]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: run discard jobs when put_super
[mirror_ubuntu-jammy-kernel.git] / fs / f2fs / segment.c
CommitLineData
7c1a000d 1// SPDX-License-Identifier: GPL-2.0
0a8165d7 2/*
351df4b2
JK		 3 * fs/f2fs/segment.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
351df4b2
JK		 7 */
8#include <linux/fs.h>
9#include <linux/f2fs_fs.h>
10#include <linux/bio.h>
11#include <linux/blkdev.h>
690e4a3e 12#include <linux/prefetch.h>
6b4afdd7 13#include <linux/kthread.h>
74de593a 14#include <linux/swap.h>
60b99b48 15#include <linux/timer.h>
1d7be270 16#include <linux/freezer.h>
1eb1ef4a 17#include <linux/sched/signal.h>
351df4b2
JK		 18
19#include "f2fs.h"
20#include "segment.h"
21#include "node.h"
5f656541 22#include "gc.h"
9e4ded3f 23#include "trace.h"
6ec178da 24#include <trace/events/f2fs.h>
351df4b2 25
9a7f143a
CL		 26#define __reverse_ffz(x) __reverse_ffs(~(x))
27
7fd9e544 28static struct kmem_cache *discard_entry_slab;
b01a9201 29static struct kmem_cache *discard_cmd_slab;
184a5cd2 30static struct kmem_cache *sit_entry_set_slab;
88b88a66 31static struct kmem_cache *inmem_entry_slab;
7fd9e544 32
f96999c3
JK		 33static unsigned long __reverse_ulong(unsigned char *str)
34{
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38#if BITS_PER_LONG == 64
39 shift = 56;
40#endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46}
47
9a7f143a
CL		 48/*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52static inline unsigned long __reverse_ffs(unsigned long word)
53{
54 int num = 0;
55
56#if BITS_PER_LONG == 64
f96999c3 57 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 58 num += 32;
f96999c3 59 else
9a7f143a 60 word >>= 32;
9a7f143a 61#endif
f96999c3 62 if ((word & 0xffff0000) == 0)
9a7f143a 63 num += 16;
f96999c3 64 else
9a7f143a 65 word >>= 16;
f96999c3
JK		 66
67 if ((word & 0xff00) == 0)
9a7f143a 68 num += 8;
f96999c3 69 else
9a7f143a 70 word >>= 8;
f96999c3 71
9a7f143a
CL		 72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
f96999c3 76
9a7f143a
CL		 77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
f96999c3 81
9a7f143a
CL		 82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85}
86
87/*
e1c42045 88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 90 * @size must be integral times of unsigned long.
9a7f143a 91 * Example:
f96999c3
JK		 92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 95 */
96static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98{
99 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 100 unsigned long result = size;
9a7f143a 101 unsigned long tmp;
9a7f143a
CL		 102
103 if (offset >= size)
104 return size;
105
692223d1 106 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 107 offset %= BITS_PER_LONG;
f96999c3 108
692223d1
FL		 109 while (1) {
110 if (*p == 0)
111 goto pass;
9a7f143a 112
f96999c3 113 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 118 if (tmp)
692223d1
FL		 119 goto found;
120pass:
121 if (size <= BITS_PER_LONG)
122 break;
9a7f143a 123 size -= BITS_PER_LONG;
692223d1 124 offset = 0;
f96999c3 125 p++;
9a7f143a 126 }
692223d1
FL		 127 return result;
128found:
129 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 130}
131
132static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134{
135 const unsigned long *p = addr + BIT_WORD(offset);
80609448 136 unsigned long result = size;
9a7f143a 137 unsigned long tmp;
9a7f143a
CL		 138
139 if (offset >= size)
140 return size;
141
80609448 142 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 143 offset %= BITS_PER_LONG;
80609448
JK		 144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
f96999c3 149 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
f96999c3 155 if (tmp != ~0UL)
80609448
JK		 156 goto found;
157pass:
158 if (size <= BITS_PER_LONG)
159 break;
9a7f143a 160 size -= BITS_PER_LONG;
80609448 161 offset = 0;
f96999c3 162 p++;
9a7f143a 163 }
80609448
JK		 164 return result;
165found:
166 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 167}
168
4d57b86d 169bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
b3a97a2a
JK		 170{
171 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
174
175 if (test_opt(sbi, LFS))
176 return false;
5b0e9539 177 if (sbi->gc_mode == GC_URGENT)
b3a97a2a 178 return true;
4354994f
DR		 179 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
180 return true;
b3a97a2a
JK		 181
182 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
a2a12b67 183 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
b3a97a2a
JK		 184}
185
4d57b86d 186void f2fs_register_inmem_page(struct inode *inode, struct page *page)
88b88a66 187{
57864ae5 188 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
88b88a66
JK		 189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
9be32d72 191
9e4ded3f 192 f2fs_trace_pid(page);
0722b101 193
decd36b6
CY		 194 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
195 SetPagePrivate(page);
196
88b88a66
JK		 197 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
198
199 /* add atomic page indices to the list */
200 new->page = page;
201 INIT_LIST_HEAD(&new->list);
decd36b6 202
88b88a66
JK		 203 /* increase reference count with clean state */
204 mutex_lock(&fi->inmem_lock);
205 get_page(page);
206 list_add_tail(&new->list, &fi->inmem_pages);
57864ae5
JK		 207 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
208 if (list_empty(&fi->inmem_ilist))
209 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
210 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
8dcf2ff7 211 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 212 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 213
214 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 215}
216
28bc106b
CY		 217static int __revoke_inmem_pages(struct inode *inode,
218 struct list_head *head, bool drop, bool recover)
29b96b54 219{
28bc106b 220 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 221 struct inmem_pages *cur, *tmp;
28bc106b 222 int err = 0;
29b96b54
CY		 223
224 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 225 struct page *page = cur->page;
226
227 if (drop)
228 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
229
230 lock_page(page);
29b96b54 231
bae0ee7a 232 f2fs_wait_on_page_writeback(page, DATA, true, true);
e5e5732d 233
28bc106b
CY		 234 if (recover) {
235 struct dnode_of_data dn;
236 struct node_info ni;
237
238 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 239retry:
28bc106b 240 set_new_dnode(&dn, inode, NULL, NULL, 0);
4d57b86d
CY		 241 err = f2fs_get_dnode_of_data(&dn, page->index,
242 LOOKUP_NODE);
7f2b4e8e
CY		 243 if (err) {
244 if (err == -ENOMEM) {
245 congestion_wait(BLK_RW_ASYNC, HZ/50);
246 cond_resched();
247 goto retry;
248 }
28bc106b
CY		 249 err = -EAGAIN;
250 goto next;
251 }
7735730d
CY		 252
253 err = f2fs_get_node_info(sbi, dn.nid, &ni);
254 if (err) {
255 f2fs_put_dnode(&dn);
256 return err;
257 }
258
f1d2564a 259 if (cur->old_addr == NEW_ADDR) {
4d57b86d 260 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
f1d2564a
DJ		 261 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
262 } else
263 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
28bc106b
CY		 264 cur->old_addr, ni.version, true, true);
265 f2fs_put_dnode(&dn);
266 }
267next:
63c52d78 268 /* we don't need to invalidate this in the sccessful status */
2baf0781 269 if (drop || recover) {
63c52d78 270 ClearPageUptodate(page);
2baf0781
CY		 271 clear_cold_data(page);
272 }
28bc106b 273 set_page_private(page, 0);
c81ced05 274 ClearPagePrivate(page);
28bc106b 275 f2fs_put_page(page, 1);
29b96b54
CY		 276
277 list_del(&cur->list);
278 kmem_cache_free(inmem_entry_slab, cur);
279 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
280 }
28bc106b 281 return err;
29b96b54
CY		 282}
283
4d57b86d 284void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
57864ae5
JK		 285{
286 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
287 struct inode *inode;
288 struct f2fs_inode_info *fi;
289next:
290 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
291 if (list_empty(head)) {
292 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
293 return;
294 }
295 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
296 inode = igrab(&fi->vfs_inode);
297 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
298
299 if (inode) {
2ef79ecb
CY		 300 if (gc_failure) {
301 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
302 goto drop;
303 goto skip;
304 }
305drop:
306 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
4d57b86d 307 f2fs_drop_inmem_pages(inode);
57864ae5
JK		 308 iput(inode);
309 }
2ef79ecb 310skip:
57864ae5
JK		 311 congestion_wait(BLK_RW_ASYNC, HZ/50);
312 cond_resched();
313 goto next;
314}
315
4d57b86d 316void f2fs_drop_inmem_pages(struct inode *inode)
29b96b54 317{
57864ae5 318 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 319 struct f2fs_inode_info *fi = F2FS_I(inode);
320
321 mutex_lock(&fi->inmem_lock);
28bc106b 322 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 323 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
324 if (!list_empty(&fi->inmem_ilist))
325 list_del_init(&fi->inmem_ilist);
326 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 327 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 328
329 clear_inode_flag(inode, FI_ATOMIC_FILE);
2ef79ecb 330 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
5fe45743 331 stat_dec_atomic_write(inode);
29b96b54
CY		 332}
333
4d57b86d 334void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
8c242db9
JK		 335{
336 struct f2fs_inode_info *fi = F2FS_I(inode);
337 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
338 struct list_head *head = &fi->inmem_pages;
339 struct inmem_pages *cur = NULL;
340
341 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
342
343 mutex_lock(&fi->inmem_lock);
344 list_for_each_entry(cur, head, list) {
345 if (cur->page == page)
346 break;
347 }
348
d0891e84 349 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
8c242db9
JK		 350 list_del(&cur->list);
351 mutex_unlock(&fi->inmem_lock);
352
353 dec_page_count(sbi, F2FS_INMEM_PAGES);
354 kmem_cache_free(inmem_entry_slab, cur);
355
356 ClearPageUptodate(page);
357 set_page_private(page, 0);
358 ClearPagePrivate(page);
359 f2fs_put_page(page, 0);
360
361 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
362}
363
4d57b86d 364static int __f2fs_commit_inmem_pages(struct inode *inode)
88b88a66
JK		 365{
366 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
367 struct f2fs_inode_info *fi = F2FS_I(inode);
368 struct inmem_pages *cur, *tmp;
88b88a66 369 struct f2fs_io_info fio = {
05ca3632 370 .sbi = sbi,
39d787be 371 .ino = inode->i_ino,
88b88a66 372 .type = DATA,
04d328de 373 .op = REQ_OP_WRITE,
70fd7614 374 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 375 .io_type = FS_DATA_IO,
88b88a66 376 };
cf52b27a 377 struct list_head revoke_list;
bab475c5 378 bool submit_bio = false;
edb27dee 379 int err = 0;
88b88a66 380
cf52b27a
CY		 381 INIT_LIST_HEAD(&revoke_list);
382
88b88a66 383 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 384 struct page *page = cur->page;
385
386 lock_page(page);
387 if (page->mapping == inode->i_mapping) {
388 trace_f2fs_commit_inmem_page(page, INMEM);
389
bae0ee7a 390 f2fs_wait_on_page_writeback(page, DATA, true, true);
8d64d365
CY		 391
392 set_page_dirty(page);
933439c8 393 if (clear_page_dirty_for_io(page)) {
29b96b54 394 inode_dec_dirty_pages(inode);
4d57b86d 395 f2fs_remove_dirty_inode(inode);
933439c8 396 }
640cc189 397retry:
28bc106b 398 fio.page = page;
e959c8f5 399 fio.old_blkaddr = NULL_ADDR;
4d978078 400 fio.encrypted_page = NULL;
cc15620b 401 fio.need_lock = LOCK_DONE;
4d57b86d 402 err = f2fs_do_write_data_page(&fio);
29b96b54 403 if (err) {
640cc189
JK		 404 if (err == -ENOMEM) {
405 congestion_wait(BLK_RW_ASYNC, HZ/50);
406 cond_resched();
407 goto retry;
408 }
28bc106b 409 unlock_page(page);
29b96b54 410 break;
70c640b1 411 }
28bc106b
CY		 412 /* record old blkaddr for revoking */
413 cur->old_addr = fio.old_blkaddr;
bab475c5 414 submit_bio = true;
28bc106b
CY		 415 }
416 unlock_page(page);
cf52b27a 417 list_move_tail(&cur->list, &revoke_list);
88b88a66 418 }
29b96b54 419
bab475c5
CY		 420 if (submit_bio)
421 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
28bc106b 422
cf52b27a
CY		 423 if (err) {
424 /*
425 * try to revoke all committed pages, but still we could fail
426 * due to no memory or other reason, if that happened, EAGAIN
427 * will be returned, which means in such case, transaction is
428 * already not integrity, caller should use journal to do the
429 * recovery or rewrite & commit last transaction. For other
430 * error number, revoking was done by filesystem itself.
431 */
432 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
433
434 /* drop all uncommitted pages */
435 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
436 } else {
437 __revoke_inmem_pages(inode, &revoke_list, false, false);
438 }
28bc106b 439
29b96b54
CY		 440 return err;
441}
442
4d57b86d 443int f2fs_commit_inmem_pages(struct inode *inode)
29b96b54
CY		 444{
445 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
446 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b 447 int err;
29b96b54
CY		 448
449 f2fs_balance_fs(sbi, true);
29b96b54 450
6f8d4455
JK		 451 down_write(&fi->i_gc_rwsem[WRITE]);
452
453 f2fs_lock_op(sbi);
5fe45743
CY		 454 set_inode_flag(inode, FI_ATOMIC_COMMIT);
455
29b96b54 456 mutex_lock(&fi->inmem_lock);
4d57b86d 457 err = __f2fs_commit_inmem_pages(inode);
28bc106b 458
57864ae5
JK		 459 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
460 if (!list_empty(&fi->inmem_ilist))
461 list_del_init(&fi->inmem_ilist);
462 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 463 mutex_unlock(&fi->inmem_lock);
464
5fe45743
CY		 465 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
466
29b96b54 467 f2fs_unlock_op(sbi);
6f8d4455
JK		 468 up_write(&fi->i_gc_rwsem[WRITE]);
469
edb27dee 470 return err;
88b88a66
JK		 471}
472
0a8165d7 473/*
351df4b2
JK		 474 * This function balances dirty node and dentry pages.
475 * In addition, it controls garbage collection.
476 */
2c4db1a6 477void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 478{
55523519
CY		 479 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
480 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 481 f2fs_stop_checkpoint(sbi, false);
55523519 482 }
0f348028 483
e589c2c4 484 /* balance_fs_bg is able to be pending */
a7881893 485 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 486 f2fs_balance_fs_bg(sbi);
487
4354994f
DR		 488 if (f2fs_is_checkpoint_ready(sbi))
489 return;
490
351df4b2 491 /*
029cd28c
JK		 492 * We should do GC or end up with checkpoint, if there are so many dirty
493 * dir/node pages without enough free segments.
