projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: fix performance issue observed with multi-thread sequential read
[linux-2.6-block.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 }
7735730d
CY		 253
254 err = f2fs_get_node_info(sbi, dn.nid, &ni);
255 if (err) {
256 f2fs_put_dnode(&dn);
257 return err;
258 }
259
f1d2564a 260 if (cur->old_addr == NEW_ADDR) {
4d57b86d 261 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
f1d2564a
DJ		 262 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
263 } else
264 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 265 cur->old_addr, ni.version, true, true);
266 f2fs_put_dnode(&dn);
267 }
268next:
63c52d78
JK		 269 /* we don't need to invalidate this in the sccessful status */
270 if (drop || recover)
271 ClearPageUptodate(page);
28bc106b 272 set_page_private(page, 0);
c81ced05 273 ClearPagePrivate(page);
28bc106b 274 f2fs_put_page(page, 1);
29b96b54
CY		 275
276 list_del(&cur->list);
277 kmem_cache_free(inmem_entry_slab, cur);
278 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
279 }
28bc106b 280 return err;
29b96b54
CY		 281}
282
4d57b86d 283void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
57864ae5
JK		 284{
285 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
286 struct inode *inode;
287 struct f2fs_inode_info *fi;
288next:
289 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
290 if (list_empty(head)) {
291 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
292 return;
293 }
294 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
295 inode = igrab(&fi->vfs_inode);
296 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
297
298 if (inode) {
2ef79ecb
CY		 299 if (gc_failure) {
300 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
301 goto drop;
302 goto skip;
303 }
304drop:
305 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
4d57b86d 306 f2fs_drop_inmem_pages(inode);
57864ae5
JK		 307 iput(inode);
308 }
2ef79ecb 309skip:
57864ae5
JK		 310 congestion_wait(BLK_RW_ASYNC, HZ/50);
311 cond_resched();
312 goto next;
313}
314
4d57b86d 315void f2fs_drop_inmem_pages(struct inode *inode)
29b96b54 316{
57864ae5 317 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 318 struct f2fs_inode_info *fi = F2FS_I(inode);
319
320 mutex_lock(&fi->inmem_lock);
28bc106b 321 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 322 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
323 if (!list_empty(&fi->inmem_ilist))
324 list_del_init(&fi->inmem_ilist);
325 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 326 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 327
328 clear_inode_flag(inode, FI_ATOMIC_FILE);
2ef79ecb 329 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
5fe45743 330 stat_dec_atomic_write(inode);
29b96b54
CY		 331}
332
4d57b86d 333void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
8c242db9
JK		 334{
335 struct f2fs_inode_info *fi = F2FS_I(inode);
336 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
337 struct list_head *head = &fi->inmem_pages;
338 struct inmem_pages *cur = NULL;
339
340 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
341
342 mutex_lock(&fi->inmem_lock);
343 list_for_each_entry(cur, head, list) {
344 if (cur->page == page)
345 break;
346 }
347
d0891e84 348 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
8c242db9
JK		 349 list_del(&cur->list);
350 mutex_unlock(&fi->inmem_lock);
351
352 dec_page_count(sbi, F2FS_INMEM_PAGES);
353 kmem_cache_free(inmem_entry_slab, cur);
354
355 ClearPageUptodate(page);
356 set_page_private(page, 0);
357 ClearPagePrivate(page);
358 f2fs_put_page(page, 0);
359
360 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
361}
362
4d57b86d 363static int __f2fs_commit_inmem_pages(struct inode *inode)
88b88a66
JK		 364{
365 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
366 struct f2fs_inode_info *fi = F2FS_I(inode);
367 struct inmem_pages *cur, *tmp;
88b88a66 368 struct f2fs_io_info fio = {
05ca3632 369 .sbi = sbi,
39d787be 370 .ino = inode->i_ino,
88b88a66 371 .type = DATA,
04d328de 372 .op = REQ_OP_WRITE,
70fd7614 373 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 374 .io_type = FS_DATA_IO,
88b88a66 375 };
cf52b27a 376 struct list_head revoke_list;
942fd319 377 pgoff_t last_idx = ULONG_MAX;
edb27dee 378 int err = 0;
88b88a66 379
cf52b27a
CY		 380 INIT_LIST_HEAD(&revoke_list);
381
88b88a66 382 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 383 struct page *page = cur->page;
384
385 lock_page(page);
386 if (page->mapping == inode->i_mapping) {
387 trace_f2fs_commit_inmem_page(page, INMEM);
388
389 set_page_dirty(page);
390 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 391 if (clear_page_dirty_for_io(page)) {
29b96b54 392 inode_dec_dirty_pages(inode);
4d57b86d 393 f2fs_remove_dirty_inode(inode);
933439c8 394 }
640cc189 395retry:
28bc106b 396 fio.page = page;
e959c8f5 397 fio.old_blkaddr = NULL_ADDR;
4d978078 398 fio.encrypted_page = NULL;
cc15620b 399 fio.need_lock = LOCK_DONE;
4d57b86d 400 err = f2fs_do_write_data_page(&fio);
29b96b54 401 if (err) {
640cc189
JK		 402 if (err == -ENOMEM) {
403 congestion_wait(BLK_RW_ASYNC, HZ/50);
404 cond_resched();
405 goto retry;
406 }
28bc106b 407 unlock_page(page);
29b96b54 408 break;
70c640b1 409 }
28bc106b
CY		 410 /* record old blkaddr for revoking */
411 cur->old_addr = fio.old_blkaddr;
942fd319 412 last_idx = page->index;
28bc106b
CY		 413 }
414 unlock_page(page);
cf52b27a 415 list_move_tail(&cur->list, &revoke_list);
88b88a66 416 }
29b96b54 417
942fd319 418 if (last_idx != ULONG_MAX)
b9109b0e 419 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b 420
cf52b27a
CY		 421 if (err) {
422 /*
423 * try to revoke all committed pages, but still we could fail
424 * due to no memory or other reason, if that happened, EAGAIN
425 * will be returned, which means in such case, transaction is
426 * already not integrity, caller should use journal to do the
427 * recovery or rewrite & commit last transaction. For other
428 * error number, revoking was done by filesystem itself.
429 */
430 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
431
432 /* drop all uncommitted pages */
433 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
434 } else {
435 __revoke_inmem_pages(inode, &revoke_list, false, false);
436 }
28bc106b 437
29b96b54
CY		 438 return err;
439}
440
4d57b86d 441int f2fs_commit_inmem_pages(struct inode *inode)
29b96b54
CY		 442{
443 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
444 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b 445 int err;
29b96b54
CY		 446
447 f2fs_balance_fs(sbi, true);
448 f2fs_lock_op(sbi);
449
5fe45743
CY		 450 set_inode_flag(inode, FI_ATOMIC_COMMIT);
451
29b96b54 452 mutex_lock(&fi->inmem_lock);
4d57b86d 453 err = __f2fs_commit_inmem_pages(inode);
28bc106b 454
57864ae5
JK		 455 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
456 if (!list_empty(&fi->inmem_ilist))
457 list_del_init(&fi->inmem_ilist);
458 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 459 mutex_unlock(&fi->inmem_lock);
460
5fe45743
CY		 461 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
462
29b96b54 463 f2fs_unlock_op(sbi);
edb27dee 464 return err;
88b88a66
JK		 465}
466
0a8165d7 467/*
351df4b2
JK		 468 * This function balances dirty node and dentry pages.
469 * In addition, it controls garbage collection.
470 */
2c4db1a6 471void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 472{
55523519
CY		 473 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
474 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 475 f2fs_stop_checkpoint(sbi, false);
55523519 476 }
0f348028 477
e589c2c4 478 /* balance_fs_bg is able to be pending */
a7881893 479 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 480 f2fs_balance_fs_bg(sbi);
481
351df4b2 482 /*
029cd28c
JK		 483 * We should do GC or end up with checkpoint, if there are so many dirty
484 * dir/node pages without enough free segments.
351df4b2 485 */
7f3037a5 486 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 487 mutex_lock(&sbi->gc_mutex);
e066b83c 488 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 489 }
490}
491
4660f9c0
JK		 492void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
493{
64c74a7a
CY		 494 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
495 return;
496
1dcc336b 497 /* try to shrink extent cache when there is no enough memory */
4d57b86d 498 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
554df79e 499 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 500
1b38dc8e 501 /* check the # of cached NAT entries */
4d57b86d
CY		 502 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
503 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
1b38dc8e 504
4d57b86d
CY		 505 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
506 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
ad4edb83 507 else
4d57b86d 508 f2fs_build_free_nids(sbi, false, false);
31696580 509
fd8c8caf
CY		 510 if (!is_idle(sbi) &&
511 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
f455c8a5 512 return;
31696580 513
1b38dc8e 514 /* checkpoint is the only way to shrink partial cached entries */
4d57b86d
CY		 515 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
516 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 517 excess_prefree_segs(sbi) ||
518 excess_dirty_nats(sbi) ||
fd8c8caf 519 excess_dirty_nodes(sbi) ||
f455c8a5 520 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 521 if (test_opt(sbi, DATA_FLUSH)) {
522 struct blk_plug plug;
523
524 blk_start_plug(&plug);
4d57b86d 525 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 526 blk_finish_plug(&plug);
527 }
4660f9c0 528 f2fs_sync_fs(sbi->sb, true);
42190d2a 529 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 530 }
4660f9c0
JK		 531}
532
20fda56b
KM		 533static int __submit_flush_wait(struct f2fs_sb_info *sbi,
534 struct block_device *bdev)
3c62be17 535{
d62fe971 536 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 537 int ret;
538
3adc5fcb 539 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 540 bio_set_dev(bio, bdev);
3c62be17
JK		 541 ret = submit_bio_wait(bio);
542 bio_put(bio);
20fda56b
KM		 543
544 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
545 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 546 return ret;
547}
548
39d787be 549static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 550{
39d787be 551 int ret = 0;
3c62be17
JK		 552 int i;
553
39d787be
CY		 554 if (!sbi->s_ndevs)
555 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 556
39d787be 557 for (i = 0; i < sbi->s_ndevs; i++) {
4d57b86d 558 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
39d787be 559 continue;
20fda56b
KM		 560 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
561 if (ret)
562 break;
3c62be17
JK		 563 }
564 return ret;
565}
566
2163d198 567static int issue_flush_thread(void *data)
6b4afdd7
JK		 568{
569 struct f2fs_sb_info *sbi = data;
b01a9201 570 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 571 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 572repeat:
573 if (kthread_should_stop())
574 return 0;
575
dc6febb6
CY		 576 sb_start_intwrite(sbi->sb);
577
721bd4d5 578 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 579 struct flush_cmd *cmd, *next;
580 int ret;
581
721bd4d5
GZ		 582 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
583 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
584
39d787be
CY		 585 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
586
587 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 588 atomic_inc(&fcc->issued_flush);
589
721bd4d5
GZ		 590 llist_for_each_entry_safe(cmd, next,
591 fcc->dispatch_list, llnode) {
6b4afdd7 592 cmd->ret = ret;
6b4afdd7
JK		 593 complete(&cmd->wait);
594 }
a688b9d9 595 fcc->dispatch_list = NULL;
6b4afdd7
JK		 596 }
597
dc6febb6
CY		 598 sb_end_intwrite(sbi->sb);
599
a688b9d9 600 wait_event_interruptible(*q,
721bd4d5 601 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 602 goto repeat;
603}
604
39d787be 605int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 606{
b01a9201 607 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 608 struct flush_cmd cmd;
8b8dd65f 609 int ret;
6b4afdd7 610
0f7b2abd
JK		 611 if (test_opt(sbi, NOBARRIER))
612 return 0;
613
8b8dd65f 614 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 615 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 616 atomic_inc(&fcc->issued_flush);
617 return ret;
618 }
740432f8 619
39d787be
CY		 620 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
621 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 622 atomic_dec(&fcc->issing_flush);
623
624 atomic_inc(&fcc->issued_flush);
740432f8
JK		 625 return ret;
626 }
6b4afdd7 627
39d787be 628 cmd.ino = ino;
adf8d90b 629 init_completion(&cmd.wait);
6b4afdd7 630
721bd4d5 631 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 632
6f890df0
CY		 633 /* update issue_list before we wake up issue_flush thread */
634 smp_mb();
635
636 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 637 wake_up(&fcc->flush_wait_queue);
6b4afdd7 638
5eba8c5d
JK		 639 if (fcc->f2fs_issue_flush) {
640 wait_for_completion(&cmd.wait);
8b8dd65f 641 atomic_dec(&fcc->issing_flush);
5eba8c5d 642 } else {
d3238691
CY		 643 struct llist_node *list;
644
645 list = llist_del_all(&fcc->issue_list);
646 if (!list) {
647 wait_for_completion(&cmd.wait);
648 atomic_dec(&fcc->issing_flush);
649 } else {
650 struct flush_cmd *tmp, *next;
651
39d787be 652 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 653
654 llist_for_each_entry_safe(tmp, next, list, llnode) {
655 if (tmp == &cmd) {
656 cmd.ret = ret;
657 atomic_dec(&fcc->issing_flush);
658 continue;
659 }
660 tmp->ret = ret;
661 complete(&tmp->wait);
662 }
663 }
5eba8c5d 664 }
adf8d90b
CY		 665
666 return cmd.ret;
6b4afdd7
JK		 667}
668
4d57b86d 669int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
2163d198
GZ		 670{
671 dev_t dev = sbi->sb->s_bdev->bd_dev;
672 struct flush_cmd_control *fcc;
673 int err = 0;
674
b01a9201
JK		 675 if (SM_I(sbi)->fcc_info) {
676 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 677 if (fcc->f2fs_issue_flush)
678 return err;
5eba8c5d
JK		 679 goto init_thread;
680 }
681
acbf054d 682 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 683 if (!fcc)
684 return -ENOMEM;
8b8dd65f
CY		 685 atomic_set(&fcc->issued_flush, 0);
686 atomic_set(&fcc->issing_flush, 0);
2163d198 687 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 688 init_llist_head(&fcc->issue_list);
b01a9201 689 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 690 if (!test_opt(sbi, FLUSH_MERGE))
691 return err;
692
5eba8c5d 693init_thread:
2163d198
GZ		 694 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
695 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
696 if (IS_ERR(fcc->f2fs_issue_flush)) {
697 err = PTR_ERR(fcc->f2fs_issue_flush);
698 kfree(fcc);
b01a9201 699 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 700 return err;
701 }
2163d198
GZ		 702
703 return err;
704}
705
4d57b86d 706void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 707{
b01a9201 708 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 709
5eba8c5d
JK		 710 if (fcc && fcc->f2fs_issue_flush) {
711 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
712
713 fcc->f2fs_issue_flush = NULL;
714 kthread_stop(flush_thread);
715 }
716 if (free) {
717 kfree(fcc);
b01a9201 718 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 719 }
2163d198
GZ		 720}
721
1228b482
CY		 722int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
723{
724 int ret = 0, i;
725
726 if (!sbi->s_ndevs)
727 return 0;
728
729 for (i = 1; i < sbi->s_ndevs; i++) {
730 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
731 continue;
732 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
733 if (ret)
734 break;
735
736 spin_lock(&sbi->dev_lock);
737 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
738 spin_unlock(&sbi->dev_lock);
739 }
740
741 return ret;
742}
743
351df4b2
JK		 744static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
745 enum dirty_type dirty_type)
746{
747 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
748
749 /* need not be added */
750 if (IS_CURSEG(sbi, segno))
751 return;
752
753 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
754 dirty_i->nr_dirty[dirty_type]++;
755
756 if (dirty_type == DIRTY) {
757 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 758 enum dirty_type t = sentry->type;
b2f2c390 759
ec325b52
JK		 760 if (unlikely(t >= DIRTY)) {
761 f2fs_bug_on(sbi, 1);
762 return;
763 }
4625d6aa
CL		 764 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
765 dirty_i->nr_dirty[t]++;
351df4b2
JK		 766 }
767}
768
769static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
770 enum dirty_type dirty_type)
771{
772 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
773
774 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
775 dirty_i->nr_dirty[dirty_type]--;
776
777 if (dirty_type == DIRTY) {
4625d6aa
CL		 778 struct seg_entry *sentry = get_seg_entry(sbi, segno);
779 enum dirty_type t = sentry->type;
780
781 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
782 dirty_i->nr_dirty[t]--;
b2f2c390 783
302bd348 784 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 785 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 786 dirty_i->victim_secmap);
351df4b2
JK		 787 }
788}
789
0a8165d7 790/*
351df4b2
JK		 791 * Should not occur error such as -ENOMEM.
