]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: fix to wait page writeback during revoking atomic write
[mirror_ubuntu-jammy-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 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;
b01a9201 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.
692223d1 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);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL		 112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL		 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);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
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)
80609448
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
b3a97a2a
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 +
a2a12b67 184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
b3a97a2a
JK		 185}
186
88b88a66
JK		 187void register_inmem_page(struct inode *inode, struct page *page)
188{
57864ae5 189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
88b88a66
JK		 190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
9be32d72 192
9e4ded3f 193 f2fs_trace_pid(page);
0722b101 194
decd36b6
CY		 195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
197
88b88a66
JK		 198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
decd36b6 203
88b88a66
JK		 204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
57864ae5
JK		 208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
8dcf2ff7 212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 213 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 214
215 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 216}
217
28bc106b
CY		 218static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover)
29b96b54 220{
28bc106b 221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 222 struct inmem_pages *cur, *tmp;
28bc106b 223 int err = 0;
29b96b54
CY		 224
225 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 226 struct page *page = cur->page;
227
228 if (drop)
229 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
230
231 lock_page(page);
29b96b54 232
e5e5732d
CY		 233 f2fs_wait_on_page_writeback(page, DATA, true);
234
28bc106b
CY		 235 if (recover) {
236 struct dnode_of_data dn;
237 struct node_info ni;
238
239 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 240retry:
28bc106b 241 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 242 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
243 if (err) {
244 if (err == -ENOMEM) {
245 congestion_wait(BLK_RW_ASYNC, HZ/50);
246 cond_resched();
247 goto retry;
248 }
28bc106b
CY		 249 err = -EAGAIN;
250 goto next;
251 }
252 get_node_info(sbi, dn.nid, &ni);
f1d2564a
DJ		 253 if (cur->old_addr == NEW_ADDR) {
254 invalidate_blocks(sbi, dn.data_blkaddr);
255 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
256 } else
257 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 258 cur->old_addr, ni.version, true, true);
259 f2fs_put_dnode(&dn);
260 }
261next:
63c52d78
JK		 262 /* we don't need to invalidate this in the sccessful status */
263 if (drop || recover)
264 ClearPageUptodate(page);
28bc106b 265 set_page_private(page, 0);
c81ced05 266 ClearPagePrivate(page);
28bc106b 267 f2fs_put_page(page, 1);
29b96b54
CY		 268
269 list_del(&cur->list);
270 kmem_cache_free(inmem_entry_slab, cur);
271 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
272 }
28bc106b 273 return err;
29b96b54
CY		 274}
275
57864ae5
JK		 276void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
277{
278 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
279 struct inode *inode;
280 struct f2fs_inode_info *fi;
281next:
282 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
283 if (list_empty(head)) {
284 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
285 return;
286 }
287 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
288 inode = igrab(&fi->vfs_inode);
289 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
290
291 if (inode) {
292 drop_inmem_pages(inode);
293 iput(inode);
294 }
295 congestion_wait(BLK_RW_ASYNC, HZ/50);
296 cond_resched();
297 goto next;
298}
299
29b96b54
CY		 300void drop_inmem_pages(struct inode *inode)
301{
57864ae5 302 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 303 struct f2fs_inode_info *fi = F2FS_I(inode);
304
305 mutex_lock(&fi->inmem_lock);
28bc106b 306 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 307 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
308 if (!list_empty(&fi->inmem_ilist))
309 list_del_init(&fi->inmem_ilist);
310 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 311 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 312
313 clear_inode_flag(inode, FI_ATOMIC_FILE);
314 stat_dec_atomic_write(inode);
29b96b54
CY		 315}
316
8c242db9
JK		 317void drop_inmem_page(struct inode *inode, struct page *page)
318{
319 struct f2fs_inode_info *fi = F2FS_I(inode);
320 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
321 struct list_head *head = &fi->inmem_pages;
322 struct inmem_pages *cur = NULL;
323
324 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
325
326 mutex_lock(&fi->inmem_lock);
327 list_for_each_entry(cur, head, list) {
328 if (cur->page == page)
329 break;
330 }
331
d0891e84 332 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
8c242db9
JK		 333 list_del(&cur->list);
334 mutex_unlock(&fi->inmem_lock);
335
336 dec_page_count(sbi, F2FS_INMEM_PAGES);
337 kmem_cache_free(inmem_entry_slab, cur);
338
339 ClearPageUptodate(page);
340 set_page_private(page, 0);
341 ClearPagePrivate(page);
342 f2fs_put_page(page, 0);
343
344 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
345}
346
cf52b27a 347static int __commit_inmem_pages(struct inode *inode)
88b88a66
JK		 348{
349 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
350 struct f2fs_inode_info *fi = F2FS_I(inode);
351 struct inmem_pages *cur, *tmp;
88b88a66 352 struct f2fs_io_info fio = {
05ca3632 353 .sbi = sbi,
39d787be 354 .ino = inode->i_ino,
88b88a66 355 .type = DATA,
04d328de 356 .op = REQ_OP_WRITE,
70fd7614 357 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 358 .io_type = FS_DATA_IO,
88b88a66 359 };
cf52b27a 360 struct list_head revoke_list;
942fd319 361 pgoff_t last_idx = ULONG_MAX;
edb27dee 362 int err = 0;
88b88a66 363
cf52b27a
CY		 364 INIT_LIST_HEAD(&revoke_list);
365
88b88a66 366 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 367 struct page *page = cur->page;
368
369 lock_page(page);
370 if (page->mapping == inode->i_mapping) {
371 trace_f2fs_commit_inmem_page(page, INMEM);
372
373 set_page_dirty(page);
374 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 375 if (clear_page_dirty_for_io(page)) {
29b96b54 376 inode_dec_dirty_pages(inode);
933439c8
CY		 377 remove_dirty_inode(inode);
378 }
640cc189 379retry:
28bc106b 380 fio.page = page;
e959c8f5 381 fio.old_blkaddr = NULL_ADDR;
4d978078 382 fio.encrypted_page = NULL;
cc15620b 383 fio.need_lock = LOCK_DONE;
29b96b54
CY		 384 err = do_write_data_page(&fio);
385 if (err) {
640cc189
JK		 386 if (err == -ENOMEM) {
387 congestion_wait(BLK_RW_ASYNC, HZ/50);
388 cond_resched();
389 goto retry;
390 }
28bc106b 391 unlock_page(page);
29b96b54 392 break;
70c640b1 393 }
28bc106b
CY		 394 /* record old blkaddr for revoking */
395 cur->old_addr = fio.old_blkaddr;
942fd319 396 last_idx = page->index;
28bc106b
CY		 397 }
398 unlock_page(page);
cf52b27a 399 list_move_tail(&cur->list, &revoke_list);
88b88a66 400 }
29b96b54 401
942fd319 402 if (last_idx != ULONG_MAX)
b9109b0e 403 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b 404
cf52b27a
CY		 405 if (err) {
406 /*
407 * try to revoke all committed pages, but still we could fail
408 * due to no memory or other reason, if that happened, EAGAIN
409 * will be returned, which means in such case, transaction is
410 * already not integrity, caller should use journal to do the
411 * recovery or rewrite & commit last transaction. For other
412 * error number, revoking was done by filesystem itself.
413 */
414 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
415
416 /* drop all uncommitted pages */
417 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
418 } else {
419 __revoke_inmem_pages(inode, &revoke_list, false, false);
420 }
28bc106b 421
29b96b54
CY		 422 return err;
423}
424
425int commit_inmem_pages(struct inode *inode)
426{
427 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
428 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b 429 int err;
29b96b54
CY		 430
431 f2fs_balance_fs(sbi, true);
432 f2fs_lock_op(sbi);
433
5fe45743
CY		 434 set_inode_flag(inode, FI_ATOMIC_COMMIT);
435
29b96b54 436 mutex_lock(&fi->inmem_lock);
cf52b27a 437 err = __commit_inmem_pages(inode);
28bc106b 438
57864ae5
JK		 439 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
440 if (!list_empty(&fi->inmem_ilist))
441 list_del_init(&fi->inmem_ilist);
442 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 443 mutex_unlock(&fi->inmem_lock);
444
5fe45743
CY		 445 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
446
29b96b54 447 f2fs_unlock_op(sbi);
edb27dee 448 return err;
88b88a66
JK		 449}
450
0a8165d7 451/*
351df4b2
JK		 452 * This function balances dirty node and dentry pages.
453 * In addition, it controls garbage collection.
454 */
2c4db1a6 455void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 456{
0f348028 457#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 458 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
459 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 460 f2fs_stop_checkpoint(sbi, false);
55523519 461 }
0f348028
CY		 462#endif
463
e589c2c4 464 /* balance_fs_bg is able to be pending */
a7881893 465 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 466 f2fs_balance_fs_bg(sbi);
467
351df4b2 468 /*
029cd28c
JK		 469 * We should do GC or end up with checkpoint, if there are so many dirty
470 * dir/node pages without enough free segments.
351df4b2 471 */
7f3037a5 472 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 473 mutex_lock(&sbi->gc_mutex);
e066b83c 474 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 475 }
476}
477
4660f9c0
JK		 478void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
479{
1dcc336b 480 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 481 if (!available_free_memory(sbi, EXTENT_CACHE))
482 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 483
1b38dc8e
JK		 484 /* check the # of cached NAT entries */
485 if (!available_free_memory(sbi, NAT_ENTRIES))
486 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
487
31696580 488 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 489 try_to_free_nids(sbi, MAX_FREE_NIDS);
490 else
22ad0b6a 491 build_free_nids(sbi, false, false);
31696580 492
1c0f4bf5 493 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 494 return;
31696580 495
1b38dc8e
JK		 496 /* checkpoint is the only way to shrink partial cached entries */
497 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 498 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 499 excess_prefree_segs(sbi) ||
500 excess_dirty_nats(sbi) ||
f455c8a5 501 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 502 if (test_opt(sbi, DATA_FLUSH)) {
503 struct blk_plug plug;
504
505 blk_start_plug(&plug);
36b35a0d 506 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 507 blk_finish_plug(&plug);
508 }
4660f9c0 509 f2fs_sync_fs(sbi->sb, true);
42190d2a 510 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 511 }
4660f9c0
JK		 512}
513
20fda56b
KM		 514static int __submit_flush_wait(struct f2fs_sb_info *sbi,
515 struct block_device *bdev)
3c62be17 516{
d62fe971 517 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 518 int ret;
519
3adc5fcb 520 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 521 bio_set_dev(bio, bdev);
3c62be17
JK		 522 ret = submit_bio_wait(bio);
523 bio_put(bio);
20fda56b
KM		 524
525 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
526 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 527 return ret;
528}
529
39d787be 530static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 531{
39d787be 532 int ret = 0;
3c62be17
JK		 533 int i;
534
39d787be
CY		 535 if (!sbi->s_ndevs)
536 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 537
39d787be
CY		 538 for (i = 0; i < sbi->s_ndevs; i++) {
539 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
540 continue;
20fda56b
KM		 541 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
542 if (ret)
543 break;
3c62be17
JK		 544 }
545 return ret;
546}
547
2163d198 548static int issue_flush_thread(void *data)
6b4afdd7
JK		 549{
550 struct f2fs_sb_info *sbi = data;
b01a9201 551 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 552 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 553repeat:
554 if (kthread_should_stop())
555 return 0;
556
dc6febb6
CY		 557 sb_start_intwrite(sbi->sb);
558
721bd4d5 559 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 560 struct flush_cmd *cmd, *next;
561 int ret;
562
721bd4d5
GZ		 563 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
564 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
565
39d787be
CY		 566 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
567
568 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 569 atomic_inc(&fcc->issued_flush);
570
721bd4d5
GZ		 571 llist_for_each_entry_safe(cmd, next,
572 fcc->dispatch_list, llnode) {
6b4afdd7 573 cmd->ret = ret;
6b4afdd7
JK		 574 complete(&cmd->wait);
575 }
a688b9d9 576 fcc->dispatch_list = NULL;
6b4afdd7
JK		 577 }
578
dc6febb6
CY		 579 sb_end_intwrite(sbi->sb);
580
a688b9d9 581 wait_event_interruptible(*q,
721bd4d5 582 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 583 goto repeat;
584}
585
39d787be 586int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 587{
b01a9201 588 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 589 struct flush_cmd cmd;
8b8dd65f 590 int ret;
6b4afdd7 591
0f7b2abd
JK		 592 if (test_opt(sbi, NOBARRIER))
593 return 0;
594
8b8dd65f 595 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 596 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 597 atomic_inc(&fcc->issued_flush);
598 return ret;
599 }
740432f8 600
39d787be
CY		 601 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
602 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 603 atomic_dec(&fcc->issing_flush);
604
605 atomic_inc(&fcc->issued_flush);
740432f8
JK		 606 return ret;
607 }
6b4afdd7 608
39d787be 609 cmd.ino = ino;
adf8d90b 610 init_completion(&cmd.wait);
6b4afdd7 611
721bd4d5 612 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 613
6f890df0
CY		 614 /* update issue_list before we wake up issue_flush thread */
615 smp_mb();
616
617 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 618 wake_up(&fcc->flush_wait_queue);
6b4afdd7 619
5eba8c5d
JK		 620 if (fcc->f2fs_issue_flush) {
621 wait_for_completion(&cmd.wait);
8b8dd65f 622 atomic_dec(&fcc->issing_flush);
5eba8c5d 623 } else {
d3238691
CY		 624 struct llist_node *list;
625
626 list = llist_del_all(&fcc->issue_list);
627 if (!list) {
628 wait_for_completion(&cmd.wait);
629 atomic_dec(&fcc->issing_flush);
630 } else {
631 struct flush_cmd *tmp, *next;
632
39d787be 633 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 634
635 llist_for_each_entry_safe(tmp, next, list, llnode) {
636 if (tmp == &cmd) {
637 cmd.ret = ret;
638 atomic_dec(&fcc->issing_flush);
639 continue;
640 }
641 tmp->ret = ret;
642 complete(&tmp->wait);
643 }
644 }
5eba8c5d 645 }
adf8d90b
CY		 646
647 return cmd.ret;
6b4afdd7
JK		 648}
649
2163d198
GZ		 650int create_flush_cmd_control(struct f2fs_sb_info *sbi)
651{
652 dev_t dev = sbi->sb->s_bdev->bd_dev;
653 struct flush_cmd_control *fcc;
654 int err = 0;
655
b01a9201
JK		 656 if (SM_I(sbi)->fcc_info) {
657 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 658 if (fcc->f2fs_issue_flush)
659 return err;
5eba8c5d
JK		 660 goto init_thread;
661 }
662
acbf054d 663 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 664 if (!fcc)
665 return -ENOMEM;
8b8dd65f
CY		 666 atomic_set(&fcc->issued_flush, 0);
667 atomic_set(&fcc->issing_flush, 0);
2163d198 668 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 669 init_llist_head(&fcc->issue_list);
b01a9201 670 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 671 if (!test_opt(sbi, FLUSH_MERGE))
672 return err;
673
5eba8c5d 674init_thread:
2163d198
GZ		 675 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
676 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
677 if (IS_ERR(fcc->f2fs_issue_flush)) {
678 err = PTR_ERR(fcc->f2fs_issue_flush);
679 kfree(fcc);
b01a9201 680 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 681 return err;
682 }
2163d198
GZ		 683
684 return err;
685}
686
5eba8c5d 687void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 688{
b01a9201 689 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 690
5eba8c5d
JK		 691 if (fcc && fcc->f2fs_issue_flush) {
692 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
693
694 fcc->f2fs_issue_flush = NULL;
695 kthread_stop(flush_thread);
696 }
697 if (free) {
698 kfree(fcc);
b01a9201 699 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 700 }
2163d198
GZ		 701}
702
1228b482
CY		 703int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
704{
705 int ret = 0, i;
706
707 if (!sbi->s_ndevs)
708 return 0;
709
710 for (i = 1; i < sbi->s_ndevs; i++) {
711 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
712 continue;
713 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
714 if (ret)
715 break;
716
717 spin_lock(&sbi->dev_lock);
718 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
719 spin_unlock(&sbi->dev_lock);
720 }
721
722 return ret;
723}
724
351df4b2
JK		 725static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
726 enum dirty_type dirty_type)
727{
728 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
729
730 /* need not be added */
731 if (IS_CURSEG(sbi, segno))
732 return;
733
734 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
735 dirty_i->nr_dirty[dirty_type]++;
736
737 if (dirty_type == DIRTY) {
738 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 739 enum dirty_type t = sentry->type;
b2f2c390 740
ec325b52
JK		 741 if (unlikely(t >= DIRTY)) {
742 f2fs_bug_on(sbi, 1);
743 return;
744 }
4625d6aa
CL		 745 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
746 dirty_i->nr_dirty[t]++;
351df4b2
JK		 747 }
748}
749
750static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
751 enum dirty_type dirty_type)
752{
753 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
754
755 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
756 dirty_i->nr_dirty[dirty_type]--;
757
758 if (dirty_type == DIRTY) {
4625d6aa
CL		 759 struct seg_entry *sentry = get_seg_entry(sbi, segno);
760 enum dirty_type t = sentry->type;
761
762 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
763 dirty_i->nr_dirty[t]--;
b2f2c390 764
302bd348 765 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 766 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 767 dirty_i->victim_secmap);
351df4b2
JK		 768 }
769}
770
0a8165d7 771/*
351df4b2
JK		 772 * Should not occur error such as -ENOMEM.