351df4b2 494 */
7f3037a5 495 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 496 mutex_lock(&sbi->gc_mutex);
e066b83c 497 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 498 }
499}
500
4660f9c0
JK		 501void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
502{
64c74a7a
CY		 503 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
504 return;
505
1dcc336b 506 /* try to shrink extent cache when there is no enough memory */
4d57b86d 507 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
554df79e 508 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 509
1b38dc8e 510 /* check the # of cached NAT entries */
4d57b86d
CY		 511 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
512 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
1b38dc8e 513
4d57b86d
CY		 514 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
515 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
ad4edb83 516 else
4d57b86d 517 f2fs_build_free_nids(sbi, false, false);
31696580 518
a7d10cf3 519 if (!is_idle(sbi, REQ_TIME) &&
fd8c8caf 520 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
f455c8a5 521 return;
31696580 522
1b38dc8e 523 /* checkpoint is the only way to shrink partial cached entries */
4d57b86d
CY		 524 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
525 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 526 excess_prefree_segs(sbi) ||
527 excess_dirty_nats(sbi) ||
fd8c8caf 528 excess_dirty_nodes(sbi) ||
f455c8a5 529 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 530 if (test_opt(sbi, DATA_FLUSH)) {
531 struct blk_plug plug;
532
533 blk_start_plug(&plug);
4d57b86d 534 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 535 blk_finish_plug(&plug);
536 }
4660f9c0 537 f2fs_sync_fs(sbi->sb, true);
42190d2a 538 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 539 }
4660f9c0
JK		 540}
541
20fda56b
KM		 542static int __submit_flush_wait(struct f2fs_sb_info *sbi,
543 struct block_device *bdev)
3c62be17 544{
d62fe971 545 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 546 int ret;
547
3adc5fcb 548 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 549 bio_set_dev(bio, bdev);
3c62be17
JK		 550 ret = submit_bio_wait(bio);
551 bio_put(bio);
20fda56b
KM		 552
553 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
554 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 555 return ret;
556}
557
39d787be 558static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 559{
39d787be 560 int ret = 0;
3c62be17
JK		 561 int i;
562
39d787be
CY		 563 if (!sbi->s_ndevs)
564 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 565
39d787be 566 for (i = 0; i < sbi->s_ndevs; i++) {
4d57b86d 567 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
39d787be 568 continue;
20fda56b
KM		 569 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
570 if (ret)
571 break;
3c62be17
JK		 572 }
573 return ret;
574}
575
2163d198 576static int issue_flush_thread(void *data)
6b4afdd7
JK		 577{
578 struct f2fs_sb_info *sbi = data;
b01a9201 579 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 580 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 581repeat:
582 if (kthread_should_stop())
583 return 0;
584
dc6febb6
CY		 585 sb_start_intwrite(sbi->sb);
586
721bd4d5 587 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 588 struct flush_cmd *cmd, *next;
589 int ret;
590
721bd4d5
GZ		 591 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
592 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
593
39d787be
CY		 594 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
595
596 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 597 atomic_inc(&fcc->issued_flush);
598
721bd4d5
GZ		 599 llist_for_each_entry_safe(cmd, next,
600 fcc->dispatch_list, llnode) {
6b4afdd7 601 cmd->ret = ret;
6b4afdd7
JK		 602 complete(&cmd->wait);
603 }
a688b9d9 604 fcc->dispatch_list = NULL;
6b4afdd7
JK		 605 }
606
dc6febb6
CY		 607 sb_end_intwrite(sbi->sb);
608
a688b9d9 609 wait_event_interruptible(*q,
721bd4d5 610 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 611 goto repeat;
612}
613
39d787be 614int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 615{
b01a9201 616 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 617 struct flush_cmd cmd;
8b8dd65f 618 int ret;
6b4afdd7 619
0f7b2abd
JK		 620 if (test_opt(sbi, NOBARRIER))
621 return 0;
622
8b8dd65f 623 if (!test_opt(sbi, FLUSH_MERGE)) {
72691af6 624 atomic_inc(&fcc->queued_flush);
39d787be 625 ret = submit_flush_wait(sbi, ino);
72691af6 626 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 627 atomic_inc(&fcc->issued_flush);
628 return ret;
629 }
740432f8 630
72691af6 631 if (atomic_inc_return(&fcc->queued_flush) == 1 || sbi->s_ndevs > 1) {
39d787be 632 ret = submit_flush_wait(sbi, ino);
72691af6 633 atomic_dec(&fcc->queued_flush);
8b8dd65f
CY		 634
635 atomic_inc(&fcc->issued_flush);
740432f8
JK		 636 return ret;
637 }
6b4afdd7 638
39d787be 639 cmd.ino = ino;
adf8d90b 640 init_completion(&cmd.wait);
6b4afdd7 641
721bd4d5 642 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 643
6f890df0
CY		 644 /* update issue_list before we wake up issue_flush thread */
645 smp_mb();
646
647 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 648 wake_up(&fcc->flush_wait_queue);
6b4afdd7 649
5eba8c5d
JK		 650 if (fcc->f2fs_issue_flush) {
651 wait_for_completion(&cmd.wait);
72691af6 652 atomic_dec(&fcc->queued_flush);
5eba8c5d 653 } else {
d3238691
CY		 654 struct llist_node *list;
655
656 list = llist_del_all(&fcc->issue_list);
657 if (!list) {
658 wait_for_completion(&cmd.wait);
72691af6 659 atomic_dec(&fcc->queued_flush);
d3238691
CY		 660 } else {
661 struct flush_cmd *tmp, *next;
662
39d787be 663 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 664
665 llist_for_each_entry_safe(tmp, next, list, llnode) {
666 if (tmp == &cmd) {
667 cmd.ret = ret;
72691af6 668 atomic_dec(&fcc->queued_flush);
d3238691
CY		 669 continue;
670 }
671 tmp->ret = ret;
672 complete(&tmp->wait);
673 }
674 }
5eba8c5d 675 }
adf8d90b
CY		 676
677 return cmd.ret;
6b4afdd7
JK		 678}
679
4d57b86d 680int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
2163d198
GZ		 681{
682 dev_t dev = sbi->sb->s_bdev->bd_dev;
683 struct flush_cmd_control *fcc;
684 int err = 0;
685
b01a9201
JK		 686 if (SM_I(sbi)->fcc_info) {
687 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 688 if (fcc->f2fs_issue_flush)
689 return err;
5eba8c5d
JK		 690 goto init_thread;
691 }
692
acbf054d 693 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
2163d198
GZ		 694 if (!fcc)
695 return -ENOMEM;
8b8dd65f 696 atomic_set(&fcc->issued_flush, 0);
72691af6 697 atomic_set(&fcc->queued_flush, 0);
2163d198 698 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 699 init_llist_head(&fcc->issue_list);
b01a9201 700 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 701 if (!test_opt(sbi, FLUSH_MERGE))
702 return err;
703
5eba8c5d 704init_thread:
2163d198
GZ		 705 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
706 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
707 if (IS_ERR(fcc->f2fs_issue_flush)) {
708 err = PTR_ERR(fcc->f2fs_issue_flush);
5222595d 709 kvfree(fcc);
b01a9201 710 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 711 return err;
712 }
2163d198
GZ		 713
714 return err;
715}
716
4d57b86d 717void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 718{
b01a9201 719 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 720
5eba8c5d
JK		 721 if (fcc && fcc->f2fs_issue_flush) {
722 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
723
724 fcc->f2fs_issue_flush = NULL;
725 kthread_stop(flush_thread);
726 }
727 if (free) {
5222595d 728 kvfree(fcc);
b01a9201 729 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 730 }
2163d198
GZ		 731}
732
1228b482
CY		 733int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
734{
735 int ret = 0, i;
736
737 if (!sbi->s_ndevs)
738 return 0;
739
740 for (i = 1; i < sbi->s_ndevs; i++) {
741 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
742 continue;
743 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
744 if (ret)
745 break;
746
747 spin_lock(&sbi->dev_lock);
748 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
749 spin_unlock(&sbi->dev_lock);
750 }
751
752 return ret;
753}
754
351df4b2
JK		 755static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
756 enum dirty_type dirty_type)
757{
758 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
759
760 /* need not be added */
761 if (IS_CURSEG(sbi, segno))
762 return;
763
764 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
765 dirty_i->nr_dirty[dirty_type]++;
766
767 if (dirty_type == DIRTY) {
768 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 769 enum dirty_type t = sentry->type;
b2f2c390 770
ec325b52
JK		 771 if (unlikely(t >= DIRTY)) {
772 f2fs_bug_on(sbi, 1);
773 return;
774 }
4625d6aa
CL		 775 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
776 dirty_i->nr_dirty[t]++;
351df4b2
JK		 777 }
778}
779
780static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
781 enum dirty_type dirty_type)
782{
783 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
784
785 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
786 dirty_i->nr_dirty[dirty_type]--;
787
788 if (dirty_type == DIRTY) {
4625d6aa
CL		 789 struct seg_entry *sentry = get_seg_entry(sbi, segno);
790 enum dirty_type t = sentry->type;
791
792 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
793 dirty_i->nr_dirty[t]--;
b2f2c390 794
302bd348 795 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 796 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 797 dirty_i->victim_secmap);
351df4b2
JK		 798 }
799}
800
0a8165d7 801/*
351df4b2
JK		 802 * Should not occur error such as -ENOMEM.
803 * Adding dirty entry into seglist is not critical operation.
804 * If a given segment is one of current working segments, it won't be added.
805 */
8d8451af 806static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 807{
808 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4354994f 809 unsigned short valid_blocks, ckpt_valid_blocks;
351df4b2
JK		 810
811 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
812 return;
813
814 mutex_lock(&dirty_i->seglist_lock);
815
302bd348 816 valid_blocks = get_valid_blocks(sbi, segno, false);
4354994f 817 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
351df4b2 818
4354994f
DR		 819 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
820 ckpt_valid_blocks == sbi->blocks_per_seg)) {
351df4b2
JK		 821 __locate_dirty_segment(sbi, segno, PRE);
822 __remove_dirty_segment(sbi, segno, DIRTY);
823 } else if (valid_blocks < sbi->blocks_per_seg) {
824 __locate_dirty_segment(sbi, segno, DIRTY);
825 } else {
826 /* Recovery routine with SSR needs this */
827 __remove_dirty_segment(sbi, segno, DIRTY);
828 }
829
830 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 831}
832
4354994f
DR		 833/* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
834void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
835{
836 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
837 unsigned int segno;
838
839 mutex_lock(&dirty_i->seglist_lock);
840 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
841 if (get_valid_blocks(sbi, segno, false))
842 continue;
843 if (IS_CURSEG(sbi, segno))
844 continue;
845 __locate_dirty_segment(sbi, segno, PRE);
846 __remove_dirty_segment(sbi, segno, DIRTY);
847 }
848 mutex_unlock(&dirty_i->seglist_lock);
849}
850
851int f2fs_disable_cp_again(struct f2fs_sb_info *sbi)
852{
853 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
854 block_t ovp = overprovision_segments(sbi) << sbi->log_blocks_per_seg;
855 block_t holes[2] = {0, 0}; /* DATA and NODE */
856 struct seg_entry *se;
857 unsigned int segno;
858
859 mutex_lock(&dirty_i->seglist_lock);
860 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
861 se = get_seg_entry(sbi, segno);
862 if (IS_NODESEG(se->type))
863 holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
864 else
865 holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
866 }
867 mutex_unlock(&dirty_i->seglist_lock);
868
869 if (holes[DATA] > ovp || holes[NODE] > ovp)
870 return -EAGAIN;
871 return 0;
872}
873
874/* This is only used by SBI_CP_DISABLED */
875static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
876{
877 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
878 unsigned int segno = 0;
879
880 mutex_lock(&dirty_i->seglist_lock);
881 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
882 if (get_valid_blocks(sbi, segno, false))
883 continue;
884 if (get_ckpt_valid_blocks(sbi, segno))
885 continue;
886 mutex_unlock(&dirty_i->seglist_lock);
887 return segno;
888 }
889 mutex_unlock(&dirty_i->seglist_lock);
890 return NULL_SEGNO;
891}
892
004b6862 893static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 894 struct block_device *bdev, block_t lstart,
895 block_t start, block_t len)
275b66b0 896{
0b54fb84 897 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 898 struct list_head *pend_list;
b01a9201 899 struct discard_cmd *dc;
275b66b0 900
ba48a33e
CY		 901 f2fs_bug_on(sbi, !len);
902
903 pend_list = &dcc->pend_list[plist_idx(len)];
904
b01a9201
JK		 905 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
906 INIT_LIST_HEAD(&dc->list);
c81abe34 907 dc->bdev = bdev;
b01a9201 908 dc->lstart = lstart;
c81abe34 909 dc->start = start;
b01a9201 910 dc->len = len;
ec9895ad 911 dc->ref = 0;
15469963 912 dc->state = D_PREP;
72691af6 913 dc->queued = 0;
c81abe34 914 dc->error = 0;
b01a9201 915 init_completion(&dc->wait);
22d375dd 916 list_add_tail(&dc->list, pend_list);
35ec7d57
CY		 917 spin_lock_init(&dc->lock);
918 dc->bio_ref = 0;
5f32366a 919 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 920 dcc->undiscard_blks += len;
004b6862
CY		 921
922 return dc;
923}
924
925static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
926 struct block_device *bdev, block_t lstart,
927 block_t start, block_t len,
4dada3fd
CY		 928 struct rb_node *parent, struct rb_node **p,
929 bool leftmost)
004b6862
CY		 930{
931 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
932 struct discard_cmd *dc;
933
934 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
935
936 rb_link_node(&dc->rb_node, parent, p);
4dada3fd 937 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
004b6862
CY		 938
939 return dc;
15469963
JK		 940}
941
004b6862
CY		 942static void __detach_discard_cmd(struct discard_cmd_control *dcc,
943 struct discard_cmd *dc)
15469963 944{
dcc9165d 945 if (dc->state == D_DONE)
72691af6 946 atomic_sub(dc->queued, &dcc->queued_discard);
004b6862
CY		 947
948 list_del(&dc->list);
4dada3fd 949 rb_erase_cached(&dc->rb_node, &dcc->root);
d84d1cbd 950 dcc->undiscard_blks -= dc->len;
004b6862
CY		 951
952 kmem_cache_free(discard_cmd_slab, dc);
953
954 atomic_dec(&dcc->discard_cmd_cnt);
955}
956
957static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
958 struct discard_cmd *dc)
959{
960 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
35ec7d57 961 unsigned long flags;
dcc9165d 962
2ec6f2ef
CY		 963 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
964
35ec7d57
CY		 965 spin_lock_irqsave(&dc->lock, flags);
966 if (dc->bio_ref) {
967 spin_unlock_irqrestore(&dc->lock, flags);
968 return;
969 }
970 spin_unlock_irqrestore(&dc->lock, flags);
971
d9703d90
CY		 972 f2fs_bug_on(sbi, dc->ref);
973
c81abe34
JK		 974 if (dc->error == -EOPNOTSUPP)
975 dc->error = 0;
15469963 976
c81abe34 977 if (dc->error)
22d7ea13
CY		 978 printk_ratelimited(
979 "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
980 KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
004b6862 981 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 982}
983
c81abe34
JK		 984static void f2fs_submit_discard_endio(struct bio *bio)
985{
986 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
35ec7d57 987 unsigned long flags;
c81abe34 988
4e4cbee9 989 dc->error = blk_status_to_errno(bio->bi_status);
35ec7d57
CY		 990
991 spin_lock_irqsave(&dc->lock, flags);
992 dc->bio_ref--;
993 if (!dc->bio_ref && dc->state == D_SUBMIT) {
994 dc->state = D_DONE;
995 complete_all(&dc->wait);
996 }
997 spin_unlock_irqrestore(&dc->lock, flags);
c81abe34
JK		 998 bio_put(bio);
999}
1000
94b1e10e 1001static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
6915ea9d
CY		 1002 block_t start, block_t end)
1003{
1004#ifdef CONFIG_F2FS_CHECK_FS
1005 struct seg_entry *sentry;
1006 unsigned int segno;
1007 block_t blk = start;
1008 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1009 unsigned long *map;
1010
1011 while (blk < end) {
1012 segno = GET_SEGNO(sbi, blk);
1013 sentry = get_seg_entry(sbi, segno);
1014 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1015
008396e1
YS		 1016 if (end < START_BLOCK(sbi, segno + 1))
1017 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1018 else
1019 size = max_blocks;
6915ea9d
CY		 1020 map = (unsigned long *)(sentry->cur_valid_map);
1021 offset = __find_rev_next_bit(map, size, offset);
1022 f2fs_bug_on(sbi, offset != size);
008396e1 1023 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 1024 }
1025#endif
1026}
1027
8bb4f253
JK		 1028static void __init_discard_policy(struct f2fs_sb_info *sbi,
1029 struct discard_policy *dpolicy,
1030 int discard_type, unsigned int granularity)
1031{
1032 /* common policy */
1033 dpolicy->type = discard_type;
1034 dpolicy->sync = true;
20ee4382 1035 dpolicy->ordered = false;
8bb4f253
JK		 1036 dpolicy->granularity = granularity;
1037
1038 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1039 dpolicy->io_aware_gran = MAX_PLIST_NUM;
03f2c02d 1040 dpolicy->timeout = 0;
8bb4f253
JK		 1041
1042 if (discard_type == DPOLICY_BG) {
1043 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1044 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1045 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1046 dpolicy->io_aware = true;
cba60849 1047 dpolicy->sync = false;
20ee4382 1048 dpolicy->ordered = true;
8bb4f253
JK		 1049 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1050 dpolicy->granularity = 1;
1051 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1052 }
1053 } else if (discard_type == DPOLICY_FORCE) {
1054 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1055 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1056 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1057 dpolicy->io_aware = false;
1058 } else if (discard_type == DPOLICY_FSTRIM) {
1059 dpolicy->io_aware = false;
1060 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 1061 dpolicy->max_requests = UINT_MAX;
8bb4f253
JK		 1062 dpolicy->io_aware = false;
1063 }
1064}
1065
35ec7d57
CY		 1066static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1067 struct block_device *bdev, block_t lstart,
1068 block_t start, block_t len);
c81abe34 1069/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
6b9cb124 1070static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1071 struct discard_policy *dpolicy,
35ec7d57
CY		 1072 struct discard_cmd *dc,
1073 unsigned int *issued)
c81abe34 1074{
35ec7d57
CY		 1075 struct block_device *bdev = dc->bdev;
1076 struct request_queue *q = bdev_get_queue(bdev);
1077 unsigned int max_discard_blocks =
1078 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
c81abe34 1079 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1080 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1081 &(dcc->fstrim_list) : &(dcc->wait_list);
78997b56 1082 int flag = dpolicy->sync ? REQ_SYNC : 0;
35ec7d57
CY		 1083 block_t lstart, start, len, total_len;
1084 int err = 0;
c81abe34
JK		 1085
1086 if (dc->state != D_PREP)
6b9cb124 1087 return 0;
c81abe34 1088
d6184774 1089 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
6b9cb124 1090 return 0;
d6184774 1091
35ec7d57
