• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * NILFS segment constructor.
4  *
5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6  *
7  * Written by Ryusuke Konishi.
8  *
9  */
10 
11 #include <linux/pagemap.h>
12 #include <linux/buffer_head.h>
13 #include <linux/writeback.h>
14 #include <linux/bitops.h>
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>
23 #include <linux/slab.h>
24 #include <linux/sched/signal.h>
25 
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"
33 #include "segbuf.h"
34 
35 
36 /*
37  * Segment constructor
38  */
39 #define SC_N_INODEVEC	16   /* Size of locally allocated inode vector */
40 
41 #define SC_MAX_SEGDELTA 64   /*
42 			      * Upper limit of the number of segments
43 			      * appended in collection retry loop
44 			      */
45 
46 /* Construction mode */
47 enum {
48 	SC_LSEG_SR = 1,	/* Make a logical segment having a super root */
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 			 */
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,
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 
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  */
nilfs_sc_cstage_inc(struct nilfs_sc_info * sci)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 
nilfs_sc_cstage_set(struct nilfs_sc_info * sci,int next_scnt)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 
nilfs_sc_cstage_get(struct nilfs_sc_info * sci)102 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
103 {
104 	return sci->sc_stage.scnt;
105 }
106 
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 */
110 #define NILFS_CF_SUFREED	0x0004	/* segment usages has been freed */
111 #define NILFS_CF_HISTORY_MASK	(NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
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 *);
135 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
136 
137 #define nilfs_cnt32_ge(a, b)   \
138 	(typecheck(__u32, a) && typecheck(__u32, b) && \
139 	 ((__s32)(a) - (__s32)(b) >= 0))
140 
nilfs_prepare_segment_lock(struct super_block * sb,struct nilfs_transaction_info * ti)141 static int nilfs_prepare_segment_lock(struct super_block *sb,
142 				      struct nilfs_transaction_info *ti)
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;
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 		 */
156 		nilfs_warn(sb, "journal info from a different FS");
157 		save = current->journal_info;
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
182  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
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
189  * instead; otherwise a new struct is assigned from a slab.
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  *
199  * %-ENOSPC - No space left on device
200  */
nilfs_transaction_begin(struct super_block * sb,struct nilfs_transaction_info * ti,int vacancy_check)201 int nilfs_transaction_begin(struct super_block *sb,
202 			    struct nilfs_transaction_info *ti,
203 			    int vacancy_check)
204 {
205 	struct the_nilfs *nilfs;
206 	int ret = nilfs_prepare_segment_lock(sb, ti);
207 	struct nilfs_transaction_info *trace_ti;
208 
209 	if (unlikely(ret < 0))
210 		return ret;
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);
217 		return 0;
218 	}
219 
220 	sb_start_intwrite(sb);
221 
222 	nilfs = sb->s_fs_info;
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 	}
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);
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);
241 	sb_end_intwrite(sb);
242 	return ret;
243 }
244 
245 /**
246  * nilfs_transaction_commit - commit indivisible file operations.
247  * @sb: super block
248  *
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.
255  */
nilfs_transaction_commit(struct super_block * sb)256 int nilfs_transaction_commit(struct super_block *sb)
257 {
258 	struct nilfs_transaction_info *ti = current->journal_info;
259 	struct the_nilfs *nilfs = sb->s_fs_info;
260 	int err = 0;
261 
262 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
263 	ti->ti_flags |= NILFS_TI_COMMIT;
264 	if (ti->ti_count > 0) {
265 		ti->ti_count--;
266 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
267 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
268 		return 0;
269 	}
270 	if (nilfs->ns_writer) {
271 		struct nilfs_sc_info *sci = nilfs->ns_writer;
272 
273 		if (ti->ti_flags & NILFS_TI_COMMIT)
274 			nilfs_segctor_start_timer(sci);
275 		if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
276 			nilfs_segctor_do_flush(sci, 0);
277 	}
278 	up_read(&nilfs->ns_segctor_sem);
279 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
280 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
281 
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);
288 	sb_end_intwrite(sb);
289 	return err;
290 }
291 
nilfs_transaction_abort(struct super_block * sb)292 void nilfs_transaction_abort(struct super_block *sb)
293 {
294 	struct nilfs_transaction_info *ti = current->journal_info;
295 	struct the_nilfs *nilfs = sb->s_fs_info;
296 
297 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
298 	if (ti->ti_count > 0) {
299 		ti->ti_count--;
300 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
301 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
302 		return;
303 	}
304 	up_read(&nilfs->ns_segctor_sem);
305 
306 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
307 		    ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
308 
309 	current->journal_info = ti->ti_save;
310 	if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
311 		kmem_cache_free(nilfs_transaction_cachep, ti);
312 	sb_end_intwrite(sb);
313 }
314 
nilfs_relax_pressure_in_lock(struct super_block * sb)315 void nilfs_relax_pressure_in_lock(struct super_block *sb)
316 {
317 	struct the_nilfs *nilfs = sb->s_fs_info;
318 	struct nilfs_sc_info *sci = nilfs->ns_writer;
319 
320 	if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request)
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 
nilfs_transaction_lock(struct super_block * sb,struct nilfs_transaction_info * ti,int gcflag)338 static void nilfs_transaction_lock(struct super_block *sb,
339 				   struct nilfs_transaction_info *ti,
340 				   int gcflag)
341 {
342 	struct nilfs_transaction_info *cur_ti = current->journal_info;
343 	struct the_nilfs *nilfs = sb->s_fs_info;
344 	struct nilfs_sc_info *sci = nilfs->ns_writer;
345 
346 	WARN_ON(cur_ti);
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;
351 	current->journal_info = ti;
352 
353 	for (;;) {
354 		trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
355 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
356 
357 		down_write(&nilfs->ns_segctor_sem);
358 		if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
359 			break;
360 
361 		nilfs_segctor_do_immediate_flush(sci);
362 
363 		up_write(&nilfs->ns_segctor_sem);
364 		cond_resched();
365 	}
366 	if (gcflag)
367 		ti->ti_flags |= NILFS_TI_GC;
368 
369 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
370 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
371 }
372 
nilfs_transaction_unlock(struct super_block * sb)373 static void nilfs_transaction_unlock(struct super_block *sb)
374 {
375 	struct nilfs_transaction_info *ti = current->journal_info;
376 	struct the_nilfs *nilfs = sb->s_fs_info;
377 
378 	BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
379 	BUG_ON(ti->ti_count > 0);
380 
381 	up_write(&nilfs->ns_segctor_sem);
382 	current->journal_info = ti->ti_save;
383 
384 	trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
385 			    ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
386 }
387 
nilfs_segctor_map_segsum_entry(struct nilfs_sc_info * sci,struct nilfs_segsum_pointer * ssp,unsigned int bytes)388 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
389 					    struct nilfs_segsum_pointer *ssp,
390 					    unsigned int bytes)
391 {
392 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
393 	unsigned int blocksize = sci->sc_super->s_blocksize;
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  */
nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info * sci)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;
415 	unsigned int sumbytes;
416 	unsigned int flags = 0;
417 	int err;
418 
419 	if (nilfs_doing_gc())
420 		flags = NILFS_SS_GC;
421 	err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
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 /**
434  * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
435  * @sci: segment constructor object
436  *
437  * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
438  * the current segment summary block.
439  */
nilfs_segctor_zeropad_segsum(struct nilfs_sc_info * sci)440 static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
441 {
442 	struct nilfs_segsum_pointer *ssp;
443 
444 	ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
445 	if (ssp->offset < ssp->bh->b_size)
446 		memset(ssp->bh->b_data + ssp->offset, 0,
447 		       ssp->bh->b_size - ssp->offset);
448 }
449 
nilfs_segctor_feed_segment(struct nilfs_sc_info * sci)450 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
451 {
452 	sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
453 	if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
454 		return -E2BIG; /*
455 				* The current segment is filled up
456 				* (internal code)
457 				*/
458 	nilfs_segctor_zeropad_segsum(sci);
459 	sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
460 	return nilfs_segctor_reset_segment_buffer(sci);
461 }
462 
nilfs_segctor_add_super_root(struct nilfs_sc_info * sci)463 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
464 {
465 	struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
466 	int err;
467 
468 	if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
469 		err = nilfs_segctor_feed_segment(sci);
470 		if (err)
471 			return err;
472 		segbuf = sci->sc_curseg;
473 	}
474 	err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
475 	if (likely(!err))
476 		segbuf->sb_sum.flags |= NILFS_SS_SR;
477 	return err;
478 }
479 
480 /*
481  * Functions for making segment summary and payloads
482  */
nilfs_segctor_segsum_block_required(struct nilfs_sc_info * sci,const struct nilfs_segsum_pointer * ssp,unsigned int binfo_size)483 static int nilfs_segctor_segsum_block_required(
484 	struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
485 	unsigned int binfo_size)
486 {
487 	unsigned int blocksize = sci->sc_super->s_blocksize;
488 	/* Size of finfo and binfo is enough small against blocksize */
489 
490 	return ssp->offset + binfo_size +
491 		(!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
492 		blocksize;
493 }
494 
nilfs_segctor_begin_finfo(struct nilfs_sc_info * sci,struct inode * inode)495 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
496 				      struct inode *inode)
497 {
498 	sci->sc_curseg->sb_sum.nfinfo++;
499 	sci->sc_binfo_ptr = sci->sc_finfo_ptr;
500 	nilfs_segctor_map_segsum_entry(
501 		sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
502 
503 	if (NILFS_I(inode)->i_root &&
504 	    !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
505 		set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
506 	/* skip finfo */
507 }
508 
nilfs_segctor_end_finfo(struct nilfs_sc_info * sci,struct inode * inode)509 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
510 				    struct inode *inode)
511 {
512 	struct nilfs_finfo *finfo;
513 	struct nilfs_inode_info *ii;
514 	struct nilfs_segment_buffer *segbuf;
515 	__u64 cno;
516 
517 	if (sci->sc_blk_cnt == 0)
518 		return;
519 
520 	ii = NILFS_I(inode);
521 
522 	if (test_bit(NILFS_I_GCINODE, &ii->i_state))
523 		cno = ii->i_cno;
524 	else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
525 		cno = 0;
526 	else
527 		cno = sci->sc_cno;
528 
529 	finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
530 						 sizeof(*finfo));
531 	finfo->fi_ino = cpu_to_le64(inode->i_ino);
532 	finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
533 	finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
534 	finfo->fi_cno = cpu_to_le64(cno);
535 
536 	segbuf = sci->sc_curseg;
537 	segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
538 		sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
539 	sci->sc_finfo_ptr = sci->sc_binfo_ptr;
540 	sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
541 }
542 
