]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: fix to do sanity check on valid block count of segment
[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
549 blk_start_plug(&plug);
4d57b86d 550 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 551 blk_finish_plug(&plug);
552 }
4660f9c0 553 f2fs_sync_fs(sbi->sb, true);
42190d2a 554 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 555 }
4660f9c0
JK		 556}
557
20fda56b
KM		 558static int __submit_flush_wait(struct f2fs_sb_info *sbi,
559 struct block_device *bdev)
3c62be17 560{
dc37910d 561 struct bio *bio;
3c62be17
JK		 562 int ret;
563
dc37910d
CY		 564 bio = f2fs_bio_alloc(sbi, 0, false);
565 if (!bio)
566 return -ENOMEM;
567
3adc5fcb 568 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 569 bio_set_dev(bio, bdev);
3c62be17
JK		 570 ret = submit_bio_wait(bio);
571 bio_put(bio);
20fda56b
KM		 572
573 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
574 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 575 return ret;
576}
577
39d787be 578static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 579{
39d787be 580 int ret = 0;
3c62be17
JK		 581 int i;
582
0916878d 583 if (!f2fs_is_multi_device(sbi))
39d787be 584 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 585
39d787be 586 for (i = 0; i < sbi->s_ndevs; i++) {
4d57b86d 587 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
39d787be 588 continue;
20fda56b
KM		 589 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
590 if (ret)
591 break;
3c62be17
JK		 592 }
593 return ret;
594}
595
2163d198 596static int issue_flush_thread(void *data)
6b4afdd7
JK		 597{
598 struct f2fs_sb_info *sbi = data;
b01a9201 599 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 600 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 601repeat:
602 if (kthread_should_stop())
603 return 0;
604
dc6febb6
CY		 605 sb_start_intwrite(sbi->sb);
606
721bd4d5 607 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 608 struct flush_cmd *cmd, *next;
609 int ret;
610
721bd4d5
GZ		 611 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
612 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
613
39d787be
CY		 614 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
615
616 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 617 atomic_inc(&fcc->issued_flush);
618
721bd4d5
GZ		 619 llist_for_each_entry_safe(cmd, next,
620 fcc->dispatch_list, llnode) {
6b4afdd7 621 cmd->ret = ret;
6b4afdd7
JK		 622 complete(&cmd->wait);
623 }
a688b9d9 624 fcc->dispatch_list = NULL;
6b4afdd7
JK		 625 }
626
dc6febb6
CY		 627 sb_end_intwrite(sbi->sb);
628
a688b9d9 629 wait_event_interruptible(*q,
721bd4d5 630 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 631 goto repeat;
632}
633
39d787be 634int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 635{
b01a9201 636 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 637 struct flush_cmd cmd;
8b8dd65f 638 int ret;
6b4afdd7 639
0f7b2abd
JK		 640 if (test_opt(sbi, NOBARRIER))
641 return 0;
642
8b8dd65f 643 if (!test_opt(sbi, FLUSH_MERGE)) {
72691af6 644 atomic_inc(&fcc->queued_flush);
39d787be 645 ret = submit_flush_wait(sbi, ino);
72691af6 646 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 647 atomic_inc(&fcc->issued_flush);
648 return ret;
649 }
740432f8 650
0916878d
DLM		 651 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
652 f2fs_is_multi_device(sbi)) {
39d787be 653 ret = submit_flush_wait(sbi, ino);
72691af6 654 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 655
656 atomic_inc(&fcc->issued_flush);
740432f8
JK		 657 return ret;
658 }
6b4afdd7 659
39d787be 660 cmd.ino = ino;
adf8d90b 661 init_completion(&cmd.wait);
6b4afdd7 662
721bd4d5 663 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 664
6f890df0
CY		 665 /* update issue_list before we wake up issue_flush thread */
666 smp_mb();
667
668 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 669 wake_up(&fcc->flush_wait_queue);
6b4afdd7 670
5eba8c5d
JK		 671 if (fcc->f2fs_issue_flush) {
672 wait_for_completion(&cmd.wait);
72691af6 673 atomic_dec(&fcc->queued_flush);
5eba8c5d 674 } else {
d3238691
CY		 675 struct llist_node *list;
676
677 list = llist_del_all(&fcc->issue_list);
678 if (!list) {
679 wait_for_completion(&cmd.wait);
72691af6 680 atomic_dec(&fcc->queued_flush);
d3238691
CY		 681 } else {
682 struct flush_cmd *tmp, *next;
683
39d787be 684 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 685
686 llist_for_each_entry_safe(tmp, next, list, llnode) {
687 if (tmp == &cmd) {
688 cmd.ret = ret;
72691af6 689 atomic_dec(&fcc->queued_flush);
d3238691
CY		 690 continue;
691 }
692 tmp->ret = ret;
693 complete(&tmp->wait);
694 }
695 }
5eba8c5d 696 }
adf8d90b
CY		 697
698 return cmd.ret;
6b4afdd7
JK		 699}
700
4d57b86d 701int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
2163d198
GZ		 702{
703 dev_t dev = sbi->sb->s_bdev->bd_dev;
704 struct flush_cmd_control *fcc;
705 int err = 0;
706
b01a9201
JK		 707 if (SM_I(sbi)->fcc_info) {
708 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 709 if (fcc->f2fs_issue_flush)
710 return err;
5eba8c5d
JK		 711 goto init_thread;
712 }
713
acbf054d 714 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 715 if (!fcc)
716 return -ENOMEM;
8b8dd65f 717 atomic_set(&fcc->issued_flush, 0);
72691af6 718 atomic_set(&fcc->queued_flush, 0);
2163d198 719 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 720 init_llist_head(&fcc->issue_list);
b01a9201 721 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 722 if (!test_opt(sbi, FLUSH_MERGE))
723 return err;
724
5eba8c5d 725init_thread:
2163d198
GZ		 726 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
727 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
728 if (IS_ERR(fcc->f2fs_issue_flush)) {
729 err = PTR_ERR(fcc->f2fs_issue_flush);
5222595d 730 kvfree(fcc);
b01a9201 731 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 732 return err;
733 }
2163d198
GZ		 734
735 return err;
736}
737
4d57b86d 738void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 739{
b01a9201 740 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 741
5eba8c5d
JK		 742 if (fcc && fcc->f2fs_issue_flush) {
743 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
744
745 fcc->f2fs_issue_flush = NULL;
746 kthread_stop(flush_thread);
747 }
748 if (free) {
5222595d 749 kvfree(fcc);
b01a9201 750 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 751 }
2163d198
GZ		 752}
753
1228b482
CY		 754int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
755{
756 int ret = 0, i;
757
0916878d 758 if (!f2fs_is_multi_device(sbi))
1228b482
CY		 759 return 0;
760
761 for (i = 1; i < sbi->s_ndevs; i++) {
762 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
763 continue;
764 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
765 if (ret)
766 break;
767
768 spin_lock(&sbi->dev_lock);
769 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
770 spin_unlock(&sbi->dev_lock);
771 }
772
773 return ret;
774}
775
351df4b2
JK		 776static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
777 enum dirty_type dirty_type)
778{
779 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
780
781 /* need not be added */
782 if (IS_CURSEG(sbi, segno))
783 return;
784
785 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
786 dirty_i->nr_dirty[dirty_type]++;
787
788 if (dirty_type == DIRTY) {
789 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 790 enum dirty_type t = sentry->type;
b2f2c390 791
ec325b52
JK		 792 if (unlikely(t >= DIRTY)) {
793 f2fs_bug_on(sbi, 1);
794 return;
795 }
4625d6aa
CL		 796 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
797 dirty_i->nr_dirty[t]++;
351df4b2
JK		 798 }
799}
800
801static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
802 enum dirty_type dirty_type)
803{
804 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
805
806 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
807 dirty_i->nr_dirty[dirty_type]--;
808
809 if (dirty_type == DIRTY) {
4625d6aa
CL		 810 struct seg_entry *sentry = get_seg_entry(sbi, segno);
811 enum dirty_type t = sentry->type;
812
813 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
814 dirty_i->nr_dirty[t]--;
b2f2c390 815
302bd348 816 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 817 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 818 dirty_i->victim_secmap);
351df4b2
JK		 819 }
820}
821
0a8165d7 822/*
351df4b2
JK		 823 * Should not occur error such as -ENOMEM.
824 * Adding dirty entry into seglist is not critical operation.
825 * If a given segment is one of current working segments, it won't be added.
826 */
8d8451af 827static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 828{
829 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4354994f 830 unsigned short valid_blocks, ckpt_valid_blocks;
351df4b2
JK		 831
832 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
833 return;
834
835 mutex_lock(&dirty_i->seglist_lock);
836
302bd348 837 valid_blocks = get_valid_blocks(sbi, segno, false);
4354994f 838 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
351df4b2 839
4354994f
DR		 840 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
841 ckpt_valid_blocks == sbi->blocks_per_seg)) {
351df4b2
JK		 842 __locate_dirty_segment(sbi, segno, PRE);
843 __remove_dirty_segment(sbi, segno, DIRTY);
844 } else if (valid_blocks < sbi->blocks_per_seg) {
845 __locate_dirty_segment(sbi, segno, DIRTY);
846 } else {
847 /* Recovery routine with SSR needs this */
848 __remove_dirty_segment(sbi, segno, DIRTY);
849 }
850
851 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 852}
853
4354994f
DR		 854/* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
855void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
856{
857 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
858 unsigned int segno;
859
860 mutex_lock(&dirty_i->seglist_lock);
861 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
862 if (get_valid_blocks(sbi, segno, false))
863 continue;
864 if (IS_CURSEG(sbi, segno))
865 continue;
866 __locate_dirty_segment(sbi, segno, PRE);
867 __remove_dirty_segment(sbi, segno, DIRTY);
868 }
869 mutex_unlock(&dirty_i->seglist_lock);
870}
871
872int f2fs_disable_cp_again(struct f2fs_sb_info *sbi)
873{
874 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
875 block_t ovp = overprovision_segments(sbi) << sbi->log_blocks_per_seg;
876 block_t holes[2] = {0, 0}; /* DATA and NODE */
877 struct seg_entry *se;
878 unsigned int segno;
879
880 mutex_lock(&dirty_i->seglist_lock);
881 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
882 se = get_seg_entry(sbi, segno);
883 if (IS_NODESEG(se->type))
884 holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
885 else
886 holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
887 }
888 mutex_unlock(&dirty_i->seglist_lock);
889
890 if (holes[DATA] > ovp || holes[NODE] > ovp)
891 return -EAGAIN;
db610a64
JK		 892 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
893 dirty_segments(sbi) > overprovision_segments(sbi))
894 return -EAGAIN;
4354994f
DR		 895 return 0;
896}
897
898/* This is only used by SBI_CP_DISABLED */
899static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
900{
901 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
902 unsigned int segno = 0;
903
904 mutex_lock(&dirty_i->seglist_lock);
905 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
906 if (get_valid_blocks(sbi, segno, false))
907 continue;
908 if (get_ckpt_valid_blocks(sbi, segno))
909 continue;
910 mutex_unlock(&dirty_i->seglist_lock);
911 return segno;
912 }
913 mutex_unlock(&dirty_i->seglist_lock);
914 return NULL_SEGNO;
915}
916
004b6862 917static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 918 struct block_device *bdev, block_t lstart,
919 block_t start, block_t len)
275b66b0 920{
0b54fb84 921 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 922 struct list_head *pend_list;
b01a9201 923 struct discard_cmd *dc;
275b66b0 924
ba48a33e
CY		 925 f2fs_bug_on(sbi, !len);
926
927 pend_list = &dcc->pend_list[plist_idx(len)];
928
b01a9201
JK		 929 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
930 INIT_LIST_HEAD(&dc->list);
c81abe34 931 dc->bdev = bdev;
b01a9201 932 dc->lstart = lstart;
c81abe34 933 dc->start = start;
b01a9201 934 dc->len = len;
ec9895ad 935 dc->ref = 0;
15469963 936 dc->state = D_PREP;
72691af6 937 dc->queued = 0;
c81abe34 938 dc->error = 0;
b01a9201 939 init_completion(&dc->wait);
22d375dd 940 list_add_tail(&dc->list, pend_list);
35ec7d57
CY		 941 spin_lock_init(&dc->lock);
942 dc->bio_ref = 0;
5f32366a 943 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 944 dcc->undiscard_blks += len;
004b6862
CY		 945
946 return dc;
947}
948
949static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
950 struct block_device *bdev, block_t lstart,
951 block_t start, block_t len,
4dada3fd
CY		 952 struct rb_node *parent, struct rb_node **p,
953 bool leftmost)
004b6862
CY		 954{
955 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
956 struct discard_cmd *dc;
957
958 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
959
960 rb_link_node(&dc->rb_node, parent, p);
4dada3fd 961 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
004b6862
CY		 962
963 return dc;
15469963
JK		 964}
965
004b6862
CY		 966static void __detach_discard_cmd(struct discard_cmd_control *dcc,
967 struct discard_cmd *dc)
15469963 968{
dcc9165d 969 if (dc->state == D_DONE)
72691af6 970 atomic_sub(dc->queued, &dcc->queued_discard);
004b6862
CY		 971
972 list_del(&dc->list);
4dada3fd 973 rb_erase_cached(&dc->rb_node, &dcc->root);
d84d1cbd 974 dcc->undiscard_blks -= dc->len;
004b6862
CY		 975
976 kmem_cache_free(discard_cmd_slab, dc);
977
978 atomic_dec(&dcc->discard_cmd_cnt);
979}
980
981static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
982 struct discard_cmd *dc)
983{
984 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
35ec7d57 985 unsigned long flags;
dcc9165d 986
2ec6f2ef
CY		 987 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
988
35ec7d57
CY		 989 spin_lock_irqsave(&dc->lock, flags);
990 if (dc->bio_ref) {
991 spin_unlock_irqrestore(&dc->lock, flags);
992 return;
993 }
994 spin_unlock_irqrestore(&dc->lock, flags);
995
d9703d90
CY		 996 f2fs_bug_on(sbi, dc->ref);
997
c81abe34
JK		 998 if (dc->error == -EOPNOTSUPP)
999 dc->error = 0;
15469963 1000
c81abe34 1001 if (dc->error)
22d7ea13
CY		 1002 printk_ratelimited(
1003 "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1004 KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
004b6862 1005 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 1006}
1007
c81abe34
JK		 1008static void f2fs_submit_discard_endio(struct bio *bio)
1009{
1010 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
35ec7d57 1011 unsigned long flags;
c81abe34 1012
4e4cbee9 1013 dc->error = blk_status_to_errno(bio->bi_status);
35ec7d57
CY		 1014
1015 spin_lock_irqsave(&dc->lock, flags);
1016 dc->bio_ref--;
1017 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1018 dc->state = D_DONE;
1019 complete_all(&dc->wait);
1020 }
1021 spin_unlock_irqrestore(&dc->lock, flags);
c81abe34
JK		 1022 bio_put(bio);
1023}
1024
94b1e10e 1025static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 1026 block_t start, block_t end)
1027{
1028#ifdef CONFIG_F2FS_CHECK_FS
1029 struct seg_entry *sentry;
1030 unsigned int segno;
1031 block_t blk = start;
1032 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1033 unsigned long *map;
1034
1035 while (blk < end) {
1036 segno = GET_SEGNO(sbi, blk);
1037 sentry = get_seg_entry(sbi, segno);
1038 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1039
008396e1
YS		 1040 if (end < START_BLOCK(sbi, segno + 1))
1041 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1042 else
1043 size = max_blocks;
6915ea9d
CY		 1044 map = (unsigned long *)(sentry->cur_valid_map);
1045 offset = __find_rev_next_bit(map, size, offset);
