]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: call update_inode_page for orphan inodes
[mirror_ubuntu-artful-kernel.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;
e589c2c4
JK		 348
349 /* balance_fs_bg is able to be pending */
350 if (excess_cached_nats(sbi))
351 f2fs_balance_fs_bg(sbi);
352
351df4b2 353 /*
029cd28c
JK		 354 * We should do GC or end up with checkpoint, if there are so many dirty
355 * dir/node pages without enough free segments.
351df4b2 356 */
43727527 357 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 358 mutex_lock(&sbi->gc_mutex);
d530d4d8 359 f2fs_gc(sbi, false);
351df4b2
JK		 360 }
361}
362
4660f9c0
JK		 363void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
364{
1dcc336b 365 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 366 if (!available_free_memory(sbi, EXTENT_CACHE))
367 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 368
1b38dc8e
JK		 369 /* check the # of cached NAT entries */
370 if (!available_free_memory(sbi, NAT_ENTRIES))
371 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
372
31696580
CY		 373 if (!available_free_memory(sbi, FREE_NIDS))
374 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
375
1b38dc8e
JK		 376 /* checkpoint is the only way to shrink partial cached entries */
377 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 378 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 379 excess_prefree_segs(sbi) ||
380 excess_dirty_nats(sbi) ||
d0239e1b 381 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
19a5f5e2 382 if (test_opt(sbi, DATA_FLUSH))
36b35a0d 383 sync_dirty_inodes(sbi, FILE_INODE);
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;
36abef4e 710 unsigned int secno, start_segno;
351df4b2
JK		 711
712 mutex_lock(&dirty_i->seglist_lock);
29e59c14 713
351df4b2 714 while (1) {
29e59c14 715 int i;
7cd8558b
JK		 716 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
717 if (start >= MAIN_SEGS(sbi))
351df4b2 718 break;
7cd8558b
JK		 719 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
720 start + 1);
29e59c14
CL		 721
722 for (i = start; i < end; i++)
723 clear_bit(i, prefree_map);
724
725 dirty_i->nr_dirty[PRE] -= end - start;
726
727 if (!test_opt(sbi, DISCARD))
728 continue;
351df4b2 729
36abef4e
JK		 730 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
731 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 732 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 733 continue;
734 }
735next:
736 secno = GET_SECNO(sbi, start);
737 start_segno = secno * sbi->segs_per_sec;
738 if (!IS_CURSEC(sbi, secno) &&
739 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
740 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
741 sbi->segs_per_sec << sbi->log_blocks_per_seg);
742
743 start = start_segno + sbi->segs_per_sec;
744 if (start < end)
745 goto next;
351df4b2
JK		 746 }
747 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 748
749 /* send small discards */
2d7b822a 750 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 751 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
752 goto skip;
37208879 753 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 754 cpc->trimmed += entry->len;
836b5a63 755skip:
b2955550
JK		 756 list_del(&entry->list);
757 SM_I(sbi)->nr_discards -= entry->len;
758 kmem_cache_free(discard_entry_slab, entry);
759 }
351df4b2
JK		 760}
761
184a5cd2 762static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 763{
764 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 765
766 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 767 sit_i->dirty_sentries++;
184a5cd2
CY		 768 return false;
769 }
770
771 return true;
351df4b2
JK		 772}
773
774static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
775 unsigned int segno, int modified)
776{
777 struct seg_entry *se = get_seg_entry(sbi, segno);
778 se->type = type;
779 if (modified)
780 __mark_sit_entry_dirty(sbi, segno);
781}
782
783static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
784{
785 struct seg_entry *se;
786 unsigned int segno, offset;
787 long int new_vblocks;
788
789 segno = GET_SEGNO(sbi, blkaddr);
790
791 se = get_seg_entry(sbi, segno);
792 new_vblocks = se->valid_blocks + del;
491c0854 793 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 794
9850cf4a 795 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 796 (new_vblocks > sbi->blocks_per_seg)));
797
798 se->valid_blocks = new_vblocks;
799 se->mtime = get_mtime(sbi);
800 SIT_I(sbi)->max_mtime = se->mtime;
801
802 /* Update valid block bitmap */
803 if (del > 0) {
52aca074 804 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 805 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 806 if (!f2fs_test_and_set_bit(offset, se->discard_map))
807 sbi->discard_blks--;
351df4b2 808 } else {
52aca074 809 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 810 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 811 if (f2fs_test_and_clear_bit(offset, se->discard_map))
812 sbi->discard_blks++;
351df4b2
JK		 813 }
814 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
815 se->ckpt_valid_blocks += del;
816
817 __mark_sit_entry_dirty(sbi, segno);
818
819 /* update total number of valid blocks to be written in ckpt area */
820 SIT_I(sbi)->written_valid_blocks += del;
821
822 if (sbi->segs_per_sec > 1)
823 get_sec_entry(sbi, segno)->valid_blocks += del;
824}
825
5e443818 826void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 827{
5e443818
JK		 828 update_sit_entry(sbi, new, 1);
829 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
830 update_sit_entry(sbi, old, -1);
831
832 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
833 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 834}
835
836void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
837{
838 unsigned int segno = GET_SEGNO(sbi, addr);
839 struct sit_info *sit_i = SIT_I(sbi);
840
9850cf4a 841 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 842 if (addr == NEW_ADDR)
843 return;
844
845 /* add it into sit main buffer */
846 mutex_lock(&sit_i->sentry_lock);
847
848 update_sit_entry(sbi, addr, -1);
849
850 /* add it into dirty seglist */
851 locate_dirty_segment(sbi, segno);
852
853 mutex_unlock(&sit_i->sentry_lock);
854}
855
6e2c64ad
JK		 856bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
857{
858 struct sit_info *sit_i = SIT_I(sbi);
859 unsigned int segno, offset;
860 struct seg_entry *se;
861 bool is_cp = false;
862
863 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
864 return true;
865
866 mutex_lock(&sit_i->sentry_lock);
867
868 segno = GET_SEGNO(sbi, blkaddr);
869 se = get_seg_entry(sbi, segno);
870 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
871
872 if (f2fs_test_bit(offset, se->ckpt_valid_map))
873 is_cp = true;
874
875 mutex_unlock(&sit_i->sentry_lock);
876
877 return is_cp;
878}
879
0a8165d7 880/*
351df4b2
JK		 881 * This function should be resided under the curseg_mutex lock
882 */
