]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: allow all the users to pin a file
[mirror_ubuntu-jammy-kernel.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) {
dcbb4c10 1760 f2fs_err(sbi, "Invalid block %x", blkstart);
95175daf
DLM		 1761 return -EIO;
1762 }
3c62be17
JK		 1763 blkstart -= FDEV(devi).start_blk;
1764 }
f46e8809 1765
95175daf
DLM		 1766 /* For sequential zones, reset the zone write pointer */
1767 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
92592285
JK		 1768 sector = SECTOR_FROM_BLOCK(blkstart);
1769 nr_sects = SECTOR_FROM_BLOCK(blklen);
1770
1771 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1772 nr_sects != bdev_zone_sectors(bdev)) {
dcbb4c10
JP		 1773 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1774 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1775 blkstart, blklen);
92592285
JK		 1776 return -EIO;
1777 }
d50aaeec 1778 trace_f2fs_issue_reset_zone(bdev, blkstart);
95175daf 1779 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
f46e8809 1780 }
95175daf
DLM		 1781
1782 /* For conventional zones, use regular discard if supported */
95175daf 1783 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1784}
1785#endif
1786
3c62be17
JK		 1787static int __issue_discard_async(struct f2fs_sb_info *sbi,
1788 struct block_device *bdev, block_t blkstart, block_t blklen)
1789{
1790#ifdef CONFIG_BLK_DEV_ZONED
7f3d7719 1791 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
3c62be17
JK		 1792 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1793#endif
c81abe34 1794 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1795}
1796
1e87a78d 1797static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1798 block_t blkstart, block_t blklen)
1799{
3c62be17
JK		 1800 sector_t start = blkstart, len = 0;
1801 struct block_device *bdev;
a66cdd98
JK		 1802 struct seg_entry *se;
1803 unsigned int offset;
1804 block_t i;
3c62be17
JK		 1805 int err = 0;
1806
1807 bdev = f2fs_target_device(sbi, blkstart, NULL);
1808
1809 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1810 if (i != start) {
1811 struct block_device *bdev2 =
1812 f2fs_target_device(sbi, i, NULL);
1813
1814 if (bdev2 != bdev) {
1815 err = __issue_discard_async(sbi, bdev,
1816 start, len);
1817 if (err)
1818 return err;
1819 bdev = bdev2;
1820 start = i;
1821 len = 0;
1822 }
1823 }
a66cdd98 1824
a66cdd98
JK		 1825 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1826 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1827
1828 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1829 sbi->discard_blks--;
1830 }
f46e8809 1831
3c62be17
JK		 1832 if (len)
1833 err = __issue_discard_async(sbi, bdev, start, len);
1834 return err;
1e87a78d
JK		 1835}
1836
25290fa5
JK		 1837static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1838 bool check_only)
adf4983b 1839{
b2955550
JK		 1840 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1841 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1842 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1843 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1844 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1845 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1846 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1847 unsigned int start = 0, end = -1;
c473f1a9 1848 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1849 struct discard_entry *de = NULL;
46f84c2c 1850 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1851 int i;
1852
7d20c8ab 1853 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
25290fa5 1854 return false;
b2955550 1855
a66cdd98 1856 if (!force) {
7d20c8ab 1857 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
0b54fb84
JK		 1858 SM_I(sbi)->dcc_info->nr_discards >=
1859 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1860 return false;
4b2fecc8
JK		 1861 }
1862
b2955550
JK		 1863 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1864 for (i = 0; i < entries; i++)
a66cdd98 1865 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1866 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1867
0b54fb84
JK		 1868 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1869 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1870 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1871 if (start >= max_blocks)
1872 break;
1873
1874 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1875 if (force && start && end != max_blocks
1876 && (end - start) < cpc->trim_minlen)
1877 continue;
1878
25290fa5
JK		 1879 if (check_only)
1880 return true;
1881
a7eeb823
CY		 1882 if (!de) {
1883 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1884 GFP_F2FS_ZERO);
1885 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1886 list_add_tail(&de->list, head);
1887 }
1888
1889 for (i = start; i < end; i++)
1890 __set_bit_le(i, (void *)de->discard_map);
1891
1892 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1893 }
25290fa5 1894 return false;
b2955550
JK		 1895}
1896
af8ff65b
CY		 1897static void release_discard_addr(struct discard_entry *entry)
1898{
1899 list_del(&entry->list);
1900 kmem_cache_free(discard_entry_slab, entry);
1901}
1902
4d57b86d 1903void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1904{
46f84c2c 1905 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1906 struct discard_entry *entry, *this;
1907
1908 /* drop caches */
af8ff65b
CY		 1909 list_for_each_entry_safe(entry, this, head, list)
1910 release_discard_addr(entry);
4b2fecc8
JK		 1911}
1912
0a8165d7 1913/*
4d57b86d 1914 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1915 */
1916static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1917{
1918 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1919 unsigned int segno;
351df4b2
JK		 1920
1921 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1922 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1923 __set_test_and_free(sbi, segno);
351df4b2
JK		 1924 mutex_unlock(&dirty_i->seglist_lock);
1925}
1926
4d57b86d
CY		 1927void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1928 struct cp_control *cpc)
351df4b2 1929{
969d1b18
CY		 1930 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1931 struct list_head *head = &dcc->entry_list;
2d7b822a 1932 struct discard_entry *entry, *this;
351df4b2 1933 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1934 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1935 unsigned int start = 0, end = -1;
36abef4e 1936 unsigned int secno, start_segno;
c473f1a9 1937 bool force = (cpc->reason & CP_DISCARD);
2c70c5e3 1938 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
351df4b2
JK		 1939
1940 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1941
351df4b2 1942 while (1) {
29e59c14 1943 int i;
ad6672bb
YS		 1944
1945 if (need_align && end != -1)
1946 end--;
7cd8558b
JK		 1947 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1948 if (start >= MAIN_SEGS(sbi))
351df4b2 1949 break;
7cd8558b
JK		 1950 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1951 start + 1);
29e59c14 1952
ad6672bb
YS		 1953 if (need_align) {
1954 start = rounddown(start, sbi->segs_per_sec);
1955 end = roundup(end, sbi->segs_per_sec);
1956 }
29e59c14 1957
ad6672bb
YS		 1958 for (i = start; i < end; i++) {
1959 if (test_and_clear_bit(i, prefree_map))
1960 dirty_i->nr_dirty[PRE]--;
1961 }
29e59c14 1962
7d20c8ab 1963 if (!f2fs_realtime_discard_enable(sbi))
29e59c14 1964 continue;
351df4b2 1965
650d3c4e
YH		 1966 if (force && start >= cpc->trim_start &&
1967 (end - 1) <= cpc->trim_end)
1968 continue;
1969
2c70c5e3 1970 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
36abef4e 1971 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1972 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1973 continue;
1974 }
1975next:
4ddb1a4d
JK		 1976 secno = GET_SEC_FROM_SEG(sbi, start);
1977 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1978 if (!IS_CURSEC(sbi, secno) &&
302bd348 1979 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1980 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1981 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1982
1983 start = start_segno + sbi->segs_per_sec;
1984 if (start < end)
1985 goto next;
8b107f5b
JK		 1986 else
1987 end = start - 1;
351df4b2
JK		 1988 }
1989 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1990
1991 /* send small discards */
2d7b822a 1992 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1993 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1994 bool is_valid = test_bit_le(0, entry->discard_map);
1995
1996find_next:
1997 if (is_valid) {
1998 next_pos = find_next_zero_bit_le(entry->discard_map,
1999 sbi->blocks_per_seg, cur_pos);
2000 len = next_pos - cur_pos;
2001
7beb01f7 2002 if (f2fs_sb_has_blkzoned(sbi) ||
acfd2810 2003 (force && len < cpc->trim_minlen))
a7eeb823
CY		 2004 goto skip;
2005
2006 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2007 len);
a7eeb823
CY		 2008 total_len += len;
2009 } else {
2010 next_pos = find_next_bit_le(entry->discard_map,
2011 sbi->blocks_per_seg, cur_pos);
2012 }
836b5a63 2013skip:
a7eeb823
CY		 2014 cur_pos = next_pos;
2015 is_valid = !is_valid;
2016
2017 if (cur_pos < sbi->blocks_per_seg)
2018 goto find_next;
2019
af8ff65b 2020 release_discard_addr(entry);
969d1b18 2021 dcc->nr_discards -= total_len;
b2955550 2022 }
34e159da 2023
01983c71 2024 wake_up_discard_thread(sbi, false);
351df4b2
JK		 2025}
2026
8ed59745 2027static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 2028{
15469963 2029 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 2030 struct discard_cmd_control *dcc;
ba48a33e 2031 int err = 0, i;
0b54fb84
JK		 2032
2033 if (SM_I(sbi)->dcc_info) {
2034 dcc = SM_I(sbi)->dcc_info;
2035 goto init_thread;
2036 }
2037
acbf054d 2038 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 2039 if (!dcc)
2040 return -ENOMEM;
2041
969d1b18 2042 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 2043 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 2044 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 2045 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 2046 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 2047 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 2048 mutex_init(&dcc->cmd_lock);
8b8dd65f 2049 atomic_set(&dcc->issued_discard, 0);
72691af6 2050 atomic_set(&dcc->queued_discard, 0);
5f32366a 2051 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 2052 dcc->nr_discards = 0;
d618ebaf 2053 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 2054 dcc->undiscard_blks = 0;
20ee4382 2055 dcc->next_pos = 0;
4dada3fd 2056 dcc->root = RB_ROOT_CACHED;
67fce70b 2057 dcc->rbtree_check = false;
0b54fb84 2058
15469963 2059 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 2060 SM_I(sbi)->dcc_info = dcc;
2061init_thread:
15469963
JK		 2062 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2063 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2064 if (IS_ERR(dcc->f2fs_issue_discard)) {
2065 err = PTR_ERR(dcc->f2fs_issue_discard);
5222595d 2066 kvfree(dcc);
15469963
JK		 2067 SM_I(sbi)->dcc_info = NULL;
2068 return err;
2069 }
2070
0b54fb84
JK		 2071 return err;
2072}
2073
f099405f 2074static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 2075{
2076 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2077
f099405f
CY		 2078 if (!dcc)
2079 return;
2080
4d57b86d 2081 f2fs_stop_discard_thread(sbi);
f099405f 2082
5222595d 2083 kvfree(dcc);
f099405f 2084 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 2085}
2086
184a5cd2 2087static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 2088{
2089 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 2090
2091 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 2092 sit_i->dirty_sentries++;
184a5cd2
CY		 2093 return false;
2094 }
2095
2096 return true;
351df4b2
JK		 2097}
2098
2099static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2100 unsigned int segno, int modified)
2101{
2102 struct seg_entry *se = get_seg_entry(sbi, segno);
2103 se->type = type;
2104 if (modified)
2105 __mark_sit_entry_dirty(sbi, segno);
2106}
2107
2108static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2109{
2110 struct seg_entry *se;
2111 unsigned int segno, offset;
2112 long int new_vblocks;
6415fedc
YS		 2113 bool exist;
2114#ifdef CONFIG_F2FS_CHECK_FS
2115 bool mir_exist;
2116#endif
351df4b2
JK		 2117
2118 segno = GET_SEGNO(sbi, blkaddr);
2119
2120 se = get_seg_entry(sbi, segno);
2121 new_vblocks = se->valid_blocks + del;
491c0854 2122 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 2123
9850cf4a 2124 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 2125 (new_vblocks > sbi->blocks_per_seg)));
2126
2127 se->valid_blocks = new_vblocks;
a1f72ac2
CY		 2128 se->mtime = get_mtime(sbi, false);
2129 if (se->mtime > SIT_I(sbi)->max_mtime)
2130 SIT_I(sbi)->max_mtime = se->mtime;
351df4b2
JK		 2131
2132 /* Update valid block bitmap */
2133 if (del > 0) {
6415fedc 2134 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 2135#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2136 mir_exist = f2fs_test_and_set_bit(offset,
2137 se->cur_valid_map_mir);
2138 if (unlikely(exist != mir_exist)) {
dcbb4c10
JP		 2139 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2140 blkaddr, exist);
05796763 2141 f2fs_bug_on(sbi, 1);
6415fedc 2142 }
355e7891 2143#endif
6415fedc 2144 if (unlikely(exist)) {
dcbb4c10
JP		 2145 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2146 blkaddr);
6415fedc 2147 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2148 se->valid_blocks--;
2149 del = 0;
355e7891 2150 }
6415fedc 2151
7d20c8ab 2152 if (!f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 2153 sbi->discard_blks--;
720037f9
JK		 2154
2155 /* don't overwrite by SSR to keep node chain */
4354994f
DR		 2156 if (IS_NODESEG(se->type) &&
2157 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
720037f9
JK		 2158 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2159 se->ckpt_valid_blocks++;
2160 }
351df4b2 2161 } else {
6415fedc 2162 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 2163#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2164 mir_exist = f2fs_test_and_clear_bit(offset,
2165 se->cur_valid_map_mir);
2166 if (unlikely(exist != mir_exist)) {
dcbb4c10
JP		 2167 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2168 blkaddr, exist);
05796763 2169 f2fs_bug_on(sbi, 1);
6415fedc 2170 }
355e7891 2171#endif
6415fedc 2172 if (unlikely(!exist)) {
dcbb4c10
JP		 2173 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2174 blkaddr);
6415fedc 2175 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2176 se->valid_blocks++;
2177 del = 0;
4354994f
DR		 2178 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2179 /*
2180 * If checkpoints are off, we must not reuse data that
2181 * was used in the previous checkpoint. If it was used
2182 * before, we must track that to know how much space we
2183 * really have.