1046 f2fs_bug_on(sbi, offset != size);
008396e1 1047 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 1048 }
1049#endif
1050}
1051
8bb4f253
JK		 1052static void __init_discard_policy(struct f2fs_sb_info *sbi,
1053 struct discard_policy *dpolicy,
1054 int discard_type, unsigned int granularity)
1055{
1056 /* common policy */
1057 dpolicy->type = discard_type;
1058 dpolicy->sync = true;
20ee4382 1059 dpolicy->ordered = false;
8bb4f253
JK		 1060 dpolicy->granularity = granularity;
1061
1062 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1063 dpolicy->io_aware_gran = MAX_PLIST_NUM;
03f2c02d 1064 dpolicy->timeout = 0;
8bb4f253
JK		 1065
1066 if (discard_type == DPOLICY_BG) {
1067 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1068 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1069 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1070 dpolicy->io_aware = true;
cba60849 1071 dpolicy->sync = false;
20ee4382 1072 dpolicy->ordered = true;
8bb4f253
JK		 1073 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1074 dpolicy->granularity = 1;
1075 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1076 }
1077 } else if (discard_type == DPOLICY_FORCE) {
1078 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1079 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1080 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1081 dpolicy->io_aware = false;
1082 } else if (discard_type == DPOLICY_FSTRIM) {
1083 dpolicy->io_aware = false;
1084 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 1085 dpolicy->max_requests = UINT_MAX;
8bb4f253 1086 dpolicy->io_aware = false;
b8623253
JK		 1087 /* we need to issue all to keep CP_TRIMMED_FLAG */
1088 dpolicy->granularity = 1;
8bb4f253
JK		 1089 }
1090}
1091
35ec7d57
CY		 1092static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1093 struct block_device *bdev, block_t lstart,
1094 block_t start, block_t len);
c81abe34 1095/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
6b9cb124 1096static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1097 struct discard_policy *dpolicy,
35ec7d57
CY		 1098 struct discard_cmd *dc,
1099 unsigned int *issued)
c81abe34 1100{
35ec7d57
CY		 1101 struct block_device *bdev = dc->bdev;
1102 struct request_queue *q = bdev_get_queue(bdev);
1103 unsigned int max_discard_blocks =
1104 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
c81abe34 1105 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1106 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1107 &(dcc->fstrim_list) : &(dcc->wait_list);
78997b56 1108 int flag = dpolicy->sync ? REQ_SYNC : 0;
35ec7d57
CY		 1109 block_t lstart, start, len, total_len;
1110 int err = 0;
c81abe34
JK		 1111
1112 if (dc->state != D_PREP)
6b9cb124 1113 return 0;
c81abe34 1114
d6184774 1115 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
6b9cb124 1116 return 0;
d6184774 1117
35ec7d57
CY		 1118 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1119
1120 lstart = dc->lstart;
1121 start = dc->start;
1122 len = dc->len;
1123 total_len = len;
1124
1125 dc->len = 0;
1126
1127 while (total_len && *issued < dpolicy->max_requests && !err) {
1128 struct bio *bio = NULL;
1129 unsigned long flags;
1130 bool last = true;
1131
1132 if (len > max_discard_blocks) {
1133 len = max_discard_blocks;
1134 last = false;
1135 }
1136
1137 (*issued)++;
1138 if (*issued == dpolicy->max_requests)
1139 last = true;
1140
1141 dc->len += len;
1142
b83dcfe6
CY		 1143 if (time_to_inject(sbi, FAULT_DISCARD)) {
1144 f2fs_show_injection_info(FAULT_DISCARD);
1145 err = -EIO;
1146 goto submit;
1147 }
35ec7d57
CY		 1148 err = __blkdev_issue_discard(bdev,
1149 SECTOR_FROM_BLOCK(start),
1150 SECTOR_FROM_BLOCK(len),
1151 GFP_NOFS, 0, &bio);
b83dcfe6 1152submit:
6b9cb124 1153 if (err) {
35ec7d57 1154 spin_lock_irqsave(&dc->lock, flags);
6b9cb124 1155 if (dc->state == D_PARTIAL)
35ec7d57 1156 dc->state = D_SUBMIT;
35ec7d57
CY		 1157 spin_unlock_irqrestore(&dc->lock, flags);
1158
6b9cb124
CY		 1159 break;
1160 }
35ec7d57 1161
6b9cb124 1162 f2fs_bug_on(sbi, !bio);
35ec7d57 1163
6b9cb124
CY		 1164 /*
1165 * should keep before submission to avoid D_DONE
1166 * right away
1167 */
1168 spin_lock_irqsave(&dc->lock, flags);
1169 if (last)
1170 dc->state = D_SUBMIT;
1171 else
1172 dc->state = D_PARTIAL;
1173 dc->bio_ref++;
1174 spin_unlock_irqrestore(&dc->lock, flags);
35ec7d57 1175
72691af6
JK		 1176 atomic_inc(&dcc->queued_discard);
1177 dc->queued++;
6b9cb124 1178 list_move_tail(&dc->list, wait_list);
b0af6d49 1179
6b9cb124 1180 /* sanity check on discard range */
9249dded 1181 __check_sit_bitmap(sbi, lstart, lstart + len);
35ec7d57 1182
6b9cb124
CY		 1183 bio->bi_private = dc;
1184 bio->bi_end_io = f2fs_submit_discard_endio;
1185 bio->bi_opf |= flag;
1186 submit_bio(bio);
1187
1188 atomic_inc(&dcc->issued_discard);
1189
1190 f2fs_update_iostat(sbi, FS_DISCARD, 1);
35ec7d57
CY		 1191
1192 lstart += len;
1193 start += len;
1194 total_len -= len;
1195 len = total_len;
c81abe34 1196 }
35ec7d57 1197
6b9cb124 1198 if (!err && len)
35ec7d57 1199 __update_discard_tree_range(sbi, bdev, lstart, start, len);
6b9cb124 1200 return err;
c81abe34
JK		 1201}
1202
004b6862
CY		 1203static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1204 struct block_device *bdev, block_t lstart,
1205 block_t start, block_t len,
1206 struct rb_node **insert_p,
1207 struct rb_node *insert_parent)
c81abe34 1208{
004b6862 1209 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1210 struct rb_node **p;
004b6862
CY		 1211 struct rb_node *parent = NULL;
1212 struct discard_cmd *dc = NULL;
4dada3fd 1213 bool leftmost = true;
004b6862
CY		 1214
1215 if (insert_p && insert_parent) {
1216 parent = insert_parent;
1217 p = insert_p;
1218 goto do_insert;
1219 }
c81abe34 1220
4dada3fd
CY		 1221 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1222 lstart, &leftmost);
004b6862 1223do_insert:
4dada3fd
CY		 1224 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1225 p, leftmost);
004b6862
CY		 1226 if (!dc)
1227 return NULL;
c81abe34 1228
004b6862 1229 return dc;
c81abe34
JK		 1230}
1231
ba48a33e
CY		 1232static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1233 struct discard_cmd *dc)
1234{
1235 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1236}
1237
3d6a650f
YH		 1238static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1239 struct discard_cmd *dc, block_t blkaddr)
1240{
ba48a33e 1241 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1242 struct discard_info di = dc->di;
1243 bool modified = false;
3d6a650f 1244
004b6862 1245 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1246 __remove_discard_cmd(sbi, dc);
1247 return;
1248 }
1249
d84d1cbd
CY		 1250 dcc->undiscard_blks -= di.len;
1251
004b6862 1252 if (blkaddr > di.lstart) {
3d6a650f 1253 dc->len = blkaddr - dc->lstart;
d84d1cbd 1254 dcc->undiscard_blks += dc->len;
ba48a33e 1255 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1256 modified = true;
1257 }
1258
1259 if (blkaddr < di.lstart + di.len - 1) {
1260 if (modified) {
1261 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1262 di.start + blkaddr + 1 - di.lstart,
1263 di.lstart + di.len - 1 - blkaddr,
1264 NULL, NULL);
1265 } else {
1266 dc->lstart++;
1267 dc->len--;
1268 dc->start++;
d84d1cbd 1269 dcc->undiscard_blks += dc->len;
ba48a33e 1270 __relocate_discard_cmd(dcc, dc);
004b6862 1271 }
3d6a650f
YH		 1272 }
1273}
1274
004b6862
CY		 1275static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1276 struct block_device *bdev, block_t lstart,
1277 block_t start, block_t len)
275b66b0 1278{
0b54fb84 1279 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1280 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1281 struct discard_cmd *dc;
1282 struct discard_info di = {0};
1283 struct rb_node **insert_p = NULL, *insert_parent = NULL;
35ec7d57
CY		 1284 struct request_queue *q = bdev_get_queue(bdev);
1285 unsigned int max_discard_blocks =
1286 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
004b6862 1287 block_t end = lstart + len;
275b66b0 1288
4d57b86d 1289 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
004b6862
CY		 1290 NULL, lstart,
1291 (struct rb_entry **)&prev_dc,
1292 (struct rb_entry **)&next_dc,
4dada3fd 1293 &insert_p, &insert_parent, true, NULL);
004b6862
CY		 1294 if (dc)
1295 prev_dc = dc;
1296
1297 if (!prev_dc) {
1298 di.lstart = lstart;
1299 di.len = next_dc ? next_dc->lstart - lstart : len;
1300 di.len = min(di.len, len);
1301 di.start = start;
22d375dd 1302 }
15469963 1303
004b6862
CY		 1304 while (1) {
1305 struct rb_node *node;
1306 bool merged = false;
1307 struct discard_cmd *tdc = NULL;
1308
1309 if (prev_dc) {
1310 di.lstart = prev_dc->lstart + prev_dc->len;
1311 if (di.lstart < lstart)
1312 di.lstart = lstart;
1313 if (di.lstart >= end)
1314 break;
1315
1316 if (!next_dc || next_dc->lstart > end)
1317 di.len = end - di.lstart;
1318 else
1319 di.len = next_dc->lstart - di.lstart;
1320 di.start = start + di.lstart - lstart;
1321 }
1322
1323 if (!di.len)
1324 goto next;
1325
1326 if (prev_dc && prev_dc->state == D_PREP &&
1327 prev_dc->bdev == bdev &&
35ec7d57
CY		 1328 __is_discard_back_mergeable(&di, &prev_dc->di,
1329 max_discard_blocks)) {
004b6862 1330 prev_dc->di.len += di.len;
d84d1cbd 1331 dcc->undiscard_blks += di.len;
ba48a33e 1332 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1333 di = prev_dc->di;
1334 tdc = prev_dc;
1335 merged = true;
1336 }
1337
1338 if (next_dc && next_dc->state == D_PREP &&
1339 next_dc->bdev == bdev &&
35ec7d57
CY		 1340 __is_discard_front_mergeable(&di, &next_dc->di,
1341 max_discard_blocks)) {
004b6862
CY		 1342 next_dc->di.lstart = di.lstart;
1343 next_dc->di.len += di.len;
1344 next_dc->di.start = di.start;
d84d1cbd 1345 dcc->undiscard_blks += di.len;
ba48a33e 1346 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1347 if (tdc)
1348 __remove_discard_cmd(sbi, tdc);
004b6862 1349 merged = true;
4e6a8d9b 1350 }
004b6862 1351
df0f6b44 1352 if (!merged) {
004b6862
CY		 1353 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1354 di.len, NULL, NULL);
df0f6b44 1355 }
004b6862
CY		 1356 next:
1357 prev_dc = next_dc;
1358 if (!prev_dc)
1359 break;
1360
1361 node = rb_next(&prev_dc->rb_node);
1362 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1363 }
004b6862
CY		 1364}
1365
1366static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1367 struct block_device *bdev, block_t blkstart, block_t blklen)
1368{
1369 block_t lblkstart = blkstart;
1370
7f3d7719
DLM		 1371 if (!f2fs_bdev_support_discard(bdev))
1372 return 0;
1373
0243a5f9 1374 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862 1375
0916878d 1376 if (f2fs_is_multi_device(sbi)) {
004b6862
CY		 1377 int devi = f2fs_target_device_index(sbi, blkstart);
1378
1379 blkstart -= FDEV(devi).start_blk;
1380 }
35ec7d57 1381 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862 1382 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
35ec7d57 1383 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862
CY		 1384 return 0;
1385}
1386
20ee4382
CY		 1387static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1388 struct discard_policy *dpolicy)
1389{
1390 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1391 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1392 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1393 struct discard_cmd *dc;
1394 struct blk_plug plug;
1395 unsigned int pos = dcc->next_pos;
1396 unsigned int issued = 0;
1397 bool io_interrupted = false;
1398
1399 mutex_lock(&dcc->cmd_lock);
1400 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1401 NULL, pos,
1402 (struct rb_entry **)&prev_dc,
1403 (struct rb_entry **)&next_dc,
4dada3fd 1404 &insert_p, &insert_parent, true, NULL);
20ee4382
CY		 1405 if (!dc)
1406 dc = next_dc;
1407
1408 blk_start_plug(&plug);
1409
1410 while (dc) {
1411 struct rb_node *node;
6b9cb124 1412 int err = 0;
20ee4382
CY		 1413
1414 if (dc->state != D_PREP)
1415 goto next;
1416
a7d10cf3 1417 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
20ee4382
CY		 1418 io_interrupted = true;
1419 break;
1420 }
1421
1422 dcc->next_pos = dc->lstart + dc->len;
6b9cb124 1423 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
20ee4382 1424
35ec7d57 1425 if (issued >= dpolicy->max_requests)
20ee4382
CY		 1426 break;
1427next:
1428 node = rb_next(&dc->rb_node);
6b9cb124
CY		 1429 if (err)
1430 __remove_discard_cmd(sbi, dc);
20ee4382
CY		 1431 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1432 }
1433
1434 blk_finish_plug(&plug);
1435
1436 if (!dc)
1437 dcc->next_pos = 0;
1438
1439 mutex_unlock(&dcc->cmd_lock);
1440
1441 if (!issued && io_interrupted)
1442 issued = -1;
1443
1444 return issued;
1445}
1446
78997b56
CY		 1447static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1448 struct discard_policy *dpolicy)
bd5b0738
CY		 1449{
1450 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1451 struct list_head *pend_list;
1452 struct discard_cmd *dc, *tmp;
1453 struct blk_plug plug;
522d1711 1454 int i, issued = 0;
e6c6de18 1455 bool io_interrupted = false;
bd5b0738 1456
03f2c02d
JK		 1457 if (dpolicy->timeout != 0)
1458 f2fs_update_time(sbi, dpolicy->timeout);
1459
78997b56 1460 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
03f2c02d
JK		 1461 if (dpolicy->timeout != 0 &&
1462 f2fs_time_over(sbi, dpolicy->timeout))
1463 break;
1464
78997b56
CY		 1465 if (i + 1 < dpolicy->granularity)
1466 break;
20ee4382
CY		 1467
1468 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1469 return __issue_discard_cmd_orderly(sbi, dpolicy);
1470
bd5b0738 1471 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1472
1473 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1474 if (list_empty(pend_list))
1475 goto next;
67fce70b
CY		 1476 if (unlikely(dcc->rbtree_check))
1477 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1478 &dcc->root));
33da62cf 1479 blk_start_plug(&plug);
bd5b0738
CY		 1480 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1481 f2fs_bug_on(sbi, dc->state != D_PREP);
1482
ecc9aa00 1483 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
a7d10cf3 1484 !is_idle(sbi, DISCARD_TIME)) {
e6c6de18 1485 io_interrupted = true;
522d1711 1486 break;
969d1b18 1487 }
e6c6de18 1488
35ec7d57 1489 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
522d1711 1490
35ec7d57 1491 if (issued >= dpolicy->max_requests)
33da62cf 1492 break;
bd5b0738 1493 }
33da62cf 1494 blk_finish_plug(&plug);
49c60c67 1495next:
33da62cf
CY		 1496 mutex_unlock(&dcc->cmd_lock);
1497
522d1711 1498 if (issued >= dpolicy->max_requests || io_interrupted)
33da62cf 1499 break;
bd5b0738 1500 }
969d1b18 1501
e6c6de18
CY		 1502 if (!issued && io_interrupted)
1503 issued = -1;
1504
969d1b18
CY		 1505 return issued;
1506}
1507
cf5c759f 1508static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1509{
1510 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1511 struct list_head *pend_list;
1512 struct discard_cmd *dc, *tmp;
1513 int i;
cf5c759f 1514 bool dropped = false;
969d1b18
CY		 1515
1516 mutex_lock(&dcc->cmd_lock);
1517 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1518 pend_list = &dcc->pend_list[i];
1519 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1520 f2fs_bug_on(sbi, dc->state != D_PREP);
1521 __remove_discard_cmd(sbi, dc);
cf5c759f 1522 dropped = true;
969d1b18
CY		 1523 }
1524 }
1525 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1526
1527 return dropped;
bd5b0738
CY		 1528}
1529
4d57b86d 1530void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
7950e9ac
CY		 1531{
1532 __drop_discard_cmd(sbi);
1533}
1534
0ea80512 1535static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1536 struct discard_cmd *dc)
1537{
1538 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1539 unsigned int len = 0;
2a510c00
CY		 1540
1541 wait_for_completion_io(&dc->wait);
1542 mutex_lock(&dcc->cmd_lock);