883static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 884 struct f2fs_summary *sum)
351df4b2
JK		 885{
886 struct curseg_info *curseg = CURSEG_I(sbi, type);
887 void *addr = curseg->sum_blk;
e79efe3b 888 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 889 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 890}
891
0a8165d7 892/*
351df4b2
JK		 893 * Calculate the number of current summary pages for writing
894 */
3fa06d7b 895int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 896{
351df4b2 897 int valid_sum_count = 0;
9a47938b 898 int i, sum_in_page;
351df4b2
JK		 899
900 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
901 if (sbi->ckpt->alloc_type[i] == SSR)
902 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 903 else {
904 if (for_ra)
905 valid_sum_count += le16_to_cpu(
906 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
907 else
908 valid_sum_count += curseg_blkoff(sbi, i);
909 }
351df4b2
JK		 910 }
911
09cbfeaf 912 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 913 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
914 if (valid_sum_count <= sum_in_page)
351df4b2 915 return 1;
9a47938b 916 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 917 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 918 return 2;
919 return 3;
920}
921
0a8165d7 922/*
351df4b2
JK		 923 * Caller should put this summary page
924 */
925struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
926{
927 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
928}
929
381722d2 930void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 931{
932 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 933 void *dst = page_address(page);
934
935 if (src)
09cbfeaf 936 memcpy(dst, src, PAGE_SIZE);
381722d2 937 else
09cbfeaf 938 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 939 set_page_dirty(page);
940 f2fs_put_page(page, 1);
941}
942
381722d2
CY		 943static void write_sum_page(struct f2fs_sb_info *sbi,
944 struct f2fs_summary_block *sum_blk, block_t blk_addr)
945{
946 update_meta_page(sbi, (void *)sum_blk, blk_addr);
947}
948
b7ad7512
CY		 949static void write_current_sum_page(struct f2fs_sb_info *sbi,
950 int type, block_t blk_addr)
951{
952 struct curseg_info *curseg = CURSEG_I(sbi, type);
953 struct page *page = grab_meta_page(sbi, blk_addr);
954 struct f2fs_summary_block *src = curseg->sum_blk;
955 struct f2fs_summary_block *dst;
956
957 dst = (struct f2fs_summary_block *)page_address(page);
958
959 mutex_lock(&curseg->curseg_mutex);
960
961 down_read(&curseg->journal_rwsem);
962 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
963 up_read(&curseg->journal_rwsem);
964
965 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
966 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
967
968 mutex_unlock(&curseg->curseg_mutex);
969
970 set_page_dirty(page);
971 f2fs_put_page(page, 1);
972}
973
60374688
JK		 974static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
975{
976 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 977 unsigned int segno = curseg->segno + 1;
60374688
JK		 978 struct free_segmap_info *free_i = FREE_I(sbi);
979
7cd8558b 980 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 981 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 982 return 0;
983}
984
0a8165d7 985/*
351df4b2
JK		 986 * Find a new segment from the free segments bitmap to right order
987 * This function should be returned with success, otherwise BUG
988 */
989static void get_new_segment(struct f2fs_sb_info *sbi,
990 unsigned int *newseg, bool new_sec, int dir)
991{
992 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 993 unsigned int segno, secno, zoneno;
7cd8558b 994 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 995 unsigned int hint = *newseg / sbi->segs_per_sec;
996 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
997 unsigned int left_start = hint;
998 bool init = true;
999 int go_left = 0;
1000 int i;
1001
1a118ccf 1002 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1003
1004 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1005 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1006 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1007 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1008 goto got_it;
1009 }
1010find_other_zone:
7cd8558b
JK		 1011 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1012 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1013 if (dir == ALLOC_RIGHT) {
1014 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1015 MAIN_SECS(sbi), 0);
1016 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1017 } else {
1018 go_left = 1;
1019 left_start = hint - 1;
1020 }
1021 }
1022 if (go_left == 0)
1023 goto skip_left;
1024
1025 while (test_bit(left_start, free_i->free_secmap)) {
1026 if (left_start > 0) {
1027 left_start--;
1028 continue;
1029 }
1030 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1031 MAIN_SECS(sbi), 0);
1032 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1033 break;
1034 }
1035 secno = left_start;
1036skip_left:
1037 hint = secno;
1038 segno = secno * sbi->segs_per_sec;
1039 zoneno = secno / sbi->secs_per_zone;
1040
1041 /* give up on finding another zone */
1042 if (!init)
1043 goto got_it;
1044 if (sbi->secs_per_zone == 1)
1045 goto got_it;
1046 if (zoneno == old_zoneno)
1047 goto got_it;
1048 if (dir == ALLOC_LEFT) {
1049 if (!go_left && zoneno + 1 >= total_zones)
1050 goto got_it;
1051 if (go_left && zoneno == 0)
1052 goto got_it;
1053 }
1054 for (i = 0; i < NR_CURSEG_TYPE; i++)
1055 if (CURSEG_I(sbi, i)->zone == zoneno)
1056 break;
1057
1058 if (i < NR_CURSEG_TYPE) {
1059 /* zone is in user, try another */
1060 if (go_left)
1061 hint = zoneno * sbi->secs_per_zone - 1;
1062 else if (zoneno + 1 >= total_zones)
1063 hint = 0;
1064 else
1065 hint = (zoneno + 1) * sbi->secs_per_zone;
1066 init = false;
1067 goto find_other_zone;
1068 }
1069got_it:
1070 /* set it as dirty segment in free segmap */
9850cf4a 1071 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1072 __set_inuse(sbi, segno);
1073 *newseg = segno;
1a118ccf 1074 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1075}
1076
1077static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1078{
1079 struct curseg_info *curseg = CURSEG_I(sbi, type);
1080 struct summary_footer *sum_footer;
1081
1082 curseg->segno = curseg->next_segno;
1083 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1084 curseg->next_blkoff = 0;
1085 curseg->next_segno = NULL_SEGNO;
1086
1087 sum_footer = &(curseg->sum_blk->footer);
1088 memset(sum_footer, 0, sizeof(struct summary_footer));
1089 if (IS_DATASEG(type))
1090 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1091 if (IS_NODESEG(type))
1092 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1093 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1094}
1095
0a8165d7 1096/*
351df4b2
JK		 1097 * Allocate a current working segment.
1098 * This function always allocates a free segment in LFS manner.
