
9ff05123 RK |
498 | |
499 | if (sci->sc_blk_cnt == 0) | |
500 | return; | |
501 | ||
502 | ii = NILFS_I(inode); | |
6c43f410 RK |
503 | |
504 | if (test_bit(NILFS_I_GCINODE, &ii->i_state)) | |
505 | cno = ii->i_cno; | |
506 | else if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) | |
507 | cno = 0; | |
508 | else | |
509 | cno = sci->sc_cno; | |
510 | ||
9ff05123 RK |
511 | finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr, |
512 | sizeof(*finfo)); | |
513 | finfo->fi_ino = cpu_to_le64(inode->i_ino); | |
514 | finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt); | |
515 | finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt); | |
6c43f410 | 516 | finfo->fi_cno = cpu_to_le64(cno); |
9ff05123 RK |
517 | |
518 | segbuf = sci->sc_curseg; | |
519 | segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset + | |
520 | sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1); | |
521 | sci->sc_finfo_ptr = sci->sc_binfo_ptr; | |
522 | sci->sc_blk_cnt = sci->sc_datablk_cnt = 0; | |
523 | } | |
524 | ||
525 | static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci, | |
526 | struct buffer_head *bh, | |
527 | struct inode *inode, | |
0c6c44cb | 528 | unsigned int binfo_size) |
9ff05123 RK |
529 | { |
530 | struct nilfs_segment_buffer *segbuf; | |
531 | int required, err = 0; | |
532 | ||
533 | retry: | |
534 | segbuf = sci->sc_curseg; | |
535 | required = nilfs_segctor_segsum_block_required( | |
536 | sci, &sci->sc_binfo_ptr, binfo_size); | |
537 | if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) { | |
538 | nilfs_segctor_end_finfo(sci, inode); | |
539 | err = nilfs_segctor_feed_segment(sci); | |
540 | if (err) | |
541 | return err; | |
542 | goto retry; | |
543 | } | |
544 | if (unlikely(required)) { | |
545 | err = nilfs_segbuf_extend_segsum(segbuf); | |
546 | if (unlikely(err)) | |
547 | goto failed; | |
548 | } | |
549 | if (sci->sc_blk_cnt == 0) | |
550 | nilfs_segctor_begin_finfo(sci, inode); | |
551 | ||
552 | nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size); | |
553 | /* Substitution to vblocknr is delayed until update_blocknr() */ | |
554 | nilfs_segbuf_add_file_buffer(segbuf, bh); | |
555 | sci->sc_blk_cnt++; | |
556 | failed: | |
557 | return err; | |
558 | } | |
559 | ||
9ff05123 RK |
560 | /* |
561 | * Callback functions that enumerate, mark, and collect dirty blocks | |
562 | */ | |
563 | static int nilfs_collect_file_data(struct nilfs_sc_info *sci, | |
564 | struct buffer_head *bh, struct inode *inode) | |
565 | { | |
566 | int err; | |
567 | ||
9ff05123 | 568 | err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh); |
e828949e RK |
569 | if (err < 0) |
570 | return err; | |
9ff05123 RK |
571 | |
572 | err = nilfs_segctor_add_file_block(sci, bh, inode, | |
573 | sizeof(struct nilfs_binfo_v)); | |
574 | if (!err) | |
575 | sci->sc_datablk_cnt++; | |
576 | return err; | |
577 | } | |
578 | ||
579 | static int nilfs_collect_file_node(struct nilfs_sc_info *sci, | |
580 | struct buffer_head *bh, | |
581 | struct inode *inode) | |
582 | { | |
e828949e | 583 | return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh); |
9ff05123 RK |
584 | } |
585 | ||
586 | static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci, | |
587 | struct buffer_head *bh, | |
588 | struct inode *inode) | |
589 | { | |
1f5abe7e | 590 | WARN_ON(!buffer_dirty(bh)); |
9ff05123 RK |
591 | return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64)); |
592 | } | |
593 | ||
594 | static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci, | |
595 | struct nilfs_segsum_pointer *ssp, | |
596 | union nilfs_binfo *binfo) | |
597 | { | |
598 | struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry( | |
599 | sci, ssp, sizeof(*binfo_v)); | |
600 | *binfo_v = binfo->bi_v; | |
601 | } | |
602 | ||
603 | static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci, | |
604 | struct nilfs_segsum_pointer *ssp, | |
605 | union nilfs_binfo *binfo) | |
606 | { | |
607 | __le64 *vblocknr = nilfs_segctor_map_segsum_entry( | |
608 | sci, ssp, sizeof(*vblocknr)); | |
609 | *vblocknr = binfo->bi_v.bi_vblocknr; | |
610 | } | |
611 | ||
1c613cb9 | 612 | static const struct nilfs_sc_operations nilfs_sc_file_ops = { |
9ff05123 RK |
613 | .collect_data = nilfs_collect_file_data, |
614 | .collect_node = nilfs_collect_file_node, | |
615 | .collect_bmap = nilfs_collect_file_bmap, | |
616 | .write_data_binfo = nilfs_write_file_data_binfo, | |
617 | .write_node_binfo = nilfs_write_file_node_binfo, | |
618 | }; | |
619 | ||
620 | static int nilfs_collect_dat_data(struct nilfs_sc_info *sci, | |
621 | struct buffer_head *bh, struct inode *inode) | |
622 | { | |
623 | int err; | |
624 | ||
625 | err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh); | |
e828949e RK |
626 | if (err < 0) |
627 | return err; | |
9ff05123 RK |
628 | |
629 | err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64)); | |
630 | if (!err) | |
631 | sci->sc_datablk_cnt++; | |
632 | return err; | |
633 | } | |
634 | ||
635 | static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci, | |
636 | struct buffer_head *bh, struct inode *inode) | |
637 | { | |
1f5abe7e | 638 | WARN_ON(!buffer_dirty(bh)); |
9ff05123 RK |
639 | return nilfs_segctor_add_file_block(sci, bh, inode, |
640 | sizeof(struct nilfs_binfo_dat)); | |
641 | } | |
642 | ||
643 | static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci, | |
644 | struct nilfs_segsum_pointer *ssp, | |
645 | union nilfs_binfo *binfo) | |
646 | { | |
647 | __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp, | |
648 | sizeof(*blkoff)); | |
649 | *blkoff = binfo->bi_dat.bi_blkoff; | |
650 | } | |
651 | ||
652 | static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci, | |
653 | struct nilfs_segsum_pointer *ssp, | |
654 | union nilfs_binfo *binfo) | |
655 | { | |
656 | struct nilfs_binfo_dat *binfo_dat = | |
657 | nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat)); | |
658 | *binfo_dat = binfo->bi_dat; | |
659 | } | |
660 | ||
1c613cb9 | 661 | static const struct nilfs_sc_operations nilfs_sc_dat_ops = { |
9ff05123 RK |
662 | .collect_data = nilfs_collect_dat_data, |
663 | .collect_node = nilfs_collect_file_node, | |
664 | .collect_bmap = nilfs_collect_dat_bmap, | |
665 | .write_data_binfo = nilfs_write_dat_data_binfo, | |
666 | .write_node_binfo = nilfs_write_dat_node_binfo, | |
667 | }; | |
668 | ||
1c613cb9 | 669 | static const struct nilfs_sc_operations nilfs_sc_dsync_ops = { |
9ff05123 RK |
670 | .collect_data = nilfs_collect_file_data, |
671 | .collect_node = NULL, | |
672 | .collect_bmap = NULL, | |
673 | .write_data_binfo = nilfs_write_file_data_binfo, | |
674 | .write_node_binfo = NULL, | |
675 | }; | |
676 | ||
f30bf3e4 RK |
677 | static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode, |
678 | struct list_head *listp, | |
679 | size_t nlimit, | |
680 | loff_t start, loff_t end) | |
9ff05123 | 681 | { |
9ff05123 | 682 | struct address_space *mapping = inode->i_mapping; |
5ee4b25c | 683 | struct folio_batch fbatch; |
f30bf3e4 RK |
684 | pgoff_t index = 0, last = ULONG_MAX; |
685 | size_t ndirties = 0; | |
686 | int i; | |
9ff05123 | 687 | |
f30bf3e4 RK |
688 | if (unlikely(start != 0 || end != LLONG_MAX)) { |
689 | /* | |
690 | * A valid range is given for sync-ing data pages. The | |
691 | * range is rounded to per-page; extra dirty buffers | |
692 | * may be included if blocksize < pagesize. | |
693 | */ | |
694 | index = start >> PAGE_SHIFT; | |
695 | last = end >> PAGE_SHIFT; | |
696 | } | |
5ee4b25c | 697 | folio_batch_init(&fbatch); |
9ff05123 | 698 | repeat: |
f30bf3e4 | 699 | if (unlikely(index > last) || |
5ee4b25c VMO |
700 | !filemap_get_folios_tag(mapping, &index, last, |
701 | PAGECACHE_TAG_DIRTY, &fbatch)) | |
f30bf3e4 | 702 | return ndirties; |
9ff05123 | 703 | |
5ee4b25c | 704 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
9ff05123 | 705 | struct buffer_head *bh, *head; |
5ee4b25c | 706 | struct folio *folio = fbatch.folios[i]; |
9ff05123 | 707 | |
5ee4b25c VMO |
708 | folio_lock(folio); |
709 | head = folio_buffers(folio); | |
710 | if (!head) { | |
711 | create_empty_buffers(&folio->page, i_blocksize(inode), 0); | |
712 | head = folio_buffers(folio); | |
713 | } | |
714 | folio_unlock(folio); | |
9ff05123 | 715 | |
5ee4b25c | 716 | bh = head; |
9ff05123 | 717 | do { |
7f42ec39 | 718 | if (!buffer_dirty(bh) || buffer_async_write(bh)) |
f30bf3e4 RK |
719 | continue; |
720 | get_bh(bh); | |
721 | list_add_tail(&bh->b_assoc_buffers, listp); | |
722 | ndirties++; | |
723 | if (unlikely(ndirties >= nlimit)) { | |
5ee4b25c | 724 | folio_batch_release(&fbatch); |
f30bf3e4 RK |
725 | cond_resched(); |
726 | return ndirties; | |
9ff05123 | 727 | } |
f30bf3e4 | 728 | } while (bh = bh->b_this_page, bh != head); |
9ff05123 | 729 | } |
5ee4b25c | 730 | folio_batch_release(&fbatch); |
9ff05123 | 731 | cond_resched(); |
f30bf3e4 | 732 | goto repeat; |
9ff05123 RK |
733 | } |
734 | ||
735 | static void nilfs_lookup_dirty_node_buffers(struct inode *inode, | |
736 | struct list_head *listp) | |
737 | { | |
738 | struct nilfs_inode_info *ii = NILFS_I(inode); | |
e897be17 | 739 | struct inode *btnc_inode = ii->i_assoc_inode; |
9ff05123 RK |
740 | struct pagevec pvec; |
741 | struct buffer_head *bh, *head; | |
742 | unsigned int i; | |
743 | pgoff_t index = 0; | |
744 | ||
e897be17 RK |
745 | if (!btnc_inode) |
746 | return; | |
747 | ||
86679820 | 748 | pagevec_init(&pvec); |
9ff05123 | 749 | |
e897be17 | 750 | while (pagevec_lookup_tag(&pvec, btnc_inode->i_mapping, &index, |
67fd707f | 751 | PAGECACHE_TAG_DIRTY)) { |
9ff05123 RK |
752 | for (i = 0; i < pagevec_count(&pvec); i++) { |
753 | bh = head = page_buffers(pvec.pages[i]); | |
754 | do { | |
7f42ec39 VD |
755 | if (buffer_dirty(bh) && |
756 | !buffer_async_write(bh)) { | |
9ff05123 RK |
757 | get_bh(bh); |
758 | list_add_tail(&bh->b_assoc_buffers, | |
759 | listp); | |
760 | } | |
761 | bh = bh->b_this_page; | |
762 | } while (bh != head); | |
763 | } | |
764 | pagevec_release(&pvec); | |
765 | cond_resched(); | |
766 | } | |
767 | } | |
768 | ||
693dd321 | 769 | static void nilfs_dispose_list(struct the_nilfs *nilfs, |
9ff05123 RK |
770 | struct list_head *head, int force) |
771 | { | |
772 | struct nilfs_inode_info *ii, *n; | |
773 | struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii; | |
0c6c44cb | 774 | unsigned int nv = 0; |
9ff05123 RK |
775 | |
776 | while (!list_empty(head)) { | |
693dd321 | 777 | spin_lock(&nilfs->ns_inode_lock); |
9ff05123 RK |
778 | list_for_each_entry_safe(ii, n, head, i_dirty) { |
779 | list_del_init(&ii->i_dirty); | |
780 | if (force) { | |
781 | if (unlikely(ii->i_bh)) { | |
782 | brelse(ii->i_bh); | |
783 | ii->i_bh = NULL; | |
784 | } | |
785 | } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) { | |
786 | set_bit(NILFS_I_QUEUED, &ii->i_state); | |
787 | list_add_tail(&ii->i_dirty, | |
693dd321 | 788 | &nilfs->ns_dirty_files); |
9ff05123 RK |
789 | continue; |
790 | } | |
791 | ivec[nv++] = ii; | |
792 | if (nv == SC_N_INODEVEC) | |
793 | break; | |
794 | } | |
693dd321 | 795 | spin_unlock(&nilfs->ns_inode_lock); |
9ff05123 RK |
796 | |
797 | for (pii = ivec; nv > 0; pii++, nv--) | |
798 | iput(&(*pii)->vfs_inode); | |
799 | } | |
800 | } | |
801 | ||
7ef3ff2f RK |