2184 */
c9c8ed50
CY		 2185 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2186 spin_lock(&sbi->stat_lock);
4354994f 2187 sbi->unusable_block_count++;
c9c8ed50
CY		 2188 spin_unlock(&sbi->stat_lock);
2189 }
355e7891 2190 }
6415fedc 2191
7d20c8ab 2192 if (f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 2193 sbi->discard_blks++;
351df4b2
JK		 2194 }
2195 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2196 se->ckpt_valid_blocks += del;
2197
2198 __mark_sit_entry_dirty(sbi, segno);
2199
2200 /* update total number of valid blocks to be written in ckpt area */
2201 SIT_I(sbi)->written_valid_blocks += del;
2202
2c70c5e3 2203 if (__is_large_section(sbi))
351df4b2
JK		 2204 get_sec_entry(sbi, segno)->valid_blocks += del;
2205}
2206
4d57b86d 2207void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 2208{
2209 unsigned int segno = GET_SEGNO(sbi, addr);
2210 struct sit_info *sit_i = SIT_I(sbi);
2211
9850cf4a 2212 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 2213 if (addr == NEW_ADDR)
2214 return;
2215
6aa58d8a
CY		 2216 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2217
351df4b2 2218 /* add it into sit main buffer */
3d26fa6b 2219 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2220
2221 update_sit_entry(sbi, addr, -1);
2222
2223 /* add it into dirty seglist */
2224 locate_dirty_segment(sbi, segno);
2225
3d26fa6b 2226 up_write(&sit_i->sentry_lock);
351df4b2
JK		 2227}
2228
4d57b86d 2229bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 2230{
2231 struct sit_info *sit_i = SIT_I(sbi);
2232 unsigned int segno, offset;
2233 struct seg_entry *se;
2234 bool is_cp = false;
2235
93770ab7 2236 if (!__is_valid_data_blkaddr(blkaddr))
6e2c64ad
JK		 2237 return true;
2238
3d26fa6b 2239 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2240
2241 segno = GET_SEGNO(sbi, blkaddr);
2242 se = get_seg_entry(sbi, segno);
2243 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2244
2245 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2246 is_cp = true;
2247
3d26fa6b 2248 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2249
2250 return is_cp;
2251}
2252
0a8165d7 2253/*
351df4b2
JK		 2254 * This function should be resided under the curseg_mutex lock
2255 */
2256static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 2257 struct f2fs_summary *sum)
351df4b2
JK		 2258{
2259 struct curseg_info *curseg = CURSEG_I(sbi, type);
2260 void *addr = curseg->sum_blk;
e79efe3b 2261 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 2262 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 2263}
2264
0a8165d7 2265/*
351df4b2
JK		 2266 * Calculate the number of current summary pages for writing
2267 */
4d57b86d 2268int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 2269{
351df4b2 2270 int valid_sum_count = 0;
9a47938b 2271 int i, sum_in_page;
351df4b2
JK		 2272
2273 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2274 if (sbi->ckpt->alloc_type[i] == SSR)
2275 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 2276 else {
2277 if (for_ra)
2278 valid_sum_count += le16_to_cpu(
2279 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2280 else
2281 valid_sum_count += curseg_blkoff(sbi, i);
2282 }
351df4b2
JK		 2283 }
2284
09cbfeaf 2285 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 2286 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2287 if (valid_sum_count <= sum_in_page)
351df4b2 2288 return 1;
9a47938b 2289 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 2290 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 2291 return 2;
2292 return 3;
2293}
2294
0a8165d7 2295/*
351df4b2
JK		 2296 * Caller should put this summary page
2297 */
4d57b86d 2298struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 2299{
7735730d 2300 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2301}
2302
4d57b86d
CY		 2303void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2304 void *src, block_t blk_addr)
351df4b2 2305{
4d57b86d 2306 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 2307
0537b811 2308 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2309 set_page_dirty(page);
2310 f2fs_put_page(page, 1);
2311}
2312
381722d2
CY		 2313static void write_sum_page(struct f2fs_sb_info *sbi,
2314 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2315{
4d57b86d 2316 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2317}
2318
b7ad7512
CY		 2319static void write_current_sum_page(struct f2fs_sb_info *sbi,
2320 int type, block_t blk_addr)
2321{
2322 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2323 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2324 struct f2fs_summary_block *src = curseg->sum_blk;
2325 struct f2fs_summary_block *dst;
2326
2327 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2328 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2329
2330 mutex_lock(&curseg->curseg_mutex);
2331
2332 down_read(&curseg->journal_rwsem);
2333 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2334 up_read(&curseg->journal_rwsem);
2335
2336 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2337 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2338
2339 mutex_unlock(&curseg->curseg_mutex);
2340
2341 set_page_dirty(page);
2342 f2fs_put_page(page, 1);
2343}
2344
a7881893
JK		 2345static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2346{
2347 struct curseg_info *curseg = CURSEG_I(sbi, type);
2348 unsigned int segno = curseg->segno + 1;
2349 struct free_segmap_info *free_i = FREE_I(sbi);
2350
2351 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2352 return !test_bit(segno, free_i->free_segmap);
2353 return 0;
2354}
2355
0a8165d7 2356/*
351df4b2
JK		 2357 * Find a new segment from the free segments bitmap to right order
2358 * This function should be returned with success, otherwise BUG
2359 */
2360static void get_new_segment(struct f2fs_sb_info *sbi,
2361 unsigned int *newseg, bool new_sec, int dir)
2362{
2363 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2364 unsigned int segno, secno, zoneno;
7cd8558b 2365 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2366 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2367 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2368 unsigned int left_start = hint;
2369 bool init = true;
2370 int go_left = 0;
2371 int i;
2372
1a118ccf 2373 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2374
2375 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2376 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2377 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2378 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2379 goto got_it;
2380 }
2381find_other_zone:
7cd8558b
JK		 2382 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2383 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2384 if (dir == ALLOC_RIGHT) {
2385 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2386 MAIN_SECS(sbi), 0);
2387 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2388 } else {
2389 go_left = 1;
2390 left_start = hint - 1;
2391 }
2392 }
2393 if (go_left == 0)
2394 goto skip_left;
2395
2396 while (test_bit(left_start, free_i->free_secmap)) {
2397 if (left_start > 0) {
2398 left_start--;
2399 continue;
2400 }
2401 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2402 MAIN_SECS(sbi), 0);
2403 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2404 break;
2405 }
2406 secno = left_start;
2407skip_left:
4ddb1a4d
JK		 2408 segno = GET_SEG_FROM_SEC(sbi, secno);
2409 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2410
2411 /* give up on finding another zone */
2412 if (!init)
2413 goto got_it;
2414 if (sbi->secs_per_zone == 1)
2415 goto got_it;
2416 if (zoneno == old_zoneno)
2417 goto got_it;
2418 if (dir == ALLOC_LEFT) {
2419 if (!go_left && zoneno + 1 >= total_zones)
2420 goto got_it;
2421 if (go_left && zoneno == 0)
2422 goto got_it;
2423 }
2424 for (i = 0; i < NR_CURSEG_TYPE; i++)
2425 if (CURSEG_I(sbi, i)->zone == zoneno)
2426 break;
2427
2428 if (i < NR_CURSEG_TYPE) {
2429 /* zone is in user, try another */
2430 if (go_left)
2431 hint = zoneno * sbi->secs_per_zone - 1;
2432 else if (zoneno + 1 >= total_zones)
2433 hint = 0;
2434 else
2435 hint = (zoneno + 1) * sbi->secs_per_zone;
2436 init = false;
2437 goto find_other_zone;
2438 }
2439got_it:
2440 /* set it as dirty segment in free segmap */
9850cf4a 2441 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2442 __set_inuse(sbi, segno);
2443 *newseg = segno;
1a118ccf 2444 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2445}
2446
2447static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2448{
2449 struct curseg_info *curseg = CURSEG_I(sbi, type);
2450 struct summary_footer *sum_footer;
2451
2452 curseg->segno = curseg->next_segno;
4ddb1a4d 2453 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2454 curseg->next_blkoff = 0;
2455 curseg->next_segno = NULL_SEGNO;
2456
2457 sum_footer = &(curseg->sum_blk->footer);
2458 memset(sum_footer, 0, sizeof(struct summary_footer));
2459 if (IS_DATASEG(type))
2460 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2461 if (IS_NODESEG(type))
2462 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2463 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2464}
2465
7a20b8a6
JK		 2466static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2467{
a7881893 2468 /* if segs_per_sec is large than 1, we need to keep original policy. */
2c70c5e3 2469 if (__is_large_section(sbi))
a7881893
JK		 2470 return CURSEG_I(sbi, type)->segno;
2471
4354994f
DR		 2472 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2473 return 0;
2474
b94929d9
YS		 2475 if (test_opt(sbi, NOHEAP) &&
2476 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2477 return 0;
2478
e066b83c
JK		 2479 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2480 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2481
2482 /* find segments from 0 to reuse freed segments */
63189b78 2483 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2484 return 0;
2485
7a20b8a6
JK		 2486 return CURSEG_I(sbi, type)->segno;
2487}
2488
0a8165d7 2489/*
351df4b2
JK		 2490 * Allocate a current working segment.
