projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: add maximum prefree segments
[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;
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 373 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 374 try_to_free_nids(sbi, MAX_FREE_NIDS);
375 else
376 build_free_nids(sbi);
31696580 377
1b38dc8e
JK		 378 /* checkpoint is the only way to shrink partial cached entries */
379 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 380 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 381 excess_prefree_segs(sbi) ||
382 excess_dirty_nats(sbi) ||
d0239e1b 383 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
19a5f5e2 384 if (test_opt(sbi, DATA_FLUSH))
36b35a0d 385 sync_dirty_inodes(sbi, FILE_INODE);
4660f9c0 386 f2fs_sync_fs(sbi->sb, true);
42190d2a 387 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 388 }
4660f9c0
JK		 389}
390
2163d198 391static int issue_flush_thread(void *data)
6b4afdd7
JK		 392{
393 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 394 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
395 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 396repeat:
397 if (kthread_should_stop())
398 return 0;
399
721bd4d5 400 if (!llist_empty(&fcc->issue_list)) {
740432f8 401 struct bio *bio;
6b4afdd7
JK		 402 struct flush_cmd *cmd, *next;
403 int ret;
404
740432f8
JK		 405 bio = f2fs_bio_alloc(0);
406
721bd4d5
GZ		 407 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
408 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
409
6b4afdd7
JK		 410 bio->bi_bdev = sbi->sb->s_bdev;
411 ret = submit_bio_wait(WRITE_FLUSH, bio);
412
721bd4d5
GZ		 413 llist_for_each_entry_safe(cmd, next,
414 fcc->dispatch_list, llnode) {
6b4afdd7 415 cmd->ret = ret;
6b4afdd7
JK		 416 complete(&cmd->wait);
417 }
a4ed23f2 418 bio_put(bio);
a688b9d9 419 fcc->dispatch_list = NULL;
6b4afdd7
JK		 420 }
421
a688b9d9 422 wait_event_interruptible(*q,
721bd4d5 423 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 424 goto repeat;
425}
426
427int f2fs_issue_flush(struct f2fs_sb_info *sbi)
428{
a688b9d9 429 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 430 struct flush_cmd cmd;
6b4afdd7 431
24a9ee0f
JK		 432 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
433 test_opt(sbi, FLUSH_MERGE));
434
0f7b2abd
JK		 435 if (test_opt(sbi, NOBARRIER))
436 return 0;
437
0a87f664 438 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
740432f8
JK		 439 struct bio *bio = f2fs_bio_alloc(0);
440 int ret;
441
0a87f664 442 atomic_inc(&fcc->submit_flush);
740432f8
JK		 443 bio->bi_bdev = sbi->sb->s_bdev;
444 ret = submit_bio_wait(WRITE_FLUSH, bio);
0a87f664 445 atomic_dec(&fcc->submit_flush);
740432f8
JK		 446 bio_put(bio);
447 return ret;
448 }
6b4afdd7 449
adf8d90b 450 init_completion(&cmd.wait);
6b4afdd7 451
0a87f664 452 atomic_inc(&fcc->submit_flush);
721bd4d5 453 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 454
a688b9d9
GZ		 455 if (!fcc->dispatch_list)
456 wake_up(&fcc->flush_wait_queue);
6b4afdd7 457
adf8d90b 458 wait_for_completion(&cmd.wait);
0a87f664 459 atomic_dec(&fcc->submit_flush);
adf8d90b
CY		 460
461 return cmd.ret;
6b4afdd7
JK		 462}
463
2163d198
GZ		 464int create_flush_cmd_control(struct f2fs_sb_info *sbi)
465{
466 dev_t dev = sbi->sb->s_bdev->bd_dev;
467 struct flush_cmd_control *fcc;
468 int err = 0;
469
470 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
471 if (!fcc)
472 return -ENOMEM;
0a87f664 473 atomic_set(&fcc->submit_flush, 0);
2163d198 474 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 475 init_llist_head(&fcc->issue_list);
6b2920a5 476 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 477 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
478 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
479 if (IS_ERR(fcc->f2fs_issue_flush)) {
480 err = PTR_ERR(fcc->f2fs_issue_flush);
481 kfree(fcc);
6b2920a5 482 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 483 return err;
484 }
2163d198
GZ		 485
486 return err;
487}
488
489void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
490{
6b2920a5 491 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 492
493 if (fcc && fcc->f2fs_issue_flush)
494 kthread_stop(fcc->f2fs_issue_flush);
495 kfree(fcc);
6b2920a5 496 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 497}
498
351df4b2
JK		 499static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
500 enum dirty_type dirty_type)
501{
502 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
503
504 /* need not be added */
505 if (IS_CURSEG(sbi, segno))
506 return;
507
508 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
509 dirty_i->nr_dirty[dirty_type]++;
510
511 if (dirty_type == DIRTY) {
512 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 513 enum dirty_type t = sentry->type;
b2f2c390 514
ec325b52
JK		 515 if (unlikely(t >= DIRTY)) {
516 f2fs_bug_on(sbi, 1);
517 return;
518 }
4625d6aa
CL		 519 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
520 dirty_i->nr_dirty[t]++;
351df4b2
JK		 521 }
522}
523
524static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
525 enum dirty_type dirty_type)
526{
527 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
528
529 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
530 dirty_i->nr_dirty[dirty_type]--;
531
532 if (dirty_type == DIRTY) {
4625d6aa
CL		 533 struct seg_entry *sentry = get_seg_entry(sbi, segno);
534 enum dirty_type t = sentry->type;
535
536 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
537 dirty_i->nr_dirty[t]--;
b2f2c390 538
5ec4e49f
JK		 539 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
540 clear_bit(GET_SECNO(sbi, segno),
541 dirty_i->victim_secmap);
351df4b2
JK		 542 }
543}
544
0a8165d7 545/*
351df4b2
JK		 546 * Should not occur error such as -ENOMEM.
547 * Adding dirty entry into seglist is not critical operation.
548 * If a given segment is one of current working segments, it won't be added.
549 */
8d8451af 550static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 551{
552 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
553 unsigned short valid_blocks;
554
555 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
556 return;
557
558 mutex_lock(&dirty_i->seglist_lock);
559
560 valid_blocks = get_valid_blocks(sbi, segno, 0);
561
562 if (valid_blocks == 0) {
563 __locate_dirty_segment(sbi, segno, PRE);
564 __remove_dirty_segment(sbi, segno, DIRTY);
565 } else if (valid_blocks < sbi->blocks_per_seg) {
566 __locate_dirty_segment(sbi, segno, DIRTY);
567 } else {
568 /* Recovery routine with SSR needs this */
569 __remove_dirty_segment(sbi, segno, DIRTY);
570 }
571
572 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 573}
574
1e87a78d 575static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 576 block_t blkstart, block_t blklen)
577{
55cf9cb6
CY		 578 sector_t start = SECTOR_FROM_BLOCK(blkstart);
579 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 580 struct seg_entry *se;
581 unsigned int offset;
582 block_t i;
583
584 for (i = blkstart; i < blkstart + blklen; i++) {
585 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
586 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
587
588 if (!f2fs_test_and_set_bit(offset, se->discard_map))
589 sbi->discard_blks--;
590 }
1661d07c 591 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 592 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
593}
594
e90c2d28 595bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 596{
60b286c4 597 int err = -EOPNOTSUPP;
40a02be1
JK		 598
599 if (test_opt(sbi, DISCARD)) {
600 struct seg_entry *se = get_seg_entry(sbi,
601 GET_SEGNO(sbi, blkaddr));
602 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
603
604 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 605 return false;
40a02be1
JK		 606
607 err = f2fs_issue_discard(sbi, blkaddr, 1);
608 }
609
e90c2d28 610 if (err) {
381722d2 611 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 612 return true;
613 }
614 return false;
37208879
JK		 615}
616
adf4983b 617static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 618 struct cp_control *cpc, struct seg_entry *se,
619 unsigned int start, unsigned int end)
b2955550
JK		 620{
621 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 622 struct discard_entry *new, *last;
623
624 if (!list_empty(head)) {
625 last = list_last_entry(head, struct discard_entry, list);
626 if (START_BLOCK(sbi, cpc->trim_start) + start ==
627 last->blkaddr + last->len) {
628 last->len += end - start;
629 goto done;
630 }
631 }
632
633 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
634 INIT_LIST_HEAD(&new->list);
635 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
636 new->len = end - start;
637 list_add_tail(&new->list, head);
638done:
639 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 640}
641
642static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
643{
b2955550
JK		 644 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
645 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 646 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 647 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
648 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 649 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 650 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 651 unsigned int start = 0, end = -1;
4b2fecc8 652 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 653 int i;
654
a66cdd98 655 if (se->valid_blocks == max_blocks)
b2955550
JK		 656 return;
657
a66cdd98
JK		 658 if (!force) {
659 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 660 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
661 return;
4b2fecc8
JK		 662 }
663
b2955550
JK		 664 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
665 for (i = 0; i < entries; i++)
a66cdd98 666 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 667 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 668
4b2fecc8 669 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 670 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
671 if (start >= max_blocks)
672 break;
673
674 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 675 if (force && start && end != max_blocks
676 && (end - start) < cpc->trim_minlen)
677 continue;
678
a66cdd98 679 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 680 }
681}
682
4b2fecc8
JK		 683void release_discard_addrs(struct f2fs_sb_info *sbi)
684{
685 struct list_head *head = &(SM_I(sbi)->discard_list);
686 struct discard_entry *entry, *this;
687
688 /* drop caches */
689 list_for_each_entry_safe(entry, this, head, list) {
690 list_del(&entry->list);
691 kmem_cache_free(discard_entry_slab, entry);
692 }
693}
694
0a8165d7 695/*
351df4b2
JK		 696 * Should call clear_prefree_segments after checkpoint is done.
