projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: catch up to v4.14-rc1
[GitHub/exynos8895/android_kernel_samsung_universal8895.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>
13f00235
JK		 19#include <linux/freezer.h>
20#include <linux/sched.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
13f00235 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL		 29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
eee3f1f5 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK		 36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL		 51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK		 69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL		 75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL		 80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL		 85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
c1286ff4 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK		 95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
c1286ff4 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
c1286ff4 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
c1286ff4
JK		 111
112 while (1) {
113 if (*p == 0)
114 goto pass;
115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
c1286ff4
JK		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
c1286ff4
JK		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
c1286ff4 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
c1286ff4
JK		 130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
c1286ff4 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
c1286ff4 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
c1286ff4
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
c1286ff4
JK		 153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
c1286ff4
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
c1286ff4 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
c1286ff4
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
13f00235
JK		 172bool need_SSR(struct f2fs_sb_info *sbi)
173{
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 2 * reserved_sections(sbi));
185}
186
88b88a66
JK		 187void register_inmem_page(struct inode *inode, struct page *page)
188{
189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
9be32d72 191
9e4ded3f 192 f2fs_trace_pid(page);
0722b101 193
decd36b6
CY		 194 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
195 SetPagePrivate(page);
196
88b88a66
JK		 197 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
198
199 /* add atomic page indices to the list */
200 new->page = page;
201 INIT_LIST_HEAD(&new->list);
decd36b6 202
88b88a66
JK		 203 /* increase reference count with clean state */
204 mutex_lock(&fi->inmem_lock);
205 get_page(page);
206 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 207 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 208 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 209
210 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 211}
212
c1286ff4
JK		 213static int __revoke_inmem_pages(struct inode *inode,
214 struct list_head *head, bool drop, bool recover)
215{
216 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
217 struct inmem_pages *cur, *tmp;
218 int err = 0;
219
220 list_for_each_entry_safe(cur, tmp, head, list) {
221 struct page *page = cur->page;
222
223 if (drop)
224 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
225
226 lock_page(page);
227
228 if (recover) {
229 struct dnode_of_data dn;
230 struct node_info ni;
231
232 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
13f00235 233retry:
c1286ff4 234 set_new_dnode(&dn, inode, NULL, NULL, 0);
13f00235
JK		 235 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
236 if (err) {
237 if (err == -ENOMEM) {
238 congestion_wait(BLK_RW_ASYNC, HZ/50);
239 cond_resched();
240 goto retry;
241 }
c1286ff4
JK		 242 err = -EAGAIN;
243 goto next;
244 }
245 get_node_info(sbi, dn.nid, &ni);
246 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
247 cur->old_addr, ni.version, true, true);
248 f2fs_put_dnode(&dn);
249 }
250next:
251 /* we don't need to invalidate this in the sccessful status */
252 if (drop || recover)
253 ClearPageUptodate(page);
254 set_page_private(page, 0);
255 ClearPagePrivate(page);
256 f2fs_put_page(page, 1);
257
258 list_del(&cur->list);
259 kmem_cache_free(inmem_entry_slab, cur);
260 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
261 }
262 return err;
263}
264
265void drop_inmem_pages(struct inode *inode)
266{
267 struct f2fs_inode_info *fi = F2FS_I(inode);
268
c1286ff4
JK		 269 mutex_lock(&fi->inmem_lock);
270 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
271 mutex_unlock(&fi->inmem_lock);
dc8b8cea
CY		 272
273 clear_inode_flag(inode, FI_ATOMIC_FILE);
13f00235 274 clear_inode_flag(inode, FI_HOT_DATA);
dc8b8cea 275 stat_dec_atomic_write(inode);
c1286ff4
JK		 276}
277
13f00235
JK		 278void drop_inmem_page(struct inode *inode, struct page *page)
279{
280 struct f2fs_inode_info *fi = F2FS_I(inode);
281 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
282 struct list_head *head = &fi->inmem_pages;
283 struct inmem_pages *cur = NULL;
284
285 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
286
287 mutex_lock(&fi->inmem_lock);
288 list_for_each_entry(cur, head, list) {
289 if (cur->page == page)
290 break;
291 }
292
293 f2fs_bug_on(sbi, !cur || cur->page != page);
294 list_del(&cur->list);
295 mutex_unlock(&fi->inmem_lock);
296
297 dec_page_count(sbi, F2FS_INMEM_PAGES);
298 kmem_cache_free(inmem_entry_slab, cur);
299
300 ClearPageUptodate(page);
301 set_page_private(page, 0);
302 ClearPagePrivate(page);
303 f2fs_put_page(page, 0);
304
305 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
306}
307
c1286ff4
JK		 308static int __commit_inmem_pages(struct inode *inode,
309 struct list_head *revoke_list)
88b88a66
JK		 310{
311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
312 struct f2fs_inode_info *fi = F2FS_I(inode);
313 struct inmem_pages *cur, *tmp;
88b88a66 314 struct f2fs_io_info fio = {
05ca3632 315 .sbi = sbi,
88b88a66 316 .type = DATA,
dc45fd9e 317 .op = REQ_OP_WRITE,
13f00235
JK		 318 .op_flags = REQ_SYNC | REQ_PRIO,
319 .io_type = FS_DATA_IO,
88b88a66 320 };
13f00235 321 pgoff_t last_idx = ULONG_MAX;
edb27dee 322 int err = 0;
88b88a66 323
88b88a66 324 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
c1286ff4
JK		 325 struct page *page = cur->page;
326
327 lock_page(page);
328 if (page->mapping == inode->i_mapping) {
329 trace_f2fs_commit_inmem_page(page, INMEM);
330
331 set_page_dirty(page);
332 f2fs_wait_on_page_writeback(page, DATA, true);
75bb19d8 333 if (clear_page_dirty_for_io(page)) {
c1286ff4 334 inode_dec_dirty_pages(inode);
75bb19d8
CY		 335 remove_dirty_inode(inode);
336 }
13f00235 337retry:
c1286ff4 338 fio.page = page;
13f00235
JK		 339 fio.old_blkaddr = NULL_ADDR;
340 fio.encrypted_page = NULL;
341 fio.need_lock = LOCK_DONE;
c1286ff4
JK		 342 err = do_write_data_page(&fio);
343 if (err) {
13f00235
JK		 344 if (err == -ENOMEM) {
345 congestion_wait(BLK_RW_ASYNC, HZ/50);
346 cond_resched();
347 goto retry;
348 }
c1286ff4
JK		 349 unlock_page(page);
350 break;
70c640b1 351 }
c1286ff4
JK		 352 /* record old blkaddr for revoking */
353 cur->old_addr = fio.old_blkaddr;
13f00235 354 last_idx = page->index;
88b88a66 355 }
c1286ff4
JK		 356 unlock_page(page);
357 list_move_tail(&cur->list, revoke_list);
358 }
decd36b6 359
13f00235
JK		 360 if (last_idx != ULONG_MAX)
361 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
c1286ff4
JK		 362
363 if (!err)
364 __revoke_inmem_pages(inode, revoke_list, false, false);
365
366 return err;
367}
368
369int commit_inmem_pages(struct inode *inode)
370{
371 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
372 struct f2fs_inode_info *fi = F2FS_I(inode);
373 struct list_head revoke_list;
374 int err;
375
376 INIT_LIST_HEAD(&revoke_list);
377 f2fs_balance_fs(sbi, true);
378 f2fs_lock_op(sbi);
379
dc8b8cea
CY		 380 set_inode_flag(inode, FI_ATOMIC_COMMIT);
381
c1286ff4
JK		 382 mutex_lock(&fi->inmem_lock);
383 err = __commit_inmem_pages(inode, &revoke_list);
384 if (err) {
385 int ret;
386 /*
387 * try to revoke all committed pages, but still we could fail
388 * due to no memory or other reason, if that happened, EAGAIN
389 * will be returned, which means in such case, transaction is
390 * already not integrity, caller should use journal to do the
391 * recovery or rewrite & commit last transaction. For other
392 * error number, revoking was done by filesystem itself.
393 */
394 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
395 if (ret)
396 err = ret;
397
398 /* drop all uncommitted pages */
399 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
88b88a66 400 }
88b88a66
JK		 401 mutex_unlock(&fi->inmem_lock);
402
dc8b8cea
CY		 403 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
404
c1286ff4 405 f2fs_unlock_op(sbi);
edb27dee 406 return err;
88b88a66
JK		 407}
408
0a8165d7 409/*
351df4b2
JK		 410 * This function balances dirty node and dentry pages.
411 * In addition, it controls garbage collection.
412 */
c1286ff4 413void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 414{
c1286ff4 415#ifdef CONFIG_F2FS_FAULT_INJECTION
13f00235
JK		 416 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
417 f2fs_show_injection_info(FAULT_CHECKPOINT);
c1286ff4 418 f2fs_stop_checkpoint(sbi, false);
13f00235 419 }
c1286ff4
JK		 420#endif
421
c1286ff4 422 /* balance_fs_bg is able to be pending */
13f00235 423 if (need && excess_cached_nats(sbi))
c1286ff4
JK		 424 f2fs_balance_fs_bg(sbi);
425
351df4b2 426 /*
029cd28c
JK		 427 * We should do GC or end up with checkpoint, if there are so many dirty
428 * dir/node pages without enough free segments.
351df4b2 429 */
c1286ff4 430 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 431 mutex_lock(&sbi->gc_mutex);
13f00235 432 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 433 }
434}
435
4660f9c0
JK		 436void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
437{
1dcc336b 438 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 439 if (!available_free_memory(sbi, EXTENT_CACHE))
440 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 441
1b38dc8e
JK		 442 /* check the # of cached NAT entries */
443 if (!available_free_memory(sbi, NAT_ENTRIES))
444 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
445
31696580 446 if (!available_free_memory(sbi, FREE_NIDS))
c1286ff4
JK		 447 try_to_free_nids(sbi, MAX_FREE_NIDS);
448 else
13f00235 449 build_free_nids(sbi, false, false);
31696580 450
13f00235 451 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
d1c2c357
JK		 452 return;
453
1b38dc8e
JK		 454 /* checkpoint is the only way to shrink partial cached entries */
455 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 456 !available_free_memory(sbi, INO_ENTRIES) ||
c1286ff4
JK		 457 excess_prefree_segs(sbi) ||
458 excess_dirty_nats(sbi) ||
d1c2c357 459 f2fs_time_over(sbi, CP_TIME)) {
c1286ff4
JK		 460 if (test_opt(sbi, DATA_FLUSH)) {
461 struct blk_plug plug;
462
463 blk_start_plug(&plug);
464 sync_dirty_inodes(sbi, FILE_INODE);
465 blk_finish_plug(&plug);
466 }
4660f9c0 467 f2fs_sync_fs(sbi->sb, true);
c1286ff4
JK		 468 stat_inc_bg_cp_count(sbi->stat_info);
469 }
4660f9c0
JK		 470}
471
13f00235
JK		 472static int __submit_flush_wait(struct f2fs_sb_info *sbi,
473 struct block_device *bdev)
07f01079
JK		 474{
475 struct bio *bio = f2fs_bio_alloc(0);
476 int ret;
477
dc45fd9e 478 bio->bi_rw = REQ_OP_WRITE;
07f01079
JK		 479 bio->bi_bdev = bdev;
480 ret = submit_bio_wait(WRITE_FLUSH, bio);
481 bio_put(bio);
13f00235
JK		 482
483 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
484 test_opt(sbi, FLUSH_MERGE), ret);
07f01079
JK		 485 return ret;
486}
487
488static int submit_flush_wait(struct f2fs_sb_info *sbi)
489{
13f00235 490 int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
07f01079
JK		 491 int i;
492
13f00235
JK		 493 if (!sbi->s_ndevs || ret)
494 return ret;
495
496 for (i = 1; i < sbi->s_ndevs; i++) {
497 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
498 if (ret)
499 break;
07f01079
JK		 500 }
501 return ret;
502}
503
2163d198 504static int issue_flush_thread(void *data)
6b4afdd7
JK		 505{
506 struct f2fs_sb_info *sbi = data;
eee3f1f5 507 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 508 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 509repeat:
510 if (kthread_should_stop())
511 return 0;
512
13f00235
JK		 513 sb_start_intwrite(sbi->sb);
514
721bd4d5 515 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 516 struct flush_cmd *cmd, *next;
517 int ret;
518
721bd4d5
GZ		 519 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
520 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
521
07f01079 522 ret = submit_flush_wait(sbi);
13f00235
JK		 523 atomic_inc(&fcc->issued_flush);
524
721bd4d5
GZ		 525 llist_for_each_entry_safe(cmd, next,
526 fcc->dispatch_list, llnode) {
6b4afdd7 527 cmd->ret = ret;
6b4afdd7
JK		 528 complete(&cmd->wait);
529 }
a688b9d9 530 fcc->dispatch_list = NULL;
6b4afdd7
JK		 531 }
532
13f00235
JK		 533 sb_end_intwrite(sbi->sb);
534
a688b9d9 535 wait_event_interruptible(*q,
721bd4d5 536 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 537 goto repeat;
538}
539
540int f2fs_issue_flush(struct f2fs_sb_info *sbi)
541{
eee3f1f5 542 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 543 struct flush_cmd cmd;
13f00235 544 int ret;
24a9ee0f 545
0f7b2abd
JK		 546 if (test_opt(sbi, NOBARRIER))
547 return 0;
548
13f00235
JK		 549 if (!test_opt(sbi, FLUSH_MERGE)) {
550 ret = submit_flush_wait(sbi);
551 atomic_inc(&fcc->issued_flush);
552 return ret;
553 }
740432f8 554
13f00235 555 if (atomic_inc_return(&fcc->issing_flush) == 1) {
07f01079 556 ret = submit_flush_wait(sbi);
13f00235
JK		 557 atomic_dec(&fcc->issing_flush);
558
559 atomic_inc(&fcc->issued_flush);
740432f8
JK		 560 return ret;
561 }
6b4afdd7 562
adf8d90b 563 init_completion(&cmd.wait);
6b4afdd7 564
721bd4d5 565 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 566
13f00235
JK		 567 /* update issue_list before we wake up issue_flush thread */
568 smp_mb();
569
570 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 571 wake_up(&fcc->flush_wait_queue);
6b4afdd7 572
be4b8492
JK		 573 if (fcc->f2fs_issue_flush) {
574 wait_for_completion(&cmd.wait);
13f00235 575 atomic_dec(&fcc->issing_flush);
be4b8492 576 } else {
13f00235
JK		 577 struct llist_node *list;
578
579 list = llist_del_all(&fcc->issue_list);
580 if (!list) {
581 wait_for_completion(&cmd.wait);
582 atomic_dec(&fcc->issing_flush);
583 } else {
584 struct flush_cmd *tmp, *next;
585
586 ret = submit_flush_wait(sbi);
587
588 llist_for_each_entry_safe(tmp, next, list, llnode) {
589 if (tmp == &cmd) {
590 cmd.ret = ret;
591 atomic_dec(&fcc->issing_flush);
592 continue;
593 }
594 tmp->ret = ret;
595 complete(&tmp->wait);
596 }
597 }
be4b8492 598 }
adf8d90b
CY		 599
600 return cmd.ret;
6b4afdd7
JK		 601}
602
2163d198
GZ		 603int create_flush_cmd_control(struct f2fs_sb_info *sbi)
604{
605 dev_t dev = sbi->sb->s_bdev->bd_dev;
606 struct flush_cmd_control *fcc;
607 int err = 0;
608
eee3f1f5
JK		 609 if (SM_I(sbi)->fcc_info) {
610 fcc = SM_I(sbi)->fcc_info;
13f00235
JK		 611 if (fcc->f2fs_issue_flush)
612 return err;
be4b8492
JK		 613 goto init_thread;
614 }
615
2163d198
GZ		 616 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
617 if (!fcc)
618 return -ENOMEM;
13f00235
JK		 619 atomic_set(&fcc->issued_flush, 0);
620 atomic_set(&fcc->issing_flush, 0);
2163d198 621 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 622 init_llist_head(&fcc->issue_list);
eee3f1f5 623 SM_I(sbi)->fcc_info = fcc;
13f00235
JK		 624 if (!test_opt(sbi, FLUSH_MERGE))
625 return err;
626
be4b8492 627init_thread:
2163d198
GZ		 628 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
629 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
630 if (IS_ERR(fcc->f2fs_issue_flush)) {
631 err = PTR_ERR(fcc->f2fs_issue_flush);
632 kfree(fcc);
eee3f1f5 633 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 634 return err;
635 }
2163d198
GZ		 636
637 return err;
638}
639
be4b8492 640void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 641{
eee3f1f5 642 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 643
be4b8492
JK		 644 if (fcc && fcc->f2fs_issue_flush) {
645 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
646
647 fcc->f2fs_issue_flush = NULL;
648 kthread_stop(flush_thread);
649 }
650 if (free) {
651 kfree(fcc);
eee3f1f5 652 SM_I(sbi)->fcc_info = NULL;
be4b8492 653 }
2163d198
GZ		 654}
655
351df4b2
JK		 656static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
657 enum dirty_type dirty_type)
658{
659 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
660
661 /* need not be added */
662 if (IS_CURSEG(sbi, segno))
663 return;
664
665 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
666 dirty_i->nr_dirty[dirty_type]++;
667
668 if (dirty_type == DIRTY) {
669 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 670 enum dirty_type t = sentry->type;
b2f2c390 671
ec325b52
JK		 672 if (unlikely(t >= DIRTY)) {
673 f2fs_bug_on(sbi, 1);
674 return;
675 }
4625d6aa
CL		 676 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
677 dirty_i->nr_dirty[t]++;
351df4b2
JK		 678 }
679}
680
681static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
682 enum dirty_type dirty_type)
683{
684 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
685
686 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
687 dirty_i->nr_dirty[dirty_type]--;
688
689 if (dirty_type == DIRTY) {
4625d6aa
CL		 690 struct seg_entry *sentry = get_seg_entry(sbi, segno);
691 enum dirty_type t = sentry->type;
692
693 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
694 dirty_i->nr_dirty[t]--;
b2f2c390 695
13f00235
JK		 696 if (get_valid_blocks(sbi, segno, true) == 0)
697 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 698 dirty_i->victim_secmap);
351df4b2
JK		 699 }
700}
701
0a8165d7 702/*
351df4b2
JK		 703 * Should not occur error such as -ENOMEM.