nilfs_segctor_add_file_block(struct nilfs_sc_info * sci,struct buffer_head * bh,struct inode * inode,unsigned int binfo_size)543 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
544 					struct buffer_head *bh,
545 					struct inode *inode,
546 					unsigned int binfo_size)
547 {
548 	struct nilfs_segment_buffer *segbuf;
549 	int required, err = 0;
550 
551  retry:
552 	segbuf = sci->sc_curseg;
553 	required = nilfs_segctor_segsum_block_required(
554 		sci, &sci->sc_binfo_ptr, binfo_size);
555 	if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
556 		nilfs_segctor_end_finfo(sci, inode);
557 		err = nilfs_segctor_feed_segment(sci);
558 		if (err)
559 			return err;
560 		goto retry;
561 	}
562 	if (unlikely(required)) {
563 		nilfs_segctor_zeropad_segsum(sci);
564 		err = nilfs_segbuf_extend_segsum(segbuf);
565 		if (unlikely(err))
566 			goto failed;
567 	}
568 	if (sci->sc_blk_cnt == 0)
569 		nilfs_segctor_begin_finfo(sci, inode);
570 
571 	nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
572 	/* Substitution to vblocknr is delayed until update_blocknr() */
573 	nilfs_segbuf_add_file_buffer(segbuf, bh);
574 	sci->sc_blk_cnt++;
575  failed:
576 	return err;
577 }
578 
579 /*
580  * Callback functions that enumerate, mark, and collect dirty blocks
581  */
nilfs_collect_file_data(struct nilfs_sc_info * sci,struct buffer_head * bh,struct inode * inode)582 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
583 				   struct buffer_head *bh, struct inode *inode)
584 {
585 	int err;
586 
587 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
588 	if (err < 0)
589 		return err;
590 
591 	err = nilfs_segctor_add_file_block(sci, bh, inode,
592 					   sizeof(struct nilfs_binfo_v));
593 	if (!err)
594 		sci->sc_datablk_cnt++;
595 	return err;
596 }
597 
nilfs_collect_file_node(struct nilfs_sc_info * sci,struct buffer_head * bh,struct inode * inode)598 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
599 				   struct buffer_head *bh,
600 				   struct inode *inode)
601 {
602 	return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
603 }
604 
nilfs_collect_file_bmap(struct nilfs_sc_info * sci,struct buffer_head * bh,struct inode * inode)605 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
606 				   struct buffer_head *bh,
607 				   struct inode *inode)
608 {
609 	WARN_ON(!buffer_dirty(bh));
610 	return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
611 }
612 
nilfs_write_file_data_binfo(struct nilfs_sc_info * sci,struct nilfs_segsum_pointer * ssp,union nilfs_binfo * binfo)613 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
614 					struct nilfs_segsum_pointer *ssp,
615 					union nilfs_binfo *binfo)
616 {
617 	struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
618 		sci, ssp, sizeof(*binfo_v));
619 	*binfo_v = binfo->bi_v;
620 }
621 
nilfs_write_file_node_binfo(struct nilfs_sc_info * sci,struct nilfs_segsum_pointer * ssp,union nilfs_binfo * binfo)622 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
623 					struct nilfs_segsum_pointer *ssp,
624 					union nilfs_binfo *binfo)
625 {
626 	__le64 *vblocknr = nilfs_segctor_map_segsum_entry(
627 		sci, ssp, sizeof(*vblocknr));
628 	*vblocknr = binfo->bi_v.bi_vblocknr;
629 }
630 
631 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
632 	.collect_data = nilfs_collect_file_data,
633 	.collect_node = nilfs_collect_file_node,
634 	.collect_bmap = nilfs_collect_file_bmap,
635 	.write_data_binfo = nilfs_write_file_data_binfo,
636 	.write_node_binfo = nilfs_write_file_node_binfo,
637 };
638 
nilfs_collect_dat_data(struct nilfs_sc_info * sci,struct buffer_head * bh,struct inode * inode)639 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
640 				  struct buffer_head *bh, struct inode *inode)
641 {
642 	int err;
643 
644 	err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
645 	if (err < 0)
646 		return err;
647 
648 	err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
649 	if (!err)
650 		sci->sc_datablk_cnt++;
651 	return err;
652 }
653 
nilfs_collect_dat_bmap(struct nilfs_sc_info * sci,struct buffer_head * bh,struct inode * inode)654 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
655 				  struct buffer_head *bh, struct inode *inode)
656 {
657 	WARN_ON(!buffer_dirty(bh));
658 	return nilfs_segctor_add_file_block(sci, bh, inode,
659 					    sizeof(struct nilfs_binfo_dat));
660 }
661 
nilfs_write_dat_data_binfo(struct nilfs_sc_info * sci,struct nilfs_segsum_pointer * ssp,union nilfs_binfo * binfo)662 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
663 				       struct nilfs_segsum_pointer *ssp,
664 				       union nilfs_binfo *binfo)
665 {
666 	__le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
667 							  sizeof(*blkoff));
668 	*blkoff = binfo->bi_dat.bi_blkoff;
669 }
670 
nilfs_write_dat_node_binfo(struct nilfs_sc_info * sci,struct nilfs_segsum_pointer * ssp,union nilfs_binfo * binfo)671 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
672 				       struct nilfs_segsum_pointer *ssp,
673 				       union nilfs_binfo *binfo)
674 {
675 	struct nilfs_binfo_dat *binfo_dat =
676 		nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
677 	*binfo_dat = binfo->bi_dat;
678 }
679 
680 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
681 	.collect_data = nilfs_collect_dat_data,
682 	.collect_node = nilfs_collect_file_node,
683 	.collect_bmap = nilfs_collect_dat_bmap,
684 	.write_data_binfo = nilfs_write_dat_data_binfo,
685 	.write_node_binfo = nilfs_write_dat_node_binfo,
686 };
687 
688 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
689 	.collect_data = nilfs_collect_file_data,
690 	.collect_node = NULL,
691 	.collect_bmap = NULL,
692 	.write_data_binfo = nilfs_write_file_data_binfo,
693 	.write_node_binfo = NULL,
694 };
695 
nilfs_lookup_dirty_data_buffers(struct inode * inode,struct list_head * listp,size_t nlimit,loff_t start,loff_t end)696 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
697 					      struct list_head *listp,
698 					      size_t nlimit,
699 					      loff_t start, loff_t end)
700 {
701 	struct address_space *mapping = inode->i_mapping;
702 	struct pagevec pvec;
703 	pgoff_t index = 0, last = ULONG_MAX;
704 	size_t ndirties = 0;
705 	int i;
706 
707 	if (unlikely(start != 0 || end != LLONG_MAX)) {
708 		/*
709 		 * A valid range is given for sync-ing data pages. The
710 		 * range is rounded to per-page; extra dirty buffers
711 		 * may be included if blocksize < pagesize.
712 		 */
713 		index = start >> PAGE_SHIFT;
714 		last = end >> PAGE_SHIFT;
715 	}
716 	pagevec_init(&pvec);
717  repeat:
718 	if (unlikely(index > last) ||
719 	    !pagevec_lookup_range_tag(&pvec, mapping, &index, last,
720 				PAGECACHE_TAG_DIRTY))
721 		return ndirties;
722 
723 	for (i = 0; i < pagevec_count(&pvec); i++) {
724 		struct buffer_head *bh, *head;
725 		struct page *page = pvec.pages[i];
726 
727 		lock_page(page);
728 		if (unlikely(page->mapping != mapping)) {
729 			/* Exclude pages removed from the address space */
730 			unlock_page(page);
731 			continue;
732 		}
733 		if (!page_has_buffers(page))
734 			create_empty_buffers(page, i_blocksize(inode), 0);
735 		unlock_page(page);
736 
737 		bh = head = page_buffers(page);
738 		do {
739 			if (!buffer_dirty(bh) || buffer_async_write(bh))
740 				continue;
741 			get_bh(bh);
742 			list_add_tail(&bh->b_assoc_buffers, listp);
743 			ndirties++;
744 			if (unlikely(ndirties >= nlimit)) {
745 				pagevec_release(&pvec);
746 				cond_resched();
747 				return ndirties;
748 			}
749 		} while (bh = bh->b_this_page, bh != head);
750 	}
751 	pagevec_release(&pvec);
752 	cond_resched();
753 	goto repeat;
754 }
755 
nilfs_lookup_dirty_node_buffers(struct inode * inode,struct list_head * listp)756 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
757 					    struct list_head *listp)
758 {
759 	struct nilfs_inode_info *ii = NILFS_I(inode);
760 	struct inode *btnc_inode = ii->i_assoc_inode;
761 	struct pagevec pvec;
762 	struct buffer_head *bh, *head;
763 	unsigned int i;
764 	pgoff_t index = 0;
765 
766 	if (!btnc_inode)
767 		return;
768 
769 	pagevec_init(&pvec);
770 
771 	while (pagevec_lookup_tag(&pvec, btnc_inode->i_mapping, &index,
772 					PAGECACHE_TAG_DIRTY)) {
773 		for (i = 0; i < pagevec_count(&pvec); i++) {
774 			bh = head = page_buffers(pvec.pages[i]);
775 			do {
776 				if (buffer_dirty(bh) &&
777 						!buffer_async_write(bh)) {
778 					get_bh(bh);
779 					list_add_tail(&bh->b_assoc_buffers,
780 						      listp);
781 				}
782 				bh = bh->b_this_page;
783 			} while (bh != head);
784 		}
785 		pagevec_release(&pvec);
786 		cond_resched();
787 	}
788 }
789 
nilfs_dispose_list(struct the_nilfs * nilfs,struct list_head * head,int force)790 static void nilfs_dispose_list(struct the_nilfs *nilfs,
791 			       struct list_head *head, int force)
792 {
793 	struct nilfs_inode_info *ii, *n;
794 	struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
795 	unsigned int nv = 0;
796 
797 	while (!list_empty(head)) {
798 		spin_lock(&nilfs->ns_inode_lock);
799 		list_for_each_entry_safe(ii, n, head, i_dirty) {
800 			list_del_init(&ii->i_dirty);
801 			if (force) {
802 				if (unlikely(ii->i_bh)) {
803 					brelse(ii->i_bh);
804 					ii->i_bh = NULL;
805 				}
806 			} else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
807 				set_bit(NILFS_I_QUEUED, &ii->i_state);
808 				list_add_tail(&ii->i_dirty,
809 					      &nilfs->ns_dirty_files);
810 				continue;
811 			}
812 			ivec[nv++] = ii;
813 			if (nv == SC_N_INODEVEC)
814 				break;
815 		}
816 		spin_unlock(&nilfs->ns_inode_lock);
817 
818 		for (pii = ivec; nv > 0; pii++, nv--)
819 			iput(&(*pii)->vfs_inode);
820 	}
821 }
822 
nilfs_iput_work_func(struct work_struct * work)823 static void nilfs_iput_work_func(struct work_struct *work)
824 {
825 	struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
826 						 sc_iput_work);
827 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
828 
829 	nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
830 }
831 
nilfs_test_metadata_dirty(struct the_nilfs * nilfs,struct nilfs_root * root)832 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
833 				     struct nilfs_root *root)
834 {
835 	int ret = 0;
836 
837 	if (nilfs_mdt_fetch_dirty(root->ifile))
838 		ret++;
839 	if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
840 		ret++;
841 	if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
842 		ret++;
843 	if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
844 		ret++;
845 	return ret;
846 }
847 
nilfs_segctor_clean(struct nilfs_sc_info * sci)848 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
849 {
850 	return list_empty(&sci->sc_dirty_files) &&
851 		!test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
852 		sci->sc_nfreesegs == 0 &&
853 		(!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
854 }
855 
nilfs_segctor_confirm(struct nilfs_sc_info * sci)856 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
857 {
858 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
859 	int ret = 0;
860 
861 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
862 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
863 
864 	spin_lock(&nilfs->ns_inode_lock);
865 	if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
866 		ret++;
867 
868 	spin_unlock(&nilfs->ns_inode_lock);
869 	return ret;
870 }
871 
nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info * sci)872 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
873 {
874 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
875 
876 	nilfs_mdt_clear_dirty(sci->sc_root->ifile);
877 	nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
878 	nilfs_mdt_clear_dirty(nilfs->ns_sufile);
879 	nilfs_mdt_clear_dirty(nilfs->ns_dat);
880 }
881 
nilfs_segctor_create_checkpoint(struct nilfs_sc_info * sci)882 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
883 {
884 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
885 	struct buffer_head *bh_cp;
886 	struct nilfs_checkpoint *raw_cp;
887 	int err;
888 
889 	/* XXX: this interface will be changed */
890 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
891 					  &raw_cp, &bh_cp);
892 	if (likely(!err)) {
893 		/*
894 		 * The following code is duplicated with cpfile.  But, it is
895 		 * needed to collect the checkpoint even if it was not newly
896 		 * created.