802 | static void nilfs_iput_work_func(struct work_struct *work) |
803 | { | |
804 | struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info, | |
805 | sc_iput_work); | |
806 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; | |
807 | ||
808 | nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0); | |
809 | } | |
810 | ||
e912a5b6 RK |
811 | static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs, |
812 | struct nilfs_root *root) | |
9ff05123 | 813 | { |
9ff05123 RK |
814 | int ret = 0; |
815 | ||
e912a5b6 | 816 | if (nilfs_mdt_fetch_dirty(root->ifile)) |
9ff05123 RK |
817 | ret++; |
818 | if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile)) | |
819 | ret++; | |
820 | if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile)) | |
821 | ret++; | |
365e215c RK |
822 | if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat)) |
823 | ret++; | |
9ff05123 RK |
824 | return ret; |
825 | } | |
826 | ||
827 | static int nilfs_segctor_clean(struct nilfs_sc_info *sci) | |
828 | { | |
829 | return list_empty(&sci->sc_dirty_files) && | |
830 | !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) && | |
071cb4b8 | 831 | sci->sc_nfreesegs == 0 && |
9ff05123 RK |
832 | (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes)); |
833 | } | |
834 | ||
835 | static int nilfs_segctor_confirm(struct nilfs_sc_info *sci) | |
836 | { | |
e3154e97 | 837 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
9ff05123 RK |
838 | int ret = 0; |
839 | ||
693dd321 | 840 | if (nilfs_test_metadata_dirty(nilfs, sci->sc_root)) |
9ff05123 RK |
841 | set_bit(NILFS_SC_DIRTY, &sci->sc_flags); |
842 | ||
693dd321 RK |
843 | spin_lock(&nilfs->ns_inode_lock); |
844 | if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci)) | |
9ff05123 RK |
845 | ret++; |
846 | ||
693dd321 | 847 | spin_unlock(&nilfs->ns_inode_lock); |
9ff05123 RK |
848 | return ret; |
849 | } | |
850 | ||
851 | static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci) | |
852 | { | |
e3154e97 | 853 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
9ff05123 | 854 | |
e912a5b6 | 855 | nilfs_mdt_clear_dirty(sci->sc_root->ifile); |
9ff05123 RK |
856 | nilfs_mdt_clear_dirty(nilfs->ns_cpfile); |
857 | nilfs_mdt_clear_dirty(nilfs->ns_sufile); | |
365e215c | 858 | nilfs_mdt_clear_dirty(nilfs->ns_dat); |
9ff05123 RK |
859 | } |
860 | ||
861 | static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci) | |
862 | { | |
e3154e97 | 863 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
9ff05123 RK |
864 | struct buffer_head *bh_cp; |
865 | struct nilfs_checkpoint *raw_cp; | |
866 | int err; | |
867 | ||
868 | /* XXX: this interface will be changed */ | |
869 | err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1, | |
870 | &raw_cp, &bh_cp); | |
871 | if (likely(!err)) { | |
076a378b RK |
872 | /* |
873 | * The following code is duplicated with cpfile. But, it is | |
874 | * needed to collect the checkpoint even if it was not newly | |
875 | * created. | |
876 | */ | |
5fc7b141 | 877 | mark_buffer_dirty(bh_cp); |
9ff05123 RK |
878 | nilfs_mdt_mark_dirty(nilfs->ns_cpfile); |
879 | nilfs_cpfile_put_checkpoint( | |
880 | nilfs->ns_cpfile, nilfs->ns_cno, bh_cp); | |
723ac751 RK |
881 | } else if (err == -EINVAL || err == -ENOENT) { |
882 | nilfs_error(sci->sc_super, | |
883 | "checkpoint creation failed due to metadata corruption."); | |
884 | err = -EIO; | |
885 | } | |
9ff05123 RK |
886 | return err; |
887 | } | |
888 | ||
889 | static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci) | |
890 | { | |
e3154e97 | 891 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
9ff05123 RK |
892 | struct buffer_head *bh_cp; |
893 | struct nilfs_checkpoint *raw_cp; | |
894 | int err; | |
895 | ||
896 | err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0, | |
897 | &raw_cp, &bh_cp); | |
898 | if (unlikely(err)) { | |
723ac751 RK |
899 | if (err == -EINVAL || err == -ENOENT) { |
900 | nilfs_error(sci->sc_super, | |
901 | "checkpoint finalization failed due to metadata corruption."); | |
902 | err = -EIO; | |
903 | } | |
9ff05123 RK |
904 | goto failed_ibh; |
905 | } | |
906 | raw_cp->cp_snapshot_list.ssl_next = 0; | |
907 | raw_cp->cp_snapshot_list.ssl_prev = 0; | |
908 | raw_cp->cp_inodes_count = | |
e5f7f848 | 909 | cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count)); |
9ff05123 | 910 | raw_cp->cp_blocks_count = |
e5f7f848 | 911 | cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count)); |
9ff05123 RK |
912 | raw_cp->cp_nblk_inc = |
913 | cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc); | |
914 | raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime); | |
915 | raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno); | |
458c5b08 | 916 | |
c96fa464 RK |
917 | if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags)) |
918 | nilfs_checkpoint_clear_minor(raw_cp); | |
919 | else | |
920 | nilfs_checkpoint_set_minor(raw_cp); | |
921 | ||
e912a5b6 RK |
922 | nilfs_write_inode_common(sci->sc_root->ifile, |
923 | &raw_cp->cp_ifile_inode, 1); | |
9ff05123 RK |
924 | nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp); |
925 | return 0; | |
926 | ||
927 | failed_ibh: | |
928 | return err; | |
929 | } | |
930 | ||
931 | static void nilfs_fill_in_file_bmap(struct inode *ifile, | |
932 | struct nilfs_inode_info *ii) | |
933 | ||
934 | { | |
935 | struct buffer_head *ibh; | |
936 | struct nilfs_inode *raw_inode; | |
937 | ||
938 | if (test_bit(NILFS_I_BMAP, &ii->i_state)) { | |
939 | ibh = ii->i_bh; | |
940 | BUG_ON(!ibh); | |
941 | raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino, | |
942 | ibh); | |
943 | nilfs_bmap_write(ii->i_bmap, raw_inode); | |
944 | nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh); | |
945 | } | |
946 | } | |
947 | ||
e912a5b6 | 948 | static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci) |
9ff05123 RK |
949 | { |
950 | struct nilfs_inode_info *ii; | |
951 | ||
952 | list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) { | |
e912a5b6 | 953 | nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii); |
9ff05123 RK |
954 | set_bit(NILFS_I_COLLECTED, &ii->i_state); |
955 | } | |
9ff05123 RK |
956 | } |
957 | ||
9ff05123 RK |
958 | static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci, |
959 | struct the_nilfs *nilfs) | |
960 | { | |
1e2b68bf RK |
961 | struct buffer_head *bh_sr; |
962 | struct nilfs_super_root *raw_sr; | |
0c6c44cb | 963 | unsigned int isz, srsz; |
9ff05123 | 964 | |
1e2b68bf RK |
965 | bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root; |
966 | raw_sr = (struct nilfs_super_root *)bh_sr->b_data; | |
56eb5538 RK |
967 | isz = nilfs->ns_inode_size; |
968 | srsz = NILFS_SR_BYTES(isz); | |
1e2b68bf | 969 | |
56eb5538 | 970 | raw_sr->sr_bytes = cpu_to_le16(srsz); |
9ff05123 RK |
971 | raw_sr->sr_nongc_ctime |
972 | = cpu_to_le64(nilfs_doing_gc() ? | |
973 | nilfs->ns_nongc_ctime : sci->sc_seg_ctime); | |
974 | raw_sr->sr_flags = 0; | |
975 | ||
365e215c | 976 | nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr + |
3961f0e2 RK |
977 | NILFS_SR_DAT_OFFSET(isz), 1); |
978 | nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr + | |
979 | NILFS_SR_CPFILE_OFFSET(isz), 1); | |
980 | nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr + | |
981 | NILFS_SR_SUFILE_OFFSET(isz), 1); | |
56eb5538 | 982 | memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz); |
9ff05123 RK |
983 | } |
984 | ||
985 | static void nilfs_redirty_inodes(struct list_head *head) | |
986 | { | |
987 | struct nilfs_inode_info *ii; | |
988 | ||
989 | list_for_each_entry(ii, head, i_dirty) { | |
990 | if (test_bit(NILFS_I_COLLECTED, &ii->i_state)) | |
991 | clear_bit(NILFS_I_COLLECTED, &ii->i_state); | |
992 | } | |
993 | } | |
994 | ||
995 | static void nilfs_drop_collected_inodes(struct list_head *head) | |
996 | { | |
997 | struct nilfs_inode_info *ii; | |
998 | ||
999 | list_for_each_entry(ii, head, i_dirty) { | |
1000 | if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state)) | |
1001 | continue; | |
1002 | ||
b9f66140 | 1003 | clear_bit(NILFS_I_INODE_SYNC, &ii->i_state); |
9ff05123 RK |
1004 | set_bit(NILFS_I_UPDATED, &ii->i_state); |
1005 | } | |
1006 | } | |
1007 | ||
9ff05123 RK |
1008 | static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci, |
1009 | struct inode *inode, | |
1010 | struct list_head *listp, | |
1011 | int (*collect)(struct nilfs_sc_info *, | |
1012 | struct buffer_head *, | |
1013 | struct inode *)) | |
1014 | { | |
1015 | struct buffer_head *bh, *n; | |
1016 | int err = 0; | |
1017 | ||
1018 | if (collect) { | |
1019 | list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) { | |
1020 | list_del_init(&bh->b_assoc_buffers); | |
1021 | err = collect(sci, bh, inode); | |
1022 | brelse(bh); | |
1023 | if (unlikely(err)) | |
1024 | goto dispose_buffers; | |
1025 | } | |
1026 | return 0; | |
1027 | } | |
1028 | ||
1029 | dispose_buffers: | |
1030 | while (!list_empty(listp)) { | |
0cc12838 RK |
1031 | bh = list_first_entry(listp, struct buffer_head, |
1032 | b_assoc_buffers); | |
9ff05123 RK |
1033 | list_del_init(&bh->b_assoc_buffers); |
1034 | brelse(bh); | |
1035 | } | |
1036 | return err; | |
1037 | } | |
1038 | ||
f30bf3e4 RK |
1039 | static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci) |
1040 | { | |
1041 | /* Remaining number of blocks within segment buffer */ | |
1042 | return sci->sc_segbuf_nblocks - | |
1043 | (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks); | |
1044 | } | |
1045 | ||
9ff05123 RK |
1046 | static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci, |
1047 | struct inode *inode, | |
1c613cb9 | 1048 | const struct nilfs_sc_operations *sc_ops) |
9ff05123 RK |
1049 | { |
1050 | LIST_HEAD(data_buffers); | |
1051 | LIST_HEAD(node_buffers); | |
f30bf3e4 | 1052 | int err; |
9ff05123 RK |
1053 | |
1054 | if (!(sci->sc_stage.flags & NILFS_CF_NODE)) { | |
f30bf3e4 RK |
1055 | size_t n, rest = nilfs_segctor_buffer_rest(sci); |
1056 | ||
1057 | n = nilfs_lookup_dirty_data_buffers( | |
1058 | inode, &data_buffers, rest + 1, 0, LLONG_MAX); | |
1059 | if (n > rest) { | |
1060 | err = nilfs_segctor_apply_buffers( | |
9ff05123 | 1061 | sci, inode, &data_buffers, |
f30bf3e4 RK |
1062 | sc_ops->collect_data); |
1063 | BUG_ON(!err); /* always receive -E2BIG or true error */ | |
9ff05123 RK |
1064 | goto break_or_fail; |
1065 | } | |
1066 | } | |
1067 | nilfs_lookup_dirty_node_buffers(inode, &node_buffers); | |
1068 | ||
1069 | if (!(sci->sc_stage.flags & NILFS_CF_NODE)) { | |
1070 | err = nilfs_segctor_apply_buffers( | |
1071 | sci, inode, &data_buffers, sc_ops->collect_data); | |
1072 | if (unlikely(err)) { | |
1073 | /* dispose node list */ | |
1074 | nilfs_segctor_apply_buffers( | |
1075 | sci, inode, &node_buffers, NULL); | |
1076 | goto break_or_fail; | |
1077 | } | |
1078 | sci->sc_stage.flags |= NILFS_CF_NODE; | |
1079 | } | |
1080 | /* Collect node */ | |
1081 | err = nilfs_segctor_apply_buffers( | |
1082 | sci, inode, &node_buffers, sc_ops->collect_node); | |
1083 | if (unlikely(err)) | |
1084 | goto break_or_fail; | |
1085 | ||
1086 | nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers); | |
1087 | err = nilfs_segctor_apply_buffers( | |
1088 | sci, inode, &node_buffers, sc_ops->collect_bmap); | |
1089 | if (unlikely(err)) | |