2491 * This function always allocates a free segment in LFS manner.
2492 */
2493static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2494{
2495 struct curseg_info *curseg = CURSEG_I(sbi, type);
2496 unsigned int segno = curseg->segno;
2497 int dir = ALLOC_LEFT;
2498
2499 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2500 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2501 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2502 dir = ALLOC_RIGHT;
2503
2504 if (test_opt(sbi, NOHEAP))
2505 dir = ALLOC_RIGHT;
2506
7a20b8a6 2507 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2508 get_new_segment(sbi, &segno, new_sec, dir);
2509 curseg->next_segno = segno;
2510 reset_curseg(sbi, type, 1);
2511 curseg->alloc_type = LFS;
2512}
2513
2514static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2515 struct curseg_info *seg, block_t start)
2516{
2517 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2518 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2519 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2520 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2521 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2522 int i, pos;
2523
2524 for (i = 0; i < entries; i++)
2525 target_map[i] = ckpt_map[i] | cur_map[i];
2526
2527 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2528
2529 seg->next_blkoff = pos;
351df4b2
JK		 2530}
2531
0a8165d7 2532/*
351df4b2
JK		 2533 * If a segment is written by LFS manner, next block offset is just obtained
2534 * by increasing the current block offset. However, if a segment is written by
2535 * SSR manner, next block offset obtained by calling __next_free_blkoff
2536 */
2537static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2538 struct curseg_info *seg)
2539{
2540 if (seg->alloc_type == SSR)
2541 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2542 else
2543 seg->next_blkoff++;
2544}
2545
0a8165d7 2546/*
e1c42045 2547 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2548 * manner, so it should recover the existing segment information of valid blocks
2549 */
025d63a4 2550static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2551{
2552 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2553 struct curseg_info *curseg = CURSEG_I(sbi, type);
2554 unsigned int new_segno = curseg->next_segno;
2555 struct f2fs_summary_block *sum_node;
2556 struct page *sum_page;
2557
2558 write_sum_page(sbi, curseg->sum_blk,
2559 GET_SUM_BLOCK(sbi, curseg->segno));
2560 __set_test_and_inuse(sbi, new_segno);
2561
2562 mutex_lock(&dirty_i->seglist_lock);
2563 __remove_dirty_segment(sbi, new_segno, PRE);
2564 __remove_dirty_segment(sbi, new_segno, DIRTY);
2565 mutex_unlock(&dirty_i->seglist_lock);
2566
2567 reset_curseg(sbi, type, 1);
2568 curseg->alloc_type = SSR;
2569 __next_free_blkoff(sbi, curseg, 0);
2570
4d57b86d 2571 sum_page = f2fs_get_sum_page(sbi, new_segno);
edc55aaf 2572 f2fs_bug_on(sbi, IS_ERR(sum_page));
025d63a4
CY		 2573 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2574 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2575 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2576}
2577
43727527
JK		 2578static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2579{
2580 struct curseg_info *curseg = CURSEG_I(sbi, type);
2581 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2582 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2583 int i, cnt;
2584 bool reversed = false;
c192f7a4 2585
4d57b86d 2586 /* f2fs_need_SSR() already forces to do this */
e066b83c
JK		 2587 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2588 curseg->next_segno = segno;
c192f7a4 2589 return 1;
e066b83c 2590 }
43727527 2591
70d625cb
JK		 2592 /* For node segments, let's do SSR more intensively */
2593 if (IS_NODESEG(type)) {
d27c3d89
CY		 2594 if (type >= CURSEG_WARM_NODE) {
2595 reversed = true;
2596 i = CURSEG_COLD_NODE;
2597 } else {
2598 i = CURSEG_HOT_NODE;
2599 }
2600 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2601 } else {
d27c3d89
CY		 2602 if (type >= CURSEG_WARM_DATA) {
2603 reversed = true;
2604 i = CURSEG_COLD_DATA;
2605 } else {
2606 i = CURSEG_HOT_DATA;
2607 }
2608 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2609 }
43727527 2610
d27c3d89 2611 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2612 if (i == type)
2613 continue;
e066b83c
JK		 2614 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2615 curseg->next_segno = segno;
43727527 2616 return 1;
e066b83c 2617 }
c192f7a4 2618 }
4354994f
DR		 2619
2620 /* find valid_blocks=0 in dirty list */
2621 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2622 segno = get_free_segment(sbi);
2623 if (segno != NULL_SEGNO) {
2624 curseg->next_segno = segno;
2625 return 1;
2626 }
2627 }
43727527
JK		 2628 return 0;
2629}
2630
351df4b2
JK		 2631/*
2632 * flush out current segment and replace it with new segment
2633 * This function should be returned with success, otherwise BUG
2634 */
2635static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2636 int type, bool force)
2637{
a7881893
JK		 2638 struct curseg_info *curseg = CURSEG_I(sbi, type);
2639
7b405275 2640 if (force)
351df4b2 2641 new_curseg(sbi, type, true);
5b6c6be2
JK		 2642 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2643 type == CURSEG_WARM_NODE)
351df4b2 2644 new_curseg(sbi, type, false);
4354994f
DR		 2645 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2646 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
a7881893 2647 new_curseg(sbi, type, false);
4d57b86d 2648 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2649 change_curseg(sbi, type);
351df4b2
JK		 2650 else
2651 new_curseg(sbi, type, false);
dcdfff65 2652
a7881893 2653 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2654}
2655
04f0b2ea
QS		 2656void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2657 unsigned int start, unsigned int end)
2658{
2659 struct curseg_info *curseg = CURSEG_I(sbi, type);
2660 unsigned int segno;
2661
2662 down_read(&SM_I(sbi)->curseg_lock);
2663 mutex_lock(&curseg->curseg_mutex);
2664 down_write(&SIT_I(sbi)->sentry_lock);
2665
2666 segno = CURSEG_I(sbi, type)->segno;
2667 if (segno < start || segno > end)
2668 goto unlock;
2669
2670 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2671 change_curseg(sbi, type);
2672 else
2673 new_curseg(sbi, type, true);
2674
2675 stat_inc_seg_type(sbi, curseg);
2676
2677 locate_dirty_segment(sbi, segno);
2678unlock:
2679 up_write(&SIT_I(sbi)->sentry_lock);
2680
2681 if (segno != curseg->segno)
dcbb4c10
JP		 2682 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2683 type, segno, curseg->segno);
04f0b2ea
QS		 2684
2685 mutex_unlock(&curseg->curseg_mutex);
2686 up_read(&SM_I(sbi)->curseg_lock);
2687}
2688
4d57b86d 2689void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
351df4b2 2690{
6ae1be13
JK		 2691 struct curseg_info *curseg;
2692 unsigned int old_segno;
351df4b2
JK		 2693 int i;
2694
3d26fa6b
CY		 2695 down_write(&SIT_I(sbi)->sentry_lock);
2696
6ae1be13
JK		 2697 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2698 curseg = CURSEG_I(sbi, i);
2699 old_segno = curseg->segno;
2700 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2701 locate_dirty_segment(sbi, old_segno);
2702 }
3d26fa6b
CY		 2703
2704 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2705}
2706
2707static const struct segment_allocation default_salloc_ops = {
2708 .allocate_segment = allocate_segment_by_default,
2709};
2710
4d57b86d
CY		 2711bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2712 struct cp_control *cpc)
25290fa5
JK		 2713{
2714 __u64 trim_start = cpc->trim_start;
2715 bool has_candidate = false;
2716
3d26fa6b 2717 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2718 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2719 if (add_discard_addrs(sbi, cpc, true)) {
2720 has_candidate = true;
2721 break;
2722 }
2723 }
3d26fa6b 2724 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2725
2726 cpc->trim_start = trim_start;
2727 return has_candidate;
2728}
2729
01f9cf6d 2730static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 2731 struct discard_policy *dpolicy,
2732 unsigned int start, unsigned int end)
2733{
2734 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2735 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2736 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2737 struct discard_cmd *dc;
2738 struct blk_plug plug;
2739 int issued;
01f9cf6d 2740 unsigned int trimmed = 0;
9a997188
JK		 2741
2742next:
2743 issued = 0;
2744
2745 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 2746 if (unlikely(dcc->rbtree_check))
2747 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2748 &dcc->root));
9a997188 2749
4d57b86d 2750 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 2751 NULL, start,
2752 (struct rb_entry **)&prev_dc,
2753 (struct rb_entry **)&next_dc,
4dada3fd 2754 &insert_p, &insert_parent, true, NULL);
9a997188
JK		 2755 if (!dc)
2756 dc = next_dc;
2757
2758 blk_start_plug(&plug);
2759
2760 while (dc && dc->lstart <= end) {
2761 struct rb_node *node;
6b9cb124 2762 int err = 0;
9a997188
JK		 2763
2764 if (dc->len < dpolicy->granularity)
2765 goto skip;
2766
2767 if (dc->state != D_PREP) {
2768 list_move_tail(&dc->list, &dcc->fstrim_list);
2769 goto skip;
2770 }
2771
6b9cb124 2772 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
9a997188 2773
35ec7d57 2774 if (issued >= dpolicy->max_requests) {
9a997188
JK		 2775 start = dc->lstart + dc->len;
2776
6b9cb124
CY		 2777 if (err)
2778 __remove_discard_cmd(sbi, dc);
2779
9a997188
JK		 2780 blk_finish_plug(&plug);
2781 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 2782 trimmed += __wait_all_discard_cmd(sbi, NULL);
9a997188
JK		 2783 congestion_wait(BLK_RW_ASYNC, HZ/50);
2784 goto next;
2785 }
2786skip:
2787 node = rb_next(&dc->rb_node);
6b9cb124
CY		 2788 if (err)
2789 __remove_discard_cmd(sbi, dc);
9a997188
JK		 2790 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2791
2792 if (fatal_signal_pending(current))
2793 break;
2794 }
2795
2796 blk_finish_plug(&plug);
2797 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 2798
2799 return trimmed;
9a997188
JK		 2800}
2801
4b2fecc8
JK		 2802int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2803{
f7ef9b83
JK		 2804 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2805 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2806 unsigned int start_segno, end_segno;
8412663d 2807 block_t start_block, end_block;
4b2fecc8 2808 struct cp_control cpc;
78997b56 2809 struct discard_policy dpolicy;
0ea80512 2810 unsigned long long trimmed = 0;
c34f42e2 2811 int err = 0;
2c70c5e3 2812 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
4b2fecc8 2813
836b5a63 2814 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2815 return -EINVAL;
2816
3f16ecd9
CY		 2817 if (end < MAIN_BLKADDR(sbi))
2818 goto out;
4b2fecc8 2819
ed214a11 2820 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
dcbb4c10 2821 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
10f966bb 2822 return -EFSCORRUPTED;
ed214a11
YH		 2823 }
2824
4b2fecc8 2825 /* start/end segment number in main_area */
7cd8558b
JK		 2826 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2827 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2828 GET_SEGNO(sbi, end);
ad6672bb
YS		 2829 if (need_align) {
2830 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2831 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2832 }
8412663d 2833
4b2fecc8 2834 cpc.reason = CP_DISCARD;
836b5a63 2835 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2836 cpc.trim_start = start_segno;
2837 cpc.trim_end = end_segno;
4b2fecc8 2838
377224c4
CY		 2839 if (sbi->discard_blks == 0)
2840 goto out;
74fa5f3d 2841
377224c4 2842 mutex_lock(&sbi->gc_mutex);
4d57b86d 2843 err = f2fs_write_checkpoint(sbi, &cpc);
377224c4
CY		 2844 mutex_unlock(&sbi->gc_mutex);
2845 if (err)
2846 goto out;
8412663d 2847
e555da9f
JK		 2848 /*
2849 * We filed discard candidates, but actually we don't need to wait for
2850 * all of them, since they'll be issued in idle time along with runtime
2851 * discard option. User configuration looks like using runtime discard
2852 * or periodic fstrim instead of it.