CY		 1092 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1093
1094 lstart = dc->lstart;
1095 start = dc->start;
1096 len = dc->len;
1097 total_len = len;
1098
1099 dc->len = 0;
1100
1101 while (total_len && *issued < dpolicy->max_requests && !err) {
1102 struct bio *bio = NULL;
1103 unsigned long flags;
1104 bool last = true;
1105
1106 if (len > max_discard_blocks) {
1107 len = max_discard_blocks;
1108 last = false;
1109 }
1110
1111 (*issued)++;
1112 if (*issued == dpolicy->max_requests)
1113 last = true;
1114
1115 dc->len += len;
1116
b83dcfe6
CY		 1117 if (time_to_inject(sbi, FAULT_DISCARD)) {
1118 f2fs_show_injection_info(FAULT_DISCARD);
1119 err = -EIO;
1120 goto submit;
1121 }
35ec7d57
CY		 1122 err = __blkdev_issue_discard(bdev,
1123 SECTOR_FROM_BLOCK(start),
1124 SECTOR_FROM_BLOCK(len),
1125 GFP_NOFS, 0, &bio);
b83dcfe6 1126submit:
6b9cb124 1127 if (err) {
35ec7d57 1128 spin_lock_irqsave(&dc->lock, flags);
6b9cb124 1129 if (dc->state == D_PARTIAL)
35ec7d57 1130 dc->state = D_SUBMIT;
35ec7d57
CY		 1131 spin_unlock_irqrestore(&dc->lock, flags);
1132
6b9cb124
CY		 1133 break;
1134 }
35ec7d57 1135
6b9cb124 1136 f2fs_bug_on(sbi, !bio);
35ec7d57 1137
6b9cb124
CY		 1138 /*
1139 * should keep before submission to avoid D_DONE
1140 * right away
1141 */
1142 spin_lock_irqsave(&dc->lock, flags);
1143 if (last)
1144 dc->state = D_SUBMIT;
1145 else
1146 dc->state = D_PARTIAL;
1147 dc->bio_ref++;
1148 spin_unlock_irqrestore(&dc->lock, flags);
35ec7d57 1149
72691af6
JK		 1150 atomic_inc(&dcc->queued_discard);
1151 dc->queued++;
6b9cb124 1152 list_move_tail(&dc->list, wait_list);
b0af6d49 1153
6b9cb124 1154 /* sanity check on discard range */
9249dded 1155 __check_sit_bitmap(sbi, lstart, lstart + len);
35ec7d57 1156
6b9cb124
CY		 1157 bio->bi_private = dc;
1158 bio->bi_end_io = f2fs_submit_discard_endio;
1159 bio->bi_opf |= flag;
1160 submit_bio(bio);
1161
1162 atomic_inc(&dcc->issued_discard);
1163
1164 f2fs_update_iostat(sbi, FS_DISCARD, 1);
35ec7d57
CY		 1165
1166 lstart += len;
1167 start += len;
1168 total_len -= len;
1169 len = total_len;
c81abe34 1170 }
35ec7d57 1171
6b9cb124 1172 if (!err && len)
35ec7d57 1173 __update_discard_tree_range(sbi, bdev, lstart, start, len);
6b9cb124 1174 return err;
c81abe34
JK		 1175}
1176
004b6862
CY		 1177static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1178 struct block_device *bdev, block_t lstart,
1179 block_t start, block_t len,
1180 struct rb_node **insert_p,
1181 struct rb_node *insert_parent)
c81abe34 1182{
004b6862 1183 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1184 struct rb_node **p;
004b6862
CY		 1185 struct rb_node *parent = NULL;
1186 struct discard_cmd *dc = NULL;
4dada3fd 1187 bool leftmost = true;
004b6862
CY		 1188
1189 if (insert_p && insert_parent) {
1190 parent = insert_parent;
1191 p = insert_p;
1192 goto do_insert;
1193 }
c81abe34 1194
4dada3fd
CY		 1195 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1196 lstart, &leftmost);
004b6862 1197do_insert:
4dada3fd
CY		 1198 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1199 p, leftmost);
004b6862
CY		 1200 if (!dc)
1201 return NULL;
c81abe34 1202
004b6862 1203 return dc;
c81abe34
JK		 1204}
1205
ba48a33e
CY		 1206static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1207 struct discard_cmd *dc)
1208{
1209 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1210}
1211
3d6a650f
YH		 1212static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1213 struct discard_cmd *dc, block_t blkaddr)
1214{
ba48a33e 1215 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1216 struct discard_info di = dc->di;
1217 bool modified = false;
3d6a650f 1218
004b6862 1219 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1220 __remove_discard_cmd(sbi, dc);
1221 return;
1222 }
1223
d84d1cbd
CY		 1224 dcc->undiscard_blks -= di.len;
1225
004b6862 1226 if (blkaddr > di.lstart) {
3d6a650f 1227 dc->len = blkaddr - dc->lstart;
d84d1cbd 1228 dcc->undiscard_blks += dc->len;
ba48a33e 1229 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1230 modified = true;
1231 }
1232
1233 if (blkaddr < di.lstart + di.len - 1) {
1234 if (modified) {
1235 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1236 di.start + blkaddr + 1 - di.lstart,
1237 di.lstart + di.len - 1 - blkaddr,
1238 NULL, NULL);
1239 } else {
1240 dc->lstart++;
1241 dc->len--;
1242 dc->start++;
d84d1cbd 1243 dcc->undiscard_blks += dc->len;
ba48a33e 1244 __relocate_discard_cmd(dcc, dc);
004b6862 1245 }
3d6a650f
YH		 1246 }
1247}
1248
004b6862
CY		 1249static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1250 struct block_device *bdev, block_t lstart,
1251 block_t start, block_t len)
275b66b0 1252{
0b54fb84 1253 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1254 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1255 struct discard_cmd *dc;
1256 struct discard_info di = {0};
1257 struct rb_node **insert_p = NULL, *insert_parent = NULL;
35ec7d57
CY		 1258 struct request_queue *q = bdev_get_queue(bdev);
1259 unsigned int max_discard_blocks =
1260 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
004b6862 1261 block_t end = lstart + len;
275b66b0 1262
4d57b86d 1263 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
004b6862
CY		 1264 NULL, lstart,
1265 (struct rb_entry **)&prev_dc,
1266 (struct rb_entry **)&next_dc,
4dada3fd 1267 &insert_p, &insert_parent, true, NULL);
004b6862
CY		 1268 if (dc)
1269 prev_dc = dc;
1270
1271 if (!prev_dc) {
1272 di.lstart = lstart;
1273 di.len = next_dc ? next_dc->lstart - lstart : len;
1274 di.len = min(di.len, len);
1275 di.start = start;
22d375dd 1276 }
15469963 1277
004b6862
CY		 1278 while (1) {
1279 struct rb_node *node;
1280 bool merged = false;
1281 struct discard_cmd *tdc = NULL;
1282
1283 if (prev_dc) {
1284 di.lstart = prev_dc->lstart + prev_dc->len;
1285 if (di.lstart < lstart)
1286 di.lstart = lstart;
1287 if (di.lstart >= end)
1288 break;
1289
1290 if (!next_dc || next_dc->lstart > end)
1291 di.len = end - di.lstart;
1292 else
1293 di.len = next_dc->lstart - di.lstart;
1294 di.start = start + di.lstart - lstart;
1295 }
1296
1297 if (!di.len)
1298 goto next;
1299
1300 if (prev_dc && prev_dc->state == D_PREP &&
1301 prev_dc->bdev == bdev &&
35ec7d57
CY		 1302 __is_discard_back_mergeable(&di, &prev_dc->di,
1303 max_discard_blocks)) {
004b6862 1304 prev_dc->di.len += di.len;
d84d1cbd 1305 dcc->undiscard_blks += di.len;
ba48a33e 1306 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1307 di = prev_dc->di;
1308 tdc = prev_dc;
1309 merged = true;
1310 }
1311
1312 if (next_dc && next_dc->state == D_PREP &&
1313 next_dc->bdev == bdev &&
35ec7d57
CY		 1314 __is_discard_front_mergeable(&di, &next_dc->di,
1315 max_discard_blocks)) {
004b6862
CY		 1316 next_dc->di.lstart = di.lstart;
1317 next_dc->di.len += di.len;
1318 next_dc->di.start = di.start;
d84d1cbd 1319 dcc->undiscard_blks += di.len;
ba48a33e 1320 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1321 if (tdc)
1322 __remove_discard_cmd(sbi, tdc);
004b6862 1323 merged = true;
4e6a8d9b 1324 }
004b6862 1325
df0f6b44 1326 if (!merged) {
004b6862
CY		 1327 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1328 di.len, NULL, NULL);
df0f6b44 1329 }
004b6862
CY		 1330 next:
1331 prev_dc = next_dc;
1332 if (!prev_dc)
1333 break;
1334
1335 node = rb_next(&prev_dc->rb_node);
1336 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1337 }
004b6862
CY		 1338}
1339
1340static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1341 struct block_device *bdev, block_t blkstart, block_t blklen)
1342{
1343 block_t lblkstart = blkstart;
1344
0243a5f9 1345 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1346
1347 if (sbi->s_ndevs) {
1348 int devi = f2fs_target_device_index(sbi, blkstart);
1349
1350 blkstart -= FDEV(devi).start_blk;
1351 }
35ec7d57 1352 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862 1353 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
35ec7d57 1354 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862
CY		 1355 return 0;
1356}
1357
20ee4382
CY		 1358static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1359 struct discard_policy *dpolicy)
1360{
1361 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1362 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1363 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1364 struct discard_cmd *dc;
1365 struct blk_plug plug;
1366 unsigned int pos = dcc->next_pos;
1367 unsigned int issued = 0;
1368 bool io_interrupted = false;
1369
1370 mutex_lock(&dcc->cmd_lock);
1371 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1372 NULL, pos,
1373 (struct rb_entry **)&prev_dc,
1374 (struct rb_entry **)&next_dc,
4dada3fd 1375 &insert_p, &insert_parent, true, NULL);
20ee4382
CY		 1376 if (!dc)
1377 dc = next_dc;
1378
1379 blk_start_plug(&plug);
1380
1381 while (dc) {
1382 struct rb_node *node;
6b9cb124 1383 int err = 0;
20ee4382
CY		 1384
1385 if (dc->state != D_PREP)
1386 goto next;
1387
a7d10cf3 1388 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
20ee4382
CY		 1389 io_interrupted = true;
1390 break;
1391 }
1392
1393 dcc->next_pos = dc->lstart + dc->len;
6b9cb124 1394 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
20ee4382 1395
35ec7d57 1396 if (issued >= dpolicy->max_requests)
20ee4382
CY		 1397 break;
1398next:
1399 node = rb_next(&dc->rb_node);
6b9cb124
CY		 1400 if (err)
1401 __remove_discard_cmd(sbi, dc);
20ee4382
CY		 1402 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1403 }
1404
1405 blk_finish_plug(&plug);
1406
1407 if (!dc)
1408 dcc->next_pos = 0;
1409
1410 mutex_unlock(&dcc->cmd_lock);
1411
1412 if (!issued && io_interrupted)
1413 issued = -1;
1414
1415 return issued;
1416}
1417
78997b56
CY		 1418static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1419 struct discard_policy *dpolicy)
bd5b0738
CY		 1420{
1421 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1422 struct list_head *pend_list;
1423 struct discard_cmd *dc, *tmp;
1424 struct blk_plug plug;
522d1711 1425 int i, issued = 0;
e6c6de18 1426 bool io_interrupted = false;
bd5b0738 1427
03f2c02d
JK		 1428 if (dpolicy->timeout != 0)
1429 f2fs_update_time(sbi, dpolicy->timeout);
1430
78997b56 1431 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
03f2c02d
JK		 1432 if (dpolicy->timeout != 0 &&
1433 f2fs_time_over(sbi, dpolicy->timeout))
1434 break;
1435
78997b56
CY		 1436 if (i + 1 < dpolicy->granularity)
1437 break;
20ee4382
CY		 1438
1439 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1440 return __issue_discard_cmd_orderly(sbi, dpolicy);
1441
bd5b0738 1442 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1443
1444 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1445 if (list_empty(pend_list))
1446 goto next;
67fce70b
CY		 1447 if (unlikely(dcc->rbtree_check))
1448 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1449 &dcc->root));
33da62cf 1450 blk_start_plug(&plug);
bd5b0738
CY		 1451 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1452 f2fs_bug_on(sbi, dc->state != D_PREP);
1453
ecc9aa00 1454 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
a7d10cf3 1455 !is_idle(sbi, DISCARD_TIME)) {
e6c6de18 1456 io_interrupted = true;
522d1711 1457 break;
969d1b18 1458 }
e6c6de18 1459
35ec7d57 1460 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
522d1711 1461
35ec7d57 1462 if (issued >= dpolicy->max_requests)
33da62cf 1463 break;
bd5b0738 1464 }
33da62cf 1465 blk_finish_plug(&plug);
49c60c67 1466next:
33da62cf
CY		 1467 mutex_unlock(&dcc->cmd_lock);
1468
522d1711 1469 if (issued >= dpolicy->max_requests || io_interrupted)
33da62cf 1470 break;
bd5b0738 1471 }
969d1b18 1472
e6c6de18
CY		 1473 if (!issued && io_interrupted)
1474 issued = -1;
1475
969d1b18
CY		 1476 return issued;
1477}
1478
cf5c759f 1479static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1480{
1481 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1482 struct list_head *pend_list;
1483 struct discard_cmd *dc, *tmp;
1484 int i;
cf5c759f 1485 bool dropped = false;
969d1b18
CY		 1486
1487 mutex_lock(&dcc->cmd_lock);
1488 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1489 pend_list = &dcc->pend_list[i];
1490 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1491 f2fs_bug_on(sbi, dc->state != D_PREP);
1492 __remove_discard_cmd(sbi, dc);
cf5c759f 1493 dropped = true;
969d1b18
CY		 1494 }
1495 }
1496 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1497
1498 return dropped;
bd5b0738
CY		 1499}
1500
4d57b86d 1501void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
7950e9ac
CY		 1502{
1503 __drop_discard_cmd(sbi);
1504}
1505
0ea80512 1506static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1507 struct discard_cmd *dc)
1508{
1509 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1510 unsigned int len = 0;
2a510c00
CY		 1511
1512 wait_for_completion_io(&dc->wait);
1513 mutex_lock(&dcc->cmd_lock);
1514 f2fs_bug_on(sbi, dc->state != D_DONE);
1515 dc->ref--;
0ea80512
CY		 1516 if (!dc->ref) {
1517 if (!dc->error)
1518 len = dc->len;
2a510c00 1519 __remove_discard_cmd(sbi, dc);
0ea80512 1520 }
2a510c00 1521 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1522
1523 return len;
2a510c00
CY		 1524}
1525
0ea80512 1526static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1527 struct discard_policy *dpolicy,
1528 block_t start, block_t end)
63a94fa1
CY		 1529{
1530 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1531 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1532 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1533 struct discard_cmd *dc, *tmp;
6afae633 1534 bool need_wait;
0ea80512 1535 unsigned int trimmed = 0;
6afae633
CY		 1536
1537next:
1538 need_wait = false;
63a94fa1
CY		 1539
1540 mutex_lock(&dcc->cmd_lock);
1541 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1542 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1543 continue;
78997b56 1544 if (dc->len < dpolicy->granularity)
8412663d 1545 continue;
78997b56 1546 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1547 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1548 if (!dc->error)
1549 trimmed += dc->len;
63a94fa1 1550 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1551 } else {
1552 dc->ref++;
1553 need_wait = true;
1554 break;
63a94fa1
CY		 1555 }
1556 }
1557 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1558
1559 if (need_wait) {
0ea80512 1560 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1561 goto next;
1562 }
0ea80512
CY		 1563
1564 return trimmed;
63a94fa1
CY		 1565}
1566
01f9cf6d 1567static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1568 struct discard_policy *dpolicy)
8412663d 1569{
9a997188 1570 struct discard_policy dp;
01f9cf6d 1571 unsigned int discard_blks;
9a997188 1572
01f9cf6d
CY		 1573 if (dpolicy)
1574 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
9a997188
JK		 1575
1576 /* wait all */
8bb4f253 1577 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
01f9cf6d 1578 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1579 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
01f9cf6d
CY		 1580 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1581
1582 return discard_blks;
8412663d
CY		 1583}
1584
004b6862 1585/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1586static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1587{
1588 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1589 struct discard_cmd *dc;
ec9895ad 1590 bool need_wait = false;
004b6862
CY		 1591
1592 mutex_lock(&dcc->cmd_lock);
4d57b86d
CY		 1593 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1594 NULL, blkaddr);
004b6862 1595 if (dc) {
ec9895ad
CY		 1596 if (dc->state == D_PREP) {
1597 __punch_discard_cmd(sbi, dc, blkaddr);
1598 } else {
1599 dc->ref++;
1600 need_wait = true;
1601 }
275b66b0 1602 }
d431413f 1603 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1604
2a510c00
CY		 1605 if (need_wait)
1606 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1607}
1608
4d57b86d 1609void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
cce13252
CY		 1610{
1611 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1612
1613 if (dcc && dcc->f2fs_issue_discard) {
1614 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1615
1616 dcc->f2fs_issue_discard = NULL;
1617 kthread_stop(discard_thread);
ec9895ad 1618 }
d431413f
CY		 1619}
1620
8412663d 1621/* This comes from f2fs_put_super */
03f2c02d 1622bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
969d1b18
CY		 1623{
1624 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1625 struct discard_policy dpolicy;
cf5c759f 1626 bool dropped;
969d1b18 1627
8bb4f253
JK		 1628 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1629 dcc->discard_granularity);
03f2c02d 1630 dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
78997b56 1631 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1632 dropped = __drop_discard_cmd(sbi);
cf5c759f 1633
9a997188
JK		 1634 /* just to make sure there is no pending discard commands */
1635 __wait_all_discard_cmd(sbi, NULL);
2482c432
CY		 1636
1637 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
cf5c759f 1638 return dropped;
969d1b18
CY		 1639}
1640
15469963
JK		 1641static int issue_discard_thread(void *data)
1642{
1643 struct f2fs_sb_info *sbi = data;
1644 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1645 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1646 struct discard_policy dpolicy;
969d1b18
CY		 1647 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1648 int issued;
15469963 1649
1d7be270 1650 set_freezable();
15469963 1651
1d7be270 1652 do {
8bb4f253 1653 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1654 dcc->discard_granularity);
1655
969d1b18
CY		 1656 wait_event_interruptible_timeout(*q,
1657 kthread_should_stop() || freezing(current) ||
1658 dcc->discard_wake,
1659 msecs_to_jiffies(wait_ms));
35a9a766
SY		 1660
1661 if (dcc->discard_wake)
1662 dcc->discard_wake = 0;
1663
76c7bfb3
JK		 1664 /* clean up pending candidates before going to sleep */
1665 if (atomic_read(&dcc->queued_discard))
1666 __wait_all_discard_cmd(sbi, NULL);
1667
1d7be270
JK		 1668 if (try_to_freeze())
1669 continue;
3b60d802
CY		 1670 if (f2fs_readonly(sbi->sb))
1671 continue;
1d7be270
JK		 1672 if (kthread_should_stop())
1673 return 0;
d6184774
YH		 1674 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1675 wait_ms = dpolicy.max_interval;
1676 continue;
1677 }
15469963 1678
5b0e9539 1679 if (sbi->gc_mode == GC_URGENT)
8bb4f253 1680 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1681
dc6febb6
CY		 1682 sb_start_intwrite(sbi->sb);
1683
78997b56 1684 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1685 if (issued > 0) {
78997b56
CY		 1686 __wait_all_discard_cmd(sbi, &dpolicy);
1687 wait_ms = dpolicy.min_interval;
f9d1dced 1688 } else if (issued == -1){
a7d10cf3
ST		 1689 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1690 if (!wait_ms)
abde73c7 1691 wait_ms = dpolicy.mid_interval;
969d1b18 1692 } else {
78997b56 1693 wait_ms = dpolicy.max_interval;
969d1b18 1694 }
1d7be270 1695
dc6febb6 1696 sb_end_intwrite(sbi->sb);
1d7be270 1697
1d7be270
JK		 1698 } while (!kthread_should_stop());
1699 return 0;
15469963
JK		 1700}
1701
f46e8809 1702#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1703static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1704 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1705{
92592285 1706 sector_t sector, nr_sects;
10a875f8 1707 block_t lblkstart = blkstart;
3c62be17
JK		 1708 int devi = 0;
1709
1710 if (sbi->s_ndevs) {
1711 devi = f2fs_target_device_index(sbi, blkstart);
1712 blkstart -= FDEV(devi).start_blk;
1713 }
f46e8809
DLM		 1714
1715 /*
1716 * We need to know the type of the zone: for conventional zones,
1717 * use regular discard if the drive supports it. For sequential
1718 * zones, reset the zone write pointer.