1543 f2fs_bug_on(sbi, dc->state != D_DONE);
1544 dc->ref--;
0ea80512
CY		 1545 if (!dc->ref) {
1546 if (!dc->error)
1547 len = dc->len;
2a510c00 1548 __remove_discard_cmd(sbi, dc);
0ea80512 1549 }
2a510c00 1550 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1551
1552 return len;
2a510c00
CY		 1553}
1554
0ea80512 1555static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1556 struct discard_policy *dpolicy,
1557 block_t start, block_t end)
63a94fa1
CY		 1558{
1559 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1560 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1561 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1562 struct discard_cmd *dc, *tmp;
6afae633 1563 bool need_wait;
0ea80512 1564 unsigned int trimmed = 0;
6afae633
CY		 1565
1566next:
1567 need_wait = false;
63a94fa1
CY		 1568
1569 mutex_lock(&dcc->cmd_lock);
1570 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1571 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1572 continue;
78997b56 1573 if (dc->len < dpolicy->granularity)
8412663d 1574 continue;
78997b56 1575 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1576 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1577 if (!dc->error)
1578 trimmed += dc->len;
63a94fa1 1579 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1580 } else {
1581 dc->ref++;
1582 need_wait = true;
1583 break;
63a94fa1
CY		 1584 }
1585 }
1586 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1587
1588 if (need_wait) {
0ea80512 1589 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1590 goto next;
1591 }
0ea80512
CY		 1592
1593 return trimmed;
63a94fa1
CY		 1594}
1595
01f9cf6d 1596static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1597 struct discard_policy *dpolicy)
8412663d 1598{
9a997188 1599 struct discard_policy dp;
01f9cf6d 1600 unsigned int discard_blks;
9a997188 1601
01f9cf6d
CY		 1602 if (dpolicy)
1603 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
9a997188
JK		 1604
1605 /* wait all */
8bb4f253 1606 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
01f9cf6d 1607 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1608 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
01f9cf6d
CY		 1609 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1610
1611 return discard_blks;
8412663d
CY		 1612}
1613
004b6862 1614/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1615static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1616{
1617 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1618 struct discard_cmd *dc;
ec9895ad 1619 bool need_wait = false;
004b6862
CY		 1620
1621 mutex_lock(&dcc->cmd_lock);
4d57b86d
CY		 1622 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1623 NULL, blkaddr);
004b6862 1624 if (dc) {
ec9895ad
CY		 1625 if (dc->state == D_PREP) {
1626 __punch_discard_cmd(sbi, dc, blkaddr);
1627 } else {
1628 dc->ref++;
1629 need_wait = true;
1630 }
275b66b0 1631 }
d431413f 1632 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1633
2a510c00
CY		 1634 if (need_wait)
1635 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1636}
1637
4d57b86d 1638void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
cce13252
CY		 1639{
1640 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1641
1642 if (dcc && dcc->f2fs_issue_discard) {
1643 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1644
1645 dcc->f2fs_issue_discard = NULL;
1646 kthread_stop(discard_thread);
ec9895ad 1647 }
d431413f
CY		 1648}
1649
8412663d 1650/* This comes from f2fs_put_super */
03f2c02d 1651bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
969d1b18
CY		 1652{
1653 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1654 struct discard_policy dpolicy;
cf5c759f 1655 bool dropped;
969d1b18 1656
8bb4f253
JK		 1657 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1658 dcc->discard_granularity);
03f2c02d 1659 dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
78997b56 1660 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1661 dropped = __drop_discard_cmd(sbi);
cf5c759f 1662
9a997188
JK		 1663 /* just to make sure there is no pending discard commands */
1664 __wait_all_discard_cmd(sbi, NULL);
2482c432
CY		 1665
1666 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
cf5c759f 1667 return dropped;
969d1b18
CY		 1668}
1669
15469963
JK		 1670static int issue_discard_thread(void *data)
1671{
1672 struct f2fs_sb_info *sbi = data;
1673 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1674 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1675 struct discard_policy dpolicy;
969d1b18
CY		 1676 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1677 int issued;
15469963 1678
1d7be270 1679 set_freezable();
15469963 1680
1d7be270 1681 do {
8bb4f253 1682 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1683 dcc->discard_granularity);
1684
969d1b18
CY		 1685 wait_event_interruptible_timeout(*q,
1686 kthread_should_stop() || freezing(current) ||
1687 dcc->discard_wake,
1688 msecs_to_jiffies(wait_ms));
35a9a766
SY		 1689
1690 if (dcc->discard_wake)
1691 dcc->discard_wake = 0;
1692
76c7bfb3
JK		 1693 /* clean up pending candidates before going to sleep */
1694 if (atomic_read(&dcc->queued_discard))
1695 __wait_all_discard_cmd(sbi, NULL);
1696
1d7be270
JK		 1697 if (try_to_freeze())
1698 continue;
3b60d802
CY		 1699 if (f2fs_readonly(sbi->sb))
1700 continue;
1d7be270
JK		 1701 if (kthread_should_stop())
1702 return 0;
d6184774
YH		 1703 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1704 wait_ms = dpolicy.max_interval;
1705 continue;
1706 }
15469963 1707
5b0e9539 1708 if (sbi->gc_mode == GC_URGENT)
8bb4f253 1709 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1710
dc6febb6
CY		 1711 sb_start_intwrite(sbi->sb);
1712
78997b56 1713 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1714 if (issued > 0) {
78997b56
CY		 1715 __wait_all_discard_cmd(sbi, &dpolicy);
1716 wait_ms = dpolicy.min_interval;
f9d1dced 1717 } else if (issued == -1){
a7d10cf3
ST		 1718 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1719 if (!wait_ms)
abde73c7 1720 wait_ms = dpolicy.mid_interval;
969d1b18 1721 } else {
78997b56 1722 wait_ms = dpolicy.max_interval;
969d1b18 1723 }
1d7be270 1724
dc6febb6 1725 sb_end_intwrite(sbi->sb);
1d7be270 1726
1d7be270
JK		 1727 } while (!kthread_should_stop());
1728 return 0;
15469963
JK		 1729}
1730
f46e8809 1731#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1732static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1733 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1734{
92592285 1735 sector_t sector, nr_sects;
10a875f8 1736 block_t lblkstart = blkstart;
3c62be17
JK		 1737 int devi = 0;
1738
0916878d 1739 if (f2fs_is_multi_device(sbi)) {
3c62be17 1740 devi = f2fs_target_device_index(sbi, blkstart);
95175daf
DLM		 1741 if (blkstart < FDEV(devi).start_blk ||
1742 blkstart > FDEV(devi).end_blk) {
1743 f2fs_msg(sbi->sb, KERN_ERR, "Invalid block %x",
1744 blkstart);
1745 return -EIO;
1746 }
3c62be17
JK		 1747 blkstart -= FDEV(devi).start_blk;
1748 }
f46e8809 1749
95175daf
DLM		 1750 /* For sequential zones, reset the zone write pointer */
1751 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
92592285
JK		 1752 sector = SECTOR_FROM_BLOCK(blkstart);
1753 nr_sects = SECTOR_FROM_BLOCK(blklen);
1754
1755 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1756 nr_sects != bdev_zone_sectors(bdev)) {
95175daf
DLM		 1757 f2fs_msg(sbi->sb, KERN_ERR,
1758 "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
92592285
JK		 1759 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1760 blkstart, blklen);
1761 return -EIO;
1762 }
d50aaeec 1763 trace_f2fs_issue_reset_zone(bdev, blkstart);
95175daf 1764 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
f46e8809 1765 }
95175daf
DLM		 1766
1767 /* For conventional zones, use regular discard if supported */
95175daf 1768 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1769}
1770#endif
1771
3c62be17
JK		 1772static int __issue_discard_async(struct f2fs_sb_info *sbi,
1773 struct block_device *bdev, block_t blkstart, block_t blklen)
1774{
1775#ifdef CONFIG_BLK_DEV_ZONED
7f3d7719 1776 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
3c62be17
JK		 1777 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1778#endif
c81abe34 1779 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1780}
1781
1e87a78d 1782static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1783 block_t blkstart, block_t blklen)
1784{
3c62be17
JK		 1785 sector_t start = blkstart, len = 0;
1786 struct block_device *bdev;
a66cdd98
JK		 1787 struct seg_entry *se;
1788 unsigned int offset;
1789 block_t i;
3c62be17
JK		 1790 int err = 0;
1791
1792 bdev = f2fs_target_device(sbi, blkstart, NULL);
1793
1794 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1795 if (i != start) {
1796 struct block_device *bdev2 =
1797 f2fs_target_device(sbi, i, NULL);
1798
1799 if (bdev2 != bdev) {
1800 err = __issue_discard_async(sbi, bdev,
1801 start, len);
1802 if (err)
1803 return err;
1804 bdev = bdev2;
1805 start = i;
1806 len = 0;
1807 }
1808 }
a66cdd98 1809
a66cdd98
JK		 1810 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1811 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1812
1813 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1814 sbi->discard_blks--;
1815 }
f46e8809 1816
3c62be17
JK		 1817 if (len)
1818 err = __issue_discard_async(sbi, bdev, start, len);
1819 return err;
1e87a78d
JK		 1820}
1821
25290fa5
JK		 1822static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1823 bool check_only)
adf4983b 1824{
b2955550
JK		 1825 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1826 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1827 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1828 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1829 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1830 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1831 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1832 unsigned int start = 0, end = -1;
c473f1a9 1833 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1834 struct discard_entry *de = NULL;
46f84c2c 1835 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1836 int i;
1837
7d20c8ab 1838 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
25290fa5 1839 return false;
b2955550 1840
a66cdd98 1841 if (!force) {
7d20c8ab 1842 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
0b54fb84
JK		 1843 SM_I(sbi)->dcc_info->nr_discards >=
1844 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1845 return false;
4b2fecc8
JK		 1846 }
1847
b2955550
JK		 1848 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1849 for (i = 0; i < entries; i++)
a66cdd98 1850 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1851 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1852
0b54fb84
JK		 1853 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1854 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1855 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1856 if (start >= max_blocks)
1857 break;
1858
1859 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1860 if (force && start && end != max_blocks
1861 && (end - start) < cpc->trim_minlen)
1862 continue;
1863
25290fa5
JK		 1864 if (check_only)
1865 return true;
1866
a7eeb823
CY		 1867 if (!de) {
1868 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1869 GFP_F2FS_ZERO);
1870 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1871 list_add_tail(&de->list, head);
1872 }
1873
1874 for (i = start; i < end; i++)
1875 __set_bit_le(i, (void *)de->discard_map);
1876
1877 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1878 }
25290fa5 1879 return false;
b2955550
JK		 1880}
1881
af8ff65b
CY		 1882static void release_discard_addr(struct discard_entry *entry)
1883{
1884 list_del(&entry->list);
1885 kmem_cache_free(discard_entry_slab, entry);
1886}
1887
4d57b86d 1888void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1889{
46f84c2c 1890 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1891 struct discard_entry *entry, *this;
1892
1893 /* drop caches */
af8ff65b
CY		 1894 list_for_each_entry_safe(entry, this, head, list)
1895 release_discard_addr(entry);
4b2fecc8
JK		 1896}
1897
0a8165d7 1898/*
4d57b86d 1899 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1900 */
1901static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1902{
1903 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1904 unsigned int segno;
351df4b2
JK		 1905
1906 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1907 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1908 __set_test_and_free(sbi, segno);
351df4b2
JK		 1909 mutex_unlock(&dirty_i->seglist_lock);
1910}
1911
4d57b86d
CY		 1912void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1913 struct cp_control *cpc)
351df4b2 1914{
969d1b18
CY		 1915 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1916 struct list_head *head = &dcc->entry_list;
2d7b822a 1917 struct discard_entry *entry, *this;
351df4b2 1918 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1919 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1920 unsigned int start = 0, end = -1;
36abef4e 1921 unsigned int secno, start_segno;
c473f1a9 1922 bool force = (cpc->reason & CP_DISCARD);
2c70c5e3 1923 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
351df4b2
JK		 1924
1925 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1926
351df4b2 1927 while (1) {
29e59c14 1928 int i;
ad6672bb
YS		 1929
1930 if (need_align && end != -1)
1931 end--;
7cd8558b
JK		 1932 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1933 if (start >= MAIN_SEGS(sbi))
351df4b2 1934 break;
7cd8558b
JK		 1935 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1936 start + 1);
29e59c14 1937
ad6672bb
YS		 1938 if (need_align) {
1939 start = rounddown(start, sbi->segs_per_sec);
1940 end = roundup(end, sbi->segs_per_sec);
1941 }
29e59c14 1942
ad6672bb
YS		 1943 for (i = start; i < end; i++) {
1944 if (test_and_clear_bit(i, prefree_map))
1945 dirty_i->nr_dirty[PRE]--;
1946 }
29e59c14 1947
7d20c8ab 1948 if (!f2fs_realtime_discard_enable(sbi))
29e59c14 1949 continue;
351df4b2 1950
650d3c4e
YH		 1951 if (force && start >= cpc->trim_start &&
1952 (end - 1) <= cpc->trim_end)
1953 continue;
1954
2c70c5e3 1955 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
36abef4e 1956 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1957 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1958 continue;
1959 }
1960next:
4ddb1a4d
JK		 1961 secno = GET_SEC_FROM_SEG(sbi, start);
1962 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1963 if (!IS_CURSEC(sbi, secno) &&
302bd348 1964 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1965 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1966 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1967
1968 start = start_segno + sbi->segs_per_sec;
1969 if (start < end)
1970 goto next;
8b107f5b
JK		 1971 else
1972 end = start - 1;
351df4b2
JK		 1973 }
1974 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1975
1976 /* send small discards */
2d7b822a 1977 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1978 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1979 bool is_valid = test_bit_le(0, entry->discard_map);
1980
1981find_next:
1982 if (is_valid) {
1983 next_pos = find_next_zero_bit_le(entry->discard_map,
1984 sbi->blocks_per_seg, cur_pos);
1985 len = next_pos - cur_pos;
1986
7beb01f7 1987 if (f2fs_sb_has_blkzoned(sbi) ||
acfd2810 1988 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1989 goto skip;
1990
1991 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1992 len);
a7eeb823
CY		 1993 total_len += len;
1994 } else {
1995 next_pos = find_next_bit_le(entry->discard_map,
1996 sbi->blocks_per_seg, cur_pos);
1997 }
836b5a63 1998skip:
a7eeb823
CY		 1999 cur_pos = next_pos;
2000 is_valid = !is_valid;
2001
2002 if (cur_pos < sbi->blocks_per_seg)
