]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: clean up with IS_INODE()
[mirror_ubuntu-jammy-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL		 29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
b01a9201 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK		 36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL		 51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK		 69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL		 75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL		 80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL		 85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK		 95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL		 112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL		 130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
80609448
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
4d57b86d 172bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
b3a97a2a
JK		 173{
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
5b0e9539 180 if (sbi->gc_mode == GC_URGENT)
b3a97a2a
JK		 181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
a2a12b67 184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
b3a97a2a
JK		 185}
186
4d57b86d 187void f2fs_register_inmem_page(struct inode *inode, struct page *page)
88b88a66 188{
57864ae5 189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
88b88a66
JK		 190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
9be32d72 192
9e4ded3f 193 f2fs_trace_pid(page);
0722b101 194
decd36b6
CY		 195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
197
88b88a66
JK		 198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
decd36b6 203
88b88a66
JK		 204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
57864ae5
JK		 208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
8dcf2ff7 212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 213 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 214
215 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 216}
217
28bc106b
CY		 218static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover)
29b96b54 220{
28bc106b 221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 222 struct inmem_pages *cur, *tmp;
28bc106b 223 int err = 0;
29b96b54
CY		 224
225 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 226 struct page *page = cur->page;
227
228 if (drop)
229 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
230
231 lock_page(page);
29b96b54 232
e5e5732d
CY		 233 f2fs_wait_on_page_writeback(page, DATA, true);
234
28bc106b
CY		 235 if (recover) {
236 struct dnode_of_data dn;
237 struct node_info ni;
238
239 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 240retry:
28bc106b 241 set_new_dnode(&dn, inode, NULL, NULL, 0);
4d57b86d
CY		 242 err = f2fs_get_dnode_of_data(&dn, page->index,
243 LOOKUP_NODE);
7f2b4e8e
CY		 244 if (err) {
245 if (err == -ENOMEM) {
246 congestion_wait(BLK_RW_ASYNC, HZ/50);
247 cond_resched();
248 goto retry;
249 }
28bc106b
CY		 250 err = -EAGAIN;
251 goto next;
252 }
4d57b86d 253 f2fs_get_node_info(sbi, dn.nid, &ni);
f1d2564a 254 if (cur->old_addr == NEW_ADDR) {
4d57b86d 255 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
f1d2564a
DJ		 256 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
257 } else
258 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 259 cur->old_addr, ni.version, true, true);
260 f2fs_put_dnode(&dn);
261 }
262next:
63c52d78
JK		 263 /* we don't need to invalidate this in the sccessful status */
264 if (drop || recover)
265 ClearPageUptodate(page);
28bc106b 266 set_page_private(page, 0);
c81ced05 267 ClearPagePrivate(page);
28bc106b 268 f2fs_put_page(page, 1);
29b96b54
CY		 269
270 list_del(&cur->list);
271 kmem_cache_free(inmem_entry_slab, cur);
272 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
273 }
28bc106b 274 return err;
29b96b54
CY		 275}
276
4d57b86d 277void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
57864ae5
JK		 278{
279 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
280 struct inode *inode;
281 struct f2fs_inode_info *fi;
282next:
283 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
284 if (list_empty(head)) {
285 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
286 return;
287 }
288 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
289 inode = igrab(&fi->vfs_inode);
290 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
291
292 if (inode) {
2ef79ecb
CY		 293 if (gc_failure) {
294 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
295 goto drop;
296 goto skip;
297 }
298drop:
299 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
4d57b86d 300 f2fs_drop_inmem_pages(inode);
57864ae5
JK		 301 iput(inode);
302 }
2ef79ecb 303skip:
57864ae5
JK		 304 congestion_wait(BLK_RW_ASYNC, HZ/50);
305 cond_resched();
306 goto next;
307}
308
4d57b86d 309void f2fs_drop_inmem_pages(struct inode *inode)
29b96b54 310{
57864ae5 311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 312 struct f2fs_inode_info *fi = F2FS_I(inode);
313
314 mutex_lock(&fi->inmem_lock);
28bc106b 315 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 316 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
317 if (!list_empty(&fi->inmem_ilist))
318 list_del_init(&fi->inmem_ilist);
319 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 320 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 321
322 clear_inode_flag(inode, FI_ATOMIC_FILE);
2ef79ecb 323 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
5fe45743 324 stat_dec_atomic_write(inode);
29b96b54
CY		 325}
326
4d57b86d 327void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
8c242db9
JK		 328{
329 struct f2fs_inode_info *fi = F2FS_I(inode);
330 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
331 struct list_head *head = &fi->inmem_pages;
332 struct inmem_pages *cur = NULL;
333
334 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
335
336 mutex_lock(&fi->inmem_lock);
337 list_for_each_entry(cur, head, list) {
338 if (cur->page == page)
339 break;
340 }
341
d0891e84 342 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
8c242db9
JK		 343 list_del(&cur->list);
344 mutex_unlock(&fi->inmem_lock);
345
346 dec_page_count(sbi, F2FS_INMEM_PAGES);
347 kmem_cache_free(inmem_entry_slab, cur);
348
349 ClearPageUptodate(page);
350 set_page_private(page, 0);
351 ClearPagePrivate(page);
352 f2fs_put_page(page, 0);
353
354 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
355}
356
4d57b86d 357static int __f2fs_commit_inmem_pages(struct inode *inode)
88b88a66
JK		 358{
359 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
360 struct f2fs_inode_info *fi = F2FS_I(inode);
361 struct inmem_pages *cur, *tmp;
88b88a66 362 struct f2fs_io_info fio = {
05ca3632 363 .sbi = sbi,
39d787be 364 .ino = inode->i_ino,
88b88a66 365 .type = DATA,
04d328de 366 .op = REQ_OP_WRITE,
70fd7614 367 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 368 .io_type = FS_DATA_IO,
88b88a66 369 };
cf52b27a 370 struct list_head revoke_list;
942fd319 371 pgoff_t last_idx = ULONG_MAX;
edb27dee 372 int err = 0;
88b88a66 373
cf52b27a
CY		 374 INIT_LIST_HEAD(&revoke_list);
375
88b88a66 376 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 377 struct page *page = cur->page;
378
379 lock_page(page);
380 if (page->mapping == inode->i_mapping) {
381 trace_f2fs_commit_inmem_page(page, INMEM);
382
383 set_page_dirty(page);
384 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 385 if (clear_page_dirty_for_io(page)) {
29b96b54 386 inode_dec_dirty_pages(inode);
4d57b86d 387 f2fs_remove_dirty_inode(inode);
933439c8 388 }
640cc189 389retry:
28bc106b 390 fio.page = page;
e959c8f5 391 fio.old_blkaddr = NULL_ADDR;
4d978078 392 fio.encrypted_page = NULL;
cc15620b 393 fio.need_lock = LOCK_DONE;
4d57b86d 394 err = f2fs_do_write_data_page(&fio);
29b96b54 395 if (err) {
640cc189
JK		 396 if (err == -ENOMEM) {
397 congestion_wait(BLK_RW_ASYNC, HZ/50);
398 cond_resched();
399 goto retry;
400 }
28bc106b 401 unlock_page(page);
29b96b54 402 break;
70c640b1 403 }
28bc106b
CY		 404 /* record old blkaddr for revoking */
405 cur->old_addr = fio.old_blkaddr;
942fd319 406 last_idx = page->index;
28bc106b
CY		 407 }
408 unlock_page(page);
cf52b27a 409 list_move_tail(&cur->list, &revoke_list);
88b88a66 410 }
29b96b54 411
942fd319 412 if (last_idx != ULONG_MAX)
b9109b0e 413 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b 414
cf52b27a
CY		 415 if (err) {
416 /*
417 * try to revoke all committed pages, but still we could fail
418 * due to no memory or other reason, if that happened, EAGAIN
419 * will be returned, which means in such case, transaction is
420 * already not integrity, caller should use journal to do the
421 * recovery or rewrite & commit last transaction. For other
422 * error number, revoking was done by filesystem itself.
423 */
424 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
425
426 /* drop all uncommitted pages */
427 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
428 } else {
429 __revoke_inmem_pages(inode, &revoke_list, false, false);
430 }
28bc106b 431
29b96b54
CY		 432 return err;
433}
434
4d57b86d 435int f2fs_commit_inmem_pages(struct inode *inode)
29b96b54
CY		 436{
437 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
438 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b 439 int err;
29b96b54
CY		 440
441 f2fs_balance_fs(sbi, true);
442 f2fs_lock_op(sbi);
443
5fe45743
CY		 444 set_inode_flag(inode, FI_ATOMIC_COMMIT);
445
29b96b54 446 mutex_lock(&fi->inmem_lock);
4d57b86d 447 err = __f2fs_commit_inmem_pages(inode);
28bc106b 448
57864ae5
JK		 449 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
450 if (!list_empty(&fi->inmem_ilist))
451 list_del_init(&fi->inmem_ilist);
452 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 453 mutex_unlock(&fi->inmem_lock);
454
5fe45743
CY		 455 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
456
29b96b54 457 f2fs_unlock_op(sbi);
edb27dee 458 return err;
88b88a66
JK		 459}
460
0a8165d7 461/*
351df4b2
JK		 462 * This function balances dirty node and dentry pages.
463 * In addition, it controls garbage collection.
464 */
2c4db1a6 465void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 466{
0f348028 467#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 468 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
469 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 470 f2fs_stop_checkpoint(sbi, false);
55523519 471 }
0f348028
CY		 472#endif
473
e589c2c4 474 /* balance_fs_bg is able to be pending */
a7881893 475 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 476 f2fs_balance_fs_bg(sbi);
477
351df4b2 478 /*
029cd28c
JK		 479 * We should do GC or end up with checkpoint, if there are so many dirty
480 * dir/node pages without enough free segments.
351df4b2 481 */
7f3037a5 482 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 483 mutex_lock(&sbi->gc_mutex);
e066b83c 484 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 485 }
486}
487
4660f9c0
JK		 488void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
489{
64c74a7a
CY		 490 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
491 return;
492
1dcc336b 493 /* try to shrink extent cache when there is no enough memory */
4d57b86d 494 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
554df79e 495 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 496
1b38dc8e 497 /* check the # of cached NAT entries */
4d57b86d
CY		 498 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
499 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
1b38dc8e 500
4d57b86d
CY		 501 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
502 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
ad4edb83 503 else
4d57b86d 504 f2fs_build_free_nids(sbi, false, false);
31696580 505
1c0f4bf5 506 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 507 return;
31696580 508
1b38dc8e 509 /* checkpoint is the only way to shrink partial cached entries */
4d57b86d
CY		 510 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
511 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 512 excess_prefree_segs(sbi) ||
513 excess_dirty_nats(sbi) ||
f455c8a5 514 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 515 if (test_opt(sbi, DATA_FLUSH)) {
516 struct blk_plug plug;
517
518 blk_start_plug(&plug);
4d57b86d 519 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 520 blk_finish_plug(&plug);
521 }
4660f9c0 522 f2fs_sync_fs(sbi->sb, true);
42190d2a 523 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 524 }
4660f9c0
JK		 525}
526
20fda56b
KM		 527static int __submit_flush_wait(struct f2fs_sb_info *sbi,
528 struct block_device *bdev)
3c62be17 529{
d62fe971 530 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 531 int ret;
532
3adc5fcb 533 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 534 bio_set_dev(bio, bdev);
3c62be17
JK		 535 ret = submit_bio_wait(bio);
536 bio_put(bio);
20fda56b
KM		 537
538 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
539 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 540 return ret;
541}
542
39d787be 543static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 544{
39d787be 545 int ret = 0;
3c62be17
JK		 546 int i;
547
39d787be
CY		 548 if (!sbi->s_ndevs)
549 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 550
39d787be 551 for (i = 0; i < sbi->s_ndevs; i++) {
4d57b86d 552 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
39d787be 553 continue;
20fda56b
KM		 554 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
555 if (ret)
556 break;
3c62be17
JK		 557 }
558 return ret;
559}
560
2163d198 561static int issue_flush_thread(void *data)
6b4afdd7
JK		 562{
563 struct f2fs_sb_info *sbi = data;
b01a9201 564 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 565 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 566repeat:
567 if (kthread_should_stop())
568 return 0;
569
dc6febb6
CY		 570 sb_start_intwrite(sbi->sb);
571
721bd4d5 572 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 573 struct flush_cmd *cmd, *next;
574 int ret;
575
721bd4d5
GZ		 576 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
577 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
578
39d787be
CY		 579 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
580
581 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 582 atomic_inc(&fcc->issued_flush);
583
721bd4d5
GZ		 584 llist_for_each_entry_safe(cmd, next,
585 fcc->dispatch_list, llnode) {
6b4afdd7 586 cmd->ret = ret;
6b4afdd7
JK		 587 complete(&cmd->wait);
588 }
a688b9d9 589 fcc->dispatch_list = NULL;
6b4afdd7
JK		 590 }
591
dc6febb6
CY		 592 sb_end_intwrite(sbi->sb);
593
a688b9d9 594 wait_event_interruptible(*q,
721bd4d5 595 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 596 goto repeat;
597}
598
39d787be 599int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 600{
b01a9201 601 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 602 struct flush_cmd cmd;
8b8dd65f 603 int ret;
6b4afdd7 604
0f7b2abd
JK		 605 if (test_opt(sbi, NOBARRIER))
606 return 0;
607
8b8dd65f 608 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 609 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 610 atomic_inc(&fcc->issued_flush);
611 return ret;
612 }
740432f8 613
39d787be
CY		 614 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
615 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 616 atomic_dec(&fcc->issing_flush);
617
618 atomic_inc(&fcc->issued_flush);
740432f8
JK		 619 return ret;
620 }
6b4afdd7 621
39d787be 622 cmd.ino = ino;
adf8d90b 623 init_completion(&cmd.wait);
6b4afdd7 624
721bd4d5 625 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 626
6f890df0
CY		 627 /* update issue_list before we wake up issue_flush thread */
628 smp_mb();
629
630 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 631 wake_up(&fcc->flush_wait_queue);
6b4afdd7 632
5eba8c5d
JK		 633 if (fcc->f2fs_issue_flush) {
634 wait_for_completion(&cmd.wait);
8b8dd65f 635 atomic_dec(&fcc->issing_flush);
5eba8c5d 636 } else {
d3238691
CY		 637 struct llist_node *list;
638
639 list = llist_del_all(&fcc->issue_list);
640 if (!list) {
641 wait_for_completion(&cmd.wait);
642 atomic_dec(&fcc->issing_flush);
643 } else {
644 struct flush_cmd *tmp, *next;
645
39d787be 646 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 647
648 llist_for_each_entry_safe(tmp, next, list, llnode) {
649 if (tmp == &cmd) {
650 cmd.ret = ret;
651 atomic_dec(&fcc->issing_flush);
652 continue;
653 }
654 tmp->ret = ret;
655 complete(&tmp->wait);
656 }
657 }
5eba8c5d 658 }
adf8d90b
CY		 659
660 return cmd.ret;
6b4afdd7
JK		 661}
662
4d57b86d 663int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
2163d198
GZ		 664{
665 dev_t dev = sbi->sb->s_bdev->bd_dev;
666 struct flush_cmd_control *fcc;
667 int err = 0;
668
b01a9201
JK		 669 if (SM_I(sbi)->fcc_info) {
670 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 671 if (fcc->f2fs_issue_flush)
672 return err;
5eba8c5d
JK		 673 goto init_thread;
674 }
675
acbf054d 676 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 677 if (!fcc)
678 return -ENOMEM;
8b8dd65f
CY		 679 atomic_set(&fcc->issued_flush, 0);
680 atomic_set(&fcc->issing_flush, 0);
2163d198 681 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 682 init_llist_head(&fcc->issue_list);
b01a9201 683 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 684 if (!test_opt(sbi, FLUSH_MERGE))
685 return err;
686
5eba8c5d 687init_thread:
2163d198
GZ		 688 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
689 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
690 if (IS_ERR(fcc->f2fs_issue_flush)) {
691 err = PTR_ERR(fcc->f2fs_issue_flush);
692 kfree(fcc);
b01a9201 693 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 694 return err;
695 }
2163d198
GZ		 696
697 return err;
698}
699
4d57b86d 700void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 701{
b01a9201 702 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 703
5eba8c5d
JK		 704 if (fcc && fcc->f2fs_issue_flush) {
705 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
706
707 fcc->f2fs_issue_flush = NULL;
708 kthread_stop(flush_thread);
709 }
710 if (free) {
711 kfree(fcc);
b01a9201 712 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 713 }
2163d198
GZ		 714}
715
1228b482
CY		 716int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
717{
718 int ret = 0, i;
719
720 if (!sbi->s_ndevs)
721 return 0;
722
723 for (i = 1; i < sbi->s_ndevs; i++) {
724 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
725 continue;
726 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
727 if (ret)
728 break;
729
730 spin_lock(&sbi->dev_lock);
731 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
732 spin_unlock(&sbi->dev_lock);
733 }
734
735 return ret;
736}
737
351df4b2
JK		 738static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
739 enum dirty_type dirty_type)
740{
741 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
742
743 /* need not be added */
744 if (IS_CURSEG(sbi, segno))
745 return;
746
747 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
748 dirty_i->nr_dirty[dirty_type]++;
749
750 if (dirty_type == DIRTY) {
751 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 752 enum dirty_type t = sentry->type;
b2f2c390 753
ec325b52
JK		 754 if (unlikely(t >= DIRTY)) {
755 f2fs_bug_on(sbi, 1);
756 return;
757 }
4625d6aa
CL		 758 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
759 dirty_i->nr_dirty[t]++;
351df4b2
JK		 760 }
761}
762
763static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
764 enum dirty_type dirty_type)
765{
766 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
767
768 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
769 dirty_i->nr_dirty[dirty_type]--;
770
771 if (dirty_type == DIRTY) {
4625d6aa
CL		 772 struct seg_entry *sentry = get_seg_entry(sbi, segno);
773 enum dirty_type t = sentry->type;
774
775 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
776 dirty_i->nr_dirty[t]--;
b2f2c390 777
302bd348 778 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 779 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 780 dirty_i->victim_secmap);
351df4b2
JK		 781 }
782}
783
0a8165d7 784/*
351df4b2
JK		 785 * Should not occur error such as -ENOMEM.
