]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block/fs/drivers: remove rw argument from submit_bio
[mirror_ubuntu-bionic-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
351df4b2
JK		 19
20#include "f2fs.h"
21#include "segment.h"
22#include "node.h"
9e4ded3f 23#include "trace.h"
6ec178da 24#include <trace/events/f2fs.h>
351df4b2 25
9a7f143a
CL		 26#define __reverse_ffz(x) __reverse_ffs(~(x))
27
7fd9e544 28static struct kmem_cache *discard_entry_slab;
184a5cd2 29static struct kmem_cache *sit_entry_set_slab;
88b88a66 30static struct kmem_cache *inmem_entry_slab;
7fd9e544 31
f96999c3
JK		 32static unsigned long __reverse_ulong(unsigned char *str)
33{
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
36
37#if BITS_PER_LONG == 64
38 shift = 56;
39#endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
43 }
44 return tmp;
45}
46
9a7f143a
CL		 47/*
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
50 */
51static inline unsigned long __reverse_ffs(unsigned long word)
52{
53 int num = 0;
54
55#if BITS_PER_LONG == 64
f96999c3 56 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 57 num += 32;
f96999c3 58 else
9a7f143a 59 word >>= 32;
9a7f143a 60#endif
f96999c3 61 if ((word & 0xffff0000) == 0)
9a7f143a 62 num += 16;
f96999c3 63 else
9a7f143a 64 word >>= 16;
f96999c3
JK		 65
66 if ((word & 0xff00) == 0)
9a7f143a 67 num += 8;
f96999c3 68 else
9a7f143a 69 word >>= 8;
f96999c3 70
9a7f143a
CL		 71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
f96999c3 75
9a7f143a
CL		 76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
f96999c3 80
9a7f143a
CL		 81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
84}
85
86/*
e1c42045 87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 89 * @size must be integral times of unsigned long.
9a7f143a 90 * Example:
f96999c3
JK		 91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 94 */
95static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
97{
98 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 99 unsigned long result = size;
9a7f143a 100 unsigned long tmp;
9a7f143a
CL		 101
102 if (offset >= size)
103 return size;
104
692223d1 105 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 106 offset %= BITS_PER_LONG;
f96999c3 107
692223d1
FL		 108 while (1) {
109 if (*p == 0)
110 goto pass;
9a7f143a 111
f96999c3 112 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 113
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 117 if (tmp)
692223d1
FL		 118 goto found;
119pass:
120 if (size <= BITS_PER_LONG)
121 break;
9a7f143a 122 size -= BITS_PER_LONG;
692223d1 123 offset = 0;
f96999c3 124 p++;
9a7f143a 125 }
692223d1
FL		 126 return result;
127found:
128 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 129}
130
131static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
133{
134 const unsigned long *p = addr + BIT_WORD(offset);
80609448 135 unsigned long result = size;
9a7f143a 136 unsigned long tmp;
9a7f143a
CL		 137
138 if (offset >= size)
139 return size;
140
80609448 141 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 142 offset %= BITS_PER_LONG;
80609448
JK		 143
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
147
f96999c3 148 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 149
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
f96999c3 154 if (tmp != ~0UL)
80609448
JK		 155 goto found;
156pass:
157 if (size <= BITS_PER_LONG)
158 break;
9a7f143a 159 size -= BITS_PER_LONG;
80609448 160 offset = 0;
f96999c3 161 p++;
9a7f143a 162 }
80609448
JK		 163 return result;
164found:
165 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 166}
167
88b88a66
JK		 168void register_inmem_page(struct inode *inode, struct page *page)
169{
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
9be32d72 172
9e4ded3f 173 f2fs_trace_pid(page);
0722b101 174
decd36b6
CY		 175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
177
88b88a66
JK		 178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
179
180 /* add atomic page indices to the list */
181 new->page = page;
182 INIT_LIST_HEAD(&new->list);
decd36b6 183
88b88a66
JK		 184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
186 get_page(page);
187 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 189 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 190
191 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 192}
193
28bc106b
CY		 194static int __revoke_inmem_pages(struct inode *inode,
195 struct list_head *head, bool drop, bool recover)
29b96b54 196{
28bc106b 197 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 198 struct inmem_pages *cur, *tmp;
28bc106b 199 int err = 0;
29b96b54
CY		 200
201 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 202 struct page *page = cur->page;
203
204 if (drop)
205 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
206
207 lock_page(page);
29b96b54 208
28bc106b
CY		 209 if (recover) {
210 struct dnode_of_data dn;
211 struct node_info ni;
212
213 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
214
215 set_new_dnode(&dn, inode, NULL, NULL, 0);
216 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
217 err = -EAGAIN;
218 goto next;
219 }
220 get_node_info(sbi, dn.nid, &ni);
221 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
222 cur->old_addr, ni.version, true, true);
223 f2fs_put_dnode(&dn);
224 }
225next:
63c52d78
JK		 226 /* we don't need to invalidate this in the sccessful status */
227 if (drop || recover)
228 ClearPageUptodate(page);
28bc106b 229 set_page_private(page, 0);
c81ced05 230 ClearPagePrivate(page);
28bc106b 231 f2fs_put_page(page, 1);
29b96b54
CY		 232
233 list_del(&cur->list);
234 kmem_cache_free(inmem_entry_slab, cur);
235 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
236 }
28bc106b 237 return err;
29b96b54
CY		 238}
239
240void drop_inmem_pages(struct inode *inode)
241{
242 struct f2fs_inode_info *fi = F2FS_I(inode);
243
26dc3d44
JK		 244 clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
245
29b96b54 246 mutex_lock(&fi->inmem_lock);
28bc106b 247 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54
CY		 248 mutex_unlock(&fi->inmem_lock);
249}
250
28bc106b
CY		 251static int __commit_inmem_pages(struct inode *inode,
252 struct list_head *revoke_list)
88b88a66
JK		 253{
254 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct inmem_pages *cur, *tmp;
88b88a66 257 struct f2fs_io_info fio = {
05ca3632 258 .sbi = sbi,
88b88a66 259 .type = DATA,
1e84371f 260 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 261 .encrypted_page = NULL,
88b88a66 262 };
29b96b54 263 bool submit_bio = false;
edb27dee 264 int err = 0;
88b88a66 265
88b88a66 266 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 267 struct page *page = cur->page;
268
269 lock_page(page);
270 if (page->mapping == inode->i_mapping) {
271 trace_f2fs_commit_inmem_page(page, INMEM);
272
273 set_page_dirty(page);
274 f2fs_wait_on_page_writeback(page, DATA, true);
275 if (clear_page_dirty_for_io(page))
29b96b54 276 inode_dec_dirty_pages(inode);
28bc106b
CY		 277
278 fio.page = page;
29b96b54
CY		 279 err = do_write_data_page(&fio);
280 if (err) {
28bc106b 281 unlock_page(page);
29b96b54 282 break;
70c640b1 283 }
29b96b54 284
28bc106b
CY		 285 /* record old blkaddr for revoking */
286 cur->old_addr = fio.old_blkaddr;
decd36b6 287
28bc106b
CY		 288 clear_cold_data(page);
289 submit_bio = true;
290 }
291 unlock_page(page);
292 list_move_tail(&cur->list, revoke_list);
88b88a66 293 }
29b96b54
CY		 294
295 if (submit_bio)
296 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
28bc106b
CY		 297
298 if (!err)
299 __revoke_inmem_pages(inode, revoke_list, false, false);
300
29b96b54
CY		 301 return err;
302}
303
304int commit_inmem_pages(struct inode *inode)
305{
306 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
307 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 308 struct list_head revoke_list;
309 int err;
29b96b54 310
28bc106b 311 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 312 f2fs_balance_fs(sbi, true);
313 f2fs_lock_op(sbi);
314
315 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 316 err = __commit_inmem_pages(inode, &revoke_list);
317 if (err) {
318 int ret;
319 /*
320 * try to revoke all committed pages, but still we could fail
321 * due to no memory or other reason, if that happened, EAGAIN
322 * will be returned, which means in such case, transaction is
323 * already not integrity, caller should use journal to do the
324 * recovery or rewrite & commit last transaction. For other
325 * error number, revoking was done by filesystem itself.
326 */
327 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
328 if (ret)
329 err = ret;
330
331 /* drop all uncommitted pages */
332 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
333 }
88b88a66
JK		 334 mutex_unlock(&fi->inmem_lock);
335
29b96b54 336 f2fs_unlock_op(sbi);
edb27dee 337 return err;
88b88a66
JK		 338}
339
0a8165d7 340/*
351df4b2
JK		 341 * This function balances dirty node and dentry pages.
342 * In addition, it controls garbage collection.
343 */
2c4db1a6 344void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 345{
2c4db1a6
JK		 346 if (!need)
347 return;
351df4b2 348 /*
029cd28c
JK		 349 * We should do GC or end up with checkpoint, if there are so many dirty
350 * dir/node pages without enough free segments.
