]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: issue discard align to section in LFS mode
[mirror_ubuntu-bionic-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL		 29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
b01a9201 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK		 36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL		 51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK		 69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL		 75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL		 80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL		 85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK		 95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL		 112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL		 130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
80609448
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
b3a97a2a
JK		 172bool need_SSR(struct f2fs_sb_info *sbi)
173{
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
a2a12b67 184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
b3a97a2a
JK		 185}
186
88b88a66
JK		 187void register_inmem_page(struct inode *inode, struct page *page)
188{
57864ae5 189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
88b88a66
JK		 190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
9be32d72 192
9e4ded3f 193 f2fs_trace_pid(page);
0722b101 194
decd36b6
CY		 195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
197
88b88a66
JK		 198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
decd36b6 203
88b88a66
JK		 204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
57864ae5
JK		 208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
8dcf2ff7 212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 213 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 214
215 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 216}
217
28bc106b
CY		 218static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover)
29b96b54 220{
28bc106b 221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 222 struct inmem_pages *cur, *tmp;
28bc106b 223 int err = 0;
29b96b54
CY		 224
225 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 226 struct page *page = cur->page;
227
228 if (drop)
229 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
230
231 lock_page(page);
29b96b54 232
b6624bc0
CY		 233 f2fs_wait_on_page_writeback(page, DATA, true);
234
28bc106b
CY		 235 if (recover) {
236 struct dnode_of_data dn;
237 struct node_info ni;
238
239 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 240retry:
28bc106b 241 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 242 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
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 }
252 get_node_info(sbi, dn.nid, &ni);
253 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
254 cur->old_addr, ni.version, true, true);
255 f2fs_put_dnode(&dn);
256 }
257next:
63c52d78
JK		 258 /* we don't need to invalidate this in the sccessful status */
259 if (drop || recover)
260 ClearPageUptodate(page);
28bc106b 261 set_page_private(page, 0);
c81ced05 262 ClearPagePrivate(page);
28bc106b 263 f2fs_put_page(page, 1);
29b96b54
CY		 264
265 list_del(&cur->list);
266 kmem_cache_free(inmem_entry_slab, cur);
267 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
268 }
28bc106b 269 return err;
29b96b54
CY		 270}
271
57864ae5
JK		 272void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
273{
274 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
275 struct inode *inode;
276 struct f2fs_inode_info *fi;
277next:
278 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
279 if (list_empty(head)) {
280 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
281 return;
282 }
283 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
284 inode = igrab(&fi->vfs_inode);
285 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
286
287 if (inode) {
288 drop_inmem_pages(inode);
289 iput(inode);
290 }
291 congestion_wait(BLK_RW_ASYNC, HZ/50);
292 cond_resched();
293 goto next;
294}
295
29b96b54
CY		 296void drop_inmem_pages(struct inode *inode)
297{
57864ae5 298 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54
CY		 299 struct f2fs_inode_info *fi = F2FS_I(inode);
300
301 mutex_lock(&fi->inmem_lock);
28bc106b 302 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
57864ae5
JK		 303 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
304 if (!list_empty(&fi->inmem_ilist))
305 list_del_init(&fi->inmem_ilist);
306 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
29b96b54 307 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 308
309 clear_inode_flag(inode, FI_ATOMIC_FILE);
84a23fbe 310 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 311 stat_dec_atomic_write(inode);
29b96b54
CY		 312}
313
8c242db9
JK		 314void drop_inmem_page(struct inode *inode, struct page *page)
315{
316 struct f2fs_inode_info *fi = F2FS_I(inode);
317 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
318 struct list_head *head = &fi->inmem_pages;
319 struct inmem_pages *cur = NULL;
320
321 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
322
323 mutex_lock(&fi->inmem_lock);
324 list_for_each_entry(cur, head, list) {
325 if (cur->page == page)
326 break;
327 }
328
329 f2fs_bug_on(sbi, !cur || cur->page != page);
330 list_del(&cur->list);
331 mutex_unlock(&fi->inmem_lock);
332
333 dec_page_count(sbi, F2FS_INMEM_PAGES);
334 kmem_cache_free(inmem_entry_slab, cur);
335
336 ClearPageUptodate(page);
337 set_page_private(page, 0);
338 ClearPagePrivate(page);
339 f2fs_put_page(page, 0);
340
341 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
342}
343
28bc106b
CY		 344static int __commit_inmem_pages(struct inode *inode,
345 struct list_head *revoke_list)
88b88a66
JK		 346{
347 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
348 struct f2fs_inode_info *fi = F2FS_I(inode);
349 struct inmem_pages *cur, *tmp;
88b88a66 350 struct f2fs_io_info fio = {
05ca3632 351 .sbi = sbi,
39d787be 352 .ino = inode->i_ino,
88b88a66 353 .type = DATA,
04d328de 354 .op = REQ_OP_WRITE,
70fd7614 355 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 356 .io_type = FS_DATA_IO,
88b88a66 357 };
942fd319 358 pgoff_t last_idx = ULONG_MAX;
edb27dee 359 int err = 0;
88b88a66 360
88b88a66 361 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 362 struct page *page = cur->page;
363
364 lock_page(page);
365 if (page->mapping == inode->i_mapping) {
366 trace_f2fs_commit_inmem_page(page, INMEM);
367
368 set_page_dirty(page);
369 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 370 if (clear_page_dirty_for_io(page)) {
29b96b54 371 inode_dec_dirty_pages(inode);
933439c8
CY		 372 remove_dirty_inode(inode);
373 }
640cc189 374retry:
28bc106b 375 fio.page = page;
e959c8f5 376 fio.old_blkaddr = NULL_ADDR;
4d978078 377 fio.encrypted_page = NULL;
cc15620b 378 fio.need_lock = LOCK_DONE;
29b96b54
CY		 379 err = do_write_data_page(&fio);
380 if (err) {
640cc189
JK		 381 if (err == -ENOMEM) {
382 congestion_wait(BLK_RW_ASYNC, HZ/50);
383 cond_resched();
384 goto retry;
385 }
28bc106b 386 unlock_page(page);
29b96b54 387 break;
70c640b1 388 }
28bc106b
CY		 389 /* record old blkaddr for revoking */
390 cur->old_addr = fio.old_blkaddr;
942fd319 391 last_idx = page->index;
28bc106b
CY		 392 }
393 unlock_page(page);
394 list_move_tail(&cur->list, revoke_list);
88b88a66 395 }
29b96b54 396
942fd319 397 if (last_idx != ULONG_MAX)
b9109b0e 398 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 399
400 if (!err)
401 __revoke_inmem_pages(inode, revoke_list, false, false);
402
29b96b54
CY		 403 return err;
404}
405
406int commit_inmem_pages(struct inode *inode)
407{
408 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
409 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 410 struct list_head revoke_list;
411 int err;
29b96b54 412
28bc106b 413 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 414 f2fs_balance_fs(sbi, true);
415 f2fs_lock_op(sbi);
416
5fe45743
CY		 417 set_inode_flag(inode, FI_ATOMIC_COMMIT);
418
29b96b54 419 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 420 err = __commit_inmem_pages(inode, &revoke_list);
421 if (err) {
422 int ret;
423 /*
424 * try to revoke all committed pages, but still we could fail
425 * due to no memory or other reason, if that happened, EAGAIN
426 * will be returned, which means in such case, transaction is
427 * already not integrity, caller should use journal to do the
428 * recovery or rewrite & commit last transaction. For other
429 * error number, revoking was done by filesystem itself.
430 */
431 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
432 if (ret)
433 err = ret;
434
435 /* drop all uncommitted pages */
436 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
437 }
57864ae5
JK		 438 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
439 if (!list_empty(&fi->inmem_ilist))
440 list_del_init(&fi->inmem_ilist);
441 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
88b88a66
JK		 442 mutex_unlock(&fi->inmem_lock);
443
5fe45743
CY		 444 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
445
29b96b54 446 f2fs_unlock_op(sbi);
edb27dee 447 return err;
88b88a66
JK		 448}
449
0a8165d7 450/*
351df4b2
JK		 451 * This function balances dirty node and dentry pages.
452 * In addition, it controls garbage collection.
453 */
2c4db1a6 454void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 455{
0f348028 456#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 457 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
458 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 459 f2fs_stop_checkpoint(sbi, false);
55523519 460 }
0f348028
CY		 461#endif
462
e589c2c4 463 /* balance_fs_bg is able to be pending */
a7881893 464 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 465 f2fs_balance_fs_bg(sbi);
466
351df4b2 467 /*
029cd28c
JK		 468 * We should do GC or end up with checkpoint, if there are so many dirty
469 * dir/node pages without enough free segments.
351df4b2 470 */
7f3037a5 471 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 472 mutex_lock(&sbi->gc_mutex);
e066b83c 473 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 474 }
475}
476
4660f9c0
JK		 477void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
478{
de0b4f9e
CY		 479 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
480 return;
481
1dcc336b 482 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 483 if (!available_free_memory(sbi, EXTENT_CACHE))
484 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 485
1b38dc8e
JK		 486 /* check the # of cached NAT entries */
487 if (!available_free_memory(sbi, NAT_ENTRIES))
488 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
489
31696580 490 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 491 try_to_free_nids(sbi, MAX_FREE_NIDS);
492 else
22ad0b6a 493 build_free_nids(sbi, false, false);
31696580 494
1c0f4bf5 495 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 496 return;
31696580 497
1b38dc8e
JK		 498 /* checkpoint is the only way to shrink partial cached entries */
499 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 500 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 501 excess_prefree_segs(sbi) ||
502 excess_dirty_nats(sbi) ||
f455c8a5 503 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 504 if (test_opt(sbi, DATA_FLUSH)) {
505 struct blk_plug plug;
506
507 blk_start_plug(&plug);
36b35a0d 508 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 509 blk_finish_plug(&plug);
510 }
4660f9c0 511 f2fs_sync_fs(sbi->sb, true);
42190d2a 512 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 513 }
4660f9c0
JK		 514}
515
20fda56b
KM		 516static int __submit_flush_wait(struct f2fs_sb_info *sbi,
517 struct block_device *bdev)
3c62be17 518{
d62fe971 519 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
3c62be17
JK		 520 int ret;
521
3adc5fcb 522 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 523 bio_set_dev(bio, bdev);
3c62be17
JK		 524 ret = submit_bio_wait(bio);
525 bio_put(bio);
20fda56b
KM		 526
527 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
528 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 529 return ret;
530}
531
39d787be 532static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 533{
39d787be 534 int ret = 0;
3c62be17
JK		 535 int i;
536
39d787be
CY		 537 if (!sbi->s_ndevs)
538 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 539
39d787be
CY		 540 for (i = 0; i < sbi->s_ndevs; i++) {
541 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
542 continue;
20fda56b
KM		 543 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
544 if (ret)
545 break;
3c62be17
JK		 546 }
547 return ret;
548}
549
2163d198 550static int issue_flush_thread(void *data)
6b4afdd7
JK		 551{
552 struct f2fs_sb_info *sbi = data;
b01a9201 553 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 554 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 555repeat:
556 if (kthread_should_stop())
557 return 0;
558
dc6febb6
CY		 559 sb_start_intwrite(sbi->sb);
560
721bd4d5 561 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 562 struct flush_cmd *cmd, *next;
563 int ret;
564
721bd4d5
GZ		 565 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
566 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
567
39d787be
CY		 568 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
569
570 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 571 atomic_inc(&fcc->issued_flush);
572
721bd4d5
GZ		 573 llist_for_each_entry_safe(cmd, next,
574 fcc->dispatch_list, llnode) {
6b4afdd7 575 cmd->ret = ret;
6b4afdd7
JK		 576 complete(&cmd->wait);
577 }
a688b9d9 578 fcc->dispatch_list = NULL;
6b4afdd7
JK		 579 }
580
dc6febb6
CY		 581 sb_end_intwrite(sbi->sb);
582
a688b9d9 583 wait_event_interruptible(*q,
721bd4d5 584 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 585 goto repeat;
586}
587
39d787be 588int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 589{
b01a9201 590 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 591 struct flush_cmd cmd;
8b8dd65f 592 int ret;
6b4afdd7 593
0f7b2abd
JK		 594 if (test_opt(sbi, NOBARRIER))
595 return 0;
596
8b8dd65f 597 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 598 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 599 atomic_inc(&fcc->issued_flush);
600 return ret;
601 }
740432f8 602
39d787be
CY		 603 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
604 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 605 atomic_dec(&fcc->issing_flush);
606
607 atomic_inc(&fcc->issued_flush);
740432f8
JK		 608 return ret;
609 }
6b4afdd7 610
39d787be 611 cmd.ino = ino;
adf8d90b 612 init_completion(&cmd.wait);
6b4afdd7 613
721bd4d5 614 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 615
6f890df0
CY		 616 /* update issue_list before we wake up issue_flush thread */
617 smp_mb();
618
619 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 620 wake_up(&fcc->flush_wait_queue);
6b4afdd7 621
5eba8c5d
JK		 622 if (fcc->f2fs_issue_flush) {
623 wait_for_completion(&cmd.wait);
8b8dd65f 624 atomic_dec(&fcc->issing_flush);
5eba8c5d 625 } else {
d3238691
CY		 626 struct llist_node *list;
627
628 list = llist_del_all(&fcc->issue_list);
629 if (!list) {
630 wait_for_completion(&cmd.wait);
631 atomic_dec(&fcc->issing_flush);
632 } else {
633 struct flush_cmd *tmp, *next;
634
39d787be 635 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 636
637 llist_for_each_entry_safe(tmp, next, list, llnode) {
638 if (tmp == &cmd) {
639 cmd.ret = ret;
640 atomic_dec(&fcc->issing_flush);
641 continue;
642 }
643 tmp->ret = ret;
644 complete(&tmp->wait);
645 }
646 }
5eba8c5d 647 }
adf8d90b
CY		 648
649 return cmd.ret;
6b4afdd7
JK		 650}
651
2163d198
GZ		 652int create_flush_cmd_control(struct f2fs_sb_info *sbi)
653{
654 dev_t dev = sbi->sb->s_bdev->bd_dev;
655 struct flush_cmd_control *fcc;
656 int err = 0;
657
b01a9201
JK		 658 if (SM_I(sbi)->fcc_info) {
659 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 660 if (fcc->f2fs_issue_flush)
661 return err;
5eba8c5d
JK		 662 goto init_thread;
663 }
664
2163d198
GZ		 665 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
666 if (!fcc)
667 return -ENOMEM;
8b8dd65f
CY		 668 atomic_set(&fcc->issued_flush, 0);
669 atomic_set(&fcc->issing_flush, 0);
2163d198 670 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 671 init_llist_head(&fcc->issue_list);
b01a9201 672 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 673 if (!test_opt(sbi, FLUSH_MERGE))
674 return err;
675
5eba8c5d 676init_thread:
2163d198
GZ		 677 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
678 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
679 if (IS_ERR(fcc->f2fs_issue_flush)) {
680 err = PTR_ERR(fcc->f2fs_issue_flush);
681 kfree(fcc);
b01a9201 682 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 683 return err;
684 }
2163d198
GZ		 685
686 return err;
687}
688
5eba8c5d 689void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 690{
b01a9201 691 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 692
5eba8c5d
JK		 693 if (fcc && fcc->f2fs_issue_flush) {
694 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
695
696 fcc->f2fs_issue_flush = NULL;
697 kthread_stop(flush_thread);
698 }
699 if (free) {
700 kfree(fcc);
b01a9201 701 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 702 }
2163d198
GZ		 703}
704
1228b482
CY		 705int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
706{
707 int ret = 0, i;
708
709 if (!sbi->s_ndevs)
710 return 0;
711
712 for (i = 1; i < sbi->s_ndevs; i++) {
713 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
714 continue;
715 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
716 if (ret)
717 break;
718
719 spin_lock(&sbi->dev_lock);
720 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
721 spin_unlock(&sbi->dev_lock);
722 }
723
724 return ret;
725}
726
351df4b2
JK		 727static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
728 enum dirty_type dirty_type)
729{
730 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
731
732 /* need not be added */
733 if (IS_CURSEG(sbi, segno))
734 return;
735
736 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
737 dirty_i->nr_dirty[dirty_type]++;
738
739 if (dirty_type == DIRTY) {
740 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 741 enum dirty_type t = sentry->type;
b2f2c390 742
ec325b52
JK		 743 if (unlikely(t >= DIRTY)) {
744 f2fs_bug_on(sbi, 1);
745 return;
746 }
4625d6aa
CL		 747 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
748 dirty_i->nr_dirty[t]++;
351df4b2
JK		 749 }
750}
751
752static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
753 enum dirty_type dirty_type)
754{
755 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
756
757 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
758 dirty_i->nr_dirty[dirty_type]--;
759
760 if (dirty_type == DIRTY) {
4625d6aa
CL		 761 struct seg_entry *sentry = get_seg_entry(sbi, segno);
762 enum dirty_type t = sentry->type;
763
764 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
765 dirty_i->nr_dirty[t]--;
b2f2c390 766
302bd348 767 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 768 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 769 dirty_i->victim_secmap);
351df4b2
JK		 770 }
771}
772
0a8165d7 773/*
351df4b2
JK		 774 * Should not occur error such as -ENOMEM.