897 		 */
898 		mark_buffer_dirty(bh_cp);
899 		nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
900 		nilfs_cpfile_put_checkpoint(
901 			nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
902 	} else if (err == -EINVAL || err == -ENOENT) {
903 		nilfs_error(sci->sc_super,
904 			    "checkpoint creation failed due to metadata corruption.");
905 		err = -EIO;
906 	}
907 	return err;
908 }
909 
nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info * sci)910 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
911 {
912 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
913 	struct buffer_head *bh_cp;
914 	struct nilfs_checkpoint *raw_cp;
915 	int err;
916 
917 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
918 					  &raw_cp, &bh_cp);
919 	if (unlikely(err)) {
920 		if (err == -EINVAL || err == -ENOENT) {
921 			nilfs_error(sci->sc_super,
922 				    "checkpoint finalization failed due to metadata corruption.");
923 			err = -EIO;
924 		}
925 		goto failed_ibh;
926 	}
927 	raw_cp->cp_snapshot_list.ssl_next = 0;
928 	raw_cp->cp_snapshot_list.ssl_prev = 0;
929 	raw_cp->cp_inodes_count =
930 		cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
931 	raw_cp->cp_blocks_count =
932 		cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
933 	raw_cp->cp_nblk_inc =
934 		cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
935 	raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
936 	raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
937 
938 	if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
939 		nilfs_checkpoint_clear_minor(raw_cp);
940 	else
941 		nilfs_checkpoint_set_minor(raw_cp);
942 
943 	nilfs_write_inode_common(sci->sc_root->ifile,
944 				 &raw_cp->cp_ifile_inode, 1);
945 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
946 	return 0;
947 
948  failed_ibh:
949 	return err;
950 }
951 
nilfs_fill_in_file_bmap(struct inode * ifile,struct nilfs_inode_info * ii)952 static void nilfs_fill_in_file_bmap(struct inode *ifile,
953 				    struct nilfs_inode_info *ii)
954 
955 {
956 	struct buffer_head *ibh;
957 	struct nilfs_inode *raw_inode;
958 
959 	if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
960 		ibh = ii->i_bh;
961 		BUG_ON(!ibh);
962 		raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
963 						  ibh);
964 		nilfs_bmap_write(ii->i_bmap, raw_inode);
965 		nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
966 	}
967 }
968 
nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info * sci)969 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
970 {
971 	struct nilfs_inode_info *ii;
972 
973 	list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
974 		nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
975 		set_bit(NILFS_I_COLLECTED, &ii->i_state);
976 	}
977 }
978 
nilfs_segctor_fill_in_super_root(struct nilfs_sc_info * sci,struct the_nilfs * nilfs)979 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
980 					     struct the_nilfs *nilfs)
981 {
982 	struct buffer_head *bh_sr;
983 	struct nilfs_super_root *raw_sr;
984 	unsigned int isz, srsz;
985 
986 	bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
987 
988 	lock_buffer(bh_sr);
989 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
990 	isz = nilfs->ns_inode_size;
991 	srsz = NILFS_SR_BYTES(isz);
992 
993 	raw_sr->sr_sum = 0;  /* Ensure initialization within this update */
994 	raw_sr->sr_bytes = cpu_to_le16(srsz);
995 	raw_sr->sr_nongc_ctime
996 		= cpu_to_le64(nilfs_doing_gc() ?
997 			      nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
998 	raw_sr->sr_flags = 0;
999 
1000 	nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
1001 				 NILFS_SR_DAT_OFFSET(isz), 1);
1002 	nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
1003 				 NILFS_SR_CPFILE_OFFSET(isz), 1);
1004 	nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
1005 				 NILFS_SR_SUFILE_OFFSET(isz), 1);
1006 	memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1007 	set_buffer_uptodate(bh_sr);
1008 	unlock_buffer(bh_sr);
1009 }
1010 
nilfs_redirty_inodes(struct list_head * head)1011 static void nilfs_redirty_inodes(struct list_head *head)
1012 {
1013 	struct nilfs_inode_info *ii;
1014 
1015 	list_for_each_entry(ii, head, i_dirty) {
1016 		if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1017 			clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1018 	}
1019 }
1020 
nilfs_drop_collected_inodes(struct list_head * head)1021 static void nilfs_drop_collected_inodes(struct list_head *head)
1022 {
1023 	struct nilfs_inode_info *ii;
1024 
1025 	list_for_each_entry(ii, head, i_dirty) {
1026 		if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1027 			continue;
1028 
1029 		clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1030 		set_bit(NILFS_I_UPDATED, &ii->i_state);
1031 	}
1032 }
1033 
nilfs_segctor_apply_buffers(struct nilfs_sc_info * sci,struct inode * inode,struct list_head * listp,int (* collect)(struct nilfs_sc_info *,struct buffer_head *,struct inode *))1034 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1035 				       struct inode *inode,
1036 				       struct list_head *listp,
1037 				       int (*collect)(struct nilfs_sc_info *,
1038 						      struct buffer_head *,
1039 						      struct inode *))
1040 {
1041 	struct buffer_head *bh, *n;
1042 	int err = 0;
1043 
1044 	if (collect) {
1045 		list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1046 			list_del_init(&bh->b_assoc_buffers);
1047 			err = collect(sci, bh, inode);
1048 			brelse(bh);
1049 			if (unlikely(err))
1050 				goto dispose_buffers;
1051 		}
1052 		return 0;
1053 	}
1054 
1055  dispose_buffers:
1056 	while (!list_empty(listp)) {
1057 		bh = list_first_entry(listp, struct buffer_head,
1058 				      b_assoc_buffers);
1059 		list_del_init(&bh->b_assoc_buffers);
1060 		brelse(bh);
1061 	}
1062 	return err;
1063 }
1064 
nilfs_segctor_buffer_rest(struct nilfs_sc_info * sci)1065 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1066 {
1067 	/* Remaining number of blocks within segment buffer */
1068 	return sci->sc_segbuf_nblocks -
1069 		(sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1070 }
1071 
nilfs_segctor_scan_file(struct nilfs_sc_info * sci,struct inode * inode,const struct nilfs_sc_operations * sc_ops)1072 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1073 				   struct inode *inode,
1074 				   const struct nilfs_sc_operations *sc_ops)
1075 {
1076 	LIST_HEAD(data_buffers);
1077 	LIST_HEAD(node_buffers);
1078 	int err;
1079 
1080 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1081 		size_t n, rest = nilfs_segctor_buffer_rest(sci);
1082 
1083 		n = nilfs_lookup_dirty_data_buffers(
1084 			inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1085 		if (n > rest) {
1086 			err = nilfs_segctor_apply_buffers(
1087 				sci, inode, &data_buffers,
1088 				sc_ops->collect_data);
1089 			BUG_ON(!err); /* always receive -E2BIG or true error */
1090 			goto break_or_fail;
1091 		}
1092 	}
1093 	nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1094 
1095 	if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1096 		err = nilfs_segctor_apply_buffers(
1097 			sci, inode, &data_buffers, sc_ops->collect_data);
1098 		if (unlikely(err)) {
1099 			/* dispose node list */
1100 			nilfs_segctor_apply_buffers(
1101 				sci, inode, &node_buffers, NULL);
1102 			goto break_or_fail;
1103 		}
1104 		sci->sc_stage.flags |= NILFS_CF_NODE;
1105 	}
1106 	/* Collect node */
1107 	err = nilfs_segctor_apply_buffers(
1108 		sci, inode, &node_buffers, sc_ops->collect_node);
1109 	if (unlikely(err))
1110 		goto break_or_fail;
1111 
1112 	nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1113 	err = nilfs_segctor_apply_buffers(
1114 		sci, inode, &node_buffers, sc_ops->collect_bmap);
1115 	if (unlikely(err))
1116 		goto break_or_fail;
1117 
1118 	nilfs_segctor_end_finfo(sci, inode);
1119 	sci->sc_stage.flags &= ~NILFS_CF_NODE;
1120 
1121  break_or_fail:
1122 	return err;
1123 }
1124 
nilfs_segctor_scan_file_dsync(struct nilfs_sc_info * sci,struct inode * inode)1125 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1126 					 struct inode *inode)
1127 {
1128 	LIST_HEAD(data_buffers);
1129 	size_t n, rest = nilfs_segctor_buffer_rest(sci);
1130 	int err;
1131 
1132 	n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1133 					    sci->sc_dsync_start,
1134 					    sci->sc_dsync_end);
1135 
1136 	err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1137 					  nilfs_collect_file_data);
1138 	if (!err) {
1139 		nilfs_segctor_end_finfo(sci, inode);
1140 		BUG_ON(n > rest);
1141 		/* always receive -E2BIG or true error if n > rest */
1142 	}
1143 	return err;
1144 }
1145 
nilfs_segctor_collect_blocks(struct nilfs_sc_info * sci,int mode)1146 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1147 {
1148 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1149 	struct list_head *head;
1150 	struct nilfs_inode_info *ii;
1151 	size_t ndone;
1152 	int err = 0;
1153 
1154 	switch (nilfs_sc_cstage_get(sci)) {
1155 	case NILFS_ST_INIT:
1156 		/* Pre-processes */
1157 		sci->sc_stage.flags = 0;
1158 
1159 		if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1160 			sci->sc_nblk_inc = 0;
1161 			sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1162 			if (mode == SC_LSEG_DSYNC) {
1163 				nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1164 				goto dsync_mode;
1165 			}
1166 		}
1167 
1168 		sci->sc_stage.dirty_file_ptr = NULL;
1169 		sci->sc_stage.gc_inode_ptr = NULL;
1170 		if (mode == SC_FLUSH_DAT) {
1171 			nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1172 			goto dat_stage;
1173 		}
1174 		nilfs_sc_cstage_inc(sci);
1175 		fallthrough;
1176 	case NILFS_ST_GC:
1177 		if (nilfs_doing_gc()) {
1178 			head = &sci->sc_gc_inodes;
1179 			ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1180 						head, i_dirty);
1181 			list_for_each_entry_continue(ii, head, i_dirty) {
1182 				err = nilfs_segctor_scan_file(
1183 					sci, &ii->vfs_inode,
1184 					&nilfs_sc_file_ops);
1185 				if (unlikely(err)) {
1186 					sci->sc_stage.gc_inode_ptr = list_entry(
1187 						ii->i_dirty.prev,
1188 						struct nilfs_inode_info,
1189 						i_dirty);
1190 					goto break_or_fail;
1191 				}
1192 				set_bit(NILFS_I_COLLECTED, &ii->i_state);
1193 			}
1194 			sci->sc_stage.gc_inode_ptr = NULL;
1195 		}
1196 		nilfs_sc_cstage_inc(sci);
1197 		fallthrough;
1198 	case NILFS_ST_FILE:
1199 		head = &sci->sc_dirty_files;
1200 		ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1201 					i_dirty);
1202 		list_for_each_entry_continue(ii, head, i_dirty) {
1203 			clear_bit(NILFS_I_DIRTY, &ii->i_state);
1204 
1205 			err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1206 						      &nilfs_sc_file_ops);
1207 			if (unlikely(err)) {
1208 				sci->sc_stage.dirty_file_ptr =
1209 					list_entry(ii->i_dirty.prev,
1210 						   struct nilfs_inode_info,
1211 						   i_dirty);
1212 				goto break_or_fail;