2853 */
7d20c8ab 2854 if (f2fs_realtime_discard_enable(sbi))
5a615492
JK		 2855 goto out;
2856
2857 start_block = START_BLOCK(sbi, start_segno);
2858 end_block = START_BLOCK(sbi, end_segno + 1);
2859
2860 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 2861 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2862 start_block, end_block);
5a615492 2863
01f9cf6d 2864 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2865 start_block, end_block);
377224c4 2866out:
6eae2694
CY		 2867 if (!err)
2868 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2869 return err;
4b2fecc8
JK		 2870}
2871
351df4b2
JK		 2872static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2873{
2874 struct curseg_info *curseg = CURSEG_I(sbi, type);
2875 if (curseg->next_blkoff < sbi->blocks_per_seg)
2876 return true;
2877 return false;
2878}
2879
4d57b86d 2880int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 2881{
2882 switch (hint) {
2883 case WRITE_LIFE_SHORT:
2884 return CURSEG_HOT_DATA;
2885 case WRITE_LIFE_EXTREME:
2886 return CURSEG_COLD_DATA;
2887 default:
2888 return CURSEG_WARM_DATA;
2889 }
2890}
2891
0cdd3195
HL		 2892/* This returns write hints for each segment type. This hints will be
2893 * passed down to block layer. There are mapping tables which depend on
2894 * the mount option 'whint_mode'.
2895 *
2896 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2897 *
2898 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2899 *
2900 * User F2FS Block
2901 * ---- ---- -----
2902 * META WRITE_LIFE_NOT_SET
2903 * HOT_NODE "
2904 * WARM_NODE "
2905 * COLD_NODE "
2906 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2907 * extension list " "
2908 *
2909 * -- buffered io
2910 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2911 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2912 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2913 * WRITE_LIFE_NONE " "
2914 * WRITE_LIFE_MEDIUM " "
2915 * WRITE_LIFE_LONG " "
2916 *
2917 * -- direct io
2918 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2919 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2920 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2921 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2922 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2923 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2924 *
f2e703f9
HL		 2925 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2926 *
2927 * User F2FS Block
2928 * ---- ---- -----
2929 * META WRITE_LIFE_MEDIUM;
2930 * HOT_NODE WRITE_LIFE_NOT_SET
2931 * WARM_NODE "
2932 * COLD_NODE WRITE_LIFE_NONE
2933 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2934 * extension list " "
2935 *
2936 * -- buffered io
2937 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2938 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2939 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2940 * WRITE_LIFE_NONE " "
2941 * WRITE_LIFE_MEDIUM " "
2942 * WRITE_LIFE_LONG " "
2943 *
2944 * -- direct io
2945 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2946 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2947 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2948 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2949 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2950 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2951 */
2952
4d57b86d 2953enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 2954 enum page_type type, enum temp_type temp)
2955{
63189b78 2956 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2957 if (type == DATA) {
f2e703f9 2958 if (temp == WARM)
0cdd3195 2959 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2960 else if (temp == HOT)
2961 return WRITE_LIFE_SHORT;
2962 else if (temp == COLD)
2963 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2964 } else {
2965 return WRITE_LIFE_NOT_SET;
2966 }
63189b78 2967 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2968 if (type == DATA) {
2969 if (temp == WARM)
2970 return WRITE_LIFE_LONG;
2971 else if (temp == HOT)
2972 return WRITE_LIFE_SHORT;
2973 else if (temp == COLD)
2974 return WRITE_LIFE_EXTREME;
2975 } else if (type == NODE) {
2976 if (temp == WARM || temp == HOT)
2977 return WRITE_LIFE_NOT_SET;
2978 else if (temp == COLD)
2979 return WRITE_LIFE_NONE;
2980 } else if (type == META) {
2981 return WRITE_LIFE_MEDIUM;
2982 }
0cdd3195 2983 }
f2e703f9 2984 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2985}
2986
81377bd6 2987static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2988{
81377bd6 2989 if (fio->type == DATA)
351df4b2
JK		 2990 return CURSEG_HOT_DATA;
2991 else
2992 return CURSEG_HOT_NODE;
2993}
2994
81377bd6 2995static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2996{
81377bd6
JK		 2997 if (fio->type == DATA) {
2998 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2999
3000 if (S_ISDIR(inode->i_mode))
3001 return CURSEG_HOT_DATA;
3002 else
3003 return CURSEG_COLD_DATA;
3004 } else {
81377bd6 3005 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 3006 return CURSEG_WARM_NODE;
351df4b2
JK		 3007 else
3008 return CURSEG_COLD_NODE;
3009 }
3010}
3011
81377bd6 3012static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 3013{
81377bd6
JK		 3014 if (fio->type == DATA) {
3015 struct inode *inode = fio->page->mapping->host;
351df4b2 3016
81377bd6 3017 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 3018 return CURSEG_COLD_DATA;
b6a06cbb 3019 if (file_is_hot(inode) ||
b4c3ca8b 3020 is_inode_flag_set(inode, FI_HOT_DATA) ||
2079f115
CY		 3021 f2fs_is_atomic_file(inode) ||
3022 f2fs_is_volatile_file(inode))
ef095d19 3023 return CURSEG_HOT_DATA;
4d57b86d 3024 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 3025 } else {
81377bd6
JK		 3026 if (IS_DNODE(fio->page))
3027 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 3028 CURSEG_HOT_NODE;
ef095d19 3029 return CURSEG_COLD_NODE;
351df4b2
JK		 3030 }
3031}
3032
81377bd6 3033static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 3034{
a912b54d
JK		 3035 int type = 0;
3036
63189b78 3037 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 3038 case 2:
a912b54d
JK		 3039 type = __get_segment_type_2(fio);
3040 break;
351df4b2 3041 case 4:
a912b54d
JK		 3042 type = __get_segment_type_4(fio);
3043 break;
3044 case 6:
3045 type = __get_segment_type_6(fio);
3046 break;
3047 default:
3048 f2fs_bug_on(fio->sbi, true);
351df4b2 3049 }
81377bd6 3050
a912b54d
JK		 3051 if (IS_HOT(type))
3052 fio->temp = HOT;
3053 else if (IS_WARM(type))
3054 fio->temp = WARM;
3055 else
3056 fio->temp = COLD;
3057 return type;
351df4b2
JK		 3058}
3059
4d57b86d 3060void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 3061 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 3062 struct f2fs_summary *sum, int type,
3063 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 3064{
3065 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 3066 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 3067
2b60311d
CY		 3068 down_read(&SM_I(sbi)->curseg_lock);
3069
351df4b2 3070 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3071 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3072
3073 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 3074
4e6a8d9b
JK		 3075 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3076
351df4b2
JK		 3077 /*
3078 * __add_sum_entry should be resided under the curseg_mutex
3079 * because, this function updates a summary entry in the
3080 * current summary block.
3081 */
e79efe3b 3082 __add_sum_entry(sbi, type, sum);
351df4b2 3083
351df4b2 3084 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 3085
3086 stat_inc_block_count(sbi, curseg);
351df4b2 3087
65f1b80b
YS		 3088 /*
3089 * SIT information should be updated before segment allocation,
3090 * since SSR needs latest valid block information.
3091 */
3092 update_sit_entry(sbi, *new_blkaddr, 1);
3093 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3094 update_sit_entry(sbi, old_blkaddr, -1);
3095
c6f82fe9
JK		 3096 if (!__has_curseg_space(sbi, type))
3097 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 3098
351df4b2 3099 /*
65f1b80b
YS		 3100 * segment dirty status should be updated after segment allocation,
3101 * so we just need to update status only one time after previous
3102 * segment being closed.