2003 goto find_next;
2004
af8ff65b 2005 release_discard_addr(entry);
969d1b18 2006 dcc->nr_discards -= total_len;
b2955550 2007 }
34e159da 2008
01983c71 2009 wake_up_discard_thread(sbi, false);
351df4b2
JK		 2010}
2011
8ed59745 2012static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 2013{
15469963 2014 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 2015 struct discard_cmd_control *dcc;
ba48a33e 2016 int err = 0, i;
0b54fb84
JK		 2017
2018 if (SM_I(sbi)->dcc_info) {
2019 dcc = SM_I(sbi)->dcc_info;
2020 goto init_thread;
2021 }
2022
acbf054d 2023 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 2024 if (!dcc)
2025 return -ENOMEM;
2026
969d1b18 2027 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 2028 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 2029 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 2030 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 2031 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 2032 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 2033 mutex_init(&dcc->cmd_lock);
8b8dd65f 2034 atomic_set(&dcc->issued_discard, 0);
72691af6 2035 atomic_set(&dcc->queued_discard, 0);
5f32366a 2036 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 2037 dcc->nr_discards = 0;
d618ebaf 2038 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 2039 dcc->undiscard_blks = 0;
20ee4382 2040 dcc->next_pos = 0;
4dada3fd 2041 dcc->root = RB_ROOT_CACHED;
67fce70b 2042 dcc->rbtree_check = false;
0b54fb84 2043
15469963 2044 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 2045 SM_I(sbi)->dcc_info = dcc;
2046init_thread:
15469963
JK		 2047 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2048 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2049 if (IS_ERR(dcc->f2fs_issue_discard)) {
2050 err = PTR_ERR(dcc->f2fs_issue_discard);
5222595d 2051 kvfree(dcc);
15469963
JK		 2052 SM_I(sbi)->dcc_info = NULL;
2053 return err;
2054 }
2055
0b54fb84
JK		 2056 return err;
2057}
2058
f099405f 2059static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 2060{
2061 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2062
f099405f
CY		 2063 if (!dcc)
2064 return;
2065
4d57b86d 2066 f2fs_stop_discard_thread(sbi);
f099405f 2067
5222595d 2068 kvfree(dcc);
f099405f 2069 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 2070}
2071
184a5cd2 2072static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 2073{
2074 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 2075
2076 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 2077 sit_i->dirty_sentries++;
184a5cd2
CY		 2078 return false;
2079 }
2080
2081 return true;
351df4b2
JK		 2082}
2083
2084static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2085 unsigned int segno, int modified)
2086{
2087 struct seg_entry *se = get_seg_entry(sbi, segno);
2088 se->type = type;
2089 if (modified)
2090 __mark_sit_entry_dirty(sbi, segno);
2091}
2092
2093static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2094{
2095 struct seg_entry *se;
2096 unsigned int segno, offset;
2097 long int new_vblocks;
6415fedc
YS		 2098 bool exist;
2099#ifdef CONFIG_F2FS_CHECK_FS
2100 bool mir_exist;
2101#endif
351df4b2
JK		 2102
2103 segno = GET_SEGNO(sbi, blkaddr);
2104
2105 se = get_seg_entry(sbi, segno);
2106 new_vblocks = se->valid_blocks + del;
491c0854 2107 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 2108
9850cf4a 2109 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 2110 (new_vblocks > sbi->blocks_per_seg)));
2111
2112 se->valid_blocks = new_vblocks;
a1f72ac2
CY		 2113 se->mtime = get_mtime(sbi, false);
2114 if (se->mtime > SIT_I(sbi)->max_mtime)
2115 SIT_I(sbi)->max_mtime = se->mtime;
351df4b2
JK		 2116
2117 /* Update valid block bitmap */
2118 if (del > 0) {
6415fedc 2119 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 2120#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2121 mir_exist = f2fs_test_and_set_bit(offset,
2122 se->cur_valid_map_mir);
2123 if (unlikely(exist != mir_exist)) {
2124 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2125 "when setting bitmap, blk:%u, old bit:%d",
2126 blkaddr, exist);
05796763 2127 f2fs_bug_on(sbi, 1);
6415fedc 2128 }
355e7891 2129#endif
6415fedc
YS		 2130 if (unlikely(exist)) {
2131 f2fs_msg(sbi->sb, KERN_ERR,
2132 "Bitmap was wrongly set, blk:%u", blkaddr);
2133 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2134 se->valid_blocks--;
2135 del = 0;
355e7891 2136 }
6415fedc 2137
7d20c8ab 2138 if (!f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 2139 sbi->discard_blks--;
720037f9
JK		 2140
2141 /* don't overwrite by SSR to keep node chain */
4354994f
DR		 2142 if (IS_NODESEG(se->type) &&
2143 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
720037f9
JK		 2144 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2145 se->ckpt_valid_blocks++;
2146 }
351df4b2 2147 } else {
6415fedc 2148 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 2149#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2150 mir_exist = f2fs_test_and_clear_bit(offset,
2151 se->cur_valid_map_mir);
2152 if (unlikely(exist != mir_exist)) {
2153 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2154 "when clearing bitmap, blk:%u, old bit:%d",
2155 blkaddr, exist);
05796763 2156 f2fs_bug_on(sbi, 1);
6415fedc 2157 }
355e7891 2158#endif
6415fedc
YS		 2159 if (unlikely(!exist)) {
2160 f2fs_msg(sbi->sb, KERN_ERR,
2161 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2162 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2163 se->valid_blocks++;
2164 del = 0;
4354994f
DR		 2165 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2166 /*
2167 * If checkpoints are off, we must not reuse data that
2168 * was used in the previous checkpoint. If it was used
2169 * before, we must track that to know how much space we
2170 * really have.
2171 */
2172 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2173 sbi->unusable_block_count++;
355e7891 2174 }
6415fedc 2175
7d20c8ab 2176 if (f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 2177 sbi->discard_blks++;
351df4b2
JK		 2178 }
2179 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2180 se->ckpt_valid_blocks += del;
2181
2182 __mark_sit_entry_dirty(sbi, segno);
2183
2184 /* update total number of valid blocks to be written in ckpt area */
2185 SIT_I(sbi)->written_valid_blocks += del;
2186
2c70c5e3 2187 if (__is_large_section(sbi))
351df4b2
JK		 2188 get_sec_entry(sbi, segno)->valid_blocks += del;
2189}
2190
4d57b86d 2191void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 2192{
2193 unsigned int segno = GET_SEGNO(sbi, addr);
2194 struct sit_info *sit_i = SIT_I(sbi);
2195
9850cf4a 2196 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 2197 if (addr == NEW_ADDR)
2198 return;
2199
6aa58d8a
CY		 2200 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2201
351df4b2 2202 /* add it into sit main buffer */
3d26fa6b 2203 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2204
2205 update_sit_entry(sbi, addr, -1);
2206
2207 /* add it into dirty seglist */
2208 locate_dirty_segment(sbi, segno);
2209
3d26fa6b 2210 up_write(&sit_i->sentry_lock);
351df4b2
JK		 2211}
2212
4d57b86d 2213bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 2214{
2215 struct sit_info *sit_i = SIT_I(sbi);
2216 unsigned int segno, offset;
2217 struct seg_entry *se;
2218 bool is_cp = false;
2219
e1da7872 2220 if (!is_valid_data_blkaddr(sbi, blkaddr))
6e2c64ad
JK		 2221 return true;
2222
3d26fa6b 2223 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2224
2225 segno = GET_SEGNO(sbi, blkaddr);
2226 se = get_seg_entry(sbi, segno);
2227 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2228
2229 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2230 is_cp = true;
2231
3d26fa6b 2232 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2233
2234 return is_cp;
2235}
2236
0a8165d7 2237/*
351df4b2
JK		 2238 * This function should be resided under the curseg_mutex lock
2239 */
2240static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 2241 struct f2fs_summary *sum)
351df4b2
JK		 2242{
2243 struct curseg_info *curseg = CURSEG_I(sbi, type);
2244 void *addr = curseg->sum_blk;
e79efe3b 2245 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 2246 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 2247}
2248
0a8165d7 2249/*
351df4b2
JK		 2250 * Calculate the number of current summary pages for writing
2251 */
4d57b86d 2252int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 2253{
351df4b2 2254 int valid_sum_count = 0;
9a47938b 2255 int i, sum_in_page;
351df4b2
JK		 2256
2257 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2258 if (sbi->ckpt->alloc_type[i] == SSR)
2259 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 2260 else {
2261 if (for_ra)
2262 valid_sum_count += le16_to_cpu(
2263 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2264 else
2265 valid_sum_count += curseg_blkoff(sbi, i);
2266 }
351df4b2
JK		 2267 }
2268
09cbfeaf 2269 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 2270 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2271 if (valid_sum_count <= sum_in_page)
351df4b2 2272 return 1;
9a47938b 2273 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 2274 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 2275 return 2;
2276 return 3;
2277}
2278
0a8165d7 2279/*
351df4b2
JK		 2280 * Caller should put this summary page
2281 */
4d57b86d 2282struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 2283{
7735730d 2284 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2285}
2286
4d57b86d
CY		 2287void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2288 void *src, block_t blk_addr)
351df4b2 2289{
4d57b86d 2290 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 2291
0537b811 2292 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2293 set_page_dirty(page);
2294 f2fs_put_page(page, 1);
2295}
2296
381722d2
CY		 2297static void write_sum_page(struct f2fs_sb_info *sbi,
2298 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2299{
4d57b86d 2300 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2301}
2302
b7ad7512
CY		 2303static void write_current_sum_page(struct f2fs_sb_info *sbi,
2304 int type, block_t blk_addr)
2305{
2306 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2307 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2308 struct f2fs_summary_block *src = curseg->sum_blk;
2309 struct f2fs_summary_block *dst;
2310
2311 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2312 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2313
2314 mutex_lock(&curseg->curseg_mutex);
2315
2316 down_read(&curseg->journal_rwsem);
2317 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2318 up_read(&curseg->journal_rwsem);
2319
2320 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2321 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2322
2323 mutex_unlock(&curseg->curseg_mutex);
2324
2325 set_page_dirty(page);
2326 f2fs_put_page(page, 1);
2327}
2328
a7881893
JK		 2329static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2330{
2331 struct curseg_info *curseg = CURSEG_I(sbi, type);
2332 unsigned int segno = curseg->segno + 1;
2333 struct free_segmap_info *free_i = FREE_I(sbi);
2334
2335 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2336 return !test_bit(segno, free_i->free_segmap);
2337 return 0;
2338}
2339
0a8165d7 2340/*
351df4b2
JK		 2341 * Find a new segment from the free segments bitmap to right order
2342 * This function should be returned with success, otherwise BUG
2343 */
2344static void get_new_segment(struct f2fs_sb_info *sbi,
2345 unsigned int *newseg, bool new_sec, int dir)
2346{
2347 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2348 unsigned int segno, secno, zoneno;
7cd8558b 2349 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2350 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2351 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2352 unsigned int left_start = hint;
2353 bool init = true;
2354 int go_left = 0;
2355 int i;
2356
1a118ccf 2357 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2358
2359 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2360 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2361 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2362 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2363 goto got_it;
2364 }
2365find_other_zone:
7cd8558b
JK		 2366 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2367 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2368 if (dir == ALLOC_RIGHT) {
2369 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2370 MAIN_SECS(sbi), 0);
2371 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2372 } else {
2373 go_left = 1;
2374 left_start = hint - 1;
2375 }
2376 }
2377 if (go_left == 0)
2378 goto skip_left;
2379
2380 while (test_bit(left_start, free_i->free_secmap)) {
2381 if (left_start > 0) {
2382 left_start--;
2383 continue;
2384 }
2385 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2386 MAIN_SECS(sbi), 0);
2387 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2388 break;
2389 }
2390 secno = left_start;
2391skip_left:
4ddb1a4d
JK		 2392 segno = GET_SEG_FROM_SEC(sbi, secno);
2393 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2394
2395 /* give up on finding another zone */
2396 if (!init)
2397 goto got_it;
2398 if (sbi->secs_per_zone == 1)
2399 goto got_it;
2400 if (zoneno == old_zoneno)
2401 goto got_it;
2402 if (dir == ALLOC_LEFT) {
2403 if (!go_left && zoneno + 1 >= total_zones)
2404 goto got_it;
2405 if (go_left && zoneno == 0)
2406 goto got_it;
2407 }
2408 for (i = 0; i < NR_CURSEG_TYPE; i++)
2409 if (CURSEG_I(sbi, i)->zone == zoneno)
2410 break;
2411
2412 if (i < NR_CURSEG_TYPE) {
2413 /* zone is in user, try another */
2414 if (go_left)
2415 hint = zoneno * sbi->secs_per_zone - 1;
2416 else if (zoneno + 1 >= total_zones)
2417 hint = 0;
2418 else
2419 hint = (zoneno + 1) * sbi->secs_per_zone;
2420 init = false;
2421 goto find_other_zone;
2422 }
2423got_it:
2424 /* set it as dirty segment in free segmap */
9850cf4a 2425 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2426 __set_inuse(sbi, segno);
2427 *newseg = segno;
1a118ccf 2428 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2429}
2430
2431static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2432{
2433 struct curseg_info *curseg = CURSEG_I(sbi, type);
2434 struct summary_footer *sum_footer;
2435
2436 curseg->segno = curseg->next_segno;
4ddb1a4d 2437 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2438 curseg->next_blkoff = 0;
2439 curseg->next_segno = NULL_SEGNO;
2440
2441 sum_footer = &(curseg->sum_blk->footer);
2442 memset(sum_footer, 0, sizeof(struct summary_footer));
2443 if (IS_DATASEG(type))
2444 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2445 if (IS_NODESEG(type))
2446 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2447 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2448}
2449
7a20b8a6
JK		 2450static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2451{
a7881893 2452 /* if segs_per_sec is large than 1, we need to keep original policy. */
2c70c5e3 2453 if (__is_large_section(sbi))
a7881893
JK		 2454 return CURSEG_I(sbi, type)->segno;
2455
4354994f
DR		 2456 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2457 return 0;
2458
b94929d9
YS		 2459 if (test_opt(sbi, NOHEAP) &&
2460 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2461 return 0;
2462
e066b83c
JK		 2463 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2464 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2465
2466 /* find segments from 0 to reuse freed segments */
63189b78 2467 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2468 return 0;
2469
7a20b8a6
JK		 2470 return CURSEG_I(sbi, type)->segno;
2471}
2472
0a8165d7 2473/*
351df4b2
JK		 2474 * Allocate a current working segment.
2475 * This function always allocates a free segment in LFS manner.