351df4b2 351 */
43727527 352 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 353 mutex_lock(&sbi->gc_mutex);
d530d4d8 354 f2fs_gc(sbi, false);
351df4b2
JK		 355 }
356}
357
4660f9c0
JK		 358void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
359{
1dcc336b 360 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 361 if (!available_free_memory(sbi, EXTENT_CACHE))
362 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 363
1b38dc8e
JK		 364 /* check the # of cached NAT entries */
365 if (!available_free_memory(sbi, NAT_ENTRIES))
366 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
367
31696580
CY		 368 if (!available_free_memory(sbi, FREE_NIDS))
369 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
370
1b38dc8e
JK		 371 /* checkpoint is the only way to shrink partial cached entries */
372 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 373 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 374 excess_prefree_segs(sbi) ||
375 excess_dirty_nats(sbi) ||
d0239e1b 376 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
e9f5b8b8
CY		 377 if (test_opt(sbi, DATA_FLUSH)) {
378 struct blk_plug plug;
379
380 blk_start_plug(&plug);
36b35a0d 381 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 382 blk_finish_plug(&plug);
383 }
4660f9c0 384 f2fs_sync_fs(sbi->sb, true);
42190d2a 385 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 386 }
4660f9c0
JK		 387}
388
2163d198 389static int issue_flush_thread(void *data)
6b4afdd7
JK		 390{
391 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 392 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
393 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 394repeat:
395 if (kthread_should_stop())
396 return 0;
397
721bd4d5 398 if (!llist_empty(&fcc->issue_list)) {
740432f8 399 struct bio *bio;
6b4afdd7
JK		 400 struct flush_cmd *cmd, *next;
401 int ret;
402
740432f8
JK		 403 bio = f2fs_bio_alloc(0);
404
721bd4d5
GZ		 405 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
406 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
407
6b4afdd7 408 bio->bi_bdev = sbi->sb->s_bdev;
4e49ea4a
MC		 409 bio->bi_rw = WRITE_FLUSH;
410 ret = submit_bio_wait(bio);
6b4afdd7 411
721bd4d5
GZ		 412 llist_for_each_entry_safe(cmd, next,
413 fcc->dispatch_list, llnode) {
6b4afdd7 414 cmd->ret = ret;
6b4afdd7
JK		 415 complete(&cmd->wait);
416 }
a4ed23f2 417 bio_put(bio);
a688b9d9 418 fcc->dispatch_list = NULL;
6b4afdd7
JK		 419 }
420
a688b9d9 421 wait_event_interruptible(*q,
721bd4d5 422 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 423 goto repeat;
424}
425
426int f2fs_issue_flush(struct f2fs_sb_info *sbi)
427{
a688b9d9 428 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 429 struct flush_cmd cmd;
6b4afdd7 430
24a9ee0f
JK		 431 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
432 test_opt(sbi, FLUSH_MERGE));
433
0f7b2abd
JK		 434 if (test_opt(sbi, NOBARRIER))
435 return 0;
436
740432f8
JK		 437 if (!test_opt(sbi, FLUSH_MERGE)) {
438 struct bio *bio = f2fs_bio_alloc(0);
439 int ret;
440
441 bio->bi_bdev = sbi->sb->s_bdev;
4e49ea4a
MC		 442 bio->bi_rw = WRITE_FLUSH;
443 ret = submit_bio_wait(bio);
740432f8
JK		 444 bio_put(bio);
445 return ret;
446 }
6b4afdd7 447
adf8d90b 448 init_completion(&cmd.wait);
6b4afdd7 449
721bd4d5 450 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 451
a688b9d9
GZ		 452 if (!fcc->dispatch_list)
453 wake_up(&fcc->flush_wait_queue);
6b4afdd7 454
adf8d90b
CY		 455 wait_for_completion(&cmd.wait);
456
457 return cmd.ret;
6b4afdd7
JK		 458}
459
2163d198
GZ		 460int create_flush_cmd_control(struct f2fs_sb_info *sbi)
461{
462 dev_t dev = sbi->sb->s_bdev->bd_dev;
463 struct flush_cmd_control *fcc;
464 int err = 0;
465
466 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
467 if (!fcc)
468 return -ENOMEM;
2163d198 469 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 470 init_llist_head(&fcc->issue_list);
6b2920a5 471 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 472 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
473 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
474 if (IS_ERR(fcc->f2fs_issue_flush)) {
475 err = PTR_ERR(fcc->f2fs_issue_flush);
476 kfree(fcc);
6b2920a5 477 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 478 return err;
479 }
2163d198
GZ		 480
481 return err;
482}
483
484void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
485{
6b2920a5 486 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 487
488 if (fcc && fcc->f2fs_issue_flush)
489 kthread_stop(fcc->f2fs_issue_flush);
490 kfree(fcc);
6b2920a5 491 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 492}
493
351df4b2
JK		 494static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
495 enum dirty_type dirty_type)
496{
497 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
498
499 /* need not be added */
500 if (IS_CURSEG(sbi, segno))
501 return;
502
503 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
504 dirty_i->nr_dirty[dirty_type]++;
505
506 if (dirty_type == DIRTY) {
507 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 508 enum dirty_type t = sentry->type;
b2f2c390 509
ec325b52
JK		 510 if (unlikely(t >= DIRTY)) {
511 f2fs_bug_on(sbi, 1);
512 return;
513 }
4625d6aa
CL		 514 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
515 dirty_i->nr_dirty[t]++;
351df4b2
JK		 516 }
517}
518
519static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
520 enum dirty_type dirty_type)
521{
522 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
523
524 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
525 dirty_i->nr_dirty[dirty_type]--;
526
527 if (dirty_type == DIRTY) {
4625d6aa
CL		 528 struct seg_entry *sentry = get_seg_entry(sbi, segno);
529 enum dirty_type t = sentry->type;
530
531 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
532 dirty_i->nr_dirty[t]--;
b2f2c390 533
5ec4e49f
JK		 534 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
535 clear_bit(GET_SECNO(sbi, segno),
536 dirty_i->victim_secmap);
351df4b2
JK		 537 }
538}
539
0a8165d7 540/*
351df4b2
JK		 541 * Should not occur error such as -ENOMEM.
542 * Adding dirty entry into seglist is not critical operation.
543 * If a given segment is one of current working segments, it won't be added.
544 */
8d8451af 545static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 546{
547 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
548 unsigned short valid_blocks;
549
550 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
551 return;
552
553 mutex_lock(&dirty_i->seglist_lock);
554
555 valid_blocks = get_valid_blocks(sbi, segno, 0);
556
557 if (valid_blocks == 0) {
558 __locate_dirty_segment(sbi, segno, PRE);
559 __remove_dirty_segment(sbi, segno, DIRTY);
560 } else if (valid_blocks < sbi->blocks_per_seg) {
561 __locate_dirty_segment(sbi, segno, DIRTY);
562 } else {
563 /* Recovery routine with SSR needs this */
564 __remove_dirty_segment(sbi, segno, DIRTY);
565 }
566
567 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 568}
569
1e87a78d 570static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 571 block_t blkstart, block_t blklen)
572{
55cf9cb6
CY		 573 sector_t start = SECTOR_FROM_BLOCK(blkstart);
574 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 575 struct seg_entry *se;
576 unsigned int offset;
577 block_t i;
578
579 for (i = blkstart; i < blkstart + blklen; i++) {
580 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
581 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
582
583 if (!f2fs_test_and_set_bit(offset, se->discard_map))
584 sbi->discard_blks--;
585 }
1661d07c 586 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 587 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
588}
589
e90c2d28 590bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 591{
60b286c4 592 int err = -EOPNOTSUPP;
40a02be1
JK		 593
594 if (test_opt(sbi, DISCARD)) {
595 struct seg_entry *se = get_seg_entry(sbi,
596 GET_SEGNO(sbi, blkaddr));
597 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
598
599 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 600 return false;
40a02be1
JK		 601
602 err = f2fs_issue_discard(sbi, blkaddr, 1);
603 }
604
e90c2d28 605 if (err) {
381722d2 606 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 607 return true;
608 }
609 return false;
37208879
JK		 610}
611
adf4983b 612static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 613 struct cp_control *cpc, struct seg_entry *se,
614 unsigned int start, unsigned int end)
b2955550
JK		 615{
616 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 617 struct discard_entry *new, *last;
618
619 if (!list_empty(head)) {
620 last = list_last_entry(head, struct discard_entry, list);
621 if (START_BLOCK(sbi, cpc->trim_start) + start ==
622 last->blkaddr + last->len) {
623 last->len += end - start;
624 goto done;
625 }
626 }
627
628 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
629 INIT_LIST_HEAD(&new->list);
630 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
631 new->len = end - start;
632 list_add_tail(&new->list, head);
633done:
634 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 635}
636
637static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
638{
b2955550
JK		 639 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
640 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 641 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 642 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
643 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 644 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 645 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 646 unsigned int start = 0, end = -1;
4b2fecc8 647 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 648 int i;
649
a66cdd98 650 if (se->valid_blocks == max_blocks)
b2955550
JK		 651 return;
652
a66cdd98
JK		 653 if (!force) {
654 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 655 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
656 return;
4b2fecc8
JK		 657 }
658
b2955550
JK		 659 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
660 for (i = 0; i < entries; i++)
a66cdd98 661 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 662 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 663
4b2fecc8 664 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 665 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
666 if (start >= max_blocks)
667 break;
668
669 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 670 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 671 }
672}
673
4b2fecc8
JK		 674void release_discard_addrs(struct f2fs_sb_info *sbi)
675{
676 struct list_head *head = &(SM_I(sbi)->discard_list);
677 struct discard_entry *entry, *this;
678
679 /* drop caches */
680 list_for_each_entry_safe(entry, this, head, list) {
681 list_del(&entry->list);
682 kmem_cache_free(discard_entry_slab, entry);
683 }
684}
685
0a8165d7 686/*
351df4b2
JK		 687 * Should call clear_prefree_segments after checkpoint is done.