351df4b2 3103 */
65f1b80b
YS		 3104 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3105 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 3106
3d26fa6b 3107 up_write(&sit_i->sentry_lock);
351df4b2 3108
704956ec 3109 if (page && IS_NODESEG(type)) {
351df4b2
JK		 3110 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3111
704956ec
CY		 3112 f2fs_inode_chksum_set(sbi, page);
3113 }
3114
fb830fc5
CY		 3115 if (add_list) {
3116 struct f2fs_bio_info *io;
3117
3118 INIT_LIST_HEAD(&fio->list);
3119 fio->in_list = true;
fe16efe6 3120 fio->retry = false;
fb830fc5
CY		 3121 io = sbi->write_io[fio->type] + fio->temp;
3122 spin_lock(&io->io_lock);
3123 list_add_tail(&fio->list, &io->io_list);
3124 spin_unlock(&io->io_lock);
3125 }
3126
bfad7c2d 3127 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 3128
3129 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 3130}
3131
39d787be
CY		 3132static void update_device_state(struct f2fs_io_info *fio)
3133{
3134 struct f2fs_sb_info *sbi = fio->sbi;
3135 unsigned int devidx;
3136
0916878d 3137 if (!f2fs_is_multi_device(sbi))
39d787be
CY		 3138 return;
3139
3140 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3141
3142 /* update device state for fsync */
4d57b86d 3143 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 3144
3145 /* update device state for checkpoint */
3146 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3147 spin_lock(&sbi->dev_lock);
3148 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3149 spin_unlock(&sbi->dev_lock);
3150 }
39d787be
CY		 3151}
3152
05ca3632 3153static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 3154{
81377bd6 3155 int type = __get_segment_type(fio);
107a805d 3156 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 3157
107a805d
CY		 3158 if (keep_order)
3159 down_read(&fio->sbi->io_order_lock);
0a595eba 3160reallocate:
4d57b86d 3161 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 3162 &fio->new_blkaddr, sum, type, fio, true);
6aa58d8a
CY		 3163 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3164 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3165 fio->old_blkaddr, fio->old_blkaddr);
bfad7c2d 3166
351df4b2 3167 /* writeout dirty page into bdev */
fe16efe6
CY		 3168 f2fs_submit_page_write(fio);
3169 if (fio->retry) {
0a595eba
JK		 3170 fio->old_blkaddr = fio->new_blkaddr;
3171 goto reallocate;
3172 }
fe16efe6
CY		 3173
3174 update_device_state(fio);
3175
107a805d
CY		 3176 if (keep_order)
3177 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 3178}
3179
4d57b86d 3180void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 3181 enum iostat_type io_type)
351df4b2 3182{
458e6197 3183 struct f2fs_io_info fio = {
05ca3632 3184 .sbi = sbi,
458e6197 3185 .type = META,
0cdd3195 3186 .temp = HOT,
04d328de 3187 .op = REQ_OP_WRITE,
70fd7614 3188 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 3189 .old_blkaddr = page->index,
3190 .new_blkaddr = page->index,
05ca3632 3191 .page = page,
4375a336 3192 .encrypted_page = NULL,
fb830fc5 3193 .in_list = false,
458e6197
JK		 3194 };
3195
2b947003 3196 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 3197 fio.op_flags &= ~REQ_META;
2b947003 3198
351df4b2 3199 set_page_writeback(page);
17c50035 3200 ClearPageError(page);
b9109b0e 3201 f2fs_submit_page_write(&fio);
b0af6d49 3202
b63e7be5 3203 stat_inc_meta_count(sbi, page->index);
b0af6d49 3204 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 3205}
3206
4d57b86d 3207void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 3208{
3209 struct f2fs_summary sum;
05ca3632 3210
351df4b2 3211 set_summary(&sum, nid, 0, 0);
05ca3632 3212 do_write_page(&sum, fio);
b0af6d49
CY		 3213
3214 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3215}
3216
4d57b86d
CY		 3217void f2fs_outplace_write_data(struct dnode_of_data *dn,
3218 struct f2fs_io_info *fio)
351df4b2 3219{
05ca3632 3220 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2 3221 struct f2fs_summary sum;
351df4b2 3222
9850cf4a 3223 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
7735730d 3224 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
05ca3632 3225 do_write_page(&sum, fio);
f28b3434 3226 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 3227
3228 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3229}
3230
4d57b86d 3231int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 3232{
b0af6d49 3233 int err;
d21b0f23 3234 struct f2fs_sb_info *sbi = fio->sbi;
05573d6c 3235 unsigned int segno;
b0af6d49 3236
7a9d7548 3237 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 3238 /* i/o temperature is needed for passing down write hints */
3239 __get_segment_type(fio);
d21b0f23 3240
05573d6c
CY		 3241 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3242
3243 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3244 set_sbi_flag(sbi, SBI_NEED_FSCK);
2d821c12
CY		 3245 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3246 __func__, segno);
10f966bb 3247 return -EFSCORRUPTED;
05573d6c 3248 }
d21b0f23 3249
05ca3632 3250 stat_inc_inplace_blocks(fio->sbi);
b0af6d49 3251
8648de2c
CY		 3252 if (fio->bio)
3253 err = f2fs_merge_page_bio(fio);
3254 else
3255 err = f2fs_submit_page_bio(fio);
e46f6bd8 3256 if (!err) {
39d787be 3257 update_device_state(fio);
e46f6bd8
CY		 3258 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3259 }
b0af6d49
CY		 3260
3261 return err;
351df4b2
JK		 3262}
3263
2b60311d
CY		 3264static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3265 unsigned int segno)
3266{
3267 int i;
3268
3269 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3270 if (CURSEG_I(sbi, i)->segno == segno)
3271 break;
3272 }
3273 return i;
3274}
3275
4d57b86d 3276void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 3277 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 3278 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 3279{
3280 struct sit_info *sit_i = SIT_I(sbi);
3281 struct curseg_info *curseg;
3282 unsigned int segno, old_cursegno;
3283 struct seg_entry *se;
3284 int type;
19f106bc 3285 unsigned short old_blkoff;
351df4b2
JK		 3286
3287 segno = GET_SEGNO(sbi, new_blkaddr);
3288 se = get_seg_entry(sbi, segno);
3289 type = se->type;
3290
2b60311d
CY		 3291 down_write(&SM_I(sbi)->curseg_lock);
3292
19f106bc
CY		 3293 if (!recover_curseg) {
3294 /* for recovery flow */
3295 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3296 if (old_blkaddr == NULL_ADDR)
3297 type = CURSEG_COLD_DATA;
3298 else
3299 type = CURSEG_WARM_DATA;
3300 }
3301 } else {
2b60311d
CY		 3302 if (IS_CURSEG(sbi, segno)) {
3303 /* se->type is volatile as SSR allocation */
3304 type = __f2fs_get_curseg(sbi, segno);
3305 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3306 } else {
351df4b2 3307 type = CURSEG_WARM_DATA;
2b60311d 3308 }
351df4b2 3309 }
19f106bc 3310
2c190504 3311 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 3312 curseg = CURSEG_I(sbi, type);
3313
3314 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3315 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3316
3317 old_cursegno = curseg->segno;
19f106bc 3318 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 3319
3320 /* change the current segment */
3321 if (segno != curseg->segno) {
3322 curseg->next_segno = segno;
025d63a4 3323 change_curseg(sbi, type);
351df4b2
JK		 3324 }
3325
491c0854 3326 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 3327 __add_sum_entry(sbi, type, sum);
351df4b2 3328
28bc106b 3329 if (!recover_curseg || recover_newaddr)
6e2c64ad 3330 update_sit_entry(sbi, new_blkaddr, 1);
6aa58d8a
CY		 3331 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3332 invalidate_mapping_pages(META_MAPPING(sbi),
3333 old_blkaddr, old_blkaddr);
6e2c64ad 3334 update_sit_entry(sbi, old_blkaddr, -1);
6aa58d8a 3335 }
6e2c64ad
JK		 3336
3337 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3338 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3339
351df4b2 3340 locate_dirty_segment(sbi, old_cursegno);
351df4b2 3341
19f106bc
CY		 3342 if (recover_curseg) {
3343 if (old_cursegno != curseg->segno) {
3344 curseg->next_segno = old_cursegno;
025d63a4 3345 change_curseg(sbi, type);
19f106bc
CY		 3346 }
3347 curseg->next_blkoff = old_blkoff;
3348 }
3349
3d26fa6b 3350 up_write(&sit_i->sentry_lock);
351df4b2 3351 mutex_unlock(&curseg->curseg_mutex);
2b60311d 3352 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 3353}
3354
528e3459
CY		 3355void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3356 block_t old_addr, block_t new_addr,
28bc106b
CY		 3357 unsigned char version, bool recover_curseg,
3358 bool recover_newaddr)
528e3459
CY		 3359{
3360 struct f2fs_summary sum;
3361
3362 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3363
4d57b86d 3364 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
28bc106b 3365 recover_curseg, recover_newaddr);
528e3459 3366
f28b3434 3367 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 3368}
3369
93dfe2ac 3370void f2fs_wait_on_page_writeback(struct page *page,
bae0ee7a 3371 enum page_type type, bool ordered, bool locked)
93dfe2ac 3372{
93dfe2ac 3373 if (PageWriteback(page)) {
4081363f
JK		 3374 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3375
bab475c5 3376 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
bae0ee7a 3377 if (ordered) {
fec1d657 3378 wait_on_page_writeback(page);
bae0ee7a
CY		 3379 f2fs_bug_on(sbi, locked && PageWriteback(page));
3380 } else {
fec1d657 3381 wait_for_stable_page(page);
bae0ee7a 3382 }
93dfe2ac
JK		 3383 }
3384}
3385
0ded69f6 3386void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
08b39fbd 3387{
0ded69f6 3388 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
08b39fbd
CY		 3389 struct page *cpage;
3390
0ded69f6
JK		 3391 if (!f2fs_post_read_required(inode))
3392 return;
3393
93770ab7 3394 if (!__is_valid_data_blkaddr(blkaddr))
08b39fbd
CY		 3395 return;
3396
08b39fbd
CY		 3397 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3398 if (cpage) {
bae0ee7a 3399 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
08b39fbd
CY		 3400 f2fs_put_page(cpage, 1);
3401 }
3402}
3403
1e78e8bd
ST		 3404void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3405 block_t len)
3406{
3407 block_t i;
3408
3409 for (i = 0; i < len; i++)
3410 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3411}
3412
7735730d 3413static int read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3414{
3415 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3416 struct curseg_info *seg_i;
3417 unsigned char *kaddr;
3418 struct page *page;
3419 block_t start;
3420 int i, j, offset;
3421
3422 start = start_sum_block(sbi);
3423
4d57b86d 3424 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3425 if (IS_ERR(page))
3426 return PTR_ERR(page);
351df4b2
JK		 3427 kaddr = (unsigned char *)page_address(page);
3428
3429 /* Step 1: restore nat cache */
3430 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3431 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3432
3433 /* Step 2: restore sit cache */
3434 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3435 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3436 offset = 2 * SUM_JOURNAL_SIZE;
3437
3438 /* Step 3: restore summary entries */
3439 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3440 unsigned short blk_off;
3441 unsigned int segno;
3442
3443 seg_i = CURSEG_I(sbi, i);
3444 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3445 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3446 seg_i->next_segno = segno;
3447 reset_curseg(sbi, i, 0);
3448 seg_i->alloc_type = ckpt->alloc_type[i];
3449 seg_i->next_blkoff = blk_off;
3450
3451 if (seg_i->alloc_type == SSR)
3452 blk_off = sbi->blocks_per_seg;
3453
3454 for (j = 0; j < blk_off; j++) {
3455 struct f2fs_summary *s;
3456 s = (struct f2fs_summary *)(kaddr + offset);
3457 seg_i->sum_blk->entries[j] = *s;
3458 offset += SUMMARY_SIZE;
09cbfeaf 3459 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3460 SUM_FOOTER_SIZE)
3461 continue;
3462
3463 f2fs_put_page(page, 1);
3464 page = NULL;
3465
4d57b86d 3466 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3467 if (IS_ERR(page))
3468 return PTR_ERR(page);
351df4b2
JK		 3469 kaddr = (unsigned char *)page_address(page);
3470 offset = 0;