697 */
698static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
699{
700 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 701 unsigned int segno;
351df4b2
JK		 702
703 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 704 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 705 __set_test_and_free(sbi, segno);
351df4b2
JK		 706 mutex_unlock(&dirty_i->seglist_lock);
707}
708
836b5a63 709void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 710{
b2955550 711 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 712 struct discard_entry *entry, *this;
351df4b2 713 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 714 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 715 unsigned int start = 0, end = -1;
36abef4e 716 unsigned int secno, start_segno;
c24a0fd6 717 bool force = (cpc->reason == CP_DISCARD);
351df4b2
JK		 718
719 mutex_lock(&dirty_i->seglist_lock);
29e59c14 720
351df4b2 721 while (1) {
29e59c14 722 int i;
7cd8558b
JK		 723 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
724 if (start >= MAIN_SEGS(sbi))
351df4b2 725 break;
7cd8558b
JK		 726 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
727 start + 1);
29e59c14
CL		 728
729 for (i = start; i < end; i++)
730 clear_bit(i, prefree_map);
731
732 dirty_i->nr_dirty[PRE] -= end - start;
733
c24a0fd6 734 if (force || !test_opt(sbi, DISCARD))
29e59c14 735 continue;
351df4b2 736
36abef4e
JK		 737 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
738 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 739 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 740 continue;
741 }
742next:
743 secno = GET_SECNO(sbi, start);
744 start_segno = secno * sbi->segs_per_sec;
745 if (!IS_CURSEC(sbi, secno) &&
746 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
747 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
748 sbi->segs_per_sec << sbi->log_blocks_per_seg);
749
750 start = start_segno + sbi->segs_per_sec;
751 if (start < end)
752 goto next;
351df4b2
JK		 753 }
754 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 755
756 /* send small discards */
2d7b822a 757 list_for_each_entry_safe(entry, this, head, list) {
c24a0fd6 758 if (force && entry->len < cpc->trim_minlen)
836b5a63 759 goto skip;
37208879 760 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 761 cpc->trimmed += entry->len;
836b5a63 762skip:
b2955550
JK		 763 list_del(&entry->list);
764 SM_I(sbi)->nr_discards -= entry->len;
765 kmem_cache_free(discard_entry_slab, entry);
766 }
351df4b2
JK		 767}
768
184a5cd2 769static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 770{
771 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 772
773 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 774 sit_i->dirty_sentries++;
184a5cd2
CY		 775 return false;
776 }
777
778 return true;
351df4b2
JK		 779}
780
781static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
782 unsigned int segno, int modified)
783{
784 struct seg_entry *se = get_seg_entry(sbi, segno);
785 se->type = type;
786 if (modified)
787 __mark_sit_entry_dirty(sbi, segno);
788}
789
790static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
791{
792 struct seg_entry *se;
793 unsigned int segno, offset;
794 long int new_vblocks;
795
796 segno = GET_SEGNO(sbi, blkaddr);
797
798 se = get_seg_entry(sbi, segno);
799 new_vblocks = se->valid_blocks + del;
491c0854 800 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 801
9850cf4a 802 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 803 (new_vblocks > sbi->blocks_per_seg)));
804
805 se->valid_blocks = new_vblocks;
806 se->mtime = get_mtime(sbi);
807 SIT_I(sbi)->max_mtime = se->mtime;
808
809 /* Update valid block bitmap */
810 if (del > 0) {
52aca074 811 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 812 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 813 if (!f2fs_test_and_set_bit(offset, se->discard_map))
814 sbi->discard_blks--;
351df4b2 815 } else {
52aca074 816 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 817 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 818 if (f2fs_test_and_clear_bit(offset, se->discard_map))
819 sbi->discard_blks++;
351df4b2
JK		 820 }
821 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
822 se->ckpt_valid_blocks += del;
823
824 __mark_sit_entry_dirty(sbi, segno);
825
826 /* update total number of valid blocks to be written in ckpt area */
827 SIT_I(sbi)->written_valid_blocks += del;
828
829 if (sbi->segs_per_sec > 1)
830 get_sec_entry(sbi, segno)->valid_blocks += del;
831}
832
5e443818 833void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 834{
5e443818
JK		 835 update_sit_entry(sbi, new, 1);
836 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
837 update_sit_entry(sbi, old, -1);
838
839 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
840 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 841}
842
843void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
844{
845 unsigned int segno = GET_SEGNO(sbi, addr);
846 struct sit_info *sit_i = SIT_I(sbi);
847
9850cf4a 848 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 849 if (addr == NEW_ADDR)
850 return;
851
852 /* add it into sit main buffer */
853 mutex_lock(&sit_i->sentry_lock);
854
855 update_sit_entry(sbi, addr, -1);
856
857 /* add it into dirty seglist */
858 locate_dirty_segment(sbi, segno);
859
860 mutex_unlock(&sit_i->sentry_lock);
861}
862
6e2c64ad
JK		 863bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
864{
865 struct sit_info *sit_i = SIT_I(sbi);
866 unsigned int segno, offset;
867 struct seg_entry *se;
868 bool is_cp = false;
869
870 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
871 return true;
872
873 mutex_lock(&sit_i->sentry_lock);
874
875 segno = GET_SEGNO(sbi, blkaddr);
876 se = get_seg_entry(sbi, segno);
877 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
878
879 if (f2fs_test_bit(offset, se->ckpt_valid_map))
880 is_cp = true;
881
882 mutex_unlock(&sit_i->sentry_lock);
883
884 return is_cp;
885}
886
0a8165d7 887/*
351df4b2
JK		 888 * This function should be resided under the curseg_mutex lock
889 */
890static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 891 struct f2fs_summary *sum)
351df4b2
JK		 892{
893 struct curseg_info *curseg = CURSEG_I(sbi, type);
894 void *addr = curseg->sum_blk;
e79efe3b 895 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 896 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 897}
898
0a8165d7 899/*
351df4b2
JK		 900 * Calculate the number of current summary pages for writing
901 */
3fa06d7b 902int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 903{
351df4b2 904 int valid_sum_count = 0;
9a47938b 905 int i, sum_in_page;
351df4b2
JK		 906
907 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
908 if (sbi->ckpt->alloc_type[i] == SSR)
909 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 910 else {
911 if (for_ra)
912 valid_sum_count += le16_to_cpu(
913 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
914 else
915 valid_sum_count += curseg_blkoff(sbi, i);
916 }
351df4b2
JK		 917 }
918
09cbfeaf 919 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 920 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
921 if (valid_sum_count <= sum_in_page)
351df4b2 922 return 1;
9a47938b 923 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 924 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 925 return 2;
926 return 3;
927}
928
0a8165d7 929/*
351df4b2
JK		 930 * Caller should put this summary page
931 */
932struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
933{
934 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
935}
936
381722d2 937void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 938{