704 * Adding dirty entry into seglist is not critical operation.
705 * If a given segment is one of current working segments, it won't be added.
706 */
8d8451af 707static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 708{
709 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
710 unsigned short valid_blocks;
711
712 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
713 return;
714
715 mutex_lock(&dirty_i->seglist_lock);
716
13f00235 717 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 718
719 if (valid_blocks == 0) {
720 __locate_dirty_segment(sbi, segno, PRE);
721 __remove_dirty_segment(sbi, segno, DIRTY);
722 } else if (valid_blocks < sbi->blocks_per_seg) {
723 __locate_dirty_segment(sbi, segno, DIRTY);
724 } else {
725 /* Recovery routine with SSR needs this */
726 __remove_dirty_segment(sbi, segno, DIRTY);
727 }
728
729 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 730}
731
13f00235
JK		 732static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
733 struct block_device *bdev, block_t lstart,
734 block_t start, block_t len)
401c465b 735{
13f00235
JK		 736 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
737 struct list_head *pend_list;
eee3f1f5 738 struct discard_cmd *dc;
401c465b 739
13f00235
JK		 740 f2fs_bug_on(sbi, !len);
741
742 pend_list = &dcc->pend_list[plist_idx(len)];
743
eee3f1f5
JK		 744 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
745 INIT_LIST_HEAD(&dc->list);
13f00235 746 dc->bdev = bdev;
eee3f1f5 747 dc->lstart = lstart;
13f00235 748 dc->start = start;
eee3f1f5 749 dc->len = len;
13f00235
JK		 750 dc->ref = 0;
751 dc->state = D_PREP;
752 dc->error = 0;
eee3f1f5 753 init_completion(&dc->wait);
13f00235
JK		 754 list_add_tail(&dc->list, pend_list);
755 atomic_inc(&dcc->discard_cmd_cnt);
756 dcc->undiscard_blks += len;
401c465b 757
eee3f1f5 758 return dc;
401c465b
JK		 759}
760
13f00235
JK		 761static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
762 struct block_device *bdev, block_t lstart,
763 block_t start, block_t len,
764 struct rb_node *parent, struct rb_node **p)
401c465b 765{
13f00235
JK		 766 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
767 struct discard_cmd *dc;
401c465b 768
13f00235 769 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
132263dd 770
13f00235
JK		 771 rb_link_node(&dc->rb_node, parent, p);
772 rb_insert_color(&dc->rb_node, &dcc->root);
401c465b 773
13f00235
JK		 774 return dc;
775}
401c465b 776
13f00235
JK		 777static void __detach_discard_cmd(struct discard_cmd_control *dcc,
778 struct discard_cmd *dc)
779{
780 if (dc->state == D_DONE)
781 atomic_dec(&dcc->issing_discard);
782
783 list_del(&dc->list);
784 rb_erase(&dc->rb_node, &dcc->root);
785 dcc->undiscard_blks -= dc->len;
786
787 kmem_cache_free(discard_cmd_slab, dc);
788
789 atomic_dec(&dcc->discard_cmd_cnt);
790}
791
792static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
793 struct discard_cmd *dc)
794{
795 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
796
797 f2fs_bug_on(sbi, dc->ref);
798
799 if (dc->error == -EOPNOTSUPP)
800 dc->error = 0;
801
802 if (dc->error)
803 f2fs_msg(sbi->sb, KERN_INFO,
804 "Issue discard(%u, %u, %u) failed, ret: %d",
805 dc->lstart, dc->start, dc->len, dc->error);
806 __detach_discard_cmd(dcc, dc);
401c465b
JK		 807}
808
eee3f1f5 809static void f2fs_submit_discard_endio(struct bio *bio)
401c465b 810{
eee3f1f5 811 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
401c465b 812
13f00235
JK		 813 dc->error = bio->bi_error;
814 dc->state = D_DONE;
815 complete_all(&dc->wait);
816 bio_put(bio);
401c465b
JK		 817}
818
819/* copied from block/blk-lib.c in 4.10-rc1 */
820static int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
821 sector_t nr_sects, gfp_t gfp_mask, int flags,
822 struct bio **biop)
823{
824 struct request_queue *q = bdev_get_queue(bdev);
825 struct bio *bio = *biop;
826 unsigned int granularity;
827 int op = REQ_WRITE | REQ_DISCARD;
828 int alignment;
829 sector_t bs_mask;
830
831 if (!q)
832 return -ENXIO;
833
834 if (!blk_queue_discard(q))
835 return -EOPNOTSUPP;
836
837 if (flags & BLKDEV_DISCARD_SECURE) {
838 if (!blk_queue_secdiscard(q))
839 return -EOPNOTSUPP;
840 op |= REQ_SECURE;
841 }
842
843 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
844 if ((sector | nr_sects) & bs_mask)
845 return -EINVAL;
846
847 /* Zero-sector (unknown) and one-sector granularities are the same. */
848 granularity = max(q->limits.discard_granularity >> 9, 1U);
849 alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
850
851 while (nr_sects) {
852 unsigned int req_sects;
853 sector_t end_sect, tmp;
854
855 /* Make sure bi_size doesn't overflow */
856 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
857
858 /**
859 * If splitting a request, and the next starting sector would be
860 * misaligned, stop the discard at the previous aligned sector.
861 */
862 end_sect = sector + req_sects;
863 tmp = end_sect;
864 if (req_sects < nr_sects &&
865 sector_div(tmp, granularity) != alignment) {
866 end_sect = end_sect - alignment;
867 sector_div(end_sect, granularity);
868 end_sect = end_sect * granularity + alignment;
869 req_sects = end_sect - sector;
870 }
871
872 if (bio) {
13f00235 873 int ret = submit_bio_wait(op, bio);
401c465b
JK		 874 bio_put(bio);
875 if (ret)
876 return ret;
877 }
13f00235 878 bio = f2fs_bio_alloc(1);
401c465b
JK		 879 bio->bi_iter.bi_sector = sector;
880 bio->bi_bdev = bdev;
881 bio_set_op_attrs(bio, op, 0);
882
883 bio->bi_iter.bi_size = req_sects << 9;
884 nr_sects -= req_sects;
885 sector = end_sect;
886
887 /*
888 * We can loop for a long time in here, if someone does
889 * full device discards (like mkfs). Be nice and allow
890 * us to schedule out to avoid softlocking if preempt
891 * is disabled.
