projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
lockd: Update the NLMv4 nlm_res arguments decoder to use struct xdr_stream
[linux-block.git] / fs / f2fs / segment.c
CommitLineData
7c1a000d 1// SPDX-License-Identifier: GPL-2.0
0a8165d7 2/*
351df4b2
JK		 3 * fs/f2fs/segment.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
351df4b2
JK		 7 */
8#include <linux/fs.h>
9#include <linux/f2fs_fs.h>
10#include <linux/bio.h>
11#include <linux/blkdev.h>
690e4a3e 12#include <linux/prefetch.h>
6b4afdd7 13#include <linux/kthread.h>
74de593a 14#include <linux/swap.h>
60b99b48 15#include <linux/timer.h>
1d7be270 16#include <linux/freezer.h>
1eb1ef4a 17#include <linux/sched/signal.h>
351df4b2
JK		 18
19#include "f2fs.h"
20#include "segment.h"
21#include "node.h"
5f656541 22#include "gc.h"
6ec178da 23#include <trace/events/f2fs.h>
351df4b2 24
9a7f143a
CL		 25#define __reverse_ffz(x) __reverse_ffs(~(x))
26
7fd9e544 27static struct kmem_cache *discard_entry_slab;
b01a9201 28static struct kmem_cache *discard_cmd_slab;
184a5cd2 29static struct kmem_cache *sit_entry_set_slab;
88b88a66 30static struct kmem_cache *inmem_entry_slab;
7fd9e544 31
f96999c3
JK		 32static unsigned long __reverse_ulong(unsigned char *str)
33{
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
36
37#if BITS_PER_LONG == 64
38 shift = 56;
39#endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
43 }
44 return tmp;
45}
46
9a7f143a
CL		 47/*
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
50 */
51static inline unsigned long __reverse_ffs(unsigned long word)
52{
53 int num = 0;
54
55#if BITS_PER_LONG == 64
f96999c3 56 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 57 num += 32;
f96999c3 58 else
9a7f143a 59 word >>= 32;
9a7f143a 60#endif
f96999c3 61 if ((word & 0xffff0000) == 0)
9a7f143a 62 num += 16;
f96999c3 63 else
9a7f143a 64 word >>= 16;
f96999c3
JK		 65
66 if ((word & 0xff00) == 0)
9a7f143a 67 num += 8;
f96999c3 68 else
9a7f143a 69 word >>= 8;
f96999c3 70
9a7f143a
CL		 71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
f96999c3 75
9a7f143a
CL		 76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
f96999c3 80
9a7f143a
CL		 81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
84}
85
86/*
e1c42045 87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 89 * @size must be integral times of unsigned long.
9a7f143a 90 * Example:
f96999c3
JK		 91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 94 */
95static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
97{
98 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 99 unsigned long result = size;
9a7f143a 100 unsigned long tmp;
9a7f143a
CL		 101
102 if (offset >= size)
103 return size;
104
692223d1 105 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 106 offset %= BITS_PER_LONG;
f96999c3 107
692223d1
FL		 108 while (1) {
109 if (*p == 0)
110 goto pass;
9a7f143a 111
f96999c3 112 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 113
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 117 if (tmp)
692223d1
FL		 118 goto found;
119pass:
120 if (size <= BITS_PER_LONG)
121 break;
9a7f143a 122 size -= BITS_PER_LONG;
692223d1 123 offset = 0;
f96999c3 124 p++;
9a7f143a 125 }
692223d1
FL		 126 return result;
127found:
128 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 129}
130
131static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
133{
134 const unsigned long *p = addr + BIT_WORD(offset);
80609448 135 unsigned long result = size;
9a7f143a 136 unsigned long tmp;
9a7f143a
CL		 137
138 if (offset >= size)
139 return size;
140
80609448 141 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 142 offset %= BITS_PER_LONG;
80609448
JK		 143
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
147
f96999c3 148 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 149
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
f96999c3 154 if (tmp != ~0UL)
80609448
JK		 155 goto found;
156pass:
157 if (size <= BITS_PER_LONG)
158 break;
9a7f143a 159 size -= BITS_PER_LONG;
80609448 160 offset = 0;
f96999c3 161 p++;
9a7f143a 162 }
80609448
JK		 163 return result;
164found:
165 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 166}
167
4d57b86d 168bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
b3a97a2a
JK		 169{
170 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
171 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
172 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
173
b0332a0f 174 if (f2fs_lfs_mode(sbi))
b3a97a2a 175 return false;
0e5e8111 176 if (sbi->gc_mode == GC_URGENT_HIGH)
b3a97a2a 177 return true;
4354994f
DR		 178 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
179 return true;
b3a97a2a
JK		 180
181 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
a2a12b67 182 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
b3a97a2a
JK		 183}
184
4d57b86d 185void f2fs_register_inmem_page(struct inode *inode, struct page *page)
88b88a66 186{
88b88a66 187 struct inmem_pages *new;
9be32d72 188
594b6d04
YC		 189 if (PagePrivate(page))
190 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
191 else
192 f2fs_set_page_private(page, ATOMIC_WRITTEN_PAGE);
decd36b6 193
88b88a66
JK		 194 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
195
196 /* add atomic page indices to the list */
197 new->page = page;
198 INIT_LIST_HEAD(&new->list);
decd36b6 199
88b88a66 200 /* increase reference count with clean state */
88b88a66 201 get_page(page);
743b620c
JK		 202 mutex_lock(&F2FS_I(inode)->inmem_lock);
203 list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
8dcf2ff7 204 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
743b620c 205 mutex_unlock(&F2FS_I(inode)->inmem_lock);
8ce67cb0
JK		 206
207 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 208}
209
28bc106b 210static int __revoke_inmem_pages(struct inode *inode,
48432984
CY		 211 struct list_head *head, bool drop, bool recover,
212 bool trylock)
29b96b54 213{
28bc106b 214 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 215 struct inmem_pages *cur, *tmp;
28bc106b 216 int err = 0;
29b96b54
CY		 217
218 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 219 struct page *page = cur->page;
220
221 if (drop)
222 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
223
48432984
CY		 224 if (trylock) {
225 /*
226 * to avoid deadlock in between page lock and
227 * inmem_lock.
228 */
229 if (!trylock_page(page))
230 continue;
231 } else {
232 lock_page(page);
233 }
29b96b54 234
bae0ee7a 235 f2fs_wait_on_page_writeback(page, DATA, true, true);
e5e5732d 236
28bc106b
CY		 237 if (recover) {
238 struct dnode_of_data dn;
239 struct node_info ni;
240
241 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 242retry:
28bc106b 243 set_new_dnode(&dn, inode, NULL, NULL, 0);
4d57b86d
CY		 244 err = f2fs_get_dnode_of_data(&dn, page->index,
245 LOOKUP_NODE);
7f2b4e8e
CY		 246 if (err) {
247 if (err == -ENOMEM) {
5df7731f
CY		 248 congestion_wait(BLK_RW_ASYNC,
249 DEFAULT_IO_TIMEOUT);
7f2b4e8e
CY		 250 cond_resched();
251 goto retry;
252 }
28bc106b
CY		 253 err = -EAGAIN;
254 goto next;
255 }
7735730d
CY		 256
257 err = f2fs_get_node_info(sbi, dn.nid, &ni);
258 if (err) {
259 f2fs_put_dnode(&dn);
260 return err;
261 }
262
f1d2564a 263 if (cur->old_addr == NEW_ADDR) {
4d57b86d 264 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
f1d2564a
DJ		 265 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
266 } else
267 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 268 cur->old_addr, ni.version, true, true);
269 f2fs_put_dnode(&dn);
270 }
271next:
63c52d78 272 /* we don't need to invalidate this in the sccessful status */
2baf0781 273 if (drop || recover) {
63c52d78 274 ClearPageUptodate(page);
2baf0781
CY		 275 clear_cold_data(page);
276 }
240a5915 277 f2fs_clear_page_private(page);
28bc106b 278 f2fs_put_page(page, 1);
29b96b54
CY		 279
280 list_del(&cur->list);
281 kmem_cache_free(inmem_entry_slab, cur);
282 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
283 }
28bc106b 284 return err;
29b96b54
CY		 285}
286
4d57b86d 287void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
57864ae5
JK		 288{
289 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
290 struct inode *inode;
291 struct f2fs_inode_info *fi;
677017d1
ST		 292 unsigned int count = sbi->atomic_files;
293 unsigned int looped = 0;
57864ae5
JK		 294next:
295 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
296 if (list_empty(head)) {
297 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
298 return;
299 }
300 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
301 inode = igrab(&fi->vfs_inode);
677017d1
ST		 302 if (inode)
303 list_move_tail(&fi->inmem_ilist, head);
57864ae5
JK		 304 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
305
306 if (inode) {
2ef79ecb 307 if (gc_failure) {
677017d1
ST		 308 if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
309 goto skip;
2ef79ecb 310 }
2ef79ecb 311 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
4d57b86d 312 f2fs_drop_inmem_pages(inode);
677017d1 313skip:
57864ae5
JK		 314 iput(inode);
315 }
5df7731f 316 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
57864ae5 317 cond_resched();
677017d1
ST		 318 if (gc_failure) {
319 if (++looped >= count)
320 return;
321 }
57864ae5
JK		 322 goto next;
323}
324
4d57b86d 325void f2fs_drop_inmem_pages(struct inode *inode)
29b96b54 326{
57864ae5 327 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 328 struct f2fs_inode_info *fi = F2FS_I(inode);
329
be1ee45d 330 do {
48432984 331 mutex_lock(&fi->inmem_lock);
be1ee45d
YZ		 332 if (list_empty(&fi->inmem_pages)) {
333 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
334
335 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
336 if (!list_empty(&fi->inmem_ilist))
337 list_del_init(&fi->inmem_ilist);
338 if (f2fs_is_atomic_file(inode)) {
339 clear_inode_flag(inode, FI_ATOMIC_FILE);
340 sbi->atomic_files--;
341 }
342 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
343
344 mutex_unlock(&fi->inmem_lock);
345 break;
346 }
48432984
CY		 347 __revoke_inmem_pages(inode, &fi->inmem_pages,
348 true, false, true);
48432984 349 mutex_unlock(&fi->inmem_lock);
be1ee45d 350 } while (1);
29b96b54
CY		 351}
352
4d57b86d 353void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
8c242db9
JK		 354{
355 struct f2fs_inode_info *fi = F2FS_I(inode);
356 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
357 struct list_head *head = &fi->inmem_pages;
358 struct inmem_pages *cur = NULL;
359
360 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
361
362 mutex_lock(&fi->inmem_lock);
363 list_for_each_entry(cur, head, list) {
364 if (cur->page == page)
365 break;
366 }
367
d0891e84 368 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
8c242db9
JK		 369 list_del(&cur->list);
370 mutex_unlock(&fi->inmem_lock);
371
372 dec_page_count(sbi, F2FS_INMEM_PAGES);
373 kmem_cache_free(inmem_entry_slab, cur);
374
375 ClearPageUptodate(page);
240a5915 376 f2fs_clear_page_private(page);
8c242db9
JK		 377 f2fs_put_page(page, 0);
378
379 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
380}
381
4d57b86d 382static int __f2fs_commit_inmem_pages(struct inode *inode)
88b88a66
JK		 383{
384 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
385 struct f2fs_inode_info *fi = F2FS_I(inode);
386 struct inmem_pages *cur, *tmp;
88b88a66 387 struct f2fs_io_info fio = {
05ca3632 388 .sbi = sbi,
39d787be 389 .ino = inode->i_ino,
88b88a66 390 .type = DATA,
04d328de 391 .op = REQ_OP_WRITE,
70fd7614 392 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 393 .io_type = FS_DATA_IO,
88b88a66 394 };
cf52b27a 395 struct list_head revoke_list;
bab475c5 396 bool submit_bio = false;
edb27dee 397 int err = 0;
88b88a66 398
cf52b27a
CY		 399 INIT_LIST_HEAD(&revoke_list);
400
88b88a66 401 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 402 struct page *page = cur->page;
403
404 lock_page(page);
405 if (page->mapping == inode->i_mapping) {
406 trace_f2fs_commit_inmem_page(page, INMEM);
407
bae0ee7a 408 f2fs_wait_on_page_writeback(page, DATA, true, true);
8d64d365
CY		 409
410 set_page_dirty(page);
933439c8 411 if (clear_page_dirty_for_io(page)) {
29b96b54 412 inode_dec_dirty_pages(inode);
4d57b86d 413 f2fs_remove_dirty_inode(inode);
933439c8 414 }
640cc189 415retry:
28bc106b 416 fio.page = page;
e959c8f5 417 fio.old_blkaddr = NULL_ADDR;
4d978078 418 fio.encrypted_page = NULL;
cc15620b 419 fio.need_lock = LOCK_DONE;
4d57b86d 420 err = f2fs_do_write_data_page(&fio);
29b96b54 421 if (err) {
640cc189 422 if (err == -ENOMEM) {
5df7731f
CY		 423 congestion_wait(BLK_RW_ASYNC,
424 DEFAULT_IO_TIMEOUT);
640cc189
JK		 425 cond_resched();
426 goto retry;
427 }
28bc106b 428 unlock_page(page);
29b96b54 429 break;
70c640b1 430 }
28bc106b
CY		 431 /* record old blkaddr for revoking */
432 cur->old_addr = fio.old_blkaddr;
bab475c5 433 submit_bio = true;
28bc106b
CY		 434 }
435 unlock_page(page);
cf52b27a 436 list_move_tail(&cur->list, &revoke_list);
88b88a66 437 }
29b96b54 438
bab475c5
CY		 439 if (submit_bio)
440 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
28bc106b 441
cf52b27a
CY		 442 if (err) {
443 /*
444 * try to revoke all committed pages, but still we could fail
445 * due to no memory or other reason, if that happened, EAGAIN
446 * will be returned, which means in such case, transaction is
447 * already not integrity, caller should use journal to do the
448 * recovery or rewrite & commit last transaction. For other
449 * error number, revoking was done by filesystem itself.
450 */
48432984
CY		 451 err = __revoke_inmem_pages(inode, &revoke_list,
452 false, true, false);
cf52b27a
CY		 453
454 /* drop all uncommitted pages */
48432984
CY		 455 __revoke_inmem_pages(inode, &fi->inmem_pages,
456 true, false, false);
cf52b27a 457 } else {
48432984
CY		 458 __revoke_inmem_pages(inode, &revoke_list,
459 false, false, false);
cf52b27a 460 }
28bc106b 461
29b96b54
CY		 462 return err;
463}
464
4d57b86d 465int f2fs_commit_inmem_pages(struct inode *inode)
29b96b54
CY		 466{
467 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
468 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b 469 int err;
29b96b54
CY		 470
471 f2fs_balance_fs(sbi, true);
29b96b54 472
6f8d4455
JK		 473 down_write(&fi->i_gc_rwsem[WRITE]);
474
475 f2fs_lock_op(sbi);
5fe45743
CY		 476 set_inode_flag(inode, FI_ATOMIC_COMMIT);
477
29b96b54 478 mutex_lock(&fi->inmem_lock);
4d57b86d 479 err = __f2fs_commit_inmem_pages(inode);
88b88a66
JK		 480 mutex_unlock(&fi->inmem_lock);
481
5fe45743
CY		 482 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
483
29b96b54 484 f2fs_unlock_op(sbi);
6f8d4455
JK		 485 up_write(&fi->i_gc_rwsem[WRITE]);
486
edb27dee 487 return err;
88b88a66
JK		 488}
489
0a8165d7 490/*
351df4b2
JK		 491 * This function balances dirty node and dentry pages.
492 * In addition, it controls garbage collection.
493 */
2c4db1a6 494void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 495{
55523519 496 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
c45d6002 497 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
0f348028 498 f2fs_stop_checkpoint(sbi, false);
55523519 499 }
0f348028 500
e589c2c4 501 /* balance_fs_bg is able to be pending */
a7881893 502 if (need && excess_cached_nats(sbi))
7bcd0cfa 503 f2fs_balance_fs_bg(sbi, false);
e589c2c4 504
00e09c0b 505 if (!f2fs_is_checkpoint_ready(sbi))
4354994f
DR		 506 return;
507
351df4b2 508 /*
029cd28c
JK		 509 * We should do GC or end up with checkpoint, if there are so many dirty
510 * dir/node pages without enough free segments.
351df4b2 511 */
7f3037a5 512 if (has_not_enough_free_secs(sbi, 0, 0)) {
5911d2d1
CY		 513 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
514 sbi->gc_thread->f2fs_gc_task) {
515 DEFINE_WAIT(wait);
516
517 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
518 TASK_UNINTERRUPTIBLE);
519 wake_up(&sbi->gc_thread->gc_wait_queue_head);
520 io_schedule();
521 finish_wait(&sbi->gc_thread->fggc_wq, &wait);
522 } else {
523 down_write(&sbi->gc_lock);
524 f2fs_gc(sbi, false, false, false, NULL_SEGNO);
525 }
351df4b2
JK		 526 }
527}
528
7bcd0cfa 529void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
4660f9c0 530{
64c74a7a
CY		 531 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
532 return;
533
1dcc336b 534 /* try to shrink extent cache when there is no enough memory */
4d57b86d 535 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
554df79e 536 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 537
1b38dc8e 538 /* check the # of cached NAT entries */
4d57b86d
CY		 539 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
540 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
1b38dc8e 541
4d57b86d
CY		 542 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
543 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
ad4edb83 544 else
4d57b86d 545 f2fs_build_free_nids(sbi, false, false);
31696580 546
493720a4
CY		 547 if (excess_dirty_nats(sbi) || excess_dirty_nodes(sbi) ||
548 excess_prefree_segs(sbi))
549 goto do_sync;
550
551 /* there is background inflight IO or foreground operation recently */
552 if (is_inflight_io(sbi, REQ_TIME) ||
553 (!f2fs_time_over(sbi, REQ_TIME) && rwsem_is_locked(&sbi->cp_rwsem)))
f455c8a5 554 return;
31696580 555
493720a4
CY		 556 /* exceed periodical checkpoint timeout threshold */
557 if (f2fs_time_over(sbi, CP_TIME))
558 goto do_sync;
559
1b38dc8e 560 /* checkpoint is the only way to shrink partial cached entries */
493720a4
CY		 561 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
562 f2fs_available_free_memory(sbi, INO_ENTRIES))
563 return;
564
565do_sync:
566 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
567 struct blk_plug plug;
568
569 mutex_lock(&sbi->flush_lock);
570
571 blk_start_plug(&plug);
572 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
573 blk_finish_plug(&plug);
574
575 mutex_unlock(&sbi->flush_lock);
36b35a0d 576 }
493720a4
CY		 577 f2fs_sync_fs(sbi->sb, true);
578 stat_inc_bg_cp_count(sbi->stat_info);
4660f9c0
JK		 579}
580
20fda56b
KM		 581static int __submit_flush_wait(struct f2fs_sb_info *sbi,
582 struct block_device *bdev)
3c62be17 583{
25ac8426 584 int ret = blkdev_issue_flush(bdev);
20fda56b
KM		 585
586 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
587 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 588 return ret;
589}
590
39d787be 591static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 592{
39d787be 593 int ret = 0;
3c62be17
JK		 594 int i;
595
0916878d 596 if (!f2fs_is_multi_device(sbi))
39d787be 597 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 598
39d787be 599 for (i = 0; i < sbi->s_ndevs; i++) {
4d57b86d 600 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
39d787be 601 continue;
20fda56b
KM		 602 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
603 if (ret)
604 break;
3c62be17
JK		 605 }
606 return ret;
607}
608
2163d198 609static int issue_flush_thread(void *data)
6b4afdd7
JK		 610{
611 struct f2fs_sb_info *sbi = data;
b01a9201 612 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 613 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 614repeat:
615 if (kthread_should_stop())
616 return 0;
617
721bd4d5 618 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 619 struct flush_cmd *cmd, *next;
620 int ret;
621
721bd4d5
GZ		 622 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
623 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
624
39d787be
CY		 625 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
626
627 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 628 atomic_inc(&fcc->issued_flush);
629
721bd4d5
GZ		 630 llist_for_each_entry_safe(cmd, next,
631 fcc->dispatch_list, llnode) {
6b4afdd7 632 cmd->ret = ret;
6b4afdd7
JK		 633 complete(&cmd->wait);
634 }
a688b9d9 635 fcc->dispatch_list = NULL;
6b4afdd7
JK		 636 }
637
a688b9d9 638 wait_event_interruptible(*q,
721bd4d5 639 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 640 goto repeat;
641}
642
39d787be 643int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 644{
b01a9201 645 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 646 struct flush_cmd cmd;
8b8dd65f 647 int ret;
6b4afdd7 648
0f7b2abd
JK		 649 if (test_opt(sbi, NOBARRIER))
650 return 0;
651
8b8dd65f 652 if (!test_opt(sbi, FLUSH_MERGE)) {
72691af6 653 atomic_inc(&fcc->queued_flush);
39d787be 654 ret = submit_flush_wait(sbi, ino);
72691af6 655 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 656 atomic_inc(&fcc->issued_flush);
657 return ret;
658 }
740432f8 659
0916878d
DLM		 660 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
661 f2fs_is_multi_device(sbi)) {
39d787be 662 ret = submit_flush_wait(sbi, ino);
72691af6 663 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 664
665 atomic_inc(&fcc->issued_flush);
740432f8
JK		 666 return ret;
667 }
6b4afdd7 668
39d787be 669 cmd.ino = ino;
adf8d90b 670 init_completion(&cmd.wait);
6b4afdd7 671
721bd4d5 672 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 673
3b42c741
CY		 674 /*
675 * update issue_list before we wake up issue_flush thread, this
676 * smp_mb() pairs with another barrier in ___wait_event(), see
677 * more details in comments of waitqueue_active().