939 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 940 void *dst = page_address(page);
941
942 if (src)
09cbfeaf 943 memcpy(dst, src, PAGE_SIZE);
381722d2 944 else
09cbfeaf 945 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 946 set_page_dirty(page);
947 f2fs_put_page(page, 1);
948}
949
381722d2
CY		 950static void write_sum_page(struct f2fs_sb_info *sbi,
951 struct f2fs_summary_block *sum_blk, block_t blk_addr)
952{
953 update_meta_page(sbi, (void *)sum_blk, blk_addr);
954}
955
b7ad7512
CY		 956static void write_current_sum_page(struct f2fs_sb_info *sbi,
957 int type, block_t blk_addr)
958{
959 struct curseg_info *curseg = CURSEG_I(sbi, type);
960 struct page *page = grab_meta_page(sbi, blk_addr);
961 struct f2fs_summary_block *src = curseg->sum_blk;
962 struct f2fs_summary_block *dst;
963
964 dst = (struct f2fs_summary_block *)page_address(page);
965
966 mutex_lock(&curseg->curseg_mutex);
967
968 down_read(&curseg->journal_rwsem);
969 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
970 up_read(&curseg->journal_rwsem);
971
972 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
973 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
974
975 mutex_unlock(&curseg->curseg_mutex);
976
977 set_page_dirty(page);
978 f2fs_put_page(page, 1);
979}
980
60374688
JK		 981static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
982{
983 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 984 unsigned int segno = curseg->segno + 1;
60374688
JK		 985 struct free_segmap_info *free_i = FREE_I(sbi);
986
7cd8558b 987 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 988 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 989 return 0;
990}
991
0a8165d7 992/*
351df4b2
JK		 993 * Find a new segment from the free segments bitmap to right order
994 * This function should be returned with success, otherwise BUG
995 */
996static void get_new_segment(struct f2fs_sb_info *sbi,
997 unsigned int *newseg, bool new_sec, int dir)
998{
999 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1000 unsigned int segno, secno, zoneno;
7cd8558b 1001 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1002 unsigned int hint = *newseg / sbi->segs_per_sec;
1003 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1004 unsigned int left_start = hint;
1005 bool init = true;
1006 int go_left = 0;
1007 int i;
1008
1a118ccf 1009 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1010
1011 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1012 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1013 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1014 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1015 goto got_it;
1016 }
1017find_other_zone:
7cd8558b
JK		 1018 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1019 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1020 if (dir == ALLOC_RIGHT) {
1021 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1022 MAIN_SECS(sbi), 0);
1023 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1024 } else {
1025 go_left = 1;
1026 left_start = hint - 1;
1027 }
1028 }
1029 if (go_left == 0)
1030 goto skip_left;
1031
1032 while (test_bit(left_start, free_i->free_secmap)) {
1033 if (left_start > 0) {
1034 left_start--;
1035 continue;
1036 }
1037 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1038 MAIN_SECS(sbi), 0);
1039 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1040 break;
1041 }
1042 secno = left_start;
1043skip_left:
1044 hint = secno;
1045 segno = secno * sbi->segs_per_sec;
1046 zoneno = secno / sbi->secs_per_zone;
1047
1048 /* give up on finding another zone */
1049 if (!init)
1050 goto got_it;
1051 if (sbi->secs_per_zone == 1)
1052 goto got_it;
1053 if (zoneno == old_zoneno)
1054 goto got_it;
1055 if (dir == ALLOC_LEFT) {
1056 if (!go_left && zoneno + 1 >= total_zones)
1057 goto got_it;
1058 if (go_left && zoneno == 0)
1059 goto got_it;
1060 }
1061 for (i = 0; i < NR_CURSEG_TYPE; i++)
1062 if (CURSEG_I(sbi, i)->zone == zoneno)
1063 break;
1064
1065 if (i < NR_CURSEG_TYPE) {
1066 /* zone is in user, try another */
1067 if (go_left)
1068 hint = zoneno * sbi->secs_per_zone - 1;
1069 else if (zoneno + 1 >= total_zones)
1070 hint = 0;
1071 else
1072 hint = (zoneno + 1) * sbi->secs_per_zone;
1073 init = false;
1074 goto find_other_zone;
1075 }
1076got_it:
1077 /* set it as dirty segment in free segmap */
9850cf4a 1078 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1079 __set_inuse(sbi, segno);
1080 *newseg = segno;
1a118ccf 1081 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1082}
1083
1084static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1085{
1086 struct curseg_info *curseg = CURSEG_I(sbi, type);
1087 struct summary_footer *sum_footer;
1088
1089 curseg->segno = curseg->next_segno;
1090 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1091 curseg->next_blkoff = 0;
1092 curseg->next_segno = NULL_SEGNO;
1093
1094 sum_footer = &(curseg->sum_blk->footer);
1095 memset(sum_footer, 0, sizeof(struct summary_footer));
1096 if (IS_DATASEG(type))
1097 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1098 if (IS_NODESEG(type))
1099 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1100 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1101}
1102
0a8165d7 1103/*
351df4b2
JK		 1104 * Allocate a current working segment.
1105 * This function always allocates a free segment in LFS manner.
1106 */
1107static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1108{
1109 struct curseg_info *curseg = CURSEG_I(sbi, type);
1110 unsigned int segno = curseg->segno;
1111 int dir = ALLOC_LEFT;
1112
1113 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1114 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1115 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1116 dir = ALLOC_RIGHT;
1117
1118 if (test_opt(sbi, NOHEAP))
1119 dir = ALLOC_RIGHT;
1120
1121 get_new_segment(sbi, &segno, new_sec, dir);
1122 curseg->next_segno = segno;
1123 reset_curseg(sbi, type, 1);
1124 curseg->alloc_type = LFS;
1125}
1126
1127static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1128 struct curseg_info *seg, block_t start)
1129{
1130 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1131 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1132 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1133 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1134 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1135 int i, pos;
1136
1137 for (i = 0; i < entries; i++)
1138 target_map[i] = ckpt_map[i] | cur_map[i];
1139
1140 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1141
1142 seg->next_blkoff = pos;
351df4b2
JK		 1143}
1144
0a8165d7 1145/*
351df4b2
JK		 1146 * If a segment is written by LFS manner, next block offset is just obtained
1147 * by increasing the current block offset. However, if a segment is written by
1148 * SSR manner, next block offset obtained by calling __next_free_blkoff
1149 */
1150static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1151 struct curseg_info *seg)
1152{
1153 if (seg->alloc_type == SSR)
1154 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1155 else
1156 seg->next_blkoff++;
1157}
1158
0a8165d7 1159/*
e1c42045 1160 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1161 * manner, so it should recover the existing segment information of valid blocks
1162 */
1163static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1164{
1165 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1166 struct curseg_info *curseg = CURSEG_I(sbi, type);
1167 unsigned int new_segno = curseg->next_segno;
1168 struct f2fs_summary_block *sum_node;
1169 struct page *sum_page;
1170
1171 write_sum_page(sbi, curseg->sum_blk,
1172 GET_SUM_BLOCK(sbi, curseg->segno));