773 * Adding dirty entry into seglist is not critical operation.
774 * If a given segment is one of current working segments, it won't be added.
775 */
8d8451af 776static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 777{
778 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
779 unsigned short valid_blocks;
780
781 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
782 return;
783
784 mutex_lock(&dirty_i->seglist_lock);
785
302bd348 786 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 787
788 if (valid_blocks == 0) {
789 __locate_dirty_segment(sbi, segno, PRE);
790 __remove_dirty_segment(sbi, segno, DIRTY);
791 } else if (valid_blocks < sbi->blocks_per_seg) {
792 __locate_dirty_segment(sbi, segno, DIRTY);
793 } else {
794 /* Recovery routine with SSR needs this */
795 __remove_dirty_segment(sbi, segno, DIRTY);
796 }
797
798 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 799}
800
004b6862 801static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 802 struct block_device *bdev, block_t lstart,
803 block_t start, block_t len)
275b66b0 804{
0b54fb84 805 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 806 struct list_head *pend_list;
b01a9201 807 struct discard_cmd *dc;
275b66b0 808
ba48a33e
CY		 809 f2fs_bug_on(sbi, !len);
810
811 pend_list = &dcc->pend_list[plist_idx(len)];
812
b01a9201
JK		 813 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
814 INIT_LIST_HEAD(&dc->list);
c81abe34 815 dc->bdev = bdev;
b01a9201 816 dc->lstart = lstart;
c81abe34 817 dc->start = start;
b01a9201 818 dc->len = len;
ec9895ad 819 dc->ref = 0;
15469963 820 dc->state = D_PREP;
c81abe34 821 dc->error = 0;
b01a9201 822 init_completion(&dc->wait);
22d375dd 823 list_add_tail(&dc->list, pend_list);
5f32366a 824 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 825 dcc->undiscard_blks += len;
004b6862
CY		 826
827 return dc;
828}
829
830static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
831 struct block_device *bdev, block_t lstart,
832 block_t start, block_t len,
833 struct rb_node *parent, struct rb_node **p)
834{
835 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
836 struct discard_cmd *dc;
837
838 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
839
840 rb_link_node(&dc->rb_node, parent, p);
841 rb_insert_color(&dc->rb_node, &dcc->root);
842
843 return dc;
15469963
JK		 844}
845
004b6862
CY		 846static void __detach_discard_cmd(struct discard_cmd_control *dcc,
847 struct discard_cmd *dc)
15469963 848{
dcc9165d 849 if (dc->state == D_DONE)
004b6862
CY		 850 atomic_dec(&dcc->issing_discard);
851
852 list_del(&dc->list);
853 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 854 dcc->undiscard_blks -= dc->len;
004b6862
CY		 855
856 kmem_cache_free(discard_cmd_slab, dc);
857
858 atomic_dec(&dcc->discard_cmd_cnt);
859}
860
861static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
862 struct discard_cmd *dc)
863{
864 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 865
2ec6f2ef
CY		 866 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
867
d9703d90
CY		 868 f2fs_bug_on(sbi, dc->ref);
869
c81abe34
JK		 870 if (dc->error == -EOPNOTSUPP)
871 dc->error = 0;
15469963 872
c81abe34 873 if (dc->error)
15469963 874 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 875 "Issue discard(%u, %u, %u) failed, ret: %d",
876 dc->lstart, dc->start, dc->len, dc->error);
004b6862 877 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 878}
879
c81abe34
JK		 880static void f2fs_submit_discard_endio(struct bio *bio)
881{
882 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
883
4e4cbee9 884 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 885 dc->state = D_DONE;
e31b9821 886 complete_all(&dc->wait);
c81abe34
JK		 887 bio_put(bio);
888}
889
94b1e10e 890static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 891 block_t start, block_t end)
892{
893#ifdef CONFIG_F2FS_CHECK_FS
894 struct seg_entry *sentry;
895 unsigned int segno;
896 block_t blk = start;
897 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
898 unsigned long *map;
899
900 while (blk < end) {
901 segno = GET_SEGNO(sbi, blk);
902 sentry = get_seg_entry(sbi, segno);
903 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
904
008396e1
YS		 905 if (end < START_BLOCK(sbi, segno + 1))
906 size = GET_BLKOFF_FROM_SEG0(sbi, end);
907 else
908 size = max_blocks;
6915ea9d
CY		 909 map = (unsigned long *)(sentry->cur_valid_map);
910 offset = __find_rev_next_bit(map, size, offset);
911 f2fs_bug_on(sbi, offset != size);
008396e1 912 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 913 }
914#endif
915}
916
8bb4f253
JK		 917static void __init_discard_policy(struct f2fs_sb_info *sbi,
918 struct discard_policy *dpolicy,
919 int discard_type, unsigned int granularity)
920{
921 /* common policy */
922 dpolicy->type = discard_type;
923 dpolicy->sync = true;
924 dpolicy->granularity = granularity;
925
926 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
927 dpolicy->io_aware_gran = MAX_PLIST_NUM;
928
929 if (discard_type == DPOLICY_BG) {
930 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
931 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
932 dpolicy->io_aware = true;
cba60849 933 dpolicy->sync = false;
8bb4f253
JK		 934 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
935 dpolicy->granularity = 1;
936 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
937 }
938 } else if (discard_type == DPOLICY_FORCE) {
939 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
940 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
941 dpolicy->io_aware = false;
942 } else if (discard_type == DPOLICY_FSTRIM) {
943 dpolicy->io_aware = false;
944 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 945 dpolicy->max_requests = UINT_MAX;
8bb4f253
JK		 946 dpolicy->io_aware = false;
947 }
948}
949
950
c81abe34
JK		 951/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
952static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56
CY		 953 struct discard_policy *dpolicy,
954 struct discard_cmd *dc)
c81abe34
JK		 955{
956 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 957 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
958 &(dcc->fstrim_list) : &(dcc->wait_list);
c81abe34 959 struct bio *bio = NULL;
78997b56 960 int flag = dpolicy->sync ? REQ_SYNC : 0;
c81abe34
JK		 961
962 if (dc->state != D_PREP)
963 return;
964
d6184774
YH		 965 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
966 return;
967
0243a5f9
CY		 968 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
969
c81abe34
JK		 970 dc->error = __blkdev_issue_discard(dc->bdev,
971 SECTOR_FROM_BLOCK(dc->start),
972 SECTOR_FROM_BLOCK(dc->len),
973 GFP_NOFS, 0, &bio);
974 if (!dc->error) {
975 /* should keep before submission to avoid D_DONE right away */
976 dc->state = D_SUBMIT;
8b8dd65f
CY		 977 atomic_inc(&dcc->issued_discard);
978 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 979 if (bio) {
980 bio->bi_private = dc;
981 bio->bi_end_io = f2fs_submit_discard_endio;
ecc9aa00 982 bio->bi_opf |= flag;
c81abe34 983 submit_bio(bio);
8412663d 984 list_move_tail(&dc->list, wait_list);
6915ea9d 985 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 986
987 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 988 }
989 } else {
990 __remove_discard_cmd(sbi, dc);
991 }
992}
993
004b6862
CY		 994static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
995 struct block_device *bdev, block_t lstart,
996 block_t start, block_t len,
997 struct rb_node **insert_p,
998 struct rb_node *insert_parent)
c81abe34 999{
004b6862 1000 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1001 struct rb_node **p;
004b6862
CY		 1002 struct rb_node *parent = NULL;
1003 struct discard_cmd *dc = NULL;
1004
1005 if (insert_p && insert_parent) {
1006 parent = insert_parent;
1007 p = insert_p;
1008 goto do_insert;
1009 }
c81abe34 1010
004b6862
CY		 1011 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1012do_insert:
1013 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1014 if (!dc)
1015 return NULL;
c81abe34 1016
004b6862 1017 return dc;
c81abe34
JK		 1018}
1019
ba48a33e
CY		 1020static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1021 struct discard_cmd *dc)
1022{
1023 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1024}
1025
3d6a650f
YH		 1026static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1027 struct discard_cmd *dc, block_t blkaddr)
1028{
ba48a33e 1029 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1030 struct discard_info di = dc->di;
1031 bool modified = false;
3d6a650f 1032
004b6862 1033 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1034 __remove_discard_cmd(sbi, dc);
1035 return;
1036 }
1037
d84d1cbd
CY		 1038 dcc->undiscard_blks -= di.len;
1039
004b6862 1040 if (blkaddr > di.lstart) {
3d6a650f 1041 dc->len = blkaddr - dc->lstart;
d84d1cbd 1042 dcc->undiscard_blks += dc->len;
ba48a33e 1043 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1044 modified = true;
1045 }
1046
1047 if (blkaddr < di.lstart + di.len - 1) {
1048 if (modified) {
1049 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1050 di.start + blkaddr + 1 - di.lstart,
1051 di.lstart + di.len - 1 - blkaddr,
1052 NULL, NULL);
1053 } else {
1054 dc->lstart++;
1055 dc->len--;
1056 dc->start++;
d84d1cbd 1057 dcc->undiscard_blks += dc->len;
ba48a33e 1058 __relocate_discard_cmd(dcc, dc);
004b6862 1059 }
3d6a650f
YH		 1060 }
1061}
1062
004b6862
CY		 1063static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1064 struct block_device *bdev, block_t lstart,
1065 block_t start, block_t len)
275b66b0 1066{
0b54fb84 1067 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1068 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1069 struct discard_cmd *dc;
1070 struct discard_info di = {0};
1071 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1072 block_t end = lstart + len;
275b66b0 1073
15469963 1074 mutex_lock(&dcc->cmd_lock);
40465257 1075
004b6862
CY		 1076 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1077 NULL, lstart,
1078 (struct rb_entry **)&prev_dc,
1079 (struct rb_entry **)&next_dc,
1080 &insert_p, &insert_parent, true);
1081 if (dc)
1082 prev_dc = dc;
1083
1084 if (!prev_dc) {
1085 di.lstart = lstart;
1086 di.len = next_dc ? next_dc->lstart - lstart : len;
1087 di.len = min(di.len, len);
1088 di.start = start;
22d375dd 1089 }
15469963 1090
004b6862
CY		 1091 while (1) {
1092 struct rb_node *node;
1093 bool merged = false;
1094 struct discard_cmd *tdc = NULL;
1095
1096 if (prev_dc) {
1097 di.lstart = prev_dc->lstart + prev_dc->len;
1098 if (di.lstart < lstart)
1099 di.lstart = lstart;
1100 if (di.lstart >= end)
1101 break;
1102
1103 if (!next_dc || next_dc->lstart > end)
1104 di.len = end - di.lstart;
1105 else
1106 di.len = next_dc->lstart - di.lstart;
1107 di.start = start + di.lstart - lstart;
1108 }
1109
1110 if (!di.len)
1111 goto next;
1112
1113 if (prev_dc && prev_dc->state == D_PREP &&
1114 prev_dc->bdev == bdev &&
1115 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1116 prev_dc->di.len += di.len;
d84d1cbd 1117 dcc->undiscard_blks += di.len;
ba48a33e 1118 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1119 di = prev_dc->di;
1120 tdc = prev_dc;
1121 merged = true;
1122 }
1123
1124 if (next_dc && next_dc->state == D_PREP &&
1125 next_dc->bdev == bdev &&
1126 __is_discard_front_mergeable(&di, &next_dc->di)) {
1127 next_dc->di.lstart = di.lstart;
1128 next_dc->di.len += di.len;
1129 next_dc->di.start = di.start;
d84d1cbd 1130 dcc->undiscard_blks += di.len;
ba48a33e 1131 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1132 if (tdc)
1133 __remove_discard_cmd(sbi, tdc);
004b6862 1134 merged = true;
4e6a8d9b 1135 }
004b6862 1136
df0f6b44 1137 if (!merged) {
004b6862
CY		 1138 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1139 di.len, NULL, NULL);
df0f6b44 1140 }
004b6862
CY		 1141 next:
1142 prev_dc = next_dc;
1143 if (!prev_dc)
1144 break;
1145
1146 node = rb_next(&prev_dc->rb_node);
1147 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1148 }
1149
1150 mutex_unlock(&dcc->cmd_lock);
1151}
1152
1153static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1154 struct block_device *bdev, block_t blkstart, block_t blklen)
1155{
1156 block_t lblkstart = blkstart;
1157
0243a5f9 1158 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1159
1160 if (sbi->s_ndevs) {
1161 int devi = f2fs_target_device_index(sbi, blkstart);
1162
1163 blkstart -= FDEV(devi).start_blk;
1164 }
1165 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1166 return 0;
1167}
1168
78997b56
CY		 1169static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1170 struct discard_policy *dpolicy)
bd5b0738
CY		 1171{
1172 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1173 struct list_head *pend_list;
1174 struct discard_cmd *dc, *tmp;
1175 struct blk_plug plug;
78997b56 1176 int i, iter = 0, issued = 0;
e6c6de18 1177 bool io_interrupted = false;
bd5b0738 1178
78997b56
CY		 1179 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1180 if (i + 1 < dpolicy->granularity)
1181 break;
bd5b0738 1182 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1183
1184 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1185 if (list_empty(pend_list))
1186 goto next;
33da62cf
CY		 1187 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1188 blk_start_plug(&plug);
bd5b0738
CY		 1189 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1190 f2fs_bug_on(sbi, dc->state != D_PREP);
1191
ecc9aa00