3471 }
3472 }
3473 f2fs_put_page(page, 1);
7735730d 3474 return 0;
351df4b2
JK		 3475}
3476
3477static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3478{
3479 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3480 struct f2fs_summary_block *sum;
3481 struct curseg_info *curseg;
3482 struct page *new;
3483 unsigned short blk_off;
3484 unsigned int segno = 0;
3485 block_t blk_addr = 0;
7735730d 3486 int err = 0;
351df4b2
JK		 3487
3488 /* get segment number and block addr */
3489 if (IS_DATASEG(type)) {
3490 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3491 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3492 CURSEG_HOT_DATA]);
119ee914 3493 if (__exist_node_summaries(sbi))
351df4b2
JK		 3494 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3495 else
3496 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3497 } else {
3498 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3499 CURSEG_HOT_NODE]);
3500 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3501 CURSEG_HOT_NODE]);
119ee914 3502 if (__exist_node_summaries(sbi))
351df4b2
JK		 3503 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3504 type - CURSEG_HOT_NODE);
3505 else
3506 blk_addr = GET_SUM_BLOCK(sbi, segno);
3507 }
3508
4d57b86d 3509 new = f2fs_get_meta_page(sbi, blk_addr);
7735730d
CY		 3510 if (IS_ERR(new))
3511 return PTR_ERR(new);
351df4b2
JK		 3512 sum = (struct f2fs_summary_block *)page_address(new);
3513
3514 if (IS_NODESEG(type)) {
119ee914 3515 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3516 struct f2fs_summary *ns = &sum->entries[0];
3517 int i;
3518 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3519 ns->version = 0;
3520 ns->ofs_in_node = 0;
3521 }
3522 } else {
7735730d
CY		 3523 err = f2fs_restore_node_summary(sbi, segno, sum);
3524 if (err)
3525 goto out;
351df4b2
JK		 3526 }
3527 }
3528
3529 /* set uncompleted segment to curseg */
3530 curseg = CURSEG_I(sbi, type);
3531 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3532
3533 /* update journal info */
3534 down_write(&curseg->journal_rwsem);
3535 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3536 up_write(&curseg->journal_rwsem);
3537
3538 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3539 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3540 curseg->next_segno = segno;
3541 reset_curseg(sbi, type, 0);
3542 curseg->alloc_type = ckpt->alloc_type[type];
3543 curseg->next_blkoff = blk_off;
3544 mutex_unlock(&curseg->curseg_mutex);
7735730d 3545out:
351df4b2 3546 f2fs_put_page(new, 1);
7735730d 3547 return err;
351df4b2
JK		 3548}
3549
3550static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3551{
21d3f8e1
JQ		 3552 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3553 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3554 int type = CURSEG_HOT_DATA;
e4fc5fbf 3555 int err;
351df4b2 3556
aaec2b1d 3557 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3558 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3559
3560 if (npages >= 2)
4d57b86d 3561 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3562 META_CP, true);
3fa06d7b 3563
351df4b2 3564 /* restore for compacted data summary */
7735730d
CY		 3565 err = read_compacted_summaries(sbi);
3566 if (err)
3567 return err;
351df4b2
JK		 3568 type = CURSEG_HOT_NODE;
3569 }
3570
119ee914 3571 if (__exist_node_summaries(sbi))
4d57b86d 3572 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3573 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3574
e4fc5fbf
CY		 3575 for (; type <= CURSEG_COLD_NODE; type++) {
3576 err = read_normal_summaries(sbi, type);
3577 if (err)
3578 return err;
3579 }
3580
21d3f8e1
JQ		 3581 /* sanity check for summary blocks */
3582 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
9227d522 3583 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
dcbb4c10
JP		 3584 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3585 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
21d3f8e1 3586 return -EINVAL;
9227d522 3587 }
21d3f8e1 3588
351df4b2
JK		 3589 return 0;
3590}
3591
3592static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3593{
3594 struct page *page;
3595 unsigned char *kaddr;
3596 struct f2fs_summary *summary;
3597 struct curseg_info *seg_i;
3598 int written_size = 0;
3599 int i, j;
3600
4d57b86d 3601 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3602 kaddr = (unsigned char *)page_address(page);
81114baa 3603 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3604
3605 /* Step 1: write nat cache */
3606 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3607 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3608 written_size += SUM_JOURNAL_SIZE;
3609
3610 /* Step 2: write sit cache */
3611 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3612 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3613 written_size += SUM_JOURNAL_SIZE;
3614
351df4b2
JK		 3615 /* Step 3: write summary entries */
3616 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3617 unsigned short blkoff;
3618 seg_i = CURSEG_I(sbi, i);
3619 if (sbi->ckpt->alloc_type[i] == SSR)
3620 blkoff = sbi->blocks_per_seg;
3621 else
3622 blkoff = curseg_blkoff(sbi, i);
3623
3624 for (j = 0; j < blkoff; j++) {
3625 if (!page) {
4d57b86d 3626 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3627 kaddr = (unsigned char *)page_address(page);
81114baa 3628 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3629 written_size = 0;
3630 }
3631 summary = (struct f2fs_summary *)(kaddr + written_size);
3632 *summary = seg_i->sum_blk->entries[j];
3633 written_size += SUMMARY_SIZE;
351df4b2 3634
09cbfeaf 3635 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3636 SUM_FOOTER_SIZE)
3637 continue;
3638
e8d61a74 3639 set_page_dirty(page);
351df4b2
JK		 3640 f2fs_put_page(page, 1);
3641 page = NULL;
3642 }
3643 }
e8d61a74
CY		 3644 if (page) {
3645 set_page_dirty(page);
351df4b2 3646 f2fs_put_page(page, 1);
e8d61a74 3647 }
351df4b2
JK		 3648}
3649
3650static void write_normal_summaries(struct f2fs_sb_info *sbi,
3651 block_t blkaddr, int type)
3652{
3653 int i, end;
3654 if (IS_DATASEG(type))
3655 end = type + NR_CURSEG_DATA_TYPE;
3656 else
3657 end = type + NR_CURSEG_NODE_TYPE;
3658
b7ad7512
CY		 3659 for (i = type; i < end; i++)
3660 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3661}
3662
4d57b86d 3663void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3664{
aaec2b1d 3665 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3666 write_compacted_summaries(sbi, start_blk);
3667 else
3668 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3669}
3670
4d57b86d 3671void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3672{
119ee914 3673 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3674}
3675
4d57b86d 3676int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3677 unsigned int val, int alloc)
3678{
3679 int i;
3680
3681 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3682 for (i = 0; i < nats_in_cursum(journal); i++) {
3683 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3684 return i;
3685 }
dfc08a12
CY		 3686 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3687 return update_nats_in_cursum(journal, 1);
351df4b2 3688 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3689 for (i = 0; i < sits_in_cursum(journal); i++)
3690 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3691 return i;
dfc08a12
CY		 3692 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3693 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3694 }
3695 return -1;
3696}
3697
3698static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3699 unsigned int segno)
3700{
7735730d 3701 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3702}
3703
3704static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3705 unsigned int start)
3706{
3707 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3708 struct page *page;
351df4b2 3709 pgoff_t src_off, dst_off;
351df4b2
JK		 3710
3711 src_off = current_sit_addr(sbi, start);
3712 dst_off = next_sit_addr(sbi, src_off);
3713
4d57b86d 3714 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 3715 seg_info_to_sit_page(sbi, page, start);
351df4b2 3716
068c3cd8 3717 set_page_dirty(page);
351df4b2
JK		 3718 set_to_next_sit(sit_i, start);
3719
068c3cd8 3720 return page;
351df4b2
JK		 3721}
3722
184a5cd2
CY		 3723static struct sit_entry_set *grab_sit_entry_set(void)
3724{
3725 struct sit_entry_set *ses =
80c54505 3726 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3727
3728 ses->entry_cnt = 0;
3729 INIT_LIST_HEAD(&ses->set_list);
3730 return ses;
3731}
3732
3733static void release_sit_entry_set(struct sit_entry_set *ses)
3734{
3735 list_del(&ses->set_list);
3736 kmem_cache_free(sit_entry_set_slab, ses);
3737}
3738
3739static void adjust_sit_entry_set(struct sit_entry_set *ses,
3740 struct list_head *head)
3741{
3742 struct sit_entry_set *next = ses;
3743
3744 if (list_is_last(&ses->set_list, head))
3745 return;
3746
3747 list_for_each_entry_continue(next, head, set_list)
3748 if (ses->entry_cnt <= next->entry_cnt)
3749 break;
3750
3751 list_move_tail(&ses->set_list, &next->set_list);
3752}
3753
3754static void add_sit_entry(unsigned int segno, struct list_head *head)
3755{
3756 struct sit_entry_set *ses;
3757 unsigned int start_segno = START_SEGNO(segno);
3758
3759 list_for_each_entry(ses, head, set_list) {
3760 if (ses->start_segno == start_segno) {
3761 ses->entry_cnt++;
3762 adjust_sit_entry_set(ses, head);
3763 return;
3764 }
3765 }
3766
3767 ses = grab_sit_entry_set();
3768
3769 ses->start_segno = start_segno;
3770 ses->entry_cnt++;
3771 list_add(&ses->set_list, head);
3772}
3773
3774static void add_sits_in_set(struct f2fs_sb_info *sbi)
3775{
3776 struct f2fs_sm_info *sm_info = SM_I(sbi);
3777 struct list_head *set_list = &sm_info->sit_entry_set;
3778 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3779 unsigned int segno;
3780
7cd8558b 3781 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3782 add_sit_entry(segno, set_list);
3783}
3784
3785static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3786{
3787 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3788 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3789 int i;
3790
b7ad7512 3791 down_write(&curseg->journal_rwsem);
dfc08a12 3792 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3793 unsigned int segno;
3794 bool dirtied;
3795
dfc08a12 3796 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3797 dirtied = __mark_sit_entry_dirty(sbi, segno);
3798
3799 if (!dirtied)
3800 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3801 }
dfc08a12 3802 update_sits_in_cursum(journal, -i);
b7ad7512 3803 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3804}
3805
0a8165d7 3806/*
351df4b2
JK		 3807 * CP calls this function, which flushes SIT entries including sit_journal,
3808 * and moves prefree segs to free segs.
3809 */
4d57b86d 3810void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3811{
3812 struct sit_info *sit_i = SIT_I(sbi);
3813 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3814 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3815 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3816 struct sit_entry_set *ses, *tmp;
3817 struct list_head *head = &SM_I(sbi)->sit_entry_set;
04f0b2ea 3818 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4b2fecc8 3819 struct seg_entry *se;
351df4b2 3820
3d26fa6b 3821 down_write(&sit_i->sentry_lock);
351df4b2 3822
2b11a74b
WL		 3823 if (!sit_i->dirty_sentries)
3824 goto out;
3825
351df4b2 3826 /*
184a5cd2
CY		 3827 * add and account sit entries of dirty bitmap in sit entry
3828 * set temporarily
351df4b2 3829 */
184a5cd2 3830 add_sits_in_set(sbi);
351df4b2 3831
184a5cd2
CY		 3832 /*
3833 * if there are no enough space in journal to store dirty sit
3834 * entries, remove all entries from journal and add and account
3835 * them in sit entry set.
3836 */
04f0b2ea
QS		 3837 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3838 !to_journal)
184a5cd2 3839 remove_sits_in_journal(sbi);
b2955550 3840
184a5cd2
CY		 3841 /*
3842 * there are two steps to flush sit entries:
3843 * #1, flush sit entries to journal in current cold data summary block.
3844 * #2, flush sit entries to sit page.