1173 __set_test_and_inuse(sbi, new_segno);
1174
1175 mutex_lock(&dirty_i->seglist_lock);
1176 __remove_dirty_segment(sbi, new_segno, PRE);
1177 __remove_dirty_segment(sbi, new_segno, DIRTY);
1178 mutex_unlock(&dirty_i->seglist_lock);
1179
1180 reset_curseg(sbi, type, 1);
1181 curseg->alloc_type = SSR;
1182 __next_free_blkoff(sbi, curseg, 0);
1183
1184 if (reuse) {
1185 sum_page = get_sum_page(sbi, new_segno);
1186 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1187 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1188 f2fs_put_page(sum_page, 1);
1189 }
1190}
1191
43727527
JK		 1192static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1193{
1194 struct curseg_info *curseg = CURSEG_I(sbi, type);
1195 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1196
1197 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1198 return v_ops->get_victim(sbi,
1199 &(curseg)->next_segno, BG_GC, type, SSR);
1200
1201 /* For data segments, let's do SSR more intensively */
1202 for (; type >= CURSEG_HOT_DATA; type--)
1203 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1204 BG_GC, type, SSR))
1205 return 1;
1206 return 0;
1207}
1208
351df4b2
JK		 1209/*
1210 * flush out current segment and replace it with new segment
1211 * This function should be returned with success, otherwise BUG
1212 */
1213static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1214 int type, bool force)
1215{
1216 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1217
7b405275 1218 if (force)
351df4b2 1219 new_curseg(sbi, type, true);
7b405275 1220 else if (type == CURSEG_WARM_NODE)
351df4b2 1221 new_curseg(sbi, type, false);
60374688
JK		 1222 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1223 new_curseg(sbi, type, false);
351df4b2
JK		 1224 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1225 change_curseg(sbi, type, true);
1226 else
1227 new_curseg(sbi, type, false);
dcdfff65
JK		 1228
1229 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1230}
1231
38aa0889
JK		 1232static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1233{
1234 struct curseg_info *curseg = CURSEG_I(sbi, type);
1235 unsigned int old_segno;
1236
1237 old_segno = curseg->segno;
1238 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1239 locate_dirty_segment(sbi, old_segno);
1240}
1241
351df4b2
JK		 1242void allocate_new_segments(struct f2fs_sb_info *sbi)
1243{
351df4b2
JK		 1244 int i;
1245
36abef4e
JK		 1246 if (test_opt(sbi, LFS))
1247 return;
1248
38aa0889
JK		 1249 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1250 __allocate_new_segments(sbi, i);
351df4b2
JK		 1251}
1252
1253static const struct segment_allocation default_salloc_ops = {
1254 .allocate_segment = allocate_segment_by_default,
1255};
1256
4b2fecc8
JK		 1257int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1258{
f7ef9b83
JK		 1259 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1260 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1261 unsigned int start_segno, end_segno;
1262 struct cp_control cpc;
c34f42e2 1263 int err = 0;
4b2fecc8 1264
836b5a63 1265 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1266 return -EINVAL;
1267
9bd27ae4 1268 cpc.trimmed = 0;
7cd8558b 1269 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1270 goto out;
1271
1272 /* start/end segment number in main_area */
7cd8558b
JK		 1273 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1274 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1275 GET_SEGNO(sbi, end);
4b2fecc8 1276 cpc.reason = CP_DISCARD;
836b5a63 1277 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1278
1279 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1280 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1281 cpc.trim_start = start_segno;
a66cdd98
JK		 1282
1283 if (sbi->discard_blks == 0)
1284 break;
1285 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1286 cpc.trim_end = end_segno;
1287 else
1288 cpc.trim_end = min_t(unsigned int,
1289 rounddown(start_segno +
bba681cb
JK		 1290 BATCHED_TRIM_SEGMENTS(sbi),
1291 sbi->segs_per_sec) - 1, end_segno);
1292
1293 mutex_lock(&sbi->gc_mutex);
c34f42e2 1294 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1295 mutex_unlock(&sbi->gc_mutex);
1296 }
4b2fecc8 1297out:
f7ef9b83 1298 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1299 return err;
4b2fecc8
JK		 1300}
1301
351df4b2
JK		 1302static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1303{
1304 struct curseg_info *curseg = CURSEG_I(sbi, type);
1305 if (curseg->next_blkoff < sbi->blocks_per_seg)
1306 return true;
1307 return false;
1308}
1309
1310static int __get_segment_type_2(struct page *page, enum page_type p_type)
1311{
1312 if (p_type == DATA)
1313 return CURSEG_HOT_DATA;
1314 else
1315 return CURSEG_HOT_NODE;
1316}
1317
1318static int __get_segment_type_4(struct page *page, enum page_type p_type)
1319{
1320 if (p_type == DATA) {
1321 struct inode *inode = page->mapping->host;
1322
1323 if (S_ISDIR(inode->i_mode))
1324 return CURSEG_HOT_DATA;
1325 else
1326 return CURSEG_COLD_DATA;
1327 } else {
a344b9fd
JK		 1328 if (IS_DNODE(page) && is_cold_node(page))
1329 return CURSEG_WARM_NODE;
351df4b2
JK		 1330 else
1331 return CURSEG_COLD_NODE;
1332 }
1333}
1334
1335static int __get_segment_type_6(struct page *page, enum page_type p_type)
1336{
1337 if (p_type == DATA) {
1338 struct inode *inode = page->mapping->host;
1339
1340 if (S_ISDIR(inode->i_mode))
1341 return CURSEG_HOT_DATA;
354a3399 1342 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1343 return CURSEG_COLD_DATA;
1344 else
1345 return CURSEG_WARM_DATA;
1346 } else {
1347 if (IS_DNODE(page))
1348 return is_cold_node(page) ? CURSEG_WARM_NODE :
1349 CURSEG_HOT_NODE;
1350 else
1351 return CURSEG_COLD_NODE;
1352 }
1353}
1354
1355static int __get_segment_type(struct page *page, enum page_type p_type)
1356{
4081363f 1357 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1358 case 2:
1359 return __get_segment_type_2(page, p_type);
1360 case 4:
1361 return __get_segment_type_4(page, p_type);
351df4b2 1362 }
12a67146 1363 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1364 f2fs_bug_on(F2FS_P_SB(page),
1365 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1366 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1367}
1368
bfad7c2d
JK		 1369void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1370 block_t old_blkaddr, block_t *new_blkaddr,
1371 struct f2fs_summary *sum, int type)
351df4b2
JK		 1372{
1373 struct sit_info *sit_i = SIT_I(sbi);
1374 struct curseg_info *curseg;
38aa0889
JK		 1375 bool direct_io = (type == CURSEG_DIRECT_IO);
1376
1377 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1378
351df4b2
JK		 1379 curseg = CURSEG_I(sbi, type);
1380
1381 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1382 mutex_lock(&sit_i->sentry_lock);
351df4b2 1383
38aa0889 1384 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1385 if (direct_io && curseg->next_blkoff &&
1386 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1387 __allocate_new_segments(sbi, type);
1388
351df4b2 1389 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1390
1391 /*
1392 * __add_sum_entry should be resided under the curseg_mutex
1393 * because, this function updates a summary entry in the
1394 * current summary block.
1395 */
e79efe3b 1396 __add_sum_entry(sbi, type, sum);
351df4b2 1397
351df4b2 1398 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1399
1400 stat_inc_block_count(sbi, curseg);
351df4b2 1401
5e443818
JK		 1402 if (!__has_curseg_space(sbi, type))
1403 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1404 /*
1405 * SIT information should be updated before segment allocation,
1406 * since SSR needs latest valid block information.