786 * Adding dirty entry into seglist is not critical operation.
787 * If a given segment is one of current working segments, it won't be added.
788 */
8d8451af 789static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 790{
791 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
792 unsigned short valid_blocks;
793
794 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
795 return;
796
797 mutex_lock(&dirty_i->seglist_lock);
798
302bd348 799 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 800
801 if (valid_blocks == 0) {
802 __locate_dirty_segment(sbi, segno, PRE);
803 __remove_dirty_segment(sbi, segno, DIRTY);
804 } else if (valid_blocks < sbi->blocks_per_seg) {
805 __locate_dirty_segment(sbi, segno, DIRTY);
806 } else {
807 /* Recovery routine with SSR needs this */
808 __remove_dirty_segment(sbi, segno, DIRTY);
809 }
810
811 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 812}
813
004b6862 814static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 815 struct block_device *bdev, block_t lstart,
816 block_t start, block_t len)
275b66b0 817{
0b54fb84 818 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 819 struct list_head *pend_list;
b01a9201 820 struct discard_cmd *dc;
275b66b0 821
ba48a33e
CY		 822 f2fs_bug_on(sbi, !len);
823
824 pend_list = &dcc->pend_list[plist_idx(len)];
825
b01a9201
JK		 826 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
827 INIT_LIST_HEAD(&dc->list);
c81abe34 828 dc->bdev = bdev;
b01a9201 829 dc->lstart = lstart;
c81abe34 830 dc->start = start;
b01a9201 831 dc->len = len;
ec9895ad 832 dc->ref = 0;
15469963 833 dc->state = D_PREP;
c81abe34 834 dc->error = 0;
b01a9201 835 init_completion(&dc->wait);
22d375dd 836 list_add_tail(&dc->list, pend_list);
5f32366a 837 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 838 dcc->undiscard_blks += len;
004b6862
CY		 839
840 return dc;
841}
842
843static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
844 struct block_device *bdev, block_t lstart,
845 block_t start, block_t len,
846 struct rb_node *parent, struct rb_node **p)
847{
848 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
849 struct discard_cmd *dc;
850
851 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
852
853 rb_link_node(&dc->rb_node, parent, p);
854 rb_insert_color(&dc->rb_node, &dcc->root);
855
856 return dc;
15469963
JK		 857}
858
004b6862
CY		 859static void __detach_discard_cmd(struct discard_cmd_control *dcc,
860 struct discard_cmd *dc)
15469963 861{
dcc9165d 862 if (dc->state == D_DONE)
004b6862
CY		 863 atomic_dec(&dcc->issing_discard);
864
865 list_del(&dc->list);
866 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 867 dcc->undiscard_blks -= dc->len;
004b6862
CY		 868
869 kmem_cache_free(discard_cmd_slab, dc);
870
871 atomic_dec(&dcc->discard_cmd_cnt);
872}
873
874static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
875 struct discard_cmd *dc)
876{
877 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 878
2ec6f2ef
CY		 879 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
880
d9703d90
CY		 881 f2fs_bug_on(sbi, dc->ref);
882
c81abe34
JK		 883 if (dc->error == -EOPNOTSUPP)
884 dc->error = 0;
15469963 885
c81abe34 886 if (dc->error)
15469963 887 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 888 "Issue discard(%u, %u, %u) failed, ret: %d",
889 dc->lstart, dc->start, dc->len, dc->error);
004b6862 890 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 891}
892
c81abe34
JK		 893static void f2fs_submit_discard_endio(struct bio *bio)
894{
895 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
896
4e4cbee9 897 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 898 dc->state = D_DONE;
e31b9821 899 complete_all(&dc->wait);
c81abe34
JK		 900 bio_put(bio);
901}
902
94b1e10e 903static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 904 block_t start, block_t end)
905{
906#ifdef CONFIG_F2FS_CHECK_FS
907 struct seg_entry *sentry;
908 unsigned int segno;
909 block_t blk = start;
910 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
911 unsigned long *map;
912
913 while (blk < end) {
914 segno = GET_SEGNO(sbi, blk);
915 sentry = get_seg_entry(sbi, segno);
916 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
917
008396e1
YS		 918 if (end < START_BLOCK(sbi, segno + 1))
919 size = GET_BLKOFF_FROM_SEG0(sbi, end);
920 else
921 size = max_blocks;
6915ea9d
CY		 922 map = (unsigned long *)(sentry->cur_valid_map);
923 offset = __find_rev_next_bit(map, size, offset);
924 f2fs_bug_on(sbi, offset != size);
008396e1 925 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 926 }
927#endif
928}
929
8bb4f253
JK		 930static void __init_discard_policy(struct f2fs_sb_info *sbi,
931 struct discard_policy *dpolicy,
932 int discard_type, unsigned int granularity)
933{
934 /* common policy */
935 dpolicy->type = discard_type;
936 dpolicy->sync = true;
937 dpolicy->granularity = granularity;
938
939 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
940 dpolicy->io_aware_gran = MAX_PLIST_NUM;
941
942 if (discard_type == DPOLICY_BG) {
943 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 944 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 945 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
946 dpolicy->io_aware = true;
cba60849 947 dpolicy->sync = false;
8bb4f253
JK		 948 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
949 dpolicy->granularity = 1;
950 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
951 }
952 } else if (discard_type == DPOLICY_FORCE) {
953 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 954 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 955 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
956 dpolicy->io_aware = false;
957 } else if (discard_type == DPOLICY_FSTRIM) {
958 dpolicy->io_aware = false;
959 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 960 dpolicy->max_requests = UINT_MAX;
8bb4f253
JK		 961 dpolicy->io_aware = false;
962 }
963}
964
965
c81abe34
JK		 966/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
967static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56
CY		 968 struct discard_policy *dpolicy,
969 struct discard_cmd *dc)
c81abe34
JK		 970{
971 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 972 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
973 &(dcc->fstrim_list) : &(dcc->wait_list);
c81abe34 974 struct bio *bio = NULL;
78997b56 975 int flag = dpolicy->sync ? REQ_SYNC : 0;
c81abe34
JK		 976
977 if (dc->state != D_PREP)
978 return;
979
d6184774
YH		 980 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
981 return;
982
0243a5f9
CY		 983 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
984
c81abe34
JK		 985 dc->error = __blkdev_issue_discard(dc->bdev,
986 SECTOR_FROM_BLOCK(dc->start),
987 SECTOR_FROM_BLOCK(dc->len),
988 GFP_NOFS, 0, &bio);
989 if (!dc->error) {
990 /* should keep before submission to avoid D_DONE right away */
991 dc->state = D_SUBMIT;
8b8dd65f
CY		 992 atomic_inc(&dcc->issued_discard);
993 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 994 if (bio) {
995 bio->bi_private = dc;
996 bio->bi_end_io = f2fs_submit_discard_endio;
ecc9aa00 997 bio->bi_opf |= flag;
c81abe34 998 submit_bio(bio);
8412663d 999 list_move_tail(&dc->list, wait_list);
6915ea9d 1000 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 1001
1002 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 1003 }
1004 } else {
1005 __remove_discard_cmd(sbi, dc);
1006 }
1007}
1008
004b6862
CY		 1009static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1010 struct block_device *bdev, block_t lstart,
1011 block_t start, block_t len,
1012 struct rb_node **insert_p,
1013 struct rb_node *insert_parent)
c81abe34 1014{
004b6862 1015 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1016 struct rb_node **p;
004b6862
CY		 1017 struct rb_node *parent = NULL;
1018 struct discard_cmd *dc = NULL;
1019
1020 if (insert_p && insert_parent) {
1021 parent = insert_parent;
1022 p = insert_p;
1023 goto do_insert;
1024 }
c81abe34 1025
4d57b86d 1026 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
004b6862
CY		 1027do_insert:
1028 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1029 if (!dc)
1030 return NULL;
c81abe34 1031
004b6862 1032 return dc;
c81abe34
JK		 1033}
1034
ba48a33e
CY		 1035static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1036 struct discard_cmd *dc)
1037{
1038 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1039}
1040
3d6a650f
YH		 1041static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1042 struct discard_cmd *dc, block_t blkaddr)
1043{
ba48a33e 1044 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1045 struct discard_info di = dc->di;
1046 bool modified = false;
3d6a650f 1047
004b6862 1048 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1049 __remove_discard_cmd(sbi, dc);
1050 return;
1051 }
1052
d84d1cbd
CY		 1053 dcc->undiscard_blks -= di.len;
1054
004b6862 1055 if (blkaddr > di.lstart) {
3d6a650f 1056 dc->len = blkaddr - dc->lstart;
d84d1cbd 1057 dcc->undiscard_blks += dc->len;
ba48a33e 1058 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1059 modified = true;
1060 }
1061
1062 if (blkaddr < di.lstart + di.len - 1) {
1063 if (modified) {
1064 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1065 di.start + blkaddr + 1 - di.lstart,
1066 di.lstart + di.len - 1 - blkaddr,
1067 NULL, NULL);
1068 } else {
1069 dc->lstart++;
1070 dc->len--;
1071 dc->start++;
d84d1cbd 1072 dcc->undiscard_blks += dc->len;
ba48a33e 1073 __relocate_discard_cmd(dcc, dc);
004b6862 1074 }
3d6a650f
YH		 1075 }
1076}
1077
004b6862
CY		 1078static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1079 struct block_device *bdev, block_t lstart,
1080 block_t start, block_t len)
275b66b0 1081{
0b54fb84 1082 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1083 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1084 struct discard_cmd *dc;
1085 struct discard_info di = {0};
1086 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1087 block_t end = lstart + len;
275b66b0 1088
15469963 1089 mutex_lock(&dcc->cmd_lock);
40465257 1090
4d57b86d 1091 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
004b6862
CY		 1092 NULL, lstart,
1093 (struct rb_entry **)&prev_dc,
1094 (struct rb_entry **)&next_dc,
1095 &insert_p, &insert_parent, true);
1096 if (dc)
1097 prev_dc = dc;
1098
1099 if (!prev_dc) {
1100 di.lstart = lstart;
1101 di.len = next_dc ? next_dc->lstart - lstart : len;
1102 di.len = min(di.len, len);
1103 di.start = start;
22d375dd 1104 }
15469963 1105
004b6862
CY		 1106 while (1) {
1107 struct rb_node *node;
1108 bool merged = false;
1109 struct discard_cmd *tdc = NULL;
1110
1111 if (prev_dc) {
1112 di.lstart = prev_dc->lstart + prev_dc->len;
1113 if (di.lstart < lstart)
1114 di.lstart = lstart;
1115 if (di.lstart >= end)
1116 break;
1117
1118 if (!next_dc || next_dc->lstart > end)
1119 di.len = end - di.lstart;
1120 else
1121 di.len = next_dc->lstart - di.lstart;
1122 di.start = start + di.lstart - lstart;
1123 }
1124
1125 if (!di.len)
1126 goto next;
1127
1128 if (prev_dc && prev_dc->state == D_PREP &&
1129 prev_dc->bdev == bdev &&
1130 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1131 prev_dc->di.len += di.len;
d84d1cbd 1132 dcc->undiscard_blks += di.len;
ba48a33e 1133 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1134 di = prev_dc->di;
1135 tdc = prev_dc;
1136 merged = true;
1137 }
1138
1139 if (next_dc && next_dc->state == D_PREP &&
1140 next_dc->bdev == bdev &&
1141 __is_discard_front_mergeable(&di, &next_dc->di)) {
1142 next_dc->di.lstart = di.lstart;
1143 next_dc->di.len += di.len;
1144 next_dc->di.start = di.start;
d84d1cbd 1145 dcc->undiscard_blks += di.len;
ba48a33e 1146 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1147 if (tdc)
1148 __remove_discard_cmd(sbi, tdc);
004b6862 1149 merged = true;
4e6a8d9b 1150 }
004b6862 1151
df0f6b44 1152 if (!merged) {
004b6862
CY		 1153 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1154 di.len, NULL, NULL);
df0f6b44 1155 }
004b6862
CY		 1156 next:
1157 prev_dc = next_dc;
1158 if (!prev_dc)
1159 break;
1160
1161 node = rb_next(&prev_dc->rb_node);
1162 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1163 }
1164
1165 mutex_unlock(&dcc->cmd_lock);
1166}
1167
1168static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1169 struct block_device *bdev, block_t blkstart, block_t blklen)
1170{
1171 block_t lblkstart = blkstart;
1172
0243a5f9 1173 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1174
1175 if (sbi->s_ndevs) {
1176 int devi = f2fs_target_device_index(sbi, blkstart);
1177
1178 blkstart -= FDEV(devi).start_blk;
1179 }
1180 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1181 return 0;
1182}
1183
78997b56
CY		 1184static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1185 struct discard_policy *dpolicy)
bd5b0738
CY		 1186{
1187 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1188 struct list_head *pend_list;
1189 struct discard_cmd *dc, *tmp;
1190 struct blk_plug plug;
78997b56 1191 int i, iter = 0, issued = 0;
e6c6de18 1192 bool io_interrupted = false;
bd5b0738 1193
78997b56
CY		 1194 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1195 if (i + 1 < dpolicy->granularity)
1196 break;
bd5b0738 1197 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1198
1199 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1200 if (list_empty(pend_list))
1201 goto next;
67fce70b
CY		 1202 if (unlikely(dcc->rbtree_check))
1203 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1204 &dcc->root));
33da62cf 1205 blk_start_plug(&plug);
bd5b0738
CY		 1206 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1207 f2fs_bug_on(sbi, dc->state != D_PREP);
1208
ecc9aa00
CY		 1209 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1210 !is_idle(sbi)) {
e6c6de18 1211 io_interrupted = true;
ecc9aa00 1212 goto skip;
969d1b18 1213 }
e6c6de18 1214
78997b56 1215 __submit_discard_cmd(sbi, dpolicy, dc);
ecc9aa00
CY		 1216 issued++;
1217skip:
1218 if (++iter >= dpolicy->max_requests)
33da62cf 1219 break;
bd5b0738 1220 }
33da62cf 1221 blk_finish_plug(&plug);
49c60c67 1222next:
33da62cf
CY		 1223 mutex_unlock(&dcc->cmd_lock);
1224
1225 if (iter >= dpolicy->max_requests)
1226 break;
bd5b0738 1227 }
969d1b18 1228
e6c6de18
CY		 1229 if (!issued && io_interrupted)
1230 issued = -1;
1231
969d1b18
CY		 1232 return issued;
1233}
1234
cf5c759f 1235static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1236{
1237 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1238 struct list_head *pend_list;
1239 struct discard_cmd *dc, *tmp;
1240 int i;
cf5c759f 1241 bool dropped = false;
969d1b18
CY		 1242
1243 mutex_lock(&dcc->cmd_lock);
1244 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1245 pend_list = &dcc->pend_list[i];
1246 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1247 f2fs_bug_on(sbi, dc->state != D_PREP);
1248 __remove_discard_cmd(sbi, dc);
cf5c759f 1249 dropped = true;
969d1b18
CY		 1250 }
1251 }
1252 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1253
1254 return dropped;
bd5b0738
CY		 1255}
1256
4d57b86d 1257void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
7950e9ac
CY		 1258{
1259 __drop_discard_cmd(sbi);
1260}
1261
0ea80512 1262static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1263 struct discard_cmd *dc)
1264{