775 * Adding dirty entry into seglist is not critical operation.
776 * If a given segment is one of current working segments, it won't be added.
777 */
8d8451af 778static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 779{
780 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
781 unsigned short valid_blocks;
782
783 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
784 return;
785
786 mutex_lock(&dirty_i->seglist_lock);
787
302bd348 788 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 789
790 if (valid_blocks == 0) {
791 __locate_dirty_segment(sbi, segno, PRE);
792 __remove_dirty_segment(sbi, segno, DIRTY);
793 } else if (valid_blocks < sbi->blocks_per_seg) {
794 __locate_dirty_segment(sbi, segno, DIRTY);
795 } else {
796 /* Recovery routine with SSR needs this */
797 __remove_dirty_segment(sbi, segno, DIRTY);
798 }
799
800 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 801}
802
004b6862 803static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 804 struct block_device *bdev, block_t lstart,
805 block_t start, block_t len)
275b66b0 806{
0b54fb84 807 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 808 struct list_head *pend_list;
b01a9201 809 struct discard_cmd *dc;
275b66b0 810
ba48a33e
CY		 811 f2fs_bug_on(sbi, !len);
812
813 pend_list = &dcc->pend_list[plist_idx(len)];
814
b01a9201
JK		 815 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
816 INIT_LIST_HEAD(&dc->list);
c81abe34 817 dc->bdev = bdev;
b01a9201 818 dc->lstart = lstart;
c81abe34 819 dc->start = start;
b01a9201 820 dc->len = len;
ec9895ad 821 dc->ref = 0;
15469963 822 dc->state = D_PREP;
c81abe34 823 dc->error = 0;
b01a9201 824 init_completion(&dc->wait);
22d375dd 825 list_add_tail(&dc->list, pend_list);
5f32366a 826 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 827 dcc->undiscard_blks += len;
004b6862
CY		 828
829 return dc;
830}
831
832static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
833 struct block_device *bdev, block_t lstart,
834 block_t start, block_t len,
835 struct rb_node *parent, struct rb_node **p)
836{
837 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
838 struct discard_cmd *dc;
839
840 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
841
842 rb_link_node(&dc->rb_node, parent, p);
843 rb_insert_color(&dc->rb_node, &dcc->root);
844
845 return dc;
15469963
JK		 846}
847
004b6862
CY		 848static void __detach_discard_cmd(struct discard_cmd_control *dcc,
849 struct discard_cmd *dc)
15469963 850{
dcc9165d 851 if (dc->state == D_DONE)
004b6862
CY		 852 atomic_dec(&dcc->issing_discard);
853
854 list_del(&dc->list);
855 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 856 dcc->undiscard_blks -= dc->len;
004b6862
CY		 857
858 kmem_cache_free(discard_cmd_slab, dc);
859
860 atomic_dec(&dcc->discard_cmd_cnt);
861}
862
863static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
864 struct discard_cmd *dc)
865{
866 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 867
2ec6f2ef
CY		 868 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
869
d9703d90
CY		 870 f2fs_bug_on(sbi, dc->ref);
871
c81abe34
JK		 872 if (dc->error == -EOPNOTSUPP)
873 dc->error = 0;
15469963 874
c81abe34 875 if (dc->error)
15469963 876 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 877 "Issue discard(%u, %u, %u) failed, ret: %d",
878 dc->lstart, dc->start, dc->len, dc->error);
004b6862 879 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 880}
881
c81abe34
JK		 882static void f2fs_submit_discard_endio(struct bio *bio)
883{
884 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
885
4e4cbee9 886 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 887 dc->state = D_DONE;
e31b9821 888 complete_all(&dc->wait);
c81abe34
JK		 889 bio_put(bio);
890}
891
6915ea9d
CY		 892void __check_sit_bitmap(struct f2fs_sb_info *sbi,
893 block_t start, block_t end)
894{
895#ifdef CONFIG_F2FS_CHECK_FS
896 struct seg_entry *sentry;
897 unsigned int segno;
898 block_t blk = start;
899 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
900 unsigned long *map;
901
902 while (blk < end) {
903 segno = GET_SEGNO(sbi, blk);
904 sentry = get_seg_entry(sbi, segno);
905 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
906
008396e1
YS		 907 if (end < START_BLOCK(sbi, segno + 1))
908 size = GET_BLKOFF_FROM_SEG0(sbi, end);
909 else
910 size = max_blocks;
6915ea9d
CY		 911 map = (unsigned long *)(sentry->cur_valid_map);
912 offset = __find_rev_next_bit(map, size, offset);
913 f2fs_bug_on(sbi, offset != size);
008396e1 914 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 915 }
916#endif
917}
918
c81abe34
JK		 919/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
920static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
78997b56
CY		 921 struct discard_policy *dpolicy,
922 struct discard_cmd *dc)
c81abe34
JK		 923{
924 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 925 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
926 &(dcc->fstrim_list) : &(dcc->wait_list);
c81abe34 927 struct bio *bio = NULL;
78997b56 928 int flag = dpolicy->sync ? REQ_SYNC : 0;
c81abe34
JK		 929
930 if (dc->state != D_PREP)
931 return;
932
0243a5f9
CY		 933 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
934
c81abe34
JK		 935 dc->error = __blkdev_issue_discard(dc->bdev,
936 SECTOR_FROM_BLOCK(dc->start),
937 SECTOR_FROM_BLOCK(dc->len),
938 GFP_NOFS, 0, &bio);
939 if (!dc->error) {
940 /* should keep before submission to avoid D_DONE right away */
941 dc->state = D_SUBMIT;
8b8dd65f
CY		 942 atomic_inc(&dcc->issued_discard);
943 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 944 if (bio) {
945 bio->bi_private = dc;
946 bio->bi_end_io = f2fs_submit_discard_endio;
ecc9aa00 947 bio->bi_opf |= flag;
c81abe34 948 submit_bio(bio);
8412663d 949 list_move_tail(&dc->list, wait_list);
6915ea9d 950 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 951
952 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 953 }
954 } else {
955 __remove_discard_cmd(sbi, dc);
956 }
957}
958
004b6862
CY		 959static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
960 struct block_device *bdev, block_t lstart,
961 block_t start, block_t len,
962 struct rb_node **insert_p,
963 struct rb_node *insert_parent)
c81abe34 964{
004b6862 965 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dca6951f 966 struct rb_node **p;
004b6862
CY		 967 struct rb_node *parent = NULL;
968 struct discard_cmd *dc = NULL;
969
970 if (insert_p && insert_parent) {
971 parent = insert_parent;
972 p = insert_p;
973 goto do_insert;
974 }
c81abe34 975
004b6862
CY		 976 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
977do_insert:
978 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
979 if (!dc)
980 return NULL;
c81abe34 981
004b6862 982 return dc;
c81abe34
JK		 983}
984
ba48a33e
CY		 985static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
986 struct discard_cmd *dc)
987{
988 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
989}
990
3d6a650f
YH		 991static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
992 struct discard_cmd *dc, block_t blkaddr)
993{
ba48a33e 994 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 995 struct discard_info di = dc->di;
996 bool modified = false;
3d6a650f 997
004b6862 998 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 999 __remove_discard_cmd(sbi, dc);
1000 return;
1001 }
1002
d84d1cbd
CY		 1003 dcc->undiscard_blks -= di.len;
1004
004b6862 1005 if (blkaddr > di.lstart) {
3d6a650f 1006 dc->len = blkaddr - dc->lstart;
d84d1cbd 1007 dcc->undiscard_blks += dc->len;
ba48a33e 1008 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 1009 modified = true;
1010 }
1011
1012 if (blkaddr < di.lstart + di.len - 1) {
1013 if (modified) {
1014 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1015 di.start + blkaddr + 1 - di.lstart,
1016 di.lstart + di.len - 1 - blkaddr,
1017 NULL, NULL);
1018 } else {
1019 dc->lstart++;
1020 dc->len--;
1021 dc->start++;
d84d1cbd 1022 dcc->undiscard_blks += dc->len;
ba48a33e 1023 __relocate_discard_cmd(dcc, dc);
004b6862 1024 }
3d6a650f
YH		 1025 }
1026}
1027
004b6862
CY		 1028static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1029 struct block_device *bdev, block_t lstart,
1030 block_t start, block_t len)
275b66b0 1031{
0b54fb84 1032 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 1033 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1034 struct discard_cmd *dc;
1035 struct discard_info di = {0};
1036 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1037 block_t end = lstart + len;
275b66b0 1038
15469963 1039 mutex_lock(&dcc->cmd_lock);
40465257 1040
004b6862
CY		 1041 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1042 NULL, lstart,
1043 (struct rb_entry **)&prev_dc,
1044 (struct rb_entry **)&next_dc,
1045 &insert_p, &insert_parent, true);
1046 if (dc)
1047 prev_dc = dc;
1048
1049 if (!prev_dc) {
1050 di.lstart = lstart;
1051 di.len = next_dc ? next_dc->lstart - lstart : len;
1052 di.len = min(di.len, len);
1053 di.start = start;
22d375dd 1054 }
15469963 1055
004b6862
CY		 1056 while (1) {
1057 struct rb_node *node;
1058 bool merged = false;
1059 struct discard_cmd *tdc = NULL;
1060
1061 if (prev_dc) {
1062 di.lstart = prev_dc->lstart + prev_dc->len;
1063 if (di.lstart < lstart)
1064 di.lstart = lstart;
1065 if (di.lstart >= end)
1066 break;
1067
1068 if (!next_dc || next_dc->lstart > end)
1069 di.len = end - di.lstart;
1070 else
1071 di.len = next_dc->lstart - di.lstart;
1072 di.start = start + di.lstart - lstart;
1073 }
1074
1075 if (!di.len)
1076 goto next;
1077
1078 if (prev_dc && prev_dc->state == D_PREP &&
1079 prev_dc->bdev == bdev &&
1080 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1081 prev_dc->di.len += di.len;
d84d1cbd 1082 dcc->undiscard_blks += di.len;
ba48a33e 1083 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1084 di = prev_dc->di;
1085 tdc = prev_dc;
1086 merged = true;
1087 }
1088
1089 if (next_dc && next_dc->state == D_PREP &&
1090 next_dc->bdev == bdev &&
1091 __is_discard_front_mergeable(&di, &next_dc->di)) {
1092 next_dc->di.lstart = di.lstart;
1093 next_dc->di.len += di.len;
1094 next_dc->di.start = di.start;
d84d1cbd 1095 dcc->undiscard_blks += di.len;
ba48a33e 1096 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1097 if (tdc)
1098 __remove_discard_cmd(sbi, tdc);
004b6862 1099 merged = true;
4e6a8d9b 1100 }
004b6862 1101
df0f6b44 1102 if (!merged) {
004b6862
CY		 1103 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1104 di.len, NULL, NULL);
df0f6b44 1105 }
004b6862
CY		 1106 next:
1107 prev_dc = next_dc;
1108 if (!prev_dc)
1109 break;
1110
1111 node = rb_next(&prev_dc->rb_node);
1112 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1113 }
1114
1115 mutex_unlock(&dcc->cmd_lock);
1116}
1117
1118static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1119 struct block_device *bdev, block_t blkstart, block_t blklen)
1120{
1121 block_t lblkstart = blkstart;
1122
0243a5f9 1123 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1124
1125 if (sbi->s_ndevs) {
1126 int devi = f2fs_target_device_index(sbi, blkstart);
1127
1128 blkstart -= FDEV(devi).start_blk;
1129 }
1130 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1131 return 0;
1132}
1133
8412663d 1134static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1135 struct discard_policy *dpolicy,
1136 unsigned int start, unsigned int end)
8412663d
CY		 1137{
1138 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1139 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1140 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1141 struct discard_cmd *dc;
1142 struct blk_plug plug;
1143 int issued;
1144
1145next:
1146 issued = 0;
1147
1148 mutex_lock(&dcc->cmd_lock);
1149 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1150
1151 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1152 NULL, start,
1153 (struct rb_entry **)&prev_dc,
1154 (struct rb_entry **)&next_dc,
1155 &insert_p, &insert_parent, true);
1156 if (!dc)
1157 dc = next_dc;
1158
1159 blk_start_plug(&plug);
1160
1161 while (dc && dc->lstart <= end) {
1162 struct rb_node *node;
1163
78997b56 1164 if (dc->len < dpolicy->granularity)
8412663d
CY		 1165 goto skip;
1166
1167 if (dc->state != D_PREP) {
1168 list_move_tail(&dc->list, &dcc->fstrim_list);
1169 goto skip;
1170 }
1171
78997b56 1172 __submit_discard_cmd(sbi, dpolicy, dc);
8412663d 1173
ecc9aa00 1174 if (++issued >= dpolicy->max_requests) {
8412663d
CY		 1175 start = dc->lstart + dc->len;
1176
1177 blk_finish_plug(&plug);
1178 mutex_unlock(&dcc->cmd_lock);
1179
1180 schedule();
1181
1182 goto next;
1183 }
1184skip:
1185 node = rb_next(&dc->rb_node);
1186 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1187
1188 if (fatal_signal_pending(current))
1189 break;
1190 }
1191
1192 blk_finish_plug(&plug);
1193 mutex_unlock(&dcc->cmd_lock);
1194}
1195
78997b56
CY		 1196static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1197 struct discard_policy *dpolicy)
bd5b0738
CY		 1198{
1199 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1200 struct list_head *pend_list;
1201 struct discard_cmd *dc, *tmp;
1202 struct blk_plug plug;
78997b56 1203 int i, iter = 0, issued = 0;