1090 | goto break_or_fail; | |
1091 | ||
1092 | nilfs_segctor_end_finfo(sci, inode); | |
1093 | sci->sc_stage.flags &= ~NILFS_CF_NODE; | |
1094 | ||
1095 | break_or_fail: | |
1096 | return err; | |
1097 | } | |
1098 | ||
1099 | static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci, | |
1100 | struct inode *inode) | |
1101 | { | |
1102 | LIST_HEAD(data_buffers); | |
f30bf3e4 RK |
1103 | size_t n, rest = nilfs_segctor_buffer_rest(sci); |
1104 | int err; | |
9ff05123 | 1105 | |
f30bf3e4 RK |
1106 | n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1, |
1107 | sci->sc_dsync_start, | |
1108 | sci->sc_dsync_end); | |
1109 | ||
1110 | err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers, | |
1111 | nilfs_collect_file_data); | |
1112 | if (!err) { | |
9ff05123 | 1113 | nilfs_segctor_end_finfo(sci, inode); |
f30bf3e4 RK |
1114 | BUG_ON(n > rest); |
1115 | /* always receive -E2BIG or true error if n > rest */ | |
1116 | } | |
9ff05123 RK |
1117 | return err; |
1118 | } | |
1119 | ||
1120 | static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode) | |
1121 | { | |
e3154e97 | 1122 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
9ff05123 RK |
1123 | struct list_head *head; |
1124 | struct nilfs_inode_info *ii; | |
071cb4b8 | 1125 | size_t ndone; |
9ff05123 RK |
1126 | int err = 0; |
1127 | ||
58497703 | 1128 | switch (nilfs_sc_cstage_get(sci)) { |
9ff05123 RK |
1129 | case NILFS_ST_INIT: |
1130 | /* Pre-processes */ | |
1131 | sci->sc_stage.flags = 0; | |
1132 | ||
1133 | if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) { | |
1134 | sci->sc_nblk_inc = 0; | |
1135 | sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN; | |
1136 | if (mode == SC_LSEG_DSYNC) { | |
58497703 | 1137 | nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC); |
9ff05123 RK |
1138 | goto dsync_mode; |
1139 | } | |
1140 | } | |
1141 | ||
1142 | sci->sc_stage.dirty_file_ptr = NULL; | |
1143 | sci->sc_stage.gc_inode_ptr = NULL; | |
1144 | if (mode == SC_FLUSH_DAT) { | |
58497703 | 1145 | nilfs_sc_cstage_set(sci, NILFS_ST_DAT); |
9ff05123 RK |
1146 | goto dat_stage; |
1147 | } | |
df561f66 GS |
1148 | nilfs_sc_cstage_inc(sci); |
1149 | fallthrough; | |
9ff05123 RK |
1150 | case NILFS_ST_GC: |
1151 | if (nilfs_doing_gc()) { | |
1152 | head = &sci->sc_gc_inodes; | |
1153 | ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr, | |
1154 | head, i_dirty); | |
1155 | list_for_each_entry_continue(ii, head, i_dirty) { | |
1156 | err = nilfs_segctor_scan_file( | |
1157 | sci, &ii->vfs_inode, | |
1158 | &nilfs_sc_file_ops); | |
1159 | if (unlikely(err)) { | |
1160 | sci->sc_stage.gc_inode_ptr = list_entry( | |
1161 | ii->i_dirty.prev, | |
1162 | struct nilfs_inode_info, | |
1163 | i_dirty); | |
1164 | goto break_or_fail; | |
1165 | } | |
1166 | set_bit(NILFS_I_COLLECTED, &ii->i_state); | |
1167 | } | |
1168 | sci->sc_stage.gc_inode_ptr = NULL; | |
1169 | } | |
df561f66 GS |
1170 | nilfs_sc_cstage_inc(sci); |
1171 | fallthrough; | |
9ff05123 RK |
1172 | case NILFS_ST_FILE: |
1173 | head = &sci->sc_dirty_files; | |
1174 | ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head, | |
1175 | i_dirty); | |
1176 | list_for_each_entry_continue(ii, head, i_dirty) { | |
1177 | clear_bit(NILFS_I_DIRTY, &ii->i_state); | |
1178 | ||
1179 | err = nilfs_segctor_scan_file(sci, &ii->vfs_inode, | |
1180 | &nilfs_sc_file_ops); | |
1181 | if (unlikely(err)) { | |
1182 | sci->sc_stage.dirty_file_ptr = | |
1183 | list_entry(ii->i_dirty.prev, | |
1184 | struct nilfs_inode_info, | |
1185 | i_dirty); | |
1186 | goto break_or_fail; | |
1187 | } | |
1188 | /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */ | |
1189 | /* XXX: required ? */ | |
1190 | } | |
1191 | sci->sc_stage.dirty_file_ptr = NULL; | |
1192 | if (mode == SC_FLUSH_FILE) { | |
58497703 | 1193 | nilfs_sc_cstage_set(sci, NILFS_ST_DONE); |
9ff05123 RK |
1194 | return 0; |
1195 | } | |
58497703 | 1196 | nilfs_sc_cstage_inc(sci); |
9ff05123 | 1197 | sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED; |
df561f66 | 1198 | fallthrough; |
9ff05123 | 1199 | case NILFS_ST_IFILE: |
e912a5b6 | 1200 | err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile, |
9ff05123 RK |
1201 | &nilfs_sc_file_ops); |
1202 | if (unlikely(err)) | |
1203 | break; | |
58497703 | 1204 | nilfs_sc_cstage_inc(sci); |
9ff05123 RK |
1205 | /* Creating a checkpoint */ |
1206 | err = nilfs_segctor_create_checkpoint(sci); | |
1207 | if (unlikely(err)) | |
1208 | break; | |
df561f66 | 1209 | fallthrough; |
9ff05123 RK |
1210 | case NILFS_ST_CPFILE: |
1211 | err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile, | |
1212 | &nilfs_sc_file_ops); | |
1213 | if (unlikely(err)) | |
1214 | break; | |
df561f66 GS |
1215 | nilfs_sc_cstage_inc(sci); |
1216 | fallthrough; | |
9ff05123 | 1217 | case NILFS_ST_SUFILE: |
071cb4b8 RK |
1218 | err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs, |
1219 | sci->sc_nfreesegs, &ndone); | |
1220 | if (unlikely(err)) { | |
1221 | nilfs_sufile_cancel_freev(nilfs->ns_sufile, | |
1222 | sci->sc_freesegs, ndone, | |
1223 | NULL); | |
9ff05123 | 1224 | break; |
071cb4b8 RK |
1225 | } |
1226 | sci->sc_stage.flags |= NILFS_CF_SUFREED; | |
1227 | ||
9ff05123 RK |
1228 | err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile, |
1229 | &nilfs_sc_file_ops); | |
1230 | if (unlikely(err)) | |
1231 | break; | |
df561f66 GS |
1232 | nilfs_sc_cstage_inc(sci); |
1233 | fallthrough; | |
9ff05123 RK |
1234 | case NILFS_ST_DAT: |
1235 | dat_stage: | |
365e215c | 1236 | err = nilfs_segctor_scan_file(sci, nilfs->ns_dat, |
9ff05123 RK |
1237 | &nilfs_sc_dat_ops); |
1238 | if (unlikely(err)) | |
1239 | break; | |
1240 | if (mode == SC_FLUSH_DAT) { | |
58497703 | 1241 | nilfs_sc_cstage_set(sci, NILFS_ST_DONE); |
9ff05123 RK |
1242 | return 0; |
1243 | } | |
df561f66 GS |
1244 | nilfs_sc_cstage_inc(sci); |
1245 | fallthrough; | |
9ff05123 RK |
1246 | case NILFS_ST_SR: |
1247 | if (mode == SC_LSEG_SR) { | |
1248 | /* Appending a super root */ | |
1249 | err = nilfs_segctor_add_super_root(sci); | |
1250 | if (unlikely(err)) | |
1251 | break; | |
1252 | } | |
1253 | /* End of a logical segment */ | |
1254 | sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND; | |
58497703 | 1255 | nilfs_sc_cstage_set(sci, NILFS_ST_DONE); |
9ff05123 RK |
1256 | return 0; |
1257 | case NILFS_ST_DSYNC: | |
1258 | dsync_mode: | |
1259 | sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT; | |
f30bf3e4 | 1260 | ii = sci->sc_dsync_inode; |
9ff05123 RK |
1261 | if (!test_bit(NILFS_I_BUSY, &ii->i_state)) |
1262 | break; | |
1263 | ||
1264 | err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode); | |
1265 | if (unlikely(err)) | |
1266 | break; | |
9ff05123 | 1267 | sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND; |
58497703 | 1268 | nilfs_sc_cstage_set(sci, NILFS_ST_DONE); |
9ff05123 RK |
1269 | return 0; |
1270 | case NILFS_ST_DONE: | |
1271 | return 0; | |
1272 | default: | |
1273 | BUG(); | |
1274 | } | |
1275 | ||
1276 | break_or_fail: | |
1277 | return err; | |
1278 | } | |
1279 | ||
a694291a RK |
1280 | /** |
1281 | * nilfs_segctor_begin_construction - setup segment buffer to make a new log | |
1282 | * @sci: nilfs_sc_info | |
1283 | * @nilfs: nilfs object | |
1284 | */ | |
9ff05123 RK |
1285 | static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci, |
1286 | struct the_nilfs *nilfs) | |
1287 | { | |
a694291a | 1288 | struct nilfs_segment_buffer *segbuf, *prev; |
9ff05123 | 1289 | __u64 nextnum; |
a694291a | 1290 | int err, alloc = 0; |
9ff05123 | 1291 | |
a694291a RK |
1292 | segbuf = nilfs_segbuf_new(sci->sc_super); |
1293 | if (unlikely(!segbuf)) | |
1294 | return -ENOMEM; | |
9ff05123 | 1295 | |
a694291a RK |
1296 | if (list_empty(&sci->sc_write_logs)) { |
1297 | nilfs_segbuf_map(segbuf, nilfs->ns_segnum, | |
1298 | nilfs->ns_pseg_offset, nilfs); | |
1299 | if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) { | |
1300 | nilfs_shift_to_next_segment(nilfs); | |
1301 | nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs); | |
1302 | } | |
9ff05123 | 1303 | |
a694291a RK |
1304 | segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq; |
1305 | nextnum = nilfs->ns_nextnum; | |
1306 | ||
1307 | if (nilfs->ns_segnum == nilfs->ns_nextnum) | |
1308 | /* Start from the head of a new full segment */ | |
1309 | alloc++; | |
1310 | } else { | |
1311 | /* Continue logs */ | |
1312 | prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs); | |
1313 | nilfs_segbuf_map_cont(segbuf, prev); | |
1314 | segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq; | |
1315 | nextnum = prev->sb_nextnum; | |
1316 | ||
1317 | if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) { | |
1318 | nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs); | |
1319 | segbuf->sb_sum.seg_seq++; | |
1320 | alloc++; | |
1321 | } | |
9ff05123 | 1322 | } |
9ff05123 | 1323 | |
61a189e9 | 1324 | err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum); |
a694291a RK |
1325 | if (err) |
1326 | goto failed; | |
9ff05123 | 1327 | |
a694291a | 1328 | if (alloc) { |
cece5520 | 1329 | err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum); |
a694291a RK |
1330 | if (err) |
1331 | goto failed; | |
1332 | } | |
9ff05123 RK |
1333 | nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs); |
1334 | ||
a694291a RK |
1335 | BUG_ON(!list_empty(&sci->sc_segbufs)); |
1336 | list_add_tail(&segbuf->sb_list, &sci->sc_segbufs); | |
1337 | sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks; | |
cece5520 | 1338 | return 0; |
a694291a RK |
1339 | |
1340 | failed: | |
1341 | nilfs_segbuf_free(segbuf); | |
1342 | return err; | |
9ff05123 RK |
1343 | } |
1344 | ||
1345 | static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci, | |
1346 | struct the_nilfs *nilfs, int nadd) | |
1347 | { | |
e29df395 | 1348 | struct nilfs_segment_buffer *segbuf, *prev; |
9ff05123 RK |
1349 | struct inode *sufile = nilfs->ns_sufile; |
1350 | __u64 nextnextnum; | |
1351 | LIST_HEAD(list); | |
1352 | int err, ret, i; | |
1353 | ||
1354 | prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs); | |
1355 | /* | |
1356 | * Since the segment specified with nextnum might be allocated during | |
1357 | * the previous construction, the buffer including its segusage may | |
1358 | * not be dirty. The following call ensures that the buffer is dirty | |
1359 | * and will pin the buffer on memory until the sufile is written. | |
1360 | */ | |
61a189e9 | 1361 | err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum); |
9ff05123 RK |
1362 | if (unlikely(err)) |
1363 | return err; | |
1364 | ||
1365 | for (i = 0; i < nadd; i++) { | |
1366 | /* extend segment info */ | |
1367 | err = -ENOMEM; | |
1368 | segbuf = nilfs_segbuf_new(sci->sc_super); | |
1369 | if (unlikely(!segbuf)) | |
1370 | goto failed; | |
1371 | ||
1372 | /* map this buffer to region of segment on-disk */ | |
cece5520 | 1373 | nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs); |
9ff05123 RK |
1374 | sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks; |
1375 | ||
1376 | /* allocate the next next full segment */ | |
1377 | err = nilfs_sufile_alloc(sufile, &nextnextnum); | |
1378 | if (unlikely(err)) | |
1379 | goto failed_segbuf; | |
1380 | ||
1381 | segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1; | |