CY		 1192 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1193 !is_idle(sbi)) {
e6c6de18 1194 io_interrupted = true;
ecc9aa00 1195 goto skip;
969d1b18 1196 }
e6c6de18 1197
78997b56 1198 __submit_discard_cmd(sbi, dpolicy, dc);
ecc9aa00
CY		 1199 issued++;
1200skip:
1201 if (++iter >= dpolicy->max_requests)
33da62cf 1202 break;
bd5b0738 1203 }
33da62cf 1204 blk_finish_plug(&plug);
49c60c67 1205next:
33da62cf
CY		 1206 mutex_unlock(&dcc->cmd_lock);
1207
1208 if (iter >= dpolicy->max_requests)
1209 break;
bd5b0738 1210 }
969d1b18 1211
e6c6de18
CY		 1212 if (!issued && io_interrupted)
1213 issued = -1;
1214
969d1b18
CY		 1215 return issued;
1216}
1217
cf5c759f 1218static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1219{
1220 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1221 struct list_head *pend_list;
1222 struct discard_cmd *dc, *tmp;
1223 int i;
cf5c759f 1224 bool dropped = false;
969d1b18
CY		 1225
1226 mutex_lock(&dcc->cmd_lock);
1227 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1228 pend_list = &dcc->pend_list[i];
1229 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1230 f2fs_bug_on(sbi, dc->state != D_PREP);
1231 __remove_discard_cmd(sbi, dc);
cf5c759f 1232 dropped = true;
969d1b18
CY		 1233 }
1234 }
1235 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1236
1237 return dropped;
bd5b0738
CY		 1238}
1239
7950e9ac
CY		 1240void drop_discard_cmd(struct f2fs_sb_info *sbi)
1241{
1242 __drop_discard_cmd(sbi);
1243}
1244
0ea80512 1245static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1246 struct discard_cmd *dc)
1247{
1248 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1249 unsigned int len = 0;
2a510c00
CY		 1250
1251 wait_for_completion_io(&dc->wait);
1252 mutex_lock(&dcc->cmd_lock);
1253 f2fs_bug_on(sbi, dc->state != D_DONE);
1254 dc->ref--;
0ea80512
CY		 1255 if (!dc->ref) {
1256 if (!dc->error)
1257 len = dc->len;
2a510c00 1258 __remove_discard_cmd(sbi, dc);
0ea80512 1259 }
2a510c00 1260 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1261
1262 return len;
2a510c00
CY		 1263}
1264
0ea80512 1265static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1266 struct discard_policy *dpolicy,
1267 block_t start, block_t end)
63a94fa1
CY		 1268{
1269 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1270 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1271 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1272 struct discard_cmd *dc, *tmp;
6afae633 1273 bool need_wait;
0ea80512 1274 unsigned int trimmed = 0;
6afae633
CY		 1275
1276next:
1277 need_wait = false;
63a94fa1
CY		 1278
1279 mutex_lock(&dcc->cmd_lock);
1280 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1281 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1282 continue;
78997b56 1283 if (dc->len < dpolicy->granularity)
8412663d 1284 continue;
78997b56 1285 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1286 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1287 if (!dc->error)
1288 trimmed += dc->len;
63a94fa1 1289 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1290 } else {
1291 dc->ref++;
1292 need_wait = true;
1293 break;
63a94fa1
CY		 1294 }
1295 }
1296 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1297
1298 if (need_wait) {
0ea80512 1299 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1300 goto next;
1301 }
0ea80512
CY		 1302
1303 return trimmed;
63a94fa1
CY		 1304}
1305
78997b56
CY		 1306static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1307 struct discard_policy *dpolicy)
8412663d 1308{
9a997188
JK		 1309 struct discard_policy dp;
1310
1311 if (dpolicy) {
1312 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1313 return;
1314 }
1315
1316 /* wait all */
8bb4f253 1317 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
9a997188 1318 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1319 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
9a997188 1320 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8412663d
CY		 1321}
1322
004b6862 1323/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1324static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1325{
1326 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1327 struct discard_cmd *dc;
ec9895ad 1328 bool need_wait = false;
004b6862
CY		 1329
1330 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1331 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1332 if (dc) {
ec9895ad
CY		 1333 if (dc->state == D_PREP) {
1334 __punch_discard_cmd(sbi, dc, blkaddr);
1335 } else {
1336 dc->ref++;
1337 need_wait = true;
1338 }
275b66b0 1339 }
d431413f 1340 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1341
2a510c00
CY		 1342 if (need_wait)
1343 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1344}
1345
cce13252
CY		 1346void stop_discard_thread(struct f2fs_sb_info *sbi)
1347{
1348 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1349
1350 if (dcc && dcc->f2fs_issue_discard) {
1351 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1352
1353 dcc->f2fs_issue_discard = NULL;
1354 kthread_stop(discard_thread);
ec9895ad 1355 }
d431413f
CY		 1356}
1357
8412663d 1358/* This comes from f2fs_put_super */
cf5c759f 1359bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1360{
1361 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1362 struct discard_policy dpolicy;
cf5c759f 1363 bool dropped;
969d1b18 1364
8bb4f253
JK		 1365 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1366 dcc->discard_granularity);
78997b56 1367 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1368 dropped = __drop_discard_cmd(sbi);
cf5c759f 1369
9a997188
JK		 1370 /* just to make sure there is no pending discard commands */
1371 __wait_all_discard_cmd(sbi, NULL);
cf5c759f 1372 return dropped;
969d1b18
CY		 1373}
1374
15469963
JK		 1375static int issue_discard_thread(void *data)
1376{
1377 struct f2fs_sb_info *sbi = data;
1378 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1379 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1380 struct discard_policy dpolicy;
969d1b18
CY		 1381 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1382 int issued;
15469963 1383
1d7be270 1384 set_freezable();
15469963 1385
1d7be270 1386 do {
8bb4f253 1387 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1388 dcc->discard_granularity);
1389
969d1b18
CY		 1390 wait_event_interruptible_timeout(*q,
1391 kthread_should_stop() || freezing(current) ||
1392 dcc->discard_wake,
1393 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1394 if (try_to_freeze())
1395 continue;
3b60d802
CY		 1396 if (f2fs_readonly(sbi->sb))
1397 continue;
1d7be270
JK		 1398 if (kthread_should_stop())
1399 return 0;
d6184774
YH		 1400 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1401 wait_ms = dpolicy.max_interval;
1402 continue;
1403 }
15469963 1404
dee02f0d 1405 if (dcc->discard_wake)
969d1b18 1406 dcc->discard_wake = 0;
dee02f0d
JK		 1407
1408 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
8bb4f253 1409 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1410
dc6febb6
CY		 1411 sb_start_intwrite(sbi->sb);
1412
78997b56 1413 issued = __issue_discard_cmd(sbi, &dpolicy);
969d1b18 1414 if (issued) {
78997b56
CY		 1415 __wait_all_discard_cmd(sbi, &dpolicy);
1416 wait_ms = dpolicy.min_interval;
969d1b18 1417 } else {
78997b56 1418 wait_ms = dpolicy.max_interval;
969d1b18 1419 }
1d7be270 1420
dc6febb6 1421 sb_end_intwrite(sbi->sb);
1d7be270 1422
1d7be270
JK		 1423 } while (!kthread_should_stop());
1424 return 0;
15469963
JK		 1425}
1426
f46e8809 1427#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1428static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1429 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1430{
92592285 1431 sector_t sector, nr_sects;
10a875f8 1432 block_t lblkstart = blkstart;
3c62be17
JK		 1433 int devi = 0;
1434
1435 if (sbi->s_ndevs) {
1436 devi = f2fs_target_device_index(sbi, blkstart);
1437 blkstart -= FDEV(devi).start_blk;
1438 }
f46e8809
DLM		 1439
1440 /*
1441 * We need to know the type of the zone: for conventional zones,
1442 * use regular discard if the drive supports it. For sequential
1443 * zones, reset the zone write pointer.
1444 */
3c62be17 1445 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1446
1447 case BLK_ZONE_TYPE_CONVENTIONAL:
1448 if (!blk_queue_discard(bdev_get_queue(bdev)))
1449 return 0;
c81abe34 1450 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1451 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1452 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1453 sector = SECTOR_FROM_BLOCK(blkstart);
1454 nr_sects = SECTOR_FROM_BLOCK(blklen);
1455
1456 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1457 nr_sects != bdev_zone_sectors(bdev)) {
1458 f2fs_msg(sbi->sb, KERN_INFO,
1459 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1460 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1461 blkstart, blklen);
1462 return -EIO;
1463 }
d50aaeec 1464 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1465 return blkdev_reset_zones(bdev, sector,
1466 nr_sects, GFP_NOFS);
1467 default:
1468 /* Unknown zone type: broken device ? */
1469 return -EIO;
1470 }
1471}
1472#endif
1473
3c62be17
JK		 1474static int __issue_discard_async(struct f2fs_sb_info *sbi,
1475 struct block_device *bdev, block_t blkstart, block_t blklen)
1476{
1477#ifdef CONFIG_BLK_DEV_ZONED
ccd31cb2 1478 if (f2fs_sb_has_blkzoned(sbi->sb) &&
3c62be17
JK		 1479 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1480 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1481#endif
c81abe34 1482 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1483}
1484
1e87a78d 1485static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1486 block_t blkstart, block_t blklen)
1487{
3c62be17
JK		 1488 sector_t start = blkstart, len = 0;
1489 struct block_device *bdev;
a66cdd98
JK		 1490 struct seg_entry *se;
1491 unsigned int offset;
1492 block_t i;
3c62be17
JK		 1493 int err = 0;
1494
1495 bdev = f2fs_target_device(sbi, blkstart, NULL);
1496
1497 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1498 if (i != start) {
1499 struct block_device *bdev2 =
1500 f2fs_target_device(sbi, i, NULL);
1501
1502 if (bdev2 != bdev) {
1503 err = __issue_discard_async(sbi, bdev,
1504 start, len);
1505 if (err)
1506 return err;
1507 bdev = bdev2;
1508 start = i;
1509 len = 0;
1510 }
1511 }
a66cdd98 1512
a66cdd98
JK		 1513 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1514 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1515
1516 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1517 sbi->discard_blks--;
1518 }
f46e8809 1519
3c62be17
JK		 1520 if (len)
1521 err = __issue_discard_async(sbi, bdev, start, len);
1522 return err;
1e87a78d
JK		 1523}
1524
25290fa5
JK		 1525static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1526 bool check_only)
adf4983b 1527{
b2955550
JK		 1528 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1529 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1530 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1531 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1532 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1533 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1534 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1535 unsigned int start = 0, end = -1;
c473f1a9 1536 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1537 struct discard_entry *de = NULL;
46f84c2c 1538 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1539 int i;
1540
3e025740 1541 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1542 return false;
b2955550 1543
a66cdd98
JK		 1544 if (!force) {
1545 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1546 SM_I(sbi)->dcc_info->nr_discards >=
1547 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1548 return false;
4b2fecc8
JK		 1549 }
1550
b2955550
JK		 1551 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1552 for (i = 0; i < entries; i++)
a66cdd98 1553 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1554 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1555
0b54fb84
JK		 1556 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1557 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1558 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1559 if (start >= max_blocks)
1560 break;
1561
1562 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1563 if (force && start && end != max_blocks
1564 && (end - start) < cpc->trim_minlen)
1565 continue;
1566
25290fa5
JK		 1567 if (check_only)
1568 return true;
1569
a7eeb823
CY		 1570 if (!de) {
1571 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1572 GFP_F2FS_ZERO);
1573 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1574 list_add_tail(&de->list, head);
1575 }
1576
1577 for (i = start; i < end; i++)
1578 __set_bit_le(i, (void *)de->discard_map);
1579
1580 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1581 }
25290fa5 1582 return false;
b2955550
JK		 1583}
1584
af8ff65b
CY		 1585static void release_discard_addr(struct discard_entry *entry)
1586{
1587 list_del(&entry->list);
1588 kmem_cache_free(discard_entry_slab, entry);
1589}
1590
4b2fecc8
JK		 1591void release_discard_addrs(struct f2fs_sb_info *sbi)
1592{
46f84c2c 1593 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1594 struct discard_entry *entry, *this;
1595
1596 /* drop caches */
af8ff65b
CY		 1597 list_for_each_entry_safe(entry, this, head, list)
1598 release_discard_addr(entry);
4b2fecc8
JK		 1599}
1600
0a8165d7 1601/*
351df4b2
JK		 1602 * Should call clear_prefree_segments after checkpoint is done.