e6c6de18 1204 bool io_interrupted = false;
bd5b0738 1205
78997b56
CY		 1206 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1207 if (i + 1 < dpolicy->granularity)
1208 break;
bd5b0738 1209 pend_list = &dcc->pend_list[i];
33da62cf
CY		 1210
1211 mutex_lock(&dcc->cmd_lock);
1212 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1213 blk_start_plug(&plug);
bd5b0738
CY		 1214 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1215 f2fs_bug_on(sbi, dc->state != D_PREP);
1216
ecc9aa00
CY		 1217 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1218 !is_idle(sbi)) {
e6c6de18 1219 io_interrupted = true;
ecc9aa00 1220 goto skip;
969d1b18 1221 }
e6c6de18 1222
78997b56 1223 __submit_discard_cmd(sbi, dpolicy, dc);
ecc9aa00
CY		 1224 issued++;
1225skip:
1226 if (++iter >= dpolicy->max_requests)
33da62cf 1227 break;
bd5b0738 1228 }
33da62cf
CY		 1229 blk_finish_plug(&plug);
1230 mutex_unlock(&dcc->cmd_lock);
1231
1232 if (iter >= dpolicy->max_requests)
1233 break;
bd5b0738 1234 }
969d1b18 1235
e6c6de18
CY		 1236 if (!issued && io_interrupted)
1237 issued = -1;
1238
969d1b18
CY		 1239 return issued;
1240}
1241
cf5c759f 1242static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
969d1b18
CY		 1243{
1244 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1245 struct list_head *pend_list;
1246 struct discard_cmd *dc, *tmp;
1247 int i;
cf5c759f 1248 bool dropped = false;
969d1b18
CY		 1249
1250 mutex_lock(&dcc->cmd_lock);
1251 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1252 pend_list = &dcc->pend_list[i];
1253 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1254 f2fs_bug_on(sbi, dc->state != D_PREP);
1255 __remove_discard_cmd(sbi, dc);
cf5c759f 1256 dropped = true;
969d1b18
CY		 1257 }
1258 }
1259 mutex_unlock(&dcc->cmd_lock);
cf5c759f
CY		 1260
1261 return dropped;
bd5b0738
CY		 1262}
1263
0ea80512 1264static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
2a510c00
CY		 1265 struct discard_cmd *dc)
1266{
1267 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0ea80512 1268 unsigned int len = 0;
2a510c00
CY		 1269
1270 wait_for_completion_io(&dc->wait);
1271 mutex_lock(&dcc->cmd_lock);
1272 f2fs_bug_on(sbi, dc->state != D_DONE);
1273 dc->ref--;
0ea80512
CY		 1274 if (!dc->ref) {
1275 if (!dc->error)
1276 len = dc->len;
2a510c00 1277 __remove_discard_cmd(sbi, dc);
0ea80512 1278 }
2a510c00 1279 mutex_unlock(&dcc->cmd_lock);
0ea80512
CY		 1280
1281 return len;
2a510c00
CY		 1282}
1283
0ea80512 1284static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
78997b56
CY		 1285 struct discard_policy *dpolicy,
1286 block_t start, block_t end)
63a94fa1
CY		 1287{
1288 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56
CY		 1289 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1290 &(dcc->fstrim_list) : &(dcc->wait_list);
63a94fa1 1291 struct discard_cmd *dc, *tmp;
6afae633 1292 bool need_wait;
0ea80512 1293 unsigned int trimmed = 0;
6afae633
CY		 1294
1295next:
1296 need_wait = false;
63a94fa1
CY		 1297
1298 mutex_lock(&dcc->cmd_lock);
1299 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1300 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1301 continue;
78997b56 1302 if (dc->len < dpolicy->granularity)
8412663d 1303 continue;
78997b56 1304 if (dc->state == D_DONE && !dc->ref) {
63a94fa1 1305 wait_for_completion_io(&dc->wait);
0ea80512
CY		 1306 if (!dc->error)
1307 trimmed += dc->len;
63a94fa1 1308 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1309 } else {
1310 dc->ref++;
1311 need_wait = true;
1312 break;
63a94fa1
CY		 1313 }
1314 }
1315 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1316
1317 if (need_wait) {
0ea80512 1318 trimmed += __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1319 goto next;
1320 }
0ea80512
CY		 1321
1322 return trimmed;
63a94fa1
CY		 1323}
1324
78997b56
CY		 1325static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1326 struct discard_policy *dpolicy)
8412663d 1327{
78997b56 1328 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
8412663d
CY		 1329}
1330
004b6862
CY		 1331/* This should be covered by global mutex, &sit_i->sentry_lock */
1332void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1333{
1334 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1335 struct discard_cmd *dc;
ec9895ad 1336 bool need_wait = false;
004b6862
CY		 1337
1338 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1339 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1340 if (dc) {
ec9895ad
CY		 1341 if (dc->state == D_PREP) {
1342 __punch_discard_cmd(sbi, dc, blkaddr);
1343 } else {
1344 dc->ref++;
1345 need_wait = true;
1346 }
275b66b0 1347 }
d431413f 1348 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1349
2a510c00
CY		 1350 if (need_wait)
1351 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1352}
1353
cce13252
CY		 1354void stop_discard_thread(struct f2fs_sb_info *sbi)
1355{
1356 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1357
1358 if (dcc && dcc->f2fs_issue_discard) {
1359 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1360
1361 dcc->f2fs_issue_discard = NULL;
1362 kthread_stop(discard_thread);
ec9895ad 1363 }
d431413f
CY		 1364}
1365
8412663d 1366/* This comes from f2fs_put_super */
cf5c759f 1367bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
969d1b18
CY		 1368{
1369 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
78997b56 1370 struct discard_policy dpolicy;
cf5c759f 1371 bool dropped;
969d1b18 1372
78997b56
CY		 1373 init_discard_policy(&dpolicy, DPOLICY_UMOUNT, dcc->discard_granularity);
1374 __issue_discard_cmd(sbi, &dpolicy);
cf5c759f 1375 dropped = __drop_discard_cmd(sbi);
78997b56 1376 __wait_all_discard_cmd(sbi, &dpolicy);
cf5c759f
CY		 1377
1378 return dropped;
969d1b18
CY		 1379}
1380
15469963
JK		 1381static int issue_discard_thread(void *data)
1382{
1383 struct f2fs_sb_info *sbi = data;
1384 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1385 wait_queue_head_t *q = &dcc->discard_wait_queue;
78997b56 1386 struct discard_policy dpolicy;
969d1b18
CY		 1387 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1388 int issued;
15469963 1389
1d7be270 1390 set_freezable();
15469963 1391
1d7be270 1392 do {
78997b56
CY		 1393 init_discard_policy(&dpolicy, DPOLICY_BG,
1394 dcc->discard_granularity);
1395
969d1b18
CY		 1396 wait_event_interruptible_timeout(*q,
1397 kthread_should_stop() || freezing(current) ||
1398 dcc->discard_wake,
1399 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1400 if (try_to_freeze())
1401 continue;
1402 if (kthread_should_stop())
1403 return 0;
15469963 1404
5f656541 1405 if (dcc->discard_wake) {
969d1b18 1406 dcc->discard_wake = 0;
5f656541 1407 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
78997b56
CY		 1408 init_discard_policy(&dpolicy,
1409 DPOLICY_FORCE, 1);
5f656541 1410 }
969d1b18 1411
dc6febb6
CY		 1412 sb_start_intwrite(sbi->sb);
1413
78997b56 1414 issued = __issue_discard_cmd(sbi, &dpolicy);
969d1b18 1415 if (issued) {
78997b56
CY		 1416 __wait_all_discard_cmd(sbi, &dpolicy);
1417 wait_ms = dpolicy.min_interval;
969d1b18 1418 } else {
78997b56 1419 wait_ms = dpolicy.max_interval;
969d1b18 1420 }
1d7be270 1421
dc6febb6 1422 sb_end_intwrite(sbi->sb);
1d7be270 1423
1d7be270
JK		 1424 } while (!kthread_should_stop());
1425 return 0;
15469963
JK		 1426}
1427
f46e8809 1428#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1429static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1430 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1431{
92592285 1432 sector_t sector, nr_sects;
10a875f8 1433 block_t lblkstart = blkstart;
3c62be17
JK		 1434 int devi = 0;
1435
1436 if (sbi->s_ndevs) {
1437 devi = f2fs_target_device_index(sbi, blkstart);
1438 blkstart -= FDEV(devi).start_blk;
1439 }
f46e8809
DLM		 1440
1441 /*
1442 * We need to know the type of the zone: for conventional zones,
1443 * use regular discard if the drive supports it. For sequential
1444 * zones, reset the zone write pointer.
1445 */
3c62be17 1446 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1447
1448 case BLK_ZONE_TYPE_CONVENTIONAL:
1449 if (!blk_queue_discard(bdev_get_queue(bdev)))
1450 return 0;
c81abe34 1451 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1452 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1453 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1454 sector = SECTOR_FROM_BLOCK(blkstart);
1455 nr_sects = SECTOR_FROM_BLOCK(blklen);
1456
1457 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1458 nr_sects != bdev_zone_sectors(bdev)) {
1459 f2fs_msg(sbi->sb, KERN_INFO,
1460 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1461 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1462 blkstart, blklen);
1463 return -EIO;
1464 }
d50aaeec 1465 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1466 return blkdev_reset_zones(bdev, sector,
1467 nr_sects, GFP_NOFS);
1468 default:
1469 /* Unknown zone type: broken device ? */
1470 return -EIO;
1471 }
1472}
1473#endif
1474
3c62be17
JK		 1475static int __issue_discard_async(struct f2fs_sb_info *sbi,
1476 struct block_device *bdev, block_t blkstart, block_t blklen)
1477{
1478#ifdef CONFIG_BLK_DEV_ZONED
1479 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1480 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1481 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1482#endif
c81abe34 1483 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1484}
1485
1e87a78d 1486static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1487 block_t blkstart, block_t blklen)
1488{
3c62be17
JK		 1489 sector_t start = blkstart, len = 0;
1490 struct block_device *bdev;
a66cdd98
JK		 1491 struct seg_entry *se;
1492 unsigned int offset;
1493 block_t i;
3c62be17
JK		 1494 int err = 0;
1495
1496 bdev = f2fs_target_device(sbi, blkstart, NULL);
1497
1498 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1499 if (i != start) {
1500 struct block_device *bdev2 =
1501 f2fs_target_device(sbi, i, NULL);
1502
1503 if (bdev2 != bdev) {
1504 err = __issue_discard_async(sbi, bdev,
1505 start, len);
1506 if (err)
1507 return err;
1508 bdev = bdev2;
1509 start = i;
1510 len = 0;
1511 }
1512 }
a66cdd98 1513
a66cdd98
JK		 1514 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1515 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1516
1517 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1518 sbi->discard_blks--;
1519 }
f46e8809 1520
3c62be17
JK		 1521 if (len)
1522 err = __issue_discard_async(sbi, bdev, start, len);
1523 return err;
1e87a78d
JK		 1524}
1525
25290fa5
JK		 1526static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1527 bool check_only)
adf4983b 1528{
b2955550
JK		 1529 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1530 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1531 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1532 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1533 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1534 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1535 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1536 unsigned int start = 0, end = -1;
c473f1a9 1537 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1538 struct discard_entry *de = NULL;
46f84c2c 1539 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1540 int i;
1541
3e025740 1542 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1543 return false;
b2955550 1544
a66cdd98
JK		 1545 if (!force) {
1546 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1547 SM_I(sbi)->dcc_info->nr_discards >=
1548 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1549 return false;
4b2fecc8
JK		 1550 }
1551
b2955550
JK		 1552 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1553 for (i = 0; i < entries; i++)
a66cdd98 1554 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1555 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1556
0b54fb84
JK		 1557 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1558 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1559 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1560 if (start >= max_blocks)
1561 break;
1562
1563 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1564 if (force && start && end != max_blocks
1565 && (end - start) < cpc->trim_minlen)
1566 continue;
1567
25290fa5
JK		 1568 if (check_only)
1569 return true;
1570
a7eeb823
CY		 1571 if (!de) {
1572 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1573 GFP_F2FS_ZERO);
1574 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1575 list_add_tail(&de->list, head);
1576 }
1577
1578 for (i = start; i < end; i++)
1579 __set_bit_le(i, (void *)de->discard_map);
1580
1581 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1582 }
25290fa5 1583 return false;
b2955550
JK		 1584}
1585
4b2fecc8
JK		 1586void release_discard_addrs(struct f2fs_sb_info *sbi)
1587{
46f84c2c 1588 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1589 struct discard_entry *entry, *this;
1590
1591 /* drop caches */
1592 list_for_each_entry_safe(entry, this, head, list) {
1593 list_del(&entry->list);
1594 kmem_cache_free(discard_entry_slab, entry);
1595 }
1596}
1597
0a8165d7 1598/*
351df4b2
JK		 1599 * Should call clear_prefree_segments after checkpoint is done.