1099 */
1100static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1101{
1102 struct curseg_info *curseg = CURSEG_I(sbi, type);
1103 unsigned int segno = curseg->segno;
1104 int dir = ALLOC_LEFT;
1105
1106 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1107 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1108 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1109 dir = ALLOC_RIGHT;
1110
1111 if (test_opt(sbi, NOHEAP))
1112 dir = ALLOC_RIGHT;
1113
1114 get_new_segment(sbi, &segno, new_sec, dir);
1115 curseg->next_segno = segno;
1116 reset_curseg(sbi, type, 1);
1117 curseg->alloc_type = LFS;
1118}
1119
1120static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1121 struct curseg_info *seg, block_t start)
1122{
1123 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1124 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1125 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1126 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1127 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1128 int i, pos;
1129
1130 for (i = 0; i < entries; i++)
1131 target_map[i] = ckpt_map[i] | cur_map[i];
1132
1133 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1134
1135 seg->next_blkoff = pos;
351df4b2
JK		 1136}
1137
0a8165d7 1138/*
351df4b2
JK		 1139 * If a segment is written by LFS manner, next block offset is just obtained
1140 * by increasing the current block offset. However, if a segment is written by
1141 * SSR manner, next block offset obtained by calling __next_free_blkoff
1142 */
1143static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1144 struct curseg_info *seg)
1145{
1146 if (seg->alloc_type == SSR)
1147 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1148 else
1149 seg->next_blkoff++;
1150}
1151
0a8165d7 1152/*
e1c42045 1153 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1154 * manner, so it should recover the existing segment information of valid blocks
1155 */
1156static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1157{
1158 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1159 struct curseg_info *curseg = CURSEG_I(sbi, type);
1160 unsigned int new_segno = curseg->next_segno;
1161 struct f2fs_summary_block *sum_node;
1162 struct page *sum_page;
1163
1164 write_sum_page(sbi, curseg->sum_blk,
1165 GET_SUM_BLOCK(sbi, curseg->segno));
1166 __set_test_and_inuse(sbi, new_segno);
1167
1168 mutex_lock(&dirty_i->seglist_lock);
1169 __remove_dirty_segment(sbi, new_segno, PRE);
1170 __remove_dirty_segment(sbi, new_segno, DIRTY);
1171 mutex_unlock(&dirty_i->seglist_lock);
1172
1173 reset_curseg(sbi, type, 1);
1174 curseg->alloc_type = SSR;
1175 __next_free_blkoff(sbi, curseg, 0);
1176
1177 if (reuse) {
1178 sum_page = get_sum_page(sbi, new_segno);
1179 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1180 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1181 f2fs_put_page(sum_page, 1);
1182 }
1183}
1184
43727527
JK		 1185static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1186{
1187 struct curseg_info *curseg = CURSEG_I(sbi, type);
1188 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1189
1190 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1191 return v_ops->get_victim(sbi,
1192 &(curseg)->next_segno, BG_GC, type, SSR);
1193
1194 /* For data segments, let's do SSR more intensively */
1195 for (; type >= CURSEG_HOT_DATA; type--)
1196 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1197 BG_GC, type, SSR))
1198 return 1;
1199 return 0;
1200}
1201
351df4b2
JK		 1202/*
1203 * flush out current segment and replace it with new segment
1204 * This function should be returned with success, otherwise BUG
1205 */
1206static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1207 int type, bool force)
1208{
1209 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1210
7b405275 1211 if (force)
351df4b2 1212 new_curseg(sbi, type, true);
7b405275 1213 else if (type == CURSEG_WARM_NODE)
351df4b2 1214 new_curseg(sbi, type, false);
60374688
JK		 1215 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1216 new_curseg(sbi, type, false);
351df4b2
JK		 1217 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1218 change_curseg(sbi, type, true);
1219 else
1220 new_curseg(sbi, type, false);
dcdfff65
JK		 1221
1222 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1223}
1224
38aa0889
JK		 1225static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1226{
1227 struct curseg_info *curseg = CURSEG_I(sbi, type);
1228 unsigned int old_segno;
1229
1230 old_segno = curseg->segno;
1231 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1232 locate_dirty_segment(sbi, old_segno);
1233}
1234
351df4b2
JK		 1235void allocate_new_segments(struct f2fs_sb_info *sbi)
1236{
351df4b2
JK		 1237 int i;
1238
36abef4e
JK		 1239 if (test_opt(sbi, LFS))
1240 return;
1241
38aa0889
JK		 1242 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1243 __allocate_new_segments(sbi, i);
351df4b2
JK		 1244}
1245
1246static const struct segment_allocation default_salloc_ops = {
1247 .allocate_segment = allocate_segment_by_default,
1248};
1249
4b2fecc8
JK		 1250int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1251{
f7ef9b83
JK		 1252 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1253 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1254 unsigned int start_segno, end_segno;
1255 struct cp_control cpc;
c34f42e2 1256 int err = 0;
4b2fecc8 1257
836b5a63 1258 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1259 return -EINVAL;
1260
9bd27ae4 1261 cpc.trimmed = 0;
7cd8558b 1262 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1263 goto out;
1264
1265 /* start/end segment number in main_area */
7cd8558b
JK		 1266 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1267 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1268 GET_SEGNO(sbi, end);
4b2fecc8 1269 cpc.reason = CP_DISCARD;
836b5a63 1270 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1271
1272 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1273 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1274 cpc.trim_start = start_segno;
a66cdd98
JK		 1275
1276 if (sbi->discard_blks == 0)
1277 break;
1278 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1279 cpc.trim_end = end_segno;
1280 else
1281 cpc.trim_end = min_t(unsigned int,
1282 rounddown(start_segno +
bba681cb
JK		 1283 BATCHED_TRIM_SEGMENTS(sbi),
1284 sbi->segs_per_sec) - 1, end_segno);
1285
1286 mutex_lock(&sbi->gc_mutex);
c34f42e2 1287 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1288 mutex_unlock(&sbi->gc_mutex);
1289 }
4b2fecc8 1290out:
f7ef9b83 1291 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1292 return err;
4b2fecc8
JK		 1293}
1294
351df4b2
JK		 1295static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1296{
1297 struct curseg_info *curseg = CURSEG_I(sbi, type);
1298 if (curseg->next_blkoff < sbi->blocks_per_seg)
1299 return true;
1300 return false;
1301}
1302
1303static int __get_segment_type_2(struct page *page, enum page_type p_type)
1304{
1305 if (p_type == DATA)
1306 return CURSEG_HOT_DATA;
1307 else
1308 return CURSEG_HOT_NODE;
1309}
1310
1311static int __get_segment_type_4(struct page *page, enum page_type p_type)
1312{
1313 if (p_type == DATA) {
1314 struct inode *inode = page->mapping->host;
1315
1316 if (S_ISDIR(inode->i_mode))
1317 return CURSEG_HOT_DATA;
1318 else
1319 return CURSEG_COLD_DATA;
1320 } else {
a344b9fd
JK		 1321 if (IS_DNODE(page) && is_cold_node(page))
1322 return CURSEG_WARM_NODE;
351df4b2
JK		 1323 else
1324 return CURSEG_COLD_NODE;
1325 }
1326}
1327
1328static int __get_segment_type_6(struct page *page, enum page_type p_type)
1329{
1330 if (p_type == DATA) {
1331 struct inode *inode = page->mapping->host;
1332
1333 if (S_ISDIR(inode->i_mode))
1334 return CURSEG_HOT_DATA;
354a3399 1335 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1336 return CURSEG_COLD_DATA;
1337 else
1338 return CURSEG_WARM_DATA;
1339 } else {
1340 if (IS_DNODE(page))
1341 return is_cold_node(page) ? CURSEG_WARM_NODE :
1342 CURSEG_HOT_NODE;
1343 else
1344 return CURSEG_COLD_NODE;
1345 }
1346}
1347
1348static int __get_segment_type(struct page *page, enum page_type p_type)
1349{
4081363f 1350 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1351 case 2:
1352 return __get_segment_type_2(page, p_type);
1353 case 4:
1354 return __get_segment_type_4(page, p_type);
351df4b2 1355 }
12a67146 1356 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1357 f2fs_bug_on(F2FS_P_SB(page),
1358 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1359 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1360}
1361
bfad7c2d
JK		 1362void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1363 block_t old_blkaddr, block_t *new_blkaddr,
1364 struct f2fs_summary *sum, int type)
351df4b2
JK		 1365{
1366 struct sit_info *sit_i = SIT_I(sbi);
1367 struct curseg_info *curseg;
38aa0889
JK		 1368 bool direct_io = (type == CURSEG_DIRECT_IO);
1369
1370 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1371
351df4b2
JK		 1372 curseg = CURSEG_I(sbi, type);
1373
1374 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1375 mutex_lock(&sit_i->sentry_lock);
351df4b2 1376
38aa0889 1377 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1378 if (direct_io && curseg->next_blkoff &&
1379 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1380 __allocate_new_segments(sbi, type);
1381
351df4b2 1382 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1383
1384 /*
1385 * __add_sum_entry should be resided under the curseg_mutex
1386 * because, this function updates a summary entry in the
1387 * current summary block.
