projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: use percpu_rw_semaphore
[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);
a66cdd98 675 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 676 }
677}
678
4b2fecc8
JK		 679void release_discard_addrs(struct f2fs_sb_info *sbi)
680{
681 struct list_head *head = &(SM_I(sbi)->discard_list);
682 struct discard_entry *entry, *this;
683
684 /* drop caches */
685 list_for_each_entry_safe(entry, this, head, list) {
686 list_del(&entry->list);
687 kmem_cache_free(discard_entry_slab, entry);
688 }
689}
690
0a8165d7 691/*
351df4b2
JK		 692 * Should call clear_prefree_segments after checkpoint is done.
693 */
694static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
695{
696 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 697 unsigned int segno;
351df4b2
JK		 698
699 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 700 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 701 __set_test_and_free(sbi, segno);
351df4b2
JK		 702 mutex_unlock(&dirty_i->seglist_lock);
703}
704
836b5a63 705void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 706{
b2955550 707 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 708 struct discard_entry *entry, *this;
351df4b2 709 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 710 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 711 unsigned int start = 0, end = -1;
36abef4e 712 unsigned int secno, start_segno;
351df4b2
JK		 713
714 mutex_lock(&dirty_i->seglist_lock);
29e59c14 715
351df4b2 716 while (1) {
29e59c14 717 int i;
7cd8558b
JK		 718 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
719 if (start >= MAIN_SEGS(sbi))
351df4b2 720 break;
7cd8558b
JK		 721 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
722 start + 1);
29e59c14
CL		 723
724 for (i = start; i < end; i++)
725 clear_bit(i, prefree_map);
726
727 dirty_i->nr_dirty[PRE] -= end - start;
728
729 if (!test_opt(sbi, DISCARD))
730 continue;
351df4b2 731
36abef4e
JK		 732 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
733 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 734 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 735 continue;
736 }
737next:
738 secno = GET_SECNO(sbi, start);
739 start_segno = secno * sbi->segs_per_sec;
740 if (!IS_CURSEC(sbi, secno) &&
741 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
742 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
743 sbi->segs_per_sec << sbi->log_blocks_per_seg);
744
745 start = start_segno + sbi->segs_per_sec;
746 if (start < end)
747 goto next;
351df4b2
JK		 748 }
749 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 750
751 /* send small discards */
2d7b822a 752 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 753 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
754 goto skip;
37208879 755 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 756 cpc->trimmed += entry->len;
836b5a63 757skip:
b2955550
JK		 758 list_del(&entry->list);
759 SM_I(sbi)->nr_discards -= entry->len;
760 kmem_cache_free(discard_entry_slab, entry);
761 }
351df4b2
JK		 762}
763
184a5cd2 764static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 765{
766 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 767
768 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 769 sit_i->dirty_sentries++;
184a5cd2
CY		 770 return false;
771 }
772
773 return true;
351df4b2
JK		 774}
775
776static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
777 unsigned int segno, int modified)
778{
779 struct seg_entry *se = get_seg_entry(sbi, segno);
780 se->type = type;
781 if (modified)
782 __mark_sit_entry_dirty(sbi, segno);
783}
784
785static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
786{
787 struct seg_entry *se;
788 unsigned int segno, offset;
789 long int new_vblocks;
790
791 segno = GET_SEGNO(sbi, blkaddr);
792
793 se = get_seg_entry(sbi, segno);
794 new_vblocks = se->valid_blocks + del;
491c0854 795 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 796
9850cf4a 797 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 798 (new_vblocks > sbi->blocks_per_seg)));
799
800 se->valid_blocks = new_vblocks;
801 se->mtime = get_mtime(sbi);
802 SIT_I(sbi)->max_mtime = se->mtime;
803
804 /* Update valid block bitmap */
805 if (del > 0) {
52aca074 806 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 807 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 808 if (!f2fs_test_and_set_bit(offset, se->discard_map))
809 sbi->discard_blks--;
351df4b2 810 } else {
52aca074 811 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 812 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 813 if (f2fs_test_and_clear_bit(offset, se->discard_map))
814 sbi->discard_blks++;
351df4b2
JK		 815 }
816 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
817 se->ckpt_valid_blocks += del;
818
819 __mark_sit_entry_dirty(sbi, segno);
820
821 /* update total number of valid blocks to be written in ckpt area */
822 SIT_I(sbi)->written_valid_blocks += del;
823
824 if (sbi->segs_per_sec > 1)
825 get_sec_entry(sbi, segno)->valid_blocks += del;
826}
827
5e443818 828void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 829{
5e443818
JK		 830 update_sit_entry(sbi, new, 1);
831 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
832 update_sit_entry(sbi, old, -1);
833
834 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
835 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 836}
837
838void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
839{
840 unsigned int segno = GET_SEGNO(sbi, addr);
841 struct sit_info *sit_i = SIT_I(sbi);
842
9850cf4a 843 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 844 if (addr == NEW_ADDR)
845 return;
846
847 /* add it into sit main buffer */
848 mutex_lock(&sit_i->sentry_lock);
849
850 update_sit_entry(sbi, addr, -1);
851
852 /* add it into dirty seglist */
853 locate_dirty_segment(sbi, segno);
854
855 mutex_unlock(&sit_i->sentry_lock);
856}
857
6e2c64ad
JK		 858bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
859{
860 struct sit_info *sit_i = SIT_I(sbi);
861 unsigned int segno, offset;
862 struct seg_entry *se;
863 bool is_cp = false;
864
865 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
866 return true;
867
868 mutex_lock(&sit_i->sentry_lock);
869
870 segno = GET_SEGNO(sbi, blkaddr);
871 se = get_seg_entry(sbi, segno);
872 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
873
874 if (f2fs_test_bit(offset, se->ckpt_valid_map))
875 is_cp = true;
876
877 mutex_unlock(&sit_i->sentry_lock);
878
879 return is_cp;
880}
881
0a8165d7 882/*
351df4b2
JK		 883 * This function should be resided under the curseg_mutex lock
884 */
885static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 886 struct f2fs_summary *sum)
351df4b2
JK		 887{
888 struct curseg_info *curseg = CURSEG_I(sbi, type);