678 */
6f890df0
CY		 679 smp_mb();
680
681 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 682 wake_up(&fcc->flush_wait_queue);
6b4afdd7 683
5eba8c5d
JK		 684 if (fcc->f2fs_issue_flush) {
685 wait_for_completion(&cmd.wait);
72691af6 686 atomic_dec(&fcc->queued_flush);
5eba8c5d 687 } else {
d3238691
CY		 688 struct llist_node *list;
689
690 list = llist_del_all(&fcc->issue_list);
691 if (!list) {
692 wait_for_completion(&cmd.wait);
72691af6 693 atomic_dec(&fcc->queued_flush);
d3238691
CY		 694 } else {
695 struct flush_cmd *tmp, *next;
696
39d787be 697 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 698
699 llist_for_each_entry_safe(tmp, next, list, llnode) {
700 if (tmp == &cmd) {
701 cmd.ret = ret;
72691af6 702 atomic_dec(&fcc->queued_flush);
d3238691
CY		 703 continue;
704 }
705 tmp->ret = ret;
706 complete(&tmp->wait);
707 }
708 }
5eba8c5d 709 }
adf8d90b
CY		 710
711 return cmd.ret;
6b4afdd7
JK		 712}
713
4d57b86d 714int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
2163d198
GZ		 715{
716 dev_t dev = sbi->sb->s_bdev->bd_dev;
717 struct flush_cmd_control *fcc;
718 int err = 0;
719
b01a9201
JK		 720 if (SM_I(sbi)->fcc_info) {
721 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 722 if (fcc->f2fs_issue_flush)
723 return err;
5eba8c5d
JK		 724 goto init_thread;
725 }
726
acbf054d 727 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 728 if (!fcc)
729 return -ENOMEM;
8b8dd65f 730 atomic_set(&fcc->issued_flush, 0);
72691af6 731 atomic_set(&fcc->queued_flush, 0);
2163d198 732 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 733 init_llist_head(&fcc->issue_list);
b01a9201 734 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 735 if (!test_opt(sbi, FLUSH_MERGE))
736 return err;
737
5eba8c5d 738init_thread:
2163d198
GZ		 739 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
740 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
741 if (IS_ERR(fcc->f2fs_issue_flush)) {
742 err = PTR_ERR(fcc->f2fs_issue_flush);
c8eb7024 743 kfree(fcc);
b01a9201 744 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 745 return err;
746 }
2163d198
GZ		 747
748 return err;
749}
750
4d57b86d 751void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 752{
b01a9201 753 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 754
5eba8c5d
JK		 755 if (fcc && fcc->f2fs_issue_flush) {
756 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
757
758 fcc->f2fs_issue_flush = NULL;
759 kthread_stop(flush_thread);
760 }
761 if (free) {
c8eb7024 762 kfree(fcc);
b01a9201 763 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 764 }
2163d198
GZ		 765}
766
1228b482
CY		 767int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
768{
769 int ret = 0, i;
770
0916878d 771 if (!f2fs_is_multi_device(sbi))
1228b482
CY		 772 return 0;
773
6ed29fe1
CY		 774 if (test_opt(sbi, NOBARRIER))
775 return 0;
776
1228b482
CY		 777 for (i = 1; i < sbi->s_ndevs; i++) {
778 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
779 continue;
780 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
781 if (ret)
782 break;
783
784 spin_lock(&sbi->dev_lock);
785 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
786 spin_unlock(&sbi->dev_lock);
787 }
788
789 return ret;
790}
791
351df4b2
JK		 792static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
793 enum dirty_type dirty_type)
794{
795 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
796
797 /* need not be added */
798 if (IS_CURSEG(sbi, segno))
799 return;
800
801 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
802 dirty_i->nr_dirty[dirty_type]++;
803
804 if (dirty_type == DIRTY) {
805 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 806 enum dirty_type t = sentry->type;
b2f2c390 807
ec325b52
JK		 808 if (unlikely(t >= DIRTY)) {
809 f2fs_bug_on(sbi, 1);
810 return;
811 }
4625d6aa
CL		 812 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
813 dirty_i->nr_dirty[t]++;
da52f8ad
JQ		 814
815 if (__is_large_section(sbi)) {
816 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
123aaf77 817 block_t valid_blocks =
da52f8ad
JQ		 818 get_valid_blocks(sbi, segno, true);
819
820 f2fs_bug_on(sbi, unlikely(!valid_blocks ||
821 valid_blocks == BLKS_PER_SEC(sbi)));
822
823 if (!IS_CURSEC(sbi, secno))
824 set_bit(secno, dirty_i->dirty_secmap);
825 }
351df4b2
JK		 826 }
827}
828
829static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
830 enum dirty_type dirty_type)
831{
832 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
123aaf77 833 block_t valid_blocks;
351df4b2
JK		 834
835 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
836 dirty_i->nr_dirty[dirty_type]--;
837
838 if (dirty_type == DIRTY) {
4625d6aa
CL		 839 struct seg_entry *sentry = get_seg_entry(sbi, segno);
840 enum dirty_type t = sentry->type;
841
842 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
843 dirty_i->nr_dirty[t]--;
b2f2c390 844
da52f8ad
JQ		 845 valid_blocks = get_valid_blocks(sbi, segno, true);
846 if (valid_blocks == 0) {
4ddb1a4d 847 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 848 dirty_i->victim_secmap);
bbf9f7d9
ST		 849#ifdef CONFIG_F2FS_CHECK_FS
850 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
851#endif
852 }
da52f8ad
JQ		 853 if (__is_large_section(sbi)) {
854 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
855
856 if (!valid_blocks ||
857 valid_blocks == BLKS_PER_SEC(sbi)) {
858 clear_bit(secno, dirty_i->dirty_secmap);
859 return;
860 }
861
862 if (!IS_CURSEC(sbi, secno))
863 set_bit(secno, dirty_i->dirty_secmap);
864 }
351df4b2
JK		 865 }
866}
867
0a8165d7 868/*
351df4b2
JK		 869 * Should not occur error such as -ENOMEM.
870 * Adding dirty entry into seglist is not critical operation.
871 * If a given segment is one of current working segments, it won't be added.
872 */
8d8451af 873static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 874{
875 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4354994f 876 unsigned short valid_blocks, ckpt_valid_blocks;
de881df9 877 unsigned int usable_blocks;
351df4b2
JK		 878
879 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
880 return;
881
de881df9 882 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
351df4b2
JK		 883 mutex_lock(&dirty_i->seglist_lock);
884
302bd348 885 valid_blocks = get_valid_blocks(sbi, segno, false);
61461fc9 886 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
351df4b2 887
4354994f 888 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
de881df9 889 ckpt_valid_blocks == usable_blocks)) {
351df4b2
JK		 890 __locate_dirty_segment(sbi, segno, PRE);
891 __remove_dirty_segment(sbi, segno, DIRTY);
de881df9 892 } else if (valid_blocks < usable_blocks) {
351df4b2
JK		 893 __locate_dirty_segment(sbi, segno, DIRTY);
894 } else {
895 /* Recovery routine with SSR needs this */
896 __remove_dirty_segment(sbi, segno, DIRTY);
897 }
898
899 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 900}
901
4354994f
DR		 902/* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
903void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
904{
905 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
906 unsigned int segno;
907
908 mutex_lock(&dirty_i->seglist_lock);
909 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
910 if (get_valid_blocks(sbi, segno, false))
911 continue;
912 if (IS_CURSEG(sbi, segno))
913 continue;
914 __locate_dirty_segment(sbi, segno, PRE);
915 __remove_dirty_segment(sbi, segno, DIRTY);
916 }
917 mutex_unlock(&dirty_i->seglist_lock);
918}
919
4d3aed70 920block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
4354994f 921{
ae4ad7ea
DR		 922 int ovp_hole_segs =
923 (overprovision_segments(sbi) - reserved_segments(sbi));
924 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
4d3aed70 925 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4354994f 926 block_t holes[2] = {0, 0}; /* DATA and NODE */
4d3aed70 927 block_t unusable;
4354994f
DR		 928 struct seg_entry *se;
929 unsigned int segno;
930
931 mutex_lock(&dirty_i->seglist_lock);
932 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
933 se = get_seg_entry(sbi, segno);
934 if (IS_NODESEG(se->type))
de881df9
AR		 935 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
936 se->valid_blocks;
4354994f 937 else
de881df9
AR		 938 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
939 se->valid_blocks;
4354994f
DR		 940 }
941 mutex_unlock(&dirty_i->seglist_lock);
942
4d3aed70
DR		 943 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
944 if (unusable > ovp_holes)
945 return unusable - ovp_holes;
946 return 0;
947}
948
949int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
950{
951 int ovp_hole_segs =
952 (overprovision_segments(sbi) - reserved_segments(sbi));
953 if (unusable > F2FS_OPTION(sbi).unusable_cap)
4354994f 954 return -EAGAIN;
db610a64 955 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
ae4ad7ea 956 dirty_segments(sbi) > ovp_hole_segs)
db610a64 957 return -EAGAIN;
4354994f
DR		 958 return 0;
959}
960
961/* This is only used by SBI_CP_DISABLED */
962static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
963{
964 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
965 unsigned int segno = 0;
966
967 mutex_lock(&dirty_i->seglist_lock);
968 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
969 if (get_valid_blocks(sbi, segno, false))
970 continue;
61461fc9 971 if (get_ckpt_valid_blocks(sbi, segno, false))
4354994f
DR		 972 continue;
973 mutex_unlock(&dirty_i->seglist_lock);
974 return segno;
975 }
976 mutex_unlock(&dirty_i->seglist_lock);
977 return NULL_SEGNO;
978}
979
004b6862 980static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 981 struct block_device *bdev, block_t lstart,
982 block_t start, block_t len)
275b66b0 983{
0b54fb84 984 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 985 struct list_head *pend_list;
b01a9201 986 struct discard_cmd *dc;
275b66b0 987
ba48a33e
CY		 988 f2fs_bug_on(sbi, !len);
989
990 pend_list = &dcc->pend_list[plist_idx(len)];
991
b01a9201
JK		 992 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
993 INIT_LIST_HEAD(&dc->list);
c81abe34 994 dc->bdev = bdev;
b01a9201 995 dc->lstart = lstart;
c81abe34 996 dc->start = start;
b01a9201 997 dc->len = len;
ec9895ad 998 dc->ref = 0;
15469963 999 dc->state = D_PREP;
72691af6 1000 dc->queued = 0;
c81abe34 1001 dc->error = 0;
b01a9201 1002 init_completion(&dc->wait);
22d375dd 1003 list_add_tail(&dc->list, pend_list);
35ec7d57
CY		 1004 spin_lock_init(&dc->lock);
1005 dc->bio_ref = 0;
5f32366a 1006 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 1007 dcc->undiscard_blks += len;
004b6862
CY		 1008
1009 return dc;
1010}
1011
1012static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
1013 struct block_device *bdev, block_t lstart,
1014 block_t start, block_t len,
4dada3fd
CY		 1015 struct rb_node *parent, struct rb_node **p,
1016 bool leftmost)
004b6862
CY		 1017{
1018 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1019 struct discard_cmd *dc;
1020
1021 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1022
1023 rb_link_node(&dc->rb_node, parent, p);
4dada3fd 1024 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
004b6862
CY		 1025
1026 return dc;
15469963
JK		 1027}
1028
004b6862
CY		 1029static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1030 struct discard_cmd *dc)
15469963 1031{
dcc9165d 1032 if (dc->state == D_DONE)
72691af6 1033 atomic_sub(dc->queued, &dcc->queued_discard);
004b6862
CY		 1034
1035 list_del(&dc->list);
4dada3fd 1036 rb_erase_cached(&dc->rb_node, &dcc->root);
d84d1cbd 1037 dcc->undiscard_blks -= dc->len;
004b6862
CY		 1038
1039 kmem_cache_free(discard_cmd_slab, dc);
1040
1041 atomic_dec(&dcc->discard_cmd_cnt);
1042}
1043
1044static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1045 struct discard_cmd *dc)
1046{
1047 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
35ec7d57 1048 unsigned long flags;
dcc9165d 1049
2ec6f2ef
CY		 1050 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1051
35ec7d57
CY		 1052 spin_lock_irqsave(&dc->lock, flags);
1053 if (dc->bio_ref) {
1054 spin_unlock_irqrestore(&dc->lock, flags);
1055 return;
1056 }
1057 spin_unlock_irqrestore(&dc->lock, flags);
1058
d9703d90
CY		 1059 f2fs_bug_on(sbi, dc->ref);
1060
c81abe34
JK		 1061 if (dc->error == -EOPNOTSUPP)
1062 dc->error = 0;
15469963 1063
c81abe34 1064 if (dc->error)
22d7ea13 1065 printk_ratelimited(
c45d6002
CY		 1066 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1067 KERN_INFO, sbi->sb->s_id,
1068 dc->lstart, dc->start, dc->len, dc->error);
004b6862 1069 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 1070}
1071
c81abe34
JK		 1072static void f2fs_submit_discard_endio(struct bio *bio)
1073{
1074 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
35ec7d57 1075 unsigned long flags;
c81abe34 1076
35ec7d57 1077 spin_lock_irqsave(&dc->lock, flags);
3fa6a8c5
ST		 1078 if (!dc->error)
1079 dc->error = blk_status_to_errno(bio->bi_status);
35ec7d57
CY		 1080 dc->bio_ref--;
1081 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1082 dc->state = D_DONE;
1083 complete_all(&dc->wait);
1084 }
1085 spin_unlock_irqrestore(&dc->lock, flags);
c81abe34
JK		 1086 bio_put(bio);
1087}
1088
94b1e10e 1089static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 1090 block_t start, block_t end)
1091{
1092#ifdef CONFIG_F2FS_CHECK_FS
1093 struct seg_entry *sentry;
1094 unsigned int segno;
1095 block_t blk = start;
1096 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1097 unsigned long *map;
1098
1099 while (blk < end) {
1100 segno = GET_SEGNO(sbi, blk);
1101 sentry = get_seg_entry(sbi, segno);
1102 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1103
008396e1
YS		 1104 if (end < START_BLOCK(sbi, segno + 1))
1105 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1106 else
1107 size = max_blocks;
6915ea9d
CY		 1108 map = (unsigned long *)(sentry->cur_valid_map);
1109 offset = __find_rev_next_bit(map, size, offset);
1110 f2fs_bug_on(sbi, offset != size);
008396e1 1111 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 1112 }
1113#endif
1114}
1115
8bb4f253
JK		 1116static void __init_discard_policy(struct f2fs_sb_info *sbi,
1117 struct discard_policy *dpolicy,
1118 int discard_type, unsigned int granularity)
1119{
c35b8d5e
ST		 1120 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1121
8bb4f253
JK		 1122 /* common policy */
1123 dpolicy->type = discard_type;
1124 dpolicy->sync = true;
20ee4382 1125 dpolicy->ordered = false;
8bb4f253
JK		 1126 dpolicy->granularity = granularity;
1127
1128 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1129 dpolicy->io_aware_gran = MAX_PLIST_NUM;
6ce48b0c 1130 dpolicy->timeout = false;
8bb4f253
JK		 1131
1132 if (discard_type == DPOLICY_BG) {
1133 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1134 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1135 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1136 dpolicy->io_aware = true;
cba60849 1137 dpolicy->sync = false;
20ee4382 1138 dpolicy->ordered = true;
8bb4f253
JK		 1139 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1140 dpolicy->granularity = 1;
c35b8d5e
ST		 1141 if (atomic_read(&dcc->discard_cmd_cnt))
1142 dpolicy->max_interval =
1143 DEF_MIN_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1144 }
1145 } else if (discard_type == DPOLICY_FORCE) {
1146 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1147 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1148 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1149 dpolicy->io_aware = false;
1150 } else if (discard_type == DPOLICY_FSTRIM) {
1151 dpolicy->io_aware = false;
1152 } else if (discard_type == DPOLICY_UMOUNT) {
1153 dpolicy->io_aware = false;
b8623253
JK		 1154 /* we need to issue all to keep CP_TRIMMED_FLAG */
1155 dpolicy->granularity = 1;
6ce48b0c 1156 dpolicy->timeout = true;
8bb4f253
JK		 1157 }
1158}
1159
35ec7d57
CY		 1160static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1161 struct block_device *bdev, block_t lstart,
1162 block_t start, block_t len);
c81abe34 1163/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
6b9cb124 1164static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1165 struct discard_policy *dpolicy,
35ec7d57
CY		 1166 struct discard_cmd *dc,
1167 unsigned int *issued)
c81abe34 1168{
35ec7d57
CY		 1169 struct block_device *bdev = dc->bdev;
1170 struct request_queue *q = bdev_get_queue(bdev);
1171 unsigned int max_discard_blocks =
1172 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
c81abe34 1173 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1174 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1175 &(dcc->fstrim_list) : &(dcc->wait_list);
78997b56 1176 int flag = dpolicy->sync ? REQ_SYNC : 0;
35ec7d57
CY		 1177 block_t lstart, start, len, total_len;
1178 int err = 0;
c81abe34
JK		 1179
1180 if (dc->state != D_PREP)
6b9cb124 1181 return 0;
c81abe34 1182
d6184774 1183 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
6b9cb124 1184 return 0;
d6184774 1185
35ec7d57
CY		 1186 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1187
1188 lstart = dc->lstart;
1189 start = dc->start;
1190 len = dc->len;
1191 total_len = len;
1192
1193 dc->len = 0;
1194
1195 while (total_len && *issued < dpolicy->max_requests && !err) {
1196 struct bio *bio = NULL;
1197 unsigned long flags;
1198 bool last = true;
1199
1200 if (len > max_discard_blocks) {
1201 len = max_discard_blocks;
1202 last = false;
1203 }
1204
1205 (*issued)++;
1206 if (*issued == dpolicy->max_requests)
1207 last = true;
1208
1209 dc->len += len;
1210
b83dcfe6 1211 if (time_to_inject(sbi, FAULT_DISCARD)) {
c45d6002 1212 f2fs_show_injection_info(sbi, FAULT_DISCARD);
b83dcfe6
CY		 1213 err = -EIO;
1214 goto submit;
1215 }
35ec7d57
CY		 1216 err = __blkdev_issue_discard(bdev,
1217 SECTOR_FROM_BLOCK(start),
1218 SECTOR_FROM_BLOCK(len),
1219 GFP_NOFS, 0, &bio);
b83dcfe6 1220submit:
6b9cb124 1221 if (err) {
35ec7d57 1222 spin_lock_irqsave(&dc->lock, flags);
6b9cb124 1223 if (dc->state == D_PARTIAL)
35ec7d57 1224 dc->state = D_SUBMIT;
35ec7d57
CY		 1225 spin_unlock_irqrestore(&dc->lock, flags);
1226
6b9cb124
CY		 1227 break;
1228 }
35ec7d57 1229
6b9cb124 1230 f2fs_bug_on(sbi, !bio);
35ec7d57 1231
6b9cb124
CY		 1232 /*
1233 * should keep before submission to avoid D_DONE
1234 * right away
1235 */
1236 spin_lock_irqsave(&dc->lock, flags);
1237 if (last)
1238 dc->state = D_SUBMIT;
1239 else
1240 dc->state = D_PARTIAL;
1241 dc->bio_ref++;
1242 spin_unlock_irqrestore(&dc->lock, flags);
35ec7d57 1243
72691af6
JK		 1244 atomic_inc(&dcc->queued_discard);
1245 dc->queued++;
6b9cb124 1246 list_move_tail(&dc->list, wait_list);
b0af6d49 1247
6b9cb124 1248 /* sanity check on discard range */
9249dded 1249 __check_sit_bitmap(sbi, lstart, lstart + len);
35ec7d57 1250
6b9cb124
CY		 1251 bio->bi_private = dc;
1252 bio->bi_end_io = f2fs_submit_discard_endio;
1253 bio->bi_opf |= flag;
1254 submit_bio(bio);
1255
1256 atomic_inc(&dcc->issued_discard);
1257
1258 f2fs_update_iostat(sbi, FS_DISCARD, 1);
35ec7d57
CY		 1259
1260 lstart += len;
1261 start += len;
1262 total_len -= len;
1263 len = total_len;
c81abe34 1264 }
35ec7d57 1265
df423399
ST		 1266 if (!err && len) {
1267 dcc->undiscard_blks -= len;
35ec7d57 1268 __update_discard_tree_range(sbi, bdev, lstart, start, len);
df423399 1269 }
6b9cb124 1270 return err;
c81abe34
JK		 1271}
1272
47d0d7d7 1273static void __insert_discard_tree(struct f2fs_sb_info *sbi,
004b6862
CY		 1274 struct block_device *bdev, block_t lstart,
1275 block_t start, block_t len,
1276 struct rb_node **insert_p,
1277 struct rb_node *insert_parent)
c81abe34 1278{
004b6862 1279 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1280 struct rb_node **p;
004b6862 1281 struct rb_node *parent = NULL;
4dada3fd 1282 bool leftmost = true;
004b6862
CY		 1283
1284 if (insert_p && insert_parent) {
1285 parent = insert_parent;
1286 p = insert_p;
1287 goto do_insert;
1288 }
c81abe34 1289
4dada3fd
CY		 1290 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1291 lstart, &leftmost);
004b6862 1292do_insert:
47d0d7d7 1293 __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
4dada3fd 1294 p, leftmost);
c81abe34
JK		 1295}
1296
ba48a33e
CY		 1297static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1298 struct discard_cmd *dc)
1299{
1300 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1301}
1302
3d6a650f
YH		 1303static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1304 struct discard_cmd *dc, block_t blkaddr)
1305{
ba48a33e 1306 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1307 struct discard_info di = dc->di;
1308 bool modified = false;
3d6a650f 1309
004b6862 1310 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1311 __remove_discard_cmd(sbi, dc);
1312 return;
1313 }
1314
d84d1cbd
CY		 1315 dcc->undiscard_blks -= di.len;
1316
004b6862 1317 if (blkaddr > di.lstart) {
3d6a650f 1318 dc->len = blkaddr - dc->lstart;
d84d1cbd 1319 dcc->undiscard_blks += dc->len;
ba48a33e 1320 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1321 modified = true;
1322 }
1323
1324 if (blkaddr < di.lstart + di.len - 1) {
1325 if (modified) {
1326 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1327 di.start + blkaddr + 1 - di.lstart,
1328 di.lstart + di.len - 1 - blkaddr,
1329 NULL, NULL);
1330 } else {
1331 dc->lstart++;
1332 dc->len--;
1333 dc->start++;
d84d1cbd 1334 dcc->undiscard_blks += dc->len;
ba48a33e 1335 __relocate_discard_cmd(dcc, dc);
004b6862 1336 }
3d6a650f
YH		 1337 }
1338}
1339
004b6862
CY		 1340static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1341 struct block_device *bdev, block_t lstart,
1342 block_t start, block_t len)
275b66b0 1343{
0b54fb84 1344 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1345 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1346 struct discard_cmd *dc;
1347 struct discard_info di = {0};
1348 struct rb_node **insert_p = NULL, *insert_parent = NULL;
35ec7d57
CY		 1349 struct request_queue *q = bdev_get_queue(bdev);
1350 unsigned int max_discard_blocks =
1351 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
004b6862 1352 block_t end = lstart + len;
275b66b0 1353
4d57b86d 1354 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
004b6862
CY		 1355 NULL, lstart,
1356 (struct rb_entry **)&prev_dc,
1357 (struct rb_entry **)&next_dc,
4dada3fd 1358 &insert_p, &insert_parent, true, NULL);
004b6862
CY		 1359 if (dc)
1360 prev_dc = dc;
1361
1362 if (!prev_dc) {
1363 di.lstart = lstart;
1364 di.len = next_dc ? next_dc->lstart - lstart : len;
1365 di.len = min(di.len, len);
1366 di.start = start;
22d375dd 1367 }
15469963 1368
004b6862
CY		 1369 while (1) {
1370 struct rb_node *node;
1371 bool merged = false;
1372 struct discard_cmd *tdc = NULL;
1373
1374 if (prev_dc) {
1375 di.lstart = prev_dc->lstart + prev_dc->len;
1376 if (di.lstart < lstart)
1377 di.lstart = lstart;
1378 if (di.lstart >= end)
1379 break;
1380
1381 if (!next_dc || next_dc->lstart > end)
1382 di.len = end - di.lstart;
1383 else
1384 di.len = next_dc->lstart - di.lstart;
1385 di.start = start + di.lstart - lstart;
1386 }
1387
1388 if (!di.len)
1389 goto next;
1390
1391 if (prev_dc && prev_dc->state == D_PREP &&
1392 prev_dc->bdev == bdev &&
35ec7d57
CY		 1393 __is_discard_back_mergeable(&di, &prev_dc->di,
1394 max_discard_blocks)) {
004b6862 1395 prev_dc->di.len += di.len;
d84d1cbd 1396 dcc->undiscard_blks += di.len;
ba48a33e 1397 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1398 di = prev_dc->di;
1399 tdc = prev_dc;
1400 merged = true;
1401 }
1402
1403 if (next_dc && next_dc->state == D_PREP &&
1404 next_dc->bdev == bdev &&
35ec7d57
CY		 1405 __is_discard_front_mergeable(&di, &next_dc->di,
1406 max_discard_blocks)) {
004b6862
CY		 1407 next_dc->di.lstart = di.lstart;
1408 next_dc->di.len += di.len;
1409 next_dc->di.start = di.start;
d84d1cbd 1410 dcc->undiscard_blks += di.len;
ba48a33e 1411 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1412 if (tdc)
1413 __remove_discard_cmd(sbi, tdc);
004b6862 1414 merged = true;
4e6a8d9b 1415 }
004b6862 1416
df0f6b44 1417 if (!merged) {
004b6862
CY		 1418 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1419 di.len, NULL, NULL);
df0f6b44 1420 }
004b6862
CY		 1421 next:
1422 prev_dc = next_dc;
1423 if (!prev_dc)
1424 break;
1425
1426 node = rb_next(&prev_dc->rb_node);
1427 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1428 }
004b6862
CY		 1429}
1430
1431static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1432 struct block_device *bdev, block_t blkstart, block_t blklen)
