projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'acpi-5.1-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
[linux-2.6-block.git] / fs / f2fs / segment.c
CommitLineData
7c1a000d 1// SPDX-License-Identifier: GPL-2.0
0a8165d7 2/*
351df4b2
JK		 3 * fs/f2fs/segment.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
351df4b2
JK		 7 */
8#include <linux/fs.h>
9#include <linux/f2fs_fs.h>
10#include <linux/bio.h>
11#include <linux/blkdev.h>
690e4a3e 12#include <linux/prefetch.h>
6b4afdd7 13#include <linux/kthread.h>
74de593a 14#include <linux/swap.h>
60b99b48 15#include <linux/timer.h>
1d7be270 16#include <linux/freezer.h>
1eb1ef4a 17#include <linux/sched/signal.h>
351df4b2
JK		 18
19#include "f2fs.h"
20#include "segment.h"
21#include "node.h"
5f656541 22#include "gc.h"
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;
1040
1041 if (discard_type == DPOLICY_BG) {
1042 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1043 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1044 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1045 dpolicy->io_aware = true;
cba60849 1046 dpolicy->sync = false;
20ee4382 1047 dpolicy->ordered = true;
8bb4f253
JK		 1048 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1049 dpolicy->granularity = 1;
1050 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1051 }
1052 } else if (discard_type == DPOLICY_FORCE) {
1053 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
f9d1dced 1054 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
8bb4f253
JK		 1055 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1056 dpolicy->io_aware = false;
1057 } else if (discard_type == DPOLICY_FSTRIM) {
1058 dpolicy->io_aware = false;
1059 } else if (discard_type == DPOLICY_UMOUNT) {
241b493d 1060 dpolicy->max_requests = UINT_MAX;
8bb4f253
JK		 1061 dpolicy->io_aware = false;
1062 }
1063}
1064
35ec7d57
CY		 1065static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1066 struct block_device *bdev, block_t lstart,
1067 block_t start, block_t len);
c81abe34 1068/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
6b9cb124 1069static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1070 struct discard_policy *dpolicy,
35ec7d57
CY		 1071 struct discard_cmd *dc,
1072 unsigned int *issued)
c81abe34 1073{
35ec7d57
CY		 1074 struct block_device *bdev = dc->bdev;
1075 struct request_queue *q = bdev_get_queue(bdev);
1076 unsigned int max_discard_blocks =
1077 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
c81abe34 1078 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1079 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1080 &(dcc->fstrim_list) : &(dcc->wait_list);
78997b56 1081 int flag = dpolicy->sync ? REQ_SYNC : 0;
35ec7d57
CY		 1082 block_t lstart, start, len, total_len;
1083 int err = 0;
c81abe34
JK		 1084
1085 if (dc->state != D_PREP)
6b9cb124 1086 return 0;
c81abe34 1087
d6184774 1088 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
6b9cb124 1089 return 0;
d6184774 1090
35ec7d57
CY		 1091 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1092
1093 lstart = dc->lstart;
1094 start = dc->start;
1095 len = dc->len;
1096 total_len = len;
1097
1098 dc->len = 0;
1099
1100 while (total_len && *issued < dpolicy->max_requests && !err) {
1101 struct bio *bio = NULL;
1102 unsigned long flags;
1103 bool last = true;
1104
1105 if (len > max_discard_blocks) {
1106 len = max_discard_blocks;
1107 last = false;
1108 }
1109
1110 (*issued)++;
1111 if (*issued == dpolicy->max_requests)
1112 last = true;
1113
1114 dc->len += len;
1115
b83dcfe6
CY		 1116 if (time_to_inject(sbi, FAULT_DISCARD)) {
1117 f2fs_show_injection_info(FAULT_DISCARD);
1118 err = -EIO;
1119 goto submit;
1120 }
35ec7d57
CY		 1121 err = __blkdev_issue_discard(bdev,
1122 SECTOR_FROM_BLOCK(start),
1123 SECTOR_FROM_BLOCK(len),
1124 GFP_NOFS, 0, &bio);
b83dcfe6 1125submit:
6b9cb124 1126 if (err) {
35ec7d57 1127 spin_lock_irqsave(&dc->lock, flags);
6b9cb124 1128 if (dc->state == D_PARTIAL)
35ec7d57 1129 dc->state = D_SUBMIT;
35ec7d57
CY		 1130 spin_unlock_irqrestore(&dc->lock, flags);
1131
6b9cb124
CY		 1132 break;
1133 }
35ec7d57 1134
6b9cb124 1135 f2fs_bug_on(sbi, !bio);
35ec7d57 1136
6b9cb124
CY		 1137 /*
1138 * should keep before submission to avoid D_DONE
1139 * right away
1140 */
1141 spin_lock_irqsave(&dc->lock, flags);
1142 if (last)
1143 dc->state = D_SUBMIT;
1144 else
1145 dc->state = D_PARTIAL;
1146 dc->bio_ref++;
1147 spin_unlock_irqrestore(&dc->lock, flags);
35ec7d57 1148
72691af6
JK		 1149 atomic_inc(&dcc->queued_discard);
1150 dc->queued++;
6b9cb124 1151 list_move_tail(&dc->list, wait_list);
b0af6d49 1152
6b9cb124 1153 /* sanity check on discard range */
9249dded 1154 __check_sit_bitmap(sbi, lstart, lstart + len);
35ec7d57 1155
6b9cb124
CY		 1156 bio->bi_private = dc;
1157 bio->bi_end_io = f2fs_submit_discard_endio;
1158 bio->bi_opf |= flag;
1159 submit_bio(bio);
1160
1161 atomic_inc(&dcc->issued_discard);
1162
1163 f2fs_update_iostat(sbi, FS_DISCARD, 1);
35ec7d57
CY		 1164
1165 lstart += len;
1166 start += len;
1167 total_len -= len;
1168 len = total_len;
c81abe34 1169 }
35ec7d57 1170
6b9cb124 1171 if (!err && len)
35ec7d57 1172 __update_discard_tree_range(sbi, bdev, lstart, start, len);
6b9cb124 1173 return err;
c81abe34
JK		 1174}
1175
004b6862
CY		 1176static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1177 struct block_device *bdev, block_t lstart,
1178 block_t start, block_t len,
1179 struct rb_node **insert_p,
1180 struct rb_node *insert_parent)
c81abe34 1181{
004b6862 1182 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 1183 struct rb_node **p;
004b6862
CY		 1184 struct rb_node *parent = NULL;
1185 struct discard_cmd *dc = NULL;
4dada3fd 1186 bool leftmost = true;
004b6862
CY		 1187
1188 if (insert_p && insert_parent) {
1189 parent = insert_parent;
1190 p = insert_p;
1191 goto do_insert;
1192 }
c81abe34 1193
4dada3fd
CY		 1194 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1195 lstart, &leftmost);
004b6862 1196do_insert:
4dada3fd
CY		 1197 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1198 p, leftmost);
004b6862
CY		 1199 if (!dc)
1200 return NULL;
c81abe34 1201
004b6862 1202 return dc;
c81abe34
JK		 1203}
1204
ba48a33e
CY		 1205static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1206 struct discard_cmd *dc)
1207{
1208 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1209}
1210
3d6a650f
YH		 1211static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1212 struct discard_cmd *dc, block_t blkaddr)
1213{
ba48a33e 1214 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1215 struct discard_info di = dc->di;
1216 bool modified = false;
3d6a650f 1217
004b6862 1218 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 1219 __remove_discard_cmd(sbi, dc);
1220 return;
1221 }
1222
d84d1cbd
CY		 1223 dcc->undiscard_blks -= di.len;
1224
004b6862 1225 if (blkaddr > di.lstart) {
3d6a650f 1226 dc->len = blkaddr - dc->lstart;
d84d1cbd 1227 dcc->undiscard_blks += dc->len;
ba48a33e 1228 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1229 modified = true;
1230 }
1231
1232 if (blkaddr < di.lstart + di.len - 1) {
1233 if (modified) {
1234 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1235 di.start + blkaddr + 1 - di.lstart,
1236 di.lstart + di.len - 1 - blkaddr,
1237 NULL, NULL);
1238 } else {
1239 dc->lstart++;
1240 dc->len--;
1241 dc->start++;
d84d1cbd 1242 dcc->undiscard_blks += dc->len;
ba48a33e 1243 __relocate_discard_cmd(dcc, dc);
004b6862 1244 }
3d6a650f
YH		 1245 }
1246}
1247
004b6862
CY		 1248static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1249 struct block_device *bdev, block_t lstart,
1250 block_t start, block_t len)
275b66b0 1251{
0b54fb84 1252 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1253 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1254 struct discard_cmd *dc;
1255 struct discard_info di = {0};
1256 struct rb_node **insert_p = NULL, *insert_parent = NULL;
35ec7d57
CY		 1257 struct request_queue *q = bdev_get_queue(bdev);
1258 unsigned int max_discard_blocks =
1259 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
004b6862 1260 block_t end = lstart + len;
275b66b0 1261
4d57b86d 1262 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
004b6862
CY		 1263 NULL, lstart,
1264 (struct rb_entry **)&prev_dc,
1265 (struct rb_entry **)&next_dc,
4dada3fd 1266 &insert_p, &insert_parent, true, NULL);
004b6862
CY		 1267 if (dc)
1268 prev_dc = dc;
1269
1270 if (!prev_dc) {
1271 di.lstart = lstart;
1272 di.len = next_dc ? next_dc->lstart - lstart : len;
1273 di.len = min(di.len, len);
1274 di.start = start;
22d375dd 1275 }
15469963 1276
004b6862
CY		 1277 while (1) {
1278 struct rb_node *node;
1279 bool merged = false;
1280 struct discard_cmd *tdc = NULL;
1281
1282 if (prev_dc) {
1283 di.lstart = prev_dc->lstart + prev_dc->len;
1284 if (di.lstart < lstart)
1285 di.lstart = lstart;
1286 if (di.lstart >= end)
1287 break;
1288
1289 if (!next_dc || next_dc->lstart > end)
1290 di.len = end - di.lstart;
1291 else
1292 di.len = next_dc->lstart - di.lstart;
1293 di.start = start + di.lstart - lstart;
1294 }
1295
1296 if (!di.len)
1297 goto next;
1298
1299 if (prev_dc && prev_dc->state == D_PREP &&
1300 prev_dc->bdev == bdev &&
35ec7d57
CY		 1301 __is_discard_back_mergeable(&di, &prev_dc->di,
1302 max_discard_blocks)) {
004b6862 1303 prev_dc->di.len += di.len;
d84d1cbd 1304 dcc->undiscard_blks += di.len;
ba48a33e 1305 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1306 di = prev_dc->di;
1307 tdc = prev_dc;
1308 merged = true;
1309 }
1310
1311 if (next_dc && next_dc->state == D_PREP &&
1312 next_dc->bdev == bdev &&
35ec7d57
CY		 1313 __is_discard_front_mergeable(&di, &next_dc->di,
1314 max_discard_blocks)) {
004b6862
CY		 1315 next_dc->di.lstart = di.lstart;
1316 next_dc->di.len += di.len;
1317 next_dc->di.start = di.start;
d84d1cbd 1318 dcc->undiscard_blks += di.len;
ba48a33e 1319 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1320 if (tdc)
1321 __remove_discard_cmd(sbi, tdc);
004b6862 1322 merged = true;
4e6a8d9b 1323 }
004b6862 1324
df0f6b44 1325 if (!merged) {
004b6862
CY		 1326 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1327 di.len, NULL, NULL);
df0f6b44 1328 }
004b6862
CY		 1329 next:
1330 prev_dc = next_dc;
1331 if (!prev_dc)
1332 break;
1333
1334 node = rb_next(&prev_dc->rb_node);
1335 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1336 }
004b6862
CY		 1337}
1338
1339static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1340 struct block_device *bdev, block_t blkstart, block_t blklen)
1341{
1342 block_t lblkstart = blkstart;
1343
0243a5f9 1344 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1345
1346 if (sbi->s_ndevs) {
1347 int devi = f2fs_target_device_index(sbi, blkstart);
1348
1349 blkstart -= FDEV(devi).start_blk;
1350 }
35ec7d57 1351 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862 1352 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
35ec7d57 1353 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
004b6862
CY		 1354 return 0;
1355}
1356
20ee4382
CY		 1357static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1358 struct discard_policy *dpolicy)
1359{
1360 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1361 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1362 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1363 struct discard_cmd *dc;
1364 struct blk_plug plug;
1365 unsigned int pos = dcc->next_pos;
1366 unsigned int issued = 0;
1367 bool io_interrupted = false;
1368
1369 mutex_lock(&dcc->cmd_lock);
1370 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1371 NULL, pos,
1372 (struct rb_entry **)&prev_dc,
1373 (struct rb_entry **)&next_dc,
4dada3fd 1374 &insert_p, &insert_parent, true, NULL);
20ee4382
CY		 1375 if (!dc)
1376 dc = next_dc;
1377
1378 blk_start_plug(&plug);
1379
1380 while (dc) {
1381 struct rb_node *node;
6b9cb124 1382 int err = 0;
20ee4382
CY		 1383
1384 if (dc->state != D_PREP)
1385 goto next;
1386
a7d10cf3 1387 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
20ee4382
CY		 1388 io_interrupted = true;
1389 break;
1390 }
1391
1392 dcc->next_pos = dc->lstart + dc->len;
6b9cb124 1393 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
20ee4382 1394
35ec7d57 1395 if (issued >= dpolicy->max_requests)
20ee4382
CY		 1396 break;
1397next:
1398 node = rb_next(&dc->rb_node);
6b9cb124
CY		 1399 if (err)
1400 __remove_discard_cmd(sbi, dc);
20ee4382
CY		 1401 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1402 }
1403
1404 blk_finish_plug(&plug);
1405
1406 if (!dc)
1407 dcc->next_pos = 0;
1408
1409 mutex_unlock(&dcc->cmd_lock);
1410
1411 if (!issued && io_interrupted)
1412 issued = -1;
1413
1414 return issued;
1415}
1416
78997b56
CY		 1417static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1418 struct discard_policy *dpolicy)
bd5b0738
CY		 1419{
1420 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1421 struct list_head *pend_list;
1422 struct discard_cmd *dc, *tmp;
1423 struct blk_plug plug;
522d1711 1424 int i, issued = 0;
e6c6de18 1425 bool io_interrupted = false;
bd5b0738 1426
78997b56
CY		 1427 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1428 if (i + 1 < dpolicy->granularity)
1429 break;
20ee4382
CY		 1430
1431 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1432 return __issue_discard_cmd_orderly(sbi, dpolicy);
1433
bd5b0738 1434 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1435
1436 mutex_lock(&dcc->cmd_lock);
49c60c67
CY		 1437 if (list_empty(pend_list))
1438 goto next;
67fce70b
CY		 1439 if (unlikely(dcc->rbtree_check))
1440 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1441 &dcc->root));
33da62cf 1442 blk_start_plug(&plug);
bd5b0738
CY		 1443 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1444 f2fs_bug_on(sbi, dc->state != D_PREP);
1445
ecc9aa00 1446 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
a7d10cf3 1447 !is_idle(sbi, DISCARD_TIME)) {
e6c6de18 1448 io_interrupted = true;
522d1711 1449 break;
969d1b18 1450 }
e6c6de18 1451
35ec7d57 1452 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