889 void *addr = curseg->sum_blk;
e79efe3b 890 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 891 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 892}
893
0a8165d7 894/*
351df4b2
JK		 895 * Calculate the number of current summary pages for writing
896 */
3fa06d7b 897int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 898{
351df4b2 899 int valid_sum_count = 0;
9a47938b 900 int i, sum_in_page;
351df4b2
JK		 901
902 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
903 if (sbi->ckpt->alloc_type[i] == SSR)
904 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 905 else {
906 if (for_ra)
907 valid_sum_count += le16_to_cpu(
908 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
909 else
910 valid_sum_count += curseg_blkoff(sbi, i);
911 }
351df4b2
JK		 912 }
913
09cbfeaf 914 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 915 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
916 if (valid_sum_count <= sum_in_page)
351df4b2 917 return 1;
9a47938b 918 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 919 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 920 return 2;
921 return 3;
922}
923
0a8165d7 924/*
351df4b2
JK		 925 * Caller should put this summary page
926 */
927struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
928{
929 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
930}
931
381722d2 932void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 933{
934 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 935 void *dst = page_address(page);
936
937 if (src)
09cbfeaf 938 memcpy(dst, src, PAGE_SIZE);
381722d2 939 else
09cbfeaf 940 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 941 set_page_dirty(page);
942 f2fs_put_page(page, 1);
943}
944
381722d2
CY		 945static void write_sum_page(struct f2fs_sb_info *sbi,
946 struct f2fs_summary_block *sum_blk, block_t blk_addr)
947{
948 update_meta_page(sbi, (void *)sum_blk, blk_addr);
949}
950
b7ad7512
CY		 951static void write_current_sum_page(struct f2fs_sb_info *sbi,
952 int type, block_t blk_addr)
953{
954 struct curseg_info *curseg = CURSEG_I(sbi, type);
955 struct page *page = grab_meta_page(sbi, blk_addr);
956 struct f2fs_summary_block *src = curseg->sum_blk;
957 struct f2fs_summary_block *dst;
958
959 dst = (struct f2fs_summary_block *)page_address(page);
960
961 mutex_lock(&curseg->curseg_mutex);
962
963 down_read(&curseg->journal_rwsem);
964 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
965 up_read(&curseg->journal_rwsem);
966
967 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
968 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
969
970 mutex_unlock(&curseg->curseg_mutex);
971
972 set_page_dirty(page);
973 f2fs_put_page(page, 1);
974}
975
60374688
JK		 976static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
977{
978 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 979 unsigned int segno = curseg->segno + 1;
60374688
JK		 980 struct free_segmap_info *free_i = FREE_I(sbi);
981
7cd8558b 982 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 983 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 984 return 0;
985}
986
0a8165d7 987/*
351df4b2
JK		 988 * Find a new segment from the free segments bitmap to right order
989 * This function should be returned with success, otherwise BUG
990 */
991static void get_new_segment(struct f2fs_sb_info *sbi,
992 unsigned int *newseg, bool new_sec, int dir)
993{
994 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 995 unsigned int segno, secno, zoneno;
7cd8558b 996 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 997 unsigned int hint = *newseg / sbi->segs_per_sec;
998 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
999 unsigned int left_start = hint;
1000 bool init = true;
1001 int go_left = 0;
1002 int i;
1003
1a118ccf 1004 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1005
1006 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1007 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1008 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1009 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1010 goto got_it;
1011 }
1012find_other_zone:
7cd8558b
JK		 1013 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1014 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1015 if (dir == ALLOC_RIGHT) {
1016 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1017 MAIN_SECS(sbi), 0);
1018 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1019 } else {
1020 go_left = 1;
1021 left_start = hint - 1;
1022 }
1023 }
1024 if (go_left == 0)
1025 goto skip_left;
1026
1027 while (test_bit(left_start, free_i->free_secmap)) {
1028 if (left_start > 0) {
1029 left_start--;
1030 continue;
1031 }
1032 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1033 MAIN_SECS(sbi), 0);
1034 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1035 break;
1036 }
1037 secno = left_start;
1038skip_left:
1039 hint = secno;
1040 segno = secno * sbi->segs_per_sec;
1041 zoneno = secno / sbi->secs_per_zone;
1042
1043 /* give up on finding another zone */
1044 if (!init)
1045 goto got_it;
1046 if (sbi->secs_per_zone == 1)
1047 goto got_it;
1048 if (zoneno == old_zoneno)
1049 goto got_it;
1050 if (dir == ALLOC_LEFT) {
1051 if (!go_left && zoneno + 1 >= total_zones)
1052 goto got_it;
1053 if (go_left && zoneno == 0)
1054 goto got_it;
1055 }
1056 for (i = 0; i < NR_CURSEG_TYPE; i++)
1057 if (CURSEG_I(sbi, i)->zone == zoneno)
1058 break;
1059
1060 if (i < NR_CURSEG_TYPE) {
1061 /* zone is in user, try another */
1062 if (go_left)
1063 hint = zoneno * sbi->secs_per_zone - 1;
1064 else if (zoneno + 1 >= total_zones)
1065 hint = 0;
1066 else
1067 hint = (zoneno + 1) * sbi->secs_per_zone;
1068 init = false;
1069 goto find_other_zone;
1070 }
1071got_it:
1072 /* set it as dirty segment in free segmap */
9850cf4a 1073 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1074 __set_inuse(sbi, segno);
1075 *newseg = segno;
1a118ccf 1076 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1077}
1078
1079static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1080{
1081 struct curseg_info *curseg = CURSEG_I(sbi, type);
1082 struct summary_footer *sum_footer;
1083
1084 curseg->segno = curseg->next_segno;
1085 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1086 curseg->next_blkoff = 0;
1087 curseg->next_segno = NULL_SEGNO;
1088
1089 sum_footer = &(curseg->sum_blk->footer);
1090 memset(sum_footer, 0, sizeof(struct summary_footer));
1091 if (IS_DATASEG(type))
1092 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1093 if (IS_NODESEG(type))
1094 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1095 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1096}
1097
0a8165d7 1098/*
351df4b2
JK		 1099 * Allocate a current working segment.
1100 * This function always allocates a free segment in LFS manner.
