projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: avoid get_valid_blocks() for cleanup
[linux-2.6-block.git] / fs / f2fs / segment.c
CommitLineData
7c1a000d 1// SPDX-License-Identifier: GPL-2.0
0a8165d7 2/*
351df4b2
JK		 3 * fs/f2fs/segment.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
351df4b2
JK		 7 */
8#include <linux/fs.h>
9#include <linux/f2fs_fs.h>
10#include <linux/bio.h>
11#include <linux/blkdev.h>
690e4a3e 12#include <linux/prefetch.h>
6b4afdd7 13#include <linux/kthread.h>
74de593a 14#include <linux/swap.h>
60b99b48 15#include <linux/timer.h>
1d7be270 16#include <linux/freezer.h>
1eb1ef4a 17#include <linux/sched/signal.h>
351df4b2
JK		 18
19#include "f2fs.h"
20#include "segment.h"
21#include "node.h"
5f656541 22#include "gc.h"
9e4ded3f 23#include "trace.h"
6ec178da 24#include <trace/events/f2fs.h>
351df4b2 25
9a7f143a
CL		 26#define __reverse_ffz(x) __reverse_ffs(~(x))
27
7fd9e544 28static struct kmem_cache *discard_entry_slab;
b01a9201 29static struct kmem_cache *discard_cmd_slab;
184a5cd2 30static struct kmem_cache *sit_entry_set_slab;
88b88a66 31static struct kmem_cache *inmem_entry_slab;
7fd9e544 32
f96999c3
JK		 33static unsigned long __reverse_ulong(unsigned char *str)
34{
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38#if BITS_PER_LONG == 64
39 shift = 56;
40#endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46}
47
9a7f143a
CL		 48/*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52static inline unsigned long __reverse_ffs(unsigned long word)
53{
54 int num = 0;
55
56#if BITS_PER_LONG == 64
f96999c3 57 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 58 num += 32;
f96999c3 59 else
9a7f143a 60 word >>= 32;
9a7f143a 61#endif
f96999c3 62 if ((word & 0xffff0000) == 0)
9a7f143a 63 num += 16;
f96999c3 64 else
9a7f143a 65 word >>= 16;
f96999c3
JK		 66
67 if ((word & 0xff00) == 0)
9a7f143a 68 num += 8;
f96999c3 69 else
9a7f143a 70 word >>= 8;
f96999c3 71
9a7f143a
CL		 72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
f96999c3 76
9a7f143a
CL		 77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
f96999c3 81
9a7f143a
CL		 82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85}
86
87/*
e1c42045 88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 90 * @size must be integral times of unsigned long.
9a7f143a 91 * Example:
f96999c3
JK		 92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 95 */
96static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98{
99 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 100 unsigned long result = size;
9a7f143a 101 unsigned long tmp;
9a7f143a
CL		 102
103 if (offset >= size)
104 return size;
105
692223d1 106 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 107 offset %= BITS_PER_LONG;
f96999c3 108
692223d1
FL		 109 while (1) {
110 if (*p == 0)
111 goto pass;
9a7f143a 112
f96999c3 113 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 118 if (tmp)
692223d1
FL		 119 goto found;
120pass:
121 if (size <= BITS_PER_LONG)
122 break;
9a7f143a 123 size -= BITS_PER_LONG;
692223d1 124 offset = 0;
f96999c3 125 p++;
9a7f143a 126 }
692223d1
FL		 127 return result;
128found:
129 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 130}
131
132static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134{
135 const unsigned long *p = addr + BIT_WORD(offset);
80609448 136 unsigned long result = size;
9a7f143a 137 unsigned long tmp;
9a7f143a
CL		 138
139 if (offset >= size)
140 return size;
141
80609448 142 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 143 offset %= BITS_PER_LONG;
80609448
JK		 144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
f96999c3 149 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
f96999c3 155 if (tmp != ~0UL)
80609448
JK		 156 goto found;
157pass:
158 if (size <= BITS_PER_LONG)
159 break;
9a7f143a 160 size -= BITS_PER_LONG;
80609448 161 offset = 0;
f96999c3 162 p++;
9a7f143a 163 }
80609448
JK		 164 return result;
165found:
166 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 167}
168
4d57b86d 169bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
b3a97a2a
JK		 170{
171 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
174
175 if (test_opt(sbi, LFS))
176 return false;
5b0e9539 177 if (sbi->gc_mode == GC_URGENT)
b3a97a2a 178 return true;
4354994f
DR		 179 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
180 return true;
b3a97a2a
JK		 181
182 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
a2a12b67 183 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
b3a97a2a
JK		 184}
185
4d57b86d 186void f2fs_register_inmem_page(struct inode *inode, struct page *page)
88b88a66 187{
57864ae5 188 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
88b88a66
JK		 189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
9be32d72 191
9e4ded3f 192 f2fs_trace_pid(page);
0722b101 193
240a5915 194 f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
decd36b6 195
88b88a66
JK		 196 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
197
198 /* add atomic page indices to the list */
199 new->page = page;
200 INIT_LIST_HEAD(&new->list);
decd36b6 201
88b88a66
JK		 202 /* increase reference count with clean state */
203 mutex_lock(&fi->inmem_lock);
204 get_page(page);
205 list_add_tail(&new->list, &fi->inmem_pages);
57864ae5
JK		 206 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
207 if (list_empty(&fi->inmem_ilist))
208 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
209 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
8dcf2ff7 210 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 211 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 212
213 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 214}
215
28bc106b 216static int __revoke_inmem_pages(struct inode *inode,
48432984
CY		 217 struct list_head *head, bool drop, bool recover,
218 bool trylock)
29b96b54 219{
28bc106b 220 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 221 struct inmem_pages *cur, *tmp;
28bc106b 222 int err = 0;
29b96b54
CY		 223
224 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 225 struct page *page = cur->page;
226
227 if (drop)
228 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
229
48432984
CY		 230 if (trylock) {
231 /*
232 * to avoid deadlock in between page lock and
233 * inmem_lock.
234 */
235 if (!trylock_page(page))
236 continue;
237 } else {
238 lock_page(page);
239 }
29b96b54 240
bae0ee7a 241 f2fs_wait_on_page_writeback(page, DATA, true, true);
e5e5732d 242
28bc106b
CY		 243 if (recover) {
244 struct dnode_of_data dn;
245 struct node_info ni;
246
247 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 248retry:
28bc106b 249 set_new_dnode(&dn, inode, NULL, NULL, 0);
4d57b86d
CY		 250 err = f2fs_get_dnode_of_data(&dn, page->index,
251 LOOKUP_NODE);
7f2b4e8e
CY		 252 if (err) {
253 if (err == -ENOMEM) {
254 congestion_wait(BLK_RW_ASYNC, HZ/50);
255 cond_resched();
256 goto retry;
257 }
28bc106b
CY		 258 err = -EAGAIN;
259 goto next;
260 }
7735730d
CY		 261
262 err = f2fs_get_node_info(sbi, dn.nid, &ni);
263 if (err) {
264 f2fs_put_dnode(&dn);
265 return err;
266 }
267
f1d2564a 268 if (cur->old_addr == NEW_ADDR) {
4d57b86d 269 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
f1d2564a
DJ		 270 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
271 } else
272 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 273 cur->old_addr, ni.version, true, true);
274 f2fs_put_dnode(&dn);
275 }
276next:
63c52d78 277 /* we don't need to invalidate this in the sccessful status */
2baf0781 278 if (drop || recover) {
63c52d78 279 ClearPageUptodate(page);
2baf0781
CY		 280 clear_cold_data(page);
281 }
240a5915 282 f2fs_clear_page_private(page);
28bc106b 283 f2fs_put_page(page, 1);
29b96b54
CY		 284
285 list_del(&cur->list);
286 kmem_cache_free(inmem_entry_slab, cur);
287 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
288 }
28bc106b 289 return err;
29b96b54
CY		 290}
291
4d57b86d 292void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
57864ae5
JK		 293{
294 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
295 struct inode *inode;
296 struct f2fs_inode_info *fi;
297next:
298 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
299 if (list_empty(head)) {
300 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
301 return;
302 }
303 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
304 inode = igrab(&fi->vfs_inode);
305 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
306
307 if (inode) {
2ef79ecb
CY		 308 if (gc_failure) {
309 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
310 goto drop;
311 goto skip;
312 }
313drop:
314 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
4d57b86d 315 f2fs_drop_inmem_pages(inode);
57864ae5
JK		 316 iput(inode);
317 }
2ef79ecb 318skip:
57864ae5
JK		 319 congestion_wait(BLK_RW_ASYNC, HZ/50);
320 cond_resched();
321 goto next;
322}
323
4d57b86d 324void f2fs_drop_inmem_pages(struct inode *inode)
29b96b54 325{
57864ae5 326 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 327 struct f2fs_inode_info *fi = F2FS_I(inode);
328
48432984
CY		 329 while (!list_empty(&fi->inmem_pages)) {
330 mutex_lock(&fi->inmem_lock);
331 __revoke_inmem_pages(inode, &fi->inmem_pages,
332 true, false, true);
333
334 if (list_empty(&fi->inmem_pages)) {
335 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
336 if (!list_empty(&fi->inmem_ilist))
337 list_del_init(&fi->inmem_ilist);
338 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
339 }
340 mutex_unlock(&fi->inmem_lock);
341 }
5fe45743
CY		 342
343 clear_inode_flag(inode, FI_ATOMIC_FILE);
2ef79ecb 344 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
5fe45743 345 stat_dec_atomic_write(inode);
29b96b54
CY		 346}
347
4d57b86d 348void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
8c242db9
JK		 349{
350 struct f2fs_inode_info *fi = F2FS_I(inode);
351 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
352 struct list_head *head = &fi->inmem_pages;
353 struct inmem_pages *cur = NULL;
354
355 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
356
357 mutex_lock(&fi->inmem_lock);
358 list_for_each_entry(cur, head, list) {
359 if (cur->page == page)
360 break;
361 }
362
d0891e84 363 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
8c242db9
JK		 364 list_del(&cur->list);
365 mutex_unlock(&fi->inmem_lock);
366
367 dec_page_count(sbi, F2FS_INMEM_PAGES);
368 kmem_cache_free(inmem_entry_slab, cur);
369
370 ClearPageUptodate(page);
240a5915 371 f2fs_clear_page_private(page);
8c242db9
JK		 372 f2fs_put_page(page, 0);
373
374 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
375}
376
4d57b86d 377static int __f2fs_commit_inmem_pages(struct inode *inode)
88b88a66
JK		 378{
379 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
380 struct f2fs_inode_info *fi = F2FS_I(inode);
381 struct inmem_pages *cur, *tmp;
88b88a66 382 struct f2fs_io_info fio = {
05ca3632 383 .sbi = sbi,
39d787be 384 .ino = inode->i_ino,
88b88a66 385 .type = DATA,
04d328de 386 .op = REQ_OP_WRITE,
70fd7614 387 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 388 .io_type = FS_DATA_IO,
88b88a66 389 };
cf52b27a 390 struct list_head revoke_list;
bab475c5 391 bool submit_bio = false;
edb27dee 392 int err = 0;
88b88a66 393
cf52b27a
CY		 394 INIT_LIST_HEAD(&revoke_list);
395
88b88a66 396 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 397 struct page *page = cur->page;
398
399 lock_page(page);
400 if (page->mapping == inode->i_mapping) {
401 trace_f2fs_commit_inmem_page(page, INMEM);
402
bae0ee7a 403 f2fs_wait_on_page_writeback(page, DATA, true, true);
8d64d365
CY		 404
405 set_page_dirty(page);
933439c8 406 if (clear_page_dirty_for_io(page)) {
29b96b54 407 inode_dec_dirty_pages(inode);
4d57b86d 408 f2fs_remove_dirty_inode(inode);
933439c8 409 }
640cc189 410retry:
28bc106b 411 fio.page = page;
e959c8f5 412 fio.old_blkaddr = NULL_ADDR;
4d978078 413 fio.encrypted_page = NULL;
cc15620b 414 fio.need_lock = LOCK_DONE;
4d57b86d 415 err = f2fs_do_write_data_page(&fio);
29b96b54 416 if (err) {
640cc189
JK		 417 if (err == -ENOMEM) {
418 congestion_wait(BLK_RW_ASYNC, HZ/50);
419 cond_resched();
420 goto retry;
421 }
28bc106b 422 unlock_page(page);
29b96b54 423 break;
70c640b1 424 }
28bc106b
CY		 425 /* record old blkaddr for revoking */
426 cur->old_addr = fio.old_blkaddr;
bab475c5 427 submit_bio = true;
28bc106b
CY		 428 }
429 unlock_page(page);
cf52b27a 430 list_move_tail(&cur->list, &revoke_list);
88b88a66 431 }
29b96b54 432
bab475c5
CY		 433 if (submit_bio)
434 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
28bc106b 435
cf52b27a
CY		 436 if (err) {
437 /*
438 * try to revoke all committed pages, but still we could fail
439 * due to no memory or other reason, if that happened, EAGAIN
440 * will be returned, which means in such case, transaction is
441 * already not integrity, caller should use journal to do the
442 * recovery or rewrite & commit last transaction. For other
443 * error number, revoking was done by filesystem itself.
444 */
48432984
CY		 445 err = __revoke_inmem_pages(inode, &revoke_list,
446 false, true, false);
cf52b27a
CY		 447
448 /* drop all uncommitted pages */
48432984
CY		 449 __revoke_inmem_pages(inode, &fi->inmem_pages,
450 true, false, false);
cf52b27a 451 } else {
48432984
CY		 452 __revoke_inmem_pages(inode, &revoke_list,
453 false, false, false);
cf52b27a 454 }
28bc106b 455
29b96b54
CY		 456 return err;
457}
458
4d57b86d 459int f2fs_commit_inmem_pages(struct inode *inode)
29b96b54
CY		 460{
461 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
462 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b 463 int err;
29b96b54
CY		 464
465 f2fs_balance_fs(sbi, true);
29b96b54 466
6f8d4455
JK		 467 down_write(&fi->i_gc_rwsem[WRITE]);
468
469 f2fs_lock_op(sbi);
5fe45743
CY		 470 set_inode_flag(inode, FI_ATOMIC_COMMIT);
471
29b96b54 472 mutex_lock(&fi->inmem_lock);
4d57b86d 473 err = __f2fs_commit_inmem_pages(inode);
28bc106b 474
57864ae5
JK		 475 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
476 if (!list_empty(&fi->inmem_ilist))
477 list_del_init(&fi->inmem_ilist);
478 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 479 mutex_unlock(&fi->inmem_lock);
480
5fe45743
CY		 481 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
482
29b96b54 483 f2fs_unlock_op(sbi);
6f8d4455
JK		 484 up_write(&fi->i_gc_rwsem[WRITE]);
485
edb27dee 486 return err;
88b88a66
JK		 487}
488
0a8165d7 489/*
351df4b2
JK		 490 * This function balances dirty node and dentry pages.
491 * In addition, it controls garbage collection.
492 */
2c4db1a6 493void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 494{
55523519
CY		 495 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
496 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 497 f2fs_stop_checkpoint(sbi, false);
55523519 498 }
0f348028 499
e589c2c4 500 /* balance_fs_bg is able to be pending */
a7881893 501 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 502 f2fs_balance_fs_bg(sbi);
503
4354994f
DR		 504 if (f2fs_is_checkpoint_ready(sbi))
505 return;
506
351df4b2 507 /*
029cd28c
JK		 508 * We should do GC or end up with checkpoint, if there are so many dirty
509 * dir/node pages without enough free segments.