1382 | nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs); | |
1383 | ||
1384 | list_add_tail(&segbuf->sb_list, &list); | |
1385 | prev = segbuf; | |
1386 | } | |
0935db74 | 1387 | list_splice_tail(&list, &sci->sc_segbufs); |
9ff05123 RK |
1388 | return 0; |
1389 | ||
1390 | failed_segbuf: | |
1391 | nilfs_segbuf_free(segbuf); | |
1392 | failed: | |
e29df395 | 1393 | list_for_each_entry(segbuf, &list, sb_list) { |
9ff05123 | 1394 | ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum); |
1f5abe7e | 1395 | WARN_ON(ret); /* never fails */ |
9ff05123 | 1396 | } |
e29df395 | 1397 | nilfs_destroy_logs(&list); |
9ff05123 RK |
1398 | return err; |
1399 | } | |
1400 | ||
a694291a RK |
1401 | static void nilfs_free_incomplete_logs(struct list_head *logs, |
1402 | struct the_nilfs *nilfs) | |
9ff05123 | 1403 | { |
a694291a RK |
1404 | struct nilfs_segment_buffer *segbuf, *prev; |
1405 | struct inode *sufile = nilfs->ns_sufile; | |
9284ad2a | 1406 | int ret; |
9ff05123 | 1407 | |
a694291a | 1408 | segbuf = NILFS_FIRST_SEGBUF(logs); |
9ff05123 | 1409 | if (nilfs->ns_nextnum != segbuf->sb_nextnum) { |
a694291a | 1410 | ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum); |
1f5abe7e | 1411 | WARN_ON(ret); /* never fails */ |
9ff05123 | 1412 | } |
9284ad2a | 1413 | if (atomic_read(&segbuf->sb_err)) { |
9ff05123 RK |
1414 | /* Case 1: The first segment failed */ |
1415 | if (segbuf->sb_pseg_start != segbuf->sb_fseg_start) | |
076a378b RK |
1416 | /* |
1417 | * Case 1a: Partial segment appended into an existing | |
1418 | * segment | |
1419 | */ | |
9ff05123 RK |
1420 | nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start, |
1421 | segbuf->sb_fseg_end); | |
1422 | else /* Case 1b: New full segment */ | |
1423 | set_nilfs_discontinued(nilfs); | |
9ff05123 RK |
1424 | } |
1425 | ||
a694291a RK |
1426 | prev = segbuf; |
1427 | list_for_each_entry_continue(segbuf, logs, sb_list) { | |
1428 | if (prev->sb_nextnum != segbuf->sb_nextnum) { | |
1429 | ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum); | |
1430 | WARN_ON(ret); /* never fails */ | |
1431 | } | |
9284ad2a RK |
1432 | if (atomic_read(&segbuf->sb_err) && |
1433 | segbuf->sb_segnum != nilfs->ns_nextnum) | |
1434 | /* Case 2: extended segment (!= next) failed */ | |
a694291a RK |
1435 | nilfs_sufile_set_error(sufile, segbuf->sb_segnum); |
1436 | prev = segbuf; | |
9ff05123 | 1437 | } |
9ff05123 RK |
1438 | } |
1439 | ||
1440 | static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci, | |
1441 | struct inode *sufile) | |
1442 | { | |
1443 | struct nilfs_segment_buffer *segbuf; | |
9ff05123 RK |
1444 | unsigned long live_blocks; |
1445 | int ret; | |
1446 | ||
1447 | list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) { | |
9ff05123 RK |
1448 | live_blocks = segbuf->sb_sum.nblocks + |
1449 | (segbuf->sb_pseg_start - segbuf->sb_fseg_start); | |
071ec54d RK |
1450 | ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum, |
1451 | live_blocks, | |
1452 | sci->sc_seg_ctime); | |
1453 | WARN_ON(ret); /* always succeed because the segusage is dirty */ | |
9ff05123 RK |
1454 | } |
1455 | } | |
1456 | ||
a694291a | 1457 | static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile) |
9ff05123 RK |
1458 | { |
1459 | struct nilfs_segment_buffer *segbuf; | |
9ff05123 RK |
1460 | int ret; |
1461 | ||
a694291a | 1462 | segbuf = NILFS_FIRST_SEGBUF(logs); |
071ec54d RK |
1463 | ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum, |
1464 | segbuf->sb_pseg_start - | |
1465 | segbuf->sb_fseg_start, 0); | |
1466 | WARN_ON(ret); /* always succeed because the segusage is dirty */ | |
9ff05123 | 1467 | |
a694291a | 1468 | list_for_each_entry_continue(segbuf, logs, sb_list) { |
071ec54d RK |
1469 | ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum, |
1470 | 0, 0); | |
1f5abe7e | 1471 | WARN_ON(ret); /* always succeed */ |
9ff05123 RK |
1472 | } |
1473 | } | |
1474 | ||
1475 | static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci, | |
1476 | struct nilfs_segment_buffer *last, | |
1477 | struct inode *sufile) | |
1478 | { | |
e29df395 | 1479 | struct nilfs_segment_buffer *segbuf = last; |
9ff05123 RK |
1480 | int ret; |
1481 | ||
e29df395 | 1482 | list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) { |
9ff05123 RK |
1483 | sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks; |
1484 | ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum); | |
1f5abe7e | 1485 | WARN_ON(ret); |
9ff05123 | 1486 | } |
e29df395 | 1487 | nilfs_truncate_logs(&sci->sc_segbufs, last); |
9ff05123 RK |
1488 | } |
1489 | ||
1490 | ||
1491 | static int nilfs_segctor_collect(struct nilfs_sc_info *sci, | |
1492 | struct the_nilfs *nilfs, int mode) | |
1493 | { | |
1494 | struct nilfs_cstage prev_stage = sci->sc_stage; | |
1495 | int err, nadd = 1; | |
1496 | ||
1497 | /* Collection retry loop */ | |
1498 | for (;;) { | |
9ff05123 RK |
1499 | sci->sc_nblk_this_inc = 0; |
1500 | sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs); | |
1501 | ||
1502 | err = nilfs_segctor_reset_segment_buffer(sci); | |
1503 | if (unlikely(err)) | |
1504 | goto failed; | |
1505 | ||
1506 | err = nilfs_segctor_collect_blocks(sci, mode); | |
1507 | sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks; | |
1508 | if (!err) | |
1509 | break; | |
1510 | ||
1511 | if (unlikely(err != -E2BIG)) | |
1512 | goto failed; | |
1513 | ||
1514 | /* The current segment is filled up */ | |
58497703 HM |
1515 | if (mode != SC_LSEG_SR || |
1516 | nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE) | |
9ff05123 RK |
1517 | break; |
1518 | ||
2d8428ac RK |
1519 | nilfs_clear_logs(&sci->sc_segbufs); |
1520 | ||
071cb4b8 RK |
1521 | if (sci->sc_stage.flags & NILFS_CF_SUFREED) { |
1522 | err = nilfs_sufile_cancel_freev(nilfs->ns_sufile, | |
1523 | sci->sc_freesegs, | |
1524 | sci->sc_nfreesegs, | |
1525 | NULL); | |
1526 | WARN_ON(err); /* do not happen */ | |
70f2fe3a | 1527 | sci->sc_stage.flags &= ~NILFS_CF_SUFREED; |
071cb4b8 | 1528 | } |
70f2fe3a AR |
1529 | |
1530 | err = nilfs_segctor_extend_segments(sci, nilfs, nadd); | |
1531 | if (unlikely(err)) | |
1532 | return err; | |
1533 | ||
9ff05123 RK |
1534 | nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA); |
1535 | sci->sc_stage = prev_stage; | |
1536 | } | |
1537 | nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile); | |
1538 | return 0; | |
1539 | ||
1540 | failed: | |
1541 | return err; | |
1542 | } | |
1543 | ||
1544 | static void nilfs_list_replace_buffer(struct buffer_head *old_bh, | |
1545 | struct buffer_head *new_bh) | |
1546 | { | |
1547 | BUG_ON(!list_empty(&new_bh->b_assoc_buffers)); | |
1548 | ||
1549 | list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers); | |
1550 | /* The caller must release old_bh */ | |
1551 | } | |
1552 | ||
1553 | static int | |
1554 | nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci, | |
1555 | struct nilfs_segment_buffer *segbuf, | |
1556 | int mode) | |
1557 | { | |
1558 | struct inode *inode = NULL; | |
1559 | sector_t blocknr; | |
1560 | unsigned long nfinfo = segbuf->sb_sum.nfinfo; | |
1561 | unsigned long nblocks = 0, ndatablk = 0; | |
1c613cb9 | 1562 | const struct nilfs_sc_operations *sc_op = NULL; |
9ff05123 RK |
1563 | struct nilfs_segsum_pointer ssp; |
1564 | struct nilfs_finfo *finfo = NULL; | |
1565 | union nilfs_binfo binfo; | |
1566 | struct buffer_head *bh, *bh_org; | |
1567 | ino_t ino = 0; | |
1568 | int err = 0; | |
1569 | ||
1570 | if (!nfinfo) | |
1571 | goto out; | |
1572 | ||
1573 | blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk; | |
1574 | ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers); | |
1575 | ssp.offset = sizeof(struct nilfs_segment_summary); | |
1576 | ||
1577 | list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) { | |
1e2b68bf | 1578 | if (bh == segbuf->sb_super_root) |
9ff05123 RK |
1579 | break; |
1580 | if (!finfo) { | |
1581 | finfo = nilfs_segctor_map_segsum_entry( | |
1582 | sci, &ssp, sizeof(*finfo)); | |
1583 | ino = le64_to_cpu(finfo->fi_ino); | |
1584 | nblocks = le32_to_cpu(finfo->fi_nblocks); | |
1585 | ndatablk = le32_to_cpu(finfo->fi_ndatablk); | |
1586 | ||
6ad4cd7f | 1587 | inode = bh->b_folio->mapping->host; |
9ff05123 RK |
1588 | |
1589 | if (mode == SC_LSEG_DSYNC) | |
1590 | sc_op = &nilfs_sc_dsync_ops; | |
1591 | else if (ino == NILFS_DAT_INO) | |
1592 | sc_op = &nilfs_sc_dat_ops; | |
1593 | else /* file blocks */ | |
1594 | sc_op = &nilfs_sc_file_ops; | |
1595 | } | |
1596 | bh_org = bh; | |
1597 | get_bh(bh_org); | |
1598 | err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr, | |
1599 | &binfo); | |
1600 | if (bh != bh_org) | |
1601 | nilfs_list_replace_buffer(bh_org, bh); | |
1602 | brelse(bh_org); | |
1603 | if (unlikely(err)) | |
1604 | goto failed_bmap; | |
1605 | ||
1606 | if (ndatablk > 0) | |
1607 | sc_op->write_data_binfo(sci, &ssp, &binfo); | |
1608 | else | |
1609 | sc_op->write_node_binfo(sci, &ssp, &binfo); | |
1610 | ||
1611 | blocknr++; | |
1612 | if (--nblocks == 0) { | |
1613 | finfo = NULL; | |
1614 | if (--nfinfo == 0) | |
1615 | break; | |
1616 | } else if (ndatablk > 0) | |
1617 | ndatablk--; | |
1618 | } | |
1619 | out: | |
1620 | return 0; | |
1621 | ||
1622 | failed_bmap: | |
9ff05123 RK |
1623 | return err; |
1624 | } | |
1625 | ||
1626 | static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode) | |
1627 | { | |
1628 | struct nilfs_segment_buffer *segbuf; | |
1629 | int err; | |
1630 | ||
1631 | list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) { | |
1632 | err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode); | |
1633 | if (unlikely(err)) | |
1634 | return err; | |
1635 | nilfs_segbuf_fill_in_segsum(segbuf); | |
1636 | } | |
1637 | return 0; | |
1638 | } | |
1639 | ||
1cb2d38c | 1640 | static void nilfs_begin_page_io(struct page *page) |
9ff05123 RK |
1641 | { |
1642 | if (!page || PageWriteback(page)) | |
076a378b RK |
1643 | /* |
1644 | * For split b-tree node pages, this function may be called | |
1645 | * twice. We ignore the 2nd or later calls by this check. | |
1646 | */ | |
1cb2d38c | 1647 | return; |
9ff05123 RK |
1648 | |
1649 | lock_page(page); | |
1650 | clear_page_dirty_for_io(page); | |
1651 | set_page_writeback(page); | |
1652 | unlock_page(page); | |
9ff05123 RK |
1653 | } |
1654 | ||
1cb2d38c | 1655 | static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci) |
9ff05123 RK |
1656 | { |
1657 | struct nilfs_segment_buffer *segbuf; | |
1658 | struct page *bd_page = NULL, *fs_page = NULL; | |
9ff05123 | 1659 | |
9ff05123 RK |
1660 | list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) { |
1661 | struct buffer_head *bh; | |
1662 | ||
1663 | list_for_each_entry(bh, &segbuf->sb_segsum_buffers, | |
1664 | b_assoc_buffers) { | |
1665 | if (bh->b_page != bd_page) { | |
1666 | if (bd_page) { | |
1667 | lock_page(bd_page); | |
1668 | clear_page_dirty_for_io(bd_page); | |
1669 | set_page_writeback(bd_page); | |
1670 | unlock_page(bd_page); | |
1671 | } | |
1672 | bd_page = bh->b_page; | |
1673 | } | |
1674 | } | |
1675 | ||
1676 | list_for_each_entry(bh, &segbuf->sb_payload_buffers, | |
1677 | b_assoc_buffers) { | |
7f42ec39 | 1678 | set_buffer_async_write(bh); |
1e2b68bf | 1679 | if (bh == segbuf->sb_super_root) { |
9ff05123 RK |
1680 | if (bh->b_page != bd_page) { |