892 */
893 cond_resched();
894 }
895
896 *biop = bio;
897 return 0;
898}
899
13f00235
JK		 900void __check_sit_bitmap(struct f2fs_sb_info *sbi,
901 block_t start, block_t end)
902{
903#ifdef CONFIG_F2FS_CHECK_FS
904 struct seg_entry *sentry;
905 unsigned int segno;
906 block_t blk = start;
907 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
908 unsigned long *map;
909
910 while (blk < end) {
911 segno = GET_SEGNO(sbi, blk);
912 sentry = get_seg_entry(sbi, segno);
913 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
914
915 if (end < START_BLOCK(sbi, segno + 1))
916 size = GET_BLKOFF_FROM_SEG0(sbi, end);
917 else
918 size = max_blocks;
919 map = (unsigned long *)(sentry->cur_valid_map);
920 offset = __find_rev_next_bit(map, size, offset);
921 f2fs_bug_on(sbi, offset != size);
922 blk = START_BLOCK(sbi, segno + 1);
923 }
924#endif
925}
926
401c465b 927/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
13f00235
JK		 928static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
929 struct discard_cmd *dc)
401c465b 930{
13f00235 931 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
401c465b 932 struct bio *bio = NULL;
13f00235
JK		 933
934 if (dc->state != D_PREP)
935 return;
936
937 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
938
939 dc->error = __blkdev_issue_discard(dc->bdev,
940 SECTOR_FROM_BLOCK(dc->start),
941 SECTOR_FROM_BLOCK(dc->len),
942 GFP_NOFS, 0, &bio);
943 if (!dc->error) {
944 /* should keep before submission to avoid D_DONE right away */
945 dc->state = D_SUBMIT;
946 atomic_inc(&dcc->issued_discard);
947 atomic_inc(&dcc->issing_discard);
948 if (bio) {
949 bio->bi_private = dc;
950 bio->bi_end_io = f2fs_submit_discard_endio;
951 submit_bio(REQ_SYNC, bio);
952 list_move_tail(&dc->list, &dcc->wait_list);
953 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
954
955 f2fs_update_iostat(sbi, FS_DISCARD, 1);
956 }
957 } else {
958 __remove_discard_cmd(sbi, dc);
959 }
960}
961
962static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
963 struct block_device *bdev, block_t lstart,
964 block_t start, block_t len,
965 struct rb_node **insert_p,
966 struct rb_node *insert_parent)
967{
968 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
969 struct rb_node **p = &dcc->root.rb_node;
970 struct rb_node *parent = NULL;
971 struct discard_cmd *dc = NULL;
972
973 if (insert_p && insert_parent) {
974 parent = insert_parent;
975 p = insert_p;
976 goto do_insert;
977 }
978
979 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
980do_insert:
981 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
982 if (!dc)
983 return NULL;
984
985 return dc;
986}
987
988static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
989 struct discard_cmd *dc)
990{
991 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
992}
993
994static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
995 struct discard_cmd *dc, block_t blkaddr)
996{
997 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
998 struct discard_info di = dc->di;
999 bool modified = false;
1000
1001 if (dc->state == D_DONE || dc->len == 1) {
1002 __remove_discard_cmd(sbi, dc);
1003 return;
1004 }
1005
1006 dcc->undiscard_blks -= di.len;
1007
1008 if (blkaddr > di.lstart) {
1009 dc->len = blkaddr - dc->lstart;
1010 dcc->undiscard_blks += dc->len;
1011 __relocate_discard_cmd(dcc, dc);
1012 modified = true;
1013 }
1014
1015 if (blkaddr < di.lstart + di.len - 1) {
1016 if (modified) {
1017 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1018 di.start + blkaddr + 1 - di.lstart,
1019 di.lstart + di.len - 1 - blkaddr,
1020 NULL, NULL);
1021 } else {
1022 dc->lstart++;
1023 dc->len--;
1024 dc->start++;
1025 dcc->undiscard_blks += dc->len;
1026 __relocate_discard_cmd(dcc, dc);
1027 }
1028 }
1029}
1030
1031static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1032 struct block_device *bdev, block_t lstart,
1033 block_t start, block_t len)
1034{
1035 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1036 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1037 struct discard_cmd *dc;
1038 struct discard_info di = {0};
1039 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1040 block_t end = lstart + len;
1041
1042 mutex_lock(&dcc->cmd_lock);
1043
1044 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1045 NULL, lstart,
1046 (struct rb_entry **)&prev_dc,
1047 (struct rb_entry **)&next_dc,
1048 &insert_p, &insert_parent, true);
1049 if (dc)
1050 prev_dc = dc;
1051
1052 if (!prev_dc) {
1053 di.lstart = lstart;
1054 di.len = next_dc ? next_dc->lstart - lstart : len;
1055 di.len = min(di.len, len);
1056 di.start = start;
1057 }
1058
1059 while (1) {
1060 struct rb_node *node;
1061 bool merged = false;
1062 struct discard_cmd *tdc = NULL;
1063
1064 if (prev_dc) {
1065 di.lstart = prev_dc->lstart + prev_dc->len;
1066 if (di.lstart < lstart)
1067 di.lstart = lstart;
1068 if (di.lstart >= end)
1069 break;
1070
1071 if (!next_dc || next_dc->lstart > end)
1072 di.len = end - di.lstart;
1073 else
1074 di.len = next_dc->lstart - di.lstart;
1075 di.start = start + di.lstart - lstart;
1076 }
1077
1078 if (!di.len)
1079 goto next;
1080
1081 if (prev_dc && prev_dc->state == D_PREP &&
1082 prev_dc->bdev == bdev &&
1083 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1084 prev_dc->di.len += di.len;
1085 dcc->undiscard_blks += di.len;
1086 __relocate_discard_cmd(dcc, prev_dc);
1087 di = prev_dc->di;
1088 tdc = prev_dc;
1089 merged = true;
1090 }
1091
1092 if (next_dc && next_dc->state == D_PREP &&
1093 next_dc->bdev == bdev &&
1094 __is_discard_front_mergeable(&di, &next_dc->di)) {
1095 next_dc->di.lstart = di.lstart;
1096 next_dc->di.len += di.len;
1097 next_dc->di.start = di.start;
1098 dcc->undiscard_blks += di.len;
1099 __relocate_discard_cmd(dcc, next_dc);
1100 if (tdc)
1101 __remove_discard_cmd(sbi, tdc);
1102 merged = true;
1103 }
1104
1105 if (!merged) {
1106 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1107 di.len, NULL, NULL);
1108 }
1109 next:
1110 prev_dc = next_dc;
1111 if (!prev_dc)
1112 break;
1113
1114 node = rb_next(&prev_dc->rb_node);
1115 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1116 }
1117
1118 mutex_unlock(&dcc->cmd_lock);
1119}
1120
1121static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1122 struct block_device *bdev, block_t blkstart, block_t blklen)
1123{
132263dd 1124 block_t lblkstart = blkstart;
401c465b 1125
13f00235 1126 trace_f2fs_queue_discard(bdev, blkstart, blklen);
401c465b
JK		 1127
1128 if (sbi->s_ndevs) {
1129 int devi = f2fs_target_device_index(sbi, blkstart);
1130
1131 blkstart -= FDEV(devi).start_blk;
1132 }
13f00235
JK		 1133 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1134 return 0;
1135}
1136
1137static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
1138{
1139 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1140 struct list_head *pend_list;
1141 struct discard_cmd *dc, *tmp;
1142 struct blk_plug plug;
1143 int iter = 0, issued = 0;
1144 int i;
1145 bool io_interrupted = false;
1146
1147 mutex_lock(&dcc->cmd_lock);
1148 f2fs_bug_on(sbi,
1149 !__check_rb_tree_consistence(sbi, &dcc->root));
1150 blk_start_plug(&plug);
1151 for (i = MAX_PLIST_NUM - 1;
1152 i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
1153 pend_list = &dcc->pend_list[i];
1154 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1155 f2fs_bug_on(sbi, dc->state != D_PREP);
1156
1157 /* Hurry up to finish fstrim */
1158 if (dcc->pend_list_tag[i] & P_TRIM) {
1159 __submit_discard_cmd(sbi, dc);
1160 issued++;
1161
1162 if (fatal_signal_pending(current))
1163 break;
1164 continue;
1165 }
1166
1167 if (!issue_cond) {
1168 __submit_discard_cmd(sbi, dc);
1169 issued++;
1170 continue;
1171 }
1172
1173 if (is_idle(sbi)) {
1174 __submit_discard_cmd(sbi, dc);
1175 issued++;
1176 } else {
1177 io_interrupted = true;
1178 }
401c465b 1179
13f00235
JK		 1180 if (++iter >= DISCARD_ISSUE_RATE)
1181 goto out;
1182 }
1183 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1184 dcc->pend_list_tag[i] &= (~P_TRIM);
401c465b 1185 }
13f00235
JK		 1186out:
1187 blk_finish_plug(&plug);
1188 mutex_unlock(&dcc->cmd_lock);
1189
1190 if (!issued && io_interrupted)
1191 issued = -1;
1192
1193 return issued;
1194}
1195
1196static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1197{
1198 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1199 struct list_head *pend_list;
1200 struct discard_cmd *dc, *tmp;
1201 int i;
1202
1203 mutex_lock(&dcc->cmd_lock);
1204 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1205 pend_list = &dcc->pend_list[i];
1206 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1207 f2fs_bug_on(sbi, dc->state != D_PREP);
1208 __remove_discard_cmd(sbi, dc);
1209 }
1210 }
1211 mutex_unlock(&dcc->cmd_lock);
1212}
1213
1214static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1215 struct discard_cmd *dc)
1216{
1217 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1218
1219 wait_for_completion_io(&dc->wait);
1220 mutex_lock(&dcc->cmd_lock);
1221 f2fs_bug_on(sbi, dc->state != D_DONE);
1222 dc->ref--;
1223 if (!dc->ref)
1224 __remove_discard_cmd(sbi, dc);
1225 mutex_unlock(&dcc->cmd_lock);
1226}
1227
1228static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1229{
1230 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1231 struct list_head *wait_list = &(dcc->wait_list);
1232 struct discard_cmd *dc, *tmp;
1233 bool need_wait;
1234
1235next:
1236 need_wait = false;
1237
1238 mutex_lock(&dcc->cmd_lock);
1239 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1240 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
1241 wait_for_completion_io(&dc->wait);
1242 __remove_discard_cmd(sbi, dc);
1243 } else {
1244 dc->ref++;
1245 need_wait = true;
1246 break;
1247 }
1248 }
1249 mutex_unlock(&dcc->cmd_lock);
1250
1251 if (need_wait) {
1252 __wait_one_discard_bio(sbi, dc);
1253 goto next;
1254 }
1255}
1256
1257/* This should be covered by global mutex, &sit_i->sentry_lock */
1258void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1259{
1260 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1261 struct discard_cmd *dc;
1262 bool need_wait = false;
1263
1264 mutex_lock(&dcc->cmd_lock);
1265 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1266 if (dc) {
1267 if (dc->state == D_PREP) {
1268 __punch_discard_cmd(sbi, dc, blkaddr);
1269 } else {
1270 dc->ref++;
1271 need_wait = true;
1272 }
1273 }
1274 mutex_unlock(&dcc->cmd_lock);
1275
1276 if (need_wait)
1277 __wait_one_discard_bio(sbi, dc);
1278}
1279
1280void stop_discard_thread(struct f2fs_sb_info *sbi)
1281{
1282 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1283
1284 if (dcc && dcc->f2fs_issue_discard) {
1285 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1286
1287 dcc->f2fs_issue_discard = NULL;
1288 kthread_stop(discard_thread);
1289 }
1290}
1291
1292/* This comes from f2fs_put_super and f2fs_trim_fs */
1293void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1294{
1295 __issue_discard_cmd(sbi, false);
1296 __drop_discard_cmd(sbi);
1297 __wait_discard_cmd(sbi, false);
1298}
1299
1300static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1301{
1302 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1303 int i;
1304
1305 mutex_lock(&dcc->cmd_lock);
1306 for (i = 0; i < MAX_PLIST_NUM; i++)
1307 dcc->pend_list_tag[i] |= P_TRIM;
1308 mutex_unlock(&dcc->cmd_lock);
1309}
1310
1311static int issue_discard_thread(void *data)
1312{
1313 struct f2fs_sb_info *sbi = data;
1314 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1315 wait_queue_head_t *q = &dcc->discard_wait_queue;
1316 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1317 int issued;
1318
1319 set_freezable();
1320
1321 do {
1322 wait_event_interruptible_timeout(*q,
1323 kthread_should_stop() || freezing(current) ||
1324 dcc->discard_wake,
1325 msecs_to_jiffies(wait_ms));
1326 if (try_to_freeze())
1327 continue;
1328 if (kthread_should_stop())
1329 return 0;
1330
1331 if (dcc->discard_wake) {
1332 dcc->discard_wake = 0;
1333 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1334 mark_discard_range_all(sbi);
1335 }
1336
1337 sb_start_intwrite(sbi->sb);
1338
1339 issued = __issue_discard_cmd(sbi, true);
1340 if (issued) {
1341 __wait_discard_cmd(sbi, true);
1342 wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1343 } else {
1344 wait_ms = DEF_MAX_DISCARD_ISSUE_TIME;
1345 }
1346
1347 sb_end_intwrite(sbi->sb);
1348
1349 } while (!kthread_should_stop());
1350 return 0;
401c465b
JK		 1351}
1352
0573aa06 1353#ifdef CONFIG_BLK_DEV_ZONED
07f01079
JK		 1354static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1355 struct block_device *bdev, block_t blkstart, block_t blklen)
0573aa06 1356{
13f00235
JK		 1357 sector_t sector, nr_sects;
1358 block_t lblkstart = blkstart;
07f01079 1359 int devi = 0;
0573aa06 1360
07f01079
JK		 1361 if (sbi->s_ndevs) {
1362 devi = f2fs_target_device_index(sbi, blkstart);
1363 blkstart -= FDEV(devi).start_blk;
1364 }
0573aa06
DLM		 1365
1366 /*
1367 * We need to know the type of the zone: for conventional zones,
1368 * use regular discard if the drive supports it. For sequential
1369 * zones, reset the zone write pointer.
1370 */
07f01079 1371 switch (get_blkz_type(sbi, bdev, blkstart)) {
0573aa06
DLM		 1372
1373 case BLK_ZONE_TYPE_CONVENTIONAL:
1374 if (!blk_queue_discard(bdev_get_queue(bdev)))
1375 return 0;
13f00235 1376 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
0573aa06
DLM		 1377 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1378 case BLK_ZONE_TYPE_SEQWRITE_PREF:
13f00235
JK		 1379 sector = SECTOR_FROM_BLOCK(blkstart);
1380 nr_sects = SECTOR_FROM_BLOCK(blklen);
1381
1382 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1383 nr_sects != bdev_zone_sectors(bdev)) {
1384 f2fs_msg(sbi->sb, KERN_INFO,
1385 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1386 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1387 blkstart, blklen);
1388 return -EIO;
1389 }
1390 trace_f2fs_issue_reset_zone(bdev, blkstart);
0573aa06
DLM		 1391 return blkdev_reset_zones(bdev, sector,
1392 nr_sects, GFP_NOFS);
1393 default:
1394 /* Unknown zone type: broken device ? */
1395 return -EIO;
1396 }
1397}
1398#endif
1399
07f01079
JK		 1400static int __issue_discard_async(struct f2fs_sb_info *sbi,
1401 struct block_device *bdev, block_t blkstart, block_t blklen)
37208879 1402{
07f01079
JK		 1403#ifdef CONFIG_BLK_DEV_ZONED
1404 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1405 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1406 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1407#endif
13f00235 1408 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
07f01079
JK		 1409}
1410
1411static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1412 block_t blkstart, block_t blklen)
1413{
1414 sector_t start = blkstart, len = 0;
1415 struct block_device *bdev;
a66cdd98
JK		 1416 struct seg_entry *se;
1417 unsigned int offset;
1418 block_t i;
07f01079
JK		 1419 int err = 0;
1420
1421 bdev = f2fs_target_device(sbi, blkstart, NULL);
1422
1423 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1424 if (i != start) {
1425 struct block_device *bdev2 =
1426 f2fs_target_device(sbi, i, NULL);
1427
1428 if (bdev2 != bdev) {
1429 err = __issue_discard_async(sbi, bdev,
1430 start, len);
1431 if (err)
1432 return err;
1433 bdev = bdev2;
1434 start = i;
1435 len = 0;
1436 }
1437 }
a66cdd98 1438
a66cdd98
JK		 1439 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1440 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1441
1442 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1443 sbi->discard_blks--;
1444 }
0573aa06 1445
07f01079
JK		 1446 if (len)
1447 err = __issue_discard_async(sbi, bdev, start, len);
07f01079 1448 return err;
1e87a78d
JK		 1449}
1450
556f5ba3
JK		 1451static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1452 bool check_only)
adf4983b 1453{
b2955550
JK		 1454 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1455 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1456 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1457 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1458 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1459 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1460 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1461 unsigned int start = 0, end = -1;
13f00235
JK		 1462 bool force = (cpc->reason & CP_DISCARD);
1463 struct discard_entry *de = NULL;
1464 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1465 int i;
1466
c1286ff4 1467 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
556f5ba3 1468 return false;
b2955550 1469
a66cdd98
JK		 1470 if (!force) {
1471 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
13f00235
JK		 1472 SM_I(sbi)->dcc_info->nr_discards >=
1473 SM_I(sbi)->dcc_info->max_discards)
556f5ba3 1474 return false;
4b2fecc8
JK		 1475 }
1476
b2955550
JK		 1477 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1478 for (i = 0; i < entries; i++)
a66cdd98 1479 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1480 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1481
13f00235
JK		 1482 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1483 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1484 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1485 if (start >= max_blocks)
1486 break;
1487
1488 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c1286ff4
JK		 1489 if (force && start && end != max_blocks
1490 && (end - start) < cpc->trim_minlen)
1491 continue;
1492
556f5ba3
JK		 1493 if (check_only)
1494 return true;
1495
13f00235
JK		 1496 if (!de) {
1497 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1498 GFP_F2FS_ZERO);
1499 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1500 list_add_tail(&de->list, head);
1501 }
1502
1503 for (i = start; i < end; i++)
1504 __set_bit_le(i, (void *)de->discard_map);
1505
1506 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1507 }
556f5ba3 1508 return false;
b2955550
JK		 1509}
1510
4b2fecc8
JK		 1511void release_discard_addrs(struct f2fs_sb_info *sbi)
1512{
13f00235 1513 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1514 struct discard_entry *entry, *this;
1515
1516 /* drop caches */
1517 list_for_each_entry_safe(entry, this, head, list) {
1518 list_del(&entry->list);
1519 kmem_cache_free(discard_entry_slab, entry);
1520 }
1521}
1522
0a8165d7 1523/*
351df4b2
JK		 1524 * Should call clear_prefree_segments after checkpoint is done.