1407 */
1408 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1409
351df4b2
JK		 1410 mutex_unlock(&sit_i->sentry_lock);
1411
bfad7c2d 1412 if (page && IS_NODESEG(type))
351df4b2
JK		 1413 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1414
bfad7c2d
JK		 1415 mutex_unlock(&curseg->curseg_mutex);
1416}
1417
05ca3632 1418static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1419{
05ca3632 1420 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1421
7dfeaa32
JK		 1422 if (fio->type == NODE || fio->type == DATA)
1423 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1424
7a9d7548
CY		 1425 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1426 &fio->new_blkaddr, sum, type);
bfad7c2d 1427
351df4b2 1428 /* writeout dirty page into bdev */
05ca3632 1429 f2fs_submit_page_mbio(fio);
7dfeaa32
JK		 1430
1431 if (fio->type == NODE || fio->type == DATA)
1432 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1433}
1434
577e3495 1435void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1436{
458e6197 1437 struct f2fs_io_info fio = {
05ca3632 1438 .sbi = sbi,
458e6197 1439 .type = META,
cf04e8eb 1440 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1441 .old_blkaddr = page->index,
1442 .new_blkaddr = page->index,
05ca3632 1443 .page = page,
4375a336 1444 .encrypted_page = NULL,
458e6197
JK		 1445 };
1446
2b947003
CY		 1447 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1448 fio.rw &= ~REQ_META;
1449
351df4b2 1450 set_page_writeback(page);
05ca3632 1451 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1452}
1453
05ca3632 1454void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1455{
1456 struct f2fs_summary sum;
05ca3632 1457
351df4b2 1458 set_summary(&sum, nid, 0, 0);
05ca3632 1459 do_write_page(&sum, fio);
351df4b2
JK		 1460}
1461
05ca3632 1462void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1463{
05ca3632 1464 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1465 struct f2fs_summary sum;
1466 struct node_info ni;
1467
9850cf4a 1468 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1469 get_node_info(sbi, dn->nid, &ni);
1470 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1471 do_write_page(&sum, fio);
f28b3434 1472 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1473}
1474
05ca3632 1475void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1476{
7a9d7548 1477 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1478 stat_inc_inplace_blocks(fio->sbi);
1479 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1480}
1481
4356e48e 1482void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1483 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1484 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1485{
1486 struct sit_info *sit_i = SIT_I(sbi);
1487 struct curseg_info *curseg;
1488 unsigned int segno, old_cursegno;
1489 struct seg_entry *se;
1490 int type;
19f106bc 1491 unsigned short old_blkoff;
351df4b2
JK		 1492
1493 segno = GET_SEGNO(sbi, new_blkaddr);
1494 se = get_seg_entry(sbi, segno);
1495 type = se->type;
1496
19f106bc
CY		 1497 if (!recover_curseg) {
1498 /* for recovery flow */
1499 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1500 if (old_blkaddr == NULL_ADDR)
1501 type = CURSEG_COLD_DATA;
1502 else
1503 type = CURSEG_WARM_DATA;
1504 }
1505 } else {
1506 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1507 type = CURSEG_WARM_DATA;
1508 }
19f106bc 1509
351df4b2
JK		 1510 curseg = CURSEG_I(sbi, type);
1511
1512 mutex_lock(&curseg->curseg_mutex);
1513 mutex_lock(&sit_i->sentry_lock);
1514
1515 old_cursegno = curseg->segno;
19f106bc 1516 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1517
1518 /* change the current segment */
1519 if (segno != curseg->segno) {
1520 curseg->next_segno = segno;
1521 change_curseg(sbi, type, true);
1522 }
1523
491c0854 1524 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1525 __add_sum_entry(sbi, type, sum);
351df4b2 1526
28bc106b 1527 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1528 update_sit_entry(sbi, new_blkaddr, 1);
1529 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1530 update_sit_entry(sbi, old_blkaddr, -1);
1531
1532 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1533 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1534
351df4b2 1535 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1536
19f106bc
CY		 1537 if (recover_curseg) {
1538 if (old_cursegno != curseg->segno) {
1539 curseg->next_segno = old_cursegno;
1540 change_curseg(sbi, type, true);
1541 }
1542 curseg->next_blkoff = old_blkoff;
1543 }
1544
351df4b2
JK		 1545 mutex_unlock(&sit_i->sentry_lock);
1546 mutex_unlock(&curseg->curseg_mutex);
1547}
1548
528e3459
CY		 1549void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1550 block_t old_addr, block_t new_addr,
28bc106b
CY		 1551 unsigned char version, bool recover_curseg,
1552 bool recover_newaddr)
528e3459
CY		 1553{
1554 struct f2fs_summary sum;
1555
1556 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1557
28bc106b
CY		 1558 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1559 recover_curseg, recover_newaddr);
528e3459 1560
f28b3434 1561 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1562}
1563
93dfe2ac 1564void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1565 enum page_type type, bool ordered)
93dfe2ac 1566{
93dfe2ac 1567 if (PageWriteback(page)) {
4081363f
JK		 1568 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1569
0c3a5797 1570 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1571 if (ordered)
1572 wait_on_page_writeback(page);
1573 else
1574 wait_for_stable_page(page);
93dfe2ac
JK		 1575 }
1576}
1577
08b39fbd
CY		 1578void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1579 block_t blkaddr)
1580{
1581 struct page *cpage;
1582
1583 if (blkaddr == NEW_ADDR)
1584 return;
1585
1586 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1587
1588 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1589 if (cpage) {
fec1d657 1590 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1591 f2fs_put_page(cpage, 1);
1592 }
1593}
1594
351df4b2
JK		 1595static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1596{
1597 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1598 struct curseg_info *seg_i;
1599 unsigned char *kaddr;
1600 struct page *page;
1601 block_t start;
1602 int i, j, offset;
1603
1604 start = start_sum_block(sbi);
1605
1606 page = get_meta_page(sbi, start++);
1607 kaddr = (unsigned char *)page_address(page);
1608
1609 /* Step 1: restore nat cache */
1610 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1611 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1612
1613 /* Step 2: restore sit cache */
1614 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1615 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1616 offset = 2 * SUM_JOURNAL_SIZE;
1617
1618 /* Step 3: restore summary entries */
1619 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1620 unsigned short blk_off;
1621 unsigned int segno;
1622
1623 seg_i = CURSEG_I(sbi, i);
1624 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1625 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1626 seg_i->next_segno = segno;
1627 reset_curseg(sbi, i, 0);
1628 seg_i->alloc_type = ckpt->alloc_type[i];
1629 seg_i->next_blkoff = blk_off;
1630
1631 if (seg_i->alloc_type == SSR)
1632 blk_off = sbi->blocks_per_seg;
1633
1634 for (j = 0; j < blk_off; j++) {
1635 struct f2fs_summary *s;
1636 s = (struct f2fs_summary *)(kaddr + offset);
1637 seg_i->sum_blk->entries[j] = *s;
1638 offset += SUMMARY_SIZE;
09cbfeaf 1639 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1640 SUM_FOOTER_SIZE)
1641 continue;
1642
1643 f2fs_put_page(page, 1);
1644 page = NULL;
1645
1646 page = get_meta_page(sbi, start++);
1647 kaddr = (unsigned char *)page_address(page);
1648 offset = 0;
1649 }
1650 }
1651 f2fs_put_page(page, 1);
1652 return 0;
1653}
1654
1655static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1656{
1657 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1658 struct f2fs_summary_block *sum;
1659 struct curseg_info *curseg;
1660 struct page *new;
1661 unsigned short blk_off;
1662 unsigned int segno = 0;
1663 block_t blk_addr = 0;
1664
1665 /* get segment number and block addr */
1666 if (IS_DATASEG(type)) {
1667 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1668 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1669 CURSEG_HOT_DATA]);
119ee914 1670 if (__exist_node_summaries(sbi))
351df4b2
JK		 1671 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1672 else
1673 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1674 } else {
1675 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1676 CURSEG_HOT_NODE]);
1677 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1678 CURSEG_HOT_NODE]);
119ee914 1679 if (__exist_node_summaries(sbi))
351df4b2
JK		 1680 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1681 type - CURSEG_HOT_NODE);
1682 else
1683 blk_addr = GET_SUM_BLOCK(sbi, segno);
1684 }
1685
1686 new = get_meta_page(sbi, blk_addr);
1687 sum = (struct f2fs_summary_block *)page_address(new);
1688
1689 if (IS_NODESEG(type)) {
119ee914 1690 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1691 struct f2fs_summary *ns = &sum->entries[0];
1692 int i;
1693 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1694 ns->version = 0;
1695 ns->ofs_in_node = 0;
1696 }
1697 } else {
d653788a
GZ		 1698 int err;
1699
1700 err = restore_node_summary(sbi, segno, sum);
1701 if (err) {
351df4b2 1702 f2fs_put_page(new, 1);
d653788a 1703 return err;
351df4b2
JK		 1704 }
1705 }
1706 }
1707
1708 /* set uncompleted segment to curseg */
1709 curseg = CURSEG_I(sbi, type);