522d1711 1453
35ec7d57 1454 if (issued >= dpolicy->max_requests)
33da62cf 1455 break;
bd5b0738 1456 }
33da62cf 1457 blk_finish_plug(&plug);
49c60c67 1458next:
33da62cf
CY		 1459 mutex_unlock(&dcc->cmd_lock);
1460
522d1711 1461 if (issued >= dpolicy->max_requests || io_interrupted)
33da62cf 1462 break;
bd5b0738 1463 }
969d1b18 1464
e6c6de18
CY		 1465 if (!issued && io_interrupted)
1466 issued = -1;
1467
969d1b18
CY		 1468 return issued;
1469}
1470
cf5c759f 1471static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1472{
1473 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1474 struct list_head *pend_list;
1475 struct discard_cmd *dc, *tmp;
1476 int i;
cf5c759f 1477 bool dropped = false;
969d1b18
CY		 1478
1479 mutex_lock(&dcc->cmd_lock);
1480 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1481 pend_list = &dcc->pend_list[i];
1482 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1483 f2fs_bug_on(sbi, dc->state != D_PREP);
1484 __remove_discard_cmd(sbi, dc);
cf5c759f 1485 dropped = true;
969d1b18
CY		 1486 }
1487 }
1488 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1489
1490 return dropped;
bd5b0738
CY		 1491}
1492
4d57b86d 1493void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
7950e9ac
CY		 1494{
1495 __drop_discard_cmd(sbi);
1496}
1497
0ea80512 1498static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1499 struct discard_cmd *dc)
1500{
1501 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1502 unsigned int len = 0;
2a510c00
CY		 1503
1504 wait_for_completion_io(&dc->wait);
1505 mutex_lock(&dcc->cmd_lock);
1506 f2fs_bug_on(sbi, dc->state != D_DONE);
1507 dc->ref--;
0ea80512
CY		 1508 if (!dc->ref) {
1509 if (!dc->error)
1510 len = dc->len;
2a510c00 1511 __remove_discard_cmd(sbi, dc);
0ea80512 1512 }
2a510c00 1513 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1514
1515 return len;
2a510c00
CY		 1516}
1517
0ea80512 1518static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1519 struct discard_policy *dpolicy,
1520 block_t start, block_t end)
63a94fa1
CY		 1521{
1522 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1523 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1524 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1525 struct discard_cmd *dc, *tmp;
6afae633 1526 bool need_wait;
0ea80512 1527 unsigned int trimmed = 0;
6afae633
CY		 1528
1529next:
1530 need_wait = false;
63a94fa1
CY		 1531
1532 mutex_lock(&dcc->cmd_lock);
1533 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1534 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1535 continue;
78997b56 1536 if (dc->len < dpolicy->granularity)
8412663d 1537 continue;
78997b56 1538 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1539 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1540 if (!dc->error)
1541 trimmed += dc->len;
63a94fa1 1542 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1543 } else {
1544 dc->ref++;
1545 need_wait = true;
1546 break;
63a94fa1
CY		 1547 }
1548 }
1549 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1550
1551 if (need_wait) {
0ea80512 1552 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1553 goto next;
1554 }
0ea80512
CY		 1555
1556 return trimmed;
63a94fa1
CY		 1557}
1558
01f9cf6d 1559static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
78997b56 1560 struct discard_policy *dpolicy)
8412663d 1561{
9a997188 1562 struct discard_policy dp;
01f9cf6d 1563 unsigned int discard_blks;
9a997188 1564
01f9cf6d
CY		 1565 if (dpolicy)
1566 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
9a997188
JK		 1567
1568 /* wait all */
8bb4f253 1569 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
01f9cf6d 1570 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
8bb4f253 1571 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
01f9cf6d
CY		 1572 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1573
1574 return discard_blks;
8412663d
CY		 1575}
1576
004b6862 1577/* This should be covered by global mutex, &sit_i->sentry_lock */
94b1e10e 1578static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
004b6862
CY		 1579{
1580 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1581 struct discard_cmd *dc;
ec9895ad 1582 bool need_wait = false;
004b6862
CY		 1583
1584 mutex_lock(&dcc->cmd_lock);
4d57b86d
CY		 1585 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1586 NULL, blkaddr);
004b6862 1587 if (dc) {
ec9895ad
CY		 1588 if (dc->state == D_PREP) {
1589 __punch_discard_cmd(sbi, dc, blkaddr);
1590 } else {
1591 dc->ref++;
1592 need_wait = true;
1593 }
275b66b0 1594 }
d431413f 1595 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1596
2a510c00
CY		 1597 if (need_wait)
1598 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1599}
1600
4d57b86d 1601void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
cce13252
CY		 1602{
1603 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1604
1605 if (dcc && dcc->f2fs_issue_discard) {
1606 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1607
1608 dcc->f2fs_issue_discard = NULL;
1609 kthread_stop(discard_thread);
ec9895ad 1610 }
d431413f
CY		 1611}
1612
8412663d 1613/* This comes from f2fs_put_super */
cf5c759f 1614bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1615{
1616 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1617 struct discard_policy dpolicy;
cf5c759f 1618 bool dropped;
969d1b18 1619
8bb4f253
JK		 1620 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1621 dcc->discard_granularity);
78997b56 1622 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1623 dropped = __drop_discard_cmd(sbi);
cf5c759f 1624
9a997188
JK		 1625 /* just to make sure there is no pending discard commands */
1626 __wait_all_discard_cmd(sbi, NULL);
2482c432
CY		 1627
1628 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
cf5c759f 1629 return dropped;
969d1b18
CY		 1630}
1631
15469963
JK		 1632static int issue_discard_thread(void *data)
1633{
1634 struct f2fs_sb_info *sbi = data;
1635 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1636 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1637 struct discard_policy dpolicy;
969d1b18
CY		 1638 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1639 int issued;
15469963 1640
1d7be270 1641 set_freezable();
15469963 1642
1d7be270 1643 do {
8bb4f253 1644 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
78997b56
CY		 1645 dcc->discard_granularity);
1646
969d1b18
CY		 1647 wait_event_interruptible_timeout(*q,
1648 kthread_should_stop() || freezing(current) ||
1649 dcc->discard_wake,
1650 msecs_to_jiffies(wait_ms));
35a9a766
SY		 1651
1652 if (dcc->discard_wake)
1653 dcc->discard_wake = 0;
1654
76c7bfb3
JK		 1655 /* clean up pending candidates before going to sleep */
1656 if (atomic_read(&dcc->queued_discard))
1657 __wait_all_discard_cmd(sbi, NULL);
1658
1d7be270
JK		 1659 if (try_to_freeze())
1660 continue;
3b60d802
CY		 1661 if (f2fs_readonly(sbi->sb))
1662 continue;
1d7be270
JK		 1663 if (kthread_should_stop())
1664 return 0;
d6184774
YH		 1665 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1666 wait_ms = dpolicy.max_interval;
1667 continue;
1668 }
15469963 1669
5b0e9539 1670 if (sbi->gc_mode == GC_URGENT)
8bb4f253 1671 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
969d1b18 1672
dc6febb6
CY		 1673 sb_start_intwrite(sbi->sb);
1674
78997b56 1675 issued = __issue_discard_cmd(sbi, &dpolicy);
f9d1dced 1676 if (issued > 0) {
78997b56
CY		 1677 __wait_all_discard_cmd(sbi, &dpolicy);
1678 wait_ms = dpolicy.min_interval;
f9d1dced 1679 } else if (issued == -1){
a7d10cf3
ST		 1680 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1681 if (!wait_ms)
abde73c7 1682 wait_ms = dpolicy.mid_interval;
969d1b18 1683 } else {
78997b56 1684 wait_ms = dpolicy.max_interval;
969d1b18 1685 }
1d7be270 1686
dc6febb6 1687 sb_end_intwrite(sbi->sb);
1d7be270 1688
1d7be270
JK		 1689 } while (!kthread_should_stop());
1690 return 0;
15469963
JK		 1691}
1692
f46e8809 1693#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1694static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1695 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1696{
92592285 1697 sector_t sector, nr_sects;
10a875f8 1698 block_t lblkstart = blkstart;
3c62be17
JK		 1699 int devi = 0;
1700
1701 if (sbi->s_ndevs) {
1702 devi = f2fs_target_device_index(sbi, blkstart);
1703 blkstart -= FDEV(devi).start_blk;
1704 }
f46e8809
DLM		 1705
1706 /*
1707 * We need to know the type of the zone: for conventional zones,
1708 * use regular discard if the drive supports it. For sequential
1709 * zones, reset the zone write pointer.
1710 */
3c62be17 1711 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1712
1713 case BLK_ZONE_TYPE_CONVENTIONAL:
1714 if (!blk_queue_discard(bdev_get_queue(bdev)))
1715 return 0;
c81abe34 1716 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1717 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1718 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1719 sector = SECTOR_FROM_BLOCK(blkstart);
1720 nr_sects = SECTOR_FROM_BLOCK(blklen);
1721
1722 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1723 nr_sects != bdev_zone_sectors(bdev)) {
1724 f2fs_msg(sbi->sb, KERN_INFO,
1725 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1726 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1727 blkstart, blklen);
1728 return -EIO;
1729 }
d50aaeec 1730 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1731 return blkdev_reset_zones(bdev, sector,
1732 nr_sects, GFP_NOFS);
1733 default:
1734 /* Unknown zone type: broken device ? */
1735 return -EIO;
1736 }
1737}
1738#endif
1739
3c62be17
JK		 1740static int __issue_discard_async(struct f2fs_sb_info *sbi,
1741 struct block_device *bdev, block_t blkstart, block_t blklen)
1742{
1743#ifdef CONFIG_BLK_DEV_ZONED
7beb01f7 1744 if (f2fs_sb_has_blkzoned(sbi) &&
3c62be17
JK		 1745 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1746 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1747#endif
c81abe34 1748 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1749}
1750
1e87a78d 1751static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1752 block_t blkstart, block_t blklen)
1753{
3c62be17
JK		 1754 sector_t start = blkstart, len = 0;
1755 struct block_device *bdev;
a66cdd98
JK		 1756 struct seg_entry *se;
1757 unsigned int offset;
1758 block_t i;
3c62be17
JK		 1759 int err = 0;
1760
1761 bdev = f2fs_target_device(sbi, blkstart, NULL);
1762
1763 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1764 if (i != start) {
1765 struct block_device *bdev2 =
1766 f2fs_target_device(sbi, i, NULL);
1767
1768 if (bdev2 != bdev) {
1769 err = __issue_discard_async(sbi, bdev,
1770 start, len);
1771 if (err)
1772 return err;
1773 bdev = bdev2;
1774 start = i;
1775 len = 0;
1776 }
1777 }
a66cdd98 1778
a66cdd98
JK		 1779 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1780 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1781
1782 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1783 sbi->discard_blks--;
1784 }
f46e8809 1785
3c62be17
JK		 1786 if (len)
1787 err = __issue_discard_async(sbi, bdev, start, len);
1788 return err;
1e87a78d
JK		 1789}
1790
25290fa5
JK		 1791static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1792 bool check_only)
adf4983b 1793{
b2955550
JK		 1794 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1795 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1796 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1797 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1798 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1799 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1800 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1801 unsigned int start = 0, end = -1;
c473f1a9 1802 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1803 struct discard_entry *de = NULL;
46f84c2c 1804 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1805 int i;
1806
7d20c8ab 1807 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
25290fa5 1808 return false;
b2955550 1809
a66cdd98 1810 if (!force) {
7d20c8ab 1811 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
0b54fb84
JK		 1812 SM_I(sbi)->dcc_info->nr_discards >=
1813 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1814 return false;
4b2fecc8
JK		 1815 }
1816
b2955550
JK		 1817 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1818 for (i = 0; i < entries; i++)
a66cdd98 1819 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1820 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1821
0b54fb84
JK		 1822 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1823 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1824 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1825 if (start >= max_blocks)
1826 break;
1827
1828 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1829 if (force && start && end != max_blocks
1830 && (end - start) < cpc->trim_minlen)
1831 continue;
1832
25290fa5
JK		 1833 if (check_only)
1834 return true;
1835
a7eeb823
CY		 1836 if (!de) {
1837 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1838 GFP_F2FS_ZERO);
1839 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1840 list_add_tail(&de->list, head);
1841 }
1842
1843 for (i = start; i < end; i++)
1844 __set_bit_le(i, (void *)de->discard_map);
1845
1846 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1847 }
25290fa5 1848 return false;
b2955550
JK		 1849}
1850
af8ff65b
CY		 1851static void release_discard_addr(struct discard_entry *entry)
1852{
1853 list_del(&entry->list);
1854 kmem_cache_free(discard_entry_slab, entry);
1855}
1856
4d57b86d 1857void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
4b2fecc8 1858{
46f84c2c 1859 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1860 struct discard_entry *entry, *this;
1861
1862 /* drop caches */
af8ff65b
CY		 1863 list_for_each_entry_safe(entry, this, head, list)
1864 release_discard_addr(entry);
4b2fecc8
JK		 1865}
1866
0a8165d7 1867/*
4d57b86d 1868 * Should call f2fs_clear_prefree_segments after checkpoint is done.