1525 */
1526static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1527{
1528 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1529 unsigned int segno;
351df4b2
JK		 1530
1531 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1532 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1533 __set_test_and_free(sbi, segno);
351df4b2
JK		 1534 mutex_unlock(&dirty_i->seglist_lock);
1535}
1536
836b5a63 1537void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1538{
13f00235
JK		 1539 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1540 struct list_head *head = &dcc->entry_list;
2d7b822a 1541 struct discard_entry *entry, *this;
351df4b2 1542 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1543 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1544 unsigned int start = 0, end = -1;
c1286ff4 1545 unsigned int secno, start_segno;
13f00235 1546 bool force = (cpc->reason & CP_DISCARD);
401c465b 1547
351df4b2 1548 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1549
351df4b2 1550 while (1) {
29e59c14 1551 int i;
7cd8558b
JK		 1552 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1553 if (start >= MAIN_SEGS(sbi))
351df4b2 1554 break;
7cd8558b
JK		 1555 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1556 start + 1);
29e59c14
CL		 1557
1558 for (i = start; i < end; i++)
1559 clear_bit(i, prefree_map);
1560
1561 dirty_i->nr_dirty[PRE] -= end - start;
1562
c70e14cd 1563 if (!test_opt(sbi, DISCARD))
29e59c14 1564 continue;
351df4b2 1565
c70e14cd
YH		 1566 if (force && start >= cpc->trim_start &&
1567 (end - 1) <= cpc->trim_end)
1568 continue;
1569
c1286ff4
JK		 1570 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1571 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1572 (end - start) << sbi->log_blocks_per_seg);
c1286ff4
JK		 1573 continue;
1574 }
1575next:
13f00235
JK		 1576 secno = GET_SEC_FROM_SEG(sbi, start);
1577 start_segno = GET_SEG_FROM_SEC(sbi, secno);
c1286ff4 1578 if (!IS_CURSEC(sbi, secno) &&
13f00235 1579 !get_valid_blocks(sbi, start, true))
c1286ff4
JK		 1580 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1581 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1582
1583 start = start_segno + sbi->segs_per_sec;
1584 if (start < end)
1585 goto next;
13f00235
JK		 1586 else
1587 end = start - 1;
351df4b2
JK		 1588 }
1589 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1590
1591 /* send small discards */
2d7b822a 1592 list_for_each_entry_safe(entry, this, head, list) {
13f00235
JK		 1593 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1594 bool is_valid = test_bit_le(0, entry->discard_map);
1595
1596find_next:
1597 if (is_valid) {
1598 next_pos = find_next_zero_bit_le(entry->discard_map,
1599 sbi->blocks_per_seg, cur_pos);
1600 len = next_pos - cur_pos;
1601
1602 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1603 (force && len < cpc->trim_minlen))
1604 goto skip;
1605
1606 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1607 len);
1608 cpc->trimmed += len;
1609 total_len += len;
1610 } else {
1611 next_pos = find_next_bit_le(entry->discard_map,
1612 sbi->blocks_per_seg, cur_pos);
1613 }
836b5a63 1614skip:
13f00235
JK		 1615 cur_pos = next_pos;
1616 is_valid = !is_valid;
1617
1618 if (cur_pos < sbi->blocks_per_seg)
1619 goto find_next;
1620
b2955550 1621 list_del(&entry->list);
13f00235 1622 dcc->nr_discards -= total_len;
b2955550
JK		 1623 kmem_cache_free(discard_entry_slab, entry);
1624 }
401c465b 1625
13f00235
JK		 1626 wake_up_discard_thread(sbi, false);
1627}
1628
1629static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1630{
1631 dev_t dev = sbi->sb->s_bdev->bd_dev;
1632 struct discard_cmd_control *dcc;
1633 int err = 0, i;
1634
1635 if (SM_I(sbi)->dcc_info) {
1636 dcc = SM_I(sbi)->dcc_info;
1637 goto init_thread;
1638 }
1639
1640 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1641 if (!dcc)
1642 return -ENOMEM;
1643
1644 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1645 INIT_LIST_HEAD(&dcc->entry_list);
1646 for (i = 0; i < MAX_PLIST_NUM; i++) {
1647 INIT_LIST_HEAD(&dcc->pend_list[i]);
1648 if (i >= dcc->discard_granularity - 1)
1649 dcc->pend_list_tag[i] |= P_ACTIVE;
1650 }
1651 INIT_LIST_HEAD(&dcc->wait_list);
1652 mutex_init(&dcc->cmd_lock);
1653 atomic_set(&dcc->issued_discard, 0);
1654 atomic_set(&dcc->issing_discard, 0);
1655 atomic_set(&dcc->discard_cmd_cnt, 0);
1656 dcc->nr_discards = 0;
1657 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1658 dcc->undiscard_blks = 0;
1659 dcc->root = RB_ROOT;
1660
1661 init_waitqueue_head(&dcc->discard_wait_queue);
1662 SM_I(sbi)->dcc_info = dcc;
1663init_thread:
1664 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1665 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1666 if (IS_ERR(dcc->f2fs_issue_discard)) {
1667 err = PTR_ERR(dcc->f2fs_issue_discard);
1668 kfree(dcc);
1669 SM_I(sbi)->dcc_info = NULL;
1670 return err;
1671 }
1672
1673 return err;
1674}
1675
1676static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1677{
1678 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1679
1680 if (!dcc)
1681 return;
1682
1683 stop_discard_thread(sbi);
1684
1685 kfree(dcc);
1686 SM_I(sbi)->dcc_info = NULL;
351df4b2
JK		 1687}
1688
184a5cd2 1689static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1690{
1691 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1692
1693 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1694 sit_i->dirty_sentries++;
184a5cd2
CY		 1695 return false;
1696 }
1697
1698 return true;
351df4b2
JK		 1699}
1700
1701static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1702 unsigned int segno, int modified)
1703{
1704 struct seg_entry *se = get_seg_entry(sbi, segno);
1705 se->type = type;
1706 if (modified)
1707 __mark_sit_entry_dirty(sbi, segno);
1708}
1709
1710static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1711{
1712 struct seg_entry *se;
1713 unsigned int segno, offset;
1714 long int new_vblocks;
13f00235
JK		 1715 bool exist;
1716#ifdef CONFIG_F2FS_CHECK_FS
1717 bool mir_exist;
1718#endif
351df4b2
JK		 1719
1720 segno = GET_SEGNO(sbi, blkaddr);
1721
1722 se = get_seg_entry(sbi, segno);
1723 new_vblocks = se->valid_blocks + del;
491c0854 1724 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1725
9850cf4a 1726 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1727 (new_vblocks > sbi->blocks_per_seg)));
1728
1729 se->valid_blocks = new_vblocks;
1730 se->mtime = get_mtime(sbi);
1731 SIT_I(sbi)->max_mtime = se->mtime;
1732
1733 /* Update valid block bitmap */
1734 if (del > 0) {
13f00235 1735 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
dd5804b2 1736#ifdef CONFIG_F2FS_CHECK_FS
13f00235
JK		 1737 mir_exist = f2fs_test_and_set_bit(offset,
1738 se->cur_valid_map_mir);
1739 if (unlikely(exist != mir_exist)) {
1740 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1741 "when setting bitmap, blk:%u, old bit:%d",
1742 blkaddr, exist);
05796763 1743 f2fs_bug_on(sbi, 1);
13f00235 1744 }
dd5804b2 1745#endif
13f00235
JK		 1746 if (unlikely(exist)) {
1747 f2fs_msg(sbi->sb, KERN_ERR,
1748 "Bitmap was wrongly set, blk:%u", blkaddr);
1749 f2fs_bug_on(sbi, 1);
1750 se->valid_blocks--;
1751 del = 0;
dd5804b2 1752 }
13f00235 1753
c1286ff4
JK		 1754 if (f2fs_discard_en(sbi) &&
1755 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1756 sbi->discard_blks--;
13f00235
JK		 1757
1758 /* don't overwrite by SSR to keep node chain */
1759 if (se->type == CURSEG_WARM_NODE) {
1760 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1761 se->ckpt_valid_blocks++;
1762 }
351df4b2 1763 } else {
13f00235 1764 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
dd5804b2 1765#ifdef CONFIG_F2FS_CHECK_FS
13f00235
JK		 1766 mir_exist = f2fs_test_and_clear_bit(offset,
1767 se->cur_valid_map_mir);
1768 if (unlikely(exist != mir_exist)) {
1769 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1770 "when clearing bitmap, blk:%u, old bit:%d",
1771 blkaddr, exist);
05796763 1772 f2fs_bug_on(sbi, 1);
13f00235 1773 }
dd5804b2 1774#endif
13f00235
JK		 1775 if (unlikely(!exist)) {
1776 f2fs_msg(sbi->sb, KERN_ERR,
1777 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1778 f2fs_bug_on(sbi, 1);
1779 se->valid_blocks++;
1780 del = 0;
dd5804b2 1781 }
13f00235 1782
c1286ff4
JK		 1783 if (f2fs_discard_en(sbi) &&
1784 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1785 sbi->discard_blks++;
351df4b2
JK		 1786 }
1787 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1788 se->ckpt_valid_blocks += del;
1789
1790 __mark_sit_entry_dirty(sbi, segno);
1791
1792 /* update total number of valid blocks to be written in ckpt area */
1793 SIT_I(sbi)->written_valid_blocks += del;
1794
1795 if (sbi->segs_per_sec > 1)
1796 get_sec_entry(sbi, segno)->valid_blocks += del;
1797}
1798
5e443818 1799void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1800{
5e443818
JK		 1801 update_sit_entry(sbi, new, 1);
1802 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1803 update_sit_entry(sbi, old, -1);
1804
1805 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1806 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1807}
1808
1809void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1810{
1811 unsigned int segno = GET_SEGNO(sbi, addr);
1812 struct sit_info *sit_i = SIT_I(sbi);
1813
9850cf4a 1814 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1815 if (addr == NEW_ADDR)
1816 return;
1817
1818 /* add it into sit main buffer */
1819 mutex_lock(&sit_i->sentry_lock);
1820
1821 update_sit_entry(sbi, addr, -1);
1822
1823 /* add it into dirty seglist */
1824 locate_dirty_segment(sbi, segno);
1825
1826 mutex_unlock(&sit_i->sentry_lock);
1827}
1828
6e2c64ad
JK		 1829bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1830{
1831 struct sit_info *sit_i = SIT_I(sbi);
1832 unsigned int segno, offset;
1833 struct seg_entry *se;
1834 bool is_cp = false;
1835
1836 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1837 return true;
1838
1839 mutex_lock(&sit_i->sentry_lock);
1840
1841 segno = GET_SEGNO(sbi, blkaddr);
1842 se = get_seg_entry(sbi, segno);
1843 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1844
1845 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1846 is_cp = true;
1847
1848 mutex_unlock(&sit_i->sentry_lock);
1849
1850 return is_cp;
1851}
1852
0a8165d7 1853/*
351df4b2
JK		 1854 * This function should be resided under the curseg_mutex lock
1855 */
1856static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1857 struct f2fs_summary *sum)
351df4b2
JK		 1858{
1859 struct curseg_info *curseg = CURSEG_I(sbi, type);
1860 void *addr = curseg->sum_blk;
e79efe3b 1861 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1862 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1863}
1864
0a8165d7 1865/*
351df4b2
JK		 1866 * Calculate the number of current summary pages for writing
1867 */
3fa06d7b 1868int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1869{
351df4b2 1870 int valid_sum_count = 0;
9a47938b 1871 int i, sum_in_page;
351df4b2
JK		 1872
1873 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1874 if (sbi->ckpt->alloc_type[i] == SSR)
1875 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1876 else {
1877 if (for_ra)
1878 valid_sum_count += le16_to_cpu(
1879 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1880 else
1881 valid_sum_count += curseg_blkoff(sbi, i);
1882 }
351df4b2
JK		 1883 }
1884
c1286ff4 1885 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1886 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1887 if (valid_sum_count <= sum_in_page)
351df4b2 1888 return 1;
9a47938b 1889 else if ((valid_sum_count - sum_in_page) <=
c1286ff4 1890 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1891 return 2;
1892 return 3;
1893}
1894
0a8165d7 1895/*
351df4b2
JK		 1896 * Caller should put this summary page
1897 */
1898struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1899{
1900 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1901}
1902
381722d2 1903void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1904{
1905 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1906 void *dst = page_address(page);
1907
1908 if (src)
c1286ff4 1909 memcpy(dst, src, PAGE_SIZE);
381722d2 1910 else
c1286ff4 1911 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1912 set_page_dirty(page);
1913 f2fs_put_page(page, 1);
1914}
1915
381722d2
CY		 1916static void write_sum_page(struct f2fs_sb_info *sbi,
1917 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1918{
1919 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1920}
1921
c1286ff4
JK		 1922static void write_current_sum_page(struct f2fs_sb_info *sbi,
1923 int type, block_t blk_addr)
1924{
1925 struct curseg_info *curseg = CURSEG_I(sbi, type);
1926 struct page *page = grab_meta_page(sbi, blk_addr);
1927 struct f2fs_summary_block *src = curseg->sum_blk;
1928 struct f2fs_summary_block *dst;
1929
1930 dst = (struct f2fs_summary_block *)page_address(page);
1931
1932 mutex_lock(&curseg->curseg_mutex);
1933
1934 down_read(&curseg->journal_rwsem);
1935 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1936 up_read(&curseg->journal_rwsem);
1937
1938 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1939 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1940
1941 mutex_unlock(&curseg->curseg_mutex);
1942
1943 set_page_dirty(page);
1944 f2fs_put_page(page, 1);
1945}
1946
60374688
JK		 1947static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1948{
1949 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 1950 unsigned int segno = curseg->segno + 1;
60374688
JK		 1951 struct free_segmap_info *free_i = FREE_I(sbi);
1952
7cd8558b 1953 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 1954 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 1955 return 0;
1956}
1957
0a8165d7 1958/*
351df4b2
JK		 1959 * Find a new segment from the free segments bitmap to right order
1960 * This function should be returned with success, otherwise BUG
1961 */
1962static void get_new_segment(struct f2fs_sb_info *sbi,
1963 unsigned int *newseg, bool new_sec, int dir)
1964{
1965 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1966 unsigned int segno, secno, zoneno;
7cd8558b 1967 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
13f00235
JK		 1968 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1969 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 1970 unsigned int left_start = hint;
1971 bool init = true;
1972 int go_left = 0;
1973 int i;
1974
1a118ccf 1975 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1976
1977 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1978 segno = find_next_zero_bit(free_i->free_segmap,
13f00235