351df4b2 510 */
7f3037a5 511 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 512 mutex_lock(&sbi->gc_mutex);
e066b83c 513 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 514 }
515}
516
4660f9c0
JK		 517void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
518{
64c74a7a
CY		 519 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
520 return;
521
1dcc336b 522 /* try to shrink extent cache when there is no enough memory */
4d57b86d 523 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
554df79e 524 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 525
1b38dc8e 526 /* check the # of cached NAT entries */
4d57b86d
CY		 527 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
528 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
1b38dc8e 529
4d57b86d
CY		 530 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
531 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
ad4edb83 532 else
4d57b86d 533 f2fs_build_free_nids(sbi, false, false);
31696580 534
a7d10cf3 535 if (!is_idle(sbi, REQ_TIME) &&
fd8c8caf 536 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
f455c8a5 537 return;
31696580 538
1b38dc8e 539 /* checkpoint is the only way to shrink partial cached entries */
4d57b86d
CY		 540 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
541 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 542 excess_prefree_segs(sbi) ||
543 excess_dirty_nats(sbi) ||
fd8c8caf 544 excess_dirty_nodes(sbi) ||
f455c8a5 545 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 546 if (test_opt(sbi, DATA_FLUSH)) {
547 struct blk_plug plug;
548
040d2bb3
CY		 549 mutex_lock(&sbi->flush_lock);
550
e9f5b8b8 551 blk_start_plug(&plug);
4d57b86d 552 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8 553 blk_finish_plug(&plug);
040d2bb3
CY		 554
555 mutex_unlock(&sbi->flush_lock);
e9f5b8b8 556 }
4660f9c0 557 f2fs_sync_fs(sbi->sb, true);
42190d2a 558 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 559 }
4660f9c0
JK		 560}
561
20fda56b
KM		 562static int __submit_flush_wait(struct f2fs_sb_info *sbi,
563 struct block_device *bdev)
3c62be17 564{
dc37910d 565 struct bio *bio;
3c62be17
JK		 566 int ret;
567
dc37910d
CY		 568 bio = f2fs_bio_alloc(sbi, 0, false);
569 if (!bio)
570 return -ENOMEM;
571
3adc5fcb 572 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 573 bio_set_dev(bio, bdev);
3c62be17
JK		 574 ret = submit_bio_wait(bio);
575 bio_put(bio);
20fda56b
KM		 576
577 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
578 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 579 return ret;
580}
581
39d787be 582static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 583{
39d787be 584 int ret = 0;
3c62be17
JK		 585 int i;
586
0916878d 587 if (!f2fs_is_multi_device(sbi))
39d787be 588 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 589
39d787be 590 for (i = 0; i < sbi->s_ndevs; i++) {
4d57b86d 591 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
39d787be 592 continue;
20fda56b
KM		 593 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
594 if (ret)
595 break;
3c62be17
JK		 596 }
597 return ret;
598}
599
2163d198 600static int issue_flush_thread(void *data)
6b4afdd7
JK		 601{
602 struct f2fs_sb_info *sbi = data;
b01a9201 603 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 604 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 605repeat:
606 if (kthread_should_stop())
607 return 0;
608
dc6febb6
CY		 609 sb_start_intwrite(sbi->sb);
610
721bd4d5 611 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 612 struct flush_cmd *cmd, *next;
613 int ret;
614
721bd4d5
GZ		 615 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
616 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
617
39d787be
CY		 618 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
619
620 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 621 atomic_inc(&fcc->issued_flush);
622
721bd4d5
GZ		 623 llist_for_each_entry_safe(cmd, next,
624 fcc->dispatch_list, llnode) {
6b4afdd7 625 cmd->ret = ret;
6b4afdd7
JK		 626 complete(&cmd->wait);
627 }
a688b9d9 628 fcc->dispatch_list = NULL;
6b4afdd7
JK		 629 }
630
dc6febb6
CY		 631 sb_end_intwrite(sbi->sb);
632
a688b9d9 633 wait_event_interruptible(*q,
721bd4d5 634 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 635 goto repeat;
636}
637
39d787be 638int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 639{
b01a9201 640 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 641 struct flush_cmd cmd;
8b8dd65f 642 int ret;
6b4afdd7 643
0f7b2abd
JK		 644 if (test_opt(sbi, NOBARRIER))
645 return 0;
646
8b8dd65f 647 if (!test_opt(sbi, FLUSH_MERGE)) {
72691af6 648 atomic_inc(&fcc->queued_flush);
39d787be 649 ret = submit_flush_wait(sbi, ino);
72691af6 650 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 651 atomic_inc(&fcc->issued_flush);
652 return ret;
653 }
740432f8 654
0916878d
DLM		 655 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
656 f2fs_is_multi_device(sbi)) {
39d787be 657 ret = submit_flush_wait(sbi, ino);
72691af6 658 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 659
660 atomic_inc(&fcc->issued_flush);
740432f8
JK		 661 return ret;
662 }
6b4afdd7 663
39d787be 664 cmd.ino = ino;
adf8d90b 665 init_completion(&cmd.wait);
6b4afdd7 666
721bd4d5 667 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 668
6f890df0
CY		 669 /* update issue_list before we wake up issue_flush thread */
670 smp_mb();
671
672 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 673 wake_up(&fcc->flush_wait_queue);
6b4afdd7 674
5eba8c5d
JK		 675 if (fcc->f2fs_issue_flush) {
676 wait_for_completion(&cmd.wait);
72691af6 677 atomic_dec(&fcc->queued_flush);
5eba8c5d 678 } else {
d3238691
CY		 679 struct llist_node *list;
680
681 list = llist_del_all(&fcc->issue_list);
682 if (!list) {
683 wait_for_completion(&cmd.wait);
72691af6 684 atomic_dec(&fcc->queued_flush);
d3238691
CY		 685 } else {
686 struct flush_cmd *tmp, *next;
687
39d787be 688 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 689
690 llist_for_each_entry_safe(tmp, next, list, llnode) {
691 if (tmp == &cmd) {
692 cmd.ret = ret;
72691af6 693 atomic_dec(&fcc->queued_flush);
d3238691
CY		 694 continue;
695 }
696 tmp->ret = ret;
697 complete(&tmp->wait);
698 }
699 }
5eba8c5d 700 }
adf8d90b
CY		 701
702 return cmd.ret;
6b4afdd7
JK		 703}
704
4d57b86d 705int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
2163d198
GZ		 706{
707 dev_t dev = sbi->sb->s_bdev->bd_dev;
708 struct flush_cmd_control *fcc;
709 int err = 0;
710
b01a9201
JK		 711 if (SM_I(sbi)->fcc_info) {
712 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 713 if (fcc->f2fs_issue_flush)
714 return err;
5eba8c5d
JK		 715 goto init_thread;
716 }
717
acbf054d 718 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 719 if (!fcc)
720 return -ENOMEM;
8b8dd65f 721 atomic_set(&fcc->issued_flush, 0);
72691af6 722 atomic_set(&fcc->queued_flush, 0);
2163d198 723 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 724 init_llist_head(&fcc->issue_list);
b01a9201 725 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 726 if (!test_opt(sbi, FLUSH_MERGE))
727 return err;
728
5eba8c5d 729init_thread:
2163d198
GZ		 730 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
731 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
732 if (IS_ERR(fcc->f2fs_issue_flush)) {
733 err = PTR_ERR(fcc->f2fs_issue_flush);
5222595d 734 kvfree(fcc);
b01a9201 735 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 736 return err;
737 }
2163d198
GZ		 738
739 return err;
740}
741
4d57b86d 742void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 743{
b01a9201 744 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 745
5eba8c5d
JK		 746 if (fcc && fcc->f2fs_issue_flush) {
747 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
748
749 fcc->f2fs_issue_flush = NULL;
750 kthread_stop(flush_thread);
751 }
752 if (free) {
5222595d 753 kvfree(fcc);
b01a9201 754 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 755 }
2163d198
GZ		 756}
757
1228b482
CY		 758int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
759{
760 int ret = 0, i;
761
0916878d 762 if (!f2fs_is_multi_device(sbi))
1228b482
CY		 763 return 0;
764
765 for (i = 1; i < sbi->s_ndevs; i++) {
766 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
767 continue;
768 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
769 if (ret)
770 break;
771
772 spin_lock(&sbi->dev_lock);
773 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
774 spin_unlock(&sbi->dev_lock);
775 }
776
777 return ret;
778}
779
351df4b2
JK		 780static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
781 enum dirty_type dirty_type)
782{
783 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
784
785 /* need not be added */
786 if (IS_CURSEG(sbi, segno))
787 return;
788
789 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
790 dirty_i->nr_dirty[dirty_type]++;
791
792 if (dirty_type == DIRTY) {
793 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 794 enum dirty_type t = sentry->type;
b2f2c390 795
ec325b52
JK		 796 if (unlikely(t >= DIRTY)) {
797 f2fs_bug_on(sbi, 1);
798 return;
799 }
4625d6aa
CL		 800 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
801 dirty_i->nr_dirty[t]++;
351df4b2
JK		 802 }
803}
804
805static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
806 enum dirty_type dirty_type)
807{
808 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
809
810 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
811 dirty_i->nr_dirty[dirty_type]--;
812
813 if (dirty_type == DIRTY) {
4625d6aa
CL		 814 struct seg_entry *sentry = get_seg_entry(sbi, segno);
815 enum dirty_type t = sentry->type;
816
817 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
818 dirty_i->nr_dirty[t]--;
b2f2c390 819
302bd348 820 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 821 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 822 dirty_i->victim_secmap);
351df4b2
JK		 823 }
824}
825
0a8165d7 826/*
351df4b2
JK		 827 * Should not occur error such as -ENOMEM.
828 * Adding dirty entry into seglist is not critical operation.
829 * If a given segment is one of current working segments, it won't be added.
830 */
8d8451af 831static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 832{
833 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4354994f 834 unsigned short valid_blocks, ckpt_valid_blocks;
351df4b2
JK		 835
836 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
837 return;
838
839 mutex_lock(&dirty_i->seglist_lock);
840
302bd348 841 valid_blocks = get_valid_blocks(sbi, segno, false);
4354994f 842 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
351df4b2 843
4354994f
DR		 844 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
845 ckpt_valid_blocks == sbi->blocks_per_seg)) {
351df4b2
JK		 846 __locate_dirty_segment(sbi, segno, PRE);
847 __remove_dirty_segment(sbi, segno, DIRTY);
848 } else if (valid_blocks < sbi->blocks_per_seg) {
849 __locate_dirty_segment(sbi, segno, DIRTY);
850 } else {
851 /* Recovery routine with SSR needs this */
852 __remove_dirty_segment(sbi, segno, DIRTY);
853 }
854
855 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 856}
857
4354994f
DR		 858/* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
859void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
860{
861 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
862 unsigned int segno;
863
864 mutex_lock(&dirty_i->seglist_lock);
865 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
866 if (get_valid_blocks(sbi, segno, false))
867 continue;
868 if (IS_CURSEG(sbi, segno))
869 continue;
870 __locate_dirty_segment(sbi, segno, PRE);
871 __remove_dirty_segment(sbi, segno, DIRTY);
872 }
873 mutex_unlock(&dirty_i->seglist_lock);
874}
875
4d3aed70 876block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
4354994f 877{
ae4ad7ea
DR		 878 int ovp_hole_segs =
879 (overprovision_segments(sbi) - reserved_segments(sbi));
880 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
4d3aed70 881 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4354994f 882 block_t holes[2] = {0, 0}; /* DATA and NODE */
4d3aed70 883 block_t unusable;
4354994f
DR		 884 struct seg_entry *se;
885 unsigned int segno;
886
887 mutex_lock(&dirty_i->seglist_lock);
888 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
889 se = get_seg_entry(sbi, segno);
890 if (IS_NODESEG(se->type))
891 holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
892 else
893 holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
894 }
895 mutex_unlock(&dirty_i->seglist_lock);
896
4d3aed70
DR		 897 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
898 if (unusable > ovp_holes)
899 return unusable - ovp_holes;
900 return 0;
901}
902
903int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
904{
905 int ovp_hole_segs =
906 (overprovision_segments(sbi) - reserved_segments(sbi));
907 if (unusable > F2FS_OPTION(sbi).unusable_cap)
4354994f 908 return -EAGAIN;
db610a64 909 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
ae4ad7ea 910 dirty_segments(sbi) > ovp_hole_segs)
db610a64 911 return -EAGAIN;
4354994f
DR		 912 return 0;
913}
914
915/* This is only used by SBI_CP_DISABLED */
916static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
917{
918 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
919 unsigned int segno = 0;
920
921 mutex_lock(&dirty_i->seglist_lock);
922 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
923 if (get_valid_blocks(sbi, segno, false))
924 continue;
925 if (get_ckpt_valid_blocks(sbi, segno))
926 continue;
927 mutex_unlock(&dirty_i->seglist_lock);
928 return segno;
929 }
930 mutex_unlock(&dirty_i->seglist_lock);
931 return NULL_SEGNO;
932}
933
004b6862 934static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 935 struct block_device *bdev, block_t lstart,
936 block_t start, block_t len)
275b66b0 937{
0b54fb84 938 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 939 struct list_head *pend_list;
b01a9201 940 struct discard_cmd *dc;
275b66b0 941
ba48a33e
CY		 942 f2fs_bug_on(sbi, !len);
943
944 pend_list = &dcc->pend_list[plist_idx(len)];
945
b01a9201
JK		 946 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
947 INIT_LIST_HEAD(&dc->list);
c81abe34 948 dc->bdev = bdev;
b01a9201 949 dc->lstart = lstart;
c81abe34 950 dc->start = start;
b01a9201 951 dc->len = len;
ec9895ad 952 dc->ref = 0;
15469963 953 dc->state = D_PREP;
72691af6 954 dc->queued = 0;
c81abe34 955 dc->error = 0;
b01a9201 956 init_completion(&dc->wait);
22d375dd 957 list_add_tail(&dc->list, pend_list);
35ec7d57
CY		 958 spin_lock_init(&dc->lock);
959 dc->bio_ref = 0;
5f32366a 960 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 961 dcc->undiscard_blks += len;
004b6862
CY		 962
963 return dc;
964}
965
966static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
967 struct block_device *bdev, block_t lstart,
968 block_t start, block_t len,
4dada3fd
CY		 969 struct rb_node *parent, struct rb_node **p,
970 bool leftmost)
004b6862
CY		 971{
972 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
973 struct discard_cmd *dc;
974
975 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
976
977 rb_link_node(&dc->rb_node, parent, p);
4dada3fd 978 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
004b6862
CY		 979
980 return dc;
15469963
JK		 981}
982
004b6862
CY		 983static void __detach_discard_cmd(struct discard_cmd_control *dcc,
984 struct discard_cmd *dc)
15469963 985{
dcc9165d 986 if (dc->state == D_DONE)
72691af6 987 atomic_sub(dc->queued, &dcc->queued_discard);
004b6862
CY		 988
989 list_del(&dc->list);
4dada3fd 990 rb_erase_cached(&dc->rb_node, &dcc->root);
d84d1cbd 991 dcc->undiscard_blks -= dc->len;
004b6862
CY		 992
993 kmem_cache_free(discard_cmd_slab, dc);
994
995 atomic_dec(&dcc->discard_cmd_cnt);
996}
997
998static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
999 struct discard_cmd *dc)
1000{
1001 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
35ec7d57 1002 unsigned long flags;
dcc9165d 1003
2ec6f2ef
CY		 1004 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1005
35ec7d57
CY		 1006 spin_lock_irqsave(&dc->lock, flags);
1007 if (dc->bio_ref) {
1008 spin_unlock_irqrestore(&dc->lock, flags);
1009 return;
1010 }
1011 spin_unlock_irqrestore(&dc->lock, flags);
1012
d9703d90
CY		 1013 f2fs_bug_on(sbi, dc->ref);
1014
c81abe34
JK		 1015 if (dc->error == -EOPNOTSUPP)
1016 dc->error = 0;
15469963 1017
c81abe34 1018 if (dc->error)
22d7ea13
CY		 1019 printk_ratelimited(
1020 "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1021 KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
004b6862 1022 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 1023}
1024
c81abe34
JK		 1025static void f2fs_submit_discard_endio(struct bio *bio)
1026{
1027 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
35ec7d57 1028 unsigned long flags;
c81abe34 1029
4e4cbee9 1030 dc->error = blk_status_to_errno(bio->bi_status);
35ec7d57
CY		 1031
1032 spin_lock_irqsave(&dc->lock, flags);
1033 dc->bio_ref--;
1034 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1035 dc->state = D_DONE;
1036 complete_all(&dc->wait);
1037 }
1038 spin_unlock_irqrestore(&dc->lock, flags);
c81abe34
JK		 1039 bio_put(bio);
1040}
1041
94b1e10e 1042static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 1043 block_t start, block_t end)
1044{
1045#ifdef CONFIG_F2FS_CHECK_FS
1046 struct seg_entry *sentry;
1047 unsigned int segno;
1048 block_t blk = start;
1049 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1050 unsigned long *map;
1051
1052 while (blk < end) {
1053 segno = GET_SEGNO(sbi, blk);
1054 sentry = get_seg_entry(sbi, segno);
1055 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1056
008396e1
YS		 1057 if (end < START_BLOCK(sbi, segno + 1))
1058 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1059 else
1060 size = max_blocks;
6915ea9d
CY		 1061 map = (unsigned long *)(sentry->cur_valid_map);
1062 offset = __find_rev_next_bit(map, size, offset);
1063 f2fs_bug_on(sbi, offset != size);
008396e1 1064 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 1065 }
1066#endif
1067}
1068
8bb4f253
JK		 1069static void __init_discard_policy(struct f2fs_sb_info *sbi,
1070 struct discard_policy *dpolicy,
1071 int discard_type, unsigned int granularity)
1072{
1073 /* common policy */
1074 dpolicy->type = discard_type;
1075 dpolicy->sync = true;
20ee4382 1076 dpolicy->ordered = false;
8bb4f253
JK		 1077 dpolicy->granularity = granularity;
1078
1079 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1080 dpolicy->io_aware_gran = MAX_PLIST_NUM;
