projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: detect congestion of flush command issues
[deliverable/linux.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
351df4b2
JK		 19
20#include "f2fs.h"
21#include "segment.h"
22#include "node.h"
9e4ded3f 23#include "trace.h"
6ec178da 24#include <trace/events/f2fs.h>
351df4b2 25
9a7f143a
CL		 26#define __reverse_ffz(x) __reverse_ffs(~(x))
27
7fd9e544 28static struct kmem_cache *discard_entry_slab;
184a5cd2 29static struct kmem_cache *sit_entry_set_slab;
88b88a66 30static struct kmem_cache *inmem_entry_slab;
7fd9e544 31
f96999c3
JK		 32static unsigned long __reverse_ulong(unsigned char *str)
33{
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
36
37#if BITS_PER_LONG == 64
38 shift = 56;
39#endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
43 }
44 return tmp;
45}
46
9a7f143a
CL		 47/*
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
50 */
51static inline unsigned long __reverse_ffs(unsigned long word)
52{
53 int num = 0;
54
55#if BITS_PER_LONG == 64
f96999c3 56 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 57 num += 32;
f96999c3 58 else
9a7f143a 59 word >>= 32;
9a7f143a 60#endif
f96999c3 61 if ((word & 0xffff0000) == 0)
9a7f143a 62 num += 16;
f96999c3 63 else
9a7f143a 64 word >>= 16;
f96999c3
JK		 65
66 if ((word & 0xff00) == 0)
9a7f143a 67 num += 8;
f96999c3 68 else
9a7f143a 69 word >>= 8;
f96999c3 70
9a7f143a
CL		 71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
f96999c3 75
9a7f143a
CL		 76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
f96999c3 80
9a7f143a
CL		 81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
84}
85
86/*
e1c42045 87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 89 * @size must be integral times of unsigned long.
9a7f143a 90 * Example:
f96999c3
JK		 91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 94 */
95static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
97{
98 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 99 unsigned long result = size;
9a7f143a 100 unsigned long tmp;
9a7f143a
CL		 101
102 if (offset >= size)
103 return size;
104
692223d1 105 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 106 offset %= BITS_PER_LONG;
f96999c3 107
692223d1
FL		 108 while (1) {
109 if (*p == 0)
110 goto pass;
9a7f143a 111
f96999c3 112 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 113
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 117 if (tmp)
692223d1
FL		 118 goto found;
119pass:
120 if (size <= BITS_PER_LONG)
121 break;
9a7f143a 122 size -= BITS_PER_LONG;
692223d1 123 offset = 0;
f96999c3 124 p++;
9a7f143a 125 }
692223d1
FL		 126 return result;
127found:
128 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 129}
130
131static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
133{
134 const unsigned long *p = addr + BIT_WORD(offset);
80609448 135 unsigned long result = size;
9a7f143a 136 unsigned long tmp;
9a7f143a
CL		 137
138 if (offset >= size)
139 return size;
140
80609448 141 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 142 offset %= BITS_PER_LONG;
80609448
JK		 143
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
147
f96999c3 148 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 149
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
f96999c3 154 if (tmp != ~0UL)
80609448
JK		 155 goto found;
156pass:
157 if (size <= BITS_PER_LONG)
158 break;
9a7f143a 159 size -= BITS_PER_LONG;
80609448 160 offset = 0;
f96999c3 161 p++;
9a7f143a 162 }
80609448
JK		 163 return result;
164found:
165 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 166}
167
88b88a66
JK		 168void register_inmem_page(struct inode *inode, struct page *page)
169{
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
9be32d72 172
9e4ded3f 173 f2fs_trace_pid(page);
0722b101 174
decd36b6
CY		 175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
177
88b88a66
JK		 178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
179
180 /* add atomic page indices to the list */
181 new->page = page;
182 INIT_LIST_HEAD(&new->list);
decd36b6 183
88b88a66
JK		 184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
186 get_page(page);
187 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 189 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 190
191 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 192}
193
28bc106b
CY		 194static int __revoke_inmem_pages(struct inode *inode,
195 struct list_head *head, bool drop, bool recover)
29b96b54 196{
28bc106b 197 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 198 struct inmem_pages *cur, *tmp;
28bc106b 199 int err = 0;
29b96b54
CY		 200
201 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 202 struct page *page = cur->page;
203
204 if (drop)
205 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
206
207 lock_page(page);
29b96b54 208
28bc106b
CY		 209 if (recover) {
210 struct dnode_of_data dn;
211 struct node_info ni;
212
213 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
214
215 set_new_dnode(&dn, inode, NULL, NULL, 0);
216 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
217 err = -EAGAIN;
218 goto next;
219 }
220 get_node_info(sbi, dn.nid, &ni);
221 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
222 cur->old_addr, ni.version, true, true);
223 f2fs_put_dnode(&dn);
224 }
225next:
63c52d78
JK		 226 /* we don't need to invalidate this in the sccessful status */
227 if (drop || recover)
228 ClearPageUptodate(page);
28bc106b 229 set_page_private(page, 0);
c81ced05 230 ClearPagePrivate(page);
28bc106b 231 f2fs_put_page(page, 1);
29b96b54
CY		 232
233 list_del(&cur->list);
234 kmem_cache_free(inmem_entry_slab, cur);
235 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
236 }
28bc106b 237 return err;
29b96b54
CY		 238}
239
240void drop_inmem_pages(struct inode *inode)
241{
242 struct f2fs_inode_info *fi = F2FS_I(inode);
243
91942321 244 clear_inode_flag(inode, FI_ATOMIC_FILE);
26dc3d44 245
29b96b54 246 mutex_lock(&fi->inmem_lock);
28bc106b 247 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54
CY		 248 mutex_unlock(&fi->inmem_lock);
249}
250
28bc106b
CY		 251static int __commit_inmem_pages(struct inode *inode,
252 struct list_head *revoke_list)
88b88a66
JK		 253{
254 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct inmem_pages *cur, *tmp;
88b88a66 257 struct f2fs_io_info fio = {
05ca3632 258 .sbi = sbi,
88b88a66 259 .type = DATA,
1e84371f 260 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 261 .encrypted_page = NULL,
88b88a66 262 };
29b96b54 263 bool submit_bio = false;
edb27dee 264 int err = 0;
88b88a66 265
88b88a66 266 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 267 struct page *page = cur->page;
268
269 lock_page(page);
270 if (page->mapping == inode->i_mapping) {
271 trace_f2fs_commit_inmem_page(page, INMEM);
272
273 set_page_dirty(page);
274 f2fs_wait_on_page_writeback(page, DATA, true);
275 if (clear_page_dirty_for_io(page))
29b96b54 276 inode_dec_dirty_pages(inode);
28bc106b
CY		 277
278 fio.page = page;
29b96b54
CY		 279 err = do_write_data_page(&fio);
280 if (err) {
28bc106b 281 unlock_page(page);
29b96b54 282 break;
70c640b1 283 }
29b96b54 284
28bc106b
CY		 285 /* record old blkaddr for revoking */
286 cur->old_addr = fio.old_blkaddr;
decd36b6 287
28bc106b
CY		 288 clear_cold_data(page);
289 submit_bio = true;
290 }
291 unlock_page(page);
292 list_move_tail(&cur->list, revoke_list);
88b88a66 293 }
29b96b54
CY		 294
295 if (submit_bio)
296 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
28bc106b
CY		 297
298 if (!err)
299 __revoke_inmem_pages(inode, revoke_list, false, false);
300
29b96b54
CY		 301 return err;
302}
303
304int commit_inmem_pages(struct inode *inode)
305{
306 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
307 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 308 struct list_head revoke_list;
309 int err;
29b96b54 310
28bc106b 311 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 312 f2fs_balance_fs(sbi, true);
313 f2fs_lock_op(sbi);
314
315 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 316 err = __commit_inmem_pages(inode, &revoke_list);
317 if (err) {
318 int ret;
319 /*
320 * try to revoke all committed pages, but still we could fail
321 * due to no memory or other reason, if that happened, EAGAIN
322 * will be returned, which means in such case, transaction is
323 * already not integrity, caller should use journal to do the
324 * recovery or rewrite & commit last transaction. For other
325 * error number, revoking was done by filesystem itself.
326 */
327 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
328 if (ret)
329 err = ret;
330
331 /* drop all uncommitted pages */
332 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
333 }
88b88a66
JK		 334 mutex_unlock(&fi->inmem_lock);
335
29b96b54 336 f2fs_unlock_op(sbi);
edb27dee 337 return err;
88b88a66
JK		 338}
339
0a8165d7 340/*
351df4b2
JK		 341 * This function balances dirty node and dentry pages.
342 * In addition, it controls garbage collection.
343 */
2c4db1a6 344void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 345{
2c4db1a6
JK		 346 if (!need)
347 return;
351df4b2 348 /*
029cd28c
JK		 349 * We should do GC or end up with checkpoint, if there are so many dirty
350 * dir/node pages without enough free segments.