1719 */
3c62be17 1720 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1721
1722 case BLK_ZONE_TYPE_CONVENTIONAL:
1723 if (!blk_queue_discard(bdev_get_queue(bdev)))
1724 return 0;
c81abe34 1725 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1726 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1727 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1728 sector = SECTOR_FROM_BLOCK(blkstart);
1729 nr_sects = SECTOR_FROM_BLOCK(blklen);
1730
1731 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1732 nr_sects != bdev_zone_sectors(bdev)) {
1733 f2fs_msg(sbi->sb, KERN_INFO,
1734 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1735 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1736 blkstart, blklen);
1737 return -EIO;
1738 }
d50aaeec 1739 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1740 return blkdev_reset_zones(bdev, sector,
1741 nr_sects, GFP_NOFS);
1742 default:
1743 /* Unknown zone type: broken device ? */
1744 return -EIO;
1745 }
1746}
1747#endif
1748
3c62be17
JK		 1749static int __issue_discard_async(struct f2fs_sb_info *sbi,
1750 struct block_device *bdev, block_t blkstart, block_t blklen)
1751{
1752#ifdef CONFIG_BLK_DEV_ZONED
7beb01f7 1753 if (f2fs_sb_has_blkzoned(sbi) &&
3c62be17
JK		 1754 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1755 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1756#endif
c81abe34 1757 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1758}
1759
1e87a78d 1760static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1761 block_t blkstart, block_t blklen)
1762{
3c62be17
JK		 1763 sector_t start = blkstart, len = 0;
1764 struct block_device *bdev;
a66cdd98
JK		 1765 struct seg_entry *se;
1766 unsigned int offset;
1767 block_t i;
3c62be17
JK		 1768 int err = 0;
1769
1770 bdev = f2fs_target_device(sbi, blkstart, NULL);
1771
1772 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1773 if (i != start) {
1774 struct block_device *bdev2 =
1775 f2fs_target_device(sbi, i, NULL);
1776
1777 if (bdev2 != bdev) {
1778 err = __issue_discard_async(sbi, bdev,
1779 start, len);
1780 if (err)
1781 return err;
1782 bdev = bdev2;
1783 start = i;
1784 len = 0;
1785 }
1786 }
a66cdd98 1787
a66cdd98
JK		 1788 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1789 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1790
1791 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1792 sbi->discard_blks--;
1793 }
f46e8809 1794
3c62be17
JK		 1795 if (len)
1796 err = __issue_discard_async(sbi, bdev, start, len);
1797 return err;
1e87a78d
JK		 1798}
1799
25290fa5
JK		 1800static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1801 bool check_only)
adf4983b 1802{
b2955550
JK		 1803 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1804 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1805 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1806 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1807 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1808 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1809 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1810 unsigned int start = 0, end = -1;
c473f1a9 1811 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1812 struct discard_entry *de = NULL;
46f84c2c 1813 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1814 int i;
1815
7d20c8ab 1816 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
25290fa5 1817 return false;
b2955550 1818
a66cdd98 1819 if (!force) {
7d20c8ab 1820 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
0b54fb84
JK		 1821 SM_I(sbi)->dcc_info->nr_discards >=
1822 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1823 return false;
4b2fecc8
JK		 1824 }
1825
b2955550
JK		 1826 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1827 for (i = 0; i < entries; i++)
a66cdd98 1828 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1829 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1830
0b54fb84
JK		 1831 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1832 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1833 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1834 if (start >= max_blocks)
1835 break;
1836
1837 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1838 if (force && start && end != max_blocks
1839 && (end - start) < cpc->trim_minlen)
1840 continue;
1841
25290fa5
JK		 1842 if (check_only)
1843 return true;
1844
a7eeb823
CY		 1845 if (!de) {
1846 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1847 GFP_F2FS_ZERO);
1848 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1849 list_add_tail(&de->list, head);
1850 }
1851
1852 for (i = start; i < end; i++)
1853 __set_bit_le(i, (void *)de->discard_map);
1854
1855 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1856 }
25290fa5 1857 return false;
b2955550
JK		 1858}
1859
af8ff65b
CY		 1860static void release_discard_addr(struct discard_entry *entry)
1861{
1862 list_del(&entry->list);
1863 kmem_cache_free(discard_entry_slab, entry);
1864}
1865
4d57b86d 1866void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1867{
46f84c2c 1868 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1869 struct discard_entry *entry, *this;
1870
1871 /* drop caches */
af8ff65b
CY		 1872 list_for_each_entry_safe(entry, this, head, list)
1873 release_discard_addr(entry);
4b2fecc8
JK		 1874}
1875
0a8165d7 1876/*
4d57b86d 1877 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1878 */
1879static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1880{
1881 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1882 unsigned int segno;
351df4b2
JK		 1883
1884 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1885 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1886 __set_test_and_free(sbi, segno);
351df4b2
JK		 1887 mutex_unlock(&dirty_i->seglist_lock);
1888}
1889
4d57b86d
CY		 1890void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1891 struct cp_control *cpc)
351df4b2 1892{
969d1b18
CY		 1893 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1894 struct list_head *head = &dcc->entry_list;
2d7b822a 1895 struct discard_entry *entry, *this;
351df4b2 1896 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1897 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1898 unsigned int start = 0, end = -1;
36abef4e 1899 unsigned int secno, start_segno;
c473f1a9 1900 bool force = (cpc->reason & CP_DISCARD);
2c70c5e3 1901 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
351df4b2
JK		 1902
1903 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1904
351df4b2 1905 while (1) {
29e59c14 1906 int i;
ad6672bb
YS		 1907
1908 if (need_align && end != -1)
1909 end--;
7cd8558b
JK		 1910 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1911 if (start >= MAIN_SEGS(sbi))
351df4b2 1912 break;
7cd8558b
JK		 1913 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1914 start + 1);
29e59c14 1915
ad6672bb
YS		 1916 if (need_align) {
1917 start = rounddown(start, sbi->segs_per_sec);
1918 end = roundup(end, sbi->segs_per_sec);
1919 }
29e59c14 1920
ad6672bb
YS		 1921 for (i = start; i < end; i++) {
1922 if (test_and_clear_bit(i, prefree_map))
1923 dirty_i->nr_dirty[PRE]--;
1924 }
29e59c14 1925
7d20c8ab 1926 if (!f2fs_realtime_discard_enable(sbi))
29e59c14 1927 continue;
351df4b2 1928
650d3c4e
YH		 1929 if (force && start >= cpc->trim_start &&
1930 (end - 1) <= cpc->trim_end)
1931 continue;
1932
2c70c5e3 1933 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
36abef4e 1934 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1935 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1936 continue;
1937 }
1938next:
4ddb1a4d
JK		 1939 secno = GET_SEC_FROM_SEG(sbi, start);
1940 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1941 if (!IS_CURSEC(sbi, secno) &&
302bd348 1942 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1943 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1944 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1945
1946 start = start_segno + sbi->segs_per_sec;
1947 if (start < end)
1948 goto next;
8b107f5b
JK		 1949 else
1950 end = start - 1;
351df4b2
JK		 1951 }
1952 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1953
1954 /* send small discards */
2d7b822a 1955 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1956 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1957 bool is_valid = test_bit_le(0, entry->discard_map);
1958
1959find_next:
1960 if (is_valid) {
1961 next_pos = find_next_zero_bit_le(entry->discard_map,
1962 sbi->blocks_per_seg, cur_pos);
1963 len = next_pos - cur_pos;
1964
7beb01f7 1965 if (f2fs_sb_has_blkzoned(sbi) ||
acfd2810 1966 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1967 goto skip;
1968
1969 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1970 len);
a7eeb823
CY		 1971 total_len += len;
1972 } else {
1973 next_pos = find_next_bit_le(entry->discard_map,
1974 sbi->blocks_per_seg, cur_pos);
1975 }
836b5a63 1976skip:
a7eeb823
CY		 1977 cur_pos = next_pos;
1978 is_valid = !is_valid;
1979
1980 if (cur_pos < sbi->blocks_per_seg)
1981 goto find_next;
1982
af8ff65b 1983 release_discard_addr(entry);
969d1b18 1984 dcc->nr_discards -= total_len;
b2955550 1985 }
34e159da 1986
01983c71 1987 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1988}
1989
8ed59745 1990static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1991{
15469963 1992 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1993 struct discard_cmd_control *dcc;
ba48a33e 1994 int err = 0, i;
0b54fb84
JK		 1995
1996 if (SM_I(sbi)->dcc_info) {
1997 dcc = SM_I(sbi)->dcc_info;
1998 goto init_thread;
1999 }
2000
acbf054d 2001 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 2002 if (!dcc)
2003 return -ENOMEM;
2004
969d1b18 2005 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 2006 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 2007 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 2008 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 2009 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 2010 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 2011 mutex_init(&dcc->cmd_lock);
8b8dd65f 2012 atomic_set(&dcc->issued_discard, 0);
72691af6 2013 atomic_set(&dcc->queued_discard, 0);
5f32366a 2014 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 2015 dcc->nr_discards = 0;
d618ebaf 2016 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 2017 dcc->undiscard_blks = 0;
20ee4382 2018 dcc->next_pos = 0;
4dada3fd 2019 dcc->root = RB_ROOT_CACHED;
67fce70b 2020 dcc->rbtree_check = false;
0b54fb84 2021
15469963 2022 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 2023 SM_I(sbi)->dcc_info = dcc;
2024init_thread:
15469963
JK		 2025 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2026 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2027 if (IS_ERR(dcc->f2fs_issue_discard)) {
2028 err = PTR_ERR(dcc->f2fs_issue_discard);
5222595d 2029 kvfree(dcc);
15469963
JK		 2030 SM_I(sbi)->dcc_info = NULL;
2031 return err;
2032 }
2033
0b54fb84
JK		 2034 return err;
2035}
2036
f099405f 2037static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 2038{
2039 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2040
f099405f
CY		 2041 if (!dcc)
2042 return;
2043
4d57b86d 2044 f2fs_stop_discard_thread(sbi);
f099405f 2045
5222595d 2046 kvfree(dcc);
f099405f 2047 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 2048}
2049
184a5cd2 2050static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 2051{
2052 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 2053
2054 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 2055 sit_i->dirty_sentries++;
184a5cd2
CY		 2056 return false;
2057 }
2058
2059 return true;
351df4b2
JK		 2060}
2061
2062static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2063 unsigned int segno, int modified)
2064{
2065 struct seg_entry *se = get_seg_entry(sbi, segno);
2066 se->type = type;
2067 if (modified)
2068 __mark_sit_entry_dirty(sbi, segno);
2069}
2070
2071static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2072{
2073 struct seg_entry *se;
2074 unsigned int segno, offset;
2075 long int new_vblocks;
6415fedc
YS		 2076 bool exist;
2077#ifdef CONFIG_F2FS_CHECK_FS
2078 bool mir_exist;
2079#endif
351df4b2
JK		 2080
2081 segno = GET_SEGNO(sbi, blkaddr);
2082
2083 se = get_seg_entry(sbi, segno);
2084 new_vblocks = se->valid_blocks + del;
491c0854 2085 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 2086
9850cf4a 2087 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 2088 (new_vblocks > sbi->blocks_per_seg)));
2089
2090 se->valid_blocks = new_vblocks;
a1f72ac2
CY		 2091 se->mtime = get_mtime(sbi, false);
2092 if (se->mtime > SIT_I(sbi)->max_mtime)
2093 SIT_I(sbi)->max_mtime = se->mtime;
351df4b2
JK		 2094
2095 /* Update valid block bitmap */
2096 if (del > 0) {
6415fedc 2097 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 2098#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2099 mir_exist = f2fs_test_and_set_bit(offset,
2100 se->cur_valid_map_mir);
2101 if (unlikely(exist != mir_exist)) {
2102 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2103 "when setting bitmap, blk:%u, old bit:%d",
2104 blkaddr, exist);
05796763 2105 f2fs_bug_on(sbi, 1);
6415fedc 2106 }
355e7891 2107#endif
6415fedc
YS		 2108 if (unlikely(exist)) {
2109 f2fs_msg(sbi->sb, KERN_ERR,
2110 "Bitmap was wrongly set, blk:%u", blkaddr);
2111 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2112 se->valid_blocks--;
2113 del = 0;
355e7891 2114 }
6415fedc 2115
7d20c8ab 2116 if (!f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 2117 sbi->discard_blks--;
720037f9
JK		 2118
2119 /* don't overwrite by SSR to keep node chain */
4354994f
DR		 2120 if (IS_NODESEG(se->type) &&
2121 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
720037f9
JK		 2122 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2123 se->ckpt_valid_blocks++;
2124 }
351df4b2 2125 } else {
6415fedc 2126 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 2127#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2128 mir_exist = f2fs_test_and_clear_bit(offset,
2129 se->cur_valid_map_mir);
2130 if (unlikely(exist != mir_exist)) {
2131 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2132 "when clearing bitmap, blk:%u, old bit:%d",
2133 blkaddr, exist);
05796763 2134 f2fs_bug_on(sbi, 1);
6415fedc 2135 }
355e7891 2136#endif
6415fedc
YS		 2137 if (unlikely(!exist)) {
2138 f2fs_msg(sbi->sb, KERN_ERR,
2139 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2140 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2141 se->valid_blocks++;
2142 del = 0;
4354994f
DR		 2143 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2144 /*
2145 * If checkpoints are off, we must not reuse data that
2146 * was used in the previous checkpoint. If it was used
2147 * before, we must track that to know how much space we
2148 * really have.