1101 */
1102static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1103{
1104 struct curseg_info *curseg = CURSEG_I(sbi, type);
1105 unsigned int segno = curseg->segno;
1106 int dir = ALLOC_LEFT;
1107
1108 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1109 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1110 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1111 dir = ALLOC_RIGHT;
1112
1113 if (test_opt(sbi, NOHEAP))
1114 dir = ALLOC_RIGHT;
1115
1116 get_new_segment(sbi, &segno, new_sec, dir);
1117 curseg->next_segno = segno;
1118 reset_curseg(sbi, type, 1);
1119 curseg->alloc_type = LFS;
1120}
1121
1122static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1123 struct curseg_info *seg, block_t start)
1124{
1125 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1126 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1127 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1128 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1129 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1130 int i, pos;
1131
1132 for (i = 0; i < entries; i++)
1133 target_map[i] = ckpt_map[i] | cur_map[i];
1134
1135 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1136
1137 seg->next_blkoff = pos;
351df4b2
JK		 1138}
1139
0a8165d7 1140/*
351df4b2
JK		 1141 * If a segment is written by LFS manner, next block offset is just obtained
1142 * by increasing the current block offset. However, if a segment is written by
1143 * SSR manner, next block offset obtained by calling __next_free_blkoff
1144 */
1145static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1146 struct curseg_info *seg)
1147{
1148 if (seg->alloc_type == SSR)
1149 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1150 else
1151 seg->next_blkoff++;
1152}
1153
0a8165d7 1154/*
e1c42045 1155 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1156 * manner, so it should recover the existing segment information of valid blocks
1157 */
1158static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1159{
1160 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1161 struct curseg_info *curseg = CURSEG_I(sbi, type);
1162 unsigned int new_segno = curseg->next_segno;
1163 struct f2fs_summary_block *sum_node;
1164 struct page *sum_page;
1165
1166 write_sum_page(sbi, curseg->sum_blk,
1167 GET_SUM_BLOCK(sbi, curseg->segno));
1168 __set_test_and_inuse(sbi, new_segno);
1169
1170 mutex_lock(&dirty_i->seglist_lock);
1171 __remove_dirty_segment(sbi, new_segno, PRE);
1172 __remove_dirty_segment(sbi, new_segno, DIRTY);
1173 mutex_unlock(&dirty_i->seglist_lock);
1174
1175 reset_curseg(sbi, type, 1);
1176 curseg->alloc_type = SSR;
1177 __next_free_blkoff(sbi, curseg, 0);
1178
1179 if (reuse) {
1180 sum_page = get_sum_page(sbi, new_segno);
1181 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1182 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1183 f2fs_put_page(sum_page, 1);
1184 }
1185}
1186
43727527
JK		 1187static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1188{
1189 struct curseg_info *curseg = CURSEG_I(sbi, type);
1190 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1191
1192 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1193 return v_ops->get_victim(sbi,
1194 &(curseg)->next_segno, BG_GC, type, SSR);
1195
1196 /* For data segments, let's do SSR more intensively */
1197 for (; type >= CURSEG_HOT_DATA; type--)
1198 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1199 BG_GC, type, SSR))
1200 return 1;
1201 return 0;
1202}
1203
351df4b2
JK		 1204/*
1205 * flush out current segment and replace it with new segment
1206 * This function should be returned with success, otherwise BUG
1207 */
1208static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1209 int type, bool force)
1210{
1211 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1212
7b405275 1213 if (force)
351df4b2 1214 new_curseg(sbi, type, true);
7b405275 1215 else if (type == CURSEG_WARM_NODE)
351df4b2 1216 new_curseg(sbi, type, false);
60374688
JK		 1217 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1218 new_curseg(sbi, type, false);
351df4b2
JK		 1219 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1220 change_curseg(sbi, type, true);
1221 else
1222 new_curseg(sbi, type, false);
dcdfff65
JK		 1223
1224 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1225}
1226
38aa0889
JK		 1227static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1228{
1229 struct curseg_info *curseg = CURSEG_I(sbi, type);
1230 unsigned int old_segno;
1231
1232 old_segno = curseg->segno;
1233 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1234 locate_dirty_segment(sbi, old_segno);
1235}
1236
351df4b2
JK		 1237void allocate_new_segments(struct f2fs_sb_info *sbi)
1238{
351df4b2
JK		 1239 int i;
1240
36abef4e
JK		 1241 if (test_opt(sbi, LFS))
1242 return;
1243
38aa0889
JK		 1244 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1245 __allocate_new_segments(sbi, i);
351df4b2
JK		 1246}
1247
1248static const struct segment_allocation default_salloc_ops = {
1249 .allocate_segment = allocate_segment_by_default,
1250};
1251
4b2fecc8
JK		 1252int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1253{
f7ef9b83
JK		 1254 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1255 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1256 unsigned int start_segno, end_segno;
1257 struct cp_control cpc;
c34f42e2 1258 int err = 0;
4b2fecc8 1259
836b5a63 1260 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1261 return -EINVAL;
1262
9bd27ae4 1263 cpc.trimmed = 0;
7cd8558b 1264 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1265 goto out;
1266
1267 /* start/end segment number in main_area */
7cd8558b
JK		 1268 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1269 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1270 GET_SEGNO(sbi, end);
4b2fecc8 1271 cpc.reason = CP_DISCARD;
836b5a63 1272 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1273
1274 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1275 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1276 cpc.trim_start = start_segno;
a66cdd98
JK		 1277
1278 if (sbi->discard_blks == 0)
1279 break;
1280 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1281 cpc.trim_end = end_segno;
1282 else
1283 cpc.trim_end = min_t(unsigned int,
1284 rounddown(start_segno +
bba681cb
JK		 1285 BATCHED_TRIM_SEGMENTS(sbi),
1286 sbi->segs_per_sec) - 1, end_segno);
1287
1288 mutex_lock(&sbi->gc_mutex);
c34f42e2 1289 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1290 mutex_unlock(&sbi->gc_mutex);
1291 }
4b2fecc8 1292out:
f7ef9b83 1293 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1294 return err;
4b2fecc8
JK		 1295}
1296
351df4b2
JK		 1297static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1298{
1299 struct curseg_info *curseg = CURSEG_I(sbi, type);
1300 if (curseg->next_blkoff < sbi->blocks_per_seg)
1301 return true;
1302 return false;
1303}
1304
1305static int __get_segment_type_2(struct page *page, enum page_type p_type)
1306{
1307 if (p_type == DATA)
1308 return CURSEG_HOT_DATA;
1309 else
1310 return CURSEG_HOT_NODE;
1311}
1312
1313static int __get_segment_type_4(struct page *page, enum page_type p_type)
1314{
1315 if (p_type == DATA) {
1316 struct inode *inode = page->mapping->host;
1317
1318 if (S_ISDIR(inode->i_mode))
1319 return CURSEG_HOT_DATA;
1320 else
1321 return CURSEG_COLD_DATA;
1322 } else {
a344b9fd
JK		 1323 if (IS_DNODE(page) && is_cold_node(page))
1324 return CURSEG_WARM_NODE;
351df4b2
JK		 1325 else
1326 return CURSEG_COLD_NODE;
1327 }
1328}
1329
1330static int __get_segment_type_6(struct page *page, enum page_type p_type)
1331{
1332 if (p_type == DATA) {
1333 struct inode *inode = page->mapping->host;
1334
1335 if (S_ISDIR(inode->i_mode))
1336 return CURSEG_HOT_DATA;
354a3399 1337 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1338 return CURSEG_COLD_DATA;
1339 else
1340 return CURSEG_WARM_DATA;
1341 } else {
1342 if (IS_DNODE(page))
1343 return is_cold_node(page) ? CURSEG_WARM_NODE :
1344 CURSEG_HOT_NODE;
1345 else
1346 return CURSEG_COLD_NODE;
1347 }
1348}
1349
1350static int __get_segment_type(struct page *page, enum page_type p_type)
1351{
4081363f 1352 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1353 case 2:
1354 return __get_segment_type_2(page, p_type);
1355 case 4:
1356 return __get_segment_type_4(page, p_type);
351df4b2 1357 }
12a67146 1358 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1359 f2fs_bug_on(F2FS_P_SB(page),
1360 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1361 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1362}
1363
bfad7c2d
JK		 1364void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1365 block_t old_blkaddr, block_t *new_blkaddr,
1366 struct f2fs_summary *sum, int type)
351df4b2
JK		 1367{
1368 struct sit_info *sit_i = SIT_I(sbi);
1369 struct curseg_info *curseg;
38aa0889
JK		 1370 bool direct_io = (type == CURSEG_DIRECT_IO);
1371
1372 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1373
351df4b2
JK		 1374 curseg = CURSEG_I(sbi, type);
1375
1376 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1377 mutex_lock(&sit_i->sentry_lock);
351df4b2 1378
38aa0889 1379 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1380 if (direct_io && curseg->next_blkoff &&
1381 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1382 __allocate_new_segments(sbi, type);
1383
351df4b2 1384 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1385
1386 /*
1387 * __add_sum_entry should be resided under the curseg_mutex
1388 * because, this function updates a summary entry in the
1389 * current summary block.