1388 */
e79efe3b 1389 __add_sum_entry(sbi, type, sum);
351df4b2 1390
351df4b2 1391 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1392
1393 stat_inc_block_count(sbi, curseg);
351df4b2 1394
5e443818
JK		 1395 if (!__has_curseg_space(sbi, type))
1396 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1397 /*
1398 * SIT information should be updated before segment allocation,
1399 * since SSR needs latest valid block information.
1400 */
1401 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1402
351df4b2
JK		 1403 mutex_unlock(&sit_i->sentry_lock);
1404
bfad7c2d 1405 if (page && IS_NODESEG(type))
351df4b2
JK		 1406 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1407
bfad7c2d
JK		 1408 mutex_unlock(&curseg->curseg_mutex);
1409}
1410
05ca3632 1411static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1412{
05ca3632 1413 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1414
7dfeaa32
JK		 1415 if (fio->type == NODE || fio->type == DATA)
1416 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1417
7a9d7548
CY		 1418 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1419 &fio->new_blkaddr, sum, type);
bfad7c2d 1420
351df4b2 1421 /* writeout dirty page into bdev */
05ca3632 1422 f2fs_submit_page_mbio(fio);
7dfeaa32
JK		 1423
1424 if (fio->type == NODE || fio->type == DATA)
1425 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1426}
1427
577e3495 1428void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1429{
458e6197 1430 struct f2fs_io_info fio = {
05ca3632 1431 .sbi = sbi,
458e6197 1432 .type = META,
cf04e8eb 1433 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1434 .old_blkaddr = page->index,
1435 .new_blkaddr = page->index,
05ca3632 1436 .page = page,
4375a336 1437 .encrypted_page = NULL,
458e6197
JK		 1438 };
1439
2b947003
CY		 1440 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1441 fio.rw &= ~REQ_META;
1442
351df4b2 1443 set_page_writeback(page);
05ca3632 1444 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1445}
1446
05ca3632 1447void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1448{
1449 struct f2fs_summary sum;
05ca3632 1450
351df4b2 1451 set_summary(&sum, nid, 0, 0);
05ca3632 1452 do_write_page(&sum, fio);
351df4b2
JK		 1453}
1454
05ca3632 1455void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1456{
05ca3632 1457 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1458 struct f2fs_summary sum;
1459 struct node_info ni;
1460
9850cf4a 1461 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1462 get_node_info(sbi, dn->nid, &ni);
1463 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1464 do_write_page(&sum, fio);
f28b3434 1465 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1466}
1467
05ca3632 1468void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1469{
7a9d7548 1470 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1471 stat_inc_inplace_blocks(fio->sbi);
1472 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1473}
1474
4356e48e 1475void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1476 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1477 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1478{
1479 struct sit_info *sit_i = SIT_I(sbi);
1480 struct curseg_info *curseg;
1481 unsigned int segno, old_cursegno;
1482 struct seg_entry *se;
1483 int type;
19f106bc 1484 unsigned short old_blkoff;
351df4b2
JK		 1485
1486 segno = GET_SEGNO(sbi, new_blkaddr);
1487 se = get_seg_entry(sbi, segno);
1488 type = se->type;
1489
19f106bc
CY		 1490 if (!recover_curseg) {
1491 /* for recovery flow */
1492 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1493 if (old_blkaddr == NULL_ADDR)
1494 type = CURSEG_COLD_DATA;
1495 else
1496 type = CURSEG_WARM_DATA;
1497 }
1498 } else {
1499 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1500 type = CURSEG_WARM_DATA;
1501 }
19f106bc 1502
351df4b2
JK		 1503 curseg = CURSEG_I(sbi, type);
1504
1505 mutex_lock(&curseg->curseg_mutex);
1506 mutex_lock(&sit_i->sentry_lock);
1507
1508 old_cursegno = curseg->segno;
19f106bc 1509 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1510
1511 /* change the current segment */
1512 if (segno != curseg->segno) {
1513 curseg->next_segno = segno;
1514 change_curseg(sbi, type, true);
1515 }
1516
491c0854 1517 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1518 __add_sum_entry(sbi, type, sum);
351df4b2 1519
28bc106b 1520 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1521 update_sit_entry(sbi, new_blkaddr, 1);
1522 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1523 update_sit_entry(sbi, old_blkaddr, -1);
1524
1525 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1526 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1527
351df4b2 1528 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1529
19f106bc
CY		 1530 if (recover_curseg) {
1531 if (old_cursegno != curseg->segno) {
1532 curseg->next_segno = old_cursegno;
1533 change_curseg(sbi, type, true);
1534 }
1535 curseg->next_blkoff = old_blkoff;
1536 }
1537
351df4b2
JK		 1538 mutex_unlock(&sit_i->sentry_lock);
1539 mutex_unlock(&curseg->curseg_mutex);
1540}
1541
528e3459
CY		 1542void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1543 block_t old_addr, block_t new_addr,
28bc106b
CY		 1544 unsigned char version, bool recover_curseg,
1545 bool recover_newaddr)
528e3459
CY		 1546{
1547 struct f2fs_summary sum;
1548
1549 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1550
28bc106b
CY		 1551 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1552 recover_curseg, recover_newaddr);
528e3459 1553
f28b3434 1554 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1555}
1556
93dfe2ac 1557void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1558 enum page_type type, bool ordered)
93dfe2ac 1559{
93dfe2ac 1560 if (PageWriteback(page)) {
4081363f
JK		 1561 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1562
0c3a5797 1563 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1564 if (ordered)
1565 wait_on_page_writeback(page);
1566 else
1567 wait_for_stable_page(page);
93dfe2ac
JK		 1568 }
1569}
1570
08b39fbd
CY		 1571void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1572 block_t blkaddr)
1573{
1574 struct page *cpage;
1575
1576 if (blkaddr == NEW_ADDR)
1577 return;
1578
1579 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1580
1581 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1582 if (cpage) {
fec1d657 1583 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1584 f2fs_put_page(cpage, 1);
1585 }
1586}
1587
351df4b2
JK		 1588static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1589{
1590 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1591 struct curseg_info *seg_i;
1592 unsigned char *kaddr;
1593 struct page *page;
1594 block_t start;
1595 int i, j, offset;
1596
1597 start = start_sum_block(sbi);
1598
1599 page = get_meta_page(sbi, start++);
1600 kaddr = (unsigned char *)page_address(page);
1601
1602 /* Step 1: restore nat cache */
1603 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1604 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1605
1606 /* Step 2: restore sit cache */
1607 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1608 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1609 offset = 2 * SUM_JOURNAL_SIZE;
1610
1611 /* Step 3: restore summary entries */
1612 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1613 unsigned short blk_off;
1614 unsigned int segno;
1615
1616 seg_i = CURSEG_I(sbi, i);
1617 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1618 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1619 seg_i->next_segno = segno;
1620 reset_curseg(sbi, i, 0);
1621 seg_i->alloc_type = ckpt->alloc_type[i];