792 * Adding dirty entry into seglist is not critical operation.
793 * If a given segment is one of current working segments, it won't be added.
794 */
8d8451af 795static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 796{
797 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
798 unsigned short valid_blocks;
799
800 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
801 return;
802
803 mutex_lock(&dirty_i->seglist_lock);
804
302bd348 805 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 806
807 if (valid_blocks == 0) {
808 __locate_dirty_segment(sbi, segno, PRE);
809 __remove_dirty_segment(sbi, segno, DIRTY);
810 } else if (valid_blocks < sbi->blocks_per_seg) {
811 __locate_dirty_segment(sbi, segno, DIRTY);
812 } else {
813 /* Recovery routine with SSR needs this */
814 __remove_dirty_segment(sbi, segno, DIRTY);
815 }
816
817 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 818}
819
004b6862 820static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 821 struct block_device *bdev, block_t lstart,
822 block_t start, block_t len)
275b66b0 823{
0b54fb84 824 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 825 struct list_head *pend_list;
b01a9201 826 struct discard_cmd *dc;
275b66b0 827
ba48a33e
CY		 828 f2fs_bug_on(sbi, !len);
829
830 pend_list = &dcc->pend_list[plist_idx(len)];
831
b01a9201
JK		 832 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
833 INIT_LIST_HEAD(&dc->list);
c81abe34 834 dc->bdev = bdev;
b01a9201 835 dc->lstart = lstart;
c81abe34 836 dc->start = start;
b01a9201 837 dc->len = len;
ec9895ad 838 dc->ref = 0;
15469963 839 dc->state = D_PREP;
35ec7d57 840 dc->issuing = 0;
c81abe34 841 dc->error = 0;
b01a9201 842 init_completion(&dc->wait);
22d375dd 843 list_add_tail(&dc->list, pend_list);
35ec7d57
CY		 844 spin_lock_init(&dc->lock);
845 dc->bio_ref = 0;
5f32366a 846 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 847 dcc->undiscard_blks += len;
004b6862
CY		 848
849 return dc;
850}
851
852static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
853 struct block_device *bdev, block_t lstart,
854 block_t start, block_t len,
855 struct rb_node *parent, struct rb_node **p)
856{
857 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
858 struct discard_cmd *dc;
859
860 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
861
862 rb_link_node(&dc->rb_node, parent, p);
863 rb_insert_color(&dc->rb_node, &dcc->root);
864
865 return dc;
15469963
JK		 866}
867
004b6862
CY		 868static void __detach_discard_cmd(struct discard_cmd_control *dcc,
869 struct discard_cmd *dc)
15469963 870{
dcc9165d 871 if (dc->state == D_DONE)
35ec7d57 872 atomic_sub(dc->issuing, &dcc->issing_discard);
004b6862
CY		 873
874 list_del(&dc->list);
875 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 876 dcc->undiscard_blks -= dc->len;
004b6862
CY		 877
878 kmem_cache_free(discard_cmd_slab, dc);
879
880 atomic_dec(&dcc->discard_cmd_cnt);
881}
882
883static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
884 struct discard_cmd *dc)
885{
886 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
35ec7d57 887 unsigned long flags;
dcc9165d 888
2ec6f2ef
CY		 889 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
890
35ec7d57
CY		 891 spin_lock_irqsave(&dc->lock, flags);
892 if (dc->bio_ref) {
893 spin_unlock_irqrestore(&dc->lock, flags);
894 return;
895 }
896 spin_unlock_irqrestore(&dc->lock, flags);
897
d9703d90
CY		 898 f2fs_bug_on(sbi, dc->ref);
899
c81abe34
JK		 900 if (dc->error == -EOPNOTSUPP)
901 dc->error = 0;
15469963 902
c81abe34 903 if (dc->error)
15469963 904 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 905 "Issue discard(%u, %u, %u) failed, ret: %d",
906 dc->lstart, dc->start, dc->len, dc->error);
004b6862 907 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 908}
909
c81abe34
JK		 910static void f2fs_submit_discard_endio(struct bio *bio)
911{
912 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
35ec7d57 913 unsigned long flags;
c81abe34 914
4e4cbee9 915 dc->error = blk_status_to_errno(bio->bi_status);
35ec7d57
CY		 916
917 spin_lock_irqsave(&dc->lock, flags);
918 dc->bio_ref--;
919 if (!dc->bio_ref && dc->state == D_SUBMIT) {
920 dc->state = D_DONE;
921 complete_all(&dc->wait);
922 }
923 spin_unlock_irqrestore(&dc->lock, flags);
c81abe34
JK		 924 bio_put(bio);
925}
926
94b1e10e 927static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 928 block_t start, block_t end)
929{
930#ifdef CONFIG_F2FS_CHECK_FS
931 struct seg_entry *sentry;
932 unsigned int segno;
933 block_t blk = start;
934 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
935 unsigned long *map;
936
937 while (blk < end) {
938 segno = GET_SEGNO(sbi, blk);
939 sentry = get_seg_entry(sbi, segno);
940 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
941
008396e1
YS		 942 if (end < START_BLOCK(sbi, segno + 1))
943 size = GET_BLKOFF_FROM_SEG0(sbi, end);
944 else
945 size = max_blocks;
6915ea9d
CY		 946 map = (unsigned long *)(sentry->cur_valid_map);
947 offset = __find_rev_next_bit(map, size, offset);
948 f2fs_bug_on(sbi, offset != size);
008396e1 949 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 950 }
951#endif
952}
953
8bb4f253
JK		 954static void __init_discard_policy(struct f2fs_sb_info *sbi,
955 struct discard_policy *dpolicy,
956 int discard_type, unsigned int granularity)
957{
958 /* common policy */
959 dpolicy->type = discard_type;
960 dpolicy->sync = true;
20ee4382 961 dpolicy->ordered = false;
8bb4f253
JK		 962 dpolicy->granularity = granularity;
963
964 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
965 dpolicy->io_aware_gran = MAX_PLIST_NUM;
966
967 if (discard_type == DPOLICY_BG) {
968 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 969 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 970 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
971 dpolicy->io_aware = true;
cba60849 972 dpolicy->sync = false;
20ee4382 973 dpolicy->ordered = true;
8bb4f253
JK		 974 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
975 dpolicy->granularity = 1;
976 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
977 }
978 } else if (discard_type == DPOLICY_FORCE) {
979 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 980 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 981 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
982 dpolicy->io_aware = false;
983 } else if (discard_type == DPOLICY_FSTRIM) {
984 dpolicy->io_aware = false;
985 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 986 dpolicy->max_requests = UINT_MAX;
8bb4f253
JK		 987 dpolicy->io_aware = false;
988 }
989}
990
35ec7d57
CY		 991static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
992 struct block_device *bdev, block_t lstart,
993 block_t start, block_t len);
c81abe34 994/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
6b9cb124 995static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 996 struct discard_policy *dpolicy,
35ec7d57
CY		 997 struct discard_cmd *dc,
998 unsigned int *issued)
c81abe34 999{
35ec7d57
CY		 1000 struct block_device *bdev = dc->bdev;
1001 struct request_queue *q = bdev_get_queue(bdev);
1002 unsigned int max_discard_blocks =
1003 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
c81abe34 1004 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1005 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1006 &(dcc->fstrim_list) : &(dcc->wait_list);
78997b56 1007 int flag = dpolicy->sync ? REQ_SYNC : 0;
35ec7d57
CY		 1008 block_t lstart, start, len, total_len;
1009 int err = 0;
c81abe34
JK		 1010
1011 if (dc->state != D_PREP)
6b9cb124 1012 return 0;
c81abe34 1013
d6184774 1014 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
6b9cb124 1015 return 0;
d6184774 1016
35ec7d57
CY		 1017 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1018
1019 lstart = dc->lstart;
1020 start = dc->start;
1021 len = dc->len;
1022 total_len = len;
1023
1024 dc->len = 0;
1025
1026 while (total_len && *issued < dpolicy->max_requests && !err) {
1027 struct bio *bio = NULL;
1028 unsigned long flags;
1029 bool last = true;
1030
1031 if (len > max_discard_blocks) {
1032 len = max_discard_blocks;
1033 last = false;
1034 }
1035
1036 (*issued)++;
1037 if (*issued == dpolicy->max_requests)
1038 last = true;
1039
1040 dc->len += len;
1041
b83dcfe6
CY		 1042 if (time_to_inject(sbi, FAULT_DISCARD)) {
1043 f2fs_show_injection_info(FAULT_DISCARD);
1044 err = -EIO;
1045 goto submit;
1046 }
35ec7d57
CY		 1047 err = __blkdev_issue_discard(bdev,
1048 SECTOR_FROM_BLOCK(start),
1049 SECTOR_FROM_BLOCK(len),
1050 GFP_NOFS, 0, &bio);
b83dcfe6 1051submit:
6b9cb124 1052 if (err) {
35ec7d57 1053 spin_lock_irqsave(&dc->lock, flags);
6b9cb124 1054 if (dc->state == D_PARTIAL)
35ec7d57 1055 dc->state = D_SUBMIT;
35ec7d57
CY		 1056 spin_unlock_irqrestore(&dc->lock, flags);
1057
6b9cb124
CY		 1058 break;
1059 }
35ec7d57 1060
6b9cb124 1061 f2fs_bug_on(sbi, !bio);
35ec7d57 1062
6b9cb124
CY		 1063 /*
1064 * should keep before submission to avoid D_DONE
1065 * right away
1066 */
1067 spin_lock_irqsave(&dc->lock, flags);
1068 if (last)
1069 dc->state = D_SUBMIT;
1070 else
1071 dc->state = D_PARTIAL;
1072 dc->bio_ref++;
1073 spin_unlock_irqrestore(&dc->lock, flags);
35ec7d57 1074
6b9cb124
CY		 1075 atomic_inc(&dcc->issing_discard);
1076 dc->issuing++;
1077 list_move_tail(&dc->list, wait_list);
b0af6d49 1078
6b9cb124
CY		 1079 /* sanity check on discard range */
1080 __check_sit_bitmap(sbi, start, start + len);
35ec7d57 1081
6b9cb124
CY		 1082 bio->bi_private = dc;
1083 bio->bi_end_io = f2fs_submit_discard_endio;
1084 bio->bi_opf |= flag;
1085 submit_bio(bio);
1086
1087 atomic_inc(&dcc->issued_discard);
1088
1089 f2fs_update_iostat(sbi, FS_DISCARD, 1);
35ec7d57
CY		 1090
1091 lstart += len;
1092 start += len;
1093 total_len -= len;
1094 len = total_len;
c81abe34 1095 }
35ec7d57 1096
6b9cb124 1097 if (!err && len)
35ec7d57 1098 __update_discard_tree_range(sbi, bdev, lstart, start, len);
6b9cb124 1099 return err;
c81abe34
JK		 1100}
1101
004b6862
CY		 1102static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1103 struct block_device *bdev, block_t lstart,
1104 block_t start, block_t len,
1105 struct rb_node **insert_p,
1106 struct rb_node *insert_parent)
c81abe34 1107{
004b6862 1108 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1109 struct rb_node **p;
004b6862
CY		 1110 struct rb_node *parent = NULL;
1111 struct discard_cmd *dc = NULL;
1112
1113 if (insert_p && insert_parent) {
1114 parent = insert_parent;
1115 p = insert_p;
1116 goto do_insert;
1117 }
c81abe34 1118
4d57b86d 1119 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
004b6862
CY		 1120do_insert:
1121 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1122 if (!dc)
1123 return NULL;
c81abe34 1124
004b6862 1125 return dc;
c81abe34
JK		 1126}
1127
ba48a33e
CY		 1128static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1129 struct discard_cmd *dc)
1130{
1131 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1132}
1133
3d6a650f
YH		 1134static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1135 struct discard_cmd *dc, block_t blkaddr)
1136{
ba48a33e 1137 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1138 struct discard_info di = dc->di;
1139 bool modified = false;
3d6a650f 1140
004b6862 1141 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1142 __remove_discard_cmd(sbi, dc);
1143 return;
1144 }
1145
d84d1cbd
CY		 1146 dcc->undiscard_blks -= di.len;
1147
004b6862 1148 if (blkaddr > di.lstart) {
3d6a650f 1149 dc->len = blkaddr - dc->lstart;
d84d1cbd 1150 dcc->undiscard_blks += dc->len;
ba48a33e 1151 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1152 modified = true;
1153 }
1154
1155 if (blkaddr < di.lstart + di.len - 1) {
1156 if (modified) {
1157 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1158 di.start + blkaddr + 1 - di.lstart,
1159 di.lstart + di.len - 1 - blkaddr,
1160 NULL, NULL);
1161 } else {
1162 dc->lstart++;
1163 dc->len--;
1164 dc->start++;
d84d1cbd 1165 dcc->undiscard_blks += dc->len;
ba48a33e 1166 __relocate_discard_cmd(dcc, dc);
004b6862 1167 }
3d6a650f
YH		 1168 }
1169}
1170
004b6862
CY		 1171static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1172 struct block_device *bdev, block_t lstart,
1173 block_t start, block_t len)
275b66b0 1174{
0b54fb84 1175 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1176 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1177 struct discard_cmd *dc;
1178 struct discard_info di = {0};
1179 struct rb_node **insert_p = NULL, *insert_parent = NULL;
35ec7d57
CY		 1180 struct request_queue *q = bdev_get_queue(bdev);
1181 unsigned int max_discard_blocks =
1182 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
004b6862 1183 block_t end = lstart + len;
275b66b0 1184
4d57b86d 1185 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
004b6862
CY		 1186 NULL, lstart,
1187 (struct rb_entry **)&prev_dc,
1188 (struct rb_entry **)&next_dc,
1189 &insert_p, &insert_parent, true);
1190 if (dc)
1191 prev_dc = dc;
1192
1193 if (!prev_dc) {
1194 di.lstart = lstart;
1195 di.len = next_dc ? next_dc->lstart - lstart : len;
1196 di.len = min(di.len, len);
1197 di.start = start;
22d375dd 1198 }
15469963 1199
004b6862
CY		 1200 while (1) {
1201 struct rb_node *node;
1202 bool merged = false;
1203 struct discard_cmd *tdc = NULL;
1204
1205 if (prev_dc) {
1206 di.lstart = prev_dc->lstart + prev_dc->len;