2149 */
2150 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2151 sbi->unusable_block_count++;
355e7891 2152 }
6415fedc 2153
7d20c8ab 2154 if (f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 2155 sbi->discard_blks++;
351df4b2
JK		 2156 }
2157 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2158 se->ckpt_valid_blocks += del;
2159
2160 __mark_sit_entry_dirty(sbi, segno);
2161
2162 /* update total number of valid blocks to be written in ckpt area */
2163 SIT_I(sbi)->written_valid_blocks += del;
2164
2c70c5e3 2165 if (__is_large_section(sbi))
351df4b2
JK		 2166 get_sec_entry(sbi, segno)->valid_blocks += del;
2167}
2168
4d57b86d 2169void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 2170{
2171 unsigned int segno = GET_SEGNO(sbi, addr);
2172 struct sit_info *sit_i = SIT_I(sbi);
2173
9850cf4a 2174 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 2175 if (addr == NEW_ADDR)
2176 return;
2177
6aa58d8a
CY		 2178 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2179
351df4b2 2180 /* add it into sit main buffer */
3d26fa6b 2181 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2182
2183 update_sit_entry(sbi, addr, -1);
2184
2185 /* add it into dirty seglist */
2186 locate_dirty_segment(sbi, segno);
2187
3d26fa6b 2188 up_write(&sit_i->sentry_lock);
351df4b2
JK		 2189}
2190
4d57b86d 2191bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 2192{
2193 struct sit_info *sit_i = SIT_I(sbi);
2194 unsigned int segno, offset;
2195 struct seg_entry *se;
2196 bool is_cp = false;
2197
e1da7872 2198 if (!is_valid_data_blkaddr(sbi, blkaddr))
6e2c64ad
JK		 2199 return true;
2200
3d26fa6b 2201 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2202
2203 segno = GET_SEGNO(sbi, blkaddr);
2204 se = get_seg_entry(sbi, segno);
2205 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2206
2207 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2208 is_cp = true;
2209
3d26fa6b 2210 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2211
2212 return is_cp;
2213}
2214
0a8165d7 2215/*
351df4b2
JK		 2216 * This function should be resided under the curseg_mutex lock
2217 */
2218static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 2219 struct f2fs_summary *sum)
351df4b2
JK		 2220{
2221 struct curseg_info *curseg = CURSEG_I(sbi, type);
2222 void *addr = curseg->sum_blk;
e79efe3b 2223 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 2224 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 2225}
2226
0a8165d7 2227/*
351df4b2
JK		 2228 * Calculate the number of current summary pages for writing
2229 */
4d57b86d 2230int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 2231{
351df4b2 2232 int valid_sum_count = 0;
9a47938b 2233 int i, sum_in_page;
351df4b2
JK		 2234
2235 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2236 if (sbi->ckpt->alloc_type[i] == SSR)
2237 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 2238 else {
2239 if (for_ra)
2240 valid_sum_count += le16_to_cpu(
2241 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2242 else
2243 valid_sum_count += curseg_blkoff(sbi, i);
2244 }
351df4b2
JK		 2245 }
2246
09cbfeaf 2247 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 2248 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2249 if (valid_sum_count <= sum_in_page)
351df4b2 2250 return 1;
9a47938b 2251 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 2252 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 2253 return 2;
2254 return 3;
2255}
2256
0a8165d7 2257/*
351df4b2
JK		 2258 * Caller should put this summary page
2259 */
4d57b86d 2260struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 2261{
7735730d 2262 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2263}
2264
4d57b86d
CY		 2265void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2266 void *src, block_t blk_addr)
351df4b2 2267{
4d57b86d 2268 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 2269
0537b811 2270 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2271 set_page_dirty(page);
2272 f2fs_put_page(page, 1);
2273}
2274
381722d2
CY		 2275static void write_sum_page(struct f2fs_sb_info *sbi,
2276 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2277{
4d57b86d 2278 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2279}
2280
b7ad7512
CY		 2281static void write_current_sum_page(struct f2fs_sb_info *sbi,
2282 int type, block_t blk_addr)
2283{
2284 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2285 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2286 struct f2fs_summary_block *src = curseg->sum_blk;
2287 struct f2fs_summary_block *dst;
2288
2289 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2290 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2291
2292 mutex_lock(&curseg->curseg_mutex);
2293
2294 down_read(&curseg->journal_rwsem);
2295 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2296 up_read(&curseg->journal_rwsem);
2297
2298 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2299 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2300
2301 mutex_unlock(&curseg->curseg_mutex);
2302
2303 set_page_dirty(page);
2304 f2fs_put_page(page, 1);
2305}
2306
a7881893
JK		 2307static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2308{
2309 struct curseg_info *curseg = CURSEG_I(sbi, type);
2310 unsigned int segno = curseg->segno + 1;
2311 struct free_segmap_info *free_i = FREE_I(sbi);
2312
2313 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2314 return !test_bit(segno, free_i->free_segmap);
2315 return 0;
2316}
2317
0a8165d7 2318/*
351df4b2
JK		 2319 * Find a new segment from the free segments bitmap to right order
2320 * This function should be returned with success, otherwise BUG
2321 */
2322static void get_new_segment(struct f2fs_sb_info *sbi,
2323 unsigned int *newseg, bool new_sec, int dir)
2324{
2325 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2326 unsigned int segno, secno, zoneno;
7cd8558b 2327 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2328 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2329 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2330 unsigned int left_start = hint;
2331 bool init = true;
2332 int go_left = 0;
2333 int i;
2334
1a118ccf 2335 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2336
2337 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2338 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2339 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2340 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2341 goto got_it;
2342 }
2343find_other_zone:
7cd8558b
JK		 2344 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2345 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2346 if (dir == ALLOC_RIGHT) {
2347 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2348 MAIN_SECS(sbi), 0);
2349 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2350 } else {
2351 go_left = 1;
2352 left_start = hint - 1;
2353 }
2354 }
2355 if (go_left == 0)
2356 goto skip_left;
2357
2358 while (test_bit(left_start, free_i->free_secmap)) {
2359 if (left_start > 0) {
2360 left_start--;
2361 continue;
2362 }
2363 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2364 MAIN_SECS(sbi), 0);
2365 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2366 break;
2367 }
2368 secno = left_start;
2369skip_left:
4ddb1a4d
JK		 2370 segno = GET_SEG_FROM_SEC(sbi, secno);
2371 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2372
2373 /* give up on finding another zone */
2374 if (!init)
2375 goto got_it;
2376 if (sbi->secs_per_zone == 1)
2377 goto got_it;
2378 if (zoneno == old_zoneno)
2379 goto got_it;
2380 if (dir == ALLOC_LEFT) {
2381 if (!go_left && zoneno + 1 >= total_zones)
2382 goto got_it;
2383 if (go_left && zoneno == 0)
2384 goto got_it;
2385 }
2386 for (i = 0; i < NR_CURSEG_TYPE; i++)
2387 if (CURSEG_I(sbi, i)->zone == zoneno)
2388 break;
2389
2390 if (i < NR_CURSEG_TYPE) {
2391 /* zone is in user, try another */
2392 if (go_left)
2393 hint = zoneno * sbi->secs_per_zone - 1;
2394 else if (zoneno + 1 >= total_zones)
2395 hint = 0;
2396 else
2397 hint = (zoneno + 1) * sbi->secs_per_zone;
2398 init = false;
2399 goto find_other_zone;
2400 }
2401got_it:
2402 /* set it as dirty segment in free segmap */
9850cf4a 2403 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2404 __set_inuse(sbi, segno);
2405 *newseg = segno;
1a118ccf 2406 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2407}
2408
2409static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2410{
2411 struct curseg_info *curseg = CURSEG_I(sbi, type);
2412 struct summary_footer *sum_footer;
2413
2414 curseg->segno = curseg->next_segno;
4ddb1a4d 2415 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2416 curseg->next_blkoff = 0;
2417 curseg->next_segno = NULL_SEGNO;
2418
2419 sum_footer = &(curseg->sum_blk->footer);
2420 memset(sum_footer, 0, sizeof(struct summary_footer));
2421 if (IS_DATASEG(type))
2422 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2423 if (IS_NODESEG(type))
2424 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2425 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2426}
2427
7a20b8a6
JK		 2428static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2429{
a7881893 2430 /* if segs_per_sec is large than 1, we need to keep original policy. */
2c70c5e3 2431 if (__is_large_section(sbi))
a7881893
JK		 2432 return CURSEG_I(sbi, type)->segno;
2433
4354994f
DR		 2434 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2435 return 0;
2436
b94929d9
YS		 2437 if (test_opt(sbi, NOHEAP) &&
2438 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2439 return 0;
2440
e066b83c
JK		 2441 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2442 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2443
2444 /* find segments from 0 to reuse freed segments */
63189b78 2445 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2446 return 0;
2447
7a20b8a6
JK		 2448 return CURSEG_I(sbi, type)->segno;
2449}
2450
0a8165d7 2451/*
351df4b2
JK		 2452 * Allocate a current working segment.
2453 * This function always allocates a free segment in LFS manner.
2454 */
2455static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2456{
2457 struct curseg_info *curseg = CURSEG_I(sbi, type);
2458 unsigned int segno = curseg->segno;
2459 int dir = ALLOC_LEFT;
2460
2461 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2462 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2463 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2464 dir = ALLOC_RIGHT;
2465
2466 if (test_opt(sbi, NOHEAP))
2467 dir = ALLOC_RIGHT;
2468
7a20b8a6 2469 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2470 get_new_segment(sbi, &segno, new_sec, dir);
2471 curseg->next_segno = segno;
2472 reset_curseg(sbi, type, 1);
2473 curseg->alloc_type = LFS;
2474}
2475
2476static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2477 struct curseg_info *seg, block_t start)
2478{
2479 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2480 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2481 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2482 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2483 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2484 int i, pos;
2485
2486 for (i = 0; i < entries; i++)
2487 target_map[i] = ckpt_map[i] | cur_map[i];
2488
2489 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2490
2491 seg->next_blkoff = pos;
351df4b2
JK		 2492}
2493
0a8165d7 2494/*
351df4b2
JK		 2495 * If a segment is written by LFS manner, next block offset is just obtained
2496 * by increasing the current block offset. However, if a segment is written by
2497 * SSR manner, next block offset obtained by calling __next_free_blkoff
2498 */
2499static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2500 struct curseg_info *seg)
2501{
2502 if (seg->alloc_type == SSR)
2503 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2504 else
2505 seg->next_blkoff++;
2506}
2507
0a8165d7 2508/*
e1c42045 2509 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2510 * manner, so it should recover the existing segment information of valid blocks
2511 */
025d63a4 2512static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2513{
2514 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2515 struct curseg_info *curseg = CURSEG_I(sbi, type);
2516 unsigned int new_segno = curseg->next_segno;
2517 struct f2fs_summary_block *sum_node;
2518 struct page *sum_page;
2519
2520 write_sum_page(sbi, curseg->sum_blk,
2521 GET_SUM_BLOCK(sbi, curseg->segno));
2522 __set_test_and_inuse(sbi, new_segno);
2523
2524 mutex_lock(&dirty_i->seglist_lock);
2525 __remove_dirty_segment(sbi, new_segno, PRE);
2526 __remove_dirty_segment(sbi, new_segno, DIRTY);
2527 mutex_unlock(&dirty_i->seglist_lock);
2528
2529 reset_curseg(sbi, type, 1);
2530 curseg->alloc_type = SSR;
2531 __next_free_blkoff(sbi, curseg, 0);
2532
4d57b86d 2533 sum_page = f2fs_get_sum_page(sbi, new_segno);
edc55aaf 2534 f2fs_bug_on(sbi, IS_ERR(sum_page));
025d63a4
CY		 2535 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2536 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2537 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2538}
2539
43727527
JK		 2540static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2541{
2542 struct curseg_info *curseg = CURSEG_I(sbi, type);
2543 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2544 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2545 int i, cnt;
2546 bool reversed = false;
c192f7a4 2547
4d57b86d 2548 /* f2fs_need_SSR() already forces to do this */
e066b83c
JK		 2549 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2550 curseg->next_segno = segno;
c192f7a4 2551 return 1;
e066b83c 2552 }
43727527 2553
70d625cb
JK		 2554 /* For node segments, let's do SSR more intensively */
2555 if (IS_NODESEG(type)) {
d27c3d89
CY		 2556 if (type >= CURSEG_WARM_NODE) {
2557 reversed = true;
2558 i = CURSEG_COLD_NODE;
2559 } else {
2560 i = CURSEG_HOT_NODE;
2561 }
2562 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2563 } else {
d27c3d89
CY		 2564 if (type >= CURSEG_WARM_DATA) {
2565 reversed = true;
2566 i = CURSEG_COLD_DATA;
2567 } else {
2568 i = CURSEG_HOT_DATA;
2569 }
2570 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2571 }
43727527 2572
d27c3d89 2573 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2574 if (i == type)
2575 continue;
e066b83c
JK		 2576 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2577 curseg->next_segno = segno;
43727527 2578 return 1;
e066b83c 2579 }
c192f7a4 2580 }
4354994f
DR		 2581
2582 /* find valid_blocks=0 in dirty list */
2583 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2584 segno = get_free_segment(sbi);
2585 if (segno != NULL_SEGNO) {
2586 curseg->next_segno = segno;
2587 return 1;
2588 }
2589 }
43727527
JK		 2590 return 0;
2591}
2592
351df4b2
JK		 2593/*
2594 * flush out current segment and replace it with new segment
2595 * This function should be returned with success, otherwise BUG
2596 */
2597static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2598 int type, bool force)
2599{
a7881893
JK		 2600 struct curseg_info *curseg = CURSEG_I(sbi, type);
2601
7b405275 2602 if (force)
351df4b2 2603 new_curseg(sbi, type, true);
5b6c6be2
JK		 2604 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2605 type == CURSEG_WARM_NODE)
351df4b2 2606 new_curseg(sbi, type, false);
4354994f
DR		 2607 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2608 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
a7881893 2609 new_curseg(sbi, type, false);
4d57b86d 2610 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2611 change_curseg(sbi, type);
351df4b2
JK		 2612 else
2613 new_curseg(sbi, type, false);
dcdfff65 2614
a7881893 2615 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2616}
2617
4d57b86d 2618void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
351df4b2 2619{
6ae1be13
JK		 2620 struct curseg_info *curseg;
2621 unsigned int old_segno;
351df4b2
JK		 2622 int i;
2623
3d26fa6b
CY		 2624 down_write(&SIT_I(sbi)->sentry_lock);
2625
6ae1be13
JK		 2626 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2627 curseg = CURSEG_I(sbi, i);
2628 old_segno = curseg->segno;
2629 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2630 locate_dirty_segment(sbi, old_segno);
2631 }
3d26fa6b
CY		 2632
2633 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2634}
2635
2636static const struct segment_allocation default_salloc_ops = {
2637 .allocate_segment = allocate_segment_by_default,
2638};
2639
4d57b86d
CY		 2640bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2641 struct cp_control *cpc)
25290fa5
JK		 2642{
2643 __u64 trim_start = cpc->trim_start;
2644 bool has_candidate = false;
2645
3d26fa6b 2646 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2647 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2648 if (add_discard_addrs(sbi, cpc, true)) {
2649 has_candidate = true;
2650 break;
2651 }
2652 }
3d26fa6b 2653 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2654
2655 cpc->trim_start = trim_start;
2656 return has_candidate;
2657}
2658
01f9cf6d 2659static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 2660 struct discard_policy *dpolicy,
2661 unsigned int start, unsigned int end)
2662{
2663 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2664 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2665 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2666 struct discard_cmd *dc;
2667 struct blk_plug plug;
2668 int issued;
01f9cf6d 2669 unsigned int trimmed = 0;
9a997188
JK		 2670
2671next:
2672 issued = 0;
2673
2674 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 2675 if (unlikely(dcc->rbtree_check))
2676 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2677 &dcc->root));
9a997188 2678
4d57b86d 2679 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 2680 NULL, start,
2681 (struct rb_entry **)&prev_dc,
2682 (struct rb_entry **)&next_dc,
4dada3fd 2683 &insert_p, &insert_parent, true, NULL);
9a997188
JK		 2684 if (!dc)
2685 dc = next_dc;
2686
2687 blk_start_plug(&plug);
2688
2689 while (dc && dc->lstart <= end) {
2690 struct rb_node *node;
6b9cb124 2691 int err = 0;
9a997188
JK		 2692
2693 if (dc->len < dpolicy->granularity)
2694 goto skip;
2695
2696 if (dc->state != D_PREP) {
2697 list_move_tail(&dc->list, &dcc->fstrim_list);
2698 goto skip;
2699 }
2700
6b9cb124 2701 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
9a997188 2702
35ec7d57 2703 if (issued >= dpolicy->max_requests) {
9a997188
JK		 2704 start = dc->lstart + dc->len;
2705
6b9cb124
CY		 2706 if (err)
2707 __remove_discard_cmd(sbi, dc);
2708
9a997188
JK		 2709 blk_finish_plug(&plug);
2710 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 2711 trimmed += __wait_all_discard_cmd(sbi, NULL);
9a997188
JK		 2712 congestion_wait(BLK_RW_ASYNC, HZ/50);
2713 goto next;
2714 }
2715skip:
2716 node = rb_next(&dc->rb_node);
6b9cb124
CY		 2717 if (err)
2718 __remove_discard_cmd(sbi, dc);
9a997188
JK		 2719 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2720
2721 if (fatal_signal_pending(current))
2722 break;
2723 }
2724
2725 blk_finish_plug(&plug);
2726 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 2727
2728 return trimmed;
9a997188
JK		 2729}
2730
4b2fecc8
JK		 2731int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2732{
f7ef9b83
JK		 2733 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2734 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2735 unsigned int start_segno, end_segno;
8412663d 2736 block_t start_block, end_block;
4b2fecc8 2737 struct cp_control cpc;
78997b56 2738 struct discard_policy dpolicy;
0ea80512 2739 unsigned long long trimmed = 0;
c34f42e2 2740 int err = 0;
2c70c5e3 2741 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
4b2fecc8 2742
836b5a63 2743 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2744 return -EINVAL;
2745
3f16ecd9
CY		 2746 if (end < MAIN_BLKADDR(sbi))
2747 goto out;
4b2fecc8 2748
ed214a11
YH		 2749 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2750 f2fs_msg(sbi->sb, KERN_WARNING,
2751 "Found FS corruption, run fsck to fix.");
3d165dc3 2752 return -EIO;
ed214a11
YH		 2753 }
2754
4b2fecc8 2755 /* start/end segment number in main_area */
7cd8558b
JK		 2756 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2757 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2758 GET_SEGNO(sbi, end);
ad6672bb
YS		 2759 if (need_align) {
2760 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2761 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2762 }
8412663d 2763
4b2fecc8 2764 cpc.reason = CP_DISCARD;
836b5a63 2765 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2766 cpc.trim_start = start_segno;
2767 cpc.trim_end = end_segno;
4b2fecc8 2768
377224c4
CY		 2769 if (sbi->discard_blks == 0)
2770 goto out;
74fa5f3d 2771
377224c4 2772 mutex_lock(&sbi->gc_mutex);
4d57b86d 2773 err = f2fs_write_checkpoint(sbi, &cpc);
377224c4
CY		 2774 mutex_unlock(&sbi->gc_mutex);
2775 if (err)
2776 goto out;
8412663d 2777
e555da9f
JK		 2778 /*
2779 * We filed discard candidates, but actually we don't need to wait for
2780 * all of them, since they'll be issued in idle time along with runtime
2781 * discard option. User configuration looks like using runtime discard
2782 * or periodic fstrim instead of it.