1600 */
1601static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1602{
1603 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1604 unsigned int segno;
351df4b2
JK		 1605
1606 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1607 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1608 __set_test_and_free(sbi, segno);
351df4b2
JK		 1609 mutex_unlock(&dirty_i->seglist_lock);
1610}
1611
836b5a63 1612void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1613{
969d1b18
CY		 1614 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1615 struct list_head *head = &dcc->entry_list;
2d7b822a 1616 struct discard_entry *entry, *this;
351df4b2 1617 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1618 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1619 unsigned int start = 0, end = -1;
36abef4e 1620 unsigned int secno, start_segno;
c473f1a9 1621 bool force = (cpc->reason & CP_DISCARD);
ad5d3c45 1622 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
351df4b2
JK		 1623
1624 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1625
351df4b2 1626 while (1) {
29e59c14 1627 int i;
ad5d3c45
YS		 1628
1629 if (need_align && end != -1)
1630 end--;
7cd8558b
JK		 1631 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1632 if (start >= MAIN_SEGS(sbi))
351df4b2 1633 break;
7cd8558b
JK		 1634 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1635 start + 1);
29e59c14 1636
ad5d3c45
YS		 1637 if (need_align) {
1638 start = rounddown(start, sbi->segs_per_sec);
1639 end = roundup(end, sbi->segs_per_sec);
1640 }
29e59c14 1641
ad5d3c45
YS		 1642 for (i = start; i < end; i++) {
1643 if (test_and_clear_bit(i, prefree_map))
1644 dirty_i->nr_dirty[PRE]--;
1645 }
29e59c14 1646
650d3c4e 1647 if (!test_opt(sbi, DISCARD))
29e59c14 1648 continue;
351df4b2 1649
650d3c4e
YH		 1650 if (force && start >= cpc->trim_start &&
1651 (end - 1) <= cpc->trim_end)
1652 continue;
1653
36abef4e
JK		 1654 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1655 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1656 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1657 continue;
1658 }
1659next:
4ddb1a4d
JK		 1660 secno = GET_SEC_FROM_SEG(sbi, start);
1661 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1662 if (!IS_CURSEC(sbi, secno) &&
302bd348 1663 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1664 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1665 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1666
1667 start = start_segno + sbi->segs_per_sec;
1668 if (start < end)
1669 goto next;
8b107f5b
JK		 1670 else
1671 end = start - 1;
351df4b2
JK		 1672 }
1673 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1674
1675 /* send small discards */
2d7b822a 1676 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1677 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1678 bool is_valid = test_bit_le(0, entry->discard_map);
1679
1680find_next:
1681 if (is_valid) {
1682 next_pos = find_next_zero_bit_le(entry->discard_map,
1683 sbi->blocks_per_seg, cur_pos);
1684 len = next_pos - cur_pos;
1685
acfd2810
DLM		 1686 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1687 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1688 goto skip;
1689
1690 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1691 len);
a7eeb823
CY		 1692 total_len += len;
1693 } else {
1694 next_pos = find_next_bit_le(entry->discard_map,
1695 sbi->blocks_per_seg, cur_pos);
1696 }
836b5a63 1697skip:
a7eeb823
CY		 1698 cur_pos = next_pos;
1699 is_valid = !is_valid;
1700
1701 if (cur_pos < sbi->blocks_per_seg)
1702 goto find_next;
1703
b2955550 1704 list_del(&entry->list);
969d1b18 1705 dcc->nr_discards -= total_len;
b2955550
JK		 1706 kmem_cache_free(discard_entry_slab, entry);
1707 }
34e159da 1708
01983c71 1709 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1710}
1711
78997b56
CY		 1712void init_discard_policy(struct discard_policy *dpolicy,
1713 int discard_type, unsigned int granularity)
ecc9aa00 1714{
78997b56
CY		 1715 /* common policy */
1716 dpolicy->type = discard_type;
ecc9aa00 1717 dpolicy->sync = true;
78997b56
CY		 1718 dpolicy->granularity = granularity;
1719
1720 if (discard_type == DPOLICY_BG) {
1721 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1722 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1723 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1724 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1725 dpolicy->io_aware = true;
1726 } else if (discard_type == DPOLICY_FORCE) {
1727 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1728 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1729 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1730 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1731 dpolicy->io_aware = true;
1732 } else if (discard_type == DPOLICY_FSTRIM) {
1733 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1734 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1735 dpolicy->io_aware = false;
1736 } else if (discard_type == DPOLICY_UMOUNT) {
1737 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1738 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1739 dpolicy->io_aware = false;
1740 }
ecc9aa00
CY		 1741}
1742
8ed59745 1743static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1744{
15469963 1745 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1746 struct discard_cmd_control *dcc;
ba48a33e 1747 int err = 0, i;
0b54fb84
JK		 1748
1749 if (SM_I(sbi)->dcc_info) {
1750 dcc = SM_I(sbi)->dcc_info;
1751 goto init_thread;
1752 }
1753
1754 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1755 if (!dcc)
1756 return -ENOMEM;
1757
969d1b18 1758 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1759 INIT_LIST_HEAD(&dcc->entry_list);
78997b56 1760 for (i = 0; i < MAX_PLIST_NUM; i++)
ba48a33e 1761 INIT_LIST_HEAD(&dcc->pend_list[i]);
46f84c2c 1762 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1763 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1764 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1765 atomic_set(&dcc->issued_discard, 0);
1766 atomic_set(&dcc->issing_discard, 0);
5f32366a 1767 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1768 dcc->nr_discards = 0;
d618ebaf 1769 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1770 dcc->undiscard_blks = 0;
004b6862 1771 dcc->root = RB_ROOT;
0b54fb84 1772
15469963 1773 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1774 SM_I(sbi)->dcc_info = dcc;
1775init_thread:
15469963
JK		 1776 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1777 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1778 if (IS_ERR(dcc->f2fs_issue_discard)) {
1779 err = PTR_ERR(dcc->f2fs_issue_discard);
1780 kfree(dcc);
1781 SM_I(sbi)->dcc_info = NULL;
1782 return err;
1783 }
1784
0b54fb84
JK		 1785 return err;
1786}
1787
f099405f 1788static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1789{
1790 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1791
f099405f
CY		 1792 if (!dcc)
1793 return;
1794
cce13252 1795 stop_discard_thread(sbi);
f099405f
CY		 1796
1797 kfree(dcc);
1798 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1799}
1800
184a5cd2 1801static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1802{
1803 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1804
1805 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1806 sit_i->dirty_sentries++;
184a5cd2
CY		 1807 return false;
1808 }
1809
1810 return true;
351df4b2
JK		 1811}
1812
1813static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1814 unsigned int segno, int modified)
1815{
1816 struct seg_entry *se = get_seg_entry(sbi, segno);
1817 se->type = type;
1818 if (modified)
1819 __mark_sit_entry_dirty(sbi, segno);
1820}
1821
1822static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1823{
1824 struct seg_entry *se;
1825 unsigned int segno, offset;
1826 long int new_vblocks;
6415fedc
YS		 1827 bool exist;
1828#ifdef CONFIG_F2FS_CHECK_FS
1829 bool mir_exist;
1830#endif
351df4b2
JK		 1831
1832 segno = GET_SEGNO(sbi, blkaddr);
1833
1834 se = get_seg_entry(sbi, segno);
1835 new_vblocks = se->valid_blocks + del;
491c0854 1836 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1837
9850cf4a 1838 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1839 (new_vblocks > sbi->blocks_per_seg)));
1840
1841 se->valid_blocks = new_vblocks;
1842 se->mtime = get_mtime(sbi);
1843 SIT_I(sbi)->max_mtime = se->mtime;
1844
1845 /* Update valid block bitmap */
1846 if (del > 0) {
6415fedc 1847 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1848#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1849 mir_exist = f2fs_test_and_set_bit(offset,
1850 se->cur_valid_map_mir);
1851 if (unlikely(exist != mir_exist)) {
1852 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1853 "when setting bitmap, blk:%u, old bit:%d",
1854 blkaddr, exist);
05796763 1855 f2fs_bug_on(sbi, 1);
6415fedc 1856 }
355e7891 1857#endif
6415fedc
YS		 1858 if (unlikely(exist)) {
1859 f2fs_msg(sbi->sb, KERN_ERR,
1860 "Bitmap was wrongly set, blk:%u", blkaddr);
1861 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1862 se->valid_blocks--;
1863 del = 0;
355e7891 1864 }
6415fedc 1865
3e025740
JK		 1866 if (f2fs_discard_en(sbi) &&
1867 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1868 sbi->discard_blks--;
720037f9
JK		 1869
1870 /* don't overwrite by SSR to keep node chain */
1871 if (se->type == CURSEG_WARM_NODE) {
1872 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1873 se->ckpt_valid_blocks++;
1874 }
351df4b2 1875 } else {
6415fedc 1876 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1877#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1878 mir_exist = f2fs_test_and_clear_bit(offset,
1879 se->cur_valid_map_mir);
1880 if (unlikely(exist != mir_exist)) {
1881 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1882 "when clearing bitmap, blk:%u, old bit:%d",
1883 blkaddr, exist);
05796763 1884 f2fs_bug_on(sbi, 1);
6415fedc 1885 }
355e7891 1886#endif
6415fedc
YS		 1887 if (unlikely(!exist)) {
1888 f2fs_msg(sbi->sb, KERN_ERR,
1889 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1890 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1891 se->valid_blocks++;
1892 del = 0;
355e7891 1893 }
6415fedc 1894
3e025740
JK		 1895 if (f2fs_discard_en(sbi) &&
1896 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1897 sbi->discard_blks++;
351df4b2
JK		 1898 }
1899 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1900 se->ckpt_valid_blocks += del;
1901
1902 __mark_sit_entry_dirty(sbi, segno);
1903
1904 /* update total number of valid blocks to be written in ckpt area */
1905 SIT_I(sbi)->written_valid_blocks += del;
1906
1907 if (sbi->segs_per_sec > 1)
1908 get_sec_entry(sbi, segno)->valid_blocks += del;
1909}
1910
351df4b2
JK		 1911void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1912{
1913 unsigned int segno = GET_SEGNO(sbi, addr);
1914 struct sit_info *sit_i = SIT_I(sbi);
1915
9850cf4a 1916 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1917 if (addr == NEW_ADDR)
1918 return;
1919
1920 /* add it into sit main buffer */
3d26fa6b 1921 down_write(&sit_i->sentry_lock);
351df4b2
JK		 1922
1923 update_sit_entry(sbi, addr, -1);
1924
1925 /* add it into dirty seglist */
1926 locate_dirty_segment(sbi, segno);
1927
3d26fa6b 1928 up_write(&sit_i->sentry_lock);
351df4b2
JK		 1929}
1930
6e2c64ad
JK		 1931bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1932{
1933 struct sit_info *sit_i = SIT_I(sbi);
1934 unsigned int segno, offset;
1935 struct seg_entry *se;
1936 bool is_cp = false;
1937
1938 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1939 return true;
1940
3d26fa6b 1941 down_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1942
1943 segno = GET_SEGNO(sbi, blkaddr);
1944 se = get_seg_entry(sbi, segno);
1945 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1946
1947 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1948 is_cp = true;
1949
3d26fa6b 1950 up_read(&sit_i->sentry_lock);
6e2c64ad
JK		 1951
1952 return is_cp;
1953}
1954
0a8165d7 1955/*
351df4b2
JK		 1956 * This function should be resided under the curseg_mutex lock
1957 */
1958static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1959 struct f2fs_summary *sum)
351df4b2
JK		 1960{
1961 struct curseg_info *curseg = CURSEG_I(sbi, type);
1962 void *addr = curseg->sum_blk;
e79efe3b 1963 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1964 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1965}
1966
0a8165d7 1967/*
351df4b2
JK		 1968 * Calculate the number of current summary pages for writing
1969 */
3fa06d7b 1970int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1971{
351df4b2 1972 int valid_sum_count = 0;
9a47938b 1973 int i, sum_in_page;
351df4b2
JK		 1974
1975 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1976 if (sbi->ckpt->alloc_type[i] == SSR)
1977 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1978 else {
1979 if (for_ra)
1980 valid_sum_count += le16_to_cpu(
1981 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1982 else
1983 valid_sum_count += curseg_blkoff(sbi, i);
1984 }
351df4b2
JK		 1985 }
1986
09cbfeaf 1987 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1988 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1989 if (valid_sum_count <= sum_in_page)
351df4b2 1990 return 1;
9a47938b 1991 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1992 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1993 return 2;
1994 return 3;
1995}
1996
0a8165d7 1997/*
351df4b2
JK		 1998 * Caller should put this summary page
1999 */
2000struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2001{
2002 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
2003}
2004
381722d2 2005void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 2006{
2007 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2 2008
0537b811 2009 memcpy(page_address(page), src, PAGE_SIZE);
351df4b2
JK		 2010 set_page_dirty(page);
2011 f2fs_put_page(page, 1);
2012}
2013
381722d2
CY		 2014static void write_sum_page(struct f2fs_sb_info *sbi,
2015 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2016{
2017 update_meta_page(sbi, (void *)sum_blk, blk_addr);
2018}
2019
b7ad7512
CY		 2020static void write_current_sum_page(struct f2fs_sb_info *sbi,
2021 int type, block_t blk_addr)
2022{
2023 struct curseg_info *curseg = CURSEG_I(sbi, type);
2024 struct page *page = grab_meta_page(sbi, blk_addr);
2025 struct f2fs_summary_block *src = curseg->sum_blk;
2026 struct f2fs_summary_block *dst;
2027
2028 dst = (struct f2fs_summary_block *)page_address(page);
fb0053ba 2029 memset(dst, 0, PAGE_SIZE);
b7ad7512
CY		 2030
2031 mutex_lock(&curseg->curseg_mutex);
2032
2033 down_read(&curseg->journal_rwsem);
2034 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2035 up_read(&curseg->journal_rwsem);
2036
2037 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2038 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2039
2040 mutex_unlock(&curseg->curseg_mutex);
2041
2042 set_page_dirty(page);
2043 f2fs_put_page(page, 1);
2044}
2045
a7881893
JK		 2046static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2047{
2048 struct curseg_info *curseg = CURSEG_I(sbi, type);
2049 unsigned int segno = curseg->segno + 1;
2050 struct free_segmap_info *free_i = FREE_I(sbi);
2051
2052 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2053 return !test_bit(segno, free_i->free_segmap);
2054 return 0;
2055}
2056
0a8165d7 2057/*
351df4b2
JK		 2058 * Find a new segment from the free segments bitmap to right order
2059 * This function should be returned with success, otherwise BUG
2060 */
2061static void get_new_segment(struct f2fs_sb_info *sbi,
2062 unsigned int *newseg, bool new_sec, int dir)
2063{
2064 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 2065 unsigned int segno, secno, zoneno;
7cd8558b 2066 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 2067 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2068 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 2069 unsigned int left_start = hint;
2070 bool init = true;
2071 int go_left = 0;
2072 int i;
2073
1a118ccf 2074 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 2075
2076 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2077 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2078 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2079 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2080 goto got_it;
2081 }
2082find_other_zone:
7cd8558b
JK		 2083 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2084 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2085 if (dir == ALLOC_RIGHT) {
2086 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2087 MAIN_SECS(sbi), 0);
2088 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2089 } else {
2090 go_left = 1;
2091 left_start = hint - 1;
2092 }
2093 }
2094 if (go_left == 0)
2095 goto skip_left;
2096
2097 while (test_bit(left_start, free_i->free_secmap)) {
2098 if (left_start > 0) {
2099 left_start--;
2100 continue;
2101 }
2102 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2103 MAIN_SECS(sbi), 0);
2104 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2105 break;
2106 }
2107 secno = left_start;
2108skip_left:
4ddb1a4d
JK		 2109 segno = GET_SEG_FROM_SEC(sbi, secno);
2110 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2111
2112 /* give up on finding another zone */
2113 if (!init)
2114 goto got_it;
2115 if (sbi->secs_per_zone == 1)
2116 goto got_it;
2117 if (zoneno == old_zoneno)
2118 goto got_it;
2119 if (dir == ALLOC_LEFT) {
2120 if (!go_left && zoneno + 1 >= total_zones)
2121 goto got_it;
2122 if (go_left && zoneno == 0)
2123 goto got_it;
2124 }
2125 for (i = 0; i < NR_CURSEG_TYPE; i++)
2126 if (CURSEG_I(sbi, i)->zone == zoneno)
2127 break;
2128
2129 if (i < NR_CURSEG_TYPE) {
2130 /* zone is in user, try another */
2131 if (go_left)
2132 hint = zoneno * sbi->secs_per_zone - 1;
2133 else if (zoneno + 1 >= total_zones)
2134 hint = 0;
2135 else
2136 hint = (zoneno + 1) * sbi->secs_per_zone;
2137 init = false;
2138 goto find_other_zone;
2139 }
2140got_it:
2141 /* set it as dirty segment in free segmap */
9850cf4a 2142 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2143 __set_inuse(sbi, segno);
2144 *newseg = segno;
1a118ccf 2145 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2146}
2147
2148static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2149{
2150 struct curseg_info *curseg = CURSEG_I(sbi, type);
2151 struct summary_footer *sum_footer;
2152
2153 curseg->segno = curseg->next_segno;
4ddb1a4d 2154 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2155 curseg->next_blkoff = 0;
2156 curseg->next_segno = NULL_SEGNO;
2157
2158 sum_footer = &(curseg->sum_blk->footer);
2159 memset(sum_footer, 0, sizeof(struct summary_footer));
2160 if (IS_DATASEG(type))
2161 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2162 if (IS_NODESEG(type))
2163 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2164 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2165}
2166
7a20b8a6
JK		 2167static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2168{
a7881893
JK		 2169 /* if segs_per_sec is large than 1, we need to keep original policy. */
2170 if (sbi->segs_per_sec != 1)
2171 return CURSEG_I(sbi, type)->segno;
2172
482eb99a
YS		 2173 if (test_opt(sbi, NOHEAP) &&
2174 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
7a20b8a6
JK		 2175 return 0;
2176
e066b83c
JK		 2177 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2178 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK		 2179 return CURSEG_I(sbi, type)->segno;
2180}
2181
0a8165d7 2182/*
351df4b2
JK		 2183 * Allocate a current working segment.