1710 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1711
1712 /* update journal info */
1713 down_write(&curseg->journal_rwsem);
1714 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1715 up_write(&curseg->journal_rwsem);
1716
1717 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1718 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1719 curseg->next_segno = segno;
1720 reset_curseg(sbi, type, 0);
1721 curseg->alloc_type = ckpt->alloc_type[type];
1722 curseg->next_blkoff = blk_off;
1723 mutex_unlock(&curseg->curseg_mutex);
1724 f2fs_put_page(new, 1);
1725 return 0;
1726}
1727
1728static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1729{
1730 int type = CURSEG_HOT_DATA;
e4fc5fbf 1731 int err;
351df4b2 1732
25ca923b 1733 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1734 int npages = npages_for_summary_flush(sbi, true);
1735
1736 if (npages >= 2)
1737 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1738 META_CP, true);
3fa06d7b 1739
351df4b2
JK		 1740 /* restore for compacted data summary */
1741 if (read_compacted_summaries(sbi))
1742 return -EINVAL;
1743 type = CURSEG_HOT_NODE;
1744 }
1745
119ee914 1746 if (__exist_node_summaries(sbi))
3fa06d7b 1747 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1748 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1749
e4fc5fbf
CY		 1750 for (; type <= CURSEG_COLD_NODE; type++) {
1751 err = read_normal_summaries(sbi, type);
1752 if (err)
1753 return err;
1754 }
1755
351df4b2
JK		 1756 return 0;
1757}
1758
1759static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1760{
1761 struct page *page;
1762 unsigned char *kaddr;
1763 struct f2fs_summary *summary;
1764 struct curseg_info *seg_i;
1765 int written_size = 0;
1766 int i, j;
1767
1768 page = grab_meta_page(sbi, blkaddr++);
1769 kaddr = (unsigned char *)page_address(page);
1770
1771 /* Step 1: write nat cache */
1772 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1773 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1774 written_size += SUM_JOURNAL_SIZE;
1775
1776 /* Step 2: write sit cache */
1777 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1778 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1779 written_size += SUM_JOURNAL_SIZE;
1780
351df4b2
JK		 1781 /* Step 3: write summary entries */
1782 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1783 unsigned short blkoff;
1784 seg_i = CURSEG_I(sbi, i);
1785 if (sbi->ckpt->alloc_type[i] == SSR)
1786 blkoff = sbi->blocks_per_seg;
1787 else
1788 blkoff = curseg_blkoff(sbi, i);
1789
1790 for (j = 0; j < blkoff; j++) {
1791 if (!page) {
1792 page = grab_meta_page(sbi, blkaddr++);
1793 kaddr = (unsigned char *)page_address(page);
1794 written_size = 0;
1795 }
1796 summary = (struct f2fs_summary *)(kaddr + written_size);
1797 *summary = seg_i->sum_blk->entries[j];
1798 written_size += SUMMARY_SIZE;
351df4b2 1799
09cbfeaf 1800 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1801 SUM_FOOTER_SIZE)
1802 continue;
1803
e8d61a74 1804 set_page_dirty(page);
351df4b2
JK		 1805 f2fs_put_page(page, 1);
1806 page = NULL;
1807 }
1808 }
e8d61a74
CY		 1809 if (page) {
1810 set_page_dirty(page);
351df4b2 1811 f2fs_put_page(page, 1);
e8d61a74 1812 }
351df4b2
JK		 1813}
1814
1815static void write_normal_summaries(struct f2fs_sb_info *sbi,
1816 block_t blkaddr, int type)
1817{
1818 int i, end;
1819 if (IS_DATASEG(type))
1820 end = type + NR_CURSEG_DATA_TYPE;
1821 else
1822 end = type + NR_CURSEG_NODE_TYPE;
1823
b7ad7512
CY		 1824 for (i = type; i < end; i++)
1825 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1826}
1827
1828void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1829{
25ca923b 1830 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1831 write_compacted_summaries(sbi, start_blk);
1832 else
1833 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1834}
1835
1836void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1837{
119ee914 1838 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1839}
1840
dfc08a12 1841int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1842 unsigned int val, int alloc)
1843{
1844 int i;
1845
1846 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1847 for (i = 0; i < nats_in_cursum(journal); i++) {
1848 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1849 return i;
1850 }
dfc08a12
CY		 1851 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1852 return update_nats_in_cursum(journal, 1);
351df4b2 1853 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1854 for (i = 0; i < sits_in_cursum(journal); i++)
1855 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1856 return i;
dfc08a12
CY		 1857 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1858 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1859 }
1860 return -1;
1861}
1862
1863static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1864 unsigned int segno)
1865{
2cc22186 1866 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1867}
1868
1869static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1870 unsigned int start)
1871{
1872 struct sit_info *sit_i = SIT_I(sbi);
1873 struct page *src_page, *dst_page;
1874 pgoff_t src_off, dst_off;
1875 void *src_addr, *dst_addr;
1876
1877 src_off = current_sit_addr(sbi, start);
1878 dst_off = next_sit_addr(sbi, src_off);
1879
1880 /* get current sit block page without lock */
1881 src_page = get_meta_page(sbi, src_off);
1882 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1883 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1884
1885 src_addr = page_address(src_page);
1886 dst_addr = page_address(dst_page);
09cbfeaf 1887 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1888
1889 set_page_dirty(dst_page);
1890 f2fs_put_page(src_page, 1);
1891
1892 set_to_next_sit(sit_i, start);
1893
1894 return dst_page;
1895}
1896
184a5cd2
CY		 1897static struct sit_entry_set *grab_sit_entry_set(void)
1898{
1899 struct sit_entry_set *ses =
80c54505 1900 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1901
1902 ses->entry_cnt = 0;
1903 INIT_LIST_HEAD(&ses->set_list);
1904 return ses;
1905}
1906
1907static void release_sit_entry_set(struct sit_entry_set *ses)
1908{
1909 list_del(&ses->set_list);
1910 kmem_cache_free(sit_entry_set_slab, ses);
1911}
1912
1913static void adjust_sit_entry_set(struct sit_entry_set *ses,
1914 struct list_head *head)
1915{
1916 struct sit_entry_set *next = ses;
1917
1918 if (list_is_last(&ses->set_list, head))
1919 return;
1920
1921 list_for_each_entry_continue(next, head, set_list)
1922 if (ses->entry_cnt <= next->entry_cnt)
1923 break;
1924
1925 list_move_tail(&ses->set_list, &next->set_list);
1926}
1927
1928static void add_sit_entry(unsigned int segno, struct list_head *head)
1929{
1930 struct sit_entry_set *ses;
1931 unsigned int start_segno = START_SEGNO(segno);
1932
1933 list_for_each_entry(ses, head, set_list) {
1934 if (ses->start_segno == start_segno) {
1935 ses->entry_cnt++;
1936 adjust_sit_entry_set(ses, head);
1937 return;
1938 }
1939 }
1940
1941 ses = grab_sit_entry_set();
1942
1943 ses->start_segno = start_segno;
1944 ses->entry_cnt++;
1945 list_add(&ses->set_list, head);
1946}
1947
1948static void add_sits_in_set(struct f2fs_sb_info *sbi)
1949{
1950 struct f2fs_sm_info *sm_info = SM_I(sbi);
1951 struct list_head *set_list = &sm_info->sit_entry_set;
1952 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1953 unsigned int segno;
1954
7cd8558b 1955 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1956 add_sit_entry(segno, set_list);
1957}
1958
1959static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1960{
1961 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1962 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1963 int i;
1964
b7ad7512 1965 down_write(&curseg->journal_rwsem);
dfc08a12 1966 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1967 unsigned int segno;
1968 bool dirtied;
1969
dfc08a12 1970 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1971 dirtied = __mark_sit_entry_dirty(sbi, segno);
1972
1973 if (!dirtied)
1974 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1975 }
dfc08a12 1976 update_sits_in_cursum(journal, -i);
b7ad7512 1977 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1978}
1979
0a8165d7 1980/*
351df4b2
JK		 1981 * CP calls this function, which flushes SIT entries including sit_journal,
1982 * and moves prefree segs to free segs.
1983 */
4b2fecc8 1984void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1985{
1986 struct sit_info *sit_i = SIT_I(sbi);
1987 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1988 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1989 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 1990 struct sit_entry_set *ses, *tmp;
1991 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1992 bool to_journal = true;
4b2fecc8 1993 struct seg_entry *se;
351df4b2 1994
351df4b2
JK		 1995 mutex_lock(&sit_i->sentry_lock);
1996
2b11a74b
WL		 1997 if (!sit_i->dirty_sentries)
1998 goto out;
1999
351df4b2 2000 /*
184a5cd2
CY		 2001 * add and account sit entries of dirty bitmap in sit entry
2002 * set temporarily
351df4b2 2003 */
184a5cd2 2004 add_sits_in_set(sbi);
351df4b2 2005
184a5cd2
CY		 2006 /*
2007 * if there are no enough space in journal to store dirty sit
2008 * entries, remove all entries from journal and add and account
2009 * them in sit entry set.
2010 */
dfc08a12 2011 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2012 remove_sits_in_journal(sbi);
b2955550 2013
184a5cd2
CY		 2014 /*
2015 * there are two steps to flush sit entries:
2016 * #1, flush sit entries to journal in current cold data summary block.
2017 * #2, flush sit entries to sit page.