03f2c02d 1081 dpolicy->timeout = 0;
8bb4f253
JK		 1082
1083 if (discard_type == DPOLICY_BG) {
1084 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1085 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1086 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1087 dpolicy->io_aware = true;
cba60849 1088 dpolicy->sync = false;
20ee4382 1089 dpolicy->ordered = true;
8bb4f253
JK		 1090 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1091 dpolicy->granularity = 1;
1092 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1093 }
1094 } else if (discard_type == DPOLICY_FORCE) {
1095 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1096 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1097 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1098 dpolicy->io_aware = false;
1099 } else if (discard_type == DPOLICY_FSTRIM) {
1100 dpolicy->io_aware = false;
1101 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 1102 dpolicy->max_requests = UINT_MAX;
8bb4f253 1103 dpolicy->io_aware = false;
b8623253
JK		 1104 /* we need to issue all to keep CP_TRIMMED_FLAG */
1105 dpolicy->granularity = 1;
8bb4f253
JK		 1106 }
1107}
1108
35ec7d57
CY		 1109static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1110 struct block_device *bdev, block_t lstart,
1111 block_t start, block_t len);
c81abe34 1112/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
6b9cb124 1113static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1114 struct discard_policy *dpolicy,
35ec7d57
CY		 1115 struct discard_cmd *dc,
1116 unsigned int *issued)
c81abe34 1117{
35ec7d57
CY		 1118 struct block_device *bdev = dc->bdev;
1119 struct request_queue *q = bdev_get_queue(bdev);
1120 unsigned int max_discard_blocks =
1121 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
c81abe34 1122 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1123 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1124 &(dcc->fstrim_list) : &(dcc->wait_list);
78997b56 1125 int flag = dpolicy->sync ? REQ_SYNC : 0;
35ec7d57
CY		 1126 block_t lstart, start, len, total_len;
1127 int err = 0;
c81abe34
JK		 1128
1129 if (dc->state != D_PREP)
6b9cb124 1130 return 0;
c81abe34 1131
d6184774 1132 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
6b9cb124 1133 return 0;
d6184774 1134
35ec7d57
CY		 1135 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1136
1137 lstart = dc->lstart;
1138 start = dc->start;
1139 len = dc->len;
1140 total_len = len;
1141
1142 dc->len = 0;
1143
1144 while (total_len && *issued < dpolicy->max_requests && !err) {
1145 struct bio *bio = NULL;
1146 unsigned long flags;
1147 bool last = true;
1148
1149 if (len > max_discard_blocks) {
1150 len = max_discard_blocks;
1151 last = false;
1152 }
1153
1154 (*issued)++;
1155 if (*issued == dpolicy->max_requests)
1156 last = true;
1157
1158 dc->len += len;
1159
b83dcfe6
CY		 1160 if (time_to_inject(sbi, FAULT_DISCARD)) {
1161 f2fs_show_injection_info(FAULT_DISCARD);
1162 err = -EIO;
1163 goto submit;
1164 }
35ec7d57
CY		 1165 err = __blkdev_issue_discard(bdev,
1166 SECTOR_FROM_BLOCK(start),
1167 SECTOR_FROM_BLOCK(len),
1168 GFP_NOFS, 0, &bio);
b83dcfe6 1169submit:
6b9cb124 1170 if (err) {
35ec7d57 1171 spin_lock_irqsave(&dc->lock, flags);
6b9cb124 1172 if (dc->state == D_PARTIAL)
35ec7d57 1173 dc->state = D_SUBMIT;
35ec7d57
CY		 1174 spin_unlock_irqrestore(&dc->lock, flags);
1175
6b9cb124
CY		 1176 break;
1177 }
35ec7d57 1178
6b9cb124 1179 f2fs_bug_on(sbi, !bio);
35ec7d57 1180
6b9cb124
CY		 1181 /*
1182 * should keep before submission to avoid D_DONE
1183 * right away
1184 */
1185 spin_lock_irqsave(&dc->lock, flags);
1186 if (last)
1187 dc->state = D_SUBMIT;
1188 else
1189 dc->state = D_PARTIAL;
1190 dc->bio_ref++;
1191 spin_unlock_irqrestore(&dc->lock, flags);
35ec7d57 1192
72691af6
JK		 1193 atomic_inc(&dcc->queued_discard);
1194 dc->queued++;
6b9cb124 1195 list_move_tail(&dc->list, wait_list);
b0af6d49 1196
6b9cb124 1197 /* sanity check on discard range */
9249dded 1198 __check_sit_bitmap(sbi, lstart, lstart + len);
35ec7d57 1199
6b9cb124
CY		 1200 bio->bi_private = dc;
1201 bio->bi_end_io = f2fs_submit_discard_endio;
1202 bio->bi_opf |= flag;
1203 submit_bio(bio);
1204
1205 atomic_inc(&dcc->issued_discard);
1206
1207 f2fs_update_iostat(sbi, FS_DISCARD, 1);
35ec7d57
CY		 1208
1209 lstart += len;
1210 start += len;
1211 total_len -= len;
1212 len = total_len;
c81abe34 1213 }
35ec7d57 1214
6b9cb124 1215 if (!err && len)
35ec7d57 1216 __update_discard_tree_range(sbi, bdev, lstart, start, len);
6b9cb124 1217 return err;
c81abe34
JK		 1218}
1219
004b6862
CY		 1220static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1221 struct block_device *bdev, block_t lstart,
1222 block_t start, block_t len,
1223 struct rb_node **insert_p,
1224 struct rb_node *insert_parent)
c81abe34 1225{
004b6862 1226 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1227 struct rb_node **p;
004b6862
CY		 1228 struct rb_node *parent = NULL;
1229 struct discard_cmd *dc = NULL;
4dada3fd 1230 bool leftmost = true;
004b6862
CY		 1231
1232 if (insert_p && insert_parent) {
1233 parent = insert_parent;
1234 p = insert_p;
1235 goto do_insert;
1236 }
c81abe34 1237
4dada3fd
CY		 1238 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1239 lstart, &leftmost);
004b6862 1240do_insert:
4dada3fd
CY		 1241 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1242 p, leftmost);
004b6862
CY		 1243 if (!dc)
1244 return NULL;
c81abe34 1245
004b6862 1246 return dc;
c81abe34
JK		 1247}
1248
ba48a33e
CY		 1249static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1250 struct discard_cmd *dc)
1251{
1252 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1253}
1254
3d6a650f
YH		 1255static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1256 struct discard_cmd *dc, block_t blkaddr)
1257{
ba48a33e 1258 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1259 struct discard_info di = dc->di;
1260 bool modified = false;
3d6a650f 1261
004b6862 1262 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1263 __remove_discard_cmd(sbi, dc);
1264 return;
1265 }
1266
d84d1cbd
CY		 1267 dcc->undiscard_blks -= di.len;
1268
004b6862 1269 if (blkaddr > di.lstart) {
3d6a650f 1270 dc->len = blkaddr - dc->lstart;
d84d1cbd 1271 dcc->undiscard_blks += dc->len;
ba48a33e 1272 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1273 modified = true;
1274 }
1275
1276 if (blkaddr < di.lstart + di.len - 1) {
1277 if (modified) {
1278 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1279 di.start + blkaddr + 1 - di.lstart,
1280 di.lstart + di.len - 1 - blkaddr,
1281 NULL, NULL);
1282 } else {
1283 dc->lstart++;
1284 dc->len--;
1285 dc->start++;
d84d1cbd 1286 dcc->undiscard_blks += dc->len;
ba48a33e 1287 __relocate_discard_cmd(dcc, dc);
004b6862 1288 }
3d6a650f
YH		 1289 }
1290}
1291
004b6862
CY		 1292static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1293 struct block_device *bdev, block_t lstart,
1294 block_t start, block_t len)
275b66b0 1295{
0b54fb84 1296 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1297 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1298 struct discard_cmd *dc;
1299 struct discard_info di = {0};
1300 struct rb_node **insert_p = NULL, *insert_parent = NULL;
35ec7d57
CY		 1301 struct request_queue *q = bdev_get_queue(bdev);
1302 unsigned int max_discard_blocks =
1303 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
004b6862 1304 block_t end = lstart + len;
275b66b0 1305
4d57b86d 1306 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
004b6862
CY		 1307 NULL, lstart,
1308 (struct rb_entry **)&prev_dc,
1309 (struct rb_entry **)&next_dc,
4dada3fd 1310 &insert_p, &insert_parent, true, NULL);
004b6862
CY		 1311 if (dc)
1312 prev_dc = dc;
1313
1314 if (!prev_dc) {
1315 di.lstart = lstart;
1316 di.len = next_dc ? next_dc->lstart - lstart : len;
1317 di.len = min(di.len, len);
1318 di.start = start;
22d375dd 1319 }
15469963 1320
004b6862
CY		 1321 while (1) {
1322 struct rb_node *node;
1323 bool merged = false;
1324 struct discard_cmd *tdc = NULL;
1325
1326 if (prev_dc) {
1327 di.lstart = prev_dc->lstart + prev_dc->len;
1328 if (di.lstart < lstart)
1329 di.lstart = lstart;
1330 if (di.lstart >= end)
1331 break;
1332
1333 if (!next_dc || next_dc->lstart > end)
1334 di.len = end - di.lstart;
1335 else
1336 di.len = next_dc->lstart - di.lstart;
1337 di.start = start + di.lstart - lstart;
1338 }
1339
1340 if (!di.len)
1341 goto next;
1342
1343 if (prev_dc && prev_dc->state == D_PREP &&
1344 prev_dc->bdev == bdev &&
35ec7d57
CY		 1345 __is_discard_back_mergeable(&di, &prev_dc->di,
1346 max_discard_blocks)) {
004b6862 1347 prev_dc->di.len += di.len;
d84d1cbd 1348 dcc->undiscard_blks += di.len;
ba48a33e 1349 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1350 di = prev_dc->di;
1351 tdc = prev_dc;
1352 merged = true;
1353 }
1354
1355 if (next_dc && next_dc->state == D_PREP &&
1356 next_dc->bdev == bdev &&
35ec7d57
CY		 1357 __is_discard_front_mergeable(&di, &next_dc->di,
1358 max_discard_blocks)) {
004b6862
CY		 1359 next_dc->di.lstart = di.lstart;
1360 next_dc->di.len += di.len;
1361 next_dc->di.start = di.start;
d84d1cbd 1362 dcc->undiscard_blks += di.len;
ba48a33e 1363 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1364 if (tdc)
1365 __remove_discard_cmd(sbi, tdc);
004b6862 1366 merged = true;
4e6a8d9b 1367 }
004b6862 1368
df0f6b44 1369 if (!merged) {
004b6862
CY		 1370 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1371 di.len, NULL, NULL);
df0f6b44 1372 }
004b6862
CY		 1373 next:
1374 prev_dc = next_dc;
1375 if (!prev_dc)
1376 break;
1377
1378 node = rb_next(&prev_dc->rb_node);
1379 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1380 }
004b6862
CY		 1381}
1382
1383static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1384 struct block_device *bdev, block_t blkstart, block_t blklen)
1385{
1386 block_t lblkstart = blkstart;
1387
7f3d7719
DLM		 1388 if (!f2fs_bdev_support_discard(bdev))
1389 return 0;
1390
0243a5f9 1391 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862 1392
0916878d 1393 if (f2fs_is_multi_device(sbi)) {
004b6862
CY		 1394 int devi = f2fs_target_device_index(sbi, blkstart);
1395
1396 blkstart -= FDEV(devi).start_blk;
1397 }
35ec7d57 1398 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862 1399 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
35ec7d57 1400 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862
CY		 1401 return 0;
1402}
1403
20ee4382
CY		 1404static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1405 struct discard_policy *dpolicy)
1406{
1407 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1408 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1409 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1410 struct discard_cmd *dc;
1411 struct blk_plug plug;
1412 unsigned int pos = dcc->next_pos;
1413 unsigned int issued = 0;
1414 bool io_interrupted = false;
1415
1416 mutex_lock(&dcc->cmd_lock);
1417 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1418 NULL, pos,
1419 (struct rb_entry **)&prev_dc,
1420 (struct rb_entry **)&next_dc,
4dada3fd 1421 &insert_p, &insert_parent, true, NULL);
20ee4382
CY		 1422 if (!dc)
1423 dc = next_dc;
1424
1425 blk_start_plug(&plug);
1426
1427 while (dc) {
1428 struct rb_node *node;
6b9cb124 1429 int err = 0;
20ee4382
CY		 1430
1431 if (dc->state != D_PREP)
1432 goto next;
1433
a7d10cf3 1434 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
20ee4382
CY		 1435 io_interrupted = true;
1436 break;
1437 }
1438
1439 dcc->next_pos = dc->lstart + dc->len;
6b9cb124 1440 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
20ee4382 1441
35ec7d57 1442 if (issued >= dpolicy->max_requests)
20ee4382
CY		 1443 break;
1444next:
1445 node = rb_next(&dc->rb_node);
6b9cb124
CY		 1446 if (err)
1447 __remove_discard_cmd(sbi, dc);
20ee4382
CY		 1448 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1449 }
1450
1451 blk_finish_plug(&plug);
1452
1453 if (!dc)
1454 dcc->next_pos = 0;
1455
1456 mutex_unlock(&dcc->cmd_lock);
1457
1458 if (!issued && io_interrupted)
1459 issued = -1;
1460
1461 return issued;
1462}
1463
78997b56
CY		 1464static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1465 struct discard_policy *dpolicy)
bd5b0738
CY		 1466{
1467 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1468 struct list_head *pend_list;
1469 struct discard_cmd *dc, *tmp;
1470 struct blk_plug plug;
522d1711 1471 int i, issued = 0;
e6c6de18 1472 bool io_interrupted = false;
bd5b0738 1473
03f2c02d
JK		 1474 if (dpolicy->timeout != 0)
1475 f2fs_update_time(sbi, dpolicy->timeout);
1476
78997b56 1477 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
03f2c02d
JK		 1478 if (dpolicy->timeout != 0 &&
1479 f2fs_time_over(sbi, dpolicy->timeout))
1480 break;
1481
78997b56
CY		 1482 if (i + 1 < dpolicy->granularity)
1483 break;
20ee4382
CY		 1484
1485 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1486 return __issue_discard_cmd_orderly(sbi, dpolicy);
1487
bd5b0738 1488 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1489
1490 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1491 if (list_empty(pend_list))
1492 goto next;
67fce70b
CY		 1493 if (unlikely(dcc->rbtree_check))
1494 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1495 &dcc->root));
33da62cf 1496 blk_start_plug(&plug);
bd5b0738
CY		 1497 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1498 f2fs_bug_on(sbi, dc->state != D_PREP);
1499
ecc9aa00 1500 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
a7d10cf3 1501 !is_idle(sbi, DISCARD_TIME)) {
e6c6de18 1502 io_interrupted = true;
522d1711 1503 break;
969d1b18 1504 }
e6c6de18 1505
35ec7d57 1506 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
522d1711 1507
35ec7d57 1508 if (issued >= dpolicy->max_requests)
33da62cf 1509 break;
bd5b0738 1510 }
33da62cf 1511 blk_finish_plug(&plug);
49c60c67 1512next:
33da62cf
CY		 1513 mutex_unlock(&dcc->cmd_lock);
1514
522d1711 1515 if (issued >= dpolicy->max_requests || io_interrupted)
33da62cf 1516 break;
bd5b0738 1517 }
969d1b18 1518
e6c6de18
CY		 1519 if (!issued && io_interrupted)
1520 issued = -1;
1521
969d1b18
CY		 1522 return issued;
1523}
1524
cf5c759f 1525static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1526{
1527 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1528 struct list_head *pend_list;
1529 struct discard_cmd *dc, *tmp;
1530 int i;
cf5c759f 1531 bool dropped = false;
969d1b18
CY		 1532
1533 mutex_lock(&dcc->cmd_lock);
1534 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1535 pend_list = &dcc->pend_list[i];
1536 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1537 f2fs_bug_on(sbi, dc->state != D_PREP);
1538 __remove_discard_cmd(sbi, dc);
cf5c759f 1539 dropped = true;
969d1b18
CY		 1540 }
1541 }
1542 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1543
1544 return dropped;
bd5b0738
CY		 1545}
1546
4d57b86d 1547void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
7950e9ac
CY		 1548{
1549 __drop_discard_cmd(sbi);
1550}
1551
0ea80512 1552static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1553 struct discard_cmd *dc)
1554{
1555 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1556 unsigned int len = 0;
2a510c00
CY		 1557
1558 wait_for_completion_io(&dc->wait);
1559 mutex_lock(&dcc->cmd_lock);
1560 f2fs_bug_on(sbi, dc->state != D_DONE);
1561 dc->ref--;
0ea80512
CY		 1562 if (!dc->ref) {
1563 if (!dc->error)
1564 len = dc->len;
2a510c00 1565 __remove_discard_cmd(sbi, dc);
0ea80512 1566 }
2a510c00 1567 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1568
1569 return len;
2a510c00
CY		 1570}
1571
0ea80512 1572static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1573 struct discard_policy *dpolicy,
1574 block_t start, block_t end)
63a94fa1
CY		 1575{
1576 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1577 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1578 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1579 struct discard_cmd *dc, *tmp;
6afae633 1580 bool need_wait;
0ea80512 1581 unsigned int trimmed = 0;
6afae633
CY		 1582
1583next:
1584 need_wait = false;
63a94fa1
CY		 1585
1586 mutex_lock(&dcc->cmd_lock);
1587 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1588 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1589 continue;
78997b56 1590 if (dc->len < dpolicy->granularity)
8412663d 1591 continue;
78997b56 1592 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1593 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1594 if (!dc->error)
1595 trimmed += dc->len;
63a94fa1 1596 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1597 } else {
1598 dc->ref++;
1599 need_wait = true;
1600 break;
63a94fa1
CY		 1601 }
1602 }
1603 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1604
1605 if (need_wait) {
0ea80512 1606 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1607 goto next;
1608 }
0ea80512
CY		 1609
1610 return trimmed;
63a94fa1
CY		 1611}
1612
01f9cf6d 1613static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1614 struct discard_policy *dpolicy)
8412663d 1615{
9a997188 1616 struct discard_policy dp;
01f9cf6d 1617 unsigned int discard_blks;
9a997188 1618
01f9cf6d
CY		 1619 if (dpolicy)
1620 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
9a997188
JK		 1621
1622 /* wait all */
8bb4f253 1623 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
01f9cf6d 1624 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1625 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
01f9cf6d
CY		 1626 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1627
1628 return discard_blks;
8412663d
CY		 1629}
1630
004b6862 1631/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1632static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1633{
1634 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1635 struct discard_cmd *dc;
ec9895ad 1636 bool need_wait = false;
004b6862
CY		 1637
1638 mutex_lock(&dcc->cmd_lock);
4d57b86d
CY		 1639 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1640 NULL, blkaddr);
004b6862 1641 if (dc) {
ec9895ad
CY		 1642 if (dc->state == D_PREP) {
1643 __punch_discard_cmd(sbi, dc, blkaddr);
1644 } else {
1645 dc->ref++;
1646 need_wait = true;
1647 }
275b66b0 1648 }
d431413f 1649 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1650
2a510c00
CY		 1651 if (need_wait)
1652 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1653}
1654
4d57b86d 1655void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
cce13252
CY		 1656{