351df4b2 351 */
43727527 352 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 353 mutex_lock(&sbi->gc_mutex);
d530d4d8 354 f2fs_gc(sbi, false);
351df4b2
JK		 355 }
356}
357
4660f9c0
JK		 358void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
359{
1dcc336b 360 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 361 if (!available_free_memory(sbi, EXTENT_CACHE))
362 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 363
1b38dc8e
JK		 364 /* check the # of cached NAT entries */
365 if (!available_free_memory(sbi, NAT_ENTRIES))
366 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
367
31696580
CY		 368 if (!available_free_memory(sbi, FREE_NIDS))
369 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
370
1b38dc8e
JK		 371 /* checkpoint is the only way to shrink partial cached entries */
372 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 373 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 374 excess_prefree_segs(sbi) ||
375 excess_dirty_nats(sbi) ||
d0239e1b 376 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
e9f5b8b8
CY		 377 if (test_opt(sbi, DATA_FLUSH)) {
378 struct blk_plug plug;
379
380 blk_start_plug(&plug);
36b35a0d 381 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 382 blk_finish_plug(&plug);
383 }
4660f9c0 384 f2fs_sync_fs(sbi->sb, true);
42190d2a 385 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 386 }
4660f9c0
JK		 387}
388
2163d198 389static int issue_flush_thread(void *data)
6b4afdd7
JK		 390{
391 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 392 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
393 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 394repeat:
395 if (kthread_should_stop())
396 return 0;
397
721bd4d5 398 if (!llist_empty(&fcc->issue_list)) {
740432f8 399 struct bio *bio;
6b4afdd7
JK		 400 struct flush_cmd *cmd, *next;
401 int ret;
402
740432f8
JK		 403 bio = f2fs_bio_alloc(0);
404
721bd4d5
GZ		 405 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
406 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
407
6b4afdd7
JK		 408 bio->bi_bdev = sbi->sb->s_bdev;
409 ret = submit_bio_wait(WRITE_FLUSH, bio);
410
721bd4d5
GZ		 411 llist_for_each_entry_safe(cmd, next,
412 fcc->dispatch_list, llnode) {
6b4afdd7 413 cmd->ret = ret;
6b4afdd7
JK		 414 complete(&cmd->wait);
415 }
a4ed23f2 416 bio_put(bio);
a688b9d9 417 fcc->dispatch_list = NULL;
6b4afdd7
JK		 418 }
419
a688b9d9 420 wait_event_interruptible(*q,
721bd4d5 421 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 422 goto repeat;
423}
424
425int f2fs_issue_flush(struct f2fs_sb_info *sbi)
426{
a688b9d9 427 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 428 struct flush_cmd cmd;
6b4afdd7 429
24a9ee0f
JK		 430 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
431 test_opt(sbi, FLUSH_MERGE));
432
0f7b2abd
JK		 433 if (test_opt(sbi, NOBARRIER))
434 return 0;
435
0a87f664 436 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
740432f8
JK		 437 struct bio *bio = f2fs_bio_alloc(0);
438 int ret;
439
0a87f664 440 atomic_inc(&fcc->submit_flush);
740432f8
JK		 441 bio->bi_bdev = sbi->sb->s_bdev;
442 ret = submit_bio_wait(WRITE_FLUSH, bio);
0a87f664 443 atomic_dec(&fcc->submit_flush);
740432f8
JK		 444 bio_put(bio);
445 return ret;
446 }
6b4afdd7 447
adf8d90b 448 init_completion(&cmd.wait);
6b4afdd7 449
0a87f664 450 atomic_inc(&fcc->submit_flush);
721bd4d5 451 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 452
a688b9d9
GZ		 453 if (!fcc->dispatch_list)
454 wake_up(&fcc->flush_wait_queue);
6b4afdd7 455
adf8d90b 456 wait_for_completion(&cmd.wait);
0a87f664 457 atomic_dec(&fcc->submit_flush);
adf8d90b
CY		 458
459 return cmd.ret;
6b4afdd7
JK		 460}
461
2163d198
GZ		 462int create_flush_cmd_control(struct f2fs_sb_info *sbi)
463{
464 dev_t dev = sbi->sb->s_bdev->bd_dev;
465 struct flush_cmd_control *fcc;
466 int err = 0;
467
468 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
469 if (!fcc)
470 return -ENOMEM;
0a87f664 471 atomic_set(&fcc->submit_flush, 0);
2163d198 472 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 473 init_llist_head(&fcc->issue_list);
6b2920a5 474 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 475 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
476 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
477 if (IS_ERR(fcc->f2fs_issue_flush)) {
478 err = PTR_ERR(fcc->f2fs_issue_flush);
479 kfree(fcc);
6b2920a5 480 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 481 return err;
482 }
2163d198
GZ		 483
484 return err;
485}
486
487void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
488{
6b2920a5 489 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 490
491 if (fcc && fcc->f2fs_issue_flush)
492 kthread_stop(fcc->f2fs_issue_flush);
493 kfree(fcc);
6b2920a5 494 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 495}
496
351df4b2
JK		 497static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
498 enum dirty_type dirty_type)
499{
500 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
501
502 /* need not be added */
503 if (IS_CURSEG(sbi, segno))
504 return;
505
506 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
507 dirty_i->nr_dirty[dirty_type]++;
508
509 if (dirty_type == DIRTY) {
510 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 511 enum dirty_type t = sentry->type;
b2f2c390 512
ec325b52
JK		 513 if (unlikely(t >= DIRTY)) {
514 f2fs_bug_on(sbi, 1);
515 return;
516 }
4625d6aa
CL		 517 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
518 dirty_i->nr_dirty[t]++;
351df4b2
JK		 519 }
520}
521
522static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
523 enum dirty_type dirty_type)
524{
525 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
526
527 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
528 dirty_i->nr_dirty[dirty_type]--;
529
530 if (dirty_type == DIRTY) {
4625d6aa
CL		 531 struct seg_entry *sentry = get_seg_entry(sbi, segno);
532 enum dirty_type t = sentry->type;
533
534 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
535 dirty_i->nr_dirty[t]--;
b2f2c390 536
5ec4e49f
JK		 537 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
538 clear_bit(GET_SECNO(sbi, segno),
539 dirty_i->victim_secmap);
351df4b2
JK		 540 }
541}
542
0a8165d7 543/*
351df4b2
JK		 544 * Should not occur error such as -ENOMEM.
545 * Adding dirty entry into seglist is not critical operation.
546 * If a given segment is one of current working segments, it won't be added.
547 */
8d8451af 548static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 549{
550 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
551 unsigned short valid_blocks;
552
553 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
554 return;
555
556 mutex_lock(&dirty_i->seglist_lock);
557
558 valid_blocks = get_valid_blocks(sbi, segno, 0);
559
560 if (valid_blocks == 0) {
561 __locate_dirty_segment(sbi, segno, PRE);
562 __remove_dirty_segment(sbi, segno, DIRTY);
563 } else if (valid_blocks < sbi->blocks_per_seg) {
564 __locate_dirty_segment(sbi, segno, DIRTY);
565 } else {
566 /* Recovery routine with SSR needs this */
567 __remove_dirty_segment(sbi, segno, DIRTY);
568 }
569
570 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 571}
572
1e87a78d 573static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 574 block_t blkstart, block_t blklen)
575{
55cf9cb6
CY		 576 sector_t start = SECTOR_FROM_BLOCK(blkstart);
577 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 578 struct seg_entry *se;
579 unsigned int offset;
580 block_t i;
581
582 for (i = blkstart; i < blkstart + blklen; i++) {
583 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
584 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
585
586 if (!f2fs_test_and_set_bit(offset, se->discard_map))
587 sbi->discard_blks--;
588 }
1661d07c 589 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 590 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
591}
592
e90c2d28 593bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 594{
60b286c4 595 int err = -EOPNOTSUPP;
40a02be1
JK		 596
597 if (test_opt(sbi, DISCARD)) {
598 struct seg_entry *se = get_seg_entry(sbi,
599 GET_SEGNO(sbi, blkaddr));
600 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
601
602 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 603 return false;
40a02be1
JK		 604
605 err = f2fs_issue_discard(sbi, blkaddr, 1);
606 }
607
e90c2d28 608 if (err) {
381722d2 609 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 610 return true;
611 }
612 return false;
37208879
JK		 613}
614
adf4983b 615static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 616 struct cp_control *cpc, struct seg_entry *se,
617 unsigned int start, unsigned int end)
b2955550
JK		 618{
619 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 620 struct discard_entry *new, *last;
621
622 if (!list_empty(head)) {
623 last = list_last_entry(head, struct discard_entry, list);
624 if (START_BLOCK(sbi, cpc->trim_start) + start ==
625 last->blkaddr + last->len) {
626 last->len += end - start;
627 goto done;
628 }
629 }
630
631 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
632 INIT_LIST_HEAD(&new->list);
633 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
634 new->len = end - start;
635 list_add_tail(&new->list, head);
636done:
637 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 638}
639
640static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
641{
b2955550
JK		 642 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
643 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 644 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 645 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
646 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 647 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 648 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 649 unsigned int start = 0, end = -1;
4b2fecc8 650 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 651 int i;
652
a66cdd98 653 if (se->valid_blocks == max_blocks)
b2955550
JK		 654 return;
655
a66cdd98
JK		 656 if (!force) {
657 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 658 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
659 return;
4b2fecc8
JK		 660 }
661
b2955550
JK		 662 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
663 for (i = 0; i < entries; i++)
a66cdd98 664 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 665 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 666
4b2fecc8 667 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 668 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
669 if (start >= max_blocks)
670 break;
671
672 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 673 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 674 }
675}
676
4b2fecc8
JK		 677void release_discard_addrs(struct f2fs_sb_info *sbi)
678{
679 struct list_head *head = &(SM_I(sbi)->discard_list);
680 struct discard_entry *entry, *this;
681
682 /* drop caches */
683 list_for_each_entry_safe(entry, this, head, list) {
684 list_del(&entry->list);
685 kmem_cache_free(discard_entry_slab, entry);
686 }
687}
688
0a8165d7 689/*
351df4b2
JK		 690 * Should call clear_prefree_segments after checkpoint is done.