2184 * This function always allocates a free segment in LFS manner.
2185 */
2186static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2187{
2188 struct curseg_info *curseg = CURSEG_I(sbi, type);
2189 unsigned int segno = curseg->segno;
2190 int dir = ALLOC_LEFT;
2191
2192 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2193 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2194 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2195 dir = ALLOC_RIGHT;
2196
2197 if (test_opt(sbi, NOHEAP))
2198 dir = ALLOC_RIGHT;
2199
7a20b8a6 2200 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2201 get_new_segment(sbi, &segno, new_sec, dir);
2202 curseg->next_segno = segno;
2203 reset_curseg(sbi, type, 1);
2204 curseg->alloc_type = LFS;
2205}
2206
2207static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2208 struct curseg_info *seg, block_t start)
2209{
2210 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2211 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2212 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2213 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2214 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2215 int i, pos;
2216
2217 for (i = 0; i < entries; i++)
2218 target_map[i] = ckpt_map[i] | cur_map[i];
2219
2220 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2221
2222 seg->next_blkoff = pos;
351df4b2
JK		 2223}
2224
0a8165d7 2225/*
351df4b2
JK		 2226 * If a segment is written by LFS manner, next block offset is just obtained
2227 * by increasing the current block offset. However, if a segment is written by
2228 * SSR manner, next block offset obtained by calling __next_free_blkoff
2229 */
2230static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2231 struct curseg_info *seg)
2232{
2233 if (seg->alloc_type == SSR)
2234 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2235 else
2236 seg->next_blkoff++;
2237}
2238
0a8165d7 2239/*
e1c42045 2240 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2241 * manner, so it should recover the existing segment information of valid blocks
2242 */
025d63a4 2243static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2244{
2245 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2246 struct curseg_info *curseg = CURSEG_I(sbi, type);
2247 unsigned int new_segno = curseg->next_segno;
2248 struct f2fs_summary_block *sum_node;
2249 struct page *sum_page;
2250
2251 write_sum_page(sbi, curseg->sum_blk,
2252 GET_SUM_BLOCK(sbi, curseg->segno));
2253 __set_test_and_inuse(sbi, new_segno);
2254
2255 mutex_lock(&dirty_i->seglist_lock);
2256 __remove_dirty_segment(sbi, new_segno, PRE);
2257 __remove_dirty_segment(sbi, new_segno, DIRTY);
2258 mutex_unlock(&dirty_i->seglist_lock);
2259
2260 reset_curseg(sbi, type, 1);
2261 curseg->alloc_type = SSR;
2262 __next_free_blkoff(sbi, curseg, 0);
2263
025d63a4
CY		 2264 sum_page = get_sum_page(sbi, new_segno);
2265 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2266 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2267 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2268}
2269
43727527
JK		 2270static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2271{
2272 struct curseg_info *curseg = CURSEG_I(sbi, type);
2273 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2274 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2275 int i, cnt;
2276 bool reversed = false;
c192f7a4
JK		 2277
2278 /* need_SSR() already forces to do this */
e066b83c
JK		 2279 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2280 curseg->next_segno = segno;
c192f7a4 2281 return 1;
e066b83c 2282 }
43727527 2283
70d625cb
JK		 2284 /* For node segments, let's do SSR more intensively */
2285 if (IS_NODESEG(type)) {
d27c3d89
CY		 2286 if (type >= CURSEG_WARM_NODE) {
2287 reversed = true;
2288 i = CURSEG_COLD_NODE;
2289 } else {
2290 i = CURSEG_HOT_NODE;
2291 }
2292 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2293 } else {
d27c3d89
CY		 2294 if (type >= CURSEG_WARM_DATA) {
2295 reversed = true;
2296 i = CURSEG_COLD_DATA;
2297 } else {
2298 i = CURSEG_HOT_DATA;
2299 }
2300 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2301 }
43727527 2302
d27c3d89 2303 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2304 if (i == type)
2305 continue;
e066b83c
JK		 2306 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2307 curseg->next_segno = segno;
43727527 2308 return 1;
e066b83c 2309 }
c192f7a4 2310 }
43727527
JK		 2311 return 0;
2312}
2313
351df4b2
JK		 2314/*
2315 * flush out current segment and replace it with new segment
2316 * This function should be returned with success, otherwise BUG
2317 */
2318static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2319 int type, bool force)
2320{
a7881893
JK		 2321 struct curseg_info *curseg = CURSEG_I(sbi, type);
2322
7b405275 2323 if (force)
351df4b2 2324 new_curseg(sbi, type, true);
5b6c6be2
JK		 2325 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2326 type == CURSEG_WARM_NODE)
351df4b2 2327 new_curseg(sbi, type, false);
a7881893
JK		 2328 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2329 new_curseg(sbi, type, false);
351df4b2 2330 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2331 change_curseg(sbi, type);
351df4b2
JK		 2332 else
2333 new_curseg(sbi, type, false);
dcdfff65 2334
a7881893 2335 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2336}
2337
2338void allocate_new_segments(struct f2fs_sb_info *sbi)
2339{
6ae1be13
JK		 2340 struct curseg_info *curseg;
2341 unsigned int old_segno;
351df4b2
JK		 2342 int i;
2343
3d26fa6b
CY		 2344 down_write(&SIT_I(sbi)->sentry_lock);
2345
6ae1be13
JK		 2346 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2347 curseg = CURSEG_I(sbi, i);
2348 old_segno = curseg->segno;
2349 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2350 locate_dirty_segment(sbi, old_segno);
2351 }
3d26fa6b
CY		 2352
2353 up_write(&SIT_I(sbi)->sentry_lock);
351df4b2
JK		 2354}
2355
2356static const struct segment_allocation default_salloc_ops = {
2357 .allocate_segment = allocate_segment_by_default,
2358};
2359
25290fa5
JK		 2360bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2361{
2362 __u64 trim_start = cpc->trim_start;
2363 bool has_candidate = false;
2364
3d26fa6b 2365 down_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2366 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2367 if (add_discard_addrs(sbi, cpc, true)) {
2368 has_candidate = true;
2369 break;
2370 }
2371 }
3d26fa6b 2372 up_write(&SIT_I(sbi)->sentry_lock);
25290fa5
JK		 2373
2374 cpc->trim_start = trim_start;
2375 return has_candidate;
2376}
2377
4b2fecc8
JK		 2378int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2379{
f7ef9b83
JK		 2380 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2381 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
8412663d
CY		 2382 unsigned int start_segno, end_segno, cur_segno;
2383 block_t start_block, end_block;
4b2fecc8 2384 struct cp_control cpc;
78997b56 2385 struct discard_policy dpolicy;
0ea80512 2386 unsigned long long trimmed = 0;
c34f42e2 2387 int err = 0;
ad5d3c45 2388 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
4b2fecc8 2389
836b5a63 2390 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2391 return -EINVAL;
2392
7cd8558b 2393 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2394 goto out;
2395
ed214a11
YH		 2396 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2397 f2fs_msg(sbi->sb, KERN_WARNING,
2398 "Found FS corruption, run fsck to fix.");
2399 goto out;
2400 }
2401
4b2fecc8 2402 /* start/end segment number in main_area */
7cd8558b
JK		 2403 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2404 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2405 GET_SEGNO(sbi, end);
ad5d3c45
YS		 2406 if (need_align) {
2407 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2408 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2409 }
8412663d 2410
4b2fecc8 2411 cpc.reason = CP_DISCARD;
836b5a63 2412 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2413
2414 /* do checkpoint to issue discard commands safely */
8412663d
CY		 2415 for (cur_segno = start_segno; cur_segno <= end_segno;
2416 cur_segno = cpc.trim_end + 1) {
2417 cpc.trim_start = cur_segno;
a66cdd98
JK		 2418
2419 if (sbi->discard_blks == 0)
2420 break;
2421 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2422 cpc.trim_end = end_segno;
2423 else
2424 cpc.trim_end = min_t(unsigned int,
8412663d 2425 rounddown(cur_segno +
bba681cb
JK		 2426 BATCHED_TRIM_SEGMENTS(sbi),
2427 sbi->segs_per_sec) - 1, end_segno);
2428
2429 mutex_lock(&sbi->gc_mutex);
c34f42e2 2430 err = write_checkpoint(sbi, &cpc);
bba681cb 2431 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 2432 if (err)
2433 break;
74fa5f3d
CY		 2434
2435 schedule();
bba681cb 2436 }
8412663d
CY		 2437
2438 start_block = START_BLOCK(sbi, start_segno);
2439 end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2440
78997b56
CY		 2441 init_discard_policy(&dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2442 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
0ea80512
CY		 2443 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2444 start_block, end_block);
4b2fecc8 2445out:
0ea80512 2446 range->len = F2FS_BLK_TO_BYTES(trimmed);
c34f42e2 2447 return err;
4b2fecc8
JK		 2448}
2449
351df4b2
JK		 2450static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2451{
2452 struct curseg_info *curseg = CURSEG_I(sbi, type);
2453 if (curseg->next_blkoff < sbi->blocks_per_seg)
2454 return true;
2455 return false;
2456}
2457
4f0a03d3
HL		 2458int rw_hint_to_seg_type(enum rw_hint hint)
2459{
2460 switch (hint) {
2461 case WRITE_LIFE_SHORT:
2462 return CURSEG_HOT_DATA;
2463 case WRITE_LIFE_EXTREME:
2464 return CURSEG_COLD_DATA;
2465 default:
2466 return CURSEG_WARM_DATA;
2467 }
2468}
2469
81377bd6 2470static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2471{
81377bd6 2472 if (fio->type == DATA)
351df4b2
JK		 2473 return CURSEG_HOT_DATA;
2474 else
2475 return CURSEG_HOT_NODE;
2476}
2477
81377bd6 2478static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2479{
81377bd6
JK		 2480 if (fio->type == DATA) {
2481 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2482
2483 if (S_ISDIR(inode->i_mode))
2484 return CURSEG_HOT_DATA;
2485 else
2486 return CURSEG_COLD_DATA;
2487 } else {
81377bd6 2488 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2489 return CURSEG_WARM_NODE;
351df4b2
JK		 2490 else
2491 return CURSEG_COLD_NODE;
2492 }
2493}
2494
81377bd6 2495static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2496{
81377bd6
JK		 2497 if (fio->type == DATA) {
2498 struct inode *inode = fio->page->mapping->host;
351df4b2 2499
81377bd6 2500 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2501 return CURSEG_COLD_DATA;
ef095d19
JK		 2502 if (is_inode_flag_set(inode, FI_HOT_DATA))
2503 return CURSEG_HOT_DATA;
4f0a03d3 2504 return rw_hint_to_seg_type(inode->i_write_hint);
351df4b2 2505 } else {
81377bd6
JK		 2506 if (IS_DNODE(fio->page))
2507 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2508 CURSEG_HOT_NODE;
ef095d19 2509 return CURSEG_COLD_NODE;
351df4b2
JK		 2510 }
2511}
2512
81377bd6 2513static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2514{
a912b54d
JK		 2515 int type = 0;
2516
81377bd6 2517 switch (fio->sbi->active_logs) {
351df4b2 2518 case 2:
a912b54d
JK		 2519 type = __get_segment_type_2(fio);
2520 break;
351df4b2 2521 case 4:
a912b54d
JK		 2522 type = __get_segment_type_4(fio);
2523 break;
2524 case 6:
2525 type = __get_segment_type_6(fio);
2526 break;
2527 default:
2528 f2fs_bug_on(fio->sbi, true);
351df4b2 2529 }
81377bd6 2530
a912b54d
JK		 2531 if (IS_HOT(type))
2532 fio->temp = HOT;
2533 else if (IS_WARM(type))
2534 fio->temp = WARM;
2535 else
2536 fio->temp = COLD;
2537 return type;
351df4b2
JK		 2538}
2539
bfad7c2d
JK		 2540void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2541 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2542 struct f2fs_summary *sum, int type,
2543 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2544{
2545 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2546 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 2547
2b60311d
CY		 2548 down_read(&SM_I(sbi)->curseg_lock);
2549
351df4b2 2550 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2551 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2552
2553 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2554
4e6a8d9b
JK		 2555 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2556
351df4b2
JK		 2557 /*
2558 * __add_sum_entry should be resided under the curseg_mutex
2559 * because, this function updates a summary entry in the
2560 * current summary block.