1265 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1266 unsigned int len = 0;
2a510c00
CY		 1267
1268 wait_for_completion_io(&dc->wait);
1269 mutex_lock(&dcc->cmd_lock);
1270 f2fs_bug_on(sbi, dc->state != D_DONE);
1271 dc->ref--;
0ea80512
CY		 1272 if (!dc->ref) {
1273 if (!dc->error)
1274 len = dc->len;
2a510c00 1275 __remove_discard_cmd(sbi, dc);
0ea80512 1276 }
2a510c00 1277 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1278
1279 return len;
2a510c00
CY		 1280}
1281
0ea80512 1282static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1283 struct discard_policy *dpolicy,
1284 block_t start, block_t end)
63a94fa1
CY		 1285{
1286 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1287 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1288 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1289 struct discard_cmd *dc, *tmp;
6afae633 1290 bool need_wait;
0ea80512 1291 unsigned int trimmed = 0;
6afae633
CY		 1292
1293next:
1294 need_wait = false;
63a94fa1
CY		 1295
1296 mutex_lock(&dcc->cmd_lock);
1297 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1298 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1299 continue;
78997b56 1300 if (dc->len < dpolicy->granularity)
8412663d 1301 continue;
78997b56 1302 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1303 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1304 if (!dc->error)
1305 trimmed += dc->len;
63a94fa1 1306 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1307 } else {
1308 dc->ref++;
1309 need_wait = true;
1310 break;
63a94fa1
CY		 1311 }
1312 }
1313 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1314
1315 if (need_wait) {
0ea80512 1316 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1317 goto next;
1318 }
0ea80512
CY		 1319
1320 return trimmed;
63a94fa1
CY		 1321}
1322
01f9cf6d 1323static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1324 struct discard_policy *dpolicy)
8412663d 1325{
9a997188 1326 struct discard_policy dp;
01f9cf6d 1327 unsigned int discard_blks;
9a997188 1328
01f9cf6d
CY		 1329 if (dpolicy)
1330 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
9a997188
JK		 1331
1332 /* wait all */
8bb4f253 1333 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
01f9cf6d 1334 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1335 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
01f9cf6d
CY		 1336 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1337
1338 return discard_blks;
8412663d
CY		 1339}
1340
004b6862 1341/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1342static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1343{
1344 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1345 struct discard_cmd *dc;
ec9895ad 1346 bool need_wait = false;
004b6862
CY		 1347
1348 mutex_lock(&dcc->cmd_lock);
4d57b86d
CY		 1349 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1350 NULL, blkaddr);
004b6862 1351 if (dc) {
ec9895ad
CY		 1352 if (dc->state == D_PREP) {
1353 __punch_discard_cmd(sbi, dc, blkaddr);
1354 } else {
1355 dc->ref++;
1356 need_wait = true;
1357 }
275b66b0 1358 }
d431413f 1359 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1360
2a510c00
CY		 1361 if (need_wait)
1362 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1363}
1364
4d57b86d 1365void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
cce13252
CY		 1366{
1367 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1368
1369 if (dcc && dcc->f2fs_issue_discard) {
1370 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1371
1372 dcc->f2fs_issue_discard = NULL;
1373 kthread_stop(discard_thread);
ec9895ad 1374 }
d431413f
CY		 1375}
1376
8412663d 1377/* This comes from f2fs_put_super */
cf5c759f 1378bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1379{
1380 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1381 struct discard_policy dpolicy;
cf5c759f 1382 bool dropped;
969d1b18 1383
8bb4f253
JK		 1384 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1385 dcc->discard_granularity);
78997b56 1386 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1387 dropped = __drop_discard_cmd(sbi);
cf5c759f 1388
9a997188
JK		 1389 /* just to make sure there is no pending discard commands */
1390 __wait_all_discard_cmd(sbi, NULL);
2482c432
CY		 1391
1392 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
cf5c759f 1393 return dropped;
969d1b18
CY		 1394}
1395
15469963
JK		 1396static int issue_discard_thread(void *data)
1397{
1398 struct f2fs_sb_info *sbi = data;
1399 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1400 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1401 struct discard_policy dpolicy;
969d1b18
CY		 1402 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1403 int issued;
15469963 1404
1d7be270 1405 set_freezable();
15469963 1406
1d7be270 1407 do {
8bb4f253 1408 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1409 dcc->discard_granularity);
1410
969d1b18
CY		 1411 wait_event_interruptible_timeout(*q,
1412 kthread_should_stop() || freezing(current) ||
1413 dcc->discard_wake,
1414 msecs_to_jiffies(wait_ms));
35a9a766
SY		 1415
1416 if (dcc->discard_wake)
1417 dcc->discard_wake = 0;
1418
1d7be270
JK		 1419 if (try_to_freeze())
1420 continue;
3b60d802
CY		 1421 if (f2fs_readonly(sbi->sb))
1422 continue;
1d7be270
JK		 1423 if (kthread_should_stop())
1424 return 0;
d6184774
YH		 1425 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1426 wait_ms = dpolicy.max_interval;
1427 continue;
1428 }
15469963 1429
5b0e9539 1430 if (sbi->gc_mode == GC_URGENT)
8bb4f253 1431 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1432
dc6febb6
CY		 1433 sb_start_intwrite(sbi->sb);
1434
78997b56 1435 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1436 if (issued > 0) {
78997b56
CY		 1437 __wait_all_discard_cmd(sbi, &dpolicy);
1438 wait_ms = dpolicy.min_interval;
f9d1dced
YH		 1439 } else if (issued == -1){
1440 wait_ms = dpolicy.mid_interval;
969d1b18 1441 } else {
78997b56 1442 wait_ms = dpolicy.max_interval;
969d1b18 1443 }
1d7be270 1444
dc6febb6 1445 sb_end_intwrite(sbi->sb);
1d7be270 1446
1d7be270
JK		 1447 } while (!kthread_should_stop());
1448 return 0;
15469963
JK		 1449}
1450
f46e8809 1451#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1452static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1453 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1454{
92592285 1455 sector_t sector, nr_sects;
10a875f8 1456 block_t lblkstart = blkstart;
3c62be17
JK		 1457 int devi = 0;
1458
1459 if (sbi->s_ndevs) {
1460 devi = f2fs_target_device_index(sbi, blkstart);
1461 blkstart -= FDEV(devi).start_blk;
1462 }
f46e8809
DLM		 1463
1464 /*
1465 * We need to know the type of the zone: for conventional zones,
1466 * use regular discard if the drive supports it. For sequential
1467 * zones, reset the zone write pointer.
1468 */
3c62be17 1469 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1470
1471 case BLK_ZONE_TYPE_CONVENTIONAL:
1472 if (!blk_queue_discard(bdev_get_queue(bdev)))
1473 return 0;
c81abe34 1474 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1475 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1476 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1477 sector = SECTOR_FROM_BLOCK(blkstart);
1478 nr_sects = SECTOR_FROM_BLOCK(blklen);
1479
1480 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1481 nr_sects != bdev_zone_sectors(bdev)) {
1482 f2fs_msg(sbi->sb, KERN_INFO,
1483 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1484 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1485 blkstart, blklen);
1486 return -EIO;
1487 }
d50aaeec 1488 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1489 return blkdev_reset_zones(bdev, sector,
1490 nr_sects, GFP_NOFS);
1491 default:
1492 /* Unknown zone type: broken device ? */
1493 return -EIO;
1494 }
1495}
1496#endif
1497
3c62be17
JK		 1498static int __issue_discard_async(struct f2fs_sb_info *sbi,
1499 struct block_device *bdev, block_t blkstart, block_t blklen)
1500{
1501#ifdef CONFIG_BLK_DEV_ZONED
ccd31cb2 1502 if (f2fs_sb_has_blkzoned(sbi->sb) &&
3c62be17
JK		 1503 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1504 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1505#endif
c81abe34 1506 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1507}
1508
1e87a78d 1509static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1510 block_t blkstart, block_t blklen)
1511{
3c62be17
JK		 1512 sector_t start = blkstart, len = 0;
1513 struct block_device *bdev;
a66cdd98
JK		 1514 struct seg_entry *se;
1515 unsigned int offset;
1516 block_t i;
3c62be17
JK		 1517 int err = 0;
1518
1519 bdev = f2fs_target_device(sbi, blkstart, NULL);
1520
1521 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1522 if (i != start) {
1523 struct block_device *bdev2 =
1524 f2fs_target_device(sbi, i, NULL);
1525
1526 if (bdev2 != bdev) {
1527 err = __issue_discard_async(sbi, bdev,
1528 start, len);
1529 if (err)
1530 return err;
1531 bdev = bdev2;
1532 start = i;
1533 len = 0;
1534 }
1535 }
a66cdd98 1536
a66cdd98
JK		 1537 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1538 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1539
1540 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1541 sbi->discard_blks--;
1542 }
f46e8809 1543
3c62be17
JK		 1544 if (len)
1545 err = __issue_discard_async(sbi, bdev, start, len);
1546 return err;
1e87a78d
JK		 1547}
1548
25290fa5
JK		 1549static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1550 bool check_only)
adf4983b 1551{
b2955550
JK		 1552 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1553 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1554 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1555 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1556 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1557 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1558 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1559 unsigned int start = 0, end = -1;
c473f1a9 1560 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1561 struct discard_entry *de = NULL;
46f84c2c 1562 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1563 int i;
1564
3e025740 1565 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1566 return false;
b2955550 1567
a66cdd98
JK		 1568 if (!force) {
1569 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1570 SM_I(sbi)->dcc_info->nr_discards >=
1571 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1572 return false;
4b2fecc8
JK		 1573 }
1574
b2955550
JK		 1575 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1576 for (i = 0; i < entries; i++)
a66cdd98 1577 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1578 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1579
0b54fb84
JK		 1580 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1581 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1582 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1583 if (start >= max_blocks)
1584 break;
1585
1586 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1587 if (force && start && end != max_blocks
1588 && (end - start) < cpc->trim_minlen)
1589 continue;
1590
25290fa5
JK		 1591 if (check_only)
1592 return true;
1593
a7eeb823
CY		 1594 if (!de) {
1595 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1596 GFP_F2FS_ZERO);
1597 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1598 list_add_tail(&de->list, head);
1599 }
1600
1601 for (i = start; i < end; i++)
1602 __set_bit_le(i, (void *)de->discard_map);
1603
1604 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1605 }
25290fa5 1606 return false;
b2955550
JK		 1607}
1608
af8ff65b
CY		 1609static void release_discard_addr(struct discard_entry *entry)
1610{
1611 list_del(&entry->list);
1612 kmem_cache_free(discard_entry_slab, entry);
1613}
1614
4d57b86d 1615void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1616{
46f84c2c 1617 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1618 struct discard_entry *entry, *this;
1619
1620 /* drop caches */
af8ff65b
CY		 1621 list_for_each_entry_safe(entry, this, head, list)
1622 release_discard_addr(entry);
4b2fecc8
JK		 1623}
1624
0a8165d7 1625/*
4d57b86d 1626 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1627 */
1628static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1629{
1630 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1631 unsigned int segno;
351df4b2
JK		 1632
1633 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1634 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1635 __set_test_and_free(sbi, segno);
351df4b2
JK		 1636 mutex_unlock(&dirty_i->seglist_lock);
1637}
1638
4d57b86d
CY		 1639void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1640 struct cp_control *cpc)
351df4b2 1641{
969d1b18
CY		 1642 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1643 struct list_head *head = &dcc->entry_list;
2d7b822a 1644 struct discard_entry *entry, *this;
351df4b2 1645 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1646 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1647 unsigned int start = 0, end = -1;
36abef4e 1648 unsigned int secno, start_segno;
c473f1a9 1649 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1650
1651 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1652
351df4b2 1653 while (1) {
29e59c14 1654 int i;
7cd8558b
JK		 1655 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1656 if (start >= MAIN_SEGS(sbi))
351df4b2 1657 break;
7cd8558b
JK		 1658 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1659 start + 1);
29e59c14
CL		 1660
1661 for (i = start; i < end; i++)
1662 clear_bit(i, prefree_map);
1663
1664 dirty_i->nr_dirty[PRE] -= end - start;
1665
650d3c4e 1666 if (!test_opt(sbi, DISCARD))
29e59c14 1667 continue;
351df4b2 1668
650d3c4e
YH		 1669 if (force && start >= cpc->trim_start &&
1670 (end - 1) <= cpc->trim_end)
1671 continue;
1672
36abef4e
JK		 1673 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1674 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1675 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1676 continue;
1677 }
1678next:
4ddb1a4d
JK		 1679 secno = GET_SEC_FROM_SEG(sbi, start);
1680 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1681 if (!IS_CURSEC(sbi, secno) &&
302bd348 1682 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1683 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1684 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1685
1686 start = start_segno + sbi->segs_per_sec;
1687 if (start < end)
1688 goto next;
8b107f5b
JK		 1689 else
1690 end = start - 1;
351df4b2
JK		 1691 }
1692 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1693
1694 /* send small discards */
2d7b822a 1695 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1696 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1697 bool is_valid = test_bit_le(0, entry->discard_map);
1698
1699find_next:
1700 if (is_valid) {
1701 next_pos = find_next_zero_bit_le(entry->discard_map,
1702 sbi->blocks_per_seg, cur_pos);
1703 len = next_pos - cur_pos;
1704
ccd31cb2 1705 if (f2fs_sb_has_blkzoned(sbi->sb) ||
acfd2810 1706 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1707 goto skip;
1708
1709 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1710 len);
a7eeb823
CY		 1711 total_len += len;
1712 } else {
1713 next_pos = find_next_bit_le(entry->discard_map,