688 */
689static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
690{
691 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 692 unsigned int segno;
351df4b2
JK		 693
694 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 695 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 696 __set_test_and_free(sbi, segno);
351df4b2
JK		 697 mutex_unlock(&dirty_i->seglist_lock);
698}
699
836b5a63 700void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 701{
b2955550 702 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 703 struct discard_entry *entry, *this;
351df4b2 704 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 705 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 706 unsigned int start = 0, end = -1;
351df4b2
JK		 707
708 mutex_lock(&dirty_i->seglist_lock);
29e59c14 709
351df4b2 710 while (1) {
29e59c14 711 int i;
7cd8558b
JK		 712 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
713 if (start >= MAIN_SEGS(sbi))
351df4b2 714 break;
7cd8558b
JK		 715 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
716 start + 1);
29e59c14
CL		 717
718 for (i = start; i < end; i++)
719 clear_bit(i, prefree_map);
720
721 dirty_i->nr_dirty[PRE] -= end - start;
722
723 if (!test_opt(sbi, DISCARD))
724 continue;
351df4b2 725
37208879
JK		 726 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
727 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 728 }
729 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 730
731 /* send small discards */
2d7b822a 732 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 733 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
734 goto skip;
37208879 735 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 736 cpc->trimmed += entry->len;
836b5a63 737skip:
b2955550
JK		 738 list_del(&entry->list);
739 SM_I(sbi)->nr_discards -= entry->len;
740 kmem_cache_free(discard_entry_slab, entry);
741 }
351df4b2
JK		 742}
743
184a5cd2 744static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 745{
746 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 747
748 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 749 sit_i->dirty_sentries++;
184a5cd2
CY		 750 return false;
751 }
752
753 return true;
351df4b2
JK		 754}
755
756static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
757 unsigned int segno, int modified)
758{
759 struct seg_entry *se = get_seg_entry(sbi, segno);
760 se->type = type;
761 if (modified)
762 __mark_sit_entry_dirty(sbi, segno);
763}
764
765static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
766{
767 struct seg_entry *se;
768 unsigned int segno, offset;
769 long int new_vblocks;
770
771 segno = GET_SEGNO(sbi, blkaddr);
772
773 se = get_seg_entry(sbi, segno);
774 new_vblocks = se->valid_blocks + del;
491c0854 775 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 776
9850cf4a 777 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 778 (new_vblocks > sbi->blocks_per_seg)));
779
780 se->valid_blocks = new_vblocks;
781 se->mtime = get_mtime(sbi);
782 SIT_I(sbi)->max_mtime = se->mtime;
783
784 /* Update valid block bitmap */
785 if (del > 0) {
52aca074 786 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 787 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 788 if (!f2fs_test_and_set_bit(offset, se->discard_map))
789 sbi->discard_blks--;
351df4b2 790 } else {
52aca074 791 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 792 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 793 if (f2fs_test_and_clear_bit(offset, se->discard_map))
794 sbi->discard_blks++;
351df4b2
JK		 795 }
796 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
797 se->ckpt_valid_blocks += del;
798
799 __mark_sit_entry_dirty(sbi, segno);
800
801 /* update total number of valid blocks to be written in ckpt area */
802 SIT_I(sbi)->written_valid_blocks += del;
803
804 if (sbi->segs_per_sec > 1)
805 get_sec_entry(sbi, segno)->valid_blocks += del;
806}
807
5e443818 808void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 809{
5e443818
JK		 810 update_sit_entry(sbi, new, 1);
811 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
812 update_sit_entry(sbi, old, -1);
813
814 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
815 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 816}
817
818void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
819{
820 unsigned int segno = GET_SEGNO(sbi, addr);
821 struct sit_info *sit_i = SIT_I(sbi);
822
9850cf4a 823 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 824 if (addr == NEW_ADDR)
825 return;
826
827 /* add it into sit main buffer */
828 mutex_lock(&sit_i->sentry_lock);
829
830 update_sit_entry(sbi, addr, -1);
831
832 /* add it into dirty seglist */
833 locate_dirty_segment(sbi, segno);
834
835 mutex_unlock(&sit_i->sentry_lock);
836}
837
6e2c64ad
JK		 838bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
839{
840 struct sit_info *sit_i = SIT_I(sbi);
841 unsigned int segno, offset;
842 struct seg_entry *se;
843 bool is_cp = false;
844
845 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
846 return true;
847
848 mutex_lock(&sit_i->sentry_lock);
849
850 segno = GET_SEGNO(sbi, blkaddr);
851 se = get_seg_entry(sbi, segno);
852 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
853
854 if (f2fs_test_bit(offset, se->ckpt_valid_map))
855 is_cp = true;
856
857 mutex_unlock(&sit_i->sentry_lock);
858
859 return is_cp;
860}
861
0a8165d7 862/*
351df4b2
JK		 863 * This function should be resided under the curseg_mutex lock
864 */
865static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 866 struct f2fs_summary *sum)
351df4b2
JK		 867{
868 struct curseg_info *curseg = CURSEG_I(sbi, type);
869 void *addr = curseg->sum_blk;
e79efe3b 870 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 871 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 872}
873
0a8165d7 874/*
351df4b2
JK		 875 * Calculate the number of current summary pages for writing
876 */
3fa06d7b 877int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 878{
351df4b2 879 int valid_sum_count = 0;
9a47938b 880 int i, sum_in_page;
351df4b2
JK		 881
882 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
883 if (sbi->ckpt->alloc_type[i] == SSR)
884 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 885 else {
886 if (for_ra)
887 valid_sum_count += le16_to_cpu(
888 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
889 else
890 valid_sum_count += curseg_blkoff(sbi, i);
891 }
351df4b2
JK		 892 }
893
09cbfeaf 894 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 895 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
896 if (valid_sum_count <= sum_in_page)
351df4b2 897 return 1;
9a47938b 898 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 899 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 900 return 2;
901 return 3;
902}
903
0a8165d7 904/*
351df4b2
JK		 905 * Caller should put this summary page
906 */
907struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
908{
909 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
910}
911
381722d2 912void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 913{
914 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 915 void *dst = page_address(page);
916
917 if (src)
09cbfeaf 918 memcpy(dst, src, PAGE_SIZE);
381722d2 919 else
09cbfeaf 920 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 921 set_page_dirty(page);
922 f2fs_put_page(page, 1);
923}
924
381722d2
CY		 925static void write_sum_page(struct f2fs_sb_info *sbi,
926 struct f2fs_summary_block *sum_blk, block_t blk_addr)
927{
928 update_meta_page(sbi, (void *)sum_blk, blk_addr);
929}
930
b7ad7512
CY		 931static void write_current_sum_page(struct f2fs_sb_info *sbi,
932 int type, block_t blk_addr)
933{
934 struct curseg_info *curseg = CURSEG_I(sbi, type);
935 struct page *page = grab_meta_page(sbi, blk_addr);
936 struct f2fs_summary_block *src = curseg->sum_blk;
937 struct f2fs_summary_block *dst;
938
939 dst = (struct f2fs_summary_block *)page_address(page);
940
941 mutex_lock(&curseg->curseg_mutex);
942
943 down_read(&curseg->journal_rwsem);
944 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
945 up_read(&curseg->journal_rwsem);
946
947 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
948 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
949
950 mutex_unlock(&curseg->curseg_mutex);
951
952 set_page_dirty(page);
953 f2fs_put_page(page, 1);
954}
955
60374688
JK		 956static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
957{
958 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 959 unsigned int segno = curseg->segno + 1;
60374688
JK		 960 struct free_segmap_info *free_i = FREE_I(sbi);
961
7cd8558b 962 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 963 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 964 return 0;
965}
966
0a8165d7 967/*
351df4b2
JK		 968 * Find a new segment from the free segments bitmap to right order
969 * This function should be returned with success, otherwise BUG
970 */
971static void get_new_segment(struct f2fs_sb_info *sbi,
972 unsigned int *newseg, bool new_sec, int dir)
973{
974 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 975 unsigned int segno, secno, zoneno;
7cd8558b 976 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 977 unsigned int hint = *newseg / sbi->segs_per_sec;
978 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
979 unsigned int left_start = hint;
980 bool init = true;
981 int go_left = 0;
982 int i;
983
1a118ccf 984 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 985
986 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
987 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 988 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
989 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 990 goto got_it;
991 }
992find_other_zone:
7cd8558b
JK		 993 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
994 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 995 if (dir == ALLOC_RIGHT) {
996 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 997 MAIN_SECS(sbi), 0);
998 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 999 } else {
1000 go_left = 1;
1001 left_start = hint - 1;
1002 }
1003 }
1004 if (go_left == 0)
1005 goto skip_left;
1006
1007 while (test_bit(left_start, free_i->free_secmap)) {
1008 if (left_start > 0) {
1009 left_start--;
1010 continue;
1011 }
1012 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1013 MAIN_SECS(sbi), 0);
1014 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1015 break;
1016 }
1017 secno = left_start;
1018skip_left:
1019 hint = secno;
1020 segno = secno * sbi->segs_per_sec;
1021 zoneno = secno / sbi->secs_per_zone;
1022
1023 /* give up on finding another zone */
1024 if (!init)
1025 goto got_it;
1026 if (sbi->secs_per_zone == 1)
1027 goto got_it;
1028 if (zoneno == old_zoneno)
1029 goto got_it;
1030 if (dir == ALLOC_LEFT) {
1031 if (!go_left && zoneno + 1 >= total_zones)
1032 goto got_it;
1033 if (go_left && zoneno == 0)
1034 goto got_it;
1035 }
1036 for (i = 0; i < NR_CURSEG_TYPE; i++)
1037 if (CURSEG_I(sbi, i)->zone == zoneno)
1038 break;
1039
1040 if (i < NR_CURSEG_TYPE) {
1041 /* zone is in user, try another */
1042 if (go_left)
1043 hint = zoneno * sbi->secs_per_zone - 1;
1044 else if (zoneno + 1 >= total_zones)
1045 hint = 0;
1046 else
1047 hint = (zoneno + 1) * sbi->secs_per_zone;
1048 init = false;
1049 goto find_other_zone;
1050 }
1051got_it:
1052 /* set it as dirty segment in free segmap */
9850cf4a 1053 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1054 __set_inuse(sbi, segno);
1055 *newseg = segno;
1a118ccf 1056 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1057}
1058
1059static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1060{
1061 struct curseg_info *curseg = CURSEG_I(sbi, type);
1062 struct summary_footer *sum_footer;
1063
1064 curseg->segno = curseg->next_segno;
1065 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1066 curseg->next_blkoff = 0;
1067 curseg->next_segno = NULL_SEGNO;
1068
1069 sum_footer = &(curseg->sum_blk->footer);
1070 memset(sum_footer, 0, sizeof(struct summary_footer));
1071 if (IS_DATASEG(type))
1072 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1073 if (IS_NODESEG(type))
1074 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1075 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1076}
1077
0a8165d7 1078/*
351df4b2
JK		 1079 * Allocate a current working segment.
