]> 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: correct return value of f2fs_trim_fs
[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
28bc106b
CY		 233 if (recover) {
234 struct dnode_of_data dn;
235 struct node_info ni;
236
237 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 238retry:
28bc106b 239 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 240 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
241 if (err) {
242 if (err == -ENOMEM) {
243 congestion_wait(BLK_RW_ASYNC, HZ/50);
244 cond_resched();
245 goto retry;
246 }
28bc106b
CY		 247 err = -EAGAIN;
248 goto next;
249 }
250 get_node_info(sbi, dn.nid, &ni);
f1d2564a
DJ		 251 if (cur->old_addr == NEW_ADDR) {
252 invalidate_blocks(sbi, dn.data_blkaddr);
253 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
254 } else
255 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 256 cur->old_addr, ni.version, true, true);
257 f2fs_put_dnode(&dn);
258 }
259next:
63c52d78
JK		 260 /* we don't need to invalidate this in the sccessful status */
261 if (drop || recover)
262 ClearPageUptodate(page);
28bc106b 263 set_page_private(page, 0);
c81ced05 264 ClearPagePrivate(page);
28bc106b 265 f2fs_put_page(page, 1);
29b96b54
CY		 266
267 list_del(&cur->list);
268 kmem_cache_free(inmem_entry_slab, cur);
269 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
270 }
28bc106b 271 return err;
29b96b54
CY		 272}
273
57864ae5
JK		 274void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
275{
276 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
277 struct inode *inode;
278 struct f2fs_inode_info *fi;
279next:
280 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
281 if (list_empty(head)) {
282 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
283 return;
284 }
285 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
286 inode = igrab(&fi->vfs_inode);
287 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
288
289 if (inode) {
290 drop_inmem_pages(inode);
291 iput(inode);
292 }
293 congestion_wait(BLK_RW_ASYNC, HZ/50);
294 cond_resched();
295 goto next;
296}
297
29b96b54
CY		 298void drop_inmem_pages(struct inode *inode)
299{
57864ae5 300 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 301 struct f2fs_inode_info *fi = F2FS_I(inode);
302
303 mutex_lock(&fi->inmem_lock);
28bc106b 304 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 305 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
306 if (!list_empty(&fi->inmem_ilist))
307 list_del_init(&fi->inmem_ilist);
308 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 309 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 310
311 clear_inode_flag(inode, FI_ATOMIC_FILE);
84a23fbe 312 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 313 stat_dec_atomic_write(inode);
29b96b54
CY		 314}
315
8c242db9
JK		 316void drop_inmem_page(struct inode *inode, struct page *page)
317{
318 struct f2fs_inode_info *fi = F2FS_I(inode);
319 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
320 struct list_head *head = &fi->inmem_pages;
321 struct inmem_pages *cur = NULL;
322
323 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
324
325 mutex_lock(&fi->inmem_lock);
326 list_for_each_entry(cur, head, list) {
327 if (cur->page == page)
328 break;
329 }
330
331 f2fs_bug_on(sbi, !cur || cur->page != page);
332 list_del(&cur->list);
333 mutex_unlock(&fi->inmem_lock);
334
335 dec_page_count(sbi, F2FS_INMEM_PAGES);
336 kmem_cache_free(inmem_entry_slab, cur);
337
338 ClearPageUptodate(page);
339 set_page_private(page, 0);
340 ClearPagePrivate(page);
341 f2fs_put_page(page, 0);
342
343 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
344}
345
28bc106b
CY		 346static int __commit_inmem_pages(struct inode *inode,
347 struct list_head *revoke_list)
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 };
942fd319 360 pgoff_t last_idx = ULONG_MAX;
edb27dee 361 int err = 0;
88b88a66 362
88b88a66 363 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 364 struct page *page = cur->page;
365
366 lock_page(page);
367 if (page->mapping == inode->i_mapping) {
368 trace_f2fs_commit_inmem_page(page, INMEM);
369
370 set_page_dirty(page);
371 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 372 if (clear_page_dirty_for_io(page)) {
29b96b54 373 inode_dec_dirty_pages(inode);
933439c8
CY		 374 remove_dirty_inode(inode);
375 }
640cc189 376retry:
28bc106b 377 fio.page = page;
e959c8f5 378 fio.old_blkaddr = NULL_ADDR;
4d978078 379 fio.encrypted_page = NULL;
cc15620b 380 fio.need_lock = LOCK_DONE;
29b96b54
CY		 381 err = do_write_data_page(&fio);
382 if (err) {
640cc189
JK		 383 if (err == -ENOMEM) {
384 congestion_wait(BLK_RW_ASYNC, HZ/50);
385 cond_resched();
386 goto retry;
387 }
28bc106b 388 unlock_page(page);
29b96b54 389 break;
70c640b1 390 }
28bc106b
CY		 391 /* record old blkaddr for revoking */
392 cur->old_addr = fio.old_blkaddr;
942fd319 393 last_idx = page->index;
28bc106b
CY		 394 }
395 unlock_page(page);
396 list_move_tail(&cur->list, revoke_list);
88b88a66 397 }
29b96b54 398
942fd319 399 if (last_idx != ULONG_MAX)
b9109b0e 400 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 401
402 if (!err)
403 __revoke_inmem_pages(inode, revoke_list, false, false);
404
29b96b54
CY		 405 return err;
406}
407
408int commit_inmem_pages(struct inode *inode)
409{
410 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
411 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 412 struct list_head revoke_list;
413 int err;
29b96b54 414
28bc106b 415 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 416 f2fs_balance_fs(sbi, true);
417 f2fs_lock_op(sbi);
418
5fe45743
CY		 419 set_inode_flag(inode, FI_ATOMIC_COMMIT);
420
29b96b54 421 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 422 err = __commit_inmem_pages(inode, &revoke_list);
423 if (err) {
424 int ret;
425 /*
426 * try to revoke all committed pages, but still we could fail
427 * due to no memory or other reason, if that happened, EAGAIN
428 * will be returned, which means in such case, transaction is
429 * already not integrity, caller should use journal to do the
430 * recovery or rewrite & commit last transaction. For other
431 * error number, revoking was done by filesystem itself.
432 */
433 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
434 if (ret)
435 err = ret;
436
437 /* drop all uncommitted pages */
438 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
439 }
57864ae5
JK		 440 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
441 if (!list_empty(&fi->inmem_ilist))
442 list_del_init(&fi->inmem_ilist);
443 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 444 mutex_unlock(&fi->inmem_lock);
445
5fe45743
CY		 446 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
447
29b96b54 448 f2fs_unlock_op(sbi);
edb27dee 449 return err;
88b88a66
JK		 450}
451
0a8165d7 452/*
351df4b2
JK		 453 * This function balances dirty node and dentry pages.
454 * In addition, it controls garbage collection.
455 */
2c4db1a6 456void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 457{
0f348028 458#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 459 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
460 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 461 f2fs_stop_checkpoint(sbi, false);
55523519 462 }
0f348028
CY		 463#endif
464
e589c2c4 465 /* balance_fs_bg is able to be pending */
a7881893 466 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 467 f2fs_balance_fs_bg(sbi);
468
351df4b2 469 /*
029cd28c
JK		 470 * We should do GC or end up with checkpoint, if there are so many dirty
471 * dir/node pages without enough free segments.
351df4b2 472 */
7f3037a5 473 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 474 mutex_lock(&sbi->gc_mutex);
e066b83c 475 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 476 }
477}
478
4660f9c0
JK		 479void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
480{
1dcc336b 481 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 482 if (!available_free_memory(sbi, EXTENT_CACHE))
483 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 484
1b38dc8e
JK		 485 /* check the # of cached NAT entries */
486 if (!available_free_memory(sbi, NAT_ENTRIES))
487 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
488
31696580 489 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 490 try_to_free_nids(sbi, MAX_FREE_NIDS);
491 else
22ad0b6a 492 build_free_nids(sbi, false, false);
31696580 493
1c0f4bf5 494 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 495 return;
31696580 496
1b38dc8e
JK		 497 /* checkpoint is the only way to shrink partial cached entries */
498 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 499 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 500 excess_prefree_segs(sbi) ||
501 excess_dirty_nats(sbi) ||
f455c8a5 502 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 503 if (test_opt(sbi, DATA_FLUSH)) {
504 struct blk_plug plug;
505
506 blk_start_plug(&plug);
36b35a0d 507 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 508 blk_finish_plug(&plug);
509 }
4660f9c0 510 f2fs_sync_fs(sbi->sb, true);
42190d2a 511 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 512 }
4660f9c0
JK		 513}
514
20fda56b
KM		 515static int __submit_flush_wait(struct f2fs_sb_info *sbi,
516 struct block_device *bdev)
3c62be17 517{
d62fe971 518 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 519 int ret;
520
3adc5fcb 521 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 522 bio_set_dev(bio, bdev);
3c62be17
JK		 523 ret = submit_bio_wait(bio);
524 bio_put(bio);
20fda56b
KM		 525
526 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
527 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 528 return ret;
529}
530
39d787be 531static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 532{
39d787be 533 int ret = 0;
3c62be17
JK		 534 int i;
535
39d787be
CY		 536 if (!sbi->s_ndevs)
537 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 538
39d787be
CY		 539 for (i = 0; i < sbi->s_ndevs; i++) {
540 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
541 continue;
20fda56b
KM		 542 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
543 if (ret)
544 break;
3c62be17
JK		 545 }
546 return ret;
547}
548
2163d198 549static int issue_flush_thread(void *data)
6b4afdd7
JK		 550{
551 struct f2fs_sb_info *sbi = data;
b01a9201 552 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 553 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 554repeat:
555 if (kthread_should_stop())
556 return 0;
557
dc6febb6
CY		 558 sb_start_intwrite(sbi->sb);
559
721bd4d5 560 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 561 struct flush_cmd *cmd, *next;
562 int ret;
563
721bd4d5
GZ		 564 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
565 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
566
39d787be
CY		 567 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
568
569 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 570 atomic_inc(&fcc->issued_flush);
571
721bd4d5
GZ		 572 llist_for_each_entry_safe(cmd, next,
573 fcc->dispatch_list, llnode) {
6b4afdd7 574 cmd->ret = ret;
6b4afdd7
JK		 575 complete(&cmd->wait);
576 }
a688b9d9 577 fcc->dispatch_list = NULL;
6b4afdd7
JK		 578 }
579
dc6febb6
CY		 580 sb_end_intwrite(sbi->sb);
581
a688b9d9 582 wait_event_interruptible(*q,
721bd4d5 583 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 584 goto repeat;
585}
586
39d787be 587int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 588{
b01a9201 589 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 590 struct flush_cmd cmd;
8b8dd65f 591 int ret;
6b4afdd7 592
0f7b2abd
JK		 593 if (test_opt(sbi, NOBARRIER))
594 return 0;
595
8b8dd65f 596 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 597 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 598 atomic_inc(&fcc->issued_flush);
599 return ret;
600 }
740432f8 601
39d787be
CY		 602 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
603 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 604 atomic_dec(&fcc->issing_flush);
605
606 atomic_inc(&fcc->issued_flush);
740432f8
JK		 607 return ret;
608 }
6b4afdd7 609
39d787be 610 cmd.ino = ino;
adf8d90b 611 init_completion(&cmd.wait);
6b4afdd7 612
721bd4d5 613 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 614
6f890df0
CY		 615 /* update issue_list before we wake up issue_flush thread */
616 smp_mb();
617
618 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 619 wake_up(&fcc->flush_wait_queue);
6b4afdd7 620
5eba8c5d
JK		 621 if (fcc->f2fs_issue_flush) {
622 wait_for_completion(&cmd.wait);
8b8dd65f 623 atomic_dec(&fcc->issing_flush);
5eba8c5d 624 } else {
d3238691
CY		 625 struct llist_node *list;
626
627 list = llist_del_all(&fcc->issue_list);
628 if (!list) {
629 wait_for_completion(&cmd.wait);
630 atomic_dec(&fcc->issing_flush);
631 } else {
632 struct flush_cmd *tmp, *next;
633
39d787be 634 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 635
636 llist_for_each_entry_safe(tmp, next, list, llnode) {
637 if (tmp == &cmd) {
638 cmd.ret = ret;
639 atomic_dec(&fcc->issing_flush);
640 continue;
641 }
642 tmp->ret = ret;
643 complete(&tmp->wait);
644 }
645 }
5eba8c5d 646 }
adf8d90b
CY		 647
648 return cmd.ret;
6b4afdd7
JK		 649}
650
2163d198
GZ		 651int create_flush_cmd_control(struct f2fs_sb_info *sbi)
652{
653 dev_t dev = sbi->sb->s_bdev->bd_dev;
654 struct flush_cmd_control *fcc;
655 int err = 0;
656
b01a9201
JK		 657 if (SM_I(sbi)->fcc_info) {
658 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 659 if (fcc->f2fs_issue_flush)
660 return err;
5eba8c5d
JK		 661 goto init_thread;
662 }
663
acbf054d 664 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 665 if (!fcc)
666 return -ENOMEM;
8b8dd65f
CY		 667 atomic_set(&fcc->issued_flush, 0);
668 atomic_set(&fcc->issing_flush, 0);
2163d198 669 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 670 init_llist_head(&fcc->issue_list);
b01a9201 671 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 672 if (!test_opt(sbi, FLUSH_MERGE))
673 return err;
674
5eba8c5d 675init_thread:
2163d198
GZ		 676 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
677 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
678 if (IS_ERR(fcc->f2fs_issue_flush)) {
679 err = PTR_ERR(fcc->f2fs_issue_flush);
680 kfree(fcc);
b01a9201 681 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 682 return err;
683 }
2163d198
GZ		 684
685 return err;
686}
687
5eba8c5d 688void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 689{
b01a9201 690 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 691
5eba8c5d
JK		 692 if (fcc && fcc->f2fs_issue_flush) {
693 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
694
695 fcc->f2fs_issue_flush = NULL;
696 kthread_stop(flush_thread);
697 }
698 if (free) {
699 kfree(fcc);
b01a9201 700 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 701 }
2163d198
GZ		 702}
703
1228b482
CY		 704int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
705{
706 int ret = 0, i;
707
708 if (!sbi->s_ndevs)
709 return 0;
710
711 for (i = 1; i < sbi->s_ndevs; i++) {
712 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
713 continue;
714 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
715 if (ret)
716 break;
717
718 spin_lock(&sbi->dev_lock);
719 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
720 spin_unlock(&sbi->dev_lock);
721 }
722
723 return ret;
724}
725
351df4b2
JK		 726static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
727 enum dirty_type dirty_type)
728{
729 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
730
731 /* need not be added */
732 if (IS_CURSEG(sbi, segno))
733 return;
734
735 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
736 dirty_i->nr_dirty[dirty_type]++;
737
738 if (dirty_type == DIRTY) {
739 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 740 enum dirty_type t = sentry->type;
b2f2c390 741
ec325b52
JK		 742 if (unlikely(t >= DIRTY)) {
743 f2fs_bug_on(sbi, 1);
744 return;
745 }
4625d6aa
CL		 746 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
747 dirty_i->nr_dirty[t]++;
351df4b2
JK		 748 }
749}
750
751static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
752 enum dirty_type dirty_type)
753{
754 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
755
756 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
757 dirty_i->nr_dirty[dirty_type]--;
758
759 if (dirty_type == DIRTY) {
4625d6aa
CL		 760 struct seg_entry *sentry = get_seg_entry(sbi, segno);
761 enum dirty_type t = sentry->type;
762
763 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
764 dirty_i->nr_dirty[t]--;
b2f2c390 765
302bd348 766 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 767 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 768 dirty_i->victim_secmap);
351df4b2
JK		 769 }
770}
771
0a8165d7 772/*
351df4b2
JK		 773 * Should not occur error such as -ENOMEM.