2783 */
7d20c8ab 2784 if (f2fs_realtime_discard_enable(sbi))
5a615492
JK		 2785 goto out;
2786
2787 start_block = START_BLOCK(sbi, start_segno);
2788 end_block = START_BLOCK(sbi, end_segno + 1);
2789
2790 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 2791 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2792 start_block, end_block);
5a615492 2793
01f9cf6d 2794 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2795 start_block, end_block);
377224c4 2796out:
6eae2694
CY		 2797 if (!err)
2798 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2799 return err;
4b2fecc8
JK		 2800}
2801
351df4b2
JK		 2802static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2803{
2804 struct curseg_info *curseg = CURSEG_I(sbi, type);
2805 if (curseg->next_blkoff < sbi->blocks_per_seg)
2806 return true;
2807 return false;
2808}
2809
4d57b86d 2810int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 2811{
2812 switch (hint) {
2813 case WRITE_LIFE_SHORT:
2814 return CURSEG_HOT_DATA;
2815 case WRITE_LIFE_EXTREME:
2816 return CURSEG_COLD_DATA;
2817 default:
2818 return CURSEG_WARM_DATA;
2819 }
2820}
2821
0cdd3195
HL		 2822/* This returns write hints for each segment type. This hints will be
2823 * passed down to block layer. There are mapping tables which depend on
2824 * the mount option 'whint_mode'.
2825 *
2826 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2827 *
2828 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2829 *
2830 * User F2FS Block
2831 * ---- ---- -----
2832 * META WRITE_LIFE_NOT_SET
2833 * HOT_NODE "
2834 * WARM_NODE "
2835 * COLD_NODE "
2836 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2837 * extension list " "
2838 *
2839 * -- buffered io
2840 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2841 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2842 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2843 * WRITE_LIFE_NONE " "
2844 * WRITE_LIFE_MEDIUM " "
2845 * WRITE_LIFE_LONG " "
2846 *
2847 * -- direct io
2848 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2849 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2850 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2851 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2852 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2853 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2854 *
f2e703f9
HL		 2855 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2856 *
2857 * User F2FS Block
2858 * ---- ---- -----
2859 * META WRITE_LIFE_MEDIUM;
2860 * HOT_NODE WRITE_LIFE_NOT_SET
2861 * WARM_NODE "
2862 * COLD_NODE WRITE_LIFE_NONE
2863 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2864 * extension list " "
2865 *
2866 * -- buffered io
2867 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2868 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2869 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2870 * WRITE_LIFE_NONE " "
2871 * WRITE_LIFE_MEDIUM " "
2872 * WRITE_LIFE_LONG " "
2873 *
2874 * -- direct io
2875 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2876 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2877 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2878 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2879 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2880 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2881 */
2882
4d57b86d 2883enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 2884 enum page_type type, enum temp_type temp)
2885{
63189b78 2886 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2887 if (type == DATA) {
f2e703f9 2888 if (temp == WARM)
0cdd3195 2889 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2890 else if (temp == HOT)
2891 return WRITE_LIFE_SHORT;
2892 else if (temp == COLD)
2893 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2894 } else {
2895 return WRITE_LIFE_NOT_SET;
2896 }
63189b78 2897 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2898 if (type == DATA) {
2899 if (temp == WARM)
2900 return WRITE_LIFE_LONG;
2901 else if (temp == HOT)
2902 return WRITE_LIFE_SHORT;
2903 else if (temp == COLD)
2904 return WRITE_LIFE_EXTREME;
2905 } else if (type == NODE) {
2906 if (temp == WARM || temp == HOT)
2907 return WRITE_LIFE_NOT_SET;
2908 else if (temp == COLD)
2909 return WRITE_LIFE_NONE;
2910 } else if (type == META) {
2911 return WRITE_LIFE_MEDIUM;
2912 }
0cdd3195 2913 }
f2e703f9 2914 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2915}
2916
81377bd6 2917static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2918{
81377bd6 2919 if (fio->type == DATA)
351df4b2
JK		 2920 return CURSEG_HOT_DATA;
2921 else
2922 return CURSEG_HOT_NODE;
2923}
2924
81377bd6 2925static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2926{
81377bd6
JK		 2927 if (fio->type == DATA) {
2928 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2929
2930 if (S_ISDIR(inode->i_mode))
2931 return CURSEG_HOT_DATA;
2932 else
2933 return CURSEG_COLD_DATA;
2934 } else {
81377bd6 2935 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2936 return CURSEG_WARM_NODE;
351df4b2
JK		 2937 else
2938 return CURSEG_COLD_NODE;
2939 }
2940}
2941
81377bd6 2942static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2943{
81377bd6
JK		 2944 if (fio->type == DATA) {
2945 struct inode *inode = fio->page->mapping->host;
351df4b2 2946
81377bd6 2947 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2948 return CURSEG_COLD_DATA;
b6a06cbb 2949 if (file_is_hot(inode) ||
b4c3ca8b 2950 is_inode_flag_set(inode, FI_HOT_DATA) ||
2079f115
CY		 2951 f2fs_is_atomic_file(inode) ||
2952 f2fs_is_volatile_file(inode))
ef095d19 2953 return CURSEG_HOT_DATA;
4d57b86d 2954 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2955 } else {
81377bd6
JK		 2956 if (IS_DNODE(fio->page))
2957 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2958 CURSEG_HOT_NODE;
ef095d19 2959 return CURSEG_COLD_NODE;
351df4b2
JK		 2960 }
2961}
2962
81377bd6 2963static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2964{
a912b54d
JK		 2965 int type = 0;
2966
63189b78 2967 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2968 case 2:
a912b54d
JK		 2969 type = __get_segment_type_2(fio);
2970 break;
351df4b2 2971 case 4:
a912b54d
JK		 2972 type = __get_segment_type_4(fio);
2973 break;
2974 case 6:
2975 type = __get_segment_type_6(fio);
2976 break;
2977 default:
2978 f2fs_bug_on(fio->sbi, true);
351df4b2 2979 }
81377bd6 2980
a912b54d
JK		 2981 if (IS_HOT(type))
2982 fio->temp = HOT;
2983 else if (IS_WARM(type))
2984 fio->temp = WARM;
2985 else
2986 fio->temp = COLD;
2987 return type;
351df4b2
JK		 2988}
2989
4d57b86d 2990void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 2991 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2992 struct f2fs_summary *sum, int type,
2993 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2994{
2995 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2996 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2997
2b60311d
CY		 2998 down_read(&SM_I(sbi)->curseg_lock);
2999
351df4b2 3000 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3001 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3002
3003 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 3004
4e6a8d9b
JK		 3005 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3006
351df4b2
JK		 3007 /*
3008 * __add_sum_entry should be resided under the curseg_mutex
3009 * because, this function updates a summary entry in the
3010 * current summary block.
3011 */
e79efe3b 3012 __add_sum_entry(sbi, type, sum);
351df4b2 3013
351df4b2 3014 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 3015
3016 stat_inc_block_count(sbi, curseg);
351df4b2 3017
65f1b80b
YS		 3018 /*
3019 * SIT information should be updated before segment allocation,
3020 * since SSR needs latest valid block information.
3021 */
3022 update_sit_entry(sbi, *new_blkaddr, 1);
3023 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3024 update_sit_entry(sbi, old_blkaddr, -1);
3025
c6f82fe9
JK		 3026 if (!__has_curseg_space(sbi, type))
3027 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 3028
351df4b2 3029 /*
65f1b80b
YS		 3030 * segment dirty status should be updated after segment allocation,
3031 * so we just need to update status only one time after previous
3032 * segment being closed.
351df4b2 3033 */
65f1b80b
YS		 3034 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3035 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 3036
3d26fa6b 3037 up_write(&sit_i->sentry_lock);
351df4b2 3038
704956ec 3039 if (page && IS_NODESEG(type)) {
351df4b2
JK		 3040 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3041
704956ec
CY		 3042 f2fs_inode_chksum_set(sbi, page);
3043 }
3044
fb830fc5
CY		 3045 if (add_list) {
3046 struct f2fs_bio_info *io;
3047
3048 INIT_LIST_HEAD(&fio->list);
3049 fio->in_list = true;
fe16efe6 3050 fio->retry = false;
fb830fc5
CY		 3051 io = sbi->write_io[fio->type] + fio->temp;
3052 spin_lock(&io->io_lock);
3053 list_add_tail(&fio->list, &io->io_list);
3054 spin_unlock(&io->io_lock);
3055 }
3056
bfad7c2d 3057 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 3058
3059 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 3060}
3061
39d787be
CY		 3062static void update_device_state(struct f2fs_io_info *fio)
3063{
3064 struct f2fs_sb_info *sbi = fio->sbi;
3065 unsigned int devidx;
3066
3067 if (!sbi->s_ndevs)
3068 return;
3069
3070 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3071
3072 /* update device state for fsync */
4d57b86d 3073 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 3074
3075 /* update device state for checkpoint */
3076 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3077 spin_lock(&sbi->dev_lock);
3078 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3079 spin_unlock(&sbi->dev_lock);
3080 }
39d787be
CY		 3081}
3082
05ca3632 3083static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 3084{
81377bd6 3085 int type = __get_segment_type(fio);
107a805d 3086 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 3087
107a805d
CY		 3088 if (keep_order)
3089 down_read(&fio->sbi->io_order_lock);
0a595eba 3090reallocate:
4d57b86d 3091 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 3092 &fio->new_blkaddr, sum, type, fio, true);
6aa58d8a
CY		 3093 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3094 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3095 fio->old_blkaddr, fio->old_blkaddr);
bfad7c2d 3096
351df4b2 3097 /* writeout dirty page into bdev */
fe16efe6
CY		 3098 f2fs_submit_page_write(fio);
3099 if (fio->retry) {
0a595eba
JK		 3100 fio->old_blkaddr = fio->new_blkaddr;
3101 goto reallocate;
3102 }
fe16efe6
CY		 3103
3104 update_device_state(fio);
3105
107a805d
CY		 3106 if (keep_order)
3107 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 3108}
3109
4d57b86d 3110void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 3111 enum iostat_type io_type)
351df4b2 3112{
458e6197 3113 struct f2fs_io_info fio = {
05ca3632 3114 .sbi = sbi,
458e6197 3115 .type = META,
0cdd3195 3116 .temp = HOT,
04d328de 3117 .op = REQ_OP_WRITE,
70fd7614 3118 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 3119 .old_blkaddr = page->index,
3120 .new_blkaddr = page->index,
05ca3632 3121 .page = page,
4375a336 3122 .encrypted_page = NULL,
fb830fc5 3123 .in_list = false,
458e6197
JK		 3124 };
3125
2b947003 3126 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 3127 fio.op_flags &= ~REQ_META;
2b947003 3128
351df4b2 3129 set_page_writeback(page);
17c50035 3130 ClearPageError(page);
b9109b0e 3131 f2fs_submit_page_write(&fio);
b0af6d49 3132
b63e7be5 3133 stat_inc_meta_count(sbi, page->index);
b0af6d49 3134 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 3135}
3136
4d57b86d 3137void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 3138{
3139 struct f2fs_summary sum;
05ca3632 3140
351df4b2 3141 set_summary(&sum, nid, 0, 0);
05ca3632 3142 do_write_page(&sum, fio);
b0af6d49
CY		 3143
3144 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3145}
3146
4d57b86d
CY		 3147void f2fs_outplace_write_data(struct dnode_of_data *dn,
3148 struct f2fs_io_info *fio)
351df4b2 3149{
05ca3632 3150 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2 3151 struct f2fs_summary sum;
351df4b2 3152
9850cf4a 3153 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
7735730d 3154 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
05ca3632 3155 do_write_page(&sum, fio);
f28b3434 3156 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 3157
3158 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3159}
3160
4d57b86d 3161int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 3162{
b0af6d49 3163 int err;
d21b0f23 3164 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 3165
7a9d7548 3166 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 3167 /* i/o temperature is needed for passing down write hints */
3168 __get_segment_type(fio);
d21b0f23
YH		 3169
3170 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3171 GET_SEGNO(sbi, fio->new_blkaddr))->type));
3172
05ca3632 3173 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 3174
3175 err = f2fs_submit_page_bio(fio);
39d787be
CY		 3176 if (!err)
3177 update_device_state(fio);
b0af6d49
CY		 3178
3179 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3180
3181 return err;
351df4b2
JK		 3182}
3183
2b60311d
CY		 3184static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3185 unsigned int segno)
3186{
3187 int i;
3188
3189 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3190 if (CURSEG_I(sbi, i)->segno == segno)
3191 break;
3192 }
3193 return i;
3194}
3195
4d57b86d 3196void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 3197 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 3198 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 3199{
3200 struct sit_info *sit_i = SIT_I(sbi);
3201 struct curseg_info *curseg;
3202 unsigned int segno, old_cursegno;
3203 struct seg_entry *se;
3204 int type;
19f106bc 3205 unsigned short old_blkoff;
351df4b2
JK		 3206
3207 segno = GET_SEGNO(sbi, new_blkaddr);
3208 se = get_seg_entry(sbi, segno);
3209 type = se->type;
3210
2b60311d
CY		 3211 down_write(&SM_I(sbi)->curseg_lock);
3212
19f106bc
CY		 3213 if (!recover_curseg) {
3214 /* for recovery flow */
3215 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3216 if (old_blkaddr == NULL_ADDR)
3217 type = CURSEG_COLD_DATA;
3218 else
3219 type = CURSEG_WARM_DATA;
3220 }
3221 } else {
2b60311d
CY		 3222 if (IS_CURSEG(sbi, segno)) {
3223 /* se->type is volatile as SSR allocation */
3224 type = __f2fs_get_curseg(sbi, segno);
3225 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3226 } else {
351df4b2 3227 type = CURSEG_WARM_DATA;
2b60311d 3228 }
351df4b2 3229 }
19f106bc 3230
2c190504 3231 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 3232 curseg = CURSEG_I(sbi, type);
3233
3234 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3235 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3236
3237 old_cursegno = curseg->segno;
19f106bc 3238 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 3239
3240 /* change the current segment */
3241 if (segno != curseg->segno) {
3242 curseg->next_segno = segno;
025d63a4 3243 change_curseg(sbi, type);
351df4b2
JK		 3244 }
3245
491c0854 3246 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 3247 __add_sum_entry(sbi, type, sum);
351df4b2 3248
28bc106b 3249 if (!recover_curseg || recover_newaddr)
6e2c64ad 3250 update_sit_entry(sbi, new_blkaddr, 1);
6aa58d8a
CY		 3251 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3252 invalidate_mapping_pages(META_MAPPING(sbi),
3253 old_blkaddr, old_blkaddr);
6e2c64ad 3254 update_sit_entry(sbi, old_blkaddr, -1);
6aa58d8a 3255 }
6e2c64ad
JK		 3256
3257 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3258 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3259
351df4b2 3260 locate_dirty_segment(sbi, old_cursegno);
351df4b2 3261
19f106bc
CY		 3262 if (recover_curseg) {
3263 if (old_cursegno != curseg->segno) {
3264 curseg->next_segno = old_cursegno;
025d63a4 3265 change_curseg(sbi, type);
19f106bc