1207 if (di.lstart < lstart)
1208 di.lstart = lstart;
1209 if (di.lstart >= end)
1210 break;
1211
1212 if (!next_dc || next_dc->lstart > end)
1213 di.len = end - di.lstart;
1214 else
1215 di.len = next_dc->lstart - di.lstart;
1216 di.start = start + di.lstart - lstart;
1217 }
1218
1219 if (!di.len)
1220 goto next;
1221
1222 if (prev_dc && prev_dc->state == D_PREP &&
1223 prev_dc->bdev == bdev &&
35ec7d57
CY		 1224 __is_discard_back_mergeable(&di, &prev_dc->di,
1225 max_discard_blocks)) {
004b6862 1226 prev_dc->di.len += di.len;
d84d1cbd 1227 dcc->undiscard_blks += di.len;
ba48a33e 1228 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1229 di = prev_dc->di;
1230 tdc = prev_dc;
1231 merged = true;
1232 }
1233
1234 if (next_dc && next_dc->state == D_PREP &&
1235 next_dc->bdev == bdev &&
35ec7d57
CY		 1236 __is_discard_front_mergeable(&di, &next_dc->di,
1237 max_discard_blocks)) {
004b6862
CY		 1238 next_dc->di.lstart = di.lstart;
1239 next_dc->di.len += di.len;
1240 next_dc->di.start = di.start;
d84d1cbd 1241 dcc->undiscard_blks += di.len;
ba48a33e 1242 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1243 if (tdc)
1244 __remove_discard_cmd(sbi, tdc);
004b6862 1245 merged = true;
4e6a8d9b 1246 }
004b6862 1247
df0f6b44 1248 if (!merged) {
004b6862
CY		 1249 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1250 di.len, NULL, NULL);
df0f6b44 1251 }
004b6862
CY		 1252 next:
1253 prev_dc = next_dc;
1254 if (!prev_dc)
1255 break;
1256
1257 node = rb_next(&prev_dc->rb_node);
1258 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1259 }
004b6862
CY		 1260}
1261
1262static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1263 struct block_device *bdev, block_t blkstart, block_t blklen)
1264{
1265 block_t lblkstart = blkstart;
1266
0243a5f9 1267 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1268
1269 if (sbi->s_ndevs) {
1270 int devi = f2fs_target_device_index(sbi, blkstart);
1271
1272 blkstart -= FDEV(devi).start_blk;
1273 }
35ec7d57 1274 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862 1275 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
35ec7d57 1276 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862
CY		 1277 return 0;
1278}
1279
20ee4382
CY		 1280static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1281 struct discard_policy *dpolicy)
1282{
1283 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1284 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1285 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1286 struct discard_cmd *dc;
1287 struct blk_plug plug;
1288 unsigned int pos = dcc->next_pos;
1289 unsigned int issued = 0;
1290 bool io_interrupted = false;
1291
1292 mutex_lock(&dcc->cmd_lock);
1293 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1294 NULL, pos,
1295 (struct rb_entry **)&prev_dc,
1296 (struct rb_entry **)&next_dc,
1297 &insert_p, &insert_parent, true);
1298 if (!dc)
1299 dc = next_dc;
1300
1301 blk_start_plug(&plug);
1302
1303 while (dc) {
1304 struct rb_node *node;
6b9cb124 1305 int err = 0;
20ee4382
CY		 1306
1307 if (dc->state != D_PREP)
1308 goto next;
1309
1310 if (dpolicy->io_aware && !is_idle(sbi)) {
1311 io_interrupted = true;
1312 break;
1313 }
1314
1315 dcc->next_pos = dc->lstart + dc->len;
6b9cb124 1316 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
20ee4382 1317
35ec7d57 1318 if (issued >= dpolicy->max_requests)
20ee4382
CY		 1319 break;
1320next:
1321 node = rb_next(&dc->rb_node);
6b9cb124
CY		 1322 if (err)
1323 __remove_discard_cmd(sbi, dc);
20ee4382
CY		 1324 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1325 }
1326
1327 blk_finish_plug(&plug);
1328
1329 if (!dc)
1330 dcc->next_pos = 0;
1331
1332 mutex_unlock(&dcc->cmd_lock);
1333
1334 if (!issued && io_interrupted)
1335 issued = -1;
1336
1337 return issued;
1338}
1339
78997b56
CY		 1340static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1341 struct discard_policy *dpolicy)
bd5b0738
CY		 1342{
1343 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1344 struct list_head *pend_list;
1345 struct discard_cmd *dc, *tmp;
1346 struct blk_plug plug;
522d1711 1347 int i, issued = 0;
e6c6de18 1348 bool io_interrupted = false;
bd5b0738 1349
78997b56
CY		 1350 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1351 if (i + 1 < dpolicy->granularity)
1352 break;
20ee4382
CY		 1353
1354 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1355 return __issue_discard_cmd_orderly(sbi, dpolicy);
1356
bd5b0738 1357 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1358
1359 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1360 if (list_empty(pend_list))
1361 goto next;
67fce70b
CY		 1362 if (unlikely(dcc->rbtree_check))
1363 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1364 &dcc->root));
33da62cf 1365 blk_start_plug(&plug);
bd5b0738
CY		 1366 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1367 f2fs_bug_on(sbi, dc->state != D_PREP);
1368
ecc9aa00
CY		 1369 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1370 !is_idle(sbi)) {
e6c6de18 1371 io_interrupted = true;
522d1711 1372 break;
969d1b18 1373 }
e6c6de18 1374
35ec7d57 1375 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
522d1711 1376
35ec7d57 1377 if (issued >= dpolicy->max_requests)
33da62cf 1378 break;
bd5b0738 1379 }
33da62cf 1380 blk_finish_plug(&plug);
49c60c67 1381next:
33da62cf
CY		 1382 mutex_unlock(&dcc->cmd_lock);
1383
522d1711 1384 if (issued >= dpolicy->max_requests || io_interrupted)
33da62cf 1385 break;
bd5b0738 1386 }
969d1b18 1387
e6c6de18
CY		 1388 if (!issued && io_interrupted)
1389 issued = -1;
1390
969d1b18
CY		 1391 return issued;
1392}
1393
cf5c759f 1394static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1395{
1396 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1397 struct list_head *pend_list;
1398 struct discard_cmd *dc, *tmp;
1399 int i;
cf5c759f 1400 bool dropped = false;
969d1b18
CY		 1401
1402 mutex_lock(&dcc->cmd_lock);
1403 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1404 pend_list = &dcc->pend_list[i];
1405 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1406 f2fs_bug_on(sbi, dc->state != D_PREP);
1407 __remove_discard_cmd(sbi, dc);
cf5c759f 1408 dropped = true;
969d1b18
CY		 1409 }
1410 }
1411 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1412
1413 return dropped;
bd5b0738
CY		 1414}
1415
4d57b86d 1416void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
7950e9ac
CY		 1417{
1418 __drop_discard_cmd(sbi);
1419}
1420
0ea80512 1421static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1422 struct discard_cmd *dc)
1423{
1424 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1425 unsigned int len = 0;
2a510c00
CY		 1426
1427 wait_for_completion_io(&dc->wait);
1428 mutex_lock(&dcc->cmd_lock);
1429 f2fs_bug_on(sbi, dc->state != D_DONE);
1430 dc->ref--;
0ea80512
CY		 1431 if (!dc->ref) {
1432 if (!dc->error)
1433 len = dc->len;
2a510c00 1434 __remove_discard_cmd(sbi, dc);
0ea80512 1435 }
2a510c00 1436 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1437
1438 return len;
2a510c00
CY		 1439}
1440
0ea80512 1441static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1442 struct discard_policy *dpolicy,
1443 block_t start, block_t end)
63a94fa1
CY		 1444{
1445 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1446 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1447 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1448 struct discard_cmd *dc, *tmp;
6afae633 1449 bool need_wait;
0ea80512 1450 unsigned int trimmed = 0;
6afae633
CY		 1451
1452next:
1453 need_wait = false;
63a94fa1
CY		 1454
1455 mutex_lock(&dcc->cmd_lock);
1456 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1457 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1458 continue;
78997b56 1459 if (dc->len < dpolicy->granularity)
8412663d 1460 continue;
78997b56 1461 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1462 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1463 if (!dc->error)
1464 trimmed += dc->len;
63a94fa1 1465 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1466 } else {
1467 dc->ref++;
1468 need_wait = true;
1469 break;
63a94fa1
CY		 1470 }
1471 }
1472 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1473
1474 if (need_wait) {
0ea80512 1475 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1476 goto next;
1477 }
0ea80512
CY		 1478
1479 return trimmed;
63a94fa1
CY		 1480}
1481
01f9cf6d 1482static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1483 struct discard_policy *dpolicy)
8412663d 1484{
9a997188 1485 struct discard_policy dp;
01f9cf6d 1486 unsigned int discard_blks;
9a997188 1487
01f9cf6d
CY		 1488 if (dpolicy)
1489 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
9a997188
JK		 1490
1491 /* wait all */
8bb4f253 1492 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
01f9cf6d 1493 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1494 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
01f9cf6d
CY		 1495 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1496
1497 return discard_blks;
8412663d
CY		 1498}
1499
004b6862 1500/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1501static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1502{
1503 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1504 struct discard_cmd *dc;
ec9895ad 1505 bool need_wait = false;
004b6862
CY		 1506
1507 mutex_lock(&dcc->cmd_lock);
4d57b86d
CY		 1508 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1509 NULL, blkaddr);
004b6862 1510 if (dc) {
ec9895ad
CY		 1511 if (dc->state == D_PREP) {
1512 __punch_discard_cmd(sbi, dc, blkaddr);
1513 } else {
1514 dc->ref++;
1515 need_wait = true;
1516 }
275b66b0 1517 }
d431413f 1518 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1519
2a510c00
CY		 1520 if (need_wait)
1521 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1522}
1523
4d57b86d 1524void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
cce13252
CY		 1525{
1526 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1527
1528 if (dcc && dcc->f2fs_issue_discard) {
1529 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1530
1531 dcc->f2fs_issue_discard = NULL;
1532 kthread_stop(discard_thread);
ec9895ad 1533 }
d431413f
CY		 1534}
1535
8412663d 1536/* This comes from f2fs_put_super */
cf5c759f 1537bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1538{
1539 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1540 struct discard_policy dpolicy;
cf5c759f 1541 bool dropped;
969d1b18 1542
8bb4f253
JK		 1543 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1544 dcc->discard_granularity);
78997b56 1545 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1546 dropped = __drop_discard_cmd(sbi);
cf5c759f 1547
9a997188
JK		 1548 /* just to make sure there is no pending discard commands */
1549 __wait_all_discard_cmd(sbi, NULL);
2482c432
CY		 1550
1551 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
cf5c759f 1552 return dropped;
969d1b18
CY		 1553}
1554
15469963
JK		 1555static int issue_discard_thread(void *data)
1556{
1557 struct f2fs_sb_info *sbi = data;
1558 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1559 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1560 struct discard_policy dpolicy;
969d1b18
CY		 1561 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1562 int issued;
15469963 1563
1d7be270 1564 set_freezable();
15469963 1565
1d7be270 1566 do {
8bb4f253 1567 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1568 dcc->discard_granularity);
1569
969d1b18
CY		 1570 wait_event_interruptible_timeout(*q,
1571 kthread_should_stop() || freezing(current) ||
1572 dcc->discard_wake,
1573 msecs_to_jiffies(wait_ms));
35a9a766
SY		 1574
1575 if (dcc->discard_wake)
1576 dcc->discard_wake = 0;
1577
1d7be270
JK		 1578 if (try_to_freeze())
1579 continue;
3b60d802
CY		 1580 if (f2fs_readonly(sbi->sb))
1581 continue;
1d7be270
JK		 1582 if (kthread_should_stop())
1583 return 0;
d6184774
YH		 1584 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1585 wait_ms = dpolicy.max_interval;
1586 continue;
1587 }
15469963 1588
5b0e9539 1589 if (sbi->gc_mode == GC_URGENT)
8bb4f253 1590 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1591
dc6febb6
CY		 1592 sb_start_intwrite(sbi->sb);
1593
78997b56 1594 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1595 if (issued > 0) {
78997b56
CY		 1596 __wait_all_discard_cmd(sbi, &dpolicy);
1597 wait_ms = dpolicy.min_interval;
f9d1dced
YH		 1598 } else if (issued == -1){
1599 wait_ms = dpolicy.mid_interval;
969d1b18 1600 } else {
78997b56 1601 wait_ms = dpolicy.max_interval;
969d1b18 1602 }
1d7be270 1603
dc6febb6 1604 sb_end_intwrite(sbi->sb);
1d7be270 1605
1d7be270
JK		 1606 } while (!kthread_should_stop());
1607 return 0;
15469963
JK		 1608}
1609
f46e8809 1610#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1611static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1612 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1613{
92592285 1614 sector_t sector, nr_sects;
10a875f8 1615 block_t lblkstart = blkstart;
3c62be17
JK		 1616 int devi = 0;
1617
1618 if (sbi->s_ndevs) {
1619 devi = f2fs_target_device_index(sbi, blkstart);
1620 blkstart -= FDEV(devi).start_blk;
1621 }
f46e8809
DLM		 1622
1623 /*
1624 * We need to know the type of the zone: for conventional zones,
1625 * use regular discard if the drive supports it. For sequential
1626 * zones, reset the zone write pointer.