1213 			}
1214 			/* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1215 			/* XXX: required ? */
1216 		}
1217 		sci->sc_stage.dirty_file_ptr = NULL;
1218 		if (mode == SC_FLUSH_FILE) {
1219 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1220 			return 0;
1221 		}
1222 		nilfs_sc_cstage_inc(sci);
1223 		sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1224 		fallthrough;
1225 	case NILFS_ST_IFILE:
1226 		err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1227 					      &nilfs_sc_file_ops);
1228 		if (unlikely(err))
1229 			break;
1230 		nilfs_sc_cstage_inc(sci);
1231 		/* Creating a checkpoint */
1232 		err = nilfs_segctor_create_checkpoint(sci);
1233 		if (unlikely(err))
1234 			break;
1235 		fallthrough;
1236 	case NILFS_ST_CPFILE:
1237 		err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1238 					      &nilfs_sc_file_ops);
1239 		if (unlikely(err))
1240 			break;
1241 		nilfs_sc_cstage_inc(sci);
1242 		fallthrough;
1243 	case NILFS_ST_SUFILE:
1244 		err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1245 					 sci->sc_nfreesegs, &ndone);
1246 		if (unlikely(err)) {
1247 			nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1248 						  sci->sc_freesegs, ndone,
1249 						  NULL);
1250 			break;
1251 		}
1252 		sci->sc_stage.flags |= NILFS_CF_SUFREED;
1253 
1254 		err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1255 					      &nilfs_sc_file_ops);
1256 		if (unlikely(err))
1257 			break;
1258 		nilfs_sc_cstage_inc(sci);
1259 		fallthrough;
1260 	case NILFS_ST_DAT:
1261  dat_stage:
1262 		err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1263 					      &nilfs_sc_dat_ops);
1264 		if (unlikely(err))
1265 			break;
1266 		if (mode == SC_FLUSH_DAT) {
1267 			nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1268 			return 0;
1269 		}
1270 		nilfs_sc_cstage_inc(sci);
1271 		fallthrough;
1272 	case NILFS_ST_SR:
1273 		if (mode == SC_LSEG_SR) {
1274 			/* Appending a super root */
1275 			err = nilfs_segctor_add_super_root(sci);
1276 			if (unlikely(err))
1277 				break;
1278 		}
1279 		/* End of a logical segment */
1280 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1281 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1282 		return 0;
1283 	case NILFS_ST_DSYNC:
1284  dsync_mode:
1285 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1286 		ii = sci->sc_dsync_inode;
1287 		if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1288 			break;
1289 
1290 		err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1291 		if (unlikely(err))
1292 			break;
1293 		sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1294 		nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1295 		return 0;
1296 	case NILFS_ST_DONE:
1297 		return 0;
1298 	default:
1299 		BUG();
1300 	}
1301 
1302  break_or_fail:
1303 	return err;
1304 }
1305 
1306 /**
1307  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1308  * @sci: nilfs_sc_info
1309  * @nilfs: nilfs object
1310  */
nilfs_segctor_begin_construction(struct nilfs_sc_info * sci,struct the_nilfs * nilfs)1311 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1312 					    struct the_nilfs *nilfs)
1313 {
1314 	struct nilfs_segment_buffer *segbuf, *prev;
1315 	__u64 nextnum;
1316 	int err, alloc = 0;
1317 
1318 	segbuf = nilfs_segbuf_new(sci->sc_super);
1319 	if (unlikely(!segbuf))
1320 		return -ENOMEM;
1321 
1322 	if (list_empty(&sci->sc_write_logs)) {
1323 		nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1324 				 nilfs->ns_pseg_offset, nilfs);
1325 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1326 			nilfs_shift_to_next_segment(nilfs);
1327 			nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1328 		}
1329 
1330 		segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1331 		nextnum = nilfs->ns_nextnum;
1332 
1333 		if (nilfs->ns_segnum == nilfs->ns_nextnum)
1334 			/* Start from the head of a new full segment */
1335 			alloc++;
1336 	} else {
1337 		/* Continue logs */
1338 		prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1339 		nilfs_segbuf_map_cont(segbuf, prev);
1340 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1341 		nextnum = prev->sb_nextnum;
1342 
1343 		if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1344 			nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1345 			segbuf->sb_sum.seg_seq++;
1346 			alloc++;
1347 		}
1348 	}
1349 
1350 	err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1351 	if (err)
1352 		goto failed;
1353 
1354 	if (alloc) {
1355 		err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1356 		if (err)
1357 			goto failed;
1358 	}
1359 	nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1360 
1361 	BUG_ON(!list_empty(&sci->sc_segbufs));
1362 	list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1363 	sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1364 	return 0;
1365 
1366  failed:
1367 	nilfs_segbuf_free(segbuf);
1368 	return err;
1369 }
1370 
nilfs_segctor_extend_segments(struct nilfs_sc_info * sci,struct the_nilfs * nilfs,int nadd)1371 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1372 					 struct the_nilfs *nilfs, int nadd)
1373 {
1374 	struct nilfs_segment_buffer *segbuf, *prev;
1375 	struct inode *sufile = nilfs->ns_sufile;
1376 	__u64 nextnextnum;
1377 	LIST_HEAD(list);
1378 	int err, ret, i;
1379 
1380 	prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1381 	/*
1382 	 * Since the segment specified with nextnum might be allocated during
1383 	 * the previous construction, the buffer including its segusage may
1384 	 * not be dirty.  The following call ensures that the buffer is dirty
1385 	 * and will pin the buffer on memory until the sufile is written.
1386 	 */
1387 	err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1388 	if (unlikely(err))
1389 		return err;
1390 
1391 	for (i = 0; i < nadd; i++) {
1392 		/* extend segment info */
1393 		err = -ENOMEM;
1394 		segbuf = nilfs_segbuf_new(sci->sc_super);
1395 		if (unlikely(!segbuf))
1396 			goto failed;
1397 
1398 		/* map this buffer to region of segment on-disk */
1399 		nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1400 		sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1401 
1402 		/* allocate the next next full segment */
1403 		err = nilfs_sufile_alloc(sufile, &nextnextnum);
1404 		if (unlikely(err))
1405 			goto failed_segbuf;
1406 
1407 		segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1408 		nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1409 
1410 		list_add_tail(&segbuf->sb_list, &list);
1411 		prev = segbuf;
1412 	}
1413 	list_splice_tail(&list, &sci->sc_segbufs);
1414 	return 0;
1415 
1416  failed_segbuf:
1417 	nilfs_segbuf_free(segbuf);
1418  failed:
1419 	list_for_each_entry(segbuf, &list, sb_list) {
1420 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1421 		WARN_ON(ret); /* never fails */
1422 	}
1423 	nilfs_destroy_logs(&list);
1424 	return err;
1425 }
1426 
nilfs_free_incomplete_logs(struct list_head * logs,struct the_nilfs * nilfs)1427 static void nilfs_free_incomplete_logs(struct list_head *logs,
1428 				       struct the_nilfs *nilfs)
1429 {
1430 	struct nilfs_segment_buffer *segbuf, *prev;
1431 	struct inode *sufile = nilfs->ns_sufile;
1432 	int ret;
1433 
1434 	segbuf = NILFS_FIRST_SEGBUF(logs);
1435 	if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1436 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1437 		WARN_ON(ret); /* never fails */
1438 	}
1439 	if (atomic_read(&segbuf->sb_err)) {
1440 		/* Case 1: The first segment failed */
1441 		if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1442 			/*
1443 			 * Case 1a:  Partial segment appended into an existing
1444 			 * segment
1445 			 */
1446 			nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1447 						segbuf->sb_fseg_end);
1448 		else /* Case 1b:  New full segment */
1449 			set_nilfs_discontinued(nilfs);
1450 	}
1451 
1452 	prev = segbuf;
1453 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1454 		if (prev->sb_nextnum != segbuf->sb_nextnum) {
1455 			ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1456 			WARN_ON(ret); /* never fails */
1457 		}
1458 		if (atomic_read(&segbuf->sb_err) &&
1459 		    segbuf->sb_segnum != nilfs->ns_nextnum)
1460 			/* Case 2: extended segment (!= next) failed */
1461 			nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1462 		prev = segbuf;
1463 	}
1464 }
1465 
nilfs_segctor_update_segusage(struct nilfs_sc_info * sci,struct inode * sufile)1466 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1467 					  struct inode *sufile)
1468 {
1469 	struct nilfs_segment_buffer *segbuf;
1470 	unsigned long live_blocks;
1471 	int ret;
1472 
1473 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1474 		live_blocks = segbuf->sb_sum.nblocks +
1475 			(segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1476 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1477 						     live_blocks,
1478 						     sci->sc_seg_ctime);
1479 		WARN_ON(ret); /* always succeed because the segusage is dirty */
1480 	}
1481 }
1482 
nilfs_cancel_segusage(struct list_head * logs,struct inode * sufile)1483 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1484 {
1485 	struct nilfs_segment_buffer *segbuf;
1486 	int ret;
1487 
1488 	segbuf = NILFS_FIRST_SEGBUF(logs);
1489 	ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1490 					     segbuf->sb_pseg_start -
1491 					     segbuf->sb_fseg_start, 0);
1492 	WARN_ON(ret); /* always succeed because the segusage is dirty */
1493 
1494 	list_for_each_entry_continue(segbuf, logs, sb_list) {
1495 		ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1496 						     0, 0);
1497 		WARN_ON(ret); /* always succeed */
1498 	}
1499 }
1500 
nilfs_segctor_truncate_segments(struct nilfs_sc_info * sci,struct nilfs_segment_buffer * last,struct inode * sufile)1501 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1502 					    struct nilfs_segment_buffer *last,
1503 					    struct inode *sufile)
1504 {
1505 	struct nilfs_segment_buffer *segbuf = last;
1506 	int ret;
1507 
1508 	list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1509 		sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1510 		ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1511 		WARN_ON(ret);
1512 	}
1513 	nilfs_truncate_logs(&sci->sc_segbufs, last);
1514 }
1515 
1516 
nilfs_segctor_collect(struct nilfs_sc_info * sci,struct the_nilfs * nilfs,int mode)1517 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1518 				 struct the_nilfs *nilfs, int mode)
1519 {
1520 	struct nilfs_cstage prev_stage = sci->sc_stage;
1521 	int err, nadd = 1;
1522 
1523 	/* Collection retry loop */
1524 	for (;;) {
1525 		sci->sc_nblk_this_inc = 0;
1526 		sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1527 
1528 		err = nilfs_segctor_reset_segment_buffer(sci);
1529 		if (unlikely(err))
1530 			goto failed;
1531 
1532 		err = nilfs_segctor_collect_blocks(sci, mode);