1080 * This function always allocates a free segment in LFS manner.
1081 */
1082static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1083{
1084 struct curseg_info *curseg = CURSEG_I(sbi, type);
1085 unsigned int segno = curseg->segno;
1086 int dir = ALLOC_LEFT;
1087
1088 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1089 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1090 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1091 dir = ALLOC_RIGHT;
1092
1093 if (test_opt(sbi, NOHEAP))
1094 dir = ALLOC_RIGHT;
1095
1096 get_new_segment(sbi, &segno, new_sec, dir);
1097 curseg->next_segno = segno;
1098 reset_curseg(sbi, type, 1);
1099 curseg->alloc_type = LFS;
1100}
1101
1102static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1103 struct curseg_info *seg, block_t start)
1104{
1105 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1106 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1107 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1108 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1109 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1110 int i, pos;
1111
1112 for (i = 0; i < entries; i++)
1113 target_map[i] = ckpt_map[i] | cur_map[i];
1114
1115 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1116
1117 seg->next_blkoff = pos;
351df4b2
JK		 1118}
1119
0a8165d7 1120/*
351df4b2
JK		 1121 * If a segment is written by LFS manner, next block offset is just obtained
1122 * by increasing the current block offset. However, if a segment is written by
1123 * SSR manner, next block offset obtained by calling __next_free_blkoff
1124 */
1125static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1126 struct curseg_info *seg)
1127{
1128 if (seg->alloc_type == SSR)
1129 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1130 else
1131 seg->next_blkoff++;
1132}
1133
0a8165d7 1134/*
e1c42045 1135 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1136 * manner, so it should recover the existing segment information of valid blocks
1137 */
1138static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1139{
1140 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1141 struct curseg_info *curseg = CURSEG_I(sbi, type);
1142 unsigned int new_segno = curseg->next_segno;
1143 struct f2fs_summary_block *sum_node;
1144 struct page *sum_page;
1145
1146 write_sum_page(sbi, curseg->sum_blk,
1147 GET_SUM_BLOCK(sbi, curseg->segno));
1148 __set_test_and_inuse(sbi, new_segno);
1149
1150 mutex_lock(&dirty_i->seglist_lock);
1151 __remove_dirty_segment(sbi, new_segno, PRE);
1152 __remove_dirty_segment(sbi, new_segno, DIRTY);
1153 mutex_unlock(&dirty_i->seglist_lock);
1154
1155 reset_curseg(sbi, type, 1);
1156 curseg->alloc_type = SSR;
1157 __next_free_blkoff(sbi, curseg, 0);
1158
1159 if (reuse) {
1160 sum_page = get_sum_page(sbi, new_segno);
1161 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1162 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1163 f2fs_put_page(sum_page, 1);
1164 }
1165}
1166
43727527
JK		 1167static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1168{
1169 struct curseg_info *curseg = CURSEG_I(sbi, type);
1170 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1171
1172 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1173 return v_ops->get_victim(sbi,
1174 &(curseg)->next_segno, BG_GC, type, SSR);
1175
1176 /* For data segments, let's do SSR more intensively */
1177 for (; type >= CURSEG_HOT_DATA; type--)
1178 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1179 BG_GC, type, SSR))
1180 return 1;
1181 return 0;
1182}
1183
351df4b2
JK		 1184/*
1185 * flush out current segment and replace it with new segment
1186 * This function should be returned with success, otherwise BUG
1187 */
1188static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1189 int type, bool force)
1190{
1191 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1192
7b405275 1193 if (force)
351df4b2 1194 new_curseg(sbi, type, true);
7b405275 1195 else if (type == CURSEG_WARM_NODE)
351df4b2 1196 new_curseg(sbi, type, false);
60374688
JK		 1197 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1198 new_curseg(sbi, type, false);
351df4b2
JK		 1199 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1200 change_curseg(sbi, type, true);
1201 else
1202 new_curseg(sbi, type, false);
dcdfff65
JK		 1203
1204 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1205}
1206
38aa0889
JK		 1207static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1208{
1209 struct curseg_info *curseg = CURSEG_I(sbi, type);
1210 unsigned int old_segno;
1211
1212 old_segno = curseg->segno;
1213 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1214 locate_dirty_segment(sbi, old_segno);
1215}
1216
351df4b2
JK		 1217void allocate_new_segments(struct f2fs_sb_info *sbi)
1218{
351df4b2
JK		 1219 int i;
1220
38aa0889
JK		 1221 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1222 __allocate_new_segments(sbi, i);
351df4b2
JK		 1223}
1224
1225static const struct segment_allocation default_salloc_ops = {
1226 .allocate_segment = allocate_segment_by_default,
1227};
1228
4b2fecc8
JK		 1229int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1230{
f7ef9b83
JK		 1231 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1232 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1233 unsigned int start_segno, end_segno;
1234 struct cp_control cpc;
c34f42e2 1235 int err = 0;
4b2fecc8 1236
836b5a63 1237 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1238 return -EINVAL;
1239
9bd27ae4 1240 cpc.trimmed = 0;
7cd8558b 1241 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1242 goto out;
1243
1244 /* start/end segment number in main_area */
7cd8558b
JK		 1245 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1246 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1247 GET_SEGNO(sbi, end);
4b2fecc8 1248 cpc.reason = CP_DISCARD;
836b5a63 1249 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1250
1251 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1252 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1253 cpc.trim_start = start_segno;
a66cdd98
JK		 1254
1255 if (sbi->discard_blks == 0)
1256 break;
1257 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1258 cpc.trim_end = end_segno;
1259 else
1260 cpc.trim_end = min_t(unsigned int,
1261 rounddown(start_segno +
bba681cb
JK		 1262 BATCHED_TRIM_SEGMENTS(sbi),
1263 sbi->segs_per_sec) - 1, end_segno);
1264
1265 mutex_lock(&sbi->gc_mutex);
c34f42e2 1266 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1267 mutex_unlock(&sbi->gc_mutex);
1268 }
4b2fecc8 1269out:
f7ef9b83 1270 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1271 return err;
4b2fecc8
JK		 1272}
1273
351df4b2
JK		 1274static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1275{
1276 struct curseg_info *curseg = CURSEG_I(sbi, type);
1277 if (curseg->next_blkoff < sbi->blocks_per_seg)
1278 return true;
1279 return false;
1280}
1281
1282static int __get_segment_type_2(struct page *page, enum page_type p_type)
1283{
1284 if (p_type == DATA)
1285 return CURSEG_HOT_DATA;
1286 else
1287 return CURSEG_HOT_NODE;
1288}
1289
1290static int __get_segment_type_4(struct page *page, enum page_type p_type)
1291{
1292 if (p_type == DATA) {
1293 struct inode *inode = page->mapping->host;
1294
1295 if (S_ISDIR(inode->i_mode))
1296 return CURSEG_HOT_DATA;
1297 else
1298 return CURSEG_COLD_DATA;
1299 } else {
a344b9fd
JK		 1300 if (IS_DNODE(page) && is_cold_node(page))
1301 return CURSEG_WARM_NODE;
351df4b2
JK		 1302 else
1303 return CURSEG_COLD_NODE;
1304 }
1305}
1306
1307static int __get_segment_type_6(struct page *page, enum page_type p_type)
1308{
1309 if (p_type == DATA) {
1310 struct inode *inode = page->mapping->host;
1311
1312 if (S_ISDIR(inode->i_mode))
1313 return CURSEG_HOT_DATA;
354a3399 1314 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1315 return CURSEG_COLD_DATA;
1316 else
1317 return CURSEG_WARM_DATA;
1318 } else {
1319 if (IS_DNODE(page))
1320 return is_cold_node(page) ? CURSEG_WARM_NODE :
1321 CURSEG_HOT_NODE;
1322 else
1323 return CURSEG_COLD_NODE;
1324 }
1325}
1326
1327static int __get_segment_type(struct page *page, enum page_type p_type)
1328{
4081363f 1329 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1330 case 2:
1331 return __get_segment_type_2(page, p_type);
1332 case 4:
1333 return __get_segment_type_4(page, p_type);
351df4b2 1334 }
12a67146 1335 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1336 f2fs_bug_on(F2FS_P_SB(page),
1337 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1338 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1339}
1340
bfad7c2d
JK		 1341void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1342 block_t old_blkaddr, block_t *new_blkaddr,
1343 struct f2fs_summary *sum, int type)
351df4b2
JK		 1344{
1345 struct sit_info *sit_i = SIT_I(sbi);
1346 struct curseg_info *curseg;
38aa0889
JK		 1347 bool direct_io = (type == CURSEG_DIRECT_IO);
1348
1349 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1350
351df4b2
JK		 1351 curseg = CURSEG_I(sbi, type);
1352
1353 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1354 mutex_lock(&sit_i->sentry_lock);
351df4b2 1355
38aa0889 1356 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1357 if (direct_io && curseg->next_blkoff &&
1358 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1359 __allocate_new_segments(sbi, type);
1360
351df4b2 1361 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1362
1363 /*
1364 * __add_sum_entry should be resided under the curseg_mutex
1365 * because, this function updates a summary entry in the
1366 * current summary block.