3845 */
3846 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3847 struct page *page = NULL;
184a5cd2
CY		 3848 struct f2fs_sit_block *raw_sit = NULL;
3849 unsigned int start_segno = ses->start_segno;
3850 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3851 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3852 unsigned int segno = start_segno;
3853
3854 if (to_journal &&
dfc08a12 3855 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3856 to_journal = false;
3857
b7ad7512
CY		 3858 if (to_journal) {
3859 down_write(&curseg->journal_rwsem);
3860 } else {
184a5cd2
CY		 3861 page = get_next_sit_page(sbi, start_segno);
3862 raw_sit = page_address(page);
351df4b2 3863 }
351df4b2 3864
184a5cd2
CY		 3865 /* flush dirty sit entries in region of current sit set */
3866 for_each_set_bit_from(segno, bitmap, end) {
3867 int offset, sit_offset;
4b2fecc8
JK		 3868
3869 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3870#ifdef CONFIG_F2FS_CHECK_FS
3871 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3872 SIT_VBLOCK_MAP_SIZE))
3873 f2fs_bug_on(sbi, 1);
3874#endif
184a5cd2
CY		 3875
3876 /* add discard candidates */
c473f1a9 3877 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3878 cpc->trim_start = segno;
25290fa5 3879 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3880 }
184a5cd2
CY		 3881
3882 if (to_journal) {
4d57b86d 3883 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 3884 SIT_JOURNAL, segno, 1);
3885 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3886 segno_in_journal(journal, offset) =
184a5cd2
CY		 3887 cpu_to_le32(segno);
3888 seg_info_to_raw_sit(se,
dfc08a12 3889 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3890 check_block_count(sbi, segno,
3891 &sit_in_journal(journal, offset));
184a5cd2
CY		 3892 } else {
3893 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3894 seg_info_to_raw_sit(se,
3895 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3896 check_block_count(sbi, segno,
3897 &raw_sit->entries[sit_offset]);
184a5cd2 3898 }
351df4b2 3899
184a5cd2
CY		 3900 __clear_bit(segno, bitmap);
3901 sit_i->dirty_sentries--;
3902 ses->entry_cnt--;
351df4b2
JK		 3903 }
3904
b7ad7512
CY		 3905 if (to_journal)
3906 up_write(&curseg->journal_rwsem);
3907 else
184a5cd2
CY		 3908 f2fs_put_page(page, 1);
3909
3910 f2fs_bug_on(sbi, ses->entry_cnt);
3911 release_sit_entry_set(ses);
351df4b2 3912 }
184a5cd2
CY		 3913
3914 f2fs_bug_on(sbi, !list_empty(head));
3915 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3916out:
c473f1a9 3917 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3918 __u64 trim_start = cpc->trim_start;
3919
4b2fecc8 3920 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3921 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3922
3923 cpc->trim_start = trim_start;
4b2fecc8 3924 }
3d26fa6b 3925 up_write(&sit_i->sentry_lock);
351df4b2 3926
351df4b2
JK		 3927 set_prefree_as_free_segments(sbi);
3928}
3929
3930static int build_sit_info(struct f2fs_sb_info *sbi)
3931{
3932 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3933 struct sit_info *sit_i;
3934 unsigned int sit_segs, start;
ae27d62e 3935 char *src_bitmap;
351df4b2
JK		 3936 unsigned int bitmap_size;
3937
3938 /* allocate memory for SIT information */
acbf054d 3939 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3940 if (!sit_i)
3941 return -ENOMEM;
3942
3943 SM_I(sbi)->sit_info = sit_i;
3944
9d2a789c
KC		 3945 sit_i->sentries =
3946 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3947 MAIN_SEGS(sbi)),
3948 GFP_KERNEL);
351df4b2
JK		 3949 if (!sit_i->sentries)
3950 return -ENOMEM;
3951
7cd8558b 3952 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3953 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3954 GFP_KERNEL);
351df4b2
JK		 3955 if (!sit_i->dirty_sentries_bitmap)
3956 return -ENOMEM;
3957
7cd8558b 3958 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3959 sit_i->sentries[start].cur_valid_map
acbf054d 3960 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3961 sit_i->sentries[start].ckpt_valid_map
acbf054d 3962 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3963 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3964 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3965 return -ENOMEM;
3e025740 3966
355e7891
CY		 3967#ifdef CONFIG_F2FS_CHECK_FS
3968 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3969 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3970 if (!sit_i->sentries[start].cur_valid_map_mir)
3971 return -ENOMEM;
3972#endif
3973
7d20c8ab
CY		 3974 sit_i->sentries[start].discard_map
3975 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3976 GFP_KERNEL);
3977 if (!sit_i->sentries[start].discard_map)
3978 return -ENOMEM;
351df4b2
JK		 3979 }
3980
acbf054d 3981 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3982 if (!sit_i->tmp_map)
3983 return -ENOMEM;
3984
2c70c5e3 3985 if (__is_large_section(sbi)) {
9d2a789c
KC		 3986 sit_i->sec_entries =
3987 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3988 MAIN_SECS(sbi)),
3989 GFP_KERNEL);
351df4b2
JK		 3990 if (!sit_i->sec_entries)
3991 return -ENOMEM;
3992 }
3993
3994 /* get information related with SIT */
3995 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3996
3997 /* setup SIT bitmap from ckeckpoint pack */
3998 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3999 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4000
ae27d62e
CY		 4001 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4002 if (!sit_i->sit_bitmap)
351df4b2 4003 return -ENOMEM;
351df4b2 4004
ae27d62e
CY		 4005#ifdef CONFIG_F2FS_CHECK_FS
4006 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4007 if (!sit_i->sit_bitmap_mir)
4008 return -ENOMEM;
4009#endif
4010
351df4b2
JK		 4011 /* init SIT information */
4012 sit_i->s_ops = &default_salloc_ops;
4013
4014 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4015 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 4016 sit_i->written_valid_blocks = 0;
351df4b2
JK		 4017 sit_i->bitmap_size = bitmap_size;
4018 sit_i->dirty_sentries = 0;
4019 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4020 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 4021 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 4022 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 4023 return 0;
4024}
4025
4026static int build_free_segmap(struct f2fs_sb_info *sbi)
4027{
351df4b2
JK		 4028 struct free_segmap_info *free_i;
4029 unsigned int bitmap_size, sec_bitmap_size;
4030
4031 /* allocate memory for free segmap information */
acbf054d 4032 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 4033 if (!free_i)
4034 return -ENOMEM;
4035
4036 SM_I(sbi)->free_info = free_i;
4037
7cd8558b 4038 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 4039 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 4040 if (!free_i->free_segmap)
4041 return -ENOMEM;
4042
7cd8558b 4043 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 4044 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 4045 if (!free_i->free_secmap)
4046 return -ENOMEM;
4047
4048 /* set all segments as dirty temporarily */
4049 memset(free_i->free_segmap, 0xff, bitmap_size);
4050 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4051
4052 /* init free segmap information */
7cd8558b 4053 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 4054 free_i->free_segments = 0;
4055 free_i->free_sections = 0;
1a118ccf 4056 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 4057 return 0;
4058}
4059
4060static int build_curseg(struct f2fs_sb_info *sbi)
4061{
1042d60f 4062 struct curseg_info *array;
351df4b2
JK		 4063 int i;
4064
026f0507
KC		 4065 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4066 GFP_KERNEL);
351df4b2
JK		 4067 if (!array)
4068 return -ENOMEM;
4069
4070 SM_I(sbi)->curseg_array = array;
4071
4072 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4073 mutex_init(&array[i].curseg_mutex);
acbf054d 4074 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 4075 if (!array[i].sum_blk)
4076 return -ENOMEM;
b7ad7512 4077 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 4078 array[i].journal = f2fs_kzalloc(sbi,
4079 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 4080 if (!array[i].journal)
4081 return -ENOMEM;
351df4b2
JK		 4082 array[i].segno = NULL_SEGNO;
4083 array[i].next_blkoff = 0;
4084 }
4085 return restore_curseg_summaries(sbi);
4086}
4087
c39a1b34 4088static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 4089{
4090 struct sit_info *sit_i = SIT_I(sbi);
4091 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 4092 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 4093 struct seg_entry *se;
4094 struct f2fs_sit_entry sit;
74de593a
CY		 4095 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4096 unsigned int i, start, end;
4097 unsigned int readed, start_blk = 0;
c39a1b34 4098 int err = 0;
8a29c126 4099 block_t total_node_blocks = 0;
351df4b2 4100
74de593a 4101 do {
4d57b86d 4102 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
664ba972 4103 META_SIT, true);
74de593a
CY		 4104
4105 start = start_blk * sit_i->sents_per_block;
4106 end = (start_blk + readed) * sit_i->sents_per_block;
4107
7cd8558b 4108 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 4109 struct f2fs_sit_block *sit_blk;
74de593a
CY		 4110 struct page *page;
4111
9c094040 4112 se = &sit_i->sentries[start];
74de593a 4113 page = get_current_sit_page(sbi, start);
edc55aaf
JK		 4114 if (IS_ERR(page))
4115 return PTR_ERR(page);
74de593a
CY		 4116 sit_blk = (struct f2fs_sit_block *)page_address(page);
4117 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4118 f2fs_put_page(page, 1);
d600af23 4119
c39a1b34
JK		 4120 err = check_block_count(sbi, start, &sit);
4121 if (err)
4122 return err;
74de593a 4123 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4124 if (IS_NODESEG(se->type))
4125 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 4126
4127 /* build discard map only one time */
7d20c8ab
CY		 4128 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4129 memset(se->discard_map, 0xff,
4130 SIT_VBLOCK_MAP_SIZE);
4131 } else {
4132 memcpy(se->discard_map,
4133 se->cur_valid_map,
4134 SIT_VBLOCK_MAP_SIZE);
4135 sbi->discard_blks +=
4136 sbi->blocks_per_seg -
4137 se->valid_blocks;
3e025740 4138 }
a66cdd98 4139
2c70c5e3 4140 if (__is_large_section(sbi))
d600af23
CY		 4141 get_sec_entry(sbi, start)->valid_blocks +=
4142 se->valid_blocks;
351df4b2 4143 }
74de593a
CY		 4144 start_blk += readed;
4145 } while (start_blk < sit_blk_cnt);
d600af23
CY		 4146
4147 down_read(&curseg->journal_rwsem);
4148 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 4149 unsigned int old_valid_blocks;
4150
4151 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f 4152 if (start >= MAIN_SEGS(sbi)) {
dcbb4c10
JP		 4153 f2fs_err(sbi, "Wrong journal entry on segno %u",
4154 start);
b2ca374f 4155 set_sbi_flag(sbi, SBI_NEED_FSCK);
10f966bb 4156 err = -EFSCORRUPTED;
b2ca374f
JK		 4157 break;
4158 }
4159
d600af23
CY		 4160 se = &sit_i->sentries[start];
4161 sit = sit_in_journal(journal, i);
4162
4163 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 4164 if (IS_NODESEG(se->type))
4165 total_node_blocks -= old_valid_blocks;
d600af23 4166
c39a1b34
JK		 4167 err = check_block_count(sbi, start, &sit);
4168 if (err)
4169 break;
d600af23 4170 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4171 if (IS_NODESEG(se->type))
4172 total_node_blocks += se->valid_blocks;
d600af23 4173
7d20c8ab
CY		 4174 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4175 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4176 } else {
4177 memcpy(se->discard_map, se->cur_valid_map,
4178 SIT_VBLOCK_MAP_SIZE);
4179 sbi->discard_blks += old_valid_blocks;
4180 sbi->discard_blks -= se->valid_blocks;
d600af23
CY		 4181 }
4182
2c70c5e3 4183 if (__is_large_section(sbi)) {
d600af23 4184 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 4185 se->valid_blocks;
4186 get_sec_entry(sbi, start)->valid_blocks -=
4187 old_valid_blocks;
4188 }
d600af23
CY		 4189 }
4190 up_read(&curseg->journal_rwsem);
8a29c126
JK		 4191
4192 if (!err && total_node_blocks != valid_node_count(sbi)) {
dcbb4c10
JP		 4193 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4194 total_node_blocks, valid_node_count(sbi));