1622 seg_i->next_blkoff = blk_off;
1623
1624 if (seg_i->alloc_type == SSR)
1625 blk_off = sbi->blocks_per_seg;
1626
1627 for (j = 0; j < blk_off; j++) {
1628 struct f2fs_summary *s;
1629 s = (struct f2fs_summary *)(kaddr + offset);
1630 seg_i->sum_blk->entries[j] = *s;
1631 offset += SUMMARY_SIZE;
09cbfeaf 1632 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1633 SUM_FOOTER_SIZE)
1634 continue;
1635
1636 f2fs_put_page(page, 1);
1637 page = NULL;
1638
1639 page = get_meta_page(sbi, start++);
1640 kaddr = (unsigned char *)page_address(page);
1641 offset = 0;
1642 }
1643 }
1644 f2fs_put_page(page, 1);
1645 return 0;
1646}
1647
1648static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1649{
1650 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1651 struct f2fs_summary_block *sum;
1652 struct curseg_info *curseg;
1653 struct page *new;
1654 unsigned short blk_off;
1655 unsigned int segno = 0;
1656 block_t blk_addr = 0;
1657
1658 /* get segment number and block addr */
1659 if (IS_DATASEG(type)) {
1660 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1661 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1662 CURSEG_HOT_DATA]);
119ee914 1663 if (__exist_node_summaries(sbi))
351df4b2
JK		 1664 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1665 else
1666 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1667 } else {
1668 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1669 CURSEG_HOT_NODE]);
1670 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1671 CURSEG_HOT_NODE]);
119ee914 1672 if (__exist_node_summaries(sbi))
351df4b2
JK		 1673 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1674 type - CURSEG_HOT_NODE);
1675 else
1676 blk_addr = GET_SUM_BLOCK(sbi, segno);
1677 }
1678
1679 new = get_meta_page(sbi, blk_addr);
1680 sum = (struct f2fs_summary_block *)page_address(new);
1681
1682 if (IS_NODESEG(type)) {
119ee914 1683 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1684 struct f2fs_summary *ns = &sum->entries[0];
1685 int i;
1686 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1687 ns->version = 0;
1688 ns->ofs_in_node = 0;
1689 }
1690 } else {
d653788a
GZ		 1691 int err;
1692
1693 err = restore_node_summary(sbi, segno, sum);
1694 if (err) {
351df4b2 1695 f2fs_put_page(new, 1);
d653788a 1696 return err;
351df4b2
JK		 1697 }
1698 }
1699 }
1700
1701 /* set uncompleted segment to curseg */
1702 curseg = CURSEG_I(sbi, type);
1703 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1704
1705 /* update journal info */
1706 down_write(&curseg->journal_rwsem);
1707 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1708 up_write(&curseg->journal_rwsem);
1709
1710 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1711 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1712 curseg->next_segno = segno;
1713 reset_curseg(sbi, type, 0);
1714 curseg->alloc_type = ckpt->alloc_type[type];
1715 curseg->next_blkoff = blk_off;
1716 mutex_unlock(&curseg->curseg_mutex);
1717 f2fs_put_page(new, 1);
1718 return 0;
1719}
1720
1721static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1722{
1723 int type = CURSEG_HOT_DATA;
e4fc5fbf 1724 int err;
351df4b2 1725
25ca923b 1726 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1727 int npages = npages_for_summary_flush(sbi, true);
1728
1729 if (npages >= 2)
1730 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1731 META_CP, true);
3fa06d7b 1732
351df4b2
JK		 1733 /* restore for compacted data summary */
1734 if (read_compacted_summaries(sbi))
1735 return -EINVAL;
1736 type = CURSEG_HOT_NODE;
1737 }
1738
119ee914 1739 if (__exist_node_summaries(sbi))
3fa06d7b 1740 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1741 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1742
e4fc5fbf
CY		 1743 for (; type <= CURSEG_COLD_NODE; type++) {
1744 err = read_normal_summaries(sbi, type);
1745 if (err)
1746 return err;
1747 }
1748
351df4b2
JK		 1749 return 0;
1750}
1751
1752static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1753{
1754 struct page *page;
1755 unsigned char *kaddr;
1756 struct f2fs_summary *summary;
1757 struct curseg_info *seg_i;
1758 int written_size = 0;
1759 int i, j;
1760
1761 page = grab_meta_page(sbi, blkaddr++);
1762 kaddr = (unsigned char *)page_address(page);
1763
1764 /* Step 1: write nat cache */
1765 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1766 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1767 written_size += SUM_JOURNAL_SIZE;
1768
1769 /* Step 2: write sit cache */
1770 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1771 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1772 written_size += SUM_JOURNAL_SIZE;
1773
351df4b2
JK		 1774 /* Step 3: write summary entries */
1775 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1776 unsigned short blkoff;
1777 seg_i = CURSEG_I(sbi, i);
1778 if (sbi->ckpt->alloc_type[i] == SSR)
1779 blkoff = sbi->blocks_per_seg;
1780 else
1781 blkoff = curseg_blkoff(sbi, i);
1782
1783 for (j = 0; j < blkoff; j++) {
1784 if (!page) {
1785 page = grab_meta_page(sbi, blkaddr++);
1786 kaddr = (unsigned char *)page_address(page);
1787 written_size = 0;
1788 }
1789 summary = (struct f2fs_summary *)(kaddr + written_size);
1790 *summary = seg_i->sum_blk->entries[j];
1791 written_size += SUMMARY_SIZE;
351df4b2 1792
09cbfeaf 1793 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1794 SUM_FOOTER_SIZE)
1795 continue;
1796
e8d61a74 1797 set_page_dirty(page);
351df4b2
JK		 1798 f2fs_put_page(page, 1);
1799 page = NULL;
1800 }
1801 }
e8d61a74
CY		 1802 if (page) {
1803 set_page_dirty(page);
351df4b2 1804 f2fs_put_page(page, 1);
e8d61a74 1805 }
351df4b2
JK		 1806}
1807
1808static void write_normal_summaries(struct f2fs_sb_info *sbi,
1809 block_t blkaddr, int type)
1810{
1811 int i, end;
1812 if (IS_DATASEG(type))
1813 end = type + NR_CURSEG_DATA_TYPE;
1814 else
1815 end = type + NR_CURSEG_NODE_TYPE;
1816
b7ad7512
CY		 1817 for (i = type; i < end; i++)
1818 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1819}
1820
1821void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1822{
25ca923b 1823 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1824 write_compacted_summaries(sbi, start_blk);
1825 else
1826 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1827}
1828
1829void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1830{
119ee914 1831 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1832}
1833
dfc08a12 1834int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1835 unsigned int val, int alloc)
1836{
1837 int i;
1838
1839 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1840 for (i = 0; i < nats_in_cursum(journal); i++) {
1841 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1842 return i;
1843 }
dfc08a12
CY		 1844 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1845 return update_nats_in_cursum(journal, 1);
351df4b2 1846 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1847 for (i = 0; i < sits_in_cursum(journal); i++)
1848 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1849 return i;
dfc08a12
CY		 1850 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1851 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1852 }
1853 return -1;
1854}
1855
1856static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1857 unsigned int segno)
1858{
2cc22186 1859 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1860}
1861
1862static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1863 unsigned int start)
1864{
1865 struct sit_info *sit_i = SIT_I(sbi);
1866 struct page *src_page, *dst_page;
1867 pgoff_t src_off, dst_off;
1868 void *src_addr, *dst_addr;
1869
1870 src_off = current_sit_addr(sbi, start);