1603 */
1604static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1605{
1606 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1607 unsigned int segno;
351df4b2
JK		 1608
1609 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1610 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1611 __set_test_and_free(sbi, segno);
351df4b2
JK		 1612 mutex_unlock(&dirty_i->seglist_lock);
1613}
1614
836b5a63 1615void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1616{
969d1b18
CY		 1617 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1618 struct list_head *head = &dcc->entry_list;
2d7b822a 1619 struct discard_entry *entry, *this;
351df4b2 1620 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1621 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1622 unsigned int start = 0, end = -1;
36abef4e 1623 unsigned int secno, start_segno;
c473f1a9 1624 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1625
1626 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1627
351df4b2 1628 while (1) {
29e59c14 1629 int i;
7cd8558b
JK		 1630 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1631 if (start >= MAIN_SEGS(sbi))
351df4b2 1632 break;
7cd8558b
JK		 1633 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1634 start + 1);
29e59c14
CL		 1635
1636 for (i = start; i < end; i++)
1637 clear_bit(i, prefree_map);
1638
1639 dirty_i->nr_dirty[PRE] -= end - start;
1640
650d3c4e 1641 if (!test_opt(sbi, DISCARD))
29e59c14 1642 continue;
351df4b2 1643
650d3c4e
YH		 1644 if (force && start >= cpc->trim_start &&
1645 (end - 1) <= cpc->trim_end)
1646 continue;
1647
36abef4e
JK		 1648 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1649 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1650 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1651 continue;
1652 }
1653next:
4ddb1a4d
JK		 1654 secno = GET_SEC_FROM_SEG(sbi, start);
1655 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1656 if (!IS_CURSEC(sbi, secno) &&
302bd348 1657 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1658 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1659 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1660
1661 start = start_segno + sbi->segs_per_sec;
1662 if (start < end)
1663 goto next;
8b107f5b
JK		 1664 else
1665 end = start - 1;
351df4b2
JK		 1666 }
1667 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1668
1669 /* send small discards */
2d7b822a 1670 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1671 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1672 bool is_valid = test_bit_le(0, entry->discard_map);
1673
1674find_next:
1675 if (is_valid) {
1676 next_pos = find_next_zero_bit_le(entry->discard_map,
1677 sbi->blocks_per_seg, cur_pos);
1678 len = next_pos - cur_pos;
1679
ccd31cb2 1680 if (f2fs_sb_has_blkzoned(sbi->sb) ||
acfd2810 1681 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1682 goto skip;
1683
1684 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1685 len);
a7eeb823
CY		 1686 total_len += len;
1687 } else {
1688 next_pos = find_next_bit_le(entry->discard_map,
1689 sbi->blocks_per_seg, cur_pos);
1690 }
836b5a63 1691skip:
a7eeb823
CY		 1692 cur_pos = next_pos;
1693 is_valid = !is_valid;
1694
1695 if (cur_pos < sbi->blocks_per_seg)
1696 goto find_next;
1697
af8ff65b 1698 release_discard_addr(entry);
969d1b18 1699 dcc->nr_discards -= total_len;
b2955550 1700 }
34e159da 1701
01983c71 1702 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1703}
1704
8ed59745 1705static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1706{
15469963 1707 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1708 struct discard_cmd_control *dcc;
ba48a33e 1709 int err = 0, i;
0b54fb84
JK		 1710
1711 if (SM_I(sbi)->dcc_info) {
1712 dcc = SM_I(sbi)->dcc_info;
1713 goto init_thread;
1714 }
1715
acbf054d 1716 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 1717 if (!dcc)
1718 return -ENOMEM;
1719
969d1b18 1720 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1721 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1722 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1723 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 1724 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1725 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1726 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1727 atomic_set(&dcc->issued_discard, 0);
1728 atomic_set(&dcc->issing_discard, 0);
5f32366a 1729 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1730 dcc->nr_discards = 0;
d618ebaf 1731 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1732 dcc->undiscard_blks = 0;
004b6862 1733 dcc->root = RB_ROOT;
0b54fb84 1734
15469963 1735 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1736 SM_I(sbi)->dcc_info = dcc;
1737init_thread:
15469963
JK		 1738 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1739 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1740 if (IS_ERR(dcc->f2fs_issue_discard)) {
1741 err = PTR_ERR(dcc->f2fs_issue_discard);
1742 kfree(dcc);
1743 SM_I(sbi)->dcc_info = NULL;
1744 return err;
1745 }
1746
0b54fb84
JK		 1747 return err;
1748}
1749
f099405f 1750static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1751{
1752 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1753
f099405f
CY		 1754 if (!dcc)
1755 return;
1756
cce13252 1757 stop_discard_thread(sbi);
f099405f
CY		 1758
1759 kfree(dcc);
1760 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1761}
1762
184a5cd2 1763static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1764{
1765 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1766
1767 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1768 sit_i->dirty_sentries++;
184a5cd2
CY		 1769 return false;
1770 }
1771
1772 return true;
351df4b2
JK		 1773}
1774
1775static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1776 unsigned int segno, int modified)
1777{
1778 struct seg_entry *se = get_seg_entry(sbi, segno);
1779 se->type = type;
1780 if (modified)
1781 __mark_sit_entry_dirty(sbi, segno);
1782}
1783
1784static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1785{
1786 struct seg_entry *se;
1787 unsigned int segno, offset;
1788 long int new_vblocks;
6415fedc
YS		 1789 bool exist;
1790#ifdef CONFIG_F2FS_CHECK_FS
1791 bool mir_exist;
1792#endif
351df4b2
JK		 1793
1794 segno = GET_SEGNO(sbi, blkaddr);
1795
1796 se = get_seg_entry(sbi, segno);
1797 new_vblocks = se->valid_blocks + del;
491c0854 1798 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1799
9850cf4a 1800 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1801 (new_vblocks > sbi->blocks_per_seg)));
1802
1803 se->valid_blocks = new_vblocks;
1804 se->mtime = get_mtime(sbi);
1805 SIT_I(sbi)->max_mtime = se->mtime;
1806
1807 /* Update valid block bitmap */
1808 if (del > 0) {
6415fedc 1809 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1810#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1811 mir_exist = f2fs_test_and_set_bit(offset,
1812 se->cur_valid_map_mir);
1813 if (unlikely(exist != mir_exist)) {
1814 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1815 "when setting bitmap, blk:%u, old bit:%d",
1816 blkaddr, exist);
05796763 1817 f2fs_bug_on(sbi, 1);
6415fedc 1818 }
355e7891 1819#endif
6415fedc
YS		 1820 if (unlikely(exist)) {
1821 f2fs_msg(sbi->sb, KERN_ERR,
1822 "Bitmap was wrongly set, blk:%u", blkaddr);
1823 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1824 se->valid_blocks--;
1825 del = 0;
355e7891 1826 }
6415fedc 1827
3e025740
JK		 1828 if (f2fs_discard_en(sbi) &&
1829 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1830 sbi->discard_blks--;
720037f9
JK		 1831
1832 /* don't overwrite by SSR to keep node chain */
5d7881ca 1833 if (IS_NODESEG(se->type)) {
720037f9
JK		 1834 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1835 se->ckpt_valid_blocks++;
1836 }
351df4b2 1837 } else {
6415fedc 1838 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1839#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1840 mir_exist = f2fs_test_and_clear_bit(offset,
1841 se->cur_valid_map_mir);
1842 if (unlikely(exist != mir_exist)) {
1843 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1844 "when clearing bitmap, blk:%u, old bit:%d",
1845 blkaddr, exist);
05796763 1846 f2fs_bug_on(sbi, 1);
6415fedc 1847 }
355e7891 1848#endif
6415fedc
YS		 1849 if (unlikely(!exist)) {
1850 f2fs_msg(sbi->sb, KERN_ERR,
1851 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1852 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1853 se->valid_blocks++;
1854 del = 0;
355e7891 1855 }
6415fedc 1856
3e025740
JK		 1857 if (f2fs_discard_en(sbi) &&
1858 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1859 sbi->discard_blks++;
351df4b2
JK		 1860 }
1861 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1862 se->ckpt_valid_blocks += del;
1863
1864 __mark_sit_entry_dirty(sbi, segno);
1865
1866 /* update total number of valid blocks to be written in ckpt area */
1867 SIT_I(sbi)->written_valid_blocks += del;
1868
1869 if (sbi->segs_per_sec > 1)
1870 get_sec_entry(sbi, segno)->valid_blocks += del;
1871}
1872
351df4b2
JK		 1873void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1874{
1875 unsigned int segno = GET_SEGNO(sbi, addr);
1876 struct sit_info *sit_i = SIT_I(sbi);
1877
9850cf4a 1878 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1879 if (addr == NEW_ADDR)
1880 return;
1881
1882 /* add it into sit main buffer */
3d26fa6b 1883 down_write(&sit_i->sentry_lock);
351df4b2
JK		 1884
1885 update_sit_entry(sbi, addr, -1);
1886
1887 /* add it into dirty seglist */
1888 locate_dirty_segment(sbi, segno);
1889
3d26fa6b 1890 up_write(&sit_i->sentry_lock);
351df4b2
JK		 1891}
1892
6e2c64ad
JK		 1893bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1894{
1895 struct sit_info *sit_i = SIT_I(sbi);
1896 unsigned int segno, offset;
1897 struct seg_entry *se;
1898 bool is_cp = false;
1899
7b525dd0 1900 if (!is_valid_blkaddr(blkaddr))
6e2c64ad
JK		 1901 return true;
1902
3d26fa6b 1903 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1904
1905 segno = GET_SEGNO(sbi, blkaddr);
1906 se = get_seg_entry(sbi, segno);
1907 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1908
1909 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1910 is_cp = true;
1911
3d26fa6b 1912 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1913
1914 return is_cp;
1915}
1916
0a8165d7 1917/*
351df4b2
JK		 1918 * This function should be resided under the curseg_mutex lock
1919 */
1920static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1921 struct f2fs_summary *sum)
351df4b2
JK		 1922{
1923 struct curseg_info *curseg = CURSEG_I(sbi, type);
1924 void *addr = curseg->sum_blk;
e79efe3b 1925 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1926 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1927}
1928
0a8165d7 1929/*
351df4b2
JK		 1930 * Calculate the number of current summary pages for writing
1931 */
3fa06d7b 1932int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1933{
351df4b2 1934 int valid_sum_count = 0;
9a47938b 1935 int i, sum_in_page;
351df4b2
JK		 1936
1937 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1938 if (sbi->ckpt->alloc_type[i] == SSR)
1939 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1940 else {
1941 if (for_ra)
1942 valid_sum_count += le16_to_cpu(
1943 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1944 else
1945 valid_sum_count += curseg_blkoff(sbi, i);
1946 }
351df4b2
JK		 1947 }
1948
09cbfeaf 1949 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1950 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1951 if (valid_sum_count <= sum_in_page)
351df4b2 1952 return 1;
9a47938b 1953 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1954 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1955 return 2;
1956 return 3;
1957}
1958
0a8165d7 1959/*
351df4b2
JK		 1960 * Caller should put this summary page
1961 */
1962struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1963{
1964 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1965}
1966
381722d2 1967void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1968{
1969 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2 1970
0537b811 1971 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 1972 set_page_dirty(page);
1973 f2fs_put_page(page, 1);
1974}
1975
381722d2
CY		 1976static void write_sum_page(struct f2fs_sb_info *sbi,
1977 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1978{
1979 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1980}
1981
b7ad7512
CY		 1982static void write_current_sum_page(struct f2fs_sb_info *sbi,
1983 int type, block_t blk_addr)
1984{
1985 struct curseg_info *curseg = CURSEG_I(sbi, type);
1986 struct page *page = grab_meta_page(sbi, blk_addr);
1987 struct f2fs_summary_block *src = curseg->sum_blk;
1988 struct f2fs_summary_block *dst;
1989
1990 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 1991 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 1992
1993 mutex_lock(&curseg->curseg_mutex);
1994
1995 down_read(&curseg->journal_rwsem);
1996 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1997 up_read(&curseg->journal_rwsem);
1998
1999 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2000 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2001
2002 mutex_unlock(&curseg->curseg_mutex);
2003
2004 set_page_dirty(page);
2005 f2fs_put_page(page, 1);
2006}
2007
a7881893
JK		 2008static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2009{
2010 struct curseg_info *curseg = CURSEG_I(sbi, type);
2011 unsigned int segno = curseg->segno + 1;
2012 struct free_segmap_info *free_i = FREE_I(sbi);
2013
2014 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2015 return !test_bit(segno, free_i->free_segmap);
2016 return 0;
2017}
2018
0a8165d7 2019/*
351df4b2
JK		 2020 * Find a new segment from the free segments bitmap to right order
2021 * This function should be returned with success, otherwise BUG
2022 */
2023static void get_new_segment(struct f2fs_sb_info *sbi,
2024 unsigned int *newseg, bool new_sec, int dir)
2025{
2026 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2027 unsigned int segno, secno, zoneno;
7cd8558b 2028 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2029 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2030 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2031 unsigned int left_start = hint;
2032 bool init = true;
2033 int go_left = 0;
2034 int i;
2035
1a118ccf 2036 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2037
2038 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2039 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2040 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2041 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2042 goto got_it;
2043 }
2044find_other_zone:
7cd8558b
JK		 2045 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2046 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2047 if (dir == ALLOC_RIGHT) {
2048 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2049 MAIN_SECS(sbi), 0);
2050 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2051 } else {
2052 go_left = 1;
2053 left_start = hint - 1;
2054 }
2055 }
2056 if (go_left == 0)
2057 goto skip_left;
2058
2059 while (test_bit(left_start, free_i->free_secmap)) {
2060 if (left_start > 0) {
2061 left_start--;
2062 continue;
2063 }
2064 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2065 MAIN_SECS(sbi), 0);
2066 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2067 break;
2068 }
2069 secno = left_start;
2070skip_left:
4ddb1a4d
JK		 2071 segno = GET_SEG_FROM_SEC(sbi, secno);
2072 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2073
2074 /* give up on finding another zone */
2075 if (!init)
2076 goto got_it;
2077 if (sbi->secs_per_zone == 1)
2078 goto got_it;
2079 if (zoneno == old_zoneno)
2080 goto got_it;
2081 if (dir == ALLOC_LEFT) {
2082 if (!go_left && zoneno + 1 >= total_zones)
2083 goto got_it;
2084 if (go_left && zoneno == 0)
2085 goto got_it;
2086 }
2087 for (i = 0; i < NR_CURSEG_TYPE; i++)
2088 if (CURSEG_I(sbi, i)->zone == zoneno)
2089 break;
2090
2091 if (i < NR_CURSEG_TYPE) {
2092 /* zone is in user, try another */
2093 if (go_left)
2094 hint = zoneno * sbi->secs_per_zone - 1;
2095 else if (zoneno + 1 >= total_zones)
2096 hint = 0;
2097 else
2098 hint = (zoneno + 1) * sbi->secs_per_zone;
2099 init = false;
2100 goto find_other_zone;
2101 }
2102got_it:
2103 /* set it as dirty segment in free segmap */
9850cf4a 2104 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2105 __set_inuse(sbi, segno);
2106 *newseg = segno;
1a118ccf 2107 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2108}
2109
2110static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2111{
2112 struct curseg_info *curseg = CURSEG_I(sbi, type);
2113 struct summary_footer *sum_footer;
2114
2115 curseg->segno = curseg->next_segno;
4ddb1a4d 2116 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2117 curseg->next_blkoff = 0;
2118 curseg->next_segno = NULL_SEGNO;
2119
2120 sum_footer = &(curseg->sum_blk->footer);
2121 memset(sum_footer, 0, sizeof(struct summary_footer));
2122 if (IS_DATASEG(type))
2123 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2124 if (IS_NODESEG(type))
2125 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2126 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2127}
2128
7a20b8a6
JK		 2129static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2130{
a7881893
JK		 2131 /* if segs_per_sec is large than 1, we need to keep original policy. */
2132 if (sbi->segs_per_sec != 1)
2133 return CURSEG_I(sbi, type)->segno;
2134
b94929d9
YS		 2135 if (test_opt(sbi, NOHEAP) &&
2136 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2137 return 0;
2138
e066b83c
JK		 2139 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2140 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2141
2142 /* find segments from 0 to reuse freed segments */
63189b78 2143 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2144 return 0;
2145
7a20b8a6
JK		 2146 return CURSEG_I(sbi, type)->segno;
2147}
2148
0a8165d7 2149/*
351df4b2
JK		 2150 * Allocate a current working segment.