1533 		sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1534 		if (!err)
1535 			break;
1536 
1537 		if (unlikely(err != -E2BIG))
1538 			goto failed;
1539 
1540 		/* The current segment is filled up */
1541 		if (mode != SC_LSEG_SR ||
1542 		    nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1543 			break;
1544 
1545 		nilfs_clear_logs(&sci->sc_segbufs);
1546 
1547 		if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1548 			err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1549 							sci->sc_freesegs,
1550 							sci->sc_nfreesegs,
1551 							NULL);
1552 			WARN_ON(err); /* do not happen */
1553 			sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1554 		}
1555 
1556 		err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1557 		if (unlikely(err))
1558 			return err;
1559 
1560 		nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1561 		sci->sc_stage = prev_stage;
1562 	}
1563 	nilfs_segctor_zeropad_segsum(sci);
1564 	nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1565 	return 0;
1566 
1567  failed:
1568 	return err;
1569 }
1570 
nilfs_list_replace_buffer(struct buffer_head * old_bh,struct buffer_head * new_bh)1571 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1572 				      struct buffer_head *new_bh)
1573 {
1574 	BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1575 
1576 	list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1577 	/* The caller must release old_bh */
1578 }
1579 
1580 static int
nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info * sci,struct nilfs_segment_buffer * segbuf,int mode)1581 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1582 				     struct nilfs_segment_buffer *segbuf,
1583 				     int mode)
1584 {
1585 	struct inode *inode = NULL;
1586 	sector_t blocknr;
1587 	unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1588 	unsigned long nblocks = 0, ndatablk = 0;
1589 	const struct nilfs_sc_operations *sc_op = NULL;
1590 	struct nilfs_segsum_pointer ssp;
1591 	struct nilfs_finfo *finfo = NULL;
1592 	union nilfs_binfo binfo;
1593 	struct buffer_head *bh, *bh_org;
1594 	ino_t ino = 0;
1595 	int err = 0;
1596 
1597 	if (!nfinfo)
1598 		goto out;
1599 
1600 	blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1601 	ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1602 	ssp.offset = sizeof(struct nilfs_segment_summary);
1603 
1604 	list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1605 		if (bh == segbuf->sb_super_root)
1606 			break;
1607 		if (!finfo) {
1608 			finfo =	nilfs_segctor_map_segsum_entry(
1609 				sci, &ssp, sizeof(*finfo));
1610 			ino = le64_to_cpu(finfo->fi_ino);
1611 			nblocks = le32_to_cpu(finfo->fi_nblocks);
1612 			ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1613 
1614 			inode = bh->b_page->mapping->host;
1615 
1616 			if (mode == SC_LSEG_DSYNC)
1617 				sc_op = &nilfs_sc_dsync_ops;
1618 			else if (ino == NILFS_DAT_INO)
1619 				sc_op = &nilfs_sc_dat_ops;
1620 			else /* file blocks */
1621 				sc_op = &nilfs_sc_file_ops;
1622 		}
1623 		bh_org = bh;
1624 		get_bh(bh_org);
1625 		err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1626 					&binfo);
1627 		if (bh != bh_org)
1628 			nilfs_list_replace_buffer(bh_org, bh);
1629 		brelse(bh_org);
1630 		if (unlikely(err))
1631 			goto failed_bmap;
1632 
1633 		if (ndatablk > 0)
1634 			sc_op->write_data_binfo(sci, &ssp, &binfo);
1635 		else
1636 			sc_op->write_node_binfo(sci, &ssp, &binfo);
1637 
1638 		blocknr++;
1639 		if (--nblocks == 0) {
1640 			finfo = NULL;
1641 			if (--nfinfo == 0)
1642 				break;
1643 		} else if (ndatablk > 0)
1644 			ndatablk--;
1645 	}
1646  out:
1647 	return 0;
1648 
1649  failed_bmap:
1650 	return err;
1651 }
1652 
nilfs_segctor_assign(struct nilfs_sc_info * sci,int mode)1653 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1654 {
1655 	struct nilfs_segment_buffer *segbuf;
1656 	int err;
1657 
1658 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1659 		err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1660 		if (unlikely(err))
1661 			return err;
1662 		nilfs_segbuf_fill_in_segsum(segbuf);
1663 	}
1664 	return 0;
1665 }
1666 
nilfs_begin_page_io(struct page * page)1667 static void nilfs_begin_page_io(struct page *page)
1668 {
1669 	if (!page || PageWriteback(page))
1670 		/*
1671 		 * For split b-tree node pages, this function may be called
1672 		 * twice.  We ignore the 2nd or later calls by this check.
1673 		 */
1674 		return;
1675 
1676 	lock_page(page);
1677 	clear_page_dirty_for_io(page);
1678 	set_page_writeback(page);
1679 	unlock_page(page);
1680 }
1681 
nilfs_segctor_prepare_write(struct nilfs_sc_info * sci)1682 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1683 {
1684 	struct nilfs_segment_buffer *segbuf;
1685 	struct page *bd_page = NULL, *fs_page = NULL;
1686 
1687 	list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1688 		struct buffer_head *bh;
1689 
1690 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1691 				    b_assoc_buffers) {
1692 			if (bh->b_page != bd_page) {
1693 				if (bd_page) {
1694 					lock_page(bd_page);
1695 					clear_page_dirty_for_io(bd_page);
1696 					set_page_writeback(bd_page);
1697 					unlock_page(bd_page);
1698 				}
1699 				bd_page = bh->b_page;
1700 			}
1701 		}
1702 
1703 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1704 				    b_assoc_buffers) {
1705 			if (bh == segbuf->sb_super_root) {
1706 				if (bh->b_page != bd_page) {
1707 					lock_page(bd_page);
1708 					clear_page_dirty_for_io(bd_page);
1709 					set_page_writeback(bd_page);
1710 					unlock_page(bd_page);
1711 					bd_page = bh->b_page;
1712 				}
1713 				break;
1714 			}
1715 			set_buffer_async_write(bh);
1716 			if (bh->b_page != fs_page) {
1717 				nilfs_begin_page_io(fs_page);
1718 				fs_page = bh->b_page;
1719 			}
1720 		}
1721 	}
1722 	if (bd_page) {
1723 		lock_page(bd_page);
1724 		clear_page_dirty_for_io(bd_page);
1725 		set_page_writeback(bd_page);
1726 		unlock_page(bd_page);
1727 	}
1728 	nilfs_begin_page_io(fs_page);
1729 }
1730 
nilfs_segctor_write(struct nilfs_sc_info * sci,struct the_nilfs * nilfs)1731 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1732 			       struct the_nilfs *nilfs)
1733 {
1734 	int ret;
1735 
1736 	ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1737 	list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1738 	return ret;
1739 }
1740 
nilfs_end_page_io(struct page * page,int err)1741 static void nilfs_end_page_io(struct page *page, int err)
1742 {
1743 	if (!page)
1744 		return;
1745 
1746 	if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1747 		/*
1748 		 * For b-tree node pages, this function may be called twice
1749 		 * or more because they might be split in a segment.
1750 		 */
1751 		if (PageDirty(page)) {
1752 			/*
1753 			 * For pages holding split b-tree node buffers, dirty
1754 			 * flag on the buffers may be cleared discretely.
1755 			 * In that case, the page is once redirtied for
1756 			 * remaining buffers, and it must be cancelled if
1757 			 * all the buffers get cleaned later.
1758 			 */
1759 			lock_page(page);
1760 			if (nilfs_page_buffers_clean(page))
1761 				__nilfs_clear_page_dirty(page);
1762 			unlock_page(page);
1763 		}
1764 		return;
1765 	}
1766 
1767 	if (!err) {
1768 		if (!nilfs_page_buffers_clean(page))
1769 			__set_page_dirty_nobuffers(page);
1770 		ClearPageError(page);
1771 	} else {
1772 		__set_page_dirty_nobuffers(page);
1773 		SetPageError(page);
1774 	}
1775 
1776 	end_page_writeback(page);
1777 }
1778 
nilfs_abort_logs(struct list_head * logs,int err)1779 static void nilfs_abort_logs(struct list_head *logs, int err)
1780 {
1781 	struct nilfs_segment_buffer *segbuf;
1782 	struct page *bd_page = NULL, *fs_page = NULL;
1783 	struct buffer_head *bh;
1784 
1785 	if (list_empty(logs))
1786 		return;
1787 
1788 	list_for_each_entry(segbuf, logs, sb_list) {
1789 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1790 				    b_assoc_buffers) {
1791 			clear_buffer_uptodate(bh);
1792 			if (bh->b_page != bd_page) {
1793 				if (bd_page)
1794 					end_page_writeback(bd_page);
1795 				bd_page = bh->b_page;
1796 			}
1797 		}
1798 
1799 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1800 				    b_assoc_buffers) {
1801 			if (bh == segbuf->sb_super_root) {
1802 				clear_buffer_uptodate(bh);
1803 				if (bh->b_page != bd_page) {
1804 					end_page_writeback(bd_page);
1805 					bd_page = bh->b_page;
1806 				}
1807 				break;
1808 			}
1809 			clear_buffer_async_write(bh);
1810 			if (bh->b_page != fs_page) {
1811 				nilfs_end_page_io(fs_page, err);
1812 				fs_page = bh->b_page;
1813 			}
1814 		}
1815 	}
1816 	if (bd_page)
1817 		end_page_writeback(bd_page);
1818 
1819 	nilfs_end_page_io(fs_page, err);
1820 }
1821 
nilfs_segctor_abort_construction(struct nilfs_sc_info * sci,struct the_nilfs * nilfs,int err)1822 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1823 					     struct the_nilfs *nilfs, int err)
1824 {
1825 	LIST_HEAD(logs);
1826 	int ret;
1827 
1828 	list_splice_tail_init(&sci->sc_write_logs, &logs);
1829 	ret = nilfs_wait_on_logs(&logs);
1830 	nilfs_abort_logs(&logs, ret ? : err);
1831 
1832 	list_splice_tail_init(&sci->sc_segbufs, &logs);
1833 	nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1834 	nilfs_free_incomplete_logs(&logs, nilfs);
1835 
1836 	if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1837 		ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1838 						sci->sc_freesegs,
1839 						sci->sc_nfreesegs,
1840 						NULL);
1841 		WARN_ON(ret); /* do not happen */
1842 	}
1843 
1844 	nilfs_destroy_logs(&logs);
1845 }
1846 
nilfs_set_next_segment(struct the_nilfs * nilfs,struct nilfs_segment_buffer * segbuf)1847 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1848 				   struct nilfs_segment_buffer *segbuf)
1849 {
1850 	nilfs->ns_segnum = segbuf->sb_segnum;
1851 	nilfs->ns_nextnum = segbuf->sb_nextnum;
1852 	nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1853 		+ segbuf->sb_sum.nblocks;
1854 	nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1855 	nilfs->ns_ctime = segbuf->sb_sum.ctime;
1856 }
1857 
nilfs_segctor_complete_write(struct nilfs_sc_info * sci)1858 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1859 {
1860 	struct nilfs_segment_buffer *segbuf;
1861 	struct page *bd_page = NULL, *fs_page = NULL;
1862 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1863 	int update_sr = false;
1864 
1865 	list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1866 		struct buffer_head *bh;
1867 
1868 		list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1869 				    b_assoc_buffers) {
1870 			set_buffer_uptodate(bh);
1871 			clear_buffer_dirty(bh);
1872 			if (bh->b_page != bd_page) {
1873 				if (bd_page)
1874 					end_page_writeback(bd_page);
1875 				bd_page = bh->b_page;
1876 			}
1877 		}
1878 		/*
1879 		 * We assume that the buffers which belong to the same page
1880 		 * continue over the buffer list.