2561 */
e79efe3b 2562 __add_sum_entry(sbi, type, sum);
351df4b2 2563
351df4b2 2564 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2565
2566 stat_inc_block_count(sbi, curseg);
351df4b2 2567
65f1b80b
YS		 2568 /*
2569 * SIT information should be updated before segment allocation,
2570 * since SSR needs latest valid block information.
2571 */
2572 update_sit_entry(sbi, *new_blkaddr, 1);
2573 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2574 update_sit_entry(sbi, old_blkaddr, -1);
2575
c6f82fe9
JK		 2576 if (!__has_curseg_space(sbi, type))
2577 sit_i->s_ops->allocate_segment(sbi, type, false);
65f1b80b 2578
351df4b2 2579 /*
65f1b80b
YS		 2580 * segment dirty status should be updated after segment allocation,
2581 * so we just need to update status only one time after previous
2582 * segment being closed.
351df4b2 2583 */
65f1b80b
YS		 2584 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2585 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
5e443818 2586
3d26fa6b 2587 up_write(&sit_i->sentry_lock);
351df4b2 2588
704956ec 2589 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2590 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2591
704956ec
CY		 2592 f2fs_inode_chksum_set(sbi, page);
2593 }
2594
fb830fc5
CY		 2595 if (add_list) {
2596 struct f2fs_bio_info *io;
2597
2598 INIT_LIST_HEAD(&fio->list);
2599 fio->in_list = true;
2600 io = sbi->write_io[fio->type] + fio->temp;
2601 spin_lock(&io->io_lock);
2602 list_add_tail(&fio->list, &io->io_list);
2603 spin_unlock(&io->io_lock);
2604 }
2605
bfad7c2d 2606 mutex_unlock(&curseg->curseg_mutex);
2b60311d
CY		 2607
2608 up_read(&SM_I(sbi)->curseg_lock);
bfad7c2d
JK		 2609}
2610
39d787be
CY		 2611static void update_device_state(struct f2fs_io_info *fio)
2612{
2613 struct f2fs_sb_info *sbi = fio->sbi;
2614 unsigned int devidx;
2615
2616 if (!sbi->s_ndevs)
2617 return;
2618
2619 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2620
2621 /* update device state for fsync */
2622 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2623
2624 /* update device state for checkpoint */
2625 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2626 spin_lock(&sbi->dev_lock);
2627 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2628 spin_unlock(&sbi->dev_lock);
2629 }
39d787be
CY		 2630}
2631
05ca3632 2632static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2633{
81377bd6 2634 int type = __get_segment_type(fio);
0a595eba 2635 int err;
bfad7c2d 2636
0a595eba 2637reallocate:
7a9d7548 2638 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2639 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2640
351df4b2 2641 /* writeout dirty page into bdev */
b9109b0e 2642 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2643 if (err == -EAGAIN) {
2644 fio->old_blkaddr = fio->new_blkaddr;
2645 goto reallocate;
39d787be
CY		 2646 } else if (!err) {
2647 update_device_state(fio);
0a595eba 2648 }
351df4b2
JK		 2649}
2650
b0af6d49
CY		 2651void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2652 enum iostat_type io_type)
351df4b2 2653{
458e6197 2654 struct f2fs_io_info fio = {
05ca3632 2655 .sbi = sbi,
458e6197 2656 .type = META,
04d328de 2657 .op = REQ_OP_WRITE,
70fd7614 2658 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2659 .old_blkaddr = page->index,
2660 .new_blkaddr = page->index,
05ca3632 2661 .page = page,
4375a336 2662 .encrypted_page = NULL,
fb830fc5 2663 .in_list = false,
458e6197
JK		 2664 };
2665
2b947003 2666 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2667 fio.op_flags &= ~REQ_META;
2b947003 2668
351df4b2 2669 set_page_writeback(page);
b9109b0e 2670 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2671
2672 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2673}
2674
05ca3632 2675void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2676{
2677 struct f2fs_summary sum;
05ca3632 2678
351df4b2 2679 set_summary(&sum, nid, 0, 0);
05ca3632 2680 do_write_page(&sum, fio);
b0af6d49
CY		 2681
2682 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2683}
2684
05ca3632 2685void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2686{
05ca3632 2687 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2688 struct f2fs_summary sum;
2689 struct node_info ni;
2690
9850cf4a 2691 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2692 get_node_info(sbi, dn->nid, &ni);
2693 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2694 do_write_page(&sum, fio);
f28b3434 2695 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2696
2697 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2698}
2699
d1b3e72d 2700int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2701{
b0af6d49
CY		 2702 int err;
2703
7a9d7548 2704 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2705 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2706
2707 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2708 if (!err)
2709 update_device_state(fio);
b0af6d49
CY		 2710
2711 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2712
2713 return err;
351df4b2
JK		 2714}
2715
2b60311d
CY		 2716static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2717 unsigned int segno)
2718{
2719 int i;
2720
2721 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2722 if (CURSEG_I(sbi, i)->segno == segno)
2723 break;
2724 }
2725 return i;
2726}
2727
4356e48e 2728void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2729 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2730 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2731{
2732 struct sit_info *sit_i = SIT_I(sbi);
2733 struct curseg_info *curseg;
2734 unsigned int segno, old_cursegno;
2735 struct seg_entry *se;
2736 int type;
19f106bc 2737 unsigned short old_blkoff;
351df4b2
JK		 2738
2739 segno = GET_SEGNO(sbi, new_blkaddr);
2740 se = get_seg_entry(sbi, segno);
2741 type = se->type;
2742
2b60311d
CY		 2743 down_write(&SM_I(sbi)->curseg_lock);
2744
19f106bc
CY		 2745 if (!recover_curseg) {
2746 /* for recovery flow */
2747 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2748 if (old_blkaddr == NULL_ADDR)
2749 type = CURSEG_COLD_DATA;
2750 else
2751 type = CURSEG_WARM_DATA;
2752 }
2753 } else {
2b60311d
CY		 2754 if (IS_CURSEG(sbi, segno)) {
2755 /* se->type is volatile as SSR allocation */
2756 type = __f2fs_get_curseg(sbi, segno);
2757 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2758 } else {
351df4b2 2759 type = CURSEG_WARM_DATA;
2b60311d 2760 }
351df4b2 2761 }
19f106bc 2762
351df4b2
JK		 2763 curseg = CURSEG_I(sbi, type);
2764
2765 mutex_lock(&curseg->curseg_mutex);
3d26fa6b 2766 down_write(&sit_i->sentry_lock);
351df4b2
JK		 2767
2768 old_cursegno = curseg->segno;
19f106bc 2769 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2770
2771 /* change the current segment */
2772 if (segno != curseg->segno) {
2773 curseg->next_segno = segno;
025d63a4 2774 change_curseg(sbi, type);
351df4b2
JK		 2775 }
2776
491c0854 2777 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2778 __add_sum_entry(sbi, type, sum);
351df4b2 2779
28bc106b 2780 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2781 update_sit_entry(sbi, new_blkaddr, 1);
2782 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2783 update_sit_entry(sbi, old_blkaddr, -1);
2784
2785 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2786 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2787
351df4b2 2788 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2789
19f106bc
CY		 2790 if (recover_curseg) {
2791 if (old_cursegno != curseg->segno) {
2792 curseg->next_segno = old_cursegno;
025d63a4 2793 change_curseg(sbi, type);
19f106bc
CY		 2794 }
2795 curseg->next_blkoff = old_blkoff;
2796 }
2797
3d26fa6b 2798 up_write(&sit_i->sentry_lock);
351df4b2 2799 mutex_unlock(&curseg->curseg_mutex);
2b60311d 2800 up_write(&SM_I(sbi)->curseg_lock);
351df4b2
JK		 2801}
2802
528e3459
CY		 2803void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2804 block_t old_addr, block_t new_addr,
28bc106b
CY		 2805 unsigned char version, bool recover_curseg,
2806 bool recover_newaddr)
528e3459
CY		 2807{
2808 struct f2fs_summary sum;
2809
2810 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2811
28bc106b
CY		 2812 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2813 recover_curseg, recover_newaddr);
528e3459 2814
f28b3434 2815 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2816}
2817
93dfe2ac 2818void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2819 enum page_type type, bool ordered)
93dfe2ac 2820{
93dfe2ac 2821 if (PageWriteback(page)) {
4081363f
JK		 2822 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2823
b9109b0e
JK		 2824 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2825 0, page->index, type);
fec1d657
JK		 2826 if (ordered)
2827 wait_on_page_writeback(page);
2828 else
2829 wait_for_stable_page(page);
93dfe2ac
JK		 2830 }
2831}
2832
d4c759ee 2833void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2834{
2835 struct page *cpage;
2836
5d4c0af4 2837 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2838 return;
2839
08b39fbd
CY		 2840 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2841 if (cpage) {
fec1d657 2842 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2843 f2fs_put_page(cpage, 1);
2844 }
2845}
2846
351df4b2
JK		 2847static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2848{
2849 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2850 struct curseg_info *seg_i;
2851 unsigned char *kaddr;
2852 struct page *page;
2853 block_t start;
2854 int i, j, offset;
2855
2856 start = start_sum_block(sbi);
2857
2858 page = get_meta_page(sbi, start++);
2859 kaddr = (unsigned char *)page_address(page);
2860
2861 /* Step 1: restore nat cache */
2862 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2863 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2864
2865 /* Step 2: restore sit cache */
2866 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2867 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2868 offset = 2 * SUM_JOURNAL_SIZE;
2869
2870 /* Step 3: restore summary entries */
2871 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2872 unsigned short blk_off;
2873 unsigned int segno;
2874
2875 seg_i = CURSEG_I(sbi, i);
2876 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2877 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2878 seg_i->next_segno = segno;
2879 reset_curseg(sbi, i, 0);
2880 seg_i->alloc_type = ckpt->alloc_type[i];
2881 seg_i->next_blkoff = blk_off;
2882
2883 if (seg_i->alloc_type == SSR)
2884 blk_off = sbi->blocks_per_seg;
2885
2886 for (j = 0; j < blk_off; j++) {
2887 struct f2fs_summary *s;
2888 s = (struct f2fs_summary *)(kaddr + offset);
2889 seg_i->sum_blk->entries[j] = *s;
2890 offset += SUMMARY_SIZE;
09cbfeaf 2891 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2892 SUM_FOOTER_SIZE)
2893 continue;
2894
2895 f2fs_put_page(page, 1);
2896 page = NULL;
2897
2898 page = get_meta_page(sbi, start++);
2899 kaddr = (unsigned char *)page_address(page);
2900 offset = 0;
2901 }
2902 }
2903 f2fs_put_page(page, 1);
2904 return 0;
2905}
2906
2907static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2908{
2909 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2910 struct f2fs_summary_block *sum;
2911 struct curseg_info *curseg;
2912 struct page *new;
2913 unsigned short blk_off;
2914 unsigned int segno = 0;
2915 block_t blk_addr = 0;
2916
2917 /* get segment number and block addr */
2918 if (IS_DATASEG(type)) {
2919 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2920 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2921 CURSEG_HOT_DATA]);
119ee914 2922 if (__exist_node_summaries(sbi))
351df4b2
JK		 2923 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2924 else
2925 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2926 } else {
2927 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2928 CURSEG_HOT_NODE]);
2929 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2930 CURSEG_HOT_NODE]);
119ee914 2931 if (__exist_node_summaries(sbi))
351df4b2
JK		 2932 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2933 type - CURSEG_HOT_NODE);
2934 else
2935 blk_addr = GET_SUM_BLOCK(sbi, segno);
2936 }
2937
2938 new = get_meta_page(sbi, blk_addr);
2939 sum = (struct f2fs_summary_block *)page_address(new);
2940
2941 if (IS_NODESEG(type)) {
119ee914 2942 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2943 struct f2fs_summary *ns = &sum->entries[0];
2944 int i;
2945 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2946 ns->version = 0;
2947 ns->ofs_in_node = 0;
2948 }
2949 } else {
d653788a
GZ		 2950 int err;
2951
2952 err = restore_node_summary(sbi, segno, sum);
2953 if (err) {
351df4b2 2954 f2fs_put_page(new, 1);
d653788a 2955 return err;
351df4b2
JK		 2956 }
2957 }
2958 }
2959
2960 /* set uncompleted segment to curseg */
2961 curseg = CURSEG_I(sbi, type);
2962 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2963
2964 /* update journal info */
2965 down_write(&curseg->journal_rwsem);
2966 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2967 up_write(&curseg->journal_rwsem);
2968
2969 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2970 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2971 curseg->next_segno = segno;
2972 reset_curseg(sbi, type, 0);
2973 curseg->alloc_type = ckpt->alloc_type[type];
2974 curseg->next_blkoff = blk_off;
2975 mutex_unlock(&curseg->curseg_mutex);
2976 f2fs_put_page(new, 1);
2977 return 0;
2978}
2979
2980static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2981{
21d3f8e1
JQ		 2982 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2983 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2984 int type = CURSEG_HOT_DATA;
e4fc5fbf 2985 int err;
351df4b2 2986
aaec2b1d 2987 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2988 int npages = npages_for_summary_flush(sbi, true);
2989
2990 if (npages >= 2)
2991 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2992 META_CP, true);
3fa06d7b 2993
351df4b2
JK		 2994 /* restore for compacted data summary */
2995 if (read_compacted_summaries(sbi))
2996 return -EINVAL;
2997 type = CURSEG_HOT_NODE;
2998 }
2999
119ee914 3000 if (__exist_node_summaries(sbi))
3fa06d7b 3001 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 3002 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 3003
e4fc5fbf
CY		 3004 for (; type <= CURSEG_COLD_NODE; type++) {
3005 err = read_normal_summaries(sbi, type);
3006 if (err)
3007 return err;
3008 }
3009
21d3f8e1
JQ		 3010 /* sanity check for summary blocks */
3011 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3012 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3013 return -EINVAL;
3014
351df4b2
JK		 3015 return 0;
3016}
3017
3018static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3019{
3020 struct page *page;
3021 unsigned char *kaddr;
3022 struct f2fs_summary *summary;
3023 struct curseg_info *seg_i;
3024 int written_size = 0;
3025 int i, j;
3026
3027 page = grab_meta_page(sbi, blkaddr++);
3028 kaddr = (unsigned char *)page_address(page);
fb0053ba 3029 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3030