CY		 3266 }
3267 curseg->next_blkoff = old_blkoff;
3268 }
3269
3d26fa6b 3270 up_write(&sit_i->sentry_lock);
351df4b2 3271 mutex_unlock(&curseg->curseg_mutex);
2b60311d 3272 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 3273}
3274
528e3459
CY		 3275void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3276 block_t old_addr, block_t new_addr,
28bc106b
CY		 3277 unsigned char version, bool recover_curseg,
3278 bool recover_newaddr)
528e3459
CY		 3279{
3280 struct f2fs_summary sum;
3281
3282 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3283
4d57b86d 3284 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
28bc106b 3285 recover_curseg, recover_newaddr);
528e3459 3286
f28b3434 3287 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 3288}
3289
93dfe2ac 3290void f2fs_wait_on_page_writeback(struct page *page,
bae0ee7a 3291 enum page_type type, bool ordered, bool locked)
93dfe2ac 3292{
93dfe2ac 3293 if (PageWriteback(page)) {
4081363f
JK		 3294 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3295
bab475c5 3296 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
bae0ee7a 3297 if (ordered) {
fec1d657 3298 wait_on_page_writeback(page);
bae0ee7a
CY		 3299 f2fs_bug_on(sbi, locked && PageWriteback(page));
3300 } else {
fec1d657 3301 wait_for_stable_page(page);
bae0ee7a 3302 }
93dfe2ac
JK		 3303 }
3304}
3305
0ded69f6 3306void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
08b39fbd 3307{
0ded69f6 3308 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
08b39fbd
CY		 3309 struct page *cpage;
3310
0ded69f6
JK		 3311 if (!f2fs_post_read_required(inode))
3312 return;
3313
e1da7872 3314 if (!is_valid_data_blkaddr(sbi, blkaddr))
08b39fbd
CY		 3315 return;
3316
08b39fbd
CY		 3317 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3318 if (cpage) {
bae0ee7a 3319 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
08b39fbd
CY		 3320 f2fs_put_page(cpage, 1);
3321 }
3322}
3323
1e78e8bd
ST		 3324void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3325 block_t len)
3326{
3327 block_t i;
3328
3329 for (i = 0; i < len; i++)
3330 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3331}
3332
7735730d 3333static int read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3334{
3335 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3336 struct curseg_info *seg_i;
3337 unsigned char *kaddr;
3338 struct page *page;
3339 block_t start;
3340 int i, j, offset;
3341
3342 start = start_sum_block(sbi);
3343
4d57b86d 3344 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3345 if (IS_ERR(page))
3346 return PTR_ERR(page);
351df4b2
JK		 3347 kaddr = (unsigned char *)page_address(page);
3348
3349 /* Step 1: restore nat cache */
3350 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3351 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3352
3353 /* Step 2: restore sit cache */
3354 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3355 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3356 offset = 2 * SUM_JOURNAL_SIZE;
3357
3358 /* Step 3: restore summary entries */
3359 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3360 unsigned short blk_off;
3361 unsigned int segno;
3362
3363 seg_i = CURSEG_I(sbi, i);
3364 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3365 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3366 seg_i->next_segno = segno;
3367 reset_curseg(sbi, i, 0);
3368 seg_i->alloc_type = ckpt->alloc_type[i];
3369 seg_i->next_blkoff = blk_off;
3370
3371 if (seg_i->alloc_type == SSR)
3372 blk_off = sbi->blocks_per_seg;
3373
3374 for (j = 0; j < blk_off; j++) {
3375 struct f2fs_summary *s;
3376 s = (struct f2fs_summary *)(kaddr + offset);
3377 seg_i->sum_blk->entries[j] = *s;
3378 offset += SUMMARY_SIZE;
09cbfeaf 3379 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3380 SUM_FOOTER_SIZE)
3381 continue;
3382
3383 f2fs_put_page(page, 1);
3384 page = NULL;
3385
4d57b86d 3386 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3387 if (IS_ERR(page))
3388 return PTR_ERR(page);
351df4b2
JK		 3389 kaddr = (unsigned char *)page_address(page);
3390 offset = 0;
3391 }
3392 }
3393 f2fs_put_page(page, 1);
7735730d 3394 return 0;
351df4b2
JK		 3395}
3396
3397static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3398{
3399 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3400 struct f2fs_summary_block *sum;
3401 struct curseg_info *curseg;
3402 struct page *new;
3403 unsigned short blk_off;
3404 unsigned int segno = 0;
3405 block_t blk_addr = 0;
7735730d 3406 int err = 0;
351df4b2
JK		 3407
3408 /* get segment number and block addr */
3409 if (IS_DATASEG(type)) {
3410 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3411 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3412 CURSEG_HOT_DATA]);
119ee914 3413 if (__exist_node_summaries(sbi))
351df4b2
JK		 3414 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3415 else
3416 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3417 } else {
3418 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3419 CURSEG_HOT_NODE]);
3420 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3421 CURSEG_HOT_NODE]);
119ee914 3422 if (__exist_node_summaries(sbi))
351df4b2
JK		 3423 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3424 type - CURSEG_HOT_NODE);
3425 else
3426 blk_addr = GET_SUM_BLOCK(sbi, segno);
3427 }
3428
4d57b86d 3429 new = f2fs_get_meta_page(sbi, blk_addr);
7735730d
CY		 3430 if (IS_ERR(new))
3431 return PTR_ERR(new);
351df4b2
JK		 3432 sum = (struct f2fs_summary_block *)page_address(new);
3433
3434 if (IS_NODESEG(type)) {
119ee914 3435 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3436 struct f2fs_summary *ns = &sum->entries[0];
3437 int i;
3438 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3439 ns->version = 0;
3440 ns->ofs_in_node = 0;
3441 }
3442 } else {
7735730d
CY		 3443 err = f2fs_restore_node_summary(sbi, segno, sum);
3444 if (err)
3445 goto out;
351df4b2
JK		 3446 }
3447 }
3448
3449 /* set uncompleted segment to curseg */
3450 curseg = CURSEG_I(sbi, type);
3451 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3452
3453 /* update journal info */
3454 down_write(&curseg->journal_rwsem);
3455 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3456 up_write(&curseg->journal_rwsem);
3457
3458 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3459 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3460 curseg->next_segno = segno;
3461 reset_curseg(sbi, type, 0);
3462 curseg->alloc_type = ckpt->alloc_type[type];
3463 curseg->next_blkoff = blk_off;
3464 mutex_unlock(&curseg->curseg_mutex);
7735730d 3465out:
351df4b2 3466 f2fs_put_page(new, 1);
7735730d 3467 return err;
351df4b2
JK		 3468}
3469
3470static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3471{
21d3f8e1
JQ		 3472 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3473 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3474 int type = CURSEG_HOT_DATA;
e4fc5fbf 3475 int err;
351df4b2 3476
aaec2b1d 3477 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3478 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3479
3480 if (npages >= 2)
4d57b86d 3481 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3482 META_CP, true);
3fa06d7b 3483
351df4b2 3484 /* restore for compacted data summary */
7735730d
CY		 3485 err = read_compacted_summaries(sbi);
3486 if (err)
3487 return err;
351df4b2
JK		 3488 type = CURSEG_HOT_NODE;
3489 }
3490
119ee914 3491 if (__exist_node_summaries(sbi))
4d57b86d 3492 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3493 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3494
e4fc5fbf
CY		 3495 for (; type <= CURSEG_COLD_NODE; type++) {
3496 err = read_normal_summaries(sbi, type);
3497 if (err)
3498 return err;
3499 }
3500
21d3f8e1
JQ		 3501 /* sanity check for summary blocks */
3502 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3503 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3504 return -EINVAL;
3505
351df4b2
JK		 3506 return 0;
3507}
3508
3509static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3510{
3511 struct page *page;
3512 unsigned char *kaddr;
3513 struct f2fs_summary *summary;
3514 struct curseg_info *seg_i;
3515 int written_size = 0;
3516 int i, j;
3517
4d57b86d 3518 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3519 kaddr = (unsigned char *)page_address(page);
81114baa 3520 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3521
3522 /* Step 1: write nat cache */
3523 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3524 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3525 written_size += SUM_JOURNAL_SIZE;
3526
3527 /* Step 2: write sit cache */
3528 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3529 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3530 written_size += SUM_JOURNAL_SIZE;
3531
351df4b2
JK		 3532 /* Step 3: write summary entries */
3533 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3534 unsigned short blkoff;
3535 seg_i = CURSEG_I(sbi, i);
3536 if (sbi->ckpt->alloc_type[i] == SSR)
3537 blkoff = sbi->blocks_per_seg;
3538 else
3539 blkoff = curseg_blkoff(sbi, i);
3540
3541 for (j = 0; j < blkoff; j++) {
3542 if (!page) {
4d57b86d 3543 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3544 kaddr = (unsigned char *)page_address(page);
81114baa 3545 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3546 written_size = 0;
3547 }
3548 summary = (struct f2fs_summary *)(kaddr + written_size);
3549 *summary = seg_i->sum_blk->entries[j];
3550 written_size += SUMMARY_SIZE;
351df4b2 3551
09cbfeaf 3552 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3553 SUM_FOOTER_SIZE)
3554 continue;
3555
e8d61a74 3556 set_page_dirty(page);
351df4b2
JK		 3557 f2fs_put_page(page, 1);
3558 page = NULL;
3559 }
3560 }
e8d61a74
CY		 3561 if (page) {
3562 set_page_dirty(page);
351df4b2 3563 f2fs_put_page(page, 1);
e8d61a74 3564 }
351df4b2
JK		 3565}
3566
3567static void write_normal_summaries(struct f2fs_sb_info *sbi,
3568 block_t blkaddr, int type)
3569{
3570 int i, end;
3571 if (IS_DATASEG(type))
3572 end = type + NR_CURSEG_DATA_TYPE;
3573 else
3574 end = type + NR_CURSEG_NODE_TYPE;
3575
b7ad7512
CY		 3576 for (i = type; i < end; i++)
3577 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3578}
3579
4d57b86d 3580void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3581{
aaec2b1d 3582 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3583 write_compacted_summaries(sbi, start_blk);
3584 else
3585 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3586}
3587
4d57b86d 3588void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3589{
119ee914 3590 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3591}
3592
4d57b86d 3593int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3594 unsigned int val, int alloc)
3595{
3596 int i;
3597
3598 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3599 for (i = 0; i < nats_in_cursum(journal); i++) {
3600 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3601 return i;
3602 }
dfc08a12
CY		 3603 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3604 return update_nats_in_cursum(journal, 1);
351df4b2 3605 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3606 for (i = 0; i < sits_in_cursum(journal); i++)
3607 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3608 return i;
dfc08a12
CY		 3609 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3610 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3611 }
3612 return -1;
3613}
3614
3615static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3616 unsigned int segno)
3617{
7735730d 3618 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3619}
3620
3621static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3622 unsigned int start)
3623{
3624 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3625 struct page *page;
351df4b2 3626 pgoff_t src_off, dst_off;
351df4b2
JK		 3627
3628 src_off = current_sit_addr(sbi, start);
3629 dst_off = next_sit_addr(sbi, src_off);
3630
4d57b86d 3631 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 3632 seg_info_to_sit_page(sbi, page, start);
351df4b2 3633
068c3cd8 3634 set_page_dirty(page);
351df4b2
JK		 3635 set_to_next_sit(sit_i, start);
3636
068c3cd8 3637 return page;
351df4b2
JK		 3638}
3639
184a5cd2
CY		 3640static struct sit_entry_set *grab_sit_entry_set(void)
3641{
3642 struct sit_entry_set *ses =
80c54505 3643 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3644
3645 ses->entry_cnt = 0;
3646 INIT_LIST_HEAD(&ses->set_list);
3647 return ses;
3648}
3649
3650static void release_sit_entry_set(struct sit_entry_set *ses)
3651{
3652 list_del(&ses->set_list);
3653 kmem_cache_free(sit_entry_set_slab, ses);
3654}
3655
3656static void adjust_sit_entry_set(struct sit_entry_set *ses,
3657 struct list_head *head)
3658{
3659 struct sit_entry_set *next = ses;
3660
3661 if (list_is_last(&ses->set_list, head))
3662 return;
3663
3664 list_for_each_entry_continue(next, head, set_list)
3665 if (ses->entry_cnt <= next->entry_cnt)
3666 break;
3667
3668 list_move_tail(&ses->set_list, &next->set_list);
3669}
3670
3671static void add_sit_entry(unsigned int segno, struct list_head *head)
3672{
3673 struct sit_entry_set *ses;
3674 unsigned int start_segno = START_SEGNO(segno);
3675
3676 list_for_each_entry(ses, head, set_list) {
3677 if (ses->start_segno == start_segno) {
3678 ses->entry_cnt++;
3679 adjust_sit_entry_set(ses, head);
3680 return;
3681 }
3682 }
3683
3684 ses = grab_sit_entry_set();
3685
3686 ses->start_segno = start_segno;
3687 ses->entry_cnt++;
3688 list_add(&ses->set_list, head);
3689}
3690
3691static void add_sits_in_set(struct f2fs_sb_info *sbi)
3692{
3693 struct f2fs_sm_info *sm_info = SM_I(sbi);
3694 struct list_head *set_list = &sm_info->sit_entry_set;
3695 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3696 unsigned int segno;
3697
7cd8558b 3698 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3699 add_sit_entry(segno, set_list);
3700}
3701
3702static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3703{
3704 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3705 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3706 int i;
3707
b7ad7512 3708 down_write(&curseg->journal_rwsem);
dfc08a12 3709 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3710 unsigned int segno;
3711 bool dirtied;
3712
dfc08a12 3713 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3714 dirtied = __mark_sit_entry_dirty(sbi, segno);
3715
3716 if (!dirtied)
3717 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3718 }
dfc08a12 3719 update_sits_in_cursum(journal, -i);
b7ad7512 3720 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3721}
3722
0a8165d7 3723/*
351df4b2
JK		 3724 * CP calls this function, which flushes SIT entries including sit_journal,
3725 * and moves prefree segs to free segs.
3726 */
4d57b86d 3727void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3728{
3729 struct sit_info *sit_i = SIT_I(sbi);
3730 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3731 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3732 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3733 struct sit_entry_set *ses, *tmp;
3734 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3735 bool to_journal = true;
4b2fecc8 3736 struct seg_entry *se;
351df4b2 3737
3d26fa6b 3738 down_write(&sit_i->sentry_lock);
351df4b2 3739
2b11a74b
WL		 3740 if (!sit_i->dirty_sentries)
3741 goto out;
3742
351df4b2 3743 /*
184a5cd2
CY		 3744 * add and account sit entries of dirty bitmap in sit entry
3745 * set temporarily
351df4b2 3746 */
184a5cd2 3747 add_sits_in_set(sbi);
351df4b2 3748
184a5cd2
CY		 3749 /*
3750 * if there are no enough space in journal to store dirty sit
3751 * entries, remove all entries from journal and add and account
3752 * them in sit entry set.
3753 */
dfc08a12 3754 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3755 remove_sits_in_journal(sbi);
b2955550 3756
184a5cd2
CY		 3757 /*
3758 * there are two steps to flush sit entries:
3759 * #1, flush sit entries to journal in current cold data summary block.
3760 * #2, flush sit entries to sit page.