1390 */
e79efe3b 1391 __add_sum_entry(sbi, type, sum);
351df4b2 1392
351df4b2 1393 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1394
1395 stat_inc_block_count(sbi, curseg);
351df4b2 1396
5e443818
JK		 1397 if (!__has_curseg_space(sbi, type))
1398 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1399 /*
1400 * SIT information should be updated before segment allocation,
1401 * since SSR needs latest valid block information.
1402 */
1403 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1404
351df4b2
JK		 1405 mutex_unlock(&sit_i->sentry_lock);
1406
bfad7c2d 1407 if (page && IS_NODESEG(type))
351df4b2
JK		 1408 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1409
bfad7c2d
JK		 1410 mutex_unlock(&curseg->curseg_mutex);
1411}
1412
05ca3632 1413static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1414{
05ca3632 1415 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1416
7dfeaa32
JK		 1417 if (fio->type == NODE || fio->type == DATA)
1418 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1419
7a9d7548
CY		 1420 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1421 &fio->new_blkaddr, sum, type);
bfad7c2d 1422
351df4b2 1423 /* writeout dirty page into bdev */
05ca3632 1424 f2fs_submit_page_mbio(fio);
7dfeaa32
JK		 1425
1426 if (fio->type == NODE || fio->type == DATA)
1427 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1428}
1429
577e3495 1430void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1431{
458e6197 1432 struct f2fs_io_info fio = {
05ca3632 1433 .sbi = sbi,
458e6197 1434 .type = META,
cf04e8eb 1435 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1436 .old_blkaddr = page->index,
1437 .new_blkaddr = page->index,
05ca3632 1438 .page = page,
4375a336 1439 .encrypted_page = NULL,
458e6197
JK		 1440 };
1441
2b947003
CY		 1442 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1443 fio.rw &= ~REQ_META;
1444
351df4b2 1445 set_page_writeback(page);
05ca3632 1446 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1447}
1448
05ca3632 1449void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1450{
1451 struct f2fs_summary sum;
05ca3632 1452
351df4b2 1453 set_summary(&sum, nid, 0, 0);
05ca3632 1454 do_write_page(&sum, fio);
351df4b2
JK		 1455}
1456
05ca3632 1457void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1458{
05ca3632 1459 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1460 struct f2fs_summary sum;
1461 struct node_info ni;
1462
9850cf4a 1463 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1464 get_node_info(sbi, dn->nid, &ni);
1465 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1466 do_write_page(&sum, fio);
f28b3434 1467 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1468}
1469
05ca3632 1470void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1471{
7a9d7548 1472 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1473 stat_inc_inplace_blocks(fio->sbi);
1474 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1475}
1476
4356e48e 1477void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1478 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1479 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1480{
1481 struct sit_info *sit_i = SIT_I(sbi);
1482 struct curseg_info *curseg;
1483 unsigned int segno, old_cursegno;
1484 struct seg_entry *se;
1485 int type;
19f106bc 1486 unsigned short old_blkoff;
351df4b2
JK		 1487
1488 segno = GET_SEGNO(sbi, new_blkaddr);
1489 se = get_seg_entry(sbi, segno);
1490 type = se->type;
1491
19f106bc
CY		 1492 if (!recover_curseg) {
1493 /* for recovery flow */
1494 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1495 if (old_blkaddr == NULL_ADDR)
1496 type = CURSEG_COLD_DATA;
1497 else
1498 type = CURSEG_WARM_DATA;
1499 }
1500 } else {
1501 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1502 type = CURSEG_WARM_DATA;
1503 }
19f106bc 1504
351df4b2
JK		 1505 curseg = CURSEG_I(sbi, type);
1506
1507 mutex_lock(&curseg->curseg_mutex);
1508 mutex_lock(&sit_i->sentry_lock);
1509
1510 old_cursegno = curseg->segno;
19f106bc 1511 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1512
1513 /* change the current segment */
1514 if (segno != curseg->segno) {
1515 curseg->next_segno = segno;
1516 change_curseg(sbi, type, true);
1517 }
1518
491c0854 1519 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1520 __add_sum_entry(sbi, type, sum);
351df4b2 1521
28bc106b 1522 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1523 update_sit_entry(sbi, new_blkaddr, 1);
1524 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1525 update_sit_entry(sbi, old_blkaddr, -1);
1526
1527 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1528 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1529
351df4b2 1530 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1531
19f106bc
CY		 1532 if (recover_curseg) {
1533 if (old_cursegno != curseg->segno) {
1534 curseg->next_segno = old_cursegno;
1535 change_curseg(sbi, type, true);
1536 }
1537 curseg->next_blkoff = old_blkoff;
1538 }
1539
351df4b2
JK		 1540 mutex_unlock(&sit_i->sentry_lock);
1541 mutex_unlock(&curseg->curseg_mutex);
1542}
1543
528e3459
CY		 1544void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1545 block_t old_addr, block_t new_addr,
28bc106b
CY		 1546 unsigned char version, bool recover_curseg,
1547 bool recover_newaddr)
528e3459
CY		 1548{
1549 struct f2fs_summary sum;
1550
1551 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1552
28bc106b
CY		 1553 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1554 recover_curseg, recover_newaddr);
528e3459 1555
f28b3434 1556 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1557}
1558
93dfe2ac 1559void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1560 enum page_type type, bool ordered)
93dfe2ac 1561{
93dfe2ac 1562 if (PageWriteback(page)) {
4081363f
JK		 1563 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1564
0c3a5797 1565 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1566 if (ordered)
1567 wait_on_page_writeback(page);
1568 else
1569 wait_for_stable_page(page);
93dfe2ac
JK		 1570 }
1571}
1572
08b39fbd
CY		 1573void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1574 block_t blkaddr)
1575{
1576 struct page *cpage;
1577
1578 if (blkaddr == NEW_ADDR)
1579 return;
1580
1581 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1582
1583 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1584 if (cpage) {
fec1d657 1585 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1586 f2fs_put_page(cpage, 1);
1587 }
1588}
1589
351df4b2
JK		 1590static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1591{
1592 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1593 struct curseg_info *seg_i;
1594 unsigned char *kaddr;
1595 struct page *page;
1596 block_t start;
1597 int i, j, offset;
1598
1599 start = start_sum_block(sbi);
1600
1601 page = get_meta_page(sbi, start++);
1602 kaddr = (unsigned char *)page_address(page);
1603
1604 /* Step 1: restore nat cache */
1605 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1606 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1607
1608 /* Step 2: restore sit cache */
1609 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1610 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1611 offset = 2 * SUM_JOURNAL_SIZE;
1612
1613 /* Step 3: restore summary entries */
1614 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1615 unsigned short blk_off;
1616 unsigned int segno;
1617
1618 seg_i = CURSEG_I(sbi, i);
1619 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1620 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1621 seg_i->next_segno = segno;
1622 reset_curseg(sbi, i, 0);
1623 seg_i->alloc_type = ckpt->alloc_type[i];
1624 seg_i->next_blkoff = blk_off;