2151 * This function always allocates a free segment in LFS manner.
2152 */
2153static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2154{
2155 struct curseg_info *curseg = CURSEG_I(sbi, type);
2156 unsigned int segno = curseg->segno;
2157 int dir = ALLOC_LEFT;
2158
2159 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2160 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2161 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2162 dir = ALLOC_RIGHT;
2163
2164 if (test_opt(sbi, NOHEAP))
2165 dir = ALLOC_RIGHT;
2166
7a20b8a6 2167 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2168 get_new_segment(sbi, &segno, new_sec, dir);
2169 curseg->next_segno = segno;
2170 reset_curseg(sbi, type, 1);
2171 curseg->alloc_type = LFS;
2172}
2173
2174static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2175 struct curseg_info *seg, block_t start)
2176{
2177 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2178 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2179 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2180 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2181 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2182 int i, pos;
2183
2184 for (i = 0; i < entries; i++)
2185 target_map[i] = ckpt_map[i] | cur_map[i];
2186
2187 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2188
2189 seg->next_blkoff = pos;
351df4b2
JK		 2190}
2191
0a8165d7 2192/*
351df4b2
JK		 2193 * If a segment is written by LFS manner, next block offset is just obtained
2194 * by increasing the current block offset. However, if a segment is written by
2195 * SSR manner, next block offset obtained by calling __next_free_blkoff
2196 */
2197static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2198 struct curseg_info *seg)
2199{
2200 if (seg->alloc_type == SSR)
2201 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2202 else
2203 seg->next_blkoff++;
2204}
2205
0a8165d7 2206/*
e1c42045 2207 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2208 * manner, so it should recover the existing segment information of valid blocks
2209 */
025d63a4 2210static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2211{
2212 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2213 struct curseg_info *curseg = CURSEG_I(sbi, type);
2214 unsigned int new_segno = curseg->next_segno;
2215 struct f2fs_summary_block *sum_node;
2216 struct page *sum_page;
2217
2218 write_sum_page(sbi, curseg->sum_blk,
2219 GET_SUM_BLOCK(sbi, curseg->segno));
2220 __set_test_and_inuse(sbi, new_segno);
2221
2222 mutex_lock(&dirty_i->seglist_lock);
2223 __remove_dirty_segment(sbi, new_segno, PRE);
2224 __remove_dirty_segment(sbi, new_segno, DIRTY);
2225 mutex_unlock(&dirty_i->seglist_lock);
2226
2227 reset_curseg(sbi, type, 1);
2228 curseg->alloc_type = SSR;
2229 __next_free_blkoff(sbi, curseg, 0);
2230
025d63a4
CY		 2231 sum_page = get_sum_page(sbi, new_segno);
2232 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2233 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2234 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2235}
2236
43727527
JK		 2237static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2238{
2239 struct curseg_info *curseg = CURSEG_I(sbi, type);
2240 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2241 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2242 int i, cnt;
2243 bool reversed = false;
c192f7a4
JK		 2244
2245 /* need_SSR() already forces to do this */
e066b83c
JK		 2246 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2247 curseg->next_segno = segno;
c192f7a4 2248 return 1;
e066b83c 2249 }
43727527 2250
70d625cb
JK		 2251 /* For node segments, let's do SSR more intensively */
2252 if (IS_NODESEG(type)) {
d27c3d89
CY		 2253 if (type >= CURSEG_WARM_NODE) {
2254 reversed = true;
2255 i = CURSEG_COLD_NODE;
2256 } else {
2257 i = CURSEG_HOT_NODE;
2258 }
2259 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2260 } else {
d27c3d89
CY		 2261 if (type >= CURSEG_WARM_DATA) {
2262 reversed = true;
2263 i = CURSEG_COLD_DATA;
2264 } else {
2265 i = CURSEG_HOT_DATA;
2266 }
2267 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2268 }
43727527 2269
d27c3d89 2270 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2271 if (i == type)
2272 continue;
e066b83c
JK		 2273 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2274 curseg->next_segno = segno;
43727527 2275 return 1;
e066b83c 2276 }
c192f7a4 2277 }
43727527
JK		 2278 return 0;
2279}
2280
351df4b2
JK		 2281/*
2282 * flush out current segment and replace it with new segment
2283 * This function should be returned with success, otherwise BUG
2284 */
2285static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2286 int type, bool force)
2287{
a7881893
JK		 2288 struct curseg_info *curseg = CURSEG_I(sbi, type);
2289
7b405275 2290 if (force)
351df4b2 2291 new_curseg(sbi, type, true);
5b6c6be2
JK		 2292 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2293 type == CURSEG_WARM_NODE)
351df4b2 2294 new_curseg(sbi, type, false);
a7881893
JK		 2295 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2296 new_curseg(sbi, type, false);
351df4b2 2297 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2298 change_curseg(sbi, type);
351df4b2
JK		 2299 else
2300 new_curseg(sbi, type, false);
dcdfff65 2301
a7881893 2302 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2303}
2304
2305void allocate_new_segments(struct f2fs_sb_info *sbi)
2306{
6ae1be13
JK		 2307 struct curseg_info *curseg;
2308 unsigned int old_segno;
351df4b2
JK		 2309 int i;
2310
3d26fa6b
CY		 2311 down_write(&SIT_I(sbi)->sentry_lock);
2312
6ae1be13
JK		 2313 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2314 curseg = CURSEG_I(sbi, i);
2315 old_segno = curseg->segno;
2316 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2317 locate_dirty_segment(sbi, old_segno);
2318 }
3d26fa6b
CY		 2319
2320 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2321}
2322
2323static const struct segment_allocation default_salloc_ops = {
2324 .allocate_segment = allocate_segment_by_default,
2325};
2326
25290fa5
JK		 2327bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2328{
2329 __u64 trim_start = cpc->trim_start;
2330 bool has_candidate = false;
2331
3d26fa6b 2332 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2333 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2334 if (add_discard_addrs(sbi, cpc, true)) {
2335 has_candidate = true;
2336 break;
2337 }
2338 }
3d26fa6b 2339 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2340
2341 cpc->trim_start = trim_start;
2342 return has_candidate;
2343}
2344
9a997188
JK		 2345static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2346 struct discard_policy *dpolicy,
2347 unsigned int start, unsigned int end)
2348{
2349 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2350 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2351 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2352 struct discard_cmd *dc;
2353 struct blk_plug plug;
2354 int issued;
2355
2356next:
2357 issued = 0;
2358
2359 mutex_lock(&dcc->cmd_lock);
2360 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2361
2362 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2363 NULL, start,
2364 (struct rb_entry **)&prev_dc,
2365 (struct rb_entry **)&next_dc,
2366 &insert_p, &insert_parent, true);
2367 if (!dc)
2368 dc = next_dc;
2369
2370 blk_start_plug(&plug);
2371
2372 while (dc && dc->lstart <= end) {
2373 struct rb_node *node;
2374
2375 if (dc->len < dpolicy->granularity)
2376 goto skip;
2377
2378 if (dc->state != D_PREP) {
2379 list_move_tail(&dc->list, &dcc->fstrim_list);
2380 goto skip;
2381 }
2382
2383 __submit_discard_cmd(sbi, dpolicy, dc);
2384
2385 if (++issued >= dpolicy->max_requests) {
2386 start = dc->lstart + dc->len;
2387
2388 blk_finish_plug(&plug);
2389 mutex_unlock(&dcc->cmd_lock);
2390 __wait_all_discard_cmd(sbi, NULL);
2391 congestion_wait(BLK_RW_ASYNC, HZ/50);
2392 goto next;
2393 }
2394skip:
2395 node = rb_next(&dc->rb_node);
2396 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2397
2398 if (fatal_signal_pending(current))
2399 break;
2400 }
2401
2402 blk_finish_plug(&plug);
2403 mutex_unlock(&dcc->cmd_lock);
2404}
2405
4b2fecc8
JK		 2406int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2407{
f7ef9b83
JK		 2408 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2409 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2410 unsigned int start_segno, end_segno;
8412663d 2411 block_t start_block, end_block;
4b2fecc8 2412 struct cp_control cpc;
78997b56 2413 struct discard_policy dpolicy;
0ea80512 2414 unsigned long long trimmed = 0;
c34f42e2 2415 int err = 0;
4b2fecc8 2416
836b5a63 2417 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2418 return -EINVAL;
2419
7cd8558b 2420 if (end <= MAIN_BLKADDR(sbi))
3d165dc3 2421 return -EINVAL;
4b2fecc8 2422
ed214a11
YH		 2423 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2424 f2fs_msg(sbi->sb, KERN_WARNING,
2425 "Found FS corruption, run fsck to fix.");
3d165dc3 2426 return -EIO;
ed214a11
YH		 2427 }
2428
4b2fecc8 2429 /* start/end segment number in main_area */
7cd8558b
JK		 2430 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2431 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2432 GET_SEGNO(sbi, end);
8412663d 2433
4b2fecc8 2434 cpc.reason = CP_DISCARD;
836b5a63 2435 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2436 cpc.trim_start = start_segno;
2437 cpc.trim_end = end_segno;
4b2fecc8 2438
377224c4
CY		 2439 if (sbi->discard_blks == 0)
2440 goto out;
74fa5f3d 2441
377224c4
CY		 2442 mutex_lock(&sbi->gc_mutex);
2443 err = write_checkpoint(sbi, &cpc);
2444 mutex_unlock(&sbi->gc_mutex);
2445 if (err)
2446 goto out;
8412663d
CY		 2447
2448 start_block = START_BLOCK(sbi, start_segno);
377224c4 2449 end_block = START_BLOCK(sbi, end_segno + 1);
8412663d 2450
8bb4f253 2451 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
78997b56 2452 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
e555da9f
JK		 2453
2454 /*
2455 * We filed discard candidates, but actually we don't need to wait for
2456 * all of them, since they'll be issued in idle time along with runtime
2457 * discard option. User configuration looks like using runtime discard
2458 * or periodic fstrim instead of it.
2459 */
2460 if (!test_opt(sbi, DISCARD)) {
2461 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2462 start_block, end_block);
e555da9f
JK		 2463 range->len = F2FS_BLK_TO_BYTES(trimmed);
2464 }
377224c4 2465out:
c34f42e2 2466 return err;
4b2fecc8
JK		 2467}
2468
351df4b2
JK		 2469static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2470{
2471 struct curseg_info *curseg = CURSEG_I(sbi, type);
2472 if (curseg->next_blkoff < sbi->blocks_per_seg)
2473 return true;
2474 return false;
2475}
2476
4f0a03d3
HL		 2477int rw_hint_to_seg_type(enum rw_hint hint)
2478{
2479 switch (hint) {
2480 case WRITE_LIFE_SHORT:
2481 return CURSEG_HOT_DATA;
2482 case WRITE_LIFE_EXTREME:
2483 return CURSEG_COLD_DATA;
2484 default:
2485 return CURSEG_WARM_DATA;
2486 }
2487}
2488
0cdd3195
HL		 2489/* This returns write hints for each segment type. This hints will be
2490 * passed down to block layer. There are mapping tables which depend on
2491 * the mount option 'whint_mode'.
2492 *
2493 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2494 *
2495 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2496 *
2497 * User F2FS Block
2498 * ---- ---- -----
2499 * META WRITE_LIFE_NOT_SET
2500 * HOT_NODE "
2501 * WARM_NODE "
2502 * COLD_NODE "
2503 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2504 * extension list " "
2505 *
2506 * -- buffered io
2507 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2508 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2509 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2510 * WRITE_LIFE_NONE " "
2511 * WRITE_LIFE_MEDIUM " "
2512 * WRITE_LIFE_LONG " "
2513 *
2514 * -- direct io
2515 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2516 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2517 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2518 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2519 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2520 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2521 *
f2e703f9
HL		 2522 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2523 *
2524 * User F2FS Block
2525 * ---- ---- -----
2526 * META WRITE_LIFE_MEDIUM;
2527 * HOT_NODE WRITE_LIFE_NOT_SET
2528 * WARM_NODE "
2529 * COLD_NODE WRITE_LIFE_NONE
2530 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2531 * extension list " "
2532 *
2533 * -- buffered io
2534 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2535 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2536 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2537 * WRITE_LIFE_NONE " "
2538 * WRITE_LIFE_MEDIUM " "
2539 * WRITE_LIFE_LONG " "
2540 *
2541 * -- direct io
2542 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2543 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2544 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2545 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2546 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2547 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2548 */
2549
2550enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2551 enum page_type type, enum temp_type temp)
2552{
63189b78 2553 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2554 if (type == DATA) {
f2e703f9 2555 if (temp == WARM)
0cdd3195 2556 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2557 else if (temp == HOT)
2558 return WRITE_LIFE_SHORT;
2559 else if (temp == COLD)
2560 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2561 } else {
2562 return WRITE_LIFE_NOT_SET;
2563 }
63189b78 2564 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2565 if (type == DATA) {
2566 if (temp == WARM)
2567 return WRITE_LIFE_LONG;
2568 else if (temp == HOT)
2569 return WRITE_LIFE_SHORT;
2570 else if (temp == COLD)
2571 return WRITE_LIFE_EXTREME;
2572 } else if (type == NODE) {
2573 if (temp == WARM || temp == HOT)
2574 return WRITE_LIFE_NOT_SET;
2575 else if (temp == COLD)
2576 return WRITE_LIFE_NONE;
2577 } else if (type == META) {
2578 return WRITE_LIFE_MEDIUM;
2579 }
0cdd3195 2580 }
f2e703f9 2581 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2582}
2583
81377bd6 2584static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2585{
81377bd6 2586 if (fio->type == DATA)
351df4b2
JK		 2587 return CURSEG_HOT_DATA;
2588 else
2589 return CURSEG_HOT_NODE;
2590}
2591
81377bd6 2592static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2593{
81377bd6
JK		 2594 if (fio->type == DATA) {
2595 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2596
2597 if (S_ISDIR(inode->i_mode))
2598 return CURSEG_HOT_DATA;
2599 else
2600 return CURSEG_COLD_DATA;
2601 } else {
81377bd6 2602 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2603 return CURSEG_WARM_NODE;
351df4b2
JK		 2604 else
2605 return CURSEG_COLD_NODE;
2606 }
2607}
2608
81377bd6 2609static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2610{
81377bd6
JK		 2611 if (fio->type == DATA) {
2612 struct inode *inode = fio->page->mapping->host;
351df4b2 2613
81377bd6 2614 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2615 return CURSEG_COLD_DATA;
b6a06cbb 2616 if (file_is_hot(inode) ||
b4c3ca8b
CY		 2617 is_inode_flag_set(inode, FI_HOT_DATA) ||
2618 is_inode_flag_set(inode, FI_ATOMIC_FILE) ||
2619 is_inode_flag_set(inode, FI_VOLATILE_FILE))
ef095d19 2620 return CURSEG_HOT_DATA;
4f0a03d3 2621 return rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2622 } else {
81377bd6
JK		 2623 if (IS_DNODE(fio->page))
2624 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2625 CURSEG_HOT_NODE;
ef095d19 2626 return CURSEG_COLD_NODE;
351df4b2
JK		 2627 }
2628}
2629
81377bd6 2630static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2631{
a912b54d
JK		 2632 int type = 0;
2633
63189b78 2634 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2635 case 2:
a912b54d
JK		 2636 type = __get_segment_type_2(fio);
2637 break;
351df4b2 2638 case 4:
a912b54d
JK		 2639 type = __get_segment_type_4(fio);
2640 break;
2641 case 6:
2642 type = __get_segment_type_6(fio);
2643 break;
2644 default:
2645 f2fs_bug_on(fio->sbi, true);
351df4b2 2646 }
81377bd6 2647
a912b54d
JK		 2648 if (IS_HOT(type))
2649 fio->temp = HOT;
2650 else if (IS_WARM(type))
2651 fio->temp = WARM;
2652 else
2653 fio->temp = COLD;
2654 return type;
351df4b2
JK		 2655}
2656
bfad7c2d
JK		 2657void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2658 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2659 struct f2fs_summary *sum, int type,
2660 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2661{
2662 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2663 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2664
2b60311d
CY		 2665 down_read(&SM_I(sbi)->curseg_lock);
2666
351df4b2 2667 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2668 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2669
2670 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2671
4e6a8d9b
JK		 2672 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2673
351df4b2
JK		 2674 /*
2675 * __add_sum_entry should be resided under the curseg_mutex
2676 * because, this function updates a summary entry in the
2677 * current summary block.