351df4b2
JK		 1869 */
1870static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1871{
1872 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1873 unsigned int segno;
351df4b2
JK		 1874
1875 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1876 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1877 __set_test_and_free(sbi, segno);
351df4b2
JK		 1878 mutex_unlock(&dirty_i->seglist_lock);
1879}
1880
4d57b86d
CY		 1881void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1882 struct cp_control *cpc)
351df4b2 1883{
969d1b18
CY		 1884 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1885 struct list_head *head = &dcc->entry_list;
2d7b822a 1886 struct discard_entry *entry, *this;
351df4b2 1887 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1888 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1889 unsigned int start = 0, end = -1;
36abef4e 1890 unsigned int secno, start_segno;
c473f1a9 1891 bool force = (cpc->reason & CP_DISCARD);
2c70c5e3 1892 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
351df4b2
JK		 1893
1894 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1895
351df4b2 1896 while (1) {
29e59c14 1897 int i;
ad6672bb
YS		 1898
1899 if (need_align && end != -1)
1900 end--;
7cd8558b
JK		 1901 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1902 if (start >= MAIN_SEGS(sbi))
351df4b2 1903 break;
7cd8558b
JK		 1904 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1905 start + 1);
29e59c14 1906
ad6672bb
YS		 1907 if (need_align) {
1908 start = rounddown(start, sbi->segs_per_sec);
1909 end = roundup(end, sbi->segs_per_sec);
1910 }
29e59c14 1911
ad6672bb
YS		 1912 for (i = start; i < end; i++) {
1913 if (test_and_clear_bit(i, prefree_map))
1914 dirty_i->nr_dirty[PRE]--;
1915 }
29e59c14 1916
7d20c8ab 1917 if (!f2fs_realtime_discard_enable(sbi))
29e59c14 1918 continue;
351df4b2 1919
650d3c4e
YH		 1920 if (force && start >= cpc->trim_start &&
1921 (end - 1) <= cpc->trim_end)
1922 continue;
1923
2c70c5e3 1924 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
36abef4e 1925 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1926 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1927 continue;
1928 }
1929next:
4ddb1a4d
JK		 1930 secno = GET_SEC_FROM_SEG(sbi, start);
1931 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1932 if (!IS_CURSEC(sbi, secno) &&
302bd348 1933 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1934 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1935 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1936
1937 start = start_segno + sbi->segs_per_sec;
1938 if (start < end)
1939 goto next;
8b107f5b
JK		 1940 else
1941 end = start - 1;
351df4b2
JK		 1942 }
1943 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1944
1945 /* send small discards */
2d7b822a 1946 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1947 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1948 bool is_valid = test_bit_le(0, entry->discard_map);
1949
1950find_next:
1951 if (is_valid) {
1952 next_pos = find_next_zero_bit_le(entry->discard_map,
1953 sbi->blocks_per_seg, cur_pos);
1954 len = next_pos - cur_pos;
1955
7beb01f7 1956 if (f2fs_sb_has_blkzoned(sbi) ||
acfd2810 1957 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1958 goto skip;
1959
1960 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1961 len);
a7eeb823
CY		 1962 total_len += len;
1963 } else {
1964 next_pos = find_next_bit_le(entry->discard_map,
1965 sbi->blocks_per_seg, cur_pos);
1966 }
836b5a63 1967skip:
a7eeb823
CY		 1968 cur_pos = next_pos;
1969 is_valid = !is_valid;
1970
1971 if (cur_pos < sbi->blocks_per_seg)
1972 goto find_next;
1973
af8ff65b 1974 release_discard_addr(entry);
969d1b18 1975 dcc->nr_discards -= total_len;
b2955550 1976 }
34e159da 1977
01983c71 1978 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1979}
1980
8ed59745 1981static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1982{
15469963 1983 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1984 struct discard_cmd_control *dcc;
ba48a33e 1985 int err = 0, i;
0b54fb84
JK		 1986
1987 if (SM_I(sbi)->dcc_info) {
1988 dcc = SM_I(sbi)->dcc_info;
1989 goto init_thread;
1990 }
1991
acbf054d 1992 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
0b54fb84
JK		 1993 if (!dcc)
1994 return -ENOMEM;
1995
969d1b18 1996 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1997 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1998 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1999 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 2000 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 2001 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 2002 mutex_init(&dcc->cmd_lock);
8b8dd65f 2003 atomic_set(&dcc->issued_discard, 0);
72691af6 2004 atomic_set(&dcc->queued_discard, 0);
5f32366a 2005 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 2006 dcc->nr_discards = 0;
d618ebaf 2007 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 2008 dcc->undiscard_blks = 0;
20ee4382 2009 dcc->next_pos = 0;
4dada3fd 2010 dcc->root = RB_ROOT_CACHED;
67fce70b 2011 dcc->rbtree_check = false;
0b54fb84 2012
15469963 2013 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 2014 SM_I(sbi)->dcc_info = dcc;
2015init_thread:
15469963
JK		 2016 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2017 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2018 if (IS_ERR(dcc->f2fs_issue_discard)) {
2019 err = PTR_ERR(dcc->f2fs_issue_discard);
5222595d 2020 kvfree(dcc);
15469963
JK		 2021 SM_I(sbi)->dcc_info = NULL;
2022 return err;
2023 }
2024
0b54fb84
JK		 2025 return err;
2026}
2027
f099405f 2028static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 2029{
2030 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2031
f099405f
CY		 2032 if (!dcc)
2033 return;
2034
4d57b86d 2035 f2fs_stop_discard_thread(sbi);
f099405f 2036
5222595d 2037 kvfree(dcc);
f099405f 2038 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 2039}
2040
184a5cd2 2041static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 2042{
2043 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 2044
2045 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 2046 sit_i->dirty_sentries++;
184a5cd2
CY		 2047 return false;
2048 }
2049
2050 return true;
351df4b2
JK		 2051}
2052
2053static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2054 unsigned int segno, int modified)
2055{
2056 struct seg_entry *se = get_seg_entry(sbi, segno);
2057 se->type = type;
2058 if (modified)
2059 __mark_sit_entry_dirty(sbi, segno);
2060}
2061
2062static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2063{
2064 struct seg_entry *se;
2065 unsigned int segno, offset;
2066 long int new_vblocks;
6415fedc
YS		 2067 bool exist;
2068#ifdef CONFIG_F2FS_CHECK_FS
2069 bool mir_exist;
2070#endif
351df4b2
JK		 2071
2072 segno = GET_SEGNO(sbi, blkaddr);
2073
2074 se = get_seg_entry(sbi, segno);
2075 new_vblocks = se->valid_blocks + del;
491c0854 2076 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 2077
9850cf4a 2078 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 2079 (new_vblocks > sbi->blocks_per_seg)));
2080
2081 se->valid_blocks = new_vblocks;
a1f72ac2
CY		 2082 se->mtime = get_mtime(sbi, false);
2083 if (se->mtime > SIT_I(sbi)->max_mtime)
2084 SIT_I(sbi)->max_mtime = se->mtime;
351df4b2
JK		 2085
2086 /* Update valid block bitmap */
2087 if (del > 0) {
6415fedc 2088 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 2089#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2090 mir_exist = f2fs_test_and_set_bit(offset,
2091 se->cur_valid_map_mir);
2092 if (unlikely(exist != mir_exist)) {
2093 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2094 "when setting bitmap, blk:%u, old bit:%d",
2095 blkaddr, exist);
05796763 2096 f2fs_bug_on(sbi, 1);
6415fedc 2097 }
355e7891 2098#endif
6415fedc
YS		 2099 if (unlikely(exist)) {
2100 f2fs_msg(sbi->sb, KERN_ERR,
2101 "Bitmap was wrongly set, blk:%u", blkaddr);
2102 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2103 se->valid_blocks--;
2104 del = 0;
355e7891 2105 }
6415fedc 2106
7d20c8ab 2107 if (!f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 2108 sbi->discard_blks--;
720037f9
JK		 2109
2110 /* don't overwrite by SSR to keep node chain */
4354994f
DR		 2111 if (IS_NODESEG(se->type) &&
2112 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
720037f9
JK		 2113 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2114 se->ckpt_valid_blocks++;
2115 }
351df4b2 2116 } else {
6415fedc 2117 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 2118#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 2119 mir_exist = f2fs_test_and_clear_bit(offset,
2120 se->cur_valid_map_mir);
2121 if (unlikely(exist != mir_exist)) {
2122 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2123 "when clearing bitmap, blk:%u, old bit:%d",
2124 blkaddr, exist);
05796763 2125 f2fs_bug_on(sbi, 1);
6415fedc 2126 }
355e7891 2127#endif
6415fedc
YS		 2128 if (unlikely(!exist)) {
2129 f2fs_msg(sbi->sb, KERN_ERR,
2130 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2131 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 2132 se->valid_blocks++;
2133 del = 0;
4354994f
DR		 2134 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2135 /*
2136 * If checkpoints are off, we must not reuse data that
2137 * was used in the previous checkpoint. If it was used
2138 * before, we must track that to know how much space we
2139 * really have.
2140 */
2141 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2142 sbi->unusable_block_count++;
355e7891 2143 }
6415fedc 2144
7d20c8ab 2145 if (f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 2146 sbi->discard_blks++;
351df4b2
JK		 2147 }
2148 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2149 se->ckpt_valid_blocks += del;
2150
2151 __mark_sit_entry_dirty(sbi, segno);
2152
2153 /* update total number of valid blocks to be written in ckpt area */
2154 SIT_I(sbi)->written_valid_blocks += del;
2155
2c70c5e3 2156 if (__is_large_section(sbi))
351df4b2
JK		 2157 get_sec_entry(sbi, segno)->valid_blocks += del;
2158}
2159
4d57b86d 2160void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
351df4b2
JK		 2161{
2162 unsigned int segno = GET_SEGNO(sbi, addr);
2163 struct sit_info *sit_i = SIT_I(sbi);
2164
9850cf4a 2165 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 2166 if (addr == NEW_ADDR)
2167 return;
2168
6aa58d8a
CY		 2169 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2170
351df4b2 2171 /* add it into sit main buffer */
3d26fa6b 2172 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2173
2174 update_sit_entry(sbi, addr, -1);
2175
2176 /* add it into dirty seglist */
2177 locate_dirty_segment(sbi, segno);
2178
3d26fa6b 2179 up_write(&sit_i->sentry_lock);
351df4b2
JK		 2180}
2181
4d57b86d 2182bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
6e2c64ad
JK		 2183{
2184 struct sit_info *sit_i = SIT_I(sbi);
2185 unsigned int segno, offset;
2186 struct seg_entry *se;
2187 bool is_cp = false;
2188
e1da7872 2189 if (!is_valid_data_blkaddr(sbi, blkaddr))
6e2c64ad
JK		 2190 return true;
2191
3d26fa6b 2192 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2193
2194 segno = GET_SEGNO(sbi, blkaddr);
2195 se = get_seg_entry(sbi, segno);
2196 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2197
2198 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2199 is_cp = true;
2200
3d26fa6b 2201 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 2202
2203 return is_cp;
2204}
2205
0a8165d7 2206/*
351df4b2
JK		 2207 * This function should be resided under the curseg_mutex lock
2208 */
2209static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 2210 struct f2fs_summary *sum)
351df4b2
JK		 2211{
2212 struct curseg_info *curseg = CURSEG_I(sbi, type);
2213 void *addr = curseg->sum_blk;
e79efe3b 2214 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 2215 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 2216}
2217
0a8165d7 2218/*
351df4b2
JK		 2219 * Calculate the number of current summary pages for writing
2220 */
4d57b86d 2221int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 2222{
351df4b2 2223 int valid_sum_count = 0;
9a47938b 2224 int i, sum_in_page;
351df4b2
JK		 2225
2226 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2227 if (sbi->ckpt->alloc_type[i] == SSR)
2228 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 2229 else {
2230 if (for_ra)
2231 valid_sum_count += le16_to_cpu(
2232 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2233 else
2234 valid_sum_count += curseg_blkoff(sbi, i);
2235 }
351df4b2
JK		 2236 }
2237
09cbfeaf 2238 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 2239 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2240 if (valid_sum_count <= sum_in_page)
351df4b2 2241 return 1;
9a47938b 2242 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 2243 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 2244 return 2;
2245 return 3;
2246}
2247
0a8165d7 2248/*
351df4b2
JK		 2249 * Caller should put this summary page
2250 */
4d57b86d 2251struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2 2252{
7735730d 2253 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2254}
2255
4d57b86d
CY		 2256void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2257 void *src, block_t blk_addr)
351df4b2 2258{
4d57b86d 2259 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
381722d2 2260
0537b811 2261 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2262 set_page_dirty(page);
2263 f2fs_put_page(page, 1);
2264}
2265
381722d2
CY		 2266static void write_sum_page(struct f2fs_sb_info *sbi,
2267 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2268{
4d57b86d 2269 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
381722d2
CY		 2270}
2271
b7ad7512
CY		 2272static void write_current_sum_page(struct f2fs_sb_info *sbi,
2273 int type, block_t blk_addr)
2274{
2275 struct curseg_info *curseg = CURSEG_I(sbi, type);
4d57b86d 2276 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
b7ad7512
CY		 2277 struct f2fs_summary_block *src = curseg->sum_blk;
2278 struct f2fs_summary_block *dst;
2279
2280 dst = (struct f2fs_summary_block *)page_address(page);
81114baa 2281 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2282
2283 mutex_lock(&curseg->curseg_mutex);
2284
2285 down_read(&curseg->journal_rwsem);
2286 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2287 up_read(&curseg->journal_rwsem);
2288
2289 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2290 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2291
2292 mutex_unlock(&curseg->curseg_mutex);
2293
2294 set_page_dirty(page);
2295 f2fs_put_page(page, 1);
2296}
2297
a7881893
JK		 2298static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2299{
2300 struct curseg_info *curseg = CURSEG_I(sbi, type);
2301 unsigned int segno = curseg->segno + 1;
2302 struct free_segmap_info *free_i = FREE_I(sbi);
2303
2304 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2305 return !test_bit(segno, free_i->free_segmap);
2306 return 0;
2307}
2308
0a8165d7 2309/*
351df4b2
JK		 2310 * Find a new segment from the free segments bitmap to right order
2311 * This function should be returned with success, otherwise BUG
2312 */
2313static void get_new_segment(struct f2fs_sb_info *sbi,
2314 unsigned int *newseg, bool new_sec, int dir)
2315{
2316 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2317 unsigned int segno, secno, zoneno;
7cd8558b 2318 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2319 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2320 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2321 unsigned int left_start = hint;
2322 bool init = true;
2323 int go_left = 0;
2324 int i;
2325
1a118ccf 2326 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2327
2328 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2329 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2330 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2331 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2332 goto got_it;
2333 }
2334find_other_zone:
7cd8558b
JK		 2335 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2336 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2337 if (dir == ALLOC_RIGHT) {
2338 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2339 MAIN_SECS(sbi), 0);
2340 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2341 } else {
2342 go_left = 1;
2343 left_start = hint - 1;
2344 }
2345 }
2346 if (go_left == 0)
2347 goto skip_left;
2348
2349 while (test_bit(left_start, free_i->free_secmap)) {
2350 if (left_start > 0) {
2351 left_start--;
2352 continue;
2353 }
2354 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2355 MAIN_SECS(sbi), 0);
2356 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2357 break;
2358 }
2359 secno = left_start;
2360skip_left:
4ddb1a4d
JK		 2361 segno = GET_SEG_FROM_SEC(sbi, secno);
2362 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2363
2364 /* give up on finding another zone */
2365 if (!init)
2366 goto got_it;
2367 if (sbi->secs_per_zone == 1)
2368 goto got_it;
2369 if (zoneno == old_zoneno)
2370 goto got_it;
2371 if (dir == ALLOC_LEFT) {
2372 if (!go_left && zoneno + 1 >= total_zones)
2373 goto got_it;
2374 if (go_left && zoneno == 0)
2375 goto got_it;
2376 }
2377 for (i = 0; i < NR_CURSEG_TYPE; i++)
2378 if (CURSEG_I(sbi, i)->zone == zoneno)
2379 break;
2380
2381 if (i < NR_CURSEG_TYPE) {
2382 /* zone is in user, try another */
2383 if (go_left)
2384 hint = zoneno * sbi->secs_per_zone - 1;
2385 else if (zoneno + 1 >= total_zones)
2386 hint = 0;
2387 else
2388 hint = (zoneno + 1) * sbi->secs_per_zone;
2389 init = false;
2390 goto find_other_zone;
2391 }
2392got_it:
2393 /* set it as dirty segment in free segmap */
9850cf4a 2394 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2395 __set_inuse(sbi, segno);
2396 *newseg = segno;
1a118ccf 2397 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2398}
2399
2400static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2401{
2402 struct curseg_info *curseg = CURSEG_I(sbi, type);
2403 struct summary_footer *sum_footer;
2404
2405 curseg->segno = curseg->next_segno;
4ddb1a4d 2406 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2407 curseg->next_blkoff = 0;
2408 curseg->next_segno = NULL_SEGNO;
2409
2410 sum_footer = &(curseg->sum_blk->footer);
2411 memset(sum_footer, 0, sizeof(struct summary_footer));
2412 if (IS_DATASEG(type))
2413 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2414 if (IS_NODESEG(type))
2415 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2416 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2417}
2418
7a20b8a6
JK		 2419static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2420{
a7881893 2421 /* if segs_per_sec is large than 1, we need to keep original policy. */
2c70c5e3 2422 if (__is_large_section(sbi))
a7881893
JK		 2423 return CURSEG_I(sbi, type)->segno;
2424
4354994f
DR		 2425 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2426 return 0;
2427
b94929d9
YS		 2428 if (test_opt(sbi, NOHEAP) &&
2429 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2430 return 0;
2431
e066b83c
JK		 2432 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2433 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
07939627
JK		 2434
2435 /* find segments from 0 to reuse freed segments */
63189b78 2436 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
07939627
JK		 2437 return 0;
2438
7a20b8a6
JK		 2439 return CURSEG_I(sbi, type)->segno;
2440}
2441
0a8165d7 2442/*
351df4b2
JK		 2443 * Allocate a current working segment.