1627 */
3c62be17 1628 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1629
1630 case BLK_ZONE_TYPE_CONVENTIONAL:
1631 if (!blk_queue_discard(bdev_get_queue(bdev)))
1632 return 0;
c81abe34 1633 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1634 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1635 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1636 sector = SECTOR_FROM_BLOCK(blkstart);
1637 nr_sects = SECTOR_FROM_BLOCK(blklen);
1638
1639 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1640 nr_sects != bdev_zone_sectors(bdev)) {
1641 f2fs_msg(sbi->sb, KERN_INFO,
1642 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1643 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1644 blkstart, blklen);
1645 return -EIO;
1646 }
d50aaeec 1647 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1648 return blkdev_reset_zones(bdev, sector,
1649 nr_sects, GFP_NOFS);
1650 default:
1651 /* Unknown zone type: broken device ? */
1652 return -EIO;
1653 }
1654}
1655#endif
1656
3c62be17
JK		 1657static int __issue_discard_async(struct f2fs_sb_info *sbi,
1658 struct block_device *bdev, block_t blkstart, block_t blklen)
1659{
1660#ifdef CONFIG_BLK_DEV_ZONED
ccd31cb2 1661 if (f2fs_sb_has_blkzoned(sbi->sb) &&
3c62be17
JK		 1662 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1663 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1664#endif
c81abe34 1665 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1666}
1667
1e87a78d 1668static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1669 block_t blkstart, block_t blklen)
1670{
3c62be17
JK		 1671 sector_t start = blkstart, len = 0;
1672 struct block_device *bdev;
a66cdd98
JK		 1673 struct seg_entry *se;
1674 unsigned int offset;
1675 block_t i;
3c62be17
JK		 1676 int err = 0;
1677
1678 bdev = f2fs_target_device(sbi, blkstart, NULL);
1679
1680 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1681 if (i != start) {
1682 struct block_device *bdev2 =
1683 f2fs_target_device(sbi, i, NULL);
1684
1685 if (bdev2 != bdev) {
1686 err = __issue_discard_async(sbi, bdev,
1687 start, len);
1688 if (err)
1689 return err;
1690 bdev = bdev2;
1691 start = i;
1692 len = 0;
1693 }
1694 }
a66cdd98 1695
a66cdd98
JK		 1696 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1697 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1698
1699 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1700 sbi->discard_blks--;
1701 }
f46e8809 1702
3c62be17
JK		 1703 if (len)
1704 err = __issue_discard_async(sbi, bdev, start, len);
1705 return err;
1e87a78d
JK		 1706}
1707
25290fa5
JK		 1708static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1709 bool check_only)
adf4983b 1710{
b2955550
JK		 1711 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1712 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1713 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1714 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1715 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1716 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1717 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1718 unsigned int start = 0, end = -1;
c473f1a9 1719 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1720 struct discard_entry *de = NULL;
46f84c2c 1721 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1722 int i;
1723
3e025740 1724 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1725 return false;
b2955550 1726
a66cdd98
JK		 1727 if (!force) {
1728 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1729 SM_I(sbi)->dcc_info->nr_discards >=
1730 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1731 return false;
4b2fecc8
JK		 1732 }
1733
b2955550
JK		 1734 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1735 for (i = 0; i < entries; i++)
a66cdd98 1736 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1737 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1738
0b54fb84
JK		 1739 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1740 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1741 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1742 if (start >= max_blocks)
1743 break;
1744
1745 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1746 if (force && start && end != max_blocks
1747 && (end - start) < cpc->trim_minlen)
1748 continue;
1749
25290fa5
JK		 1750 if (check_only)
1751 return true;
1752
a7eeb823
CY		 1753 if (!de) {
1754 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1755 GFP_F2FS_ZERO);
1756 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1757 list_add_tail(&de->list, head);
1758 }
1759
1760 for (i = start; i < end; i++)
1761 __set_bit_le(i, (void *)de->discard_map);
1762
1763 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1764 }
25290fa5 1765 return false;
b2955550
JK		 1766}
1767
af8ff65b
CY		 1768static void release_discard_addr(struct discard_entry *entry)
1769{
1770 list_del(&entry->list);
1771 kmem_cache_free(discard_entry_slab, entry);
1772}
1773
4d57b86d 1774void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1775{
46f84c2c 1776 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1777 struct discard_entry *entry, *this;
1778
1779 /* drop caches */
af8ff65b
CY		 1780 list_for_each_entry_safe(entry, this, head, list)
1781 release_discard_addr(entry);
4b2fecc8
JK		 1782}
1783
0a8165d7 1784/*
4d57b86d 1785 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1786 */
1787static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1788{
1789 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1790 unsigned int segno;
351df4b2
JK		 1791
1792 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1793 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1794 __set_test_and_free(sbi, segno);
351df4b2
JK		 1795 mutex_unlock(&dirty_i->seglist_lock);
1796}
1797
4d57b86d
CY		 1798void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1799 struct cp_control *cpc)
351df4b2 1800{
969d1b18
CY		 1801 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1802 struct list_head *head = &dcc->entry_list;
2d7b822a 1803 struct discard_entry *entry, *this;
351df4b2 1804 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1805 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1806 unsigned int start = 0, end = -1;
36abef4e 1807 unsigned int secno, start_segno;
c473f1a9 1808 bool force = (cpc->reason & CP_DISCARD);
ad6672bb 1809 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
351df4b2
JK		 1810
1811 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1812
351df4b2 1813 while (1) {
29e59c14 1814 int i;
ad6672bb
YS		 1815
1816 if (need_align && end != -1)
1817 end--;
7cd8558b
JK		 1818 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1819 if (start >= MAIN_SEGS(sbi))
351df4b2 1820 break;
7cd8558b
JK		 1821 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1822 start + 1);
29e59c14 1823
ad6672bb
YS		 1824 if (need_align) {
1825 start = rounddown(start, sbi->segs_per_sec);
1826 end = roundup(end, sbi->segs_per_sec);
1827 }
29e59c14 1828
ad6672bb
YS		 1829 for (i = start; i < end; i++) {
1830 if (test_and_clear_bit(i, prefree_map))
1831 dirty_i->nr_dirty[PRE]--;
1832 }
29e59c14 1833
650d3c4e 1834 if (!test_opt(sbi, DISCARD))
29e59c14 1835 continue;
351df4b2 1836
650d3c4e
YH		 1837 if (force && start >= cpc->trim_start &&
1838 (end - 1) <= cpc->trim_end)
1839 continue;
1840
36abef4e
JK		 1841 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1842 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1843 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1844 continue;
1845 }
1846next:
4ddb1a4d
JK		 1847 secno = GET_SEC_FROM_SEG(sbi, start);
1848 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1849 if (!IS_CURSEC(sbi, secno) &&
302bd348 1850 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1851 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1852 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1853
1854 start = start_segno + sbi->segs_per_sec;
1855 if (start < end)
1856 goto next;
8b107f5b
JK		 1857 else
1858 end = start - 1;
351df4b2
JK		 1859 }
1860 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1861
1862 /* send small discards */
2d7b822a 1863 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1864 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1865 bool is_valid = test_bit_le(0, entry->discard_map);
1866
1867find_next:
1868 if (is_valid) {
1869 next_pos = find_next_zero_bit_le(entry->discard_map,
1870 sbi->blocks_per_seg, cur_pos);
1871 len = next_pos - cur_pos;
1872
ccd31cb2 1873 if (f2fs_sb_has_blkzoned(sbi->sb) ||
acfd2810 1874 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1875 goto skip;
1876
1877 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1878 len);
a7eeb823
CY		 1879 total_len += len;
1880 } else {
1881 next_pos = find_next_bit_le(entry->discard_map,
1882 sbi->blocks_per_seg, cur_pos);
1883 }
836b5a63 1884skip:
a7eeb823
CY		 1885 cur_pos = next_pos;
1886 is_valid = !is_valid;
1887
1888 if (cur_pos < sbi->blocks_per_seg)
1889 goto find_next;
1890
af8ff65b 1891 release_discard_addr(entry);
969d1b18 1892 dcc->nr_discards -= total_len;
b2955550 1893 }
34e159da 1894
01983c71 1895 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1896}
1897
8ed59745 1898static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1899{
15469963 1900 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1901 struct discard_cmd_control *dcc;
ba48a33e 1902 int err = 0, i;
0b54fb84
JK		 1903
1904 if (SM_I(sbi)->dcc_info) {
1905 dcc = SM_I(sbi)->dcc_info;
1906 goto init_thread;
1907 }
1908
acbf054d 1909 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 1910 if (!dcc)
1911 return -ENOMEM;
1912
969d1b18 1913 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1914 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1915 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1916 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 1917 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1918 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1919 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1920 atomic_set(&dcc->issued_discard, 0);
1921 atomic_set(&dcc->issing_discard, 0);
5f32366a 1922 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1923 dcc->nr_discards = 0;
d618ebaf 1924 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1925 dcc->undiscard_blks = 0;
20ee4382 1926 dcc->next_pos = 0;
004b6862 1927 dcc->root = RB_ROOT;
67fce70b 1928 dcc->rbtree_check = false;
0b54fb84 1929
15469963 1930 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1931 SM_I(sbi)->dcc_info = dcc;
1932init_thread:
15469963
JK		 1933 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1934 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1935 if (IS_ERR(dcc->f2fs_issue_discard)) {
1936 err = PTR_ERR(dcc->f2fs_issue_discard);
1937 kfree(dcc);
1938 SM_I(sbi)->dcc_info = NULL;
1939 return err;
1940 }
1941
0b54fb84
JK		 1942 return err;
1943}
1944
f099405f 1945static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1946{
1947 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1948
f099405f
CY		 1949 if (!dcc)
1950 return;
1951
4d57b86d 1952 f2fs_stop_discard_thread(sbi);
f099405f
CY		 1953
1954 kfree(dcc);
1955 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1956}
1957
184a5cd2 1958static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1959{
1960 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1961
1962 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1963 sit_i->dirty_sentries++;
184a5cd2
CY		 1964 return false;
1965 }
1966
1967 return true;
351df4b2
JK		 1968}
1969
1970static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1971 unsigned int segno, int modified)
1972{
1973 struct seg_entry *se = get_seg_entry(sbi, segno);
1974 se->type = type;
1975 if (modified)
1976 __mark_sit_entry_dirty(sbi, segno);
1977}
1978
1979static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1980{
1981 struct seg_entry *se;
1982 unsigned int segno, offset;
1983 long int new_vblocks;
6415fedc
YS		 1984 bool exist;
1985#ifdef CONFIG_F2FS_CHECK_FS
1986 bool mir_exist;
1987#endif
351df4b2
JK		 1988
1989 segno = GET_SEGNO(sbi, blkaddr);
1990
1991 se = get_seg_entry(sbi, segno);
1992 new_vblocks = se->valid_blocks + del;
491c0854 1993 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1994
9850cf4a 1995 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1996 (new_vblocks > sbi->blocks_per_seg)));
1997
1998 se->valid_blocks = new_vblocks;
a1f72ac2
CY		 1999 se->mtime = get_mtime(sbi, false);
2000 if (se->mtime > SIT_I(sbi)->max_mtime)
2001 SIT_I(sbi)->max_mtime = se->mtime;
351df4b2
JK		 2002
2003 /* Update valid block bitmap */
2004 if (del > 0) {
6415fedc 2005 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 2006#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2007 mir_exist = f2fs_test_and_set_bit(offset,
2008 se->cur_valid_map_mir);
2009 if (unlikely(exist != mir_exist)) {
2010 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2011 "when setting bitmap, blk:%u, old bit:%d",
2012 blkaddr, exist);
05796763 2013 f2fs_bug_on(sbi, 1);
6415fedc 2014 }
355e7891 2015#endif
6415fedc
YS		 2016 if (unlikely(exist)) {
2017 f2fs_msg(sbi->sb, KERN_ERR,
2018 "Bitmap was wrongly set, blk:%u", blkaddr);
2019 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2020 se->valid_blocks--;
2021 del = 0;
355e7891 2022 }
6415fedc 2023
3e025740
JK		 2024 if (f2fs_discard_en(sbi) &&
2025 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 2026 sbi->discard_blks--;
720037f9
JK		 2027
2028 /* don't overwrite by SSR to keep node chain */
5d7881ca 2029 if (IS_NODESEG(se->type)) {
720037f9
JK		 2030 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2031 se->ckpt_valid_blocks++;
2032 }
351df4b2 2033 } else {
6415fedc 2034 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 2035#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2036 mir_exist = f2fs_test_and_clear_bit(offset,
2037 se->cur_valid_map_mir);
2038 if (unlikely(exist != mir_exist)) {
2039 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2040 "when clearing bitmap, blk:%u, old bit:%d",
2041 blkaddr, exist);
05796763 2042 f2fs_bug_on(sbi, 1);
6415fedc 2043 }
355e7891 2044#endif
6415fedc
YS		 2045 if (unlikely(!exist)) {
2046 f2fs_msg(sbi->sb, KERN_ERR,
2047 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2048 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2049 se->valid_blocks++;
2050 del = 0;
355e7891 2051 }
6415fedc 2052
3e025740
JK		 2053 if (f2fs_discard_en(sbi) &&
2054 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 2055 sbi->discard_blks++;
351df4b2
JK		 2056 }
2057 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2058 se->ckpt_valid_blocks += del;
2059
2060 __mark_sit_entry_dirty(sbi, segno);
2061
2062 /* update total number of valid blocks to be written in ckpt area */
2063 SIT_I(sbi)->written_valid_blocks += del;
2064
2065 if (sbi->segs_per_sec > 1)
2066 get_sec_entry(sbi, segno)->valid_blocks += del;
2067}
2068
4d57b86d 2069void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 2070{
2071 unsigned int segno = GET_SEGNO(sbi, addr);
2072 struct sit_info *sit_i = SIT_I(sbi);
2073
9850cf4a 2074 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 2075 if (addr == NEW_ADDR)
2076 return;
2077
2078 /* add it into sit main buffer */
3d26fa6b 2079 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2080
2081 update_sit_entry(sbi, addr, -1);
2082
2083 /* add it into dirty seglist */
2084 locate_dirty_segment(sbi, segno);
2085
3d26fa6b 2086 up_write(&sit_i->sentry_lock);
351df4b2
JK		 2087}
2088
4d57b86d 2089bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 2090{
2091 struct sit_info *sit_i = SIT_I(sbi);
2092 unsigned int segno, offset;
2093 struct seg_entry *se;
2094 bool is_cp = false;
2095
e1da7872 2096 if (!is_valid_data_blkaddr(sbi, blkaddr))
6e2c64ad
JK		 2097 return true;
2098
3d26fa6b 2099 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2100
2101 segno = GET_SEGNO(sbi, blkaddr);
2102 se = get_seg_entry(sbi, segno);
2103 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2104
2105 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2106 is_cp = true;
2107
3d26fa6b 2108 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2109
2110 return is_cp;
2111}
2112
0a8165d7 2113/*
351df4b2
JK		 2114 * This function should be resided under the curseg_mutex lock
2115 */
2116static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 2117 struct f2fs_summary *sum)
351df4b2
JK		 2118{
2119 struct curseg_info *curseg = CURSEG_I(sbi, type);
2120 void *addr = curseg->sum_blk;
e79efe3b 2121 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 2122 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 2123}
2124
0a8165d7 2125/*
351df4b2
JK		 2126 * Calculate the number of current summary pages for writing
2127 */
4d57b86d 2128int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 2129{
351df4b2 2130 int valid_sum_count = 0;
9a47938b 2131 int i, sum_in_page;
351df4b2
JK		 2132
2133 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2134 if (sbi->ckpt->alloc_type[i] == SSR)
2135 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 2136 else {
2137 if (for_ra)
2138 valid_sum_count += le16_to_cpu(
2139 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2140 else
2141 valid_sum_count += curseg_blkoff(sbi, i);
2142 }
351df4b2
JK		 2143 }
2144
09cbfeaf 2145 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 2146 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2147 if (valid_sum_count <= sum_in_page)
351df4b2 2148 return 1;
9a47938b 2149 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 2150 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 2151 return 2;
2152 return 3;
2153}
2154
0a8165d7 2155/*
351df4b2
JK		 2156 * Caller should put this summary page
2157 */
4d57b86d 2158struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 2159{
7735730d 2160 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2161}
2162
4d57b86d
CY		 2163void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2164 void *src, block_t blk_addr)
351df4b2 2165{
4d57b86d 2166 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 2167
0537b811 2168 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2169 set_page_dirty(page);
2170 f2fs_put_page(page, 1);
2171}
2172
381722d2
CY		 2173static void write_sum_page(struct f2fs_sb_info *sbi,
2174 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2175{
4d57b86d 2176 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2177}
2178
b7ad7512
CY		 2179static void write_current_sum_page(struct f2fs_sb_info *sbi,
2180 int type, block_t blk_addr)
2181{
2182 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2183 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2184 struct f2fs_summary_block *src = curseg->sum_blk;
2185 struct f2fs_summary_block *dst;
2186
2187 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2188 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2189
2190 mutex_lock(&curseg->curseg_mutex);
2191
2192 down_read(&curseg->journal_rwsem);
2193 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2194 up_read(&curseg->journal_rwsem);
2195
2196 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2197 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2198
2199 mutex_unlock(&curseg->curseg_mutex);
2200
2201 set_page_dirty(page);
2202 f2fs_put_page(page, 1);
2203}
2204
a7881893
JK		 2205static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2206{
2207 struct curseg_info *curseg = CURSEG_I(sbi, type);
2208 unsigned int segno = curseg->segno + 1;
2209 struct free_segmap_info *free_i = FREE_I(sbi);
2210
2211 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2212 return !test_bit(segno, free_i->free_segmap);
2213 return 0;
2214}
2215
0a8165d7 2216/*
351df4b2
JK		 2217 * Find a new segment from the free segments bitmap to right order
2218 * This function should be returned with success, otherwise BUG
2219 */
2220static void get_new_segment(struct f2fs_sb_info *sbi,
2221 unsigned int *newseg, bool new_sec, int dir)
2222{
2223 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2224 unsigned int segno, secno, zoneno;
7cd8558b 2225 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2226 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2227 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2228 unsigned int left_start = hint;
2229 bool init = true;
2230 int go_left = 0;
2231 int i;
2232
1a118ccf 2233 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2234
2235 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2236 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2237 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2238 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2239 goto got_it;
2240 }
2241find_other_zone:
7cd8558b
JK		 2242 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2243 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2244 if (dir == ALLOC_RIGHT) {
2245 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2246 MAIN_SECS(sbi), 0);
2247 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2248 } else {
2249 go_left = 1;
2250 left_start = hint - 1;
2251 }
2252 }
2253 if (go_left == 0)
2254 goto skip_left;
2255
2256 while (test_bit(left_start, free_i->free_secmap)) {
2257 if (left_start > 0) {
2258 left_start--;
2259 continue;
2260 }
2261 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2262 MAIN_SECS(sbi), 0);
2263 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2264 break;
2265 }
2266 secno = left_start;
2267skip_left:
4ddb1a4d
JK		 2268 segno = GET_SEG_FROM_SEC(sbi, secno);
2269 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2270
2271 /* give up on finding another zone */
2272 if (!init)
2273 goto got_it;
2274 if (sbi->secs_per_zone == 1)
2275 goto got_it;
2276 if (zoneno == old_zoneno)
2277 goto got_it;
2278 if (dir == ALLOC_LEFT) {
2279 if (!go_left && zoneno + 1 >= total_zones)
2280 goto got_it;
2281 if (go_left && zoneno == 0)
2282 goto got_it;
2283 }
2284 for (i = 0; i < NR_CURSEG_TYPE; i++)
2285 if (CURSEG_I(sbi, i)->zone == zoneno)
2286 break;
2287
2288 if (i < NR_CURSEG_TYPE) {
2289 /* zone is in user, try another */
2290 if (go_left)
2291 hint = zoneno * sbi->secs_per_zone - 1;
2292 else if (zoneno + 1 >= total_zones)
2293 hint = 0;
2294 else
2295 hint = (zoneno + 1) * sbi->secs_per_zone;
2296 init = false;
2297 goto find_other_zone;
2298 }
2299got_it:
2300 /* set it as dirty segment in free segmap */
9850cf4a 2301 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2302 __set_inuse(sbi, segno);
2303 *newseg = segno;
1a118ccf 2304 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2305}
2306
2307static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2308{
2309 struct curseg_info *curseg = CURSEG_I(sbi, type);
2310 struct summary_footer *sum_footer;
2311
2312 curseg->segno = curseg->next_segno;
4ddb1a4d 2313 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2314 curseg->next_blkoff = 0;
2315 curseg->next_segno = NULL_SEGNO;
2316
2317 sum_footer = &(curseg->sum_blk->footer);
2318 memset(sum_footer, 0, sizeof(struct summary_footer));
2319 if (IS_DATASEG(type))
2320 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2321 if (IS_NODESEG(type))
2322 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2323 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2324}
2325
7a20b8a6
JK		 2326static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2327{
a7881893
JK		 2328 /* if segs_per_sec is large than 1, we need to keep original policy. */
2329 if (sbi->segs_per_sec != 1)
2330 return CURSEG_I(sbi, type)->segno;
2331
b94929d9
YS		 2332 if (test_opt(sbi, NOHEAP) &&
2333 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2334 return 0;
2335
e066b83c
JK		 2336 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2337 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2338
2339 /* find segments from 0 to reuse freed segments */
63189b78 2340 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2341 return 0;
2342
7a20b8a6
JK		 2343 return CURSEG_I(sbi, type)->segno;
2344}
2345
0a8165d7 2346/*
351df4b2
JK		 2347 * Allocate a current working segment.