1657 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1658
1659 if (dcc && dcc->f2fs_issue_discard) {
1660 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1661
1662 dcc->f2fs_issue_discard = NULL;
1663 kthread_stop(discard_thread);
ec9895ad 1664 }
d431413f
CY		 1665}
1666
8412663d 1667/* This comes from f2fs_put_super */
03f2c02d 1668bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
969d1b18
CY		 1669{
1670 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1671 struct discard_policy dpolicy;
cf5c759f 1672 bool dropped;
969d1b18 1673
8bb4f253
JK		 1674 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1675 dcc->discard_granularity);
03f2c02d 1676 dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
78997b56 1677 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1678 dropped = __drop_discard_cmd(sbi);
cf5c759f 1679
9a997188
JK		 1680 /* just to make sure there is no pending discard commands */
1681 __wait_all_discard_cmd(sbi, NULL);
2482c432
CY		 1682
1683 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
cf5c759f 1684 return dropped;
969d1b18
CY		 1685}
1686
15469963
JK		 1687static int issue_discard_thread(void *data)
1688{
1689 struct f2fs_sb_info *sbi = data;
1690 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1691 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1692 struct discard_policy dpolicy;
969d1b18
CY		 1693 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1694 int issued;
15469963 1695
1d7be270 1696 set_freezable();
15469963 1697
1d7be270 1698 do {
8bb4f253 1699 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1700 dcc->discard_granularity);
1701
969d1b18
CY		 1702 wait_event_interruptible_timeout(*q,
1703 kthread_should_stop() || freezing(current) ||
1704 dcc->discard_wake,
1705 msecs_to_jiffies(wait_ms));
35a9a766
SY		 1706
1707 if (dcc->discard_wake)
1708 dcc->discard_wake = 0;
1709
76c7bfb3
JK		 1710 /* clean up pending candidates before going to sleep */
1711 if (atomic_read(&dcc->queued_discard))
1712 __wait_all_discard_cmd(sbi, NULL);
1713
1d7be270
JK		 1714 if (try_to_freeze())
1715 continue;
3b60d802
CY		 1716 if (f2fs_readonly(sbi->sb))
1717 continue;
1d7be270
JK		 1718 if (kthread_should_stop())
1719 return 0;
d6184774
YH		 1720 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1721 wait_ms = dpolicy.max_interval;
1722 continue;
1723 }
15469963 1724
5b0e9539 1725 if (sbi->gc_mode == GC_URGENT)
8bb4f253 1726 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1727
dc6febb6
CY		 1728 sb_start_intwrite(sbi->sb);
1729
78997b56 1730 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1731 if (issued > 0) {
78997b56
CY		 1732 __wait_all_discard_cmd(sbi, &dpolicy);
1733 wait_ms = dpolicy.min_interval;
f9d1dced 1734 } else if (issued == -1){
a7d10cf3
ST		 1735 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1736 if (!wait_ms)
abde73c7 1737 wait_ms = dpolicy.mid_interval;
969d1b18 1738 } else {
78997b56 1739 wait_ms = dpolicy.max_interval;
969d1b18 1740 }
1d7be270 1741
dc6febb6 1742 sb_end_intwrite(sbi->sb);
1d7be270 1743
1d7be270
JK		 1744 } while (!kthread_should_stop());
1745 return 0;
15469963
JK		 1746}
1747
f46e8809 1748#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1749static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1750 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1751{
92592285 1752 sector_t sector, nr_sects;
10a875f8 1753 block_t lblkstart = blkstart;
3c62be17
JK		 1754 int devi = 0;
1755
0916878d 1756 if (f2fs_is_multi_device(sbi)) {
3c62be17 1757 devi = f2fs_target_device_index(sbi, blkstart);
95175daf
DLM		 1758 if (blkstart < FDEV(devi).start_blk ||
1759 blkstart > FDEV(devi).end_blk) {
1760 f2fs_msg(sbi->sb, KERN_ERR, "Invalid block %x",
1761 blkstart);
1762 return -EIO;
1763 }
3c62be17
JK		 1764 blkstart -= FDEV(devi).start_blk;
1765 }
f46e8809 1766
95175daf
DLM		 1767 /* For sequential zones, reset the zone write pointer */
1768 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
92592285
JK		 1769 sector = SECTOR_FROM_BLOCK(blkstart);
1770 nr_sects = SECTOR_FROM_BLOCK(blklen);
1771
1772 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1773 nr_sects != bdev_zone_sectors(bdev)) {
95175daf
DLM		 1774 f2fs_msg(sbi->sb, KERN_ERR,
1775 "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
92592285
JK		 1776 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1777 blkstart, blklen);
1778 return -EIO;
1779 }
d50aaeec 1780 trace_f2fs_issue_reset_zone(bdev, blkstart);
95175daf 1781 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
f46e8809 1782 }
95175daf
DLM		 1783
1784 /* For conventional zones, use regular discard if supported */
95175daf 1785 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1786}
1787#endif
1788
3c62be17
JK		 1789static int __issue_discard_async(struct f2fs_sb_info *sbi,
1790 struct block_device *bdev, block_t blkstart, block_t blklen)
1791{
1792#ifdef CONFIG_BLK_DEV_ZONED
7f3d7719 1793 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
3c62be17
JK		 1794 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1795#endif
c81abe34 1796 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1797}
1798
1e87a78d 1799static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1800 block_t blkstart, block_t blklen)
1801{
3c62be17
JK		 1802 sector_t start = blkstart, len = 0;
1803 struct block_device *bdev;
a66cdd98
JK		 1804 struct seg_entry *se;
1805 unsigned int offset;
1806 block_t i;
3c62be17
JK		 1807 int err = 0;
1808
1809 bdev = f2fs_target_device(sbi, blkstart, NULL);
1810
1811 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1812 if (i != start) {
1813 struct block_device *bdev2 =
1814 f2fs_target_device(sbi, i, NULL);
1815
1816 if (bdev2 != bdev) {
1817 err = __issue_discard_async(sbi, bdev,
1818 start, len);
1819 if (err)
1820 return err;
1821 bdev = bdev2;
1822 start = i;
1823 len = 0;
1824 }
1825 }
a66cdd98 1826
a66cdd98
JK		 1827 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1828 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1829
1830 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1831 sbi->discard_blks--;
1832 }
f46e8809 1833
3c62be17
JK		 1834 if (len)
1835 err = __issue_discard_async(sbi, bdev, start, len);
1836 return err;
1e87a78d
JK		 1837}
1838
25290fa5
JK		 1839static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1840 bool check_only)
adf4983b 1841{
b2955550
JK		 1842 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1843 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1844 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1845 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1846 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1847 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1848 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1849 unsigned int start = 0, end = -1;
c473f1a9 1850 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1851 struct discard_entry *de = NULL;
46f84c2c 1852 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1853 int i;
1854
7d20c8ab 1855 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
25290fa5 1856 return false;
b2955550 1857
a66cdd98 1858 if (!force) {
7d20c8ab 1859 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
0b54fb84
JK		 1860 SM_I(sbi)->dcc_info->nr_discards >=
1861 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1862 return false;
4b2fecc8
JK		 1863 }
1864
b2955550
JK		 1865 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1866 for (i = 0; i < entries; i++)
a66cdd98 1867 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1868 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1869
0b54fb84
JK		 1870 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1871 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1872 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1873 if (start >= max_blocks)
1874 break;
1875
1876 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1877 if (force && start && end != max_blocks
1878 && (end - start) < cpc->trim_minlen)
1879 continue;
1880
25290fa5
JK		 1881 if (check_only)
1882 return true;
1883
a7eeb823
CY		 1884 if (!de) {
1885 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1886 GFP_F2FS_ZERO);
1887 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1888 list_add_tail(&de->list, head);
1889 }
1890
1891 for (i = start; i < end; i++)
1892 __set_bit_le(i, (void *)de->discard_map);
1893
1894 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1895 }
25290fa5 1896 return false;
b2955550
JK		 1897}
1898
af8ff65b
CY		 1899static void release_discard_addr(struct discard_entry *entry)
1900{
1901 list_del(&entry->list);
1902 kmem_cache_free(discard_entry_slab, entry);
1903}
1904
4d57b86d 1905void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1906{
46f84c2c 1907 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1908 struct discard_entry *entry, *this;
1909
1910 /* drop caches */
af8ff65b
CY		 1911 list_for_each_entry_safe(entry, this, head, list)
1912 release_discard_addr(entry);
4b2fecc8
JK		 1913}
1914
0a8165d7 1915/*
4d57b86d 1916 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1917 */
1918static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1919{
1920 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1921 unsigned int segno;
351df4b2
JK		 1922
1923 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1924 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1925 __set_test_and_free(sbi, segno);
351df4b2
JK		 1926 mutex_unlock(&dirty_i->seglist_lock);
1927}
1928
4d57b86d
CY		 1929void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1930 struct cp_control *cpc)
351df4b2 1931{
969d1b18
CY		 1932 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1933 struct list_head *head = &dcc->entry_list;
2d7b822a 1934 struct discard_entry *entry, *this;
351df4b2 1935 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1936 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1937 unsigned int start = 0, end = -1;
36abef4e 1938 unsigned int secno, start_segno;
c473f1a9 1939 bool force = (cpc->reason & CP_DISCARD);
2c70c5e3 1940 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
351df4b2
JK		 1941
1942 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1943
351df4b2 1944 while (1) {
29e59c14 1945 int i;
ad6672bb
YS		 1946
1947 if (need_align && end != -1)
1948 end--;
7cd8558b
JK		 1949 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1950 if (start >= MAIN_SEGS(sbi))
351df4b2 1951 break;
7cd8558b
JK		 1952 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1953 start + 1);
29e59c14 1954
ad6672bb
YS		 1955 if (need_align) {
1956 start = rounddown(start, sbi->segs_per_sec);
1957 end = roundup(end, sbi->segs_per_sec);
1958 }
29e59c14 1959
ad6672bb
YS		 1960 for (i = start; i < end; i++) {
1961 if (test_and_clear_bit(i, prefree_map))
1962 dirty_i->nr_dirty[PRE]--;
1963 }
29e59c14 1964
7d20c8ab 1965 if (!f2fs_realtime_discard_enable(sbi))
29e59c14 1966 continue;
351df4b2 1967
650d3c4e
YH		 1968 if (force && start >= cpc->trim_start &&
1969 (end - 1) <= cpc->trim_end)
1970 continue;
1971
2c70c5e3 1972 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
36abef4e 1973 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1974 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1975 continue;
1976 }
1977next:
4ddb1a4d
JK		 1978 secno = GET_SEC_FROM_SEG(sbi, start);
1979 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1980 if (!IS_CURSEC(sbi, secno) &&
302bd348 1981 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1982 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1983 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1984
1985 start = start_segno + sbi->segs_per_sec;
1986 if (start < end)
1987 goto next;
8b107f5b
JK		 1988 else
1989 end = start - 1;
351df4b2
JK		 1990 }
1991 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1992
1993 /* send small discards */
2d7b822a 1994 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1995 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1996 bool is_valid = test_bit_le(0, entry->discard_map);
1997
1998find_next:
1999 if (is_valid) {
2000 next_pos = find_next_zero_bit_le(entry->discard_map,
2001 sbi->blocks_per_seg, cur_pos);
2002 len = next_pos - cur_pos;
2003
7beb01f7 2004 if (f2fs_sb_has_blkzoned(sbi) ||
acfd2810 2005 (force && len < cpc->trim_minlen))
a7eeb823
CY		 2006 goto skip;
2007
2008 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2009 len);
a7eeb823
CY		 2010 total_len += len;
2011 } else {
2012 next_pos = find_next_bit_le(entry->discard_map,
2013 sbi->blocks_per_seg, cur_pos);
2014 }
836b5a63 2015skip:
a7eeb823
CY		 2016 cur_pos = next_pos;
2017 is_valid = !is_valid;
2018
2019 if (cur_pos < sbi->blocks_per_seg)
2020 goto find_next;
2021
af8ff65b 2022 release_discard_addr(entry);
969d1b18 2023 dcc->nr_discards -= total_len;
b2955550 2024 }
34e159da 2025
01983c71 2026 wake_up_discard_thread(sbi, false);
351df4b2
JK		 2027}
2028
8ed59745 2029static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 2030{
15469963 2031 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 2032 struct discard_cmd_control *dcc;
ba48a33e 2033 int err = 0, i;
0b54fb84
JK		 2034
2035 if (SM_I(sbi)->dcc_info) {
2036 dcc = SM_I(sbi)->dcc_info;
2037 goto init_thread;
2038 }
2039
acbf054d 2040 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 2041 if (!dcc)
2042 return -ENOMEM;
2043
969d1b18 2044 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 2045 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 2046 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 2047 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 2048 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 2049 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 2050 mutex_init(&dcc->cmd_lock);
8b8dd65f 2051 atomic_set(&dcc->issued_discard, 0);
72691af6 2052 atomic_set(&dcc->queued_discard, 0);
5f32366a 2053 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 2054 dcc->nr_discards = 0;
d618ebaf 2055 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 2056 dcc->undiscard_blks = 0;
20ee4382 2057 dcc->next_pos = 0;
4dada3fd 2058 dcc->root = RB_ROOT_CACHED;
67fce70b 2059 dcc->rbtree_check = false;
0b54fb84 2060
15469963 2061 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 2062 SM_I(sbi)->dcc_info = dcc;
2063init_thread:
15469963
JK		 2064 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2065 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2066 if (IS_ERR(dcc->f2fs_issue_discard)) {
2067 err = PTR_ERR(dcc->f2fs_issue_discard);
5222595d 2068 kvfree(dcc);
15469963
JK		 2069 SM_I(sbi)->dcc_info = NULL;
2070 return err;
2071 }
2072
0b54fb84
JK		 2073 return err;
2074}
2075
f099405f 2076static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 2077{
2078 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2079
f099405f
CY		 2080 if (!dcc)
2081 return;
2082
4d57b86d 2083 f2fs_stop_discard_thread(sbi);
f099405f 2084
5222595d 2085 kvfree(dcc);
f099405f 2086 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 2087}
2088
184a5cd2 2089static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 2090{
2091 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 2092
2093 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 2094 sit_i->dirty_sentries++;
184a5cd2
CY		 2095 return false;
2096 }
2097
2098 return true;
351df4b2
JK		 2099}
2100
2101static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2102 unsigned int segno, int modified)
2103{
2104 struct seg_entry *se = get_seg_entry(sbi, segno);
2105 se->type = type;
2106 if (modified)
2107 __mark_sit_entry_dirty(sbi, segno);
2108}
2109
2110static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2111{
2112 struct seg_entry *se;
2113 unsigned int segno, offset;
2114 long int new_vblocks;
6415fedc
YS		 2115 bool exist;
2116#ifdef CONFIG_F2FS_CHECK_FS
2117 bool mir_exist;
2118#endif
351df4b2
JK		 2119
2120 segno = GET_SEGNO(sbi, blkaddr);
2121
2122 se = get_seg_entry(sbi, segno);
2123 new_vblocks = se->valid_blocks + del;
491c0854 2124 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 2125
9850cf4a 2126 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 2127 (new_vblocks > sbi->blocks_per_seg)));
2128
2129 se->valid_blocks = new_vblocks;
a1f72ac2
CY		 2130 se->mtime = get_mtime(sbi, false);
2131 if (se->mtime > SIT_I(sbi)->max_mtime)
2132 SIT_I(sbi)->max_mtime = se->mtime;
351df4b2
JK		 2133
2134 /* Update valid block bitmap */
2135 if (del > 0) {
6415fedc 2136 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 2137#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2138 mir_exist = f2fs_test_and_set_bit(offset,
2139 se->cur_valid_map_mir);
2140 if (unlikely(exist != mir_exist)) {
2141 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2142 "when setting bitmap, blk:%u, old bit:%d",
2143 blkaddr, exist);
05796763 2144 f2fs_bug_on(sbi, 1);
6415fedc 2145 }
355e7891 2146#endif
6415fedc
YS		 2147 if (unlikely(exist)) {
2148 f2fs_msg(sbi->sb, KERN_ERR,
2149 "Bitmap was wrongly set, blk:%u", blkaddr);
2150 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2151 se->valid_blocks--;
2152 del = 0;
355e7891 2153 }
6415fedc 2154
7d20c8ab 2155 if (!f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 2156 sbi->discard_blks--;
720037f9
JK		 2157
2158 /* don't overwrite by SSR to keep node chain */
4354994f
DR		 2159 if (IS_NODESEG(se->type) &&
2160 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
720037f9
JK		 2161 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2162 se->ckpt_valid_blocks++;
2163 }
351df4b2 2164 } else {
6415fedc 2165 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 2166#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2167 mir_exist = f2fs_test_and_clear_bit(offset,
2168 se->cur_valid_map_mir);
2169 if (unlikely(exist != mir_exist)) {
2170 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2171 "when clearing bitmap, blk:%u, old bit:%d",
2172 blkaddr, exist);
05796763 2173 f2fs_bug_on(sbi, 1);
6415fedc 2174 }
355e7891 2175#endif
6415fedc
YS		 2176 if (unlikely(!exist)) {
2177 f2fs_msg(sbi->sb, KERN_ERR,
2178 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2179 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2180 se->valid_blocks++;
2181 del = 0;
4354994f
DR		 2182 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2183 /*
2184 * If checkpoints are off, we must not reuse data that
2185 * was used in the previous checkpoint. If it was used
2186 * before, we must track that to know how much space we
2187 * really have.