691 */
692static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
693{
694 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 695 unsigned int segno;
351df4b2
JK		 696
697 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 698 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 699 __set_test_and_free(sbi, segno);
351df4b2
JK		 700 mutex_unlock(&dirty_i->seglist_lock);
701}
702
836b5a63 703void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 704{
b2955550 705 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 706 struct discard_entry *entry, *this;
351df4b2 707 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 708 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 709 unsigned int start = 0, end = -1;
351df4b2
JK		 710
711 mutex_lock(&dirty_i->seglist_lock);
29e59c14 712
351df4b2 713 while (1) {
29e59c14 714 int i;
7cd8558b
JK		 715 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
716 if (start >= MAIN_SEGS(sbi))
351df4b2 717 break;
7cd8558b
JK		 718 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
719 start + 1);
29e59c14
CL		 720
721 for (i = start; i < end; i++)
722 clear_bit(i, prefree_map);
723
724 dirty_i->nr_dirty[PRE] -= end - start;
725
726 if (!test_opt(sbi, DISCARD))
727 continue;
351df4b2 728
37208879
JK		 729 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
730 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 731 }
732 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 733
734 /* send small discards */
2d7b822a 735 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 736 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
737 goto skip;
37208879 738 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 739 cpc->trimmed += entry->len;
836b5a63 740skip:
b2955550
JK		 741 list_del(&entry->list);
742 SM_I(sbi)->nr_discards -= entry->len;
743 kmem_cache_free(discard_entry_slab, entry);
744 }
351df4b2
JK		 745}
746
184a5cd2 747static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 748{
749 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 750
751 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 752 sit_i->dirty_sentries++;
184a5cd2
CY		 753 return false;
754 }
755
756 return true;
351df4b2
JK		 757}
758
759static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
760 unsigned int segno, int modified)
761{
762 struct seg_entry *se = get_seg_entry(sbi, segno);
763 se->type = type;
764 if (modified)
765 __mark_sit_entry_dirty(sbi, segno);
766}
767
768static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
769{
770 struct seg_entry *se;
771 unsigned int segno, offset;
772 long int new_vblocks;
773
774 segno = GET_SEGNO(sbi, blkaddr);
775
776 se = get_seg_entry(sbi, segno);
777 new_vblocks = se->valid_blocks + del;
491c0854 778 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 779
9850cf4a 780 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 781 (new_vblocks > sbi->blocks_per_seg)));
782
783 se->valid_blocks = new_vblocks;
784 se->mtime = get_mtime(sbi);
785 SIT_I(sbi)->max_mtime = se->mtime;
786
787 /* Update valid block bitmap */
788 if (del > 0) {
52aca074 789 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 790 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 791 if (!f2fs_test_and_set_bit(offset, se->discard_map))
792 sbi->discard_blks--;
351df4b2 793 } else {
52aca074 794 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 795 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 796 if (f2fs_test_and_clear_bit(offset, se->discard_map))
797 sbi->discard_blks++;
351df4b2
JK		 798 }
799 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
800 se->ckpt_valid_blocks += del;
801
802 __mark_sit_entry_dirty(sbi, segno);
803
804 /* update total number of valid blocks to be written in ckpt area */
805 SIT_I(sbi)->written_valid_blocks += del;
806
807 if (sbi->segs_per_sec > 1)
808 get_sec_entry(sbi, segno)->valid_blocks += del;
809}
810
5e443818 811void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 812{
5e443818
JK		 813 update_sit_entry(sbi, new, 1);
814 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
815 update_sit_entry(sbi, old, -1);
816
817 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
818 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 819}
820
821void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
822{
823 unsigned int segno = GET_SEGNO(sbi, addr);
824 struct sit_info *sit_i = SIT_I(sbi);
825
9850cf4a 826 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 827 if (addr == NEW_ADDR)
828 return;
829
830 /* add it into sit main buffer */
831 mutex_lock(&sit_i->sentry_lock);
832
833 update_sit_entry(sbi, addr, -1);
834
835 /* add it into dirty seglist */
836 locate_dirty_segment(sbi, segno);
837
838 mutex_unlock(&sit_i->sentry_lock);
839}
840
6e2c64ad
JK		 841bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
842{
843 struct sit_info *sit_i = SIT_I(sbi);
844 unsigned int segno, offset;
845 struct seg_entry *se;
846 bool is_cp = false;
847
848 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
849 return true;
850
851 mutex_lock(&sit_i->sentry_lock);
852
853 segno = GET_SEGNO(sbi, blkaddr);
854 se = get_seg_entry(sbi, segno);
855 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
856
857 if (f2fs_test_bit(offset, se->ckpt_valid_map))
858 is_cp = true;
859
860 mutex_unlock(&sit_i->sentry_lock);
861
862 return is_cp;
863}
864
0a8165d7 865/*
351df4b2
JK		 866 * This function should be resided under the curseg_mutex lock
867 */
868static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 869 struct f2fs_summary *sum)
351df4b2
JK		 870{
871 struct curseg_info *curseg = CURSEG_I(sbi, type);
872 void *addr = curseg->sum_blk;
e79efe3b 873 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 874 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 875}
876
0a8165d7 877/*
351df4b2
JK		 878 * Calculate the number of current summary pages for writing
879 */
3fa06d7b 880int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 881{
351df4b2 882 int valid_sum_count = 0;
9a47938b 883 int i, sum_in_page;
351df4b2
JK		 884
885 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
886 if (sbi->ckpt->alloc_type[i] == SSR)
887 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 888 else {
889 if (for_ra)
890 valid_sum_count += le16_to_cpu(
891 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
892 else
893 valid_sum_count += curseg_blkoff(sbi, i);
894 }
351df4b2
JK		 895 }
896
09cbfeaf 897 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 898 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
899 if (valid_sum_count <= sum_in_page)
351df4b2 900 return 1;
9a47938b 901 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 902 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 903 return 2;
904 return 3;
905}
906
0a8165d7 907/*
351df4b2
JK		 908 * Caller should put this summary page
909 */
910struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
911{
912 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
913}
914
381722d2 915void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 916{
917 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 918 void *dst = page_address(page);
919
920 if (src)
09cbfeaf 921 memcpy(dst, src, PAGE_SIZE);
381722d2 922 else
09cbfeaf 923 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 924 set_page_dirty(page);
925 f2fs_put_page(page, 1);
926}
927
381722d2
CY		 928static void write_sum_page(struct f2fs_sb_info *sbi,
929 struct f2fs_summary_block *sum_blk, block_t blk_addr)
930{
931 update_meta_page(sbi, (void *)sum_blk, blk_addr);
932}
933
b7ad7512
CY		 934static void write_current_sum_page(struct f2fs_sb_info *sbi,
935 int type, block_t blk_addr)
936{
937 struct curseg_info *curseg = CURSEG_I(sbi, type);
938 struct page *page = grab_meta_page(sbi, blk_addr);
939 struct f2fs_summary_block *src = curseg->sum_blk;
940 struct f2fs_summary_block *dst;
941
942 dst = (struct f2fs_summary_block *)page_address(page);
943
944 mutex_lock(&curseg->curseg_mutex);
945
946 down_read(&curseg->journal_rwsem);
947 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
948 up_read(&curseg->journal_rwsem);
949
950 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
951 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
952
953 mutex_unlock(&curseg->curseg_mutex);
954
955 set_page_dirty(page);
956 f2fs_put_page(page, 1);
957}
958
60374688
JK		 959static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
960{
961 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 962 unsigned int segno = curseg->segno + 1;
60374688
JK		 963 struct free_segmap_info *free_i = FREE_I(sbi);
964
7cd8558b 965 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 966 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 967 return 0;
968}
969
0a8165d7 970/*
351df4b2
JK		 971 * Find a new segment from the free segments bitmap to right order
972 * This function should be returned with success, otherwise BUG
973 */
974static void get_new_segment(struct f2fs_sb_info *sbi,
975 unsigned int *newseg, bool new_sec, int dir)
976{
977 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 978 unsigned int segno, secno, zoneno;
7cd8558b 979 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 980 unsigned int hint = *newseg / sbi->segs_per_sec;
981 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
982 unsigned int left_start = hint;
983 bool init = true;
984 int go_left = 0;
985 int i;
986
1a118ccf 987 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 988
989 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
990 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 991 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
992 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 993 goto got_it;
994 }
995find_other_zone:
7cd8558b
JK		 996 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
997 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 998 if (dir == ALLOC_RIGHT) {
999 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1000 MAIN_SECS(sbi), 0);
1001 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1002 } else {
1003 go_left = 1;
1004 left_start = hint - 1;
1005 }
1006 }
1007 if (go_left == 0)
1008 goto skip_left;
1009
1010 while (test_bit(left_start, free_i->free_secmap)) {
1011 if (left_start > 0) {
1012 left_start--;
1013 continue;
1014 }
1015 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1016 MAIN_SECS(sbi), 0);
1017 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1018 break;
1019 }
1020 secno = left_start;
1021skip_left:
1022 hint = secno;
1023 segno = secno * sbi->segs_per_sec;
1024 zoneno = secno / sbi->secs_per_zone;
1025
1026 /* give up on finding another zone */
1027 if (!init)
1028 goto got_it;
1029 if (sbi->secs_per_zone == 1)
1030 goto got_it;
1031 if (zoneno == old_zoneno)
1032 goto got_it;
1033 if (dir == ALLOC_LEFT) {
1034 if (!go_left && zoneno + 1 >= total_zones)
1035 goto got_it;
1036 if (go_left && zoneno == 0)
1037 goto got_it;
1038 }
1039 for (i = 0; i < NR_CURSEG_TYPE; i++)
1040 if (CURSEG_I(sbi, i)->zone == zoneno)
1041 break;
1042
1043 if (i < NR_CURSEG_TYPE) {
1044 /* zone is in user, try another */
1045 if (go_left)
1046 hint = zoneno * sbi->secs_per_zone - 1;
1047 else if (zoneno + 1 >= total_zones)
1048 hint = 0;
1049 else
1050 hint = (zoneno + 1) * sbi->secs_per_zone;
1051 init = false;
1052 goto find_other_zone;
1053 }
1054got_it:
1055 /* set it as dirty segment in free segmap */
9850cf4a 1056 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1057 __set_inuse(sbi, segno);
1058 *newseg = segno;
1a118ccf 1059 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1060}
1061
1062static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1063{
1064 struct curseg_info *curseg = CURSEG_I(sbi, type);
1065 struct summary_footer *sum_footer;
1066
1067 curseg->segno = curseg->next_segno;
1068 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1069 curseg->next_blkoff = 0;
1070 curseg->next_segno = NULL_SEGNO;
1071
1072 sum_footer = &(curseg->sum_blk->footer);
1073 memset(sum_footer, 0, sizeof(struct summary_footer));
1074 if (IS_DATASEG(type))
1075 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1076 if (IS_NODESEG(type))
1077 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1078 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1079}
1080
0a8165d7 1081/*
351df4b2
JK		 1082 * Allocate a current working segment.