1625
1626 if (seg_i->alloc_type == SSR)
1627 blk_off = sbi->blocks_per_seg;
1628
1629 for (j = 0; j < blk_off; j++) {
1630 struct f2fs_summary *s;
1631 s = (struct f2fs_summary *)(kaddr + offset);
1632 seg_i->sum_blk->entries[j] = *s;
1633 offset += SUMMARY_SIZE;
09cbfeaf 1634 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1635 SUM_FOOTER_SIZE)
1636 continue;
1637
1638 f2fs_put_page(page, 1);
1639 page = NULL;
1640
1641 page = get_meta_page(sbi, start++);
1642 kaddr = (unsigned char *)page_address(page);
1643 offset = 0;
1644 }
1645 }
1646 f2fs_put_page(page, 1);
1647 return 0;
1648}
1649
1650static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1651{
1652 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1653 struct f2fs_summary_block *sum;
1654 struct curseg_info *curseg;
1655 struct page *new;
1656 unsigned short blk_off;
1657 unsigned int segno = 0;
1658 block_t blk_addr = 0;
1659
1660 /* get segment number and block addr */
1661 if (IS_DATASEG(type)) {
1662 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1663 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1664 CURSEG_HOT_DATA]);
119ee914 1665 if (__exist_node_summaries(sbi))
351df4b2
JK		 1666 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1667 else
1668 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1669 } else {
1670 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1671 CURSEG_HOT_NODE]);
1672 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1673 CURSEG_HOT_NODE]);
119ee914 1674 if (__exist_node_summaries(sbi))
351df4b2
JK		 1675 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1676 type - CURSEG_HOT_NODE);
1677 else
1678 blk_addr = GET_SUM_BLOCK(sbi, segno);
1679 }
1680
1681 new = get_meta_page(sbi, blk_addr);
1682 sum = (struct f2fs_summary_block *)page_address(new);
1683
1684 if (IS_NODESEG(type)) {
119ee914 1685 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1686 struct f2fs_summary *ns = &sum->entries[0];
1687 int i;
1688 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1689 ns->version = 0;
1690 ns->ofs_in_node = 0;
1691 }
1692 } else {
d653788a
GZ		 1693 int err;
1694
1695 err = restore_node_summary(sbi, segno, sum);
1696 if (err) {
351df4b2 1697 f2fs_put_page(new, 1);
d653788a 1698 return err;
351df4b2
JK		 1699 }
1700 }
1701 }
1702
1703 /* set uncompleted segment to curseg */
1704 curseg = CURSEG_I(sbi, type);
1705 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1706
1707 /* update journal info */
1708 down_write(&curseg->journal_rwsem);
1709 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1710 up_write(&curseg->journal_rwsem);
1711
1712 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1713 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1714 curseg->next_segno = segno;
1715 reset_curseg(sbi, type, 0);
1716 curseg->alloc_type = ckpt->alloc_type[type];
1717 curseg->next_blkoff = blk_off;
1718 mutex_unlock(&curseg->curseg_mutex);
1719 f2fs_put_page(new, 1);
1720 return 0;
1721}
1722
1723static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1724{
1725 int type = CURSEG_HOT_DATA;
e4fc5fbf 1726 int err;
351df4b2 1727
25ca923b 1728 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1729 int npages = npages_for_summary_flush(sbi, true);
1730
1731 if (npages >= 2)
1732 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1733 META_CP, true);
3fa06d7b 1734
351df4b2
JK		 1735 /* restore for compacted data summary */
1736 if (read_compacted_summaries(sbi))
1737 return -EINVAL;
1738 type = CURSEG_HOT_NODE;
1739 }
1740
119ee914 1741 if (__exist_node_summaries(sbi))
3fa06d7b 1742 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1743 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1744
e4fc5fbf
CY		 1745 for (; type <= CURSEG_COLD_NODE; type++) {
1746 err = read_normal_summaries(sbi, type);
1747 if (err)
1748 return err;
1749 }
1750
351df4b2
JK		 1751 return 0;
1752}
1753
1754static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1755{
1756 struct page *page;
1757 unsigned char *kaddr;
1758 struct f2fs_summary *summary;
1759 struct curseg_info *seg_i;
1760 int written_size = 0;
1761 int i, j;
1762
1763 page = grab_meta_page(sbi, blkaddr++);
1764 kaddr = (unsigned char *)page_address(page);
1765
1766 /* Step 1: write nat cache */
1767 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1768 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1769 written_size += SUM_JOURNAL_SIZE;
1770
1771 /* Step 2: write sit cache */
1772 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1773 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1774 written_size += SUM_JOURNAL_SIZE;
1775
351df4b2
JK		 1776 /* Step 3: write summary entries */
1777 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1778 unsigned short blkoff;
1779 seg_i = CURSEG_I(sbi, i);
1780 if (sbi->ckpt->alloc_type[i] == SSR)
1781 blkoff = sbi->blocks_per_seg;
1782 else
1783 blkoff = curseg_blkoff(sbi, i);
1784
1785 for (j = 0; j < blkoff; j++) {
1786 if (!page) {
1787 page = grab_meta_page(sbi, blkaddr++);
1788 kaddr = (unsigned char *)page_address(page);
1789 written_size = 0;
1790 }
1791 summary = (struct f2fs_summary *)(kaddr + written_size);
1792 *summary = seg_i->sum_blk->entries[j];
1793 written_size += SUMMARY_SIZE;
351df4b2 1794
09cbfeaf 1795 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1796 SUM_FOOTER_SIZE)
1797 continue;
1798
e8d61a74 1799 set_page_dirty(page);
351df4b2
JK		 1800 f2fs_put_page(page, 1);
1801 page = NULL;
1802 }
1803 }
e8d61a74
CY		 1804 if (page) {
1805 set_page_dirty(page);
351df4b2 1806 f2fs_put_page(page, 1);
e8d61a74 1807 }
351df4b2
JK		 1808}
1809
1810static void write_normal_summaries(struct f2fs_sb_info *sbi,
1811 block_t blkaddr, int type)
1812{
1813 int i, end;
1814 if (IS_DATASEG(type))
1815 end = type + NR_CURSEG_DATA_TYPE;
1816 else
1817 end = type + NR_CURSEG_NODE_TYPE;
1818
b7ad7512
CY		 1819 for (i = type; i < end; i++)
1820 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1821}
1822
1823void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1824{
25ca923b 1825 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1826 write_compacted_summaries(sbi, start_blk);
1827 else
1828 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1829}
1830
1831void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1832{
119ee914 1833 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1834}
1835
dfc08a12 1836int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1837 unsigned int val, int alloc)
1838{
1839 int i;
1840
1841 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1842 for (i = 0; i < nats_in_cursum(journal); i++) {
1843 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1844 return i;
1845 }
dfc08a12
CY		 1846 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1847 return update_nats_in_cursum(journal, 1);
351df4b2 1848 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1849 for (i = 0; i < sits_in_cursum(journal); i++)
1850 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1851 return i;
dfc08a12
CY		 1852 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1853 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1854 }
1855 return -1;
1856}
1857
1858static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1859 unsigned int segno)
1860{
2cc22186 1861 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1862}
1863
1864static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1865 unsigned int start)
1866{
1867 struct sit_info *sit_i = SIT_I(sbi);
1868 struct page *src_page, *dst_page;
1869 pgoff_t src_off, dst_off;
1870 void *src_addr, *dst_addr;
1871
1872 src_off = current_sit_addr(sbi, start);
1873 dst_off = next_sit_addr(sbi, src_off);
1874