2348 * This function always allocates a free segment in LFS manner.
2349 */
2350static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2351{
2352 struct curseg_info *curseg = CURSEG_I(sbi, type);
2353 unsigned int segno = curseg->segno;
2354 int dir = ALLOC_LEFT;
2355
2356 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2357 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2358 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2359 dir = ALLOC_RIGHT;
2360
2361 if (test_opt(sbi, NOHEAP))
2362 dir = ALLOC_RIGHT;
2363
7a20b8a6 2364 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2365 get_new_segment(sbi, &segno, new_sec, dir);
2366 curseg->next_segno = segno;
2367 reset_curseg(sbi, type, 1);
2368 curseg->alloc_type = LFS;
2369}
2370
2371static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2372 struct curseg_info *seg, block_t start)
2373{
2374 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2375 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2376 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2377 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2378 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2379 int i, pos;
2380
2381 for (i = 0; i < entries; i++)
2382 target_map[i] = ckpt_map[i] | cur_map[i];
2383
2384 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2385
2386 seg->next_blkoff = pos;
351df4b2
JK		 2387}
2388
0a8165d7 2389/*
351df4b2
JK		 2390 * If a segment is written by LFS manner, next block offset is just obtained
2391 * by increasing the current block offset. However, if a segment is written by
2392 * SSR manner, next block offset obtained by calling __next_free_blkoff
2393 */
2394static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2395 struct curseg_info *seg)
2396{
2397 if (seg->alloc_type == SSR)
2398 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2399 else
2400 seg->next_blkoff++;
2401}
2402
0a8165d7 2403/*
e1c42045 2404 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2405 * manner, so it should recover the existing segment information of valid blocks
2406 */
025d63a4 2407static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2408{
2409 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2410 struct curseg_info *curseg = CURSEG_I(sbi, type);
2411 unsigned int new_segno = curseg->next_segno;
2412 struct f2fs_summary_block *sum_node;
2413 struct page *sum_page;
2414
2415 write_sum_page(sbi, curseg->sum_blk,
2416 GET_SUM_BLOCK(sbi, curseg->segno));
2417 __set_test_and_inuse(sbi, new_segno);
2418
2419 mutex_lock(&dirty_i->seglist_lock);
2420 __remove_dirty_segment(sbi, new_segno, PRE);
2421 __remove_dirty_segment(sbi, new_segno, DIRTY);
2422 mutex_unlock(&dirty_i->seglist_lock);
2423
2424 reset_curseg(sbi, type, 1);
2425 curseg->alloc_type = SSR;
2426 __next_free_blkoff(sbi, curseg, 0);
2427
4d57b86d 2428 sum_page = f2fs_get_sum_page(sbi, new_segno);
025d63a4
CY		 2429 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2430 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2431 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2432}
2433
43727527
JK		 2434static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2435{
2436 struct curseg_info *curseg = CURSEG_I(sbi, type);
2437 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2438 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2439 int i, cnt;
2440 bool reversed = false;
c192f7a4 2441
4d57b86d 2442 /* f2fs_need_SSR() already forces to do this */
e066b83c
JK		 2443 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2444 curseg->next_segno = segno;
c192f7a4 2445 return 1;
e066b83c 2446 }
43727527 2447
70d625cb
JK		 2448 /* For node segments, let's do SSR more intensively */
2449 if (IS_NODESEG(type)) {
d27c3d89
CY		 2450 if (type >= CURSEG_WARM_NODE) {
2451 reversed = true;
2452 i = CURSEG_COLD_NODE;
2453 } else {
2454 i = CURSEG_HOT_NODE;
2455 }
2456 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2457 } else {
d27c3d89
CY		 2458 if (type >= CURSEG_WARM_DATA) {
2459 reversed = true;
2460 i = CURSEG_COLD_DATA;
2461 } else {
2462 i = CURSEG_HOT_DATA;
2463 }
2464 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2465 }
43727527 2466
d27c3d89 2467 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2468 if (i == type)
2469 continue;
e066b83c
JK		 2470 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2471 curseg->next_segno = segno;
43727527 2472 return 1;
e066b83c 2473 }
c192f7a4 2474 }
43727527
JK		 2475 return 0;
2476}
2477
351df4b2
JK		 2478/*
2479 * flush out current segment and replace it with new segment
2480 * This function should be returned with success, otherwise BUG
2481 */
2482static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2483 int type, bool force)
2484{
a7881893
JK		 2485 struct curseg_info *curseg = CURSEG_I(sbi, type);
2486
7b405275 2487 if (force)
351df4b2 2488 new_curseg(sbi, type, true);
5b6c6be2
JK		 2489 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2490 type == CURSEG_WARM_NODE)
351df4b2 2491 new_curseg(sbi, type, false);
a7881893
JK		 2492 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2493 new_curseg(sbi, type, false);
4d57b86d 2494 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2495 change_curseg(sbi, type);
351df4b2
JK		 2496 else
2497 new_curseg(sbi, type, false);
dcdfff65 2498
a7881893 2499 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2500}
2501
4d57b86d 2502void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
351df4b2 2503{
6ae1be13
JK		 2504 struct curseg_info *curseg;
2505 unsigned int old_segno;
351df4b2
JK		 2506 int i;
2507
3d26fa6b
CY		 2508 down_write(&SIT_I(sbi)->sentry_lock);
2509
6ae1be13
JK		 2510 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2511 curseg = CURSEG_I(sbi, i);
2512 old_segno = curseg->segno;
2513 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2514 locate_dirty_segment(sbi, old_segno);
2515 }
3d26fa6b
CY		 2516
2517 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2518}
2519
2520static const struct segment_allocation default_salloc_ops = {
2521 .allocate_segment = allocate_segment_by_default,
2522};
2523
4d57b86d
CY		 2524bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2525 struct cp_control *cpc)
25290fa5
JK		 2526{
2527 __u64 trim_start = cpc->trim_start;
2528 bool has_candidate = false;
2529
3d26fa6b 2530 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2531 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2532 if (add_discard_addrs(sbi, cpc, true)) {
2533 has_candidate = true;
2534 break;
2535 }
2536 }
3d26fa6b 2537 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2538
2539 cpc->trim_start = trim_start;
2540 return has_candidate;
2541}
2542
01f9cf6d 2543static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 2544 struct discard_policy *dpolicy,
2545 unsigned int start, unsigned int end)
2546{
2547 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2548 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2549 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2550 struct discard_cmd *dc;
2551 struct blk_plug plug;
2552 int issued;
01f9cf6d 2553 unsigned int trimmed = 0;
9a997188
JK		 2554
2555next:
2556 issued = 0;
2557
2558 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 2559 if (unlikely(dcc->rbtree_check))
2560 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2561 &dcc->root));
9a997188 2562
4d57b86d 2563 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 2564 NULL, start,
2565 (struct rb_entry **)&prev_dc,
2566 (struct rb_entry **)&next_dc,
2567 &insert_p, &insert_parent, true);
2568 if (!dc)
2569 dc = next_dc;
2570
2571 blk_start_plug(&plug);
2572
2573 while (dc && dc->lstart <= end) {
2574 struct rb_node *node;
6b9cb124 2575 int err = 0;
9a997188
JK		 2576
2577 if (dc->len < dpolicy->granularity)
2578 goto skip;
2579
2580 if (dc->state != D_PREP) {
2581 list_move_tail(&dc->list, &dcc->fstrim_list);
2582 goto skip;
2583 }
2584
6b9cb124 2585 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
9a997188 2586
35ec7d57 2587 if (issued >= dpolicy->max_requests) {
9a997188
JK		 2588 start = dc->lstart + dc->len;
2589
6b9cb124
CY		 2590 if (err)
2591 __remove_discard_cmd(sbi, dc);
2592
9a997188
JK		 2593 blk_finish_plug(&plug);
2594 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 2595 trimmed += __wait_all_discard_cmd(sbi, NULL);
9a997188
JK		 2596 congestion_wait(BLK_RW_ASYNC, HZ/50);
2597 goto next;
2598 }
2599skip:
2600 node = rb_next(&dc->rb_node);
6b9cb124
CY		 2601 if (err)
2602 __remove_discard_cmd(sbi, dc);
9a997188
JK		 2603 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2604
2605 if (fatal_signal_pending(current))
2606 break;
2607 }
2608
2609 blk_finish_plug(&plug);
2610 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 2611
2612 return trimmed;
9a997188
JK		 2613}
2614
4b2fecc8
JK		 2615int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2616{
f7ef9b83
JK		 2617 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2618 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2619 unsigned int start_segno, end_segno;
8412663d 2620 block_t start_block, end_block;
4b2fecc8 2621 struct cp_control cpc;
78997b56 2622 struct discard_policy dpolicy;
0ea80512 2623 unsigned long long trimmed = 0;
c34f42e2 2624 int err = 0;
ad6672bb 2625 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
4b2fecc8 2626
836b5a63 2627 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2628 return -EINVAL;
2629
3f16ecd9
CY		 2630 if (end < MAIN_BLKADDR(sbi))
2631 goto out;
4b2fecc8 2632
ed214a11
YH		 2633 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2634 f2fs_msg(sbi->sb, KERN_WARNING,
2635 "Found FS corruption, run fsck to fix.");
3d165dc3 2636 return -EIO;
ed214a11
YH		 2637 }
2638
4b2fecc8 2639 /* start/end segment number in main_area */
7cd8558b
JK		 2640 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2641 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2642 GET_SEGNO(sbi, end);
ad6672bb
YS		 2643 if (need_align) {
2644 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2645 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2646 }
8412663d 2647
4b2fecc8 2648 cpc.reason = CP_DISCARD;
836b5a63 2649 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2650 cpc.trim_start = start_segno;
2651 cpc.trim_end = end_segno;
4b2fecc8 2652
377224c4
CY		 2653 if (sbi->discard_blks == 0)
2654 goto out;
74fa5f3d 2655
377224c4 2656 mutex_lock(&sbi->gc_mutex);
4d57b86d 2657 err = f2fs_write_checkpoint(sbi, &cpc);
377224c4
CY		 2658 mutex_unlock(&sbi->gc_mutex);
2659 if (err)
2660 goto out;
8412663d 2661
e555da9f
JK		 2662 /*
2663 * We filed discard candidates, but actually we don't need to wait for
2664 * all of them, since they'll be issued in idle time along with runtime
2665 * discard option. User configuration looks like using runtime discard
2666 * or periodic fstrim instead of it.
2667 */
5a615492
JK		 2668 if (test_opt(sbi, DISCARD))
2669 goto out;
2670
2671 start_block = START_BLOCK(sbi, start_segno);
2672 end_block = START_BLOCK(sbi, end_segno + 1);
2673
2674 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 2675 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2676 start_block, end_block);
5a615492 2677
01f9cf6d 2678 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2679 start_block, end_block);
377224c4 2680out:
6eae2694
CY		 2681 if (!err)
2682 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2683 return err;
4b2fecc8
JK		 2684}
2685
351df4b2
JK		 2686static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2687{
2688 struct curseg_info *curseg = CURSEG_I(sbi, type);
2689 if (curseg->next_blkoff < sbi->blocks_per_seg)
2690 return true;
2691 return false;
2692}
2693
4d57b86d 2694int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 2695{
2696 switch (hint) {
2697 case WRITE_LIFE_SHORT:
2698 return CURSEG_HOT_DATA;
2699 case WRITE_LIFE_EXTREME:
2700 return CURSEG_COLD_DATA;
2701 default:
2702 return CURSEG_WARM_DATA;
2703 }
2704}
2705
0cdd3195
HL		 2706/* This returns write hints for each segment type. This hints will be
2707 * passed down to block layer. There are mapping tables which depend on
2708 * the mount option 'whint_mode'.