2018 */
2019 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2020 struct page *page = NULL;
184a5cd2
CY		 2021 struct f2fs_sit_block *raw_sit = NULL;
2022 unsigned int start_segno = ses->start_segno;
2023 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2024 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2025 unsigned int segno = start_segno;
2026
2027 if (to_journal &&
dfc08a12 2028 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2029 to_journal = false;
2030
b7ad7512
CY		 2031 if (to_journal) {
2032 down_write(&curseg->journal_rwsem);
2033 } else {
184a5cd2
CY		 2034 page = get_next_sit_page(sbi, start_segno);
2035 raw_sit = page_address(page);
351df4b2 2036 }
351df4b2 2037
184a5cd2
CY		 2038 /* flush dirty sit entries in region of current sit set */
2039 for_each_set_bit_from(segno, bitmap, end) {
2040 int offset, sit_offset;
4b2fecc8
JK		 2041
2042 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2043
2044 /* add discard candidates */
d7bc2484 2045 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2046 cpc->trim_start = segno;
2047 add_discard_addrs(sbi, cpc);
2048 }
184a5cd2
CY		 2049
2050 if (to_journal) {
dfc08a12 2051 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2052 SIT_JOURNAL, segno, 1);
2053 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2054 segno_in_journal(journal, offset) =
184a5cd2
CY		 2055 cpu_to_le32(segno);
2056 seg_info_to_raw_sit(se,
dfc08a12 2057 &sit_in_journal(journal, offset));
184a5cd2
CY		 2058 } else {
2059 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2060 seg_info_to_raw_sit(se,
2061 &raw_sit->entries[sit_offset]);
2062 }
351df4b2 2063
184a5cd2
CY		 2064 __clear_bit(segno, bitmap);
2065 sit_i->dirty_sentries--;
2066 ses->entry_cnt--;
351df4b2
JK		 2067 }
2068
b7ad7512
CY		 2069 if (to_journal)
2070 up_write(&curseg->journal_rwsem);
2071 else
184a5cd2
CY		 2072 f2fs_put_page(page, 1);
2073
2074 f2fs_bug_on(sbi, ses->entry_cnt);
2075 release_sit_entry_set(ses);
351df4b2 2076 }
184a5cd2
CY		 2077
2078 f2fs_bug_on(sbi, !list_empty(head));
2079 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2080out:
4b2fecc8
JK		 2081 if (cpc->reason == CP_DISCARD) {
2082 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2083 add_discard_addrs(sbi, cpc);
2084 }
351df4b2 2085 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2086
351df4b2
JK		 2087 set_prefree_as_free_segments(sbi);
2088}
2089
2090static int build_sit_info(struct f2fs_sb_info *sbi)
2091{
2092 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2093 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2094 struct sit_info *sit_i;
2095 unsigned int sit_segs, start;
2096 char *src_bitmap, *dst_bitmap;
2097 unsigned int bitmap_size;
2098
2099 /* allocate memory for SIT information */
2100 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2101 if (!sit_i)
2102 return -ENOMEM;
2103
2104 SM_I(sbi)->sit_info = sit_i;
2105
39307a8e
JK		 2106 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2107 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2108 if (!sit_i->sentries)
2109 return -ENOMEM;
2110
7cd8558b 2111 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2112 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2113 if (!sit_i->dirty_sentries_bitmap)
2114 return -ENOMEM;
2115
7cd8558b 2116 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2117 sit_i->sentries[start].cur_valid_map
2118 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2119 sit_i->sentries[start].ckpt_valid_map
2120 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2121 sit_i->sentries[start].discard_map
2122 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2123 if (!sit_i->sentries[start].cur_valid_map ||
2124 !sit_i->sentries[start].ckpt_valid_map ||
2125 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2126 return -ENOMEM;
2127 }
2128
60a3b782
JK		 2129 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2130 if (!sit_i->tmp_map)
2131 return -ENOMEM;
2132
351df4b2 2133 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2134 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2135 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2136 if (!sit_i->sec_entries)
2137 return -ENOMEM;
2138 }
2139
2140 /* get information related with SIT */
2141 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2142
2143 /* setup SIT bitmap from ckeckpoint pack */
2144 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2145 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2146
79b5793b 2147 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2148 if (!dst_bitmap)
2149 return -ENOMEM;
351df4b2
JK		 2150
2151 /* init SIT information */
2152 sit_i->s_ops = &default_salloc_ops;
2153
2154 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2155 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2156 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2157 sit_i->sit_bitmap = dst_bitmap;
2158 sit_i->bitmap_size = bitmap_size;
2159 sit_i->dirty_sentries = 0;
2160 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2161 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2162 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2163 mutex_init(&sit_i->sentry_lock);
2164 return 0;
2165}
2166
2167static int build_free_segmap(struct f2fs_sb_info *sbi)
2168{
351df4b2
JK		 2169 struct free_segmap_info *free_i;
2170 unsigned int bitmap_size, sec_bitmap_size;
2171
2172 /* allocate memory for free segmap information */
2173 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2174 if (!free_i)
2175 return -ENOMEM;
2176
2177 SM_I(sbi)->free_info = free_i;
2178
7cd8558b 2179 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2180 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2181 if (!free_i->free_segmap)
2182 return -ENOMEM;
2183
7cd8558b 2184 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2185 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2186 if (!free_i->free_secmap)
2187 return -ENOMEM;
2188
2189 /* set all segments as dirty temporarily */
2190 memset(free_i->free_segmap, 0xff, bitmap_size);
2191 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2192
2193 /* init free segmap information */
7cd8558b 2194 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2195 free_i->free_segments = 0;
2196 free_i->free_sections = 0;
1a118ccf 2197 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2198 return 0;
2199}
2200
2201static int build_curseg(struct f2fs_sb_info *sbi)
2202{
1042d60f 2203 struct curseg_info *array;
351df4b2
JK		 2204 int i;
2205
b434babf 2206 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2207 if (!array)
2208 return -ENOMEM;
2209
2210 SM_I(sbi)->curseg_array = array;
2211
2212 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2213 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2214 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2215 if (!array[i].sum_blk)
2216 return -ENOMEM;
b7ad7512
CY		 2217 init_rwsem(&array[i].journal_rwsem);
2218 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2219 GFP_KERNEL);
2220 if (!array[i].journal)
2221 return -ENOMEM;
351df4b2
JK		 2222 array[i].segno = NULL_SEGNO;
2223 array[i].next_blkoff = 0;
2224 }
2225 return restore_curseg_summaries(sbi);
2226}
2227
2228static void build_sit_entries(struct f2fs_sb_info *sbi)
2229{
2230 struct sit_info *sit_i = SIT_I(sbi);
2231 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2232 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2233 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2234 unsigned int i, start, end;
2235 unsigned int readed, start_blk = 0;
e9f5b8b8 2236 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2237
74de593a 2238 do {
26879fb1 2239 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2240
2241 start = start_blk * sit_i->sents_per_block;
2242 end = (start_blk + readed) * sit_i->sents_per_block;
2243
7cd8558b 2244 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2245 struct seg_entry *se = &sit_i->sentries[start];
2246 struct f2fs_sit_block *sit_blk;
2247 struct f2fs_sit_entry sit;
2248 struct page *page;
2249
b7ad7512 2250 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2251 for (i = 0; i < sits_in_cursum(journal); i++) {
2252 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2253 == start) {
dfc08a12 2254 sit = sit_in_journal(journal, i);
b7ad7512 2255 up_read(&curseg->journal_rwsem);
74de593a
CY		 2256 goto got_it;
2257 }
351df4b2 2258 }
b7ad7512 2259 up_read(&curseg->journal_rwsem);
74de593a
CY		 2260
2261 page = get_current_sit_page(sbi, start);
2262 sit_blk = (struct f2fs_sit_block *)page_address(page);
2263 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2264 f2fs_put_page(page, 1);
351df4b2 2265got_it:
74de593a
CY		 2266 check_block_count(sbi, start, &sit);
2267 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2268
2269 /* build discard map only one time */
2270 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2271 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2272
74de593a
CY		 2273 if (sbi->segs_per_sec > 1) {
2274 struct sec_entry *e = get_sec_entry(sbi, start);
2275 e->valid_blocks += se->valid_blocks;
2276 }
351df4b2 2277 }
74de593a
CY		 2278 start_blk += readed;
2279 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2280}
2281
2282static void init_free_segmap(struct f2fs_sb_info *sbi)
2283{
2284 unsigned int start;
2285 int type;
2286
7cd8558b 2287 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2288 struct seg_entry *sentry = get_seg_entry(sbi, start);