2188 */
c9c8ed50
CY		 2189 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2190 spin_lock(&sbi->stat_lock);
4354994f 2191 sbi->unusable_block_count++;
c9c8ed50
CY		 2192 spin_unlock(&sbi->stat_lock);
2193 }
355e7891 2194 }
6415fedc 2195
7d20c8ab 2196 if (f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 2197 sbi->discard_blks++;
351df4b2
JK		 2198 }
2199 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2200 se->ckpt_valid_blocks += del;
2201
2202 __mark_sit_entry_dirty(sbi, segno);
2203
2204 /* update total number of valid blocks to be written in ckpt area */
2205 SIT_I(sbi)->written_valid_blocks += del;
2206
2c70c5e3 2207 if (__is_large_section(sbi))
351df4b2
JK		 2208 get_sec_entry(sbi, segno)->valid_blocks += del;
2209}
2210
4d57b86d 2211void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 2212{
2213 unsigned int segno = GET_SEGNO(sbi, addr);
2214 struct sit_info *sit_i = SIT_I(sbi);
2215
9850cf4a 2216 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 2217 if (addr == NEW_ADDR)
2218 return;
2219
6aa58d8a
CY		 2220 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2221
351df4b2 2222 /* add it into sit main buffer */
3d26fa6b 2223 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2224
2225 update_sit_entry(sbi, addr, -1);
2226
2227 /* add it into dirty seglist */
2228 locate_dirty_segment(sbi, segno);
2229
3d26fa6b 2230 up_write(&sit_i->sentry_lock);
351df4b2
JK		 2231}
2232
4d57b86d 2233bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 2234{
2235 struct sit_info *sit_i = SIT_I(sbi);
2236 unsigned int segno, offset;
2237 struct seg_entry *se;
2238 bool is_cp = false;
2239
93770ab7 2240 if (!__is_valid_data_blkaddr(blkaddr))
6e2c64ad
JK		 2241 return true;
2242
3d26fa6b 2243 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2244
2245 segno = GET_SEGNO(sbi, blkaddr);
2246 se = get_seg_entry(sbi, segno);
2247 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2248
2249 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2250 is_cp = true;
2251
3d26fa6b 2252 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2253
2254 return is_cp;
2255}
2256
0a8165d7 2257/*
351df4b2
JK		 2258 * This function should be resided under the curseg_mutex lock
2259 */
2260static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 2261 struct f2fs_summary *sum)
351df4b2
JK		 2262{
2263 struct curseg_info *curseg = CURSEG_I(sbi, type);
2264 void *addr = curseg->sum_blk;
e79efe3b 2265 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 2266 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 2267}
2268
0a8165d7 2269/*
351df4b2
JK		 2270 * Calculate the number of current summary pages for writing
2271 */
4d57b86d 2272int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 2273{
351df4b2 2274 int valid_sum_count = 0;
9a47938b 2275 int i, sum_in_page;
351df4b2
JK		 2276
2277 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2278 if (sbi->ckpt->alloc_type[i] == SSR)
2279 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 2280 else {
2281 if (for_ra)
2282 valid_sum_count += le16_to_cpu(
2283 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2284 else
2285 valid_sum_count += curseg_blkoff(sbi, i);
2286 }
351df4b2
JK		 2287 }
2288
09cbfeaf 2289 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 2290 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2291 if (valid_sum_count <= sum_in_page)
351df4b2 2292 return 1;
9a47938b 2293 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 2294 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 2295 return 2;
2296 return 3;
2297}
2298
0a8165d7 2299/*
351df4b2
JK		 2300 * Caller should put this summary page
2301 */
4d57b86d 2302struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 2303{
7735730d 2304 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2305}
2306
4d57b86d
CY		 2307void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2308 void *src, block_t blk_addr)
351df4b2 2309{
4d57b86d 2310 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 2311
0537b811 2312 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2313 set_page_dirty(page);
2314 f2fs_put_page(page, 1);
2315}
2316
381722d2
CY		 2317static void write_sum_page(struct f2fs_sb_info *sbi,
2318 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2319{
4d57b86d 2320 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2321}
2322
b7ad7512
CY		 2323static void write_current_sum_page(struct f2fs_sb_info *sbi,
2324 int type, block_t blk_addr)
2325{
2326 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2327 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2328 struct f2fs_summary_block *src = curseg->sum_blk;
2329 struct f2fs_summary_block *dst;
2330
2331 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2332 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2333
2334 mutex_lock(&curseg->curseg_mutex);
2335
2336 down_read(&curseg->journal_rwsem);
2337 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2338 up_read(&curseg->journal_rwsem);
2339
2340 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2341 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2342
2343 mutex_unlock(&curseg->curseg_mutex);
2344
2345 set_page_dirty(page);
2346 f2fs_put_page(page, 1);
2347}
2348
a7881893
JK		 2349static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2350{
2351 struct curseg_info *curseg = CURSEG_I(sbi, type);
2352 unsigned int segno = curseg->segno + 1;
2353 struct free_segmap_info *free_i = FREE_I(sbi);
2354
2355 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2356 return !test_bit(segno, free_i->free_segmap);
2357 return 0;
2358}
2359
0a8165d7 2360/*
351df4b2
JK		 2361 * Find a new segment from the free segments bitmap to right order
2362 * This function should be returned with success, otherwise BUG
2363 */
2364static void get_new_segment(struct f2fs_sb_info *sbi,
2365 unsigned int *newseg, bool new_sec, int dir)
2366{
2367 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2368 unsigned int segno, secno, zoneno;
7cd8558b 2369 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2370 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2371 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2372 unsigned int left_start = hint;
2373 bool init = true;
2374 int go_left = 0;
2375 int i;
2376
1a118ccf 2377 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2378
2379 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2380 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2381 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2382 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2383 goto got_it;
2384 }
2385find_other_zone:
7cd8558b
JK		 2386 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2387 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2388 if (dir == ALLOC_RIGHT) {
2389 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2390 MAIN_SECS(sbi), 0);
2391 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2392 } else {
2393 go_left = 1;
2394 left_start = hint - 1;
2395 }
2396 }
2397 if (go_left == 0)
2398 goto skip_left;
2399
2400 while (test_bit(left_start, free_i->free_secmap)) {
2401 if (left_start > 0) {
2402 left_start--;
2403 continue;
2404 }
2405 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2406 MAIN_SECS(sbi), 0);
2407 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2408 break;
2409 }
2410 secno = left_start;
2411skip_left:
4ddb1a4d
JK		 2412 segno = GET_SEG_FROM_SEC(sbi, secno);
2413 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2414
2415 /* give up on finding another zone */
2416 if (!init)
2417 goto got_it;
2418 if (sbi->secs_per_zone == 1)
2419 goto got_it;
2420 if (zoneno == old_zoneno)
2421 goto got_it;
2422 if (dir == ALLOC_LEFT) {
2423 if (!go_left && zoneno + 1 >= total_zones)
2424 goto got_it;
2425 if (go_left && zoneno == 0)
2426 goto got_it;
2427 }
2428 for (i = 0; i < NR_CURSEG_TYPE; i++)
2429 if (CURSEG_I(sbi, i)->zone == zoneno)
2430 break;
2431
2432 if (i < NR_CURSEG_TYPE) {
2433 /* zone is in user, try another */
2434 if (go_left)
2435 hint = zoneno * sbi->secs_per_zone - 1;
2436 else if (zoneno + 1 >= total_zones)
2437 hint = 0;
2438 else
2439 hint = (zoneno + 1) * sbi->secs_per_zone;
2440 init = false;
2441 goto find_other_zone;
2442 }
2443got_it:
2444 /* set it as dirty segment in free segmap */
9850cf4a 2445 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2446 __set_inuse(sbi, segno);
2447 *newseg = segno;
1a118ccf 2448 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2449}
2450
2451static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2452{
2453 struct curseg_info *curseg = CURSEG_I(sbi, type);
2454 struct summary_footer *sum_footer;
2455
2456 curseg->segno = curseg->next_segno;
4ddb1a4d 2457 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2458 curseg->next_blkoff = 0;
2459 curseg->next_segno = NULL_SEGNO;
2460
2461 sum_footer = &(curseg->sum_blk->footer);
2462 memset(sum_footer, 0, sizeof(struct summary_footer));
2463 if (IS_DATASEG(type))
2464 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2465 if (IS_NODESEG(type))
2466 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2467 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2468}
2469
7a20b8a6
JK		 2470static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2471{
a7881893 2472 /* if segs_per_sec is large than 1, we need to keep original policy. */
2c70c5e3 2473 if (__is_large_section(sbi))
a7881893
JK		 2474 return CURSEG_I(sbi, type)->segno;
2475
4354994f
DR		 2476 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2477 return 0;
2478
b94929d9
YS		 2479 if (test_opt(sbi, NOHEAP) &&
2480 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2481 return 0;
2482
e066b83c
JK		 2483 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2484 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2485
2486 /* find segments from 0 to reuse freed segments */
63189b78 2487 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2488 return 0;
2489
7a20b8a6
JK		 2490 return CURSEG_I(sbi, type)->segno;
2491}
2492
0a8165d7 2493/*
351df4b2
JK		 2494 * Allocate a current working segment.
2495 * This function always allocates a free segment in LFS manner.
2496 */
2497static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2498{
2499 struct curseg_info *curseg = CURSEG_I(sbi, type);
2500 unsigned int segno = curseg->segno;
2501 int dir = ALLOC_LEFT;
2502
2503 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2504 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2505 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2506 dir = ALLOC_RIGHT;
2507
2508 if (test_opt(sbi, NOHEAP))
2509 dir = ALLOC_RIGHT;
2510
7a20b8a6 2511 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2512 get_new_segment(sbi, &segno, new_sec, dir);
2513 curseg->next_segno = segno;
2514 reset_curseg(sbi, type, 1);
2515 curseg->alloc_type = LFS;
2516}
2517
2518static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2519 struct curseg_info *seg, block_t start)
2520{
2521 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2522 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2523 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2524 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2525 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2526 int i, pos;
2527
2528 for (i = 0; i < entries; i++)
2529 target_map[i] = ckpt_map[i] | cur_map[i];
2530
2531 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2532
2533 seg->next_blkoff = pos;
351df4b2
JK		 2534}
2535
0a8165d7 2536/*
351df4b2
JK		 2537 * If a segment is written by LFS manner, next block offset is just obtained
2538 * by increasing the current block offset. However, if a segment is written by
2539 * SSR manner, next block offset obtained by calling __next_free_blkoff
2540 */
2541static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2542 struct curseg_info *seg)
2543{
2544 if (seg->alloc_type == SSR)
2545 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2546 else
2547 seg->next_blkoff++;
2548}
2549
0a8165d7 2550/*
e1c42045 2551 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2552 * manner, so it should recover the existing segment information of valid blocks
2553 */
025d63a4 2554static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2555{
2556 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2557 struct curseg_info *curseg = CURSEG_I(sbi, type);
2558 unsigned int new_segno = curseg->next_segno;
2559 struct f2fs_summary_block *sum_node;
2560 struct page *sum_page;
2561
2562 write_sum_page(sbi, curseg->sum_blk,
2563 GET_SUM_BLOCK(sbi, curseg->segno));
2564 __set_test_and_inuse(sbi, new_segno);
2565
2566 mutex_lock(&dirty_i->seglist_lock);
2567 __remove_dirty_segment(sbi, new_segno, PRE);
2568 __remove_dirty_segment(sbi, new_segno, DIRTY);
2569 mutex_unlock(&dirty_i->seglist_lock);
2570
2571 reset_curseg(sbi, type, 1);
2572 curseg->alloc_type = SSR;
2573 __next_free_blkoff(sbi, curseg, 0);
2574
4d57b86d 2575 sum_page = f2fs_get_sum_page(sbi, new_segno);
edc55aaf 2576 f2fs_bug_on(sbi, IS_ERR(sum_page));
025d63a4
CY		 2577 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2578 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2579 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2580}
2581
43727527
JK		 2582static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2583{
2584 struct curseg_info *curseg = CURSEG_I(sbi, type);
2585 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2586 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2587 int i, cnt;
2588 bool reversed = false;
c192f7a4 2589
4d57b86d 2590 /* f2fs_need_SSR() already forces to do this */
e066b83c
JK		 2591 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2592 curseg->next_segno = segno;
c192f7a4 2593 return 1;
e066b83c 2594 }
43727527 2595
70d625cb
JK		 2596 /* For node segments, let's do SSR more intensively */
2597 if (IS_NODESEG(type)) {
d27c3d89
CY		 2598 if (type >= CURSEG_WARM_NODE) {
2599 reversed = true;
2600 i = CURSEG_COLD_NODE;
2601 } else {
2602 i = CURSEG_HOT_NODE;
2603 }
2604 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2605 } else {
d27c3d89
CY		 2606 if (type >= CURSEG_WARM_DATA) {
2607 reversed = true;
2608 i = CURSEG_COLD_DATA;
2609 } else {
2610 i = CURSEG_HOT_DATA;
2611 }
2612 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2613 }
43727527 2614
d27c3d89 2615 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2616 if (i == type)
2617 continue;
e066b83c
JK		 2618 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2619 curseg->next_segno = segno;
43727527 2620 return 1;
e066b83c 2621 }
c192f7a4 2622 }
4354994f
DR		 2623
2624 /* find valid_blocks=0 in dirty list */
2625 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2626 segno = get_free_segment(sbi);
2627 if (segno != NULL_SEGNO) {
2628 curseg->next_segno = segno;
2629 return 1;
2630 }
2631 }
43727527
JK		 2632 return 0;
2633}
2634
351df4b2
JK		 2635/*
2636 * flush out current segment and replace it with new segment
2637 * This function should be returned with success, otherwise BUG
2638 */
2639static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2640 int type, bool force)
2641{
a7881893
JK		 2642 struct curseg_info *curseg = CURSEG_I(sbi, type);
2643
7b405275 2644 if (force)
351df4b2 2645 new_curseg(sbi, type, true);
5b6c6be2
JK		 2646 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2647 type == CURSEG_WARM_NODE)
351df4b2 2648 new_curseg(sbi, type, false);
4354994f
DR		 2649 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2650 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
a7881893 2651 new_curseg(sbi, type, false);
4d57b86d 2652 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2653 change_curseg(sbi, type);
351df4b2
JK		 2654 else
2655 new_curseg(sbi, type, false);
dcdfff65 2656
a7881893 2657 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2658}
2659
04f0b2ea
QS		 2660void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2661 unsigned int start, unsigned int end)
2662{
2663 struct curseg_info *curseg = CURSEG_I(sbi, type);
2664 unsigned int segno;
2665
2666 down_read(&SM_I(sbi)->curseg_lock);
2667 mutex_lock(&curseg->curseg_mutex);
2668 down_write(&SIT_I(sbi)->sentry_lock);
2669
2670 segno = CURSEG_I(sbi, type)->segno;
2671 if (segno < start || segno > end)
2672 goto unlock;
2673
2674 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2675 change_curseg(sbi, type);
2676 else
2677 new_curseg(sbi, type, true);
2678
2679 stat_inc_seg_type(sbi, curseg);
2680
2681 locate_dirty_segment(sbi, segno);
2682unlock:
2683 up_write(&SIT_I(sbi)->sentry_lock);
2684
2685 if (segno != curseg->segno)
2686 f2fs_msg(sbi->sb, KERN_NOTICE,
2687 "For resize: curseg of type %d: %u ==> %u",
2688 type, segno, curseg->segno);
2689
2690 mutex_unlock(&curseg->curseg_mutex);
2691 up_read(&SM_I(sbi)->curseg_lock);
2692}
2693
4d57b86d 2694void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
351df4b2 2695{
6ae1be13
JK		 2696 struct curseg_info *curseg;
2697 unsigned int old_segno;
351df4b2
JK		 2698 int i;
2699
3d26fa6b
CY		 2700 down_write(&SIT_I(sbi)->sentry_lock);
2701
6ae1be13
JK		 2702 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2703 curseg = CURSEG_I(sbi, i);
2704 old_segno = curseg->segno;
2705 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2706 locate_dirty_segment(sbi, old_segno);
2707 }
3d26fa6b
CY		 2708
2709 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2710}
2711
2712static const struct segment_allocation default_salloc_ops = {
2713 .allocate_segment = allocate_segment_by_default,
2714};
2715
4d57b86d
CY		 2716bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2717 struct cp_control *cpc)
25290fa5
JK		 2718{
2719 __u64 trim_start = cpc->trim_start;
2720 bool has_candidate = false;
2721
3d26fa6b 2722 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2723 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2724 if (add_discard_addrs(sbi, cpc, true)) {
2725 has_candidate = true;
2726 break;
2727 }
2728 }
3d26fa6b 2729 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2730
2731 cpc->trim_start = trim_start;
2732 return has_candidate;
2733}
2734
01f9cf6d 2735static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 2736 struct discard_policy *dpolicy,
2737 unsigned int start, unsigned int end)
2738{
2739 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2740 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2741 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2742 struct discard_cmd *dc;
2743 struct blk_plug plug;
2744 int issued;
01f9cf6d 2745 unsigned int trimmed = 0;
9a997188
JK		 2746
2747next:
2748 issued = 0;
2749
2750 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 2751 if (unlikely(dcc->rbtree_check))
2752 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2753 &dcc->root));
9a997188 2754
4d57b86d 2755 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 2756 NULL, start,
2757 (struct rb_entry **)&prev_dc,
2758 (struct rb_entry **)&next_dc,
4dada3fd 2759 &insert_p, &insert_parent, true, NULL);
9a997188
JK		 2760 if (!dc)
2761 dc = next_dc;
2762
2763 blk_start_plug(&plug);
2764
2765 while (dc && dc->lstart <= end) {
2766 struct rb_node *node;
6b9cb124 2767 int err = 0;
9a997188
JK		 2768
2769 if (dc->len < dpolicy->granularity)
2770 goto skip;
2771
2772 if (dc->state != D_PREP) {
2773 list_move_tail(&dc->list, &dcc->fstrim_list);
2774 goto skip;
2775 }
2776
6b9cb124 2777 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
9a997188 2778
35ec7d57 2779 if (issued >= dpolicy->max_requests) {
9a997188
JK		 2780 start = dc->lstart + dc->len;
2781
6b9cb124
CY		 2782 if (err)
2783 __remove_discard_cmd(sbi, dc);
2784
9a997188
JK		 2785 blk_finish_plug(&plug);
2786 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 2787 trimmed += __wait_all_discard_cmd(sbi, NULL);
9a997188
JK		 2788 congestion_wait(BLK_RW_ASYNC, HZ/50);
2789 goto next;
2790 }
2791skip:
2792 node = rb_next(&dc->rb_node);
6b9cb124
CY		 2793 if (err)
2794 __remove_discard_cmd(sbi, dc);
9a997188
JK		 2795 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2796
2797 if (fatal_signal_pending(current))
2798 break;
2799 }
2800
2801 blk_finish_plug(&plug);
2802 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 2803
2804 return trimmed;
9a997188
JK		 2805}
2806
4b2fecc8
JK		 2807int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2808{
f7ef9b83
JK		 2809 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2810 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2811 unsigned int start_segno, end_segno;
8412663d 2812 block_t start_block, end_block;
4b2fecc8 2813 struct cp_control cpc;
78997b56 2814 struct discard_policy dpolicy;
0ea80512 2815 unsigned long long trimmed = 0;
c34f42e2 2816 int err = 0;
2c70c5e3 2817 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
4b2fecc8 2818
836b5a63 2819 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2820 return -EINVAL;
2821
3f16ecd9
CY		 2822 if (end < MAIN_BLKADDR(sbi))
2823 goto out;
4b2fecc8 2824
ed214a11
YH		 2825 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2826 f2fs_msg(sbi->sb, KERN_WARNING,
2827 "Found FS corruption, run fsck to fix.");
3d165dc3 2828 return -EIO;
ed214a11
YH		 2829 }
2830
4b2fecc8 2831 /* start/end segment number in main_area */
7cd8558b
JK		 2832 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2833 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2834 GET_SEGNO(sbi, end);
ad6672bb
YS		 2835 if (need_align) {
2836 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2837 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2838 }
8412663d 2839
4b2fecc8 2840 cpc.reason = CP_DISCARD;
836b5a63 2841 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2842 cpc.trim_start = start_segno;
2843 cpc.trim_end = end_segno;
4b2fecc8 2844
377224c4
CY		 2845 if (sbi->discard_blks == 0)
2846 goto out;
74fa5f3d 2847
377224c4 2848 mutex_lock(&sbi->gc_mutex);
4d57b86d 2849 err = f2fs_write_checkpoint(sbi, &cpc);
377224c4
CY		 2850 mutex_unlock(&sbi->gc_mutex);
2851 if (err)
2852 goto out;
8412663d 2853
e555da9f
JK		 2854 /*
2855 * We filed discard candidates, but actually we don't need to wait for
2856 * all of them, since they'll be issued in idle time along with runtime
2857 * discard option. User configuration looks like using runtime discard
2858 * or periodic fstrim instead of it.