1083 * This function always allocates a free segment in LFS manner.
1084 */
1085static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1086{
1087 struct curseg_info *curseg = CURSEG_I(sbi, type);
1088 unsigned int segno = curseg->segno;
1089 int dir = ALLOC_LEFT;
1090
1091 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1092 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1093 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1094 dir = ALLOC_RIGHT;
1095
1096 if (test_opt(sbi, NOHEAP))
1097 dir = ALLOC_RIGHT;
1098
1099 get_new_segment(sbi, &segno, new_sec, dir);
1100 curseg->next_segno = segno;
1101 reset_curseg(sbi, type, 1);
1102 curseg->alloc_type = LFS;
1103}
1104
1105static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1106 struct curseg_info *seg, block_t start)
1107{
1108 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1109 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1110 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1111 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1112 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1113 int i, pos;
1114
1115 for (i = 0; i < entries; i++)
1116 target_map[i] = ckpt_map[i] | cur_map[i];
1117
1118 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1119
1120 seg->next_blkoff = pos;
351df4b2
JK		 1121}
1122
0a8165d7 1123/*
351df4b2
JK		 1124 * If a segment is written by LFS manner, next block offset is just obtained
1125 * by increasing the current block offset. However, if a segment is written by
1126 * SSR manner, next block offset obtained by calling __next_free_blkoff
1127 */
1128static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1129 struct curseg_info *seg)
1130{
1131 if (seg->alloc_type == SSR)
1132 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1133 else
1134 seg->next_blkoff++;
1135}
1136
0a8165d7 1137/*
e1c42045 1138 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1139 * manner, so it should recover the existing segment information of valid blocks
1140 */
1141static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1142{
1143 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1144 struct curseg_info *curseg = CURSEG_I(sbi, type);
1145 unsigned int new_segno = curseg->next_segno;
1146 struct f2fs_summary_block *sum_node;
1147 struct page *sum_page;
1148
1149 write_sum_page(sbi, curseg->sum_blk,
1150 GET_SUM_BLOCK(sbi, curseg->segno));
1151 __set_test_and_inuse(sbi, new_segno);
1152
1153 mutex_lock(&dirty_i->seglist_lock);
1154 __remove_dirty_segment(sbi, new_segno, PRE);
1155 __remove_dirty_segment(sbi, new_segno, DIRTY);
1156 mutex_unlock(&dirty_i->seglist_lock);
1157
1158 reset_curseg(sbi, type, 1);
1159 curseg->alloc_type = SSR;
1160 __next_free_blkoff(sbi, curseg, 0);
1161
1162 if (reuse) {
1163 sum_page = get_sum_page(sbi, new_segno);
1164 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1165 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1166 f2fs_put_page(sum_page, 1);
1167 }
1168}
1169
43727527
JK		 1170static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1171{
1172 struct curseg_info *curseg = CURSEG_I(sbi, type);
1173 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1174
1175 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1176 return v_ops->get_victim(sbi,
1177 &(curseg)->next_segno, BG_GC, type, SSR);
1178
1179 /* For data segments, let's do SSR more intensively */
1180 for (; type >= CURSEG_HOT_DATA; type--)
1181 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1182 BG_GC, type, SSR))
1183 return 1;
1184 return 0;
1185}
1186
351df4b2
JK		 1187/*
1188 * flush out current segment and replace it with new segment
1189 * This function should be returned with success, otherwise BUG
1190 */
1191static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1192 int type, bool force)
1193{
1194 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1195
7b405275 1196 if (force)
351df4b2 1197 new_curseg(sbi, type, true);
7b405275 1198 else if (type == CURSEG_WARM_NODE)
351df4b2 1199 new_curseg(sbi, type, false);
60374688
JK		 1200 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1201 new_curseg(sbi, type, false);
351df4b2
JK		 1202 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1203 change_curseg(sbi, type, true);
1204 else
1205 new_curseg(sbi, type, false);
dcdfff65
JK		 1206
1207 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1208}
1209
38aa0889
JK		 1210static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1211{
1212 struct curseg_info *curseg = CURSEG_I(sbi, type);
1213 unsigned int old_segno;
1214
1215 old_segno = curseg->segno;
1216 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1217 locate_dirty_segment(sbi, old_segno);
1218}
1219
351df4b2
JK		 1220void allocate_new_segments(struct f2fs_sb_info *sbi)
1221{
351df4b2
JK		 1222 int i;
1223
38aa0889
JK		 1224 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1225 __allocate_new_segments(sbi, i);
351df4b2
JK		 1226}
1227
1228static const struct segment_allocation default_salloc_ops = {
1229 .allocate_segment = allocate_segment_by_default,
1230};
1231
4b2fecc8
JK		 1232int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1233{
f7ef9b83
JK		 1234 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1235 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1236 unsigned int start_segno, end_segno;
1237 struct cp_control cpc;
c34f42e2 1238 int err = 0;
4b2fecc8 1239
836b5a63 1240 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1241 return -EINVAL;
1242
9bd27ae4 1243 cpc.trimmed = 0;
7cd8558b 1244 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1245 goto out;
1246
1247 /* start/end segment number in main_area */
7cd8558b
JK		 1248 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1249 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1250 GET_SEGNO(sbi, end);
4b2fecc8 1251 cpc.reason = CP_DISCARD;
836b5a63 1252 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1253
1254 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1255 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1256 cpc.trim_start = start_segno;
a66cdd98
JK		 1257
1258 if (sbi->discard_blks == 0)
1259 break;
1260 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1261 cpc.trim_end = end_segno;
1262 else
1263 cpc.trim_end = min_t(unsigned int,
1264 rounddown(start_segno +
bba681cb
JK		 1265 BATCHED_TRIM_SEGMENTS(sbi),
1266 sbi->segs_per_sec) - 1, end_segno);
1267
1268 mutex_lock(&sbi->gc_mutex);
c34f42e2 1269 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1270 mutex_unlock(&sbi->gc_mutex);
1271 }
4b2fecc8 1272out:
f7ef9b83 1273 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1274 return err;
4b2fecc8
JK		 1275}
1276
351df4b2
JK		 1277static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1278{
1279 struct curseg_info *curseg = CURSEG_I(sbi, type);
1280 if (curseg->next_blkoff < sbi->blocks_per_seg)
1281 return true;
1282 return false;
1283}
1284
1285static int __get_segment_type_2(struct page *page, enum page_type p_type)
1286{
1287 if (p_type == DATA)
1288 return CURSEG_HOT_DATA;
1289 else
1290 return CURSEG_HOT_NODE;
1291}
1292
1293static int __get_segment_type_4(struct page *page, enum page_type p_type)
1294{
1295 if (p_type == DATA) {
1296 struct inode *inode = page->mapping->host;
1297
1298 if (S_ISDIR(inode->i_mode))
1299 return CURSEG_HOT_DATA;
1300 else
1301 return CURSEG_COLD_DATA;
1302 } else {
a344b9fd
JK		 1303 if (IS_DNODE(page) && is_cold_node(page))
1304 return CURSEG_WARM_NODE;
351df4b2
JK		 1305 else
1306 return CURSEG_COLD_NODE;
1307 }
1308}
1309
1310static int __get_segment_type_6(struct page *page, enum page_type p_type)
1311{
1312 if (p_type == DATA) {
1313 struct inode *inode = page->mapping->host;
1314
1315 if (S_ISDIR(inode->i_mode))
1316 return CURSEG_HOT_DATA;
354a3399 1317 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1318 return CURSEG_COLD_DATA;
1319 else
1320 return CURSEG_WARM_DATA;
1321 } else {
1322 if (IS_DNODE(page))
1323 return is_cold_node(page) ? CURSEG_WARM_NODE :
1324 CURSEG_HOT_NODE;
1325 else
1326 return CURSEG_COLD_NODE;
1327 }
1328}
1329
1330static int __get_segment_type(struct page *page, enum page_type p_type)
1331{
4081363f 1332 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1333 case 2:
1334 return __get_segment_type_2(page, p_type);
1335 case 4:
1336 return __get_segment_type_4(page, p_type);
351df4b2 1337 }
12a67146 1338 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1339 f2fs_bug_on(F2FS_P_SB(page),
1340 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1341 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1342}
1343
bfad7c2d
JK		 1344void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1345 block_t old_blkaddr, block_t *new_blkaddr,
1346 struct f2fs_summary *sum, int type)
351df4b2
JK		 1347{
1348 struct sit_info *sit_i = SIT_I(sbi);
1349 struct curseg_info *curseg;
38aa0889
JK		 1350 bool direct_io = (type == CURSEG_DIRECT_IO);
1351
1352 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1353
351df4b2
JK		 1354 curseg = CURSEG_I(sbi, type);
1355
1356 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1357 mutex_lock(&sit_i->sentry_lock);
351df4b2 1358
38aa0889 1359 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1360 if (direct_io && curseg->next_blkoff &&
1361 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1362 __allocate_new_segments(sbi, type);
1363
351df4b2 1364 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1365
1366 /*
1367 * __add_sum_entry should be resided under the curseg_mutex
1368 * because, this function updates a summary entry in the
1369 * current summary block.