2289 if (!sentry->valid_blocks)
2290 __set_free(sbi, start);
2291 }
2292
2293 /* set use the current segments */
2294 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2295 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2296 __set_test_and_inuse(sbi, curseg_t->segno);
2297 }
2298}
2299
2300static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2301{
2302 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2303 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2304 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2305 unsigned short valid_blocks;
2306
8736fbf0 2307 while (1) {
351df4b2 2308 /* find dirty segment based on free segmap */
7cd8558b
JK		 2309 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2310 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2311 break;
2312 offset = segno + 1;
2313 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2314 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2315 continue;
ec325b52
JK		 2316 if (valid_blocks > sbi->blocks_per_seg) {
2317 f2fs_bug_on(sbi, 1);
2318 continue;
2319 }
351df4b2
JK		 2320 mutex_lock(&dirty_i->seglist_lock);
2321 __locate_dirty_segment(sbi, segno, DIRTY);
2322 mutex_unlock(&dirty_i->seglist_lock);
2323 }
2324}
2325
5ec4e49f 2326static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2327{
2328 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2329 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2330
39307a8e 2331 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2332 if (!dirty_i->victim_secmap)
351df4b2
JK		 2333 return -ENOMEM;
2334 return 0;
2335}
2336
2337static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2338{
2339 struct dirty_seglist_info *dirty_i;
2340 unsigned int bitmap_size, i;
2341
2342 /* allocate memory for dirty segments list information */
2343 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2344 if (!dirty_i)
2345 return -ENOMEM;
2346
2347 SM_I(sbi)->dirty_info = dirty_i;
2348 mutex_init(&dirty_i->seglist_lock);
2349
7cd8558b 2350 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2351
2352 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2353 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2354 if (!dirty_i->dirty_segmap[i])
2355 return -ENOMEM;
2356 }
2357
2358 init_dirty_segmap(sbi);
5ec4e49f 2359 return init_victim_secmap(sbi);
351df4b2
JK		 2360}
2361
0a8165d7 2362/*
351df4b2
JK		 2363 * Update min, max modified time for cost-benefit GC algorithm
2364 */
2365static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2366{
2367 struct sit_info *sit_i = SIT_I(sbi);
2368 unsigned int segno;
2369
2370 mutex_lock(&sit_i->sentry_lock);
2371
2372 sit_i->min_mtime = LLONG_MAX;
2373
7cd8558b 2374 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2375 unsigned int i;
2376 unsigned long long mtime = 0;
2377
2378 for (i = 0; i < sbi->segs_per_sec; i++)
2379 mtime += get_seg_entry(sbi, segno + i)->mtime;
2380
2381 mtime = div_u64(mtime, sbi->segs_per_sec);
2382
2383 if (sit_i->min_mtime > mtime)
2384 sit_i->min_mtime = mtime;
2385 }
2386 sit_i->max_mtime = get_mtime(sbi);
2387 mutex_unlock(&sit_i->sentry_lock);
2388}
2389
2390int build_segment_manager(struct f2fs_sb_info *sbi)
2391{
2392 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2393 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2394 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2395 int err;
2396
2397 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2398 if (!sm_info)
2399 return -ENOMEM;
2400
2401 /* init sm info */
2402 sbi->sm_info = sm_info;
351df4b2
JK		 2403 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2404 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2405 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2406 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2407 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2408 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2409 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2410 sm_info->rec_prefree_segments = sm_info->main_segments *
2411 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2412 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2413 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2414
52763a4b
JK		 2415 if (!test_opt(sbi, LFS))
2416 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2417 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2418 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2419
7fd9e544
JK		 2420 INIT_LIST_HEAD(&sm_info->discard_list);
2421 sm_info->nr_discards = 0;
2422 sm_info->max_discards = 0;
2423
bba681cb
JK		 2424 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2425
184a5cd2
CY		 2426 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2427
b270ad6f 2428 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2429 err = create_flush_cmd_control(sbi);
2430 if (err)
a688b9d9 2431 return err;
6b4afdd7
JK		 2432 }
2433
351df4b2
JK		 2434 err = build_sit_info(sbi);
2435 if (err)
2436 return err;
2437 err = build_free_segmap(sbi);
2438 if (err)
2439 return err;
2440 err = build_curseg(sbi);
2441 if (err)
2442 return err;
2443
2444 /* reinit free segmap based on SIT */
2445 build_sit_entries(sbi);
2446
2447 init_free_segmap(sbi);
2448 err = build_dirty_segmap(sbi);
2449 if (err)
2450 return err;
2451
2452 init_min_max_mtime(sbi);
2453 return 0;
2454}
2455
2456static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2457 enum dirty_type dirty_type)
2458{
2459 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2460
2461 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2462 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2463 dirty_i->nr_dirty[dirty_type] = 0;
2464 mutex_unlock(&dirty_i->seglist_lock);
2465}
2466
5ec4e49f 2467static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2468{
2469 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2470 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2471}
2472
2473static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2474{
2475 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2476 int i;
2477
2478 if (!dirty_i)
2479 return;
2480
2481 /* discard pre-free/dirty segments list */
2482 for (i = 0; i < NR_DIRTY_TYPE; i++)
2483 discard_dirty_segmap(sbi, i);
2484
5ec4e49f 2485 destroy_victim_secmap(sbi);
351df4b2
JK		 2486 SM_I(sbi)->dirty_info = NULL;
2487 kfree(dirty_i);
2488}
2489
2490static void destroy_curseg(struct f2fs_sb_info *sbi)
2491{
2492 struct curseg_info *array = SM_I(sbi)->curseg_array;
2493 int i;
2494
2495 if (!array)
2496 return;
2497 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2498 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2499 kfree(array[i].sum_blk);
b7ad7512
CY		 2500 kfree(array[i].journal);
2501 }
351df4b2
JK		 2502 kfree(array);
2503}
2504
2505static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2506{
2507 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2508 if (!free_i)
2509 return;
2510 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2511 kvfree(free_i->free_segmap);
2512 kvfree(free_i->free_secmap);
351df4b2
JK		 2513 kfree(free_i);
2514}
2515
2516static void destroy_sit_info(struct f2fs_sb_info *sbi)
2517{
2518 struct sit_info *sit_i = SIT_I(sbi);
2519 unsigned int start;
2520
2521 if (!sit_i)
2522 return;
2523
2524 if (sit_i->sentries) {
7cd8558b 2525 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2526 kfree(sit_i->sentries[start].cur_valid_map);
2527 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2528 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2529 }
2530 }
60a3b782
JK		 2531 kfree(sit_i->tmp_map);
2532
39307a8e
JK		 2533 kvfree(sit_i->sentries);
2534 kvfree(sit_i->sec_entries);
2535 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2536
2537 SM_I(sbi)->sit_info = NULL;
2538 kfree(sit_i->sit_bitmap);
2539 kfree(sit_i);
2540}
2541
2542void destroy_segment_manager(struct f2fs_sb_info *sbi)
2543{
2544 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2545
3b03f724
CY		 2546 if (!sm_info)
2547 return;
2163d198 2548 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2549 destroy_dirty_segmap(sbi);
2550 destroy_curseg(sbi);
2551 destroy_free_segmap(sbi);
2552 destroy_sit_info(sbi);
2553 sbi->sm_info = NULL;
2554 kfree(sm_info);
2555}
7fd9e544
JK		 2556
2557int __init create_segment_manager_caches(void)
2558{
2559 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2560 sizeof(struct discard_entry));
7fd9e544 2561 if (!discard_entry_slab)
184a5cd2
CY		 2562 goto fail;
2563
2564 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2565 sizeof(struct sit_entry_set));
184a5cd2
CY		 2566 if (!sit_entry_set_slab)
2567 goto destory_discard_entry;
88b88a66
JK		 2568
2569 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2570 sizeof(struct inmem_pages));
2571 if (!inmem_entry_slab)
2572 goto destroy_sit_entry_set;
7fd9e544 2573 return 0;
184a5cd2 2574
88b88a66
JK		 2575destroy_sit_entry_set:
2576 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2577destory_discard_entry:
2578 kmem_cache_destroy(discard_entry_slab);
2579fail:
2580 return -ENOMEM;
7fd9e544
JK		 2581}
2582
2583void destroy_segment_manager_caches(void)
2584{
184a5cd2 2585 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2586 kmem_cache_destroy(discard_entry_slab);
88b88a66 2587 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2588}
Linux kernel - Deliverables
This page took 0.301467 seconds and 5 git commands to generate.