2678 */
e79efe3b 2679 __add_sum_entry(sbi, type, sum);
351df4b2 2680
351df4b2 2681 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2682
2683 stat_inc_block_count(sbi, curseg);
351df4b2 2684
65f1b80b
YS		 2685 /*
2686 * SIT information should be updated before segment allocation,
2687 * since SSR needs latest valid block information.
2688 */
2689 update_sit_entry(sbi, *new_blkaddr, 1);
2690 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2691 update_sit_entry(sbi, old_blkaddr, -1);
2692
c6f82fe9
JK		 2693 if (!__has_curseg_space(sbi, type))
2694 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 2695
351df4b2 2696 /*
65f1b80b
YS		 2697 * segment dirty status should be updated after segment allocation,
2698 * so we just need to update status only one time after previous
2699 * segment being closed.
351df4b2 2700 */
65f1b80b
YS		 2701 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2702 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 2703
3d26fa6b 2704 up_write(&sit_i->sentry_lock);
351df4b2 2705
704956ec 2706 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2707 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2708
704956ec
CY		 2709 f2fs_inode_chksum_set(sbi, page);
2710 }
2711
fb830fc5
CY		 2712 if (add_list) {
2713 struct f2fs_bio_info *io;
2714
2715 INIT_LIST_HEAD(&fio->list);
2716 fio->in_list = true;
2717 io = sbi->write_io[fio->type] + fio->temp;
2718 spin_lock(&io->io_lock);
2719 list_add_tail(&fio->list, &io->io_list);
2720 spin_unlock(&io->io_lock);
2721 }
2722
bfad7c2d 2723 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 2724
2725 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 2726}
2727
39d787be
CY		 2728static void update_device_state(struct f2fs_io_info *fio)
2729{
2730 struct f2fs_sb_info *sbi = fio->sbi;
2731 unsigned int devidx;
2732
2733 if (!sbi->s_ndevs)
2734 return;
2735
2736 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2737
2738 /* update device state for fsync */
2739 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2740
2741 /* update device state for checkpoint */
2742 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2743 spin_lock(&sbi->dev_lock);
2744 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2745 spin_unlock(&sbi->dev_lock);
2746 }
39d787be
CY		 2747}
2748
05ca3632 2749static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2750{
81377bd6 2751 int type = __get_segment_type(fio);
0a595eba 2752 int err;
bfad7c2d 2753
0a595eba 2754reallocate:
7a9d7548 2755 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2756 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2757
351df4b2 2758 /* writeout dirty page into bdev */
b9109b0e 2759 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2760 if (err == -EAGAIN) {
2761 fio->old_blkaddr = fio->new_blkaddr;
2762 goto reallocate;
39d787be
CY		 2763 } else if (!err) {
2764 update_device_state(fio);
0a595eba 2765 }
351df4b2
JK		 2766}
2767
b0af6d49
CY		 2768void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2769 enum iostat_type io_type)
351df4b2 2770{
458e6197 2771 struct f2fs_io_info fio = {
05ca3632 2772 .sbi = sbi,
458e6197 2773 .type = META,
0cdd3195 2774 .temp = HOT,
04d328de 2775 .op = REQ_OP_WRITE,
70fd7614 2776 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2777 .old_blkaddr = page->index,
2778 .new_blkaddr = page->index,
05ca3632 2779 .page = page,
4375a336 2780 .encrypted_page = NULL,
fb830fc5 2781 .in_list = false,
458e6197
JK		 2782 };
2783
2b947003 2784 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2785 fio.op_flags &= ~REQ_META;
2b947003 2786
351df4b2 2787 set_page_writeback(page);
17c50035 2788 ClearPageError(page);
b9109b0e 2789 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2790
2791 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2792}
2793
05ca3632 2794void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2795{
2796 struct f2fs_summary sum;
05ca3632 2797
351df4b2 2798 set_summary(&sum, nid, 0, 0);
05ca3632 2799 do_write_page(&sum, fio);
b0af6d49
CY		 2800
2801 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2802}
2803
05ca3632 2804void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2805{
05ca3632 2806 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2807 struct f2fs_summary sum;
2808 struct node_info ni;
2809
9850cf4a 2810 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2811 get_node_info(sbi, dn->nid, &ni);
2812 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2813 do_write_page(&sum, fio);
f28b3434 2814 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2815
2816 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2817}
2818
d1b3e72d 2819int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2820{
b0af6d49 2821 int err;
d21b0f23 2822 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 2823
7a9d7548 2824 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 2825 /* i/o temperature is needed for passing down write hints */
2826 __get_segment_type(fio);
d21b0f23
YH		 2827
2828 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2829 GET_SEGNO(sbi, fio->new_blkaddr))->type));
2830
05ca3632 2831 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2832
2833 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2834 if (!err)
2835 update_device_state(fio);
b0af6d49
CY		 2836
2837 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2838
2839 return err;
351df4b2
JK		 2840}
2841
2b60311d
CY		 2842static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2843 unsigned int segno)
2844{
2845 int i;
2846
2847 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2848 if (CURSEG_I(sbi, i)->segno == segno)
2849 break;
2850 }
2851 return i;
2852}
2853
4356e48e 2854void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2855 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2856 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2857{
2858 struct sit_info *sit_i = SIT_I(sbi);
2859 struct curseg_info *curseg;
2860 unsigned int segno, old_cursegno;
2861 struct seg_entry *se;
2862 int type;
19f106bc 2863 unsigned short old_blkoff;
351df4b2
JK		 2864
2865 segno = GET_SEGNO(sbi, new_blkaddr);
2866 se = get_seg_entry(sbi, segno);
2867 type = se->type;
2868
2b60311d
CY		 2869 down_write(&SM_I(sbi)->curseg_lock);
2870
19f106bc
CY		 2871 if (!recover_curseg) {
2872 /* for recovery flow */
2873 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2874 if (old_blkaddr == NULL_ADDR)
2875 type = CURSEG_COLD_DATA;
2876 else
2877 type = CURSEG_WARM_DATA;
2878 }
2879 } else {
2b60311d
CY		 2880 if (IS_CURSEG(sbi, segno)) {
2881 /* se->type is volatile as SSR allocation */
2882 type = __f2fs_get_curseg(sbi, segno);
2883 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2884 } else {
351df4b2 2885 type = CURSEG_WARM_DATA;
2b60311d 2886 }
351df4b2 2887 }
19f106bc 2888
2c190504 2889 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 2890 curseg = CURSEG_I(sbi, type);
2891
2892 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2893 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2894
2895 old_cursegno = curseg->segno;
19f106bc 2896 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2897
2898 /* change the current segment */
2899 if (segno != curseg->segno) {
2900 curseg->next_segno = segno;
025d63a4 2901 change_curseg(sbi, type);
351df4b2
JK		 2902 }
2903
491c0854 2904 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2905 __add_sum_entry(sbi, type, sum);
351df4b2 2906
28bc106b 2907 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2908 update_sit_entry(sbi, new_blkaddr, 1);
2909 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2910 update_sit_entry(sbi, old_blkaddr, -1);
2911
2912 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2913 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2914
351df4b2 2915 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2916
19f106bc
CY		 2917 if (recover_curseg) {
2918 if (old_cursegno != curseg->segno) {
2919 curseg->next_segno = old_cursegno;
025d63a4 2920 change_curseg(sbi, type);
19f106bc
CY		 2921 }
2922 curseg->next_blkoff = old_blkoff;
2923 }
2924
3d26fa6b 2925 up_write(&sit_i->sentry_lock);
351df4b2 2926 mutex_unlock(&curseg->curseg_mutex);
2b60311d 2927 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 2928}
2929
528e3459
CY		 2930void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2931 block_t old_addr, block_t new_addr,
28bc106b
CY		 2932 unsigned char version, bool recover_curseg,
2933 bool recover_newaddr)
528e3459
CY		 2934{
2935 struct f2fs_summary sum;
2936
2937 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2938
28bc106b
CY		 2939 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2940 recover_curseg, recover_newaddr);
528e3459 2941
f28b3434 2942 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2943}
2944
93dfe2ac 2945void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2946 enum page_type type, bool ordered)
93dfe2ac 2947{
93dfe2ac 2948 if (PageWriteback(page)) {
4081363f
JK		 2949 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2950
b9109b0e
JK		 2951 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2952 0, page->index, type);
fec1d657
JK		 2953 if (ordered)
2954 wait_on_page_writeback(page);
2955 else
2956 wait_for_stable_page(page);
93dfe2ac
JK		 2957 }
2958}
2959
d4c759ee 2960void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2961{
2962 struct page *cpage;
2963
7b525dd0 2964 if (!is_valid_blkaddr(blkaddr))
08b39fbd
CY		 2965 return;
2966
08b39fbd
CY		 2967 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2968 if (cpage) {
fec1d657 2969 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2970 f2fs_put_page(cpage, 1);
2971 }
2972}
2973
c376fc0f 2974static void read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 2975{
2976 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2977 struct curseg_info *seg_i;
2978 unsigned char *kaddr;
2979 struct page *page;
2980 block_t start;
2981 int i, j, offset;
2982
2983 start = start_sum_block(sbi);
2984
2985 page = get_meta_page(sbi, start++);
2986 kaddr = (unsigned char *)page_address(page);
2987
2988 /* Step 1: restore nat cache */
2989 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2990 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2991
2992 /* Step 2: restore sit cache */
2993 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2994 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2995 offset = 2 * SUM_JOURNAL_SIZE;
2996
2997 /* Step 3: restore summary entries */
2998 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2999 unsigned short blk_off;
3000 unsigned int segno;
3001
3002 seg_i = CURSEG_I(sbi, i);
3003 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3004 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3005 seg_i->next_segno = segno;
3006 reset_curseg(sbi, i, 0);
3007 seg_i->alloc_type = ckpt->alloc_type[i];
3008 seg_i->next_blkoff = blk_off;
3009
3010 if (seg_i->alloc_type == SSR)
3011 blk_off = sbi->blocks_per_seg;
3012
3013 for (j = 0; j < blk_off; j++) {
3014 struct f2fs_summary *s;
3015 s = (struct f2fs_summary *)(kaddr + offset);
3016 seg_i->sum_blk->entries[j] = *s;
3017 offset += SUMMARY_SIZE;
09cbfeaf 3018 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3019 SUM_FOOTER_SIZE)
3020 continue;
3021
3022 f2fs_put_page(page, 1);
3023 page = NULL;
3024
3025 page = get_meta_page(sbi, start++);
3026 kaddr = (unsigned char *)page_address(page);
3027 offset = 0;
3028 }
3029 }
3030 f2fs_put_page(page, 1);
351df4b2
JK		 3031}
3032
3033static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3034{
3035 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3036 struct f2fs_summary_block *sum;
3037 struct curseg_info *curseg;
3038 struct page *new;
3039 unsigned short blk_off;
3040 unsigned int segno = 0;
3041 block_t blk_addr = 0;
3042
3043 /* get segment number and block addr */
3044 if (IS_DATASEG(type)) {
3045 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3046 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3047 CURSEG_HOT_DATA]);
119ee914 3048 if (__exist_node_summaries(sbi))
351df4b2
JK		 3049 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3050 else
3051 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3052 } else {
3053 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3054 CURSEG_HOT_NODE]);
3055 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3056 CURSEG_HOT_NODE]);
119ee914 3057 if (__exist_node_summaries(sbi))
351df4b2
JK		 3058 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3059 type - CURSEG_HOT_NODE);
3060 else
3061 blk_addr = GET_SUM_BLOCK(sbi, segno);
3062 }
3063
3064 new = get_meta_page(sbi, blk_addr);
3065 sum = (struct f2fs_summary_block *)page_address(new);
3066
3067 if (IS_NODESEG(type)) {
119ee914 3068 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3069 struct f2fs_summary *ns = &sum->entries[0];
3070 int i;
3071 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3072 ns->version = 0;
3073 ns->ofs_in_node = 0;
3074 }
3075 } else {
c376fc0f 3076 restore_node_summary(sbi, segno, sum);
351df4b2
JK		 3077 }
3078 }
3079
3080 /* set uncompleted segment to curseg */
3081 curseg = CURSEG_I(sbi, type);
3082 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3083
3084 /* update journal info */
3085 down_write(&curseg->journal_rwsem);
3086 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3087 up_write(&curseg->journal_rwsem);
3088
3089 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3090 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3091 curseg->next_segno = segno;
3092 reset_curseg(sbi, type, 0);
3093 curseg->alloc_type = ckpt->alloc_type[type];
3094 curseg->next_blkoff = blk_off;
3095 mutex_unlock(&curseg->curseg_mutex);
3096 f2fs_put_page(new, 1);
3097 return 0;
3098}
3099
3100static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3101{
21d3f8e1
JQ		 3102 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3103 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3104 int type = CURSEG_HOT_DATA;
e4fc5fbf 3105 int err;
351df4b2 3106
aaec2b1d 3107 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 3108 int npages = npages_for_summary_flush(sbi, true);
3109
3110 if (npages >= 2)
3111 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3112 META_CP, true);
3fa06d7b 3113
351df4b2 3114 /* restore for compacted data summary */
c376fc0f 3115 read_compacted_summaries(sbi);
351df4b2
JK		 3116 type = CURSEG_HOT_NODE;
3117 }
3118
119ee914 3119 if (__exist_node_summaries(sbi))
3fa06d7b 3120 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3121 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3122
e4fc5fbf
CY		 3123 for (; type <= CURSEG_COLD_NODE; type++) {
3124 err = read_normal_summaries(sbi, type);
3125 if (err)
3126 return err;
3127 }
3128
21d3f8e1
JQ		 3129 /* sanity check for summary blocks */
3130 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3131 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3132 return -EINVAL;
3133
351df4b2
JK		 3134 return 0;
3135}
3136
3137static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3138{
3139 struct page *page;
3140 unsigned char *kaddr;
3141 struct f2fs_summary *summary;
3142 struct curseg_info *seg_i;
3143 int written_size = 0;
3144 int i, j;
3145
3146 page = grab_meta_page(sbi, blkaddr++);
3147 kaddr = (unsigned char *)page_address(page);
81114baa 3148 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3149