JK		 1979 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1980 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 1981 goto got_it;
1982 }
1983find_other_zone:
7cd8558b
JK		 1984 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1985 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1986 if (dir == ALLOC_RIGHT) {
1987 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1988 MAIN_SECS(sbi), 0);
1989 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1990 } else {
1991 go_left = 1;
1992 left_start = hint - 1;
1993 }
1994 }
1995 if (go_left == 0)
1996 goto skip_left;
1997
1998 while (test_bit(left_start, free_i->free_secmap)) {
1999 if (left_start > 0) {
2000 left_start--;
2001 continue;
2002 }
2003 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2004 MAIN_SECS(sbi), 0);
2005 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2006 break;
2007 }
2008 secno = left_start;
2009skip_left:
2010 hint = secno;
13f00235
JK		 2011 segno = GET_SEG_FROM_SEC(sbi, secno);
2012 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2013
2014 /* give up on finding another zone */
2015 if (!init)
2016 goto got_it;
2017 if (sbi->secs_per_zone == 1)
2018 goto got_it;
2019 if (zoneno == old_zoneno)
2020 goto got_it;
2021 if (dir == ALLOC_LEFT) {
2022 if (!go_left && zoneno + 1 >= total_zones)
2023 goto got_it;
2024 if (go_left && zoneno == 0)
2025 goto got_it;
2026 }
2027 for (i = 0; i < NR_CURSEG_TYPE; i++)
2028 if (CURSEG_I(sbi, i)->zone == zoneno)
2029 break;
2030
2031 if (i < NR_CURSEG_TYPE) {
2032 /* zone is in user, try another */
2033 if (go_left)
2034 hint = zoneno * sbi->secs_per_zone - 1;
2035 else if (zoneno + 1 >= total_zones)
2036 hint = 0;
2037 else
2038 hint = (zoneno + 1) * sbi->secs_per_zone;
2039 init = false;
2040 goto find_other_zone;
2041 }
2042got_it:
2043 /* set it as dirty segment in free segmap */
9850cf4a 2044 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2045 __set_inuse(sbi, segno);
2046 *newseg = segno;
1a118ccf 2047 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2048}
2049
2050static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2051{
2052 struct curseg_info *curseg = CURSEG_I(sbi, type);
2053 struct summary_footer *sum_footer;
2054
2055 curseg->segno = curseg->next_segno;
13f00235 2056 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2057 curseg->next_blkoff = 0;
2058 curseg->next_segno = NULL_SEGNO;
2059
2060 sum_footer = &(curseg->sum_blk->footer);
2061 memset(sum_footer, 0, sizeof(struct summary_footer));
2062 if (IS_DATASEG(type))
2063 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2064 if (IS_NODESEG(type))
2065 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2066 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2067}
2068
13f00235
JK		 2069static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2070{
2071 /* if segs_per_sec is large than 1, we need to keep original policy. */
2072 if (sbi->segs_per_sec != 1)
2073 return CURSEG_I(sbi, type)->segno;
2074
2075 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
2076 return 0;
2077
2078 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2079 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2080 return CURSEG_I(sbi, type)->segno;
2081}
2082
0a8165d7 2083/*
351df4b2
JK		 2084 * Allocate a current working segment.
2085 * This function always allocates a free segment in LFS manner.
2086 */
2087static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2088{
2089 struct curseg_info *curseg = CURSEG_I(sbi, type);
2090 unsigned int segno = curseg->segno;
2091 int dir = ALLOC_LEFT;
2092
2093 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2094 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2095 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2096 dir = ALLOC_RIGHT;
2097
2098 if (test_opt(sbi, NOHEAP))
2099 dir = ALLOC_RIGHT;
2100
13f00235 2101 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2102 get_new_segment(sbi, &segno, new_sec, dir);
2103 curseg->next_segno = segno;
2104 reset_curseg(sbi, type, 1);
2105 curseg->alloc_type = LFS;
2106}
2107
2108static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2109 struct curseg_info *seg, block_t start)
2110{
2111 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2112 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2113 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2114 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2115 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2116 int i, pos;
2117
2118 for (i = 0; i < entries; i++)
2119 target_map[i] = ckpt_map[i] | cur_map[i];
2120
2121 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2122
2123 seg->next_blkoff = pos;
351df4b2
JK		 2124}
2125
0a8165d7 2126/*
351df4b2
JK		 2127 * If a segment is written by LFS manner, next block offset is just obtained
2128 * by increasing the current block offset. However, if a segment is written by
2129 * SSR manner, next block offset obtained by calling __next_free_blkoff
2130 */
2131static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2132 struct curseg_info *seg)
2133{
2134 if (seg->alloc_type == SSR)
2135 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2136 else
2137 seg->next_blkoff++;
2138}
2139
0a8165d7 2140/*
e1c42045 2141 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2142 * manner, so it should recover the existing segment information of valid blocks
2143 */
13f00235 2144static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2145{
2146 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2147 struct curseg_info *curseg = CURSEG_I(sbi, type);
2148 unsigned int new_segno = curseg->next_segno;
2149 struct f2fs_summary_block *sum_node;
2150 struct page *sum_page;
2151
2152 write_sum_page(sbi, curseg->sum_blk,
2153 GET_SUM_BLOCK(sbi, curseg->segno));
2154 __set_test_and_inuse(sbi, new_segno);
2155
2156 mutex_lock(&dirty_i->seglist_lock);
2157 __remove_dirty_segment(sbi, new_segno, PRE);
2158 __remove_dirty_segment(sbi, new_segno, DIRTY);
2159 mutex_unlock(&dirty_i->seglist_lock);
2160
2161 reset_curseg(sbi, type, 1);
2162 curseg->alloc_type = SSR;
2163 __next_free_blkoff(sbi, curseg, 0);
2164
13f00235
JK		 2165 sum_page = get_sum_page(sbi, new_segno);
2166 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2167 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2168 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2169}
2170
43727527
JK		 2171static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2172{
2173 struct curseg_info *curseg = CURSEG_I(sbi, type);
2174 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
13f00235
JK		 2175 unsigned segno = NULL_SEGNO;
2176 int i, cnt;
2177 bool reversed = false;
2178
2179 /* need_SSR() already forces to do this */
2180 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2181 curseg->next_segno = segno;
2182 return 1;
2183 }
43727527 2184
13f00235
JK		 2185 /* For node segments, let's do SSR more intensively */
2186 if (IS_NODESEG(type)) {
2187 if (type >= CURSEG_WARM_NODE) {
2188 reversed = true;
2189 i = CURSEG_COLD_NODE;
2190 } else {
2191 i = CURSEG_HOT_NODE;
2192 }
2193 cnt = NR_CURSEG_NODE_TYPE;
2194 } else {
2195 if (type >= CURSEG_WARM_DATA) {
2196 reversed = true;
2197 i = CURSEG_COLD_DATA;
2198 } else {
2199 i = CURSEG_HOT_DATA;
2200 }
2201 cnt = NR_CURSEG_DATA_TYPE;
2202 }
43727527 2203
13f00235
JK		 2204 for (; cnt-- > 0; reversed ? i-- : i++) {
2205 if (i == type)
2206 continue;
2207 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2208 curseg->next_segno = segno;
43727527 2209 return 1;
13f00235
JK		 2210 }
2211 }
43727527
JK		 2212 return 0;
2213}
2214
351df4b2
JK		 2215/*
2216 * flush out current segment and replace it with new segment
2217 * This function should be returned with success, otherwise BUG
2218 */
2219static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2220 int type, bool force)
2221{
2222 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2223
7b405275 2224 if (force)
351df4b2 2225 new_curseg(sbi, type, true);
13f00235
JK		 2226 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2227 type == CURSEG_WARM_NODE)
351df4b2 2228 new_curseg(sbi, type, false);
60374688
JK		 2229 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2230 new_curseg(sbi, type, false);
351df4b2 2231 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
13f00235 2232 change_curseg(sbi, type);
351df4b2
JK		 2233 else
2234 new_curseg(sbi, type, false);
dcdfff65
JK		 2235
2236 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2237}
2238
2239void allocate_new_segments(struct f2fs_sb_info *sbi)
2240{
00e5a211
JK		 2241 struct curseg_info *curseg;
2242 unsigned int old_segno;
351df4b2
JK		 2243 int i;
2244
00e5a211
JK		 2245 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2246 curseg = CURSEG_I(sbi, i);
2247 old_segno = curseg->segno;
2248 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2249 locate_dirty_segment(sbi, old_segno);
2250 }
351df4b2
JK		 2251}
2252
2253static const struct segment_allocation default_salloc_ops = {
2254 .allocate_segment = allocate_segment_by_default,
2255};
2256
556f5ba3
JK		 2257bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2258{
2259 __u64 trim_start = cpc->trim_start;
13f00235 2260 bool has_candidate = false;
556f5ba3
JK		 2261
2262 mutex_lock(&SIT_I(sbi)->sentry_lock);
13f00235
JK		 2263 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2264 if (add_discard_addrs(sbi, cpc, true)) {
2265 has_candidate = true;
556f5ba3 2266 break;
13f00235
JK		 2267 }
2268 }
556f5ba3
JK		 2269 mutex_unlock(&SIT_I(sbi)->sentry_lock);
2270
13f00235
JK		 2271 cpc->trim_start = trim_start;
2272 return has_candidate;
556f5ba3
JK		 2273}
2274
4b2fecc8
JK		 2275int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2276{
f7ef9b83
JK		 2277 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2278 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 2279 unsigned int start_segno, end_segno;
2280 struct cp_control cpc;
c1286ff4 2281 int err = 0;
4b2fecc8 2282
836b5a63 2283 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2284 return -EINVAL;
2285
9bd27ae4 2286 cpc.trimmed = 0;
7cd8558b 2287 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2288 goto out;
2289
c1286ff4
JK		 2290 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2291 f2fs_msg(sbi->sb, KERN_WARNING,
2292 "Found FS corruption, run fsck to fix.");
2293 goto out;
2294 }
2295
4b2fecc8 2296 /* start/end segment number in main_area */
7cd8558b
JK		 2297 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2298 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2299 GET_SEGNO(sbi, end);
4b2fecc8 2300 cpc.reason = CP_DISCARD;
836b5a63 2301 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2302
2303 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 2304 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2305 cpc.trim_start = start_segno;
a66cdd98
JK		 2306
2307 if (sbi->discard_blks == 0)
2308 break;
2309 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2310 cpc.trim_end = end_segno;
2311 else
2312 cpc.trim_end = min_t(unsigned int,
2313 rounddown(start_segno +
bba681cb
JK		 2314 BATCHED_TRIM_SEGMENTS(sbi),
2315 sbi->segs_per_sec) - 1, end_segno);
2316
2317 mutex_lock(&sbi->gc_mutex);
c1286ff4 2318 err = write_checkpoint(sbi, &cpc);
bba681cb 2319 mutex_unlock(&sbi->gc_mutex);
c1286ff4
JK		 2320 if (err)
2321 break;
2322
2323 schedule();
bba681cb 2324 }
13f00235
JK		 2325 /* It's time to issue all the filed discards */
2326 mark_discard_range_all(sbi);
2327 f2fs_wait_discard_bios(sbi);
4b2fecc8 2328out:
f7ef9b83 2329 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c1286ff4 2330 return err;
4b2fecc8
JK		 2331}
2332
351df4b2
JK		 2333static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2334{
2335 struct curseg_info *curseg = CURSEG_I(sbi, type);
2336 if (curseg->next_blkoff < sbi->blocks_per_seg)
2337 return true;
2338 return false;
2339}
2340
13f00235 2341static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2342{
13f00235 2343 if (fio->type == DATA)
351df4b2
JK		 2344 return CURSEG_HOT_DATA;
2345 else
2346 return CURSEG_HOT_NODE;
2347}
2348
13f00235 2349static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2350{
13f00235
JK		 2351 if (fio->type == DATA) {
2352 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2353
2354 if (S_ISDIR(inode->i_mode))
2355 return CURSEG_HOT_DATA;
2356 else
2357 return CURSEG_COLD_DATA;
2358 } else {
13f00235 2359 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2360 return CURSEG_WARM_NODE;
351df4b2
JK		 2361 else
2362 return CURSEG_COLD_NODE;
2363 }
2364}
2365
13f00235 2366static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2367{
13f00235
JK		 2368 if (fio->type == DATA) {
2369 struct inode *inode = fio->page->mapping->host;
351df4b2 2370
13f00235 2371 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2372 return CURSEG_COLD_DATA;
13f00235
JK		 2373 if (is_inode_flag_set(inode, FI_HOT_DATA))
2374 return CURSEG_HOT_DATA;
2375 return CURSEG_WARM_DATA;
351df4b2 2376 } else {
13f00235
JK		 2377 if (IS_DNODE(fio->page))
2378 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2379 CURSEG_HOT_NODE;
13f00235 2380 return CURSEG_COLD_NODE;
351df4b2
JK		 2381 }
2382}
2383
13f00235 2384static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2385{
13f00235
JK		 2386 int type = 0;
2387
2388 switch (fio->sbi->active_logs) {
351df4b2 2389 case 2:
13f00235
JK		 2390 type = __get_segment_type_2(fio);
2391 break;
351df4b2 2392 case 4:
13f00235
JK		 2393 type = __get_segment_type_4(fio);
2394 break;
2395 case 6:
2396 type = __get_segment_type_6(fio);
2397 break;
2398 default:
2399 f2fs_bug_on(fio->sbi, true);
351df4b2 2400 }
13f00235
JK		 2401
2402 if (IS_HOT(type))
2403 fio->temp = HOT;
2404 else if (IS_WARM(type))
2405 fio->temp = WARM;
2406 else
2407 fio->temp = COLD;
2408 return type;
351df4b2
JK		 2409}
2410
bfad7c2d
JK		 2411void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2412 block_t old_blkaddr, block_t *new_blkaddr,
13f00235
JK		 2413 struct f2fs_summary *sum, int type,
2414 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2415{
2416 struct sit_info *sit_i = SIT_I(sbi);
00e5a211 2417 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 2418
2419 mutex_lock(&curseg->curseg_mutex);
21cb1d99 2420 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 2421
2422 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2423
132263dd
JK		 2424 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2425
351df4b2
JK		 2426 /*
2427 * __add_sum_entry should be resided under the curseg_mutex
2428 * because, this function updates a summary entry in the
2429 * current summary block.