1681 | lock_page(bd_page); | |
1682 | clear_page_dirty_for_io(bd_page); | |
1683 | set_page_writeback(bd_page); | |
1684 | unlock_page(bd_page); | |
1685 | bd_page = bh->b_page; | |
1686 | } | |
1687 | break; | |
1688 | } | |
1689 | if (bh->b_page != fs_page) { | |
1cb2d38c | 1690 | nilfs_begin_page_io(fs_page); |
9ff05123 RK |
1691 | fs_page = bh->b_page; |
1692 | } | |
1693 | } | |
1694 | } | |
1695 | if (bd_page) { | |
1696 | lock_page(bd_page); | |
1697 | clear_page_dirty_for_io(bd_page); | |
1698 | set_page_writeback(bd_page); | |
1699 | unlock_page(bd_page); | |
1700 | } | |
1cb2d38c | 1701 | nilfs_begin_page_io(fs_page); |
9ff05123 RK |
1702 | } |
1703 | ||
1704 | static int nilfs_segctor_write(struct nilfs_sc_info *sci, | |
9c965bac | 1705 | struct the_nilfs *nilfs) |
9ff05123 | 1706 | { |
d1c6b72a | 1707 | int ret; |
9ff05123 | 1708 | |
d1c6b72a | 1709 | ret = nilfs_write_logs(&sci->sc_segbufs, nilfs); |
a694291a RK |
1710 | list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs); |
1711 | return ret; | |
9ff05123 RK |
1712 | } |
1713 | ||
9ff05123 RK |
1714 | static void nilfs_end_page_io(struct page *page, int err) |
1715 | { | |
1716 | if (!page) | |
1717 | return; | |
1718 | ||
a9777845 | 1719 | if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) { |
8227b297 RK |
1720 | /* |
1721 | * For b-tree node pages, this function may be called twice | |
1722 | * or more because they might be split in a segment. | |
1723 | */ | |
a9777845 RK |
1724 | if (PageDirty(page)) { |
1725 | /* | |
1726 | * For pages holding split b-tree node buffers, dirty | |
1727 | * flag on the buffers may be cleared discretely. | |
1728 | * In that case, the page is once redirtied for | |
1729 | * remaining buffers, and it must be cancelled if | |
1730 | * all the buffers get cleaned later. | |
1731 | */ | |
1732 | lock_page(page); | |
1733 | if (nilfs_page_buffers_clean(page)) | |
1734 | __nilfs_clear_page_dirty(page); | |
1735 | unlock_page(page); | |
1736 | } | |
9ff05123 | 1737 | return; |
a9777845 | 1738 | } |
9ff05123 | 1739 | |
1cb2d38c RK |
1740 | if (!err) { |
1741 | if (!nilfs_page_buffers_clean(page)) | |
1742 | __set_page_dirty_nobuffers(page); | |
1743 | ClearPageError(page); | |
1744 | } else { | |
1745 | __set_page_dirty_nobuffers(page); | |
1746 | SetPageError(page); | |
9ff05123 | 1747 | } |
1cb2d38c RK |
1748 | |
1749 | end_page_writeback(page); | |
9ff05123 RK |
1750 | } |
1751 | ||
1cb2d38c | 1752 | static void nilfs_abort_logs(struct list_head *logs, int err) |
9ff05123 RK |
1753 | { |
1754 | struct nilfs_segment_buffer *segbuf; | |
1755 | struct page *bd_page = NULL, *fs_page = NULL; | |
a694291a | 1756 | struct buffer_head *bh; |
9ff05123 | 1757 | |
a694291a RK |
1758 | if (list_empty(logs)) |
1759 | return; | |
9ff05123 | 1760 | |
a694291a | 1761 | list_for_each_entry(segbuf, logs, sb_list) { |
9ff05123 RK |
1762 | list_for_each_entry(bh, &segbuf->sb_segsum_buffers, |
1763 | b_assoc_buffers) { | |
1764 | if (bh->b_page != bd_page) { | |
1765 | if (bd_page) | |
1766 | end_page_writeback(bd_page); | |
1767 | bd_page = bh->b_page; | |
1768 | } | |
1769 | } | |
1770 | ||
1771 | list_for_each_entry(bh, &segbuf->sb_payload_buffers, | |
1772 | b_assoc_buffers) { | |
7f42ec39 | 1773 | clear_buffer_async_write(bh); |
1e2b68bf | 1774 | if (bh == segbuf->sb_super_root) { |
9ff05123 RK |
1775 | if (bh->b_page != bd_page) { |
1776 | end_page_writeback(bd_page); | |
1777 | bd_page = bh->b_page; | |
1778 | } | |
1779 | break; | |
1780 | } | |
1781 | if (bh->b_page != fs_page) { | |
1782 | nilfs_end_page_io(fs_page, err); | |
9ff05123 RK |
1783 | fs_page = bh->b_page; |
1784 | } | |
1785 | } | |
1786 | } | |
1787 | if (bd_page) | |
1788 | end_page_writeback(bd_page); | |
1789 | ||
1790 | nilfs_end_page_io(fs_page, err); | |
a694291a RK |
1791 | } |
1792 | ||
1793 | static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci, | |
1794 | struct the_nilfs *nilfs, int err) | |
1795 | { | |
1796 | LIST_HEAD(logs); | |
1797 | int ret; | |
1798 | ||
1799 | list_splice_tail_init(&sci->sc_write_logs, &logs); | |
1800 | ret = nilfs_wait_on_logs(&logs); | |
1cb2d38c | 1801 | nilfs_abort_logs(&logs, ret ? : err); |
a694291a RK |
1802 | |
1803 | list_splice_tail_init(&sci->sc_segbufs, &logs); | |
1804 | nilfs_cancel_segusage(&logs, nilfs->ns_sufile); | |
1805 | nilfs_free_incomplete_logs(&logs, nilfs); | |
a694291a RK |
1806 | |
1807 | if (sci->sc_stage.flags & NILFS_CF_SUFREED) { | |
1808 | ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile, | |
1809 | sci->sc_freesegs, | |
1810 | sci->sc_nfreesegs, | |
1811 | NULL); | |
1812 | WARN_ON(ret); /* do not happen */ | |
1813 | } | |
1814 | ||
1815 | nilfs_destroy_logs(&logs); | |
9ff05123 RK |
1816 | } |
1817 | ||
1818 | static void nilfs_set_next_segment(struct the_nilfs *nilfs, | |
1819 | struct nilfs_segment_buffer *segbuf) | |
1820 | { | |
1821 | nilfs->ns_segnum = segbuf->sb_segnum; | |
1822 | nilfs->ns_nextnum = segbuf->sb_nextnum; | |
1823 | nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start | |
1824 | + segbuf->sb_sum.nblocks; | |
1825 | nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq; | |
1826 | nilfs->ns_ctime = segbuf->sb_sum.ctime; | |
1827 | } | |
1828 | ||
1829 | static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci) | |
1830 | { | |
1831 | struct nilfs_segment_buffer *segbuf; | |
1832 | struct page *bd_page = NULL, *fs_page = NULL; | |
e3154e97 | 1833 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
1e2b68bf | 1834 | int update_sr = false; |
9ff05123 | 1835 | |
a694291a | 1836 | list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) { |
9ff05123 RK |
1837 | struct buffer_head *bh; |
1838 | ||
1839 | list_for_each_entry(bh, &segbuf->sb_segsum_buffers, | |
1840 | b_assoc_buffers) { | |
1841 | set_buffer_uptodate(bh); | |
1842 | clear_buffer_dirty(bh); | |
1843 | if (bh->b_page != bd_page) { | |
1844 | if (bd_page) | |
1845 | end_page_writeback(bd_page); | |
1846 | bd_page = bh->b_page; | |
1847 | } | |
1848 | } | |
1849 | /* | |
1850 | * We assume that the buffers which belong to the same page | |
1851 | * continue over the buffer list. | |
1852 | * Under this assumption, the last BHs of pages is | |
1853 | * identifiable by the discontinuity of bh->b_page | |
1854 | * (page != fs_page). | |
1855 | * | |
1856 | * For B-tree node blocks, however, this assumption is not | |
1857 | * guaranteed. The cleanup code of B-tree node pages needs | |
1858 | * special care. | |
1859 | */ | |
1860 | list_for_each_entry(bh, &segbuf->sb_payload_buffers, | |
1861 | b_assoc_buffers) { | |
4ce5c342 | 1862 | const unsigned long set_bits = BIT(BH_Uptodate); |
ead8ecff | 1863 | const unsigned long clear_bits = |
4ce5c342 RK |
1864 | (BIT(BH_Dirty) | BIT(BH_Async_Write) | |
1865 | BIT(BH_Delay) | BIT(BH_NILFS_Volatile) | | |
1866 | BIT(BH_NILFS_Redirected)); | |
ead8ecff RK |
1867 | |
1868 | set_mask_bits(&bh->b_state, clear_bits, set_bits); | |
1e2b68bf | 1869 | if (bh == segbuf->sb_super_root) { |
9ff05123 RK |
1870 | if (bh->b_page != bd_page) { |
1871 | end_page_writeback(bd_page); | |
1872 | bd_page = bh->b_page; | |
1873 | } | |
1e2b68bf | 1874 | update_sr = true; |
9ff05123 RK |
1875 | break; |
1876 | } | |
1877 | if (bh->b_page != fs_page) { | |
1878 | nilfs_end_page_io(fs_page, 0); | |
1879 | fs_page = bh->b_page; | |
1880 | } | |
1881 | } | |
1882 | ||
4762077c RK |
1883 | if (!nilfs_segbuf_simplex(segbuf)) { |
1884 | if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) { | |
9ff05123 RK |
1885 | set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags); |
1886 | sci->sc_lseg_stime = jiffies; | |
1887 | } | |
4762077c | 1888 | if (segbuf->sb_sum.flags & NILFS_SS_LOGEND) |
9ff05123 RK |
1889 | clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags); |
1890 | } | |
1891 | } | |
1892 | /* | |
1893 | * Since pages may continue over multiple segment buffers, | |
1894 | * end of the last page must be checked outside of the loop. | |
1895 | */ | |
1896 | if (bd_page) | |
1897 | end_page_writeback(bd_page); | |
1898 | ||
1899 | nilfs_end_page_io(fs_page, 0); | |
1900 | ||
9ff05123 RK |
1901 | nilfs_drop_collected_inodes(&sci->sc_dirty_files); |
1902 | ||
c1c1d709 | 1903 | if (nilfs_doing_gc()) |
9ff05123 | 1904 | nilfs_drop_collected_inodes(&sci->sc_gc_inodes); |
c1c1d709 | 1905 | else |
9ff05123 | 1906 | nilfs->ns_nongc_ctime = sci->sc_seg_ctime; |
9ff05123 RK |
1907 | |
1908 | sci->sc_nblk_inc += sci->sc_nblk_this_inc; | |
1909 | ||
a694291a | 1910 | segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs); |
9ff05123 RK |
1911 | nilfs_set_next_segment(nilfs, segbuf); |
1912 | ||
1913 | if (update_sr) { | |
e2c7617a | 1914 | nilfs->ns_flushed_device = 0; |
9ff05123 | 1915 | nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start, |
e339ad31 | 1916 | segbuf->sb_sum.seg_seq, nilfs->ns_cno++); |
9ff05123 | 1917 | |
c96fa464 | 1918 | clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags); |
9ff05123 RK |
1919 | clear_bit(NILFS_SC_DIRTY, &sci->sc_flags); |
1920 | set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags); | |
a694291a | 1921 | nilfs_segctor_clear_metadata_dirty(sci); |
9ff05123 RK |
1922 | } else |
1923 | clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags); | |
1924 | } | |
1925 | ||
a694291a RK |
1926 | static int nilfs_segctor_wait(struct nilfs_sc_info *sci) |
1927 | { | |
1928 | int ret; | |
1929 | ||
1930 | ret = nilfs_wait_on_logs(&sci->sc_write_logs); | |
1931 | if (!ret) { | |
1932 | nilfs_segctor_complete_write(sci); | |
1933 | nilfs_destroy_logs(&sci->sc_write_logs); | |
1934 | } | |
1935 | return ret; | |
1936 | } | |
1937 | ||
693dd321 RK |
1938 | static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci, |
1939 | struct the_nilfs *nilfs) | |
9ff05123 RK |
1940 | { |
1941 | struct nilfs_inode_info *ii, *n; | |
e912a5b6 | 1942 | struct inode *ifile = sci->sc_root->ifile; |
9ff05123 | 1943 | |
693dd321 | 1944 | spin_lock(&nilfs->ns_inode_lock); |
9ff05123 | 1945 | retry: |
693dd321 | 1946 | list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) { |
9ff05123 RK |
1947 | if (!ii->i_bh) { |
1948 | struct buffer_head *ibh; | |
1949 | int err; | |
1950 | ||
693dd321 | 1951 | spin_unlock(&nilfs->ns_inode_lock); |
9ff05123 | 1952 | err = nilfs_ifile_get_inode_block( |
e912a5b6 | 1953 | ifile, ii->vfs_inode.i_ino, &ibh); |
9ff05123 | 1954 | if (unlikely(err)) { |
a1d0747a JP |
1955 | nilfs_warn(sci->sc_super, |
1956 | "log writer: error %d getting inode block (ino=%lu)", | |
1957 | err, ii->vfs_inode.i_ino); | |
9ff05123 RK |
1958 | return err; |
1959 | } | |
693dd321 | 1960 | spin_lock(&nilfs->ns_inode_lock); |
9ff05123 RK |
1961 | if (likely(!ii->i_bh)) |
1962 | ii->i_bh = ibh; | |
1963 | else | |
1964 | brelse(ibh); | |
1965 | goto retry; | |
1966 | } | |
9ff05123 | 1967 | |
31ccb1f7 AR |
1968 | // Always redirty the buffer to avoid race condition |
1969 | mark_buffer_dirty(ii->i_bh); | |
1970 | nilfs_mdt_mark_dirty(ifile); | |
1971 | ||
9ff05123 RK |
1972 | clear_bit(NILFS_I_QUEUED, &ii->i_state); |
1973 | set_bit(NILFS_I_BUSY, &ii->i_state); | |
eaae0f37 | 1974 | list_move_tail(&ii->i_dirty, &sci->sc_dirty_files); |
9ff05123 | 1975 | } |
693dd321 | 1976 | spin_unlock(&nilfs->ns_inode_lock); |
9ff05123 | 1977 | |
9ff05123 RK |