3150 /* Step 1: write nat cache */
3151 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3152 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3153 written_size += SUM_JOURNAL_SIZE;
3154
3155 /* Step 2: write sit cache */
3156 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3157 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3158 written_size += SUM_JOURNAL_SIZE;
3159
351df4b2
JK		 3160 /* Step 3: write summary entries */
3161 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3162 unsigned short blkoff;
3163 seg_i = CURSEG_I(sbi, i);
3164 if (sbi->ckpt->alloc_type[i] == SSR)
3165 blkoff = sbi->blocks_per_seg;
3166 else
3167 blkoff = curseg_blkoff(sbi, i);
3168
3169 for (j = 0; j < blkoff; j++) {
3170 if (!page) {
3171 page = grab_meta_page(sbi, blkaddr++);
3172 kaddr = (unsigned char *)page_address(page);
81114baa 3173 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3174 written_size = 0;
3175 }
3176 summary = (struct f2fs_summary *)(kaddr + written_size);
3177 *summary = seg_i->sum_blk->entries[j];
3178 written_size += SUMMARY_SIZE;
351df4b2 3179
09cbfeaf 3180 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3181 SUM_FOOTER_SIZE)
3182 continue;
3183
e8d61a74 3184 set_page_dirty(page);
351df4b2
JK		 3185 f2fs_put_page(page, 1);
3186 page = NULL;
3187 }
3188 }
e8d61a74
CY		 3189 if (page) {
3190 set_page_dirty(page);
351df4b2 3191 f2fs_put_page(page, 1);
e8d61a74 3192 }
351df4b2
JK		 3193}
3194
3195static void write_normal_summaries(struct f2fs_sb_info *sbi,
3196 block_t blkaddr, int type)
3197{
3198 int i, end;
3199 if (IS_DATASEG(type))
3200 end = type + NR_CURSEG_DATA_TYPE;
3201 else
3202 end = type + NR_CURSEG_NODE_TYPE;
3203
b7ad7512
CY		 3204 for (i = type; i < end; i++)
3205 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3206}
3207
3208void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3209{
aaec2b1d 3210 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3211 write_compacted_summaries(sbi, start_blk);
3212 else
3213 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3214}
3215
3216void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3217{
119ee914 3218 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3219}
3220
dfc08a12 3221int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3222 unsigned int val, int alloc)
3223{
3224 int i;
3225
3226 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3227 for (i = 0; i < nats_in_cursum(journal); i++) {
3228 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3229 return i;
3230 }
dfc08a12
CY		 3231 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3232 return update_nats_in_cursum(journal, 1);
351df4b2 3233 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3234 for (i = 0; i < sits_in_cursum(journal); i++)
3235 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3236 return i;
dfc08a12
CY		 3237 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3238 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3239 }
3240 return -1;
3241}
3242
3243static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3244 unsigned int segno)
3245{
2cc22186 3246 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3247}
3248
3249static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3250 unsigned int start)
3251{
3252 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3253 struct page *page;
351df4b2 3254 pgoff_t src_off, dst_off;
351df4b2
JK		 3255
3256 src_off = current_sit_addr(sbi, start);
3257 dst_off = next_sit_addr(sbi, src_off);
3258
068c3cd8
YH		 3259 page = grab_meta_page(sbi, dst_off);
3260 seg_info_to_sit_page(sbi, page, start);
351df4b2 3261
068c3cd8 3262 set_page_dirty(page);
351df4b2
JK		 3263 set_to_next_sit(sit_i, start);
3264
068c3cd8 3265 return page;
351df4b2
JK		 3266}
3267
184a5cd2
CY		 3268static struct sit_entry_set *grab_sit_entry_set(void)
3269{
3270 struct sit_entry_set *ses =
80c54505 3271 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3272
3273 ses->entry_cnt = 0;
3274 INIT_LIST_HEAD(&ses->set_list);
3275 return ses;
3276}
3277
3278static void release_sit_entry_set(struct sit_entry_set *ses)
3279{
3280 list_del(&ses->set_list);
3281 kmem_cache_free(sit_entry_set_slab, ses);
3282}
3283
3284static void adjust_sit_entry_set(struct sit_entry_set *ses,
3285 struct list_head *head)
3286{
3287 struct sit_entry_set *next = ses;
3288
3289 if (list_is_last(&ses->set_list, head))
3290 return;
3291
3292 list_for_each_entry_continue(next, head, set_list)
3293 if (ses->entry_cnt <= next->entry_cnt)
3294 break;
3295
3296 list_move_tail(&ses->set_list, &next->set_list);
3297}
3298
3299static void add_sit_entry(unsigned int segno, struct list_head *head)
3300{
3301 struct sit_entry_set *ses;
3302 unsigned int start_segno = START_SEGNO(segno);
3303
3304 list_for_each_entry(ses, head, set_list) {
3305 if (ses->start_segno == start_segno) {
3306 ses->entry_cnt++;
3307 adjust_sit_entry_set(ses, head);
3308 return;
3309 }
3310 }
3311
3312 ses = grab_sit_entry_set();
3313
3314 ses->start_segno = start_segno;
3315 ses->entry_cnt++;
3316 list_add(&ses->set_list, head);
3317}
3318
3319static void add_sits_in_set(struct f2fs_sb_info *sbi)
3320{
3321 struct f2fs_sm_info *sm_info = SM_I(sbi);
3322 struct list_head *set_list = &sm_info->sit_entry_set;
3323 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3324 unsigned int segno;
3325
7cd8558b 3326 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3327 add_sit_entry(segno, set_list);
3328}
3329
3330static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3331{
3332 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3333 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3334 int i;
3335
b7ad7512 3336 down_write(&curseg->journal_rwsem);
dfc08a12 3337 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3338 unsigned int segno;
3339 bool dirtied;
3340
dfc08a12 3341 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3342 dirtied = __mark_sit_entry_dirty(sbi, segno);
3343
3344 if (!dirtied)
3345 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3346 }
dfc08a12 3347 update_sits_in_cursum(journal, -i);
b7ad7512 3348 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3349}
3350
0a8165d7 3351/*
351df4b2
JK		 3352 * CP calls this function, which flushes SIT entries including sit_journal,
3353 * and moves prefree segs to free segs.
3354 */
4b2fecc8 3355void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3356{
3357 struct sit_info *sit_i = SIT_I(sbi);
3358 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3359 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3360 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3361 struct sit_entry_set *ses, *tmp;
3362 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3363 bool to_journal = true;
4b2fecc8 3364 struct seg_entry *se;
351df4b2 3365
3d26fa6b 3366 down_write(&sit_i->sentry_lock);
351df4b2 3367
2b11a74b
WL		 3368 if (!sit_i->dirty_sentries)
3369 goto out;
3370
351df4b2 3371 /*
184a5cd2
CY		 3372 * add and account sit entries of dirty bitmap in sit entry
3373 * set temporarily
351df4b2 3374 */
184a5cd2 3375 add_sits_in_set(sbi);
351df4b2 3376
184a5cd2
CY		 3377 /*
3378 * if there are no enough space in journal to store dirty sit
3379 * entries, remove all entries from journal and add and account
3380 * them in sit entry set.
3381 */
dfc08a12 3382 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3383 remove_sits_in_journal(sbi);
b2955550 3384
184a5cd2
CY		 3385 /*
3386 * there are two steps to flush sit entries:
3387 * #1, flush sit entries to journal in current cold data summary block.
3388 * #2, flush sit entries to sit page.
3389 */
3390 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3391 struct page *page = NULL;
184a5cd2
CY		 3392 struct f2fs_sit_block *raw_sit = NULL;
3393 unsigned int start_segno = ses->start_segno;
3394 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3395 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3396 unsigned int segno = start_segno;
3397
3398 if (to_journal &&
dfc08a12 3399 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3400 to_journal = false;
3401
b7ad7512
CY		 3402 if (to_journal) {
3403 down_write(&curseg->journal_rwsem);
3404 } else {
184a5cd2
CY		 3405 page = get_next_sit_page(sbi, start_segno);
3406 raw_sit = page_address(page);
351df4b2 3407 }
351df4b2 3408
184a5cd2
CY		 3409 /* flush dirty sit entries in region of current sit set */
3410 for_each_set_bit_from(segno, bitmap, end) {
3411 int offset, sit_offset;
4b2fecc8
JK		 3412
3413 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3414#ifdef CONFIG_F2FS_CHECK_FS
3415 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3416 SIT_VBLOCK_MAP_SIZE))
3417 f2fs_bug_on(sbi, 1);
3418#endif
184a5cd2
CY		 3419
3420 /* add discard candidates */
c473f1a9 3421 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3422 cpc->trim_start = segno;
25290fa5 3423 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3424 }
184a5cd2
CY		 3425
3426 if (to_journal) {
dfc08a12 3427 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3428 SIT_JOURNAL, segno, 1);
3429 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3430 segno_in_journal(journal, offset) =
184a5cd2
CY		 3431 cpu_to_le32(segno);
3432 seg_info_to_raw_sit(se,
dfc08a12 3433 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3434 check_block_count(sbi, segno,
3435 &sit_in_journal(journal, offset));
184a5cd2
CY		 3436 } else {
3437 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3438 seg_info_to_raw_sit(se,
3439 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3440 check_block_count(sbi, segno,
3441 &raw_sit->entries[sit_offset]);
184a5cd2 3442 }
351df4b2 3443
184a5cd2
CY		 3444 __clear_bit(segno, bitmap);
3445 sit_i->dirty_sentries--;
3446 ses->entry_cnt--;
351df4b2
JK		 3447 }
3448
b7ad7512
CY		 3449 if (to_journal)
3450 up_write(&curseg->journal_rwsem);
3451 else
184a5cd2
CY		 3452 f2fs_put_page(page, 1);
3453
3454 f2fs_bug_on(sbi, ses->entry_cnt);
3455 release_sit_entry_set(ses);
351df4b2 3456 }
184a5cd2
CY		 3457
3458 f2fs_bug_on(sbi, !list_empty(head));
3459 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3460out:
c473f1a9 3461 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3462 __u64 trim_start = cpc->trim_start;
3463
4b2fecc8 3464 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3465 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3466
3467 cpc->trim_start = trim_start;
4b2fecc8 3468 }
3d26fa6b 3469 up_write(&sit_i->sentry_lock);
351df4b2 3470
351df4b2
JK		 3471 set_prefree_as_free_segments(sbi);
3472}
3473
3474static int build_sit_info(struct f2fs_sb_info *sbi)
3475{
3476 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3477 struct sit_info *sit_i;
3478 unsigned int sit_segs, start;
ae27d62e 3479 char *src_bitmap;
351df4b2
JK		 3480 unsigned int bitmap_size;
3481
3482 /* allocate memory for SIT information */
acbf054d 3483 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3484 if (!sit_i)
3485 return -ENOMEM;
3486
3487 SM_I(sbi)->sit_info = sit_i;
3488
628b3d14 3489 sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
39307a8e 3490 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3491 if (!sit_i->sentries)
3492 return -ENOMEM;
3493
7cd8558b 3494 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3495 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3496 GFP_KERNEL);
351df4b2
JK		 3497 if (!sit_i->dirty_sentries_bitmap)
3498 return -ENOMEM;
3499
7cd8558b 3500 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3501 sit_i->sentries[start].cur_valid_map
acbf054d 3502 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3503 sit_i->sentries[start].ckpt_valid_map
acbf054d 3504 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3505 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3506 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3507 return -ENOMEM;
3e025740 3508
355e7891
CY		 3509#ifdef CONFIG_F2FS_CHECK_FS
3510 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3511 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3512 if (!sit_i->sentries[start].cur_valid_map_mir)
3513 return -ENOMEM;
3514#endif
3515
3e025740
JK		 3516 if (f2fs_discard_en(sbi)) {
3517 sit_i->sentries[start].discard_map
acbf054d
CY		 3518 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3519 GFP_KERNEL);
3e025740
JK		 3520 if (!sit_i->sentries[start].discard_map)
3521 return -ENOMEM;
3522 }
351df4b2
JK		 3523 }
3524
acbf054d 3525 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3526 if (!sit_i->tmp_map)
3527 return -ENOMEM;
3528
351df4b2 3529 if (sbi->segs_per_sec > 1) {
628b3d14 3530 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
39307a8e 3531 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3532 if (!sit_i->sec_entries)
3533 return -ENOMEM;
3534 }
3535
3536 /* get information related with SIT */
3537 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3538
3539 /* setup SIT bitmap from ckeckpoint pack */
3540 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3541 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3542
ae27d62e
CY		 3543 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3544 if (!sit_i->sit_bitmap)
351df4b2 3545 return -ENOMEM;
351df4b2 3546
ae27d62e
CY		 3547#ifdef CONFIG_F2FS_CHECK_FS
3548 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3549 if (!sit_i->sit_bitmap_mir)
3550 return -ENOMEM;
3551#endif
3552
351df4b2
JK		 3553 /* init SIT information */
3554 sit_i->s_ops = &default_salloc_ops;
3555
3556 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3557 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3558 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3559 sit_i->bitmap_size = bitmap_size;
3560 sit_i->dirty_sentries = 0;
3561 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3562 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3563 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3564 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3565 return 0;
3566}
3567
3568static int build_free_segmap(struct f2fs_sb_info *sbi)
3569{
351df4b2
JK		 3570 struct free_segmap_info *free_i;
3571 unsigned int bitmap_size, sec_bitmap_size;
3572
3573 /* allocate memory for free segmap information */
acbf054d 3574 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3575 if (!free_i)
3576 return -ENOMEM;
3577
3578 SM_I(sbi)->free_info = free_i;
3579
7cd8558b 3580 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3581 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3582 if (!free_i->free_segmap)
3583 return -ENOMEM;
3584
7cd8558b 3585 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3586 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3587 if (!free_i->free_secmap)
3588 return -ENOMEM;
3589
3590 /* set all segments as dirty temporarily */
3591 memset(free_i->free_segmap, 0xff, bitmap_size);
3592 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3593
3594 /* init free segmap information */
7cd8558b 3595 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3596 free_i->free_segments = 0;
3597 free_i->free_sections = 0;
1a118ccf 3598 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3599 return 0;
3600}
3601
3602static int build_curseg(struct f2fs_sb_info *sbi)
3603{
1042d60f 3604 struct curseg_info *array;
351df4b2
JK		 3605 int i;
3606
4e6aad29 3607 array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