1367 */
e79efe3b 1368 __add_sum_entry(sbi, type, sum);
351df4b2 1369
351df4b2 1370 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1371
1372 stat_inc_block_count(sbi, curseg);
351df4b2 1373
5e443818
JK		 1374 if (!__has_curseg_space(sbi, type))
1375 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1376 /*
1377 * SIT information should be updated before segment allocation,
1378 * since SSR needs latest valid block information.
1379 */
1380 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1381
351df4b2
JK		 1382 mutex_unlock(&sit_i->sentry_lock);
1383
bfad7c2d 1384 if (page && IS_NODESEG(type))
351df4b2
JK		 1385 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1386
bfad7c2d
JK		 1387 mutex_unlock(&curseg->curseg_mutex);
1388}
1389
05ca3632 1390static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1391{
05ca3632 1392 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1393
7a9d7548
CY		 1394 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1395 &fio->new_blkaddr, sum, type);
bfad7c2d 1396
351df4b2 1397 /* writeout dirty page into bdev */
05ca3632 1398 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1399}
1400
577e3495 1401void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1402{
458e6197 1403 struct f2fs_io_info fio = {
05ca3632 1404 .sbi = sbi,
458e6197 1405 .type = META,
cf04e8eb 1406 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1407 .old_blkaddr = page->index,
1408 .new_blkaddr = page->index,
05ca3632 1409 .page = page,
4375a336 1410 .encrypted_page = NULL,
458e6197
JK		 1411 };
1412
2b947003
CY		 1413 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1414 fio.rw &= ~REQ_META;
1415
351df4b2 1416 set_page_writeback(page);
05ca3632 1417 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1418}
1419
05ca3632 1420void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1421{
1422 struct f2fs_summary sum;
05ca3632 1423
351df4b2 1424 set_summary(&sum, nid, 0, 0);
05ca3632 1425 do_write_page(&sum, fio);
351df4b2
JK		 1426}
1427
05ca3632 1428void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1429{
05ca3632 1430 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1431 struct f2fs_summary sum;
1432 struct node_info ni;
1433
9850cf4a 1434 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1435 get_node_info(sbi, dn->nid, &ni);
1436 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1437 do_write_page(&sum, fio);
f28b3434 1438 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1439}
1440
05ca3632 1441void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1442{
7a9d7548 1443 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1444 stat_inc_inplace_blocks(fio->sbi);
1445 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1446}
1447
4356e48e 1448void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1449 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1450 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1451{
1452 struct sit_info *sit_i = SIT_I(sbi);
1453 struct curseg_info *curseg;
1454 unsigned int segno, old_cursegno;
1455 struct seg_entry *se;
1456 int type;
19f106bc 1457 unsigned short old_blkoff;
351df4b2
JK		 1458
1459 segno = GET_SEGNO(sbi, new_blkaddr);
1460 se = get_seg_entry(sbi, segno);
1461 type = se->type;
1462
19f106bc
CY		 1463 if (!recover_curseg) {
1464 /* for recovery flow */
1465 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1466 if (old_blkaddr == NULL_ADDR)
1467 type = CURSEG_COLD_DATA;
1468 else
1469 type = CURSEG_WARM_DATA;
1470 }
1471 } else {
1472 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1473 type = CURSEG_WARM_DATA;
1474 }
19f106bc 1475
351df4b2
JK		 1476 curseg = CURSEG_I(sbi, type);
1477
1478 mutex_lock(&curseg->curseg_mutex);
1479 mutex_lock(&sit_i->sentry_lock);
1480
1481 old_cursegno = curseg->segno;
19f106bc 1482 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1483
1484 /* change the current segment */
1485 if (segno != curseg->segno) {
1486 curseg->next_segno = segno;
1487 change_curseg(sbi, type, true);
1488 }
1489
491c0854 1490 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1491 __add_sum_entry(sbi, type, sum);
351df4b2 1492
28bc106b 1493 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1494 update_sit_entry(sbi, new_blkaddr, 1);
1495 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1496 update_sit_entry(sbi, old_blkaddr, -1);
1497
1498 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1499 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1500
351df4b2 1501 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1502
19f106bc
CY		 1503 if (recover_curseg) {
1504 if (old_cursegno != curseg->segno) {
1505 curseg->next_segno = old_cursegno;
1506 change_curseg(sbi, type, true);
1507 }
1508 curseg->next_blkoff = old_blkoff;
1509 }
1510
351df4b2
JK		 1511 mutex_unlock(&sit_i->sentry_lock);
1512 mutex_unlock(&curseg->curseg_mutex);
1513}
1514
528e3459
CY		 1515void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1516 block_t old_addr, block_t new_addr,
28bc106b
CY		 1517 unsigned char version, bool recover_curseg,
1518 bool recover_newaddr)
528e3459
CY		 1519{
1520 struct f2fs_summary sum;
1521
1522 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1523
28bc106b
CY		 1524 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1525 recover_curseg, recover_newaddr);
528e3459 1526
f28b3434 1527 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1528}
1529
93dfe2ac 1530void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1531 enum page_type type, bool ordered)
93dfe2ac 1532{
93dfe2ac 1533 if (PageWriteback(page)) {
4081363f
JK		 1534 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1535
0c3a5797 1536 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1537 if (ordered)
1538 wait_on_page_writeback(page);
1539 else
1540 wait_for_stable_page(page);
93dfe2ac
JK		 1541 }
1542}
1543
08b39fbd
CY		 1544void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1545 block_t blkaddr)
1546{
1547 struct page *cpage;
1548
1549 if (blkaddr == NEW_ADDR)
1550 return;
1551
1552 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1553
1554 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1555 if (cpage) {
fec1d657 1556 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1557 f2fs_put_page(cpage, 1);
1558 }
1559}
1560
351df4b2
JK		 1561static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1562{
1563 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1564 struct curseg_info *seg_i;
1565 unsigned char *kaddr;
1566 struct page *page;
1567 block_t start;
1568 int i, j, offset;
1569
1570 start = start_sum_block(sbi);
1571
1572 page = get_meta_page(sbi, start++);
1573 kaddr = (unsigned char *)page_address(page);
1574
1575 /* Step 1: restore nat cache */
1576 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1577 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1578
1579 /* Step 2: restore sit cache */
1580 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1581 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1582 offset = 2 * SUM_JOURNAL_SIZE;
1583
1584 /* Step 3: restore summary entries */
1585 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1586 unsigned short blk_off;
1587 unsigned int segno;
1588
1589 seg_i = CURSEG_I(sbi, i);
1590 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1591 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1592 seg_i->next_segno = segno;
1593 reset_curseg(sbi, i, 0);
1594 seg_i->alloc_type = ckpt->alloc_type[i];
1595 seg_i->next_blkoff = blk_off;
1596
1597 if (seg_i->alloc_type == SSR)
1598 blk_off = sbi->blocks_per_seg;
1599
1600 for (j = 0; j < blk_off; j++) {
1601 struct f2fs_summary *s;
1602 s = (struct f2fs_summary *)(kaddr + offset);
1603 seg_i->sum_blk->entries[j] = *s;
1604 offset += SUMMARY_SIZE;
09cbfeaf 1605 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1606 SUM_FOOTER_SIZE)
1607 continue;
1608
1609 f2fs_put_page(page, 1);
1610 page = NULL;
1611
1612 page = get_meta_page(sbi, start++);
1613 kaddr = (unsigned char *)page_address(page);
1614 offset = 0;
1615 }
1616 }
1617 f2fs_put_page(page, 1);
1618 return 0;
1619}
1620
1621static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1622{
1623 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1624 struct f2fs_summary_block *sum;
1625 struct curseg_info *curseg;
1626 struct page *new;
1627 unsigned short blk_off;
1628 unsigned int segno = 0;
1629 block_t blk_addr = 0;
1630
1631 /* get segment number and block addr */
1632 if (IS_DATASEG(type)) {
1633 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1634 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1635 CURSEG_HOT_DATA]);
119ee914 1636 if (__exist_node_summaries(sbi))
351df4b2
JK		 1637 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1638 else
1639 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1640 } else {
1641 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1642 CURSEG_HOT_NODE]);
1643 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1644 CURSEG_HOT_NODE]);
119ee914 1645 if (__exist_node_summaries(sbi))
351df4b2
JK		 1646 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1647 type - CURSEG_HOT_NODE);
