]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: support issuing/waiting discard in range
[mirror_ubuntu-jammy-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
1d7be270 19#include <linux/freezer.h>
1eb1ef4a 20#include <linux/sched/signal.h>
351df4b2
JK		 21
22#include "f2fs.h"
23#include "segment.h"
24#include "node.h"
5f656541 25#include "gc.h"
9e4ded3f 26#include "trace.h"
6ec178da 27#include <trace/events/f2fs.h>
351df4b2 28
9a7f143a
CL		 29#define __reverse_ffz(x) __reverse_ffs(~(x))
30
7fd9e544 31static struct kmem_cache *discard_entry_slab;
b01a9201 32static struct kmem_cache *discard_cmd_slab;
184a5cd2 33static struct kmem_cache *sit_entry_set_slab;
88b88a66 34static struct kmem_cache *inmem_entry_slab;
7fd9e544 35
f96999c3
JK		 36static unsigned long __reverse_ulong(unsigned char *str)
37{
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41#if BITS_PER_LONG == 64
42 shift = 56;
43#endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49}
50
9a7f143a
CL		 51/*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55static inline unsigned long __reverse_ffs(unsigned long word)
56{
57 int num = 0;
58
59#if BITS_PER_LONG == 64
f96999c3 60 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 61 num += 32;
f96999c3 62 else
9a7f143a 63 word >>= 32;
9a7f143a 64#endif
f96999c3 65 if ((word & 0xffff0000) == 0)
9a7f143a 66 num += 16;
f96999c3 67 else
9a7f143a 68 word >>= 16;
f96999c3
JK		 69
70 if ((word & 0xff00) == 0)
9a7f143a 71 num += 8;
f96999c3 72 else
9a7f143a 73 word >>= 8;
f96999c3 74
9a7f143a
CL		 75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
f96999c3 79
9a7f143a
CL		 80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
f96999c3 84
9a7f143a
CL		 85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88}
89
90/*
e1c42045 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 93 * @size must be integral times of unsigned long.
9a7f143a 94 * Example:
f96999c3
JK		 95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 98 */
99static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101{
102 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 103 unsigned long result = size;
9a7f143a 104 unsigned long tmp;
9a7f143a
CL		 105
106 if (offset >= size)
107 return size;
108
692223d1 109 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 110 offset %= BITS_PER_LONG;
f96999c3 111
692223d1
FL		 112 while (1) {
113 if (*p == 0)
114 goto pass;
9a7f143a 115
f96999c3 116 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 121 if (tmp)
692223d1
FL		 122 goto found;
123pass:
124 if (size <= BITS_PER_LONG)
125 break;
9a7f143a 126 size -= BITS_PER_LONG;
692223d1 127 offset = 0;
f96999c3 128 p++;
9a7f143a 129 }
692223d1
FL		 130 return result;
131found:
132 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
138 const unsigned long *p = addr + BIT_WORD(offset);
80609448 139 unsigned long result = size;
9a7f143a 140 unsigned long tmp;
9a7f143a
CL		 141
142 if (offset >= size)
143 return size;
144
80609448 145 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 146 offset %= BITS_PER_LONG;
80609448
JK		 147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
f96999c3 152 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
f96999c3 158 if (tmp != ~0UL)
80609448
JK		 159 goto found;
160pass:
161 if (size <= BITS_PER_LONG)
162 break;
9a7f143a 163 size -= BITS_PER_LONG;
80609448 164 offset = 0;
f96999c3 165 p++;
9a7f143a 166 }
80609448
JK		 167 return result;
168found:
169 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 170}
171
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 +
184 2 * reserved_sections(sbi));
185}
186
88b88a66
JK		 187void register_inmem_page(struct inode *inode, struct page *page)
188{
189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
9be32d72 191
9e4ded3f 192 f2fs_trace_pid(page);
0722b101 193
decd36b6