1875 /* get current sit block page without lock */
1876 src_page = get_meta_page(sbi, src_off);
1877 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1878 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1879
1880 src_addr = page_address(src_page);
1881 dst_addr = page_address(dst_page);
09cbfeaf 1882 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1883
1884 set_page_dirty(dst_page);
1885 f2fs_put_page(src_page, 1);
1886
1887 set_to_next_sit(sit_i, start);
1888
1889 return dst_page;
1890}
1891
184a5cd2
CY		 1892static struct sit_entry_set *grab_sit_entry_set(void)
1893{
1894 struct sit_entry_set *ses =
80c54505 1895 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1896
1897 ses->entry_cnt = 0;
1898 INIT_LIST_HEAD(&ses->set_list);
1899 return ses;
1900}
1901
1902static void release_sit_entry_set(struct sit_entry_set *ses)
1903{
1904 list_del(&ses->set_list);
1905 kmem_cache_free(sit_entry_set_slab, ses);
1906}
1907
1908static void adjust_sit_entry_set(struct sit_entry_set *ses,
1909 struct list_head *head)
1910{
1911 struct sit_entry_set *next = ses;
1912
1913 if (list_is_last(&ses->set_list, head))
1914 return;
1915
1916 list_for_each_entry_continue(next, head, set_list)
1917 if (ses->entry_cnt <= next->entry_cnt)
1918 break;
1919
1920 list_move_tail(&ses->set_list, &next->set_list);
1921}
1922
1923static void add_sit_entry(unsigned int segno, struct list_head *head)
1924{
1925 struct sit_entry_set *ses;
1926 unsigned int start_segno = START_SEGNO(segno);
1927
1928 list_for_each_entry(ses, head, set_list) {
1929 if (ses->start_segno == start_segno) {
1930 ses->entry_cnt++;
1931 adjust_sit_entry_set(ses, head);
1932 return;
1933 }
1934 }
1935
1936 ses = grab_sit_entry_set();
1937
1938 ses->start_segno = start_segno;
1939 ses->entry_cnt++;
1940 list_add(&ses->set_list, head);
1941}
1942
1943static void add_sits_in_set(struct f2fs_sb_info *sbi)
1944{
1945 struct f2fs_sm_info *sm_info = SM_I(sbi);
1946 struct list_head *set_list = &sm_info->sit_entry_set;
1947 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1948 unsigned int segno;
1949
7cd8558b 1950 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1951 add_sit_entry(segno, set_list);
1952}
1953
1954static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1955{
1956 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1957 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1958 int i;
1959
b7ad7512 1960 down_write(&curseg->journal_rwsem);
dfc08a12 1961 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1962 unsigned int segno;
1963 bool dirtied;
1964
dfc08a12 1965 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1966 dirtied = __mark_sit_entry_dirty(sbi, segno);
1967
1968 if (!dirtied)
1969 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1970 }
dfc08a12 1971 update_sits_in_cursum(journal, -i);
b7ad7512 1972 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1973}
1974
0a8165d7 1975/*
351df4b2
JK		 1976 * CP calls this function, which flushes SIT entries including sit_journal,
1977 * and moves prefree segs to free segs.
1978 */
4b2fecc8 1979void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1980{
1981 struct sit_info *sit_i = SIT_I(sbi);
1982 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1983 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1984 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 1985 struct sit_entry_set *ses, *tmp;
1986 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1987 bool to_journal = true;
4b2fecc8 1988 struct seg_entry *se;
351df4b2 1989
351df4b2
JK		 1990 mutex_lock(&sit_i->sentry_lock);
1991
2b11a74b
WL		 1992 if (!sit_i->dirty_sentries)
1993 goto out;
1994
351df4b2 1995 /*
184a5cd2
CY		 1996 * add and account sit entries of dirty bitmap in sit entry
1997 * set temporarily
351df4b2 1998 */
184a5cd2 1999 add_sits_in_set(sbi);
351df4b2 2000
184a5cd2
CY		 2001 /*
2002 * if there are no enough space in journal to store dirty sit
2003 * entries, remove all entries from journal and add and account
2004 * them in sit entry set.
2005 */
dfc08a12 2006 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2007 remove_sits_in_journal(sbi);
b2955550 2008
184a5cd2
CY		 2009 /*
2010 * there are two steps to flush sit entries:
2011 * #1, flush sit entries to journal in current cold data summary block.
2012 * #2, flush sit entries to sit page.
2013 */
2014 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2015 struct page *page = NULL;
184a5cd2
CY		 2016 struct f2fs_sit_block *raw_sit = NULL;
2017 unsigned int start_segno = ses->start_segno;
2018 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2019 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2020 unsigned int segno = start_segno;
2021
2022 if (to_journal &&
dfc08a12 2023 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2024 to_journal = false;
2025
b7ad7512
CY		 2026 if (to_journal) {
2027 down_write(&curseg->journal_rwsem);
2028 } else {
184a5cd2
CY		 2029 page = get_next_sit_page(sbi, start_segno);
2030 raw_sit = page_address(page);
351df4b2 2031 }
351df4b2 2032
184a5cd2
CY		 2033 /* flush dirty sit entries in region of current sit set */
2034 for_each_set_bit_from(segno, bitmap, end) {
2035 int offset, sit_offset;
4b2fecc8
JK		 2036
2037 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2038
2039 /* add discard candidates */
d7bc2484 2040 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2041 cpc->trim_start = segno;
2042 add_discard_addrs(sbi, cpc);
2043 }
184a5cd2
CY		 2044
2045 if (to_journal) {
dfc08a12 2046 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2047 SIT_JOURNAL, segno, 1);
2048 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2049 segno_in_journal(journal, offset) =
184a5cd2
CY		 2050 cpu_to_le32(segno);
2051 seg_info_to_raw_sit(se,
dfc08a12 2052 &sit_in_journal(journal, offset));
184a5cd2
CY		 2053 } else {
2054 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2055 seg_info_to_raw_sit(se,
2056 &raw_sit->entries[sit_offset]);
2057 }
351df4b2 2058
184a5cd2
CY		 2059 __clear_bit(segno, bitmap);
2060 sit_i->dirty_sentries--;
2061 ses->entry_cnt--;
351df4b2
JK		 2062 }
2063
b7ad7512
CY		 2064 if (to_journal)
2065 up_write(&curseg->journal_rwsem);
2066 else
184a5cd2
CY		 2067 f2fs_put_page(page, 1);
2068
2069 f2fs_bug_on(sbi, ses->entry_cnt);
2070 release_sit_entry_set(ses);
351df4b2 2071 }
184a5cd2
CY		 2072
2073 f2fs_bug_on(sbi, !list_empty(head));
2074 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2075out:
4b2fecc8
JK		 2076 if (cpc->reason == CP_DISCARD) {
2077 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2078 add_discard_addrs(sbi, cpc);
2079 }
351df4b2 2080 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2081
351df4b2
JK		 2082 set_prefree_as_free_segments(sbi);
2083}
2084
2085static int build_sit_info(struct f2fs_sb_info *sbi)
2086{
2087 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2088 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2089 struct sit_info *sit_i;
2090 unsigned int sit_segs, start;
2091 char *src_bitmap, *dst_bitmap;
2092 unsigned int bitmap_size;
2093
2094 /* allocate memory for SIT information */
2095 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2096 if (!sit_i)
2097 return -ENOMEM;
2098
2099 SM_I(sbi)->sit_info = sit_i;
2100
39307a8e
JK		 2101 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2102 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2103 if (!sit_i->sentries)
2104 return -ENOMEM;
2105
7cd8558b 2106 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2107 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2108 if (!sit_i->dirty_sentries_bitmap)
2109 return -ENOMEM;
2110
7cd8558b 2111 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2112 sit_i->sentries[start].cur_valid_map