1881 		 * Under this assumption, the last BHs of pages is
1882 		 * identifiable by the discontinuity of bh->b_page
1883 		 * (page != fs_page).
1884 		 *
1885 		 * For B-tree node blocks, however, this assumption is not
1886 		 * guaranteed.  The cleanup code of B-tree node pages needs
1887 		 * special care.
1888 		 */
1889 		list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1890 				    b_assoc_buffers) {
1891 			const unsigned long set_bits = BIT(BH_Uptodate);
1892 			const unsigned long clear_bits =
1893 				(BIT(BH_Dirty) | BIT(BH_Async_Write) |
1894 				 BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1895 				 BIT(BH_NILFS_Redirected));
1896 
1897 			if (bh == segbuf->sb_super_root) {
1898 				set_buffer_uptodate(bh);
1899 				clear_buffer_dirty(bh);
1900 				if (bh->b_page != bd_page) {
1901 					end_page_writeback(bd_page);
1902 					bd_page = bh->b_page;
1903 				}
1904 				update_sr = true;
1905 				break;
1906 			}
1907 			set_mask_bits(&bh->b_state, clear_bits, set_bits);
1908 			if (bh->b_page != fs_page) {
1909 				nilfs_end_page_io(fs_page, 0);
1910 				fs_page = bh->b_page;
1911 			}
1912 		}
1913 
1914 		if (!nilfs_segbuf_simplex(segbuf)) {
1915 			if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1916 				set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1917 				sci->sc_lseg_stime = jiffies;
1918 			}
1919 			if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1920 				clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1921 		}
1922 	}
1923 	/*
1924 	 * Since pages may continue over multiple segment buffers,
1925 	 * end of the last page must be checked outside of the loop.
1926 	 */
1927 	if (bd_page)
1928 		end_page_writeback(bd_page);
1929 
1930 	nilfs_end_page_io(fs_page, 0);
1931 
1932 	nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1933 
1934 	if (nilfs_doing_gc())
1935 		nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1936 	else
1937 		nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1938 
1939 	sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1940 
1941 	segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1942 	nilfs_set_next_segment(nilfs, segbuf);
1943 
1944 	if (update_sr) {
1945 		nilfs->ns_flushed_device = 0;
1946 		nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1947 				       segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1948 
1949 		clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1950 		clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1951 		set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1952 		nilfs_segctor_clear_metadata_dirty(sci);
1953 	} else
1954 		clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1955 }
1956 
nilfs_segctor_wait(struct nilfs_sc_info * sci)1957 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1958 {
1959 	int ret;
1960 
1961 	ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1962 	if (!ret) {
1963 		nilfs_segctor_complete_write(sci);
1964 		nilfs_destroy_logs(&sci->sc_write_logs);
1965 	}
1966 	return ret;
1967 }
1968 
nilfs_segctor_collect_dirty_files(struct nilfs_sc_info * sci,struct the_nilfs * nilfs)1969 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1970 					     struct the_nilfs *nilfs)
1971 {
1972 	struct nilfs_inode_info *ii, *n;
1973 	struct inode *ifile = sci->sc_root->ifile;
1974 
1975 	spin_lock(&nilfs->ns_inode_lock);
1976  retry:
1977 	list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1978 		if (!ii->i_bh) {
1979 			struct buffer_head *ibh;
1980 			int err;
1981 
1982 			spin_unlock(&nilfs->ns_inode_lock);
1983 			err = nilfs_ifile_get_inode_block(
1984 				ifile, ii->vfs_inode.i_ino, &ibh);
1985 			if (unlikely(err)) {
1986 				nilfs_warn(sci->sc_super,
1987 					   "log writer: error %d getting inode block (ino=%lu)",
1988 					   err, ii->vfs_inode.i_ino);
1989 				return err;
1990 			}
1991 			spin_lock(&nilfs->ns_inode_lock);
1992 			if (likely(!ii->i_bh))
1993 				ii->i_bh = ibh;
1994 			else
1995 				brelse(ibh);
1996 			goto retry;
1997 		}
1998 
1999 		// Always redirty the buffer to avoid race condition
2000 		mark_buffer_dirty(ii->i_bh);
2001 		nilfs_mdt_mark_dirty(ifile);
2002 
2003 		clear_bit(NILFS_I_QUEUED, &ii->i_state);
2004 		set_bit(NILFS_I_BUSY, &ii->i_state);
2005 		list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2006 	}
2007 	spin_unlock(&nilfs->ns_inode_lock);
2008 
2009 	return 0;
2010 }
2011 
nilfs_segctor_drop_written_files(struct nilfs_sc_info * sci,struct the_nilfs * nilfs)2012 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2013 					     struct the_nilfs *nilfs)
2014 {
2015 	struct nilfs_inode_info *ii, *n;
2016 	int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2017 	int defer_iput = false;
2018 
2019 	spin_lock(&nilfs->ns_inode_lock);
2020 	list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2021 		if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2022 		    test_bit(NILFS_I_DIRTY, &ii->i_state))
2023 			continue;
2024 
2025 		clear_bit(NILFS_I_BUSY, &ii->i_state);
2026 		brelse(ii->i_bh);
2027 		ii->i_bh = NULL;
2028 		list_del_init(&ii->i_dirty);
2029 		if (!ii->vfs_inode.i_nlink || during_mount) {
2030 			/*
2031 			 * Defer calling iput() to avoid deadlocks if
2032 			 * i_nlink == 0 or mount is not yet finished.
2033 			 */
2034 			list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2035 			defer_iput = true;
2036 		} else {
2037 			spin_unlock(&nilfs->ns_inode_lock);
2038 			iput(&ii->vfs_inode);
2039 			spin_lock(&nilfs->ns_inode_lock);
2040 		}
2041 	}
2042 	spin_unlock(&nilfs->ns_inode_lock);
2043 
2044 	if (defer_iput)
2045 		schedule_work(&sci->sc_iput_work);
2046 }
2047 
2048 /*
2049  * Main procedure of segment constructor
2050  */
nilfs_segctor_do_construct(struct nilfs_sc_info * sci,int mode)2051 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2052 {
2053 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2054 	int err;
2055 
2056 	if (sb_rdonly(sci->sc_super))
2057 		return -EROFS;
2058 
2059 	nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2060 	sci->sc_cno = nilfs->ns_cno;
2061 
2062 	err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2063 	if (unlikely(err))
2064 		goto out;
2065 
2066 	if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2067 		set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2068 
2069 	if (nilfs_segctor_clean(sci))
2070 		goto out;
2071 
2072 	do {
2073 		sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2074 
2075 		err = nilfs_segctor_begin_construction(sci, nilfs);
2076 		if (unlikely(err))
2077 			goto out;
2078 
2079 		/* Update time stamp */
2080 		sci->sc_seg_ctime = ktime_get_real_seconds();
2081 
2082 		err = nilfs_segctor_collect(sci, nilfs, mode);
2083 		if (unlikely(err))
2084 			goto failed;
2085 
2086 		/* Avoid empty segment */
2087 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2088 		    nilfs_segbuf_empty(sci->sc_curseg)) {
2089 			nilfs_segctor_abort_construction(sci, nilfs, 1);
2090 			goto out;
2091 		}
2092 
2093 		err = nilfs_segctor_assign(sci, mode);
2094 		if (unlikely(err))
2095 			goto failed;
2096 
2097 		if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2098 			nilfs_segctor_fill_in_file_bmap(sci);
2099 
2100 		if (mode == SC_LSEG_SR &&
2101 		    nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2102 			err = nilfs_segctor_fill_in_checkpoint(sci);
2103 			if (unlikely(err))
2104 				goto failed_to_write;
2105 
2106 			nilfs_segctor_fill_in_super_root(sci, nilfs);
2107 		}
2108 		nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2109 
2110 		/* Write partial segments */
2111 		nilfs_segctor_prepare_write(sci);
2112 
2113 		nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2114 					    nilfs->ns_crc_seed);
2115 
2116 		err = nilfs_segctor_write(sci, nilfs);
2117 		if (unlikely(err))
2118 			goto failed_to_write;
2119 
2120 		if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2121 		    nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2122 			/*
2123 			 * At this point, we avoid double buffering
2124 			 * for blocksize < pagesize because page dirty
2125 			 * flag is turned off during write and dirty
2126 			 * buffers are not properly collected for
2127 			 * pages crossing over segments.
2128 			 */
2129 			err = nilfs_segctor_wait(sci);
2130 			if (err)
2131 				goto failed_to_write;
2132 		}
2133 	} while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2134 
2135  out:
2136 	nilfs_segctor_drop_written_files(sci, nilfs);
2137 	return err;
2138 
2139  failed_to_write:
2140 	if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2141 		nilfs_redirty_inodes(&sci->sc_dirty_files);
2142 
2143  failed:
2144 	if (nilfs_doing_gc())
2145 		nilfs_redirty_inodes(&sci->sc_gc_inodes);
2146 	nilfs_segctor_abort_construction(sci, nilfs, err);
2147 	goto out;
2148 }
2149 
2150 /**
2151  * nilfs_segctor_start_timer - set timer of background write
2152  * @sci: nilfs_sc_info
2153  *
2154  * If the timer has already been set, it ignores the new request.
2155  * This function MUST be called within a section locking the segment
2156  * semaphore.
2157  */
nilfs_segctor_start_timer(struct nilfs_sc_info * sci)2158 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2159 {
2160 	spin_lock(&sci->sc_state_lock);
2161 	if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2162 		sci->sc_timer.expires = jiffies + sci->sc_interval;
2163 		add_timer(&sci->sc_timer);
2164 		sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2165 	}
2166 	spin_unlock(&sci->sc_state_lock);
2167 }
2168 
nilfs_segctor_do_flush(struct nilfs_sc_info * sci,int bn)2169 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2170 {
2171 	spin_lock(&sci->sc_state_lock);
2172 	if (!(sci->sc_flush_request & BIT(bn))) {
2173 		unsigned long prev_req = sci->sc_flush_request;
2174 
2175 		sci->sc_flush_request |= BIT(bn);
2176 		if (!prev_req)
2177 			wake_up(&sci->sc_wait_daemon);
2178 	}
2179 	spin_unlock(&sci->sc_state_lock);
2180 }
2181 
2182 /**
2183  * nilfs_flush_segment - trigger a segment construction for resource control
2184  * @sb: super block
2185  * @ino: inode number of the file to be flushed out.