3031 /* Step 1: write nat cache */
3032 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 3033 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3034 written_size += SUM_JOURNAL_SIZE;
3035
3036 /* Step 2: write sit cache */
3037 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3038 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 3039 written_size += SUM_JOURNAL_SIZE;
3040
351df4b2
JK		 3041 /* Step 3: write summary entries */
3042 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3043 unsigned short blkoff;
3044 seg_i = CURSEG_I(sbi, i);
3045 if (sbi->ckpt->alloc_type[i] == SSR)
3046 blkoff = sbi->blocks_per_seg;
3047 else
3048 blkoff = curseg_blkoff(sbi, i);
3049
3050 for (j = 0; j < blkoff; j++) {
3051 if (!page) {
3052 page = grab_meta_page(sbi, blkaddr++);
3053 kaddr = (unsigned char *)page_address(page);
fb0053ba 3054 memset(kaddr, 0, PAGE_SIZE);
351df4b2
JK		 3055 written_size = 0;
3056 }
3057 summary = (struct f2fs_summary *)(kaddr + written_size);
3058 *summary = seg_i->sum_blk->entries[j];
3059 written_size += SUMMARY_SIZE;
351df4b2 3060
09cbfeaf 3061 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 3062 SUM_FOOTER_SIZE)
3063 continue;
3064
e8d61a74 3065 set_page_dirty(page);
351df4b2
JK		 3066 f2fs_put_page(page, 1);
3067 page = NULL;
3068 }
3069 }
e8d61a74
CY		 3070 if (page) {
3071 set_page_dirty(page);
351df4b2 3072 f2fs_put_page(page, 1);
e8d61a74 3073 }
351df4b2
JK		 3074}
3075
3076static void write_normal_summaries(struct f2fs_sb_info *sbi,
3077 block_t blkaddr, int type)
3078{
3079 int i, end;
3080 if (IS_DATASEG(type))
3081 end = type + NR_CURSEG_DATA_TYPE;
3082 else
3083 end = type + NR_CURSEG_NODE_TYPE;
3084
b7ad7512
CY		 3085 for (i = type; i < end; i++)
3086 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 3087}
3088
3089void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3090{
aaec2b1d 3091 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 3092 write_compacted_summaries(sbi, start_blk);
3093 else
3094 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3095}
3096
3097void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3098{
119ee914 3099 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 3100}
3101
dfc08a12 3102int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 3103 unsigned int val, int alloc)
3104{
3105 int i;
3106
3107 if (type == NAT_JOURNAL) {
dfc08a12
CY		 3108 for (i = 0; i < nats_in_cursum(journal); i++) {
3109 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 3110 return i;
3111 }
dfc08a12
CY		 3112 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3113 return update_nats_in_cursum(journal, 1);
351df4b2 3114 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 3115 for (i = 0; i < sits_in_cursum(journal); i++)
3116 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 3117 return i;
dfc08a12
CY		 3118 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3119 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 3120 }
3121 return -1;
3122}
3123
3124static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3125 unsigned int segno)
3126{
2cc22186 3127 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 3128}
3129
3130static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3131 unsigned int start)
3132{
3133 struct sit_info *sit_i = SIT_I(sbi);
3134 struct page *src_page, *dst_page;
3135 pgoff_t src_off, dst_off;
3136 void *src_addr, *dst_addr;
3137
3138 src_off = current_sit_addr(sbi, start);
3139 dst_off = next_sit_addr(sbi, src_off);
3140
3141 /* get current sit block page without lock */
3142 src_page = get_meta_page(sbi, src_off);
3143 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 3144 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 3145
3146 src_addr = page_address(src_page);
3147 dst_addr = page_address(dst_page);
09cbfeaf 3148 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 3149
3150 set_page_dirty(dst_page);
3151 f2fs_put_page(src_page, 1);
3152
3153 set_to_next_sit(sit_i, start);
3154
3155 return dst_page;
3156}
3157
184a5cd2
CY		 3158static struct sit_entry_set *grab_sit_entry_set(void)
3159{
3160 struct sit_entry_set *ses =
80c54505 3161 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3162
3163 ses->entry_cnt = 0;
3164 INIT_LIST_HEAD(&ses->set_list);
3165 return ses;
3166}
3167
3168static void release_sit_entry_set(struct sit_entry_set *ses)
3169{
3170 list_del(&ses->set_list);
3171 kmem_cache_free(sit_entry_set_slab, ses);
3172}
3173
3174static void adjust_sit_entry_set(struct sit_entry_set *ses,
3175 struct list_head *head)
3176{
3177 struct sit_entry_set *next = ses;
3178
3179 if (list_is_last(&ses->set_list, head))
3180 return;
3181
3182 list_for_each_entry_continue(next, head, set_list)
3183 if (ses->entry_cnt <= next->entry_cnt)
3184 break;
3185
3186 list_move_tail(&ses->set_list, &next->set_list);
3187}
3188
3189static void add_sit_entry(unsigned int segno, struct list_head *head)
3190{
3191 struct sit_entry_set *ses;
3192 unsigned int start_segno = START_SEGNO(segno);
3193
3194 list_for_each_entry(ses, head, set_list) {
3195 if (ses->start_segno == start_segno) {
3196 ses->entry_cnt++;
3197 adjust_sit_entry_set(ses, head);
3198 return;
3199 }
3200 }
3201
3202 ses = grab_sit_entry_set();
3203
3204 ses->start_segno = start_segno;
3205 ses->entry_cnt++;
3206 list_add(&ses->set_list, head);
3207}
3208
3209static void add_sits_in_set(struct f2fs_sb_info *sbi)
3210{
3211 struct f2fs_sm_info *sm_info = SM_I(sbi);
3212 struct list_head *set_list = &sm_info->sit_entry_set;
3213 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3214 unsigned int segno;
3215
7cd8558b 3216 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3217 add_sit_entry(segno, set_list);
3218}
3219
3220static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3221{
3222 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3223 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3224 int i;
3225
b7ad7512 3226 down_write(&curseg->journal_rwsem);
dfc08a12 3227 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3228 unsigned int segno;
3229 bool dirtied;
3230
dfc08a12 3231 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3232 dirtied = __mark_sit_entry_dirty(sbi, segno);
3233
3234 if (!dirtied)
3235 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3236 }
dfc08a12 3237 update_sits_in_cursum(journal, -i);
b7ad7512 3238 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3239}
3240
0a8165d7 3241/*
351df4b2
JK		 3242 * CP calls this function, which flushes SIT entries including sit_journal,
3243 * and moves prefree segs to free segs.
3244 */
4b2fecc8 3245void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3246{
3247 struct sit_info *sit_i = SIT_I(sbi);
3248 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3249 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3250 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3251 struct sit_entry_set *ses, *tmp;
3252 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3253 bool to_journal = true;
4b2fecc8 3254 struct seg_entry *se;
351df4b2 3255
3d26fa6b 3256 down_write(&sit_i->sentry_lock);
351df4b2 3257
2b11a74b
WL		 3258 if (!sit_i->dirty_sentries)
3259 goto out;
3260
351df4b2 3261 /*
184a5cd2
CY		 3262 * add and account sit entries of dirty bitmap in sit entry
3263 * set temporarily
351df4b2 3264 */
184a5cd2 3265 add_sits_in_set(sbi);
351df4b2 3266
184a5cd2
CY		 3267 /*
3268 * if there are no enough space in journal to store dirty sit
3269 * entries, remove all entries from journal and add and account
3270 * them in sit entry set.
3271 */
dfc08a12 3272 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3273 remove_sits_in_journal(sbi);
b2955550 3274
184a5cd2
CY		 3275 /*
3276 * there are two steps to flush sit entries:
3277 * #1, flush sit entries to journal in current cold data summary block.
3278 * #2, flush sit entries to sit page.
3279 */
3280 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3281 struct page *page = NULL;
184a5cd2
CY		 3282 struct f2fs_sit_block *raw_sit = NULL;
3283 unsigned int start_segno = ses->start_segno;
3284 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3285 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3286 unsigned int segno = start_segno;
3287
3288 if (to_journal &&
dfc08a12 3289 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3290 to_journal = false;
3291
b7ad7512
CY		 3292 if (to_journal) {
3293 down_write(&curseg->journal_rwsem);
3294 } else {
184a5cd2
CY		 3295 page = get_next_sit_page(sbi, start_segno);
3296 raw_sit = page_address(page);
351df4b2 3297 }
351df4b2 3298
184a5cd2
CY		 3299 /* flush dirty sit entries in region of current sit set */
3300 for_each_set_bit_from(segno, bitmap, end) {
3301 int offset, sit_offset;
4b2fecc8
JK		 3302
3303 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 3304
3305 /* add discard candidates */
c473f1a9 3306 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3307 cpc->trim_start = segno;
25290fa5 3308 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3309 }
184a5cd2
CY		 3310
3311 if (to_journal) {
dfc08a12 3312 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3313 SIT_JOURNAL, segno, 1);
3314 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3315 segno_in_journal(journal, offset) =
184a5cd2
CY		 3316 cpu_to_le32(segno);
3317 seg_info_to_raw_sit(se,
dfc08a12 3318 &sit_in_journal(journal, offset));
184a5cd2
CY		 3319 } else {
3320 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3321 seg_info_to_raw_sit(se,
3322 &raw_sit->entries[sit_offset]);
3323 }
351df4b2 3324
184a5cd2
CY		 3325 __clear_bit(segno, bitmap);
3326 sit_i->dirty_sentries--;
3327 ses->entry_cnt--;
351df4b2
JK		 3328 }
3329
b7ad7512
CY		 3330 if (to_journal)
3331 up_write(&curseg->journal_rwsem);
3332 else
184a5cd2
CY		 3333 f2fs_put_page(page, 1);
3334
3335 f2fs_bug_on(sbi, ses->entry_cnt);
3336 release_sit_entry_set(ses);
351df4b2 3337 }
184a5cd2
CY		 3338
3339 f2fs_bug_on(sbi, !list_empty(head));
3340 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3341out:
c473f1a9 3342 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3343 __u64 trim_start = cpc->trim_start;
3344
4b2fecc8 3345 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3346 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3347
3348 cpc->trim_start = trim_start;
4b2fecc8 3349 }
3d26fa6b 3350 up_write(&sit_i->sentry_lock);
351df4b2 3351
351df4b2
JK		 3352 set_prefree_as_free_segments(sbi);
3353}
3354
3355static int build_sit_info(struct f2fs_sb_info *sbi)
3356{
3357 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3358 struct sit_info *sit_i;
3359 unsigned int sit_segs, start;
ae27d62e 3360 char *src_bitmap;
351df4b2
JK		 3361 unsigned int bitmap_size;
3362
3363 /* allocate memory for SIT information */
3364 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3365 if (!sit_i)
3366 return -ENOMEM;
3367
3368 SM_I(sbi)->sit_info = sit_i;
3369
a7c3e901 3370 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3371 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3372 if (!sit_i->sentries)
3373 return -ENOMEM;
3374
7cd8558b 3375 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3376 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3377 if (!sit_i->dirty_sentries_bitmap)
3378 return -ENOMEM;
3379
7cd8558b 3380 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3381 sit_i->sentries[start].cur_valid_map
3382 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3383 sit_i->sentries[start].ckpt_valid_map
3384 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3385 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3386 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3387 return -ENOMEM;
3e025740 3388
355e7891
CY		 3389#ifdef CONFIG_F2FS_CHECK_FS
3390 sit_i->sentries[start].cur_valid_map_mir
3391 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3392 if (!sit_i->sentries[start].cur_valid_map_mir)
3393 return -ENOMEM;
3394#endif
3395
3e025740
JK		 3396 if (f2fs_discard_en(sbi)) {
3397 sit_i->sentries[start].discard_map
3398 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3399 if (!sit_i->sentries[start].discard_map)
3400 return -ENOMEM;
3401 }
351df4b2
JK		 3402 }
3403
60a3b782
JK		 3404 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3405 if (!sit_i->tmp_map)
3406 return -ENOMEM;
3407
351df4b2 3408 if (sbi->segs_per_sec > 1) {
a7c3e901 3409 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3410 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3411 if (!sit_i->sec_entries)
3412 return -ENOMEM;
3413 }
3414
3415 /* get information related with SIT */
3416 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3417
3418 /* setup SIT bitmap from ckeckpoint pack */
3419 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3420 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3421
ae27d62e
CY		 3422 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3423 if (!sit_i->sit_bitmap)
351df4b2 3424 return -ENOMEM;
351df4b2 3425
ae27d62e
CY		 3426#ifdef CONFIG_F2FS_CHECK_FS
3427 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3428 if (!sit_i->sit_bitmap_mir)
3429 return -ENOMEM;
3430#endif
3431
351df4b2
JK		 3432 /* init SIT information */
3433 sit_i->s_ops = &default_salloc_ops;
3434
3435 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3436 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3437 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3438 sit_i->bitmap_size = bitmap_size;
3439 sit_i->dirty_sentries = 0;
3440 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3441 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3442 sit_i->mounted_time = ktime_get_real_seconds();
3d26fa6b 3443 init_rwsem(&sit_i->sentry_lock);
351df4b2
JK		 3444 return 0;
3445}
3446
3447static int build_free_segmap(struct f2fs_sb_info *sbi)
3448{
351df4b2
JK		 3449 struct free_segmap_info *free_i;
3450 unsigned int bitmap_size, sec_bitmap_size;
3451
3452 /* allocate memory for free segmap information */
3453 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3454 if (!free_i)
3455 return -ENOMEM;
3456
3457 SM_I(sbi)->free_info = free_i;
3458
7cd8558b 3459 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3460 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3461 if (!free_i->free_segmap)
3462 return -ENOMEM;
3463
7cd8558b 3464 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3465 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3466 if (!free_i->free_secmap)
3467 return -ENOMEM;
3468
3469 /* set all segments as dirty temporarily */
3470 memset(free_i->free_segmap, 0xff, bitmap_size);
3471 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3472
3473 /* init free segmap information */
7cd8558b 3474 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3475 free_i->free_segments = 0;
3476 free_i->free_sections = 0;
1a118ccf 3477 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3478 return 0;
3479}
3480
3481static int build_curseg(struct f2fs_sb_info *sbi)
3482{
1042d60f 3483 struct curseg_info *array;
351df4b2
JK		 3484 int i;
3485
b434babf 3486 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3487 if (!array)
3488 return -ENOMEM;