2709 *
2710 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2711 *
2712 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2713 *
2714 * User F2FS Block
2715 * ---- ---- -----
2716 * META WRITE_LIFE_NOT_SET
2717 * HOT_NODE "
2718 * WARM_NODE "
2719 * COLD_NODE "
2720 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2721 * extension list " "
2722 *
2723 * -- buffered io
2724 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2725 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2726 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2727 * WRITE_LIFE_NONE " "
2728 * WRITE_LIFE_MEDIUM " "
2729 * WRITE_LIFE_LONG " "
2730 *
2731 * -- direct io
2732 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2733 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2734 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2735 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2736 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2737 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2738 *
f2e703f9
HL		 2739 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2740 *
2741 * User F2FS Block
2742 * ---- ---- -----
2743 * META WRITE_LIFE_MEDIUM;
2744 * HOT_NODE WRITE_LIFE_NOT_SET
2745 * WARM_NODE "
2746 * COLD_NODE WRITE_LIFE_NONE
2747 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2748 * extension list " "
2749 *
2750 * -- buffered io
2751 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2752 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2753 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2754 * WRITE_LIFE_NONE " "
2755 * WRITE_LIFE_MEDIUM " "
2756 * WRITE_LIFE_LONG " "
2757 *
2758 * -- direct io
2759 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2760 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2761 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2762 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2763 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2764 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2765 */
2766
4d57b86d 2767enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 2768 enum page_type type, enum temp_type temp)
2769{
63189b78 2770 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2771 if (type == DATA) {
f2e703f9 2772 if (temp == WARM)
0cdd3195 2773 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2774 else if (temp == HOT)
2775 return WRITE_LIFE_SHORT;
2776 else if (temp == COLD)
2777 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2778 } else {
2779 return WRITE_LIFE_NOT_SET;
2780 }
63189b78 2781 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2782 if (type == DATA) {
2783 if (temp == WARM)
2784 return WRITE_LIFE_LONG;
2785 else if (temp == HOT)
2786 return WRITE_LIFE_SHORT;
2787 else if (temp == COLD)
2788 return WRITE_LIFE_EXTREME;
2789 } else if (type == NODE) {
2790 if (temp == WARM || temp == HOT)
2791 return WRITE_LIFE_NOT_SET;
2792 else if (temp == COLD)
2793 return WRITE_LIFE_NONE;
2794 } else if (type == META) {
2795 return WRITE_LIFE_MEDIUM;
2796 }
0cdd3195 2797 }
f2e703f9 2798 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2799}
2800
81377bd6 2801static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2802{
81377bd6 2803 if (fio->type == DATA)
351df4b2
JK		 2804 return CURSEG_HOT_DATA;
2805 else
2806 return CURSEG_HOT_NODE;
2807}
2808
81377bd6 2809static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2810{
81377bd6
JK		 2811 if (fio->type == DATA) {
2812 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2813
2814 if (S_ISDIR(inode->i_mode))
2815 return CURSEG_HOT_DATA;
2816 else
2817 return CURSEG_COLD_DATA;
2818 } else {
81377bd6 2819 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2820 return CURSEG_WARM_NODE;
351df4b2
JK		 2821 else
2822 return CURSEG_COLD_NODE;
2823 }
2824}
2825
81377bd6 2826static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2827{
81377bd6
JK		 2828 if (fio->type == DATA) {
2829 struct inode *inode = fio->page->mapping->host;
351df4b2 2830
81377bd6 2831 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2832 return CURSEG_COLD_DATA;
b6a06cbb 2833 if (file_is_hot(inode) ||
b4c3ca8b 2834 is_inode_flag_set(inode, FI_HOT_DATA) ||
2079f115
CY		 2835 f2fs_is_atomic_file(inode) ||
2836 f2fs_is_volatile_file(inode))
ef095d19 2837 return CURSEG_HOT_DATA;
4d57b86d 2838 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2839 } else {
81377bd6
JK		 2840 if (IS_DNODE(fio->page))
2841 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2842 CURSEG_HOT_NODE;
ef095d19 2843 return CURSEG_COLD_NODE;
351df4b2
JK		 2844 }
2845}
2846
81377bd6 2847static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2848{
a912b54d
JK		 2849 int type = 0;
2850
63189b78 2851 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2852 case 2:
a912b54d
JK		 2853 type = __get_segment_type_2(fio);
2854 break;
351df4b2 2855 case 4:
a912b54d
JK		 2856 type = __get_segment_type_4(fio);
2857 break;
2858 case 6:
2859 type = __get_segment_type_6(fio);
2860 break;
2861 default:
2862 f2fs_bug_on(fio->sbi, true);
351df4b2 2863 }
81377bd6 2864
a912b54d
JK		 2865 if (IS_HOT(type))
2866 fio->temp = HOT;
2867 else if (IS_WARM(type))
2868 fio->temp = WARM;
2869 else
2870 fio->temp = COLD;
2871 return type;
351df4b2
JK		 2872}
2873
4d57b86d 2874void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 2875 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2876 struct f2fs_summary *sum, int type,
2877 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2878{
2879 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2880 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2881
2b60311d
CY		 2882 down_read(&SM_I(sbi)->curseg_lock);
2883
351df4b2 2884 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2885 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2886
2887 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2888
4e6a8d9b
JK		 2889 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2890
351df4b2
JK		 2891 /*
2892 * __add_sum_entry should be resided under the curseg_mutex
2893 * because, this function updates a summary entry in the
2894 * current summary block.
2895 */
e79efe3b 2896 __add_sum_entry(sbi, type, sum);
351df4b2 2897
351df4b2 2898 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2899
2900 stat_inc_block_count(sbi, curseg);
351df4b2 2901
65f1b80b
YS		 2902 /*
2903 * SIT information should be updated before segment allocation,
2904 * since SSR needs latest valid block information.
2905 */
2906 update_sit_entry(sbi, *new_blkaddr, 1);
2907 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2908 update_sit_entry(sbi, old_blkaddr, -1);
2909
c6f82fe9
JK		 2910 if (!__has_curseg_space(sbi, type))
2911 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 2912
351df4b2 2913 /*
65f1b80b
YS		 2914 * segment dirty status should be updated after segment allocation,
2915 * so we just need to update status only one time after previous
2916 * segment being closed.
351df4b2 2917 */
65f1b80b
YS		 2918 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2919 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 2920
3d26fa6b 2921 up_write(&sit_i->sentry_lock);
351df4b2 2922
704956ec 2923 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2924 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2925
704956ec
CY		 2926 f2fs_inode_chksum_set(sbi, page);
2927 }
2928
fb830fc5
CY		 2929 if (add_list) {
2930 struct f2fs_bio_info *io;
2931
2932 INIT_LIST_HEAD(&fio->list);
2933 fio->in_list = true;
fe16efe6 2934 fio->retry = false;
fb830fc5
CY		 2935 io = sbi->write_io[fio->type] + fio->temp;
2936 spin_lock(&io->io_lock);
2937 list_add_tail(&fio->list, &io->io_list);
2938 spin_unlock(&io->io_lock);
2939 }
2940
bfad7c2d 2941 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 2942
2943 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 2944}
2945
39d787be
CY		 2946static void update_device_state(struct f2fs_io_info *fio)
2947{
2948 struct f2fs_sb_info *sbi = fio->sbi;
2949 unsigned int devidx;
2950
2951 if (!sbi->s_ndevs)
2952 return;
2953
2954 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2955
2956 /* update device state for fsync */
4d57b86d 2957 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2958
2959 /* update device state for checkpoint */
2960 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2961 spin_lock(&sbi->dev_lock);
2962 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2963 spin_unlock(&sbi->dev_lock);
2964 }
39d787be
CY		 2965}
2966
05ca3632 2967static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2968{
81377bd6 2969 int type = __get_segment_type(fio);
107a805d 2970 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 2971
107a805d
CY		 2972 if (keep_order)
2973 down_read(&fio->sbi->io_order_lock);
0a595eba 2974reallocate:
4d57b86d 2975 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2976 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2977
351df4b2 2978 /* writeout dirty page into bdev */
fe16efe6
CY		 2979 f2fs_submit_page_write(fio);
2980 if (fio->retry) {
0a595eba
JK		 2981 fio->old_blkaddr = fio->new_blkaddr;
2982 goto reallocate;
2983 }
fe16efe6
CY		 2984
2985 update_device_state(fio);
2986
107a805d
CY		 2987 if (keep_order)
2988 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 2989}
2990
4d57b86d 2991void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 2992 enum iostat_type io_type)
351df4b2 2993{
458e6197 2994 struct f2fs_io_info fio = {
05ca3632 2995 .sbi = sbi,
458e6197 2996 .type = META,
0cdd3195 2997 .temp = HOT,
04d328de 2998 .op = REQ_OP_WRITE,
70fd7614 2999 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 3000 .old_blkaddr = page->index,
3001 .new_blkaddr = page->index,
05ca3632 3002 .page = page,
4375a336 3003 .encrypted_page = NULL,
fb830fc5 3004 .in_list = false,
458e6197
JK		 3005 };
3006
2b947003 3007 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 3008 fio.op_flags &= ~REQ_META;
2b947003 3009
351df4b2 3010 set_page_writeback(page);
17c50035 3011 ClearPageError(page);
b9109b0e 3012 f2fs_submit_page_write(&fio);
b0af6d49
CY		 3013
3014 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 3015}
3016
4d57b86d 3017void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 3018{
3019 struct f2fs_summary sum;
05ca3632 3020
351df4b2 3021 set_summary(&sum, nid, 0, 0);
05ca3632 3022 do_write_page(&sum, fio);
b0af6d49
CY		 3023
3024 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3025}
3026
4d57b86d
CY		 3027void f2fs_outplace_write_data(struct dnode_of_data *dn,
3028 struct f2fs_io_info *fio)
351df4b2 3029{
05ca3632 3030 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2 3031 struct f2fs_summary sum;
351df4b2 3032
9850cf4a 3033 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
7735730d 3034 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
05ca3632 3035 do_write_page(&sum, fio);
f28b3434 3036 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 3037
3038 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3039}
3040
4d57b86d 3041int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 3042{
b0af6d49 3043 int err;
d21b0f23 3044 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 3045
7a9d7548 3046 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 3047 /* i/o temperature is needed for passing down write hints */
3048 __get_segment_type(fio);
d21b0f23
YH		 3049
3050 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3051 GET_SEGNO(sbi, fio->new_blkaddr))->type));
3052
05ca3632 3053 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 3054
3055 err = f2fs_submit_page_bio(fio);
39d787be
CY		 3056 if (!err)
3057 update_device_state(fio);
b0af6d49
CY		 3058
3059 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3060
3061 return err;
351df4b2
JK		 3062}
3063
2b60311d
CY		 3064static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3065 unsigned int segno)
3066{
3067 int i;
3068
3069 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3070 if (CURSEG_I(sbi, i)->segno == segno)
3071 break;
3072 }
3073 return i;
3074}
3075
4d57b86d 3076void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 3077 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 3078 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 3079{
3080 struct sit_info *sit_i = SIT_I(sbi);
3081 struct curseg_info *curseg;
3082 unsigned int segno, old_cursegno;
3083 struct seg_entry *se;
3084 int type;
19f106bc 3085 unsigned short old_blkoff;
351df4b2
JK		 3086
3087 segno = GET_SEGNO(sbi, new_blkaddr);
3088 se = get_seg_entry(sbi, segno);
3089 type = se->type;
3090
2b60311d
CY		 3091 down_write(&SM_I(sbi)->curseg_lock);
3092
19f106bc
CY		 3093 if (!recover_curseg) {
3094 /* for recovery flow */
3095 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3096 if (old_blkaddr == NULL_ADDR)
3097 type = CURSEG_COLD_DATA;
3098 else
3099 type = CURSEG_WARM_DATA;
3100 }
3101 } else {
2b60311d
CY		 3102 if (IS_CURSEG(sbi, segno)) {
3103 /* se->type is volatile as SSR allocation */
3104 type = __f2fs_get_curseg(sbi, segno);
3105 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3106 } else {
351df4b2 3107 type = CURSEG_WARM_DATA;
2b60311d 3108 }
351df4b2 3109 }
19f106bc 3110
2c190504 3111 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 3112 curseg = CURSEG_I(sbi, type);
3113
3114 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3115 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3116
3117 old_cursegno = curseg->segno;
19f106bc 3118 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 3119
3120 /* change the current segment */
3121 if (segno != curseg->segno) {
3122 curseg->next_segno = segno;
025d63a4 3123 change_curseg(sbi, type);
351df4b2
JK		 3124 }
3125
491c0854 3126 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 3127 __add_sum_entry(sbi, type, sum);
351df4b2 3128
28bc106b 3129 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 3130 update_sit_entry(sbi, new_blkaddr, 1);
3131 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3132 update_sit_entry(sbi, old_blkaddr, -1);
3133
3134 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3135 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3136
351df4b2 3137 locate_dirty_segment(sbi, old_cursegno);
351df4b2 3138
19f106bc
CY		 3139 if (recover_curseg) {
3140 if (old_cursegno != curseg->segno) {
3141 curseg->next_segno = old_cursegno;
025d63a4 3142 change_curseg(sbi, type);
19f106bc
CY		 3143 }
3144 curseg->next_blkoff = old_blkoff;
3145 }
3146
3d26fa6b 3147 up_write(&sit_i->sentry_lock);
351df4b2 3148 mutex_unlock(&curseg->curseg_mutex);
2b60311d 3149 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 3150}
3151
528e3459
CY		 3152void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3153 block_t old_addr, block_t new_addr,
28bc106b
CY		 3154 unsigned char version, bool recover_curseg,
3155 bool recover_newaddr)
528e3459
CY		 3156{
3157 struct f2fs_summary sum;
3158
3159 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3160
4d57b86d 3161 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
28bc106b 3162 recover_curseg, recover_newaddr);
528e3459 3163
f28b3434 3164 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 3165}
3166
93dfe2ac 3167void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 3168 enum page_type type, bool ordered)
93dfe2ac 3169{
93dfe2ac 3170 if (PageWriteback(page)) {
4081363f
JK		 3171 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3172
b9109b0e
JK		 3173 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
3174 0, page->index, type);
fec1d657
JK		 3175 if (ordered)
3176 wait_on_page_writeback(page);
3177 else
3178 wait_for_stable_page(page);
93dfe2ac
JK		 3179 }
3180}
3181
d4c759ee 3182void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 3183{
3184 struct page *cpage;
3185
e1da7872 3186 if (!is_valid_data_blkaddr(sbi, blkaddr))
08b39fbd
CY		 3187 return;
3188
08b39fbd
CY		 3189 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3190 if (cpage) {
fec1d657 3191 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 3192 f2fs_put_page(cpage, 1);
3193 }
3194}
3195
7735730d 3196static int read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3197{
3198 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3199 struct curseg_info *seg_i;
3200 unsigned char *kaddr;
3201 struct page *page;
3202 block_t start;
3203 int i, j, offset;
3204
3205 start = start_sum_block(sbi);
3206
4d57b86d 3207 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3208 if (IS_ERR(page))
3209 return PTR_ERR(page);
351df4b2
JK		 3210 kaddr = (unsigned char *)page_address(page);
3211
3212 /* Step 1: restore nat cache */
3213 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3214 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3215
3216 /* Step 2: restore sit cache */
3217 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3218 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3219 offset = 2 * SUM_JOURNAL_SIZE;
3220
3221 /* Step 3: restore summary entries */
3222 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3223 unsigned short blk_off;
3224 unsigned int segno;
3225
3226 seg_i = CURSEG_I(sbi, i);
3227 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3228 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3229 seg_i->next_segno = segno;
3230 reset_curseg(sbi, i, 0);
3231 seg_i->alloc_type = ckpt->alloc_type[i];
3232 seg_i->next_blkoff = blk_off;
3233
3234 if (seg_i->alloc_type == SSR)
3235 blk_off = sbi->blocks_per_seg;
3236
3237 for (j = 0; j < blk_off; j++) {
3238 struct f2fs_summary *s;
3239 s = (struct f2fs_summary *)(kaddr + offset);
3240 seg_i->sum_blk->entries[j] = *s;
3241 offset += SUMMARY_SIZE;
09cbfeaf 3242 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3243 SUM_FOOTER_SIZE)
3244 continue;
3245
3246 f2fs_put_page(page, 1);