1433{
1434 block_t lblkstart = blkstart;
1435
7f3d7719
DLM		 1436 if (!f2fs_bdev_support_discard(bdev))
1437 return 0;
1438
0243a5f9 1439 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862 1440
0916878d 1441 if (f2fs_is_multi_device(sbi)) {
004b6862
CY		 1442 int devi = f2fs_target_device_index(sbi, blkstart);
1443
1444 blkstart -= FDEV(devi).start_blk;
1445 }
35ec7d57 1446 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862 1447 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
35ec7d57 1448 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862
CY		 1449 return 0;
1450}
1451
20ee4382
CY		 1452static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1453 struct discard_policy *dpolicy)
1454{
1455 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1456 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1457 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1458 struct discard_cmd *dc;
1459 struct blk_plug plug;
1460 unsigned int pos = dcc->next_pos;
1461 unsigned int issued = 0;
1462 bool io_interrupted = false;
1463
1464 mutex_lock(&dcc->cmd_lock);
1465 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1466 NULL, pos,
1467 (struct rb_entry **)&prev_dc,
1468 (struct rb_entry **)&next_dc,
4dada3fd 1469 &insert_p, &insert_parent, true, NULL);
20ee4382
CY		 1470 if (!dc)
1471 dc = next_dc;
1472
1473 blk_start_plug(&plug);
1474
1475 while (dc) {
1476 struct rb_node *node;
6b9cb124 1477 int err = 0;
20ee4382
CY		 1478
1479 if (dc->state != D_PREP)
1480 goto next;
1481
a7d10cf3 1482 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
20ee4382
CY		 1483 io_interrupted = true;
1484 break;
1485 }
1486
1487 dcc->next_pos = dc->lstart + dc->len;
6b9cb124 1488 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
20ee4382 1489
35ec7d57 1490 if (issued >= dpolicy->max_requests)
20ee4382
CY		 1491 break;
1492next:
1493 node = rb_next(&dc->rb_node);
6b9cb124
CY		 1494 if (err)
1495 __remove_discard_cmd(sbi, dc);
20ee4382
CY		 1496 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1497 }
1498
1499 blk_finish_plug(&plug);
1500
1501 if (!dc)
1502 dcc->next_pos = 0;
1503
1504 mutex_unlock(&dcc->cmd_lock);
1505
1506 if (!issued && io_interrupted)
1507 issued = -1;
1508
1509 return issued;
1510}
141af6ba
ST		 1511static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1512 struct discard_policy *dpolicy);
20ee4382 1513
78997b56
CY		 1514static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1515 struct discard_policy *dpolicy)
bd5b0738
CY		 1516{
1517 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1518 struct list_head *pend_list;
1519 struct discard_cmd *dc, *tmp;
1520 struct blk_plug plug;
141af6ba 1521 int i, issued;
e6c6de18 1522 bool io_interrupted = false;
bd5b0738 1523
6ce48b0c
CY		 1524 if (dpolicy->timeout)
1525 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
03f2c02d 1526
141af6ba
ST		 1527retry:
1528 issued = 0;
78997b56 1529 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
6ce48b0c
CY		 1530 if (dpolicy->timeout &&
1531 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
03f2c02d
JK		 1532 break;
1533
78997b56
CY		 1534 if (i + 1 < dpolicy->granularity)
1535 break;
20ee4382
CY		 1536
1537 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1538 return __issue_discard_cmd_orderly(sbi, dpolicy);
1539
bd5b0738 1540 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1541
1542 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1543 if (list_empty(pend_list))
1544 goto next;
67fce70b
CY		 1545 if (unlikely(dcc->rbtree_check))
1546 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2e9b2bb2 1547 &dcc->root, false));
33da62cf 1548 blk_start_plug(&plug);
bd5b0738
CY		 1549 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1550 f2fs_bug_on(sbi, dc->state != D_PREP);
1551
6ce48b0c
CY		 1552 if (dpolicy->timeout &&
1553 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
6e0cd4a9
HX		 1554 break;
1555
ecc9aa00 1556 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
a7d10cf3 1557 !is_idle(sbi, DISCARD_TIME)) {
e6c6de18 1558 io_interrupted = true;
522d1711 1559 break;
969d1b18 1560 }
e6c6de18 1561
35ec7d57 1562 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
522d1711 1563
35ec7d57 1564 if (issued >= dpolicy->max_requests)
33da62cf 1565 break;
bd5b0738 1566 }
33da62cf 1567 blk_finish_plug(&plug);
49c60c67 1568next:
33da62cf
CY		 1569 mutex_unlock(&dcc->cmd_lock);
1570
522d1711 1571 if (issued >= dpolicy->max_requests || io_interrupted)
33da62cf 1572 break;
bd5b0738 1573 }
969d1b18 1574
141af6ba
ST		 1575 if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1576 __wait_all_discard_cmd(sbi, dpolicy);
1577 goto retry;
1578 }
1579
e6c6de18
CY		 1580 if (!issued && io_interrupted)
1581 issued = -1;
1582
969d1b18
CY		 1583 return issued;
1584}
1585
cf5c759f 1586static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1587{
1588 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1589 struct list_head *pend_list;
1590 struct discard_cmd *dc, *tmp;
1591 int i;
cf5c759f 1592 bool dropped = false;
969d1b18
CY		 1593
1594 mutex_lock(&dcc->cmd_lock);
1595 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1596 pend_list = &dcc->pend_list[i];
1597 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1598 f2fs_bug_on(sbi, dc->state != D_PREP);
1599 __remove_discard_cmd(sbi, dc);
cf5c759f 1600 dropped = true;
969d1b18
CY		 1601 }
1602 }
1603 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1604
1605 return dropped;
bd5b0738
CY		 1606}
1607
4d57b86d 1608void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
7950e9ac
CY		 1609{
1610 __drop_discard_cmd(sbi);
1611}
1612
0ea80512 1613static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1614 struct discard_cmd *dc)
1615{
1616 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1617 unsigned int len = 0;
2a510c00
CY		 1618
1619 wait_for_completion_io(&dc->wait);
1620 mutex_lock(&dcc->cmd_lock);
1621 f2fs_bug_on(sbi, dc->state != D_DONE);
1622 dc->ref--;
0ea80512
CY		 1623 if (!dc->ref) {
1624 if (!dc->error)
1625 len = dc->len;
2a510c00 1626 __remove_discard_cmd(sbi, dc);
0ea80512 1627 }
2a510c00 1628 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1629
1630 return len;
2a510c00
CY		 1631}
1632
0ea80512 1633static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1634 struct discard_policy *dpolicy,
1635 block_t start, block_t end)
63a94fa1
CY		 1636{
1637 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1638 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1639 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1640 struct discard_cmd *dc, *tmp;
6afae633 1641 bool need_wait;
0ea80512 1642 unsigned int trimmed = 0;
6afae633
CY		 1643
1644next:
1645 need_wait = false;
63a94fa1
CY		 1646
1647 mutex_lock(&dcc->cmd_lock);
1648 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1649 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1650 continue;
78997b56 1651 if (dc->len < dpolicy->granularity)
8412663d 1652 continue;
78997b56 1653 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1654 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1655 if (!dc->error)
1656 trimmed += dc->len;
63a94fa1 1657 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1658 } else {
1659 dc->ref++;
1660 need_wait = true;
1661 break;
63a94fa1
CY		 1662 }
1663 }
1664 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1665
1666 if (need_wait) {
0ea80512 1667 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1668 goto next;
1669 }
0ea80512
CY		 1670
1671 return trimmed;
63a94fa1
CY		 1672}
1673
01f9cf6d 1674static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1675 struct discard_policy *dpolicy)
8412663d 1676{
9a997188 1677 struct discard_policy dp;
01f9cf6d 1678 unsigned int discard_blks;
9a997188 1679
01f9cf6d
CY		 1680 if (dpolicy)
1681 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
9a997188
JK		 1682
1683 /* wait all */
8bb4f253 1684 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
01f9cf6d 1685 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1686 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
01f9cf6d
CY		 1687 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1688
1689 return discard_blks;
8412663d
CY		 1690}
1691
004b6862 1692/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1693static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1694{
1695 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1696 struct discard_cmd *dc;
ec9895ad 1697 bool need_wait = false;
004b6862
CY		 1698
1699 mutex_lock(&dcc->cmd_lock);
4d57b86d
CY		 1700 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1701 NULL, blkaddr);
004b6862 1702 if (dc) {
ec9895ad
CY		 1703 if (dc->state == D_PREP) {
1704 __punch_discard_cmd(sbi, dc, blkaddr);
1705 } else {
1706 dc->ref++;
1707 need_wait = true;
1708 }
275b66b0 1709 }
d431413f 1710 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1711
2a510c00
CY		 1712 if (need_wait)
1713 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1714}
1715
4d57b86d 1716void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
cce13252
CY		 1717{
1718 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1719
1720 if (dcc && dcc->f2fs_issue_discard) {
1721 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1722
1723 dcc->f2fs_issue_discard = NULL;
1724 kthread_stop(discard_thread);
ec9895ad 1725 }
d431413f
CY		 1726}
1727
8412663d 1728/* This comes from f2fs_put_super */
03f2c02d 1729bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
969d1b18
CY		 1730{
1731 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1732 struct discard_policy dpolicy;
cf5c759f 1733 bool dropped;
969d1b18 1734
8bb4f253
JK		 1735 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1736 dcc->discard_granularity);
78997b56 1737 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1738 dropped = __drop_discard_cmd(sbi);
cf5c759f 1739
9a997188
JK		 1740 /* just to make sure there is no pending discard commands */
1741 __wait_all_discard_cmd(sbi, NULL);
2482c432
CY		 1742
1743 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
cf5c759f 1744 return dropped;
969d1b18
CY		 1745}
1746
15469963
JK		 1747static int issue_discard_thread(void *data)
1748{
1749 struct f2fs_sb_info *sbi = data;
1750 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1751 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1752 struct discard_policy dpolicy;
969d1b18
CY		 1753 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1754 int issued;
15469963 1755
1d7be270 1756 set_freezable();
15469963 1757
1d7be270 1758 do {
c35b8d5e
ST		 1759 if (sbi->gc_mode == GC_URGENT_HIGH ||
1760 !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1761 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1762 else
1763 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1764 dcc->discard_granularity);
1765
1766 if (!atomic_read(&dcc->discard_cmd_cnt))
1767 wait_ms = dpolicy.max_interval;
78997b56 1768
969d1b18
CY		 1769 wait_event_interruptible_timeout(*q,
1770 kthread_should_stop() || freezing(current) ||
1771 dcc->discard_wake,
1772 msecs_to_jiffies(wait_ms));
35a9a766
SY		 1773
1774 if (dcc->discard_wake)
1775 dcc->discard_wake = 0;
1776
76c7bfb3
JK		 1777 /* clean up pending candidates before going to sleep */
1778 if (atomic_read(&dcc->queued_discard))
1779 __wait_all_discard_cmd(sbi, NULL);
1780
1d7be270
JK		 1781 if (try_to_freeze())
1782 continue;
3b60d802
CY		 1783 if (f2fs_readonly(sbi->sb))
1784 continue;
1d7be270
JK		 1785 if (kthread_should_stop())
1786 return 0;
d6184774
YH		 1787 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1788 wait_ms = dpolicy.max_interval;
1789 continue;
1790 }
43f8c47e
CY		 1791 if (!atomic_read(&dcc->discard_cmd_cnt))
1792 continue;
969d1b18 1793
dc6febb6
CY		 1794 sb_start_intwrite(sbi->sb);
1795
78997b56 1796 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1797 if (issued > 0) {
78997b56
CY		 1798 __wait_all_discard_cmd(sbi, &dpolicy);
1799 wait_ms = dpolicy.min_interval;
5f029c04 1800 } else if (issued == -1) {
a7d10cf3
ST		 1801 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1802 if (!wait_ms)
abde73c7 1803 wait_ms = dpolicy.mid_interval;
969d1b18 1804 } else {
78997b56 1805 wait_ms = dpolicy.max_interval;
969d1b18 1806 }
1d7be270 1807
dc6febb6 1808 sb_end_intwrite(sbi->sb);
1d7be270 1809
1d7be270
JK		 1810 } while (!kthread_should_stop());
1811 return 0;
15469963
JK		 1812}
1813
f46e8809 1814#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1815static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1816 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1817{
92592285 1818 sector_t sector, nr_sects;
10a875f8 1819 block_t lblkstart = blkstart;
3c62be17
JK		 1820 int devi = 0;
1821
0916878d 1822 if (f2fs_is_multi_device(sbi)) {
3c62be17 1823 devi = f2fs_target_device_index(sbi, blkstart);
95175daf
DLM		 1824 if (blkstart < FDEV(devi).start_blk ||
1825 blkstart > FDEV(devi).end_blk) {
dcbb4c10 1826 f2fs_err(sbi, "Invalid block %x", blkstart);
95175daf
DLM		 1827 return -EIO;
1828 }
3c62be17
JK		 1829 blkstart -= FDEV(devi).start_blk;
1830 }
f46e8809 1831
95175daf
DLM		 1832 /* For sequential zones, reset the zone write pointer */
1833 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
92592285
JK		 1834 sector = SECTOR_FROM_BLOCK(blkstart);
1835 nr_sects = SECTOR_FROM_BLOCK(blklen);
1836
1837 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1838 nr_sects != bdev_zone_sectors(bdev)) {
dcbb4c10
JP		 1839 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1840 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1841 blkstart, blklen);
92592285
JK		 1842 return -EIO;
1843 }
d50aaeec 1844 trace_f2fs_issue_reset_zone(bdev, blkstart);
6c1b1da5
AJ		 1845 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1846 sector, nr_sects, GFP_NOFS);
f46e8809 1847 }
95175daf
DLM		 1848
1849 /* For conventional zones, use regular discard if supported */
95175daf 1850 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1851}
1852#endif
1853
3c62be17
JK		 1854static int __issue_discard_async(struct f2fs_sb_info *sbi,
1855 struct block_device *bdev, block_t blkstart, block_t blklen)
1856{
1857#ifdef CONFIG_BLK_DEV_ZONED
7f3d7719 1858 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
3c62be17
JK		 1859 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1860#endif
c81abe34 1861 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1862}
1863
1e87a78d 1864static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1865 block_t blkstart, block_t blklen)
1866{
3c62be17
JK		 1867 sector_t start = blkstart, len = 0;
1868 struct block_device *bdev;
a66cdd98
JK		 1869 struct seg_entry *se;
1870 unsigned int offset;
1871 block_t i;
3c62be17
JK		 1872 int err = 0;
1873
1874 bdev = f2fs_target_device(sbi, blkstart, NULL);
1875
1876 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1877 if (i != start) {
1878 struct block_device *bdev2 =
1879 f2fs_target_device(sbi, i, NULL);
1880
1881 if (bdev2 != bdev) {
1882 err = __issue_discard_async(sbi, bdev,
1883 start, len);
1884 if (err)
1885 return err;
1886 bdev = bdev2;
1887 start = i;
1888 len = 0;
1889 }
1890 }
a66cdd98 1891
a66cdd98
JK		 1892 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1893 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1894
1895 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1896 sbi->discard_blks--;
1897 }
f46e8809 1898
3c62be17
JK		 1899 if (len)
1900 err = __issue_discard_async(sbi, bdev, start, len);
1901 return err;
1e87a78d
JK		 1902}
1903
25290fa5
JK		 1904static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1905 bool check_only)
adf4983b 1906{
b2955550
JK		 1907 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1908 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1909 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1910 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1911 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1912 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1913 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1914 unsigned int start = 0, end = -1;
c473f1a9 1915 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1916 struct discard_entry *de = NULL;
46f84c2c 1917 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1918 int i;
1919
7d20c8ab 1920 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
25290fa5 1921 return false;
b2955550 1922
a66cdd98 1923 if (!force) {
7d20c8ab 1924 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
0b54fb84
JK		 1925 SM_I(sbi)->dcc_info->nr_discards >=
1926 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1927 return false;
4b2fecc8
JK		 1928 }
1929
b2955550
JK		 1930 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1931 for (i = 0; i < entries; i++)
a66cdd98 1932 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1933 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1934
0b54fb84
JK		 1935 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1936 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1937 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1938 if (start >= max_blocks)
1939 break;
1940
1941 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1942 if (force && start && end != max_blocks
1943 && (end - start) < cpc->trim_minlen)
1944 continue;
1945
25290fa5
JK		 1946 if (check_only)
1947 return true;
1948
a7eeb823
CY		 1949 if (!de) {
1950 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1951 GFP_F2FS_ZERO);
1952 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1953 list_add_tail(&de->list, head);
1954 }
1955
1956 for (i = start; i < end; i++)
1957 __set_bit_le(i, (void *)de->discard_map);
1958
1959 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1960 }
25290fa5 1961 return false;
b2955550
JK		 1962}
1963
af8ff65b
CY		 1964static void release_discard_addr(struct discard_entry *entry)
1965{
1966 list_del(&entry->list);
1967 kmem_cache_free(discard_entry_slab, entry);
1968}
1969
4d57b86d 1970void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1971{
46f84c2c 1972 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1973 struct discard_entry *entry, *this;
1974
1975 /* drop caches */
af8ff65b
CY		 1976 list_for_each_entry_safe(entry, this, head, list)
1977 release_discard_addr(entry);
4b2fecc8
JK		 1978}
1979
0a8165d7 1980/*
4d57b86d 1981 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1982 */
1983static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1984{
1985 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1986 unsigned int segno;
351df4b2
JK		 1987
1988 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1989 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
d0b9e42a 1990 __set_test_and_free(sbi, segno, false);
351df4b2
JK		 1991 mutex_unlock(&dirty_i->seglist_lock);
1992}
1993
4d57b86d
CY		 1994void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1995 struct cp_control *cpc)
351df4b2 1996{
969d1b18
CY		 1997 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1998 struct list_head *head = &dcc->entry_list;
2d7b822a 1999 struct discard_entry *entry, *this;
351df4b2 2000 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 2001 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 2002 unsigned int start = 0, end = -1;
36abef4e 2003 unsigned int secno, start_segno;
c473f1a9 2004 bool force = (cpc->reason & CP_DISCARD);
b0332a0f 2005 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
351df4b2
JK		 2006
2007 mutex_lock(&dirty_i->seglist_lock);
29e59c14 2008
351df4b2 2009 while (1) {
29e59c14 2010 int i;
ad6672bb
YS		 2011
2012 if (need_align && end != -1)
2013 end--;
7cd8558b
JK		 2014 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2015 if (start >= MAIN_SEGS(sbi))
351df4b2 2016 break;
7cd8558b
JK		 2017 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2018 start + 1);
29e59c14 2019
ad6672bb
YS		 2020 if (need_align) {
2021 start = rounddown(start, sbi->segs_per_sec);
2022 end = roundup(end, sbi->segs_per_sec);
2023 }
29e59c14 2024
ad6672bb
YS		 2025 for (i = start; i < end; i++) {
2026 if (test_and_clear_bit(i, prefree_map))
2027 dirty_i->nr_dirty[PRE]--;
2028 }
29e59c14 2029
7d20c8ab 2030 if (!f2fs_realtime_discard_enable(sbi))
29e59c14 2031 continue;
351df4b2 2032
650d3c4e
YH		 2033 if (force && start >= cpc->trim_start &&
2034 (end - 1) <= cpc->trim_end)
2035 continue;
2036
b0332a0f 2037 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
36abef4e 2038 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 2039 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 2040 continue;
2041 }
2042next:
4ddb1a4d
JK		 2043 secno = GET_SEC_FROM_SEG(sbi, start);
2044 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 2045 if (!IS_CURSEC(sbi, secno) &&
302bd348 2046 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 2047 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2048 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2049
2050 start = start_segno + sbi->segs_per_sec;
2051 if (start < end)
2052 goto next;
8b107f5b
JK		 2053 else
2054 end = start - 1;
351df4b2
JK		 2055 }
2056 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 2057
2058 /* send small discards */
2d7b822a 2059 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 2060 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2061 bool is_valid = test_bit_le(0, entry->discard_map);
2062
2063find_next:
2064 if (is_valid) {
2065 next_pos = find_next_zero_bit_le(entry->discard_map,
2066 sbi->blocks_per_seg, cur_pos);
2067 len = next_pos - cur_pos;
2068
7beb01f7 2069 if (f2fs_sb_has_blkzoned(sbi) ||
acfd2810 2070 (force && len < cpc->trim_minlen))
a7eeb823
CY		 2071 goto skip;
2072
2073 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2074 len);
a7eeb823
CY		 2075 total_len += len;
2076 } else {
2077 next_pos = find_next_bit_le(entry->discard_map,
2078 sbi->blocks_per_seg, cur_pos);
2079 }
836b5a63 2080skip:
a7eeb823
CY		 2081 cur_pos = next_pos;
2082 is_valid = !is_valid;
2083
2084 if (cur_pos < sbi->blocks_per_seg)
2085 goto find_next;
2086
af8ff65b 2087 release_discard_addr(entry);
969d1b18 2088 dcc->nr_discards -= total_len;
b2955550 2089 }
34e159da 2090
01983c71 2091 wake_up_discard_thread(sbi, false);
351df4b2
JK		 2092}
2093
8ed59745 2094static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 2095{
15469963 2096 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 2097 struct discard_cmd_control *dcc;
ba48a33e 2098 int err = 0, i;
0b54fb84
JK		 2099
2100 if (SM_I(sbi)->dcc_info) {
2101 dcc = SM_I(sbi)->dcc_info;
2102 goto init_thread;
2103 }
2104
acbf054d 2105 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 2106 if (!dcc)
2107 return -ENOMEM;
2108
969d1b18 2109 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 2110 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 2111 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 2112 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 2113 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 2114 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 2115 mutex_init(&dcc->cmd_lock);
8b8dd65f 2116 atomic_set(&dcc->issued_discard, 0);
72691af6 2117 atomic_set(&dcc->queued_discard, 0);
5f32366a 2118 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 2119 dcc->nr_discards = 0;
d618ebaf 2120 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 2121 dcc->undiscard_blks = 0;
20ee4382 2122 dcc->next_pos = 0;
4dada3fd 2123 dcc->root = RB_ROOT_CACHED;
67fce70b 2124 dcc->rbtree_check = false;
0b54fb84 2125
15469963 2126 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 2127 SM_I(sbi)->dcc_info = dcc;
2128init_thread:
15469963
JK		 2129 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2130 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2131 if (IS_ERR(dcc->f2fs_issue_discard)) {
2132 err = PTR_ERR(dcc->f2fs_issue_discard);
c8eb7024 2133 kfree(dcc);
15469963
JK		 2134 SM_I(sbi)->dcc_info = NULL;
2135 return err;
2136 }
2137
0b54fb84
JK		 2138 return err;
2139}
2140
f099405f 2141static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 2142{
2143 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2144
f099405f
CY		 2145 if (!dcc)
2146 return;
2147
4d57b86d 2148 f2fs_stop_discard_thread(sbi);
f099405f 2149
04f9287a
CY		 2150 /*
2151 * Recovery can cache discard commands, so in error path of
2152 * fill_super(), it needs to give a chance to handle them.