774 * Adding dirty entry into seglist is not critical operation.
775 * If a given segment is one of current working segments, it won't be added.
776 */
8d8451af 777static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 778{
779 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
780 unsigned short valid_blocks;
781
782 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
783 return;
784
785 mutex_lock(&dirty_i->seglist_lock);
786
302bd348 787 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 788
789 if (valid_blocks == 0) {
790 __locate_dirty_segment(sbi, segno, PRE);
791 __remove_dirty_segment(sbi, segno, DIRTY);
792 } else if (valid_blocks < sbi->blocks_per_seg) {
793 __locate_dirty_segment(sbi, segno, DIRTY);
794 } else {
795 /* Recovery routine with SSR needs this */
796 __remove_dirty_segment(sbi, segno, DIRTY);
797 }
798
799 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 800}
801
004b6862 802static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 803 struct block_device *bdev, block_t lstart,
804 block_t start, block_t len)
275b66b0 805{
0b54fb84 806 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 807 struct list_head *pend_list;
b01a9201 808 struct discard_cmd *dc;
275b66b0 809
ba48a33e
CY		 810 f2fs_bug_on(sbi, !len);
811
812 pend_list = &dcc->pend_list[plist_idx(len)];
813
b01a9201
JK		 814 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
815 INIT_LIST_HEAD(&dc->list);
c81abe34 816 dc->bdev = bdev;
b01a9201 817 dc->lstart = lstart;
c81abe34 818 dc->start = start;
b01a9201 819 dc->len = len;
ec9895ad 820 dc->ref = 0;
15469963 821 dc->state = D_PREP;
c81abe34 822 dc->error = 0;
b01a9201 823 init_completion(&dc->wait);
22d375dd 824 list_add_tail(&dc->list, pend_list);
5f32366a 825 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 826 dcc->undiscard_blks += len;
004b6862
CY		 827
828 return dc;
829}
830
831static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
832 struct block_device *bdev, block_t lstart,
833 block_t start, block_t len,
834 struct rb_node *parent, struct rb_node **p)
835{
836 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
837 struct discard_cmd *dc;
838
839 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
840
841 rb_link_node(&dc->rb_node, parent, p);
842 rb_insert_color(&dc->rb_node, &dcc->root);
843
844 return dc;
15469963
JK		 845}
846
004b6862
CY		 847static void __detach_discard_cmd(struct discard_cmd_control *dcc,
848 struct discard_cmd *dc)
15469963 849{
dcc9165d 850 if (dc->state == D_DONE)
004b6862
CY		 851 atomic_dec(&dcc->issing_discard);
852
853 list_del(&dc->list);
854 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 855 dcc->undiscard_blks -= dc->len;
004b6862
CY		 856
857 kmem_cache_free(discard_cmd_slab, dc);
858
859 atomic_dec(&dcc->discard_cmd_cnt);
860}
861
862static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
863 struct discard_cmd *dc)
864{
865 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 866
2ec6f2ef
CY		 867 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
868
d9703d90
CY		 869 f2fs_bug_on(sbi, dc->ref);
870
c81abe34
JK		 871 if (dc->error == -EOPNOTSUPP)
872 dc->error = 0;
15469963 873
c81abe34 874 if (dc->error)
15469963 875 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 876 "Issue discard(%u, %u, %u) failed, ret: %d",
877 dc->lstart, dc->start, dc->len, dc->error);
004b6862 878 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 879}
880
c81abe34
JK		 881static void f2fs_submit_discard_endio(struct bio *bio)
882{
883 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
884
4e4cbee9 885 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 886 dc->state = D_DONE;
e31b9821 887 complete_all(&dc->wait);
c81abe34
JK		 888 bio_put(bio);
889}
890
94b1e10e 891static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 892 block_t start, block_t end)
893{
894#ifdef CONFIG_F2FS_CHECK_FS
895 struct seg_entry *sentry;
896 unsigned int segno;
897 block_t blk = start;
898 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
899 unsigned long *map;
900
901 while (blk < end) {
902 segno = GET_SEGNO(sbi, blk);
903 sentry = get_seg_entry(sbi, segno);
904 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
905
008396e1
YS		 906 if (end < START_BLOCK(sbi, segno + 1))
907 size = GET_BLKOFF_FROM_SEG0(sbi, end);
908 else
909 size = max_blocks;
6915ea9d
CY		 910 map = (unsigned long *)(sentry->cur_valid_map);
911 offset = __find_rev_next_bit(map, size, offset);
912 f2fs_bug_on(sbi, offset != size);
008396e1 913 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 914 }
915#endif
916}
917
8bb4f253
JK		 918static void __init_discard_policy(struct f2fs_sb_info *sbi,
919 struct discard_policy *dpolicy,
920 int discard_type, unsigned int granularity)
921{
922 /* common policy */
923 dpolicy->type = discard_type;
924 dpolicy->sync = true;
925 dpolicy->granularity = granularity;
926
927 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
928 dpolicy->io_aware_gran = MAX_PLIST_NUM;
929
930 if (discard_type == DPOLICY_BG) {
931 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
932 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
933 dpolicy->io_aware = true;
cba60849 934 dpolicy->sync = false;
8bb4f253
JK		 935 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
936 dpolicy->granularity = 1;
937 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
938 }
939 } else if (discard_type == DPOLICY_FORCE) {
940 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
941 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
942 dpolicy->io_aware = false;
943 } else if (discard_type == DPOLICY_FSTRIM) {
944 dpolicy->io_aware = false;
945 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 946 dpolicy->max_requests = UINT_MAX;
8bb4f253
JK		 947 dpolicy->io_aware = false;
948 }
949}
950
951
c81abe34
JK		 952/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
953static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56
CY		 954 struct discard_policy *dpolicy,
955 struct discard_cmd *dc)
c81abe34
JK		 956{
957 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 958 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
959 &(dcc->fstrim_list) : &(dcc->wait_list);
c81abe34 960 struct bio *bio = NULL;
78997b56 961 int flag = dpolicy->sync ? REQ_SYNC : 0;
c81abe34
JK		 962
963 if (dc->state != D_PREP)
964 return;
965
0243a5f9
CY		 966 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
967
c81abe34
JK		 968 dc->error = __blkdev_issue_discard(dc->bdev,
969 SECTOR_FROM_BLOCK(dc->start),
970 SECTOR_FROM_BLOCK(dc->len),
971 GFP_NOFS, 0, &bio);
972 if (!dc->error) {
973 /* should keep before submission to avoid D_DONE right away */
974 dc->state = D_SUBMIT;
8b8dd65f
CY		 975 atomic_inc(&dcc->issued_discard);
976 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 977 if (bio) {
978 bio->bi_private = dc;
979 bio->bi_end_io = f2fs_submit_discard_endio;
ecc9aa00 980 bio->bi_opf |= flag;
c81abe34 981 submit_bio(bio);
8412663d 982 list_move_tail(&dc->list, wait_list);
6915ea9d 983 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 984
985 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 986 }
987 } else {
988 __remove_discard_cmd(sbi, dc);
989 }
990}
991
004b6862
CY		 992static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
993 struct block_device *bdev, block_t lstart,
994 block_t start, block_t len,
995 struct rb_node **insert_p,
996 struct rb_node *insert_parent)
c81abe34 997{
004b6862 998 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 999 struct rb_node **p;
004b6862
CY		 1000 struct rb_node *parent = NULL;
1001 struct discard_cmd *dc = NULL;
1002
1003 if (insert_p && insert_parent) {
1004 parent = insert_parent;
1005 p = insert_p;
1006 goto do_insert;
1007 }
c81abe34 1008
004b6862
CY		 1009 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1010do_insert:
1011 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1012 if (!dc)
1013 return NULL;
c81abe34 1014
004b6862 1015 return dc;
c81abe34
JK		 1016}
1017
ba48a33e
CY		 1018static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1019 struct discard_cmd *dc)
1020{
1021 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1022}
1023
3d6a650f
YH		 1024static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1025 struct discard_cmd *dc, block_t blkaddr)
1026{
ba48a33e 1027 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1028 struct discard_info di = dc->di;
1029 bool modified = false;
3d6a650f 1030
004b6862 1031 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1032 __remove_discard_cmd(sbi, dc);
1033 return;
1034 }
1035
d84d1cbd
CY		 1036 dcc->undiscard_blks -= di.len;
1037
004b6862 1038 if (blkaddr > di.lstart) {
3d6a650f 1039 dc->len = blkaddr - dc->lstart;
d84d1cbd 1040 dcc->undiscard_blks += dc->len;
ba48a33e 1041 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1042 modified = true;
1043 }
1044
1045 if (blkaddr < di.lstart + di.len - 1) {
1046 if (modified) {
1047 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1048 di.start + blkaddr + 1 - di.lstart,
1049 di.lstart + di.len - 1 - blkaddr,
1050 NULL, NULL);
1051 } else {
1052 dc->lstart++;
1053 dc->len--;
1054 dc->start++;
d84d1cbd 1055 dcc->undiscard_blks += dc->len;
ba48a33e 1056 __relocate_discard_cmd(dcc, dc);
004b6862 1057 }
3d6a650f
YH		 1058 }
1059}
1060
004b6862
CY		 1061static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1062 struct block_device *bdev, block_t lstart,
1063 block_t start, block_t len)
275b66b0 1064{
0b54fb84 1065 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1066 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1067 struct discard_cmd *dc;
1068 struct discard_info di = {0};
1069 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1070 block_t end = lstart + len;
275b66b0 1071
15469963 1072 mutex_lock(&dcc->cmd_lock);
40465257 1073
004b6862
CY		 1074 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1075 NULL, lstart,
1076 (struct rb_entry **)&prev_dc,
1077 (struct rb_entry **)&next_dc,
1078 &insert_p, &insert_parent, true);
1079 if (dc)
1080 prev_dc = dc;
1081
1082 if (!prev_dc) {
1083 di.lstart = lstart;
1084 di.len = next_dc ? next_dc->lstart - lstart : len;
1085 di.len = min(di.len, len);
1086 di.start = start;
22d375dd 1087 }
15469963 1088
004b6862
CY		 1089 while (1) {
1090 struct rb_node *node;
1091 bool merged = false;
1092 struct discard_cmd *tdc = NULL;
1093
1094 if (prev_dc) {
1095 di.lstart = prev_dc->lstart + prev_dc->len;
1096 if (di.lstart < lstart)
1097 di.lstart = lstart;
1098 if (di.lstart >= end)
1099 break;
1100
1101 if (!next_dc || next_dc->lstart > end)
1102 di.len = end - di.lstart;
1103 else
1104 di.len = next_dc->lstart - di.lstart;
1105 di.start = start + di.lstart - lstart;
1106 }
1107
1108 if (!di.len)
1109 goto next;
1110
1111 if (prev_dc && prev_dc->state == D_PREP &&
1112 prev_dc->bdev == bdev &&
1113 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1114 prev_dc->di.len += di.len;
d84d1cbd 1115 dcc->undiscard_blks += di.len;
ba48a33e 1116 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1117 di = prev_dc->di;
1118 tdc = prev_dc;
1119 merged = true;
1120 }
1121
1122 if (next_dc && next_dc->state == D_PREP &&
1123 next_dc->bdev == bdev &&
1124 __is_discard_front_mergeable(&di, &next_dc->di)) {
1125 next_dc->di.lstart = di.lstart;
1126 next_dc->di.len += di.len;
1127 next_dc->di.start = di.start;
d84d1cbd 1128 dcc->undiscard_blks += di.len;
ba48a33e 1129 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1130 if (tdc)
1131 __remove_discard_cmd(sbi, tdc);
004b6862 1132 merged = true;
4e6a8d9b 1133 }
004b6862 1134
df0f6b44 1135 if (!merged) {
004b6862
CY		 1136 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1137 di.len, NULL, NULL);
df0f6b44 1138 }
004b6862
CY		 1139 next:
1140 prev_dc = next_dc;
1141 if (!prev_dc)
1142 break;
1143
1144 node = rb_next(&prev_dc->rb_node);
1145 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1146 }
1147
1148 mutex_unlock(&dcc->cmd_lock);
1149}
1150
1151static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1152 struct block_device *bdev, block_t blkstart, block_t blklen)
1153{
1154 block_t lblkstart = blkstart;
1155
0243a5f9 1156 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1157
1158 if (sbi->s_ndevs) {
1159 int devi = f2fs_target_device_index(sbi, blkstart);
1160
1161 blkstart -= FDEV(devi).start_blk;
1162 }
1163 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1164 return 0;
1165}
1166
78997b56
CY		 1167static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1168 struct discard_policy *dpolicy)
bd5b0738
CY		 1169{
1170 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1171 struct list_head *pend_list;
1172 struct discard_cmd *dc, *tmp;
1173 struct blk_plug plug;
78997b56 1174 int i, iter = 0, issued = 0;
e6c6de18 1175 bool io_interrupted = false;
bd5b0738 1176
78997b56
CY		 1177 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1178 if (i + 1 < dpolicy->granularity)
1179 break;
bd5b0738 1180 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1181
1182 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1183 if (list_empty(pend_list))
1184 goto next;
33da62cf
CY		 1185 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1186 blk_start_plug(&plug);
bd5b0738
CY		 1187 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1188 f2fs_bug_on(sbi, dc->state != D_PREP);