3247 page = NULL;
3248
4d57b86d 3249 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3250 if (IS_ERR(page))
3251 return PTR_ERR(page);
351df4b2
JK		 3252 kaddr = (unsigned char *)page_address(page);
3253 offset = 0;
3254 }
3255 }
3256 f2fs_put_page(page, 1);
7735730d 3257 return 0;
351df4b2
JK		 3258}
3259
3260static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3261{
3262 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3263 struct f2fs_summary_block *sum;
3264 struct curseg_info *curseg;
3265 struct page *new;
3266 unsigned short blk_off;
3267 unsigned int segno = 0;
3268 block_t blk_addr = 0;
7735730d 3269 int err = 0;
351df4b2
JK		 3270
3271 /* get segment number and block addr */
3272 if (IS_DATASEG(type)) {
3273 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3274 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3275 CURSEG_HOT_DATA]);
119ee914 3276 if (__exist_node_summaries(sbi))
351df4b2
JK		 3277 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3278 else
3279 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3280 } else {
3281 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3282 CURSEG_HOT_NODE]);
3283 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3284 CURSEG_HOT_NODE]);
119ee914 3285 if (__exist_node_summaries(sbi))
351df4b2
JK		 3286 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3287 type - CURSEG_HOT_NODE);
3288 else
3289 blk_addr = GET_SUM_BLOCK(sbi, segno);
3290 }
3291
4d57b86d 3292 new = f2fs_get_meta_page(sbi, blk_addr);
7735730d
CY		 3293 if (IS_ERR(new))
3294 return PTR_ERR(new);
351df4b2
JK		 3295 sum = (struct f2fs_summary_block *)page_address(new);
3296
3297 if (IS_NODESEG(type)) {
119ee914 3298 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3299 struct f2fs_summary *ns = &sum->entries[0];
3300 int i;
3301 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3302 ns->version = 0;
3303 ns->ofs_in_node = 0;
3304 }
3305 } else {
7735730d
CY		 3306 err = f2fs_restore_node_summary(sbi, segno, sum);
3307 if (err)
3308 goto out;
351df4b2
JK		 3309 }
3310 }
3311
3312 /* set uncompleted segment to curseg */
3313 curseg = CURSEG_I(sbi, type);
3314 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3315
3316 /* update journal info */
3317 down_write(&curseg->journal_rwsem);
3318 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3319 up_write(&curseg->journal_rwsem);
3320
3321 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3322 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3323 curseg->next_segno = segno;
3324 reset_curseg(sbi, type, 0);
3325 curseg->alloc_type = ckpt->alloc_type[type];
3326 curseg->next_blkoff = blk_off;
3327 mutex_unlock(&curseg->curseg_mutex);
7735730d 3328out:
351df4b2 3329 f2fs_put_page(new, 1);
7735730d 3330 return err;
351df4b2
JK		 3331}
3332
3333static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3334{
21d3f8e1
JQ		 3335 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3336 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3337 int type = CURSEG_HOT_DATA;
e4fc5fbf 3338 int err;
351df4b2 3339
aaec2b1d 3340 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3341 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3342
3343 if (npages >= 2)
4d57b86d 3344 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3345 META_CP, true);
3fa06d7b 3346
351df4b2 3347 /* restore for compacted data summary */
7735730d
CY		 3348 err = read_compacted_summaries(sbi);
3349 if (err)
3350 return err;
351df4b2
JK		 3351 type = CURSEG_HOT_NODE;
3352 }
3353
119ee914 3354 if (__exist_node_summaries(sbi))
4d57b86d 3355 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3356 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3357
e4fc5fbf
CY		 3358 for (; type <= CURSEG_COLD_NODE; type++) {
3359 err = read_normal_summaries(sbi, type);
3360 if (err)
3361 return err;
3362 }
3363
21d3f8e1
JQ		 3364 /* sanity check for summary blocks */
3365 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3366 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3367 return -EINVAL;
3368
351df4b2
JK		 3369 return 0;
3370}
3371
3372static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3373{
3374 struct page *page;
3375 unsigned char *kaddr;
3376 struct f2fs_summary *summary;
3377 struct curseg_info *seg_i;
3378 int written_size = 0;
3379 int i, j;
3380
4d57b86d 3381 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3382 kaddr = (unsigned char *)page_address(page);
81114baa 3383 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3384
3385 /* Step 1: write nat cache */
3386 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3387 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3388 written_size += SUM_JOURNAL_SIZE;
3389
3390 /* Step 2: write sit cache */
3391 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3392 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3393 written_size += SUM_JOURNAL_SIZE;
3394
351df4b2
JK		 3395 /* Step 3: write summary entries */
3396 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3397 unsigned short blkoff;
3398 seg_i = CURSEG_I(sbi, i);
3399 if (sbi->ckpt->alloc_type[i] == SSR)
3400 blkoff = sbi->blocks_per_seg;
3401 else
3402 blkoff = curseg_blkoff(sbi, i);
3403
3404 for (j = 0; j < blkoff; j++) {
3405 if (!page) {
4d57b86d 3406 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3407 kaddr = (unsigned char *)page_address(page);
81114baa 3408 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3409 written_size = 0;
3410 }
3411 summary = (struct f2fs_summary *)(kaddr + written_size);
3412 *summary = seg_i->sum_blk->entries[j];
3413 written_size += SUMMARY_SIZE;
351df4b2 3414
09cbfeaf 3415 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3416 SUM_FOOTER_SIZE)
3417 continue;
3418
e8d61a74 3419 set_page_dirty(page);
351df4b2
JK		 3420 f2fs_put_page(page, 1);
3421 page = NULL;
3422 }
3423 }
e8d61a74
CY		 3424 if (page) {
3425 set_page_dirty(page);
351df4b2 3426 f2fs_put_page(page, 1);
e8d61a74 3427 }
351df4b2
JK		 3428}
3429
3430static void write_normal_summaries(struct f2fs_sb_info *sbi,
3431 block_t blkaddr, int type)
3432{
3433 int i, end;
3434 if (IS_DATASEG(type))
3435 end = type + NR_CURSEG_DATA_TYPE;
3436 else
3437 end = type + NR_CURSEG_NODE_TYPE;
3438
b7ad7512
CY		 3439 for (i = type; i < end; i++)
3440 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3441}
3442
4d57b86d 3443void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3444{
aaec2b1d 3445 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3446 write_compacted_summaries(sbi, start_blk);
3447 else
3448 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3449}
3450
4d57b86d 3451void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3452{
119ee914 3453 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3454}
3455
4d57b86d 3456int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3457 unsigned int val, int alloc)
3458{
3459 int i;
3460
3461 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3462 for (i = 0; i < nats_in_cursum(journal); i++) {
3463 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3464 return i;
3465 }
dfc08a12
CY		 3466 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3467 return update_nats_in_cursum(journal, 1);
351df4b2 3468 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3469 for (i = 0; i < sits_in_cursum(journal); i++)
3470 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3471 return i;
dfc08a12
CY		 3472 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3473 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3474 }
3475 return -1;
3476}
3477
3478static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3479 unsigned int segno)
3480{
7735730d 3481 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3482}
3483
3484static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3485 unsigned int start)
3486{
3487 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3488 struct page *page;
351df4b2 3489 pgoff_t src_off, dst_off;
351df4b2
JK		 3490
3491 src_off = current_sit_addr(sbi, start);
3492 dst_off = next_sit_addr(sbi, src_off);
3493
4d57b86d 3494 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 3495 seg_info_to_sit_page(sbi, page, start);
351df4b2 3496
068c3cd8 3497 set_page_dirty(page);
351df4b2
JK		 3498 set_to_next_sit(sit_i, start);
3499
068c3cd8 3500 return page;
351df4b2
JK		 3501}
3502
184a5cd2
CY		 3503static struct sit_entry_set *grab_sit_entry_set(void)
3504{
3505 struct sit_entry_set *ses =
80c54505 3506 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3507
3508 ses->entry_cnt = 0;
3509 INIT_LIST_HEAD(&ses->set_list);
3510 return ses;
3511}
3512
3513static void release_sit_entry_set(struct sit_entry_set *ses)
3514{
3515 list_del(&ses->set_list);
3516 kmem_cache_free(sit_entry_set_slab, ses);
3517}
3518
3519static void adjust_sit_entry_set(struct sit_entry_set *ses,
3520 struct list_head *head)
3521{
3522 struct sit_entry_set *next = ses;
3523
3524 if (list_is_last(&ses->set_list, head))
3525 return;
3526
3527 list_for_each_entry_continue(next, head, set_list)
3528 if (ses->entry_cnt <= next->entry_cnt)
3529 break;
3530
3531 list_move_tail(&ses->set_list, &next->set_list);
3532}
3533
3534static void add_sit_entry(unsigned int segno, struct list_head *head)
3535{
3536 struct sit_entry_set *ses;
3537 unsigned int start_segno = START_SEGNO(segno);
3538
3539 list_for_each_entry(ses, head, set_list) {
3540 if (ses->start_segno == start_segno) {
3541 ses->entry_cnt++;
3542 adjust_sit_entry_set(ses, head);
3543 return;
3544 }
3545 }
3546
3547 ses = grab_sit_entry_set();
3548
3549 ses->start_segno = start_segno;
3550 ses->entry_cnt++;
3551 list_add(&ses->set_list, head);
3552}
3553
3554static void add_sits_in_set(struct f2fs_sb_info *sbi)
3555{
3556 struct f2fs_sm_info *sm_info = SM_I(sbi);
3557 struct list_head *set_list = &sm_info->sit_entry_set;
3558 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3559 unsigned int segno;
3560
7cd8558b 3561 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3562 add_sit_entry(segno, set_list);
3563}
3564
3565static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3566{
3567 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3568 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3569 int i;
3570
b7ad7512 3571 down_write(&curseg->journal_rwsem);
dfc08a12 3572 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3573 unsigned int segno;
3574 bool dirtied;
3575
dfc08a12 3576 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3577 dirtied = __mark_sit_entry_dirty(sbi, segno);
3578
3579 if (!dirtied)
3580 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3581 }
dfc08a12 3582 update_sits_in_cursum(journal, -i);
b7ad7512 3583 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3584}
3585
0a8165d7 3586/*
351df4b2
JK		 3587 * CP calls this function, which flushes SIT entries including sit_journal,
3588 * and moves prefree segs to free segs.
3589 */
4d57b86d 3590void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3591{
3592 struct sit_info *sit_i = SIT_I(sbi);
3593 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3594 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3595 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3596 struct sit_entry_set *ses, *tmp;
3597 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3598 bool to_journal = true;
4b2fecc8 3599 struct seg_entry *se;
351df4b2 3600
3d26fa6b 3601 down_write(&sit_i->sentry_lock);
351df4b2 3602
2b11a74b
WL		 3603 if (!sit_i->dirty_sentries)
3604 goto out;
3605
351df4b2 3606 /*
184a5cd2
CY		 3607 * add and account sit entries of dirty bitmap in sit entry
3608 * set temporarily
351df4b2 3609 */
184a5cd2 3610 add_sits_in_set(sbi);
351df4b2 3611
184a5cd2
CY		 3612 /*
3613 * if there are no enough space in journal to store dirty sit
3614 * entries, remove all entries from journal and add and account
3615 * them in sit entry set.
3616 */
dfc08a12 3617 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3618 remove_sits_in_journal(sbi);
b2955550 3619
184a5cd2
CY		 3620 /*
3621 * there are two steps to flush sit entries:
3622 * #1, flush sit entries to journal in current cold data summary block.
3623 * #2, flush sit entries to sit page.
3624 */
3625 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3626 struct page *page = NULL;
184a5cd2
CY		 3627 struct f2fs_sit_block *raw_sit = NULL;
3628 unsigned int start_segno = ses->start_segno;
3629 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3630 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3631 unsigned int segno = start_segno;
3632
3633 if (to_journal &&
dfc08a12 3634 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3635 to_journal = false;
3636
b7ad7512
CY		 3637 if (to_journal) {
3638 down_write(&curseg->journal_rwsem);
3639 } else {
184a5cd2
CY		 3640 page = get_next_sit_page(sbi, start_segno);
3641 raw_sit = page_address(page);
351df4b2 3642 }
351df4b2 3643
184a5cd2
CY		 3644 /* flush dirty sit entries in region of current sit set */
3645 for_each_set_bit_from(segno, bitmap, end) {
3646 int offset, sit_offset;
4b2fecc8
JK		 3647
3648 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3649#ifdef CONFIG_F2FS_CHECK_FS
3650 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3651 SIT_VBLOCK_MAP_SIZE))
3652 f2fs_bug_on(sbi, 1);
3653#endif
184a5cd2
CY		 3654
3655 /* add discard candidates */
c473f1a9 3656 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3657 cpc->trim_start = segno;
25290fa5 3658 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3659 }
184a5cd2
CY		 3660
3661 if (to_journal) {
4d57b86d 3662 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 3663 SIT_JOURNAL, segno, 1);
3664 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3665 segno_in_journal(journal, offset) =
184a5cd2
CY		 3666 cpu_to_le32(segno);
3667 seg_info_to_raw_sit(se,
dfc08a12 3668 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3669 check_block_count(sbi, segno,
3670 &sit_in_journal(journal, offset));
184a5cd2
CY		 3671 } else {
3672 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3673 seg_info_to_raw_sit(se,
3674 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3675 check_block_count(sbi, segno,
3676 &raw_sit->entries[sit_offset]);
184a5cd2 3677 }
351df4b2 3678
184a5cd2
CY		 3679 __clear_bit(segno, bitmap);
3680 sit_i->dirty_sentries--;
3681 ses->entry_cnt--;
351df4b2
JK		 3682 }
3683
b7ad7512
CY		 3684 if (to_journal)
3685 up_write(&curseg->journal_rwsem);
3686 else
184a5cd2
CY		 3687 f2fs_put_page(page, 1);
3688
3689 f2fs_bug_on(sbi, ses->entry_cnt);
3690 release_sit_entry_set(ses);
351df4b2 3691 }
184a5cd2
CY		 3692
3693 f2fs_bug_on(sbi, !list_empty(head));
3694 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3695out:
c473f1a9 3696 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3697 __u64 trim_start = cpc->trim_start;
3698
4b2fecc8 3699 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3700 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3701
3702 cpc->trim_start = trim_start;
4b2fecc8 3703 }
3d26fa6b 3704 up_write(&sit_i->sentry_lock);
351df4b2 3705
351df4b2
JK		 3706 set_prefree_as_free_segments(sbi);
3707}
3708
3709static int build_sit_info(struct f2fs_sb_info *sbi)
3710{
3711 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3712 struct sit_info *sit_i;
3713 unsigned int sit_segs, start;
ae27d62e 3714 char *src_bitmap;
351df4b2
JK		 3715 unsigned int bitmap_size;
3716
3717 /* allocate memory for SIT information */
acbf054d 3718 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3719 if (!sit_i)
3720 return -ENOMEM;
3721
3722 SM_I(sbi)->sit_info = sit_i;
3723
9d2a789c
KC		 3724 sit_i->sentries =
3725 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3726 MAIN_SEGS(sbi)),
3727 GFP_KERNEL);
351df4b2
JK		 3728 if (!sit_i->sentries)
3729 return -ENOMEM;
3730
7cd8558b 3731 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3732 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3733 GFP_KERNEL);
351df4b2
JK		 3734 if (!sit_i->dirty_sentries_bitmap)
3735 return -ENOMEM;
3736
7cd8558b 3737 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3738 sit_i->sentries[start].cur_valid_map
acbf054d 3739 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3740 sit_i->sentries[start].ckpt_valid_map
acbf054d 3741 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3742 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3743 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3744 return -ENOMEM;
3e025740 3745
355e7891
CY		 3746#ifdef CONFIG_F2FS_CHECK_FS
3747 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3748 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3749 if (!sit_i->sentries[start].cur_valid_map_mir)
3750 return -ENOMEM;
3751#endif
3752
3e025740
JK		 3753 if (f2fs_discard_en(sbi)) {
3754 sit_i->sentries[start].discard_map
acbf054d
CY		 3755 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3756 GFP_KERNEL);
3e025740
JK		 3757 if (!sit_i->sentries[start].discard_map)
3758 return -ENOMEM;
3759 }
351df4b2
JK		 3760 }
3761
acbf054d 3762 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3763 if (!sit_i->tmp_map)
3764 return -ENOMEM;
3765
351df4b2 3766 if (sbi->segs_per_sec > 1) {
9d2a789c
KC		 3767 sit_i->sec_entries =
3768 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3769 MAIN_SECS(sbi)),
3770 GFP_KERNEL);
351df4b2
JK		 3771 if (!sit_i->sec_entries)
3772 return -ENOMEM;
3773 }
3774
3775 /* get information related with SIT */
3776 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3777
3778 /* setup SIT bitmap from ckeckpoint pack */
3779 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3780 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3781
ae27d62e
CY		 3782 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3783 if (!sit_i->sit_bitmap)