2444 * This function always allocates a free segment in LFS manner.
2445 */
2446static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2447{
2448 struct curseg_info *curseg = CURSEG_I(sbi, type);
2449 unsigned int segno = curseg->segno;
2450 int dir = ALLOC_LEFT;
2451
2452 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2453 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2454 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2455 dir = ALLOC_RIGHT;
2456
2457 if (test_opt(sbi, NOHEAP))
2458 dir = ALLOC_RIGHT;
2459
7a20b8a6 2460 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2461 get_new_segment(sbi, &segno, new_sec, dir);
2462 curseg->next_segno = segno;
2463 reset_curseg(sbi, type, 1);
2464 curseg->alloc_type = LFS;
2465}
2466
2467static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2468 struct curseg_info *seg, block_t start)
2469{
2470 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2471 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2472 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2473 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2474 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2475 int i, pos;
2476
2477 for (i = 0; i < entries; i++)
2478 target_map[i] = ckpt_map[i] | cur_map[i];
2479
2480 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2481
2482 seg->next_blkoff = pos;
351df4b2
JK		 2483}
2484
0a8165d7 2485/*
351df4b2
JK		 2486 * If a segment is written by LFS manner, next block offset is just obtained
2487 * by increasing the current block offset. However, if a segment is written by
2488 * SSR manner, next block offset obtained by calling __next_free_blkoff
2489 */
2490static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2491 struct curseg_info *seg)
2492{
2493 if (seg->alloc_type == SSR)
2494 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2495 else
2496 seg->next_blkoff++;
2497}
2498
0a8165d7 2499/*
e1c42045 2500 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2501 * manner, so it should recover the existing segment information of valid blocks
2502 */
025d63a4 2503static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2504{
2505 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2506 struct curseg_info *curseg = CURSEG_I(sbi, type);
2507 unsigned int new_segno = curseg->next_segno;
2508 struct f2fs_summary_block *sum_node;
2509 struct page *sum_page;
2510
2511 write_sum_page(sbi, curseg->sum_blk,
2512 GET_SUM_BLOCK(sbi, curseg->segno));
2513 __set_test_and_inuse(sbi, new_segno);
2514
2515 mutex_lock(&dirty_i->seglist_lock);
2516 __remove_dirty_segment(sbi, new_segno, PRE);
2517 __remove_dirty_segment(sbi, new_segno, DIRTY);
2518 mutex_unlock(&dirty_i->seglist_lock);
2519
2520 reset_curseg(sbi, type, 1);
2521 curseg->alloc_type = SSR;
2522 __next_free_blkoff(sbi, curseg, 0);
2523
4d57b86d 2524 sum_page = f2fs_get_sum_page(sbi, new_segno);
edc55aaf 2525 f2fs_bug_on(sbi, IS_ERR(sum_page));
025d63a4
CY		 2526 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2527 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2528 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2529}
2530
43727527
JK		 2531static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2532{
2533 struct curseg_info *curseg = CURSEG_I(sbi, type);
2534 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2535 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2536 int i, cnt;
2537 bool reversed = false;
c192f7a4 2538
4d57b86d 2539 /* f2fs_need_SSR() already forces to do this */
e066b83c
JK		 2540 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2541 curseg->next_segno = segno;
c192f7a4 2542 return 1;
e066b83c 2543 }
43727527 2544
70d625cb
JK		 2545 /* For node segments, let's do SSR more intensively */
2546 if (IS_NODESEG(type)) {
d27c3d89
CY		 2547 if (type >= CURSEG_WARM_NODE) {
2548 reversed = true;
2549 i = CURSEG_COLD_NODE;
2550 } else {
2551 i = CURSEG_HOT_NODE;
2552 }
2553 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2554 } else {
d27c3d89
CY		 2555 if (type >= CURSEG_WARM_DATA) {
2556 reversed = true;
2557 i = CURSEG_COLD_DATA;
2558 } else {
2559 i = CURSEG_HOT_DATA;
2560 }
2561 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2562 }
43727527 2563
d27c3d89 2564 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2565 if (i == type)
2566 continue;
e066b83c
JK		 2567 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2568 curseg->next_segno = segno;
43727527 2569 return 1;
e066b83c 2570 }
c192f7a4 2571 }
4354994f
DR		 2572
2573 /* find valid_blocks=0 in dirty list */
2574 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2575 segno = get_free_segment(sbi);
2576 if (segno != NULL_SEGNO) {
2577 curseg->next_segno = segno;
2578 return 1;
2579 }
2580 }
43727527
JK		 2581 return 0;
2582}
2583
351df4b2
JK		 2584/*
2585 * flush out current segment and replace it with new segment
2586 * This function should be returned with success, otherwise BUG
2587 */
2588static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2589 int type, bool force)
2590{
a7881893
JK		 2591 struct curseg_info *curseg = CURSEG_I(sbi, type);
2592
7b405275 2593 if (force)
351df4b2 2594 new_curseg(sbi, type, true);
5b6c6be2
JK		 2595 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2596 type == CURSEG_WARM_NODE)
351df4b2 2597 new_curseg(sbi, type, false);
4354994f
DR		 2598 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2599 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
a7881893 2600 new_curseg(sbi, type, false);
4d57b86d 2601 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2602 change_curseg(sbi, type);
351df4b2
JK		 2603 else
2604 new_curseg(sbi, type, false);
dcdfff65 2605
a7881893 2606 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2607}
2608
4d57b86d 2609void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
351df4b2 2610{
6ae1be13
JK		 2611 struct curseg_info *curseg;
2612 unsigned int old_segno;
351df4b2
JK		 2613 int i;
2614
3d26fa6b
CY		 2615 down_write(&SIT_I(sbi)->sentry_lock);
2616
6ae1be13
JK		 2617 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2618 curseg = CURSEG_I(sbi, i);
2619 old_segno = curseg->segno;
2620 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2621 locate_dirty_segment(sbi, old_segno);
2622 }
3d26fa6b
CY		 2623
2624 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2625}
2626
2627static const struct segment_allocation default_salloc_ops = {
2628 .allocate_segment = allocate_segment_by_default,
2629};
2630
4d57b86d
CY		 2631bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2632 struct cp_control *cpc)
25290fa5
JK		 2633{
2634 __u64 trim_start = cpc->trim_start;
2635 bool has_candidate = false;
2636
3d26fa6b 2637 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2638 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2639 if (add_discard_addrs(sbi, cpc, true)) {
2640 has_candidate = true;
2641 break;
2642 }
2643 }
3d26fa6b 2644 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2645
2646 cpc->trim_start = trim_start;
2647 return has_candidate;
2648}
2649
01f9cf6d 2650static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
9a997188
JK		 2651 struct discard_policy *dpolicy,
2652 unsigned int start, unsigned int end)
2653{
2654 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2655 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2656 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2657 struct discard_cmd *dc;
2658 struct blk_plug plug;
2659 int issued;
01f9cf6d 2660 unsigned int trimmed = 0;
9a997188
JK		 2661
2662next:
2663 issued = 0;
2664
2665 mutex_lock(&dcc->cmd_lock);
67fce70b
CY		 2666 if (unlikely(dcc->rbtree_check))
2667 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2668 &dcc->root));
9a997188 2669
4d57b86d 2670 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
9a997188
JK		 2671 NULL, start,
2672 (struct rb_entry **)&prev_dc,
2673 (struct rb_entry **)&next_dc,
4dada3fd 2674 &insert_p, &insert_parent, true, NULL);
9a997188
JK		 2675 if (!dc)
2676 dc = next_dc;
2677
2678 blk_start_plug(&plug);
2679
2680 while (dc && dc->lstart <= end) {
2681 struct rb_node *node;
6b9cb124 2682 int err = 0;
9a997188
JK		 2683
2684 if (dc->len < dpolicy->granularity)
2685 goto skip;
2686
2687 if (dc->state != D_PREP) {
2688 list_move_tail(&dc->list, &dcc->fstrim_list);
2689 goto skip;
2690 }
2691
6b9cb124 2692 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
9a997188 2693
35ec7d57 2694 if (issued >= dpolicy->max_requests) {
9a997188
JK		 2695 start = dc->lstart + dc->len;
2696
6b9cb124
CY		 2697 if (err)
2698 __remove_discard_cmd(sbi, dc);
2699
9a997188
JK		 2700 blk_finish_plug(&plug);
2701 mutex_unlock(&dcc->cmd_lock);
01f9cf6d 2702 trimmed += __wait_all_discard_cmd(sbi, NULL);
9a997188
JK		 2703 congestion_wait(BLK_RW_ASYNC, HZ/50);
2704 goto next;
2705 }
2706skip:
2707 node = rb_next(&dc->rb_node);
6b9cb124
CY		 2708 if (err)
2709 __remove_discard_cmd(sbi, dc);
9a997188
JK		 2710 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2711
2712 if (fatal_signal_pending(current))
2713 break;
2714 }
2715
2716 blk_finish_plug(&plug);
2717 mutex_unlock(&dcc->cmd_lock);
01f9cf6d
CY		 2718
2719 return trimmed;
9a997188
JK		 2720}
2721
4b2fecc8
JK		 2722int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2723{
f7ef9b83
JK		 2724 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2725 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
377224c4 2726 unsigned int start_segno, end_segno;
8412663d 2727 block_t start_block, end_block;
4b2fecc8 2728 struct cp_control cpc;
78997b56 2729 struct discard_policy dpolicy;
0ea80512 2730 unsigned long long trimmed = 0;
c34f42e2 2731 int err = 0;
2c70c5e3 2732 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
4b2fecc8 2733
836b5a63 2734 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2735 return -EINVAL;
2736
3f16ecd9
CY		 2737 if (end < MAIN_BLKADDR(sbi))
2738 goto out;
4b2fecc8 2739
ed214a11
YH		 2740 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2741 f2fs_msg(sbi->sb, KERN_WARNING,
2742 "Found FS corruption, run fsck to fix.");
3d165dc3 2743 return -EIO;
ed214a11
YH		 2744 }
2745
4b2fecc8 2746 /* start/end segment number in main_area */
7cd8558b
JK		 2747 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2748 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2749 GET_SEGNO(sbi, end);
ad6672bb
YS		 2750 if (need_align) {
2751 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2752 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2753 }
8412663d 2754
4b2fecc8 2755 cpc.reason = CP_DISCARD;
836b5a63 2756 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
377224c4
CY		 2757 cpc.trim_start = start_segno;
2758 cpc.trim_end = end_segno;
4b2fecc8 2759
377224c4
CY		 2760 if (sbi->discard_blks == 0)
2761 goto out;
74fa5f3d 2762
377224c4 2763 mutex_lock(&sbi->gc_mutex);
4d57b86d 2764 err = f2fs_write_checkpoint(sbi, &cpc);
377224c4
CY		 2765 mutex_unlock(&sbi->gc_mutex);
2766 if (err)
2767 goto out;
8412663d 2768
e555da9f
JK		 2769 /*
2770 * We filed discard candidates, but actually we don't need to wait for
2771 * all of them, since they'll be issued in idle time along with runtime
2772 * discard option. User configuration looks like using runtime discard
2773 * or periodic fstrim instead of it.