2153 */
2154 if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2155 f2fs_issue_discard_timeout(sbi);
2156
c8eb7024 2157 kfree(dcc);
f099405f 2158 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 2159}
2160
184a5cd2 2161static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 2162{
2163 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 2164
2165 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 2166 sit_i->dirty_sentries++;
184a5cd2
CY		 2167 return false;
2168 }
2169
2170 return true;
351df4b2
JK		 2171}
2172
2173static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2174 unsigned int segno, int modified)
2175{
2176 struct seg_entry *se = get_seg_entry(sbi, segno);
5f029c04 2177
351df4b2
JK		 2178 se->type = type;
2179 if (modified)
2180 __mark_sit_entry_dirty(sbi, segno);
2181}
2182
c5d02785
CY		 2183static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2184 block_t blkaddr)
6f3a01ae
CY		 2185{
2186 unsigned int segno = GET_SEGNO(sbi, blkaddr);
c5d02785
CY		 2187
2188 if (segno == NULL_SEGNO)
2189 return 0;
2190 return get_seg_entry(sbi, segno)->mtime;
2191}
2192
2193static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2194 unsigned long long old_mtime)
2195{
2196 struct seg_entry *se;
2197 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2198 unsigned long long ctime = get_mtime(sbi, false);
2199 unsigned long long mtime = old_mtime ? old_mtime : ctime;
2200
2201 if (segno == NULL_SEGNO)
2202 return;
2203
2204 se = get_seg_entry(sbi, segno);
6f3a01ae
CY		 2205
2206 if (!se->mtime)
2207 se->mtime = mtime;
2208 else
2209 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2210 se->valid_blocks + 1);
2211
c5d02785
CY		 2212 if (ctime > SIT_I(sbi)->max_mtime)
2213 SIT_I(sbi)->max_mtime = ctime;
6f3a01ae
CY		 2214}
2215
351df4b2
JK		 2216static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2217{
2218 struct seg_entry *se;
2219 unsigned int segno, offset;
2220 long int new_vblocks;
6415fedc
YS		 2221 bool exist;
2222#ifdef CONFIG_F2FS_CHECK_FS
2223 bool mir_exist;
2224#endif
351df4b2
JK		 2225
2226 segno = GET_SEGNO(sbi, blkaddr);
2227
2228 se = get_seg_entry(sbi, segno);
2229 new_vblocks = se->valid_blocks + del;
491c0854 2230 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 2231
9feffe14 2232 f2fs_bug_on(sbi, (new_vblocks < 0 ||
de881df9 2233 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
351df4b2
JK		 2234
2235 se->valid_blocks = new_vblocks;
351df4b2
JK		 2236
2237 /* Update valid block bitmap */
2238 if (del > 0) {
6415fedc 2239 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 2240#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2241 mir_exist = f2fs_test_and_set_bit(offset,
2242 se->cur_valid_map_mir);
2243 if (unlikely(exist != mir_exist)) {
dcbb4c10
JP		 2244 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2245 blkaddr, exist);
05796763 2246 f2fs_bug_on(sbi, 1);
6415fedc 2247 }
355e7891 2248#endif
6415fedc 2249 if (unlikely(exist)) {
dcbb4c10
JP		 2250 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2251 blkaddr);
6415fedc 2252 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2253 se->valid_blocks--;
2254 del = 0;
355e7891 2255 }
6415fedc 2256
7d20c8ab 2257 if (!f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 2258 sbi->discard_blks--;
720037f9 2259
899fee36
CY		 2260 /*
2261 * SSR should never reuse block which is checkpointed
2262 * or newly invalidated.
2263 */
2264 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
720037f9
JK		 2265 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2266 se->ckpt_valid_blocks++;
2267 }
351df4b2 2268 } else {
6415fedc 2269 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 2270#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2271 mir_exist = f2fs_test_and_clear_bit(offset,
2272 se->cur_valid_map_mir);
2273 if (unlikely(exist != mir_exist)) {
dcbb4c10
JP		 2274 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2275 blkaddr, exist);
05796763 2276 f2fs_bug_on(sbi, 1);
6415fedc 2277 }
355e7891 2278#endif
6415fedc 2279 if (unlikely(!exist)) {
dcbb4c10
JP		 2280 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2281 blkaddr);
6415fedc 2282 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2283 se->valid_blocks++;
2284 del = 0;
4354994f
DR		 2285 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2286 /*
2287 * If checkpoints are off, we must not reuse data that
2288 * was used in the previous checkpoint. If it was used
2289 * before, we must track that to know how much space we
2290 * really have.
2291 */
c9c8ed50
CY		 2292 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2293 spin_lock(&sbi->stat_lock);
4354994f 2294 sbi->unusable_block_count++;
c9c8ed50
CY		 2295 spin_unlock(&sbi->stat_lock);
2296 }
355e7891 2297 }
6415fedc 2298
7d20c8ab 2299 if (f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 2300 sbi->discard_blks++;
351df4b2
JK		 2301 }
2302 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2303 se->ckpt_valid_blocks += del;
2304
2305 __mark_sit_entry_dirty(sbi, segno);
2306
2307 /* update total number of valid blocks to be written in ckpt area */
2308 SIT_I(sbi)->written_valid_blocks += del;
2309
2c70c5e3 2310 if (__is_large_section(sbi))
351df4b2
JK		 2311 get_sec_entry(sbi, segno)->valid_blocks += del;
2312}
2313
4d57b86d 2314void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 2315{
2316 unsigned int segno = GET_SEGNO(sbi, addr);
2317 struct sit_info *sit_i = SIT_I(sbi);
2318
9850cf4a 2319 f2fs_bug_on(sbi, addr == NULL_ADDR);
4c8ff709 2320 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
351df4b2
JK		 2321 return;
2322
6aa58d8a
CY		 2323 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2324
351df4b2 2325 /* add it into sit main buffer */
3d26fa6b 2326 down_write(&sit_i->sentry_lock);
351df4b2 2327
c5d02785 2328 update_segment_mtime(sbi, addr, 0);
351df4b2
JK		 2329 update_sit_entry(sbi, addr, -1);
2330
2331 /* add it into dirty seglist */
2332 locate_dirty_segment(sbi, segno);
2333
3d26fa6b 2334 up_write(&sit_i->sentry_lock);
351df4b2
JK		 2335}
2336
4d57b86d 2337bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 2338{
2339 struct sit_info *sit_i = SIT_I(sbi);
2340 unsigned int segno, offset;
2341 struct seg_entry *se;
2342 bool is_cp = false;
2343
93770ab7 2344 if (!__is_valid_data_blkaddr(blkaddr))
6e2c64ad
JK		 2345 return true;
2346
3d26fa6b 2347 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2348
2349 segno = GET_SEGNO(sbi, blkaddr);
2350 se = get_seg_entry(sbi, segno);
2351 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2352
2353 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2354 is_cp = true;
2355
3d26fa6b 2356 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2357
2358 return is_cp;
2359}
2360
0a8165d7 2361/*
351df4b2
JK		 2362 * This function should be resided under the curseg_mutex lock
2363 */
2364static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 2365 struct f2fs_summary *sum)
351df4b2
JK		 2366{
2367 struct curseg_info *curseg = CURSEG_I(sbi, type);
2368 void *addr = curseg->sum_blk;
5f029c04 2369
e79efe3b 2370 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 2371 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 2372}
2373
0a8165d7 2374/*
351df4b2
JK		 2375 * Calculate the number of current summary pages for writing
2376 */
4d57b86d 2377int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 2378{
351df4b2 2379 int valid_sum_count = 0;
9a47938b 2380 int i, sum_in_page;
351df4b2
JK		 2381
2382 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2383 if (sbi->ckpt->alloc_type[i] == SSR)
2384 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 2385 else {
2386 if (for_ra)
2387 valid_sum_count += le16_to_cpu(
2388 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2389 else
2390 valid_sum_count += curseg_blkoff(sbi, i);
2391 }
351df4b2
JK		 2392 }
2393
09cbfeaf 2394 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 2395 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2396 if (valid_sum_count <= sum_in_page)
351df4b2 2397 return 1;
9a47938b 2398 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 2399 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 2400 return 2;
2401 return 3;
2402}
2403
0a8165d7 2404/*
351df4b2
JK		 2405 * Caller should put this summary page
2406 */
4d57b86d 2407struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 2408{
86f33603
JK		 2409 if (unlikely(f2fs_cp_error(sbi)))
2410 return ERR_PTR(-EIO);
2411 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2412}
2413
4d57b86d
CY		 2414void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2415 void *src, block_t blk_addr)
351df4b2 2416{
4d57b86d 2417 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 2418
0537b811 2419 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2420 set_page_dirty(page);
2421 f2fs_put_page(page, 1);
2422}
2423
381722d2
CY		 2424static void write_sum_page(struct f2fs_sb_info *sbi,
2425 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2426{
4d57b86d 2427 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2428}
2429
b7ad7512
CY		 2430static void write_current_sum_page(struct f2fs_sb_info *sbi,
2431 int type, block_t blk_addr)
2432{
2433 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2434 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2435 struct f2fs_summary_block *src = curseg->sum_blk;
2436 struct f2fs_summary_block *dst;
2437
2438 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2439 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2440
2441 mutex_lock(&curseg->curseg_mutex);
2442
2443 down_read(&curseg->journal_rwsem);
2444 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2445 up_read(&curseg->journal_rwsem);
2446
2447 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2448 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2449
2450 mutex_unlock(&curseg->curseg_mutex);
2451
2452 set_page_dirty(page);
2453 f2fs_put_page(page, 1);
2454}
2455
093749e2
CY		 2456static int is_next_segment_free(struct f2fs_sb_info *sbi,
2457 struct curseg_info *curseg, int type)
a7881893 2458{
a7881893
JK		 2459 unsigned int segno = curseg->segno + 1;
2460 struct free_segmap_info *free_i = FREE_I(sbi);
2461
2462 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2463 return !test_bit(segno, free_i->free_segmap);
2464 return 0;
2465}
2466
0a8165d7 2467/*
351df4b2
JK		 2468 * Find a new segment from the free segments bitmap to right order
2469 * This function should be returned with success, otherwise BUG
2470 */
2471static void get_new_segment(struct f2fs_sb_info *sbi,
2472 unsigned int *newseg, bool new_sec, int dir)
2473{
2474 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2475 unsigned int segno, secno, zoneno;
7cd8558b 2476 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2477 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2478 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2479 unsigned int left_start = hint;
2480 bool init = true;
2481 int go_left = 0;
2482 int i;
2483
1a118ccf 2484 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2485
2486 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2487 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2488 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2489 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2490 goto got_it;
2491 }
2492find_other_zone:
7cd8558b
JK		 2493 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2494 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2495 if (dir == ALLOC_RIGHT) {
2496 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2497 MAIN_SECS(sbi), 0);
2498 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2499 } else {
2500 go_left = 1;
2501 left_start = hint - 1;
2502 }
2503 }
2504 if (go_left == 0)
2505 goto skip_left;
2506
2507 while (test_bit(left_start, free_i->free_secmap)) {
2508 if (left_start > 0) {
2509 left_start--;
2510 continue;
2511 }
2512 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2513 MAIN_SECS(sbi), 0);
2514 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2515 break;
2516 }
2517 secno = left_start;
2518skip_left:
4ddb1a4d
JK		 2519 segno = GET_SEG_FROM_SEC(sbi, secno);
2520 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2521
2522 /* give up on finding another zone */
2523 if (!init)
2524 goto got_it;
2525 if (sbi->secs_per_zone == 1)
2526 goto got_it;
2527 if (zoneno == old_zoneno)
2528 goto got_it;
2529 if (dir == ALLOC_LEFT) {
2530 if (!go_left && zoneno + 1 >= total_zones)
2531 goto got_it;
2532 if (go_left && zoneno == 0)
2533 goto got_it;
2534 }
2535 for (i = 0; i < NR_CURSEG_TYPE; i++)
2536 if (CURSEG_I(sbi, i)->zone == zoneno)
2537 break;
2538
2539 if (i < NR_CURSEG_TYPE) {
2540 /* zone is in user, try another */
2541 if (go_left)
2542 hint = zoneno * sbi->secs_per_zone - 1;
2543 else if (zoneno + 1 >= total_zones)
2544 hint = 0;
2545 else
2546 hint = (zoneno + 1) * sbi->secs_per_zone;
2547 init = false;
2548 goto find_other_zone;
2549 }
2550got_it:
2551 /* set it as dirty segment in free segmap */
9850cf4a 2552 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2553 __set_inuse(sbi, segno);
2554 *newseg = segno;
1a118ccf 2555 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2556}
2557
2558static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2559{
2560 struct curseg_info *curseg = CURSEG_I(sbi, type);
2561 struct summary_footer *sum_footer;
093749e2 2562 unsigned short seg_type = curseg->seg_type;
351df4b2 2563
d0b9e42a 2564 curseg->inited = true;
351df4b2 2565 curseg->segno = curseg->next_segno;
4ddb1a4d 2566 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2567 curseg->next_blkoff = 0;
2568 curseg->next_segno = NULL_SEGNO;
2569
2570 sum_footer = &(curseg->sum_blk->footer);
2571 memset(sum_footer, 0, sizeof(struct summary_footer));
093749e2
CY		 2572
2573 sanity_check_seg_type(sbi, seg_type);
2574
2575 if (IS_DATASEG(seg_type))
351df4b2 2576 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
093749e2 2577 if (IS_NODESEG(seg_type))
351df4b2 2578 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
093749e2 2579 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
351df4b2
JK		 2580}
2581
7a20b8a6
JK		 2582static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2583{
d0b9e42a 2584 struct curseg_info *curseg = CURSEG_I(sbi, type);
093749e2
CY		 2585 unsigned short seg_type = curseg->seg_type;
2586
2587 sanity_check_seg_type(sbi, seg_type);
d0b9e42a 2588
a7881893 2589 /* if segs_per_sec is large than 1, we need to keep original policy. */
2c70c5e3 2590 if (__is_large_section(sbi))
d0b9e42a
CY		 2591 return curseg->segno;
2592
2593 /* inmem log may not locate on any segment after mount */
2594 if (!curseg->inited)
2595 return 0;
a7881893 2596
4354994f
DR		 2597 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2598 return 0;
2599
b94929d9 2600 if (test_opt(sbi, NOHEAP) &&
093749e2 2601 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
7a20b8a6
JK		 2602 return 0;
2603
e066b83c
JK		 2604 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2605 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2606
2607 /* find segments from 0 to reuse freed segments */
63189b78 2608 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2609 return 0;
2610
d0b9e42a 2611 return curseg->segno;
7a20b8a6
JK		 2612}
2613
0a8165d7 2614/*
351df4b2
JK		 2615 * Allocate a current working segment.
2616 * This function always allocates a free segment in LFS manner.