1714 sbi->blocks_per_seg, cur_pos);
1715 }
836b5a63 1716skip:
a7eeb823
CY		 1717 cur_pos = next_pos;
1718 is_valid = !is_valid;
1719
1720 if (cur_pos < sbi->blocks_per_seg)
1721 goto find_next;
1722
af8ff65b 1723 release_discard_addr(entry);
969d1b18 1724 dcc->nr_discards -= total_len;
b2955550 1725 }
34e159da 1726
01983c71 1727 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1728}
1729
8ed59745 1730static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1731{
15469963 1732 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1733 struct discard_cmd_control *dcc;
ba48a33e 1734 int err = 0, i;
0b54fb84
JK		 1735
1736 if (SM_I(sbi)->dcc_info) {
1737 dcc = SM_I(sbi)->dcc_info;
1738 goto init_thread;
1739 }
1740
acbf054d 1741 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 1742 if (!dcc)
1743 return -ENOMEM;
1744
969d1b18 1745 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1746 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1747 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1748 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 1749 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1750 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1751 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1752 atomic_set(&dcc->issued_discard, 0);
1753 atomic_set(&dcc->issing_discard, 0);
5f32366a 1754 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1755 dcc->nr_discards = 0;
d618ebaf 1756 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1757 dcc->undiscard_blks = 0;
004b6862 1758 dcc->root = RB_ROOT;
67fce70b 1759 dcc->rbtree_check = false;
0b54fb84 1760
15469963 1761 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1762 SM_I(sbi)->dcc_info = dcc;
1763init_thread:
15469963
JK		 1764 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1765 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1766 if (IS_ERR(dcc->f2fs_issue_discard)) {
1767 err = PTR_ERR(dcc->f2fs_issue_discard);
1768 kfree(dcc);
1769 SM_I(sbi)->dcc_info = NULL;
1770 return err;
1771 }
1772
0b54fb84
JK		 1773 return err;
1774}
1775
f099405f 1776static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1777{
1778 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1779
f099405f
CY		 1780 if (!dcc)
1781 return;
1782
4d57b86d 1783 f2fs_stop_discard_thread(sbi);
f099405f
CY		 1784
1785 kfree(dcc);
1786 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1787}
1788
184a5cd2 1789static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1790{
1791 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1792
1793 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1794 sit_i->dirty_sentries++;
184a5cd2
CY		 1795 return false;
1796 }
1797
1798 return true;
351df4b2
JK		 1799}
1800
1801static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1802 unsigned int segno, int modified)
1803{
1804 struct seg_entry *se = get_seg_entry(sbi, segno);
1805 se->type = type;
1806 if (modified)
1807 __mark_sit_entry_dirty(sbi, segno);
1808}
1809
1810static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1811{
1812 struct seg_entry *se;
1813 unsigned int segno, offset;
1814 long int new_vblocks;
6415fedc
YS		 1815 bool exist;
1816#ifdef CONFIG_F2FS_CHECK_FS
1817 bool mir_exist;
1818#endif
351df4b2
JK		 1819
1820 segno = GET_SEGNO(sbi, blkaddr);
1821
1822 se = get_seg_entry(sbi, segno);
1823 new_vblocks = se->valid_blocks + del;
491c0854 1824 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1825
9850cf4a 1826 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1827 (new_vblocks > sbi->blocks_per_seg)));
1828
1829 se->valid_blocks = new_vblocks;
a1f72ac2
CY		 1830 se->mtime = get_mtime(sbi, false);
1831 if (se->mtime > SIT_I(sbi)->max_mtime)
1832 SIT_I(sbi)->max_mtime = se->mtime;
351df4b2
JK		 1833
1834 /* Update valid block bitmap */
1835 if (del > 0) {
6415fedc 1836 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1837#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1838 mir_exist = f2fs_test_and_set_bit(offset,
1839 se->cur_valid_map_mir);
1840 if (unlikely(exist != mir_exist)) {
1841 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1842 "when setting bitmap, blk:%u, old bit:%d",
1843 blkaddr, exist);
05796763 1844 f2fs_bug_on(sbi, 1);
6415fedc 1845 }
355e7891 1846#endif
6415fedc
YS		 1847 if (unlikely(exist)) {
1848 f2fs_msg(sbi->sb, KERN_ERR,
1849 "Bitmap was wrongly set, blk:%u", blkaddr);
1850 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1851 se->valid_blocks--;
1852 del = 0;
355e7891 1853 }
6415fedc 1854
3e025740
JK		 1855 if (f2fs_discard_en(sbi) &&
1856 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1857 sbi->discard_blks--;
720037f9
JK		 1858
1859 /* don't overwrite by SSR to keep node chain */
5d7881ca 1860 if (IS_NODESEG(se->type)) {
720037f9
JK		 1861 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1862 se->ckpt_valid_blocks++;
1863 }
351df4b2 1864 } else {
6415fedc 1865 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1866#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1867 mir_exist = f2fs_test_and_clear_bit(offset,
1868 se->cur_valid_map_mir);
1869 if (unlikely(exist != mir_exist)) {
1870 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1871 "when clearing bitmap, blk:%u, old bit:%d",
1872 blkaddr, exist);
05796763 1873 f2fs_bug_on(sbi, 1);
6415fedc 1874 }
355e7891 1875#endif
6415fedc
YS		 1876 if (unlikely(!exist)) {
1877 f2fs_msg(sbi->sb, KERN_ERR,
1878 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1879 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1880 se->valid_blocks++;
1881 del = 0;
355e7891 1882 }
6415fedc 1883
3e025740
JK		 1884 if (f2fs_discard_en(sbi) &&
1885 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1886 sbi->discard_blks++;
351df4b2
JK		 1887 }
1888 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1889 se->ckpt_valid_blocks += del;
1890
1891 __mark_sit_entry_dirty(sbi, segno);
1892
1893 /* update total number of valid blocks to be written in ckpt area */
1894 SIT_I(sbi)->written_valid_blocks += del;
1895
1896 if (sbi->segs_per_sec > 1)
1897 get_sec_entry(sbi, segno)->valid_blocks += del;
1898}
1899
4d57b86d 1900void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 1901{
1902 unsigned int segno = GET_SEGNO(sbi, addr);
1903 struct sit_info *sit_i = SIT_I(sbi);
1904
9850cf4a 1905 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1906 if (addr == NEW_ADDR)
1907 return;
1908
1909 /* add it into sit main buffer */
3d26fa6b 1910 down_write(&sit_i->sentry_lock);
351df4b2
JK		 1911
1912 update_sit_entry(sbi, addr, -1);
1913
1914 /* add it into dirty seglist */
1915 locate_dirty_segment(sbi, segno);
1916
3d26fa6b 1917 up_write(&sit_i->sentry_lock);
351df4b2
JK		 1918}
1919
4d57b86d 1920bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 1921{
1922 struct sit_info *sit_i = SIT_I(sbi);
1923 unsigned int segno, offset;
1924 struct seg_entry *se;
1925 bool is_cp = false;
1926
e1da7872 1927 if (!is_valid_data_blkaddr(sbi, blkaddr))
6e2c64ad
JK		 1928 return true;
1929
3d26fa6b 1930 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1931
1932 segno = GET_SEGNO(sbi, blkaddr);
1933 se = get_seg_entry(sbi, segno);
1934 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1935
1936 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1937 is_cp = true;
1938
3d26fa6b 1939 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1940
1941 return is_cp;
1942}
1943
0a8165d7 1944/*
351df4b2
JK		 1945 * This function should be resided under the curseg_mutex lock
1946 */
1947static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1948 struct f2fs_summary *sum)
351df4b2
JK		 1949{
1950 struct curseg_info *curseg = CURSEG_I(sbi, type);
1951 void *addr = curseg->sum_blk;
e79efe3b 1952 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1953 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1954}
1955
0a8165d7 1956/*
351df4b2
JK		 1957 * Calculate the number of current summary pages for writing
1958 */
4d57b86d 1959int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1960{
351df4b2 1961 int valid_sum_count = 0;
9a47938b 1962 int i, sum_in_page;
351df4b2
JK		 1963
1964 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1965 if (sbi->ckpt->alloc_type[i] == SSR)
1966 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1967 else {
1968 if (for_ra)
1969 valid_sum_count += le16_to_cpu(
1970 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1971 else
1972 valid_sum_count += curseg_blkoff(sbi, i);
1973 }
351df4b2
JK		 1974 }
1975
09cbfeaf 1976 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1977 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1978 if (valid_sum_count <= sum_in_page)
351df4b2 1979 return 1;
9a47938b 1980 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1981 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1982 return 2;
1983 return 3;
1984}
1985
0a8165d7 1986/*
351df4b2
JK		 1987 * Caller should put this summary page
1988 */
4d57b86d 1989struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 1990{
4d57b86d 1991 return f2fs_get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1992}
1993
4d57b86d
CY		 1994void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
1995 void *src, block_t blk_addr)
351df4b2 1996{
4d57b86d 1997 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 1998
0537b811 1999 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2000 set_page_dirty(page);
2001 f2fs_put_page(page, 1);
2002}
2003
381722d2
CY		 2004static void write_sum_page(struct f2fs_sb_info *sbi,
2005 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2006{
4d57b86d 2007 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2008}
2009
b7ad7512
CY		 2010static void write_current_sum_page(struct f2fs_sb_info *sbi,
2011 int type, block_t blk_addr)
2012{
2013 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2014 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2015 struct f2fs_summary_block *src = curseg->sum_blk;
2016 struct f2fs_summary_block *dst;
2017
2018 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2019 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2020
2021 mutex_lock(&curseg->curseg_mutex);
2022
2023 down_read(&curseg->journal_rwsem);
2024 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2025 up_read(&curseg->journal_rwsem);
2026
2027 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2028 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2029
2030 mutex_unlock(&curseg->curseg_mutex);
2031
2032 set_page_dirty(page);
2033 f2fs_put_page(page, 1);
2034}
2035
a7881893
JK		 2036static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2037{
2038 struct curseg_info *curseg = CURSEG_I(sbi, type);
2039 unsigned int segno = curseg->segno + 1;
2040 struct free_segmap_info *free_i = FREE_I(sbi);
2041
2042 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2043 return !test_bit(segno, free_i->free_segmap);
2044 return 0;
2045}
2046
0a8165d7 2047/*
351df4b2
JK		 2048 * Find a new segment from the free segments bitmap to right order
2049 * This function should be returned with success, otherwise BUG
2050 */
2051static void get_new_segment(struct f2fs_sb_info *sbi,
2052 unsigned int *newseg, bool new_sec, int dir)
2053{
2054 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2055 unsigned int segno, secno, zoneno;
7cd8558b 2056 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2057 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2058 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2059 unsigned int left_start = hint;
2060 bool init = true;
2061 int go_left = 0;
2062 int i;
2063
1a118ccf 2064 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2065
2066 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2067 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2068 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2069 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2070 goto got_it;
2071 }
2072find_other_zone:
7cd8558b
JK		 2073 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2074 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2075 if (dir == ALLOC_RIGHT) {
2076 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2077 MAIN_SECS(sbi), 0);
2078 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2079 } else {
2080 go_left = 1;
2081 left_start = hint - 1;
2082 }
2083 }
2084 if (go_left == 0)
2085 goto skip_left;
2086
2087 while (test_bit(left_start, free_i->free_secmap)) {
2088 if (left_start > 0) {
2089 left_start--;
2090 continue;
2091 }
2092 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2093 MAIN_SECS(sbi), 0);
2094 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2095 break;
2096 }
2097 secno = left_start;
2098skip_left:
4ddb1a4d
JK		 2099 segno = GET_SEG_FROM_SEC(sbi, secno);
2100 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2101
2102 /* give up on finding another zone */
2103 if (!init)
2104 goto got_it;
2105 if (sbi->secs_per_zone == 1)
2106 goto got_it;
2107 if (zoneno == old_zoneno)
2108 goto got_it;
2109 if (dir == ALLOC_LEFT) {
2110 if (!go_left && zoneno + 1 >= total_zones)
2111 goto got_it;
2112 if (go_left && zoneno == 0)
2113 goto got_it;
2114 }
2115 for (i = 0; i < NR_CURSEG_TYPE; i++)
2116 if (CURSEG_I(sbi, i)->zone == zoneno)
2117 break;
2118
2119 if (i < NR_CURSEG_TYPE) {
2120 /* zone is in user, try another */
2121 if (go_left)
2122 hint = zoneno * sbi->secs_per_zone - 1;
2123 else if (zoneno + 1 >= total_zones)
2124 hint = 0;
2125 else
2126 hint = (zoneno + 1) * sbi->secs_per_zone;
2127 init = false;
2128 goto find_other_zone;
2129 }
2130got_it:
2131 /* set it as dirty segment in free segmap */
9850cf4a 2132 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2133 __set_inuse(sbi, segno);
2134 *newseg = segno;
1a118ccf 2135 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2136}
2137
2138static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2139{
2140 struct curseg_info *curseg = CURSEG_I(sbi, type);
2141 struct summary_footer *sum_footer;
2142
2143 curseg->segno = curseg->next_segno;
4ddb1a4d 2144 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2145 curseg->next_blkoff = 0;
2146 curseg->next_segno = NULL_SEGNO;
2147
2148 sum_footer = &(curseg->sum_blk->footer);
2149 memset(sum_footer, 0, sizeof(struct summary_footer));
2150 if (IS_DATASEG(type))
2151 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2152 if (IS_NODESEG(type))
2153 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2154 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2155}
2156
7a20b8a6
JK		 2157static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2158{
a7881893
JK		 2159 /* if segs_per_sec is large than 1, we need to keep original policy. */
2160 if (sbi->segs_per_sec != 1)
2161 return CURSEG_I(sbi, type)->segno;
2162
b94929d9
YS		 2163 if (test_opt(sbi, NOHEAP) &&
2164 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2165 return 0;
2166
e066b83c
JK		 2167 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2168 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2169
2170 /* find segments from 0 to reuse freed segments */
63189b78 2171 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2172 return 0;
2173
7a20b8a6
JK		 2174 return CURSEG_I(sbi, type)->segno;
2175}
2176
0a8165d7 2177/*
351df4b2
JK		 2178 * Allocate a current working segment.