1189
ecc9aa00
CY		 1190 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1191 !is_idle(sbi)) {
e6c6de18 1192 io_interrupted = true;
ecc9aa00 1193 goto skip;
969d1b18 1194 }
e6c6de18 1195
78997b56 1196 __submit_discard_cmd(sbi, dpolicy, dc);
ecc9aa00
CY		 1197 issued++;
1198skip:
1199 if (++iter >= dpolicy->max_requests)
33da62cf 1200 break;
bd5b0738 1201 }
33da62cf 1202 blk_finish_plug(&plug);
49c60c67 1203next:
33da62cf
CY		 1204 mutex_unlock(&dcc->cmd_lock);
1205
1206 if (iter >= dpolicy->max_requests)
1207 break;
bd5b0738 1208 }
969d1b18 1209
e6c6de18
CY		 1210 if (!issued && io_interrupted)
1211 issued = -1;
1212
969d1b18
CY		 1213 return issued;
1214}
1215
cf5c759f 1216static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1217{
1218 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1219 struct list_head *pend_list;
1220 struct discard_cmd *dc, *tmp;
1221 int i;
cf5c759f 1222 bool dropped = false;
969d1b18
CY		 1223
1224 mutex_lock(&dcc->cmd_lock);
1225 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1226 pend_list = &dcc->pend_list[i];
1227 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1228 f2fs_bug_on(sbi, dc->state != D_PREP);
1229 __remove_discard_cmd(sbi, dc);
cf5c759f 1230 dropped = true;
969d1b18
CY		 1231 }
1232 }
1233 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1234
1235 return dropped;
bd5b0738
CY		 1236}
1237
7950e9ac
CY		 1238void drop_discard_cmd(struct f2fs_sb_info *sbi)
1239{
1240 __drop_discard_cmd(sbi);
1241}
1242
0ea80512 1243static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1244 struct discard_cmd *dc)
1245{
1246 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1247 unsigned int len = 0;
2a510c00
CY		 1248
1249 wait_for_completion_io(&dc->wait);
1250 mutex_lock(&dcc->cmd_lock);
1251 f2fs_bug_on(sbi, dc->state != D_DONE);
1252 dc->ref--;
0ea80512
CY		 1253 if (!dc->ref) {
1254 if (!dc->error)
1255 len = dc->len;
2a510c00 1256 __remove_discard_cmd(sbi, dc);
0ea80512 1257 }
2a510c00 1258 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1259
1260 return len;
2a510c00
CY		 1261}
1262
0ea80512 1263static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1264 struct discard_policy *dpolicy,
1265 block_t start, block_t end)
63a94fa1
CY		 1266{
1267 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1268 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1269 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1270 struct discard_cmd *dc, *tmp;
6afae633 1271 bool need_wait;
0ea80512 1272 unsigned int trimmed = 0;
6afae633
CY		 1273
1274next:
1275 need_wait = false;
63a94fa1
CY		 1276
1277 mutex_lock(&dcc->cmd_lock);
1278 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1279 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1280 continue;
78997b56 1281 if (dc->len < dpolicy->granularity)
8412663d 1282 continue;
78997b56 1283 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1284 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1285 if (!dc->error)
1286 trimmed += dc->len;
63a94fa1 1287 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1288 } else {
1289 dc->ref++;
1290 need_wait = true;
1291 break;
63a94fa1
CY		 1292 }
1293 }
1294 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1295
1296 if (need_wait) {
0ea80512 1297 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1298 goto next;
1299 }
0ea80512
CY		 1300
1301 return trimmed;
63a94fa1
CY		 1302}
1303
78997b56
CY		 1304static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1305 struct discard_policy *dpolicy)
8412663d 1306{
9a997188
JK		 1307 struct discard_policy dp;
1308
1309 if (dpolicy) {
1310 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1311 return;
1312 }
1313
1314 /* wait all */
8bb4f253 1315 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
9a997188 1316 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1317 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
9a997188 1318 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8412663d
CY		 1319}
1320
004b6862 1321/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1322static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1323{
1324 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1325 struct discard_cmd *dc;
ec9895ad 1326 bool need_wait = false;
004b6862
CY		 1327
1328 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1329 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1330 if (dc) {
ec9895ad
CY		 1331 if (dc->state == D_PREP) {
1332 __punch_discard_cmd(sbi, dc, blkaddr);
1333 } else {
1334 dc->ref++;
1335 need_wait = true;
1336 }
275b66b0 1337 }
d431413f 1338 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1339
2a510c00
CY		 1340 if (need_wait)
1341 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1342}
1343
cce13252
CY		 1344void stop_discard_thread(struct f2fs_sb_info *sbi)
1345{
1346 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1347
1348 if (dcc && dcc->f2fs_issue_discard) {
1349 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1350
1351 dcc->f2fs_issue_discard = NULL;
1352 kthread_stop(discard_thread);
ec9895ad 1353 }
d431413f
CY		 1354}
1355
8412663d 1356/* This comes from f2fs_put_super */
cf5c759f 1357bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1358{
1359 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1360 struct discard_policy dpolicy;
cf5c759f 1361 bool dropped;
969d1b18 1362
8bb4f253
JK		 1363 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1364 dcc->discard_granularity);
78997b56 1365 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1366 dropped = __drop_discard_cmd(sbi);
cf5c759f 1367
9a997188
JK		 1368 /* just to make sure there is no pending discard commands */
1369 __wait_all_discard_cmd(sbi, NULL);
cf5c759f 1370 return dropped;
969d1b18
CY		 1371}
1372
15469963
JK		 1373static int issue_discard_thread(void *data)
1374{
1375 struct f2fs_sb_info *sbi = data;
1376 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1377 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1378 struct discard_policy dpolicy;
969d1b18
CY		 1379 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1380 int issued;
15469963 1381
1d7be270 1382 set_freezable();
15469963 1383
1d7be270 1384 do {
8bb4f253 1385 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1386 dcc->discard_granularity);
1387
969d1b18
CY		 1388 wait_event_interruptible_timeout(*q,
1389 kthread_should_stop() || freezing(current) ||
1390 dcc->discard_wake,
1391 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1392 if (try_to_freeze())
1393 continue;
3b60d802
CY		 1394 if (f2fs_readonly(sbi->sb))
1395 continue;
1d7be270
JK		 1396 if (kthread_should_stop())
1397 return 0;
15469963 1398
dee02f0d 1399 if (dcc->discard_wake)
969d1b18 1400 dcc->discard_wake = 0;
dee02f0d
JK		 1401
1402 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
8bb4f253 1403 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1404
dc6febb6
CY		 1405 sb_start_intwrite(sbi->sb);
1406
78997b56 1407 issued = __issue_discard_cmd(sbi, &dpolicy);
969d1b18 1408 if (issued) {
78997b56
CY		 1409 __wait_all_discard_cmd(sbi, &dpolicy);
1410 wait_ms = dpolicy.min_interval;
969d1b18 1411 } else {
78997b56 1412 wait_ms = dpolicy.max_interval;
969d1b18 1413 }
1d7be270 1414
dc6febb6 1415 sb_end_intwrite(sbi->sb);
1d7be270 1416
1d7be270
JK		 1417 } while (!kthread_should_stop());
1418 return 0;
15469963
JK		 1419}
1420
f46e8809 1421#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1422static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1423 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1424{
92592285 1425 sector_t sector, nr_sects;
10a875f8 1426 block_t lblkstart = blkstart;
3c62be17
JK		 1427 int devi = 0;
1428
1429 if (sbi->s_ndevs) {
1430 devi = f2fs_target_device_index(sbi, blkstart);
1431 blkstart -= FDEV(devi).start_blk;
1432 }
f46e8809
DLM		 1433
1434 /*
1435 * We need to know the type of the zone: for conventional zones,
1436 * use regular discard if the drive supports it. For sequential
1437 * zones, reset the zone write pointer.
1438 */
3c62be17 1439 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1440
1441 case BLK_ZONE_TYPE_CONVENTIONAL:
1442 if (!blk_queue_discard(bdev_get_queue(bdev)))
1443 return 0;
c81abe34 1444 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1445 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1446 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1447 sector = SECTOR_FROM_BLOCK(blkstart);
1448 nr_sects = SECTOR_FROM_BLOCK(blklen);
1449
1450 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1451 nr_sects != bdev_zone_sectors(bdev)) {
1452 f2fs_msg(sbi->sb, KERN_INFO,
1453 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1454 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1455 blkstart, blklen);
1456 return -EIO;
1457 }
d50aaeec 1458 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1459 return blkdev_reset_zones(bdev, sector,
1460 nr_sects, GFP_NOFS);
1461 default:
1462 /* Unknown zone type: broken device ? */
1463 return -EIO;
1464 }
1465}
1466#endif
1467
3c62be17
JK		 1468static int __issue_discard_async(struct f2fs_sb_info *sbi,
1469 struct block_device *bdev, block_t blkstart, block_t blklen)
1470{
1471#ifdef CONFIG_BLK_DEV_ZONED
ccd31cb2 1472 if (f2fs_sb_has_blkzoned(sbi->sb) &&
3c62be17
JK		 1473 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1474 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1475#endif
c81abe34 1476 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1477}
1478
1e87a78d 1479static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1480 block_t blkstart, block_t blklen)
1481{
3c62be17
JK		 1482 sector_t start = blkstart, len = 0;
1483 struct block_device *bdev;
a66cdd98
JK		 1484 struct seg_entry *se;
1485 unsigned int offset;
1486 block_t i;
3c62be17
JK		 1487 int err = 0;
1488
1489 bdev = f2fs_target_device(sbi, blkstart, NULL);
1490
1491 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1492 if (i != start) {
1493 struct block_device *bdev2 =
1494 f2fs_target_device(sbi, i, NULL);
1495
1496 if (bdev2 != bdev) {
1497 err = __issue_discard_async(sbi, bdev,
1498 start, len);
1499 if (err)
1500 return err;
1501 bdev = bdev2;
1502 start = i;
1503 len = 0;
1504 }
1505 }
a66cdd98 1506
a66cdd98
JK		 1507 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1508 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1509
1510 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1511 sbi->discard_blks--;
1512 }
f46e8809 1513
3c62be17
JK		 1514 if (len)
1515 err = __issue_discard_async(sbi, bdev, start, len);
1516 return err;
1e87a78d
JK		 1517}
1518
25290fa5
JK		 1519static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1520 bool check_only)
adf4983b 1521{
b2955550
JK		 1522 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1523 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1524 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1525 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1526 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1527 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1528 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1529 unsigned int start = 0, end = -1;
c473f1a9 1530 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1531 struct discard_entry *de = NULL;
46f84c2c 1532 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1533 int i;
1534
3e025740 1535 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1536 return false;
b2955550 1537
a66cdd98
JK		 1538 if (!force) {
1539 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1540 SM_I(sbi)->dcc_info->nr_discards >=
1541 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1542 return false;
4b2fecc8
JK		 1543 }
1544
b2955550
JK		 1545 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1546 for (i = 0; i < entries; i++)
a66cdd98 1547 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1548 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1549
0b54fb84
JK		 1550 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1551 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1552 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1553 if (start >= max_blocks)
1554 break;
1555
1556 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1557 if (force && start && end != max_blocks
1558 && (end - start) < cpc->trim_minlen)
1559 continue;
1560
25290fa5
JK		 1561 if (check_only)
1562 return true;
1563
a7eeb823
CY		 1564 if (!de) {
1565 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1566 GFP_F2FS_ZERO);
1567 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1568 list_add_tail(&de->list, head);
1569 }
1570
1571 for (i = start; i < end; i++)
1572 __set_bit_le(i, (void *)de->discard_map);
1573
1574 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1575 }
25290fa5 1576 return false;
b2955550
JK		 1577}
1578
4b2fecc8
JK		 1579void release_discard_addrs(struct f2fs_sb_info *sbi)
1580{
46f84c2c 1581 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1582 struct discard_entry *entry, *this;
1583
1584 /* drop caches */
1585 list_for_each_entry_safe(entry, this, head, list) {
1586 list_del(&entry->list);
1587 kmem_cache_free(discard_entry_slab, entry);
1588 }
1589}
1590
0a8165d7 1591/*
351df4b2
JK		 1592 * Should call clear_prefree_segments after checkpoint is done.