1370 */
e79efe3b 1371 __add_sum_entry(sbi, type, sum);
351df4b2 1372
351df4b2 1373 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1374
1375 stat_inc_block_count(sbi, curseg);
351df4b2 1376
5e443818
JK		 1377 if (!__has_curseg_space(sbi, type))
1378 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1379 /*
1380 * SIT information should be updated before segment allocation,
1381 * since SSR needs latest valid block information.
1382 */
1383 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1384
351df4b2
JK		 1385 mutex_unlock(&sit_i->sentry_lock);
1386
bfad7c2d 1387 if (page && IS_NODESEG(type))
351df4b2
JK		 1388 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1389
bfad7c2d
JK		 1390 mutex_unlock(&curseg->curseg_mutex);
1391}
1392
05ca3632 1393static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1394{
05ca3632 1395 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1396
7a9d7548
CY		 1397 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1398 &fio->new_blkaddr, sum, type);
bfad7c2d 1399
351df4b2 1400 /* writeout dirty page into bdev */
05ca3632 1401 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1402}
1403
577e3495 1404void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1405{
458e6197 1406 struct f2fs_io_info fio = {
05ca3632 1407 .sbi = sbi,
458e6197 1408 .type = META,
cf04e8eb 1409 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1410 .old_blkaddr = page->index,
1411 .new_blkaddr = page->index,
05ca3632 1412 .page = page,
4375a336 1413 .encrypted_page = NULL,
458e6197
JK		 1414 };
1415
2b947003
CY		 1416 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1417 fio.rw &= ~REQ_META;
1418
351df4b2 1419 set_page_writeback(page);
05ca3632 1420 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1421}
1422
05ca3632 1423void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1424{
1425 struct f2fs_summary sum;
05ca3632 1426
351df4b2 1427 set_summary(&sum, nid, 0, 0);
05ca3632 1428 do_write_page(&sum, fio);
351df4b2
JK		 1429}
1430
05ca3632 1431void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1432{
05ca3632 1433 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1434 struct f2fs_summary sum;
1435 struct node_info ni;
1436
9850cf4a 1437 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1438 get_node_info(sbi, dn->nid, &ni);
1439 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1440 do_write_page(&sum, fio);
f28b3434 1441 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1442}
1443
05ca3632 1444void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1445{
7a9d7548 1446 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1447 stat_inc_inplace_blocks(fio->sbi);
1448 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1449}
1450
4356e48e 1451void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1452 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1453 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1454{
1455 struct sit_info *sit_i = SIT_I(sbi);
1456 struct curseg_info *curseg;
1457 unsigned int segno, old_cursegno;
1458 struct seg_entry *se;
1459 int type;
19f106bc 1460 unsigned short old_blkoff;
351df4b2
JK		 1461
1462 segno = GET_SEGNO(sbi, new_blkaddr);
1463 se = get_seg_entry(sbi, segno);
1464 type = se->type;
1465
19f106bc
CY		 1466 if (!recover_curseg) {
1467 /* for recovery flow */
1468 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1469 if (old_blkaddr == NULL_ADDR)
1470 type = CURSEG_COLD_DATA;
1471 else
1472 type = CURSEG_WARM_DATA;
1473 }
1474 } else {
1475 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1476 type = CURSEG_WARM_DATA;
1477 }
19f106bc 1478
351df4b2
JK		 1479 curseg = CURSEG_I(sbi, type);
1480
1481 mutex_lock(&curseg->curseg_mutex);
1482 mutex_lock(&sit_i->sentry_lock);
1483
1484 old_cursegno = curseg->segno;
19f106bc 1485 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1486
1487 /* change the current segment */
1488 if (segno != curseg->segno) {
1489 curseg->next_segno = segno;
1490 change_curseg(sbi, type, true);
1491 }
1492
491c0854 1493 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1494 __add_sum_entry(sbi, type, sum);
351df4b2 1495
28bc106b 1496 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1497 update_sit_entry(sbi, new_blkaddr, 1);
1498 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1499 update_sit_entry(sbi, old_blkaddr, -1);
1500
1501 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1502 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1503
351df4b2 1504 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1505
19f106bc
CY		 1506 if (recover_curseg) {
1507 if (old_cursegno != curseg->segno) {
1508 curseg->next_segno = old_cursegno;
1509 change_curseg(sbi, type, true);
1510 }
1511 curseg->next_blkoff = old_blkoff;
1512 }
1513
351df4b2
JK		 1514 mutex_unlock(&sit_i->sentry_lock);
1515 mutex_unlock(&curseg->curseg_mutex);
1516}
1517
528e3459
CY		 1518void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1519 block_t old_addr, block_t new_addr,
28bc106b
CY		 1520 unsigned char version, bool recover_curseg,
1521 bool recover_newaddr)
528e3459
CY		 1522{
1523 struct f2fs_summary sum;
1524
1525 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1526
28bc106b
CY		 1527 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1528 recover_curseg, recover_newaddr);
528e3459 1529
f28b3434 1530 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1531}
1532
93dfe2ac 1533void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1534 enum page_type type, bool ordered)
93dfe2ac 1535{
93dfe2ac 1536 if (PageWriteback(page)) {
4081363f
JK		 1537 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1538
0c3a5797 1539 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1540 if (ordered)
1541 wait_on_page_writeback(page);
1542 else
1543 wait_for_stable_page(page);
93dfe2ac
JK		 1544 }
1545}
1546
08b39fbd
CY		 1547void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1548 block_t blkaddr)
1549{
1550 struct page *cpage;
1551
1552 if (blkaddr == NEW_ADDR)
1553 return;
1554
1555 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1556
1557 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1558 if (cpage) {
fec1d657 1559 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1560 f2fs_put_page(cpage, 1);
1561 }
1562}
1563
351df4b2
JK		 1564static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1565{
1566 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1567 struct curseg_info *seg_i;
1568 unsigned char *kaddr;
1569 struct page *page;
1570 block_t start;
1571 int i, j, offset;
1572
1573 start = start_sum_block(sbi);
1574
1575 page = get_meta_page(sbi, start++);
1576 kaddr = (unsigned char *)page_address(page);
1577
1578 /* Step 1: restore nat cache */
1579 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1580 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1581
1582 /* Step 2: restore sit cache */
1583 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1584 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1585 offset = 2 * SUM_JOURNAL_SIZE;
1586
1587 /* Step 3: restore summary entries */
1588 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1589 unsigned short blk_off;
1590 unsigned int segno;
1591
1592 seg_i = CURSEG_I(sbi, i);
1593 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1594 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1595 seg_i->next_segno = segno;
1596 reset_curseg(sbi, i, 0);
1597 seg_i->alloc_type = ckpt->alloc_type[i];
1598 seg_i->next_blkoff = blk_off;
1599
1600 if (seg_i->alloc_type == SSR)
1601 blk_off = sbi->blocks_per_seg;
1602
1603 for (j = 0; j < blk_off; j++) {
1604 struct f2fs_summary *s;
1605 s = (struct f2fs_summary *)(kaddr + offset);
1606 seg_i->sum_blk->entries[j] = *s;
1607 offset += SUMMARY_SIZE;
09cbfeaf 1608 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1609 SUM_FOOTER_SIZE)
1610 continue;
1611
1612 f2fs_put_page(page, 1);
1613 page = NULL;
1614
1615 page = get_meta_page(sbi, start++);
1616 kaddr = (unsigned char *)page_address(page);
1617 offset = 0;
1618 }
1619 }
1620 f2fs_put_page(page, 1);
1621 return 0;
1622}
1623
1624static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1625{
1626 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1627 struct f2fs_summary_block *sum;
1628 struct curseg_info *curseg;
1629 struct page *new;
1630 unsigned short blk_off;
1631 unsigned int segno = 0;
1632 block_t blk_addr = 0;
1633
1634 /* get segment number and block addr */
1635 if (IS_DATASEG(type)) {
1636 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1637 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1638 CURSEG_HOT_DATA]);
119ee914 1639 if (__exist_node_summaries(sbi))
351df4b2
JK		 1640 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1641 else
1642 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1643 } else {
1644 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1645 CURSEG_HOT_NODE]);
1646 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1647 CURSEG_HOT_NODE]);
119ee914 1648 if (__exist_node_summaries(sbi))
351df4b2
JK		 1649 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1650 type - CURSEG_HOT_NODE);