2430 */
e79efe3b 2431 __add_sum_entry(sbi, type, sum);
351df4b2 2432
351df4b2 2433 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2434
2435 stat_inc_block_count(sbi, curseg);
351df4b2 2436
5e443818
JK		 2437 if (!__has_curseg_space(sbi, type))
2438 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2 2439 /*
13f00235
JK		 2440 * SIT information should be updated after segment allocation,
2441 * since we need to keep dirty segments precisely under SSR.
351df4b2
JK		 2442 */
2443 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 2444
351df4b2
JK		 2445 mutex_unlock(&sit_i->sentry_lock);
2446
13f00235 2447 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2448 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2449
13f00235
JK		 2450 f2fs_inode_chksum_set(sbi, page);
2451 }
2452
2453 if (add_list) {
2454 struct f2fs_bio_info *io;
2455
2456 INIT_LIST_HEAD(&fio->list);
2457 fio->in_list = true;
2458 io = sbi->write_io[fio->type] + fio->temp;
2459 spin_lock(&io->io_lock);
2460 list_add_tail(&fio->list, &io->io_list);
2461 spin_unlock(&io->io_lock);
2462 }
2463
bfad7c2d
JK		 2464 mutex_unlock(&curseg->curseg_mutex);
2465}
2466
05ca3632 2467static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2468{
13f00235 2469 int type = __get_segment_type(fio);
b3fcb700 2470 int err;
bfad7c2d 2471
b3fcb700 2472reallocate:
c1286ff4 2473 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
13f00235 2474 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2475
351df4b2 2476 /* writeout dirty page into bdev */
13f00235 2477 err = f2fs_submit_page_write(fio);
b3fcb700
JK		 2478 if (err == -EAGAIN) {
2479 fio->old_blkaddr = fio->new_blkaddr;
2480 goto reallocate;
2481 }
351df4b2
JK		 2482}
2483
13f00235
JK		 2484void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2485 enum iostat_type io_type)
351df4b2 2486{
458e6197 2487 struct f2fs_io_info fio = {
05ca3632 2488 .sbi = sbi,
458e6197 2489 .type = META,
dc45fd9e
JK		 2490 .op = REQ_OP_WRITE,
2491 .op_flags = REQ_SYNC | REQ_NOIDLE | REQ_META | REQ_PRIO,
c1286ff4
JK		 2492 .old_blkaddr = page->index,
2493 .new_blkaddr = page->index,
05ca3632 2494 .page = page,
4375a336 2495 .encrypted_page = NULL,
13f00235 2496 .in_list = false,
458e6197
JK		 2497 };
2498
2b947003 2499 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
dc45fd9e 2500 fio.op_flags &= ~REQ_META;
2b947003 2501
351df4b2 2502 set_page_writeback(page);
13f00235
JK		 2503 f2fs_submit_page_write(&fio);
2504
2505 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2506}
2507
05ca3632 2508void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2509{
2510 struct f2fs_summary sum;
05ca3632 2511
351df4b2 2512 set_summary(&sum, nid, 0, 0);
05ca3632 2513 do_write_page(&sum, fio);
13f00235
JK		 2514
2515 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2516}
2517
05ca3632 2518void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2519{
05ca3632 2520 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2521 struct f2fs_summary sum;
2522 struct node_info ni;
2523
9850cf4a 2524 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2525 get_node_info(sbi, dn->nid, &ni);
2526 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2527 do_write_page(&sum, fio);
c1286ff4 2528 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
13f00235
JK		 2529
2530 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2531}
2532
13f00235 2533int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2534{
13f00235
JK		 2535 int err;
2536
c1286ff4 2537 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2538 stat_inc_inplace_blocks(fio->sbi);
13f00235
JK		 2539
2540 err = f2fs_submit_page_bio(fio);
2541
2542 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2543
2544 return err;
351df4b2
JK		 2545}
2546
c1286ff4 2547void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2548 block_t old_blkaddr, block_t new_blkaddr,
c1286ff4 2549 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2550{
2551 struct sit_info *sit_i = SIT_I(sbi);
2552 struct curseg_info *curseg;
2553 unsigned int segno, old_cursegno;
2554 struct seg_entry *se;
2555 int type;
19f106bc 2556 unsigned short old_blkoff;
351df4b2
JK		 2557
2558 segno = GET_SEGNO(sbi, new_blkaddr);
2559 se = get_seg_entry(sbi, segno);
2560 type = se->type;
2561
19f106bc
CY		 2562 if (!recover_curseg) {
2563 /* for recovery flow */
2564 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2565 if (old_blkaddr == NULL_ADDR)
2566 type = CURSEG_COLD_DATA;
2567 else
2568 type = CURSEG_WARM_DATA;
2569 }
2570 } else {
2571 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 2572 type = CURSEG_WARM_DATA;
2573 }
19f106bc 2574
351df4b2
JK		 2575 curseg = CURSEG_I(sbi, type);
2576
2577 mutex_lock(&curseg->curseg_mutex);
2578 mutex_lock(&sit_i->sentry_lock);
2579
2580 old_cursegno = curseg->segno;
19f106bc 2581 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2582
2583 /* change the current segment */
2584 if (segno != curseg->segno) {
2585 curseg->next_segno = segno;
13f00235 2586 change_curseg(sbi, type);
351df4b2
JK		 2587 }
2588
491c0854 2589 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2590 __add_sum_entry(sbi, type, sum);
351df4b2 2591
c1286ff4 2592 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2593 update_sit_entry(sbi, new_blkaddr, 1);
2594 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2595 update_sit_entry(sbi, old_blkaddr, -1);
2596
2597 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2598 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2599
351df4b2 2600 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2601
19f106bc
CY		 2602 if (recover_curseg) {
2603 if (old_cursegno != curseg->segno) {
2604 curseg->next_segno = old_cursegno;
13f00235 2605 change_curseg(sbi, type);
19f106bc
CY		 2606 }
2607 curseg->next_blkoff = old_blkoff;
2608 }
2609
351df4b2
JK		 2610 mutex_unlock(&sit_i->sentry_lock);
2611 mutex_unlock(&curseg->curseg_mutex);
2612}
2613
528e3459
CY		 2614void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2615 block_t old_addr, block_t new_addr,
c1286ff4
JK		 2616 unsigned char version, bool recover_curseg,
2617 bool recover_newaddr)
528e3459
CY		 2618{
2619 struct f2fs_summary sum;
2620
2621 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2622
c1286ff4
JK		 2623 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2624 recover_curseg, recover_newaddr);
528e3459 2625
c1286ff4 2626 f2fs_update_data_blkaddr(dn, new_addr);
df0f8dc0
CY		 2627}
2628
93dfe2ac 2629void f2fs_wait_on_page_writeback(struct page *page,
c1286ff4 2630 enum page_type type, bool ordered)
93dfe2ac 2631{
93dfe2ac 2632 if (PageWriteback(page)) {
4081363f
JK		 2633 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2634
13f00235
JK		 2635 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2636 0, page->index, type);
c1286ff4
JK		 2637 if (ordered)
2638 wait_on_page_writeback(page);
2639 else
2640 wait_for_stable_page(page);
93dfe2ac
JK		 2641 }
2642}
2643
13f00235 2644void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2645{
2646 struct page *cpage;
2647
c1286ff4 2648 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2649 return;
2650
08b39fbd
CY		 2651 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2652 if (cpage) {
c1286ff4 2653 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2654 f2fs_put_page(cpage, 1);
2655 }
2656}
2657
351df4b2
JK		 2658static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2659{
2660 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2661 struct curseg_info *seg_i;
2662 unsigned char *kaddr;
2663 struct page *page;
2664 block_t start;
2665 int i, j, offset;
2666
2667 start = start_sum_block(sbi);
2668
2669 page = get_meta_page(sbi, start++);
2670 kaddr = (unsigned char *)page_address(page);
2671
2672 /* Step 1: restore nat cache */
2673 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
c1286ff4 2674 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2675
2676 /* Step 2: restore sit cache */
2677 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
c1286ff4 2678 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2679 offset = 2 * SUM_JOURNAL_SIZE;
2680
2681 /* Step 3: restore summary entries */
2682 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2683 unsigned short blk_off;
2684 unsigned int segno;
2685
2686 seg_i = CURSEG_I(sbi, i);
2687 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2688 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2689 seg_i->next_segno = segno;
2690 reset_curseg(sbi, i, 0);
2691 seg_i->alloc_type = ckpt->alloc_type[i];
2692 seg_i->next_blkoff = blk_off;
2693
2694 if (seg_i->alloc_type == SSR)
2695 blk_off = sbi->blocks_per_seg;
2696
2697 for (j = 0; j < blk_off; j++) {
2698 struct f2fs_summary *s;
2699 s = (struct f2fs_summary *)(kaddr + offset);
2700 seg_i->sum_blk->entries[j] = *s;
2701 offset += SUMMARY_SIZE;
c1286ff4 2702 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2703 SUM_FOOTER_SIZE)
2704 continue;
2705
2706 f2fs_put_page(page, 1);
2707 page = NULL;
2708
2709 page = get_meta_page(sbi, start++);
2710 kaddr = (unsigned char *)page_address(page);
2711 offset = 0;
2712 }
2713 }
2714 f2fs_put_page(page, 1);
2715 return 0;
2716}
2717
2718static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2719{
2720 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2721 struct f2fs_summary_block *sum;
2722 struct curseg_info *curseg;
2723 struct page *new;
2724 unsigned short blk_off;
2725 unsigned int segno = 0;
2726 block_t blk_addr = 0;
2727
2728 /* get segment number and block addr */
2729 if (IS_DATASEG(type)) {
2730 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2731 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2732 CURSEG_HOT_DATA]);
119ee914 2733 if (__exist_node_summaries(sbi))
351df4b2
JK		 2734 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2735 else
2736 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2737 } else {
2738 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2739 CURSEG_HOT_NODE]);
2740 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2741 CURSEG_HOT_NODE]);
119ee914 2742 if (__exist_node_summaries(sbi))
351df4b2
JK		 2743 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2744 type - CURSEG_HOT_NODE);
2745 else
2746 blk_addr = GET_SUM_BLOCK(sbi, segno);
2747 }
2748
2749 new = get_meta_page(sbi, blk_addr);
2750 sum = (struct f2fs_summary_block *)page_address(new);
2751
2752 if (IS_NODESEG(type)) {
119ee914 2753 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2754 struct f2fs_summary *ns = &sum->entries[0];
2755 int i;
2756 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2757 ns->version = 0;
2758 ns->ofs_in_node = 0;
2759 }
2760 } else {
d653788a
GZ		 2761 int err;
2762
2763 err = restore_node_summary(sbi, segno, sum);
2764 if (err) {
351df4b2 2765 f2fs_put_page(new, 1);
d653788a 2766 return err;
351df4b2
JK		 2767 }
2768 }
2769 }
2770
2771 /* set uncompleted segment to curseg */
2772 curseg = CURSEG_I(sbi, type);
2773 mutex_lock(&curseg->curseg_mutex);
c1286ff4
JK		 2774
2775 /* update journal info */
2776 down_write(&curseg->journal_rwsem);
2777 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2778 up_write(&curseg->journal_rwsem);
2779
2780 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2781 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2782 curseg->next_segno = segno;
2783 reset_curseg(sbi, type, 0);
2784 curseg->alloc_type = ckpt->alloc_type[type];
2785 curseg->next_blkoff = blk_off;
2786 mutex_unlock(&curseg->curseg_mutex);
2787 f2fs_put_page(new, 1);
2788 return 0;
2789}
2790
2791static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2792{
13f00235
JK		 2793 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2794 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2795 int type = CURSEG_HOT_DATA;
e4fc5fbf 2796 int err;
351df4b2 2797
c1286ff4 2798 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2799 int npages = npages_for_summary_flush(sbi, true);
2800
2801 if (npages >= 2)
2802 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2803 META_CP, true);
3fa06d7b 2804
351df4b2
JK		 2805 /* restore for compacted data summary */
2806 if (read_compacted_summaries(sbi))
2807 return -EINVAL;
2808 type = CURSEG_HOT_NODE;
2809 }
2810
119ee914 2811 if (__exist_node_summaries(sbi))
3fa06d7b 2812 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2813 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2814
e4fc5fbf
CY		 2815 for (; type <= CURSEG_COLD_NODE; type++) {
2816 err = read_normal_summaries(sbi, type);
2817 if (err)
2818 return err;
2819 }
2820
13f00235
JK		 2821 /* sanity check for summary blocks */
2822 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2823 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2824 return -EINVAL;
2825
351df4b2
JK		 2826 return 0;
2827}
2828
2829static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2830{
2831 struct page *page;
2832 unsigned char *kaddr;
2833 struct f2fs_summary *summary;
2834 struct curseg_info *seg_i;
2835 int written_size = 0;
2836 int i, j;
2837
2838 page = grab_meta_page(sbi, blkaddr++);
2839 kaddr = (unsigned char *)page_address(page);
2840
2841 /* Step 1: write nat cache */
2842 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
c1286ff4 2843 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2844 written_size += SUM_JOURNAL_SIZE;
2845
2846 /* Step 2: write sit cache */
2847 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
c1286ff4 2848 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2849 written_size += SUM_JOURNAL_SIZE;
2850
351df4b2
JK		 2851 /* Step 3: write summary entries */
2852 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2853 unsigned short blkoff;
2854 seg_i = CURSEG_I(sbi, i);
2855 if (sbi->ckpt->alloc_type[i] == SSR)
2856 blkoff = sbi->blocks_per_seg;
2857 else
2858 blkoff = curseg_blkoff(sbi, i);
2859
2860 for (j = 0; j < blkoff; j++) {
2861 if (!page) {
2862 page = grab_meta_page(sbi, blkaddr++);
2863 kaddr = (unsigned char *)page_address(page);