2179 * This function always allocates a free segment in LFS manner.
2180 */
2181static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2182{
2183 struct curseg_info *curseg = CURSEG_I(sbi, type);
2184 unsigned int segno = curseg->segno;
2185 int dir = ALLOC_LEFT;
2186
2187 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2188 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2189 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2190 dir = ALLOC_RIGHT;
2191
2192 if (test_opt(sbi, NOHEAP))
2193 dir = ALLOC_RIGHT;
2194
7a20b8a6 2195 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2196 get_new_segment(sbi, &segno, new_sec, dir);
2197 curseg->next_segno = segno;
2198 reset_curseg(sbi, type, 1);
2199 curseg->alloc_type = LFS;
2200}
2201
2202static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2203 struct curseg_info *seg, block_t start)
2204{
2205 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2206 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2207 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2208 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2209 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2210 int i, pos;
2211
2212 for (i = 0; i < entries; i++)
2213 target_map[i] = ckpt_map[i] | cur_map[i];
2214
2215 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2216
2217 seg->next_blkoff = pos;
351df4b2
JK		 2218}
2219
0a8165d7 2220/*
351df4b2
JK		 2221 * If a segment is written by LFS manner, next block offset is just obtained
2222 * by increasing the current block offset. However, if a segment is written by
2223 * SSR manner, next block offset obtained by calling __next_free_blkoff
2224 */
2225static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2226 struct curseg_info *seg)
2227{
2228 if (seg->alloc_type == SSR)
2229 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2230 else
2231 seg->next_blkoff++;
2232}
2233
0a8165d7 2234/*
e1c42045 2235 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2236 * manner, so it should recover the existing segment information of valid blocks
2237 */
025d63a4 2238static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2239{
2240 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2241 struct curseg_info *curseg = CURSEG_I(sbi, type);
2242 unsigned int new_segno = curseg->next_segno;
2243 struct f2fs_summary_block *sum_node;
2244 struct page *sum_page;
2245
2246 write_sum_page(sbi, curseg->sum_blk,
2247 GET_SUM_BLOCK(sbi, curseg->segno));
2248 __set_test_and_inuse(sbi, new_segno);
2249
2250 mutex_lock(&dirty_i->seglist_lock);
2251 __remove_dirty_segment(sbi, new_segno, PRE);
2252 __remove_dirty_segment(sbi, new_segno, DIRTY);
2253 mutex_unlock(&dirty_i->seglist_lock);
2254
2255 reset_curseg(sbi, type, 1);
2256 curseg->alloc_type = SSR;
2257 __next_free_blkoff(sbi, curseg, 0);
2258
4d57b86d 2259 sum_page = f2fs_get_sum_page(sbi, new_segno);
025d63a4
CY		 2260 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2261 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2262 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2263}
2264
43727527
JK		 2265static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2266{
2267 struct curseg_info *curseg = CURSEG_I(sbi, type);
2268 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2269 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2270 int i, cnt;
2271 bool reversed = false;
c192f7a4 2272
4d57b86d 2273 /* f2fs_need_SSR() already forces to do this */
e066b83c
JK		 2274 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2275 curseg->next_segno = segno;
c192f7a4 2276 return 1;
e066b83c 2277 }
43727527 2278
70d625cb
JK		 2279 /* For node segments, let's do SSR more intensively */
2280 if (IS_NODESEG(type)) {
d27c3d89
CY		 2281 if (type >= CURSEG_WARM_NODE) {
2282 reversed = true;
2283 i = CURSEG_COLD_NODE;
2284 } else {
2285 i = CURSEG_HOT_NODE;
2286 }
2287 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2288 } else {
d27c3d89
CY		 2289 if (type >= CURSEG_WARM_DATA) {
2290 reversed = true;
2291 i = CURSEG_COLD_DATA;
2292 } else {
2293 i = CURSEG_HOT_DATA;
2294 }
2295 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2296 }
43727527 2297
d27c3d89 2298 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2299 if (i == type)
2300 continue;
e066b83c
JK		 2301 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2302 curseg->next_segno = segno;
43727527 2303 return 1;
e066b83c 2304 }
c192f7a4 2305 }
43727527
JK		 2306 return 0;
2307}
2308
351df4b2
JK		 2309/*
2310 * flush out current segment and replace it with new segment
2311 * This function should be returned with success, otherwise BUG
2312 */
2313static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2314 int type, bool force)
2315{
a7881893
JK		 2316 struct curseg_info *curseg = CURSEG_I(sbi, type);
2317
7b405275 2318 if (force)
351df4b2 2319 new_curseg(sbi, type, true);
5b6c6be2
JK		 2320 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2321 type == CURSEG_WARM_NODE)
351df4b2 2322 new_curseg(sbi, type, false);
a7881893
JK		 2323 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2324 new_curseg(sbi, type, false);
4d57b86d 2325 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2326 change_curseg(sbi, type);
351df4b2
JK		 2327 else
2328 new_curseg(sbi, type, false);
dcdfff65 2329
a7881893 2330 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2331}
2332
4d57b86d 2333void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
351df4b2 2334{
6ae1be13
JK		 2335 struct curseg_info *curseg;
2336 unsigned int old_segno;
351df4b2
JK		 2337 int i;
2338
3d26fa6b
CY		 2339 down_write(&SIT_I(sbi)->sentry_lock);
2340
6ae1be13
JK		 2341 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2342 curseg = CURSEG_I(sbi, i);
2343 old_segno = curseg->segno;
2344 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2345 locate_dirty_segment(sbi, old_segno);
2346 }
3d26fa6b
CY		 2347
2348 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2349}
2350
2351static const struct segment_allocation default_salloc_ops = {
2352 .allocate_segment = allocate_segment_by_default,
2353};
2354
4d57b86d
CY		 2355bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2356 struct cp_control *cpc)
25290fa5
JK		 2357{
2358 __u64 trim_start = cpc->trim_start;
2359 bool has_candidate = false;
2360
3d26fa6b 2361 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2362 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2363 if (add_discard_addrs(sbi, cpc, true)) {
2364 has_candidate = true;
2365 break;
2366 }
2367 }
3d26fa6b 2368 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2369
2370 cpc->trim_start = trim_start;
2371 return has_candidate;
2372}
2373
01f9cf6d 2374static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 2375 struct discard_policy *dpolicy,
2376 unsigned int start, unsigned int end)
2377{
2378 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2379 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2380 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2381 struct discard_cmd *dc;
2382 struct blk_plug plug;
2383 int issued;
01f9cf6d 2384 unsigned int trimmed = 0;
9a997188
JK		 2385
2386next:
2387 issued = 0;
2388
2389 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 2390 if (unlikely(dcc->rbtree_check))
2391 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2392 &dcc->root));
9a997188 2393
4d57b86d 2394 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 2395 NULL, start,
2396 (struct rb_entry **)&prev_dc,
2397 (struct rb_entry **)&next_dc,
2398 &insert_p, &insert_parent, true);
2399 if (!dc)
2400 dc = next_dc;
2401
2402 blk_start_plug(&plug);
2403
2404 while (dc && dc->lstart <= end) {
2405 struct rb_node *node;
2406
2407 if (dc->len < dpolicy->granularity)
2408 goto skip;
2409
2410 if (dc->state != D_PREP) {
2411 list_move_tail(&dc->list, &dcc->fstrim_list);
2412 goto skip;
2413 }
2414
2415 __submit_discard_cmd(sbi, dpolicy, dc);
2416
2417 if (++issued >= dpolicy->max_requests) {
2418 start = dc->lstart + dc->len;
2419
2420 blk_finish_plug(&plug);
2421 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 2422 trimmed += __wait_all_discard_cmd(sbi, NULL);
9a997188
JK		 2423 congestion_wait(BLK_RW_ASYNC, HZ/50);
2424 goto next;
2425 }
2426skip:
2427 node = rb_next(&dc->rb_node);
2428 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2429
2430 if (fatal_signal_pending(current))
2431 break;
2432 }
2433
2434 blk_finish_plug(&plug);
2435 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 2436
2437 return trimmed;
9a997188
JK		 2438}
2439
4b2fecc8
JK		 2440int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2441{
f7ef9b83
JK		 2442 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2443 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2444 unsigned int start_segno, end_segno;
8412663d 2445 block_t start_block, end_block;
4b2fecc8 2446 struct cp_control cpc;
78997b56 2447 struct discard_policy dpolicy;
0ea80512 2448 unsigned long long trimmed = 0;
c34f42e2 2449 int err = 0;
4b2fecc8 2450
836b5a63 2451 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2452 return -EINVAL;
2453
7cd8558b 2454 if (end <= MAIN_BLKADDR(sbi))
3d165dc3 2455 return -EINVAL;
4b2fecc8 2456
ed214a11
YH		 2457 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2458 f2fs_msg(sbi->sb, KERN_WARNING,
2459 "Found FS corruption, run fsck to fix.");
3d165dc3 2460 return -EIO;
ed214a11
YH		 2461 }
2462
4b2fecc8 2463 /* start/end segment number in main_area */
7cd8558b
JK		 2464 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2465 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2466 GET_SEGNO(sbi, end);
8412663d 2467
4b2fecc8 2468 cpc.reason = CP_DISCARD;
836b5a63 2469 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2470 cpc.trim_start = start_segno;
2471 cpc.trim_end = end_segno;
4b2fecc8 2472
377224c4
CY		 2473 if (sbi->discard_blks == 0)
2474 goto out;
74fa5f3d 2475
377224c4 2476 mutex_lock(&sbi->gc_mutex);
4d57b86d 2477 err = f2fs_write_checkpoint(sbi, &cpc);
377224c4
CY		 2478 mutex_unlock(&sbi->gc_mutex);
2479 if (err)
2480 goto out;
8412663d 2481
e555da9f
JK		 2482 /*
2483 * We filed discard candidates, but actually we don't need to wait for
2484 * all of them, since they'll be issued in idle time along with runtime
2485 * discard option. User configuration looks like using runtime discard
2486 * or periodic fstrim instead of it.
2487 */
5a615492
JK		 2488 if (test_opt(sbi, DISCARD))
2489 goto out;
2490
2491 start_block = START_BLOCK(sbi, start_segno);
2492 end_block = START_BLOCK(sbi, end_segno + 1);
2493
2494 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 2495 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2496 start_block, end_block);
5a615492 2497
01f9cf6d 2498 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2499 start_block, end_block);
5a615492 2500 range->len = F2FS_BLK_TO_BYTES(trimmed);
377224c4 2501out:
c34f42e2 2502 return err;
4b2fecc8
JK		 2503}
2504
351df4b2
JK		 2505static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2506{
2507 struct curseg_info *curseg = CURSEG_I(sbi, type);
2508 if (curseg->next_blkoff < sbi->blocks_per_seg)
2509 return true;
2510 return false;
2511}
2512
4d57b86d 2513int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 2514{
2515 switch (hint) {
2516 case WRITE_LIFE_SHORT:
2517 return CURSEG_HOT_DATA;
2518 case WRITE_LIFE_EXTREME:
2519 return CURSEG_COLD_DATA;
2520 default:
2521 return CURSEG_WARM_DATA;
2522 }
2523}
2524
0cdd3195
HL		 2525/* This returns write hints for each segment type. This hints will be
2526 * passed down to block layer. There are mapping tables which depend on
2527 * the mount option 'whint_mode'.
2528 *
2529 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2530 *
2531 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2532 *
2533 * User F2FS Block
2534 * ---- ---- -----
2535 * META WRITE_LIFE_NOT_SET
2536 * HOT_NODE "
2537 * WARM_NODE "
2538 * COLD_NODE "
2539 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2540 * extension list " "
2541 *
2542 * -- buffered io
2543 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2544 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2545 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2546 * WRITE_LIFE_NONE " "
2547 * WRITE_LIFE_MEDIUM " "
2548 * WRITE_LIFE_LONG " "
2549 *
2550 * -- direct io
2551 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2552 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2553 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2554 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2555 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2556 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2557 *
f2e703f9
HL		 2558 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2559 *
2560 * User F2FS Block
2561 * ---- ---- -----
2562 * META WRITE_LIFE_MEDIUM;
2563 * HOT_NODE WRITE_LIFE_NOT_SET
2564 * WARM_NODE "
2565 * COLD_NODE WRITE_LIFE_NONE
2566 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2567 * extension list " "
2568 *
2569 * -- buffered io
2570 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2571 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2572 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2573 * WRITE_LIFE_NONE " "
2574 * WRITE_LIFE_MEDIUM " "
2575 * WRITE_LIFE_LONG " "
2576 *
2577 * -- direct io
2578 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2579 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2580 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2581 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2582 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2583 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2584 */
2585
4d57b86d 2586enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 2587 enum page_type type, enum temp_type temp)
2588{
63189b78 2589 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2590 if (type == DATA) {
f2e703f9 2591 if (temp == WARM)
0cdd3195 2592 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2593 else if (temp == HOT)
2594 return WRITE_LIFE_SHORT;
2595 else if (temp == COLD)
2596 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2597 } else {
2598 return WRITE_LIFE_NOT_SET;
2599 }
63189b78 2600 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2601 if (type == DATA) {
2602 if (temp == WARM)
2603 return WRITE_LIFE_LONG;
2604 else if (temp == HOT)
2605 return WRITE_LIFE_SHORT;
2606 else if (temp == COLD)
2607 return WRITE_LIFE_EXTREME;
2608 } else if (type == NODE) {
2609 if (temp == WARM || temp == HOT)
2610 return WRITE_LIFE_NOT_SET;
2611 else if (temp == COLD)
2612 return WRITE_LIFE_NONE;
2613 } else if (type == META) {
2614 return WRITE_LIFE_MEDIUM;
2615 }
0cdd3195 2616 }
f2e703f9 2617 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2618}
2619
81377bd6 2620static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2621{
81377bd6 2622 if (fio->type == DATA)
351df4b2
JK		 2623 return CURSEG_HOT_DATA;
2624 else
2625 return CURSEG_HOT_NODE;
2626}
2627
81377bd6 2628static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2629{
81377bd6
JK		 2630 if (fio->type == DATA) {
2631 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2632
2633 if (S_ISDIR(inode->i_mode))
2634 return CURSEG_HOT_DATA;
2635 else
2636 return CURSEG_COLD_DATA;
2637 } else {
81377bd6 2638 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2639 return CURSEG_WARM_NODE;
351df4b2
JK		 2640 else
2641 return CURSEG_COLD_NODE;
2642 }
2643}
2644
81377bd6 2645static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2646{
81377bd6
JK		 2647 if (fio->type == DATA) {
2648 struct inode *inode = fio->page->mapping->host;
351df4b2 2649
81377bd6 2650 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2651 return CURSEG_COLD_DATA;
b6a06cbb 2652 if (file_is_hot(inode) ||
b4c3ca8b
CY		 2653 is_inode_flag_set(inode, FI_HOT_DATA) ||
2654 is_inode_flag_set(inode, FI_ATOMIC_FILE) ||
2655 is_inode_flag_set(inode, FI_VOLATILE_FILE))
ef095d19 2656 return CURSEG_HOT_DATA;
4d57b86d 2657 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2658 } else {
81377bd6
JK		 2659 if (IS_DNODE(fio->page))
2660 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2661 CURSEG_HOT_NODE;
ef095d19 2662 return CURSEG_COLD_NODE;
351df4b2
JK		 2663 }
2664}
2665
81377bd6 2666static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2667{
a912b54d
JK		 2668 int type = 0;
2669
63189b78 2670 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2671 case 2:
a912b54d
JK		 2672 type = __get_segment_type_2(fio);
2673 break;
351df4b2 2674 case 4:
a912b54d
JK		 2675 type = __get_segment_type_4(fio);
2676 break;
2677 case 6:
2678 type = __get_segment_type_6(fio);
2679 break;
2680 default:
2681 f2fs_bug_on(fio->sbi, true);
351df4b2 2682 }
81377bd6 2683
a912b54d
JK		 2684 if (IS_HOT(type))
2685 fio->temp = HOT;
2686 else if (IS_WARM(type))
2687 fio->temp = WARM;
2688 else
2689 fio->temp = COLD;
2690 return type;
351df4b2
JK		 2691}
2692
4d57b86d 2693void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 2694 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2695 struct f2fs_summary *sum, int type,
2696 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2697{
2698 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2699 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2700
2b60311d
CY		 2701 down_read(&SM_I(sbi)->curseg_lock);
2702
351df4b2 2703 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2704 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2705
2706 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2707
4e6a8d9b
JK		 2708 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2709
351df4b2
JK		 2710 /*
2711 * __add_sum_entry should be resided under the curseg_mutex
2712 * because, this function updates a summary entry in the
2713 * current summary block.