1651 else
1652 blk_addr = GET_SUM_BLOCK(sbi, segno);
1653 }
1654
1655 new = get_meta_page(sbi, blk_addr);
1656 sum = (struct f2fs_summary_block *)page_address(new);
1657
1658 if (IS_NODESEG(type)) {
119ee914 1659 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1660 struct f2fs_summary *ns = &sum->entries[0];
1661 int i;
1662 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1663 ns->version = 0;
1664 ns->ofs_in_node = 0;
1665 }
1666 } else {
d653788a
GZ		 1667 int err;
1668
1669 err = restore_node_summary(sbi, segno, sum);
1670 if (err) {
351df4b2 1671 f2fs_put_page(new, 1);
d653788a 1672 return err;
351df4b2
JK		 1673 }
1674 }
1675 }
1676
1677 /* set uncompleted segment to curseg */
1678 curseg = CURSEG_I(sbi, type);
1679 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1680
1681 /* update journal info */
1682 down_write(&curseg->journal_rwsem);
1683 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1684 up_write(&curseg->journal_rwsem);
1685
1686 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1687 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1688 curseg->next_segno = segno;
1689 reset_curseg(sbi, type, 0);
1690 curseg->alloc_type = ckpt->alloc_type[type];
1691 curseg->next_blkoff = blk_off;
1692 mutex_unlock(&curseg->curseg_mutex);
1693 f2fs_put_page(new, 1);
1694 return 0;
1695}
1696
1697static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1698{
1699 int type = CURSEG_HOT_DATA;
e4fc5fbf 1700 int err;
351df4b2 1701
25ca923b 1702 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1703 int npages = npages_for_summary_flush(sbi, true);
1704
1705 if (npages >= 2)
1706 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1707 META_CP, true);
3fa06d7b 1708
351df4b2
JK		 1709 /* restore for compacted data summary */
1710 if (read_compacted_summaries(sbi))
1711 return -EINVAL;
1712 type = CURSEG_HOT_NODE;
1713 }
1714
119ee914 1715 if (__exist_node_summaries(sbi))
3fa06d7b 1716 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1717 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1718
e4fc5fbf
CY		 1719 for (; type <= CURSEG_COLD_NODE; type++) {
1720 err = read_normal_summaries(sbi, type);
1721 if (err)
1722 return err;
1723 }
1724
351df4b2
JK		 1725 return 0;
1726}
1727
1728static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1729{
1730 struct page *page;
1731 unsigned char *kaddr;
1732 struct f2fs_summary *summary;
1733 struct curseg_info *seg_i;
1734 int written_size = 0;
1735 int i, j;
1736
1737 page = grab_meta_page(sbi, blkaddr++);
1738 kaddr = (unsigned char *)page_address(page);
1739
1740 /* Step 1: write nat cache */
1741 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1742 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1743 written_size += SUM_JOURNAL_SIZE;
1744
1745 /* Step 2: write sit cache */
1746 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1747 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1748 written_size += SUM_JOURNAL_SIZE;
1749
351df4b2
JK		 1750 /* Step 3: write summary entries */
1751 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1752 unsigned short blkoff;
1753 seg_i = CURSEG_I(sbi, i);
1754 if (sbi->ckpt->alloc_type[i] == SSR)
1755 blkoff = sbi->blocks_per_seg;
1756 else
1757 blkoff = curseg_blkoff(sbi, i);
1758
1759 for (j = 0; j < blkoff; j++) {
1760 if (!page) {
1761 page = grab_meta_page(sbi, blkaddr++);
1762 kaddr = (unsigned char *)page_address(page);
1763 written_size = 0;
1764 }
1765 summary = (struct f2fs_summary *)(kaddr + written_size);
1766 *summary = seg_i->sum_blk->entries[j];
1767 written_size += SUMMARY_SIZE;
351df4b2 1768
09cbfeaf 1769 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1770 SUM_FOOTER_SIZE)
1771 continue;
1772
e8d61a74 1773 set_page_dirty(page);
351df4b2
JK		 1774 f2fs_put_page(page, 1);
1775 page = NULL;
1776 }
1777 }
e8d61a74
CY		 1778 if (page) {
1779 set_page_dirty(page);
351df4b2 1780 f2fs_put_page(page, 1);
e8d61a74 1781 }
351df4b2
JK		 1782}
1783
1784static void write_normal_summaries(struct f2fs_sb_info *sbi,
1785 block_t blkaddr, int type)
1786{
1787 int i, end;
1788 if (IS_DATASEG(type))
1789 end = type + NR_CURSEG_DATA_TYPE;
1790 else
1791 end = type + NR_CURSEG_NODE_TYPE;
1792
b7ad7512
CY		 1793 for (i = type; i < end; i++)
1794 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1795}
1796
1797void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1798{
25ca923b 1799 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1800 write_compacted_summaries(sbi, start_blk);
1801 else
1802 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1803}
1804
1805void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1806{
119ee914 1807 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1808}
1809
dfc08a12 1810int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1811 unsigned int val, int alloc)
1812{
1813 int i;
1814
1815 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1816 for (i = 0; i < nats_in_cursum(journal); i++) {
1817 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1818 return i;
1819 }
dfc08a12
CY		 1820 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1821 return update_nats_in_cursum(journal, 1);
351df4b2 1822 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1823 for (i = 0; i < sits_in_cursum(journal); i++)
1824 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1825 return i;
dfc08a12
CY		 1826 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1827 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1828 }
1829 return -1;
1830}
1831
1832static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1833 unsigned int segno)
1834{
2cc22186 1835 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1836}
1837
1838static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1839 unsigned int start)
1840{
1841 struct sit_info *sit_i = SIT_I(sbi);
1842 struct page *src_page, *dst_page;
1843 pgoff_t src_off, dst_off;
1844 void *src_addr, *dst_addr;
1845
1846 src_off = current_sit_addr(sbi, start);
1847 dst_off = next_sit_addr(sbi, src_off);
1848
1849 /* get current sit block page without lock */
1850 src_page = get_meta_page(sbi, src_off);
1851 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1852 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1853
1854 src_addr = page_address(src_page);
1855 dst_addr = page_address(dst_page);
09cbfeaf 1856 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1857
1858 set_page_dirty(dst_page);
1859 f2fs_put_page(src_page, 1);
1860
1861 set_to_next_sit(sit_i, start);
1862
1863 return dst_page;
1864}
1865
184a5cd2
CY		 1866static struct sit_entry_set *grab_sit_entry_set(void)
1867{
1868 struct sit_entry_set *ses =
80c54505 1869 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1870
1871 ses->entry_cnt = 0;
1872 INIT_LIST_HEAD(&ses->set_list);
1873 return ses;
1874}
1875
1876static void release_sit_entry_set(struct sit_entry_set *ses)
1877{
1878 list_del(&ses->set_list);
1879 kmem_cache_free(sit_entry_set_slab, ses);
1880}
1881
1882static void adjust_sit_entry_set(struct sit_entry_set *ses,
1883 struct list_head *head)
1884{
1885 struct sit_entry_set *next = ses;
1886
1887 if (list_is_last(&ses->set_list, head))
1888 return;
1889
1890 list_for_each_entry_continue(next, head, set_list)
1891 if (ses->entry_cnt <= next->entry_cnt)
1892 break;
1893
1894 list_move_tail(&ses->set_list, &next->set_list);
1895}
1896
1897static void add_sit_entry(unsigned int segno, struct list_head *head)
1898{
1899 struct sit_entry_set *ses;
1900 unsigned int start_segno = START_SEGNO(segno);
1901
1902 list_for_each_entry(ses, head, set_list) {
1903 if (ses->start_segno == start_segno) {
1904 ses->entry_cnt++;
1905 adjust_sit_entry_set(ses, head);
1906 return;
1907 }
1908 }
1909
1910 ses = grab_sit_entry_set();
1911
1912 ses->start_segno = start_segno;
1913 ses->entry_cnt++;
1914 list_add(&ses->set_list, head);
1915}
1916
1917static void add_sits_in_set(struct f2fs_sb_info *sbi)
1918{
1919 struct f2fs_sm_info *sm_info = SM_I(sbi);
1920 struct list_head *set_list = &sm_info->sit_entry_set;
1921 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1922 unsigned int segno;
1923
7cd8558b 1924 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1925 add_sit_entry(segno, set_list);
1926}
1927
1928static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1929{
1930 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1931 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1932 int i;
1933
b7ad7512 1934 down_write(&curseg->journal_rwsem);
dfc08a12 1935 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1936 unsigned int segno;
1937 bool dirtied;
1938
dfc08a12 1939 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1940 dirtied = __mark_sit_entry_dirty(sbi, segno);
1941
1942 if (!dirtied)
1943 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1944 }
dfc08a12 1945 update_sits_in_cursum(journal, -i);
b7ad7512 1946 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1947}
1948
0a8165d7 1949/*
351df4b2
JK		 1950 * CP calls this function, which flushes SIT entries including sit_journal,
1951 * and moves prefree segs to free segs.