1648 else
1649 blk_addr = GET_SUM_BLOCK(sbi, segno);
1650 }
1651
1652 new = get_meta_page(sbi, blk_addr);
1653 sum = (struct f2fs_summary_block *)page_address(new);
1654
1655 if (IS_NODESEG(type)) {
119ee914 1656 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1657 struct f2fs_summary *ns = &sum->entries[0];
1658 int i;
1659 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1660 ns->version = 0;
1661 ns->ofs_in_node = 0;
1662 }
1663 } else {
d653788a
GZ		 1664 int err;
1665
1666 err = restore_node_summary(sbi, segno, sum);
1667 if (err) {
351df4b2 1668 f2fs_put_page(new, 1);
d653788a 1669 return err;
351df4b2
JK		 1670 }
1671 }
1672 }
1673
1674 /* set uncompleted segment to curseg */
1675 curseg = CURSEG_I(sbi, type);
1676 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1677
1678 /* update journal info */
1679 down_write(&curseg->journal_rwsem);
1680 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1681 up_write(&curseg->journal_rwsem);
1682
1683 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1684 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1685 curseg->next_segno = segno;
1686 reset_curseg(sbi, type, 0);
1687 curseg->alloc_type = ckpt->alloc_type[type];
1688 curseg->next_blkoff = blk_off;
1689 mutex_unlock(&curseg->curseg_mutex);
1690 f2fs_put_page(new, 1);
1691 return 0;
1692}
1693
1694static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1695{
1696 int type = CURSEG_HOT_DATA;
e4fc5fbf 1697 int err;
351df4b2 1698
25ca923b 1699 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1700 int npages = npages_for_summary_flush(sbi, true);
1701
1702 if (npages >= 2)
1703 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1704 META_CP, true);
3fa06d7b 1705
351df4b2
JK		 1706 /* restore for compacted data summary */
1707 if (read_compacted_summaries(sbi))
1708 return -EINVAL;
1709 type = CURSEG_HOT_NODE;
1710 }
1711
119ee914 1712 if (__exist_node_summaries(sbi))
3fa06d7b 1713 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1714 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1715
e4fc5fbf
CY		 1716 for (; type <= CURSEG_COLD_NODE; type++) {
1717 err = read_normal_summaries(sbi, type);
1718 if (err)
1719 return err;
1720 }
1721
351df4b2
JK		 1722 return 0;
1723}
1724
1725static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1726{
1727 struct page *page;
1728 unsigned char *kaddr;
1729 struct f2fs_summary *summary;
1730 struct curseg_info *seg_i;
1731 int written_size = 0;
1732 int i, j;
1733
1734 page = grab_meta_page(sbi, blkaddr++);
1735 kaddr = (unsigned char *)page_address(page);
1736
1737 /* Step 1: write nat cache */
1738 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1739 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1740 written_size += SUM_JOURNAL_SIZE;
1741
1742 /* Step 2: write sit cache */
1743 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1744 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1745 written_size += SUM_JOURNAL_SIZE;
1746
351df4b2
JK		 1747 /* Step 3: write summary entries */
1748 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1749 unsigned short blkoff;
1750 seg_i = CURSEG_I(sbi, i);
1751 if (sbi->ckpt->alloc_type[i] == SSR)
1752 blkoff = sbi->blocks_per_seg;
1753 else
1754 blkoff = curseg_blkoff(sbi, i);
1755
1756 for (j = 0; j < blkoff; j++) {
1757 if (!page) {
1758 page = grab_meta_page(sbi, blkaddr++);
1759 kaddr = (unsigned char *)page_address(page);
1760 written_size = 0;
1761 }
1762 summary = (struct f2fs_summary *)(kaddr + written_size);
1763 *summary = seg_i->sum_blk->entries[j];
1764 written_size += SUMMARY_SIZE;
351df4b2 1765
09cbfeaf 1766 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1767 SUM_FOOTER_SIZE)
1768 continue;
1769
e8d61a74 1770 set_page_dirty(page);
351df4b2
JK		 1771 f2fs_put_page(page, 1);
1772 page = NULL;
1773 }
1774 }
e8d61a74
CY		 1775 if (page) {
1776 set_page_dirty(page);
351df4b2 1777 f2fs_put_page(page, 1);
e8d61a74 1778 }
351df4b2
JK		 1779}
1780
1781static void write_normal_summaries(struct f2fs_sb_info *sbi,
1782 block_t blkaddr, int type)
1783{
1784 int i, end;
1785 if (IS_DATASEG(type))
1786 end = type + NR_CURSEG_DATA_TYPE;
1787 else
1788 end = type + NR_CURSEG_NODE_TYPE;
1789
b7ad7512
CY		 1790 for (i = type; i < end; i++)
1791 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1792}
1793
1794void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1795{
25ca923b 1796 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1797 write_compacted_summaries(sbi, start_blk);
1798 else
1799 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1800}
1801
1802void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1803{
119ee914 1804 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1805}
1806
dfc08a12 1807int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1808 unsigned int val, int alloc)
1809{
1810 int i;
1811
1812 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1813 for (i = 0; i < nats_in_cursum(journal); i++) {
1814 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1815 return i;
1816 }
dfc08a12
CY		 1817 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1818 return update_nats_in_cursum(journal, 1);
351df4b2 1819 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1820 for (i = 0; i < sits_in_cursum(journal); i++)
1821 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1822 return i;
dfc08a12
CY		 1823 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1824 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1825 }
1826 return -1;
1827}
1828
1829static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1830 unsigned int segno)
1831{
2cc22186 1832 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1833}
1834
1835static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1836 unsigned int start)
1837{
1838 struct sit_info *sit_i = SIT_I(sbi);
1839 struct page *src_page, *dst_page;
1840 pgoff_t src_off, dst_off;
1841 void *src_addr, *dst_addr;
1842
1843 src_off = current_sit_addr(sbi, start);
1844 dst_off = next_sit_addr(sbi, src_off);
1845
1846 /* get current sit block page without lock */
1847 src_page = get_meta_page(sbi, src_off);
1848 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1849 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1850
1851 src_addr = page_address(src_page);
1852 dst_addr = page_address(dst_page);
09cbfeaf 1853 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1854
1855 set_page_dirty(dst_page);
1856 f2fs_put_page(src_page, 1);
1857
1858 set_to_next_sit(sit_i, start);
1859
1860 return dst_page;
1861}
1862
184a5cd2
CY		 1863static struct sit_entry_set *grab_sit_entry_set(void)
1864{
1865 struct sit_entry_set *ses =
80c54505 1866 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1867
1868 ses->entry_cnt = 0;
1869 INIT_LIST_HEAD(&ses->set_list);
1870 return ses;
1871}
1872
1873static void release_sit_entry_set(struct sit_entry_set *ses)
1874{
1875 list_del(&ses->set_list);
1876 kmem_cache_free(sit_entry_set_slab, ses);
1877}
1878
1879static void adjust_sit_entry_set(struct sit_entry_set *ses,
1880 struct list_head *head)
1881{
1882 struct sit_entry_set *next = ses;
1883
1884 if (list_is_last(&ses->set_list, head))
1885 return;
1886
1887 list_for_each_entry_continue(next, head, set_list)
1888 if (ses->entry_cnt <= next->entry_cnt)
1889 break;
1890
1891 list_move_tail(&ses->set_list, &next->set_list);
1892}
1893
1894static void add_sit_entry(unsigned int segno, struct list_head *head)
1895{
1896 struct sit_entry_set *ses;
1897 unsigned int start_segno = START_SEGNO(segno);
1898
1899 list_for_each_entry(ses, head, set_list) {
1900 if (ses->start_segno == start_segno) {
1901 ses->entry_cnt++;
1902 adjust_sit_entry_set(ses, head);
1903 return;
1904 }
1905 }
1906
1907 ses = grab_sit_entry_set();
1908
1909 ses->start_segno = start_segno;
1910 ses->entry_cnt++;
1911 list_add(&ses->set_list, head);
1912}
1913
1914static void add_sits_in_set(struct f2fs_sb_info *sbi)
1915{
1916 struct f2fs_sm_info *sm_info = SM_I(sbi);
1917 struct list_head *set_list = &sm_info->sit_entry_set;
1918 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1919 unsigned int segno;
1920
7cd8558b 1921 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1922 add_sit_entry(segno, set_list);
1923}
1924
1925static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1926{
1927 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1928 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1929 int i;
1930
b7ad7512 1931 down_write(&curseg->journal_rwsem);
dfc08a12 1932 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1933 unsigned int segno;
1934 bool dirtied;
1935
dfc08a12 1936 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1937 dirtied = __mark_sit_entry_dirty(sbi, segno);
1938
1939 if (!dirtied)
1940 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1941 }
dfc08a12 1942 update_sits_in_cursum(journal, -i);
b7ad7512 1943 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1944}
1945
0a8165d7 1946/*
351df4b2
JK		 1947 * CP calls this function, which flushes SIT entries including sit_journal,
1948 * and moves prefree segs to free segs.