2714 */
e79efe3b 2715 __add_sum_entry(sbi, type, sum);
351df4b2 2716
351df4b2 2717 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2718
2719 stat_inc_block_count(sbi, curseg);
351df4b2 2720
65f1b80b
YS		 2721 /*
2722 * SIT information should be updated before segment allocation,
2723 * since SSR needs latest valid block information.
2724 */
2725 update_sit_entry(sbi, *new_blkaddr, 1);
2726 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2727 update_sit_entry(sbi, old_blkaddr, -1);
2728
c6f82fe9
JK		 2729 if (!__has_curseg_space(sbi, type))
2730 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 2731
351df4b2 2732 /*
65f1b80b
YS		 2733 * segment dirty status should be updated after segment allocation,
2734 * so we just need to update status only one time after previous
2735 * segment being closed.
351df4b2 2736 */
65f1b80b
YS		 2737 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2738 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 2739
3d26fa6b 2740 up_write(&sit_i->sentry_lock);
351df4b2 2741
704956ec 2742 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2743 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2744
704956ec
CY		 2745 f2fs_inode_chksum_set(sbi, page);
2746 }
2747
fb830fc5
CY		 2748 if (add_list) {
2749 struct f2fs_bio_info *io;
2750
2751 INIT_LIST_HEAD(&fio->list);
2752 fio->in_list = true;
fe16efe6 2753 fio->retry = false;
fb830fc5
CY		 2754 io = sbi->write_io[fio->type] + fio->temp;
2755 spin_lock(&io->io_lock);
2756 list_add_tail(&fio->list, &io->io_list);
2757 spin_unlock(&io->io_lock);
2758 }
2759
bfad7c2d 2760 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 2761
2762 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 2763}
2764
39d787be
CY		 2765static void update_device_state(struct f2fs_io_info *fio)
2766{
2767 struct f2fs_sb_info *sbi = fio->sbi;
2768 unsigned int devidx;
2769
2770 if (!sbi->s_ndevs)
2771 return;
2772
2773 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2774
2775 /* update device state for fsync */
4d57b86d 2776 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2777
2778 /* update device state for checkpoint */
2779 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2780 spin_lock(&sbi->dev_lock);
2781 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2782 spin_unlock(&sbi->dev_lock);
2783 }
39d787be
CY		 2784}
2785
05ca3632 2786static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2787{
81377bd6 2788 int type = __get_segment_type(fio);
107a805d 2789 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 2790
107a805d
CY		 2791 if (keep_order)
2792 down_read(&fio->sbi->io_order_lock);
0a595eba 2793reallocate:
4d57b86d 2794 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2795 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2796
351df4b2 2797 /* writeout dirty page into bdev */
fe16efe6
CY		 2798 f2fs_submit_page_write(fio);
2799 if (fio->retry) {
0a595eba
JK		 2800 fio->old_blkaddr = fio->new_blkaddr;
2801 goto reallocate;
2802 }
fe16efe6
CY		 2803
2804 update_device_state(fio);
2805
107a805d
CY		 2806 if (keep_order)
2807 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 2808}
2809
4d57b86d 2810void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 2811 enum iostat_type io_type)
351df4b2 2812{
458e6197 2813 struct f2fs_io_info fio = {
05ca3632 2814 .sbi = sbi,
458e6197 2815 .type = META,
0cdd3195 2816 .temp = HOT,
04d328de 2817 .op = REQ_OP_WRITE,
70fd7614 2818 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2819 .old_blkaddr = page->index,
2820 .new_blkaddr = page->index,
05ca3632 2821 .page = page,
4375a336 2822 .encrypted_page = NULL,
fb830fc5 2823 .in_list = false,
458e6197
JK		 2824 };
2825
2b947003 2826 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2827 fio.op_flags &= ~REQ_META;
2b947003 2828
351df4b2 2829 set_page_writeback(page);
17c50035 2830 ClearPageError(page);
b9109b0e 2831 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2832
2833 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2834}
2835
4d57b86d 2836void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2837{
2838 struct f2fs_summary sum;
05ca3632 2839
351df4b2 2840 set_summary(&sum, nid, 0, 0);
05ca3632 2841 do_write_page(&sum, fio);
b0af6d49
CY		 2842
2843 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2844}
2845
4d57b86d
CY		 2846void f2fs_outplace_write_data(struct dnode_of_data *dn,
2847 struct f2fs_io_info *fio)
351df4b2 2848{
05ca3632 2849 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2850 struct f2fs_summary sum;
2851 struct node_info ni;
2852
9850cf4a 2853 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
4d57b86d 2854 f2fs_get_node_info(sbi, dn->nid, &ni);
351df4b2 2855 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2856 do_write_page(&sum, fio);
f28b3434 2857 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2858
2859 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2860}
2861
4d57b86d 2862int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 2863{
b0af6d49 2864 int err;
d21b0f23 2865 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 2866
7a9d7548 2867 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 2868 /* i/o temperature is needed for passing down write hints */
2869 __get_segment_type(fio);
d21b0f23
YH		 2870
2871 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2872 GET_SEGNO(sbi, fio->new_blkaddr))->type));
2873
05ca3632 2874 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2875
2876 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2877 if (!err)
2878 update_device_state(fio);
b0af6d49
CY		 2879
2880 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2881
2882 return err;
351df4b2
JK		 2883}
2884
2b60311d
CY		 2885static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2886 unsigned int segno)
2887{
2888 int i;
2889
2890 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2891 if (CURSEG_I(sbi, i)->segno == segno)
2892 break;
2893 }
2894 return i;
2895}
2896
4d57b86d 2897void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2898 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2899 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2900{
2901 struct sit_info *sit_i = SIT_I(sbi);
2902 struct curseg_info *curseg;
2903 unsigned int segno, old_cursegno;
2904 struct seg_entry *se;
2905 int type;
19f106bc 2906 unsigned short old_blkoff;
351df4b2
JK		 2907
2908 segno = GET_SEGNO(sbi, new_blkaddr);
2909 se = get_seg_entry(sbi, segno);
2910 type = se->type;
2911
2b60311d
CY		 2912 down_write(&SM_I(sbi)->curseg_lock);
2913
19f106bc
CY		 2914 if (!recover_curseg) {
2915 /* for recovery flow */
2916 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2917 if (old_blkaddr == NULL_ADDR)
2918 type = CURSEG_COLD_DATA;
2919 else
2920 type = CURSEG_WARM_DATA;
2921 }
2922 } else {
2b60311d
CY		 2923 if (IS_CURSEG(sbi, segno)) {
2924 /* se->type is volatile as SSR allocation */
2925 type = __f2fs_get_curseg(sbi, segno);
2926 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2927 } else {
351df4b2 2928 type = CURSEG_WARM_DATA;
2b60311d 2929 }
351df4b2 2930 }
19f106bc 2931
2c190504 2932 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 2933 curseg = CURSEG_I(sbi, type);
2934
2935 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2936 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2937
2938 old_cursegno = curseg->segno;
19f106bc 2939 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2940
2941 /* change the current segment */
2942 if (segno != curseg->segno) {
2943 curseg->next_segno = segno;
025d63a4 2944 change_curseg(sbi, type);
351df4b2
JK		 2945 }
2946
491c0854 2947 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2948 __add_sum_entry(sbi, type, sum);
351df4b2 2949
28bc106b 2950 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2951 update_sit_entry(sbi, new_blkaddr, 1);
2952 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2953 update_sit_entry(sbi, old_blkaddr, -1);
2954
2955 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2956 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2957
351df4b2 2958 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2959
19f106bc
CY		 2960 if (recover_curseg) {
2961 if (old_cursegno != curseg->segno) {
2962 curseg->next_segno = old_cursegno;
025d63a4 2963 change_curseg(sbi, type);
19f106bc
CY		 2964 }
2965 curseg->next_blkoff = old_blkoff;
2966 }
2967
3d26fa6b 2968 up_write(&sit_i->sentry_lock);
351df4b2 2969 mutex_unlock(&curseg->curseg_mutex);
2b60311d 2970 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 2971}
2972
528e3459
CY		 2973void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2974 block_t old_addr, block_t new_addr,
28bc106b
CY		 2975 unsigned char version, bool recover_curseg,
2976 bool recover_newaddr)
528e3459
CY		 2977{
2978 struct f2fs_summary sum;
2979
2980 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2981
4d57b86d 2982 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
28bc106b 2983 recover_curseg, recover_newaddr);
528e3459 2984
f28b3434 2985 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2986}
2987
93dfe2ac 2988void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2989 enum page_type type, bool ordered)
93dfe2ac 2990{
93dfe2ac 2991 if (PageWriteback(page)) {
4081363f
JK		 2992 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2993
b9109b0e
JK		 2994 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2995 0, page->index, type);
fec1d657
JK		 2996 if (ordered)
2997 wait_on_page_writeback(page);
2998 else
2999 wait_for_stable_page(page);
93dfe2ac
JK		 3000 }
3001}
3002
d4c759ee 3003void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 3004{
3005 struct page *cpage;
3006
e1da7872 3007 if (!is_valid_data_blkaddr(sbi, blkaddr))
08b39fbd
CY		 3008 return;
3009
08b39fbd
CY		 3010 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3011 if (cpage) {
fec1d657 3012 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 3013 f2fs_put_page(cpage, 1);
3014 }
3015}
3016
c376fc0f 3017static void read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3018{
3019 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3020 struct curseg_info *seg_i;
3021 unsigned char *kaddr;
3022 struct page *page;
3023 block_t start;
3024 int i, j, offset;
3025
3026 start = start_sum_block(sbi);
3027
4d57b86d 3028 page = f2fs_get_meta_page(sbi, start++);
351df4b2
JK		 3029 kaddr = (unsigned char *)page_address(page);
3030
3031 /* Step 1: restore nat cache */
3032 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3033 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3034
3035 /* Step 2: restore sit cache */
3036 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3037 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3038 offset = 2 * SUM_JOURNAL_SIZE;
3039
3040 /* Step 3: restore summary entries */
3041 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3042 unsigned short blk_off;
3043 unsigned int segno;
3044
3045 seg_i = CURSEG_I(sbi, i);
3046 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3047 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3048 seg_i->next_segno = segno;
3049 reset_curseg(sbi, i, 0);
3050 seg_i->alloc_type = ckpt->alloc_type[i];
3051 seg_i->next_blkoff = blk_off;
3052
3053 if (seg_i->alloc_type == SSR)
3054 blk_off = sbi->blocks_per_seg;
3055
3056 for (j = 0; j < blk_off; j++) {
3057 struct f2fs_summary *s;
3058 s = (struct f2fs_summary *)(kaddr + offset);
3059 seg_i->sum_blk->entries[j] = *s;
3060 offset += SUMMARY_SIZE;
09cbfeaf 3061 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3062 SUM_FOOTER_SIZE)
3063 continue;
3064
3065 f2fs_put_page(page, 1);
3066 page = NULL;
3067
4d57b86d 3068 page = f2fs_get_meta_page(sbi, start++);
351df4b2
JK		 3069 kaddr = (unsigned char *)page_address(page);
3070 offset = 0;
3071 }
3072 }
3073 f2fs_put_page(page, 1);
351df4b2
JK		 3074}
3075
3076static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3077{
3078 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3079 struct f2fs_summary_block *sum;
3080 struct curseg_info *curseg;
3081 struct page *new;
3082 unsigned short blk_off;
3083 unsigned int segno = 0;
3084 block_t blk_addr = 0;
3085
3086 /* get segment number and block addr */
3087 if (IS_DATASEG(type)) {
3088 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3089 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3090 CURSEG_HOT_DATA]);
119ee914 3091 if (__exist_node_summaries(sbi))
351df4b2
JK		 3092 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3093 else
3094 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3095 } else {
3096 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3097 CURSEG_HOT_NODE]);
3098 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3099 CURSEG_HOT_NODE]);
119ee914 3100 if (__exist_node_summaries(sbi))
351df4b2
JK		 3101 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3102 type - CURSEG_HOT_NODE);
3103 else
3104 blk_addr = GET_SUM_BLOCK(sbi, segno);
3105 }
3106
4d57b86d 3107 new = f2fs_get_meta_page(sbi, blk_addr);
351df4b2
JK		 3108 sum = (struct f2fs_summary_block *)page_address(new);
3109
3110 if (IS_NODESEG(type)) {
119ee914 3111 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3112 struct f2fs_summary *ns = &sum->entries[0];
3113 int i;
3114 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3115 ns->version = 0;
3116 ns->ofs_in_node = 0;
3117 }
3118 } else {
4d57b86d 3119 f2fs_restore_node_summary(sbi, segno, sum);
351df4b2
JK		 3120 }
3121 }
3122
3123 /* set uncompleted segment to curseg */
3124 curseg = CURSEG_I(sbi, type);
3125 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3126
3127 /* update journal info */
3128 down_write(&curseg->journal_rwsem);
3129 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3130 up_write(&curseg->journal_rwsem);
3131
3132 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3133 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3134 curseg->next_segno = segno;
3135 reset_curseg(sbi, type, 0);
3136 curseg->alloc_type = ckpt->alloc_type[type];
3137 curseg->next_blkoff = blk_off;
3138 mutex_unlock(&curseg->curseg_mutex);
3139 f2fs_put_page(new, 1);
3140 return 0;
3141}
3142
3143static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3144{
21d3f8e1
JQ		 3145 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3146 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3147 int type = CURSEG_HOT_DATA;
e4fc5fbf 3148 int err;
351df4b2 3149
aaec2b1d 3150 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3151 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3152
3153 if (npages >= 2)
4d57b86d 3154 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3155 META_CP, true);
3fa06d7b 3156
351df4b2 3157 /* restore for compacted data summary */
c376fc0f 3158 read_compacted_summaries(sbi);
351df4b2
JK		 3159 type = CURSEG_HOT_NODE;
3160 }
3161
119ee914 3162 if (__exist_node_summaries(sbi))
4d57b86d 3163 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3164 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3165
e4fc5fbf
CY		 3166 for (; type <= CURSEG_COLD_NODE; type++) {
3167 err = read_normal_summaries(sbi, type);
3168 if (err)
3169 return err;
3170 }
3171
21d3f8e1
JQ		 3172 /* sanity check for summary blocks */
3173 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3174 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3175 return -EINVAL;
3176
351df4b2
JK		 3177 return 0;
3178}
3179
3180static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3181{
3182 struct page *page;
3183 unsigned char *kaddr;
3184 struct f2fs_summary *summary;
3185 struct curseg_info *seg_i;