1593 */
1594static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1595{
1596 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1597 unsigned int segno;
351df4b2
JK		 1598
1599 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1600 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1601 __set_test_and_free(sbi, segno);
351df4b2
JK		 1602 mutex_unlock(&dirty_i->seglist_lock);
1603}
1604
836b5a63 1605void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1606{
969d1b18
CY		 1607 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1608 struct list_head *head = &dcc->entry_list;
2d7b822a 1609 struct discard_entry *entry, *this;
351df4b2 1610 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1611 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1612 unsigned int start = 0, end = -1;
36abef4e 1613 unsigned int secno, start_segno;
c473f1a9 1614 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1615
1616 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1617
351df4b2 1618 while (1) {
29e59c14 1619 int i;
7cd8558b
JK		 1620 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1621 if (start >= MAIN_SEGS(sbi))
351df4b2 1622 break;
7cd8558b
JK		 1623 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1624 start + 1);
29e59c14
CL		 1625
1626 for (i = start; i < end; i++)
1627 clear_bit(i, prefree_map);
1628
1629 dirty_i->nr_dirty[PRE] -= end - start;
1630
650d3c4e 1631 if (!test_opt(sbi, DISCARD))
29e59c14 1632 continue;
351df4b2 1633
650d3c4e
YH		 1634 if (force && start >= cpc->trim_start &&
1635 (end - 1) <= cpc->trim_end)
1636 continue;
1637
36abef4e
JK		 1638 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1639 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1640 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1641 continue;
1642 }
1643next:
4ddb1a4d
JK		 1644 secno = GET_SEC_FROM_SEG(sbi, start);
1645 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1646 if (!IS_CURSEC(sbi, secno) &&
302bd348 1647 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1648 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1649 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1650
1651 start = start_segno + sbi->segs_per_sec;
1652 if (start < end)
1653 goto next;
8b107f5b
JK		 1654 else
1655 end = start - 1;
351df4b2
JK		 1656 }
1657 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1658
1659 /* send small discards */
2d7b822a 1660 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1661 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1662 bool is_valid = test_bit_le(0, entry->discard_map);
1663
1664find_next:
1665 if (is_valid) {
1666 next_pos = find_next_zero_bit_le(entry->discard_map,
1667 sbi->blocks_per_seg, cur_pos);
1668 len = next_pos - cur_pos;
1669
ccd31cb2 1670 if (f2fs_sb_has_blkzoned(sbi->sb) ||
acfd2810 1671 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1672 goto skip;
1673
1674 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1675 len);
a7eeb823
CY		 1676 total_len += len;
1677 } else {
1678 next_pos = find_next_bit_le(entry->discard_map,
1679 sbi->blocks_per_seg, cur_pos);
1680 }
836b5a63 1681skip:
a7eeb823
CY		 1682 cur_pos = next_pos;
1683 is_valid = !is_valid;
1684
1685 if (cur_pos < sbi->blocks_per_seg)
1686 goto find_next;
1687
b2955550 1688 list_del(&entry->list);
969d1b18 1689 dcc->nr_discards -= total_len;
b2955550
JK		 1690 kmem_cache_free(discard_entry_slab, entry);
1691 }
34e159da 1692
01983c71 1693 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1694}
1695
8ed59745 1696static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1697{
15469963 1698 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1699 struct discard_cmd_control *dcc;
ba48a33e 1700 int err = 0, i;
0b54fb84
JK		 1701
1702 if (SM_I(sbi)->dcc_info) {
1703 dcc = SM_I(sbi)->dcc_info;
1704 goto init_thread;
1705 }
1706
acbf054d 1707 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 1708 if (!dcc)
1709 return -ENOMEM;
1710
969d1b18 1711 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1712 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1713 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1714 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 1715 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1716 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1717 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1718 atomic_set(&dcc->issued_discard, 0);
1719 atomic_set(&dcc->issing_discard, 0);
5f32366a 1720 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1721 dcc->nr_discards = 0;
d618ebaf 1722 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1723 dcc->undiscard_blks = 0;
004b6862 1724 dcc->root = RB_ROOT;
0b54fb84 1725
15469963 1726 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1727 SM_I(sbi)->dcc_info = dcc;
1728init_thread:
15469963
JK		 1729 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1730 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1731 if (IS_ERR(dcc->f2fs_issue_discard)) {
1732 err = PTR_ERR(dcc->f2fs_issue_discard);
1733 kfree(dcc);
1734 SM_I(sbi)->dcc_info = NULL;
1735 return err;
1736 }
1737
0b54fb84
JK		 1738 return err;
1739}
1740
f099405f 1741static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1742{
1743 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1744
f099405f
CY		 1745 if (!dcc)
1746 return;
1747
cce13252 1748 stop_discard_thread(sbi);
f099405f
CY		 1749
1750 kfree(dcc);
1751 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1752}
1753
184a5cd2 1754static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1755{
1756 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1757
1758 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1759 sit_i->dirty_sentries++;
184a5cd2
CY		 1760 return false;
1761 }
1762
1763 return true;
351df4b2
JK		 1764}
1765
1766static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1767 unsigned int segno, int modified)
1768{
1769 struct seg_entry *se = get_seg_entry(sbi, segno);
1770 se->type = type;
1771 if (modified)
1772 __mark_sit_entry_dirty(sbi, segno);
1773}
1774
1775static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1776{
1777 struct seg_entry *se;
1778 unsigned int segno, offset;
1779 long int new_vblocks;
6415fedc
YS		 1780 bool exist;
1781#ifdef CONFIG_F2FS_CHECK_FS
1782 bool mir_exist;
1783#endif
351df4b2
JK		 1784
1785 segno = GET_SEGNO(sbi, blkaddr);
1786
1787 se = get_seg_entry(sbi, segno);
1788 new_vblocks = se->valid_blocks + del;
491c0854 1789 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1790
9850cf4a 1791 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1792 (new_vblocks > sbi->blocks_per_seg)));
1793
1794 se->valid_blocks = new_vblocks;
1795 se->mtime = get_mtime(sbi);
1796 SIT_I(sbi)->max_mtime = se->mtime;
1797
1798 /* Update valid block bitmap */
1799 if (del > 0) {
6415fedc 1800 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1801#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1802 mir_exist = f2fs_test_and_set_bit(offset,
1803 se->cur_valid_map_mir);
1804 if (unlikely(exist != mir_exist)) {
1805 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1806 "when setting bitmap, blk:%u, old bit:%d",
1807 blkaddr, exist);
05796763 1808 f2fs_bug_on(sbi, 1);
6415fedc 1809 }
355e7891 1810#endif
6415fedc
YS		 1811 if (unlikely(exist)) {
1812 f2fs_msg(sbi->sb, KERN_ERR,
1813 "Bitmap was wrongly set, blk:%u", blkaddr);
1814 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1815 se->valid_blocks--;
1816 del = 0;
355e7891 1817 }
6415fedc 1818
3e025740
JK		 1819 if (f2fs_discard_en(sbi) &&
1820 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1821 sbi->discard_blks--;
720037f9
JK		 1822
1823 /* don't overwrite by SSR to keep node chain */
5d7881ca 1824 if (IS_NODESEG(se->type)) {
720037f9
JK		 1825 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1826 se->ckpt_valid_blocks++;
1827 }
351df4b2 1828 } else {
6415fedc 1829 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1830#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1831 mir_exist = f2fs_test_and_clear_bit(offset,
1832 se->cur_valid_map_mir);
1833 if (unlikely(exist != mir_exist)) {
1834 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1835 "when clearing bitmap, blk:%u, old bit:%d",
1836 blkaddr, exist);
05796763 1837 f2fs_bug_on(sbi, 1);
6415fedc 1838 }
355e7891 1839#endif
6415fedc
YS		 1840 if (unlikely(!exist)) {
1841 f2fs_msg(sbi->sb, KERN_ERR,
1842 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1843 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1844 se->valid_blocks++;
1845 del = 0;
355e7891 1846 }
6415fedc 1847
3e025740
JK		 1848 if (f2fs_discard_en(sbi) &&
1849 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1850 sbi->discard_blks++;
351df4b2
JK		 1851 }
1852 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1853 se->ckpt_valid_blocks += del;
1854
1855 __mark_sit_entry_dirty(sbi, segno);
1856
1857 /* update total number of valid blocks to be written in ckpt area */
1858 SIT_I(sbi)->written_valid_blocks += del;
1859
1860 if (sbi->segs_per_sec > 1)
1861 get_sec_entry(sbi, segno)->valid_blocks += del;
1862}
1863
351df4b2
JK		 1864void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1865{
1866 unsigned int segno = GET_SEGNO(sbi, addr);
1867 struct sit_info *sit_i = SIT_I(sbi);
1868
9850cf4a 1869 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1870 if (addr == NEW_ADDR)
1871 return;
1872
1873 /* add it into sit main buffer */
3d26fa6b 1874 down_write(&sit_i->sentry_lock);
351df4b2
JK		 1875
1876 update_sit_entry(sbi, addr, -1);
1877
1878 /* add it into dirty seglist */
1879 locate_dirty_segment(sbi, segno);
1880
3d26fa6b 1881 up_write(&sit_i->sentry_lock);
351df4b2
JK		 1882}
1883
6e2c64ad
JK		 1884bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1885{
1886 struct sit_info *sit_i = SIT_I(sbi);
1887 unsigned int segno, offset;
1888 struct seg_entry *se;
1889 bool is_cp = false;
1890
1891 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1892 return true;
1893
3d26fa6b 1894 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1895
1896 segno = GET_SEGNO(sbi, blkaddr);
1897 se = get_seg_entry(sbi, segno);
1898 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1899
1900 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1901 is_cp = true;
1902
3d26fa6b 1903 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1904
1905 return is_cp;
1906}
1907
0a8165d7 1908/*
351df4b2
JK		 1909 * This function should be resided under the curseg_mutex lock
1910 */
1911static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1912 struct f2fs_summary *sum)
351df4b2
JK		 1913{
1914 struct curseg_info *curseg = CURSEG_I(sbi, type);
1915 void *addr = curseg->sum_blk;
e79efe3b 1916 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1917 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1918}
1919
0a8165d7 1920/*
351df4b2
JK		 1921 * Calculate the number of current summary pages for writing
1922 */
3fa06d7b 1923int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1924{
351df4b2 1925 int valid_sum_count = 0;
9a47938b 1926 int i, sum_in_page;
351df4b2
JK		 1927
1928 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1929 if (sbi->ckpt->alloc_type[i] == SSR)
1930 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1931 else {
1932 if (for_ra)
1933 valid_sum_count += le16_to_cpu(
1934 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1935 else
1936 valid_sum_count += curseg_blkoff(sbi, i);
1937 }
351df4b2
JK		 1938 }
1939
09cbfeaf 1940 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1941 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1942 if (valid_sum_count <= sum_in_page)
351df4b2 1943 return 1;
9a47938b 1944 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1945 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1946 return 2;
1947 return 3;
1948}
1949
0a8165d7 1950/*
351df4b2
JK		 1951 * Caller should put this summary page
1952 */
1953struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1954{
1955 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1956}
1957
381722d2 1958void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1959{
1960 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2 1961
0537b811 1962 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 1963 set_page_dirty(page);
1964 f2fs_put_page(page, 1);
1965}
1966
381722d2
CY		 1967static void write_sum_page(struct f2fs_sb_info *sbi,
1968 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1969{
1970 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1971}
1972
b7ad7512
CY		 1973static void write_current_sum_page(struct f2fs_sb_info *sbi,
1974 int type, block_t blk_addr)
1975{
1976 struct curseg_info *curseg = CURSEG_I(sbi, type);
1977 struct page *page = grab_meta_page(sbi, blk_addr);
1978 struct f2fs_summary_block *src = curseg->sum_blk;
1979 struct f2fs_summary_block *dst;
1980
1981 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 1982 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 1983
1984 mutex_lock(&curseg->curseg_mutex);
1985
1986 down_read(&curseg->journal_rwsem);
1987 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1988 up_read(&curseg->journal_rwsem);
1989
1990 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1991 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1992
1993 mutex_unlock(&curseg->curseg_mutex);
1994
1995 set_page_dirty(page);
1996 f2fs_put_page(page, 1);
1997}
1998
a7881893
JK		 1999static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2000{
2001 struct curseg_info *curseg = CURSEG_I(sbi, type);
2002 unsigned int segno = curseg->segno + 1;
2003 struct free_segmap_info *free_i = FREE_I(sbi);
2004
2005 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2006 return !test_bit(segno, free_i->free_segmap);
2007 return 0;
2008}
2009
0a8165d7 2010/*
351df4b2
JK		 2011 * Find a new segment from the free segments bitmap to right order
2012 * This function should be returned with success, otherwise BUG
2013 */
2014static void get_new_segment(struct f2fs_sb_info *sbi,
2015 unsigned int *newseg, bool new_sec, int dir)
2016{
2017 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2018 unsigned int segno, secno, zoneno;
7cd8558b 2019 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2020 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2021 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2022 unsigned int left_start = hint;
2023 bool init = true;
2024 int go_left = 0;
2025 int i;
2026
1a118ccf 2027 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2028
2029 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2030 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2031 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2032 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2033 goto got_it;
2034 }
2035find_other_zone:
7cd8558b
JK		 2036 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2037 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2038 if (dir == ALLOC_RIGHT) {
2039 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2040 MAIN_SECS(sbi), 0);
2041 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2042 } else {
2043 go_left = 1;
2044 left_start = hint - 1;
2045 }
2046 }
2047 if (go_left == 0)
2048 goto skip_left;
2049
2050 while (test_bit(left_start, free_i->free_secmap)) {
2051 if (left_start > 0) {
2052 left_start--;
2053 continue;
2054 }
2055 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2056 MAIN_SECS(sbi), 0);
2057 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2058 break;
2059 }
2060 secno = left_start;
2061skip_left:
4ddb1a4d
JK		 2062 segno = GET_SEG_FROM_SEC(sbi, secno);
2063 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2064
2065 /* give up on finding another zone */
2066 if (!init)
2067 goto got_it;
2068 if (sbi->secs_per_zone == 1)
2069 goto got_it;
2070 if (zoneno == old_zoneno)
2071 goto got_it;
2072 if (dir == ALLOC_LEFT) {
2073 if (!go_left && zoneno + 1 >= total_zones)
2074 goto got_it;
2075 if (go_left && zoneno == 0)
2076 goto got_it;
2077 }
2078 for (i = 0; i < NR_CURSEG_TYPE; i++)
2079 if (CURSEG_I(sbi, i)->zone == zoneno)
2080 break;
2081
2082 if (i < NR_CURSEG_TYPE) {
2083 /* zone is in user, try another */
2084 if (go_left)
2085 hint = zoneno * sbi->secs_per_zone - 1;
2086 else if (zoneno + 1 >= total_zones)
2087 hint = 0;
2088 else
2089 hint = (zoneno + 1) * sbi->secs_per_zone;
2090 init = false;
2091 goto find_other_zone;
2092 }
2093got_it:
2094 /* set it as dirty segment in free segmap */
9850cf4a 2095 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2096 __set_inuse(sbi, segno);
2097 *newseg = segno;
1a118ccf 2098 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2099}
2100
2101static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2102{
2103 struct curseg_info *curseg = CURSEG_I(sbi, type);
2104 struct summary_footer *sum_footer;
2105
2106 curseg->segno = curseg->next_segno;
4ddb1a4d 2107 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2108 curseg->next_blkoff = 0;
2109 curseg->next_segno = NULL_SEGNO;
2110
2111 sum_footer = &(curseg->sum_blk->footer);
2112 memset(sum_footer, 0, sizeof(struct summary_footer));
2113 if (IS_DATASEG(type))
2114 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2115 if (IS_NODESEG(type))
2116 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2117 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2118}
2119
7a20b8a6
JK		 2120static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2121{
a7881893
JK		 2122 /* if segs_per_sec is large than 1, we need to keep original policy. */
2123 if (sbi->segs_per_sec != 1)
2124 return CURSEG_I(sbi, type)->segno;
2125
b94929d9
YS		 2126 if (test_opt(sbi, NOHEAP) &&
2127 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2128 return 0;
2129
e066b83c
JK		 2130 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2131 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2132
2133 /* find segments from 0 to reuse freed segments */
63189b78 2134 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2135 return 0;
2136
7a20b8a6
JK		 2137 return CURSEG_I(sbi, type)->segno;
2138}
2139
0a8165d7 2140/*
351df4b2
JK		 2141 * Allocate a current working segment.