1871 dst_off = next_sit_addr(sbi, src_off);
1872
1873 /* get current sit block page without lock */
1874 src_page = get_meta_page(sbi, src_off);
1875 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1876 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1877
1878 src_addr = page_address(src_page);
1879 dst_addr = page_address(dst_page);
09cbfeaf 1880 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1881
1882 set_page_dirty(dst_page);
1883 f2fs_put_page(src_page, 1);
1884
1885 set_to_next_sit(sit_i, start);
1886
1887 return dst_page;
1888}
1889
184a5cd2
CY		 1890static struct sit_entry_set *grab_sit_entry_set(void)
1891{
1892 struct sit_entry_set *ses =
80c54505 1893 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1894
1895 ses->entry_cnt = 0;
1896 INIT_LIST_HEAD(&ses->set_list);
1897 return ses;
1898}
1899
1900static void release_sit_entry_set(struct sit_entry_set *ses)
1901{
1902 list_del(&ses->set_list);
1903 kmem_cache_free(sit_entry_set_slab, ses);
1904}
1905
1906static void adjust_sit_entry_set(struct sit_entry_set *ses,
1907 struct list_head *head)
1908{
1909 struct sit_entry_set *next = ses;
1910
1911 if (list_is_last(&ses->set_list, head))
1912 return;
1913
1914 list_for_each_entry_continue(next, head, set_list)
1915 if (ses->entry_cnt <= next->entry_cnt)
1916 break;
1917
1918 list_move_tail(&ses->set_list, &next->set_list);
1919}
1920
1921static void add_sit_entry(unsigned int segno, struct list_head *head)
1922{
1923 struct sit_entry_set *ses;
1924 unsigned int start_segno = START_SEGNO(segno);
1925
1926 list_for_each_entry(ses, head, set_list) {
1927 if (ses->start_segno == start_segno) {
1928 ses->entry_cnt++;
1929 adjust_sit_entry_set(ses, head);
1930 return;
1931 }
1932 }
1933
1934 ses = grab_sit_entry_set();
1935
1936 ses->start_segno = start_segno;
1937 ses->entry_cnt++;
1938 list_add(&ses->set_list, head);
1939}
1940
1941static void add_sits_in_set(struct f2fs_sb_info *sbi)
1942{
1943 struct f2fs_sm_info *sm_info = SM_I(sbi);
1944 struct list_head *set_list = &sm_info->sit_entry_set;
1945 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1946 unsigned int segno;
1947
7cd8558b 1948 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1949 add_sit_entry(segno, set_list);
1950}
1951
1952static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1953{
1954 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1955 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1956 int i;
1957
b7ad7512 1958 down_write(&curseg->journal_rwsem);
dfc08a12 1959 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1960 unsigned int segno;
1961 bool dirtied;
1962
dfc08a12 1963 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1964 dirtied = __mark_sit_entry_dirty(sbi, segno);
1965
1966 if (!dirtied)
1967 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1968 }
dfc08a12 1969 update_sits_in_cursum(journal, -i);
b7ad7512 1970 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1971}
1972
0a8165d7 1973/*
351df4b2
JK		 1974 * CP calls this function, which flushes SIT entries including sit_journal,
1975 * and moves prefree segs to free segs.
1976 */
4b2fecc8 1977void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1978{
1979 struct sit_info *sit_i = SIT_I(sbi);
1980 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1981 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1982 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 1983 struct sit_entry_set *ses, *tmp;
1984 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1985 bool to_journal = true;
4b2fecc8 1986 struct seg_entry *se;
351df4b2 1987
351df4b2
JK		 1988 mutex_lock(&sit_i->sentry_lock);
1989
2b11a74b
WL		 1990 if (!sit_i->dirty_sentries)
1991 goto out;
1992
351df4b2 1993 /*
184a5cd2
CY		 1994 * add and account sit entries of dirty bitmap in sit entry
1995 * set temporarily
351df4b2 1996 */
184a5cd2 1997 add_sits_in_set(sbi);
351df4b2 1998
184a5cd2
CY		 1999 /*
2000 * if there are no enough space in journal to store dirty sit
2001 * entries, remove all entries from journal and add and account
2002 * them in sit entry set.
2003 */
dfc08a12 2004 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2005 remove_sits_in_journal(sbi);
b2955550 2006
184a5cd2
CY		 2007 /*
2008 * there are two steps to flush sit entries:
2009 * #1, flush sit entries to journal in current cold data summary block.
2010 * #2, flush sit entries to sit page.
2011 */
2012 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2013 struct page *page = NULL;
184a5cd2
CY		 2014 struct f2fs_sit_block *raw_sit = NULL;
2015 unsigned int start_segno = ses->start_segno;
2016 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2017 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2018 unsigned int segno = start_segno;
2019
2020 if (to_journal &&
dfc08a12 2021 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2022 to_journal = false;
2023
b7ad7512
CY		 2024 if (to_journal) {
2025 down_write(&curseg->journal_rwsem);
2026 } else {
184a5cd2
CY		 2027 page = get_next_sit_page(sbi, start_segno);
2028 raw_sit = page_address(page);
351df4b2 2029 }
351df4b2 2030
184a5cd2
CY		 2031 /* flush dirty sit entries in region of current sit set */
2032 for_each_set_bit_from(segno, bitmap, end) {
2033 int offset, sit_offset;
4b2fecc8
JK		 2034
2035 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2036
2037 /* add discard candidates */
d7bc2484 2038 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2039 cpc->trim_start = segno;
2040 add_discard_addrs(sbi, cpc);
2041 }
184a5cd2
CY		 2042
2043 if (to_journal) {
dfc08a12 2044 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2045 SIT_JOURNAL, segno, 1);
2046 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2047 segno_in_journal(journal, offset) =
184a5cd2
CY		 2048 cpu_to_le32(segno);
2049 seg_info_to_raw_sit(se,
dfc08a12 2050 &sit_in_journal(journal, offset));
184a5cd2
CY		 2051 } else {
2052 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2053 seg_info_to_raw_sit(se,
2054 &raw_sit->entries[sit_offset]);
2055 }
351df4b2 2056
184a5cd2
CY		 2057 __clear_bit(segno, bitmap);
2058 sit_i->dirty_sentries--;
2059 ses->entry_cnt--;
351df4b2
JK		 2060 }
2061
b7ad7512
CY		 2062 if (to_journal)
2063 up_write(&curseg->journal_rwsem);
2064 else
184a5cd2
CY		 2065 f2fs_put_page(page, 1);
2066
2067 f2fs_bug_on(sbi, ses->entry_cnt);
2068 release_sit_entry_set(ses);
351df4b2 2069 }
184a5cd2
CY		 2070
2071 f2fs_bug_on(sbi, !list_empty(head));
2072 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2073out:
4b2fecc8
JK		 2074 if (cpc->reason == CP_DISCARD) {
2075 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2076 add_discard_addrs(sbi, cpc);
2077 }
351df4b2 2078 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2079
351df4b2
JK		 2080 set_prefree_as_free_segments(sbi);
2081}
2082
2083static int build_sit_info(struct f2fs_sb_info *sbi)
2084{
2085 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2086 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2087 struct sit_info *sit_i;
2088 unsigned int sit_segs, start;
2089 char *src_bitmap, *dst_bitmap;
2090 unsigned int bitmap_size;
2091
2092 /* allocate memory for SIT information */
2093 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2094 if (!sit_i)
2095 return -ENOMEM;
2096
2097 SM_I(sbi)->sit_info = sit_i;
2098
39307a8e
JK		 2099 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2100 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2101 if (!sit_i->sentries)
2102 return -ENOMEM;
2103
7cd8558b 2104 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2105 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2106 if (!sit_i->dirty_sentries_bitmap)
2107 return -ENOMEM;
2108