2617 */
2618static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2619{
2620 struct curseg_info *curseg = CURSEG_I(sbi, type);
d0b9e42a 2621 unsigned short seg_type = curseg->seg_type;
351df4b2
JK		 2622 unsigned int segno = curseg->segno;
2623 int dir = ALLOC_LEFT;
2624
d0b9e42a
CY		 2625 if (curseg->inited)
2626 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2627 GET_SUM_BLOCK(sbi, segno));
d0b9e42a 2628 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
351df4b2
JK		 2629 dir = ALLOC_RIGHT;
2630
2631 if (test_opt(sbi, NOHEAP))
2632 dir = ALLOC_RIGHT;
2633
7a20b8a6 2634 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2635 get_new_segment(sbi, &segno, new_sec, dir);
2636 curseg->next_segno = segno;
2637 reset_curseg(sbi, type, 1);
2638 curseg->alloc_type = LFS;
2639}
2640
453e2ff8
CY		 2641static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2642 int segno, block_t start)
351df4b2 2643{
453e2ff8 2644 struct seg_entry *se = get_seg_entry(sbi, segno);
e81c93cf 2645 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2646 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2647 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2648 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
453e2ff8 2649 int i;
e81c93cf
CL		 2650
2651 for (i = 0; i < entries; i++)
2652 target_map[i] = ckpt_map[i] | cur_map[i];
2653
453e2ff8 2654 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
351df4b2
JK		 2655}
2656
0a8165d7 2657/*
351df4b2
JK		 2658 * If a segment is written by LFS manner, next block offset is just obtained
2659 * by increasing the current block offset. However, if a segment is written by
2660 * SSR manner, next block offset obtained by calling __next_free_blkoff
2661 */
2662static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2663 struct curseg_info *seg)
2664{
2665 if (seg->alloc_type == SSR)
453e2ff8
CY		 2666 seg->next_blkoff =
2667 __next_free_blkoff(sbi, seg->segno,
2668 seg->next_blkoff + 1);
351df4b2
JK		 2669 else
2670 seg->next_blkoff++;
2671}
2672
61461fc9
CY		 2673bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2674{
453e2ff8 2675 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
61461fc9
CY		 2676}
2677
0a8165d7 2678/*
e1c42045 2679 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2680 * manner, so it should recover the existing segment information of valid blocks
2681 */
093749e2 2682static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
351df4b2
JK		 2683{
2684 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2685 struct curseg_info *curseg = CURSEG_I(sbi, type);
2686 unsigned int new_segno = curseg->next_segno;
2687 struct f2fs_summary_block *sum_node;
2688 struct page *sum_page;
2689
093749e2
CY		 2690 if (flush)
2691 write_sum_page(sbi, curseg->sum_blk,
2692 GET_SUM_BLOCK(sbi, curseg->segno));
2693
351df4b2
JK		 2694 __set_test_and_inuse(sbi, new_segno);
2695
2696 mutex_lock(&dirty_i->seglist_lock);
2697 __remove_dirty_segment(sbi, new_segno, PRE);
2698 __remove_dirty_segment(sbi, new_segno, DIRTY);
2699 mutex_unlock(&dirty_i->seglist_lock);
2700
2701 reset_curseg(sbi, type, 1);
2702 curseg->alloc_type = SSR;
453e2ff8 2703 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
351df4b2 2704
4d57b86d 2705 sum_page = f2fs_get_sum_page(sbi, new_segno);
86f33603
JK		 2706 if (IS_ERR(sum_page)) {
2707 /* GC won't be able to use stale summary pages by cp_error */
2708 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2709 return;
2710 }
025d63a4
CY		 2711 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2712 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2713 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2714}
2715
093749e2
CY		 2716static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2717 int alloc_mode, unsigned long long age);
2718
2719static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2720 int target_type, int alloc_mode,
2721 unsigned long long age)
2722{
2723 struct curseg_info *curseg = CURSEG_I(sbi, type);
2724
2725 curseg->seg_type = target_type;
2726
2727 if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2728 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2729
2730 curseg->seg_type = se->type;
2731 change_curseg(sbi, type, true);
2732 } else {
2733 /* allocate cold segment by default */
2734 curseg->seg_type = CURSEG_COLD_DATA;
2735 new_curseg(sbi, type, true);
2736 }
2737 stat_inc_seg_type(sbi, curseg);
2738}
2739
2740static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2741{
2742 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2743
2744 if (!sbi->am.atgc_enabled)
2745 return;
2746
2747 down_read(&SM_I(sbi)->curseg_lock);
2748
2749 mutex_lock(&curseg->curseg_mutex);
2750 down_write(&SIT_I(sbi)->sentry_lock);
2751
2752 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2753
2754 up_write(&SIT_I(sbi)->sentry_lock);
2755 mutex_unlock(&curseg->curseg_mutex);
2756
2757 up_read(&SM_I(sbi)->curseg_lock);
2758
2759}
2760void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2761{
2762 __f2fs_init_atgc_curseg(sbi);
2763}
2764
2765static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
d0b9e42a
CY		 2766{
2767 struct curseg_info *curseg = CURSEG_I(sbi, type);
2768
2769 mutex_lock(&curseg->curseg_mutex);
2770 if (!curseg->inited)
2771 goto out;
2772
2773 if (get_valid_blocks(sbi, curseg->segno, false)) {
2774 write_sum_page(sbi, curseg->sum_blk,
2775 GET_SUM_BLOCK(sbi, curseg->segno));
2776 } else {
2777 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2778 __set_test_and_free(sbi, curseg->segno, true);
2779 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2780 }
2781out:
2782 mutex_unlock(&curseg->curseg_mutex);
2783}
2784
093749e2
CY		 2785void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2786{
2787 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2788
2789 if (sbi->am.atgc_enabled)
2790 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2791}
2792
2793static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
d0b9e42a
CY		 2794{
2795 struct curseg_info *curseg = CURSEG_I(sbi, type);
2796
2797 mutex_lock(&curseg->curseg_mutex);
2798 if (!curseg->inited)
2799 goto out;
2800 if (get_valid_blocks(sbi, curseg->segno, false))
2801 goto out;
2802
2803 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2804 __set_test_and_inuse(sbi, curseg->segno);
2805 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2806out:
2807 mutex_unlock(&curseg->curseg_mutex);
2808}
2809
093749e2
CY		 2810void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2811{
2812 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2813
2814 if (sbi->am.atgc_enabled)
2815 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2816}
2817
2818static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2819 int alloc_mode, unsigned long long age)
43727527
JK		 2820{
2821 struct curseg_info *curseg = CURSEG_I(sbi, type);
2822 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2823 unsigned segno = NULL_SEGNO;
093749e2 2824 unsigned short seg_type = curseg->seg_type;
d27c3d89
CY		 2825 int i, cnt;
2826 bool reversed = false;
c192f7a4 2827
093749e2
CY		 2828 sanity_check_seg_type(sbi, seg_type);
2829
4d57b86d 2830 /* f2fs_need_SSR() already forces to do this */
093749e2 2831 if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
e066b83c 2832 curseg->next_segno = segno;
c192f7a4 2833 return 1;
e066b83c 2834 }
43727527 2835
70d625cb 2836 /* For node segments, let's do SSR more intensively */
093749e2
CY		 2837 if (IS_NODESEG(seg_type)) {
2838 if (seg_type >= CURSEG_WARM_NODE) {
d27c3d89
CY		 2839 reversed = true;
2840 i = CURSEG_COLD_NODE;
2841 } else {
2842 i = CURSEG_HOT_NODE;
2843 }
2844 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2845 } else {
093749e2 2846 if (seg_type >= CURSEG_WARM_DATA) {
d27c3d89
CY		 2847 reversed = true;
2848 i = CURSEG_COLD_DATA;
2849 } else {
2850 i = CURSEG_HOT_DATA;
2851 }
2852 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2853 }
43727527 2854
d27c3d89 2855 for (; cnt-- > 0; reversed ? i-- : i++) {
093749e2 2856 if (i == seg_type)
c192f7a4 2857 continue;
093749e2 2858 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
e066b83c 2859 curseg->next_segno = segno;
43727527 2860 return 1;
e066b83c 2861 }
c192f7a4 2862 }
4354994f
DR		 2863
2864 /* find valid_blocks=0 in dirty list */
2865 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2866 segno = get_free_segment(sbi);
2867 if (segno != NULL_SEGNO) {
2868 curseg->next_segno = segno;
2869 return 1;
2870 }
2871 }
43727527
JK		 2872 return 0;
2873}
2874
351df4b2
JK		 2875/*
2876 * flush out current segment and replace it with new segment
2877 * This function should be returned with success, otherwise BUG
2878 */
2879static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2880 int type, bool force)
2881{
a7881893
JK		 2882 struct curseg_info *curseg = CURSEG_I(sbi, type);
2883
7b405275 2884 if (force)
351df4b2 2885 new_curseg(sbi, type, true);
5b6c6be2 2886 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
093749e2 2887 curseg->seg_type == CURSEG_WARM_NODE)
351df4b2 2888 new_curseg(sbi, type, false);
093749e2
CY		 2889 else if (curseg->alloc_type == LFS &&
2890 is_next_segment_free(sbi, curseg, type) &&
4354994f 2891 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
a7881893 2892 new_curseg(sbi, type, false);
093749e2
CY		 2893 else if (f2fs_need_SSR(sbi) &&
2894 get_ssr_segment(sbi, type, SSR, 0))
2895 change_curseg(sbi, type, true);
351df4b2
JK		 2896 else
2897 new_curseg(sbi, type, false);
dcdfff65 2898
a7881893 2899 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2900}
2901
0ef81833 2902void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
04f0b2ea
QS		 2903 unsigned int start, unsigned int end)
2904{
2905 struct curseg_info *curseg = CURSEG_I(sbi, type);
2906 unsigned int segno;
2907
2908 down_read(&SM_I(sbi)->curseg_lock);
2909 mutex_lock(&curseg->curseg_mutex);
2910 down_write(&SIT_I(sbi)->sentry_lock);
2911
2912 segno = CURSEG_I(sbi, type)->segno;
2913 if (segno < start || segno > end)
2914 goto unlock;
2915
093749e2
CY		 2916 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2917 change_curseg(sbi, type, true);
04f0b2ea
QS		 2918 else
2919 new_curseg(sbi, type, true);
2920
2921 stat_inc_seg_type(sbi, curseg);
2922
2923 locate_dirty_segment(sbi, segno);
2924unlock:
2925 up_write(&SIT_I(sbi)->sentry_lock);
2926
2927 if (segno != curseg->segno)
dcbb4c10
JP		 2928 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2929 type, segno, curseg->segno);
04f0b2ea
QS		 2930
2931 mutex_unlock(&curseg->curseg_mutex);
2932 up_read(&SM_I(sbi)->curseg_lock);
2933}
2934
e1175f02 2935static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
509f1010 2936 bool new_sec, bool force)
351df4b2 2937{
901d745f 2938 struct curseg_info *curseg = CURSEG_I(sbi, type);
6ae1be13 2939 unsigned int old_segno;
351df4b2 2940
d0b9e42a
CY		 2941 if (!curseg->inited)
2942 goto alloc;
2943
509f1010 2944 if (force || curseg->next_blkoff ||
e1175f02
CY		 2945 get_valid_blocks(sbi, curseg->segno, new_sec))
2946 goto alloc;
3d26fa6b 2947
61461fc9
CY		 2948 if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
2949 return;
d0b9e42a 2950alloc:
901d745f
CY		 2951 old_segno = curseg->segno;
2952 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2953 locate_dirty_segment(sbi, old_segno);
2954}
f5a53edc 2955
509f1010
CY		 2956static void __allocate_new_section(struct f2fs_sb_info *sbi,
2957 int type, bool force)
e1175f02 2958{
509f1010 2959 __allocate_new_segment(sbi, type, true, force);
e1175f02
CY		 2960}
2961
509f1010 2962void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
901d745f 2963{
823d13e1 2964 down_read(&SM_I(sbi)->curseg_lock);
901d745f 2965 down_write(&SIT_I(sbi)->sentry_lock);
509f1010 2966 __allocate_new_section(sbi, type, force);
901d745f 2967 up_write(&SIT_I(sbi)->sentry_lock);
823d13e1 2968 up_read(&SM_I(sbi)->curseg_lock);
901d745f 2969}
3d26fa6b 2970
901d745f
CY		 2971void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2972{
2973 int i;
2974
823d13e1 2975 down_read(&SM_I(sbi)->curseg_lock);
901d745f
CY		 2976 down_write(&SIT_I(sbi)->sentry_lock);
2977 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
509f1010 2978 __allocate_new_segment(sbi, i, false, false);
3d26fa6b 2979 up_write(&SIT_I(sbi)->sentry_lock);
823d13e1 2980 up_read(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 2981}
2982
2983static const struct segment_allocation default_salloc_ops = {
2984 .allocate_segment = allocate_segment_by_default,
2985};
2986
4d57b86d
CY		 2987bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2988 struct cp_control *cpc)
25290fa5
JK		 2989{
2990 __u64 trim_start = cpc->trim_start;
2991 bool has_candidate = false;
2992
3d26fa6b 2993 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2994 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2995 if (add_discard_addrs(sbi, cpc, true)) {
2996 has_candidate = true;
2997 break;
2998 }
2999 }
3d26fa6b 3000 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 3001
3002 cpc->trim_start = trim_start;
3003 return has_candidate;
3004}
3005
01f9cf6d 3006static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 3007 struct discard_policy *dpolicy,
3008 unsigned int start, unsigned int end)
3009{
3010 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3011 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3012 struct rb_node **insert_p = NULL, *insert_parent = NULL;
3013 struct discard_cmd *dc;
3014 struct blk_plug plug;
3015 int issued;
01f9cf6d 3016 unsigned int trimmed = 0;
9a997188
JK		 3017
3018next:
3019 issued = 0;
3020
3021 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 3022 if (unlikely(dcc->rbtree_check))
3023 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2e9b2bb2 3024 &dcc->root, false));
9a997188 3025
4d57b86d 3026 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 3027 NULL, start,
3028 (struct rb_entry **)&prev_dc,
3029 (struct rb_entry **)&next_dc,
4dada3fd 3030 &insert_p, &insert_parent, true, NULL);
9a997188
JK		 3031 if (!dc)
3032 dc = next_dc;
3033
3034 blk_start_plug(&plug);
3035
3036 while (dc && dc->lstart <= end) {
3037 struct rb_node *node;
6b9cb124 3038 int err = 0;
9a997188
JK		 3039
3040 if (dc->len < dpolicy->granularity)
3041 goto skip;
3042
3043 if (dc->state != D_PREP) {
3044 list_move_tail(&dc->list, &dcc->fstrim_list);
3045 goto skip;
3046 }
3047
6b9cb124 3048 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
9a997188 3049
35ec7d57 3050 if (issued >= dpolicy->max_requests) {
9a997188
JK		 3051 start = dc->lstart + dc->len;
3052
6b9cb124
CY		 3053 if (err)
3054 __remove_discard_cmd(sbi, dc);
3055
9a997188
JK		 3056 blk_finish_plug(&plug);
3057 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 3058 trimmed += __wait_all_discard_cmd(sbi, NULL);
5df7731f 3059 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
9a997188
JK		 3060 goto next;
3061 }
3062skip:
3063 node = rb_next(&dc->rb_node);
6b9cb124
CY		 3064 if (err)
3065 __remove_discard_cmd(sbi, dc);
9a997188
JK		 3066 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3067
3068 if (fatal_signal_pending(current))
3069 break;
3070 }
3071
3072 blk_finish_plug(&plug);
3073 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 3074
3075 return trimmed;
9a997188
JK		 3076}
3077
4b2fecc8
JK		 3078int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3079{
f7ef9b83
JK		 3080 __u64 start = F2FS_BYTES_TO_BLK(range->start);
3081 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 3082 unsigned int start_segno, end_segno;
8412663d 3083 block_t start_block, end_block;
4b2fecc8 3084 struct cp_control cpc;
78997b56 3085 struct discard_policy dpolicy;
0ea80512 3086 unsigned long long trimmed = 0;
c34f42e2 3087 int err = 0;
b0332a0f 3088 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
4b2fecc8 3089
836b5a63 3090 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 3091 return -EINVAL;
3092
3f16ecd9
CY		 3093 if (end < MAIN_BLKADDR(sbi))
3094 goto out;
4b2fecc8 3095
ed214a11 3096 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
dcbb4c10 3097 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
10f966bb 3098 return -EFSCORRUPTED;
ed214a11
YH		 3099 }
3100
4b2fecc8 3101 /* start/end segment number in main_area */
7cd8558b
JK		 3102 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3103 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3104 GET_SEGNO(sbi, end);
ad6672bb
YS		 3105 if (need_align) {
3106 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3107 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3108 }
8412663d 3109
4b2fecc8 3110 cpc.reason = CP_DISCARD;
836b5a63 3111 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 3112 cpc.trim_start = start_segno;
3113 cpc.trim_end = end_segno;
4b2fecc8 3114
377224c4
CY		 3115 if (sbi->discard_blks == 0)
3116 goto out;
74fa5f3d 3117
fb24fea7 3118 down_write(&sbi->gc_lock);
4d57b86d 3119 err = f2fs_write_checkpoint(sbi, &cpc);
fb24fea7 3120 up_write(&sbi->gc_lock);
377224c4
CY		 3121 if (err)
3122 goto out;
8412663d 3123
e555da9f
JK		 3124 /*
3125 * We filed discard candidates, but actually we don't need to wait for
3126 * all of them, since they'll be issued in idle time along with runtime
3127 * discard option. User configuration looks like using runtime discard
3128 * or periodic fstrim instead of it.
3129 */
7d20c8ab 3130 if (f2fs_realtime_discard_enable(sbi))
5a615492
JK		 3131 goto out;
3132
3133 start_block = START_BLOCK(sbi, start_segno);
3134 end_block = START_BLOCK(sbi, end_segno + 1);
3135
3136 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 3137 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3138 start_block, end_block);
5a615492 3139
01f9cf6d 3140 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 3141 start_block, end_block);
377224c4 3142out:
6eae2694
CY		 3143 if (!err)
3144 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 3145 return err;
4b2fecc8
JK		 3146}
3147
093749e2
CY		 3148static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3149 struct curseg_info *curseg)
351df4b2 3150{
de881df9
AR		 3151 return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3152 curseg->segno);
351df4b2
JK		 3153}
3154
4d57b86d 3155int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 3156{
3157 switch (hint) {
3158 case WRITE_LIFE_SHORT:
3159 return CURSEG_HOT_DATA;
3160 case WRITE_LIFE_EXTREME:
3161 return CURSEG_COLD_DATA;
3162 default:
3163 return CURSEG_WARM_DATA;
3164 }
3165}
3166
0cdd3195
HL		 3167/* This returns write hints for each segment type. This hints will be
3168 * passed down to block layer. There are mapping tables which depend on
3169 * the mount option 'whint_mode'.
3170 *
3171 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3172 *
3173 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3174 *
3175 * User F2FS Block
3176 * ---- ---- -----
3177 * META WRITE_LIFE_NOT_SET
3178 * HOT_NODE "
3179 * WARM_NODE "
3180 * COLD_NODE "
3181 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
3182 * extension list " "
3183 *
3184 * -- buffered io
3185 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3186 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3187 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3188 * WRITE_LIFE_NONE " "
3189 * WRITE_LIFE_MEDIUM " "
3190 * WRITE_LIFE_LONG " "
3191 *
3192 * -- direct io
3193 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3194 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3195 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3196 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
3197 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
3198 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
3199 *
f2e703f9
HL		 3200 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3201 *
3202 * User F2FS Block
3203 * ---- ---- -----
3204 * META WRITE_LIFE_MEDIUM;
3205 * HOT_NODE WRITE_LIFE_NOT_SET
3206 * WARM_NODE "
3207 * COLD_NODE WRITE_LIFE_NONE
3208 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
3209 * extension list " "
3210 *
3211 * -- buffered io
3212 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3213 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3214 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
3215 * WRITE_LIFE_NONE " "
3216 * WRITE_LIFE_MEDIUM " "
3217 * WRITE_LIFE_LONG " "
3218 *
3219 * -- direct io
3220 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3221 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3222 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3223 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
3224 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
3225 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 3226 */
3227
4d57b86d 3228enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 3229 enum page_type type, enum temp_type temp)
3230{
63189b78 3231 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 3232 if (type == DATA) {
f2e703f9 3233 if (temp == WARM)
0cdd3195 3234 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 3235 else if (temp == HOT)
3236 return WRITE_LIFE_SHORT;
3237 else if (temp == COLD)
3238 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 3239 } else {
3240 return WRITE_LIFE_NOT_SET;
3241 }
63189b78 3242 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 3243 if (type == DATA) {
3244 if (temp == WARM)
3245 return WRITE_LIFE_LONG;
3246 else if (temp == HOT)
3247 return WRITE_LIFE_SHORT;
3248 else if (temp == COLD)
3249 return WRITE_LIFE_EXTREME;
3250 } else if (type == NODE) {
3251 if (temp == WARM || temp == HOT)
3252 return WRITE_LIFE_NOT_SET;
3253 else if (temp == COLD)
3254 return WRITE_LIFE_NONE;
3255 } else if (type == META) {
3256 return WRITE_LIFE_MEDIUM;
3257 }
0cdd3195 3258 }
f2e703f9 3259 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 3260}
3261
81377bd6 3262static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 3263{
81377bd6 3264 if (fio->type == DATA)
351df4b2
JK		 3265 return CURSEG_HOT_DATA;
3266 else
3267 return CURSEG_HOT_NODE;
3268}
3269
81377bd6 3270static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 3271{
81377bd6
JK		 3272 if (fio->type == DATA) {
3273 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 3274
3275 if (S_ISDIR(inode->i_mode))
3276 return CURSEG_HOT_DATA;
3277 else
3278 return CURSEG_COLD_DATA;
3279 } else {
81377bd6 3280 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 3281 return CURSEG_WARM_NODE;
351df4b2
JK		 3282 else
3283 return CURSEG_COLD_NODE;
3284 }
3285}
3286
81377bd6 3287static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 3288{
81377bd6
JK		 3289 if (fio->type == DATA) {
3290 struct inode *inode = fio->page->mapping->host;
351df4b2 3291
093749e2 3292 if (is_cold_data(fio->page)) {
ac2d750b
WG		 3293 if (fio->sbi->am.atgc_enabled &&
3294 (fio->io_type == FS_DATA_IO) &&
3295 (fio->sbi->gc_mode != GC_URGENT_HIGH))
093749e2
CY		 3296 return CURSEG_ALL_DATA_ATGC;
3297 else
3298 return CURSEG_COLD_DATA;
3299 }
602a16d5 3300 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
351df4b2 3301 return CURSEG_COLD_DATA;
b6a06cbb 3302 if (file_is_hot(inode) ||
b4c3ca8b 3303 is_inode_flag_set(inode, FI_HOT_DATA) ||
2079f115
CY		 3304 f2fs_is_atomic_file(inode) ||
3305 f2fs_is_volatile_file(inode))
ef095d19 3306 return CURSEG_HOT_DATA;
4d57b86d 3307 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 3308 } else {
81377bd6
JK		 3309 if (IS_DNODE(fio->page))
3310 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 3311 CURSEG_HOT_NODE;
ef095d19 3312 return CURSEG_COLD_NODE;
351df4b2
JK		 3313 }
3314}
3315
81377bd6 3316static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 3317{
a912b54d
JK		 3318 int type = 0;
3319
63189b78 3320 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 3321 case 2:
a912b54d
JK		 3322 type = __get_segment_type_2(fio);
3323 break;
351df4b2 3324 case 4:
a912b54d
JK		 3325 type = __get_segment_type_4(fio);
3326 break;
3327 case 6:
3328 type = __get_segment_type_6(fio);
3329 break;
3330 default:
3331 f2fs_bug_on(fio->sbi, true);
351df4b2 3332 }
81377bd6 3333
a912b54d
JK		 3334 if (IS_HOT(type))
3335 fio->temp = HOT;
3336 else if (IS_WARM(type))
3337 fio->temp = WARM;
3338 else
3339 fio->temp = COLD;
3340 return type;
351df4b2
JK		 3341}
3342
4d57b86d 3343void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 3344 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5 3345 struct f2fs_summary *sum, int type,
093749e2 3346 struct f2fs_io_info *fio)
351df4b2
JK		 3347{
3348 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 3349 struct curseg_info *curseg = CURSEG_I(sbi, type);
c5d02785 3350 unsigned long long old_mtime;
093749e2
CY		 3351 bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3352 struct seg_entry *se = NULL;
351df4b2 3353
2b60311d
CY		 3354 down_read(&SM_I(sbi)->curseg_lock);
3355
351df4b2 3356 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3357 down_write(&sit_i->sentry_lock);
351df4b2 3358
093749e2
CY		 3359 if (from_gc) {
3360 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3361 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3362 sanity_check_seg_type(sbi, se->type);
3363 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3364 }
351df4b2 3365 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 3366
093749e2
CY		 3367 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3368
4e6a8d9b
JK		 3369 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3370
351df4b2
JK		 3371 /*
3372 * __add_sum_entry should be resided under the curseg_mutex
3373 * because, this function updates a summary entry in the
3374 * current summary block.