2774 */
7d20c8ab 2775 if (f2fs_realtime_discard_enable(sbi))
5a615492
JK		 2776 goto out;
2777
2778 start_block = START_BLOCK(sbi, start_segno);
2779 end_block = START_BLOCK(sbi, end_segno + 1);
2780
2781 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
01f9cf6d
CY		 2782 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2783 start_block, end_block);
5a615492 2784
01f9cf6d 2785 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
0ea80512 2786 start_block, end_block);
377224c4 2787out:
6eae2694
CY		 2788 if (!err)
2789 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2790 return err;
4b2fecc8
JK		 2791}
2792
351df4b2
JK		 2793static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2794{
2795 struct curseg_info *curseg = CURSEG_I(sbi, type);
2796 if (curseg->next_blkoff < sbi->blocks_per_seg)
2797 return true;
2798 return false;
2799}
2800
4d57b86d 2801int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
4f0a03d3
HL		 2802{
2803 switch (hint) {
2804 case WRITE_LIFE_SHORT:
2805 return CURSEG_HOT_DATA;
2806 case WRITE_LIFE_EXTREME:
2807 return CURSEG_COLD_DATA;
2808 default:
2809 return CURSEG_WARM_DATA;
2810 }
2811}
2812
0cdd3195
HL		 2813/* This returns write hints for each segment type. This hints will be
2814 * passed down to block layer. There are mapping tables which depend on
2815 * the mount option 'whint_mode'.
2816 *
2817 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2818 *
2819 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2820 *
2821 * User F2FS Block
2822 * ---- ---- -----
2823 * META WRITE_LIFE_NOT_SET
2824 * HOT_NODE "
2825 * WARM_NODE "
2826 * COLD_NODE "
2827 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2828 * extension list " "
2829 *
2830 * -- buffered io
2831 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2832 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2833 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2834 * WRITE_LIFE_NONE " "
2835 * WRITE_LIFE_MEDIUM " "
2836 * WRITE_LIFE_LONG " "
2837 *
2838 * -- direct io
2839 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2840 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2841 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2842 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2843 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2844 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2845 *
f2e703f9
HL		 2846 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2847 *
2848 * User F2FS Block
2849 * ---- ---- -----
2850 * META WRITE_LIFE_MEDIUM;
2851 * HOT_NODE WRITE_LIFE_NOT_SET
2852 * WARM_NODE "
2853 * COLD_NODE WRITE_LIFE_NONE
2854 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2855 * extension list " "
2856 *
2857 * -- buffered io
2858 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2859 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2860 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2861 * WRITE_LIFE_NONE " "
2862 * WRITE_LIFE_MEDIUM " "
2863 * WRITE_LIFE_LONG " "
2864 *
2865 * -- direct io
2866 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2867 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2868 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2869 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2870 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2871 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
0cdd3195
HL		 2872 */
2873
4d57b86d 2874enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
0cdd3195
HL		 2875 enum page_type type, enum temp_type temp)
2876{
63189b78 2877 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
0cdd3195 2878 if (type == DATA) {
f2e703f9 2879 if (temp == WARM)
0cdd3195 2880 return WRITE_LIFE_NOT_SET;
f2e703f9
HL		 2881 else if (temp == HOT)
2882 return WRITE_LIFE_SHORT;
2883 else if (temp == COLD)
2884 return WRITE_LIFE_EXTREME;
0cdd3195
HL		 2885 } else {
2886 return WRITE_LIFE_NOT_SET;
2887 }
63189b78 2888 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
f2e703f9
HL		 2889 if (type == DATA) {
2890 if (temp == WARM)
2891 return WRITE_LIFE_LONG;
2892 else if (temp == HOT)
2893 return WRITE_LIFE_SHORT;
2894 else if (temp == COLD)
2895 return WRITE_LIFE_EXTREME;
2896 } else if (type == NODE) {
2897 if (temp == WARM || temp == HOT)
2898 return WRITE_LIFE_NOT_SET;
2899 else if (temp == COLD)
2900 return WRITE_LIFE_NONE;
2901 } else if (type == META) {
2902 return WRITE_LIFE_MEDIUM;
2903 }
0cdd3195 2904 }
f2e703f9 2905 return WRITE_LIFE_NOT_SET;
0cdd3195
HL		 2906}
2907
81377bd6 2908static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2909{
81377bd6 2910 if (fio->type == DATA)
351df4b2
JK		 2911 return CURSEG_HOT_DATA;
2912 else
2913 return CURSEG_HOT_NODE;
2914}
2915
81377bd6 2916static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2917{
81377bd6
JK		 2918 if (fio->type == DATA) {
2919 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2920
2921 if (S_ISDIR(inode->i_mode))
2922 return CURSEG_HOT_DATA;
2923 else
2924 return CURSEG_COLD_DATA;
2925 } else {
81377bd6 2926 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2927 return CURSEG_WARM_NODE;
351df4b2
JK		 2928 else
2929 return CURSEG_COLD_NODE;
2930 }
2931}
2932
81377bd6 2933static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2934{
81377bd6
JK		 2935 if (fio->type == DATA) {
2936 struct inode *inode = fio->page->mapping->host;
351df4b2 2937
81377bd6 2938 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2939 return CURSEG_COLD_DATA;
b6a06cbb 2940 if (file_is_hot(inode) ||
b4c3ca8b 2941 is_inode_flag_set(inode, FI_HOT_DATA) ||
2079f115
CY		 2942 f2fs_is_atomic_file(inode) ||
2943 f2fs_is_volatile_file(inode))
ef095d19 2944 return CURSEG_HOT_DATA;
4d57b86d 2945 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2946 } else {
81377bd6
JK		 2947 if (IS_DNODE(fio->page))
2948 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2949 CURSEG_HOT_NODE;
ef095d19 2950 return CURSEG_COLD_NODE;
351df4b2
JK		 2951 }
2952}
2953
81377bd6 2954static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2955{
a912b54d
JK		 2956 int type = 0;
2957
63189b78 2958 switch (F2FS_OPTION(fio->sbi).active_logs) {
351df4b2 2959 case 2:
a912b54d
JK		 2960 type = __get_segment_type_2(fio);
2961 break;
351df4b2 2962 case 4:
a912b54d
JK		 2963 type = __get_segment_type_4(fio);
2964 break;
2965 case 6:
2966 type = __get_segment_type_6(fio);
2967 break;
2968 default:
2969 f2fs_bug_on(fio->sbi, true);
351df4b2 2970 }
81377bd6 2971
a912b54d
JK		 2972 if (IS_HOT(type))
2973 fio->temp = HOT;
2974 else if (IS_WARM(type))
2975 fio->temp = WARM;
2976 else
2977 fio->temp = COLD;
2978 return type;
351df4b2
JK		 2979}
2980
4d57b86d 2981void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
bfad7c2d 2982 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2983 struct f2fs_summary *sum, int type,
2984 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2985{
2986 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2987 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2988
2b60311d
CY		 2989 down_read(&SM_I(sbi)->curseg_lock);
2990
351df4b2 2991 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2992 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2993
2994 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2995
4e6a8d9b
JK		 2996 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2997
351df4b2
JK		 2998 /*
2999 * __add_sum_entry should be resided under the curseg_mutex
3000 * because, this function updates a summary entry in the
3001 * current summary block.
3002 */
e79efe3b 3003 __add_sum_entry(sbi, type, sum);
351df4b2 3004
351df4b2 3005 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 3006
3007 stat_inc_block_count(sbi, curseg);
351df4b2 3008
65f1b80b
YS		 3009 /*
3010 * SIT information should be updated before segment allocation,
3011 * since SSR needs latest valid block information.
3012 */
3013 update_sit_entry(sbi, *new_blkaddr, 1);
3014 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3015 update_sit_entry(sbi, old_blkaddr, -1);
3016
c6f82fe9
JK		 3017 if (!__has_curseg_space(sbi, type))
3018 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 3019
351df4b2 3020 /*
65f1b80b
YS		 3021 * segment dirty status should be updated after segment allocation,
3022 * so we just need to update status only one time after previous
3023 * segment being closed.
351df4b2 3024 */
65f1b80b
YS		 3025 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3026 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 3027
3d26fa6b 3028 up_write(&sit_i->sentry_lock);
351df4b2 3029
704956ec 3030 if (page && IS_NODESEG(type)) {
351df4b2
JK		 3031 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3032
704956ec
CY		 3033 f2fs_inode_chksum_set(sbi, page);
3034 }
3035
fb830fc5
CY		 3036 if (add_list) {
3037 struct f2fs_bio_info *io;
3038
3039 INIT_LIST_HEAD(&fio->list);
3040 fio->in_list = true;
fe16efe6 3041 fio->retry = false;
fb830fc5
CY		 3042 io = sbi->write_io[fio->type] + fio->temp;
3043 spin_lock(&io->io_lock);
3044 list_add_tail(&fio->list, &io->io_list);
3045 spin_unlock(&io->io_lock);
3046 }
3047
bfad7c2d 3048 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 3049
3050 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 3051}
3052
39d787be
CY		 3053static void update_device_state(struct f2fs_io_info *fio)
3054{
3055 struct f2fs_sb_info *sbi = fio->sbi;
3056 unsigned int devidx;
3057
3058 if (!sbi->s_ndevs)
3059 return;
3060
3061 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3062
3063 /* update device state for fsync */
4d57b86d 3064 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 3065
3066 /* update device state for checkpoint */
3067 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3068 spin_lock(&sbi->dev_lock);
3069 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3070 spin_unlock(&sbi->dev_lock);
3071 }
39d787be
CY		 3072}
3073
05ca3632 3074static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 3075{
81377bd6 3076 int type = __get_segment_type(fio);
107a805d 3077 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
bfad7c2d 3078
107a805d
CY		 3079 if (keep_order)
3080 down_read(&fio->sbi->io_order_lock);
0a595eba 3081reallocate:
4d57b86d 3082 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 3083 &fio->new_blkaddr, sum, type, fio, true);
6aa58d8a
CY		 3084 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3085 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3086 fio->old_blkaddr, fio->old_blkaddr);
bfad7c2d 3087
351df4b2 3088 /* writeout dirty page into bdev */
fe16efe6
CY		 3089 f2fs_submit_page_write(fio);
3090 if (fio->retry) {
0a595eba
JK		 3091 fio->old_blkaddr = fio->new_blkaddr;
3092 goto reallocate;
3093 }
fe16efe6
CY		 3094
3095 update_device_state(fio);
3096
107a805d
CY		 3097 if (keep_order)
3098 up_read(&fio->sbi->io_order_lock);
351df4b2
JK		 3099}
3100
4d57b86d 3101void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
b0af6d49 3102 enum iostat_type io_type)
351df4b2 3103{
458e6197 3104 struct f2fs_io_info fio = {
05ca3632 3105 .sbi = sbi,
458e6197 3106 .type = META,
0cdd3195 3107 .temp = HOT,
04d328de 3108 .op = REQ_OP_WRITE,
70fd7614 3109 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 3110 .old_blkaddr = page->index,
3111 .new_blkaddr = page->index,
05ca3632 3112 .page = page,
4375a336 3113 .encrypted_page = NULL,
fb830fc5 3114 .in_list = false,
458e6197
JK		 3115 };
3116
2b947003 3117 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 3118 fio.op_flags &= ~REQ_META;
2b947003 3119
351df4b2 3120 set_page_writeback(page);
17c50035 3121 ClearPageError(page);
b9109b0e 3122 f2fs_submit_page_write(&fio);
b0af6d49 3123
b63e7be5 3124 stat_inc_meta_count(sbi, page->index);
b0af6d49 3125 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 3126}
3127
4d57b86d 3128void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 3129{
3130 struct f2fs_summary sum;
05ca3632 3131
351df4b2 3132 set_summary(&sum, nid, 0, 0);
05ca3632 3133 do_write_page(&sum, fio);
b0af6d49
CY		 3134
3135 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3136}
3137
4d57b86d
CY		 3138void f2fs_outplace_write_data(struct dnode_of_data *dn,
3139 struct f2fs_io_info *fio)
351df4b2 3140{
05ca3632 3141 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2 3142 struct f2fs_summary sum;
351df4b2 3143
9850cf4a 3144 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
7735730d 3145 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
05ca3632 3146 do_write_page(&sum, fio);
f28b3434 3147 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 3148
3149 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 3150}
3151
4d57b86d 3152int f2fs_inplace_write_data(struct f2fs_io_info *fio)
351df4b2 3153{
b0af6d49 3154 int err;
d21b0f23 3155 struct f2fs_sb_info *sbi = fio->sbi;
b0af6d49 3156
7a9d7548 3157 fio->new_blkaddr = fio->old_blkaddr;
0cdd3195
HL		 3158 /* i/o temperature is needed for passing down write hints */
3159 __get_segment_type(fio);
d21b0f23
YH		 3160
3161 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3162 GET_SEGNO(sbi, fio->new_blkaddr))->type));
3163
05ca3632 3164 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 3165
3166 err = f2fs_submit_page_bio(fio);
39d787be
CY		 3167 if (!err)
3168 update_device_state(fio);
b0af6d49
CY		 3169
3170 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3171
3172 return err;
351df4b2
JK		 3173}
3174
2b60311d
CY		 3175static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3176 unsigned int segno)
3177{
3178 int i;
3179
3180 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3181 if (CURSEG_I(sbi, i)->segno == segno)
3182 break;
3183 }
3184 return i;
3185}
3186
4d57b86d 3187void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 3188 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 3189 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 3190{
3191 struct sit_info *sit_i = SIT_I(sbi);
3192 struct curseg_info *curseg;
3193 unsigned int segno, old_cursegno;
3194 struct seg_entry *se;
3195 int type;
19f106bc 3196 unsigned short old_blkoff;
351df4b2
JK		 3197
3198 segno = GET_SEGNO(sbi, new_blkaddr);
3199 se = get_seg_entry(sbi, segno);
3200 type = se->type;
3201
2b60311d
CY		 3202 down_write(&SM_I(sbi)->curseg_lock);
3203
19f106bc
CY		 3204 if (!recover_curseg) {
3205 /* for recovery flow */
3206 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3207 if (old_blkaddr == NULL_ADDR)
3208 type = CURSEG_COLD_DATA;
3209 else
3210 type = CURSEG_WARM_DATA;
3211 }
3212 } else {
2b60311d
CY		 3213 if (IS_CURSEG(sbi, segno)) {
3214 /* se->type is volatile as SSR allocation */
3215 type = __f2fs_get_curseg(sbi, segno);
3216 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3217 } else {
351df4b2 3218 type = CURSEG_WARM_DATA;
2b60311d 3219 }
351df4b2 3220 }
19f106bc 3221
2c190504 3222 f2fs_bug_on(sbi, !IS_DATASEG(type));
351df4b2
JK		 3223 curseg = CURSEG_I(sbi, type);
3224
3225 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 3226 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3227
3228 old_cursegno = curseg->segno;
19f106bc 3229 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 3230
3231 /* change the current segment */
3232 if (segno != curseg->segno) {
3233 curseg->next_segno = segno;
025d63a4 3234 change_curseg(sbi, type);
351df4b2
JK		 3235 }
3236
491c0854 3237 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 3238 __add_sum_entry(sbi, type, sum);
351df4b2 3239
28bc106b 3240 if (!recover_curseg || recover_newaddr)