1978 | return 0; |
1979 | } | |
1980 | ||
693dd321 RK |
1981 | static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci, |
1982 | struct the_nilfs *nilfs) | |
9ff05123 | 1983 | { |
9ff05123 | 1984 | struct nilfs_inode_info *ii, *n; |
1751e8a6 | 1985 | int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE); |
7ef3ff2f | 1986 | int defer_iput = false; |
9ff05123 | 1987 | |
693dd321 | 1988 | spin_lock(&nilfs->ns_inode_lock); |
9ff05123 RK |
1989 | list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) { |
1990 | if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) || | |
6c43f410 | 1991 | test_bit(NILFS_I_DIRTY, &ii->i_state)) |
9ff05123 | 1992 | continue; |
6c43f410 | 1993 | |
9ff05123 RK |
1994 | clear_bit(NILFS_I_BUSY, &ii->i_state); |
1995 | brelse(ii->i_bh); | |
1996 | ii->i_bh = NULL; | |
7ef3ff2f | 1997 | list_del_init(&ii->i_dirty); |
283ee148 | 1998 | if (!ii->vfs_inode.i_nlink || during_mount) { |
7ef3ff2f | 1999 | /* |
283ee148 RK |
2000 | * Defer calling iput() to avoid deadlocks if |
2001 | * i_nlink == 0 or mount is not yet finished. | |
7ef3ff2f RK |
2002 | */ |
2003 | list_add_tail(&ii->i_dirty, &sci->sc_iput_queue); | |
2004 | defer_iput = true; | |
2005 | } else { | |
2006 | spin_unlock(&nilfs->ns_inode_lock); | |
2007 | iput(&ii->vfs_inode); | |
2008 | spin_lock(&nilfs->ns_inode_lock); | |
2009 | } | |
9ff05123 | 2010 | } |
693dd321 | 2011 | spin_unlock(&nilfs->ns_inode_lock); |
7ef3ff2f RK |
2012 | |
2013 | if (defer_iput) | |
2014 | schedule_work(&sci->sc_iput_work); | |
9ff05123 RK |
2015 | } |
2016 | ||
9ff05123 RK |
2017 | /* |
2018 | * Main procedure of segment constructor | |
2019 | */ | |
2020 | static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode) | |
2021 | { | |
e3154e97 | 2022 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
1e2b68bf | 2023 | int err; |
9ff05123 | 2024 | |
58497703 | 2025 | nilfs_sc_cstage_set(sci, NILFS_ST_INIT); |
6c43f410 | 2026 | sci->sc_cno = nilfs->ns_cno; |
9ff05123 | 2027 | |
693dd321 | 2028 | err = nilfs_segctor_collect_dirty_files(sci, nilfs); |
9ff05123 RK |
2029 | if (unlikely(err)) |
2030 | goto out; | |
2031 | ||
e912a5b6 | 2032 | if (nilfs_test_metadata_dirty(nilfs, sci->sc_root)) |
9ff05123 RK |
2033 | set_bit(NILFS_SC_DIRTY, &sci->sc_flags); |
2034 | ||
2035 | if (nilfs_segctor_clean(sci)) | |
2036 | goto out; | |
2037 | ||
2038 | do { | |
2039 | sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK; | |
2040 | ||
2041 | err = nilfs_segctor_begin_construction(sci, nilfs); | |
2042 | if (unlikely(err)) | |
2043 | goto out; | |
2044 | ||
2045 | /* Update time stamp */ | |
fb04b91b | 2046 | sci->sc_seg_ctime = ktime_get_real_seconds(); |
9ff05123 RK |
2047 | |
2048 | err = nilfs_segctor_collect(sci, nilfs, mode); | |
2049 | if (unlikely(err)) | |
2050 | goto failed; | |
2051 | ||
9ff05123 | 2052 | /* Avoid empty segment */ |
58497703 | 2053 | if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE && |
4762077c | 2054 | nilfs_segbuf_empty(sci->sc_curseg)) { |
a694291a | 2055 | nilfs_segctor_abort_construction(sci, nilfs, 1); |
9ff05123 RK |
2056 | goto out; |
2057 | } | |
2058 | ||
2059 | err = nilfs_segctor_assign(sci, mode); | |
2060 | if (unlikely(err)) | |
2061 | goto failed; | |
2062 | ||
9ff05123 | 2063 | if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED) |
e912a5b6 | 2064 | nilfs_segctor_fill_in_file_bmap(sci); |
9ff05123 | 2065 | |
1e2b68bf | 2066 | if (mode == SC_LSEG_SR && |
58497703 | 2067 | nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) { |
9ff05123 RK |
2068 | err = nilfs_segctor_fill_in_checkpoint(sci); |
2069 | if (unlikely(err)) | |
a694291a | 2070 | goto failed_to_write; |
9ff05123 RK |
2071 | |
2072 | nilfs_segctor_fill_in_super_root(sci, nilfs); | |
2073 | } | |
2074 | nilfs_segctor_update_segusage(sci, nilfs->ns_sufile); | |
2075 | ||
2076 | /* Write partial segments */ | |
1cb2d38c | 2077 | nilfs_segctor_prepare_write(sci); |
aaed1d5b RK |
2078 | |
2079 | nilfs_add_checksums_on_logs(&sci->sc_segbufs, | |
2080 | nilfs->ns_crc_seed); | |
9ff05123 | 2081 | |
9c965bac | 2082 | err = nilfs_segctor_write(sci, nilfs); |
9ff05123 RK |
2083 | if (unlikely(err)) |
2084 | goto failed_to_write; | |
2085 | ||
58497703 | 2086 | if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE || |
09cbfeaf | 2087 | nilfs->ns_blocksize_bits != PAGE_SHIFT) { |
a694291a RK |
2088 | /* |
2089 | * At this point, we avoid double buffering | |
2090 | * for blocksize < pagesize because page dirty | |
2091 | * flag is turned off during write and dirty | |
2092 | * buffers are not properly collected for | |
2093 | * pages crossing over segments. | |
2094 | */ | |
2095 | err = nilfs_segctor_wait(sci); | |
2096 | if (err) | |
2097 | goto failed_to_write; | |
2098 | } | |
58497703 | 2099 | } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE); |
9ff05123 | 2100 | |
9ff05123 | 2101 | out: |
693dd321 | 2102 | nilfs_segctor_drop_written_files(sci, nilfs); |
9ff05123 RK |
2103 | return err; |
2104 | ||
2105 | failed_to_write: | |
9ff05123 RK |
2106 | if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED) |
2107 | nilfs_redirty_inodes(&sci->sc_dirty_files); | |
9ff05123 RK |
2108 | |
2109 | failed: | |
2110 | if (nilfs_doing_gc()) | |
2111 | nilfs_redirty_inodes(&sci->sc_gc_inodes); | |
a694291a | 2112 | nilfs_segctor_abort_construction(sci, nilfs, err); |
9ff05123 RK |
2113 | goto out; |
2114 | } | |
2115 | ||
2116 | /** | |
9ccf56c1 | 2117 | * nilfs_segctor_start_timer - set timer of background write |
9ff05123 RK |
2118 | * @sci: nilfs_sc_info |
2119 | * | |
2120 | * If the timer has already been set, it ignores the new request. | |
2121 | * This function MUST be called within a section locking the segment | |
2122 | * semaphore. | |
2123 | */ | |
2124 | static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci) | |
2125 | { | |
2126 | spin_lock(&sci->sc_state_lock); | |
fdce895e LH |
2127 | if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) { |
2128 | sci->sc_timer.expires = jiffies + sci->sc_interval; | |
2129 | add_timer(&sci->sc_timer); | |
9ff05123 RK |
2130 | sci->sc_state |= NILFS_SEGCTOR_COMMIT; |
2131 | } | |
2132 | spin_unlock(&sci->sc_state_lock); | |
2133 | } | |
2134 | ||
2135 | static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn) | |
2136 | { | |
2137 | spin_lock(&sci->sc_state_lock); | |
4ce5c342 | 2138 | if (!(sci->sc_flush_request & BIT(bn))) { |
9ff05123 RK |
2139 | unsigned long prev_req = sci->sc_flush_request; |
2140 | ||
4ce5c342 | 2141 | sci->sc_flush_request |= BIT(bn); |
9ff05123 RK |
2142 | if (!prev_req) |
2143 | wake_up(&sci->sc_wait_daemon); | |
2144 | } | |
2145 | spin_unlock(&sci->sc_state_lock); | |
2146 | } | |
2147 | ||
2148 | /** | |
2149 | * nilfs_flush_segment - trigger a segment construction for resource control | |
2150 | * @sb: super block | |
2151 | * @ino: inode number of the file to be flushed out. | |
2152 | */ | |
2153 | void nilfs_flush_segment(struct super_block *sb, ino_t ino) | |
2154 | { | |
e3154e97 | 2155 | struct the_nilfs *nilfs = sb->s_fs_info; |
3fd3fe5a | 2156 | struct nilfs_sc_info *sci = nilfs->ns_writer; |
9ff05123 RK |
2157 | |
2158 | if (!sci || nilfs_doing_construction()) | |
2159 | return; | |
2160 | nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0); | |
2161 | /* assign bit 0 to data files */ | |
2162 | } | |
2163 | ||
9ff05123 | 2164 | struct nilfs_segctor_wait_request { |
ac6424b9 | 2165 | wait_queue_entry_t wq; |
9ff05123 RK |
2166 | __u32 seq; |
2167 | int err; | |
2168 | atomic_t done; | |
2169 | }; | |
2170 | ||
2171 | static int nilfs_segctor_sync(struct nilfs_sc_info *sci) | |
2172 | { | |
2173 | struct nilfs_segctor_wait_request wait_req; | |
2174 | int err = 0; | |
2175 | ||
2176 | spin_lock(&sci->sc_state_lock); | |
2177 | init_wait(&wait_req.wq); | |
2178 | wait_req.err = 0; | |
2179 | atomic_set(&wait_req.done, 0); | |
2180 | wait_req.seq = ++sci->sc_seq_request; | |
2181 | spin_unlock(&sci->sc_state_lock); | |
2182 | ||
2183 | init_waitqueue_entry(&wait_req.wq, current); | |
2184 | add_wait_queue(&sci->sc_wait_request, &wait_req.wq); | |
2185 | set_current_state(TASK_INTERRUPTIBLE); | |
2186 | wake_up(&sci->sc_wait_daemon); | |
2187 | ||
2188 | for (;;) { | |
2189 | if (atomic_read(&wait_req.done)) { | |
2190 | err = wait_req.err; | |
2191 | break; | |
2192 | } | |
2193 | if (!signal_pending(current)) { | |
2194 | schedule(); | |
2195 | continue; | |
2196 | } | |
2197 | err = -ERESTARTSYS; | |
2198 | break; | |
2199 | } | |
2200 | finish_wait(&sci->sc_wait_request, &wait_req.wq); | |
2201 | return err; | |
2202 | } | |
2203 | ||
2204 | static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err) | |
2205 | { | |
2206 | struct nilfs_segctor_wait_request *wrq, *n; | |
2207 | unsigned long flags; | |
2208 | ||
2209 | spin_lock_irqsave(&sci->sc_wait_request.lock, flags); | |
2055da97 | 2210 | list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) { |
9ff05123 RK |
2211 | if (!atomic_read(&wrq->done) && |
2212 | nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) { | |
2213 | wrq->err = err; | |
2214 | atomic_set(&wrq->done, 1); | |
2215 | } | |
2216 | if (atomic_read(&wrq->done)) { | |
2217 | wrq->wq.func(&wrq->wq, | |
2218 | TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
2219 | 0, NULL); | |
2220 | } | |
2221 | } | |
2222 | spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags); | |
2223 | } | |
2224 | ||
2225 | /** | |
2226 | * nilfs_construct_segment - construct a logical segment | |
2227 | * @sb: super block | |
2228 | * | |
300563e6 | 2229 | * Return Value: On success, 0 is returned. On errors, one of the following |
9ff05123 RK |
2230 | * negative error code is returned. |
2231 | * | |
2232 | * %-EROFS - Read only filesystem. | |
2233 | * | |
2234 | * %-EIO - I/O error | |
2235 | * | |
2236 | * %-ENOSPC - No space left on device (only in a panic state). | |
2237 | * | |
2238 | * %-ERESTARTSYS - Interrupted. | |
2239 | * | |
2240 | * %-ENOMEM - Insufficient memory available. | |
2241 | */ | |
2242 | int nilfs_construct_segment(struct super_block *sb) | |
2243 | { | |
e3154e97 | 2244 | struct the_nilfs *nilfs = sb->s_fs_info; |
3fd3fe5a | 2245 | struct nilfs_sc_info *sci = nilfs->ns_writer; |
9ff05123 | 2246 | struct nilfs_transaction_info *ti; |
9ff05123 | 2247 | |
8cccf05f | 2248 | if (sb_rdonly(sb) || unlikely(!sci)) |
9ff05123 RK |
2249 | return -EROFS; |
2250 | ||
2251 | /* A call inside transactions causes a deadlock. */ | |
2252 | BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC); | |
2253 | ||
da6f7916 | 2254 | return nilfs_segctor_sync(sci); |
9ff05123 RK |
2255 | } |
2256 | ||
2257 | /** | |
2258 | * nilfs_construct_dsync_segment - construct a data-only logical segment | |
2259 | * @sb: super block | |
f30bf3e4 RK |
2260 | * @inode: inode whose data blocks should be written out |
2261 | * @start: start byte offset | |
2262 | * @end: end byte offset (inclusive) | |
9ff05123 | 2263 | * |
300563e6 | 2264 | * Return Value: On success, 0 is returned. On errors, one of the following |
9ff05123 RK |
2265 | * negative error code is returned. |
2266 | * | |
2267 | * %-EROFS - Read only filesystem. | |
2268 | * | |
2269 | * %-EIO - I/O error | |
2270 | * | |
2271 | * %-ENOSPC - No space left on device (only in a panic state). | |