2476 */
2477static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2478{
2479 struct curseg_info *curseg = CURSEG_I(sbi, type);
2480 unsigned int segno = curseg->segno;
2481 int dir = ALLOC_LEFT;
2482
2483 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2484 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2485 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2486 dir = ALLOC_RIGHT;
2487
2488 if (test_opt(sbi, NOHEAP))
2489 dir = ALLOC_RIGHT;
2490
7a20b8a6 2491 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2492 get_new_segment(sbi, &segno, new_sec, dir);
2493 curseg->next_segno = segno;
2494 reset_curseg(sbi, type, 1);
2495 curseg->alloc_type = LFS;
2496}
2497
2498static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2499 struct curseg_info *seg, block_t start)
2500{
2501 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2502 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2503 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2504 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2505 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2506 int i, pos;
2507
2508 for (i = 0; i < entries; i++)
2509 target_map[i] = ckpt_map[i] | cur_map[i];
2510
2511 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2512
2513 seg->next_blkoff = pos;
351df4b2
JK		 2514}
2515
0a8165d7 2516/*
351df4b2
JK		 2517 * If a segment is written by LFS manner, next block offset is just obtained
2518 * by increasing the current block offset. However, if a segment is written by
2519 * SSR manner, next block offset obtained by calling __next_free_blkoff
2520 */
2521static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2522 struct curseg_info *seg)
2523{
2524 if (seg->alloc_type == SSR)
2525 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2526 else
2527 seg->next_blkoff++;
2528}
2529
0a8165d7 2530/*
e1c42045 2531 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2532 * manner, so it should recover the existing segment information of valid blocks
2533 */
025d63a4 2534static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2535{
2536 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2537 struct curseg_info *curseg = CURSEG_I(sbi, type);
2538 unsigned int new_segno = curseg->next_segno;
2539 struct f2fs_summary_block *sum_node;
2540 struct page *sum_page;
2541
2542 write_sum_page(sbi, curseg->sum_blk,
2543 GET_SUM_BLOCK(sbi, curseg->segno));
2544 __set_test_and_inuse(sbi, new_segno);
2545
2546 mutex_lock(&dirty_i->seglist_lock);
2547 __remove_dirty_segment(sbi, new_segno, PRE);
2548 __remove_dirty_segment(sbi, new_segno, DIRTY);
2549 mutex_unlock(&dirty_i->seglist_lock);
2550
2551 reset_curseg(sbi, type, 1);
2552 curseg->alloc_type = SSR;
2553 __next_free_blkoff(sbi, curseg, 0);
2554
4d57b86d 2555 sum_page = f2fs_get_sum_page(sbi, new_segno);
edc55aaf 2556 f2fs_bug_on(sbi, IS_ERR(sum_page));
025d63a4
CY		 2557 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2558 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2559 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2560}
2561
43727527
JK		 2562static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2563{
2564 struct curseg_info *curseg = CURSEG_I(sbi, type);
2565 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2566 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2567 int i, cnt;
2568 bool reversed = false;
c192f7a4 2569
4d57b86d 2570 /* f2fs_need_SSR() already forces to do this */
e066b83c
JK		 2571 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2572 curseg->next_segno = segno;
c192f7a4 2573 return 1;
e066b83c 2574 }
43727527 2575
70d625cb
JK		 2576 /* For node segments, let's do SSR more intensively */
2577 if (IS_NODESEG(type)) {
d27c3d89
CY		 2578 if (type >= CURSEG_WARM_NODE) {
2579 reversed = true;
2580 i = CURSEG_COLD_NODE;
2581 } else {
2582 i = CURSEG_HOT_NODE;
2583 }
2584 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2585 } else {
d27c3d89
CY		 2586 if (type >= CURSEG_WARM_DATA) {
2587 reversed = true;
2588 i = CURSEG_COLD_DATA;
2589 } else {
2590 i = CURSEG_HOT_DATA;
2591 }
2592 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2593 }
43727527 2594
d27c3d89 2595 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2596 if (i == type)
2597 continue;
e066b83c
JK		 2598 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2599 curseg->next_segno = segno;
43727527 2600 return 1;
e066b83c 2601 }
c192f7a4 2602 }
4354994f
DR		 2603
2604 /* find valid_blocks=0 in dirty list */
2605 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2606 segno = get_free_segment(sbi);
2607 if (segno != NULL_SEGNO) {
2608 curseg->next_segno = segno;
2609 return 1;
2610 }
2611 }
43727527
JK		 2612 return 0;
2613}
2614
351df4b2
JK		 2615/*
2616 * flush out current segment and replace it with new segment
2617 * This function should be returned with success, otherwise BUG
2618 */
2619static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2620 int type, bool force)
2621{
a7881893
JK		 2622 struct curseg_info *curseg = CURSEG_I(sbi, type);
2623
7b405275 2624 if (force)
351df4b2 2625 new_curseg(sbi, type, true);
5b6c6be2
JK		 2626 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2627 type == CURSEG_WARM_NODE)
351df4b2 2628 new_curseg(sbi, type, false);
4354994f
DR		 2629 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2630 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
a7881893 2631 new_curseg(sbi, type, false);
4d57b86d 2632 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2633 change_curseg(sbi, type);
351df4b2
JK		 2634 else
2635 new_curseg(sbi, type, false);
dcdfff65 2636
a7881893 2637 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2638}
2639
4d57b86d 2640void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
351df4b2 2641{
6ae1be13
JK		 2642 struct curseg_info *curseg;
2643 unsigned int old_segno;
351df4b2
JK		 2644 int i;
2645
3d26fa6b
CY		 2646 down_write(&SIT_I(sbi)->sentry_lock);
2647
6ae1be13
JK		 2648 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2649 curseg = CURSEG_I(sbi, i);
2650 old_segno = curseg->segno;
2651 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2652 locate_dirty_segment(sbi, old_segno);
2653 }
3d26fa6b
CY		 2654
2655 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2656}
2657
2658static const struct segment_allocation default_salloc_ops = {
2659 .allocate_segment = allocate_segment_by_default,
2660};
2661
4d57b86d
CY		 2662bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2663 struct cp_control *cpc)
25290fa5
JK		 2664{
2665 __u64 trim_start = cpc->trim_start;
2666 bool has_candidate = false;
2667
3d26fa6b 2668 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2669 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2670 if (add_discard_addrs(sbi, cpc, true)) {
2671 has_candidate = true;
2672 break;
2673 }
2674 }
3d26fa6b 2675 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2676
2677 cpc->trim_start = trim_start;
2678 return has_candidate;
2679}
2680
01f9cf6d 2681static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 2682 struct discard_policy *dpolicy,
2683 unsigned int start, unsigned int end)
2684{
2685 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2686 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2687 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2688 struct discard_cmd *dc;
2689 struct blk_plug plug;
2690 int issued;
01f9cf6d 2691 unsigned int trimmed = 0;
9a997188
JK		 2692
2693next:
2694 issued = 0;
2695
2696 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 2697 if (unlikely(dcc->rbtree_check))
2698 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2699 &dcc->root));
9a997188 2700
4d57b86d 2701 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 2702 NULL, start,
2703 (struct rb_entry **)&prev_dc,
2704 (struct rb_entry **)&next_dc,
4dada3fd 2705 &insert_p, &insert_parent, true, NULL);
9a997188
JK		 2706 if (!dc)
2707 dc = next_dc;
2708
2709 blk_start_plug(&plug);
2710
2711 while (dc && dc->lstart <= end) {
2712 struct rb_node *node;
6b9cb124 2713 int err = 0;
9a997188
JK		 2714
2715 if (dc->len < dpolicy->granularity)
2716 goto skip;
2717
2718 if (dc->state != D_PREP) {
2719 list_move_tail(&dc->list, &dcc->fstrim_list);
2720 goto skip;
2721 }
2722
6b9cb124 2723 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
9a997188 2724
35ec7d57 2725 if (issued >= dpolicy->max_requests) {
9a997188
JK		 2726 start = dc->lstart + dc->len;
2727
6b9cb124
CY		 2728 if (err)
2729 __remove_discard_cmd(sbi, dc);
2730
9a997188
JK		 2731 blk_finish_plug(&plug);
2732 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 2733 trimmed += __wait_all_discard_cmd(sbi, NULL);
9a997188
JK		 2734 congestion_wait(BLK_RW_ASYNC, HZ/50);
2735 goto next;
2736 }
2737skip:
2738 node = rb_next(&dc->rb_node);
6b9cb124
CY		 2739 if (err)
2740 __remove_discard_cmd(sbi, dc);
9a997188
JK		 2741 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2742
2743 if (fatal_signal_pending(current))
2744 break;
2745 }
2746
2747 blk_finish_plug(&plug);
2748 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 2749
2750 return trimmed;
9a997188
JK		 2751}
2752
4b2fecc8
JK		 2753int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2754{
f7ef9b83
JK		 2755 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2756 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2757 unsigned int start_segno, end_segno;
8412663d 2758 block_t start_block, end_block;
4b2fecc8 2759 struct cp_control cpc;
78997b56 2760 struct discard_policy dpolicy;
0ea80512 2761 unsigned long long trimmed = 0;
c34f42e2 2762 int err = 0;
2c70c5e3 2763 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
4b2fecc8 2764
836b5a63 2765 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2766 return -EINVAL;
2767
3f16ecd9
CY		 2768 if (end < MAIN_BLKADDR(sbi))
2769 goto out;
4b2fecc8 2770
ed214a11
YH		 2771 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2772 f2fs_msg(sbi->sb, KERN_WARNING,
2773 "Found FS corruption, run fsck to fix.");
3d165dc3 2774 return -EIO;
ed214a11
YH		 2775 }
2776
4b2fecc8 2777 /* start/end segment number in main_area */
7cd8558b
JK		 2778 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2779 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2780 GET_SEGNO(sbi, end);
ad6672bb
YS		 2781 if (need_align) {
2782 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2783 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2784 }
8412663d 2785
4b2fecc8 2786 cpc.reason = CP_DISCARD;
836b5a63 2787 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2788 cpc.trim_start = start_segno;
2789 cpc.trim_end = end_segno;
4b2fecc8 2790
377224c4
CY		 2791 if (sbi->discard_blks == 0)
2792 goto out;
74fa5f3d 2793
377224c4 2794 mutex_lock(&sbi->gc_mutex);
4d57b86d 2795 err = f2fs_write_checkpoint(sbi, &cpc);
377224c4
CY		 2796 mutex_unlock(&sbi->gc_mutex);
2797 if (err)
2798 goto out;
8412663d 2799
e555da9f
JK		 2800 /*
2801 * We filed discard candidates, but actually we don't need to wait for
2802 * all of them, since they'll be issued in idle time along with runtime
2803 * discard option. User configuration looks like using runtime discard
2804 * or periodic fstrim instead of it.
2805 */
7d20c8ab 2806 if (f2fs_realtime_discard_enable(sbi))
5a615492
JK		 2807 goto out;
2808
2809 start_block = START_BLOCK(sbi, start_segno);
2810 end_block = START_BLOCK(sbi, end_segno + 1);
2811
2812 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 2813 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2814 start_block, end_block);
5a615492 2815
01f9cf6d 2816 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2817 start_block, end_block);
377224c4 2818out:
6eae2694
CY		 2819 if (!err)
2820 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2821 return err;
4b2fecc8
JK		 2822}
2823
351df4b2
JK		 2824static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2825{
2826 struct curseg_info *curseg = CURSEG_I(sbi, type);
2827 if (curseg->next_blkoff < sbi->blocks_per_seg)
2828 return true;
2829 return false;
2830}
2831
4d57b86d 2832int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 2833{
2834 switch (hint) {
2835 case WRITE_LIFE_SHORT:
2836 return CURSEG_HOT_DATA;
2837 case WRITE_LIFE_EXTREME:
2838 return CURSEG_COLD_DATA;
2839 default:
2840 return CURSEG_WARM_DATA;
2841 }
2842}
2843
0cdd3195
HL		 2844/* This returns write hints for each segment type. This hints will be
2845 * passed down to block layer. There are mapping tables which depend on
2846 * the mount option 'whint_mode'.
2847 *
2848 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2849 *
2850 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2851 *
2852 * User F2FS Block
2853 * ---- ---- -----
2854 * META WRITE_LIFE_NOT_SET
2855 * HOT_NODE "
2856 * WARM_NODE "
2857 * COLD_NODE "
2858 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2859 * extension list " "
2860 *
2861 * -- buffered io
2862 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2863 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2864 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2865 * WRITE_LIFE_NONE " "
2866 * WRITE_LIFE_MEDIUM " "
2867 * WRITE_LIFE_LONG " "
2868 *
2869 * -- direct io
2870 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2871 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2872 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2873 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2874 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2875 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2876 *
f2e703f9
HL		 2877 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2878 *
2879 * User F2FS Block
2880 * ---- ---- -----
2881 * META WRITE_LIFE_MEDIUM;
2882 * HOT_NODE WRITE_LIFE_NOT_SET
2883 * WARM_NODE "
2884 * COLD_NODE WRITE_LIFE_NONE
2885 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2886 * extension list " "
2887 *
2888 * -- buffered io
2889 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2890 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2891 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2892 * WRITE_LIFE_NONE " "
2893 * WRITE_LIFE_MEDIUM " "
2894 * WRITE_LIFE_LONG " "
2895 *
2896 * -- direct io
2897 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2898 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2899 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2900 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2901 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2902 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2903 */
2904
4d57b86d 2905enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 2906 enum page_type type, enum temp_type temp)
2907{
63189b78 2908 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2909 if (type == DATA) {
f2e703f9 2910 if (temp == WARM)
0cdd3195 2911 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2912 else if (temp == HOT)
2913 return WRITE_LIFE_SHORT;
2914 else if (temp == COLD)
2915 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2916 } else {
2917 return WRITE_LIFE_NOT_SET;
2918 }
63189b78 2919 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2920 if (type == DATA) {
2921 if (temp == WARM)
2922 return WRITE_LIFE_LONG;
2923 else if (temp == HOT)
2924 return WRITE_LIFE_SHORT;
2925 else if (temp == COLD)
2926 return WRITE_LIFE_EXTREME;
2927 } else if (type == NODE) {
2928 if (temp == WARM || temp == HOT)
2929 return WRITE_LIFE_NOT_SET;
2930 else if (temp == COLD)
2931 return WRITE_LIFE_NONE;
2932 } else if (type == META) {
2933 return WRITE_LIFE_MEDIUM;
2934 }
0cdd3195 2935 }
f2e703f9 2936 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2937}
2938
81377bd6 2939static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2940{
81377bd6 2941 if (fio->type == DATA)
351df4b2
JK		 2942 return CURSEG_HOT_DATA;
2943 else
2944 return CURSEG_HOT_NODE;
2945}
2946
81377bd6 2947static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2948{
81377bd6
JK		 2949 if (fio->type == DATA) {
2950 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2951
2952 if (S_ISDIR(inode->i_mode))
2953 return CURSEG_HOT_DATA;
2954 else
2955 return CURSEG_COLD_DATA;
2956 } else {
81377bd6 2957 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2958 return CURSEG_WARM_NODE;
351df4b2
JK		 2959 else
2960 return CURSEG_COLD_NODE;
2961 }
2962}
2963
81377bd6 2964static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2965{
81377bd6
JK		 2966 if (fio->type == DATA) {
2967 struct inode *inode = fio->page->mapping->host;
351df4b2 2968
81377bd6 2969 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2970 return CURSEG_COLD_DATA;
b6a06cbb 2971 if (file_is_hot(inode) ||
b4c3ca8b 2972 is_inode_flag_set(inode, FI_HOT_DATA) ||
2079f115
CY		 2973 f2fs_is_atomic_file(inode) ||
2974 f2fs_is_volatile_file(inode))
ef095d19 2975 return CURSEG_HOT_DATA;
4d57b86d 2976 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2977 } else {
81377bd6
JK		 2978 if (IS_DNODE(fio->page))
2979 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2980 CURSEG_HOT_NODE;
ef095d19 2981 return CURSEG_COLD_NODE;
351df4b2
JK		 2982 }
2983}
2984
81377bd6 2985static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2986{
a912b54d
JK		 2987 int type = 0;
2988
63189b78 2989 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2990 case 2:
a912b54d
JK		 2991 type = __get_segment_type_2(fio);
2992 break;
351df4b2 2993 case 4:
a912b54d
JK		 2994 type = __get_segment_type_4(fio);
2995 break;
2996 case 6:
2997 type = __get_segment_type_6(fio);
2998 break;
2999 default:
3000 f2fs_bug_on(fio->sbi, true);
351df4b2 3001 }
81377bd6 3002
a912b54d
JK		 3003 if (IS_HOT(type))
3004 fio->temp = HOT;
3005 else if (IS_WARM(type))
3006 fio->temp = WARM;
3007 else
3008 fio->temp = COLD;
3009 return type;
351df4b2
JK		 3010}
3011
4d57b86d 3012void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 3013 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 3014 struct f2fs_summary *sum, int type,
3015 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 3016{
3017 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 3018 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 3019
2b60311d
CY		 3020 down_read(&SM_I(sbi)->curseg_lock);
3021
351df4b2 3022 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3023 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3024
3025 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 3026
4e6a8d9b
JK		 3027 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3028
351df4b2
JK		 3029 /*
3030 * __add_sum_entry should be resided under the curseg_mutex
3031 * because, this function updates a summary entry in the
3032 * current summary block.