7cd8558b 2109 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2110 sit_i->sentries[start].cur_valid_map
2111 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2112 sit_i->sentries[start].ckpt_valid_map
2113 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2114 sit_i->sentries[start].discard_map
2115 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2116 if (!sit_i->sentries[start].cur_valid_map ||
2117 !sit_i->sentries[start].ckpt_valid_map ||
2118 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2119 return -ENOMEM;
2120 }
2121
60a3b782
JK		 2122 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2123 if (!sit_i->tmp_map)
2124 return -ENOMEM;
2125
351df4b2 2126 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2127 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2128 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2129 if (!sit_i->sec_entries)
2130 return -ENOMEM;
2131 }
2132
2133 /* get information related with SIT */
2134 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2135
2136 /* setup SIT bitmap from ckeckpoint pack */
2137 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2138 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2139
79b5793b 2140 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2141 if (!dst_bitmap)
2142 return -ENOMEM;
351df4b2
JK		 2143
2144 /* init SIT information */
2145 sit_i->s_ops = &default_salloc_ops;
2146
2147 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2148 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2149 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2150 sit_i->sit_bitmap = dst_bitmap;
2151 sit_i->bitmap_size = bitmap_size;
2152 sit_i->dirty_sentries = 0;
2153 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2154 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2155 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2156 mutex_init(&sit_i->sentry_lock);
2157 return 0;
2158}
2159
2160static int build_free_segmap(struct f2fs_sb_info *sbi)
2161{
351df4b2
JK		 2162 struct free_segmap_info *free_i;
2163 unsigned int bitmap_size, sec_bitmap_size;
2164
2165 /* allocate memory for free segmap information */
2166 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2167 if (!free_i)
2168 return -ENOMEM;
2169
2170 SM_I(sbi)->free_info = free_i;
2171
7cd8558b 2172 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2173 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2174 if (!free_i->free_segmap)
2175 return -ENOMEM;
2176
7cd8558b 2177 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2178 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2179 if (!free_i->free_secmap)
2180 return -ENOMEM;
2181
2182 /* set all segments as dirty temporarily */
2183 memset(free_i->free_segmap, 0xff, bitmap_size);
2184 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2185
2186 /* init free segmap information */
7cd8558b 2187 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2188 free_i->free_segments = 0;
2189 free_i->free_sections = 0;
1a118ccf 2190 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2191 return 0;
2192}
2193
2194static int build_curseg(struct f2fs_sb_info *sbi)
2195{
1042d60f 2196 struct curseg_info *array;
351df4b2
JK		 2197 int i;
2198
b434babf 2199 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2200 if (!array)
2201 return -ENOMEM;
2202
2203 SM_I(sbi)->curseg_array = array;
2204
2205 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2206 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2207 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2208 if (!array[i].sum_blk)
2209 return -ENOMEM;
b7ad7512
CY		 2210 init_rwsem(&array[i].journal_rwsem);
2211 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2212 GFP_KERNEL);
2213 if (!array[i].journal)
2214 return -ENOMEM;
351df4b2
JK		 2215 array[i].segno = NULL_SEGNO;
2216 array[i].next_blkoff = 0;
2217 }
2218 return restore_curseg_summaries(sbi);
2219}
2220
2221static void build_sit_entries(struct f2fs_sb_info *sbi)
2222{
2223 struct sit_info *sit_i = SIT_I(sbi);
2224 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2225 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2226 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2227 unsigned int i, start, end;
2228 unsigned int readed, start_blk = 0;
e9f5b8b8 2229 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2230
74de593a 2231 do {
26879fb1 2232 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2233
2234 start = start_blk * sit_i->sents_per_block;
2235 end = (start_blk + readed) * sit_i->sents_per_block;
2236
7cd8558b 2237 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2238 struct seg_entry *se = &sit_i->sentries[start];
2239 struct f2fs_sit_block *sit_blk;
2240 struct f2fs_sit_entry sit;
2241 struct page *page;
2242
b7ad7512 2243 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2244 for (i = 0; i < sits_in_cursum(journal); i++) {
2245 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2246 == start) {
dfc08a12 2247 sit = sit_in_journal(journal, i);
b7ad7512 2248 up_read(&curseg->journal_rwsem);
74de593a
CY		 2249 goto got_it;
2250 }
351df4b2 2251 }
b7ad7512 2252 up_read(&curseg->journal_rwsem);
74de593a
CY		 2253
2254 page = get_current_sit_page(sbi, start);
2255 sit_blk = (struct f2fs_sit_block *)page_address(page);
2256 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2257 f2fs_put_page(page, 1);
351df4b2 2258got_it:
74de593a
CY		 2259 check_block_count(sbi, start, &sit);
2260 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2261
2262 /* build discard map only one time */
2263 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2264 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2265
74de593a
CY		 2266 if (sbi->segs_per_sec > 1) {
2267 struct sec_entry *e = get_sec_entry(sbi, start);
2268 e->valid_blocks += se->valid_blocks;
2269 }
351df4b2 2270 }
74de593a
CY		 2271 start_blk += readed;
2272 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2273}
2274
2275static void init_free_segmap(struct f2fs_sb_info *sbi)
2276{
2277 unsigned int start;
2278 int type;
2279
7cd8558b 2280 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2281 struct seg_entry *sentry = get_seg_entry(sbi, start);
2282 if (!sentry->valid_blocks)
2283 __set_free(sbi, start);
2284 }
2285
2286 /* set use the current segments */
2287 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2288 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2289 __set_test_and_inuse(sbi, curseg_t->segno);
2290 }
2291}
2292
2293static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2294{
2295 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2296 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2297 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2298 unsigned short valid_blocks;
2299
8736fbf0 2300 while (1) {
351df4b2 2301 /* find dirty segment based on free segmap */
7cd8558b
JK		 2302 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2303 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2304 break;
2305 offset = segno + 1;
2306 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2307 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2308 continue;
ec325b52
JK		 2309 if (valid_blocks > sbi->blocks_per_seg) {
2310 f2fs_bug_on(sbi, 1);
2311 continue;
2312 }
351df4b2
JK		 2313 mutex_lock(&dirty_i->seglist_lock);
2314 __locate_dirty_segment(sbi, segno, DIRTY);
2315 mutex_unlock(&dirty_i->seglist_lock);
2316 }
2317}
2318
5ec4e49f 2319static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2320{
2321 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2322 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2323
39307a8e 2324 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2325 if (!dirty_i->victim_secmap)