2142 * This function always allocates a free segment in LFS manner.
2143 */
2144static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2145{
2146 struct curseg_info *curseg = CURSEG_I(sbi, type);
2147 unsigned int segno = curseg->segno;
2148 int dir = ALLOC_LEFT;
2149
2150 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2151 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2152 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2153 dir = ALLOC_RIGHT;
2154
2155 if (test_opt(sbi, NOHEAP))
2156 dir = ALLOC_RIGHT;
2157
7a20b8a6 2158 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2159 get_new_segment(sbi, &segno, new_sec, dir);
2160 curseg->next_segno = segno;
2161 reset_curseg(sbi, type, 1);
2162 curseg->alloc_type = LFS;
2163}
2164
2165static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2166 struct curseg_info *seg, block_t start)
2167{
2168 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2169 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2170 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2171 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2172 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2173 int i, pos;
2174
2175 for (i = 0; i < entries; i++)
2176 target_map[i] = ckpt_map[i] | cur_map[i];
2177
2178 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2179
2180 seg->next_blkoff = pos;
351df4b2
JK		 2181}
2182
0a8165d7 2183/*
351df4b2
JK		 2184 * If a segment is written by LFS manner, next block offset is just obtained
2185 * by increasing the current block offset. However, if a segment is written by
2186 * SSR manner, next block offset obtained by calling __next_free_blkoff
2187 */
2188static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2189 struct curseg_info *seg)
2190{
2191 if (seg->alloc_type == SSR)
2192 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2193 else
2194 seg->next_blkoff++;
2195}
2196
0a8165d7 2197/*
e1c42045 2198 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2199 * manner, so it should recover the existing segment information of valid blocks
2200 */
025d63a4 2201static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2202{
2203 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2204 struct curseg_info *curseg = CURSEG_I(sbi, type);
2205 unsigned int new_segno = curseg->next_segno;
2206 struct f2fs_summary_block *sum_node;
2207 struct page *sum_page;
2208
2209 write_sum_page(sbi, curseg->sum_blk,
2210 GET_SUM_BLOCK(sbi, curseg->segno));
2211 __set_test_and_inuse(sbi, new_segno);
2212
2213 mutex_lock(&dirty_i->seglist_lock);
2214 __remove_dirty_segment(sbi, new_segno, PRE);
2215 __remove_dirty_segment(sbi, new_segno, DIRTY);
2216 mutex_unlock(&dirty_i->seglist_lock);
2217
2218 reset_curseg(sbi, type, 1);
2219 curseg->alloc_type = SSR;
2220 __next_free_blkoff(sbi, curseg, 0);
2221
025d63a4
CY		 2222 sum_page = get_sum_page(sbi, new_segno);
2223 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2224 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2225 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2226}
2227
43727527
JK		 2228static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2229{
2230 struct curseg_info *curseg = CURSEG_I(sbi, type);
2231 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2232 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2233 int i, cnt;
2234 bool reversed = false;
c192f7a4
JK		 2235
2236 /* need_SSR() already forces to do this */
e066b83c
JK		 2237 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2238 curseg->next_segno = segno;
c192f7a4 2239 return 1;
e066b83c 2240 }
43727527 2241
70d625cb
JK		 2242 /* For node segments, let's do SSR more intensively */
2243 if (IS_NODESEG(type)) {
d27c3d89
CY		 2244 if (type >= CURSEG_WARM_NODE) {
2245 reversed = true;
2246 i = CURSEG_COLD_NODE;
2247 } else {
2248 i = CURSEG_HOT_NODE;
2249 }
2250 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2251 } else {
d27c3d89
CY		 2252 if (type >= CURSEG_WARM_DATA) {
2253 reversed = true;
2254 i = CURSEG_COLD_DATA;
2255 } else {
2256 i = CURSEG_HOT_DATA;
2257 }
2258 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2259 }
43727527 2260
d27c3d89 2261 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2262 if (i == type)
2263 continue;
e066b83c
JK		 2264 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2265 curseg->next_segno = segno;
43727527 2266 return 1;
e066b83c 2267 }
c192f7a4 2268 }
43727527
JK		 2269 return 0;
2270}
2271
351df4b2
JK		 2272/*
2273 * flush out current segment and replace it with new segment
2274 * This function should be returned with success, otherwise BUG
2275 */
2276static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2277 int type, bool force)
2278{
a7881893
JK		 2279 struct curseg_info *curseg = CURSEG_I(sbi, type);
2280
7b405275 2281 if (force)
351df4b2 2282 new_curseg(sbi, type, true);
5b6c6be2
JK		 2283 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2284 type == CURSEG_WARM_NODE)
351df4b2 2285 new_curseg(sbi, type, false);
a7881893
JK		 2286 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2287 new_curseg(sbi, type, false);
351df4b2 2288 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2289 change_curseg(sbi, type);
351df4b2
JK		 2290 else
2291 new_curseg(sbi, type, false);
dcdfff65 2292
a7881893 2293 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2294}
2295
2296void allocate_new_segments(struct f2fs_sb_info *sbi)
2297{
6ae1be13
JK		 2298 struct curseg_info *curseg;
2299 unsigned int old_segno;
351df4b2
JK		 2300 int i;
2301
3d26fa6b
CY		 2302 down_write(&SIT_I(sbi)->sentry_lock);
2303
6ae1be13
JK		 2304 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2305 curseg = CURSEG_I(sbi, i);
2306 old_segno = curseg->segno;
2307 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2308 locate_dirty_segment(sbi, old_segno);
2309 }
3d26fa6b
CY		 2310
2311 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2312}
2313
2314static const struct segment_allocation default_salloc_ops = {
2315 .allocate_segment = allocate_segment_by_default,
2316};
2317
25290fa5
JK		 2318bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2319{
2320 __u64 trim_start = cpc->trim_start;
2321 bool has_candidate = false;
2322
3d26fa6b 2323 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2324 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2325 if (add_discard_addrs(sbi, cpc, true)) {
2326 has_candidate = true;
2327 break;
2328 }
2329 }
3d26fa6b 2330 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2331
2332 cpc->trim_start = trim_start;
2333 return has_candidate;
2334}
2335
9a997188
JK		 2336static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2337 struct discard_policy *dpolicy,
2338 unsigned int start, unsigned int end)
2339{
2340 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2341 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2342 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2343 struct discard_cmd *dc;
2344 struct blk_plug plug;
2345 int issued;
2346
2347next:
2348 issued = 0;
2349
2350 mutex_lock(&dcc->cmd_lock);
2351 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2352
2353 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2354 NULL, start,
2355 (struct rb_entry **)&prev_dc,
2356 (struct rb_entry **)&next_dc,
2357 &insert_p, &insert_parent, true);
2358 if (!dc)
2359 dc = next_dc;
2360
2361 blk_start_plug(&plug);
2362
2363 while (dc && dc->lstart <= end) {
2364 struct rb_node *node;
2365
2366 if (dc->len < dpolicy->granularity)
2367 goto skip;
2368
2369 if (dc->state != D_PREP) {
2370 list_move_tail(&dc->list, &dcc->fstrim_list);
2371 goto skip;
2372 }
2373
2374 __submit_discard_cmd(sbi, dpolicy, dc);
2375
2376 if (++issued >= dpolicy->max_requests) {
2377 start = dc->lstart + dc->len;
2378
2379 blk_finish_plug(&plug);
2380 mutex_unlock(&dcc->cmd_lock);
2381 __wait_all_discard_cmd(sbi, NULL);
2382 congestion_wait(BLK_RW_ASYNC, HZ/50);
2383 goto next;
2384 }
2385skip:
2386 node = rb_next(&dc->rb_node);
2387 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2388
2389 if (fatal_signal_pending(current))
2390 break;
2391 }
2392
2393 blk_finish_plug(&plug);
2394 mutex_unlock(&dcc->cmd_lock);
2395}
2396
4b2fecc8
JK		 2397int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2398{
f7ef9b83
JK		 2399 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2400 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2401 unsigned int start_segno, end_segno;
8412663d 2402 block_t start_block, end_block;
4b2fecc8 2403 struct cp_control cpc;
78997b56 2404 struct discard_policy dpolicy;
0ea80512 2405 unsigned long long trimmed = 0;
c34f42e2 2406 int err = 0;
4b2fecc8 2407
836b5a63 2408 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2409 return -EINVAL;
2410
7cd8558b 2411 if (end <= MAIN_BLKADDR(sbi))
3d165dc3 2412 return -EINVAL;
4b2fecc8 2413
ed214a11
YH		 2414 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2415 f2fs_msg(sbi->sb, KERN_WARNING,
2416 "Found FS corruption, run fsck to fix.");
3d165dc3 2417 return -EIO;
ed214a11
YH		 2418 }
2419
4b2fecc8 2420 /* start/end segment number in main_area */
7cd8558b
JK		 2421 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2422 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2423 GET_SEGNO(sbi, end);
8412663d 2424
4b2fecc8 2425 cpc.reason = CP_DISCARD;
836b5a63 2426 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2427 cpc.trim_start = start_segno;
2428 cpc.trim_end = end_segno;
4b2fecc8 2429
377224c4
CY		 2430 if (sbi->discard_blks == 0)
2431 goto out;
74fa5f3d 2432
377224c4
CY		 2433 mutex_lock(&sbi->gc_mutex);
2434 err = write_checkpoint(sbi, &cpc);
2435 mutex_unlock(&sbi->gc_mutex);
2436 if (err)
2437 goto out;
8412663d
CY		 2438
2439 start_block = START_BLOCK(sbi, start_segno);
377224c4 2440 end_block = START_BLOCK(sbi, end_segno + 1);
8412663d 2441
8bb4f253 2442 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
78997b56 2443 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
e555da9f
JK		 2444
2445 /*
2446 * We filed discard candidates, but actually we don't need to wait for
2447 * all of them, since they'll be issued in idle time along with runtime
2448 * discard option. User configuration looks like using runtime discard
2449 * or periodic fstrim instead of it.
2450 */
2451 if (!test_opt(sbi, DISCARD)) {
2452 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2453 start_block, end_block);
e555da9f
JK		 2454 range->len = F2FS_BLK_TO_BYTES(trimmed);
2455 }
377224c4 2456out:
c34f42e2 2457 return err;
4b2fecc8
JK		 2458}
2459
351df4b2
JK		 2460static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2461{
2462 struct curseg_info *curseg = CURSEG_I(sbi, type);
2463 if (curseg->next_blkoff < sbi->blocks_per_seg)
2464 return true;
2465 return false;
2466}
2467
4f0a03d3
HL		 2468int rw_hint_to_seg_type(enum rw_hint hint)
2469{
2470 switch (hint) {
2471 case WRITE_LIFE_SHORT:
2472 return CURSEG_HOT_DATA;
2473 case WRITE_LIFE_EXTREME:
2474 return CURSEG_COLD_DATA;
2475 default:
2476 return CURSEG_WARM_DATA;
2477 }
2478}
2479
0cdd3195
HL		 2480/* This returns write hints for each segment type. This hints will be
2481 * passed down to block layer. There are mapping tables which depend on
2482 * the mount option 'whint_mode'.
2483 *
2484 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2485 *
2486 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2487 *
2488 * User F2FS Block
2489 * ---- ---- -----
2490 * META WRITE_LIFE_NOT_SET
2491 * HOT_NODE "
2492 * WARM_NODE "
2493 * COLD_NODE "
2494 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2495 * extension list " "
2496 *
2497 * -- buffered io
2498 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2499 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2500 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2501 * WRITE_LIFE_NONE " "
2502 * WRITE_LIFE_MEDIUM " "
2503 * WRITE_LIFE_LONG " "
2504 *
2505 * -- direct io
2506 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2507 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2508 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2509 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2510 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2511 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2512 *
f2e703f9
HL		 2513 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2514 *
2515 * User F2FS Block
2516 * ---- ---- -----
2517 * META WRITE_LIFE_MEDIUM;
2518 * HOT_NODE WRITE_LIFE_NOT_SET
2519 * WARM_NODE "
2520 * COLD_NODE WRITE_LIFE_NONE
2521 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2522 * extension list " "
2523 *
2524 * -- buffered io
2525 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2526 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2527 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2528 * WRITE_LIFE_NONE " "
2529 * WRITE_LIFE_MEDIUM " "
2530 * WRITE_LIFE_LONG " "
2531 *
2532 * -- direct io
2533 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2534 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2535 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2536 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2537 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2538 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2539 */
2540
2541enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2542 enum page_type type, enum temp_type temp)
2543{
63189b78 2544 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2545 if (type == DATA) {
f2e703f9 2546 if (temp == WARM)
0cdd3195 2547 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2548 else if (temp == HOT)
2549 return WRITE_LIFE_SHORT;
2550 else if (temp == COLD)
2551 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2552 } else {
2553 return WRITE_LIFE_NOT_SET;
2554 }
63189b78 2555 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2556 if (type == DATA) {
2557 if (temp == WARM)
2558 return WRITE_LIFE_LONG;
2559 else if (temp == HOT)
2560 return WRITE_LIFE_SHORT;
2561 else if (temp == COLD)
2562 return WRITE_LIFE_EXTREME;
2563 } else if (type == NODE) {
2564 if (temp == WARM || temp == HOT)
2565 return WRITE_LIFE_NOT_SET;
2566 else if (temp == COLD)
2567 return WRITE_LIFE_NONE;
2568 } else if (type == META) {
2569 return WRITE_LIFE_MEDIUM;
2570 }
0cdd3195 2571 }
f2e703f9 2572 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2573}
2574
81377bd6 2575static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2576{
81377bd6 2577 if (fio->type == DATA)
351df4b2
JK		 2578 return CURSEG_HOT_DATA;
2579 else
2580 return CURSEG_HOT_NODE;
2581}
2582
81377bd6 2583static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2584{
81377bd6
JK		 2585 if (fio->type == DATA) {
2586 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2587
2588 if (S_ISDIR(inode->i_mode))
2589 return CURSEG_HOT_DATA;
2590 else
2591 return CURSEG_COLD_DATA;
2592 } else {
81377bd6 2593 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2594 return CURSEG_WARM_NODE;
351df4b2
JK		 2595 else
2596 return CURSEG_COLD_NODE;
2597 }
2598}
2599
81377bd6 2600static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2601{
81377bd6
JK		 2602 if (fio->type == DATA) {
2603 struct inode *inode = fio->page->mapping->host;
351df4b2 2604
81377bd6 2605 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2606 return CURSEG_COLD_DATA;
b6a06cbb
CY		 2607 if (file_is_hot(inode) ||
2608 is_inode_flag_set(inode, FI_HOT_DATA))
ef095d19 2609 return CURSEG_HOT_DATA;
4f0a03d3 2610 return rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2611 } else {
81377bd6
JK		 2612 if (IS_DNODE(fio->page))
2613 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2614 CURSEG_HOT_NODE;
ef095d19 2615 return CURSEG_COLD_NODE;
351df4b2
JK		 2616 }
2617}
2618
81377bd6 2619static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2620{
a912b54d
JK		 2621 int type = 0;
2622
63189b78 2623 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2624 case 2:
a912b54d
JK		 2625 type = __get_segment_type_2(fio);
2626 break;
351df4b2 2627 case 4:
a912b54d
JK		 2628 type = __get_segment_type_4(fio);
2629 break;
2630 case 6:
2631 type = __get_segment_type_6(fio);
2632 break;
2633 default:
2634 f2fs_bug_on(fio->sbi, true);
351df4b2 2635 }
81377bd6 2636
a912b54d
JK		 2637 if (IS_HOT(type))
2638 fio->temp = HOT;
2639 else if (IS_WARM(type))
2640 fio->temp = WARM;
2641 else
2642 fio->temp = COLD;
2643 return type;
351df4b2
JK		 2644}
2645
bfad7c2d
JK		 2646void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2647 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2648 struct f2fs_summary *sum, int type,
2649 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2650{
2651 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2652 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2653
2b60311d
CY		 2654 down_read(&SM_I(sbi)->curseg_lock);
2655
351df4b2 2656 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2657 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2658
2659 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2660
4e6a8d9b
JK		 2661 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2662
351df4b2
JK		 2663 /*
2664 * __add_sum_entry should be resided under the curseg_mutex
2665 * because, this function updates a summary entry in the
2666 * current summary block.