2186  */
nilfs_flush_segment(struct super_block * sb,ino_t ino)2187 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2188 {
2189 	struct the_nilfs *nilfs = sb->s_fs_info;
2190 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2191 
2192 	if (!sci || nilfs_doing_construction())
2193 		return;
2194 	nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2195 					/* assign bit 0 to data files */
2196 }
2197 
2198 struct nilfs_segctor_wait_request {
2199 	wait_queue_entry_t	wq;
2200 	__u32		seq;
2201 	int		err;
2202 	atomic_t	done;
2203 };
2204 
nilfs_segctor_sync(struct nilfs_sc_info * sci)2205 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2206 {
2207 	struct nilfs_segctor_wait_request wait_req;
2208 	int err = 0;
2209 
2210 	spin_lock(&sci->sc_state_lock);
2211 	init_wait(&wait_req.wq);
2212 	wait_req.err = 0;
2213 	atomic_set(&wait_req.done, 0);
2214 	wait_req.seq = ++sci->sc_seq_request;
2215 	spin_unlock(&sci->sc_state_lock);
2216 
2217 	init_waitqueue_entry(&wait_req.wq, current);
2218 	add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2219 	set_current_state(TASK_INTERRUPTIBLE);
2220 	wake_up(&sci->sc_wait_daemon);
2221 
2222 	for (;;) {
2223 		if (atomic_read(&wait_req.done)) {
2224 			err = wait_req.err;
2225 			break;
2226 		}
2227 		if (!signal_pending(current)) {
2228 			schedule();
2229 			continue;
2230 		}
2231 		err = -ERESTARTSYS;
2232 		break;
2233 	}
2234 	finish_wait(&sci->sc_wait_request, &wait_req.wq);
2235 	return err;
2236 }
2237 
nilfs_segctor_wakeup(struct nilfs_sc_info * sci,int err)2238 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2239 {
2240 	struct nilfs_segctor_wait_request *wrq, *n;
2241 	unsigned long flags;
2242 
2243 	spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2244 	list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2245 		if (!atomic_read(&wrq->done) &&
2246 		    nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2247 			wrq->err = err;
2248 			atomic_set(&wrq->done, 1);
2249 		}
2250 		if (atomic_read(&wrq->done)) {
2251 			wrq->wq.func(&wrq->wq,
2252 				     TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2253 				     0, NULL);
2254 		}
2255 	}
2256 	spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2257 }
2258 
2259 /**
2260  * nilfs_construct_segment - construct a logical segment
2261  * @sb: super block
2262  *
2263  * Return Value: On success, 0 is returned. On errors, one of the following
2264  * negative error code is returned.
2265  *
2266  * %-EROFS - Read only filesystem.
2267  *
2268  * %-EIO - I/O error
2269  *
2270  * %-ENOSPC - No space left on device (only in a panic state).
2271  *
2272  * %-ERESTARTSYS - Interrupted.
2273  *
2274  * %-ENOMEM - Insufficient memory available.
2275  */
nilfs_construct_segment(struct super_block * sb)2276 int nilfs_construct_segment(struct super_block *sb)
2277 {
2278 	struct the_nilfs *nilfs = sb->s_fs_info;
2279 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2280 	struct nilfs_transaction_info *ti;
2281 
2282 	if (sb_rdonly(sb) || unlikely(!sci))
2283 		return -EROFS;
2284 
2285 	/* A call inside transactions causes a deadlock. */
2286 	BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2287 
2288 	return nilfs_segctor_sync(sci);
2289 }
2290 
2291 /**
2292  * nilfs_construct_dsync_segment - construct a data-only logical segment
2293  * @sb: super block
2294  * @inode: inode whose data blocks should be written out
2295  * @start: start byte offset
2296  * @end: end byte offset (inclusive)
2297  *
2298  * Return Value: On success, 0 is returned. On errors, one of the following
2299  * negative error code is returned.
2300  *
2301  * %-EROFS - Read only filesystem.
2302  *
2303  * %-EIO - I/O error
2304  *
2305  * %-ENOSPC - No space left on device (only in a panic state).
2306  *
2307  * %-ERESTARTSYS - Interrupted.
2308  *
2309  * %-ENOMEM - Insufficient memory available.
2310  */
nilfs_construct_dsync_segment(struct super_block * sb,struct inode * inode,loff_t start,loff_t end)2311 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2312 				  loff_t start, loff_t end)
2313 {
2314 	struct the_nilfs *nilfs = sb->s_fs_info;
2315 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2316 	struct nilfs_inode_info *ii;
2317 	struct nilfs_transaction_info ti;
2318 	int err = 0;
2319 
2320 	if (sb_rdonly(sb) || unlikely(!sci))
2321 		return -EROFS;
2322 
2323 	nilfs_transaction_lock(sb, &ti, 0);
2324 
2325 	ii = NILFS_I(inode);
2326 	if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2327 	    nilfs_test_opt(nilfs, STRICT_ORDER) ||
2328 	    test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2329 	    nilfs_discontinued(nilfs)) {
2330 		nilfs_transaction_unlock(sb);
2331 		err = nilfs_segctor_sync(sci);
2332 		return err;
2333 	}
2334 
2335 	spin_lock(&nilfs->ns_inode_lock);
2336 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2337 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2338 		spin_unlock(&nilfs->ns_inode_lock);
2339 		nilfs_transaction_unlock(sb);
2340 		return 0;
2341 	}
2342 	spin_unlock(&nilfs->ns_inode_lock);
2343 	sci->sc_dsync_inode = ii;
2344 	sci->sc_dsync_start = start;
2345 	sci->sc_dsync_end = end;
2346 
2347 	err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2348 	if (!err)
2349 		nilfs->ns_flushed_device = 0;
2350 
2351 	nilfs_transaction_unlock(sb);
2352 	return err;
2353 }
2354 
2355 #define FLUSH_FILE_BIT	(0x1) /* data file only */
2356 #define FLUSH_DAT_BIT	BIT(NILFS_DAT_INO) /* DAT only */
2357 
2358 /**
2359  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2360  * @sci: segment constructor object
2361  */
nilfs_segctor_accept(struct nilfs_sc_info * sci)2362 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2363 {
2364 	spin_lock(&sci->sc_state_lock);
2365 	sci->sc_seq_accepted = sci->sc_seq_request;
2366 	spin_unlock(&sci->sc_state_lock);
2367 	del_timer_sync(&sci->sc_timer);
2368 }
2369 
2370 /**
2371  * nilfs_segctor_notify - notify the result of request to caller threads
2372  * @sci: segment constructor object
2373  * @mode: mode of log forming
2374  * @err: error code to be notified
2375  */
nilfs_segctor_notify(struct nilfs_sc_info * sci,int mode,int err)2376 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2377 {
2378 	/* Clear requests (even when the construction failed) */
2379 	spin_lock(&sci->sc_state_lock);
2380 
2381 	if (mode == SC_LSEG_SR) {
2382 		sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2383 		sci->sc_seq_done = sci->sc_seq_accepted;
2384 		nilfs_segctor_wakeup(sci, err);
2385 		sci->sc_flush_request = 0;
2386 	} else {
2387 		if (mode == SC_FLUSH_FILE)
2388 			sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2389 		else if (mode == SC_FLUSH_DAT)
2390 			sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2391 
2392 		/* re-enable timer if checkpoint creation was not done */
2393 		if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2394 		    time_before(jiffies, sci->sc_timer.expires))
2395 			add_timer(&sci->sc_timer);
2396 	}
2397 	spin_unlock(&sci->sc_state_lock);
2398 }
2399 
2400 /**
2401  * nilfs_segctor_construct - form logs and write them to disk
2402  * @sci: segment constructor object
2403  * @mode: mode of log forming
2404  */
nilfs_segctor_construct(struct nilfs_sc_info * sci,int mode)2405 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2406 {
2407 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2408 	struct nilfs_super_block **sbp;
2409 	int err = 0;
2410 
2411 	nilfs_segctor_accept(sci);
2412 
2413 	if (nilfs_discontinued(nilfs))
2414 		mode = SC_LSEG_SR;
2415 	if (!nilfs_segctor_confirm(sci))
2416 		err = nilfs_segctor_do_construct(sci, mode);
2417 
2418 	if (likely(!err)) {
2419 		if (mode != SC_FLUSH_DAT)
2420 			atomic_set(&nilfs->ns_ndirtyblks, 0);
2421 		if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2422 		    nilfs_discontinued(nilfs)) {
2423 			down_write(&nilfs->ns_sem);
2424 			err = -EIO;
2425 			sbp = nilfs_prepare_super(sci->sc_super,
2426 						  nilfs_sb_will_flip(nilfs));
2427 			if (likely(sbp)) {
2428 				nilfs_set_log_cursor(sbp[0], nilfs);
2429 				err = nilfs_commit_super(sci->sc_super,
2430 							 NILFS_SB_COMMIT);
2431 			}
2432 			up_write(&nilfs->ns_sem);
2433 		}
2434 	}
2435 
2436 	nilfs_segctor_notify(sci, mode, err);
2437 	return err;
2438 }
2439 
nilfs_construction_timeout(struct timer_list * t)2440 static void nilfs_construction_timeout(struct timer_list *t)
2441 {
2442 	struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2443 
2444 	wake_up_process(sci->sc_timer_task);
2445 }
2446 
2447 static void
nilfs_remove_written_gcinodes(struct the_nilfs * nilfs,struct list_head * head)2448 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2449 {
2450 	struct nilfs_inode_info *ii, *n;
2451 
2452 	list_for_each_entry_safe(ii, n, head, i_dirty) {
2453 		if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2454 			continue;
2455 		list_del_init(&ii->i_dirty);
2456 		truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2457 		nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2458 		iput(&ii->vfs_inode);
2459 	}
2460 }
2461 
nilfs_clean_segments(struct super_block * sb,struct nilfs_argv * argv,void ** kbufs)2462 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2463 			 void **kbufs)
2464 {
2465 	struct the_nilfs *nilfs = sb->s_fs_info;
2466 	struct nilfs_sc_info *sci = nilfs->ns_writer;
2467 	struct nilfs_transaction_info ti;
2468 	int err;
2469 
2470 	if (unlikely(!sci))
2471 		return -EROFS;
2472 
2473 	nilfs_transaction_lock(sb, &ti, 1);
2474 
2475 	err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2476 	if (unlikely(err))
2477 		goto out_unlock;
2478 
2479 	err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2480 	if (unlikely(err)) {
2481 		nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2482 		goto out_unlock;
2483 	}
2484 
2485 	sci->sc_freesegs = kbufs[4];
2486 	sci->sc_nfreesegs = argv[4].v_nmembs;
2487 	list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2488 
2489 	for (;;) {
2490 		err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2491 		nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2492 
2493 		if (likely(!err))
2494 			break;
2495 
2496 		nilfs_warn(sb, "error %d cleaning segments", err);
2497 		set_current_state(TASK_INTERRUPTIBLE);
2498 		schedule_timeout(sci->sc_interval);
2499 	}
2500 	if (nilfs_test_opt(nilfs, DISCARD)) {
2501 		int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2502 						 sci->sc_nfreesegs);
2503 		if (ret) {
2504 			nilfs_warn(sb,
2505 				   "error %d on discard request, turning discards off for the device",
2506 				   ret);
2507 			nilfs_clear_opt(nilfs, DISCARD);
2508 		}
2509 	}
2510 
2511  out_unlock:
2512 	sci->sc_freesegs = NULL;
2513 	sci->sc_nfreesegs = 0;
2514 	nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2515 	nilfs_transaction_unlock(sb);
2516 	return err;
2517 }
2518 
nilfs_segctor_thread_construct(struct nilfs_sc_info * sci,int mode)2519 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2520 {
2521 	struct nilfs_transaction_info ti;
2522 
2523 	nilfs_transaction_lock(sci->sc_super, &ti, 0);
2524 	nilfs_segctor_construct(sci, mode);
2525 
2526 	/*
2527 	 * Unclosed segment should be retried.  We do this using sc_timer.
2528 	 * Timeout of sc_timer will invoke complete construction which leads
2529 	 * to close the current logical segment.