1952 */
4b2fecc8 1953void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1954{
1955 struct sit_info *sit_i = SIT_I(sbi);
1956 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1957 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1958 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 1959 struct sit_entry_set *ses, *tmp;
1960 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1961 bool to_journal = true;
4b2fecc8 1962 struct seg_entry *se;
351df4b2 1963
351df4b2
JK		 1964 mutex_lock(&sit_i->sentry_lock);
1965
2b11a74b
WL		 1966 if (!sit_i->dirty_sentries)
1967 goto out;
1968
351df4b2 1969 /*
184a5cd2
CY		 1970 * add and account sit entries of dirty bitmap in sit entry
1971 * set temporarily
351df4b2 1972 */
184a5cd2 1973 add_sits_in_set(sbi);
351df4b2 1974
184a5cd2
CY		 1975 /*
1976 * if there are no enough space in journal to store dirty sit
1977 * entries, remove all entries from journal and add and account
1978 * them in sit entry set.
1979 */
dfc08a12 1980 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 1981 remove_sits_in_journal(sbi);
b2955550 1982
184a5cd2
CY		 1983 /*
1984 * there are two steps to flush sit entries:
1985 * #1, flush sit entries to journal in current cold data summary block.
1986 * #2, flush sit entries to sit page.
1987 */
1988 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1989 struct page *page = NULL;
184a5cd2
CY		 1990 struct f2fs_sit_block *raw_sit = NULL;
1991 unsigned int start_segno = ses->start_segno;
1992 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1993 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1994 unsigned int segno = start_segno;
1995
1996 if (to_journal &&
dfc08a12 1997 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 1998 to_journal = false;
1999
b7ad7512
CY		 2000 if (to_journal) {
2001 down_write(&curseg->journal_rwsem);
2002 } else {
184a5cd2
CY		 2003 page = get_next_sit_page(sbi, start_segno);
2004 raw_sit = page_address(page);
351df4b2 2005 }
351df4b2 2006
184a5cd2
CY		 2007 /* flush dirty sit entries in region of current sit set */
2008 for_each_set_bit_from(segno, bitmap, end) {
2009 int offset, sit_offset;
4b2fecc8
JK		 2010
2011 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2012
2013 /* add discard candidates */
d7bc2484 2014 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2015 cpc->trim_start = segno;
2016 add_discard_addrs(sbi, cpc);
2017 }
184a5cd2
CY		 2018
2019 if (to_journal) {
dfc08a12 2020 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2021 SIT_JOURNAL, segno, 1);
2022 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2023 segno_in_journal(journal, offset) =
184a5cd2
CY		 2024 cpu_to_le32(segno);
2025 seg_info_to_raw_sit(se,
dfc08a12 2026 &sit_in_journal(journal, offset));
184a5cd2
CY		 2027 } else {
2028 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2029 seg_info_to_raw_sit(se,
2030 &raw_sit->entries[sit_offset]);
2031 }
351df4b2 2032
184a5cd2
CY		 2033 __clear_bit(segno, bitmap);
2034 sit_i->dirty_sentries--;
2035 ses->entry_cnt--;
351df4b2
JK		 2036 }
2037
b7ad7512
CY		 2038 if (to_journal)
2039 up_write(&curseg->journal_rwsem);
2040 else
184a5cd2
CY		 2041 f2fs_put_page(page, 1);
2042
2043 f2fs_bug_on(sbi, ses->entry_cnt);
2044 release_sit_entry_set(ses);
351df4b2 2045 }
184a5cd2
CY		 2046
2047 f2fs_bug_on(sbi, !list_empty(head));
2048 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2049out:
4b2fecc8
JK		 2050 if (cpc->reason == CP_DISCARD) {
2051 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2052 add_discard_addrs(sbi, cpc);
2053 }
351df4b2 2054 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2055
351df4b2
JK		 2056 set_prefree_as_free_segments(sbi);
2057}
2058
2059static int build_sit_info(struct f2fs_sb_info *sbi)
2060{
2061 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2062 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2063 struct sit_info *sit_i;
2064 unsigned int sit_segs, start;
2065 char *src_bitmap, *dst_bitmap;
2066 unsigned int bitmap_size;
2067
2068 /* allocate memory for SIT information */
2069 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2070 if (!sit_i)
2071 return -ENOMEM;
2072
2073 SM_I(sbi)->sit_info = sit_i;
2074
39307a8e
JK		 2075 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2076 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2077 if (!sit_i->sentries)
2078 return -ENOMEM;
2079
7cd8558b 2080 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2081 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2082 if (!sit_i->dirty_sentries_bitmap)
2083 return -ENOMEM;
2084
7cd8558b 2085 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2086 sit_i->sentries[start].cur_valid_map
2087 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2088 sit_i->sentries[start].ckpt_valid_map
2089 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2090 sit_i->sentries[start].discard_map
2091 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2092 if (!sit_i->sentries[start].cur_valid_map ||
2093 !sit_i->sentries[start].ckpt_valid_map ||
2094 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2095 return -ENOMEM;
2096 }
2097
60a3b782
JK		 2098 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2099 if (!sit_i->tmp_map)
2100 return -ENOMEM;
2101
351df4b2 2102 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2103 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2104 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2105 if (!sit_i->sec_entries)
2106 return -ENOMEM;
2107 }
2108
2109 /* get information related with SIT */
2110 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2111
2112 /* setup SIT bitmap from ckeckpoint pack */
2113 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2114 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2115
79b5793b 2116 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2117 if (!dst_bitmap)
2118 return -ENOMEM;
351df4b2
JK		 2119
2120 /* init SIT information */
2121 sit_i->s_ops = &default_salloc_ops;
2122
2123 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2124 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2125 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2126 sit_i->sit_bitmap = dst_bitmap;
2127 sit_i->bitmap_size = bitmap_size;
2128 sit_i->dirty_sentries = 0;
2129 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2130 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2131 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2132 mutex_init(&sit_i->sentry_lock);
2133 return 0;
2134}
2135
2136static int build_free_segmap(struct f2fs_sb_info *sbi)
2137{
351df4b2
JK		 2138 struct free_segmap_info *free_i;
2139 unsigned int bitmap_size, sec_bitmap_size;
2140
2141 /* allocate memory for free segmap information */
2142 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2143 if (!free_i)
2144 return -ENOMEM;
2145
2146 SM_I(sbi)->free_info = free_i;
2147
7cd8558b 2148 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2149 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2150 if (!free_i->free_segmap)
2151 return -ENOMEM;
2152
7cd8558b 2153 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2154 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2155 if (!free_i->free_secmap)
2156 return -ENOMEM;
2157
2158 /* set all segments as dirty temporarily */
2159 memset(free_i->free_segmap, 0xff, bitmap_size);
2160 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2161
2162 /* init free segmap information */
7cd8558b 2163 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2164 free_i->free_segments = 0;
2165 free_i->free_sections = 0;
1a118ccf 2166 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2167 return 0;
2168}
2169
2170static int build_curseg(struct f2fs_sb_info *sbi)
2171{
1042d60f 2172 struct curseg_info *array;
351df4b2
JK		 2173 int i;
2174
b434babf 2175 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2176 if (!array)
2177 return -ENOMEM;
2178
2179 SM_I(sbi)->curseg_array = array;
2180
2181 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2182 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2183 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2184 if (!array[i].sum_blk)
2185 return -ENOMEM;
b7ad7512
CY		 2186 init_rwsem(&array[i].journal_rwsem);
2187 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2188 GFP_KERNEL);
2189 if (!array[i].journal)
2190 return -ENOMEM;
351df4b2
JK		 2191 array[i].segno = NULL_SEGNO;
2192 array[i].next_blkoff = 0;
2193 }
2194 return restore_curseg_summaries(sbi);
2195}
2196
2197static void build_sit_entries(struct f2fs_sb_info *sbi)
2198{
2199 struct sit_info *sit_i = SIT_I(sbi);
2200 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2201 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2202 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2203 unsigned int i, start, end;
2204 unsigned int readed, start_blk = 0;
e9f5b8b8 2205 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2206
74de593a 2207 do {
26879fb1 2208 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2209
2210 start = start_blk * sit_i->sents_per_block;
2211 end = (start_blk + readed) * sit_i->sents_per_block;
2212
7cd8558b 2213 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2214 struct seg_entry *se = &sit_i->sentries[start];
2215 struct f2fs_sit_block *sit_blk;
2216 struct f2fs_sit_entry sit;
2217 struct page *page;
2218
b7ad7512 2219 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2220 for (i = 0; i < sits_in_cursum(journal); i++) {
2221 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2222 == start) {
dfc08a12 2223 sit = sit_in_journal(journal, i);
b7ad7512 2224 up_read(&curseg->journal_rwsem);
74de593a
CY		 2225 goto got_it;
2226 }
351df4b2 2227 }
b7ad7512 2228 up_read(&curseg->journal_rwsem);
74de593a
CY		 2229
2230 page = get_current_sit_page(sbi, start);
2231 sit_blk = (struct f2fs_sit_block *)page_address(page);
2232 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2233 f2fs_put_page(page, 1);
351df4b2 2234got_it:
74de593a