8a29c126 4195 set_sbi_flag(sbi, SBI_NEED_FSCK);
10f966bb 4196 err = -EFSCORRUPTED;
8a29c126
JK		 4197 }
4198
c39a1b34 4199 return err;
351df4b2
JK		 4200}
4201
4202static void init_free_segmap(struct f2fs_sb_info *sbi)
4203{
4204 unsigned int start;
4205 int type;
4206
7cd8558b 4207 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 4208 struct seg_entry *sentry = get_seg_entry(sbi, start);
4209 if (!sentry->valid_blocks)
4210 __set_free(sbi, start);
c79b7ff1
JK		 4211 else
4212 SIT_I(sbi)->written_valid_blocks +=
4213 sentry->valid_blocks;
351df4b2
JK		 4214 }
4215
4216 /* set use the current segments */
4217 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4218 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4219 __set_test_and_inuse(sbi, curseg_t->segno);
4220 }
4221}
4222
4223static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4224{
4225 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4226 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 4227 unsigned int segno = 0, offset = 0;
351df4b2
JK		 4228 unsigned short valid_blocks;
4229
8736fbf0 4230 while (1) {
351df4b2 4231 /* find dirty segment based on free segmap */
7cd8558b
JK		 4232 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4233 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 4234 break;
4235 offset = segno + 1;
302bd348 4236 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 4237 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 4238 continue;
ec325b52
JK		 4239 if (valid_blocks > sbi->blocks_per_seg) {
4240 f2fs_bug_on(sbi, 1);
4241 continue;
4242 }
351df4b2
JK		 4243 mutex_lock(&dirty_i->seglist_lock);
4244 __locate_dirty_segment(sbi, segno, DIRTY);
4245 mutex_unlock(&dirty_i->seglist_lock);
4246 }
4247}
4248
5ec4e49f 4249static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4250{
4251 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 4252 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 4253
628b3d14 4254 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 4255 if (!dirty_i->victim_secmap)
351df4b2
JK		 4256 return -ENOMEM;
4257 return 0;
4258}
4259
4260static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4261{
4262 struct dirty_seglist_info *dirty_i;
4263 unsigned int bitmap_size, i;
4264
4265 /* allocate memory for dirty segments list information */
acbf054d
CY		 4266 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4267 GFP_KERNEL);
351df4b2
JK		 4268 if (!dirty_i)
4269 return -ENOMEM;
4270
4271 SM_I(sbi)->dirty_info = dirty_i;
4272 mutex_init(&dirty_i->seglist_lock);
4273
7cd8558b 4274 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 4275
4276 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 4277 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4278 GFP_KERNEL);
351df4b2
JK		 4279 if (!dirty_i->dirty_segmap[i])
4280 return -ENOMEM;
4281 }
4282
4283 init_dirty_segmap(sbi);
5ec4e49f 4284 return init_victim_secmap(sbi);
351df4b2
JK		 4285}
4286
c854f4d6
CY		 4287static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4288{
4289 int i;
4290
4291 /*
4292 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4293 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4294 */
4295 for (i = 0; i < NO_CHECK_TYPE; i++) {
4296 struct curseg_info *curseg = CURSEG_I(sbi, i);
4297 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4298 unsigned int blkofs = curseg->next_blkoff;
4299
4300 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4301 goto out;
4302
4303 if (curseg->alloc_type == SSR)
4304 continue;
4305
4306 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4307 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4308 continue;
4309out:
dcbb4c10
JP		 4310 f2fs_err(sbi,
4311 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4312 i, curseg->segno, curseg->alloc_type,
4313 curseg->next_blkoff, blkofs);
10f966bb 4314 return -EFSCORRUPTED;
c854f4d6
CY		 4315 }
4316 }
4317 return 0;
4318}
4319
0a8165d7 4320/*
351df4b2
JK		 4321 * Update min, max modified time for cost-benefit GC algorithm
4322 */
4323static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4324{
4325 struct sit_info *sit_i = SIT_I(sbi);
4326 unsigned int segno;
4327
3d26fa6b 4328 down_write(&sit_i->sentry_lock);
351df4b2 4329
5ad25442 4330 sit_i->min_mtime = ULLONG_MAX;
351df4b2 4331
7cd8558b 4332 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 4333 unsigned int i;
4334 unsigned long long mtime = 0;
4335
4336 for (i = 0; i < sbi->segs_per_sec; i++)
4337 mtime += get_seg_entry(sbi, segno + i)->mtime;
4338
4339 mtime = div_u64(mtime, sbi->segs_per_sec);
4340
4341 if (sit_i->min_mtime > mtime)
4342 sit_i->min_mtime = mtime;
4343 }
a1f72ac2 4344 sit_i->max_mtime = get_mtime(sbi, false);
3d26fa6b 4345 up_write(&sit_i->sentry_lock);
351df4b2
JK		 4346}
4347
4d57b86d 4348int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4349{
4350 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4351 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 4352 struct f2fs_sm_info *sm_info;
351df4b2
JK		 4353 int err;
4354
acbf054d 4355 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 4356 if (!sm_info)
4357 return -ENOMEM;
4358
4359 /* init sm info */
4360 sbi->sm_info = sm_info;
351df4b2
JK		 4361 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4362 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4363 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4364 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4365 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4366 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4367 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 4368 sm_info->rec_prefree_segments = sm_info->main_segments *
4369 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 4370 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4371 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4372
52763a4b
JK		 4373 if (!test_opt(sbi, LFS))
4374 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 4375 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 4376 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
853137ce 4377 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
ef095d19 4378 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 4379 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 4380
184a5cd2
CY		 4381 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4382
2b60311d
CY		 4383 init_rwsem(&sm_info->curseg_lock);
4384
d4fdf8ba 4385 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 4386 err = f2fs_create_flush_cmd_control(sbi);
2163d198 4387 if (err)
a688b9d9 4388 return err;
6b4afdd7
JK		 4389 }
4390
0b54fb84
JK		 4391 err = create_discard_cmd_control(sbi);
4392 if (err)
4393 return err;
4394
351df4b2
JK		 4395 err = build_sit_info(sbi);
4396 if (err)
4397 return err;
4398 err = build_free_segmap(sbi);
4399 if (err)
4400 return err;
4401 err = build_curseg(sbi);
4402 if (err)
4403 return err;
4404
4405 /* reinit free segmap based on SIT */
c39a1b34
JK		 4406 err = build_sit_entries(sbi);
4407 if (err)
4408 return err;
351df4b2
JK		 4409
4410 init_free_segmap(sbi);
4411 err = build_dirty_segmap(sbi);
4412 if (err)
4413 return err;
4414
c854f4d6
CY		 4415 err = sanity_check_curseg(sbi);
4416 if (err)
4417 return err;
4418
351df4b2
JK		 4419 init_min_max_mtime(sbi);
4420 return 0;
4421}
4422
4423static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4424 enum dirty_type dirty_type)
4425{
4426 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4427
4428 mutex_lock(&dirty_i->seglist_lock);
39307a8e 4429 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 4430 dirty_i->nr_dirty[dirty_type] = 0;
4431 mutex_unlock(&dirty_i->seglist_lock);
4432}
4433
5ec4e49f 4434static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4435{
4436 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 4437 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 4438}
4439
4440static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4441{
4442 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4443 int i;
4444
4445 if (!dirty_i)
4446 return;
4447
4448 /* discard pre-free/dirty segments list */
4449 for (i = 0; i < NR_DIRTY_TYPE; i++)
4450 discard_dirty_segmap(sbi, i);
4451
5ec4e49f 4452 destroy_victim_secmap(sbi);
351df4b2 4453 SM_I(sbi)->dirty_info = NULL;
5222595d 4454 kvfree(dirty_i);
351df4b2
JK		 4455}
4456
4457static void destroy_curseg(struct f2fs_sb_info *sbi)
4458{
4459 struct curseg_info *array = SM_I(sbi)->curseg_array;
4460 int i;
4461
4462 if (!array)
4463 return;
4464 SM_I(sbi)->curseg_array = NULL;
b7ad7512 4465 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5222595d
JK		 4466 kvfree(array[i].sum_blk);
4467 kvfree(array[i].journal);
b7ad7512 4468 }
5222595d 4469 kvfree(array);
351df4b2
JK		 4470}
4471
4472static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4473{
4474 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4475 if (!free_i)
4476 return;
4477 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4478 kvfree(free_i->free_segmap);
4479 kvfree(free_i->free_secmap);
5222595d 4480 kvfree(free_i);
351df4b2
JK		 4481}
4482
4483static void destroy_sit_info(struct f2fs_sb_info *sbi)
4484{
4485 struct sit_info *sit_i = SIT_I(sbi);
4486 unsigned int start;
4487
4488 if (!sit_i)
4489 return;
4490
4491 if (sit_i->sentries) {
7cd8558b 4492 for (start = 0; start < MAIN_SEGS(sbi); start++) {
5222595d 4493 kvfree(sit_i->sentries[start].cur_valid_map);
355e7891 4494#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4495 kvfree(sit_i->sentries[start].cur_valid_map_mir);
355e7891 4496#endif
5222595d
JK		 4497 kvfree(sit_i->sentries[start].ckpt_valid_map);
4498 kvfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4499 }
4500 }
5222595d 4501 kvfree(sit_i->tmp_map);
60a3b782 4502
39307a8e
JK		 4503 kvfree(sit_i->sentries);
4504 kvfree(sit_i->sec_entries);
4505 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4506
4507 SM_I(sbi)->sit_info = NULL;
5222595d 4508 kvfree(sit_i->sit_bitmap);
ae27d62e 4509#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4510 kvfree(sit_i->sit_bitmap_mir);
ae27d62e 4511#endif
5222595d 4512 kvfree(sit_i);
351df4b2
JK		 4513}
4514
4d57b86d 4515void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4516{
4517 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4518
3b03f724
CY		 4519 if (!sm_info)
4520 return;
4d57b86d 4521 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 4522 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4523 destroy_dirty_segmap(sbi);
4524 destroy_curseg(sbi);
4525 destroy_free_segmap(sbi);
4526 destroy_sit_info(sbi);
4527 sbi->sm_info = NULL;
5222595d 4528 kvfree(sm_info);
351df4b2 4529}
7fd9e544 4530
4d57b86d 4531int __init f2fs_create_segment_manager_caches(void)
7fd9e544
JK		 4532{
4533 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4534 sizeof(struct discard_entry));
7fd9e544 4535 if (!discard_entry_slab)
184a5cd2
CY		 4536 goto fail;
4537
b01a9201
JK		 4538 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4539 sizeof(struct discard_cmd));
4540 if (!discard_cmd_slab)
6ab2a308 4541 goto destroy_discard_entry;
275b66b0 4542
184a5cd2 4543 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4544 sizeof(struct sit_entry_set));
184a5cd2 4545 if (!sit_entry_set_slab)
b01a9201 4546 goto destroy_discard_cmd;
88b88a66
JK		 4547
4548 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4549 sizeof(struct inmem_pages));
4550 if (!inmem_entry_slab)
4551 goto destroy_sit_entry_set;
7fd9e544 4552 return 0;
184a5cd2 4553
88b88a66
JK		 4554destroy_sit_entry_set:
4555 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4556destroy_discard_cmd:
4557 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4558destroy_discard_entry:
184a5cd2
CY		 4559 kmem_cache_destroy(discard_entry_slab);
4560fail:
4561 return -ENOMEM;
7fd9e544
JK		 4562}
4563
4d57b86d 4564void f2fs_destroy_segment_manager_caches(void)
7fd9e544 4565{
184a5cd2 4566 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4567 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4568 kmem_cache_destroy(discard_entry_slab);
88b88a66 4569 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4570}
Ubuntu Jammy Linux kernel mirror