2113 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2114 sit_i->sentries[start].ckpt_valid_map
2115 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2116 sit_i->sentries[start].discard_map
2117 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2118 if (!sit_i->sentries[start].cur_valid_map ||
2119 !sit_i->sentries[start].ckpt_valid_map ||
2120 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2121 return -ENOMEM;
2122 }
2123
60a3b782
JK		 2124 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2125 if (!sit_i->tmp_map)
2126 return -ENOMEM;
2127
351df4b2 2128 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2129 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2130 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2131 if (!sit_i->sec_entries)
2132 return -ENOMEM;
2133 }
2134
2135 /* get information related with SIT */
2136 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2137
2138 /* setup SIT bitmap from ckeckpoint pack */
2139 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2140 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2141
79b5793b 2142 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2143 if (!dst_bitmap)
2144 return -ENOMEM;
351df4b2
JK		 2145
2146 /* init SIT information */
2147 sit_i->s_ops = &default_salloc_ops;
2148
2149 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2150 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2151 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2152 sit_i->sit_bitmap = dst_bitmap;
2153 sit_i->bitmap_size = bitmap_size;
2154 sit_i->dirty_sentries = 0;
2155 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2156 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2157 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2158 mutex_init(&sit_i->sentry_lock);
2159 return 0;
2160}
2161
2162static int build_free_segmap(struct f2fs_sb_info *sbi)
2163{
351df4b2
JK		 2164 struct free_segmap_info *free_i;
2165 unsigned int bitmap_size, sec_bitmap_size;
2166
2167 /* allocate memory for free segmap information */
2168 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2169 if (!free_i)
2170 return -ENOMEM;
2171
2172 SM_I(sbi)->free_info = free_i;
2173
7cd8558b 2174 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2175 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2176 if (!free_i->free_segmap)
2177 return -ENOMEM;
2178
7cd8558b 2179 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2180 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2181 if (!free_i->free_secmap)
2182 return -ENOMEM;
2183
2184 /* set all segments as dirty temporarily */
2185 memset(free_i->free_segmap, 0xff, bitmap_size);
2186 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2187
2188 /* init free segmap information */
7cd8558b 2189 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2190 free_i->free_segments = 0;
2191 free_i->free_sections = 0;
1a118ccf 2192 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2193 return 0;
2194}
2195
2196static int build_curseg(struct f2fs_sb_info *sbi)
2197{
1042d60f 2198 struct curseg_info *array;
351df4b2
JK		 2199 int i;
2200
b434babf 2201 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2202 if (!array)
2203 return -ENOMEM;
2204
2205 SM_I(sbi)->curseg_array = array;
2206
2207 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2208 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2209 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2210 if (!array[i].sum_blk)
2211 return -ENOMEM;
b7ad7512
CY		 2212 init_rwsem(&array[i].journal_rwsem);
2213 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2214 GFP_KERNEL);
2215 if (!array[i].journal)
2216 return -ENOMEM;
351df4b2
JK		 2217 array[i].segno = NULL_SEGNO;
2218 array[i].next_blkoff = 0;
2219 }
2220 return restore_curseg_summaries(sbi);
2221}
2222
2223static void build_sit_entries(struct f2fs_sb_info *sbi)
2224{
2225 struct sit_info *sit_i = SIT_I(sbi);
2226 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2227 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2228 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2229 unsigned int i, start, end;
2230 unsigned int readed, start_blk = 0;
e9f5b8b8 2231 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2232
74de593a 2233 do {
26879fb1 2234 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2235
2236 start = start_blk * sit_i->sents_per_block;
2237 end = (start_blk + readed) * sit_i->sents_per_block;
2238
7cd8558b 2239 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2240 struct seg_entry *se = &sit_i->sentries[start];
2241 struct f2fs_sit_block *sit_blk;
2242 struct f2fs_sit_entry sit;
2243 struct page *page;
2244
b7ad7512 2245 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2246 for (i = 0; i < sits_in_cursum(journal); i++) {
2247 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2248 == start) {
dfc08a12 2249 sit = sit_in_journal(journal, i);
b7ad7512 2250 up_read(&curseg->journal_rwsem);
74de593a
CY		 2251 goto got_it;
2252 }
351df4b2 2253 }
b7ad7512 2254 up_read(&curseg->journal_rwsem);
74de593a
CY		 2255
2256 page = get_current_sit_page(sbi, start);
2257 sit_blk = (struct f2fs_sit_block *)page_address(page);
2258 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2259 f2fs_put_page(page, 1);
351df4b2 2260got_it:
74de593a
CY		 2261 check_block_count(sbi, start, &sit);
2262 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2263
2264 /* build discard map only one time */
2265 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2266 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2267
74de593a
CY		 2268 if (sbi->segs_per_sec > 1) {
2269 struct sec_entry *e = get_sec_entry(sbi, start);
2270 e->valid_blocks += se->valid_blocks;
2271 }
351df4b2 2272 }
74de593a
CY		 2273 start_blk += readed;
2274 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2275}
2276
2277static void init_free_segmap(struct f2fs_sb_info *sbi)
2278{
2279 unsigned int start;
2280 int type;
2281
7cd8558b 2282 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2283 struct seg_entry *sentry = get_seg_entry(sbi, start);
2284 if (!sentry->valid_blocks)
2285 __set_free(sbi, start);
2286 }
2287
2288 /* set use the current segments */
2289 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2290 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2291 __set_test_and_inuse(sbi, curseg_t->segno);
2292 }
2293}
2294
2295static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2296{
2297 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2298 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2299 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2300 unsigned short valid_blocks;
2301
8736fbf0 2302 while (1) {
351df4b2 2303 /* find dirty segment based on free segmap */
7cd8558b
JK		 2304 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2305 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2306 break;
2307 offset = segno + 1;
2308 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2309 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2310 continue;
ec325b52
JK		 2311 if (valid_blocks > sbi->blocks_per_seg) {
2312 f2fs_bug_on(sbi, 1);
2313 continue;
2314 }
351df4b2
JK		 2315 mutex_lock(&dirty_i->seglist_lock);
2316 __locate_dirty_segment(sbi, segno, DIRTY);
2317 mutex_unlock(&dirty_i->seglist_lock);
2318 }
2319}
2320
5ec4e49f 2321static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2322{
2323 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2324 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2325
39307a8e 2326 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2327 if (!dirty_i->victim_secmap)
351df4b2
JK		 2328 return -ENOMEM;
2329 return 0;