3186 int written_size = 0;
3187 int i, j;
3188
4d57b86d 3189 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3190 kaddr = (unsigned char *)page_address(page);
81114baa 3191 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3192
3193 /* Step 1: write nat cache */
3194 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3195 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3196 written_size += SUM_JOURNAL_SIZE;
3197
3198 /* Step 2: write sit cache */
3199 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3200 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3201 written_size += SUM_JOURNAL_SIZE;
3202
351df4b2
JK		 3203 /* Step 3: write summary entries */
3204 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3205 unsigned short blkoff;
3206 seg_i = CURSEG_I(sbi, i);
3207 if (sbi->ckpt->alloc_type[i] == SSR)
3208 blkoff = sbi->blocks_per_seg;
3209 else
3210 blkoff = curseg_blkoff(sbi, i);
3211
3212 for (j = 0; j < blkoff; j++) {
3213 if (!page) {
4d57b86d 3214 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3215 kaddr = (unsigned char *)page_address(page);
81114baa 3216 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3217 written_size = 0;
3218 }
3219 summary = (struct f2fs_summary *)(kaddr + written_size);
3220 *summary = seg_i->sum_blk->entries[j];
3221 written_size += SUMMARY_SIZE;
351df4b2 3222
09cbfeaf 3223 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3224 SUM_FOOTER_SIZE)
3225 continue;
3226
e8d61a74 3227 set_page_dirty(page);
351df4b2
JK		 3228 f2fs_put_page(page, 1);
3229 page = NULL;
3230 }
3231 }
e8d61a74
CY		 3232 if (page) {
3233 set_page_dirty(page);
351df4b2 3234 f2fs_put_page(page, 1);
e8d61a74 3235 }
351df4b2
JK		 3236}
3237
3238static void write_normal_summaries(struct f2fs_sb_info *sbi,
3239 block_t blkaddr, int type)
3240{
3241 int i, end;
3242 if (IS_DATASEG(type))
3243 end = type + NR_CURSEG_DATA_TYPE;
3244 else
3245 end = type + NR_CURSEG_NODE_TYPE;
3246
b7ad7512
CY		 3247 for (i = type; i < end; i++)
3248 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3249}
3250
4d57b86d 3251void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3252{
aaec2b1d 3253 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3254 write_compacted_summaries(sbi, start_blk);
3255 else
3256 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3257}
3258
4d57b86d 3259void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3260{
119ee914 3261 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3262}
3263
4d57b86d 3264int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3265 unsigned int val, int alloc)
3266{
3267 int i;
3268
3269 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3270 for (i = 0; i < nats_in_cursum(journal); i++) {
3271 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3272 return i;
3273 }
dfc08a12
CY		 3274 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3275 return update_nats_in_cursum(journal, 1);
351df4b2 3276 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3277 for (i = 0; i < sits_in_cursum(journal); i++)
3278 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3279 return i;
dfc08a12
CY		 3280 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3281 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3282 }
3283 return -1;
3284}
3285
3286static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3287 unsigned int segno)
3288{
4d57b86d 3289 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3290}
3291
3292static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3293 unsigned int start)
3294{
3295 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3296 struct page *page;
351df4b2 3297 pgoff_t src_off, dst_off;
351df4b2
JK		 3298
3299 src_off = current_sit_addr(sbi, start);
3300 dst_off = next_sit_addr(sbi, src_off);
3301
4d57b86d 3302 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 3303 seg_info_to_sit_page(sbi, page, start);
351df4b2 3304
068c3cd8 3305 set_page_dirty(page);
351df4b2
JK		 3306 set_to_next_sit(sit_i, start);
3307
068c3cd8 3308 return page;
351df4b2
JK		 3309}
3310
184a5cd2
CY		 3311static struct sit_entry_set *grab_sit_entry_set(void)
3312{
3313 struct sit_entry_set *ses =
80c54505 3314 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3315
3316 ses->entry_cnt = 0;
3317 INIT_LIST_HEAD(&ses->set_list);
3318 return ses;
3319}
3320
3321static void release_sit_entry_set(struct sit_entry_set *ses)
3322{
3323 list_del(&ses->set_list);
3324 kmem_cache_free(sit_entry_set_slab, ses);
3325}
3326
3327static void adjust_sit_entry_set(struct sit_entry_set *ses,
3328 struct list_head *head)
3329{
3330 struct sit_entry_set *next = ses;
3331
3332 if (list_is_last(&ses->set_list, head))
3333 return;
3334
3335 list_for_each_entry_continue(next, head, set_list)
3336 if (ses->entry_cnt <= next->entry_cnt)
3337 break;
3338
3339 list_move_tail(&ses->set_list, &next->set_list);
3340}
3341
3342static void add_sit_entry(unsigned int segno, struct list_head *head)
3343{
3344 struct sit_entry_set *ses;
3345 unsigned int start_segno = START_SEGNO(segno);
3346
3347 list_for_each_entry(ses, head, set_list) {
3348 if (ses->start_segno == start_segno) {
3349 ses->entry_cnt++;
3350 adjust_sit_entry_set(ses, head);
3351 return;
3352 }
3353 }
3354
3355 ses = grab_sit_entry_set();
3356
3357 ses->start_segno = start_segno;
3358 ses->entry_cnt++;
3359 list_add(&ses->set_list, head);
3360}
3361
3362static void add_sits_in_set(struct f2fs_sb_info *sbi)
3363{
3364 struct f2fs_sm_info *sm_info = SM_I(sbi);
3365 struct list_head *set_list = &sm_info->sit_entry_set;
3366 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3367 unsigned int segno;
3368
7cd8558b 3369 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3370 add_sit_entry(segno, set_list);
3371}
3372
3373static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3374{
3375 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3376 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3377 int i;
3378
b7ad7512 3379 down_write(&curseg->journal_rwsem);
dfc08a12 3380 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3381 unsigned int segno;
3382 bool dirtied;
3383
dfc08a12 3384 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3385 dirtied = __mark_sit_entry_dirty(sbi, segno);
3386
3387 if (!dirtied)
3388 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3389 }
dfc08a12 3390 update_sits_in_cursum(journal, -i);
b7ad7512 3391 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3392}
3393
0a8165d7 3394/*
351df4b2
JK		 3395 * CP calls this function, which flushes SIT entries including sit_journal,
3396 * and moves prefree segs to free segs.
3397 */
4d57b86d 3398void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3399{
3400 struct sit_info *sit_i = SIT_I(sbi);
3401 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3402 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3403 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3404 struct sit_entry_set *ses, *tmp;
3405 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3406 bool to_journal = true;
4b2fecc8 3407 struct seg_entry *se;
351df4b2 3408
3d26fa6b 3409 down_write(&sit_i->sentry_lock);
351df4b2 3410
2b11a74b
WL		 3411 if (!sit_i->dirty_sentries)
3412 goto out;
3413
351df4b2 3414 /*
184a5cd2
CY		 3415 * add and account sit entries of dirty bitmap in sit entry
3416 * set temporarily
351df4b2 3417 */
184a5cd2 3418 add_sits_in_set(sbi);
351df4b2 3419
184a5cd2
CY		 3420 /*
3421 * if there are no enough space in journal to store dirty sit
3422 * entries, remove all entries from journal and add and account
3423 * them in sit entry set.
3424 */
dfc08a12 3425 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3426 remove_sits_in_journal(sbi);
b2955550 3427
184a5cd2
CY		 3428 /*
3429 * there are two steps to flush sit entries:
3430 * #1, flush sit entries to journal in current cold data summary block.
3431 * #2, flush sit entries to sit page.
3432 */
3433 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3434 struct page *page = NULL;
184a5cd2
CY		 3435 struct f2fs_sit_block *raw_sit = NULL;
3436 unsigned int start_segno = ses->start_segno;
3437 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3438 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3439 unsigned int segno = start_segno;
3440
3441 if (to_journal &&
dfc08a12 3442 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3443 to_journal = false;
3444
b7ad7512
CY		 3445 if (to_journal) {
3446 down_write(&curseg->journal_rwsem);
3447 } else {
184a5cd2
CY		 3448 page = get_next_sit_page(sbi, start_segno);
3449 raw_sit = page_address(page);
351df4b2 3450 }
351df4b2 3451
184a5cd2
CY		 3452 /* flush dirty sit entries in region of current sit set */
3453 for_each_set_bit_from(segno, bitmap, end) {
3454 int offset, sit_offset;
4b2fecc8
JK		 3455
3456 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3457#ifdef CONFIG_F2FS_CHECK_FS
3458 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3459 SIT_VBLOCK_MAP_SIZE))
3460 f2fs_bug_on(sbi, 1);
3461#endif
184a5cd2
CY		 3462
3463 /* add discard candidates */
c473f1a9 3464 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3465 cpc->trim_start = segno;
25290fa5 3466 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3467 }
184a5cd2
CY		 3468
3469 if (to_journal) {
4d57b86d 3470 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 3471 SIT_JOURNAL, segno, 1);
3472 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3473 segno_in_journal(journal, offset) =
184a5cd2
CY		 3474 cpu_to_le32(segno);
3475 seg_info_to_raw_sit(se,
dfc08a12 3476 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3477 check_block_count(sbi, segno,
3478 &sit_in_journal(journal, offset));
184a5cd2
CY		 3479 } else {
3480 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3481 seg_info_to_raw_sit(se,
3482 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3483 check_block_count(sbi, segno,
3484 &raw_sit->entries[sit_offset]);
184a5cd2 3485 }
351df4b2 3486
184a5cd2
CY		 3487 __clear_bit(segno, bitmap);
3488 sit_i->dirty_sentries--;
3489 ses->entry_cnt--;
351df4b2
JK		 3490 }
3491
b7ad7512
CY		 3492 if (to_journal)
3493 up_write(&curseg->journal_rwsem);
3494 else
184a5cd2
CY		 3495 f2fs_put_page(page, 1);
3496
3497 f2fs_bug_on(sbi, ses->entry_cnt);
3498 release_sit_entry_set(ses);
351df4b2 3499 }
184a5cd2
CY		 3500
3501 f2fs_bug_on(sbi, !list_empty(head));
3502 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3503out:
c473f1a9 3504 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3505 __u64 trim_start = cpc->trim_start;
3506
4b2fecc8 3507 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3508 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3509
3510 cpc->trim_start = trim_start;
4b2fecc8 3511 }
3d26fa6b 3512 up_write(&sit_i->sentry_lock);
351df4b2 3513
351df4b2
JK		 3514 set_prefree_as_free_segments(sbi);
3515}
3516
3517static int build_sit_info(struct f2fs_sb_info *sbi)
3518{
3519 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3520 struct sit_info *sit_i;
3521 unsigned int sit_segs, start;
ae27d62e 3522 char *src_bitmap;
351df4b2
JK		 3523 unsigned int bitmap_size;
3524
3525 /* allocate memory for SIT information */
acbf054d 3526 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3527 if (!sit_i)
3528 return -ENOMEM;
3529
3530 SM_I(sbi)->sit_info = sit_i;
3531
9d2a789c
KC		 3532 sit_i->sentries =
3533 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3534 MAIN_SEGS(sbi)),
3535 GFP_KERNEL);
351df4b2
JK		 3536 if (!sit_i->sentries)
3537 return -ENOMEM;
3538
7cd8558b 3539 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3540 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3541 GFP_KERNEL);
351df4b2
JK		 3542 if (!sit_i->dirty_sentries_bitmap)
3543 return -ENOMEM;
3544
7cd8558b 3545 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3546 sit_i->sentries[start].cur_valid_map
acbf054d 3547 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3548 sit_i->sentries[start].ckpt_valid_map
acbf054d 3549 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3550 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3551 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3552 return -ENOMEM;
3e025740 3553
355e7891
CY		 3554#ifdef CONFIG_F2FS_CHECK_FS
3555 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3556 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3557 if (!sit_i->sentries[start].cur_valid_map_mir)
3558 return -ENOMEM;
3559#endif
3560
3e025740
JK		 3561 if (f2fs_discard_en(sbi)) {
3562 sit_i->sentries[start].discard_map
acbf054d
CY		 3563 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3564 GFP_KERNEL);
3e025740
JK		 3565 if (!sit_i->sentries[start].discard_map)
3566 return -ENOMEM;
3567 }
351df4b2
JK		 3568 }
3569
acbf054d 3570 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3571 if (!sit_i->tmp_map)
3572 return -ENOMEM;
3573
351df4b2 3574 if (sbi->segs_per_sec > 1) {
9d2a789c
KC		 3575 sit_i->sec_entries =
3576 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3577 MAIN_SECS(sbi)),
3578 GFP_KERNEL);
351df4b2
JK		 3579 if (!sit_i->sec_entries)
3580 return -ENOMEM;
3581 }
3582
3583 /* get information related with SIT */
3584 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3585
3586 /* setup SIT bitmap from ckeckpoint pack */
3587 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3588 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3589
ae27d62e
CY		 3590 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3591 if (!sit_i->sit_bitmap)
351df4b2 3592 return -ENOMEM;
351df4b2 3593
ae27d62e
CY		 3594#ifdef CONFIG_F2FS_CHECK_FS
3595 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3596 if (!sit_i->sit_bitmap_mir)
3597 return -ENOMEM;
3598#endif
3599
351df4b2
JK		 3600 /* init SIT information */
3601 sit_i->s_ops = &default_salloc_ops;
3602
3603 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3604 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3605 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3606 sit_i->bitmap_size = bitmap_size;
3607 sit_i->dirty_sentries = 0;
3608 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3609 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3610 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3611 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3612 return 0;
3613}
3614
3615static int build_free_segmap(struct f2fs_sb_info *sbi)
3616{
351df4b2
JK		 3617 struct free_segmap_info *free_i;
3618 unsigned int bitmap_size, sec_bitmap_size;
3619
3620 /* allocate memory for free segmap information */
acbf054d 3621 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3622 if (!free_i)
3623 return -ENOMEM;
3624
3625 SM_I(sbi)->free_info = free_i;
3626
7cd8558b 3627 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3628 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3629 if (!free_i->free_segmap)
3630 return -ENOMEM;
3631
7cd8558b 3632 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3633 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3634 if (!free_i->free_secmap)
3635 return -ENOMEM;
3636
3637 /* set all segments as dirty temporarily */
3638 memset(free_i->free_segmap, 0xff, bitmap_size);
3639 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3640
3641 /* init free segmap information */
7cd8558b 3642 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3643 free_i->free_segments = 0;
3644 free_i->free_sections = 0;
1a118ccf 3645 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3646 return 0;
3647}
3648
3649static int build_curseg(struct f2fs_sb_info *sbi)
3650{
1042d60f 3651 struct curseg_info *array;