6e2c64ad 3241 update_sit_entry(sbi, new_blkaddr, 1);
6aa58d8a
CY		 3242 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3243 invalidate_mapping_pages(META_MAPPING(sbi),
3244 old_blkaddr, old_blkaddr);
6e2c64ad 3245 update_sit_entry(sbi, old_blkaddr, -1);
6aa58d8a 3246 }
6e2c64ad
JK		 3247
3248 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3249 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3250
351df4b2 3251 locate_dirty_segment(sbi, old_cursegno);
351df4b2 3252
19f106bc
CY		 3253 if (recover_curseg) {
3254 if (old_cursegno != curseg->segno) {
3255 curseg->next_segno = old_cursegno;
025d63a4 3256 change_curseg(sbi, type);
19f106bc
CY		 3257 }
3258 curseg->next_blkoff = old_blkoff;
3259 }
3260
3d26fa6b 3261 up_write(&sit_i->sentry_lock);
351df4b2 3262 mutex_unlock(&curseg->curseg_mutex);
2b60311d 3263 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 3264}
3265
528e3459
CY		 3266void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3267 block_t old_addr, block_t new_addr,
28bc106b
CY		 3268 unsigned char version, bool recover_curseg,
3269 bool recover_newaddr)
528e3459
CY		 3270{
3271 struct f2fs_summary sum;
3272
3273 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3274
4d57b86d 3275 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
28bc106b 3276 recover_curseg, recover_newaddr);
528e3459 3277
f28b3434 3278 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 3279}
3280
93dfe2ac 3281void f2fs_wait_on_page_writeback(struct page *page,
bae0ee7a 3282 enum page_type type, bool ordered, bool locked)
93dfe2ac 3283{
93dfe2ac 3284 if (PageWriteback(page)) {
4081363f
JK		 3285 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3286
bab475c5 3287 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
bae0ee7a 3288 if (ordered) {
fec1d657 3289 wait_on_page_writeback(page);
bae0ee7a
CY		 3290 f2fs_bug_on(sbi, locked && PageWriteback(page));
3291 } else {
fec1d657 3292 wait_for_stable_page(page);
bae0ee7a 3293 }
93dfe2ac
JK		 3294 }
3295}
3296
0ded69f6 3297void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
08b39fbd 3298{
0ded69f6 3299 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
08b39fbd
CY		 3300 struct page *cpage;
3301
0ded69f6
JK		 3302 if (!f2fs_post_read_required(inode))
3303 return;
3304
e1da7872 3305 if (!is_valid_data_blkaddr(sbi, blkaddr))
08b39fbd
CY		 3306 return;
3307
08b39fbd
CY		 3308 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3309 if (cpage) {
bae0ee7a 3310 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
08b39fbd
CY		 3311 f2fs_put_page(cpage, 1);
3312 }
3313}
3314
1e78e8bd
ST		 3315void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3316 block_t len)
3317{
3318 block_t i;
3319
3320 for (i = 0; i < len; i++)
3321 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3322}
3323
7735730d 3324static int read_compacted_summaries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3325{
3326 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3327 struct curseg_info *seg_i;
3328 unsigned char *kaddr;
3329 struct page *page;
3330 block_t start;
3331 int i, j, offset;
3332
3333 start = start_sum_block(sbi);
3334
4d57b86d 3335 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3336 if (IS_ERR(page))
3337 return PTR_ERR(page);
351df4b2
JK		 3338 kaddr = (unsigned char *)page_address(page);
3339
3340 /* Step 1: restore nat cache */
3341 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3342 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 3343
3344 /* Step 2: restore sit cache */
3345 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3346 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 3347 offset = 2 * SUM_JOURNAL_SIZE;
3348
3349 /* Step 3: restore summary entries */
3350 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3351 unsigned short blk_off;
3352 unsigned int segno;
3353
3354 seg_i = CURSEG_I(sbi, i);
3355 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3356 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3357 seg_i->next_segno = segno;
3358 reset_curseg(sbi, i, 0);
3359 seg_i->alloc_type = ckpt->alloc_type[i];
3360 seg_i->next_blkoff = blk_off;
3361
3362 if (seg_i->alloc_type == SSR)
3363 blk_off = sbi->blocks_per_seg;
3364
3365 for (j = 0; j < blk_off; j++) {
3366 struct f2fs_summary *s;
3367 s = (struct f2fs_summary *)(kaddr + offset);
3368 seg_i->sum_blk->entries[j] = *s;
3369 offset += SUMMARY_SIZE;
09cbfeaf 3370 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3371 SUM_FOOTER_SIZE)
3372 continue;
3373
3374 f2fs_put_page(page, 1);
3375 page = NULL;
3376
4d57b86d 3377 page = f2fs_get_meta_page(sbi, start++);
7735730d
CY		 3378 if (IS_ERR(page))
3379 return PTR_ERR(page);
351df4b2
JK		 3380 kaddr = (unsigned char *)page_address(page);
3381 offset = 0;
3382 }
3383 }
3384 f2fs_put_page(page, 1);
7735730d 3385 return 0;
351df4b2
JK		 3386}
3387
3388static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3389{
3390 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3391 struct f2fs_summary_block *sum;
3392 struct curseg_info *curseg;
3393 struct page *new;
3394 unsigned short blk_off;
3395 unsigned int segno = 0;
3396 block_t blk_addr = 0;
7735730d 3397 int err = 0;
351df4b2
JK		 3398
3399 /* get segment number and block addr */
3400 if (IS_DATASEG(type)) {
3401 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3402 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3403 CURSEG_HOT_DATA]);
119ee914 3404 if (__exist_node_summaries(sbi))
351df4b2
JK		 3405 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3406 else
3407 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3408 } else {
3409 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3410 CURSEG_HOT_NODE]);
3411 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3412 CURSEG_HOT_NODE]);
119ee914 3413 if (__exist_node_summaries(sbi))
351df4b2
JK		 3414 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3415 type - CURSEG_HOT_NODE);
3416 else
3417 blk_addr = GET_SUM_BLOCK(sbi, segno);
3418 }
3419
4d57b86d 3420 new = f2fs_get_meta_page(sbi, blk_addr);
7735730d
CY		 3421 if (IS_ERR(new))
3422 return PTR_ERR(new);
351df4b2
JK		 3423 sum = (struct f2fs_summary_block *)page_address(new);
3424
3425 if (IS_NODESEG(type)) {
119ee914 3426 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 3427 struct f2fs_summary *ns = &sum->entries[0];
3428 int i;
3429 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3430 ns->version = 0;
3431 ns->ofs_in_node = 0;
3432 }
3433 } else {
7735730d
CY		 3434 err = f2fs_restore_node_summary(sbi, segno, sum);
3435 if (err)
3436 goto out;
351df4b2
JK		 3437 }
3438 }
3439
3440 /* set uncompleted segment to curseg */
3441 curseg = CURSEG_I(sbi, type);
3442 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 3443
3444 /* update journal info */
3445 down_write(&curseg->journal_rwsem);
3446 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3447 up_write(&curseg->journal_rwsem);
3448
3449 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3450 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 3451 curseg->next_segno = segno;
3452 reset_curseg(sbi, type, 0);
3453 curseg->alloc_type = ckpt->alloc_type[type];
3454 curseg->next_blkoff = blk_off;
3455 mutex_unlock(&curseg->curseg_mutex);
7735730d 3456out:
351df4b2 3457 f2fs_put_page(new, 1);
7735730d 3458 return err;
351df4b2
JK		 3459}
3460
3461static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3462{
21d3f8e1
JQ		 3463 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3464 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 3465 int type = CURSEG_HOT_DATA;
e4fc5fbf 3466 int err;
351df4b2 3467
aaec2b1d 3468 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4d57b86d 3469 int npages = f2fs_npages_for_summary_flush(sbi, true);
3fa06d7b
CY		 3470
3471 if (npages >= 2)
4d57b86d 3472 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 3473 META_CP, true);
3fa06d7b 3474
351df4b2 3475 /* restore for compacted data summary */
7735730d
CY		 3476 err = read_compacted_summaries(sbi);
3477 if (err)
3478 return err;
351df4b2
JK		 3479 type = CURSEG_HOT_NODE;
3480 }
3481
119ee914 3482 if (__exist_node_summaries(sbi))
4d57b86d 3483 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3484 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3485
e4fc5fbf
CY		 3486 for (; type <= CURSEG_COLD_NODE; type++) {
3487 err = read_normal_summaries(sbi, type);
3488 if (err)
3489 return err;
3490 }
3491
21d3f8e1
JQ		 3492 /* sanity check for summary blocks */
3493 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3494 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3495 return -EINVAL;
3496
351df4b2
JK		 3497 return 0;
3498}
3499
3500static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3501{
3502 struct page *page;
3503 unsigned char *kaddr;
3504 struct f2fs_summary *summary;
3505 struct curseg_info *seg_i;
3506 int written_size = 0;
3507 int i, j;
3508
4d57b86d 3509 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3510 kaddr = (unsigned char *)page_address(page);
81114baa 3511 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3512
3513 /* Step 1: write nat cache */
3514 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3515 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3516 written_size += SUM_JOURNAL_SIZE;
3517
3518 /* Step 2: write sit cache */
3519 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3520 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3521 written_size += SUM_JOURNAL_SIZE;
3522
351df4b2
JK		 3523 /* Step 3: write summary entries */
3524 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3525 unsigned short blkoff;
3526 seg_i = CURSEG_I(sbi, i);
3527 if (sbi->ckpt->alloc_type[i] == SSR)
3528 blkoff = sbi->blocks_per_seg;
3529 else
3530 blkoff = curseg_blkoff(sbi, i);
3531
3532 for (j = 0; j < blkoff; j++) {
3533 if (!page) {
4d57b86d 3534 page = f2fs_grab_meta_page(sbi, blkaddr++);
351df4b2 3535 kaddr = (unsigned char *)page_address(page);
81114baa 3536 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3537 written_size = 0;
3538 }
3539 summary = (struct f2fs_summary *)(kaddr + written_size);
3540 *summary = seg_i->sum_blk->entries[j];
3541 written_size += SUMMARY_SIZE;
351df4b2 3542
09cbfeaf 3543 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3544 SUM_FOOTER_SIZE)
3545 continue;
3546
e8d61a74 3547 set_page_dirty(page);
351df4b2
JK		 3548 f2fs_put_page(page, 1);
3549 page = NULL;
3550 }
3551 }
e8d61a74
CY		 3552 if (page) {
3553 set_page_dirty(page);
351df4b2 3554 f2fs_put_page(page, 1);
e8d61a74 3555 }
351df4b2
JK		 3556}
3557
3558static void write_normal_summaries(struct f2fs_sb_info *sbi,
3559 block_t blkaddr, int type)
3560{
3561 int i, end;
3562 if (IS_DATASEG(type))
3563 end = type + NR_CURSEG_DATA_TYPE;
3564 else
3565 end = type + NR_CURSEG_NODE_TYPE;
3566
b7ad7512
CY		 3567 for (i = type; i < end; i++)
3568 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3569}
3570
4d57b86d 3571void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3572{
aaec2b1d 3573 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3574 write_compacted_summaries(sbi, start_blk);
3575 else
3576 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3577}
3578
4d57b86d 3579void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
351df4b2 3580{
119ee914 3581 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3582}
3583
4d57b86d 3584int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3585 unsigned int val, int alloc)
3586{
3587 int i;
3588
3589 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3590 for (i = 0; i < nats_in_cursum(journal); i++) {
3591 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3592 return i;
3593 }
dfc08a12
CY		 3594 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3595 return update_nats_in_cursum(journal, 1);
351df4b2 3596 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3597 for (i = 0; i < sits_in_cursum(journal); i++)
3598 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3599 return i;
dfc08a12
CY		 3600 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3601 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3602 }
3603 return -1;
3604}
3605
3606static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3607 unsigned int segno)
3608{
7735730d 3609 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3610}
3611
3612static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3613 unsigned int start)
3614{
3615 struct sit_info *sit_i = SIT_I(sbi);
068c3cd8 3616 struct page *page;
351df4b2 3617 pgoff_t src_off, dst_off;
351df4b2
JK		 3618
3619 src_off = current_sit_addr(sbi, start);
3620 dst_off = next_sit_addr(sbi, src_off);
3621
4d57b86d 3622 page = f2fs_grab_meta_page(sbi, dst_off);
068c3cd8 3623 seg_info_to_sit_page(sbi, page, start);
351df4b2 3624
068c3cd8 3625 set_page_dirty(page);
351df4b2
JK		 3626 set_to_next_sit(sit_i, start);
3627
068c3cd8 3628 return page;
351df4b2
JK		 3629}
3630
184a5cd2
CY		 3631static struct sit_entry_set *grab_sit_entry_set(void)
3632{
3633 struct sit_entry_set *ses =
80c54505 3634 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3635
3636 ses->entry_cnt = 0;
3637 INIT_LIST_HEAD(&ses->set_list);
3638 return ses;
3639}
3640
3641static void release_sit_entry_set(struct sit_entry_set *ses)
3642{
3643 list_del(&ses->set_list);
3644 kmem_cache_free(sit_entry_set_slab, ses);
3645}
3646
3647static void adjust_sit_entry_set(struct sit_entry_set *ses,
3648 struct list_head *head)
3649{
3650 struct sit_entry_set *next = ses;
3651
3652 if (list_is_last(&ses->set_list, head))
3653 return;
3654
3655 list_for_each_entry_continue(next, head, set_list)
3656 if (ses->entry_cnt <= next->entry_cnt)
3657 break;
3658
3659 list_move_tail(&ses->set_list, &next->set_list);
3660}
3661
3662static void add_sit_entry(unsigned int segno, struct list_head *head)
3663{
3664 struct sit_entry_set *ses;
3665 unsigned int start_segno = START_SEGNO(segno);
3666
3667 list_for_each_entry(ses, head, set_list) {
3668 if (ses->start_segno == start_segno) {
3669 ses->entry_cnt++;
3670 adjust_sit_entry_set(ses, head);
3671 return;
3672 }
3673 }
3674
3675 ses = grab_sit_entry_set();
3676
3677 ses->start_segno = start_segno;
3678 ses->entry_cnt++;
3679 list_add(&ses->set_list, head);
3680}
3681
3682static void add_sits_in_set(struct f2fs_sb_info *sbi)
3683{
3684 struct f2fs_sm_info *sm_info = SM_I(sbi);
3685 struct list_head *set_list = &sm_info->sit_entry_set;
3686 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3687 unsigned int segno;
3688
7cd8558b 3689 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3690 add_sit_entry(segno, set_list);
3691}
3692
3693static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3694{
3695 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3696 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3697 int i;
3698
b7ad7512 3699 down_write(&curseg->journal_rwsem);
dfc08a12 3700 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3701 unsigned int segno;
3702 bool dirtied;
3703
dfc08a12 3704 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3705 dirtied = __mark_sit_entry_dirty(sbi, segno);
3706
3707 if (!dirtied)
3708 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3709 }
dfc08a12 3710 update_sits_in_cursum(journal, -i);
b7ad7512 3711 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3712}
3713
0a8165d7 3714/*
351df4b2
JK		 3715 * CP calls this function, which flushes SIT entries including sit_journal,
3716 * and moves prefree segs to free segs.