2330}
2331
2332static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2333{
2334 struct dirty_seglist_info *dirty_i;
2335 unsigned int bitmap_size, i;
2336
2337 /* allocate memory for dirty segments list information */
2338 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2339 if (!dirty_i)
2340 return -ENOMEM;
2341
2342 SM_I(sbi)->dirty_info = dirty_i;
2343 mutex_init(&dirty_i->seglist_lock);
2344
7cd8558b 2345 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2346
2347 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2348 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2349 if (!dirty_i->dirty_segmap[i])
2350 return -ENOMEM;
2351 }
2352
2353 init_dirty_segmap(sbi);
5ec4e49f 2354 return init_victim_secmap(sbi);
351df4b2
JK		 2355}
2356
0a8165d7 2357/*
351df4b2
JK		 2358 * Update min, max modified time for cost-benefit GC algorithm
2359 */
2360static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2361{
2362 struct sit_info *sit_i = SIT_I(sbi);
2363 unsigned int segno;
2364
2365 mutex_lock(&sit_i->sentry_lock);
2366
2367 sit_i->min_mtime = LLONG_MAX;
2368
7cd8558b 2369 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2370 unsigned int i;
2371 unsigned long long mtime = 0;
2372
2373 for (i = 0; i < sbi->segs_per_sec; i++)
2374 mtime += get_seg_entry(sbi, segno + i)->mtime;
2375
2376 mtime = div_u64(mtime, sbi->segs_per_sec);
2377
2378 if (sit_i->min_mtime > mtime)
2379 sit_i->min_mtime = mtime;
2380 }
2381 sit_i->max_mtime = get_mtime(sbi);
2382 mutex_unlock(&sit_i->sentry_lock);
2383}
2384
2385int build_segment_manager(struct f2fs_sb_info *sbi)
2386{
2387 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2388 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2389 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2390 int err;
2391
2392 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2393 if (!sm_info)
2394 return -ENOMEM;
2395
2396 /* init sm info */
2397 sbi->sm_info = sm_info;
351df4b2
JK		 2398 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2399 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2400 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2401 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2402 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2403 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2404 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2405 sm_info->rec_prefree_segments = sm_info->main_segments *
2406 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
52763a4b
JK		 2407 if (!test_opt(sbi, LFS))
2408 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2409 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2410 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2411
7fd9e544
JK		 2412 INIT_LIST_HEAD(&sm_info->discard_list);
2413 sm_info->nr_discards = 0;
2414 sm_info->max_discards = 0;
2415
bba681cb
JK		 2416 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2417
184a5cd2
CY		 2418 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2419
b270ad6f 2420 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2421 err = create_flush_cmd_control(sbi);
2422 if (err)
a688b9d9 2423 return err;
6b4afdd7
JK		 2424 }
2425
351df4b2
JK		 2426 err = build_sit_info(sbi);
2427 if (err)
2428 return err;
2429 err = build_free_segmap(sbi);
2430 if (err)
2431 return err;
2432 err = build_curseg(sbi);
2433 if (err)
2434 return err;
2435
2436 /* reinit free segmap based on SIT */
2437 build_sit_entries(sbi);
2438
2439 init_free_segmap(sbi);
2440 err = build_dirty_segmap(sbi);
2441 if (err)
2442 return err;
2443
2444 init_min_max_mtime(sbi);
2445 return 0;
2446}
2447
2448static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2449 enum dirty_type dirty_type)
2450{
2451 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2452
2453 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2454 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2455 dirty_i->nr_dirty[dirty_type] = 0;
2456 mutex_unlock(&dirty_i->seglist_lock);
2457}
2458
5ec4e49f 2459static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2460{
2461 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2462 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2463}
2464
2465static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2466{
2467 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2468 int i;
2469
2470 if (!dirty_i)
2471 return;
2472
2473 /* discard pre-free/dirty segments list */
2474 for (i = 0; i < NR_DIRTY_TYPE; i++)
2475 discard_dirty_segmap(sbi, i);
2476
5ec4e49f 2477 destroy_victim_secmap(sbi);
351df4b2
JK		 2478 SM_I(sbi)->dirty_info = NULL;
2479 kfree(dirty_i);
2480}
2481
2482static void destroy_curseg(struct f2fs_sb_info *sbi)
2483{
2484 struct curseg_info *array = SM_I(sbi)->curseg_array;
2485 int i;
2486
2487 if (!array)
2488 return;
2489 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2490 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2491 kfree(array[i].sum_blk);
b7ad7512
CY		 2492 kfree(array[i].journal);
2493 }
351df4b2
JK		 2494 kfree(array);
2495}
2496
2497static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2498{
2499 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2500 if (!free_i)
2501 return;
2502 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2503 kvfree(free_i->free_segmap);
2504 kvfree(free_i->free_secmap);
351df4b2
JK		 2505 kfree(free_i);
2506}
2507
2508static void destroy_sit_info(struct f2fs_sb_info *sbi)
2509{
2510 struct sit_info *sit_i = SIT_I(sbi);
2511 unsigned int start;
2512
2513 if (!sit_i)
2514 return;
2515
2516 if (sit_i->sentries) {
7cd8558b 2517 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2518 kfree(sit_i->sentries[start].cur_valid_map);
2519 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2520 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2521 }
2522 }
60a3b782
JK		 2523 kfree(sit_i->tmp_map);
2524
39307a8e
JK		 2525 kvfree(sit_i->sentries);
2526 kvfree(sit_i->sec_entries);
2527 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2528
2529 SM_I(sbi)->sit_info = NULL;
2530 kfree(sit_i->sit_bitmap);
2531 kfree(sit_i);
2532}
2533
2534void destroy_segment_manager(struct f2fs_sb_info *sbi)
2535{
2536 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2537
3b03f724
CY		 2538 if (!sm_info)
2539 return;
2163d198 2540 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2541 destroy_dirty_segmap(sbi);
2542 destroy_curseg(sbi);
2543 destroy_free_segmap(sbi);
2544 destroy_sit_info(sbi);
2545 sbi->sm_info = NULL;
2546 kfree(sm_info);
2547}
7fd9e544
JK		 2548
2549int __init create_segment_manager_caches(void)
2550{
2551 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2552 sizeof(struct discard_entry));
7fd9e544 2553 if (!discard_entry_slab)
184a5cd2
CY		 2554 goto fail;
2555
2556 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2557 sizeof(struct sit_entry_set));
184a5cd2
CY		 2558 if (!sit_entry_set_slab)
2559 goto destory_discard_entry;
88b88a66
JK		 2560
2561 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2562 sizeof(struct inmem_pages));
2563 if (!inmem_entry_slab)
2564 goto destroy_sit_entry_set;
7fd9e544 2565 return 0;
184a5cd2 2566
88b88a66
JK		 2567destroy_sit_entry_set:
2568 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2569destory_discard_entry:
2570 kmem_cache_destroy(discard_entry_slab);
2571fail:
2572 return -ENOMEM;
7fd9e544
JK		 2573}
2574
2575void destroy_segment_manager_caches(void)
2576{
184a5cd2 2577 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2578 kmem_cache_destroy(discard_entry_slab);
88b88a66 2579 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2580}
Linux kernel - Deliverables
This page took 0.270111 seconds and 5 git commands to generate.