1949 */
4b2fecc8 1950void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1951{
1952 struct sit_info *sit_i = SIT_I(sbi);
1953 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1954 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1955 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 1956 struct sit_entry_set *ses, *tmp;
1957 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1958 bool to_journal = true;
4b2fecc8 1959 struct seg_entry *se;
351df4b2 1960
351df4b2
JK		 1961 mutex_lock(&sit_i->sentry_lock);
1962
2b11a74b
WL		 1963 if (!sit_i->dirty_sentries)
1964 goto out;
1965
351df4b2 1966 /*
184a5cd2
CY		 1967 * add and account sit entries of dirty bitmap in sit entry
1968 * set temporarily
351df4b2 1969 */
184a5cd2 1970 add_sits_in_set(sbi);
351df4b2 1971
184a5cd2
CY		 1972 /*
1973 * if there are no enough space in journal to store dirty sit
1974 * entries, remove all entries from journal and add and account
1975 * them in sit entry set.
1976 */
dfc08a12 1977 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 1978 remove_sits_in_journal(sbi);
b2955550 1979
184a5cd2
CY		 1980 /*
1981 * there are two steps to flush sit entries:
1982 * #1, flush sit entries to journal in current cold data summary block.
1983 * #2, flush sit entries to sit page.
1984 */
1985 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1986 struct page *page = NULL;
184a5cd2
CY		 1987 struct f2fs_sit_block *raw_sit = NULL;
1988 unsigned int start_segno = ses->start_segno;
1989 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1990 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1991 unsigned int segno = start_segno;
1992
1993 if (to_journal &&
dfc08a12 1994 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 1995 to_journal = false;
1996
b7ad7512
CY		 1997 if (to_journal) {
1998 down_write(&curseg->journal_rwsem);
1999 } else {
184a5cd2
CY		 2000 page = get_next_sit_page(sbi, start_segno);
2001 raw_sit = page_address(page);
351df4b2 2002 }
351df4b2 2003
184a5cd2
CY		 2004 /* flush dirty sit entries in region of current sit set */
2005 for_each_set_bit_from(segno, bitmap, end) {
2006 int offset, sit_offset;
4b2fecc8
JK		 2007
2008 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2009
2010 /* add discard candidates */
d7bc2484 2011 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2012 cpc->trim_start = segno;
2013 add_discard_addrs(sbi, cpc);
2014 }
184a5cd2
CY		 2015
2016 if (to_journal) {
dfc08a12 2017 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2018 SIT_JOURNAL, segno, 1);
2019 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2020 segno_in_journal(journal, offset) =
184a5cd2
CY		 2021 cpu_to_le32(segno);
2022 seg_info_to_raw_sit(se,
dfc08a12 2023 &sit_in_journal(journal, offset));
184a5cd2
CY		 2024 } else {
2025 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2026 seg_info_to_raw_sit(se,
2027 &raw_sit->entries[sit_offset]);
2028 }
351df4b2 2029
184a5cd2
CY		 2030 __clear_bit(segno, bitmap);
2031 sit_i->dirty_sentries--;
2032 ses->entry_cnt--;
351df4b2
JK		 2033 }
2034
b7ad7512
CY		 2035 if (to_journal)
2036 up_write(&curseg->journal_rwsem);
2037 else
184a5cd2
CY		 2038 f2fs_put_page(page, 1);
2039
2040 f2fs_bug_on(sbi, ses->entry_cnt);
2041 release_sit_entry_set(ses);
351df4b2 2042 }
184a5cd2
CY		 2043
2044 f2fs_bug_on(sbi, !list_empty(head));
2045 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2046out:
4b2fecc8
JK		 2047 if (cpc->reason == CP_DISCARD) {
2048 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2049 add_discard_addrs(sbi, cpc);
2050 }
351df4b2 2051 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2052
351df4b2
JK		 2053 set_prefree_as_free_segments(sbi);
2054}
2055
2056static int build_sit_info(struct f2fs_sb_info *sbi)
2057{
2058 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2059 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2060 struct sit_info *sit_i;
2061 unsigned int sit_segs, start;
2062 char *src_bitmap, *dst_bitmap;
2063 unsigned int bitmap_size;
2064
2065 /* allocate memory for SIT information */
2066 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2067 if (!sit_i)
2068 return -ENOMEM;
2069
2070 SM_I(sbi)->sit_info = sit_i;
2071
39307a8e
JK		 2072 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2073 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2074 if (!sit_i->sentries)
2075 return -ENOMEM;
2076
7cd8558b 2077 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2078 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2079 if (!sit_i->dirty_sentries_bitmap)
2080 return -ENOMEM;
2081
7cd8558b 2082 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2083 sit_i->sentries[start].cur_valid_map
2084 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2085 sit_i->sentries[start].ckpt_valid_map
2086 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2087 sit_i->sentries[start].discard_map
2088 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2089 if (!sit_i->sentries[start].cur_valid_map ||
2090 !sit_i->sentries[start].ckpt_valid_map ||
2091 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2092 return -ENOMEM;
2093 }
2094
60a3b782
JK		 2095 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2096 if (!sit_i->tmp_map)
2097 return -ENOMEM;
2098
351df4b2 2099 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2100 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2101 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2102 if (!sit_i->sec_entries)
2103 return -ENOMEM;
2104 }
2105
2106 /* get information related with SIT */
2107 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2108
2109 /* setup SIT bitmap from ckeckpoint pack */
2110 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2111 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2112
79b5793b 2113 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2114 if (!dst_bitmap)
2115 return -ENOMEM;
351df4b2
JK		 2116
2117 /* init SIT information */
2118 sit_i->s_ops = &default_salloc_ops;
2119
2120 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2121 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2122 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2123 sit_i->sit_bitmap = dst_bitmap;
2124 sit_i->bitmap_size = bitmap_size;
2125 sit_i->dirty_sentries = 0;
2126 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2127 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2128 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2129 mutex_init(&sit_i->sentry_lock);
2130 return 0;
2131}
2132
2133static int build_free_segmap(struct f2fs_sb_info *sbi)
2134{
351df4b2
JK		 2135 struct free_segmap_info *free_i;
2136 unsigned int bitmap_size, sec_bitmap_size;
2137
2138 /* allocate memory for free segmap information */
2139 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2140 if (!free_i)
2141 return -ENOMEM;
2142
2143 SM_I(sbi)->free_info = free_i;
2144
7cd8558b 2145 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2146 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2147 if (!free_i->free_segmap)
2148 return -ENOMEM;
2149
7cd8558b 2150 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2151 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2152 if (!free_i->free_secmap)
2153 return -ENOMEM;
2154
2155 /* set all segments as dirty temporarily */
2156 memset(free_i->free_segmap, 0xff, bitmap_size);
2157 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2158
2159 /* init free segmap information */
7cd8558b 2160 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2161 free_i->free_segments = 0;
2162 free_i->free_sections = 0;
1a118ccf 2163 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2164 return 0;
2165}
2166
2167static int build_curseg(struct f2fs_sb_info *sbi)
2168{
1042d60f 2169 struct curseg_info *array;
351df4b2
JK		 2170 int i;
2171
b434babf 2172 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2173 if (!array)
2174 return -ENOMEM;
2175
2176 SM_I(sbi)->curseg_array = array;
2177
2178 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2179 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2180 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2181 if (!array[i].sum_blk)
2182 return -ENOMEM;
b7ad7512
CY		 2183 init_rwsem(&array[i].journal_rwsem);
2184 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2185 GFP_KERNEL);
2186 if (!array[i].journal)
2187 return -ENOMEM;
351df4b2
JK		 2188 array[i].segno = NULL_SEGNO;
2189 array[i].next_blkoff = 0;
2190 }
2191 return restore_curseg_summaries(sbi);
2192}
2193
2194static void build_sit_entries(struct f2fs_sb_info *sbi)
2195{
2196 struct sit_info *sit_i = SIT_I(sbi);
2197 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2198 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2199 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2200 unsigned int i, start, end;
2201 unsigned int readed, start_blk = 0;
e9f5b8b8 2202 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2203
74de593a 2204 do {
26879fb1 2205 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2206
2207 start = start_blk * sit_i->sents_per_block;
2208 end = (start_blk + readed) * sit_i->sents_per_block;
2209
7cd8558b 2210 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2211 struct seg_entry *se = &sit_i->sentries[start];
2212 struct f2fs_sit_block *sit_blk;
2213 struct f2fs_sit_entry sit;
2214 struct page *page;
2215
b7ad7512 2216 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2217 for (i = 0; i < sits_in_cursum(journal); i++) {
2218 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2219 == start) {
dfc08a12 2220 sit = sit_in_journal(journal, i);
b7ad7512 2221 up_read(&curseg->journal_rwsem);
74de593a
CY		 2222 goto got_it;
2223 }
351df4b2 2224 }
b7ad7512 2225 up_read(&curseg->journal_rwsem);
74de593a
CY		 2226
2227 page = get_current_sit_page(sbi, start);
2228 sit_blk = (struct f2fs_sit_block *)page_address(page);
2229 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2230 f2fs_put_page(page, 1);
351df4b2 2231got_it:
74de593a
CY		 2232 check_block_count(sbi, start, &sit);
2233 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2234
2235 /* build discard map only one time */
2236 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2237 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2238
74de593a
CY		 2239 if (sbi->segs_per_sec > 1) {
2240 struct sec_entry *e = get_sec_entry(sbi, start);
2241 e->valid_blocks += se->valid_blocks;
2242 }
351df4b2 2243 }
74de593a
CY		 2244 start_blk += readed;
2245 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2246}
2247
2248static void init_free_segmap(struct f2fs_sb_info *sbi)
2249{
2250 unsigned int start;
2251 int type;
2252
7cd8558b 2253 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2254 struct seg_entry *sentry = get_seg_entry(sbi, start);
2255 if (!sentry->valid_blocks)
2256 __set_free(sbi, start);
2257 }
2258
2259 /* set use the current segments */
2260 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2261 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2262 __set_test_and_inuse(sbi, curseg_t->segno);
2263 }
2264}
2265
2266static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2267{
2268 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2269 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2270 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2271 unsigned short valid_blocks;
2272
8736fbf0 2273 while (1) {
351df4b2 2274 /* find dirty segment based on free segmap */
7cd8558b
JK		 2275 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2276 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2277 break;
2278 offset = segno + 1;
2279 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2280 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2281 continue;
ec325b52
JK		 2282 if (valid_blocks > sbi->blocks_per_seg) {
2283 f2fs_bug_on(sbi, 1);
2284 continue;
2285 }
351df4b2
JK		 2286 mutex_lock(&dirty_i->seglist_lock);
2287 __locate_dirty_segment(sbi, segno, DIRTY);
2288 mutex_unlock(&dirty_i->seglist_lock);
2289 }
2290}
2291
5ec4e49f 2292static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2293{
2294 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2295 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2296
39307a8e 2297 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2298 if (!dirty_i->victim_secmap)
351df4b2
JK		 2299 return -ENOMEM;
2300 return 0;
2301}
2302
2303static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2304{
2305 struct dirty_seglist_info *dirty_i;
2306 unsigned int bitmap_size, i;
2307
2308 /* allocate memory for dirty segments list information */
2309 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2310 if (!dirty_i)
2311 return -ENOMEM;
2312
2313 SM_I(sbi)->dirty_info = dirty_i;
2314 mutex_init(&dirty_i->seglist_lock);
2315
7cd8558b 2316 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2317
2318 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2319 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2320 if (!dirty_i->dirty_segmap[i])
2321 return -ENOMEM;
2322 }
2323
2324 init_dirty_segmap(sbi);
5ec4e49f 2325 return init_victim_secmap(sbi);
351df4b2
JK		 2326}
2327
0a8165d7 2328/*
351df4b2
JK		 2329 * Update min, max modified time for cost-benefit GC algorithm
2330 */
2331static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2332{
2333 struct sit_info *sit_i = SIT_I(sbi);
2334 unsigned int segno;
2335
2336 mutex_lock(&sit_i->sentry_lock);
2337
2338 sit_i->min_mtime = LLONG_MAX;
2339
7cd8558b 2340 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2341 unsigned int i;
2342 unsigned long long mtime = 0;
2343
2344 for (i = 0; i < sbi->segs_per_sec; i++)
2345 mtime += get_seg_entry(sbi, segno + i)->mtime;
2346
2347 mtime = div_u64(mtime, sbi->segs_per_sec);
2348
2349 if (sit_i->min_mtime > mtime)
2350 sit_i->min_mtime = mtime;
2351 }
2352 sit_i->max_mtime = get_mtime(sbi);
2353 mutex_unlock(&sit_i->sentry_lock);
2354}
2355
2356int build_segment_manager(struct f2fs_sb_info *sbi)
2357{
2358 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2359 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2360 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2361 int err;
2362
2363 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2364 if (!sm_info)
2365 return -ENOMEM;
2366
2367 /* init sm info */
2368 sbi->sm_info = sm_info;
351df4b2
JK		 2369 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2370 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2371 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2372 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2373 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2374 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2375 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2376 sm_info->rec_prefree_segments = sm_info->main_segments *
2377 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2378 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2379 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2380 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2381
7fd9e544
JK		 2382 INIT_LIST_HEAD(&sm_info->discard_list);
2383 sm_info->nr_discards = 0;
2384 sm_info->max_discards = 0;
2385
bba681cb
JK		 2386 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2387
184a5cd2
CY		 2388 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2389
b270ad6f 2390 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2391 err = create_flush_cmd_control(sbi);
2392 if (err)
a688b9d9 2393 return err;
6b4afdd7
JK		 2394 }
2395
351df4b2
JK		 2396 err = build_sit_info(sbi);
2397 if (err)
2398 return err;
2399 err = build_free_segmap(sbi);
2400 if (err)
2401 return err;
2402 err = build_curseg(sbi);
2403 if (err)
2404 return err;
2405
2406 /* reinit free segmap based on SIT */
2407 build_sit_entries(sbi);
2408
2409 init_free_segmap(sbi);
2410 err = build_dirty_segmap(sbi);
2411 if (err)
2412 return err;
2413
2414 init_min_max_mtime(sbi);
2415 return 0;
2416}
2417
2418static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2419 enum dirty_type dirty_type)
2420{
2421 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2422
2423 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2424 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2425 dirty_i->nr_dirty[dirty_type] = 0;
2426 mutex_unlock(&dirty_i->seglist_lock);
2427}
2428
5ec4e49f 2429static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2430{
2431 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2432 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2433}
2434
2435static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2436{
2437 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2438 int i;
2439
2440 if (!dirty_i)
2441 return;
2442
2443 /* discard pre-free/dirty segments list */
2444 for (i = 0; i < NR_DIRTY_TYPE; i++)
2445 discard_dirty_segmap(sbi, i);
2446
5ec4e49f 2447 destroy_victim_secmap(sbi);
351df4b2
JK		 2448 SM_I(sbi)->dirty_info = NULL;
2449 kfree(dirty_i);
2450}
2451
2452static void destroy_curseg(struct f2fs_sb_info *sbi)
2453{
2454 struct curseg_info *array = SM_I(sbi)->curseg_array;
2455 int i;
2456
2457 if (!array)
2458 return;
2459 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2460 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2461 kfree(array[i].sum_blk);
b7ad7512
CY		 2462 kfree(array[i].journal);
2463 }
351df4b2
JK		 2464 kfree(array);
2465}
2466
2467static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2468{
2469 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2470 if (!free_i)
2471 return;
2472 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2473 kvfree(free_i->free_segmap);
2474 kvfree(free_i->free_secmap);
351df4b2
JK		 2475 kfree(free_i);
2476}
2477
2478static void destroy_sit_info(struct f2fs_sb_info *sbi)
2479{
2480 struct sit_info *sit_i = SIT_I(sbi);
2481 unsigned int start;
2482
2483 if (!sit_i)
2484 return;
2485
2486 if (sit_i->sentries) {
7cd8558b 2487 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2488 kfree(sit_i->sentries[start].cur_valid_map);
2489 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2490 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2491 }
2492 }
60a3b782
JK		 2493 kfree(sit_i->tmp_map);
2494
39307a8e
JK		 2495 kvfree(sit_i->sentries);
2496 kvfree(sit_i->sec_entries);
2497 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2498
2499 SM_I(sbi)->sit_info = NULL;
2500 kfree(sit_i->sit_bitmap);
2501 kfree(sit_i);
2502}
2503
2504void destroy_segment_manager(struct f2fs_sb_info *sbi)
2505{
2506 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2507
3b03f724
CY		 2508 if (!sm_info)
2509 return;
2163d198 2510 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2511 destroy_dirty_segmap(sbi);
2512 destroy_curseg(sbi);
2513 destroy_free_segmap(sbi);
2514 destroy_sit_info(sbi);
2515 sbi->sm_info = NULL;
2516 kfree(sm_info);
2517}
7fd9e544
JK		 2518
2519int __init create_segment_manager_caches(void)
2520{
2521 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2522 sizeof(struct discard_entry));
7fd9e544 2523 if (!discard_entry_slab)
184a5cd2
CY		 2524 goto fail;
2525
2526 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2527 sizeof(struct sit_entry_set));
184a5cd2
CY		 2528 if (!sit_entry_set_slab)
2529 goto destory_discard_entry;
88b88a66
JK		 2530
2531 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2532 sizeof(struct inmem_pages));
2533 if (!inmem_entry_slab)
2534 goto destroy_sit_entry_set;
7fd9e544 2535 return 0;
184a5cd2 2536
88b88a66
JK		 2537destroy_sit_entry_set:
2538 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2539destory_discard_entry:
2540 kmem_cache_destroy(discard_entry_slab);
2541fail:
2542 return -ENOMEM;
7fd9e544
JK		 2543}
2544
2545void destroy_segment_manager_caches(void)
2546{
184a5cd2 2547 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2548 kmem_cache_destroy(discard_entry_slab);
88b88a66 2549 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2550}
ubuntu-bionic-kernel mirror