CY		 194 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
195 SetPagePrivate(page);
196
88b88a66
JK		 197 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
198
199 /* add atomic page indices to the list */
200 new->page = page;
201 INIT_LIST_HEAD(&new->list);
decd36b6 202
88b88a66
JK		 203 /* increase reference count with clean state */
204 mutex_lock(&fi->inmem_lock);
205 get_page(page);
206 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 207 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 208 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 209
210 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 211}
212
28bc106b
CY		 213static int __revoke_inmem_pages(struct inode *inode,
214 struct list_head *head, bool drop, bool recover)
29b96b54 215{
28bc106b 216 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 217 struct inmem_pages *cur, *tmp;
28bc106b 218 int err = 0;
29b96b54
CY		 219
220 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 221 struct page *page = cur->page;
222
223 if (drop)
224 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
225
226 lock_page(page);
29b96b54 227
28bc106b
CY		 228 if (recover) {
229 struct dnode_of_data dn;
230 struct node_info ni;
231
232 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
7f2b4e8e 233retry:
28bc106b 234 set_new_dnode(&dn, inode, NULL, NULL, 0);
7f2b4e8e
CY		 235 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
236 if (err) {
237 if (err == -ENOMEM) {
238 congestion_wait(BLK_RW_ASYNC, HZ/50);
239 cond_resched();
240 goto retry;
241 }
28bc106b
CY		 242 err = -EAGAIN;
243 goto next;
244 }
245 get_node_info(sbi, dn.nid, &ni);
246 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
247 cur->old_addr, ni.version, true, true);
248 f2fs_put_dnode(&dn);
249 }
250next:
63c52d78
JK		 251 /* we don't need to invalidate this in the sccessful status */
252 if (drop || recover)
253 ClearPageUptodate(page);
28bc106b 254 set_page_private(page, 0);
c81ced05 255 ClearPagePrivate(page);
28bc106b 256 f2fs_put_page(page, 1);
29b96b54
CY		 257
258 list_del(&cur->list);
259 kmem_cache_free(inmem_entry_slab, cur);
260 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
261 }
28bc106b 262 return err;
29b96b54
CY		 263}
264
265void drop_inmem_pages(struct inode *inode)
266{
267 struct f2fs_inode_info *fi = F2FS_I(inode);
268
269 mutex_lock(&fi->inmem_lock);
28bc106b 270 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 271 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 272
273 clear_inode_flag(inode, FI_ATOMIC_FILE);
84a23fbe 274 clear_inode_flag(inode, FI_HOT_DATA);
5fe45743 275 stat_dec_atomic_write(inode);
29b96b54
CY		 276}
277
8c242db9
JK		 278void drop_inmem_page(struct inode *inode, struct page *page)
279{
280 struct f2fs_inode_info *fi = F2FS_I(inode);
281 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
282 struct list_head *head = &fi->inmem_pages;
283 struct inmem_pages *cur = NULL;
284
285 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
286
287 mutex_lock(&fi->inmem_lock);
288 list_for_each_entry(cur, head, list) {
289 if (cur->page == page)
290 break;
291 }
292
293 f2fs_bug_on(sbi, !cur || cur->page != page);
294 list_del(&cur->list);
295 mutex_unlock(&fi->inmem_lock);
296
297 dec_page_count(sbi, F2FS_INMEM_PAGES);
298 kmem_cache_free(inmem_entry_slab, cur);
299
300 ClearPageUptodate(page);
301 set_page_private(page, 0);
302 ClearPagePrivate(page);
303 f2fs_put_page(page, 0);
304
305 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
306}
307
28bc106b
CY		 308static int __commit_inmem_pages(struct inode *inode,
309 struct list_head *revoke_list)
88b88a66
JK		 310{
311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
312 struct f2fs_inode_info *fi = F2FS_I(inode);
313 struct inmem_pages *cur, *tmp;
88b88a66 314 struct f2fs_io_info fio = {