2272 | * | |
2273 | * %-ERESTARTSYS - Interrupted. | |
2274 | * | |
2275 | * %-ENOMEM - Insufficient memory available. | |
2276 | */ | |
f30bf3e4 RK |
2277 | int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode, |
2278 | loff_t start, loff_t end) | |
9ff05123 | 2279 | { |
e3154e97 | 2280 | struct the_nilfs *nilfs = sb->s_fs_info; |
3fd3fe5a | 2281 | struct nilfs_sc_info *sci = nilfs->ns_writer; |
9ff05123 RK |
2282 | struct nilfs_inode_info *ii; |
2283 | struct nilfs_transaction_info ti; | |
2284 | int err = 0; | |
2285 | ||
8cccf05f | 2286 | if (sb_rdonly(sb) || unlikely(!sci)) |
9ff05123 RK |
2287 | return -EROFS; |
2288 | ||
f7545144 | 2289 | nilfs_transaction_lock(sb, &ti, 0); |
9ff05123 RK |
2290 | |
2291 | ii = NILFS_I(inode); | |
b9f66140 | 2292 | if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) || |
3b2ce58b | 2293 | nilfs_test_opt(nilfs, STRICT_ORDER) || |
9ff05123 | 2294 | test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) || |
3b2ce58b | 2295 | nilfs_discontinued(nilfs)) { |
f7545144 | 2296 | nilfs_transaction_unlock(sb); |
9ff05123 RK |
2297 | err = nilfs_segctor_sync(sci); |
2298 | return err; | |
2299 | } | |
2300 | ||
693dd321 | 2301 | spin_lock(&nilfs->ns_inode_lock); |
9ff05123 RK |
2302 | if (!test_bit(NILFS_I_QUEUED, &ii->i_state) && |
2303 | !test_bit(NILFS_I_BUSY, &ii->i_state)) { | |
693dd321 | 2304 | spin_unlock(&nilfs->ns_inode_lock); |
f7545144 | 2305 | nilfs_transaction_unlock(sb); |
9ff05123 RK |
2306 | return 0; |
2307 | } | |
693dd321 | 2308 | spin_unlock(&nilfs->ns_inode_lock); |
f30bf3e4 RK |
2309 | sci->sc_dsync_inode = ii; |
2310 | sci->sc_dsync_start = start; | |
2311 | sci->sc_dsync_end = end; | |
9ff05123 RK |
2312 | |
2313 | err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC); | |
e2c7617a AR |
2314 | if (!err) |
2315 | nilfs->ns_flushed_device = 0; | |
9ff05123 | 2316 | |
f7545144 | 2317 | nilfs_transaction_unlock(sb); |
9ff05123 RK |
2318 | return err; |
2319 | } | |
2320 | ||
9ff05123 | 2321 | #define FLUSH_FILE_BIT (0x1) /* data file only */ |
4ce5c342 | 2322 | #define FLUSH_DAT_BIT BIT(NILFS_DAT_INO) /* DAT only */ |
9ff05123 | 2323 | |
dcd76186 RK |
2324 | /** |
2325 | * nilfs_segctor_accept - record accepted sequence count of log-write requests | |
2326 | * @sci: segment constructor object | |
2327 | */ | |
2328 | static void nilfs_segctor_accept(struct nilfs_sc_info *sci) | |
9ff05123 | 2329 | { |
9ff05123 | 2330 | spin_lock(&sci->sc_state_lock); |
dcd76186 | 2331 | sci->sc_seq_accepted = sci->sc_seq_request; |
9ff05123 | 2332 | spin_unlock(&sci->sc_state_lock); |
fdce895e | 2333 | del_timer_sync(&sci->sc_timer); |
9ff05123 RK |
2334 | } |
2335 | ||
dcd76186 RK |
2336 | /** |
2337 | * nilfs_segctor_notify - notify the result of request to caller threads | |
2338 | * @sci: segment constructor object | |
2339 | * @mode: mode of log forming | |
2340 | * @err: error code to be notified | |
2341 | */ | |
2342 | static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err) | |
9ff05123 RK |
2343 | { |
2344 | /* Clear requests (even when the construction failed) */ | |
2345 | spin_lock(&sci->sc_state_lock); | |
2346 | ||
dcd76186 | 2347 | if (mode == SC_LSEG_SR) { |
aeda7f63 | 2348 | sci->sc_state &= ~NILFS_SEGCTOR_COMMIT; |
dcd76186 RK |
2349 | sci->sc_seq_done = sci->sc_seq_accepted; |
2350 | nilfs_segctor_wakeup(sci, err); | |
9ff05123 | 2351 | sci->sc_flush_request = 0; |
aeda7f63 | 2352 | } else { |
dcd76186 | 2353 | if (mode == SC_FLUSH_FILE) |
aeda7f63 | 2354 | sci->sc_flush_request &= ~FLUSH_FILE_BIT; |
dcd76186 | 2355 | else if (mode == SC_FLUSH_DAT) |
aeda7f63 RK |
2356 | sci->sc_flush_request &= ~FLUSH_DAT_BIT; |
2357 | ||
2358 | /* re-enable timer if checkpoint creation was not done */ | |
fdce895e LH |
2359 | if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) && |
2360 | time_before(jiffies, sci->sc_timer.expires)) | |
2361 | add_timer(&sci->sc_timer); | |
aeda7f63 | 2362 | } |
9ff05123 RK |
2363 | spin_unlock(&sci->sc_state_lock); |
2364 | } | |
2365 | ||
dcd76186 RK |
2366 | /** |
2367 | * nilfs_segctor_construct - form logs and write them to disk | |
2368 | * @sci: segment constructor object | |
2369 | * @mode: mode of log forming | |
2370 | */ | |
2371 | static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode) | |
9ff05123 | 2372 | { |
e3154e97 | 2373 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
d26493b6 | 2374 | struct nilfs_super_block **sbp; |
9ff05123 RK |
2375 | int err = 0; |
2376 | ||
dcd76186 RK |
2377 | nilfs_segctor_accept(sci); |
2378 | ||
9ff05123 | 2379 | if (nilfs_discontinued(nilfs)) |
dcd76186 RK |
2380 | mode = SC_LSEG_SR; |
2381 | if (!nilfs_segctor_confirm(sci)) | |
2382 | err = nilfs_segctor_do_construct(sci, mode); | |
2383 | ||
9ff05123 | 2384 | if (likely(!err)) { |
dcd76186 | 2385 | if (mode != SC_FLUSH_DAT) |
9ff05123 RK |
2386 | atomic_set(&nilfs->ns_ndirtyblks, 0); |
2387 | if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) && | |
2388 | nilfs_discontinued(nilfs)) { | |
2389 | down_write(&nilfs->ns_sem); | |
d26493b6 | 2390 | err = -EIO; |
f7545144 | 2391 | sbp = nilfs_prepare_super(sci->sc_super, |
b2ac86e1 JS |
2392 | nilfs_sb_will_flip(nilfs)); |
2393 | if (likely(sbp)) { | |
2394 | nilfs_set_log_cursor(sbp[0], nilfs); | |
f7545144 RK |
2395 | err = nilfs_commit_super(sci->sc_super, |
2396 | NILFS_SB_COMMIT); | |
b2ac86e1 | 2397 | } |
9ff05123 RK |
2398 | up_write(&nilfs->ns_sem); |
2399 | } | |
2400 | } | |
dcd76186 RK |
2401 | |
2402 | nilfs_segctor_notify(sci, mode, err); | |
9ff05123 RK |
2403 | return err; |
2404 | } | |
2405 | ||
7554e9c4 | 2406 | static void nilfs_construction_timeout(struct timer_list *t) |
9ff05123 | 2407 | { |
7554e9c4 | 2408 | struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer); |
4ad364ca | 2409 | |
7554e9c4 | 2410 | wake_up_process(sci->sc_timer_task); |
9ff05123 RK |
2411 | } |
2412 | ||
2413 | static void | |
2414 | nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head) | |
2415 | { | |
2416 | struct nilfs_inode_info *ii, *n; | |
2417 | ||
2418 | list_for_each_entry_safe(ii, n, head, i_dirty) { | |
2419 | if (!test_bit(NILFS_I_UPDATED, &ii->i_state)) | |
2420 | continue; | |
9ff05123 | 2421 | list_del_init(&ii->i_dirty); |
fbb24a3a | 2422 | truncate_inode_pages(&ii->vfs_inode.i_data, 0); |
e897be17 | 2423 | nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping); |
263d90ce | 2424 | iput(&ii->vfs_inode); |
9ff05123 RK |
2425 | } |
2426 | } | |
2427 | ||
4f6b8288 RK |
2428 | int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv, |
2429 | void **kbufs) | |
9ff05123 | 2430 | { |
e3154e97 | 2431 | struct the_nilfs *nilfs = sb->s_fs_info; |
3fd3fe5a | 2432 | struct nilfs_sc_info *sci = nilfs->ns_writer; |
9ff05123 | 2433 | struct nilfs_transaction_info ti; |
9ff05123 RK |
2434 | int err; |
2435 | ||
2436 | if (unlikely(!sci)) | |
2437 | return -EROFS; | |
2438 | ||
f7545144 | 2439 | nilfs_transaction_lock(sb, &ti, 1); |
9ff05123 | 2440 | |
c1c1d709 | 2441 | err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat); |
9ff05123 RK |
2442 | if (unlikely(err)) |
2443 | goto out_unlock; | |
071cb4b8 | 2444 | |
4f6b8288 | 2445 | err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs); |
c1c1d709 RK |
2446 | if (unlikely(err)) { |
2447 | nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat); | |
9ff05123 | 2448 | goto out_unlock; |
c1c1d709 | 2449 | } |
9ff05123 | 2450 | |
071cb4b8 RK |
2451 | sci->sc_freesegs = kbufs[4]; |
2452 | sci->sc_nfreesegs = argv[4].v_nmembs; | |
0935db74 | 2453 | list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes); |
9ff05123 RK |
2454 | |
2455 | for (;;) { | |
dcd76186 | 2456 | err = nilfs_segctor_construct(sci, SC_LSEG_SR); |
9ff05123 | 2457 | nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes); |
9ff05123 RK |
2458 | |
2459 | if (likely(!err)) | |
2460 | break; | |
2461 | ||
a1d0747a | 2462 | nilfs_warn(sb, "error %d cleaning segments", err); |
9ff05123 RK |
2463 | set_current_state(TASK_INTERRUPTIBLE); |
2464 | schedule_timeout(sci->sc_interval); | |
2465 | } | |
3b2ce58b | 2466 | if (nilfs_test_opt(nilfs, DISCARD)) { |
e902ec99 JS |
2467 | int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs, |
2468 | sci->sc_nfreesegs); | |
2469 | if (ret) { | |
a1d0747a JP |
2470 | nilfs_warn(sb, |
2471 | "error %d on discard request, turning discards off for the device", | |
2472 | ret); | |
3b2ce58b | 2473 | nilfs_clear_opt(nilfs, DISCARD); |
e902ec99 JS |
2474 | } |
2475 | } | |
9ff05123 RK |
2476 | |
2477 | out_unlock: | |
071cb4b8 RK |
2478 | sci->sc_freesegs = NULL; |
2479 | sci->sc_nfreesegs = 0; | |
c1c1d709 | 2480 | nilfs_mdt_clear_shadow_map(nilfs->ns_dat); |
f7545144 | 2481 | nilfs_transaction_unlock(sb); |
9ff05123 RK |
2482 | return err; |
2483 | } | |
2484 | ||
2485 | static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode) | |
2486 | { | |
9ff05123 | 2487 | struct nilfs_transaction_info ti; |
9ff05123 | 2488 | |
f7545144 | 2489 | nilfs_transaction_lock(sci->sc_super, &ti, 0); |
dcd76186 | 2490 | nilfs_segctor_construct(sci, mode); |
9ff05123 RK |
2491 | |
2492 | /* | |
2493 | * Unclosed segment should be retried. We do this using sc_timer. | |
2494 | * Timeout of sc_timer will invoke complete construction which leads | |
2495 | * to close the current logical segment. | |
2496 | */ | |
2497 | if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) | |
2498 | nilfs_segctor_start_timer(sci); | |
2499 | ||
f7545144 | 2500 | nilfs_transaction_unlock(sci->sc_super); |
9ff05123 RK |
2501 | } |
2502 | ||
2503 | static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci) | |
2504 | { | |
2505 | int mode = 0; | |
9ff05123 RK |
2506 | |
2507 | spin_lock(&sci->sc_state_lock); | |
2508 | mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ? | |
2509 | SC_FLUSH_DAT : SC_FLUSH_FILE; | |
2510 | spin_unlock(&sci->sc_state_lock); | |
2511 | ||
2512 | if (mode) { | |
09ef29e0 | 2513 | nilfs_segctor_do_construct(sci, mode); |
9ff05123 RK |
2514 | |
2515 | spin_lock(&sci->sc_state_lock); | |
2516 | sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ? | |
2517 | ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT; | |
2518 | spin_unlock(&sci->sc_state_lock); | |
2519 | } | |
2520 | clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags); | |
2521 | } | |
2522 | ||
2523 | static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci) | |
2524 | { | |
2525 | if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) || | |
2526 | time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) { | |
2527 | if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT)) | |
2528 | return SC_FLUSH_FILE; | |
2529 | else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT)) | |
2530 | return SC_FLUSH_DAT; | |
2531 | } | |
2532 | return SC_LSEG_SR; | |
2533 | } | |
2534 | ||
2535 | /** | |
2536 | * nilfs_segctor_thread - main loop of the segment constructor thread. | |
2537 | * @arg: pointer to a struct nilfs_sc_info. | |
2538 | * | |
2539 | * nilfs_segctor_thread() initializes a timer and serves as a daemon | |
2540 | * to execute segment constructions. | |
2541 | */ | |
2542 | static int nilfs_segctor_thread(void *arg) | |
2543 | { | |
2544 | struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg; | |