351df4b2
JK		 3652 int i;
3653
026f0507
KC		 3654 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3655 GFP_KERNEL);
351df4b2
JK		 3656 if (!array)
3657 return -ENOMEM;
3658
3659 SM_I(sbi)->curseg_array = array;
3660
3661 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3662 mutex_init(&array[i].curseg_mutex);
acbf054d 3663 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3664 if (!array[i].sum_blk)
3665 return -ENOMEM;
b7ad7512 3666 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 3667 array[i].journal = f2fs_kzalloc(sbi,
3668 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 3669 if (!array[i].journal)
3670 return -ENOMEM;
351df4b2
JK		 3671 array[i].segno = NULL_SEGNO;
3672 array[i].next_blkoff = 0;
3673 }
3674 return restore_curseg_summaries(sbi);
3675}
3676
c39a1b34 3677static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3678{
3679 struct sit_info *sit_i = SIT_I(sbi);
3680 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3681 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3682 struct seg_entry *se;
3683 struct f2fs_sit_entry sit;
74de593a
CY		 3684 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3685 unsigned int i, start, end;
3686 unsigned int readed, start_blk = 0;
c39a1b34 3687 int err = 0;
8a29c126 3688 block_t total_node_blocks = 0;
351df4b2 3689
74de593a 3690 do {
4d57b86d 3691 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
664ba972 3692 META_SIT, true);
74de593a
CY		 3693
3694 start = start_blk * sit_i->sents_per_block;
3695 end = (start_blk + readed) * sit_i->sents_per_block;
3696
7cd8558b 3697 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3698 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3699 struct page *page;
3700
9c094040 3701 se = &sit_i->sentries[start];
74de593a
CY		 3702 page = get_current_sit_page(sbi, start);
3703 sit_blk = (struct f2fs_sit_block *)page_address(page);
3704 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3705 f2fs_put_page(page, 1);
d600af23 3706
c39a1b34
JK		 3707 err = check_block_count(sbi, start, &sit);
3708 if (err)
3709 return err;
74de593a 3710 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3711 if (IS_NODESEG(se->type))
3712 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 3713
3714 /* build discard map only one time */
3e025740 3715 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3716 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3717 memset(se->discard_map, 0xff,
3718 SIT_VBLOCK_MAP_SIZE);
3719 } else {
3720 memcpy(se->discard_map,
3721 se->cur_valid_map,
3722 SIT_VBLOCK_MAP_SIZE);
3723 sbi->discard_blks +=
3724 sbi->blocks_per_seg -
3725 se->valid_blocks;
3726 }
3e025740 3727 }
a66cdd98 3728
d600af23
CY		 3729 if (sbi->segs_per_sec > 1)
3730 get_sec_entry(sbi, start)->valid_blocks +=
3731 se->valid_blocks;
351df4b2 3732 }
74de593a
CY		 3733 start_blk += readed;
3734 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3735
3736 down_read(&curseg->journal_rwsem);
3737 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3738 unsigned int old_valid_blocks;
3739
3740 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 3741 if (start >= MAIN_SEGS(sbi)) {
3742 f2fs_msg(sbi->sb, KERN_ERR,
3743 "Wrong journal entry on segno %u",
3744 start);
3745 set_sbi_flag(sbi, SBI_NEED_FSCK);
3746 err = -EINVAL;
3747 break;
3748 }
3749
d600af23
CY		 3750 se = &sit_i->sentries[start];
3751 sit = sit_in_journal(journal, i);
3752
3753 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 3754 if (IS_NODESEG(se->type))
3755 total_node_blocks -= old_valid_blocks;
d600af23 3756
c39a1b34
JK		 3757 err = check_block_count(sbi, start, &sit);
3758 if (err)
3759 break;
d600af23 3760 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3761 if (IS_NODESEG(se->type))
3762 total_node_blocks += se->valid_blocks;
d600af23
CY		 3763
3764 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3765 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3766 memset(se->discard_map, 0xff,
3767 SIT_VBLOCK_MAP_SIZE);
3768 } else {
3769 memcpy(se->discard_map, se->cur_valid_map,
3770 SIT_VBLOCK_MAP_SIZE);
a9af3fdc
CY		 3771 sbi->discard_blks += old_valid_blocks;
3772 sbi->discard_blks -= se->valid_blocks;
1f43e2ad 3773 }
d600af23
CY		 3774 }
3775
a9af3fdc 3776 if (sbi->segs_per_sec > 1) {
d600af23 3777 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 3778 se->valid_blocks;
3779 get_sec_entry(sbi, start)->valid_blocks -=
3780 old_valid_blocks;
3781 }
d600af23
CY		 3782 }
3783 up_read(&curseg->journal_rwsem);
8a29c126
JK		 3784
3785 if (!err && total_node_blocks != valid_node_count(sbi)) {
3786 f2fs_msg(sbi->sb, KERN_ERR,
3787 "SIT is corrupted node# %u vs %u",
3788 total_node_blocks, valid_node_count(sbi));
3789 set_sbi_flag(sbi, SBI_NEED_FSCK);
3790 err = -EINVAL;
3791 }
3792
c39a1b34 3793 return err;
351df4b2
JK		 3794}
3795
3796static void init_free_segmap(struct f2fs_sb_info *sbi)
3797{
3798 unsigned int start;
3799 int type;
3800
7cd8558b 3801 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3802 struct seg_entry *sentry = get_seg_entry(sbi, start);
3803 if (!sentry->valid_blocks)
3804 __set_free(sbi, start);
c79b7ff1
JK		 3805 else
3806 SIT_I(sbi)->written_valid_blocks +=
3807 sentry->valid_blocks;
351df4b2
JK		 3808 }
3809
3810 /* set use the current segments */
3811 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3812 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3813 __set_test_and_inuse(sbi, curseg_t->segno);
3814 }
3815}
3816
3817static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3818{
3819 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3820 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3821 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3822 unsigned short valid_blocks;
3823
8736fbf0 3824 while (1) {
351df4b2 3825 /* find dirty segment based on free segmap */
7cd8558b
JK		 3826 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3827 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3828 break;
3829 offset = segno + 1;
302bd348 3830 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3831 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3832 continue;
ec325b52
JK		 3833 if (valid_blocks > sbi->blocks_per_seg) {
3834 f2fs_bug_on(sbi, 1);
3835 continue;
3836 }
351df4b2
JK		 3837 mutex_lock(&dirty_i->seglist_lock);
3838 __locate_dirty_segment(sbi, segno, DIRTY);
3839 mutex_unlock(&dirty_i->seglist_lock);
3840 }
3841}
3842
5ec4e49f 3843static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3844{
3845 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3846 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3847
628b3d14 3848 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 3849 if (!dirty_i->victim_secmap)
351df4b2
JK		 3850 return -ENOMEM;
3851 return 0;
3852}
3853
3854static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3855{
3856 struct dirty_seglist_info *dirty_i;
3857 unsigned int bitmap_size, i;
3858
3859 /* allocate memory for dirty segments list information */
acbf054d
CY		 3860 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3861 GFP_KERNEL);
351df4b2
JK		 3862 if (!dirty_i)
3863 return -ENOMEM;
3864
3865 SM_I(sbi)->dirty_info = dirty_i;
3866 mutex_init(&dirty_i->seglist_lock);
3867
7cd8558b 3868 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3869
3870 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 3871 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3872 GFP_KERNEL);
351df4b2
JK		 3873 if (!dirty_i->dirty_segmap[i])
3874 return -ENOMEM;
3875 }
3876
3877 init_dirty_segmap(sbi);
5ec4e49f 3878 return init_victim_secmap(sbi);
351df4b2
JK		 3879}
3880
0a8165d7 3881/*
351df4b2
JK		 3882 * Update min, max modified time for cost-benefit GC algorithm
3883 */
3884static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3885{
3886 struct sit_info *sit_i = SIT_I(sbi);
3887 unsigned int segno;
3888
3d26fa6b 3889 down_write(&sit_i->sentry_lock);
351df4b2 3890
5ad25442 3891 sit_i->min_mtime = ULLONG_MAX;
351df4b2 3892
7cd8558b 3893 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3894 unsigned int i;
3895 unsigned long long mtime = 0;
3896
3897 for (i = 0; i < sbi->segs_per_sec; i++)
3898 mtime += get_seg_entry(sbi, segno + i)->mtime;
3899
3900 mtime = div_u64(mtime, sbi->segs_per_sec);
3901
3902 if (sit_i->min_mtime > mtime)
3903 sit_i->min_mtime = mtime;
3904 }
a1f72ac2 3905 sit_i->max_mtime = get_mtime(sbi, false);
3d26fa6b 3906 up_write(&sit_i->sentry_lock);
351df4b2
JK		 3907}
3908
4d57b86d 3909int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 3910{
3911 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3912 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3913 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3914 int err;
3915
acbf054d 3916 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 3917 if (!sm_info)
3918 return -ENOMEM;
3919
3920 /* init sm info */
3921 sbi->sm_info = sm_info;
351df4b2
JK		 3922 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3923 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3924 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3925 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3926 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3927 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3928 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3929 sm_info->rec_prefree_segments = sm_info->main_segments *
3930 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3931 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3932 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3933
52763a4b
JK		 3934 if (!test_opt(sbi, LFS))
3935 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3936 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3937 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3938 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 3939 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 3940
184a5cd2
CY		 3941 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3942
2b60311d
CY		 3943 init_rwsem(&sm_info->curseg_lock);
3944
d4fdf8ba 3945 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 3946 err = f2fs_create_flush_cmd_control(sbi);
2163d198 3947 if (err)
a688b9d9 3948 return err;
6b4afdd7
JK		 3949 }
3950
0b54fb84
JK		 3951 err = create_discard_cmd_control(sbi);
3952 if (err)
3953 return err;
3954
351df4b2
JK		 3955 err = build_sit_info(sbi);
3956 if (err)
3957 return err;
3958 err = build_free_segmap(sbi);
3959 if (err)
3960 return err;
3961 err = build_curseg(sbi);
3962 if (err)
3963 return err;
3964
3965 /* reinit free segmap based on SIT */
c39a1b34
JK		 3966 err = build_sit_entries(sbi);
3967 if (err)
3968 return err;
351df4b2
JK		 3969
3970 init_free_segmap(sbi);
3971 err = build_dirty_segmap(sbi);
3972 if (err)
3973 return err;
3974
3975 init_min_max_mtime(sbi);
3976 return 0;
3977}
3978
3979static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3980 enum dirty_type dirty_type)
3981{
3982 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3983
3984 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3985 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3986 dirty_i->nr_dirty[dirty_type] = 0;
3987 mutex_unlock(&dirty_i->seglist_lock);
3988}
3989
5ec4e49f 3990static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3991{
3992 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3993 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3994}
3995
3996static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3997{
3998 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3999 int i;
4000
4001 if (!dirty_i)
4002 return;
4003
4004 /* discard pre-free/dirty segments list */
4005 for (i = 0; i < NR_DIRTY_TYPE; i++)
4006 discard_dirty_segmap(sbi, i);
4007
5ec4e49f 4008 destroy_victim_secmap(sbi);
351df4b2
JK		 4009 SM_I(sbi)->dirty_info = NULL;
4010 kfree(dirty_i);
4011}
4012
4013static void destroy_curseg(struct f2fs_sb_info *sbi)
4014{
4015 struct curseg_info *array = SM_I(sbi)->curseg_array;
4016 int i;
4017
4018 if (!array)
4019 return;
4020 SM_I(sbi)->curseg_array = NULL;
b7ad7512 4021 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 4022 kfree(array[i].sum_blk);
b7ad7512
CY		 4023 kfree(array[i].journal);
4024 }
351df4b2
JK		 4025 kfree(array);
4026}
4027
4028static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4029{
4030 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4031 if (!free_i)
4032 return;
4033 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4034 kvfree(free_i->free_segmap);
4035 kvfree(free_i->free_secmap);
351df4b2
JK		 4036 kfree(free_i);
4037}
4038
4039static void destroy_sit_info(struct f2fs_sb_info *sbi)
4040{
4041 struct sit_info *sit_i = SIT_I(sbi);
4042 unsigned int start;
4043
4044 if (!sit_i)
4045 return;
4046
4047 if (sit_i->sentries) {
7cd8558b 4048 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 4049 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 4050#ifdef CONFIG_F2FS_CHECK_FS
4051 kfree(sit_i->sentries[start].cur_valid_map_mir);
4052#endif
351df4b2 4053 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 4054 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4055 }
4056 }
60a3b782
JK		 4057 kfree(sit_i->tmp_map);
4058
39307a8e
JK		 4059 kvfree(sit_i->sentries);
4060 kvfree(sit_i->sec_entries);
4061 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4062
4063 SM_I(sbi)->sit_info = NULL;
4064 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 4065#ifdef CONFIG_F2FS_CHECK_FS
4066 kfree(sit_i->sit_bitmap_mir);
4067#endif
351df4b2
JK		 4068 kfree(sit_i);
4069}
4070
4d57b86d 4071void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4072{
4073 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4074
3b03f724
CY		 4075 if (!sm_info)
4076 return;
4d57b86d 4077 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 4078 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4079 destroy_dirty_segmap(sbi);
4080 destroy_curseg(sbi);
4081 destroy_free_segmap(sbi);
4082 destroy_sit_info(sbi);
4083 sbi->sm_info = NULL;
4084 kfree(sm_info);
4085}
7fd9e544 4086
4d57b86d 4087int __init f2fs_create_segment_manager_caches(void)
7fd9e544
JK		 4088{
4089 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4090 sizeof(struct discard_entry));
7fd9e544 4091 if (!discard_entry_slab)
184a5cd2
CY		 4092 goto fail;
4093
b01a9201
JK		 4094 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4095 sizeof(struct discard_cmd));
4096 if (!discard_cmd_slab)
6ab2a308 4097 goto destroy_discard_entry;
275b66b0 4098
184a5cd2 4099 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4100 sizeof(struct sit_entry_set));
184a5cd2 4101 if (!sit_entry_set_slab)
b01a9201 4102 goto destroy_discard_cmd;
88b88a66
JK		 4103
4104 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4105 sizeof(struct inmem_pages));
4106 if (!inmem_entry_slab)
4107 goto destroy_sit_entry_set;
7fd9e544 4108 return 0;
184a5cd2 4109
88b88a66
JK		 4110destroy_sit_entry_set:
4111 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4112destroy_discard_cmd:
4113 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4114destroy_discard_entry:
184a5cd2
CY		 4115 kmem_cache_destroy(discard_entry_slab);
4116fail:
4117 return -ENOMEM;
7fd9e544
JK		 4118}
4119
4d57b86d 4120void f2fs_destroy_segment_manager_caches(void)
7fd9e544 4121{
184a5cd2 4122 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4123 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4124 kmem_cache_destroy(discard_entry_slab);
88b88a66 4125 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4126}
Ubuntu Jammy Linux kernel mirror