351df4b2 3784 return -ENOMEM;
351df4b2 3785
ae27d62e
CY		 3786#ifdef CONFIG_F2FS_CHECK_FS
3787 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3788 if (!sit_i->sit_bitmap_mir)
3789 return -ENOMEM;
3790#endif
3791
351df4b2
JK		 3792 /* init SIT information */
3793 sit_i->s_ops = &default_salloc_ops;
3794
3795 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3796 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3797 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3798 sit_i->bitmap_size = bitmap_size;
3799 sit_i->dirty_sentries = 0;
3800 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3801 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3802 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3803 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3804 return 0;
3805}
3806
3807static int build_free_segmap(struct f2fs_sb_info *sbi)
3808{
351df4b2
JK		 3809 struct free_segmap_info *free_i;
3810 unsigned int bitmap_size, sec_bitmap_size;
3811
3812 /* allocate memory for free segmap information */
acbf054d 3813 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3814 if (!free_i)
3815 return -ENOMEM;
3816
3817 SM_I(sbi)->free_info = free_i;
3818
7cd8558b 3819 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3820 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3821 if (!free_i->free_segmap)
3822 return -ENOMEM;
3823
7cd8558b 3824 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3825 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3826 if (!free_i->free_secmap)
3827 return -ENOMEM;
3828
3829 /* set all segments as dirty temporarily */
3830 memset(free_i->free_segmap, 0xff, bitmap_size);
3831 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3832
3833 /* init free segmap information */
7cd8558b 3834 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3835 free_i->free_segments = 0;
3836 free_i->free_sections = 0;
1a118ccf 3837 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3838 return 0;
3839}
3840
3841static int build_curseg(struct f2fs_sb_info *sbi)
3842{
1042d60f 3843 struct curseg_info *array;
351df4b2
JK		 3844 int i;
3845
026f0507
KC		 3846 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3847 GFP_KERNEL);
351df4b2
JK		 3848 if (!array)
3849 return -ENOMEM;
3850
3851 SM_I(sbi)->curseg_array = array;
3852
3853 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3854 mutex_init(&array[i].curseg_mutex);
acbf054d 3855 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3856 if (!array[i].sum_blk)
3857 return -ENOMEM;
b7ad7512 3858 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 3859 array[i].journal = f2fs_kzalloc(sbi,
3860 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 3861 if (!array[i].journal)
3862 return -ENOMEM;
351df4b2
JK		 3863 array[i].segno = NULL_SEGNO;
3864 array[i].next_blkoff = 0;
3865 }
3866 return restore_curseg_summaries(sbi);
3867}
3868
c39a1b34 3869static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3870{
3871 struct sit_info *sit_i = SIT_I(sbi);
3872 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3873 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3874 struct seg_entry *se;
3875 struct f2fs_sit_entry sit;
74de593a
CY		 3876 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3877 unsigned int i, start, end;
3878 unsigned int readed, start_blk = 0;
c39a1b34 3879 int err = 0;
8a29c126 3880 block_t total_node_blocks = 0;
351df4b2 3881
74de593a 3882 do {
4d57b86d 3883 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
664ba972 3884 META_SIT, true);
74de593a
CY		 3885
3886 start = start_blk * sit_i->sents_per_block;
3887 end = (start_blk + readed) * sit_i->sents_per_block;
3888
7cd8558b 3889 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3890 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3891 struct page *page;
3892
9c094040 3893 se = &sit_i->sentries[start];
74de593a
CY		 3894 page = get_current_sit_page(sbi, start);
3895 sit_blk = (struct f2fs_sit_block *)page_address(page);
3896 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3897 f2fs_put_page(page, 1);
d600af23 3898
c39a1b34
JK		 3899 err = check_block_count(sbi, start, &sit);
3900 if (err)
3901 return err;
74de593a 3902 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3903 if (IS_NODESEG(se->type))
3904 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 3905
3906 /* build discard map only one time */
3e025740 3907 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3908 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3909 memset(se->discard_map, 0xff,
3910 SIT_VBLOCK_MAP_SIZE);
3911 } else {
3912 memcpy(se->discard_map,
3913 se->cur_valid_map,
3914 SIT_VBLOCK_MAP_SIZE);
3915 sbi->discard_blks +=
3916 sbi->blocks_per_seg -
3917 se->valid_blocks;
3918 }
3e025740 3919 }
a66cdd98 3920
d600af23
CY		 3921 if (sbi->segs_per_sec > 1)
3922 get_sec_entry(sbi, start)->valid_blocks +=
3923 se->valid_blocks;
351df4b2 3924 }
74de593a
CY		 3925 start_blk += readed;
3926 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3927
3928 down_read(&curseg->journal_rwsem);
3929 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3930 unsigned int old_valid_blocks;
3931
3932 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 3933 if (start >= MAIN_SEGS(sbi)) {
3934 f2fs_msg(sbi->sb, KERN_ERR,
3935 "Wrong journal entry on segno %u",
3936 start);
3937 set_sbi_flag(sbi, SBI_NEED_FSCK);
3938 err = -EINVAL;
3939 break;
3940 }
3941
d600af23
CY		 3942 se = &sit_i->sentries[start];
3943 sit = sit_in_journal(journal, i);
3944
3945 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 3946 if (IS_NODESEG(se->type))
3947 total_node_blocks -= old_valid_blocks;
d600af23 3948
c39a1b34
JK		 3949 err = check_block_count(sbi, start, &sit);
3950 if (err)
3951 break;
d600af23 3952 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 3953 if (IS_NODESEG(se->type))
3954 total_node_blocks += se->valid_blocks;
d600af23
CY		 3955
3956 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3957 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3958 memset(se->discard_map, 0xff,
3959 SIT_VBLOCK_MAP_SIZE);
3960 } else {
3961 memcpy(se->discard_map, se->cur_valid_map,
3962 SIT_VBLOCK_MAP_SIZE);
a9af3fdc
CY		 3963 sbi->discard_blks += old_valid_blocks;
3964 sbi->discard_blks -= se->valid_blocks;
1f43e2ad 3965 }
d600af23
CY		 3966 }
3967
a9af3fdc 3968 if (sbi->segs_per_sec > 1) {
d600af23 3969 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 3970 se->valid_blocks;
3971 get_sec_entry(sbi, start)->valid_blocks -=
3972 old_valid_blocks;
3973 }
d600af23
CY		 3974 }
3975 up_read(&curseg->journal_rwsem);
8a29c126
JK		 3976
3977 if (!err && total_node_blocks != valid_node_count(sbi)) {
3978 f2fs_msg(sbi->sb, KERN_ERR,
3979 "SIT is corrupted node# %u vs %u",
3980 total_node_blocks, valid_node_count(sbi));
3981 set_sbi_flag(sbi, SBI_NEED_FSCK);
3982 err = -EINVAL;
3983 }
3984
c39a1b34 3985 return err;
351df4b2
JK		 3986}
3987
3988static void init_free_segmap(struct f2fs_sb_info *sbi)
3989{
3990 unsigned int start;
3991 int type;
3992
7cd8558b 3993 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3994 struct seg_entry *sentry = get_seg_entry(sbi, start);
3995 if (!sentry->valid_blocks)
3996 __set_free(sbi, start);
c79b7ff1
JK		 3997 else
3998 SIT_I(sbi)->written_valid_blocks +=
3999 sentry->valid_blocks;
351df4b2
JK		 4000 }
4001
4002 /* set use the current segments */
4003 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4004 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4005 __set_test_and_inuse(sbi, curseg_t->segno);
4006 }
4007}
4008
4009static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4010{
4011 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4012 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 4013 unsigned int segno = 0, offset = 0;
351df4b2
JK		 4014 unsigned short valid_blocks;
4015
8736fbf0 4016 while (1) {
351df4b2 4017 /* find dirty segment based on free segmap */
7cd8558b
JK		 4018 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4019 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 4020 break;
4021 offset = segno + 1;
302bd348 4022 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 4023 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 4024 continue;
ec325b52
JK		 4025 if (valid_blocks > sbi->blocks_per_seg) {
4026 f2fs_bug_on(sbi, 1);
4027 continue;
4028 }
351df4b2
JK		 4029 mutex_lock(&dirty_i->seglist_lock);
4030 __locate_dirty_segment(sbi, segno, DIRTY);
4031 mutex_unlock(&dirty_i->seglist_lock);
4032 }
4033}
4034
5ec4e49f 4035static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4036{
4037 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 4038 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 4039
628b3d14 4040 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 4041 if (!dirty_i->victim_secmap)
351df4b2
JK		 4042 return -ENOMEM;
4043 return 0;
4044}
4045
4046static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4047{
4048 struct dirty_seglist_info *dirty_i;
4049 unsigned int bitmap_size, i;
4050
4051 /* allocate memory for dirty segments list information */
acbf054d
CY		 4052 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4053 GFP_KERNEL);
351df4b2
JK		 4054 if (!dirty_i)
4055 return -ENOMEM;
4056
4057 SM_I(sbi)->dirty_info = dirty_i;
4058 mutex_init(&dirty_i->seglist_lock);
4059
7cd8558b 4060 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 4061
4062 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 4063 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4064 GFP_KERNEL);
351df4b2
JK		 4065 if (!dirty_i->dirty_segmap[i])
4066 return -ENOMEM;
4067 }
4068
4069 init_dirty_segmap(sbi);
5ec4e49f 4070 return init_victim_secmap(sbi);
351df4b2
JK		 4071}
4072
0a8165d7 4073/*
351df4b2
JK		 4074 * Update min, max modified time for cost-benefit GC algorithm
4075 */
4076static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4077{
4078 struct sit_info *sit_i = SIT_I(sbi);
4079 unsigned int segno;
4080
3d26fa6b 4081 down_write(&sit_i->sentry_lock);
351df4b2 4082
5ad25442 4083 sit_i->min_mtime = ULLONG_MAX;
351df4b2 4084
7cd8558b 4085 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 4086 unsigned int i;
4087 unsigned long long mtime = 0;
4088
4089 for (i = 0; i < sbi->segs_per_sec; i++)
4090 mtime += get_seg_entry(sbi, segno + i)->mtime;
4091
4092 mtime = div_u64(mtime, sbi->segs_per_sec);
4093
4094 if (sit_i->min_mtime > mtime)
4095 sit_i->min_mtime = mtime;
4096 }
a1f72ac2 4097 sit_i->max_mtime = get_mtime(sbi, false);
3d26fa6b 4098 up_write(&sit_i->sentry_lock);
351df4b2
JK		 4099}
4100
4d57b86d 4101int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4102{
4103 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4104 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 4105 struct f2fs_sm_info *sm_info;
351df4b2
JK		 4106 int err;
4107
acbf054d 4108 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 4109 if (!sm_info)
4110 return -ENOMEM;
4111
4112 /* init sm info */
4113 sbi->sm_info = sm_info;
351df4b2
JK		 4114 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4115 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4116 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4117 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4118 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4119 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4120 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 4121 sm_info->rec_prefree_segments = sm_info->main_segments *
4122 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 4123 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4124 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4125
52763a4b
JK		 4126 if (!test_opt(sbi, LFS))
4127 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 4128 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 4129 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
853137ce 4130 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
ef095d19 4131 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 4132 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 4133
184a5cd2
CY		 4134 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4135
2b60311d
CY		 4136 init_rwsem(&sm_info->curseg_lock);
4137
d4fdf8ba 4138 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 4139 err = f2fs_create_flush_cmd_control(sbi);
2163d198 4140 if (err)
a688b9d9 4141 return err;
6b4afdd7
JK		 4142 }
4143
0b54fb84
JK		 4144 err = create_discard_cmd_control(sbi);
4145 if (err)
4146 return err;
4147
351df4b2
JK		 4148 err = build_sit_info(sbi);
4149 if (err)
4150 return err;
4151 err = build_free_segmap(sbi);
4152 if (err)
4153 return err;
4154 err = build_curseg(sbi);
4155 if (err)
4156 return err;
4157
4158 /* reinit free segmap based on SIT */
c39a1b34
JK		 4159 err = build_sit_entries(sbi);
4160 if (err)
4161 return err;
351df4b2
JK		 4162
4163 init_free_segmap(sbi);
4164 err = build_dirty_segmap(sbi);
4165 if (err)
4166 return err;
4167
4168 init_min_max_mtime(sbi);
4169 return 0;
4170}
4171
4172static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4173 enum dirty_type dirty_type)
4174{
4175 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4176
4177 mutex_lock(&dirty_i->seglist_lock);
39307a8e 4178 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 4179 dirty_i->nr_dirty[dirty_type] = 0;
4180 mutex_unlock(&dirty_i->seglist_lock);
4181}
4182
5ec4e49f 4183static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4184{
4185 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 4186 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 4187}
4188
4189static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4190{
4191 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4192 int i;
4193
4194 if (!dirty_i)
4195 return;
4196
4197 /* discard pre-free/dirty segments list */
4198 for (i = 0; i < NR_DIRTY_TYPE; i++)
4199 discard_dirty_segmap(sbi, i);
4200
5ec4e49f 4201 destroy_victim_secmap(sbi);
351df4b2
JK		 4202 SM_I(sbi)->dirty_info = NULL;
4203 kfree(dirty_i);
4204}
4205
4206static void destroy_curseg(struct f2fs_sb_info *sbi)
4207{
4208 struct curseg_info *array = SM_I(sbi)->curseg_array;
4209 int i;
4210
4211 if (!array)
4212 return;
4213 SM_I(sbi)->curseg_array = NULL;
b7ad7512 4214 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 4215 kfree(array[i].sum_blk);
b7ad7512
CY		 4216 kfree(array[i].journal);
4217 }
351df4b2
JK		 4218 kfree(array);
4219}
4220
4221static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4222{
4223 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4224 if (!free_i)
4225 return;
4226 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4227 kvfree(free_i->free_segmap);
4228 kvfree(free_i->free_secmap);
351df4b2
JK		 4229 kfree(free_i);
4230}
4231
4232static void destroy_sit_info(struct f2fs_sb_info *sbi)
4233{
4234 struct sit_info *sit_i = SIT_I(sbi);
4235 unsigned int start;
4236
4237 if (!sit_i)
4238 return;
4239
4240 if (sit_i->sentries) {
7cd8558b 4241 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 4242 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 4243#ifdef CONFIG_F2FS_CHECK_FS
4244 kfree(sit_i->sentries[start].cur_valid_map_mir);
4245#endif
351df4b2 4246 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 4247 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4248 }
4249 }
60a3b782
JK		 4250 kfree(sit_i->tmp_map);
4251
39307a8e
JK		 4252 kvfree(sit_i->sentries);
4253 kvfree(sit_i->sec_entries);
4254 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4255
4256 SM_I(sbi)->sit_info = NULL;
4257 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 4258#ifdef CONFIG_F2FS_CHECK_FS
4259 kfree(sit_i->sit_bitmap_mir);
4260#endif
351df4b2
JK		 4261 kfree(sit_i);
4262}
4263
4d57b86d 4264void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4265{
4266 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4267
3b03f724
CY		 4268 if (!sm_info)
4269 return;
4d57b86d 4270 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 4271 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4272 destroy_dirty_segmap(sbi);
4273 destroy_curseg(sbi);
4274 destroy_free_segmap(sbi);
4275 destroy_sit_info(sbi);
4276 sbi->sm_info = NULL;
4277 kfree(sm_info);
4278}
7fd9e544 4279
4d57b86d 4280int __init f2fs_create_segment_manager_caches(void)
7fd9e544
JK		 4281{
4282 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4283 sizeof(struct discard_entry));
7fd9e544 4284 if (!discard_entry_slab)
184a5cd2
CY		 4285 goto fail;
4286
b01a9201
JK		 4287 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4288 sizeof(struct discard_cmd));
4289 if (!discard_cmd_slab)
6ab2a308 4290 goto destroy_discard_entry;
275b66b0 4291
184a5cd2 4292 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4293 sizeof(struct sit_entry_set));
184a5cd2 4294 if (!sit_entry_set_slab)
b01a9201 4295 goto destroy_discard_cmd;
88b88a66
JK		 4296
4297 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4298 sizeof(struct inmem_pages));
4299 if (!inmem_entry_slab)
4300 goto destroy_sit_entry_set;
7fd9e544 4301 return 0;
184a5cd2 4302
88b88a66
JK		 4303destroy_sit_entry_set:
4304 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4305destroy_discard_cmd:
4306 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4307destroy_discard_entry:
184a5cd2
CY		 4308 kmem_cache_destroy(discard_entry_slab);
4309fail:
4310 return -ENOMEM;
7fd9e544
JK		 4311}
4312
4d57b86d 4313void f2fs_destroy_segment_manager_caches(void)
7fd9e544 4314{
184a5cd2 4315 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4316 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4317 kmem_cache_destroy(discard_entry_slab);
88b88a66 4318 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4319}
Linux 6.x block layer and io_uring tree(s)