3761 */
3762 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3763 struct page *page = NULL;
184a5cd2
CY		 3764 struct f2fs_sit_block *raw_sit = NULL;
3765 unsigned int start_segno = ses->start_segno;
3766 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3767 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3768 unsigned int segno = start_segno;
3769
3770 if (to_journal &&
dfc08a12 3771 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3772 to_journal = false;
3773
b7ad7512
CY		 3774 if (to_journal) {
3775 down_write(&curseg->journal_rwsem);
3776 } else {
184a5cd2
CY		 3777 page = get_next_sit_page(sbi, start_segno);
3778 raw_sit = page_address(page);
351df4b2 3779 }
351df4b2 3780
184a5cd2
CY		 3781 /* flush dirty sit entries in region of current sit set */
3782 for_each_set_bit_from(segno, bitmap, end) {
3783 int offset, sit_offset;
4b2fecc8
JK		 3784
3785 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3786#ifdef CONFIG_F2FS_CHECK_FS
3787 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3788 SIT_VBLOCK_MAP_SIZE))
3789 f2fs_bug_on(sbi, 1);
3790#endif
184a5cd2
CY		 3791
3792 /* add discard candidates */
c473f1a9 3793 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3794 cpc->trim_start = segno;
25290fa5 3795 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3796 }
184a5cd2
CY		 3797
3798 if (to_journal) {
4d57b86d 3799 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 3800 SIT_JOURNAL, segno, 1);
3801 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3802 segno_in_journal(journal, offset) =
184a5cd2
CY		 3803 cpu_to_le32(segno);
3804 seg_info_to_raw_sit(se,
dfc08a12 3805 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3806 check_block_count(sbi, segno,
3807 &sit_in_journal(journal, offset));
184a5cd2
CY		 3808 } else {
3809 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3810 seg_info_to_raw_sit(se,
3811 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3812 check_block_count(sbi, segno,
3813 &raw_sit->entries[sit_offset]);
184a5cd2 3814 }
351df4b2 3815
184a5cd2
CY		 3816 __clear_bit(segno, bitmap);
3817 sit_i->dirty_sentries--;
3818 ses->entry_cnt--;
351df4b2
JK		 3819 }
3820
b7ad7512
CY		 3821 if (to_journal)
3822 up_write(&curseg->journal_rwsem);
3823 else
184a5cd2
CY		 3824 f2fs_put_page(page, 1);
3825
3826 f2fs_bug_on(sbi, ses->entry_cnt);
3827 release_sit_entry_set(ses);
351df4b2 3828 }
184a5cd2
CY		 3829
3830 f2fs_bug_on(sbi, !list_empty(head));
3831 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3832out:
c473f1a9 3833 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3834 __u64 trim_start = cpc->trim_start;
3835
4b2fecc8 3836 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3837 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3838
3839 cpc->trim_start = trim_start;
4b2fecc8 3840 }
3d26fa6b 3841 up_write(&sit_i->sentry_lock);
351df4b2 3842
351df4b2
JK		 3843 set_prefree_as_free_segments(sbi);
3844}
3845
3846static int build_sit_info(struct f2fs_sb_info *sbi)
3847{
3848 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3849 struct sit_info *sit_i;
3850 unsigned int sit_segs, start;
ae27d62e 3851 char *src_bitmap;
351df4b2
JK		 3852 unsigned int bitmap_size;
3853
3854 /* allocate memory for SIT information */
acbf054d 3855 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3856 if (!sit_i)
3857 return -ENOMEM;
3858
3859 SM_I(sbi)->sit_info = sit_i;
3860
9d2a789c
KC		 3861 sit_i->sentries =
3862 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3863 MAIN_SEGS(sbi)),
3864 GFP_KERNEL);
351df4b2
JK		 3865 if (!sit_i->sentries)
3866 return -ENOMEM;
3867
7cd8558b 3868 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3869 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3870 GFP_KERNEL);
351df4b2
JK		 3871 if (!sit_i->dirty_sentries_bitmap)
3872 return -ENOMEM;
3873
7cd8558b 3874 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3875 sit_i->sentries[start].cur_valid_map
acbf054d 3876 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3877 sit_i->sentries[start].ckpt_valid_map
acbf054d 3878 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3879 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3880 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3881 return -ENOMEM;
3e025740 3882
355e7891
CY		 3883#ifdef CONFIG_F2FS_CHECK_FS
3884 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3885 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3886 if (!sit_i->sentries[start].cur_valid_map_mir)
3887 return -ENOMEM;
3888#endif
3889
7d20c8ab
CY		 3890 sit_i->sentries[start].discard_map
3891 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3892 GFP_KERNEL);
3893 if (!sit_i->sentries[start].discard_map)
3894 return -ENOMEM;
351df4b2
JK		 3895 }
3896
acbf054d 3897 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3898 if (!sit_i->tmp_map)
3899 return -ENOMEM;
3900
2c70c5e3 3901 if (__is_large_section(sbi)) {
9d2a789c
KC		 3902 sit_i->sec_entries =
3903 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3904 MAIN_SECS(sbi)),
3905 GFP_KERNEL);
351df4b2
JK		 3906 if (!sit_i->sec_entries)
3907 return -ENOMEM;
3908 }
3909
3910 /* get information related with SIT */
3911 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3912
3913 /* setup SIT bitmap from ckeckpoint pack */
3914 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3915 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3916
ae27d62e
CY		 3917 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3918 if (!sit_i->sit_bitmap)
351df4b2 3919 return -ENOMEM;
351df4b2 3920
ae27d62e
CY		 3921#ifdef CONFIG_F2FS_CHECK_FS
3922 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3923 if (!sit_i->sit_bitmap_mir)
3924 return -ENOMEM;
3925#endif
3926
351df4b2
JK		 3927 /* init SIT information */
3928 sit_i->s_ops = &default_salloc_ops;
3929
3930 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3931 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3932 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3933 sit_i->bitmap_size = bitmap_size;
3934 sit_i->dirty_sentries = 0;
3935 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3936 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3937 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3938 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3939 return 0;
3940}
3941
3942static int build_free_segmap(struct f2fs_sb_info *sbi)
3943{
351df4b2
JK		 3944 struct free_segmap_info *free_i;
3945 unsigned int bitmap_size, sec_bitmap_size;
3946
3947 /* allocate memory for free segmap information */
acbf054d 3948 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3949 if (!free_i)
3950 return -ENOMEM;
3951
3952 SM_I(sbi)->free_info = free_i;
3953
7cd8558b 3954 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3955 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3956 if (!free_i->free_segmap)
3957 return -ENOMEM;
3958
7cd8558b 3959 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3960 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3961 if (!free_i->free_secmap)
3962 return -ENOMEM;
3963
3964 /* set all segments as dirty temporarily */
3965 memset(free_i->free_segmap, 0xff, bitmap_size);
3966 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3967
3968 /* init free segmap information */
7cd8558b 3969 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3970 free_i->free_segments = 0;
3971 free_i->free_sections = 0;
1a118ccf 3972 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3973 return 0;
3974}
3975
3976static int build_curseg(struct f2fs_sb_info *sbi)
3977{
1042d60f 3978 struct curseg_info *array;
351df4b2
JK		 3979 int i;
3980
026f0507
KC		 3981 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3982 GFP_KERNEL);
351df4b2
JK		 3983 if (!array)
3984 return -ENOMEM;
3985
3986 SM_I(sbi)->curseg_array = array;
3987
3988 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3989 mutex_init(&array[i].curseg_mutex);
acbf054d 3990 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3991 if (!array[i].sum_blk)
3992 return -ENOMEM;
b7ad7512 3993 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 3994 array[i].journal = f2fs_kzalloc(sbi,
3995 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 3996 if (!array[i].journal)
3997 return -ENOMEM;
351df4b2
JK		 3998 array[i].segno = NULL_SEGNO;
3999 array[i].next_blkoff = 0;
4000 }
4001 return restore_curseg_summaries(sbi);
4002}
4003
c39a1b34 4004static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 4005{
4006 struct sit_info *sit_i = SIT_I(sbi);
4007 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 4008 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 4009 struct seg_entry *se;
4010 struct f2fs_sit_entry sit;
74de593a
CY		 4011 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4012 unsigned int i, start, end;
4013 unsigned int readed, start_blk = 0;
c39a1b34 4014 int err = 0;
8a29c126 4015 block_t total_node_blocks = 0;
351df4b2 4016
74de593a 4017 do {
4d57b86d 4018 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
664ba972 4019 META_SIT, true);
74de593a
CY		 4020
4021 start = start_blk * sit_i->sents_per_block;
4022 end = (start_blk + readed) * sit_i->sents_per_block;
4023
7cd8558b 4024 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 4025 struct f2fs_sit_block *sit_blk;
74de593a
CY		 4026 struct page *page;
4027
9c094040 4028 se = &sit_i->sentries[start];
74de593a 4029 page = get_current_sit_page(sbi, start);
edc55aaf
JK		 4030 if (IS_ERR(page))
4031 return PTR_ERR(page);
74de593a
CY		 4032 sit_blk = (struct f2fs_sit_block *)page_address(page);
4033 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4034 f2fs_put_page(page, 1);
d600af23 4035
c39a1b34
JK		 4036 err = check_block_count(sbi, start, &sit);
4037 if (err)
4038 return err;
74de593a 4039 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4040 if (IS_NODESEG(se->type))
4041 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 4042
4043 /* build discard map only one time */
7d20c8ab
CY		 4044 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4045 memset(se->discard_map, 0xff,
4046 SIT_VBLOCK_MAP_SIZE);
4047 } else {
4048 memcpy(se->discard_map,
4049 se->cur_valid_map,
4050 SIT_VBLOCK_MAP_SIZE);
4051 sbi->discard_blks +=
4052 sbi->blocks_per_seg -
4053 se->valid_blocks;
3e025740 4054 }
a66cdd98 4055
2c70c5e3 4056 if (__is_large_section(sbi))
d600af23
CY		 4057 get_sec_entry(sbi, start)->valid_blocks +=
4058 se->valid_blocks;
351df4b2 4059 }
74de593a
CY		 4060 start_blk += readed;
4061 } while (start_blk < sit_blk_cnt);
d600af23
CY		 4062
4063 down_read(&curseg->journal_rwsem);
4064 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 4065 unsigned int old_valid_blocks;
4066
4067 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 4068 if (start >= MAIN_SEGS(sbi)) {
4069 f2fs_msg(sbi->sb, KERN_ERR,
4070 "Wrong journal entry on segno %u",
4071 start);
4072 set_sbi_flag(sbi, SBI_NEED_FSCK);
4073 err = -EINVAL;
4074 break;
4075 }
4076
d600af23
CY		 4077 se = &sit_i->sentries[start];
4078 sit = sit_in_journal(journal, i);
4079
4080 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 4081 if (IS_NODESEG(se->type))
4082 total_node_blocks -= old_valid_blocks;
d600af23 4083
c39a1b34
JK		 4084 err = check_block_count(sbi, start, &sit);
4085 if (err)
4086 break;
d600af23 4087 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4088 if (IS_NODESEG(se->type))
4089 total_node_blocks += se->valid_blocks;
d600af23 4090
7d20c8ab
CY		 4091 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4092 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4093 } else {
4094 memcpy(se->discard_map, se->cur_valid_map,
4095 SIT_VBLOCK_MAP_SIZE);
4096 sbi->discard_blks += old_valid_blocks;
4097 sbi->discard_blks -= se->valid_blocks;
d600af23
CY		 4098 }
4099
2c70c5e3 4100 if (__is_large_section(sbi)) {
d600af23 4101 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 4102 se->valid_blocks;
4103 get_sec_entry(sbi, start)->valid_blocks -=
4104 old_valid_blocks;
4105 }
d600af23
CY		 4106 }
4107 up_read(&curseg->journal_rwsem);
8a29c126
JK		 4108
4109 if (!err && total_node_blocks != valid_node_count(sbi)) {
4110 f2fs_msg(sbi->sb, KERN_ERR,
4111 "SIT is corrupted node# %u vs %u",
4112 total_node_blocks, valid_node_count(sbi));
4113 set_sbi_flag(sbi, SBI_NEED_FSCK);
4114 err = -EINVAL;
4115 }
4116
c39a1b34 4117 return err;
351df4b2
JK		 4118}
4119
4120static void init_free_segmap(struct f2fs_sb_info *sbi)
4121{
4122 unsigned int start;
4123 int type;
4124
7cd8558b 4125 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 4126 struct seg_entry *sentry = get_seg_entry(sbi, start);
4127 if (!sentry->valid_blocks)
4128 __set_free(sbi, start);
c79b7ff1
JK		 4129 else
4130 SIT_I(sbi)->written_valid_blocks +=
4131 sentry->valid_blocks;
351df4b2
JK		 4132 }
4133
4134 /* set use the current segments */
4135 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4136 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4137 __set_test_and_inuse(sbi, curseg_t->segno);
4138 }
4139}
4140
4141static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4142{
4143 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4144 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 4145 unsigned int segno = 0, offset = 0;
351df4b2
JK		 4146 unsigned short valid_blocks;
4147
8736fbf0 4148 while (1) {
351df4b2 4149 /* find dirty segment based on free segmap */
7cd8558b
JK		 4150 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4151 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 4152 break;
4153 offset = segno + 1;
302bd348 4154 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 4155 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 4156 continue;
ec325b52
JK		 4157 if (valid_blocks > sbi->blocks_per_seg) {
4158 f2fs_bug_on(sbi, 1);
4159 continue;
4160 }
351df4b2
JK		 4161 mutex_lock(&dirty_i->seglist_lock);
4162 __locate_dirty_segment(sbi, segno, DIRTY);
4163 mutex_unlock(&dirty_i->seglist_lock);
4164 }
4165}
4166
5ec4e49f 4167static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4168{
4169 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 4170 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 4171
628b3d14 4172 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 4173 if (!dirty_i->victim_secmap)
351df4b2
JK		 4174 return -ENOMEM;
4175 return 0;
4176}
4177
4178static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4179{
4180 struct dirty_seglist_info *dirty_i;
4181 unsigned int bitmap_size, i;
4182
4183 /* allocate memory for dirty segments list information */
acbf054d
CY		 4184 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4185 GFP_KERNEL);
351df4b2
JK		 4186 if (!dirty_i)
4187 return -ENOMEM;
4188
4189 SM_I(sbi)->dirty_info = dirty_i;
4190 mutex_init(&dirty_i->seglist_lock);
4191
7cd8558b 4192 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 4193
4194 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 4195 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4196 GFP_KERNEL);
351df4b2
JK		 4197 if (!dirty_i->dirty_segmap[i])
4198 return -ENOMEM;
4199 }
4200
4201 init_dirty_segmap(sbi);
5ec4e49f 4202 return init_victim_secmap(sbi);
351df4b2
JK		 4203}
4204
0a8165d7 4205/*
351df4b2
JK		 4206 * Update min, max modified time for cost-benefit GC algorithm
4207 */
4208static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4209{
4210 struct sit_info *sit_i = SIT_I(sbi);
4211 unsigned int segno;
4212
3d26fa6b 4213 down_write(&sit_i->sentry_lock);
351df4b2 4214
5ad25442 4215 sit_i->min_mtime = ULLONG_MAX;
351df4b2 4216
7cd8558b 4217 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 4218 unsigned int i;
4219 unsigned long long mtime = 0;
4220
4221 for (i = 0; i < sbi->segs_per_sec; i++)
4222 mtime += get_seg_entry(sbi, segno + i)->mtime;
4223
4224 mtime = div_u64(mtime, sbi->segs_per_sec);
4225
4226 if (sit_i->min_mtime > mtime)
4227 sit_i->min_mtime = mtime;
4228 }
a1f72ac2 4229 sit_i->max_mtime = get_mtime(sbi, false);
3d26fa6b 4230 up_write(&sit_i->sentry_lock);
351df4b2
JK		 4231}
4232
4d57b86d 4233int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4234{
4235 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4236 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 4237 struct f2fs_sm_info *sm_info;
351df4b2
JK		 4238 int err;
4239
acbf054d 4240 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 4241 if (!sm_info)
4242 return -ENOMEM;
4243
4244 /* init sm info */
4245 sbi->sm_info = sm_info;
351df4b2
JK		 4246 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4247 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4248 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4249 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4250 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4251 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4252 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 4253 sm_info->rec_prefree_segments = sm_info->main_segments *
4254 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 4255 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4256 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4257
52763a4b
JK		 4258 if (!test_opt(sbi, LFS))
4259 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 4260 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 4261 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
853137ce 4262 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
ef095d19 4263 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 4264 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 4265
184a5cd2
CY		 4266 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4267
2b60311d
CY		 4268 init_rwsem(&sm_info->curseg_lock);
4269
d4fdf8ba 4270 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 4271 err = f2fs_create_flush_cmd_control(sbi);
2163d198 4272 if (err)
a688b9d9 4273 return err;
6b4afdd7
JK		 4274 }
4275
0b54fb84
JK		 4276 err = create_discard_cmd_control(sbi);
4277 if (err)
4278 return err;
4279
351df4b2
JK		 4280 err = build_sit_info(sbi);
4281 if (err)
4282 return err;
4283 err = build_free_segmap(sbi);
4284 if (err)
4285 return err;
4286 err = build_curseg(sbi);
4287 if (err)
4288 return err;
4289
4290 /* reinit free segmap based on SIT */
c39a1b34
JK		 4291 err = build_sit_entries(sbi);
4292 if (err)
4293 return err;
351df4b2
JK		 4294
4295 init_free_segmap(sbi);
4296 err = build_dirty_segmap(sbi);
4297 if (err)
4298 return err;
4299
4300 init_min_max_mtime(sbi);
4301 return 0;
4302}
4303
4304static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4305 enum dirty_type dirty_type)
4306{
4307 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4308
4309 mutex_lock(&dirty_i->seglist_lock);
39307a8e 4310 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 4311 dirty_i->nr_dirty[dirty_type] = 0;
4312 mutex_unlock(&dirty_i->seglist_lock);
4313}
4314
5ec4e49f 4315static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4316{
4317 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 4318 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 4319}
4320
4321static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4322{
4323 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4324 int i;
4325
4326 if (!dirty_i)
4327 return;
4328
4329 /* discard pre-free/dirty segments list */
4330 for (i = 0; i < NR_DIRTY_TYPE; i++)
4331 discard_dirty_segmap(sbi, i);
4332
5ec4e49f 4333 destroy_victim_secmap(sbi);
351df4b2 4334 SM_I(sbi)->dirty_info = NULL;
5222595d 4335 kvfree(dirty_i);
351df4b2
JK		 4336}
4337
4338static void destroy_curseg(struct f2fs_sb_info *sbi)
4339{
4340 struct curseg_info *array = SM_I(sbi)->curseg_array;
4341 int i;
4342
4343 if (!array)
4344 return;
4345 SM_I(sbi)->curseg_array = NULL;
b7ad7512 4346 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5222595d
JK		 4347 kvfree(array[i].sum_blk);
4348 kvfree(array[i].journal);
b7ad7512 4349 }
5222595d 4350 kvfree(array);
351df4b2
JK		 4351}
4352
4353static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4354{
4355 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4356 if (!free_i)
4357 return;
4358 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4359 kvfree(free_i->free_segmap);
4360 kvfree(free_i->free_secmap);
5222595d 4361 kvfree(free_i);
351df4b2
JK		 4362}
4363
4364static void destroy_sit_info(struct f2fs_sb_info *sbi)
4365{
4366 struct sit_info *sit_i = SIT_I(sbi);
4367 unsigned int start;
4368
4369 if (!sit_i)
4370 return;
4371
4372 if (sit_i->sentries) {
7cd8558b 4373 for (start = 0; start < MAIN_SEGS(sbi); start++) {
5222595d 4374 kvfree(sit_i->sentries[start].cur_valid_map);
355e7891 4375#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4376 kvfree(sit_i->sentries[start].cur_valid_map_mir);
355e7891 4377#endif
5222595d
JK		 4378 kvfree(sit_i->sentries[start].ckpt_valid_map);
4379 kvfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4380 }
4381 }
5222595d 4382 kvfree(sit_i->tmp_map);
60a3b782 4383
39307a8e
JK		 4384 kvfree(sit_i->sentries);
4385 kvfree(sit_i->sec_entries);
4386 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4387
4388 SM_I(sbi)->sit_info = NULL;
5222595d 4389 kvfree(sit_i->sit_bitmap);
ae27d62e 4390#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4391 kvfree(sit_i->sit_bitmap_mir);
ae27d62e 4392#endif
5222595d 4393 kvfree(sit_i);
351df4b2
JK		 4394}
4395
4d57b86d 4396void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4397{
4398 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4399
3b03f724
CY		 4400 if (!sm_info)
4401 return;
4d57b86d 4402 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 4403 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4404 destroy_dirty_segmap(sbi);
4405 destroy_curseg(sbi);
4406 destroy_free_segmap(sbi);
4407 destroy_sit_info(sbi);
4408 sbi->sm_info = NULL;
5222595d 4409 kvfree(sm_info);
351df4b2 4410}
7fd9e544 4411
4d57b86d 4412int __init f2fs_create_segment_manager_caches(void)
7fd9e544
JK		 4413{
4414 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4415 sizeof(struct discard_entry));
7fd9e544 4416 if (!discard_entry_slab)
184a5cd2
CY		 4417 goto fail;
4418
b01a9201
JK		 4419 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4420 sizeof(struct discard_cmd));
4421 if (!discard_cmd_slab)
6ab2a308 4422 goto destroy_discard_entry;
275b66b0 4423
184a5cd2 4424 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4425 sizeof(struct sit_entry_set));
184a5cd2 4426 if (!sit_entry_set_slab)
b01a9201 4427 goto destroy_discard_cmd;
88b88a66
JK		 4428
4429 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4430 sizeof(struct inmem_pages));
4431 if (!inmem_entry_slab)
4432 goto destroy_sit_entry_set;
7fd9e544 4433 return 0;
184a5cd2 4434
88b88a66
JK		 4435destroy_sit_entry_set:
4436 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4437destroy_discard_cmd:
4438 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4439destroy_discard_entry:
184a5cd2
CY		 4440 kmem_cache_destroy(discard_entry_slab);
4441fail:
4442 return -ENOMEM;
7fd9e544
JK		 4443}
4444
4d57b86d 4445void f2fs_destroy_segment_manager_caches(void)
7fd9e544 4446{
184a5cd2 4447 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4448 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4449 kmem_cache_destroy(discard_entry_slab);
88b88a66 4450 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4451}
Ubuntu Jammy Linux kernel mirror