CY		 2235 check_block_count(sbi, start, &sit);
2236 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2237
2238 /* build discard map only one time */
2239 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2240 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2241
74de593a
CY		 2242 if (sbi->segs_per_sec > 1) {
2243 struct sec_entry *e = get_sec_entry(sbi, start);
2244 e->valid_blocks += se->valid_blocks;
2245 }
351df4b2 2246 }
74de593a
CY		 2247 start_blk += readed;
2248 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2249}
2250
2251static void init_free_segmap(struct f2fs_sb_info *sbi)
2252{
2253 unsigned int start;
2254 int type;
2255
7cd8558b 2256 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2257 struct seg_entry *sentry = get_seg_entry(sbi, start);
2258 if (!sentry->valid_blocks)
2259 __set_free(sbi, start);
2260 }
2261
2262 /* set use the current segments */
2263 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2264 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2265 __set_test_and_inuse(sbi, curseg_t->segno);
2266 }
2267}
2268
2269static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2270{
2271 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2272 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2273 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2274 unsigned short valid_blocks;
2275
8736fbf0 2276 while (1) {
351df4b2 2277 /* find dirty segment based on free segmap */
7cd8558b
JK		 2278 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2279 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2280 break;
2281 offset = segno + 1;
2282 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2283 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2284 continue;
ec325b52
JK		 2285 if (valid_blocks > sbi->blocks_per_seg) {
2286 f2fs_bug_on(sbi, 1);
2287 continue;
2288 }
351df4b2
JK		 2289 mutex_lock(&dirty_i->seglist_lock);
2290 __locate_dirty_segment(sbi, segno, DIRTY);
2291 mutex_unlock(&dirty_i->seglist_lock);
2292 }
2293}
2294
5ec4e49f 2295static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2296{
2297 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2298 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2299
39307a8e 2300 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2301 if (!dirty_i->victim_secmap)
351df4b2
JK		 2302 return -ENOMEM;
2303 return 0;
2304}
2305
2306static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2307{
2308 struct dirty_seglist_info *dirty_i;
2309 unsigned int bitmap_size, i;
2310
2311 /* allocate memory for dirty segments list information */
2312 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2313 if (!dirty_i)
2314 return -ENOMEM;
2315
2316 SM_I(sbi)->dirty_info = dirty_i;
2317 mutex_init(&dirty_i->seglist_lock);
2318
7cd8558b 2319 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2320
2321 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2322 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2323 if (!dirty_i->dirty_segmap[i])
2324 return -ENOMEM;
2325 }
2326
2327 init_dirty_segmap(sbi);
5ec4e49f 2328 return init_victim_secmap(sbi);
351df4b2
JK		 2329}
2330
0a8165d7 2331/*
351df4b2
JK		 2332 * Update min, max modified time for cost-benefit GC algorithm
2333 */
2334static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2335{
2336 struct sit_info *sit_i = SIT_I(sbi);
2337 unsigned int segno;
2338
2339 mutex_lock(&sit_i->sentry_lock);
2340
2341 sit_i->min_mtime = LLONG_MAX;
2342
7cd8558b 2343 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2344 unsigned int i;
2345 unsigned long long mtime = 0;
2346
2347 for (i = 0; i < sbi->segs_per_sec; i++)
2348 mtime += get_seg_entry(sbi, segno + i)->mtime;
2349
2350 mtime = div_u64(mtime, sbi->segs_per_sec);
2351
2352 if (sit_i->min_mtime > mtime)
2353 sit_i->min_mtime = mtime;
2354 }
2355 sit_i->max_mtime = get_mtime(sbi);
2356 mutex_unlock(&sit_i->sentry_lock);
2357}
2358
2359int build_segment_manager(struct f2fs_sb_info *sbi)
2360{
2361 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2362 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2363 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2364 int err;
2365
2366 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2367 if (!sm_info)
2368 return -ENOMEM;
2369
2370 /* init sm info */
2371 sbi->sm_info = sm_info;
351df4b2
JK		 2372 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2373 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2374 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2375 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2376 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2377 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2378 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2379 sm_info->rec_prefree_segments = sm_info->main_segments *
2380 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2381 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2382 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2383 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2384
7fd9e544
JK		 2385 INIT_LIST_HEAD(&sm_info->discard_list);
2386 sm_info->nr_discards = 0;
2387 sm_info->max_discards = 0;
2388
bba681cb
JK		 2389 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2390
184a5cd2
CY		 2391 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2392
b270ad6f 2393 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2394 err = create_flush_cmd_control(sbi);
2395 if (err)
a688b9d9 2396 return err;
6b4afdd7
JK		 2397 }
2398
351df4b2
JK		 2399 err = build_sit_info(sbi);
2400 if (err)
2401 return err;
2402 err = build_free_segmap(sbi);
2403 if (err)
2404 return err;
2405 err = build_curseg(sbi);
2406 if (err)
2407 return err;
2408
2409 /* reinit free segmap based on SIT */
2410 build_sit_entries(sbi);
2411
2412 init_free_segmap(sbi);
2413 err = build_dirty_segmap(sbi);
2414 if (err)
2415 return err;
2416
2417 init_min_max_mtime(sbi);
2418 return 0;
2419}
2420
2421static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2422 enum dirty_type dirty_type)
2423{
2424 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2425
2426 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2427 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2428 dirty_i->nr_dirty[dirty_type] = 0;
2429 mutex_unlock(&dirty_i->seglist_lock);
2430}
2431
5ec4e49f 2432static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2433{
2434 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2435 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2436}
2437
2438static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2439{
2440 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2441 int i;
2442
2443 if (!dirty_i)
2444 return;
2445
2446 /* discard pre-free/dirty segments list */
2447 for (i = 0; i < NR_DIRTY_TYPE; i++)
2448 discard_dirty_segmap(sbi, i);
2449
5ec4e49f 2450 destroy_victim_secmap(sbi);
351df4b2
JK		 2451 SM_I(sbi)->dirty_info = NULL;
2452 kfree(dirty_i);
2453}
2454
2455static void destroy_curseg(struct f2fs_sb_info *sbi)
2456{
2457 struct curseg_info *array = SM_I(sbi)->curseg_array;
2458 int i;
2459
2460 if (!array)
2461 return;
2462 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2463 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2464 kfree(array[i].sum_blk);
b7ad7512
CY		 2465 kfree(array[i].journal);
2466 }
351df4b2
JK		 2467 kfree(array);
2468}
2469
2470static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2471{
2472 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2473 if (!free_i)
2474 return;
2475 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2476 kvfree(free_i->free_segmap);
2477 kvfree(free_i->free_secmap);
351df4b2
JK		 2478 kfree(free_i);
2479}
2480
2481static void destroy_sit_info(struct f2fs_sb_info *sbi)
2482{
2483 struct sit_info *sit_i = SIT_I(sbi);
2484 unsigned int start;
2485
2486 if (!sit_i)
2487 return;
2488
2489 if (sit_i->sentries) {
7cd8558b 2490 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2491 kfree(sit_i->sentries[start].cur_valid_map);
2492 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2493 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2494 }
2495 }
60a3b782
JK		 2496 kfree(sit_i->tmp_map);
2497
39307a8e
JK		 2498 kvfree(sit_i->sentries);
2499 kvfree(sit_i->sec_entries);
2500 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2501
2502 SM_I(sbi)->sit_info = NULL;
2503 kfree(sit_i->sit_bitmap);
2504 kfree(sit_i);
2505}
2506
2507void destroy_segment_manager(struct f2fs_sb_info *sbi)
2508{
2509 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2510
3b03f724
CY		 2511 if (!sm_info)
2512 return;
2163d198 2513 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2514 destroy_dirty_segmap(sbi);
2515 destroy_curseg(sbi);
2516 destroy_free_segmap(sbi);
2517 destroy_sit_info(sbi);
2518 sbi->sm_info = NULL;
2519 kfree(sm_info);
2520}
7fd9e544
JK		 2521
2522int __init create_segment_manager_caches(void)
2523{
2524 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2525 sizeof(struct discard_entry));
7fd9e544 2526 if (!discard_entry_slab)
184a5cd2
CY		 2527 goto fail;
2528
2529 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2530 sizeof(struct sit_entry_set));
184a5cd2
CY		 2531 if (!sit_entry_set_slab)
2532 goto destory_discard_entry;
88b88a66
JK		 2533
2534 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2535 sizeof(struct inmem_pages));
2536 if (!inmem_entry_slab)
2537 goto destroy_sit_entry_set;
7fd9e544 2538 return 0;
184a5cd2 2539
88b88a66
JK		 2540destroy_sit_entry_set:
2541 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2542destory_discard_entry:
2543 kmem_cache_destroy(discard_entry_slab);
2544fail:
2545 return -ENOMEM;
7fd9e544
JK		 2546}
2547
2548void destroy_segment_manager_caches(void)
2549{
184a5cd2 2550 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2551 kmem_cache_destroy(discard_entry_slab);
88b88a66 2552 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2553}
Linux kernel - Deliverables
This page took 0.26985 seconds and 5 git commands to generate.