2667 */
e79efe3b 2668 __add_sum_entry(sbi, type, sum);
351df4b2 2669
351df4b2 2670 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2671
2672 stat_inc_block_count(sbi, curseg);
351df4b2 2673
65f1b80b
YS		 2674 /*
2675 * SIT information should be updated before segment allocation,
2676 * since SSR needs latest valid block information.
2677 */
2678 update_sit_entry(sbi, *new_blkaddr, 1);
2679 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2680 update_sit_entry(sbi, old_blkaddr, -1);
2681
c6f82fe9
JK		 2682 if (!__has_curseg_space(sbi, type))
2683 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 2684
351df4b2 2685 /*
65f1b80b
YS		 2686 * segment dirty status should be updated after segment allocation,
2687 * so we just need to update status only one time after previous
2688 * segment being closed.
351df4b2 2689 */
65f1b80b
YS		 2690 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2691 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 2692
3d26fa6b 2693 up_write(&sit_i->sentry_lock);
351df4b2 2694
704956ec 2695 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2696 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2697
704956ec
CY		 2698 f2fs_inode_chksum_set(sbi, page);
2699 }
2700
fb830fc5
CY		 2701 if (add_list) {
2702 struct f2fs_bio_info *io;
2703
2704 INIT_LIST_HEAD(&fio->list);
2705 fio->in_list = true;
2706 io = sbi->write_io[fio->type] + fio->temp;
2707 spin_lock(&io->io_lock);
2708 list_add_tail(&fio->list, &io->io_list);
2709 spin_unlock(&io->io_lock);
2710 }
2711
bfad7c2d 2712 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 2713
2714 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 2715}
2716
39d787be
CY		 2717static void update_device_state(struct f2fs_io_info *fio)
2718{
2719 struct f2fs_sb_info *sbi = fio->sbi;
2720 unsigned int devidx;
2721
2722 if (!sbi->s_ndevs)
2723 return;
2724
2725 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2726
2727 /* update device state for fsync */
2728 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2729
2730 /* update device state for checkpoint */
2731 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2732 spin_lock(&sbi->dev_lock);
2733 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2734 spin_unlock(&sbi->dev_lock);
2735 }
39d787be
CY		 2736}
2737
05ca3632 2738static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2739{
81377bd6 2740 int type = __get_segment_type(fio);
0a595eba 2741 int err;
bfad7c2d 2742
0a595eba 2743reallocate:
7a9d7548 2744 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2745 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2746
351df4b2 2747 /* writeout dirty page into bdev */
b9109b0e 2748 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2749 if (err == -EAGAIN) {
2750 fio->old_blkaddr = fio->new_blkaddr;
2751 goto reallocate;
39d787be
CY		 2752 } else if (!err) {
2753 update_device_state(fio);
0a595eba 2754 }
351df4b2
JK		 2755}
2756
b0af6d49
CY		 2757void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2758 enum iostat_type io_type)
351df4b2 2759{
458e6197 2760 struct f2fs_io_info fio = {
05ca3632 2761 .sbi = sbi,
458e6197 2762 .type = META,
0cdd3195 2763 .temp = HOT,
04d328de 2764 .op = REQ_OP_WRITE,
70fd7614 2765 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2766 .old_blkaddr = page->index,
2767 .new_blkaddr = page->index,
05ca3632 2768 .page = page,
4375a336 2769 .encrypted_page = NULL,
fb830fc5 2770 .in_list = false,
458e6197
JK		 2771 };
2772
2b947003 2773 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2774 fio.op_flags &= ~REQ_META;
2b947003 2775
351df4b2 2776 set_page_writeback(page);
17c50035 2777 ClearPageError(page);
b9109b0e 2778 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2779
2780 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2781}
2782
05ca3632 2783void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2784{
2785 struct f2fs_summary sum;
05ca3632 2786
351df4b2 2787 set_summary(&sum, nid, 0, 0);
05ca3632 2788 do_write_page(&sum, fio);
b0af6d49
CY		 2789
2790 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2791}
2792
05ca3632 2793void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2794{
05ca3632 2795 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2796 struct f2fs_summary sum;
2797 struct node_info ni;
2798
9850cf4a 2799 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2800 get_node_info(sbi, dn->nid, &ni);
2801 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2802 do_write_page(&sum, fio);
f28b3434 2803 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2804
2805 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2806}
2807
d1b3e72d 2808int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2809{
b0af6d49 2810 int err;
d21b0f23 2811 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 2812
7a9d7548 2813 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 2814 /* i/o temperature is needed for passing down write hints */
2815 __get_segment_type(fio);
d21b0f23
YH		 2816
2817 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2818 GET_SEGNO(sbi, fio->new_blkaddr))->type));
2819
05ca3632 2820 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2821
2822 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2823 if (!err)
2824 update_device_state(fio);
b0af6d49
CY		 2825
2826 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2827
2828 return err;
351df4b2
JK		 2829}
2830
2b60311d
CY		 2831static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2832 unsigned int segno)
2833{
2834 int i;
2835
2836 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2837 if (CURSEG_I(sbi, i)->segno == segno)
2838 break;
2839 }
2840 return i;
2841}
2842
4356e48e 2843void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2844 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2845 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2846{
2847 struct sit_info *sit_i = SIT_I(sbi);
2848 struct curseg_info *curseg;
2849 unsigned int segno, old_cursegno;
2850 struct seg_entry *se;
2851 int type;
19f106bc 2852 unsigned short old_blkoff;
351df4b2
JK		 2853
2854 segno = GET_SEGNO(sbi, new_blkaddr);
2855 se = get_seg_entry(sbi, segno);
2856 type = se->type;
2857
2b60311d
CY		 2858 down_write(&SM_I(sbi)->curseg_lock);
2859
19f106bc
CY		 2860 if (!recover_curseg) {
2861 /* for recovery flow */
2862 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2863 if (old_blkaddr == NULL_ADDR)
2864 type = CURSEG_COLD_DATA;
2865 else
2866 type = CURSEG_WARM_DATA;
2867 }
2868 } else {
2b60311d
CY		 2869 if (IS_CURSEG(sbi, segno)) {
2870 /* se->type is volatile as SSR allocation */
2871 type = __f2fs_get_curseg(sbi, segno);
2872 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2873 } else {
351df4b2 2874 type = CURSEG_WARM_DATA;
2b60311d 2875 }
351df4b2 2876 }
19f106bc 2877
2c190504 2878 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 2879 curseg = CURSEG_I(sbi, type);
2880
2881 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2882 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2883
2884 old_cursegno = curseg->segno;
19f106bc 2885 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2886
2887 /* change the current segment */
2888 if (segno != curseg->segno) {
2889 curseg->next_segno = segno;
025d63a4 2890 change_curseg(sbi, type);
351df4b2
JK		 2891 }
2892
491c0854 2893 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2894 __add_sum_entry(sbi, type, sum);
351df4b2 2895
28bc106b 2896 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2897 update_sit_entry(sbi, new_blkaddr, 1);
2898 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2899 update_sit_entry(sbi, old_blkaddr, -1);
2900
2901 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2902 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2903
351df4b2 2904 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2905
19f106bc
CY		 2906 if (recover_curseg) {
2907 if (old_cursegno != curseg->segno) {
2908 curseg->next_segno = old_cursegno;
025d63a4 2909 change_curseg(sbi, type);
19f106bc
CY		 2910 }
2911 curseg->next_blkoff = old_blkoff;
2912 }
2913
3d26fa6b 2914 up_write(&sit_i->sentry_lock);
351df4b2 2915 mutex_unlock(&curseg->curseg_mutex);
2b60311d 2916 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 2917}
2918
528e3459
CY		 2919void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2920 block_t old_addr, block_t new_addr,
28bc106b
CY		 2921 unsigned char version, bool recover_curseg,
2922 bool recover_newaddr)
528e3459
CY		 2923{
2924 struct f2fs_summary sum;
2925
2926 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2927
28bc106b
CY		 2928 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2929 recover_curseg, recover_newaddr);
528e3459 2930
f28b3434 2931 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2932}
2933
93dfe2ac 2934void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2935 enum page_type type, bool ordered)
93dfe2ac 2936{
93dfe2ac 2937 if (PageWriteback(page)) {
4081363f
JK		 2938 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2939
b9109b0e
JK		 2940 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2941 0, page->index, type);
fec1d657
JK		 2942 if (ordered)
2943 wait_on_page_writeback(page);
2944 else
2945 wait_for_stable_page(page);
93dfe2ac
JK		 2946 }
2947}
2948
d4c759ee 2949void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2950{
2951 struct page *cpage;
2952
5d4c0af4 2953 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2954 return;
2955
08b39fbd
CY		 2956 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2957 if (cpage) {
fec1d657 2958 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2959 f2fs_put_page(cpage, 1);
2960 }
2961}
2962
c376fc0f 2963static void read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 2964{
2965 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2966 struct curseg_info *seg_i;
2967 unsigned char *kaddr;
2968 struct page *page;
2969 block_t start;
2970 int i, j, offset;
2971
2972 start = start_sum_block(sbi);
2973
2974 page = get_meta_page(sbi, start++);
2975 kaddr = (unsigned char *)page_address(page);
2976
2977 /* Step 1: restore nat cache */
2978 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2979 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2980
2981 /* Step 2: restore sit cache */
2982 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2983 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2984 offset = 2 * SUM_JOURNAL_SIZE;
2985
2986 /* Step 3: restore summary entries */
2987 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2988 unsigned short blk_off;
2989 unsigned int segno;
2990
2991 seg_i = CURSEG_I(sbi, i);
2992 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2993 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2994 seg_i->next_segno = segno;
2995 reset_curseg(sbi, i, 0);
2996 seg_i->alloc_type = ckpt->alloc_type[i];
2997 seg_i->next_blkoff = blk_off;
2998
2999 if (seg_i->alloc_type == SSR)
3000 blk_off = sbi->blocks_per_seg;
3001
3002 for (j = 0; j < blk_off; j++) {
3003 struct f2fs_summary *s;
3004 s = (struct f2fs_summary *)(kaddr + offset);
3005 seg_i->sum_blk->entries[j] = *s;
3006 offset += SUMMARY_SIZE;
09cbfeaf 3007 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3008 SUM_FOOTER_SIZE)
3009 continue;
3010
3011 f2fs_put_page(page, 1);
3012 page = NULL;
3013
3014 page = get_meta_page(sbi, start++);
3015 kaddr = (unsigned char *)page_address(page);
3016 offset = 0;
3017 }
3018 }
3019 f2fs_put_page(page, 1);
351df4b2
JK		 3020}
3021
3022static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3023{
3024 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3025 struct f2fs_summary_block *sum;
3026 struct curseg_info *curseg;
3027 struct page *new;
3028 unsigned short blk_off;
3029 unsigned int segno = 0;
3030 block_t blk_addr = 0;
3031
3032 /* get segment number and block addr */
3033 if (IS_DATASEG(type)) {
3034 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3035 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3036 CURSEG_HOT_DATA]);
119ee914 3037 if (__exist_node_summaries(sbi))
351df4b2
JK		 3038 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3039 else
3040 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3041 } else {
3042 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3043 CURSEG_HOT_NODE]);
3044 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3045 CURSEG_HOT_NODE]);
119ee914 3046 if (__exist_node_summaries(sbi))
351df4b2
JK		 3047 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3048 type - CURSEG_HOT_NODE);
3049 else
3050 blk_addr = GET_SUM_BLOCK(sbi, segno);
3051 }
3052
3053 new = get_meta_page(sbi, blk_addr);
3054 sum = (struct f2fs_summary_block *)page_address(new);
3055
3056 if (IS_NODESEG(type)) {
119ee914 3057 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3058 struct f2fs_summary *ns = &sum->entries[0];
3059 int i;
3060 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3061 ns->version = 0;
3062 ns->ofs_in_node = 0;
3063 }
3064 } else {
c376fc0f 3065 restore_node_summary(sbi, segno, sum);
351df4b2
JK		 3066 }
3067 }
3068
3069 /* set uncompleted segment to curseg */
3070 curseg = CURSEG_I(sbi, type);
3071 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3072
3073 /* update journal info */
3074 down_write(&curseg->journal_rwsem);
3075 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3076 up_write(&curseg->journal_rwsem);
3077
3078 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3079 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3080 curseg->next_segno = segno;
3081 reset_curseg(sbi, type, 0);
3082 curseg->alloc_type = ckpt->alloc_type[type];
3083 curseg->next_blkoff = blk_off;
3084 mutex_unlock(&curseg->curseg_mutex);
3085 f2fs_put_page(new, 1);
3086 return 0;
3087}
3088
3089static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3090{
21d3f8e1
JQ		 3091 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3092 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3093 int type = CURSEG_HOT_DATA;
e4fc5fbf 3094 int err;
351df4b2 3095
aaec2b1d 3096 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 3097 int npages = npages_for_summary_flush(sbi, true);
3098
3099 if (npages >= 2)
3100 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3101 META_CP, true);
3fa06d7b 3102
351df4b2 3103 /* restore for compacted data summary */
c376fc0f 3104 read_compacted_summaries(sbi);
351df4b2
JK		 3105 type = CURSEG_HOT_NODE;
3106 }
3107
119ee914 3108 if (__exist_node_summaries(sbi))
3fa06d7b 3109 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3110 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3111
e4fc5fbf
CY		 3112 for (; type <= CURSEG_COLD_NODE; type++) {
3113 err = read_normal_summaries(sbi, type);
3114 if (err)
3115 return err;
3116 }
3117
21d3f8e1
JQ		 3118 /* sanity check for summary blocks */
3119 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3120 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3121 return -EINVAL;