351df4b2
JK		 2326 return -ENOMEM;
2327 return 0;
2328}
2329
2330static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2331{
2332 struct dirty_seglist_info *dirty_i;
2333 unsigned int bitmap_size, i;
2334
2335 /* allocate memory for dirty segments list information */
2336 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2337 if (!dirty_i)
2338 return -ENOMEM;
2339
2340 SM_I(sbi)->dirty_info = dirty_i;
2341 mutex_init(&dirty_i->seglist_lock);
2342
7cd8558b 2343 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2344
2345 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2346 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2347 if (!dirty_i->dirty_segmap[i])
2348 return -ENOMEM;
2349 }
2350
2351 init_dirty_segmap(sbi);
5ec4e49f 2352 return init_victim_secmap(sbi);
351df4b2
JK		 2353}
2354
0a8165d7 2355/*
351df4b2
JK		 2356 * Update min, max modified time for cost-benefit GC algorithm
2357 */
2358static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2359{
2360 struct sit_info *sit_i = SIT_I(sbi);
2361 unsigned int segno;
2362
2363 mutex_lock(&sit_i->sentry_lock);
2364
2365 sit_i->min_mtime = LLONG_MAX;
2366
7cd8558b 2367 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2368 unsigned int i;
2369 unsigned long long mtime = 0;
2370
2371 for (i = 0; i < sbi->segs_per_sec; i++)
2372 mtime += get_seg_entry(sbi, segno + i)->mtime;
2373
2374 mtime = div_u64(mtime, sbi->segs_per_sec);
2375
2376 if (sit_i->min_mtime > mtime)
2377 sit_i->min_mtime = mtime;
2378 }
2379 sit_i->max_mtime = get_mtime(sbi);
2380 mutex_unlock(&sit_i->sentry_lock);
2381}
2382
2383int build_segment_manager(struct f2fs_sb_info *sbi)
2384{
2385 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2386 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2387 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2388 int err;
2389
2390 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2391 if (!sm_info)
2392 return -ENOMEM;
2393
2394 /* init sm info */
2395 sbi->sm_info = sm_info;
351df4b2
JK		 2396 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2397 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2398 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2399 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2400 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2401 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2402 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2403 sm_info->rec_prefree_segments = sm_info->main_segments *
2404 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2405 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2406 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2407 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2408
7fd9e544
JK		 2409 INIT_LIST_HEAD(&sm_info->discard_list);
2410 sm_info->nr_discards = 0;
2411 sm_info->max_discards = 0;
2412
bba681cb
JK		 2413 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2414
184a5cd2
CY		 2415 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2416
b270ad6f 2417 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2418 err = create_flush_cmd_control(sbi);
2419 if (err)
a688b9d9 2420 return err;
6b4afdd7
JK		 2421 }
2422
351df4b2
JK		 2423 err = build_sit_info(sbi);
2424 if (err)
2425 return err;
2426 err = build_free_segmap(sbi);
2427 if (err)
2428 return err;
2429 err = build_curseg(sbi);
2430 if (err)
2431 return err;
2432
2433 /* reinit free segmap based on SIT */
2434 build_sit_entries(sbi);
2435
2436 init_free_segmap(sbi);
2437 err = build_dirty_segmap(sbi);
2438 if (err)
2439 return err;
2440
2441 init_min_max_mtime(sbi);
2442 return 0;
2443}
2444
2445static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2446 enum dirty_type dirty_type)
2447{
2448 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2449
2450 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2451 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2452 dirty_i->nr_dirty[dirty_type] = 0;
2453 mutex_unlock(&dirty_i->seglist_lock);
2454}
2455
5ec4e49f 2456static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2457{
2458 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2459 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2460}
2461
2462static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2463{
2464 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2465 int i;
2466
2467 if (!dirty_i)
2468 return;
2469
2470 /* discard pre-free/dirty segments list */
2471 for (i = 0; i < NR_DIRTY_TYPE; i++)
2472 discard_dirty_segmap(sbi, i);
2473
5ec4e49f 2474 destroy_victim_secmap(sbi);
351df4b2
JK		 2475 SM_I(sbi)->dirty_info = NULL;
2476 kfree(dirty_i);
2477}
2478
2479static void destroy_curseg(struct f2fs_sb_info *sbi)
2480{
2481 struct curseg_info *array = SM_I(sbi)->curseg_array;
2482 int i;
2483
2484 if (!array)
2485 return;
2486 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2487 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2488 kfree(array[i].sum_blk);
b7ad7512
CY		 2489 kfree(array[i].journal);
2490 }
351df4b2
JK		 2491 kfree(array);
2492}
2493
2494static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2495{
2496 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2497 if (!free_i)
2498 return;
2499 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2500 kvfree(free_i->free_segmap);
2501 kvfree(free_i->free_secmap);
351df4b2
JK		 2502 kfree(free_i);
2503}
2504
2505static void destroy_sit_info(struct f2fs_sb_info *sbi)
2506{
2507 struct sit_info *sit_i = SIT_I(sbi);
2508 unsigned int start;
2509
2510 if (!sit_i)
2511 return;
2512
2513 if (sit_i->sentries) {
7cd8558b 2514 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2515 kfree(sit_i->sentries[start].cur_valid_map);
2516 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2517 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2518 }
2519 }
60a3b782
JK		 2520 kfree(sit_i->tmp_map);
2521
39307a8e
JK		 2522 kvfree(sit_i->sentries);
2523 kvfree(sit_i->sec_entries);
2524 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2525
2526 SM_I(sbi)->sit_info = NULL;
2527 kfree(sit_i->sit_bitmap);
2528 kfree(sit_i);
2529}
2530
2531void destroy_segment_manager(struct f2fs_sb_info *sbi)
2532{
2533 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2534
3b03f724
CY		 2535 if (!sm_info)
2536 return;
2163d198 2537 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2538 destroy_dirty_segmap(sbi);
2539 destroy_curseg(sbi);
2540 destroy_free_segmap(sbi);
2541 destroy_sit_info(sbi);
2542 sbi->sm_info = NULL;
2543 kfree(sm_info);
2544}
7fd9e544
JK		 2545
2546int __init create_segment_manager_caches(void)
2547{
2548 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2549 sizeof(struct discard_entry));
7fd9e544 2550 if (!discard_entry_slab)
184a5cd2
CY		 2551 goto fail;
2552
2553 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2554 sizeof(struct sit_entry_set));
184a5cd2
CY		 2555 if (!sit_entry_set_slab)
2556 goto destory_discard_entry;
88b88a66
JK		 2557
2558 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2559 sizeof(struct inmem_pages));
2560 if (!inmem_entry_slab)
2561 goto destroy_sit_entry_set;
7fd9e544 2562 return 0;
184a5cd2 2563
88b88a66
JK		 2564destroy_sit_entry_set:
2565 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2566destory_discard_entry:
2567 kmem_cache_destroy(discard_entry_slab);
2568fail:
2569 return -ENOMEM;
7fd9e544
JK		 2570}
2571
2572void destroy_segment_manager_caches(void)
2573{
184a5cd2 2574 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2575 kmem_cache_destroy(discard_entry_slab);
88b88a66 2576 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2577}
Ubuntu Artful kernel mirror