3033 */
e79efe3b 3034 __add_sum_entry(sbi, type, sum);
351df4b2 3035
351df4b2 3036 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 3037
3038 stat_inc_block_count(sbi, curseg);
351df4b2 3039
65f1b80b
YS		 3040 /*
3041 * SIT information should be updated before segment allocation,
3042 * since SSR needs latest valid block information.
3043 */
3044 update_sit_entry(sbi, *new_blkaddr, 1);
3045 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3046 update_sit_entry(sbi, old_blkaddr, -1);
3047
c6f82fe9
JK		 3048 if (!__has_curseg_space(sbi, type))
3049 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 3050
351df4b2 3051 /*
65f1b80b
YS		 3052 * segment dirty status should be updated after segment allocation,
3053 * so we just need to update status only one time after previous
3054 * segment being closed.
351df4b2 3055 */
65f1b80b
YS		 3056 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3057 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 3058
3d26fa6b 3059 up_write(&sit_i->sentry_lock);
351df4b2 3060
704956ec 3061 if (page && IS_NODESEG(type)) {
351df4b2
JK		 3062 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3063
704956ec
CY		 3064 f2fs_inode_chksum_set(sbi, page);
3065 }
3066
fb830fc5
CY		 3067 if (add_list) {
3068 struct f2fs_bio_info *io;
3069
3070 INIT_LIST_HEAD(&fio->list);
3071 fio->in_list = true;
fe16efe6 3072 fio->retry = false;
fb830fc5
CY		 3073 io = sbi->write_io[fio->type] + fio->temp;
3074 spin_lock(&io->io_lock);
3075 list_add_tail(&fio->list, &io->io_list);
3076 spin_unlock(&io->io_lock);
3077 }
3078
bfad7c2d 3079 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 3080
3081 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 3082}
3083
39d787be
CY		 3084static void update_device_state(struct f2fs_io_info *fio)
3085{
3086 struct f2fs_sb_info *sbi = fio->sbi;
3087 unsigned int devidx;
3088
0916878d 3089 if (!f2fs_is_multi_device(sbi))
39d787be
CY		 3090 return;
3091
3092 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3093
3094 /* update device state for fsync */
4d57b86d 3095 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 3096
3097 /* update device state for checkpoint */
3098 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3099 spin_lock(&sbi->dev_lock);
3100 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3101 spin_unlock(&sbi->dev_lock);
3102 }
39d787be
CY		 3103}
3104
05ca3632 3105static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 3106{
81377bd6 3107 int type = __get_segment_type(fio);
107a805d 3108 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 3109
107a805d
CY		 3110 if (keep_order)
3111 down_read(&fio->sbi->io_order_lock);
0a595eba 3112reallocate:
4d57b86d 3113 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 3114 &fio->new_blkaddr, sum, type, fio, true);
6aa58d8a
CY		 3115 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3116 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3117 fio->old_blkaddr, fio->old_blkaddr);
bfad7c2d 3118
351df4b2 3119 /* writeout dirty page into bdev */
fe16efe6
CY		 3120 f2fs_submit_page_write(fio);
3121 if (fio->retry) {
0a595eba
JK		 3122 fio->old_blkaddr = fio->new_blkaddr;
3123 goto reallocate;
3124 }
fe16efe6
CY		 3125
3126 update_device_state(fio);
3127
107a805d
CY		 3128 if (keep_order)
3129 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 3130}
3131
4d57b86d 3132void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 3133 enum iostat_type io_type)
351df4b2 3134{
458e6197 3135 struct f2fs_io_info fio = {
05ca3632 3136 .sbi = sbi,
458e6197 3137 .type = META,
0cdd3195 3138 .temp = HOT,
04d328de 3139 .op = REQ_OP_WRITE,
70fd7614 3140 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 3141 .old_blkaddr = page->index,
3142 .new_blkaddr = page->index,
05ca3632 3143 .page = page,
4375a336 3144 .encrypted_page = NULL,
fb830fc5 3145 .in_list = false,
458e6197
JK		 3146 };
3147
2b947003 3148 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 3149 fio.op_flags &= ~REQ_META;
2b947003 3150
351df4b2 3151 set_page_writeback(page);
17c50035 3152 ClearPageError(page);
b9109b0e 3153 f2fs_submit_page_write(&fio);
b0af6d49 3154
b63e7be5 3155 stat_inc_meta_count(sbi, page->index);
b0af6d49 3156 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 3157}
3158
4d57b86d 3159void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 3160{
3161 struct f2fs_summary sum;
05ca3632 3162
351df4b2 3163 set_summary(&sum, nid, 0, 0);
05ca3632 3164 do_write_page(&sum, fio);
b0af6d49
CY		 3165
3166 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3167}
3168
4d57b86d
CY		 3169void f2fs_outplace_write_data(struct dnode_of_data *dn,
3170 struct f2fs_io_info *fio)
351df4b2 3171{
05ca3632 3172 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2 3173 struct f2fs_summary sum;
351df4b2 3174
9850cf4a 3175 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
7735730d 3176 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
05ca3632 3177 do_write_page(&sum, fio);
f28b3434 3178 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 3179
3180 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3181}
3182
4d57b86d 3183int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 3184{
b0af6d49 3185 int err;
d21b0f23 3186 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 3187
7a9d7548 3188 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 3189 /* i/o temperature is needed for passing down write hints */
3190 __get_segment_type(fio);
d21b0f23
YH		 3191
3192 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3193 GET_SEGNO(sbi, fio->new_blkaddr))->type));
3194
05ca3632 3195 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 3196
3197 err = f2fs_submit_page_bio(fio);
e46f6bd8 3198 if (!err) {
39d787be 3199 update_device_state(fio);
e46f6bd8
CY		 3200 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3201 }
b0af6d49
CY		 3202
3203 return err;
351df4b2
JK		 3204}
3205
2b60311d
CY		 3206static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3207 unsigned int segno)
3208{
3209 int i;
3210
3211 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3212 if (CURSEG_I(sbi, i)->segno == segno)
3213 break;
3214 }
3215 return i;
3216}
3217
4d57b86d 3218void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 3219 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 3220 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 3221{
3222 struct sit_info *sit_i = SIT_I(sbi);
3223 struct curseg_info *curseg;
3224 unsigned int segno, old_cursegno;
3225 struct seg_entry *se;
3226 int type;
19f106bc 3227 unsigned short old_blkoff;
351df4b2
JK		 3228
3229 segno = GET_SEGNO(sbi, new_blkaddr);
3230 se = get_seg_entry(sbi, segno);
3231 type = se->type;
3232
2b60311d
CY		 3233 down_write(&SM_I(sbi)->curseg_lock);
3234
19f106bc
CY		 3235 if (!recover_curseg) {
3236 /* for recovery flow */
3237 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3238 if (old_blkaddr == NULL_ADDR)
3239 type = CURSEG_COLD_DATA;
3240 else
3241 type = CURSEG_WARM_DATA;
3242 }
3243 } else {
2b60311d
CY		 3244 if (IS_CURSEG(sbi, segno)) {
3245 /* se->type is volatile as SSR allocation */
3246 type = __f2fs_get_curseg(sbi, segno);
3247 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3248 } else {
351df4b2 3249 type = CURSEG_WARM_DATA;
2b60311d 3250 }
351df4b2 3251 }
19f106bc 3252
2c190504 3253 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 3254 curseg = CURSEG_I(sbi, type);
3255
3256 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3257 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3258
3259 old_cursegno = curseg->segno;
19f106bc 3260 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 3261
3262 /* change the current segment */
3263 if (segno != curseg->segno) {
3264 curseg->next_segno = segno;
025d63a4 3265 change_curseg(sbi, type);
351df4b2
JK		 3266 }
3267
491c0854 3268 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 3269 __add_sum_entry(sbi, type, sum);
351df4b2 3270
28bc106b 3271 if (!recover_curseg || recover_newaddr)
6e2c64ad 3272 update_sit_entry(sbi, new_blkaddr, 1);
6aa58d8a
CY		 3273 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3274 invalidate_mapping_pages(META_MAPPING(sbi),
3275 old_blkaddr, old_blkaddr);
6e2c64ad 3276 update_sit_entry(sbi, old_blkaddr, -1);
6aa58d8a 3277 }
6e2c64ad
JK		 3278
3279 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3280 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3281
351df4b2 3282 locate_dirty_segment(sbi, old_cursegno);
351df4b2 3283
19f106bc
CY		 3284 if (recover_curseg) {
3285 if (old_cursegno != curseg->segno) {
3286 curseg->next_segno = old_cursegno;
025d63a4 3287 change_curseg(sbi, type);
19f106bc
CY		 3288 }
3289 curseg->next_blkoff = old_blkoff;
3290 }
3291
3d26fa6b 3292 up_write(&sit_i->sentry_lock);
351df4b2 3293 mutex_unlock(&curseg->curseg_mutex);
2b60311d 3294 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 3295}
3296
528e3459
CY		 3297void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3298 block_t old_addr, block_t new_addr,
28bc106b
CY		 3299 unsigned char version, bool recover_curseg,
3300 bool recover_newaddr)
528e3459
CY		 3301{
3302 struct f2fs_summary sum;
3303
3304 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3305
4d57b86d 3306 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
28bc106b 3307 recover_curseg, recover_newaddr);
528e3459 3308
f28b3434 3309 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 3310}
3311
93dfe2ac 3312void f2fs_wait_on_page_writeback(struct page *page,
bae0ee7a 3313 enum page_type type, bool ordered, bool locked)
93dfe2ac 3314{
93dfe2ac 3315 if (PageWriteback(page)) {
4081363f
JK		 3316 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3317
bab475c5 3318 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
bae0ee7a 3319 if (ordered) {
fec1d657 3320 wait_on_page_writeback(page);
bae0ee7a
CY		 3321 f2fs_bug_on(sbi, locked && PageWriteback(page));
3322 } else {
fec1d657 3323 wait_for_stable_page(page);
bae0ee7a 3324 }
93dfe2ac
JK		 3325 }
3326}
3327
0ded69f6 3328void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
08b39fbd 3329{
0ded69f6 3330 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
08b39fbd
CY		 3331 struct page *cpage;
3332
0ded69f6
JK		 3333 if (!f2fs_post_read_required(inode))
3334 return;
3335
e1da7872 3336 if (!is_valid_data_blkaddr(sbi, blkaddr))
08b39fbd
CY		 3337 return;
3338
08b39fbd
CY		 3339 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3340 if (cpage) {
bae0ee7a 3341 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
08b39fbd
CY		 3342 f2fs_put_page(cpage, 1);
3343 }
3344}
3345
1e78e8bd
ST		 3346void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3347 block_t len)
3348{
3349 block_t i;
3350
3351 for (i = 0; i < len; i++)
3352 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3353}
3354
7735730d 3355static int read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3356{
3357 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3358 struct curseg_info *seg_i;
3359 unsigned char *kaddr;
3360 struct page *page;
3361 block_t start;
3362 int i, j, offset;
3363
3364 start = start_sum_block(sbi);
3365
4d57b86d 3366 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3367 if (IS_ERR(page))
3368 return PTR_ERR(page);
351df4b2
JK		 3369 kaddr = (unsigned char *)page_address(page);
3370
3371 /* Step 1: restore nat cache */
3372 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3373 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3374
3375 /* Step 2: restore sit cache */
3376 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3377 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3378 offset = 2 * SUM_JOURNAL_SIZE;
3379
3380 /* Step 3: restore summary entries */
3381 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3382 unsigned short blk_off;
3383 unsigned int segno;
3384
3385 seg_i = CURSEG_I(sbi, i);
3386 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3387 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3388 seg_i->next_segno = segno;
3389 reset_curseg(sbi, i, 0);
3390 seg_i->alloc_type = ckpt->alloc_type[i];
3391 seg_i->next_blkoff = blk_off;
3392
3393 if (seg_i->alloc_type == SSR)
3394 blk_off = sbi->blocks_per_seg;
3395
3396 for (j = 0; j < blk_off; j++) {
3397 struct f2fs_summary *s;
3398 s = (struct f2fs_summary *)(kaddr + offset);
3399 seg_i->sum_blk->entries[j] = *s;
3400 offset += SUMMARY_SIZE;
09cbfeaf 3401 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3402 SUM_FOOTER_SIZE)
3403 continue;
3404
3405 f2fs_put_page(page, 1);
3406 page = NULL;
3407
4d57b86d 3408 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3409 if (IS_ERR(page))
3410 return PTR_ERR(page);
351df4b2
JK		 3411 kaddr = (unsigned char *)page_address(page);
3412 offset = 0;
3413 }
3414 }
3415 f2fs_put_page(page, 1);
7735730d 3416 return 0;
351df4b2
JK		 3417}
3418
3419static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3420{
3421 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3422 struct f2fs_summary_block *sum;
3423 struct curseg_info *curseg;
3424 struct page *new;
3425 unsigned short blk_off;
3426 unsigned int segno = 0;
3427 block_t blk_addr = 0;
7735730d 3428 int err = 0;
351df4b2
JK		 3429
3430 /* get segment number and block addr */
3431 if (IS_DATASEG(type)) {
3432 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3433 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3434 CURSEG_HOT_DATA]);
119ee914 3435 if (__exist_node_summaries(sbi))
351df4b2
JK		 3436 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3437 else
3438 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3439 } else {
3440 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3441 CURSEG_HOT_NODE]);
3442 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3443 CURSEG_HOT_NODE]);
119ee914 3444 if (__exist_node_summaries(sbi))
351df4b2
JK		 3445 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3446 type - CURSEG_HOT_NODE);
3447 else
3448 blk_addr = GET_SUM_BLOCK(sbi, segno);
3449 }
3450
4d57b86d 3451 new = f2fs_get_meta_page(sbi, blk_addr);
7735730d
CY		 3452 if (IS_ERR(new))
3453 return PTR_ERR(new);
351df4b2
JK		 3454 sum = (struct f2fs_summary_block *)page_address(new);
3455
3456 if (IS_NODESEG(type)) {
119ee914 3457 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3458 struct f2fs_summary *ns = &sum->entries[0];
3459 int i;
3460 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3461 ns->version = 0;
3462 ns->ofs_in_node = 0;
3463 }
3464 } else {
7735730d
CY		 3465 err = f2fs_restore_node_summary(sbi, segno, sum);
3466 if (err)
3467 goto out;
351df4b2
JK		 3468 }
3469 }
3470
3471 /* set uncompleted segment to curseg */
3472 curseg = CURSEG_I(sbi, type);
3473 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3474
3475 /* update journal info */
3476 down_write(&curseg->journal_rwsem);
3477 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3478 up_write(&curseg->journal_rwsem);
3479
3480 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3481 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3482 curseg->next_segno = segno;
3483 reset_curseg(sbi, type, 0);
3484 curseg->alloc_type = ckpt->alloc_type[type];
3485 curseg->next_blkoff = blk_off;
3486 mutex_unlock(&curseg->curseg_mutex);
7735730d 3487out:
351df4b2 3488 f2fs_put_page(new, 1);
7735730d 3489 return err;
351df4b2
JK		 3490}
3491
3492static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3493{
21d3f8e1
JQ		 3494 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3495 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3496 int type = CURSEG_HOT_DATA;
e4fc5fbf 3497 int err;
351df4b2 3498
aaec2b1d 3499 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3500 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3501
3502 if (npages >= 2)
4d57b86d 3503 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3504 META_CP, true);
3fa06d7b 3505
351df4b2 3506 /* restore for compacted data summary */
7735730d
CY		 3507 err = read_compacted_summaries(sbi);
3508 if (err)
3509 return err;
351df4b2