3375 */
e79efe3b 3376 __add_sum_entry(sbi, type, sum);
351df4b2 3377
351df4b2 3378 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 3379
3380 stat_inc_block_count(sbi, curseg);
351df4b2 3381
c5d02785
CY		 3382 if (from_gc) {
3383 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3384 } else {
3385 update_segment_mtime(sbi, old_blkaddr, 0);
3386 old_mtime = 0;
3387 }
3388 update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3389
65f1b80b
YS		 3390 /*
3391 * SIT information should be updated before segment allocation,
3392 * since SSR needs latest valid block information.
3393 */
3394 update_sit_entry(sbi, *new_blkaddr, 1);
3395 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3396 update_sit_entry(sbi, old_blkaddr, -1);
3397
093749e2
CY		 3398 if (!__has_curseg_space(sbi, curseg)) {
3399 if (from_gc)
3400 get_atssr_segment(sbi, type, se->type,
3401 AT_SSR, se->mtime);
3402 else
3403 sit_i->s_ops->allocate_segment(sbi, type, false);
3404 }
351df4b2 3405 /*
65f1b80b
YS		 3406 * segment dirty status should be updated after segment allocation,
3407 * so we just need to update status only one time after previous
3408 * segment being closed.
351df4b2 3409 */
65f1b80b
YS		 3410 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3411 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 3412
3d26fa6b 3413 up_write(&sit_i->sentry_lock);
351df4b2 3414
704956ec 3415 if (page && IS_NODESEG(type)) {
351df4b2
JK		 3416 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3417
704956ec
CY		 3418 f2fs_inode_chksum_set(sbi, page);
3419 }
3420
f608c38c 3421 if (fio) {
fb830fc5
CY		 3422 struct f2fs_bio_info *io;
3423
25ae837e
CY		 3424 if (F2FS_IO_ALIGNED(sbi))
3425 fio->retry = false;
3426
fb830fc5
CY		 3427 INIT_LIST_HEAD(&fio->list);
3428 fio->in_list = true;
3429 io = sbi->write_io[fio->type] + fio->temp;
3430 spin_lock(&io->io_lock);
3431 list_add_tail(&fio->list, &io->io_list);
3432 spin_unlock(&io->io_lock);
3433 }
3434
bfad7c2d 3435 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 3436
3437 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 3438}
3439
39d787be
CY		 3440static void update_device_state(struct f2fs_io_info *fio)
3441{
3442 struct f2fs_sb_info *sbi = fio->sbi;
3443 unsigned int devidx;
3444
0916878d 3445 if (!f2fs_is_multi_device(sbi))
39d787be
CY		 3446 return;
3447
3448 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3449
3450 /* update device state for fsync */
4d57b86d 3451 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 3452
3453 /* update device state for checkpoint */
3454 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3455 spin_lock(&sbi->dev_lock);
3456 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3457 spin_unlock(&sbi->dev_lock);
3458 }
39d787be
CY		 3459}
3460
05ca3632 3461static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 3462{
81377bd6 3463 int type = __get_segment_type(fio);
b0332a0f 3464 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
bfad7c2d 3465
107a805d
CY		 3466 if (keep_order)
3467 down_read(&fio->sbi->io_order_lock);
0a595eba 3468reallocate:
4d57b86d 3469 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
093749e2 3470 &fio->new_blkaddr, sum, type, fio);
6aa58d8a
CY		 3471 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3472 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3473 fio->old_blkaddr, fio->old_blkaddr);
bfad7c2d 3474
351df4b2 3475 /* writeout dirty page into bdev */
fe16efe6
CY		 3476 f2fs_submit_page_write(fio);
3477 if (fio->retry) {
0a595eba
JK		 3478 fio->old_blkaddr = fio->new_blkaddr;
3479 goto reallocate;
3480 }
fe16efe6
CY		 3481
3482 update_device_state(fio);
3483
107a805d
CY		 3484 if (keep_order)
3485 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 3486}
3487
4d57b86d 3488void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 3489 enum iostat_type io_type)
351df4b2 3490{
458e6197 3491 struct f2fs_io_info fio = {
05ca3632 3492 .sbi = sbi,
458e6197 3493 .type = META,
0cdd3195 3494 .temp = HOT,
04d328de 3495 .op = REQ_OP_WRITE,
70fd7614 3496 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 3497 .old_blkaddr = page->index,
3498 .new_blkaddr = page->index,
05ca3632 3499 .page = page,
4375a336 3500 .encrypted_page = NULL,
fb830fc5 3501 .in_list = false,
458e6197
JK		 3502 };
3503
2b947003 3504 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 3505 fio.op_flags &= ~REQ_META;
2b947003 3506
351df4b2 3507 set_page_writeback(page);
17c50035 3508 ClearPageError(page);
b9109b0e 3509 f2fs_submit_page_write(&fio);
b0af6d49 3510
b63e7be5 3511 stat_inc_meta_count(sbi, page->index);
b0af6d49 3512 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 3513}
3514
4d57b86d 3515void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 3516{
3517 struct f2fs_summary sum;
05ca3632 3518
351df4b2 3519 set_summary(&sum, nid, 0, 0);
05ca3632 3520 do_write_page(&sum, fio);
b0af6d49
CY		 3521
3522 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3523}
3524
4d57b86d
CY		 3525void f2fs_outplace_write_data(struct dnode_of_data *dn,
3526 struct f2fs_io_info *fio)
351df4b2 3527{
05ca3632 3528 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2 3529 struct f2fs_summary sum;
351df4b2 3530
9850cf4a 3531 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
7735730d 3532 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
05ca3632 3533 do_write_page(&sum, fio);
f28b3434 3534 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 3535
3536 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3537}
3538
4d57b86d 3539int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 3540{
b0af6d49 3541 int err;
d21b0f23 3542 struct f2fs_sb_info *sbi = fio->sbi;
05573d6c 3543 unsigned int segno;
b0af6d49 3544
7a9d7548 3545 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 3546 /* i/o temperature is needed for passing down write hints */
3547 __get_segment_type(fio);
d21b0f23 3548
05573d6c
CY		 3549 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3550
3551 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3552 set_sbi_flag(sbi, SBI_NEED_FSCK);
2d821c12
CY		 3553 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3554 __func__, segno);
95577278
CY		 3555 err = -EFSCORRUPTED;
3556 goto drop_bio;
3557 }
3558
3559 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || f2fs_cp_error(sbi)) {
3560 err = -EIO;
3561 goto drop_bio;
05573d6c 3562 }
d21b0f23 3563
05ca3632 3564 stat_inc_inplace_blocks(fio->sbi);
b0af6d49 3565
0e7f4197 3566 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
8648de2c
CY		 3567 err = f2fs_merge_page_bio(fio);
3568 else
3569 err = f2fs_submit_page_bio(fio);
e46f6bd8 3570 if (!err) {
39d787be 3571 update_device_state(fio);
e46f6bd8
CY		 3572 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3573 }
b0af6d49 3574
95577278
CY		 3575 return err;
3576drop_bio:
349c4d6c 3577 if (fio->bio && *(fio->bio)) {
95577278
CY		 3578 struct bio *bio = *(fio->bio);
3579
3580 bio->bi_status = BLK_STS_IOERR;
3581 bio_endio(bio);
349c4d6c 3582 *(fio->bio) = NULL;
95577278 3583 }
b0af6d49 3584 return err;
351df4b2
JK		 3585}
3586
2b60311d
CY		 3587static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3588 unsigned int segno)
3589{
3590 int i;
3591
3592 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3593 if (CURSEG_I(sbi, i)->segno == segno)
3594 break;
3595 }
3596 return i;
3597}
3598
4d57b86d 3599void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 3600 block_t old_blkaddr, block_t new_blkaddr,
c5d02785
CY		 3601 bool recover_curseg, bool recover_newaddr,
3602 bool from_gc)
351df4b2
JK		 3603{
3604 struct sit_info *sit_i = SIT_I(sbi);
3605 struct curseg_info *curseg;
3606 unsigned int segno, old_cursegno;
3607 struct seg_entry *se;
3608 int type;
19f106bc 3609 unsigned short old_blkoff;
753a8ed0 3610 unsigned char old_alloc_type;
351df4b2
JK		 3611
3612 segno = GET_SEGNO(sbi, new_blkaddr);
3613 se = get_seg_entry(sbi, segno);
3614 type = se->type;
3615
2b60311d
CY		 3616 down_write(&SM_I(sbi)->curseg_lock);
3617
19f106bc
CY		 3618 if (!recover_curseg) {
3619 /* for recovery flow */
3620 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3621 if (old_blkaddr == NULL_ADDR)
3622 type = CURSEG_COLD_DATA;
3623 else
3624 type = CURSEG_WARM_DATA;
3625 }
3626 } else {
2b60311d
CY		 3627 if (IS_CURSEG(sbi, segno)) {
3628 /* se->type is volatile as SSR allocation */
3629 type = __f2fs_get_curseg(sbi, segno);
3630 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3631 } else {
351df4b2 3632 type = CURSEG_WARM_DATA;
2b60311d 3633 }
351df4b2 3634 }
19f106bc 3635
2c190504 3636 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 3637 curseg = CURSEG_I(sbi, type);
3638
3639 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3640 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3641
3642 old_cursegno = curseg->segno;
19f106bc 3643 old_blkoff = curseg->next_blkoff;
753a8ed0 3644 old_alloc_type = curseg->alloc_type;
351df4b2
JK		 3645
3646 /* change the current segment */
3647 if (segno != curseg->segno) {
3648 curseg->next_segno = segno;
093749e2 3649 change_curseg(sbi, type, true);
351df4b2
JK		 3650 }
3651
491c0854 3652 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 3653 __add_sum_entry(sbi, type, sum);
351df4b2 3654
c5d02785
CY		 3655 if (!recover_curseg || recover_newaddr) {
3656 if (!from_gc)
3657 update_segment_mtime(sbi, new_blkaddr, 0);
6e2c64ad 3658 update_sit_entry(sbi, new_blkaddr, 1);
c5d02785 3659 }
6aa58d8a
CY		 3660 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3661 invalidate_mapping_pages(META_MAPPING(sbi),
3662 old_blkaddr, old_blkaddr);
c5d02785
CY		 3663 if (!from_gc)
3664 update_segment_mtime(sbi, old_blkaddr, 0);
6e2c64ad 3665 update_sit_entry(sbi, old_blkaddr, -1);
6aa58d8a 3666 }
6e2c64ad
JK		 3667
3668 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3669 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3670
351df4b2 3671 locate_dirty_segment(sbi, old_cursegno);
351df4b2 3672
19f106bc
CY		 3673 if (recover_curseg) {
3674 if (old_cursegno != curseg->segno) {
3675 curseg->next_segno = old_cursegno;
093749e2 3676 change_curseg(sbi, type, true);
19f106bc
CY		 3677 }
3678 curseg->next_blkoff = old_blkoff;
753a8ed0 3679 curseg->alloc_type = old_alloc_type;
19f106bc
CY		 3680 }
3681
3d26fa6b 3682 up_write(&sit_i->sentry_lock);
351df4b2 3683 mutex_unlock(&curseg->curseg_mutex);
2b60311d 3684 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 3685}
3686
528e3459
CY		 3687void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3688 block_t old_addr, block_t new_addr,
28bc106b
CY		 3689 unsigned char version, bool recover_curseg,
3690 bool recover_newaddr)
528e3459
CY		 3691{
3692 struct f2fs_summary sum;
3693
3694 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3695
4d57b86d 3696 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
c5d02785 3697 recover_curseg, recover_newaddr, false);
528e3459 3698
f28b3434 3699 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 3700}
3701
93dfe2ac 3702void f2fs_wait_on_page_writeback(struct page *page,
bae0ee7a 3703 enum page_type type, bool ordered, bool locked)
93dfe2ac 3704{
93dfe2ac 3705 if (PageWriteback(page)) {
4081363f
JK		 3706 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3707
0b20fcec 3708 /* submit cached LFS IO */
bab475c5 3709 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
0b20fcec
CY		 3710 /* sbumit cached IPU IO */
3711 f2fs_submit_merged_ipu_write(sbi, NULL, page);
bae0ee7a 3712 if (ordered) {
fec1d657 3713 wait_on_page_writeback(page);
bae0ee7a
CY		 3714 f2fs_bug_on(sbi, locked && PageWriteback(page));
3715 } else {
fec1d657 3716 wait_for_stable_page(page);
bae0ee7a 3717 }
93dfe2ac
JK		 3718 }
3719}
3720
0ded69f6 3721void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
08b39fbd 3722{
0ded69f6 3723 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
08b39fbd
CY		 3724 struct page *cpage;
3725
0ded69f6
JK		 3726 if (!f2fs_post_read_required(inode))
3727 return;
3728
93770ab7 3729 if (!__is_valid_data_blkaddr(blkaddr))
08b39fbd
CY		 3730 return;
3731
08b39fbd
CY		 3732 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3733 if (cpage) {
bae0ee7a 3734 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
08b39fbd
CY		 3735 f2fs_put_page(cpage, 1);
3736 }
3737}
3738
1e78e8bd
ST		 3739void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3740 block_t len)
3741{
3742 block_t i;
3743
3744 for (i = 0; i < len; i++)
3745 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3746}
3747
7735730d 3748static int read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3749{
3750 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3751 struct curseg_info *seg_i;
3752 unsigned char *kaddr;
3753 struct page *page;
3754 block_t start;
3755 int i, j, offset;
3756
3757 start = start_sum_block(sbi);
3758
4d57b86d 3759 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3760 if (IS_ERR(page))
3761 return PTR_ERR(page);
351df4b2
JK		 3762 kaddr = (unsigned char *)page_address(page);
3763
3764 /* Step 1: restore nat cache */
3765 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3766 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3767
3768 /* Step 2: restore sit cache */
3769 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3770 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3771 offset = 2 * SUM_JOURNAL_SIZE;
3772
3773 /* Step 3: restore summary entries */
3774 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3775 unsigned short blk_off;
3776 unsigned int segno;
3777
3778 seg_i = CURSEG_I(sbi, i);
3779 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3780 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3781 seg_i->next_segno = segno;
3782 reset_curseg(sbi, i, 0);
3783 seg_i->alloc_type = ckpt->alloc_type[i];
3784 seg_i->next_blkoff = blk_off;
3785
3786 if (seg_i->alloc_type == SSR)
3787 blk_off = sbi->blocks_per_seg;
3788
3789 for (j = 0; j < blk_off; j++) {
3790 struct f2fs_summary *s;
5f029c04 3791
351df4b2
JK		 3792 s = (struct f2fs_summary *)(kaddr + offset);
3793 seg_i->sum_blk->entries[j] = *s;
3794 offset += SUMMARY_SIZE;
09cbfeaf 3795 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3796 SUM_FOOTER_SIZE)
3797 continue;
3798
3799 f2fs_put_page(page, 1);
3800 page = NULL;
3801
4d57b86d 3802 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3803 if (IS_ERR(page))
3804 return PTR_ERR(page);
351df4b2
JK		 3805 kaddr = (unsigned char *)page_address(page);
3806 offset = 0;
3807 }
3808 }
3809 f2fs_put_page(page, 1);
7735730d 3810 return 0;
351df4b2
JK		 3811}
3812
3813static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3814{
3815 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3816 struct f2fs_summary_block *sum;
3817 struct curseg_info *curseg;
3818 struct page *new;
3819 unsigned short blk_off;
3820 unsigned int segno = 0;
3821 block_t blk_addr = 0;
7735730d 3822 int err = 0;
351df4b2
JK		 3823
3824 /* get segment number and block addr */
3825 if (IS_DATASEG(type)) {
3826 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3827 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3828 CURSEG_HOT_DATA]);
119ee914 3829 if (__exist_node_summaries(sbi))
d0b9e42a 3830 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
351df4b2
JK		 3831 else
3832 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3833 } else {
3834 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3835 CURSEG_HOT_NODE]);
3836 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3837 CURSEG_HOT_NODE]);
119ee914 3838 if (__exist_node_summaries(sbi))
351df4b2
JK		 3839 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3840 type - CURSEG_HOT_NODE);
3841 else
3842 blk_addr = GET_SUM_BLOCK(sbi, segno);
3843 }
3844
4d57b86d 3845 new = f2fs_get_meta_page(sbi, blk_addr);
7735730d
CY		 3846 if (IS_ERR(new))
3847 return PTR_ERR(new);
351df4b2
JK		 3848 sum = (struct f2fs_summary_block *)page_address(new);
3849
3850 if (IS_NODESEG(type)) {
119ee914 3851 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3852 struct f2fs_summary *ns = &sum->entries[0];
3853 int i;
5f029c04 3854
351df4b2
JK		 3855 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3856 ns->version = 0;
3857 ns->ofs_in_node = 0;
3858 }
3859 } else {
7735730d
CY		 3860 err = f2fs_restore_node_summary(sbi, segno, sum);
3861 if (err)
3862 goto out;
351df4b2
JK		 3863 }
3864 }
3865
3866 /* set uncompleted segment to curseg */
3867 curseg = CURSEG_I(sbi, type);
3868 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3869
3870 /* update journal info */
3871 down_write(&curseg->journal_rwsem);
3872 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3873 up_write(&curseg->journal_rwsem);
3874
3875 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3876 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3877 curseg->next_segno = segno;
3878 reset_curseg(sbi, type, 0);
3879 curseg->alloc_type = ckpt->alloc_type[type];
3880 curseg->next_blkoff = blk_off;
3881 mutex_unlock(&curseg->curseg_mutex);
7735730d 3882out:
351df4b2 3883 f2fs_put_page(new, 1);
7735730d 3884 return err;
351df4b2
JK		 3885}
3886
3887static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3888{
21d3f8e1
JQ		 3889 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3890 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3891 int type = CURSEG_HOT_DATA;
e4fc5fbf 3892 int err;
351df4b2 3893
aaec2b1d 3894 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3895 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3896
3897 if (npages >= 2)
4d57b86d 3898 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3899 META_CP, true);
3fa06d7b 3900
351df4b2 3901 /* restore for compacted data summary */
7735730d
CY		 3902 err = read_compacted_summaries(sbi);
3903 if (err)
3904 return err;
351df4b2
JK		 3905 type = CURSEG_HOT_NODE;
3906 }
3907
119ee914 3908 if (__exist_node_summaries(sbi))
d0b9e42a
CY		 3909 f2fs_ra_meta_pages(sbi,
3910 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3911 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3fa06d7b 3912
e4fc5fbf
CY		 3913 for (; type <= CURSEG_COLD_NODE; type++) {
3914 err = read_normal_summaries(sbi, type);
3915 if (err)
3916 return err;
3917 }
3918
21d3f8e1
JQ		 3919 /* sanity check for summary blocks */
3920 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
9227d522 3921 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
dcbb4c10
JP		 3922 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3923 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
21d3f8e1 3924 return -EINVAL;
9227d522 3925 }
21d3f8e1 3926
351df4b2
JK		 3927 return 0;
3928}
3929
3930static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3931{
3932 struct page *page;
3933 unsigned char *kaddr;
3934 struct f2fs_summary *summary;
3935 struct curseg_info *seg_i;
3936 int written_size = 0;
3937 int i, j;
3938
4d57b86d 3939 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3940 kaddr = (unsigned char *)page_address(page);
81114baa 3941 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3942
3943 /* Step 1: write nat cache */
3944 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3945 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3946 written_size += SUM_JOURNAL_SIZE;
3947
3948 /* Step 2: write sit cache */
3949 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3950 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3951 written_size += SUM_JOURNAL_SIZE;
3952
351df4b2
JK		 3953 /* Step 3: write summary entries */
3954 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3955 unsigned short blkoff;
5f029c04 3956
351df4b2
JK		 3957 seg_i = CURSEG_I(sbi, i);
3958 if (sbi->ckpt->alloc_type[i] == SSR)
3959 blkoff = sbi->blocks_per_seg;
3960 else
3961 blkoff = curseg_blkoff(sbi, i);
3962
3963 for (j = 0; j < blkoff; j++) {
3964 if (!page) {
4d57b86d 3965 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3966 kaddr = (unsigned char *)page_address(page);
81114baa 3967 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3968 written_size = 0;
3969 }
3970 summary = (struct f2fs_summary *)(kaddr + written_size);
3971 *summary = seg_i->sum_blk->entries[j];
3972 written_size += SUMMARY_SIZE;
351df4b2 3973
09cbfeaf 3974 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3975 SUM_FOOTER_SIZE)
3976 continue;
3977
e8d61a74 3978 set_page_dirty(page);
351df4b2
JK		 3979 f2fs_put_page(page, 1);
3980 page = NULL;
3981 }
3982 }
e8d61a74
CY		 3983 if (page) {
3984 set_page_dirty(page);
351df4b2 3985 f2fs_put_page(page, 1);
e8d61a74 3986 }
351df4b2
JK		 3987}
3988
3989static void write_normal_summaries(struct f2fs_sb_info *sbi,
3990 block_t blkaddr, int type)
3991{
3992 int i, end;
5f029c04 3993
351df4b2
JK		 3994 if (IS_DATASEG(type))
3995 end = type + NR_CURSEG_DATA_TYPE;
3996 else
3997 end = type + NR_CURSEG_NODE_TYPE;
3998
b7ad7512
CY		 3999 for (i = type; i < end; i++)
4000 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 4001}
4002
4d57b86d 4003void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 4004{
aaec2b1d 4005 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 4006 write_compacted_summaries(sbi, start_blk);
4007 else
4008 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4009}
4010
4d57b86d 4011void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 4012{
119ee914 4013 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 4014}
4015
4d57b86d 4016int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 4017 unsigned int val, int alloc)
4018{
4019 int i;
4020
4021 if (type == NAT_JOURNAL) {
dfc08a12
CY		 4022 for (i = 0; i < nats_in_cursum(journal); i++) {
4023 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 4024 return i;
4025 }
dfc08a12
CY		 4026 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4027 return update_nats_in_cursum(journal, 1);
351df4b2 4028 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 4029 for (i = 0; i < sits_in_cursum(journal); i++)
4030 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 4031 return i;
dfc08a12
CY		 4032 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4033 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 4034 }
4035 return -1;
4036}
4037
4038static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4039 unsigned int segno)
4040{
86f33603 4041 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 4042}
4043
4044static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4045 unsigned int start)
4046{
4047 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 4048 struct page *page;
351df4b2 4049 pgoff_t src_off, dst_off;
351df4b2
JK		 4050
4051 src_off = current_sit_addr(sbi, start);
4052 dst_off = next_sit_addr(sbi, src_off);
4053
4d57b86d 4054 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 4055 seg_info_to_sit_page(sbi, page, start);
351df4b2 4056
068c3cd8 4057 set_page_dirty(page);
351df4b2
JK		 4058 set_to_next_sit(sit_i, start);
4059
068c3cd8 4060 return page;
351df4b2
JK		 4061}
4062
184a5cd2
CY		 4063static struct sit_entry_set *grab_sit_entry_set(void)
4064{
4065 struct sit_entry_set *ses =
80c54505 4066 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 4067
4068 ses->entry_cnt = 0;
4069 INIT_LIST_HEAD(&ses->set_list);
4070 return ses;
4071}
4072
4073static void release_sit_entry_set(struct sit_entry_set *ses)
4074{
4075 list_del(&ses->set_list);
4076 kmem_cache_free(sit_entry_set_slab, ses);
4077}
4078
4079static void adjust_sit_entry_set(struct sit_entry_set *ses,
4080 struct list_head *head)
4081{
4082 struct sit_entry_set *next = ses;
4083
4084 if (list_is_last(&ses->set_list, head))
4085 return;
4086
4087 list_for_each_entry_continue(next, head, set_list)
4088 if (ses->entry_cnt <= next->entry_cnt)
4089 break;
4090
4091 list_move_tail(&ses->set_list, &next->set_list);
4092}
4093
4094static void add_sit_entry(unsigned int segno, struct list_head *head)
4095{
4096 struct sit_entry_set *ses;
4097 unsigned int start_segno = START_SEGNO(segno);
4098
4099 list_for_each_entry(ses, head, set_list) {
4100 if (ses->start_segno == start_segno) {
4101 ses->entry_cnt++;
4102 adjust_sit_entry_set(ses, head);
4103 return;
4104 }
4105 }
4106
4107 ses = grab_sit_entry_set();
4108
4109 ses->start_segno = start_segno;
4110 ses->entry_cnt++;
4111 list_add(&ses->set_list, head);
4112}
4113
4114static void add_sits_in_set(struct f2fs_sb_info *sbi)
4115{
4116 struct f2fs_sm_info *sm_info = SM_I(sbi);
4117 struct list_head *set_list = &sm_info->sit_entry_set;
4118 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 4119 unsigned int segno;
4120
7cd8558b 4121 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 4122 add_sit_entry(segno, set_list);
4123}
4124
4125static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 4126{
4127 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 4128 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 4129 int i;
4130
b7ad7512 4131 down_write(&curseg->journal_rwsem);
dfc08a12 4132 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 4133 unsigned int segno;
4134 bool dirtied;
4135
dfc08a12 4136 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 4137 dirtied = __mark_sit_entry_dirty(sbi, segno);
4138
4139 if (!dirtied)
4140 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 4141 }
dfc08a12 4142 update_sits_in_cursum(journal, -i);
b7ad7512 4143 up_write(&curseg->journal_rwsem);
351df4b2
JK		 4144}
4145
0a8165d7 4146/*
351df4b2
JK		 4147 * CP calls this function, which flushes SIT entries including sit_journal,
4148 * and moves prefree segs to free segs.