2859 */
7d20c8ab 2860 if (f2fs_realtime_discard_enable(sbi))
5a615492
JK		 2861 goto out;
2862
2863 start_block = START_BLOCK(sbi, start_segno);
2864 end_block = START_BLOCK(sbi, end_segno + 1);
2865
2866 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 2867 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2868 start_block, end_block);
5a615492 2869
01f9cf6d 2870 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2871 start_block, end_block);
377224c4 2872out:
6eae2694
CY		 2873 if (!err)
2874 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2875 return err;
4b2fecc8
JK		 2876}
2877
351df4b2
JK		 2878static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2879{
2880 struct curseg_info *curseg = CURSEG_I(sbi, type);
2881 if (curseg->next_blkoff < sbi->blocks_per_seg)
2882 return true;
2883 return false;
2884}
2885
4d57b86d 2886int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 2887{
2888 switch (hint) {
2889 case WRITE_LIFE_SHORT:
2890 return CURSEG_HOT_DATA;
2891 case WRITE_LIFE_EXTREME:
2892 return CURSEG_COLD_DATA;
2893 default:
2894 return CURSEG_WARM_DATA;
2895 }
2896}
2897
0cdd3195
HL		 2898/* This returns write hints for each segment type. This hints will be
2899 * passed down to block layer. There are mapping tables which depend on
2900 * the mount option 'whint_mode'.
2901 *
2902 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2903 *
2904 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2905 *
2906 * User F2FS Block
2907 * ---- ---- -----
2908 * META WRITE_LIFE_NOT_SET
2909 * HOT_NODE "
2910 * WARM_NODE "
2911 * COLD_NODE "
2912 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2913 * extension list " "
2914 *
2915 * -- buffered io
2916 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2917 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2918 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2919 * WRITE_LIFE_NONE " "
2920 * WRITE_LIFE_MEDIUM " "
2921 * WRITE_LIFE_LONG " "
2922 *
2923 * -- direct io
2924 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2925 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2926 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2927 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2928 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2929 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2930 *
f2e703f9
HL		 2931 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2932 *
2933 * User F2FS Block
2934 * ---- ---- -----
2935 * META WRITE_LIFE_MEDIUM;
2936 * HOT_NODE WRITE_LIFE_NOT_SET
2937 * WARM_NODE "
2938 * COLD_NODE WRITE_LIFE_NONE
2939 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2940 * extension list " "
2941 *
2942 * -- buffered io
2943 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2944 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2945 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2946 * WRITE_LIFE_NONE " "
2947 * WRITE_LIFE_MEDIUM " "
2948 * WRITE_LIFE_LONG " "
2949 *
2950 * -- direct io
2951 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2952 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2953 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2954 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2955 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2956 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2957 */
2958
4d57b86d 2959enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 2960 enum page_type type, enum temp_type temp)
2961{
63189b78 2962 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2963 if (type == DATA) {
f2e703f9 2964 if (temp == WARM)
0cdd3195 2965 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2966 else if (temp == HOT)
2967 return WRITE_LIFE_SHORT;
2968 else if (temp == COLD)
2969 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2970 } else {
2971 return WRITE_LIFE_NOT_SET;
2972 }
63189b78 2973 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2974 if (type == DATA) {
2975 if (temp == WARM)
2976 return WRITE_LIFE_LONG;
2977 else if (temp == HOT)
2978 return WRITE_LIFE_SHORT;
2979 else if (temp == COLD)
2980 return WRITE_LIFE_EXTREME;
2981 } else if (type == NODE) {
2982 if (temp == WARM || temp == HOT)
2983 return WRITE_LIFE_NOT_SET;
2984 else if (temp == COLD)
2985 return WRITE_LIFE_NONE;
2986 } else if (type == META) {
2987 return WRITE_LIFE_MEDIUM;
2988 }
0cdd3195 2989 }
f2e703f9 2990 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2991}
2992
81377bd6 2993static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2994{
81377bd6 2995 if (fio->type == DATA)
351df4b2
JK		 2996 return CURSEG_HOT_DATA;
2997 else
2998 return CURSEG_HOT_NODE;
2999}
3000
81377bd6 3001static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 3002{
81377bd6
JK		 3003 if (fio->type == DATA) {
3004 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 3005
3006 if (S_ISDIR(inode->i_mode))
3007 return CURSEG_HOT_DATA;
3008 else
3009 return CURSEG_COLD_DATA;
3010 } else {
81377bd6 3011 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 3012 return CURSEG_WARM_NODE;
351df4b2
JK		 3013 else
3014 return CURSEG_COLD_NODE;
3015 }
3016}
3017
81377bd6 3018static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 3019{
81377bd6
JK		 3020 if (fio->type == DATA) {
3021 struct inode *inode = fio->page->mapping->host;
351df4b2 3022
81377bd6 3023 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 3024 return CURSEG_COLD_DATA;
b6a06cbb 3025 if (file_is_hot(inode) ||
b4c3ca8b 3026 is_inode_flag_set(inode, FI_HOT_DATA) ||
2079f115
CY		 3027 f2fs_is_atomic_file(inode) ||
3028 f2fs_is_volatile_file(inode))
ef095d19 3029 return CURSEG_HOT_DATA;
4d57b86d 3030 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 3031 } else {
81377bd6
JK		 3032 if (IS_DNODE(fio->page))
3033 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 3034 CURSEG_HOT_NODE;
ef095d19 3035 return CURSEG_COLD_NODE;
351df4b2
JK		 3036 }
3037}
3038
81377bd6 3039static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 3040{
a912b54d
JK		 3041 int type = 0;
3042
63189b78 3043 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 3044 case 2:
a912b54d
JK		 3045 type = __get_segment_type_2(fio);
3046 break;
351df4b2 3047 case 4:
a912b54d
JK		 3048 type = __get_segment_type_4(fio);
3049 break;
3050 case 6:
3051 type = __get_segment_type_6(fio);
3052 break;
3053 default:
3054 f2fs_bug_on(fio->sbi, true);
351df4b2 3055 }
81377bd6 3056
a912b54d
JK		 3057 if (IS_HOT(type))
3058 fio->temp = HOT;
3059 else if (IS_WARM(type))
3060 fio->temp = WARM;
3061 else
3062 fio->temp = COLD;
3063 return type;
351df4b2
JK		 3064}
3065
4d57b86d 3066void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 3067 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 3068 struct f2fs_summary *sum, int type,
3069 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 3070{
3071 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 3072 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 3073
2b60311d
CY		 3074 down_read(&SM_I(sbi)->curseg_lock);
3075
351df4b2 3076 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3077 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3078
3079 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 3080
4e6a8d9b
JK		 3081 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3082
351df4b2
JK		 3083 /*
3084 * __add_sum_entry should be resided under the curseg_mutex
3085 * because, this function updates a summary entry in the
3086 * current summary block.
3087 */
e79efe3b 3088 __add_sum_entry(sbi, type, sum);
351df4b2 3089
351df4b2 3090 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 3091
3092 stat_inc_block_count(sbi, curseg);
351df4b2 3093
65f1b80b
YS		 3094 /*
3095 * SIT information should be updated before segment allocation,
3096 * since SSR needs latest valid block information.
3097 */
3098 update_sit_entry(sbi, *new_blkaddr, 1);
3099 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3100 update_sit_entry(sbi, old_blkaddr, -1);
3101
c6f82fe9
JK		 3102 if (!__has_curseg_space(sbi, type))
3103 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 3104
351df4b2 3105 /*
65f1b80b
YS		 3106 * segment dirty status should be updated after segment allocation,
3107 * so we just need to update status only one time after previous
3108 * segment being closed.
351df4b2 3109 */
65f1b80b
YS		 3110 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3111 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 3112
3d26fa6b 3113 up_write(&sit_i->sentry_lock);
351df4b2 3114
704956ec 3115 if (page && IS_NODESEG(type)) {
351df4b2
JK		 3116 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3117
704956ec
CY		 3118 f2fs_inode_chksum_set(sbi, page);
3119 }
3120
fb830fc5
CY		 3121 if (add_list) {
3122 struct f2fs_bio_info *io;
3123
3124 INIT_LIST_HEAD(&fio->list);
3125 fio->in_list = true;
fe16efe6 3126 fio->retry = false;
fb830fc5
CY		 3127 io = sbi->write_io[fio->type] + fio->temp;
3128 spin_lock(&io->io_lock);
3129 list_add_tail(&fio->list, &io->io_list);
3130 spin_unlock(&io->io_lock);
3131 }
3132
bfad7c2d 3133 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 3134
3135 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 3136}
3137
39d787be
CY		 3138static void update_device_state(struct f2fs_io_info *fio)
3139{
3140 struct f2fs_sb_info *sbi = fio->sbi;
3141 unsigned int devidx;
3142
0916878d 3143 if (!f2fs_is_multi_device(sbi))
39d787be
CY		 3144 return;
3145
3146 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3147
3148 /* update device state for fsync */
4d57b86d 3149 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 3150
3151 /* update device state for checkpoint */
3152 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3153 spin_lock(&sbi->dev_lock);
3154 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3155 spin_unlock(&sbi->dev_lock);
3156 }
39d787be
CY		 3157}
3158
05ca3632 3159static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 3160{
81377bd6 3161 int type = __get_segment_type(fio);
107a805d 3162 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 3163
107a805d
CY		 3164 if (keep_order)
3165 down_read(&fio->sbi->io_order_lock);
0a595eba 3166reallocate:
4d57b86d 3167 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 3168 &fio->new_blkaddr, sum, type, fio, true);
6aa58d8a
CY		 3169 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3170 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3171 fio->old_blkaddr, fio->old_blkaddr);
bfad7c2d 3172
351df4b2 3173 /* writeout dirty page into bdev */
fe16efe6
CY		 3174 f2fs_submit_page_write(fio);
3175 if (fio->retry) {
0a595eba
JK		 3176 fio->old_blkaddr = fio->new_blkaddr;
3177 goto reallocate;
3178 }
fe16efe6
CY		 3179
3180 update_device_state(fio);
3181
107a805d
CY		 3182 if (keep_order)
3183 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 3184}
3185
4d57b86d 3186void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 3187 enum iostat_type io_type)
351df4b2 3188{
458e6197 3189 struct f2fs_io_info fio = {
05ca3632 3190 .sbi = sbi,
458e6197 3191 .type = META,
0cdd3195 3192 .temp = HOT,
04d328de 3193 .op = REQ_OP_WRITE,
70fd7614 3194 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 3195 .old_blkaddr = page->index,
3196 .new_blkaddr = page->index,
05ca3632 3197 .page = page,
4375a336 3198 .encrypted_page = NULL,
fb830fc5 3199 .in_list = false,
458e6197
JK		 3200 };
3201
2b947003 3202 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 3203 fio.op_flags &= ~REQ_META;
2b947003 3204
351df4b2 3205 set_page_writeback(page);
17c50035 3206 ClearPageError(page);
b9109b0e 3207 f2fs_submit_page_write(&fio);
b0af6d49 3208
b63e7be5 3209 stat_inc_meta_count(sbi, page->index);
b0af6d49 3210 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 3211}
3212
4d57b86d 3213void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 3214{
3215 struct f2fs_summary sum;
05ca3632 3216
351df4b2 3217 set_summary(&sum, nid, 0, 0);
05ca3632 3218 do_write_page(&sum, fio);
b0af6d49
CY		 3219
3220 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3221}
3222
4d57b86d
CY		 3223void f2fs_outplace_write_data(struct dnode_of_data *dn,
3224 struct f2fs_io_info *fio)
351df4b2 3225{
05ca3632 3226 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2 3227 struct f2fs_summary sum;
351df4b2 3228
9850cf4a 3229 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
7735730d 3230 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
05ca3632 3231 do_write_page(&sum, fio);
f28b3434 3232 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 3233
3234 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3235}
3236
4d57b86d 3237int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 3238{
b0af6d49 3239 int err;
d21b0f23 3240 struct f2fs_sb_info *sbi = fio->sbi;
05573d6c 3241 unsigned int segno;
b0af6d49 3242
7a9d7548 3243 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 3244 /* i/o temperature is needed for passing down write hints */
3245 __get_segment_type(fio);
d21b0f23 3246
05573d6c
CY		 3247 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3248
3249 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3250 set_sbi_flag(sbi, SBI_NEED_FSCK);
3251 return -EFAULT;
3252 }
d21b0f23 3253
05ca3632 3254 stat_inc_inplace_blocks(fio->sbi);
b0af6d49 3255
8648de2c
CY		 3256 if (fio->bio)
3257 err = f2fs_merge_page_bio(fio);
3258 else
3259 err = f2fs_submit_page_bio(fio);
e46f6bd8 3260 if (!err) {
39d787be 3261 update_device_state(fio);
e46f6bd8
CY		 3262 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3263 }
b0af6d49
CY		 3264
3265 return err;
351df4b2
JK		 3266}
3267
2b60311d
CY		 3268static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3269 unsigned int segno)
3270{
3271 int i;
3272
3273 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3274 if (CURSEG_I(sbi, i)->segno == segno)
3275 break;
3276 }
3277 return i;
3278}
3279
4d57b86d 3280void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 3281 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 3282 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 3283{
3284 struct sit_info *sit_i = SIT_I(sbi);
3285 struct curseg_info *curseg;
3286 unsigned int segno, old_cursegno;
3287 struct seg_entry *se;
3288 int type;
19f106bc 3289 unsigned short old_blkoff;
351df4b2
JK		 3290
3291 segno = GET_SEGNO(sbi, new_blkaddr);
3292 se = get_seg_entry(sbi, segno);
3293 type = se->type;
3294
2b60311d
CY		 3295 down_write(&SM_I(sbi)->curseg_lock);
3296
19f106bc
CY		 3297 if (!recover_curseg) {
3298 /* for recovery flow */
3299 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3300 if (old_blkaddr == NULL_ADDR)
3301 type = CURSEG_COLD_DATA;
3302 else
3303 type = CURSEG_WARM_DATA;
3304 }
3305 } else {
2b60311d
CY		 3306 if (IS_CURSEG(sbi, segno)) {
3307 /* se->type is volatile as SSR allocation */
3308 type = __f2fs_get_curseg(sbi, segno);
3309 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3310 } else {
351df4b2 3311 type = CURSEG_WARM_DATA;
2b60311d 3312 }
351df4b2 3313 }
19f106bc 3314
2c190504 3315 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 3316 curseg = CURSEG_I(sbi, type);
3317
3318 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3319 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3320
3321 old_cursegno = curseg->segno;
19f106bc 3322 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 3323
3324 /* change the current segment */
3325 if (segno != curseg->segno) {
3326 curseg->next_segno = segno;
025d63a4 3327 change_curseg(sbi, type);
351df4b2
JK		 3328 }
3329
491c0854 3330 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 3331 __add_sum_entry(sbi, type, sum);
351df4b2 3332
28bc106b 3333 if (!recover_curseg || recover_newaddr)
6e2c64ad 3334 update_sit_entry(sbi, new_blkaddr, 1);
6aa58d8a
CY		 3335 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3336 invalidate_mapping_pages(META_MAPPING(sbi),
3337 old_blkaddr, old_blkaddr);
6e2c64ad 3338 update_sit_entry(sbi, old_blkaddr, -1);
6aa58d8a 3339 }
6e2c64ad
JK		 3340
3341 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3342 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3343
351df4b2 3344 locate_dirty_segment(sbi, old_cursegno);
351df4b2 3345
19f106bc
CY		 3346 if (recover_curseg) {
3347 if (old_cursegno != curseg->segno) {
3348 curseg->next_segno = old_cursegno;
025d63a4 3349 change_curseg(sbi, type);