JK		 3510 type = CURSEG_HOT_NODE;
3511 }
3512
119ee914 3513 if (__exist_node_summaries(sbi))
4d57b86d 3514 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3515 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3516
e4fc5fbf
CY		 3517 for (; type <= CURSEG_COLD_NODE; type++) {
3518 err = read_normal_summaries(sbi, type);
3519 if (err)
3520 return err;
3521 }
3522
21d3f8e1
JQ		 3523 /* sanity check for summary blocks */
3524 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3525 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3526 return -EINVAL;
3527
351df4b2
JK		 3528 return 0;
3529}
3530
3531static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3532{
3533 struct page *page;
3534 unsigned char *kaddr;
3535 struct f2fs_summary *summary;
3536 struct curseg_info *seg_i;
3537 int written_size = 0;
3538 int i, j;
3539
4d57b86d 3540 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3541 kaddr = (unsigned char *)page_address(page);
81114baa 3542 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3543
3544 /* Step 1: write nat cache */
3545 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3546 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3547 written_size += SUM_JOURNAL_SIZE;
3548
3549 /* Step 2: write sit cache */
3550 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3551 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3552 written_size += SUM_JOURNAL_SIZE;
3553
351df4b2
JK		 3554 /* Step 3: write summary entries */
3555 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3556 unsigned short blkoff;
3557 seg_i = CURSEG_I(sbi, i);
3558 if (sbi->ckpt->alloc_type[i] == SSR)
3559 blkoff = sbi->blocks_per_seg;
3560 else
3561 blkoff = curseg_blkoff(sbi, i);
3562
3563 for (j = 0; j < blkoff; j++) {
3564 if (!page) {
4d57b86d 3565 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3566 kaddr = (unsigned char *)page_address(page);
81114baa 3567 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3568 written_size = 0;
3569 }
3570 summary = (struct f2fs_summary *)(kaddr + written_size);
3571 *summary = seg_i->sum_blk->entries[j];
3572 written_size += SUMMARY_SIZE;
351df4b2 3573
09cbfeaf 3574 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3575 SUM_FOOTER_SIZE)
3576 continue;
3577
e8d61a74 3578 set_page_dirty(page);
351df4b2
JK		 3579 f2fs_put_page(page, 1);
3580 page = NULL;
3581 }
3582 }
e8d61a74
CY		 3583 if (page) {
3584 set_page_dirty(page);
351df4b2 3585 f2fs_put_page(page, 1);
e8d61a74 3586 }
351df4b2
JK		 3587}
3588
3589static void write_normal_summaries(struct f2fs_sb_info *sbi,
3590 block_t blkaddr, int type)
3591{
3592 int i, end;
3593 if (IS_DATASEG(type))
3594 end = type + NR_CURSEG_DATA_TYPE;
3595 else
3596 end = type + NR_CURSEG_NODE_TYPE;
3597
b7ad7512
CY		 3598 for (i = type; i < end; i++)
3599 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3600}
3601
4d57b86d 3602void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3603{
aaec2b1d 3604 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3605 write_compacted_summaries(sbi, start_blk);
3606 else
3607 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3608}
3609
4d57b86d 3610void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3611{
119ee914 3612 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3613}
3614
4d57b86d 3615int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3616 unsigned int val, int alloc)
3617{
3618 int i;
3619
3620 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3621 for (i = 0; i < nats_in_cursum(journal); i++) {
3622 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3623 return i;
3624 }
dfc08a12
CY		 3625 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3626 return update_nats_in_cursum(journal, 1);
351df4b2 3627 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3628 for (i = 0; i < sits_in_cursum(journal); i++)
3629 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3630 return i;
dfc08a12
CY		 3631 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3632 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3633 }
3634 return -1;
3635}
3636
3637static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3638 unsigned int segno)
3639{
7735730d 3640 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3641}
3642
3643static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3644 unsigned int start)
3645{
3646 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3647 struct page *page;
351df4b2 3648 pgoff_t src_off, dst_off;
351df4b2
JK		 3649
3650 src_off = current_sit_addr(sbi, start);
3651 dst_off = next_sit_addr(sbi, src_off);
3652
4d57b86d 3653 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 3654 seg_info_to_sit_page(sbi, page, start);
351df4b2 3655
068c3cd8 3656 set_page_dirty(page);
351df4b2
JK		 3657 set_to_next_sit(sit_i, start);
3658
068c3cd8 3659 return page;
351df4b2
JK		 3660}
3661
184a5cd2
CY		 3662static struct sit_entry_set *grab_sit_entry_set(void)
3663{
3664 struct sit_entry_set *ses =
80c54505 3665 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3666
3667 ses->entry_cnt = 0;
3668 INIT_LIST_HEAD(&ses->set_list);
3669 return ses;
3670}
3671
3672static void release_sit_entry_set(struct sit_entry_set *ses)
3673{
3674 list_del(&ses->set_list);
3675 kmem_cache_free(sit_entry_set_slab, ses);
3676}
3677
3678static void adjust_sit_entry_set(struct sit_entry_set *ses,
3679 struct list_head *head)
3680{
3681 struct sit_entry_set *next = ses;
3682
3683 if (list_is_last(&ses->set_list, head))
3684 return;
3685
3686 list_for_each_entry_continue(next, head, set_list)
3687 if (ses->entry_cnt <= next->entry_cnt)
3688 break;
3689
3690 list_move_tail(&ses->set_list, &next->set_list);
3691}
3692
3693static void add_sit_entry(unsigned int segno, struct list_head *head)
3694{
3695 struct sit_entry_set *ses;
3696 unsigned int start_segno = START_SEGNO(segno);
3697
3698 list_for_each_entry(ses, head, set_list) {
3699 if (ses->start_segno == start_segno) {
3700 ses->entry_cnt++;
3701 adjust_sit_entry_set(ses, head);
3702 return;
3703 }
3704 }
3705
3706 ses = grab_sit_entry_set();
3707
3708 ses->start_segno = start_segno;
3709 ses->entry_cnt++;
3710 list_add(&ses->set_list, head);
3711}
3712
3713static void add_sits_in_set(struct f2fs_sb_info *sbi)
3714{
3715 struct f2fs_sm_info *sm_info = SM_I(sbi);
3716 struct list_head *set_list = &sm_info->sit_entry_set;
3717 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3718 unsigned int segno;
3719
7cd8558b 3720 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3721 add_sit_entry(segno, set_list);
3722}
3723
3724static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3725{
3726 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3727 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3728 int i;
3729
b7ad7512 3730 down_write(&curseg->journal_rwsem);
dfc08a12 3731 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3732 unsigned int segno;
3733 bool dirtied;
3734
dfc08a12 3735 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3736 dirtied = __mark_sit_entry_dirty(sbi, segno);
3737
3738 if (!dirtied)
3739 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3740 }
dfc08a12 3741 update_sits_in_cursum(journal, -i);
b7ad7512 3742 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3743}
3744
0a8165d7 3745/*
351df4b2
JK		 3746 * CP calls this function, which flushes SIT entries including sit_journal,
3747 * and moves prefree segs to free segs.
3748 */
4d57b86d 3749void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3750{
3751 struct sit_info *sit_i = SIT_I(sbi);
3752 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3753 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3754 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3755 struct sit_entry_set *ses, *tmp;
3756 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3757 bool to_journal = true;
4b2fecc8 3758 struct seg_entry *se;
351df4b2 3759
3d26fa6b 3760 down_write(&sit_i->sentry_lock);
351df4b2 3761
2b11a74b
WL		 3762 if (!sit_i->dirty_sentries)
3763 goto out;
3764
351df4b2 3765 /*
184a5cd2
CY		 3766 * add and account sit entries of dirty bitmap in sit entry
3767 * set temporarily
351df4b2 3768 */
184a5cd2 3769 add_sits_in_set(sbi);
351df4b2 3770
184a5cd2
CY		 3771 /*
3772 * if there are no enough space in journal to store dirty sit
3773 * entries, remove all entries from journal and add and account
3774 * them in sit entry set.
3775 */
dfc08a12 3776 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3777 remove_sits_in_journal(sbi);
b2955550 3778
184a5cd2
CY		 3779 /*
3780 * there are two steps to flush sit entries:
3781 * #1, flush sit entries to journal in current cold data summary block.
3782 * #2, flush sit entries to sit page.
3783 */
3784 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3785 struct page *page = NULL;
184a5cd2
CY		 3786 struct f2fs_sit_block *raw_sit = NULL;
3787 unsigned int start_segno = ses->start_segno;
3788 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3789 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3790 unsigned int segno = start_segno;
3791
3792 if (to_journal &&
dfc08a12 3793 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3794 to_journal = false;
3795
b7ad7512
CY		 3796 if (to_journal) {
3797 down_write(&curseg->journal_rwsem);
3798 } else {
184a5cd2
CY		 3799 page = get_next_sit_page(sbi, start_segno);
3800 raw_sit = page_address(page);
351df4b2 3801 }
351df4b2 3802
184a5cd2
CY		 3803 /* flush dirty sit entries in region of current sit set */
3804 for_each_set_bit_from(segno, bitmap, end) {
3805 int offset, sit_offset;
4b2fecc8
JK		 3806
3807 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3808#ifdef CONFIG_F2FS_CHECK_FS
3809 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3810 SIT_VBLOCK_MAP_SIZE))
3811 f2fs_bug_on(sbi, 1);
3812#endif
184a5cd2
CY		 3813
3814 /* add discard candidates */
c473f1a9 3815 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3816 cpc->trim_start = segno;
25290fa5 3817 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3818 }
184a5cd2
CY		 3819
3820 if (to_journal) {
4d57b86d 3821 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 3822 SIT_JOURNAL, segno, 1);
3823 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3824 segno_in_journal(journal, offset) =
184a5cd2
CY		 3825 cpu_to_le32(segno);
3826 seg_info_to_raw_sit(se,
dfc08a12 3827 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3828 check_block_count(sbi, segno,
3829 &sit_in_journal(journal, offset));
184a5cd2
CY		 3830 } else {
3831 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3832 seg_info_to_raw_sit(se,
3833 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3834 check_block_count(sbi, segno,
3835 &raw_sit->entries[sit_offset]);
184a5cd2 3836 }
351df4b2 3837
184a5cd2
CY		 3838 __clear_bit(segno, bitmap);
3839 sit_i->dirty_sentries--;
3840 ses->entry_cnt--;
351df4b2
JK		 3841 }
3842
b7ad7512
CY		 3843 if (to_journal)
3844 up_write(&curseg->journal_rwsem);
3845 else
184a5cd2
CY		 3846 f2fs_put_page(page, 1);
3847
3848 f2fs_bug_on(sbi, ses->entry_cnt);
3849 release_sit_entry_set(ses);
351df4b2 3850 }
184a5cd2
CY		 3851
3852 f2fs_bug_on(sbi, !list_empty(head));
3853 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3854out:
c473f1a9 3855 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3856 __u64 trim_start = cpc->trim_start;
3857
4b2fecc8 3858 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3859 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3860
3861 cpc->trim_start = trim_start;
4b2fecc8 3862 }
3d26fa6b 3863 up_write(&sit_i->sentry_lock);
351df4b2 3864
351df4b2
JK		 3865 set_prefree_as_free_segments(sbi);
3866}
3867
3868static int build_sit_info(struct f2fs_sb_info *sbi)
3869{
3870 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3871 struct sit_info *sit_i;
3872 unsigned int sit_segs, start;
ae27d62e 3873 char *src_bitmap;
351df4b2
JK		 3874 unsigned int bitmap_size;
3875
3876 /* allocate memory for SIT information */
acbf054d 3877 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3878 if (!sit_i)
3879 return -ENOMEM;
3880
3881 SM_I(sbi)->sit_info = sit_i;
3882
9d2a789c
KC		 3883 sit_i->sentries =
3884 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3885 MAIN_SEGS(sbi)),
3886 GFP_KERNEL);
351df4b2
JK		 3887 if (!sit_i->sentries)
3888 return -ENOMEM;
3889
7cd8558b 3890 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3891 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3892 GFP_KERNEL);
351df4b2
JK		 3893 if (!sit_i->dirty_sentries_bitmap)
3894 return -ENOMEM;
3895
7cd8558b 3896 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3897 sit_i->sentries[start].cur_valid_map
acbf054d 3898 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3899 sit_i->sentries[start].ckpt_valid_map
acbf054d 3900 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3901 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3902 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3903 return -ENOMEM;
3e025740 3904
355e7891
CY		 3905#ifdef CONFIG_F2FS_CHECK_FS
3906 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3907 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3908 if (!sit_i->sentries[start].cur_valid_map_mir)
3909 return -ENOMEM;
3910#endif
3911
7d20c8ab
CY		 3912 sit_i->sentries[start].discard_map
3913 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3914 GFP_KERNEL);
3915 if (!sit_i->sentries[start].discard_map)
3916 return -ENOMEM;
351df4b2
JK		 3917 }
3918
acbf054d 3919 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3920 if (!sit_i->tmp_map)
3921 return -ENOMEM;
3922
2c70c5e3 3923 if (__is_large_section(sbi)) {
9d2a789c
KC		 3924 sit_i->sec_entries =
3925 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3926 MAIN_SECS(sbi)),
3927 GFP_KERNEL);
351df4b2
JK		 3928 if (!sit_i->sec_entries)
3929 return -ENOMEM;
3930 }
3931
3932 /* get information related with SIT */
3933 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3934
3935 /* setup SIT bitmap from ckeckpoint pack */
3936 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3937 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3938
ae27d62e
CY		 3939 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3940 if (!sit_i->sit_bitmap)
351df4b2 3941 return -ENOMEM;
351df4b2 3942
ae27d62e
CY		 3943#ifdef CONFIG_F2FS_CHECK_FS
3944 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3945 if (!sit_i->sit_bitmap_mir)
3946 return -ENOMEM;
3947#endif
3948
351df4b2
JK		 3949 /* init SIT information */
3950 sit_i->s_ops = &default_salloc_ops;
3951
3952 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3953 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3954 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3955 sit_i->bitmap_size = bitmap_size;
3956 sit_i->dirty_sentries = 0;
3957 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3958 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3959 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3960 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3961 return 0;
3962}
3963
3964static int build_free_segmap(struct f2fs_sb_info *sbi)
3965{
351df4b2
JK		 3966 struct free_segmap_info *free_i;
3967 unsigned int bitmap_size, sec_bitmap_size;
3968
3969 /* allocate memory for free segmap information */
acbf054d 3970 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3971 if (!free_i)
3972 return -ENOMEM;
3973
3974 SM_I(sbi)->free_info = free_i;
3975
7cd8558b 3976 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3977 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3978 if (!free_i->free_segmap)
3979 return -ENOMEM;
3980
7cd8558b 3981 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3982 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3983 if (!free_i->free_secmap)
3984 return -ENOMEM;
3985
3986 /* set all segments as dirty temporarily */
3987 memset(free_i->free_segmap, 0xff, bitmap_size);
3988 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3989
3990 /* init free segmap information */