3717 */
4d57b86d 3718void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3719{
3720 struct sit_info *sit_i = SIT_I(sbi);
3721 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3722 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3723 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3724 struct sit_entry_set *ses, *tmp;
3725 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3726 bool to_journal = true;
4b2fecc8 3727 struct seg_entry *se;
351df4b2 3728
3d26fa6b 3729 down_write(&sit_i->sentry_lock);
351df4b2 3730
2b11a74b
WL		 3731 if (!sit_i->dirty_sentries)
3732 goto out;
3733
351df4b2 3734 /*
184a5cd2
CY		 3735 * add and account sit entries of dirty bitmap in sit entry
3736 * set temporarily
351df4b2 3737 */
184a5cd2 3738 add_sits_in_set(sbi);
351df4b2 3739
184a5cd2
CY		 3740 /*
3741 * if there are no enough space in journal to store dirty sit
3742 * entries, remove all entries from journal and add and account
3743 * them in sit entry set.
3744 */
dfc08a12 3745 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3746 remove_sits_in_journal(sbi);
b2955550 3747
184a5cd2
CY		 3748 /*
3749 * there are two steps to flush sit entries:
3750 * #1, flush sit entries to journal in current cold data summary block.
3751 * #2, flush sit entries to sit page.
3752 */
3753 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3754 struct page *page = NULL;
184a5cd2
CY		 3755 struct f2fs_sit_block *raw_sit = NULL;
3756 unsigned int start_segno = ses->start_segno;
3757 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3758 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3759 unsigned int segno = start_segno;
3760
3761 if (to_journal &&
dfc08a12 3762 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3763 to_journal = false;
3764
b7ad7512
CY		 3765 if (to_journal) {
3766 down_write(&curseg->journal_rwsem);
3767 } else {
184a5cd2
CY		 3768 page = get_next_sit_page(sbi, start_segno);
3769 raw_sit = page_address(page);
351df4b2 3770 }
351df4b2 3771
184a5cd2
CY		 3772 /* flush dirty sit entries in region of current sit set */
3773 for_each_set_bit_from(segno, bitmap, end) {
3774 int offset, sit_offset;
4b2fecc8
JK		 3775
3776 se = get_seg_entry(sbi, segno);
56b07e7e
ZZ		 3777#ifdef CONFIG_F2FS_CHECK_FS
3778 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3779 SIT_VBLOCK_MAP_SIZE))
3780 f2fs_bug_on(sbi, 1);
3781#endif
184a5cd2
CY		 3782
3783 /* add discard candidates */
c473f1a9 3784 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3785 cpc->trim_start = segno;
25290fa5 3786 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3787 }
184a5cd2
CY		 3788
3789 if (to_journal) {
4d57b86d 3790 offset = f2fs_lookup_journal_in_cursum(journal,
184a5cd2
CY		 3791 SIT_JOURNAL, segno, 1);
3792 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3793 segno_in_journal(journal, offset) =
184a5cd2
CY		 3794 cpu_to_le32(segno);
3795 seg_info_to_raw_sit(se,
dfc08a12 3796 &sit_in_journal(journal, offset));
56b07e7e
ZZ		 3797 check_block_count(sbi, segno,
3798 &sit_in_journal(journal, offset));
184a5cd2
CY		 3799 } else {
3800 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3801 seg_info_to_raw_sit(se,
3802 &raw_sit->entries[sit_offset]);
56b07e7e
ZZ		 3803 check_block_count(sbi, segno,
3804 &raw_sit->entries[sit_offset]);
184a5cd2 3805 }
351df4b2 3806
184a5cd2
CY		 3807 __clear_bit(segno, bitmap);
3808 sit_i->dirty_sentries--;
3809 ses->entry_cnt--;
351df4b2
JK		 3810 }
3811
b7ad7512
CY		 3812 if (to_journal)
3813 up_write(&curseg->journal_rwsem);
3814 else
184a5cd2
CY		 3815 f2fs_put_page(page, 1);
3816
3817 f2fs_bug_on(sbi, ses->entry_cnt);
3818 release_sit_entry_set(ses);
351df4b2 3819 }
184a5cd2
CY		 3820
3821 f2fs_bug_on(sbi, !list_empty(head));
3822 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3823out:
c473f1a9 3824 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3825 __u64 trim_start = cpc->trim_start;
3826
4b2fecc8 3827 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3828 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3829
3830 cpc->trim_start = trim_start;
4b2fecc8 3831 }
3d26fa6b 3832 up_write(&sit_i->sentry_lock);
351df4b2 3833
351df4b2
JK		 3834 set_prefree_as_free_segments(sbi);
3835}
3836
3837static int build_sit_info(struct f2fs_sb_info *sbi)
3838{
3839 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3840 struct sit_info *sit_i;
3841 unsigned int sit_segs, start;
ae27d62e 3842 char *src_bitmap;
351df4b2
JK		 3843 unsigned int bitmap_size;
3844
3845 /* allocate memory for SIT information */
acbf054d 3846 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
351df4b2
JK		 3847 if (!sit_i)
3848 return -ENOMEM;
3849
3850 SM_I(sbi)->sit_info = sit_i;
3851
9d2a789c
KC		 3852 sit_i->sentries =
3853 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3854 MAIN_SEGS(sbi)),
3855 GFP_KERNEL);
351df4b2
JK		 3856 if (!sit_i->sentries)
3857 return -ENOMEM;
3858
7cd8558b 3859 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14
CY		 3860 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3861 GFP_KERNEL);
351df4b2
JK		 3862 if (!sit_i->dirty_sentries_bitmap)
3863 return -ENOMEM;
3864
7cd8558b 3865 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3866 sit_i->sentries[start].cur_valid_map
acbf054d 3867 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
351df4b2 3868 sit_i->sentries[start].ckpt_valid_map
acbf054d 3869 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3870 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3871 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3872 return -ENOMEM;
3e025740 3873
355e7891
CY		 3874#ifdef CONFIG_F2FS_CHECK_FS
3875 sit_i->sentries[start].cur_valid_map_mir
acbf054d 3876 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
355e7891
CY		 3877 if (!sit_i->sentries[start].cur_valid_map_mir)
3878 return -ENOMEM;
3879#endif
3880
7d20c8ab
CY		 3881 sit_i->sentries[start].discard_map
3882 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3883 GFP_KERNEL);
3884 if (!sit_i->sentries[start].discard_map)
3885 return -ENOMEM;
351df4b2
JK		 3886 }
3887
acbf054d 3888 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
60a3b782
JK		 3889 if (!sit_i->tmp_map)
3890 return -ENOMEM;
3891
2c70c5e3 3892 if (__is_large_section(sbi)) {
9d2a789c
KC		 3893 sit_i->sec_entries =
3894 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3895 MAIN_SECS(sbi)),
3896 GFP_KERNEL);
351df4b2
JK		 3897 if (!sit_i->sec_entries)
3898 return -ENOMEM;
3899 }
3900
3901 /* get information related with SIT */
3902 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3903
3904 /* setup SIT bitmap from ckeckpoint pack */
3905 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3906 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3907
ae27d62e
CY		 3908 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3909 if (!sit_i->sit_bitmap)
351df4b2 3910 return -ENOMEM;
351df4b2 3911
ae27d62e
CY		 3912#ifdef CONFIG_F2FS_CHECK_FS
3913 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3914 if (!sit_i->sit_bitmap_mir)
3915 return -ENOMEM;
3916#endif
3917
351df4b2
JK		 3918 /* init SIT information */
3919 sit_i->s_ops = &default_salloc_ops;
3920
3921 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3922 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3923 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3924 sit_i->bitmap_size = bitmap_size;
3925 sit_i->dirty_sentries = 0;
3926 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3927 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3928 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3929 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3930 return 0;
3931}
3932
3933static int build_free_segmap(struct f2fs_sb_info *sbi)
3934{
351df4b2
JK		 3935 struct free_segmap_info *free_i;
3936 unsigned int bitmap_size, sec_bitmap_size;
3937
3938 /* allocate memory for free segmap information */
acbf054d 3939 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
351df4b2
JK		 3940 if (!free_i)
3941 return -ENOMEM;
3942
3943 SM_I(sbi)->free_info = free_i;
3944
7cd8558b 3945 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
628b3d14 3946 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
351df4b2
JK		 3947 if (!free_i->free_segmap)
3948 return -ENOMEM;
3949
7cd8558b 3950 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
628b3d14 3951 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3952 if (!free_i->free_secmap)
3953 return -ENOMEM;
3954
3955 /* set all segments as dirty temporarily */
3956 memset(free_i->free_segmap, 0xff, bitmap_size);
3957 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3958
3959 /* init free segmap information */
7cd8558b 3960 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3961 free_i->free_segments = 0;
3962 free_i->free_sections = 0;
1a118ccf 3963 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3964 return 0;
3965}
3966
3967static int build_curseg(struct f2fs_sb_info *sbi)
3968{
1042d60f 3969 struct curseg_info *array;
351df4b2
JK		 3970 int i;
3971
026f0507
KC		 3972 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3973 GFP_KERNEL);
351df4b2
JK		 3974 if (!array)
3975 return -ENOMEM;
3976
3977 SM_I(sbi)->curseg_array = array;
3978
3979 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3980 mutex_init(&array[i].curseg_mutex);
acbf054d 3981 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3982 if (!array[i].sum_blk)
3983 return -ENOMEM;
b7ad7512 3984 init_rwsem(&array[i].journal_rwsem);
acbf054d
CY		 3985 array[i].journal = f2fs_kzalloc(sbi,
3986 sizeof(struct f2fs_journal), GFP_KERNEL);
b7ad7512
CY		 3987 if (!array[i].journal)
3988 return -ENOMEM;
351df4b2
JK		 3989 array[i].segno = NULL_SEGNO;
3990 array[i].next_blkoff = 0;
3991 }
3992 return restore_curseg_summaries(sbi);
3993}
3994
c39a1b34 3995static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3996{
3997 struct sit_info *sit_i = SIT_I(sbi);
3998 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3999 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 4000 struct seg_entry *se;
4001 struct f2fs_sit_entry sit;
74de593a
CY		 4002 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4003 unsigned int i, start, end;
4004 unsigned int readed, start_blk = 0;
c39a1b34 4005 int err = 0;
8a29c126 4006 block_t total_node_blocks = 0;
351df4b2 4007
74de593a 4008 do {
4d57b86d 4009 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
664ba972 4010 META_SIT, true);
74de593a
CY		 4011
4012 start = start_blk * sit_i->sents_per_block;
4013 end = (start_blk + readed) * sit_i->sents_per_block;
4014
7cd8558b 4015 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 4016 struct f2fs_sit_block *sit_blk;
74de593a
CY		 4017 struct page *page;
4018
9c094040 4019 se = &sit_i->sentries[start];
74de593a 4020 page = get_current_sit_page(sbi, start);
edc55aaf
JK		 4021 if (IS_ERR(page))
4022 return PTR_ERR(page);
74de593a
CY		 4023 sit_blk = (struct f2fs_sit_block *)page_address(page);
4024 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4025 f2fs_put_page(page, 1);
d600af23 4026
c39a1b34
JK		 4027 err = check_block_count(sbi, start, &sit);
4028 if (err)
4029 return err;
74de593a 4030 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4031 if (IS_NODESEG(se->type))
4032 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 4033
4034 /* build discard map only one time */
7d20c8ab
CY		 4035 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4036 memset(se->discard_map, 0xff,
4037 SIT_VBLOCK_MAP_SIZE);
4038 } else {
4039 memcpy(se->discard_map,
4040 se->cur_valid_map,
4041 SIT_VBLOCK_MAP_SIZE);
4042 sbi->discard_blks +=
4043 sbi->blocks_per_seg -
4044 se->valid_blocks;
3e025740 4045 }
a66cdd98 4046
2c70c5e3 4047 if (__is_large_section(sbi))
d600af23
CY		 4048 get_sec_entry(sbi, start)->valid_blocks +=
4049 se->valid_blocks;
351df4b2 4050 }
74de593a
CY		 4051 start_blk += readed;
4052 } while (start_blk < sit_blk_cnt);
d600af23
CY		 4053
4054 down_read(&curseg->journal_rwsem);
4055 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 4056 unsigned int old_valid_blocks;
4057
4058 start = le32_to_cpu(segno_in_journal(journal, i));
b2ca374f
JK		 4059 if (start >= MAIN_SEGS(sbi)) {
4060 f2fs_msg(sbi->sb, KERN_ERR,
4061 "Wrong journal entry on segno %u",
4062 start);
4063 set_sbi_flag(sbi, SBI_NEED_FSCK);
4064 err = -EINVAL;
4065 break;
4066 }
4067
d600af23
CY		 4068 se = &sit_i->sentries[start];