3122
351df4b2
JK		 3123 return 0;
3124}
3125
3126static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3127{
3128 struct page *page;
3129 unsigned char *kaddr;
3130 struct f2fs_summary *summary;
3131 struct curseg_info *seg_i;
3132 int written_size = 0;
3133 int i, j;
3134
3135 page = grab_meta_page(sbi, blkaddr++);
3136 kaddr = (unsigned char *)page_address(page);
81114baa 3137 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3138
3139 /* Step 1: write nat cache */
3140 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3141 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3142 written_size += SUM_JOURNAL_SIZE;
3143
3144 /* Step 2: write sit cache */
3145 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3146 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3147 written_size += SUM_JOURNAL_SIZE;
3148
351df4b2
JK		 3149 /* Step 3: write summary entries */
3150 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3151 unsigned short blkoff;
3152 seg_i = CURSEG_I(sbi, i);
3153 if (sbi->ckpt->alloc_type[i] == SSR)
3154 blkoff = sbi->blocks_per_seg;
3155 else
3156 blkoff = curseg_blkoff(sbi, i);
3157
3158 for (j = 0; j < blkoff; j++) {
3159 if (!page) {
3160 page = grab_meta_page(sbi, blkaddr++);
3161 kaddr = (unsigned char *)page_address(page);
81114baa 3162 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3163 written_size = 0;
3164 }
3165 summary = (struct f2fs_summary *)(kaddr + written_size);
3166 *summary = seg_i->sum_blk->entries[j];
3167 written_size += SUMMARY_SIZE;
351df4b2 3168
09cbfeaf 3169 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3170 SUM_FOOTER_SIZE)
3171 continue;
3172
e8d61a74 3173 set_page_dirty(page);
351df4b2
JK		 3174 f2fs_put_page(page, 1);
3175 page = NULL;
3176 }
3177 }
e8d61a74
CY		 3178 if (page) {
3179 set_page_dirty(page);
351df4b2 3180 f2fs_put_page(page, 1);
e8d61a74 3181 }
351df4b2
JK		 3182}
3183
3184static void write_normal_summaries(struct f2fs_sb_info *sbi,
3185 block_t blkaddr, int type)
3186{
3187 int i, end;
3188 if (IS_DATASEG(type))
3189 end = type + NR_CURSEG_DATA_TYPE;
3190 else
3191 end = type + NR_CURSEG_NODE_TYPE;
3192
b7ad7512
CY		 3193 for (i = type; i < end; i++)
3194 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3195}
3196
3197void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3198{
aaec2b1d 3199 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3200 write_compacted_summaries(sbi, start_blk);
3201 else
3202 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3203}
3204
3205void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3206{
119ee914 3207 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3208}
3209
dfc08a12 3210int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3211 unsigned int val, int alloc)
3212{
3213 int i;
3214
3215 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3216 for (i = 0; i < nats_in_cursum(journal); i++) {
3217 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3218 return i;
3219 }
dfc08a12
CY		 3220 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3221 return update_nats_in_cursum(journal, 1);
351df4b2 3222 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3223 for (i = 0; i < sits_in_cursum(journal); i++)
3224 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3225 return i;
dfc08a12
CY		 3226 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3227 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3228 }
3229 return -1;
3230}
3231
3232static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3233 unsigned int segno)
3234{
2cc22186 3235 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3236}
3237
3238static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3239 unsigned int start)
3240{
3241 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3242 struct page *page;
351df4b2 3243 pgoff_t src_off, dst_off;
351df4b2
JK		 3244
3245 src_off = current_sit_addr(sbi, start);
3246 dst_off = next_sit_addr(sbi, src_off);
3247
068c3cd8
YH		 3248 page = grab_meta_page(sbi, dst_off);
3249 seg_info_to_sit_page(sbi, page, start);
351df4b2 3250
068c3cd8 3251 set_page_dirty(page);
351df4b2
JK		 3252 set_to_next_sit(sit_i, start);
3253
068c3cd8 3254 return page;
351df4b2
JK		 3255}
3256
184a5cd2
CY		 3257static struct sit_entry_set *grab_sit_entry_set(void)
3258{
3259 struct sit_entry_set *ses =
80c54505 3260 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3261
3262 ses->entry_cnt = 0;
3263 INIT_LIST_HEAD(&ses->set_list);
3264 return ses;
3265}
3266
3267static void release_sit_entry_set(struct sit_entry_set *ses)
3268{
3269 list_del(&ses->set_list);
3270 kmem_cache_free(sit_entry_set_slab, ses);
3271}
3272
3273static void adjust_sit_entry_set(struct sit_entry_set *ses,
3274 struct list_head *head)
3275{
3276 struct sit_entry_set *next = ses;
3277
3278 if (list_is_last(&ses->set_list, head))
3279 return;
3280
3281 list_for_each_entry_continue(next, head, set_list)
3282 if (ses->entry_cnt <= next->entry_cnt)
3283 break;
3284
3285 list_move_tail(&ses->set_list, &next->set_list);
3286}
3287
3288static void add_sit_entry(unsigned int segno, struct list_head *head)
3289{
3290 struct sit_entry_set *ses;
3291 unsigned int start_segno = START_SEGNO(segno);
3292
3293 list_for_each_entry(ses, head, set_list) {
3294 if (ses->start_segno == start_segno) {
3295 ses->entry_cnt++;
3296 adjust_sit_entry_set(ses, head);
3297 return;
3298 }
3299 }
3300
3301 ses = grab_sit_entry_set();
3302
3303 ses->start_segno = start_segno;
3304 ses->entry_cnt++;
3305 list_add(&ses->set_list, head);
3306}
3307
3308static void add_sits_in_set(struct f2fs_sb_info *sbi)
3309{
3310 struct f2fs_sm_info *sm_info = SM_I(sbi);
3311 struct list_head *set_list = &sm_info->sit_entry_set;
3312 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3313 unsigned int segno;
3314
7cd8558b 3315 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3316 add_sit_entry(segno, set_list);
3317}
3318
3319static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3320{
3321 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3322 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3323 int i;
3324
b7ad7512 3325 down_write(&curseg->journal_rwsem);
dfc08a12 3326 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3327 unsigned int segno;
3328 bool dirtied;
3329
dfc08a12 3330 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3331 dirtied = __mark_sit_entry_dirty(sbi, segno);
3332
3333 if (!dirtied)
3334 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3335 }
dfc08a12 3336 update_sits_in_cursum(journal, -i);
b7ad7512 3337 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3338}
3339
0a8165d7 3340/*
351df4b2
JK		 3341 * CP calls this function, which flushes SIT entries including sit_journal,
3342 * and moves prefree segs to free segs.
3343 */
4b2fecc8 3344void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3345{
3346 struct sit_info *sit_i = SIT_I(sbi);
3347 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3348 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3349 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3350 struct sit_entry_set *ses, *tmp;
3351 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3352 bool to_journal = true;
4b2fecc8 3353 struct seg_entry *se;
351df4b2 3354
3d26fa6b 3355 down_write(&sit_i->sentry_lock);
351df4b2 3356
2b11a74b
WL		 3357 if (!sit_i->dirty_sentries)
3358 goto out;
3359
351df4b2 3360 /*
184a5cd2
CY		 3361 * add and account sit entries of dirty bitmap in sit entry
3362 * set temporarily
351df4b2 3363 */
184a5cd2 3364 add_sits_in_set(sbi);
351df4b2 3365
184a5cd2
CY		 3366 /*
3367 * if there are no enough space in journal to store dirty sit
3368 * entries, remove all entries from journal and add and account
3369 * them in sit entry set.
3370 */
dfc08a12 3371 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3372 remove_sits_in_journal(sbi);
b2955550 3373
184a5cd2
CY		 3374 /*
3375 * there are two steps to flush sit entries:
3376 * #1, flush sit entries to journal in current cold data summary block.
3377 * #2, flush sit entries to sit page.
3378 */
3379 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3380 struct page *page = NULL;
184a5cd2
CY		 3381 struct f2fs_sit_block *raw_sit = NULL;
3382 unsigned int start_segno = ses->start_segno;
3383 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3384 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3385 unsigned int segno = start_segno;
3386
3387 if (to_journal &&
dfc08a12 3388 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3389 to_journal = false;
3390
b7ad7512
CY		 3391 if (to_journal) {
3392 down_write(&curseg->journal_rwsem);
3393 } else {
184a5cd2
CY		 3394 page = get_next_sit_page(sbi, start_segno);
3395 raw_sit = page_address(page);
351df4b2 3396 }
351df4b2 3397
184a5cd2
CY		 3398 /* flush dirty sit entries in region of current sit set */
3399 for_each_set_bit_from(segno, bitmap, end) {
3400 int offset, sit_offset;
4b2fecc8
JK		 3401
3402 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 3403
3404 /* add discard candidates */
c473f1a9 3405 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3406 cpc->trim_start = segno;
25290fa5 3407 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3408 }
184a5cd2
CY		 3409
3410 if (to_journal) {
dfc08a12 3411 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3412 SIT_JOURNAL, segno, 1);
3413 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3414 segno_in_journal(journal, offset) =
184a5cd2
CY		 3415 cpu_to_le32(segno);
3416 seg_info_to_raw_sit(se,
dfc08a12 3417 &sit_in_journal(journal, offset));
184a5cd2
CY		 3418 } else {
3419 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3420 seg_info_to_raw_sit(se,
3421 &raw_sit->entries[sit_offset]);
3422 }
351df4b2 3423
184a5cd2
CY		 3424 __clear_bit(segno, bitmap);
3425 sit_i->dirty_sentries--;
3426 ses->entry_cnt--;
351df4b2
JK		 3427 }
3428
b7ad7512
CY		 3429 if (to_journal)
3430 up_write(&curseg->journal_rwsem);
3431 else
184a5cd2
CY		 3432 f2fs_put_page(page, 1);
3433
3434 f2fs_bug_on(sbi, ses->entry_cnt);
3435 release_sit_entry_set(ses);
351df4b2 3436 }
184a5cd2
CY		 3437
3438 f2fs_bug_on(sbi, !list_empty(head));
3439 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3440out:
c473f1a9 3441 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3442 __u64 trim_start = cpc->trim_start;
3443
4b2fecc8 3444 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3445 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3446
3447 cpc->trim_start = trim_start;
4b2fecc8 3448 }
3d26fa6b 3449 up_write(&sit_i->sentry_lock);
351df4b2 3450
351df4b2
JK		 3451 set_prefree_as_free_segments(sbi);
3452}
3453
3454static int build_sit_info(struct f2fs_sb_info *sbi)
3455{
3456 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3457 struct sit_info *sit_i;
3458 unsigned int sit_segs, start;
ae27d62e 3459 char *src_bitmap;
351df4b2
JK		 3460 unsigned int bitmap_size;
3461
3462 /* allocate memory for SIT information */
acbf054d 3463 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3464 if (!sit_i)
3465 return -ENOMEM;
3466
3467 SM_I(sbi)->sit_info = sit_i;
3468
628b3d14 3469 sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
39307a8e 3470 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3471 if (!sit_i->sentries)
3472 return -ENOMEM;
3473
7cd8558b 3474 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3475 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3476 GFP_KERNEL);
351df4b2
JK		 3477 if (!sit_i->dirty_sentries_bitmap)
3478 return -ENOMEM;
3479
7cd8558b 3480 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3481 sit_i->sentries[start].cur_valid_map
acbf054d 3482 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3483 sit_i->sentries[start].ckpt_valid_map
acbf054d 3484 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3485 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3486 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3487 return -ENOMEM;
3e025740 3488
355e7891
CY		 3489#ifdef CONFIG_F2FS_CHECK_FS
3490 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3491 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3492 if (!sit_i->sentries[start].cur_valid_map_mir)
3493 return -ENOMEM;
3494#endif
3495
3e025740
JK		 3496 if (f2fs_discard_en(sbi)) {
3497 sit_i->sentries[start].discard_map
acbf054d
CY		 3498 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3499 GFP_KERNEL);
3e025740
JK		 3500 if (!sit_i->sentries[start].discard_map)
3501 return -ENOMEM;
3502 }
351df4b2
JK		 3503 }
3504
acbf054d 3505 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3506 if (!sit_i->tmp_map)
3507 return -ENOMEM;
3508
351df4b2 3509 if (sbi->segs_per_sec > 1) {
628b3d14 3510 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
39307a8e 3511 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3512 if (!sit_i->sec_entries)
3513 return -ENOMEM;
3514 }
3515
3516 /* get information related with SIT */
3517 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3518
3519 /* setup SIT bitmap from ckeckpoint pack */
3520 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3521 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3522
ae27d62e
CY		 3523 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3524 if (!sit_i->sit_bitmap)
351df4b2 3525 return -ENOMEM;
351df4b2 3526
ae27d62e
CY		 3527#ifdef CONFIG_F2FS_CHECK_FS
3528 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3529 if (!sit_i->sit_bitmap_mir)
3530 return -ENOMEM;
3531#endif
3532
351df4b2
JK		 3533 /* init SIT information */
3534 sit_i->s_ops = &default_salloc_ops;
3535
3536 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3537 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3538 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3539 sit_i->bitmap_size = bitmap_size;
3540 sit_i->dirty_sentries = 0;
3541 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3542 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3543 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3544 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3545 return 0;
3546}
3547
3548static int build_free_segmap(struct f2fs_sb_info *sbi)
3549{
351df4b2
JK		 3550 struct free_segmap_info *free_i;
3551 unsigned int bitmap_size, sec_bitmap_size;
3552
3553 /* allocate memory for free segmap information */
acbf054d 3554 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3555 if (!free_i)
3556 return -ENOMEM;
3557
3558 SM_I(sbi)->free_info = free_i;
3559
7cd8558b 3560 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3561 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3562 if (!free_i->free_segmap)
3563 return -ENOMEM;
3564
7cd8558b 3565 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3566 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3567 if (!free_i->free_secmap)
3568 return -ENOMEM;
3569
3570 /* set all segments as dirty temporarily */