2864 written_size = 0;
2865 }
2866 summary = (struct f2fs_summary *)(kaddr + written_size);
2867 *summary = seg_i->sum_blk->entries[j];
2868 written_size += SUMMARY_SIZE;
351df4b2 2869
c1286ff4 2870 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2871 SUM_FOOTER_SIZE)
2872 continue;
2873
e8d61a74 2874 set_page_dirty(page);
351df4b2
JK		 2875 f2fs_put_page(page, 1);
2876 page = NULL;
2877 }
2878 }
e8d61a74
CY		 2879 if (page) {
2880 set_page_dirty(page);
351df4b2 2881 f2fs_put_page(page, 1);
e8d61a74 2882 }
351df4b2
JK		 2883}
2884
2885static void write_normal_summaries(struct f2fs_sb_info *sbi,
2886 block_t blkaddr, int type)
2887{
2888 int i, end;
2889 if (IS_DATASEG(type))
2890 end = type + NR_CURSEG_DATA_TYPE;
2891 else
2892 end = type + NR_CURSEG_NODE_TYPE;
2893
c1286ff4
JK		 2894 for (i = type; i < end; i++)
2895 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2896}
2897
2898void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2899{
c1286ff4 2900 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2901 write_compacted_summaries(sbi, start_blk);
2902 else
2903 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2904}
2905
2906void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2907{
119ee914 2908 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2909}
2910
c1286ff4 2911int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2912 unsigned int val, int alloc)
2913{
2914 int i;
2915
2916 if (type == NAT_JOURNAL) {
c1286ff4
JK		 2917 for (i = 0; i < nats_in_cursum(journal); i++) {
2918 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2919 return i;
2920 }
c1286ff4
JK		 2921 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2922 return update_nats_in_cursum(journal, 1);
351df4b2 2923 } else if (type == SIT_JOURNAL) {
c1286ff4
JK		 2924 for (i = 0; i < sits_in_cursum(journal); i++)
2925 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2926 return i;
c1286ff4
JK		 2927 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2928 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2929 }
2930 return -1;
2931}
2932
2933static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2934 unsigned int segno)
2935{
2cc22186 2936 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2937}
2938
2939static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2940 unsigned int start)
2941{
2942 struct sit_info *sit_i = SIT_I(sbi);
2943 struct page *src_page, *dst_page;
2944 pgoff_t src_off, dst_off;
2945 void *src_addr, *dst_addr;
2946
2947 src_off = current_sit_addr(sbi, start);
2948 dst_off = next_sit_addr(sbi, src_off);
2949
2950 /* get current sit block page without lock */
2951 src_page = get_meta_page(sbi, src_off);
2952 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2953 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2954
2955 src_addr = page_address(src_page);
2956 dst_addr = page_address(dst_page);
c1286ff4 2957 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2958
2959 set_page_dirty(dst_page);
2960 f2fs_put_page(src_page, 1);
2961
2962 set_to_next_sit(sit_i, start);
2963
2964 return dst_page;
2965}
2966
184a5cd2
CY		 2967static struct sit_entry_set *grab_sit_entry_set(void)
2968{
2969 struct sit_entry_set *ses =
80c54505 2970 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2971
2972 ses->entry_cnt = 0;
2973 INIT_LIST_HEAD(&ses->set_list);
2974 return ses;
2975}
2976
2977static void release_sit_entry_set(struct sit_entry_set *ses)
2978{
2979 list_del(&ses->set_list);
2980 kmem_cache_free(sit_entry_set_slab, ses);
2981}
2982
2983static void adjust_sit_entry_set(struct sit_entry_set *ses,
2984 struct list_head *head)
2985{
2986 struct sit_entry_set *next = ses;
2987
2988 if (list_is_last(&ses->set_list, head))
2989 return;
2990
2991 list_for_each_entry_continue(next, head, set_list)
2992 if (ses->entry_cnt <= next->entry_cnt)
2993 break;
2994
2995 list_move_tail(&ses->set_list, &next->set_list);
2996}
2997
2998static void add_sit_entry(unsigned int segno, struct list_head *head)
2999{
3000 struct sit_entry_set *ses;
3001 unsigned int start_segno = START_SEGNO(segno);
3002
3003 list_for_each_entry(ses, head, set_list) {
3004 if (ses->start_segno == start_segno) {
3005 ses->entry_cnt++;
3006 adjust_sit_entry_set(ses, head);
3007 return;
3008 }
3009 }
3010
3011 ses = grab_sit_entry_set();
3012
3013 ses->start_segno = start_segno;
3014 ses->entry_cnt++;
3015 list_add(&ses->set_list, head);
3016}
3017
3018static void add_sits_in_set(struct f2fs_sb_info *sbi)
3019{
3020 struct f2fs_sm_info *sm_info = SM_I(sbi);
3021 struct list_head *set_list = &sm_info->sit_entry_set;
3022 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3023 unsigned int segno;
3024
7cd8558b 3025 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3026 add_sit_entry(segno, set_list);
3027}
3028
3029static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3030{
3031 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
c1286ff4 3032 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3033 int i;
3034
c1286ff4
JK		 3035 down_write(&curseg->journal_rwsem);
3036 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3037 unsigned int segno;
3038 bool dirtied;
3039
c1286ff4 3040 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3041 dirtied = __mark_sit_entry_dirty(sbi, segno);
3042
3043 if (!dirtied)
3044 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3045 }
c1286ff4
JK		 3046 update_sits_in_cursum(journal, -i);
3047 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3048}
3049
0a8165d7 3050/*
351df4b2
JK		 3051 * CP calls this function, which flushes SIT entries including sit_journal,
3052 * and moves prefree segs to free segs.
3053 */
4b2fecc8 3054void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3055{
3056 struct sit_info *sit_i = SIT_I(sbi);
3057 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3058 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
c1286ff4 3059 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3060 struct sit_entry_set *ses, *tmp;
3061 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3062 bool to_journal = true;
4b2fecc8 3063 struct seg_entry *se;
351df4b2 3064
351df4b2
JK		 3065 mutex_lock(&sit_i->sentry_lock);
3066
2b11a74b
WL		 3067 if (!sit_i->dirty_sentries)
3068 goto out;
3069
351df4b2 3070 /*
184a5cd2
CY		 3071 * add and account sit entries of dirty bitmap in sit entry
3072 * set temporarily
351df4b2 3073 */
184a5cd2 3074 add_sits_in_set(sbi);
351df4b2 3075
184a5cd2
CY		 3076 /*
3077 * if there are no enough space in journal to store dirty sit
3078 * entries, remove all entries from journal and add and account
3079 * them in sit entry set.
3080 */
c1286ff4 3081 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3082 remove_sits_in_journal(sbi);
b2955550 3083
184a5cd2
CY		 3084 /*
3085 * there are two steps to flush sit entries:
3086 * #1, flush sit entries to journal in current cold data summary block.
3087 * #2, flush sit entries to sit page.
3088 */
3089 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3090 struct page *page = NULL;
184a5cd2
CY		 3091 struct f2fs_sit_block *raw_sit = NULL;
3092 unsigned int start_segno = ses->start_segno;
3093 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3094 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3095 unsigned int segno = start_segno;
3096
3097 if (to_journal &&
c1286ff4 3098 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3099 to_journal = false;
3100
c1286ff4
JK		 3101 if (to_journal) {
3102 down_write(&curseg->journal_rwsem);
3103 } else {
184a5cd2
CY		 3104 page = get_next_sit_page(sbi, start_segno);
3105 raw_sit = page_address(page);
351df4b2 3106 }
351df4b2 3107
184a5cd2
CY		 3108 /* flush dirty sit entries in region of current sit set */
3109 for_each_set_bit_from(segno, bitmap, end) {
3110 int offset, sit_offset;
4b2fecc8
JK		 3111
3112 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 3113
3114 /* add discard candidates */
13f00235 3115 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3116 cpc->trim_start = segno;
556f5ba3 3117 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3118 }
184a5cd2
CY		 3119
3120 if (to_journal) {
c1286ff4 3121 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3122 SIT_JOURNAL, segno, 1);
3123 f2fs_bug_on(sbi, offset < 0);
c1286ff4 3124 segno_in_journal(journal, offset) =
184a5cd2
CY		 3125 cpu_to_le32(segno);
3126 seg_info_to_raw_sit(se,
c1286ff4 3127 &sit_in_journal(journal, offset));
184a5cd2
CY		 3128 } else {
3129 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3130 seg_info_to_raw_sit(se,
3131 &raw_sit->entries[sit_offset]);
3132 }
351df4b2 3133
184a5cd2
CY		 3134 __clear_bit(segno, bitmap);
3135 sit_i->dirty_sentries--;
3136 ses->entry_cnt--;
351df4b2
JK		 3137 }
3138
c1286ff4
JK		 3139 if (to_journal)
3140 up_write(&curseg->journal_rwsem);
3141 else
184a5cd2
CY		 3142 f2fs_put_page(page, 1);
3143
3144 f2fs_bug_on(sbi, ses->entry_cnt);
3145 release_sit_entry_set(ses);
351df4b2 3146 }
184a5cd2
CY		 3147
3148 f2fs_bug_on(sbi, !list_empty(head));
3149 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3150out:
13f00235 3151 if (cpc->reason & CP_DISCARD) {
c70e14cd
YH		 3152 __u64 trim_start = cpc->trim_start;
3153
4b2fecc8 3154 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
556f5ba3 3155 add_discard_addrs(sbi, cpc, false);
c70e14cd
YH		 3156
3157 cpc->trim_start = trim_start;
4b2fecc8 3158 }
351df4b2 3159 mutex_unlock(&sit_i->sentry_lock);
351df4b2 3160
351df4b2
JK		 3161 set_prefree_as_free_segments(sbi);
3162}
3163
3164static int build_sit_info(struct f2fs_sb_info *sbi)
3165{
3166 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3167 struct sit_info *sit_i;
3168 unsigned int sit_segs, start;
fb2e2f44 3169 char *src_bitmap;
351df4b2
JK		 3170 unsigned int bitmap_size;
3171
3172 /* allocate memory for SIT information */
3173 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3174 if (!sit_i)
3175 return -ENOMEM;
3176
3177 SM_I(sbi)->sit_info = sit_i;
3178
13f00235 3179 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3180 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3181 if (!sit_i->sentries)
3182 return -ENOMEM;
3183
7cd8558b 3184 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
13f00235 3185 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3186 if (!sit_i->dirty_sentries_bitmap)
3187 return -ENOMEM;
3188
7cd8558b 3189 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3190 sit_i->sentries[start].cur_valid_map
3191 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3192 sit_i->sentries[start].ckpt_valid_map
3193 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3194 if (!sit_i->sentries[start].cur_valid_map ||
c1286ff4 3195 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3196 return -ENOMEM;
c1286ff4 3197
dd5804b2
CY		 3198#ifdef CONFIG_F2FS_CHECK_FS
3199 sit_i->sentries[start].cur_valid_map_mir
3200 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3201 if (!sit_i->sentries[start].cur_valid_map_mir)
3202 return -ENOMEM;
3203#endif
3204
c1286ff4
JK		 3205 if (f2fs_discard_en(sbi)) {
3206 sit_i->sentries[start].discard_map
3207 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3208 if (!sit_i->sentries[start].discard_map)
3209 return -ENOMEM;
3210 }
351df4b2
JK		 3211 }
3212
60a3b782
JK		 3213 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3214 if (!sit_i->tmp_map)
3215 return -ENOMEM;
3216
351df4b2 3217 if (sbi->segs_per_sec > 1) {
13f00235 3218 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3219 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3220 if (!sit_i->sec_entries)
3221 return -ENOMEM;
3222 }
3223
3224 /* get information related with SIT */
3225 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3226
3227 /* setup SIT bitmap from ckeckpoint pack */
3228 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3229 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3230
fb2e2f44
CY		 3231 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3232 if (!sit_i->sit_bitmap)
351df4b2 3233 return -ENOMEM;
351df4b2 3234
fb2e2f44
CY		 3235#ifdef CONFIG_F2FS_CHECK_FS
3236 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3237 if (!sit_i->sit_bitmap_mir)
3238 return -ENOMEM;
3239#endif
3240
351df4b2
JK		 3241 /* init SIT information */
3242 sit_i->s_ops = &default_salloc_ops;
3243
3244 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3245 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
df3f20f1 3246 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3247 sit_i->bitmap_size = bitmap_size;
3248 sit_i->dirty_sentries = 0;
3249 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3250 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3251 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
3252 mutex_init(&sit_i->sentry_lock);
3253 return 0;
3254}
3255
3256static int build_free_segmap(struct f2fs_sb_info *sbi)
3257{
351df4b2
JK		 3258 struct free_segmap_info *free_i;
3259 unsigned int bitmap_size, sec_bitmap_size;
3260
3261 /* allocate memory for free segmap information */
3262 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3263 if (!free_i)
3264 return -ENOMEM;
3265
3266 SM_I(sbi)->free_info = free_i;
3267
7cd8558b 3268 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
13f00235 3269 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3270 if (!free_i->free_segmap)
3271 return -ENOMEM;
3272
7cd8558b 3273 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
13f00235 3274 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3275 if (!free_i->free_secmap)
3276 return -ENOMEM;
3277
3278 /* set all segments as dirty temporarily */
3279 memset(free_i->free_segmap, 0xff, bitmap_size);
3280 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3281
3282 /* init free segmap information */
7cd8558b 3283 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3284 free_i->free_segments = 0;
3285 free_i->free_sections = 0;
1a118ccf 3286 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3287 return 0;
3288}