2530 	 */
2531 	if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2532 		nilfs_segctor_start_timer(sci);
2533 
2534 	nilfs_transaction_unlock(sci->sc_super);
2535 }
2536 
nilfs_segctor_do_immediate_flush(struct nilfs_sc_info * sci)2537 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2538 {
2539 	int mode = 0;
2540 
2541 	spin_lock(&sci->sc_state_lock);
2542 	mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2543 		SC_FLUSH_DAT : SC_FLUSH_FILE;
2544 	spin_unlock(&sci->sc_state_lock);
2545 
2546 	if (mode) {
2547 		nilfs_segctor_do_construct(sci, mode);
2548 
2549 		spin_lock(&sci->sc_state_lock);
2550 		sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2551 			~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2552 		spin_unlock(&sci->sc_state_lock);
2553 	}
2554 	clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2555 }
2556 
nilfs_segctor_flush_mode(struct nilfs_sc_info * sci)2557 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2558 {
2559 	if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2560 	    time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2561 		if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2562 			return SC_FLUSH_FILE;
2563 		else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2564 			return SC_FLUSH_DAT;
2565 	}
2566 	return SC_LSEG_SR;
2567 }
2568 
2569 /**
2570  * nilfs_segctor_thread - main loop of the segment constructor thread.
2571  * @arg: pointer to a struct nilfs_sc_info.
2572  *
2573  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2574  * to execute segment constructions.
2575  */
nilfs_segctor_thread(void * arg)2576 static int nilfs_segctor_thread(void *arg)
2577 {
2578 	struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2579 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2580 	int timeout = 0;
2581 
2582 	sci->sc_timer_task = current;
2583 
2584 	/* start sync. */
2585 	sci->sc_task = current;
2586 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2587 	nilfs_info(sci->sc_super,
2588 		   "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2589 		   sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2590 
2591 	spin_lock(&sci->sc_state_lock);
2592  loop:
2593 	for (;;) {
2594 		int mode;
2595 
2596 		if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2597 			goto end_thread;
2598 
2599 		if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2600 			mode = SC_LSEG_SR;
2601 		else if (sci->sc_flush_request)
2602 			mode = nilfs_segctor_flush_mode(sci);
2603 		else
2604 			break;
2605 
2606 		spin_unlock(&sci->sc_state_lock);
2607 		nilfs_segctor_thread_construct(sci, mode);
2608 		spin_lock(&sci->sc_state_lock);
2609 		timeout = 0;
2610 	}
2611 
2612 
2613 	if (freezing(current)) {
2614 		spin_unlock(&sci->sc_state_lock);
2615 		try_to_freeze();
2616 		spin_lock(&sci->sc_state_lock);
2617 	} else {
2618 		DEFINE_WAIT(wait);
2619 		int should_sleep = 1;
2620 
2621 		prepare_to_wait(&sci->sc_wait_daemon, &wait,
2622 				TASK_INTERRUPTIBLE);
2623 
2624 		if (sci->sc_seq_request != sci->sc_seq_done)
2625 			should_sleep = 0;
2626 		else if (sci->sc_flush_request)
2627 			should_sleep = 0;
2628 		else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2629 			should_sleep = time_before(jiffies,
2630 					sci->sc_timer.expires);
2631 
2632 		if (should_sleep) {
2633 			spin_unlock(&sci->sc_state_lock);
2634 			schedule();
2635 			spin_lock(&sci->sc_state_lock);
2636 		}
2637 		finish_wait(&sci->sc_wait_daemon, &wait);
2638 		timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2639 			   time_after_eq(jiffies, sci->sc_timer.expires));
2640 
2641 		if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2642 			set_nilfs_discontinued(nilfs);
2643 	}
2644 	goto loop;
2645 
2646  end_thread:
2647 	/* end sync. */
2648 	sci->sc_task = NULL;
2649 	wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2650 	spin_unlock(&sci->sc_state_lock);
2651 	return 0;
2652 }
2653 
nilfs_segctor_start_thread(struct nilfs_sc_info * sci)2654 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2655 {
2656 	struct task_struct *t;
2657 
2658 	t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2659 	if (IS_ERR(t)) {
2660 		int err = PTR_ERR(t);
2661 
2662 		nilfs_err(sci->sc_super, "error %d creating segctord thread",
2663 			  err);
2664 		return err;
2665 	}
2666 	wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2667 	return 0;
2668 }
2669 
nilfs_segctor_kill_thread(struct nilfs_sc_info * sci)2670 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2671 	__acquires(&sci->sc_state_lock)
2672 	__releases(&sci->sc_state_lock)
2673 {
2674 	sci->sc_state |= NILFS_SEGCTOR_QUIT;
2675 
2676 	while (sci->sc_task) {
2677 		wake_up(&sci->sc_wait_daemon);
2678 		spin_unlock(&sci->sc_state_lock);
2679 		wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2680 		spin_lock(&sci->sc_state_lock);
2681 	}
2682 }
2683 
2684 /*
2685  * Setup & clean-up functions
2686  */
nilfs_segctor_new(struct super_block * sb,struct nilfs_root * root)2687 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2688 					       struct nilfs_root *root)
2689 {
2690 	struct the_nilfs *nilfs = sb->s_fs_info;
2691 	struct nilfs_sc_info *sci;
2692 
2693 	sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2694 	if (!sci)
2695 		return NULL;
2696 
2697 	sci->sc_super = sb;
2698 
2699 	nilfs_get_root(root);
2700 	sci->sc_root = root;
2701 
2702 	init_waitqueue_head(&sci->sc_wait_request);
2703 	init_waitqueue_head(&sci->sc_wait_daemon);
2704 	init_waitqueue_head(&sci->sc_wait_task);
2705 	spin_lock_init(&sci->sc_state_lock);
2706 	INIT_LIST_HEAD(&sci->sc_dirty_files);
2707 	INIT_LIST_HEAD(&sci->sc_segbufs);
2708 	INIT_LIST_HEAD(&sci->sc_write_logs);
2709 	INIT_LIST_HEAD(&sci->sc_gc_inodes);
2710 	INIT_LIST_HEAD(&sci->sc_iput_queue);
2711 	INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2712 	timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2713 
2714 	sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2715 	sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2716 	sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2717 
2718 	if (nilfs->ns_interval)
2719 		sci->sc_interval = HZ * nilfs->ns_interval;
2720 	if (nilfs->ns_watermark)
2721 		sci->sc_watermark = nilfs->ns_watermark;
2722 	return sci;
2723 }
2724 
nilfs_segctor_write_out(struct nilfs_sc_info * sci)2725 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2726 {
2727 	int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2728 
2729 	/*
2730 	 * The segctord thread was stopped and its timer was removed.
2731 	 * But some tasks remain.
2732 	 */
2733 	do {
2734 		struct nilfs_transaction_info ti;
2735 
2736 		nilfs_transaction_lock(sci->sc_super, &ti, 0);
2737 		ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2738 		nilfs_transaction_unlock(sci->sc_super);
2739 
2740 		flush_work(&sci->sc_iput_work);
2741 
2742 	} while (ret && ret != -EROFS && retrycount-- > 0);
2743 }
2744 
2745 /**
2746  * nilfs_segctor_destroy - destroy the segment constructor.
2747  * @sci: nilfs_sc_info
2748  *
2749  * nilfs_segctor_destroy() kills the segctord thread and frees
2750  * the nilfs_sc_info struct.
2751  * Caller must hold the segment semaphore.
2752  */
nilfs_segctor_destroy(struct nilfs_sc_info * sci)2753 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2754 {
2755 	struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2756 	int flag;
2757 
2758 	up_write(&nilfs->ns_segctor_sem);
2759 
2760 	spin_lock(&sci->sc_state_lock);
2761 	nilfs_segctor_kill_thread(sci);
2762 	flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2763 		|| sci->sc_seq_request != sci->sc_seq_done);
2764 	spin_unlock(&sci->sc_state_lock);
2765 
2766 	if (flush_work(&sci->sc_iput_work))
2767 		flag = true;
2768 
2769 	if (flag || !nilfs_segctor_confirm(sci))
2770 		nilfs_segctor_write_out(sci);
2771 
2772 	if (!list_empty(&sci->sc_dirty_files)) {
2773 		nilfs_warn(sci->sc_super,
2774 			   "disposed unprocessed dirty file(s) when stopping log writer");
2775 		nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2776 	}
2777 
2778 	if (!list_empty(&sci->sc_iput_queue)) {
2779 		nilfs_warn(sci->sc_super,
2780 			   "disposed unprocessed inode(s) in iput queue when stopping log writer");
2781 		nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2782 	}
2783 
2784 	WARN_ON(!list_empty(&sci->sc_segbufs));
2785 	WARN_ON(!list_empty(&sci->sc_write_logs));
2786 
2787 	nilfs_put_root(sci->sc_root);
2788 
2789 	down_write(&nilfs->ns_segctor_sem);
2790 
2791 	del_timer_sync(&sci->sc_timer);
2792 	kfree(sci);
2793 }
2794 
2795 /**
2796  * nilfs_attach_log_writer - attach log writer
2797  * @sb: super block instance
2798  * @root: root object of the current filesystem tree
2799  *
2800  * This allocates a log writer object, initializes it, and starts the
2801  * log writer.
2802  *
2803  * Return Value: On success, 0 is returned. On error, one of the following
2804  * negative error code is returned.
2805  *
2806  * %-ENOMEM - Insufficient memory available.
2807  */
nilfs_attach_log_writer(struct super_block * sb,struct nilfs_root * root)2808 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2809 {
2810 	struct the_nilfs *nilfs = sb->s_fs_info;
2811 	int err;
2812 
2813 	if (nilfs->ns_writer) {
2814 		/*
2815 		 * This happens if the filesystem is made read-only by
2816 		 * __nilfs_error or nilfs_remount and then remounted
2817 		 * read/write.  In these cases, reuse the existing
2818 		 * writer.
2819 		 */
2820 		return 0;
2821 	}
2822 
2823 	nilfs->ns_writer = nilfs_segctor_new(sb, root);
2824 	if (!nilfs->ns_writer)
2825 		return -ENOMEM;
2826 
2827 	inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2828 
2829 	err = nilfs_segctor_start_thread(nilfs->ns_writer);
2830 	if (unlikely(err))
2831 		nilfs_detach_log_writer(sb);
2832 
2833 	return err;
2834 }
2835 
2836 /**
2837  * nilfs_detach_log_writer - destroy log writer
2838  * @sb: super block instance
2839  *
2840  * This kills log writer daemon, frees the log writer object, and
2841  * destroys list of dirty files.
2842  */
nilfs_detach_log_writer(struct super_block * sb)2843 void nilfs_detach_log_writer(struct super_block *sb)
2844 {
2845 	struct the_nilfs *nilfs = sb->s_fs_info;
2846 	LIST_HEAD(garbage_list);
2847 
2848 	down_write(&nilfs->ns_segctor_sem);
2849 	if (nilfs->ns_writer) {
2850 		nilfs_segctor_destroy(nilfs->ns_writer);
2851 		nilfs->ns_writer = NULL;
2852 	}
2853 	set_nilfs_purging(nilfs);
2854 
2855 	/* Force to free the list of dirty files */
2856 	spin_lock(&nilfs->ns_inode_lock);
2857 	if (!list_empty(&nilfs->ns_dirty_files)) {
2858 		list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2859 		nilfs_warn(sb,
2860 			   "disposed unprocessed dirty file(s) when detaching log writer");
2861 	}
2862 	spin_unlock(&nilfs->ns_inode_lock);
2863 	up_write(&nilfs->ns_segctor_sem);
2864 
2865 	nilfs_dispose_list(nilfs, &garbage_list, 1);
2866 	clear_nilfs_purging(nilfs);
2867 }
2868