4149 */
4d57b86d 4150void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 4151{
4152 struct sit_info *sit_i = SIT_I(sbi);
4153 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4154 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 4155 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 4156 struct sit_entry_set *ses, *tmp;
4157 struct list_head *head = &SM_I(sbi)->sit_entry_set;
04f0b2ea 4158 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4b2fecc8 4159 struct seg_entry *se;
351df4b2 4160
3d26fa6b 4161 down_write(&sit_i->sentry_lock);
351df4b2 4162
2b11a74b
WL		 4163 if (!sit_i->dirty_sentries)
4164 goto out;
4165
351df4b2 4166 /*
184a5cd2
CY		 4167 * add and account sit entries of dirty bitmap in sit entry
4168 * set temporarily
351df4b2 4169 */
184a5cd2 4170 add_sits_in_set(sbi);
351df4b2 4171
184a5cd2
CY		 4172 /*
4173 * if there are no enough space in journal to store dirty sit
4174 * entries, remove all entries from journal and add and account
4175 * them in sit entry set.
4176 */
04f0b2ea
QS		 4177 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4178 !to_journal)
184a5cd2 4179 remove_sits_in_journal(sbi);
b2955550 4180
184a5cd2
CY		 4181 /*
4182 * there are two steps to flush sit entries:
4183 * #1, flush sit entries to journal in current cold data summary block.
4184 * #2, flush sit entries to sit page.
4185 */
4186 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 4187 struct page *page = NULL;
184a5cd2
CY		 4188 struct f2fs_sit_block *raw_sit = NULL;
4189 unsigned int start_segno = ses->start_segno;
4190 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 4191 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 4192 unsigned int segno = start_segno;
4193
4194 if (to_journal &&
dfc08a12 4195 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 4196 to_journal = false;
4197
b7ad7512
CY		 4198 if (to_journal) {
4199 down_write(&curseg->journal_rwsem);
4200 } else {
184a5cd2
CY		 4201 page = get_next_sit_page(sbi, start_segno);
4202 raw_sit = page_address(page);
351df4b2 4203 }
351df4b2 4204
184a5cd2
CY		 4205 /* flush dirty sit entries in region of current sit set */
4206 for_each_set_bit_from(segno, bitmap, end) {
4207 int offset, sit_offset;
4b2fecc8
JK		 4208
4209 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 4210#ifdef CONFIG_F2FS_CHECK_FS
4211 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4212 SIT_VBLOCK_MAP_SIZE))
4213 f2fs_bug_on(sbi, 1);
4214#endif
184a5cd2
CY		 4215
4216 /* add discard candidates */
c473f1a9 4217 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 4218 cpc->trim_start = segno;
25290fa5 4219 add_discard_addrs(sbi, cpc, false);
4b2fecc8 4220 }
184a5cd2
CY		 4221
4222 if (to_journal) {
4d57b86d 4223 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 4224 SIT_JOURNAL, segno, 1);
4225 f2fs_bug_on(sbi, offset < 0);
dfc08a12 4226 segno_in_journal(journal, offset) =
184a5cd2
CY		 4227 cpu_to_le32(segno);
4228 seg_info_to_raw_sit(se,
dfc08a12 4229 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 4230 check_block_count(sbi, segno,
4231 &sit_in_journal(journal, offset));
184a5cd2
CY		 4232 } else {
4233 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4234 seg_info_to_raw_sit(se,
4235 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 4236 check_block_count(sbi, segno,
4237 &raw_sit->entries[sit_offset]);
184a5cd2 4238 }
351df4b2 4239
184a5cd2
CY		 4240 __clear_bit(segno, bitmap);
4241 sit_i->dirty_sentries--;
4242 ses->entry_cnt--;
351df4b2
JK		 4243 }
4244
b7ad7512
CY		 4245 if (to_journal)
4246 up_write(&curseg->journal_rwsem);
4247 else
184a5cd2
CY		 4248 f2fs_put_page(page, 1);
4249
4250 f2fs_bug_on(sbi, ses->entry_cnt);
4251 release_sit_entry_set(ses);
351df4b2 4252 }
184a5cd2
CY		 4253
4254 f2fs_bug_on(sbi, !list_empty(head));
4255 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 4256out:
c473f1a9 4257 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 4258 __u64 trim_start = cpc->trim_start;
4259
4b2fecc8 4260 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 4261 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 4262
4263 cpc->trim_start = trim_start;
4b2fecc8 4264 }
3d26fa6b 4265 up_write(&sit_i->sentry_lock);
351df4b2 4266
351df4b2
JK		 4267 set_prefree_as_free_segments(sbi);
4268}
4269
4270static int build_sit_info(struct f2fs_sb_info *sbi)
4271{
4272 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 4273 struct sit_info *sit_i;
4274 unsigned int sit_segs, start;
2fde3dd1 4275 char *src_bitmap, *bitmap;
bbf9f7d9 4276 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
351df4b2
JK		 4277
4278 /* allocate memory for SIT information */
acbf054d 4279 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 4280 if (!sit_i)
4281 return -ENOMEM;
4282
4283 SM_I(sbi)->sit_info = sit_i;
4284
9d2a789c
KC		 4285 sit_i->sentries =
4286 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4287 MAIN_SEGS(sbi)),
4288 GFP_KERNEL);
351df4b2
JK		 4289 if (!sit_i->sentries)
4290 return -ENOMEM;
4291
bbf9f7d9
ST		 4292 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4293 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
628b3d14 4294 GFP_KERNEL);
351df4b2
JK		 4295 if (!sit_i->dirty_sentries_bitmap)
4296 return -ENOMEM;
4297
2fde3dd1
CY		 4298#ifdef CONFIG_F2FS_CHECK_FS
4299 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4300#else
4301 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4302#endif
4303 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4304 if (!sit_i->bitmap)
4305 return -ENOMEM;
4306
4307 bitmap = sit_i->bitmap;
4308
7cd8558b 4309 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2fde3dd1
CY		 4310 sit_i->sentries[start].cur_valid_map = bitmap;
4311 bitmap += SIT_VBLOCK_MAP_SIZE;
4312
4313 sit_i->sentries[start].ckpt_valid_map = bitmap;
4314 bitmap += SIT_VBLOCK_MAP_SIZE;
3e025740 4315
355e7891 4316#ifdef CONFIG_F2FS_CHECK_FS
2fde3dd1
CY		 4317 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4318 bitmap += SIT_VBLOCK_MAP_SIZE;
355e7891
CY		 4319#endif
4320
2fde3dd1
CY		 4321 sit_i->sentries[start].discard_map = bitmap;
4322 bitmap += SIT_VBLOCK_MAP_SIZE;
351df4b2
JK		 4323 }
4324
acbf054d 4325 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 4326 if (!sit_i->tmp_map)
4327 return -ENOMEM;
4328
2c70c5e3 4329 if (__is_large_section(sbi)) {
9d2a789c
KC		 4330 sit_i->sec_entries =
4331 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4332 MAIN_SECS(sbi)),
4333 GFP_KERNEL);
351df4b2
JK		 4334 if (!sit_i->sec_entries)
4335 return -ENOMEM;
4336 }
4337
4338 /* get information related with SIT */
4339 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4340
4341 /* setup SIT bitmap from ckeckpoint pack */
bbf9f7d9 4342 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
351df4b2
JK		 4343 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4344
bbf9f7d9 4345 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
ae27d62e 4346 if (!sit_i->sit_bitmap)
351df4b2 4347 return -ENOMEM;
351df4b2 4348
ae27d62e 4349#ifdef CONFIG_F2FS_CHECK_FS
bbf9f7d9
ST		 4350 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4351 sit_bitmap_size, GFP_KERNEL);
ae27d62e
CY		 4352 if (!sit_i->sit_bitmap_mir)
4353 return -ENOMEM;
bbf9f7d9
ST		 4354
4355 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4356 main_bitmap_size, GFP_KERNEL);
4357 if (!sit_i->invalid_segmap)
4358 return -ENOMEM;
ae27d62e
CY		 4359#endif
4360
351df4b2
JK		 4361 /* init SIT information */
4362 sit_i->s_ops = &default_salloc_ops;
4363
4364 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4365 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 4366 sit_i->written_valid_blocks = 0;
bbf9f7d9 4367 sit_i->bitmap_size = sit_bitmap_size;
351df4b2
JK		 4368 sit_i->dirty_sentries = 0;
4369 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4370 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
a7e679b5 4371 sit_i->mounted_time = ktime_get_boottime_seconds();
3d26fa6b 4372 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 4373 return 0;
4374}
4375
4376static int build_free_segmap(struct f2fs_sb_info *sbi)
4377{
351df4b2
JK		 4378 struct free_segmap_info *free_i;
4379 unsigned int bitmap_size, sec_bitmap_size;
4380
4381 /* allocate memory for free segmap information */
acbf054d 4382 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 4383 if (!free_i)
4384 return -ENOMEM;
4385
4386 SM_I(sbi)->free_info = free_i;
4387
7cd8558b 4388 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 4389 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 4390 if (!free_i->free_segmap)
4391 return -ENOMEM;
4392
7cd8558b 4393 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 4394 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 4395 if (!free_i->free_secmap)
4396 return -ENOMEM;
4397
4398 /* set all segments as dirty temporarily */
4399 memset(free_i->free_segmap, 0xff, bitmap_size);
4400 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4401
4402 /* init free segmap information */
7cd8558b 4403 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 4404 free_i->free_segments = 0;
4405 free_i->free_sections = 0;
1a118ccf 4406 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 4407 return 0;
4408}
4409
4410static int build_curseg(struct f2fs_sb_info *sbi)
4411{
1042d60f 4412 struct curseg_info *array;
351df4b2
JK		 4413 int i;
4414
d0b9e42a
CY		 4415 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4416 sizeof(*array)), GFP_KERNEL);
351df4b2
JK		 4417 if (!array)
4418 return -ENOMEM;
4419
4420 SM_I(sbi)->curseg_array = array;
4421
d0b9e42a 4422 for (i = 0; i < NO_CHECK_TYPE; i++) {
351df4b2 4423 mutex_init(&array[i].curseg_mutex);
acbf054d 4424 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 4425 if (!array[i].sum_blk)
4426 return -ENOMEM;
b7ad7512 4427 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 4428 array[i].journal = f2fs_kzalloc(sbi,
4429 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 4430 if (!array[i].journal)
4431 return -ENOMEM;
d0b9e42a
CY		 4432 if (i < NR_PERSISTENT_LOG)
4433 array[i].seg_type = CURSEG_HOT_DATA + i;
4434 else if (i == CURSEG_COLD_DATA_PINNED)
4435 array[i].seg_type = CURSEG_COLD_DATA;
093749e2
CY		 4436 else if (i == CURSEG_ALL_DATA_ATGC)
4437 array[i].seg_type = CURSEG_COLD_DATA;
351df4b2
JK		 4438 array[i].segno = NULL_SEGNO;
4439 array[i].next_blkoff = 0;
d0b9e42a 4440 array[i].inited = false;
351df4b2
JK		 4441 }
4442 return restore_curseg_summaries(sbi);
4443}
4444
c39a1b34 4445static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 4446{
4447 struct sit_info *sit_i = SIT_I(sbi);
4448 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 4449 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 4450 struct seg_entry *se;
4451 struct f2fs_sit_entry sit;
74de593a
CY		 4452 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4453 unsigned int i, start, end;
4454 unsigned int readed, start_blk = 0;
c39a1b34 4455 int err = 0;
8a29c126 4456 block_t total_node_blocks = 0;
351df4b2 4457
74de593a 4458 do {
a8affc03 4459 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
664ba972 4460 META_SIT, true);
74de593a
CY		 4461
4462 start = start_blk * sit_i->sents_per_block;
4463 end = (start_blk + readed) * sit_i->sents_per_block;
4464
7cd8558b 4465 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 4466 struct f2fs_sit_block *sit_blk;
74de593a
CY		 4467 struct page *page;
4468
9c094040 4469 se = &sit_i->sentries[start];
74de593a 4470 page = get_current_sit_page(sbi, start);
edc55aaf
JK		 4471 if (IS_ERR(page))
4472 return PTR_ERR(page);
74de593a
CY		 4473 sit_blk = (struct f2fs_sit_block *)page_address(page);
4474 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4475 f2fs_put_page(page, 1);
d600af23 4476
c39a1b34
JK		 4477 err = check_block_count(sbi, start, &sit);
4478 if (err)
4479 return err;
74de593a 4480 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4481 if (IS_NODESEG(se->type))
4482 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 4483
4484 /* build discard map only one time */
7d20c8ab
CY		 4485 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4486 memset(se->discard_map, 0xff,
4487 SIT_VBLOCK_MAP_SIZE);
4488 } else {
4489 memcpy(se->discard_map,
4490 se->cur_valid_map,
4491 SIT_VBLOCK_MAP_SIZE);
4492 sbi->discard_blks +=
4493 sbi->blocks_per_seg -
4494 se->valid_blocks;
3e025740 4495 }
a66cdd98 4496
2c70c5e3 4497 if (__is_large_section(sbi))
d600af23
CY		 4498 get_sec_entry(sbi, start)->valid_blocks +=
4499 se->valid_blocks;
351df4b2 4500 }
74de593a
CY		 4501 start_blk += readed;
4502 } while (start_blk < sit_blk_cnt);
d600af23
CY		 4503
4504 down_read(&curseg->journal_rwsem);
4505 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 4506 unsigned int old_valid_blocks;
4507
4508 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f 4509 if (start >= MAIN_SEGS(sbi)) {
dcbb4c10
JP		 4510 f2fs_err(sbi, "Wrong journal entry on segno %u",
4511 start);
10f966bb 4512 err = -EFSCORRUPTED;
b2ca374f
JK		 4513 break;
4514 }
4515
d600af23
CY		 4516 se = &sit_i->sentries[start];
4517 sit = sit_in_journal(journal, i);
4518
4519 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 4520 if (IS_NODESEG(se->type))
4521 total_node_blocks -= old_valid_blocks;
d600af23 4522
c39a1b34
JK		 4523 err = check_block_count(sbi, start, &sit);
4524 if (err)
4525 break;
d600af23 4526 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4527 if (IS_NODESEG(se->type))
4528 total_node_blocks += se->valid_blocks;
d600af23 4529
7d20c8ab
CY		 4530 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4531 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4532 } else {
4533 memcpy(se->discard_map, se->cur_valid_map,
4534 SIT_VBLOCK_MAP_SIZE);
4535 sbi->discard_blks += old_valid_blocks;
4536 sbi->discard_blks -= se->valid_blocks;
d600af23
CY		 4537 }
4538
2c70c5e3 4539 if (__is_large_section(sbi)) {
d600af23 4540 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 4541 se->valid_blocks;
4542 get_sec_entry(sbi, start)->valid_blocks -=
4543 old_valid_blocks;
4544 }
d600af23
CY		 4545 }
4546 up_read(&curseg->journal_rwsem);
8a29c126
JK		 4547
4548 if (!err && total_node_blocks != valid_node_count(sbi)) {
dcbb4c10
JP		 4549 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4550 total_node_blocks, valid_node_count(sbi));
10f966bb 4551 err = -EFSCORRUPTED;
8a29c126
JK		 4552 }
4553
c39a1b34 4554 return err;
351df4b2
JK		 4555}
4556
4557static void init_free_segmap(struct f2fs_sb_info *sbi)
4558{
4559 unsigned int start;
4560 int type;
de881df9 4561 struct seg_entry *sentry;
351df4b2 4562
7cd8558b 4563 for (start = 0; start < MAIN_SEGS(sbi); start++) {
de881df9
AR		 4564 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4565 continue;
4566 sentry = get_seg_entry(sbi, start);
351df4b2
JK		 4567 if (!sentry->valid_blocks)
4568 __set_free(sbi, start);
c79b7ff1
JK		 4569 else
4570 SIT_I(sbi)->written_valid_blocks +=
4571 sentry->valid_blocks;
351df4b2
JK		 4572 }
4573
4574 /* set use the current segments */
4575 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4576 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
5f029c04 4577
351df4b2
JK		 4578 __set_test_and_inuse(sbi, curseg_t->segno);
4579 }
4580}
4581
4582static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4583{
4584 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4585 struct free_segmap_info *free_i = FREE_I(sbi);
da52f8ad 4586 unsigned int segno = 0, offset = 0, secno;
de881df9 4587 block_t valid_blocks, usable_blks_in_seg;
123aaf77 4588 block_t blks_per_sec = BLKS_PER_SEC(sbi);
351df4b2 4589
8736fbf0 4590 while (1) {
351df4b2 4591 /* find dirty segment based on free segmap */
7cd8558b
JK		 4592 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4593 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 4594 break;
4595 offset = segno + 1;
302bd348 4596 valid_blocks = get_valid_blocks(sbi, segno, false);
de881df9
AR		 4597 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4598 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
351df4b2 4599 continue;
de881df9 4600 if (valid_blocks > usable_blks_in_seg) {
ec325b52
JK		 4601 f2fs_bug_on(sbi, 1);
4602 continue;
4603 }
351df4b2
JK		 4604 mutex_lock(&dirty_i->seglist_lock);
4605 __locate_dirty_segment(sbi, segno, DIRTY);
4606 mutex_unlock(&dirty_i->seglist_lock);
4607 }
da52f8ad
JQ		 4608
4609 if (!__is_large_section(sbi))
4610 return;
4611
4612 mutex_lock(&dirty_i->seglist_lock);
5335bfc6 4613 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
da52f8ad
JQ		 4614 valid_blocks = get_valid_blocks(sbi, segno, true);
4615 secno = GET_SEC_FROM_SEG(sbi, segno);
4616
4617 if (!valid_blocks || valid_blocks == blks_per_sec)
4618 continue;
4619 if (IS_CURSEC(sbi, secno))
4620 continue;
4621 set_bit(secno, dirty_i->dirty_secmap);
4622 }
4623 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 4624}
4625
5ec4e49f 4626static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4627{
4628 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 4629 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 4630
628b3d14 4631 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 4632 if (!dirty_i->victim_secmap)
351df4b2
JK		 4633 return -ENOMEM;
4634 return 0;
4635}
4636
4637static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4638{
4639 struct dirty_seglist_info *dirty_i;
4640 unsigned int bitmap_size, i;
4641
4642 /* allocate memory for dirty segments list information */
acbf054d
CY		 4643 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4644 GFP_KERNEL);
351df4b2
JK		 4645 if (!dirty_i)
4646 return -ENOMEM;
4647
4648 SM_I(sbi)->dirty_info = dirty_i;
4649 mutex_init(&dirty_i->seglist_lock);
4650
7cd8558b 4651 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 4652
4653 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 4654 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4655 GFP_KERNEL);
351df4b2
JK		 4656 if (!dirty_i->dirty_segmap[i])
4657 return -ENOMEM;
4658 }
4659
da52f8ad
JQ		 4660 if (__is_large_section(sbi)) {
4661 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4662 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4663 bitmap_size, GFP_KERNEL);
4664 if (!dirty_i->dirty_secmap)
4665 return -ENOMEM;
4666 }
4667
351df4b2 4668 init_dirty_segmap(sbi);
5ec4e49f 4669 return init_victim_secmap(sbi);
351df4b2
JK		 4670}
4671
c854f4d6
CY		 4672static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4673{
4674 int i;
4675
4676 /*
4677 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4678 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4679 */
d0b9e42a 4680 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
c854f4d6
CY		 4681 struct curseg_info *curseg = CURSEG_I(sbi, i);
4682 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4683 unsigned int blkofs = curseg->next_blkoff;
4684
093749e2
CY		 4685 sanity_check_seg_type(sbi, curseg->seg_type);
4686
c854f4d6
CY		 4687 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4688 goto out;
4689
4690 if (curseg->alloc_type == SSR)
4691 continue;
4692
4693 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4694 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4695 continue;
4696out:
dcbb4c10
JP		 4697 f2fs_err(sbi,
4698 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4699 i, curseg->segno, curseg->alloc_type,
4700 curseg->next_blkoff, blkofs);
10f966bb 4701 return -EFSCORRUPTED;
c854f4d6
CY		 4702 }
4703 }
4704 return 0;
4705}
4706
c426d991
SK		 4707#ifdef CONFIG_BLK_DEV_ZONED
4708
d508c94e
SK		 4709static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4710 struct f2fs_dev_info *fdev,
4711 struct blk_zone *zone)
4712{
4713 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4714 block_t zone_block, wp_block, last_valid_block;
4715 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4716 int i, s, b, ret;
4717 struct seg_entry *se;
4718
4719 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4720 return 0;
4721
4722 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4723 wp_segno = GET_SEGNO(sbi, wp_block);
4724 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4725 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4726 zone_segno = GET_SEGNO(sbi, zone_block);
4727 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4728
4729 if (zone_segno >= MAIN_SEGS(sbi))
4730 return 0;
4731
4732 /*
4733 * Skip check of zones cursegs point to, since
4734 * fix_curseg_write_pointer() checks them.