e3154e97 | 2545 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
9ff05123 RK |
2546 | int timeout = 0; |
2547 | ||
7554e9c4 | 2548 | sci->sc_timer_task = current; |
9ff05123 RK |
2549 | |
2550 | /* start sync. */ | |
2551 | sci->sc_task = current; | |
2552 | wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */ | |
a1d0747a JP |
2553 | nilfs_info(sci->sc_super, |
2554 | "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds", | |
2555 | sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ); | |
9ff05123 RK |
2556 | |
2557 | spin_lock(&sci->sc_state_lock); | |
2558 | loop: | |
2559 | for (;;) { | |
2560 | int mode; | |
2561 | ||
2562 | if (sci->sc_state & NILFS_SEGCTOR_QUIT) | |
2563 | goto end_thread; | |
2564 | ||
2565 | if (timeout || sci->sc_seq_request != sci->sc_seq_done) | |
2566 | mode = SC_LSEG_SR; | |
7f00184e | 2567 | else if (sci->sc_flush_request) |
9ff05123 | 2568 | mode = nilfs_segctor_flush_mode(sci); |
7f00184e RK |
2569 | else |
2570 | break; | |
9ff05123 RK |
2571 | |
2572 | spin_unlock(&sci->sc_state_lock); | |
2573 | nilfs_segctor_thread_construct(sci, mode); | |
2574 | spin_lock(&sci->sc_state_lock); | |
2575 | timeout = 0; | |
2576 | } | |
2577 | ||
2578 | ||
2579 | if (freezing(current)) { | |
2580 | spin_unlock(&sci->sc_state_lock); | |
a0acae0e | 2581 | try_to_freeze(); |
9ff05123 RK |
2582 | spin_lock(&sci->sc_state_lock); |
2583 | } else { | |
2584 | DEFINE_WAIT(wait); | |
2585 | int should_sleep = 1; | |
2586 | ||
2587 | prepare_to_wait(&sci->sc_wait_daemon, &wait, | |
2588 | TASK_INTERRUPTIBLE); | |
2589 | ||
2590 | if (sci->sc_seq_request != sci->sc_seq_done) | |
2591 | should_sleep = 0; | |
2592 | else if (sci->sc_flush_request) | |
2593 | should_sleep = 0; | |
2594 | else if (sci->sc_state & NILFS_SEGCTOR_COMMIT) | |
2595 | should_sleep = time_before(jiffies, | |
fdce895e | 2596 | sci->sc_timer.expires); |
9ff05123 RK |
2597 | |
2598 | if (should_sleep) { | |
2599 | spin_unlock(&sci->sc_state_lock); | |
2600 | schedule(); | |
2601 | spin_lock(&sci->sc_state_lock); | |
2602 | } | |
2603 | finish_wait(&sci->sc_wait_daemon, &wait); | |
2604 | timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) && | |
fdce895e | 2605 | time_after_eq(jiffies, sci->sc_timer.expires)); |
e605f0a7 RK |
2606 | |
2607 | if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs)) | |
1dfa2710 | 2608 | set_nilfs_discontinued(nilfs); |
9ff05123 RK |
2609 | } |
2610 | goto loop; | |
2611 | ||
2612 | end_thread: | |
2613 | spin_unlock(&sci->sc_state_lock); | |
9ff05123 RK |
2614 | |
2615 | /* end sync. */ | |
2616 | sci->sc_task = NULL; | |
2617 | wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */ | |
2618 | return 0; | |
2619 | } | |
2620 | ||
2621 | static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci) | |
2622 | { | |
2623 | struct task_struct *t; | |
2624 | ||
2625 | t = kthread_run(nilfs_segctor_thread, sci, "segctord"); | |
2626 | if (IS_ERR(t)) { | |
2627 | int err = PTR_ERR(t); | |
2628 | ||
a1d0747a JP |
2629 | nilfs_err(sci->sc_super, "error %d creating segctord thread", |
2630 | err); | |
9ff05123 RK |
2631 | return err; |
2632 | } | |
2633 | wait_event(sci->sc_wait_task, sci->sc_task != NULL); | |
2634 | return 0; | |
2635 | } | |
2636 | ||
2637 | static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci) | |
6b81e14e JS |
2638 | __acquires(&sci->sc_state_lock) |
2639 | __releases(&sci->sc_state_lock) | |
9ff05123 RK |
2640 | { |
2641 | sci->sc_state |= NILFS_SEGCTOR_QUIT; | |
2642 | ||
2643 | while (sci->sc_task) { | |
2644 | wake_up(&sci->sc_wait_daemon); | |
2645 | spin_unlock(&sci->sc_state_lock); | |
2646 | wait_event(sci->sc_wait_task, sci->sc_task == NULL); | |
2647 | spin_lock(&sci->sc_state_lock); | |
2648 | } | |
2649 | } | |
2650 | ||
9ff05123 RK |
2651 | /* |
2652 | * Setup & clean-up functions | |
2653 | */ | |
f7545144 | 2654 | static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb, |
e912a5b6 | 2655 | struct nilfs_root *root) |
9ff05123 | 2656 | { |
e3154e97 | 2657 | struct the_nilfs *nilfs = sb->s_fs_info; |
9ff05123 RK |
2658 | struct nilfs_sc_info *sci; |
2659 | ||
2660 | sci = kzalloc(sizeof(*sci), GFP_KERNEL); | |
2661 | if (!sci) | |
2662 | return NULL; | |
2663 | ||
f7545144 | 2664 | sci->sc_super = sb; |
9ff05123 | 2665 | |
e912a5b6 RK |
2666 | nilfs_get_root(root); |
2667 | sci->sc_root = root; | |
2668 | ||
9ff05123 RK |
2669 | init_waitqueue_head(&sci->sc_wait_request); |
2670 | init_waitqueue_head(&sci->sc_wait_daemon); | |
2671 | init_waitqueue_head(&sci->sc_wait_task); | |
2672 | spin_lock_init(&sci->sc_state_lock); | |
2673 | INIT_LIST_HEAD(&sci->sc_dirty_files); | |
2674 | INIT_LIST_HEAD(&sci->sc_segbufs); | |
a694291a | 2675 | INIT_LIST_HEAD(&sci->sc_write_logs); |
9ff05123 | 2676 | INIT_LIST_HEAD(&sci->sc_gc_inodes); |
7ef3ff2f RK |
2677 | INIT_LIST_HEAD(&sci->sc_iput_queue); |
2678 | INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func); | |
7554e9c4 | 2679 | timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0); |
9ff05123 RK |
2680 | |
2681 | sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT; | |
2682 | sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ; | |
2683 | sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK; | |
2684 | ||
574e6c31 | 2685 | if (nilfs->ns_interval) |
071d73cf | 2686 | sci->sc_interval = HZ * nilfs->ns_interval; |
574e6c31 RK |
2687 | if (nilfs->ns_watermark) |
2688 | sci->sc_watermark = nilfs->ns_watermark; | |
9ff05123 RK |
2689 | return sci; |
2690 | } | |
2691 | ||
2692 | static void nilfs_segctor_write_out(struct nilfs_sc_info *sci) | |
2693 | { | |
2694 | int ret, retrycount = NILFS_SC_CLEANUP_RETRY; | |
2695 | ||
076a378b RK |
2696 | /* |
2697 | * The segctord thread was stopped and its timer was removed. | |
2698 | * But some tasks remain. | |
2699 | */ | |
9ff05123 | 2700 | do { |
9ff05123 | 2701 | struct nilfs_transaction_info ti; |
9ff05123 | 2702 | |
f7545144 | 2703 | nilfs_transaction_lock(sci->sc_super, &ti, 0); |
dcd76186 | 2704 | ret = nilfs_segctor_construct(sci, SC_LSEG_SR); |
f7545144 | 2705 | nilfs_transaction_unlock(sci->sc_super); |
9ff05123 | 2706 | |
7ef3ff2f RK |
2707 | flush_work(&sci->sc_iput_work); |
2708 | ||
9ff05123 RK |
2709 | } while (ret && retrycount-- > 0); |
2710 | } | |
2711 | ||
2712 | /** | |
2713 | * nilfs_segctor_destroy - destroy the segment constructor. | |
2714 | * @sci: nilfs_sc_info | |
2715 | * | |
2716 | * nilfs_segctor_destroy() kills the segctord thread and frees | |
2717 | * the nilfs_sc_info struct. | |
2718 | * Caller must hold the segment semaphore. | |
2719 | */ | |
2720 | static void nilfs_segctor_destroy(struct nilfs_sc_info *sci) | |
2721 | { | |
e3154e97 | 2722 | struct the_nilfs *nilfs = sci->sc_super->s_fs_info; |
9ff05123 RK |
2723 | int flag; |
2724 | ||
693dd321 | 2725 | up_write(&nilfs->ns_segctor_sem); |
9ff05123 RK |
2726 | |
2727 | spin_lock(&sci->sc_state_lock); | |
2728 | nilfs_segctor_kill_thread(sci); | |
2729 | flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request | |
2730 | || sci->sc_seq_request != sci->sc_seq_done); | |
2731 | spin_unlock(&sci->sc_state_lock); | |
2732 | ||
7ef3ff2f RK |
2733 | if (flush_work(&sci->sc_iput_work)) |
2734 | flag = true; | |
2735 | ||
3256a055 | 2736 | if (flag || !nilfs_segctor_confirm(sci)) |
9ff05123 RK |
2737 | nilfs_segctor_write_out(sci); |
2738 | ||
9ff05123 | 2739 | if (!list_empty(&sci->sc_dirty_files)) { |
a1d0747a JP |
2740 | nilfs_warn(sci->sc_super, |
2741 | "disposed unprocessed dirty file(s) when stopping log writer"); | |
693dd321 | 2742 | nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1); |
9ff05123 | 2743 | } |
9ff05123 | 2744 | |
7ef3ff2f | 2745 | if (!list_empty(&sci->sc_iput_queue)) { |
a1d0747a JP |
2746 | nilfs_warn(sci->sc_super, |
2747 | "disposed unprocessed inode(s) in iput queue when stopping log writer"); | |
7ef3ff2f RK |
2748 | nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1); |
2749 | } | |
2750 | ||
1f5abe7e | 2751 | WARN_ON(!list_empty(&sci->sc_segbufs)); |
a694291a | 2752 | WARN_ON(!list_empty(&sci->sc_write_logs)); |
9ff05123 | 2753 | |
e912a5b6 RK |
2754 | nilfs_put_root(sci->sc_root); |
2755 | ||
693dd321 | 2756 | down_write(&nilfs->ns_segctor_sem); |
9ff05123 | 2757 | |
292a089d | 2758 | timer_shutdown_sync(&sci->sc_timer); |
9ff05123 RK |
2759 | kfree(sci); |
2760 | } | |
2761 | ||
2762 | /** | |
f7545144 RK |
2763 | * nilfs_attach_log_writer - attach log writer |
2764 | * @sb: super block instance | |
e912a5b6 | 2765 | * @root: root object of the current filesystem tree |
9ff05123 | 2766 | * |
f7545144 RK |
2767 | * This allocates a log writer object, initializes it, and starts the |
2768 | * log writer. | |
9ff05123 RK |
2769 | * |
2770 | * Return Value: On success, 0 is returned. On error, one of the following | |
2771 | * negative error code is returned. | |
2772 | * | |
2773 | * %-ENOMEM - Insufficient memory available. | |
2774 | */ | |
f7545144 | 2775 | int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root) |
9ff05123 | 2776 | { |
e3154e97 | 2777 | struct the_nilfs *nilfs = sb->s_fs_info; |
9ff05123 RK |
2778 | int err; |
2779 | ||
3fd3fe5a | 2780 | if (nilfs->ns_writer) { |
fe5f171b | 2781 | /* |
8cccf05f RK |
2782 | * This happens if the filesystem is made read-only by |
2783 | * __nilfs_error or nilfs_remount and then remounted | |
2784 | * read/write. In these cases, reuse the existing | |
2785 | * writer. | |
fe5f171b | 2786 | */ |
8cccf05f | 2787 | return 0; |
fe5f171b RK |
2788 | } |
2789 | ||
f7545144 | 2790 | nilfs->ns_writer = nilfs_segctor_new(sb, root); |
3fd3fe5a | 2791 | if (!nilfs->ns_writer) |
9ff05123 RK |
2792 | return -ENOMEM; |
2793 | ||
8301c719 RK |
2794 | inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL); |
2795 | ||
3fd3fe5a | 2796 | err = nilfs_segctor_start_thread(nilfs->ns_writer); |
d0d51a97 RK |
2797 | if (unlikely(err)) |
2798 | nilfs_detach_log_writer(sb); | |
2799 | ||
9ff05123 RK |
2800 | return err; |
2801 | } | |
2802 | ||
2803 | /** | |
f7545144 RK |
2804 | * nilfs_detach_log_writer - destroy log writer |
2805 | * @sb: super block instance | |
9ff05123 | 2806 | * |
f7545144 RK |
2807 | * This kills log writer daemon, frees the log writer object, and |
2808 | * destroys list of dirty files. | |
9ff05123 | 2809 | */ |
f7545144 | 2810 | void nilfs_detach_log_writer(struct super_block *sb) |
9ff05123 | 2811 | { |
e3154e97 | 2812 | struct the_nilfs *nilfs = sb->s_fs_info; |
9ff05123 RK |
2813 | LIST_HEAD(garbage_list); |
2814 | ||
2815 | down_write(&nilfs->ns_segctor_sem); | |
3fd3fe5a RK |
2816 | if (nilfs->ns_writer) { |
2817 | nilfs_segctor_destroy(nilfs->ns_writer); | |
2818 | nilfs->ns_writer = NULL; | |
9ff05123 RK |
2819 | } |
2820 | ||
2821 | /* Force to free the list of dirty files */ | |
693dd321 RK |
2822 | spin_lock(&nilfs->ns_inode_lock); |
2823 | if (!list_empty(&nilfs->ns_dirty_files)) { | |
2824 | list_splice_init(&nilfs->ns_dirty_files, &garbage_list); | |
a1d0747a JP |
2825 | nilfs_warn(sb, |
2826 | "disposed unprocessed dirty file(s) when detaching log writer"); | |
9ff05123 | 2827 | } |
693dd321 | 2828 | spin_unlock(&nilfs->ns_inode_lock); |
9ff05123 RK |
2829 | up_write(&nilfs->ns_segctor_sem); |
2830 | ||
693dd321 | 2831 | nilfs_dispose_list(nilfs, &garbage_list, 1); |
9ff05123 | 2832 | } |