3571 memset(free_i->free_segmap, 0xff, bitmap_size);
3572 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3573
3574 /* init free segmap information */
7cd8558b 3575 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3576 free_i->free_segments = 0;
3577 free_i->free_sections = 0;
1a118ccf 3578 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3579 return 0;
3580}
3581
3582static int build_curseg(struct f2fs_sb_info *sbi)
3583{
1042d60f 3584 struct curseg_info *array;
351df4b2
JK		 3585 int i;
3586
4e6aad29 3587 array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
351df4b2
JK		 3588 if (!array)
3589 return -ENOMEM;
3590
3591 SM_I(sbi)->curseg_array = array;
3592
3593 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3594 mutex_init(&array[i].curseg_mutex);
acbf054d 3595 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3596 if (!array[i].sum_blk)
3597 return -ENOMEM;
b7ad7512 3598 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 3599 array[i].journal = f2fs_kzalloc(sbi,
3600 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 3601 if (!array[i].journal)
3602 return -ENOMEM;
351df4b2
JK		 3603 array[i].segno = NULL_SEGNO;
3604 array[i].next_blkoff = 0;
3605 }
3606 return restore_curseg_summaries(sbi);
3607}
3608
c39a1b34 3609static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3610{
3611 struct sit_info *sit_i = SIT_I(sbi);
3612 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3613 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3614 struct seg_entry *se;
3615 struct f2fs_sit_entry sit;
74de593a
CY		 3616 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3617 unsigned int i, start, end;
3618 unsigned int readed, start_blk = 0;
c39a1b34 3619 int err = 0;
351df4b2 3620
74de593a 3621 do {
664ba972
JK		 3622 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3623 META_SIT, true);
74de593a
CY		 3624
3625 start = start_blk * sit_i->sents_per_block;
3626 end = (start_blk + readed) * sit_i->sents_per_block;
3627
7cd8558b 3628 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3629 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3630 struct page *page;
3631
9c094040 3632 se = &sit_i->sentries[start];
74de593a
CY		 3633 page = get_current_sit_page(sbi, start);
3634 sit_blk = (struct f2fs_sit_block *)page_address(page);
3635 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3636 f2fs_put_page(page, 1);
d600af23 3637
c39a1b34
JK		 3638 err = check_block_count(sbi, start, &sit);
3639 if (err)
3640 return err;
74de593a 3641 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3642
3643 /* build discard map only one time */
3e025740 3644 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3645 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3646 memset(se->discard_map, 0xff,
3647 SIT_VBLOCK_MAP_SIZE);
3648 } else {
3649 memcpy(se->discard_map,
3650 se->cur_valid_map,
3651 SIT_VBLOCK_MAP_SIZE);
3652 sbi->discard_blks +=
3653 sbi->blocks_per_seg -
3654 se->valid_blocks;
3655 }
3e025740 3656 }
a66cdd98 3657
d600af23
CY		 3658 if (sbi->segs_per_sec > 1)
3659 get_sec_entry(sbi, start)->valid_blocks +=
3660 se->valid_blocks;
351df4b2 3661 }
74de593a
CY		 3662 start_blk += readed;
3663 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3664
3665 down_read(&curseg->journal_rwsem);
3666 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3667 unsigned int old_valid_blocks;
3668
3669 start = le32_to_cpu(segno_in_journal(journal, i));
3670 se = &sit_i->sentries[start];
3671 sit = sit_in_journal(journal, i);
3672
3673 old_valid_blocks = se->valid_blocks;
3674
c39a1b34
JK		 3675 err = check_block_count(sbi, start, &sit);
3676 if (err)
3677 break;
d600af23
CY		 3678 seg_info_from_raw_sit(se, &sit);
3679
3680 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3681 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3682 memset(se->discard_map, 0xff,
3683 SIT_VBLOCK_MAP_SIZE);
3684 } else {
3685 memcpy(se->discard_map, se->cur_valid_map,
3686 SIT_VBLOCK_MAP_SIZE);
3687 sbi->discard_blks += old_valid_blocks -
3688 se->valid_blocks;
3689 }
d600af23
CY		 3690 }
3691
3692 if (sbi->segs_per_sec > 1)
3693 get_sec_entry(sbi, start)->valid_blocks +=
3694 se->valid_blocks - old_valid_blocks;
3695 }
3696 up_read(&curseg->journal_rwsem);
c39a1b34 3697 return err;
351df4b2
JK		 3698}
3699
3700static void init_free_segmap(struct f2fs_sb_info *sbi)
3701{
3702 unsigned int start;
3703 int type;
3704
7cd8558b 3705 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3706 struct seg_entry *sentry = get_seg_entry(sbi, start);
3707 if (!sentry->valid_blocks)
3708 __set_free(sbi, start);
c79b7ff1
JK		 3709 else
3710 SIT_I(sbi)->written_valid_blocks +=
3711 sentry->valid_blocks;
351df4b2
JK		 3712 }
3713
3714 /* set use the current segments */
3715 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3716 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3717 __set_test_and_inuse(sbi, curseg_t->segno);
3718 }
3719}
3720
3721static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3722{
3723 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3724 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3725 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3726 unsigned short valid_blocks;
3727
8736fbf0 3728 while (1) {
351df4b2 3729 /* find dirty segment based on free segmap */
7cd8558b
JK		 3730 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3731 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3732 break;
3733 offset = segno + 1;
302bd348 3734 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3735 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3736 continue;
ec325b52
JK		 3737 if (valid_blocks > sbi->blocks_per_seg) {
3738 f2fs_bug_on(sbi, 1);
3739 continue;
3740 }
351df4b2
JK		 3741 mutex_lock(&dirty_i->seglist_lock);
3742 __locate_dirty_segment(sbi, segno, DIRTY);
3743 mutex_unlock(&dirty_i->seglist_lock);
3744 }
3745}
3746
5ec4e49f 3747static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3748{
3749 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3750 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3751
628b3d14 3752 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 3753 if (!dirty_i->victim_secmap)
351df4b2
JK		 3754 return -ENOMEM;
3755 return 0;
3756}
3757
3758static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3759{
3760 struct dirty_seglist_info *dirty_i;
3761 unsigned int bitmap_size, i;
3762
3763 /* allocate memory for dirty segments list information */
acbf054d
CY		 3764 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3765 GFP_KERNEL);
351df4b2
JK		 3766 if (!dirty_i)
3767 return -ENOMEM;
3768
3769 SM_I(sbi)->dirty_info = dirty_i;
3770 mutex_init(&dirty_i->seglist_lock);
3771
7cd8558b 3772 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3773
3774 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 3775 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3776 GFP_KERNEL);
351df4b2
JK		 3777 if (!dirty_i->dirty_segmap[i])
3778 return -ENOMEM;
3779 }
3780
3781 init_dirty_segmap(sbi);
5ec4e49f 3782 return init_victim_secmap(sbi);
351df4b2
JK		 3783}
3784
0a8165d7 3785/*
351df4b2
JK		 3786 * Update min, max modified time for cost-benefit GC algorithm
3787 */
3788static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3789{
3790 struct sit_info *sit_i = SIT_I(sbi);
3791 unsigned int segno;
3792
3d26fa6b 3793 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3794
3795 sit_i->min_mtime = LLONG_MAX;
3796
7cd8558b 3797 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3798 unsigned int i;
3799 unsigned long long mtime = 0;
3800
3801 for (i = 0; i < sbi->segs_per_sec; i++)
3802 mtime += get_seg_entry(sbi, segno + i)->mtime;
3803
3804 mtime = div_u64(mtime, sbi->segs_per_sec);
3805
3806 if (sit_i->min_mtime > mtime)
3807 sit_i->min_mtime = mtime;
3808 }
3809 sit_i->max_mtime = get_mtime(sbi);
3d26fa6b 3810 up_write(&sit_i->sentry_lock);
351df4b2
JK		 3811}
3812
3813int build_segment_manager(struct f2fs_sb_info *sbi)
3814{
3815 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3816 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3817 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3818 int err;
3819
acbf054d 3820 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 3821 if (!sm_info)
3822 return -ENOMEM;
3823
3824 /* init sm info */
3825 sbi->sm_info = sm_info;
351df4b2
JK		 3826 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3827 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3828 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3829 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3830 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3831 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3832 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3833 sm_info->rec_prefree_segments = sm_info->main_segments *
3834 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3835 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3836 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3837
52763a4b
JK		 3838 if (!test_opt(sbi, LFS))
3839 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3840 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3841 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3842 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 3843 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 3844
184a5cd2
CY		 3845 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3846
2b60311d
CY		 3847 init_rwsem(&sm_info->curseg_lock);
3848
d4fdf8ba 3849 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3850 err = create_flush_cmd_control(sbi);
3851 if (err)
a688b9d9 3852 return err;
6b4afdd7
JK		 3853 }
3854
0b54fb84
JK		 3855 err = create_discard_cmd_control(sbi);
3856 if (err)
3857 return err;
3858
351df4b2
JK		 3859 err = build_sit_info(sbi);
3860 if (err)
3861 return err;
3862 err = build_free_segmap(sbi);
3863 if (err)
3864 return err;
3865 err = build_curseg(sbi);
3866 if (err)
3867 return err;
3868
3869 /* reinit free segmap based on SIT */
c39a1b34
JK		 3870 err = build_sit_entries(sbi);
3871 if (err)
3872 return err;
351df4b2
JK		 3873
3874 init_free_segmap(sbi);
3875 err = build_dirty_segmap(sbi);
3876 if (err)
3877 return err;
3878
3879 init_min_max_mtime(sbi);
3880 return 0;
3881}
3882
3883static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3884 enum dirty_type dirty_type)
3885{
3886 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3887
3888 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3889 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3890 dirty_i->nr_dirty[dirty_type] = 0;
3891 mutex_unlock(&dirty_i->seglist_lock);
3892}
3893
5ec4e49f 3894static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3895{
3896 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3897 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3898}
3899
3900static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3901{
3902 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3903 int i;
3904
3905 if (!dirty_i)
3906 return;
3907
3908 /* discard pre-free/dirty segments list */
3909 for (i = 0; i < NR_DIRTY_TYPE; i++)
3910 discard_dirty_segmap(sbi, i);
3911
5ec4e49f 3912 destroy_victim_secmap(sbi);
351df4b2
JK		 3913 SM_I(sbi)->dirty_info = NULL;
3914 kfree(dirty_i);
3915}
3916
3917static void destroy_curseg(struct f2fs_sb_info *sbi)
3918{
3919 struct curseg_info *array = SM_I(sbi)->curseg_array;
3920 int i;
3921
3922 if (!array)
3923 return;
3924 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3925 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3926 kfree(array[i].sum_blk);
b7ad7512
CY		 3927 kfree(array[i].journal);
3928 }
351df4b2
JK		 3929 kfree(array);
3930}
3931
3932static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3933{
3934 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3935 if (!free_i)
3936 return;
3937 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3938 kvfree(free_i->free_segmap);
3939 kvfree(free_i->free_secmap);
351df4b2
JK		 3940 kfree(free_i);
3941}
3942
3943static void destroy_sit_info(struct f2fs_sb_info *sbi)
3944{
3945 struct sit_info *sit_i = SIT_I(sbi);
3946 unsigned int start;
3947
3948 if (!sit_i)
3949 return;
3950
3951 if (sit_i->sentries) {
7cd8558b 3952 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3953 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3954#ifdef CONFIG_F2FS_CHECK_FS
3955 kfree(sit_i->sentries[start].cur_valid_map_mir);
3956#endif
351df4b2 3957 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3958 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3959 }
3960 }
60a3b782
JK		 3961 kfree(sit_i->tmp_map);
3962
39307a8e
JK		 3963 kvfree(sit_i->sentries);
3964 kvfree(sit_i->sec_entries);
3965 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3966
3967 SM_I(sbi)->sit_info = NULL;
3968 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3969#ifdef CONFIG_F2FS_CHECK_FS
3970 kfree(sit_i->sit_bitmap_mir);
3971#endif
351df4b2
JK		 3972 kfree(sit_i);
3973}
3974
3975void destroy_segment_manager(struct f2fs_sb_info *sbi)
3976{
3977 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3978
3b03f724
CY		 3979 if (!sm_info)
3980 return;
5eba8c5d 3981 destroy_flush_cmd_control(sbi, true);
f099405f 3982 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3983 destroy_dirty_segmap(sbi);
3984 destroy_curseg(sbi);
3985 destroy_free_segmap(sbi);
3986 destroy_sit_info(sbi);
3987 sbi->sm_info = NULL;
3988 kfree(sm_info);
3989}
7fd9e544
JK		 3990
3991int __init create_segment_manager_caches(void)
3992{
3993 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3994 sizeof(struct discard_entry));
7fd9e544 3995 if (!discard_entry_slab)
184a5cd2
CY		 3996 goto fail;
3997
b01a9201
JK		 3998 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3999 sizeof(struct discard_cmd));
4000 if (!discard_cmd_slab)
6ab2a308 4001 goto destroy_discard_entry;
275b66b0 4002
184a5cd2 4003 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4004 sizeof(struct sit_entry_set));
184a5cd2 4005 if (!sit_entry_set_slab)
b01a9201 4006 goto destroy_discard_cmd;
88b88a66
JK		 4007
4008 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4009 sizeof(struct inmem_pages));
4010 if (!inmem_entry_slab)
4011 goto destroy_sit_entry_set;
7fd9e544 4012 return 0;
184a5cd2 4013
88b88a66
JK		 4014destroy_sit_entry_set:
4015 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4016destroy_discard_cmd:
4017 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4018destroy_discard_entry:
184a5cd2
CY		 4019 kmem_cache_destroy(discard_entry_slab);
4020fail:
4021 return -ENOMEM;
7fd9e544
JK		 4022}
4023
4024void destroy_segment_manager_caches(void)
4025{
184a5cd2 4026 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4027 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4028 kmem_cache_destroy(discard_entry_slab);
88b88a66 4029 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4030}
Ubuntu Jammy Linux kernel mirror