4735 */
4736 for (i = 0; i < NO_CHECK_TYPE; i++)
4737 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4738 CURSEG_I(sbi, i)->segno))
4739 return 0;
4740
4741 /*
4742 * Get last valid block of the zone.
4743 */
4744 last_valid_block = zone_block - 1;
4745 for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4746 segno = zone_segno + s;
4747 se = get_seg_entry(sbi, segno);
4748 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4749 if (f2fs_test_bit(b, se->cur_valid_map)) {
4750 last_valid_block = START_BLOCK(sbi, segno) + b;
4751 break;
4752 }
4753 if (last_valid_block >= zone_block)
4754 break;
4755 }
4756
4757 /*
4758 * If last valid block is beyond the write pointer, report the
4759 * inconsistency. This inconsistency does not cause write error
4760 * because the zone will not be selected for write operation until
4761 * it get discarded. Just report it.
4762 */
4763 if (last_valid_block >= wp_block) {
4764 f2fs_notice(sbi, "Valid block beyond write pointer: "
4765 "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4766 GET_SEGNO(sbi, last_valid_block),
4767 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4768 wp_segno, wp_blkoff);
4769 return 0;
4770 }
4771
4772 /*
4773 * If there is no valid block in the zone and if write pointer is
4774 * not at zone start, reset the write pointer.
4775 */
4776 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4777 f2fs_notice(sbi,
4778 "Zone without valid block has non-zero write "
4779 "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4780 wp_segno, wp_blkoff);
4781 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4782 zone->len >> log_sectors_per_block);
4783 if (ret) {
4784 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4785 fdev->path, ret);
4786 return ret;
4787 }
4788 }
4789
4790 return 0;
4791}
4792
c426d991
SK		 4793static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4794 block_t zone_blkaddr)
4795{
4796 int i;
4797
4798 for (i = 0; i < sbi->s_ndevs; i++) {
4799 if (!bdev_is_zoned(FDEV(i).bdev))
4800 continue;
4801 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4802 zone_blkaddr <= FDEV(i).end_blk))
4803 return &FDEV(i);
4804 }
4805
4806 return NULL;
4807}
4808
4809static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
5f029c04
YZ		 4810 void *data)
4811{
c426d991
SK		 4812 memcpy(data, zone, sizeof(struct blk_zone));
4813 return 0;
4814}
4815
4816static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4817{
4818 struct curseg_info *cs = CURSEG_I(sbi, type);
4819 struct f2fs_dev_info *zbd;
4820 struct blk_zone zone;
4821 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4822 block_t cs_zone_block, wp_block;
4823 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4824 sector_t zone_sector;
4825 int err;
4826
4827 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4828 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4829
4830 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4831 if (!zbd)
4832 return 0;
4833
4834 /* report zone for the sector the curseg points to */
4835 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4836 << log_sectors_per_block;
4837 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4838 report_one_zone_cb, &zone);
4839 if (err != 1) {
4840 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4841 zbd->path, err);
4842 return err;
4843 }
4844
4845 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4846 return 0;
4847
4848 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4849 wp_segno = GET_SEGNO(sbi, wp_block);
4850 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4851 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4852
4853 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4854 wp_sector_off == 0)
4855 return 0;
4856
4857 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4858 "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4859 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4860
4861 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4862 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
509f1010
CY		 4863
4864 f2fs_allocate_new_section(sbi, type, true);
c426d991 4865
d508c94e
SK		 4866 /* check consistency of the zone curseg pointed to */
4867 if (check_zone_write_pointer(sbi, zbd, &zone))
4868 return -EIO;
4869
c426d991
SK		 4870 /* check newly assigned zone */
4871 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4872 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4873
4874 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4875 if (!zbd)
4876 return 0;
4877
4878 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4879 << log_sectors_per_block;
4880 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4881 report_one_zone_cb, &zone);
4882 if (err != 1) {
4883 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4884 zbd->path, err);
4885 return err;
4886 }
4887
4888 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4889 return 0;
4890
4891 if (zone.wp != zone.start) {
4892 f2fs_notice(sbi,
4893 "New zone for curseg[%d] is not yet discarded. "
4894 "Reset the zone: curseg[0x%x,0x%x]",
4895 type, cs->segno, cs->next_blkoff);
4896 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4897 zone_sector >> log_sectors_per_block,
4898 zone.len >> log_sectors_per_block);
4899 if (err) {
4900 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4901 zbd->path, err);
4902 return err;
4903 }
4904 }
4905
4906 return 0;
4907}
4908
4909int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4910{
4911 int i, ret;
4912
d0b9e42a 4913 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
c426d991
SK		 4914 ret = fix_curseg_write_pointer(sbi, i);
4915 if (ret)
4916 return ret;
4917 }
4918
4919 return 0;
4920}
d508c94e
SK		 4921
4922struct check_zone_write_pointer_args {
4923 struct f2fs_sb_info *sbi;
4924 struct f2fs_dev_info *fdev;
4925};
4926
4927static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5f029c04
YZ		 4928 void *data)
4929{
d508c94e 4930 struct check_zone_write_pointer_args *args;
5f029c04 4931
d508c94e
SK		 4932 args = (struct check_zone_write_pointer_args *)data;
4933
4934 return check_zone_write_pointer(args->sbi, args->fdev, zone);
4935}
4936
4937int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4938{
4939 int i, ret;
4940 struct check_zone_write_pointer_args args;
4941
4942 for (i = 0; i < sbi->s_ndevs; i++) {
4943 if (!bdev_is_zoned(FDEV(i).bdev))
4944 continue;
4945
4946 args.sbi = sbi;
4947 args.fdev = &FDEV(i);
4948 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4949 check_zone_write_pointer_cb, &args);
4950 if (ret < 0)
4951 return ret;
4952 }
4953
4954 return 0;
4955}
de881df9
AR		 4956
4957static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4958 unsigned int dev_idx)
4959{
4960 if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4961 return true;
4962 return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4963}
4964
4965/* Return the zone index in the given device */
4966static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4967 int dev_idx)
4968{
4969 block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4970
4971 return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4972 sbi->log_blocks_per_blkz;
4973}
4974
4975/*
4976 * Return the usable segments in a section based on the zone's
4977 * corresponding zone capacity. Zone is equal to a section.
4978 */
4979static inline unsigned int f2fs_usable_zone_segs_in_sec(
4980 struct f2fs_sb_info *sbi, unsigned int segno)
4981{
4982 unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
4983
4984 dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4985 zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4986
4987 /* Conventional zone's capacity is always equal to zone size */
4988 if (is_conv_zone(sbi, zone_idx, dev_idx))
4989 return sbi->segs_per_sec;
4990
4991 /*
4992 * If the zone_capacity_blocks array is NULL, then zone capacity
4993 * is equal to the zone size for all zones
4994 */
4995 if (!FDEV(dev_idx).zone_capacity_blocks)
4996 return sbi->segs_per_sec;
4997
4998 /* Get the segment count beyond zone capacity block */
4999 unusable_segs_in_sec = (sbi->blocks_per_blkz -
5000 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
5001 sbi->log_blocks_per_seg;
5002 return sbi->segs_per_sec - unusable_segs_in_sec;
5003}
5004
5005/*
5006 * Return the number of usable blocks in a segment. The number of blocks
5007 * returned is always equal to the number of blocks in a segment for
5008 * segments fully contained within a sequential zone capacity or a
5009 * conventional zone. For segments partially contained in a sequential
5010 * zone capacity, the number of usable blocks up to the zone capacity
5011 * is returned. 0 is returned in all other cases.
5012 */
5013static inline unsigned int f2fs_usable_zone_blks_in_seg(
5014 struct f2fs_sb_info *sbi, unsigned int segno)
5015{
5016 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5017 unsigned int zone_idx, dev_idx, secno;
5018
5019 secno = GET_SEC_FROM_SEG(sbi, segno);
5020 seg_start = START_BLOCK(sbi, segno);
5021 dev_idx = f2fs_target_device_index(sbi, seg_start);
5022 zone_idx = get_zone_idx(sbi, secno, dev_idx);
5023
5024 /*
5025 * Conventional zone's capacity is always equal to zone size,
5026 * so, blocks per segment is unchanged.
5027 */
5028 if (is_conv_zone(sbi, zone_idx, dev_idx))
5029 return sbi->blocks_per_seg;
5030
5031 if (!FDEV(dev_idx).zone_capacity_blocks)
5032 return sbi->blocks_per_seg;
5033
5034 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5035 sec_cap_blkaddr = sec_start_blkaddr +
5036 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
5037
5038 /*
5039 * If segment starts before zone capacity and spans beyond
5040 * zone capacity, then usable blocks are from seg start to
5041 * zone capacity. If the segment starts after the zone capacity,
5042 * then there are no usable blocks.
5043 */
5044 if (seg_start >= sec_cap_blkaddr)
5045 return 0;
5046 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5047 return sec_cap_blkaddr - seg_start;
5048
5049 return sbi->blocks_per_seg;
5050}
c426d991
SK		 5051#else
5052int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5053{
5054 return 0;
5055}
d508c94e
SK		 5056
5057int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5058{
5059 return 0;
5060}
de881df9
AR		 5061
5062static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5063 unsigned int segno)
5064{
5065 return 0;
5066}
5067
5068static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5069 unsigned int segno)
5070{
5071 return 0;
5072}
c426d991 5073#endif
de881df9
AR		 5074unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5075 unsigned int segno)
5076{
5077 if (f2fs_sb_has_blkzoned(sbi))
5078 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5079
5080 return sbi->blocks_per_seg;
5081}
5082
5083unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5084 unsigned int segno)
5085{
5086 if (f2fs_sb_has_blkzoned(sbi))
5087 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5088
5089 return sbi->segs_per_sec;
5090}
c426d991 5091
0a8165d7 5092/*
351df4b2
JK		 5093 * Update min, max modified time for cost-benefit GC algorithm
5094 */
5095static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5096{
5097 struct sit_info *sit_i = SIT_I(sbi);
5098 unsigned int segno;
5099
3d26fa6b 5100 down_write(&sit_i->sentry_lock);
351df4b2 5101
5ad25442 5102 sit_i->min_mtime = ULLONG_MAX;
351df4b2 5103
7cd8558b 5104 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 5105 unsigned int i;
5106 unsigned long long mtime = 0;
5107
5108 for (i = 0; i < sbi->segs_per_sec; i++)
5109 mtime += get_seg_entry(sbi, segno + i)->mtime;
5110
5111 mtime = div_u64(mtime, sbi->segs_per_sec);
5112
5113 if (sit_i->min_mtime > mtime)
5114 sit_i->min_mtime = mtime;
5115 }
a1f72ac2 5116 sit_i->max_mtime = get_mtime(sbi, false);
093749e2 5117 sit_i->dirty_max_mtime = 0;
3d26fa6b 5118 up_write(&sit_i->sentry_lock);
351df4b2
JK		 5119}
5120
4d57b86d 5121int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 5122{
5123 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5124 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 5125 struct f2fs_sm_info *sm_info;
351df4b2
JK		 5126 int err;
5127
acbf054d 5128 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 5129 if (!sm_info)
5130 return -ENOMEM;
5131
5132 /* init sm info */
5133 sbi->sm_info = sm_info;
351df4b2
JK		 5134 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5135 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5136 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5137 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5138 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5139 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5140 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 5141 sm_info->rec_prefree_segments = sm_info->main_segments *
5142 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 5143 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5144 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5145
b0332a0f 5146 if (!f2fs_lfs_mode(sbi))
52763a4b 5147 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 5148 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 5149 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
853137ce 5150 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
ef095d19 5151 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 5152 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 5153
184a5cd2
CY		 5154 INIT_LIST_HEAD(&sm_info->sit_entry_set);
5155
2b60311d
CY		 5156 init_rwsem(&sm_info->curseg_lock);
5157
d4fdf8ba 5158 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 5159 err = f2fs_create_flush_cmd_control(sbi);
2163d198 5160 if (err)
a688b9d9 5161 return err;
6b4afdd7
JK		 5162 }
5163
0b54fb84
JK		 5164 err = create_discard_cmd_control(sbi);
5165 if (err)
5166 return err;
5167
351df4b2
JK		 5168 err = build_sit_info(sbi);
5169 if (err)
5170 return err;
5171 err = build_free_segmap(sbi);
5172 if (err)
5173 return err;
5174 err = build_curseg(sbi);
5175 if (err)
5176 return err;
5177
5178 /* reinit free segmap based on SIT */
c39a1b34
JK		 5179 err = build_sit_entries(sbi);
5180 if (err)
5181 return err;
351df4b2
JK		 5182
5183 init_free_segmap(sbi);
5184 err = build_dirty_segmap(sbi);
5185 if (err)
5186 return err;
5187
c854f4d6
CY		 5188 err = sanity_check_curseg(sbi);
5189 if (err)
5190 return err;
5191
351df4b2
JK		 5192 init_min_max_mtime(sbi);
5193 return 0;
5194}
5195
5196static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5197 enum dirty_type dirty_type)
5198{
5199 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5200
5201 mutex_lock(&dirty_i->seglist_lock);
39307a8e 5202 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 5203 dirty_i->nr_dirty[dirty_type] = 0;
5204 mutex_unlock(&dirty_i->seglist_lock);
5205}
5206
5ec4e49f 5207static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 5208{
5209 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5f029c04 5210
39307a8e 5211 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 5212}
5213
5214static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5215{
5216 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5217 int i;
5218
5219 if (!dirty_i)
5220 return;
5221
5222 /* discard pre-free/dirty segments list */
5223 for (i = 0; i < NR_DIRTY_TYPE; i++)
5224 discard_dirty_segmap(sbi, i);
5225
da52f8ad
JQ		 5226 if (__is_large_section(sbi)) {
5227 mutex_lock(&dirty_i->seglist_lock);
5228 kvfree(dirty_i->dirty_secmap);
5229 mutex_unlock(&dirty_i->seglist_lock);
5230 }
5231
5ec4e49f 5232 destroy_victim_secmap(sbi);
351df4b2 5233 SM_I(sbi)->dirty_info = NULL;
c8eb7024 5234 kfree(dirty_i);
351df4b2
JK		 5235}
5236
5237static void destroy_curseg(struct f2fs_sb_info *sbi)
5238{
5239 struct curseg_info *array = SM_I(sbi)->curseg_array;
5240 int i;
5241
5242 if (!array)
5243 return;
5244 SM_I(sbi)->curseg_array = NULL;
b7ad7512 5245 for (i = 0; i < NR_CURSEG_TYPE; i++) {
c8eb7024
CY		 5246 kfree(array[i].sum_blk);
5247 kfree(array[i].journal);
b7ad7512 5248 }
c8eb7024 5249 kfree(array);
351df4b2
JK		 5250}
5251
5252static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5253{
5254 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5f029c04 5255
351df4b2
JK		 5256 if (!free_i)
5257 return;
5258 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 5259 kvfree(free_i->free_segmap);
5260 kvfree(free_i->free_secmap);
c8eb7024 5261 kfree(free_i);
351df4b2
JK		 5262}
5263
5264static void destroy_sit_info(struct f2fs_sb_info *sbi)
5265{
5266 struct sit_info *sit_i = SIT_I(sbi);
351df4b2
JK		 5267
5268 if (!sit_i)
5269 return;
5270
2fde3dd1
CY		 5271 if (sit_i->sentries)
5272 kvfree(sit_i->bitmap);
c8eb7024 5273 kfree(sit_i->tmp_map);
60a3b782 5274
39307a8e
JK		 5275 kvfree(sit_i->sentries);
5276 kvfree(sit_i->sec_entries);
5277 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 5278
5279 SM_I(sbi)->sit_info = NULL;
5222595d 5280 kvfree(sit_i->sit_bitmap);
ae27d62e 5281#ifdef CONFIG_F2FS_CHECK_FS
5222595d 5282 kvfree(sit_i->sit_bitmap_mir);
bbf9f7d9 5283 kvfree(sit_i->invalid_segmap);
ae27d62e 5284#endif
c8eb7024 5285 kfree(sit_i);
351df4b2
JK		 5286}
5287
4d57b86d 5288void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 5289{
5290 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 5291
3b03f724
CY		 5292 if (!sm_info)
5293 return;
4d57b86d 5294 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 5295 destroy_discard_cmd_control(sbi);
351df4b2
JK		 5296 destroy_dirty_segmap(sbi);
5297 destroy_curseg(sbi);
5298 destroy_free_segmap(sbi);
5299 destroy_sit_info(sbi);
5300 sbi->sm_info = NULL;
c8eb7024 5301 kfree(sm_info);
351df4b2 5302}
7fd9e544 5303
4d57b86d 5304int __init f2fs_create_segment_manager_caches(void)
7fd9e544 5305{
98510003 5306 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
e8512d2e 5307 sizeof(struct discard_entry));
7fd9e544 5308 if (!discard_entry_slab)
184a5cd2
CY		 5309 goto fail;
5310
98510003 5311 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
b01a9201
JK		 5312 sizeof(struct discard_cmd));
5313 if (!discard_cmd_slab)
6ab2a308 5314 goto destroy_discard_entry;
275b66b0 5315
98510003 5316 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
c9ee0085 5317 sizeof(struct sit_entry_set));
184a5cd2 5318 if (!sit_entry_set_slab)
b01a9201 5319 goto destroy_discard_cmd;
88b88a66 5320
98510003 5321 inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
88b88a66
JK		 5322 sizeof(struct inmem_pages));
5323 if (!inmem_entry_slab)
5324 goto destroy_sit_entry_set;
7fd9e544 5325 return 0;
184a5cd2 5326
88b88a66
JK		 5327destroy_sit_entry_set:
5328 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 5329destroy_discard_cmd:
5330 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 5331destroy_discard_entry:
184a5cd2
CY		 5332 kmem_cache_destroy(discard_entry_slab);
5333fail:
5334 return -ENOMEM;
7fd9e544
JK		 5335}
5336
4d57b86d 5337void f2fs_destroy_segment_manager_caches(void)
7fd9e544 5338{
184a5cd2 5339 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 5340 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 5341 kmem_cache_destroy(discard_entry_slab);
88b88a66 5342 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 5343}
Linux 6.x block layer and io_uring tree(s)