4069 sit = sit_in_journal(journal, i);
4070
4071 old_valid_blocks = se->valid_blocks;
8a29c126
JK		 4072 if (IS_NODESEG(se->type))
4073 total_node_blocks -= old_valid_blocks;
d600af23 4074
c39a1b34
JK		 4075 err = check_block_count(sbi, start, &sit);
4076 if (err)
4077 break;
d600af23 4078 seg_info_from_raw_sit(se, &sit);
8a29c126
JK		 4079 if (IS_NODESEG(se->type))
4080 total_node_blocks += se->valid_blocks;
d600af23 4081
7d20c8ab
CY		 4082 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4083 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4084 } else {
4085 memcpy(se->discard_map, se->cur_valid_map,
4086 SIT_VBLOCK_MAP_SIZE);
4087 sbi->discard_blks += old_valid_blocks;
4088 sbi->discard_blks -= se->valid_blocks;
d600af23
CY		 4089 }
4090
2c70c5e3 4091 if (__is_large_section(sbi)) {
d600af23 4092 get_sec_entry(sbi, start)->valid_blocks +=
a9af3fdc
CY		 4093 se->valid_blocks;
4094 get_sec_entry(sbi, start)->valid_blocks -=
4095 old_valid_blocks;
4096 }
d600af23
CY		 4097 }
4098 up_read(&curseg->journal_rwsem);
8a29c126
JK		 4099
4100 if (!err && total_node_blocks != valid_node_count(sbi)) {
4101 f2fs_msg(sbi->sb, KERN_ERR,
4102 "SIT is corrupted node# %u vs %u",
4103 total_node_blocks, valid_node_count(sbi));
4104 set_sbi_flag(sbi, SBI_NEED_FSCK);
4105 err = -EINVAL;
4106 }
4107
c39a1b34 4108 return err;
351df4b2
JK		 4109}
4110
4111static void init_free_segmap(struct f2fs_sb_info *sbi)
4112{
4113 unsigned int start;
4114 int type;
4115
7cd8558b 4116 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 4117 struct seg_entry *sentry = get_seg_entry(sbi, start);
4118 if (!sentry->valid_blocks)
4119 __set_free(sbi, start);
c79b7ff1
JK		 4120 else
4121 SIT_I(sbi)->written_valid_blocks +=
4122 sentry->valid_blocks;
351df4b2
JK		 4123 }
4124
4125 /* set use the current segments */
4126 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4127 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4128 __set_test_and_inuse(sbi, curseg_t->segno);
4129 }
4130}
4131
4132static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4133{
4134 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4135 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 4136 unsigned int segno = 0, offset = 0;
351df4b2
JK		 4137 unsigned short valid_blocks;
4138
8736fbf0 4139 while (1) {
351df4b2 4140 /* find dirty segment based on free segmap */
7cd8558b
JK		 4141 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4142 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 4143 break;
4144 offset = segno + 1;
302bd348 4145 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 4146 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 4147 continue;
ec325b52
JK		 4148 if (valid_blocks > sbi->blocks_per_seg) {
4149 f2fs_bug_on(sbi, 1);
4150 continue;
4151 }
351df4b2
JK		 4152 mutex_lock(&dirty_i->seglist_lock);
4153 __locate_dirty_segment(sbi, segno, DIRTY);
4154 mutex_unlock(&dirty_i->seglist_lock);
4155 }
4156}
4157
5ec4e49f 4158static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4159{
4160 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 4161 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 4162
628b3d14 4163 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5ec4e49f 4164 if (!dirty_i->victim_secmap)
351df4b2
JK		 4165 return -ENOMEM;
4166 return 0;
4167}
4168
4169static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4170{
4171 struct dirty_seglist_info *dirty_i;
4172 unsigned int bitmap_size, i;
4173
4174 /* allocate memory for dirty segments list information */
acbf054d
CY		 4175 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4176 GFP_KERNEL);
351df4b2
JK		 4177 if (!dirty_i)
4178 return -ENOMEM;
4179
4180 SM_I(sbi)->dirty_info = dirty_i;
4181 mutex_init(&dirty_i->seglist_lock);
4182
7cd8558b 4183 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 4184
4185 for (i = 0; i < NR_DIRTY_TYPE; i++) {
628b3d14
CY		 4186 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4187 GFP_KERNEL);
351df4b2
JK		 4188 if (!dirty_i->dirty_segmap[i])
4189 return -ENOMEM;
4190 }
4191
4192 init_dirty_segmap(sbi);
5ec4e49f 4193 return init_victim_secmap(sbi);
351df4b2
JK		 4194}
4195
0a8165d7 4196/*
351df4b2
JK		 4197 * Update min, max modified time for cost-benefit GC algorithm
4198 */
4199static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4200{
4201 struct sit_info *sit_i = SIT_I(sbi);
4202 unsigned int segno;
4203
3d26fa6b 4204 down_write(&sit_i->sentry_lock);
351df4b2 4205
5ad25442 4206 sit_i->min_mtime = ULLONG_MAX;
351df4b2 4207
7cd8558b 4208 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 4209 unsigned int i;
4210 unsigned long long mtime = 0;
4211
4212 for (i = 0; i < sbi->segs_per_sec; i++)
4213 mtime += get_seg_entry(sbi, segno + i)->mtime;
4214
4215 mtime = div_u64(mtime, sbi->segs_per_sec);
4216
4217 if (sit_i->min_mtime > mtime)
4218 sit_i->min_mtime = mtime;
4219 }
a1f72ac2 4220 sit_i->max_mtime = get_mtime(sbi, false);
3d26fa6b 4221 up_write(&sit_i->sentry_lock);
351df4b2
JK		 4222}
4223
4d57b86d 4224int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4225{
4226 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4227 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 4228 struct f2fs_sm_info *sm_info;
351df4b2
JK		 4229 int err;
4230
acbf054d 4231 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
351df4b2
JK		 4232 if (!sm_info)
4233 return -ENOMEM;
4234
4235 /* init sm info */
4236 sbi->sm_info = sm_info;
351df4b2
JK		 4237 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4238 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4239 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4240 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4241 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4242 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4243 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 4244 sm_info->rec_prefree_segments = sm_info->main_segments *
4245 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 4246 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4247 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4248
52763a4b
JK		 4249 if (!test_opt(sbi, LFS))
4250 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 4251 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 4252 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
853137ce 4253 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
ef095d19 4254 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 4255 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 4256
184a5cd2
CY		 4257 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4258
2b60311d
CY		 4259 init_rwsem(&sm_info->curseg_lock);
4260
d4fdf8ba 4261 if (!f2fs_readonly(sbi->sb)) {
4d57b86d 4262 err = f2fs_create_flush_cmd_control(sbi);
2163d198 4263 if (err)
a688b9d9 4264 return err;
6b4afdd7
JK		 4265 }
4266
0b54fb84
JK		 4267 err = create_discard_cmd_control(sbi);
4268 if (err)
4269 return err;
4270
351df4b2
JK		 4271 err = build_sit_info(sbi);
4272 if (err)
4273 return err;
4274 err = build_free_segmap(sbi);
4275 if (err)
4276 return err;
4277 err = build_curseg(sbi);
4278 if (err)
4279 return err;
4280
4281 /* reinit free segmap based on SIT */
c39a1b34
JK		 4282 err = build_sit_entries(sbi);
4283 if (err)
4284 return err;
351df4b2
JK		 4285
4286 init_free_segmap(sbi);
4287 err = build_dirty_segmap(sbi);
4288 if (err)
4289 return err;
4290
4291 init_min_max_mtime(sbi);
4292 return 0;
4293}
4294
4295static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4296 enum dirty_type dirty_type)
4297{
4298 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4299
4300 mutex_lock(&dirty_i->seglist_lock);
39307a8e 4301 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 4302 dirty_i->nr_dirty[dirty_type] = 0;
4303 mutex_unlock(&dirty_i->seglist_lock);
4304}
4305
5ec4e49f 4306static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 4307{
4308 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 4309 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 4310}
4311
4312static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4313{
4314 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4315 int i;
4316
4317 if (!dirty_i)
4318 return;
4319
4320 /* discard pre-free/dirty segments list */
4321 for (i = 0; i < NR_DIRTY_TYPE; i++)
4322 discard_dirty_segmap(sbi, i);
4323
5ec4e49f 4324 destroy_victim_secmap(sbi);
351df4b2 4325 SM_I(sbi)->dirty_info = NULL;
5222595d 4326 kvfree(dirty_i);
351df4b2
JK		 4327}
4328
4329static void destroy_curseg(struct f2fs_sb_info *sbi)
4330{
4331 struct curseg_info *array = SM_I(sbi)->curseg_array;
4332 int i;
4333
4334 if (!array)
4335 return;
4336 SM_I(sbi)->curseg_array = NULL;
b7ad7512 4337 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5222595d
JK		 4338 kvfree(array[i].sum_blk);
4339 kvfree(array[i].journal);
b7ad7512 4340 }
5222595d 4341 kvfree(array);
351df4b2
JK		 4342}
4343
4344static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4345{
4346 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4347 if (!free_i)
4348 return;
4349 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 4350 kvfree(free_i->free_segmap);
4351 kvfree(free_i->free_secmap);
5222595d 4352 kvfree(free_i);
351df4b2
JK		 4353}
4354
4355static void destroy_sit_info(struct f2fs_sb_info *sbi)
4356{
4357 struct sit_info *sit_i = SIT_I(sbi);
4358 unsigned int start;
4359
4360 if (!sit_i)
4361 return;
4362
4363 if (sit_i->sentries) {
7cd8558b 4364 for (start = 0; start < MAIN_SEGS(sbi); start++) {
5222595d 4365 kvfree(sit_i->sentries[start].cur_valid_map);
355e7891 4366#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4367 kvfree(sit_i->sentries[start].cur_valid_map_mir);
355e7891 4368#endif
5222595d
JK		 4369 kvfree(sit_i->sentries[start].ckpt_valid_map);
4370 kvfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 4371 }
4372 }
5222595d 4373 kvfree(sit_i->tmp_map);
60a3b782 4374
39307a8e
JK		 4375 kvfree(sit_i->sentries);
4376 kvfree(sit_i->sec_entries);
4377 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 4378
4379 SM_I(sbi)->sit_info = NULL;
5222595d 4380 kvfree(sit_i->sit_bitmap);
ae27d62e 4381#ifdef CONFIG_F2FS_CHECK_FS
5222595d 4382 kvfree(sit_i->sit_bitmap_mir);
ae27d62e 4383#endif
5222595d 4384 kvfree(sit_i);
351df4b2
JK		 4385}
4386
4d57b86d 4387void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
351df4b2
JK		 4388{
4389 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 4390
3b03f724
CY		 4391 if (!sm_info)
4392 return;
4d57b86d 4393 f2fs_destroy_flush_cmd_control(sbi, true);
f099405f 4394 destroy_discard_cmd_control(sbi);
351df4b2
JK		 4395 destroy_dirty_segmap(sbi);
4396 destroy_curseg(sbi);
4397 destroy_free_segmap(sbi);
4398 destroy_sit_info(sbi);
4399 sbi->sm_info = NULL;
5222595d 4400 kvfree(sm_info);
351df4b2 4401}
7fd9e544 4402
4d57b86d 4403int __init f2fs_create_segment_manager_caches(void)
7fd9e544
JK		 4404{
4405 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 4406 sizeof(struct discard_entry));
7fd9e544 4407 if (!discard_entry_slab)
184a5cd2
CY		 4408 goto fail;
4409
b01a9201
JK		 4410 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4411 sizeof(struct discard_cmd));
4412 if (!discard_cmd_slab)
6ab2a308 4413 goto destroy_discard_entry;
275b66b0 4414
184a5cd2 4415 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 4416 sizeof(struct sit_entry_set));
184a5cd2 4417 if (!sit_entry_set_slab)
b01a9201 4418 goto destroy_discard_cmd;
88b88a66
JK		 4419
4420 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4421 sizeof(struct inmem_pages));
4422 if (!inmem_entry_slab)
4423 goto destroy_sit_entry_set;
7fd9e544 4424 return 0;
184a5cd2 4425
88b88a66
JK		 4426destroy_sit_entry_set:
4427 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 4428destroy_discard_cmd:
4429 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 4430destroy_discard_entry:
184a5cd2
CY		 4431 kmem_cache_destroy(discard_entry_slab);
4432fail:
4433 return -ENOMEM;
7fd9e544
JK		 4434}
4435
4d57b86d 4436void f2fs_destroy_segment_manager_caches(void)
7fd9e544 4437{
184a5cd2 4438 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 4439 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 4440 kmem_cache_destroy(discard_entry_slab);
88b88a66 4441 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 4442}
Linux 6.x block layer and io_uring tree(s)