351df4b2
JK		 3608 if (!array)
3609 return -ENOMEM;
3610
3611 SM_I(sbi)->curseg_array = array;
3612
3613 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3614 mutex_init(&array[i].curseg_mutex);
acbf054d 3615 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3616 if (!array[i].sum_blk)
3617 return -ENOMEM;
b7ad7512 3618 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 3619 array[i].journal = f2fs_kzalloc(sbi,
3620 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 3621 if (!array[i].journal)
3622 return -ENOMEM;
351df4b2
JK		 3623 array[i].segno = NULL_SEGNO;
3624 array[i].next_blkoff = 0;
3625 }
3626 return restore_curseg_summaries(sbi);
3627}
3628
c39a1b34 3629static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3630{
3631 struct sit_info *sit_i = SIT_I(sbi);
3632 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3633 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3634 struct seg_entry *se;
3635 struct f2fs_sit_entry sit;
74de593a
CY		 3636 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3637 unsigned int i, start, end;
3638 unsigned int readed, start_blk = 0;
c39a1b34 3639 int err = 0;
8a29c126 3640 block_t total_node_blocks = 0;
351df4b2 3641
74de593a 3642 do {
664ba972
JK		 3643 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3644 META_SIT, true);
74de593a
CY		 3645
3646 start = start_blk * sit_i->sents_per_block;
3647 end = (start_blk + readed) * sit_i->sents_per_block;
3648
7cd8558b 3649 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3650 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3651 struct page *page;
3652
9c094040 3653 se = &sit_i->sentries[start];
74de593a
CY		 3654 page = get_current_sit_page(sbi, start);
3655 sit_blk = (struct f2fs_sit_block *)page_address(page);
3656 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3657 f2fs_put_page(page, 1);
d600af23 3658
c39a1b34
JK		 3659 err = check_block_count(sbi, start, &sit);
3660 if (err)
3661 return err;
74de593a 3662 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3663 if (IS_NODESEG(se->type))
3664 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 3665
3666 /* build discard map only one time */
3e025740 3667 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3668 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3669 memset(se->discard_map, 0xff,
3670 SIT_VBLOCK_MAP_SIZE);
3671 } else {
3672 memcpy(se->discard_map,
3673 se->cur_valid_map,
3674 SIT_VBLOCK_MAP_SIZE);
3675 sbi->discard_blks +=
3676 sbi->blocks_per_seg -
3677 se->valid_blocks;
3678 }
3e025740 3679 }
a66cdd98 3680
d600af23
CY		 3681 if (sbi->segs_per_sec > 1)
3682 get_sec_entry(sbi, start)->valid_blocks +=
3683 se->valid_blocks;
351df4b2 3684 }
74de593a
CY		 3685 start_blk += readed;
3686 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3687
3688 down_read(&curseg->journal_rwsem);
3689 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3690 unsigned int old_valid_blocks;
3691
3692 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 3693 if (start >= MAIN_SEGS(sbi)) {
3694 f2fs_msg(sbi->sb, KERN_ERR,
3695 "Wrong journal entry on segno %u",
3696 start);
3697 set_sbi_flag(sbi, SBI_NEED_FSCK);
3698 err = -EINVAL;
3699 break;
3700 }
3701
d600af23
CY		 3702 se = &sit_i->sentries[start];
3703 sit = sit_in_journal(journal, i);
3704
3705 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 3706 if (IS_NODESEG(se->type))
3707 total_node_blocks -= old_valid_blocks;
d600af23 3708
c39a1b34
JK		 3709 err = check_block_count(sbi, start, &sit);
3710 if (err)
3711 break;
d600af23 3712 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3713 if (IS_NODESEG(se->type))
3714 total_node_blocks += se->valid_blocks;
d600af23
CY		 3715
3716 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3717 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3718 memset(se->discard_map, 0xff,
3719 SIT_VBLOCK_MAP_SIZE);
3720 } else {
3721 memcpy(se->discard_map, se->cur_valid_map,
3722 SIT_VBLOCK_MAP_SIZE);
a9af3fdc
CY		 3723 sbi->discard_blks += old_valid_blocks;
3724 sbi->discard_blks -= se->valid_blocks;
1f43e2ad 3725 }
d600af23
CY		 3726 }
3727
a9af3fdc 3728 if (sbi->segs_per_sec > 1) {
d600af23 3729 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 3730 se->valid_blocks;
3731 get_sec_entry(sbi, start)->valid_blocks -=
3732 old_valid_blocks;
3733 }
d600af23
CY		 3734 }
3735 up_read(&curseg->journal_rwsem);
8a29c126
JK		 3736
3737 if (!err && total_node_blocks != valid_node_count(sbi)) {
3738 f2fs_msg(sbi->sb, KERN_ERR,
3739 "SIT is corrupted node# %u vs %u",
3740 total_node_blocks, valid_node_count(sbi));
3741 set_sbi_flag(sbi, SBI_NEED_FSCK);
3742 err = -EINVAL;
3743 }
3744
c39a1b34 3745 return err;
351df4b2
JK		 3746}
3747
3748static void init_free_segmap(struct f2fs_sb_info *sbi)
3749{
3750 unsigned int start;
3751 int type;
3752
7cd8558b 3753 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3754 struct seg_entry *sentry = get_seg_entry(sbi, start);
3755 if (!sentry->valid_blocks)
3756 __set_free(sbi, start);
c79b7ff1
JK		 3757 else
3758 SIT_I(sbi)->written_valid_blocks +=
3759 sentry->valid_blocks;
351df4b2
JK		 3760 }
3761
3762 /* set use the current segments */
3763 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3764 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3765 __set_test_and_inuse(sbi, curseg_t->segno);
3766 }
3767}
3768
3769static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3770{
3771 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3772 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3773 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3774 unsigned short valid_blocks;
3775
8736fbf0 3776 while (1) {
351df4b2 3777 /* find dirty segment based on free segmap */
7cd8558b
JK		 3778 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3779 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3780 break;
3781 offset = segno + 1;
302bd348 3782 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3783 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3784 continue;
ec325b52
JK		 3785 if (valid_blocks > sbi->blocks_per_seg) {
3786 f2fs_bug_on(sbi, 1);
3787 continue;
3788 }
351df4b2
JK		 3789 mutex_lock(&dirty_i->seglist_lock);
3790 __locate_dirty_segment(sbi, segno, DIRTY);
3791 mutex_unlock(&dirty_i->seglist_lock);
3792 }
3793}
3794
5ec4e49f 3795static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3796{
3797 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3798 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3799
628b3d14 3800 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 3801 if (!dirty_i->victim_secmap)
351df4b2
JK		 3802 return -ENOMEM;
3803 return 0;
3804}
3805
3806static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3807{
3808 struct dirty_seglist_info *dirty_i;
3809 unsigned int bitmap_size, i;
3810
3811 /* allocate memory for dirty segments list information */
acbf054d
CY		 3812 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3813 GFP_KERNEL);
351df4b2
JK		 3814 if (!dirty_i)
3815 return -ENOMEM;
3816
3817 SM_I(sbi)->dirty_info = dirty_i;
3818 mutex_init(&dirty_i->seglist_lock);
3819
7cd8558b 3820 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3821
3822 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 3823 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3824 GFP_KERNEL);
351df4b2
JK		 3825 if (!dirty_i->dirty_segmap[i])
3826 return -ENOMEM;
3827 }
3828
3829 init_dirty_segmap(sbi);
5ec4e49f 3830 return init_victim_secmap(sbi);
351df4b2
JK		 3831}
3832
0a8165d7 3833/*
351df4b2
JK		 3834 * Update min, max modified time for cost-benefit GC algorithm
3835 */
3836static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3837{
3838 struct sit_info *sit_i = SIT_I(sbi);
3839 unsigned int segno;
3840
3d26fa6b 3841 down_write(&sit_i->sentry_lock);
351df4b2 3842
5ad25442 3843 sit_i->min_mtime = ULLONG_MAX;
351df4b2 3844
7cd8558b 3845 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3846 unsigned int i;
3847 unsigned long long mtime = 0;
3848
3849 for (i = 0; i < sbi->segs_per_sec; i++)
3850 mtime += get_seg_entry(sbi, segno + i)->mtime;
3851
3852 mtime = div_u64(mtime, sbi->segs_per_sec);
3853
3854 if (sit_i->min_mtime > mtime)
3855 sit_i->min_mtime = mtime;
3856 }
3857 sit_i->max_mtime = get_mtime(sbi);
3d26fa6b 3858 up_write(&sit_i->sentry_lock);
351df4b2
JK		 3859}
3860
3861int build_segment_manager(struct f2fs_sb_info *sbi)
3862{
3863 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3864 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3865 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3866 int err;
3867
acbf054d 3868 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 3869 if (!sm_info)
3870 return -ENOMEM;
3871
3872 /* init sm info */
3873 sbi->sm_info = sm_info;
351df4b2
JK		 3874 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3875 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3876 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3877 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3878 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3879 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3880 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3881 sm_info->rec_prefree_segments = sm_info->main_segments *
3882 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3883 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3884 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3885
52763a4b
JK		 3886 if (!test_opt(sbi, LFS))
3887 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3888 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3889 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3890 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 3891 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 3892
184a5cd2
CY		 3893 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3894
2b60311d
CY		 3895 init_rwsem(&sm_info->curseg_lock);
3896
d4fdf8ba 3897 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3898 err = create_flush_cmd_control(sbi);
3899 if (err)
a688b9d9 3900 return err;
6b4afdd7
JK		 3901 }
3902
0b54fb84
JK		 3903 err = create_discard_cmd_control(sbi);
3904 if (err)
3905 return err;
3906
351df4b2
JK		 3907 err = build_sit_info(sbi);
3908 if (err)
3909 return err;
3910 err = build_free_segmap(sbi);
3911 if (err)
3912 return err;
3913 err = build_curseg(sbi);
3914 if (err)
3915 return err;
3916
3917 /* reinit free segmap based on SIT */
c39a1b34
JK		 3918 err = build_sit_entries(sbi);
3919 if (err)
3920 return err;
351df4b2
JK		 3921
3922 init_free_segmap(sbi);
3923 err = build_dirty_segmap(sbi);
3924 if (err)
3925 return err;
3926
3927 init_min_max_mtime(sbi);
3928 return 0;
3929}
3930
3931static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3932 enum dirty_type dirty_type)
3933{
3934 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3935
3936 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3937 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3938 dirty_i->nr_dirty[dirty_type] = 0;
3939 mutex_unlock(&dirty_i->seglist_lock);
3940}
3941
5ec4e49f 3942static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3943{
3944 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3945 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3946}
3947
3948static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3949{
3950 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3951 int i;
3952
3953 if (!dirty_i)
3954 return;
3955
3956 /* discard pre-free/dirty segments list */
3957 for (i = 0; i < NR_DIRTY_TYPE; i++)
3958 discard_dirty_segmap(sbi, i);
3959
5ec4e49f 3960 destroy_victim_secmap(sbi);
351df4b2
JK		 3961 SM_I(sbi)->dirty_info = NULL;
3962 kfree(dirty_i);
3963}
3964
3965static void destroy_curseg(struct f2fs_sb_info *sbi)
3966{
3967 struct curseg_info *array = SM_I(sbi)->curseg_array;
3968 int i;
3969
3970 if (!array)
3971 return;
3972 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3973 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3974 kfree(array[i].sum_blk);
b7ad7512
CY		 3975 kfree(array[i].journal);
3976 }
351df4b2
JK		 3977 kfree(array);
3978}
3979
3980static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3981{
3982 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3983 if (!free_i)
3984 return;
3985 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3986 kvfree(free_i->free_segmap);
3987 kvfree(free_i->free_secmap);
351df4b2
JK		 3988 kfree(free_i);
3989}
3990
3991static void destroy_sit_info(struct f2fs_sb_info *sbi)
3992{
3993 struct sit_info *sit_i = SIT_I(sbi);
3994 unsigned int start;
3995
3996 if (!sit_i)
3997 return;
3998
3999 if (sit_i->sentries) {
7cd8558b 4000 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 4001 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 4002#ifdef CONFIG_F2FS_CHECK_FS
4003 kfree(sit_i->sentries[start].cur_valid_map_mir);
4004#endif
351df4b2 4005 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 4006 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4007 }
4008 }
60a3b782
JK		 4009 kfree(sit_i->tmp_map);
4010
39307a8e
JK		 4011 kvfree(sit_i->sentries);
4012 kvfree(sit_i->sec_entries);
4013 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4014
4015 SM_I(sbi)->sit_info = NULL;
4016 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 4017#ifdef CONFIG_F2FS_CHECK_FS
4018 kfree(sit_i->sit_bitmap_mir);
4019#endif
351df4b2
JK		 4020 kfree(sit_i);
4021}
4022
4023void destroy_segment_manager(struct f2fs_sb_info *sbi)
4024{
4025 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4026
3b03f724
CY		 4027 if (!sm_info)
4028 return;
5eba8c5d 4029 destroy_flush_cmd_control(sbi, true);
f099405f 4030 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4031 destroy_dirty_segmap(sbi);
4032 destroy_curseg(sbi);
4033 destroy_free_segmap(sbi);
4034 destroy_sit_info(sbi);
4035 sbi->sm_info = NULL;
4036 kfree(sm_info);
4037}
7fd9e544
JK		 4038
4039int __init create_segment_manager_caches(void)
4040{
4041 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4042 sizeof(struct discard_entry));
7fd9e544 4043 if (!discard_entry_slab)
184a5cd2
CY		 4044 goto fail;
4045
b01a9201
JK		 4046 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4047 sizeof(struct discard_cmd));
4048 if (!discard_cmd_slab)
6ab2a308 4049 goto destroy_discard_entry;
275b66b0 4050
184a5cd2 4051 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4052 sizeof(struct sit_entry_set));
184a5cd2 4053 if (!sit_entry_set_slab)
b01a9201 4054 goto destroy_discard_cmd;
88b88a66
JK		 4055
4056 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4057 sizeof(struct inmem_pages));
4058 if (!inmem_entry_slab)
4059 goto destroy_sit_entry_set;
7fd9e544 4060 return 0;
184a5cd2 4061
88b88a66
JK		 4062destroy_sit_entry_set:
4063 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4064destroy_discard_cmd:
4065 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4066destroy_discard_entry:
184a5cd2
CY		 4067 kmem_cache_destroy(discard_entry_slab);
4068fail:
4069 return -ENOMEM;
7fd9e544
JK		 4070}
4071
4072void destroy_segment_manager_caches(void)
4073{
184a5cd2 4074 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4075 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4076 kmem_cache_destroy(discard_entry_slab);
88b88a66 4077 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4078}
Ubuntu Jammy Linux kernel mirror