7cd8558b 3991 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3992 free_i->free_segments = 0;
3993 free_i->free_sections = 0;
1a118ccf 3994 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3995 return 0;
3996}
3997
3998static int build_curseg(struct f2fs_sb_info *sbi)
3999{
1042d60f 4000 struct curseg_info *array;
351df4b2
JK		 4001 int i;
4002
026f0507
KC		 4003 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4004 GFP_KERNEL);
351df4b2
JK		 4005 if (!array)
4006 return -ENOMEM;
4007
4008 SM_I(sbi)->curseg_array = array;
4009
4010 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4011 mutex_init(&array[i].curseg_mutex);
acbf054d 4012 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 4013 if (!array[i].sum_blk)
4014 return -ENOMEM;
b7ad7512 4015 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 4016 array[i].journal = f2fs_kzalloc(sbi,
4017 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 4018 if (!array[i].journal)
4019 return -ENOMEM;
351df4b2
JK		 4020 array[i].segno = NULL_SEGNO;
4021 array[i].next_blkoff = 0;
4022 }
4023 return restore_curseg_summaries(sbi);
4024}
4025
c39a1b34 4026static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 4027{
4028 struct sit_info *sit_i = SIT_I(sbi);
4029 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 4030 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 4031 struct seg_entry *se;
4032 struct f2fs_sit_entry sit;
74de593a
CY		 4033 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4034 unsigned int i, start, end;
4035 unsigned int readed, start_blk = 0;
c39a1b34 4036 int err = 0;
8a29c126 4037 block_t total_node_blocks = 0;
351df4b2 4038
74de593a 4039 do {
4d57b86d 4040 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
664ba972 4041 META_SIT, true);
74de593a
CY		 4042
4043 start = start_blk * sit_i->sents_per_block;
4044 end = (start_blk + readed) * sit_i->sents_per_block;
4045
7cd8558b 4046 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 4047 struct f2fs_sit_block *sit_blk;
74de593a
CY		 4048 struct page *page;
4049
9c094040 4050 se = &sit_i->sentries[start];
74de593a 4051 page = get_current_sit_page(sbi, start);
edc55aaf
JK		 4052 if (IS_ERR(page))
4053 return PTR_ERR(page);
74de593a
CY		 4054 sit_blk = (struct f2fs_sit_block *)page_address(page);
4055 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4056 f2fs_put_page(page, 1);
d600af23 4057
c39a1b34
JK		 4058 err = check_block_count(sbi, start, &sit);
4059 if (err)
4060 return err;
74de593a 4061 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4062 if (IS_NODESEG(se->type))
4063 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 4064
4065 /* build discard map only one time */
7d20c8ab
CY		 4066 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4067 memset(se->discard_map, 0xff,
4068 SIT_VBLOCK_MAP_SIZE);
4069 } else {
4070 memcpy(se->discard_map,
4071 se->cur_valid_map,
4072 SIT_VBLOCK_MAP_SIZE);
4073 sbi->discard_blks +=
4074 sbi->blocks_per_seg -
4075 se->valid_blocks;
3e025740 4076 }
a66cdd98 4077
2c70c5e3 4078 if (__is_large_section(sbi))
d600af23
CY		 4079 get_sec_entry(sbi, start)->valid_blocks +=
4080 se->valid_blocks;
351df4b2 4081 }
74de593a
CY		 4082 start_blk += readed;
4083 } while (start_blk < sit_blk_cnt);
d600af23
CY		 4084
4085 down_read(&curseg->journal_rwsem);
4086 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 4087 unsigned int old_valid_blocks;
4088
4089 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 4090 if (start >= MAIN_SEGS(sbi)) {
4091 f2fs_msg(sbi->sb, KERN_ERR,
4092 "Wrong journal entry on segno %u",
4093 start);
4094 set_sbi_flag(sbi, SBI_NEED_FSCK);
4095 err = -EINVAL;
4096 break;
4097 }
4098
d600af23
CY		 4099 se = &sit_i->sentries[start];
4100 sit = sit_in_journal(journal, i);
4101
4102 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 4103 if (IS_NODESEG(se->type))
4104 total_node_blocks -= old_valid_blocks;
d600af23 4105
c39a1b34
JK		 4106 err = check_block_count(sbi, start, &sit);
4107 if (err)
4108 break;
d600af23 4109 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4110 if (IS_NODESEG(se->type))
4111 total_node_blocks += se->valid_blocks;
d600af23 4112
7d20c8ab
CY		 4113 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4114 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4115 } else {
4116 memcpy(se->discard_map, se->cur_valid_map,
4117 SIT_VBLOCK_MAP_SIZE);
4118 sbi->discard_blks += old_valid_blocks;
4119 sbi->discard_blks -= se->valid_blocks;
d600af23
CY		 4120 }
4121
2c70c5e3 4122 if (__is_large_section(sbi)) {
d600af23 4123 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 4124 se->valid_blocks;
4125 get_sec_entry(sbi, start)->valid_blocks -=
4126 old_valid_blocks;
4127 }
d600af23
CY		 4128 }
4129 up_read(&curseg->journal_rwsem);
8a29c126
JK		 4130
4131 if (!err && total_node_blocks != valid_node_count(sbi)) {
4132 f2fs_msg(sbi->sb, KERN_ERR,
4133 "SIT is corrupted node# %u vs %u",
4134 total_node_blocks, valid_node_count(sbi));
4135 set_sbi_flag(sbi, SBI_NEED_FSCK);
4136 err = -EINVAL;
4137 }
4138
c39a1b34 4139 return err;
351df4b2
JK		 4140}
4141
4142static void init_free_segmap(struct f2fs_sb_info *sbi)
4143{
4144 unsigned int start;
4145 int type;
4146
7cd8558b 4147 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 4148 struct seg_entry *sentry = get_seg_entry(sbi, start);
4149 if (!sentry->valid_blocks)
4150 __set_free(sbi, start);
c79b7ff1
JK		 4151 else
4152 SIT_I(sbi)->written_valid_blocks +=
4153 sentry->valid_blocks;
351df4b2
JK		 4154 }
4155
4156 /* set use the current segments */
4157 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4158 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4159 __set_test_and_inuse(sbi, curseg_t->segno);
4160 }
4161}
4162
4163static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4164{
4165 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4166 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 4167 unsigned int segno = 0, offset = 0;
351df4b2
JK		 4168 unsigned short valid_blocks;
4169
8736fbf0 4170 while (1) {
351df4b2 4171 /* find dirty segment based on free segmap */
7cd8558b
JK		 4172 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4173 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 4174 break;
4175 offset = segno + 1;
302bd348 4176 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 4177 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 4178 continue;
ec325b52
JK		 4179 if (valid_blocks > sbi->blocks_per_seg) {
4180 f2fs_bug_on(sbi, 1);
4181 continue;
4182 }
351df4b2
JK		 4183 mutex_lock(&dirty_i->seglist_lock);
4184 __locate_dirty_segment(sbi, segno, DIRTY);
4185 mutex_unlock(&dirty_i->seglist_lock);
4186 }
4187}
4188
5ec4e49f 4189static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4190{
4191 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 4192 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 4193
628b3d14 4194 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 4195 if (!dirty_i->victim_secmap)
351df4b2
JK		 4196 return -ENOMEM;
4197 return 0;
4198}
4199
4200static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4201{
4202 struct dirty_seglist_info *dirty_i;
4203 unsigned int bitmap_size, i;
4204
4205 /* allocate memory for dirty segments list information */
acbf054d
CY		 4206 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4207 GFP_KERNEL);
351df4b2
JK		 4208 if (!dirty_i)
4209 return -ENOMEM;
4210
4211 SM_I(sbi)->dirty_info = dirty_i;
4212 mutex_init(&dirty_i->seglist_lock);
4213
7cd8558b 4214 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 4215
4216 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 4217 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4218 GFP_KERNEL);
351df4b2
JK		 4219 if (!dirty_i->dirty_segmap[i])
4220 return -ENOMEM;
4221 }
4222
4223 init_dirty_segmap(sbi);
5ec4e49f 4224 return init_victim_secmap(sbi);
351df4b2
JK		 4225}
4226
0a8165d7 4227/*
351df4b2
JK		 4228 * Update min, max modified time for cost-benefit GC algorithm
4229 */
4230static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4231{
4232 struct sit_info *sit_i = SIT_I(sbi);
4233 unsigned int segno;
4234
3d26fa6b 4235 down_write(&sit_i->sentry_lock);
351df4b2 4236
5ad25442 4237 sit_i->min_mtime = ULLONG_MAX;
351df4b2 4238
7cd8558b 4239 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 4240 unsigned int i;
4241 unsigned long long mtime = 0;
4242
4243 for (i = 0; i < sbi->segs_per_sec; i++)
4244 mtime += get_seg_entry(sbi, segno + i)->mtime;
4245
4246 mtime = div_u64(mtime, sbi->segs_per_sec);
4247
4248 if (sit_i->min_mtime > mtime)
4249 sit_i->min_mtime = mtime;
4250 }
a1f72ac2 4251 sit_i->max_mtime = get_mtime(sbi, false);
3d26fa6b 4252 up_write(&sit_i->sentry_lock);
351df4b2
JK		 4253}
4254
4d57b86d 4255int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4256{
4257 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4258 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 4259 struct f2fs_sm_info *sm_info;
351df4b2
JK		 4260 int err;
4261
acbf054d 4262 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 4263 if (!sm_info)
4264 return -ENOMEM;
4265
4266 /* init sm info */
4267 sbi->sm_info = sm_info;
351df4b2
JK		 4268 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4269 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4270 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4271 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4272 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4273 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4274 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 4275 sm_info->rec_prefree_segments = sm_info->main_segments *
4276 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 4277 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4278 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4279
52763a4b
JK		 4280 if (!test_opt(sbi, LFS))
4281 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 4282 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 4283 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
853137ce 4284 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
ef095d19 4285 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 4286 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 4287
184a5cd2
CY		 4288 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4289
2b60311d
CY		 4290 init_rwsem(&sm_info->curseg_lock);
4291
d4fdf8ba 4292 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 4293 err = f2fs_create_flush_cmd_control(sbi);
2163d198 4294 if (err)
a688b9d9 4295 return err;
6b4afdd7
JK		 4296 }
4297
0b54fb84
JK		 4298 err = create_discard_cmd_control(sbi);
4299 if (err)
4300 return err;
4301
351df4b2
JK		 4302 err = build_sit_info(sbi);
4303 if (err)
4304 return err;
4305 err = build_free_segmap(sbi);
4306 if (err)
4307 return err;
4308 err = build_curseg(sbi);
4309 if (err)
4310 return err;
4311
4312 /* reinit free segmap based on SIT */
c39a1b34
JK		 4313 err = build_sit_entries(sbi);
4314 if (err)
4315 return err;
351df4b2
JK		 4316
4317 init_free_segmap(sbi);
4318 err = build_dirty_segmap(sbi);
4319 if (err)
4320 return err;
4321
4322 init_min_max_mtime(sbi);
4323 return 0;
4324}
4325
4326static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4327 enum dirty_type dirty_type)
4328{
4329 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4330
4331 mutex_lock(&dirty_i->seglist_lock);
39307a8e 4332 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 4333 dirty_i->nr_dirty[dirty_type] = 0;
4334 mutex_unlock(&dirty_i->seglist_lock);
4335}
4336
5ec4e49f 4337static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4338{
4339 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 4340 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 4341}
4342
4343static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4344{
4345 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4346 int i;
4347
4348 if (!dirty_i)
4349 return;
4350
4351 /* discard pre-free/dirty segments list */
4352 for (i = 0; i < NR_DIRTY_TYPE; i++)
4353 discard_dirty_segmap(sbi, i);
4354
5ec4e49f 4355 destroy_victim_secmap(sbi);
351df4b2 4356 SM_I(sbi)->dirty_info = NULL;
5222595d 4357 kvfree(dirty_i);
351df4b2
JK		 4358}
4359
4360static void destroy_curseg(struct f2fs_sb_info *sbi)
4361{
4362 struct curseg_info *array = SM_I(sbi)->curseg_array;
4363 int i;
4364
4365 if (!array)
4366 return;
4367 SM_I(sbi)->curseg_array = NULL;
b7ad7512 4368 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5222595d
JK		 4369 kvfree(array[i].sum_blk);
4370 kvfree(array[i].journal);
b7ad7512 4371 }
5222595d 4372 kvfree(array);
351df4b2
JK		 4373}
4374
4375static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4376{
4377 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4378 if (!free_i)
4379 return;
4380 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4381 kvfree(free_i->free_segmap);
4382 kvfree(free_i->free_secmap);
5222595d 4383 kvfree(free_i);
351df4b2
JK		 4384}
4385
4386static void destroy_sit_info(struct f2fs_sb_info *sbi)
4387{
4388 struct sit_info *sit_i = SIT_I(sbi);
4389 unsigned int start;
4390
4391 if (!sit_i)
4392 return;
4393
4394 if (sit_i->sentries) {
7cd8558b 4395 for (start = 0; start < MAIN_SEGS(sbi); start++) {
5222595d 4396 kvfree(sit_i->sentries[start].cur_valid_map);
355e7891 4397#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4398 kvfree(sit_i->sentries[start].cur_valid_map_mir);
355e7891 4399#endif
5222595d
JK		 4400 kvfree(sit_i->sentries[start].ckpt_valid_map);
4401 kvfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4402 }
4403 }
5222595d 4404 kvfree(sit_i->tmp_map);
60a3b782 4405
39307a8e
JK		 4406 kvfree(sit_i->sentries);
4407 kvfree(sit_i->sec_entries);
4408 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4409
4410 SM_I(sbi)->sit_info = NULL;
5222595d 4411 kvfree(sit_i->sit_bitmap);
ae27d62e 4412#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4413 kvfree(sit_i->sit_bitmap_mir);
ae27d62e 4414#endif
5222595d 4415 kvfree(sit_i);
351df4b2
JK		 4416}
4417
4d57b86d 4418void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4419{
4420 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4421
3b03f724
CY		 4422 if (!sm_info)
4423 return;
4d57b86d 4424 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 4425 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4426 destroy_dirty_segmap(sbi);
4427 destroy_curseg(sbi);
4428 destroy_free_segmap(sbi);
4429 destroy_sit_info(sbi);
4430 sbi->sm_info = NULL;
5222595d 4431 kvfree(sm_info);
351df4b2 4432}
7fd9e544 4433
4d57b86d 4434int __init f2fs_create_segment_manager_caches(void)
7fd9e544
JK		 4435{
4436 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4437 sizeof(struct discard_entry));
7fd9e544 4438 if (!discard_entry_slab)
184a5cd2
CY		 4439 goto fail;
4440
b01a9201
JK		 4441 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4442 sizeof(struct discard_cmd));
4443 if (!discard_cmd_slab)
6ab2a308 4444 goto destroy_discard_entry;
275b66b0 4445
184a5cd2 4446 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4447 sizeof(struct sit_entry_set));
184a5cd2 4448 if (!sit_entry_set_slab)
b01a9201 4449 goto destroy_discard_cmd;
88b88a66
JK		 4450
4451 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4452 sizeof(struct inmem_pages));
4453 if (!inmem_entry_slab)
4454 goto destroy_sit_entry_set;
7fd9e544 4455 return 0;
184a5cd2 4456
88b88a66
JK		 4457destroy_sit_entry_set:
4458 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4459destroy_discard_cmd:
4460 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4461destroy_discard_entry:
184a5cd2
CY		 4462 kmem_cache_destroy(discard_entry_slab);
4463fail:
4464 return -ENOMEM;
7fd9e544
JK		 4465}
4466
4d57b86d 4467void f2fs_destroy_segment_manager_caches(void)
7fd9e544 4468{
184a5cd2 4469 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4470 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4471 kmem_cache_destroy(discard_entry_slab);
88b88a66 4472 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4473}
Ubuntu Jammy Linux kernel mirror