3489
3490 SM_I(sbi)->curseg_array = array;
3491
3492 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3493 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3494 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3495 if (!array[i].sum_blk)
3496 return -ENOMEM;
b7ad7512
CY		 3497 init_rwsem(&array[i].journal_rwsem);
3498 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3499 GFP_KERNEL);
3500 if (!array[i].journal)
3501 return -ENOMEM;
351df4b2
JK		 3502 array[i].segno = NULL_SEGNO;
3503 array[i].next_blkoff = 0;
3504 }
3505 return restore_curseg_summaries(sbi);
3506}
3507
5c09c3e3 3508static int build_sit_entries(struct f2fs_sb_info *sbi)
351df4b2
JK		 3509{
3510 struct sit_info *sit_i = SIT_I(sbi);
3511 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3512 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3513 struct seg_entry *se;
3514 struct f2fs_sit_entry sit;
74de593a
CY		 3515 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3516 unsigned int i, start, end;
3517 unsigned int readed, start_blk = 0;
5c09c3e3 3518 int err = 0;
c523964d 3519 block_t total_node_blocks = 0;
351df4b2 3520
74de593a 3521 do {
664ba972
JK		 3522 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3523 META_SIT, true);
74de593a
CY		 3524
3525 start = start_blk * sit_i->sents_per_block;
3526 end = (start_blk + readed) * sit_i->sents_per_block;
3527
7cd8558b 3528 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3529 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3530 struct page *page;
3531
9c094040 3532 se = &sit_i->sentries[start];
74de593a
CY		 3533 page = get_current_sit_page(sbi, start);
3534 sit_blk = (struct f2fs_sit_block *)page_address(page);
3535 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3536 f2fs_put_page(page, 1);
d600af23 3537
5c09c3e3
JK		 3538 err = check_block_count(sbi, start, &sit);
3539 if (err)
3540 return err;
74de593a 3541 seg_info_from_raw_sit(se, &sit);
c523964d
JK		 3542 if (IS_NODESEG(se->type))
3543 total_node_blocks += se->valid_blocks;
a66cdd98
JK		 3544
3545 /* build discard map only one time */
3e025740 3546 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3547 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3548 memset(se->discard_map, 0xff,
3549 SIT_VBLOCK_MAP_SIZE);
3550 } else {
3551 memcpy(se->discard_map,
3552 se->cur_valid_map,
3553 SIT_VBLOCK_MAP_SIZE);
3554 sbi->discard_blks +=
3555 sbi->blocks_per_seg -
3556 se->valid_blocks;
3557 }
3e025740 3558 }
a66cdd98 3559
d600af23
CY		 3560 if (sbi->segs_per_sec > 1)
3561 get_sec_entry(sbi, start)->valid_blocks +=
3562 se->valid_blocks;
351df4b2 3563 }
74de593a
CY		 3564 start_blk += readed;
3565 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3566
3567 down_read(&curseg->journal_rwsem);
3568 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3569 unsigned int old_valid_blocks;
3570
3571 start = le32_to_cpu(segno_in_journal(journal, i));
3a80c3c3
JK		 3572 if (start >= MAIN_SEGS(sbi)) {
3573 f2fs_msg(sbi->sb, KERN_ERR,
3574 "Wrong journal entry on segno %u",
3575 start);
3576 set_sbi_flag(sbi, SBI_NEED_FSCK);
3577 err = -EINVAL;
3578 break;
3579 }
3580
d600af23
CY		 3581 se = &sit_i->sentries[start];
3582 sit = sit_in_journal(journal, i);
3583
3584 old_valid_blocks = se->valid_blocks;
c523964d
JK		 3585 if (IS_NODESEG(se->type))
3586 total_node_blocks -= old_valid_blocks;
d600af23 3587
5c09c3e3
JK		 3588 err = check_block_count(sbi, start, &sit);
3589 if (err)
3590 break;
d600af23 3591 seg_info_from_raw_sit(se, &sit);
c523964d
JK		 3592 if (IS_NODESEG(se->type))
3593 total_node_blocks += se->valid_blocks;
d600af23
CY		 3594
3595 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3596 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3597 memset(se->discard_map, 0xff,
3598 SIT_VBLOCK_MAP_SIZE);
3599 } else {
3600 memcpy(se->discard_map, se->cur_valid_map,
3601 SIT_VBLOCK_MAP_SIZE);
3602 sbi->discard_blks += old_valid_blocks -
3603 se->valid_blocks;
3604 }
d600af23
CY		 3605 }
3606
3607 if (sbi->segs_per_sec > 1)
3608 get_sec_entry(sbi, start)->valid_blocks +=
3609 se->valid_blocks - old_valid_blocks;
3610 }
3611 up_read(&curseg->journal_rwsem);
c523964d
JK		 3612
3613 if (!err && total_node_blocks != valid_node_count(sbi)) {
3614 f2fs_msg(sbi->sb, KERN_ERR,
3615 "SIT is corrupted node# %u vs %u",
3616 total_node_blocks, valid_node_count(sbi));
3617 set_sbi_flag(sbi, SBI_NEED_FSCK);
3618 err = -EINVAL;
3619 }
3620
5c09c3e3 3621 return err;
351df4b2
JK		 3622}
3623
3624static void init_free_segmap(struct f2fs_sb_info *sbi)
3625{
3626 unsigned int start;
3627 int type;
3628
7cd8558b 3629 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3630 struct seg_entry *sentry = get_seg_entry(sbi, start);
3631 if (!sentry->valid_blocks)
3632 __set_free(sbi, start);
c79b7ff1
JK		 3633 else
3634 SIT_I(sbi)->written_valid_blocks +=
3635 sentry->valid_blocks;
351df4b2
JK		 3636 }
3637
3638 /* set use the current segments */
3639 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3640 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3641 __set_test_and_inuse(sbi, curseg_t->segno);
3642 }
3643}
3644
3645static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3646{
3647 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3648 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3649 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3650 unsigned short valid_blocks;
3651
8736fbf0 3652 while (1) {
351df4b2 3653 /* find dirty segment based on free segmap */
7cd8558b
JK		 3654 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3655 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3656 break;
3657 offset = segno + 1;
302bd348 3658 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3659 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3660 continue;
ec325b52
JK		 3661 if (valid_blocks > sbi->blocks_per_seg) {
3662 f2fs_bug_on(sbi, 1);
3663 continue;
3664 }
351df4b2
JK		 3665 mutex_lock(&dirty_i->seglist_lock);
3666 __locate_dirty_segment(sbi, segno, DIRTY);
3667 mutex_unlock(&dirty_i->seglist_lock);
3668 }
3669}
3670
5ec4e49f 3671static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3672{
3673 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3674 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3675
a7c3e901 3676 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3677 if (!dirty_i->victim_secmap)
351df4b2
JK		 3678 return -ENOMEM;
3679 return 0;
3680}
3681
3682static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3683{
3684 struct dirty_seglist_info *dirty_i;
3685 unsigned int bitmap_size, i;
3686
3687 /* allocate memory for dirty segments list information */
3688 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3689 if (!dirty_i)
3690 return -ENOMEM;
3691
3692 SM_I(sbi)->dirty_info = dirty_i;
3693 mutex_init(&dirty_i->seglist_lock);
3694
7cd8558b 3695 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3696
3697 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3698 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3699 if (!dirty_i->dirty_segmap[i])
3700 return -ENOMEM;
3701 }
3702
3703 init_dirty_segmap(sbi);
5ec4e49f 3704 return init_victim_secmap(sbi);
351df4b2
JK		 3705}
3706
0a8165d7 3707/*
351df4b2
JK		 3708 * Update min, max modified time for cost-benefit GC algorithm
3709 */
3710static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3711{
3712 struct sit_info *sit_i = SIT_I(sbi);
3713 unsigned int segno;
3714
3d26fa6b 3715 down_write(&sit_i->sentry_lock);
351df4b2
JK		 3716
3717 sit_i->min_mtime = LLONG_MAX;
3718
7cd8558b 3719 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3720 unsigned int i;
3721 unsigned long long mtime = 0;
3722
3723 for (i = 0; i < sbi->segs_per_sec; i++)
3724 mtime += get_seg_entry(sbi, segno + i)->mtime;
3725
3726 mtime = div_u64(mtime, sbi->segs_per_sec);
3727
3728 if (sit_i->min_mtime > mtime)
3729 sit_i->min_mtime = mtime;
3730 }
3731 sit_i->max_mtime = get_mtime(sbi);
3d26fa6b 3732 up_write(&sit_i->sentry_lock);
351df4b2
JK		 3733}
3734
3735int build_segment_manager(struct f2fs_sb_info *sbi)
3736{
3737 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3738 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3739 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3740 int err;
3741
3742 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3743 if (!sm_info)
3744 return -ENOMEM;
3745
3746 /* init sm info */
3747 sbi->sm_info = sm_info;
351df4b2
JK		 3748 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3749 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3750 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3751 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3752 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3753 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3754 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3755 sm_info->rec_prefree_segments = sm_info->main_segments *
3756 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3757 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3758 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3759
52763a4b
JK		 3760 if (!test_opt(sbi, LFS))
3761 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3762 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3763 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3764 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
a2a12b67 3765 sm_info->min_ssr_sections = reserved_sections(sbi);
351df4b2 3766
bba681cb
JK		 3767 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3768
184a5cd2
CY		 3769 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3770
2b60311d
CY		 3771 init_rwsem(&sm_info->curseg_lock);
3772
d4fdf8ba 3773 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3774 err = create_flush_cmd_control(sbi);
3775 if (err)
a688b9d9 3776 return err;
6b4afdd7
JK		 3777 }
3778
0b54fb84
JK		 3779 err = create_discard_cmd_control(sbi);
3780 if (err)
3781 return err;
3782
351df4b2
JK		 3783 err = build_sit_info(sbi);
3784 if (err)
3785 return err;
3786 err = build_free_segmap(sbi);
3787 if (err)
3788 return err;
3789 err = build_curseg(sbi);
3790 if (err)
3791 return err;
3792
3793 /* reinit free segmap based on SIT */
5c09c3e3
JK		 3794 err = build_sit_entries(sbi);
3795 if (err)
3796 return err;
351df4b2
JK		 3797
3798 init_free_segmap(sbi);
3799 err = build_dirty_segmap(sbi);
3800 if (err)
3801 return err;
3802
3803 init_min_max_mtime(sbi);
3804 return 0;
3805}
3806
3807static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3808 enum dirty_type dirty_type)
3809{
3810 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3811
3812 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3813 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3814 dirty_i->nr_dirty[dirty_type] = 0;
3815 mutex_unlock(&dirty_i->seglist_lock);
3816}
3817
5ec4e49f 3818static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3819{
3820 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3821 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3822}
3823
3824static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3825{
3826 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3827 int i;
3828
3829 if (!dirty_i)
3830 return;
3831
3832 /* discard pre-free/dirty segments list */
3833 for (i = 0; i < NR_DIRTY_TYPE; i++)
3834 discard_dirty_segmap(sbi, i);
3835
5ec4e49f 3836 destroy_victim_secmap(sbi);
351df4b2
JK		 3837 SM_I(sbi)->dirty_info = NULL;
3838 kfree(dirty_i);
3839}
3840
3841static void destroy_curseg(struct f2fs_sb_info *sbi)
3842{
3843 struct curseg_info *array = SM_I(sbi)->curseg_array;
3844 int i;
3845
3846 if (!array)
3847 return;
3848 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3849 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3850 kfree(array[i].sum_blk);
b7ad7512
CY		 3851 kfree(array[i].journal);
3852 }
351df4b2
JK		 3853 kfree(array);
3854}
3855
3856static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3857{
3858 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3859 if (!free_i)
3860 return;
3861 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3862 kvfree(free_i->free_segmap);
3863 kvfree(free_i->free_secmap);
351df4b2
JK		 3864 kfree(free_i);
3865}
3866
3867static void destroy_sit_info(struct f2fs_sb_info *sbi)
3868{
3869 struct sit_info *sit_i = SIT_I(sbi);
3870 unsigned int start;
3871
3872 if (!sit_i)
3873 return;
3874
3875 if (sit_i->sentries) {
7cd8558b 3876 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3877 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3878#ifdef CONFIG_F2FS_CHECK_FS
3879 kfree(sit_i->sentries[start].cur_valid_map_mir);
3880#endif
351df4b2 3881 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3882 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3883 }
3884 }
60a3b782
JK		 3885 kfree(sit_i->tmp_map);
3886
39307a8e
JK		 3887 kvfree(sit_i->sentries);
3888 kvfree(sit_i->sec_entries);
3889 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3890
3891 SM_I(sbi)->sit_info = NULL;
3892 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3893#ifdef CONFIG_F2FS_CHECK_FS
3894 kfree(sit_i->sit_bitmap_mir);
3895#endif
351df4b2
JK		 3896 kfree(sit_i);
3897}
3898
3899void destroy_segment_manager(struct f2fs_sb_info *sbi)
3900{
3901 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3902
3b03f724
CY		 3903 if (!sm_info)
3904 return;
5eba8c5d 3905 destroy_flush_cmd_control(sbi, true);
f099405f 3906 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3907 destroy_dirty_segmap(sbi);
3908 destroy_curseg(sbi);
3909 destroy_free_segmap(sbi);
3910 destroy_sit_info(sbi);
3911 sbi->sm_info = NULL;
3912 kfree(sm_info);
3913}
7fd9e544
JK		 3914
3915int __init create_segment_manager_caches(void)
3916{
3917 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3918 sizeof(struct discard_entry));
7fd9e544 3919 if (!discard_entry_slab)
184a5cd2
CY		 3920 goto fail;
3921
b01a9201
JK		 3922 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3923 sizeof(struct discard_cmd));
3924 if (!discard_cmd_slab)
6ab2a308 3925 goto destroy_discard_entry;
275b66b0 3926
184a5cd2 3927 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3928 sizeof(struct sit_entry_set));
184a5cd2 3929 if (!sit_entry_set_slab)
b01a9201 3930 goto destroy_discard_cmd;
88b88a66
JK		 3931
3932 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3933 sizeof(struct inmem_pages));
3934 if (!inmem_entry_slab)
3935 goto destroy_sit_entry_set;
7fd9e544 3936 return 0;
184a5cd2 3937
88b88a66
JK		 3938destroy_sit_entry_set:
3939 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3940destroy_discard_cmd:
3941 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3942destroy_discard_entry:
184a5cd2
CY		 3943 kmem_cache_destroy(discard_entry_slab);
3944fail:
3945 return -ENOMEM;
7fd9e544
JK		 3946}
3947
3948void destroy_segment_manager_caches(void)
3949{
184a5cd2 3950 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3951 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3952 kmem_cache_destroy(discard_entry_slab);
88b88a66 3953 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3954}
ubuntu-bionic-kernel mirror