19f106bc
CY		 3350 }
3351 curseg->next_blkoff = old_blkoff;
3352 }
3353
3d26fa6b 3354 up_write(&sit_i->sentry_lock);
351df4b2 3355 mutex_unlock(&curseg->curseg_mutex);
2b60311d 3356 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 3357}
3358
528e3459
CY		 3359void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3360 block_t old_addr, block_t new_addr,
28bc106b
CY		 3361 unsigned char version, bool recover_curseg,
3362 bool recover_newaddr)
528e3459
CY		 3363{
3364 struct f2fs_summary sum;
3365
3366 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3367
4d57b86d 3368 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
28bc106b 3369 recover_curseg, recover_newaddr);
528e3459 3370
f28b3434 3371 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 3372}
3373
93dfe2ac 3374void f2fs_wait_on_page_writeback(struct page *page,
bae0ee7a 3375 enum page_type type, bool ordered, bool locked)
93dfe2ac 3376{
93dfe2ac 3377 if (PageWriteback(page)) {
4081363f
JK		 3378 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3379
bab475c5 3380 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
bae0ee7a 3381 if (ordered) {
fec1d657 3382 wait_on_page_writeback(page);
bae0ee7a
CY		 3383 f2fs_bug_on(sbi, locked && PageWriteback(page));
3384 } else {
fec1d657 3385 wait_for_stable_page(page);
bae0ee7a 3386 }
93dfe2ac
JK		 3387 }
3388}
3389
0ded69f6 3390void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
08b39fbd 3391{
0ded69f6 3392 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
08b39fbd
CY		 3393 struct page *cpage;
3394
0ded69f6
JK		 3395 if (!f2fs_post_read_required(inode))
3396 return;
3397
93770ab7 3398 if (!__is_valid_data_blkaddr(blkaddr))
08b39fbd
CY		 3399 return;
3400
08b39fbd
CY		 3401 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3402 if (cpage) {
bae0ee7a 3403 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
08b39fbd
CY		 3404 f2fs_put_page(cpage, 1);
3405 }
3406}
3407
1e78e8bd
ST		 3408void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3409 block_t len)
3410{
3411 block_t i;
3412
3413 for (i = 0; i < len; i++)
3414 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3415}
3416
7735730d 3417static int read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3418{
3419 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3420 struct curseg_info *seg_i;
3421 unsigned char *kaddr;
3422 struct page *page;
3423 block_t start;
3424 int i, j, offset;
3425
3426 start = start_sum_block(sbi);
3427
4d57b86d 3428 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3429 if (IS_ERR(page))
3430 return PTR_ERR(page);
351df4b2
JK		 3431 kaddr = (unsigned char *)page_address(page);
3432
3433 /* Step 1: restore nat cache */
3434 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3435 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3436
3437 /* Step 2: restore sit cache */
3438 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3439 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3440 offset = 2 * SUM_JOURNAL_SIZE;
3441
3442 /* Step 3: restore summary entries */
3443 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3444 unsigned short blk_off;
3445 unsigned int segno;
3446
3447 seg_i = CURSEG_I(sbi, i);
3448 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3449 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3450 seg_i->next_segno = segno;
3451 reset_curseg(sbi, i, 0);
3452 seg_i->alloc_type = ckpt->alloc_type[i];
3453 seg_i->next_blkoff = blk_off;
3454
3455 if (seg_i->alloc_type == SSR)
3456 blk_off = sbi->blocks_per_seg;
3457
3458 for (j = 0; j < blk_off; j++) {
3459 struct f2fs_summary *s;
3460 s = (struct f2fs_summary *)(kaddr + offset);
3461 seg_i->sum_blk->entries[j] = *s;
3462 offset += SUMMARY_SIZE;
09cbfeaf 3463 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3464 SUM_FOOTER_SIZE)
3465 continue;
3466
3467 f2fs_put_page(page, 1);
3468 page = NULL;
3469
4d57b86d 3470 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3471 if (IS_ERR(page))
3472 return PTR_ERR(page);
351df4b2
JK		 3473 kaddr = (unsigned char *)page_address(page);
3474 offset = 0;
3475 }
3476 }
3477 f2fs_put_page(page, 1);
7735730d 3478 return 0;
351df4b2
JK		 3479}
3480
3481static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3482{
3483 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3484 struct f2fs_summary_block *sum;
3485 struct curseg_info *curseg;
3486 struct page *new;
3487 unsigned short blk_off;
3488 unsigned int segno = 0;
3489 block_t blk_addr = 0;
7735730d 3490 int err = 0;
351df4b2
JK		 3491
3492 /* get segment number and block addr */
3493 if (IS_DATASEG(type)) {
3494 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3495 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3496 CURSEG_HOT_DATA]);
119ee914 3497 if (__exist_node_summaries(sbi))
351df4b2
JK		 3498 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3499 else
3500 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3501 } else {
3502 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3503 CURSEG_HOT_NODE]);
3504 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3505 CURSEG_HOT_NODE]);
119ee914 3506 if (__exist_node_summaries(sbi))
351df4b2
JK		 3507 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3508 type - CURSEG_HOT_NODE);
3509 else
3510 blk_addr = GET_SUM_BLOCK(sbi, segno);
3511 }
3512
4d57b86d 3513 new = f2fs_get_meta_page(sbi, blk_addr);
7735730d
CY		 3514 if (IS_ERR(new))
3515 return PTR_ERR(new);
351df4b2
JK		 3516 sum = (struct f2fs_summary_block *)page_address(new);
3517
3518 if (IS_NODESEG(type)) {
119ee914 3519 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3520 struct f2fs_summary *ns = &sum->entries[0];
3521 int i;
3522 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3523 ns->version = 0;
3524 ns->ofs_in_node = 0;
3525 }
3526 } else {
7735730d
CY		 3527 err = f2fs_restore_node_summary(sbi, segno, sum);
3528 if (err)
3529 goto out;
351df4b2
JK		 3530 }
3531 }
3532
3533 /* set uncompleted segment to curseg */
3534 curseg = CURSEG_I(sbi, type);
3535 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3536
3537 /* update journal info */
3538 down_write(&curseg->journal_rwsem);
3539 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3540 up_write(&curseg->journal_rwsem);
3541
3542 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3543 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3544 curseg->next_segno = segno;
3545 reset_curseg(sbi, type, 0);
3546 curseg->alloc_type = ckpt->alloc_type[type];
3547 curseg->next_blkoff = blk_off;
3548 mutex_unlock(&curseg->curseg_mutex);
7735730d 3549out:
351df4b2 3550 f2fs_put_page(new, 1);
7735730d 3551 return err;
351df4b2
JK		 3552}
3553
3554static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3555{
21d3f8e1
JQ		 3556 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3557 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3558 int type = CURSEG_HOT_DATA;
e4fc5fbf 3559 int err;
351df4b2 3560
aaec2b1d 3561 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3562 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3563
3564 if (npages >= 2)
4d57b86d 3565 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3566 META_CP, true);
3fa06d7b 3567
351df4b2 3568 /* restore for compacted data summary */
7735730d
CY		 3569 err = read_compacted_summaries(sbi);
3570 if (err)
3571 return err;
351df4b2
JK		 3572 type = CURSEG_HOT_NODE;
3573 }
3574
119ee914 3575 if (__exist_node_summaries(sbi))
4d57b86d 3576 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3577 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3578
e4fc5fbf
CY		 3579 for (; type <= CURSEG_COLD_NODE; type++) {
3580 err = read_normal_summaries(sbi, type);
3581 if (err)
3582 return err;
3583 }
3584
21d3f8e1
JQ		 3585 /* sanity check for summary blocks */
3586 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
9227d522
ST		 3587 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3588 f2fs_msg(sbi->sb, KERN_ERR,
3589 "invalid journal entries nats %u sits %u\n",
3590 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
21d3f8e1 3591 return -EINVAL;
9227d522 3592 }
21d3f8e1 3593
351df4b2
JK		 3594 return 0;
3595}
3596
3597static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3598{
3599 struct page *page;
3600 unsigned char *kaddr;
3601 struct f2fs_summary *summary;
3602 struct curseg_info *seg_i;
3603 int written_size = 0;
3604 int i, j;
3605
4d57b86d 3606 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3607 kaddr = (unsigned char *)page_address(page);
81114baa 3608 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3609
3610 /* Step 1: write nat cache */
3611 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3612 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3613 written_size += SUM_JOURNAL_SIZE;
3614
3615 /* Step 2: write sit cache */
3616 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3617 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3618 written_size += SUM_JOURNAL_SIZE;
3619
351df4b2
JK		 3620 /* Step 3: write summary entries */
3621 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3622 unsigned short blkoff;
3623 seg_i = CURSEG_I(sbi, i);
3624 if (sbi->ckpt->alloc_type[i] == SSR)
3625 blkoff = sbi->blocks_per_seg;
3626 else
3627 blkoff = curseg_blkoff(sbi, i);
3628
3629 for (j = 0; j < blkoff; j++) {
3630 if (!page) {
4d57b86d 3631 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3632 kaddr = (unsigned char *)page_address(page);
81114baa 3633 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3634 written_size = 0;
3635 }
3636 summary = (struct f2fs_summary *)(kaddr + written_size);
3637 *summary = seg_i->sum_blk->entries[j];
3638 written_size += SUMMARY_SIZE;
351df4b2 3639
09cbfeaf 3640 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3641 SUM_FOOTER_SIZE)
3642 continue;
3643
e8d61a74 3644 set_page_dirty(page);
351df4b2
JK		 3645 f2fs_put_page(page, 1);
3646 page = NULL;
3647 }
3648 }
e8d61a74
CY		 3649 if (page) {
3650 set_page_dirty(page);
351df4b2 3651 f2fs_put_page(page, 1);
e8d61a74 3652 }
351df4b2
JK		 3653}
3654
3655static void write_normal_summaries(struct f2fs_sb_info *sbi,
3656 block_t blkaddr, int type)
3657{
3658 int i, end;
3659 if (IS_DATASEG(type))
3660 end = type + NR_CURSEG_DATA_TYPE;
3661 else
3662 end = type + NR_CURSEG_NODE_TYPE;
3663
b7ad7512
CY		 3664 for (i = type; i < end; i++)
3665 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3666}
3667
4d57b86d 3668void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3669{
aaec2b1d 3670 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3671 write_compacted_summaries(sbi, start_blk);
3672 else
3673 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3674}
3675
4d57b86d 3676void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3677{
119ee914 3678 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3679}
3680
4d57b86d 3681int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3682 unsigned int val, int alloc)
3683{
3684 int i;
3685
3686 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3687 for (i = 0; i < nats_in_cursum(journal); i++) {
3688 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3689 return i;
3690 }
dfc08a12
CY		 3691 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3692 return update_nats_in_cursum(journal, 1);
351df4b2 3693 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3694 for (i = 0; i < sits_in_cursum(journal); i++)
3695 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3696 return i;
dfc08a12
CY		 3697 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3698 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3699 }
3700 return -1;
3701}
3702
3703static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3704 unsigned int segno)
3705{
7735730d 3706 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3707}
3708
3709static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3710 unsigned int start)
3711{
3712 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3713 struct page *page;
351df4b2 3714 pgoff_t src_off, dst_off;
351df4b2
JK		 3715
3716 src_off = current_sit_addr(sbi, start);
3717 dst_off = next_sit_addr(sbi, src_off);
3718
4d57b86d 3719 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 3720 seg_info_to_sit_page(sbi, page, start);
351df4b2 3721
068c3cd8 3722 set_page_dirty(page);
351df4b2
JK		 3723 set_to_next_sit(sit_i, start);
3724
068c3cd8 3725 return page;
351df4b2
JK		 3726}
3727
184a5cd2
CY		 3728static struct sit_entry_set *grab_sit_entry_set(void)
3729{
3730 struct sit_entry_set *ses =
80c54505 3731 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3732
3733 ses->entry_cnt = 0;
3734 INIT_LIST_HEAD(&ses->set_list);
3735 return ses;
3736}
3737
3738static void release_sit_entry_set(struct sit_entry_set *ses)
3739{
3740 list_del(&ses->set_list);
3741 kmem_cache_free(sit_entry_set_slab, ses);
3742}
3743
3744static void adjust_sit_entry_set(struct sit_entry_set *ses,
3745 struct list_head *head)
3746{
3747 struct sit_entry_set *next = ses;
3748
3749 if (list_is_last(&ses->set_list, head))
3750 return;
3751
3752 list_for_each_entry_continue(next, head, set_list)
3753 if (ses->entry_cnt <= next->entry_cnt)
3754 break;
3755
3756 list_move_tail(&ses->set_list, &next->set_list);
3757}
3758
3759static void add_sit_entry(unsigned int segno, struct list_head *head)
3760{
3761 struct sit_entry_set *ses;
3762 unsigned int start_segno = START_SEGNO(segno);
3763
3764 list_for_each_entry(ses, head, set_list) {
3765 if (ses->start_segno == start_segno) {
3766 ses->entry_cnt++;
3767 adjust_sit_entry_set(ses, head);
3768 return;
3769 }
3770 }
3771
3772 ses = grab_sit_entry_set();
3773
3774 ses->start_segno = start_segno;
3775 ses->entry_cnt++;
3776 list_add(&ses->set_list, head);
3777}
3778
3779static void add_sits_in_set(struct f2fs_sb_info *sbi)
3780{
3781 struct f2fs_sm_info *sm_info = SM_I(sbi);
3782 struct list_head *set_list = &sm_info->sit_entry_set;
3783 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3784 unsigned int segno;
3785
7cd8558b 3786 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3787 add_sit_entry(segno, set_list);
3788}
3789
3790static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3791{
3792 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3793 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3794 int i;
3795
b7ad7512 3796 down_write(&curseg->journal_rwsem);
dfc08a12 3797 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3798 unsigned int segno;
3799 bool dirtied;
3800
dfc08a12 3801 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3802 dirtied = __mark_sit_entry_dirty(sbi, segno);
3803
3804 if (!dirtied)
3805 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3806 }
dfc08a12 3807 update_sits_in_cursum(journal, -i);
b7ad7512 3808 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3809}
3810
0a8165d7 3811/*
351df4b2
JK		 3812 * CP calls this function, which flushes SIT entries including sit_journal,
3813 * and moves prefree segs to free segs.
3814 */
4d57b86d 3815void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3816{
3817 struct sit_info *sit_i = SIT_I(sbi);
3818 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3819 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3820 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3821 struct sit_entry_set *ses, *tmp;
3822 struct list_head *head = &SM_I(sbi)->sit_entry_set;
04f0b2ea 3823 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4b2fecc8 3824 struct seg_entry *se;
351df4b2 3825
3d26fa6b 3826 down_write(&sit_i->sentry_lock);
351df4b2 3827
2b11a74b
WL		 3828 if (!sit_i->dirty_sentries)
3829 goto out;
3830
351df4b2 3831 /*
184a5cd2
CY		 3832 * add and account sit entries of dirty bitmap in sit entry
3833 * set temporarily
351df4b2 3834 */
184a5cd2 3835 add_sits_in_set(sbi);
351df4b2 3836
184a5cd2
CY		 3837 /*
3838 * if there are no enough space in journal to store dirty sit
3839 * entries, remove all entries from journal and add and account
3840 * them in sit entry set.
3841 */
04f0b2ea
QS		 3842 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3843 !to_journal)
184a5cd2 3844 remove_sits_in_journal(sbi);
b2955550 3845
184a5cd2
CY		 3846 /*
3847 * there are two steps to flush sit entries:
3848 * #1, flush sit entries to journal in current cold data summary block.
3849 * #2, flush sit entries to sit page.