3289
3290static int build_curseg(struct f2fs_sb_info *sbi)
3291{
1042d60f 3292 struct curseg_info *array;
351df4b2
JK		 3293 int i;
3294
b434babf 3295 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3296 if (!array)
3297 return -ENOMEM;
3298
3299 SM_I(sbi)->curseg_array = array;
3300
3301 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3302 mutex_init(&array[i].curseg_mutex);
c1286ff4 3303 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3304 if (!array[i].sum_blk)
3305 return -ENOMEM;
c1286ff4
JK		 3306 init_rwsem(&array[i].journal_rwsem);
3307 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3308 GFP_KERNEL);
3309 if (!array[i].journal)
3310 return -ENOMEM;
351df4b2
JK		 3311 array[i].segno = NULL_SEGNO;
3312 array[i].next_blkoff = 0;
3313 }
3314 return restore_curseg_summaries(sbi);
3315}
3316
3317static void build_sit_entries(struct f2fs_sb_info *sbi)
3318{
3319 struct sit_info *sit_i = SIT_I(sbi);
3320 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
c1286ff4
JK		 3321 struct f2fs_journal *journal = curseg->journal;
3322 struct seg_entry *se;
3323 struct f2fs_sit_entry sit;
74de593a
CY		 3324 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3325 unsigned int i, start, end;
3326 unsigned int readed, start_blk = 0;
351df4b2 3327
74de593a 3328 do {
04030d21
JK		 3329 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3330 META_SIT, true);
74de593a
CY		 3331
3332 start = start_blk * sit_i->sents_per_block;
3333 end = (start_blk + readed) * sit_i->sents_per_block;
3334
7cd8558b 3335 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3336 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3337 struct page *page;
3338
c1286ff4 3339 se = &sit_i->sentries[start];
74de593a
CY		 3340 page = get_current_sit_page(sbi, start);
3341 sit_blk = (struct f2fs_sit_block *)page_address(page);
3342 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3343 f2fs_put_page(page, 1);
c1286ff4 3344
74de593a
CY		 3345 check_block_count(sbi, start, &sit);
3346 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3347
3348 /* build discard map only one time */
c1286ff4 3349 if (f2fs_discard_en(sbi)) {
13f00235
JK		 3350 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3351 memset(se->discard_map, 0xff,
3352 SIT_VBLOCK_MAP_SIZE);
3353 } else {
3354 memcpy(se->discard_map,
3355 se->cur_valid_map,
3356 SIT_VBLOCK_MAP_SIZE);
3357 sbi->discard_blks +=
3358 sbi->blocks_per_seg -
3359 se->valid_blocks;
3360 }
74de593a 3361 }
c1286ff4
JK		 3362
3363 if (sbi->segs_per_sec > 1)
3364 get_sec_entry(sbi, start)->valid_blocks +=
3365 se->valid_blocks;
351df4b2 3366 }
74de593a
CY		 3367 start_blk += readed;
3368 } while (start_blk < sit_blk_cnt);
c1286ff4
JK		 3369
3370 down_read(&curseg->journal_rwsem);
3371 for (i = 0; i < sits_in_cursum(journal); i++) {
3372 unsigned int old_valid_blocks;
3373
3374 start = le32_to_cpu(segno_in_journal(journal, i));
3375 se = &sit_i->sentries[start];
3376 sit = sit_in_journal(journal, i);
3377
3378 old_valid_blocks = se->valid_blocks;
3379
3380 check_block_count(sbi, start, &sit);
3381 seg_info_from_raw_sit(se, &sit);
3382
3383 if (f2fs_discard_en(sbi)) {
13f00235
JK		 3384 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3385 memset(se->discard_map, 0xff,
3386 SIT_VBLOCK_MAP_SIZE);
3387 } else {
3388 memcpy(se->discard_map, se->cur_valid_map,
3389 SIT_VBLOCK_MAP_SIZE);
3390 sbi->discard_blks += old_valid_blocks -
3391 se->valid_blocks;
3392 }
c1286ff4
JK		 3393 }
3394
3395 if (sbi->segs_per_sec > 1)
3396 get_sec_entry(sbi, start)->valid_blocks +=
3397 se->valid_blocks - old_valid_blocks;
3398 }
3399 up_read(&curseg->journal_rwsem);
351df4b2
JK		 3400}
3401
3402static void init_free_segmap(struct f2fs_sb_info *sbi)
3403{
3404 unsigned int start;
3405 int type;
3406
7cd8558b 3407 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3408 struct seg_entry *sentry = get_seg_entry(sbi, start);
3409 if (!sentry->valid_blocks)
3410 __set_free(sbi, start);
df3f20f1
JK		 3411 else
3412 SIT_I(sbi)->written_valid_blocks +=
3413 sentry->valid_blocks;
351df4b2
JK		 3414 }
3415
3416 /* set use the current segments */
3417 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3418 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3419 __set_test_and_inuse(sbi, curseg_t->segno);
3420 }
3421}
3422
3423static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3424{
3425 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3426 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3427 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3428 unsigned short valid_blocks;
3429
8736fbf0 3430 while (1) {
351df4b2 3431 /* find dirty segment based on free segmap */
7cd8558b
JK		 3432 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3433 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3434 break;
3435 offset = segno + 1;
13f00235 3436 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3437 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3438 continue;
ec325b52
JK		 3439 if (valid_blocks > sbi->blocks_per_seg) {
3440 f2fs_bug_on(sbi, 1);
3441 continue;
3442 }
351df4b2
JK		 3443 mutex_lock(&dirty_i->seglist_lock);
3444 __locate_dirty_segment(sbi, segno, DIRTY);
3445 mutex_unlock(&dirty_i->seglist_lock);
3446 }
3447}
3448
5ec4e49f 3449static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3450{
3451 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3452 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3453
13f00235 3454 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3455 if (!dirty_i->victim_secmap)
351df4b2
JK		 3456 return -ENOMEM;
3457 return 0;
3458}
3459
3460static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3461{
3462 struct dirty_seglist_info *dirty_i;
3463 unsigned int bitmap_size, i;
3464
3465 /* allocate memory for dirty segments list information */
3466 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3467 if (!dirty_i)
3468 return -ENOMEM;
3469
3470 SM_I(sbi)->dirty_info = dirty_i;
3471 mutex_init(&dirty_i->seglist_lock);
3472
7cd8558b 3473 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3474
3475 for (i = 0; i < NR_DIRTY_TYPE; i++) {
13f00235 3476 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3477 if (!dirty_i->dirty_segmap[i])
3478 return -ENOMEM;
3479 }
3480
3481 init_dirty_segmap(sbi);
5ec4e49f 3482 return init_victim_secmap(sbi);
351df4b2
JK		 3483}
3484
0a8165d7 3485/*
351df4b2
JK		 3486 * Update min, max modified time for cost-benefit GC algorithm
3487 */
3488static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3489{
3490 struct sit_info *sit_i = SIT_I(sbi);
3491 unsigned int segno;
3492
3493 mutex_lock(&sit_i->sentry_lock);
3494
3495 sit_i->min_mtime = LLONG_MAX;
3496
7cd8558b 3497 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3498 unsigned int i;
3499 unsigned long long mtime = 0;
3500
3501 for (i = 0; i < sbi->segs_per_sec; i++)
3502 mtime += get_seg_entry(sbi, segno + i)->mtime;
3503
3504 mtime = div_u64(mtime, sbi->segs_per_sec);
3505
3506 if (sit_i->min_mtime > mtime)
3507 sit_i->min_mtime = mtime;
3508 }
3509 sit_i->max_mtime = get_mtime(sbi);
3510 mutex_unlock(&sit_i->sentry_lock);
3511}
3512
3513int build_segment_manager(struct f2fs_sb_info *sbi)
3514{
3515 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3516 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3517 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3518 int err;
3519
3520 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3521 if (!sm_info)
3522 return -ENOMEM;
3523
3524 /* init sm info */
3525 sbi->sm_info = sm_info;
351df4b2
JK		 3526 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3527 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3528 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3529 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3530 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3531 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3532 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3533 sm_info->rec_prefree_segments = sm_info->main_segments *
3534 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
c1286ff4
JK		 3535 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3536 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3537
3538 if (!test_opt(sbi, LFS))
3539 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3540 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3541 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
13f00235 3542 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
7fd9e544 3543
bba681cb
JK		 3544 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3545
184a5cd2
CY		 3546 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3547
13f00235 3548 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3549 err = create_flush_cmd_control(sbi);
3550 if (err)
a688b9d9 3551 return err;
6b4afdd7
JK		 3552 }
3553
13f00235
JK		 3554 err = create_discard_cmd_control(sbi);
3555 if (err)
3556 return err;
3557
351df4b2
JK		 3558 err = build_sit_info(sbi);
3559 if (err)
3560 return err;
3561 err = build_free_segmap(sbi);
3562 if (err)
3563 return err;
3564 err = build_curseg(sbi);
3565 if (err)
3566 return err;
3567
3568 /* reinit free segmap based on SIT */
3569 build_sit_entries(sbi);
3570
3571 init_free_segmap(sbi);
3572 err = build_dirty_segmap(sbi);
3573 if (err)
3574 return err;
3575
3576 init_min_max_mtime(sbi);
3577 return 0;
3578}
3579
3580static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3581 enum dirty_type dirty_type)
3582{
3583 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3584
3585 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3586 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3587 dirty_i->nr_dirty[dirty_type] = 0;
3588 mutex_unlock(&dirty_i->seglist_lock);
3589}
3590
5ec4e49f 3591static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3592{
3593 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3594 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3595}
3596
3597static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3598{
3599 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3600 int i;
3601
3602 if (!dirty_i)
3603 return;
3604
3605 /* discard pre-free/dirty segments list */
3606 for (i = 0; i < NR_DIRTY_TYPE; i++)
3607 discard_dirty_segmap(sbi, i);
3608
5ec4e49f 3609 destroy_victim_secmap(sbi);
351df4b2
JK		 3610 SM_I(sbi)->dirty_info = NULL;
3611 kfree(dirty_i);
3612}
3613
3614static void destroy_curseg(struct f2fs_sb_info *sbi)
3615{
3616 struct curseg_info *array = SM_I(sbi)->curseg_array;
3617 int i;
3618
3619 if (!array)
3620 return;
3621 SM_I(sbi)->curseg_array = NULL;
c1286ff4 3622 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3623 kfree(array[i].sum_blk);
c1286ff4
JK		 3624 kfree(array[i].journal);
3625 }
351df4b2
JK		 3626 kfree(array);
3627}
3628
3629static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3630{
3631 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3632 if (!free_i)
3633 return;
3634 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3635 kvfree(free_i->free_segmap);
3636 kvfree(free_i->free_secmap);
351df4b2
JK		 3637 kfree(free_i);
3638}
3639
3640static void destroy_sit_info(struct f2fs_sb_info *sbi)
3641{
3642 struct sit_info *sit_i = SIT_I(sbi);
3643 unsigned int start;
3644
3645 if (!sit_i)
3646 return;
3647
3648 if (sit_i->sentries) {
7cd8558b 3649 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3650 kfree(sit_i->sentries[start].cur_valid_map);
dd5804b2
CY		 3651#ifdef CONFIG_F2FS_CHECK_FS
3652 kfree(sit_i->sentries[start].cur_valid_map_mir);
3653#endif
351df4b2 3654 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3655 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3656 }
3657 }
60a3b782
JK		 3658 kfree(sit_i->tmp_map);
3659
39307a8e
JK		 3660 kvfree(sit_i->sentries);
3661 kvfree(sit_i->sec_entries);
3662 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3663
3664 SM_I(sbi)->sit_info = NULL;
3665 kfree(sit_i->sit_bitmap);
fb2e2f44
CY		 3666#ifdef CONFIG_F2FS_CHECK_FS
3667 kfree(sit_i->sit_bitmap_mir);
3668#endif
351df4b2
JK		 3669 kfree(sit_i);
3670}
3671
3672void destroy_segment_manager(struct f2fs_sb_info *sbi)
3673{
3674 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3675
3b03f724
CY		 3676 if (!sm_info)
3677 return;
be4b8492 3678 destroy_flush_cmd_control(sbi, true);
13f00235 3679 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3680 destroy_dirty_segmap(sbi);
3681 destroy_curseg(sbi);
3682 destroy_free_segmap(sbi);
3683 destroy_sit_info(sbi);
3684 sbi->sm_info = NULL;
3685 kfree(sm_info);
3686}
7fd9e544
JK		 3687
3688int __init create_segment_manager_caches(void)
3689{
3690 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3691 sizeof(struct discard_entry));
7fd9e544 3692 if (!discard_entry_slab)
184a5cd2
CY		 3693 goto fail;
3694
eee3f1f5
JK		 3695 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3696 sizeof(struct discard_cmd));
3697 if (!discard_cmd_slab)
401c465b
JK		 3698 goto destroy_discard_entry;
3699
184a5cd2 3700 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3701 sizeof(struct sit_entry_set));
184a5cd2 3702 if (!sit_entry_set_slab)
eee3f1f5 3703 goto destroy_discard_cmd;
88b88a66
JK		 3704
3705 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3706 sizeof(struct inmem_pages));
3707 if (!inmem_entry_slab)
3708 goto destroy_sit_entry_set;
7fd9e544 3709 return 0;
184a5cd2 3710
88b88a66
JK		 3711destroy_sit_entry_set:
3712 kmem_cache_destroy(sit_entry_set_slab);
eee3f1f5
JK		 3713destroy_discard_cmd:
3714 kmem_cache_destroy(discard_cmd_slab);
c1286ff4 3715destroy_discard_entry:
184a5cd2
CY		 3716 kmem_cache_destroy(discard_entry_slab);
3717fail:
3718 return -ENOMEM;
7fd9e544
JK		 3719}
3720
3721void destroy_segment_manager_caches(void)
3722{
184a5cd2 3723 kmem_cache_destroy(sit_entry_set_slab);
eee3f1f5 3724 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3725 kmem_cache_destroy(discard_entry_slab);
88b88a66 3726 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3727}