3850 */
3851 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3852 struct page *page = NULL;
184a5cd2
CY		 3853 struct f2fs_sit_block *raw_sit = NULL;
3854 unsigned int start_segno = ses->start_segno;
3855 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3856 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3857 unsigned int segno = start_segno;
3858
3859 if (to_journal &&
dfc08a12 3860 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3861 to_journal = false;
3862
b7ad7512
CY		 3863 if (to_journal) {
3864 down_write(&curseg->journal_rwsem);
3865 } else {
184a5cd2
CY		 3866 page = get_next_sit_page(sbi, start_segno);
3867 raw_sit = page_address(page);
351df4b2 3868 }
351df4b2 3869
184a5cd2
CY		 3870 /* flush dirty sit entries in region of current sit set */
3871 for_each_set_bit_from(segno, bitmap, end) {
3872 int offset, sit_offset;
4b2fecc8
JK		 3873
3874 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3875#ifdef CONFIG_F2FS_CHECK_FS
3876 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3877 SIT_VBLOCK_MAP_SIZE))
3878 f2fs_bug_on(sbi, 1);
3879#endif
184a5cd2
CY		 3880
3881 /* add discard candidates */
c473f1a9 3882 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3883 cpc->trim_start = segno;
25290fa5 3884 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3885 }
184a5cd2
CY		 3886
3887 if (to_journal) {
4d57b86d 3888 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 3889 SIT_JOURNAL, segno, 1);
3890 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3891 segno_in_journal(journal, offset) =
184a5cd2
CY		 3892 cpu_to_le32(segno);
3893 seg_info_to_raw_sit(se,
dfc08a12 3894 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3895 check_block_count(sbi, segno,
3896 &sit_in_journal(journal, offset));
184a5cd2
CY		 3897 } else {
3898 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3899 seg_info_to_raw_sit(se,
3900 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3901 check_block_count(sbi, segno,
3902 &raw_sit->entries[sit_offset]);
184a5cd2 3903 }
351df4b2 3904
184a5cd2
CY		 3905 __clear_bit(segno, bitmap);
3906 sit_i->dirty_sentries--;
3907 ses->entry_cnt--;
351df4b2
JK		 3908 }
3909
b7ad7512
CY		 3910 if (to_journal)
3911 up_write(&curseg->journal_rwsem);
3912 else
184a5cd2
CY		 3913 f2fs_put_page(page, 1);
3914
3915 f2fs_bug_on(sbi, ses->entry_cnt);
3916 release_sit_entry_set(ses);
351df4b2 3917 }
184a5cd2
CY		 3918
3919 f2fs_bug_on(sbi, !list_empty(head));
3920 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3921out:
c473f1a9 3922 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3923 __u64 trim_start = cpc->trim_start;
3924
4b2fecc8 3925 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3926 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3927
3928 cpc->trim_start = trim_start;
4b2fecc8 3929 }
3d26fa6b 3930 up_write(&sit_i->sentry_lock);
351df4b2 3931
351df4b2
JK		 3932 set_prefree_as_free_segments(sbi);
3933}
3934
3935static int build_sit_info(struct f2fs_sb_info *sbi)
3936{
3937 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3938 struct sit_info *sit_i;
3939 unsigned int sit_segs, start;
ae27d62e 3940 char *src_bitmap;
351df4b2
JK		 3941 unsigned int bitmap_size;
3942
3943 /* allocate memory for SIT information */
acbf054d 3944 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3945 if (!sit_i)
3946 return -ENOMEM;
3947
3948 SM_I(sbi)->sit_info = sit_i;
3949
9d2a789c
KC		 3950 sit_i->sentries =
3951 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3952 MAIN_SEGS(sbi)),
3953 GFP_KERNEL);
351df4b2
JK		 3954 if (!sit_i->sentries)
3955 return -ENOMEM;
3956
7cd8558b 3957 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3958 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3959 GFP_KERNEL);
351df4b2
JK		 3960 if (!sit_i->dirty_sentries_bitmap)
3961 return -ENOMEM;
3962
7cd8558b 3963 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3964 sit_i->sentries[start].cur_valid_map
acbf054d 3965 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3966 sit_i->sentries[start].ckpt_valid_map
acbf054d 3967 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3968 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3969 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3970 return -ENOMEM;
3e025740 3971
355e7891
CY		 3972#ifdef CONFIG_F2FS_CHECK_FS
3973 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3974 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3975 if (!sit_i->sentries[start].cur_valid_map_mir)
3976 return -ENOMEM;
3977#endif
3978
7d20c8ab
CY		 3979 sit_i->sentries[start].discard_map
3980 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3981 GFP_KERNEL);
3982 if (!sit_i->sentries[start].discard_map)
3983 return -ENOMEM;
351df4b2
JK		 3984 }
3985
acbf054d 3986 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3987 if (!sit_i->tmp_map)
3988 return -ENOMEM;
3989
2c70c5e3 3990 if (__is_large_section(sbi)) {
9d2a789c
KC		 3991 sit_i->sec_entries =
3992 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3993 MAIN_SECS(sbi)),
3994 GFP_KERNEL);
351df4b2
JK		 3995 if (!sit_i->sec_entries)
3996 return -ENOMEM;
3997 }
3998
3999 /* get information related with SIT */
4000 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4001
4002 /* setup SIT bitmap from ckeckpoint pack */
4003 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4004 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4005
ae27d62e
CY		 4006 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4007 if (!sit_i->sit_bitmap)
351df4b2 4008 return -ENOMEM;
351df4b2 4009
ae27d62e
CY		 4010#ifdef CONFIG_F2FS_CHECK_FS
4011 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4012 if (!sit_i->sit_bitmap_mir)
4013 return -ENOMEM;
4014#endif
4015
351df4b2
JK		 4016 /* init SIT information */
4017 sit_i->s_ops = &default_salloc_ops;
4018
4019 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4020 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 4021 sit_i->written_valid_blocks = 0;
351df4b2
JK		 4022 sit_i->bitmap_size = bitmap_size;
4023 sit_i->dirty_sentries = 0;
4024 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4025 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 4026 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 4027 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 4028 return 0;
4029}
4030
4031static int build_free_segmap(struct f2fs_sb_info *sbi)
4032{
351df4b2
JK		 4033 struct free_segmap_info *free_i;
4034 unsigned int bitmap_size, sec_bitmap_size;
4035
4036 /* allocate memory for free segmap information */
acbf054d 4037 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 4038 if (!free_i)
4039 return -ENOMEM;
4040
4041 SM_I(sbi)->free_info = free_i;
4042
7cd8558b 4043 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 4044 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 4045 if (!free_i->free_segmap)
4046 return -ENOMEM;
4047
7cd8558b 4048 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 4049 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 4050 if (!free_i->free_secmap)
4051 return -ENOMEM;
4052
4053 /* set all segments as dirty temporarily */
4054 memset(free_i->free_segmap, 0xff, bitmap_size);
4055 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4056
4057 /* init free segmap information */
7cd8558b 4058 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 4059 free_i->free_segments = 0;
4060 free_i->free_sections = 0;
1a118ccf 4061 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 4062 return 0;
4063}
4064
4065static int build_curseg(struct f2fs_sb_info *sbi)
4066{
1042d60f 4067 struct curseg_info *array;
351df4b2
JK		 4068 int i;
4069
026f0507
KC		 4070 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4071 GFP_KERNEL);
351df4b2
JK		 4072 if (!array)
4073 return -ENOMEM;
4074
4075 SM_I(sbi)->curseg_array = array;
4076
4077 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4078 mutex_init(&array[i].curseg_mutex);
acbf054d 4079 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 4080 if (!array[i].sum_blk)
4081 return -ENOMEM;
b7ad7512 4082 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 4083 array[i].journal = f2fs_kzalloc(sbi,
4084 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 4085 if (!array[i].journal)
4086 return -ENOMEM;
351df4b2
JK		 4087 array[i].segno = NULL_SEGNO;
4088 array[i].next_blkoff = 0;
4089 }
4090 return restore_curseg_summaries(sbi);
4091}
4092
c39a1b34 4093static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 4094{
4095 struct sit_info *sit_i = SIT_I(sbi);
4096 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 4097 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 4098 struct seg_entry *se;
4099 struct f2fs_sit_entry sit;
74de593a
CY		 4100 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4101 unsigned int i, start, end;
4102 unsigned int readed, start_blk = 0;
c39a1b34 4103 int err = 0;
8a29c126 4104 block_t total_node_blocks = 0;
351df4b2 4105
74de593a 4106 do {
4d57b86d 4107 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
664ba972 4108 META_SIT, true);
74de593a
CY		 4109
4110 start = start_blk * sit_i->sents_per_block;
4111 end = (start_blk + readed) * sit_i->sents_per_block;
4112
7cd8558b 4113 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 4114 struct f2fs_sit_block *sit_blk;
74de593a
CY		 4115 struct page *page;
4116
9c094040 4117 se = &sit_i->sentries[start];
74de593a 4118 page = get_current_sit_page(sbi, start);
edc55aaf
JK		 4119 if (IS_ERR(page))
4120 return PTR_ERR(page);
74de593a
CY		 4121 sit_blk = (struct f2fs_sit_block *)page_address(page);
4122 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4123 f2fs_put_page(page, 1);
d600af23 4124
c39a1b34
JK		 4125 err = check_block_count(sbi, start, &sit);
4126 if (err)
4127 return err;
74de593a 4128 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4129 if (IS_NODESEG(se->type))
4130 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 4131
4132 /* build discard map only one time */
7d20c8ab
CY		 4133 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4134 memset(se->discard_map, 0xff,
4135 SIT_VBLOCK_MAP_SIZE);
4136 } else {
4137 memcpy(se->discard_map,
4138 se->cur_valid_map,
4139 SIT_VBLOCK_MAP_SIZE);
4140 sbi->discard_blks +=
4141 sbi->blocks_per_seg -
4142 se->valid_blocks;
3e025740 4143 }
a66cdd98 4144
2c70c5e3 4145 if (__is_large_section(sbi))
d600af23
CY		 4146 get_sec_entry(sbi, start)->valid_blocks +=
4147 se->valid_blocks;
351df4b2 4148 }
74de593a
CY		 4149 start_blk += readed;
4150 } while (start_blk < sit_blk_cnt);
d600af23
CY		 4151
4152 down_read(&curseg->journal_rwsem);
4153 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 4154 unsigned int old_valid_blocks;
4155
4156 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 4157 if (start >= MAIN_SEGS(sbi)) {
4158 f2fs_msg(sbi->sb, KERN_ERR,
4159 "Wrong journal entry on segno %u",
4160 start);
4161 set_sbi_flag(sbi, SBI_NEED_FSCK);
4162 err = -EINVAL;
4163 break;
4164 }
4165
d600af23
CY		 4166 se = &sit_i->sentries[start];
4167 sit = sit_in_journal(journal, i);
4168
4169 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 4170 if (IS_NODESEG(se->type))
4171 total_node_blocks -= old_valid_blocks;
d600af23 4172
c39a1b34
JK		 4173 err = check_block_count(sbi, start, &sit);
4174 if (err)
4175 break;
d600af23 4176 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4177 if (IS_NODESEG(se->type))
4178 total_node_blocks += se->valid_blocks;
d600af23 4179
7d20c8ab
CY		 4180 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4181 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4182 } else {
4183 memcpy(se->discard_map, se->cur_valid_map,
4184 SIT_VBLOCK_MAP_SIZE);
4185 sbi->discard_blks += old_valid_blocks;
4186 sbi->discard_blks -= se->valid_blocks;
d600af23
CY		 4187 }
4188
2c70c5e3 4189 if (__is_large_section(sbi)) {
d600af23 4190 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 4191 se->valid_blocks;
4192 get_sec_entry(sbi, start)->valid_blocks -=
4193 old_valid_blocks;
4194 }
d600af23
CY		 4195 }
4196 up_read(&curseg->journal_rwsem);
8a29c126
JK		 4197
4198 if (!err && total_node_blocks != valid_node_count(sbi)) {
4199 f2fs_msg(sbi->sb, KERN_ERR,
4200 "SIT is corrupted node# %u vs %u",
4201 total_node_blocks, valid_node_count(sbi));
4202 set_sbi_flag(sbi, SBI_NEED_FSCK);
4203 err = -EINVAL;
4204 }
4205
c39a1b34 4206 return err;
351df4b2
JK		 4207}
4208
4209static void init_free_segmap(struct f2fs_sb_info *sbi)
4210{
4211 unsigned int start;
4212 int type;
4213
7cd8558b 4214 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 4215 struct seg_entry *sentry = get_seg_entry(sbi, start);
4216 if (!sentry->valid_blocks)
4217 __set_free(sbi, start);
c79b7ff1
JK		 4218 else
4219 SIT_I(sbi)->written_valid_blocks +=
4220 sentry->valid_blocks;
351df4b2
JK		 4221 }
4222
4223 /* set use the current segments */
4224 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4225 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4226 __set_test_and_inuse(sbi, curseg_t->segno);
4227 }
4228}
4229
4230static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4231{
4232 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4233 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 4234 unsigned int segno = 0, offset = 0;
351df4b2
JK		 4235 unsigned short valid_blocks;
4236
8736fbf0 4237 while (1) {
351df4b2 4238 /* find dirty segment based on free segmap */
7cd8558b
JK		 4239 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4240 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 4241 break;
4242 offset = segno + 1;
302bd348 4243 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 4244 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 4245 continue;
ec325b52
JK		 4246 if (valid_blocks > sbi->blocks_per_seg) {
4247 f2fs_bug_on(sbi, 1);
4248 continue;
4249 }
351df4b2
JK		 4250 mutex_lock(&dirty_i->seglist_lock);
4251 __locate_dirty_segment(sbi, segno, DIRTY);
4252 mutex_unlock(&dirty_i->seglist_lock);
4253 }
4254}
4255
5ec4e49f 4256static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4257{
4258 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 4259 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 4260
628b3d14 4261 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 4262 if (!dirty_i->victim_secmap)
351df4b2
JK		 4263 return -ENOMEM;
4264 return 0;
4265}
4266
4267static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4268{
4269 struct dirty_seglist_info *dirty_i;
4270 unsigned int bitmap_size, i;
4271
4272 /* allocate memory for dirty segments list information */
acbf054d
CY		 4273 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4274 GFP_KERNEL);
351df4b2
JK		 4275 if (!dirty_i)
4276 return -ENOMEM;
4277
4278 SM_I(sbi)->dirty_info = dirty_i;
4279 mutex_init(&dirty_i->seglist_lock);
4280
7cd8558b 4281 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 4282
4283 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 4284 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4285 GFP_KERNEL);
351df4b2
JK		 4286 if (!dirty_i->dirty_segmap[i])
4287 return -ENOMEM;
4288 }
4289
4290 init_dirty_segmap(sbi);
5ec4e49f 4291 return init_victim_secmap(sbi);
351df4b2
JK		 4292}
4293
c854f4d6
CY		 4294static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4295{
4296 int i;
4297
4298 /*
4299 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4300 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4301 */
4302 for (i = 0; i < NO_CHECK_TYPE; i++) {
4303 struct curseg_info *curseg = CURSEG_I(sbi, i);
4304 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4305 unsigned int blkofs = curseg->next_blkoff;
4306
4307 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4308 goto out;
4309
4310 if (curseg->alloc_type == SSR)
4311 continue;
4312
4313 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4314 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4315 continue;
4316out:
4317 f2fs_msg(sbi->sb, KERN_ERR,
4318 "Current segment's next free block offset is "
4319 "inconsistent with bitmap, logtype:%u, "
4320 "segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4321 i, curseg->segno, curseg->alloc_type,
4322 curseg->next_blkoff, blkofs);
4323 return -EINVAL;
4324 }
4325 }
4326 return 0;
4327}
4328
0a8165d7 4329/*
351df4b2
JK		 4330 * Update min, max modified time for cost-benefit GC algorithm
4331 */
4332static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4333{
4334 struct sit_info *sit_i = SIT_I(sbi);
4335 unsigned int segno;
4336
3d26fa6b 4337 down_write(&sit_i->sentry_lock);
351df4b2 4338
5ad25442 4339 sit_i->min_mtime = ULLONG_MAX;
351df4b2 4340
7cd8558b 4341 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 4342 unsigned int i;
4343 unsigned long long mtime = 0;
4344
4345 for (i = 0; i < sbi->segs_per_sec; i++)
4346 mtime += get_seg_entry(sbi, segno + i)->mtime;
4347
4348 mtime = div_u64(mtime, sbi->segs_per_sec);
4349
4350 if (sit_i->min_mtime > mtime)
4351 sit_i->min_mtime = mtime;
4352 }
a1f72ac2 4353 sit_i->max_mtime = get_mtime(sbi, false);
3d26fa6b 4354 up_write(&sit_i->sentry_lock);
351df4b2
JK		 4355}
4356
4d57b86d 4357int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4358{
4359 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4360 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 4361 struct f2fs_sm_info *sm_info;
351df4b2
JK		 4362 int err;
4363
acbf054d 4364 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 4365 if (!sm_info)
4366 return -ENOMEM;
4367
4368 /* init sm info */
4369 sbi->sm_info = sm_info;
351df4b2
JK		 4370 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4371 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4372 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4373 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4374 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4375 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4376 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 4377 sm_info->rec_prefree_segments = sm_info->main_segments *
4378 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 4379 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4380 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4381
52763a4b
JK		 4382 if (!test_opt(sbi, LFS))
4383 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 4384 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 4385 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
853137ce 4386 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
ef095d19 4387 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 4388 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 4389
184a5cd2
CY		 4390 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4391
2b60311d
CY		 4392 init_rwsem(&sm_info->curseg_lock);
4393
d4fdf8ba 4394 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 4395 err = f2fs_create_flush_cmd_control(sbi);
2163d198 4396 if (err)
a688b9d9 4397 return err;
6b4afdd7
JK		 4398 }
4399
0b54fb84
JK		 4400 err = create_discard_cmd_control(sbi);
4401 if (err)
4402 return err;
4403
351df4b2
JK		 4404 err = build_sit_info(sbi);
4405 if (err)
4406 return err;
4407 err = build_free_segmap(sbi);
4408 if (err)
4409 return err;
4410 err = build_curseg(sbi);
4411 if (err)
4412 return err;
4413
4414 /* reinit free segmap based on SIT */
c39a1b34
JK		 4415 err = build_sit_entries(sbi);
4416 if (err)
4417 return err;
351df4b2
JK		 4418
4419 init_free_segmap(sbi);
4420 err = build_dirty_segmap(sbi);
4421 if (err)
4422 return err;
4423
c854f4d6
CY		 4424 err = sanity_check_curseg(sbi);
4425 if (err)
4426 return err;
4427
351df4b2
JK		 4428 init_min_max_mtime(sbi);
4429 return 0;
4430}
4431
4432static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4433 enum dirty_type dirty_type)
4434{
4435 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4436
4437 mutex_lock(&dirty_i->seglist_lock);
39307a8e 4438 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 4439 dirty_i->nr_dirty[dirty_type] = 0;
4440 mutex_unlock(&dirty_i->seglist_lock);
4441}
4442
5ec4e49f 4443static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4444{
4445 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 4446 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 4447}
4448
4449static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4450{
4451 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4452 int i;
4453
4454 if (!dirty_i)
4455 return;
4456
4457 /* discard pre-free/dirty segments list */
4458 for (i = 0; i < NR_DIRTY_TYPE; i++)
4459 discard_dirty_segmap(sbi, i);
4460
5ec4e49f 4461 destroy_victim_secmap(sbi);
351df4b2 4462 SM_I(sbi)->dirty_info = NULL;
5222595d 4463 kvfree(dirty_i);
351df4b2
JK		 4464}
4465
4466static void destroy_curseg(struct f2fs_sb_info *sbi)
4467{
4468 struct curseg_info *array = SM_I(sbi)->curseg_array;
4469 int i;
4470
4471 if (!array)
4472 return;
4473 SM_I(sbi)->curseg_array = NULL;
b7ad7512 4474 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5222595d
JK		 4475 kvfree(array[i].sum_blk);
4476 kvfree(array[i].journal);
b7ad7512 4477 }
5222595d 4478 kvfree(array);
351df4b2
JK		 4479}
4480
4481static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4482{
4483 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4484 if (!free_i)
4485 return;
4486 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4487 kvfree(free_i->free_segmap);
4488 kvfree(free_i->free_secmap);
5222595d 4489 kvfree(free_i);
351df4b2
JK		 4490}
4491
4492static void destroy_sit_info(struct f2fs_sb_info *sbi)
4493{
4494 struct sit_info *sit_i = SIT_I(sbi);
4495 unsigned int start;
4496
4497 if (!sit_i)
4498 return;
4499
4500 if (sit_i->sentries) {
7cd8558b 4501 for (start = 0; start < MAIN_SEGS(sbi); start++) {
5222595d 4502 kvfree(sit_i->sentries[start].cur_valid_map);
355e7891 4503#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4504 kvfree(sit_i->sentries[start].cur_valid_map_mir);
355e7891 4505#endif
5222595d
JK		 4506 kvfree(sit_i->sentries[start].ckpt_valid_map);
4507 kvfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4508 }
4509 }
5222595d 4510 kvfree(sit_i->tmp_map);
60a3b782 4511
39307a8e
JK		 4512 kvfree(sit_i->sentries);
4513 kvfree(sit_i->sec_entries);
4514 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4515
4516 SM_I(sbi)->sit_info = NULL;
5222595d 4517 kvfree(sit_i->sit_bitmap);
ae27d62e 4518#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4519 kvfree(sit_i->sit_bitmap_mir);
ae27d62e 4520#endif
5222595d 4521 kvfree(sit_i);
351df4b2
JK		 4522}
4523
4d57b86d 4524void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4525{
4526 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4527
3b03f724
CY		 4528 if (!sm_info)
4529 return;
4d57b86d 4530 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 4531 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4532 destroy_dirty_segmap(sbi);
4533 destroy_curseg(sbi);
4534 destroy_free_segmap(sbi);
4535 destroy_sit_info(sbi);
4536 sbi->sm_info = NULL;
5222595d 4537 kvfree(sm_info);
351df4b2 4538}
7fd9e544 4539
4d57b86d 4540int __init f2fs_create_segment_manager_caches(void)
7fd9e544
JK		 4541{
4542 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4543 sizeof(struct discard_entry));
7fd9e544 4544 if (!discard_entry_slab)
184a5cd2
CY		 4545 goto fail;
4546
b01a9201
JK		 4547 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4548 sizeof(struct discard_cmd));
4549 if (!discard_cmd_slab)
6ab2a308 4550 goto destroy_discard_entry;
275b66b0 4551
184a5cd2 4552 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4553 sizeof(struct sit_entry_set));
184a5cd2 4554 if (!sit_entry_set_slab)
b01a9201 4555 goto destroy_discard_cmd;
88b88a66
JK		 4556
4557 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4558 sizeof(struct inmem_pages));
4559 if (!inmem_entry_slab)
4560 goto destroy_sit_entry_set;
7fd9e544 4561 return 0;
184a5cd2 4562
88b88a66
JK		 4563destroy_sit_entry_set:
4564 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4565destroy_discard_cmd:
4566 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4567destroy_discard_entry:
184a5cd2
CY		 4568 kmem_cache_destroy(discard_entry_slab);
4569fail:
4570 return -ENOMEM;
7fd9e544
JK		 4571}
4572
4d57b86d 4573void f2fs_destroy_segment_manager_caches(void)
7fd9e544 4574{
184a5cd2 4575 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4576 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4577 kmem_cache_destroy(discard_entry_slab);
88b88a66 4578 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4579}
Linux 6.x block layer and io_uring tree(s)