05ca3632 315 .sbi = sbi,
39d787be 316 .ino = inode->i_ino,
88b88a66 317 .type = DATA,
04d328de 318 .op = REQ_OP_WRITE,
70fd7614 319 .op_flags = REQ_SYNC | REQ_PRIO,
b0af6d49 320 .io_type = FS_DATA_IO,
88b88a66 321 };
942fd319 322 pgoff_t last_idx = ULONG_MAX;
edb27dee 323 int err = 0;
88b88a66 324
88b88a66 325 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 326 struct page *page = cur->page;
327
328 lock_page(page);
329 if (page->mapping == inode->i_mapping) {
330 trace_f2fs_commit_inmem_page(page, INMEM);
331
332 set_page_dirty(page);
333 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 334 if (clear_page_dirty_for_io(page)) {
29b96b54 335 inode_dec_dirty_pages(inode);
933439c8
CY		 336 remove_dirty_inode(inode);
337 }
640cc189 338retry:
28bc106b 339 fio.page = page;
e959c8f5 340 fio.old_blkaddr = NULL_ADDR;
4d978078 341 fio.encrypted_page = NULL;
cc15620b 342 fio.need_lock = LOCK_DONE;
29b96b54
CY		 343 err = do_write_data_page(&fio);
344 if (err) {
640cc189
JK		 345 if (err == -ENOMEM) {
346 congestion_wait(BLK_RW_ASYNC, HZ/50);
347 cond_resched();
348 goto retry;
349 }
28bc106b 350 unlock_page(page);
29b96b54 351 break;
70c640b1 352 }
28bc106b
CY		 353 /* record old blkaddr for revoking */
354 cur->old_addr = fio.old_blkaddr;
942fd319 355 last_idx = page->index;
28bc106b
CY		 356 }
357 unlock_page(page);
358 list_move_tail(&cur->list, revoke_list);
88b88a66 359 }
29b96b54 360
942fd319 361 if (last_idx != ULONG_MAX)
b9109b0e 362 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
28bc106b
CY		 363
364 if (!err)
365 __revoke_inmem_pages(inode, revoke_list, false, false);
366
29b96b54
CY		 367 return err;
368}
369
370int commit_inmem_pages(struct inode *inode)
371{
372 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
373 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 374 struct list_head revoke_list;
375 int err;
29b96b54 376
28bc106b 377 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 378 f2fs_balance_fs(sbi, true);
379 f2fs_lock_op(sbi);
380
5fe45743
CY		 381 set_inode_flag(inode, FI_ATOMIC_COMMIT);
382
29b96b54 383 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 384 err = __commit_inmem_pages(inode, &revoke_list);
385 if (err) {
386 int ret;
387 /*
388 * try to revoke all committed pages, but still we could fail
389 * due to no memory or other reason, if that happened, EAGAIN
390 * will be returned, which means in such case, transaction is
391 * already not integrity, caller should use journal to do the
392 * recovery or rewrite & commit last transaction. For other
393 * error number, revoking was done by filesystem itself.
394 */
395 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
396 if (ret)
397 err = ret;
398
399 /* drop all uncommitted pages */
400 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
401 }
88b88a66
JK		 402 mutex_unlock(&fi->inmem_lock);
403
5fe45743
CY		 404 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
405
29b96b54 406 f2fs_unlock_op(sbi);
edb27dee 407 return err;
88b88a66
JK		 408}
409
0a8165d7 410/*
351df4b2
JK		 411 * This function balances dirty node and dentry pages.
412 * In addition, it controls garbage collection.
413 */
2c4db1a6 414void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 415{
0f348028 416#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 417 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
418 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 419 f2fs_stop_checkpoint(sbi, false);
55523519 420 }
0f348028
CY		 421#endif
422
e589c2c4 423 /* balance_fs_bg is able to be pending */
a7881893 424 if (need && excess_cached_nats(sbi))
e589c2c4
JK		 425 f2fs_balance_fs_bg(sbi);
426
351df4b2 427 /*
029cd28c
JK		 428 * We should do GC or end up with checkpoint, if there are so many dirty
429 * dir/node pages without enough free segments.
351df4b2 430 */
7f3037a5 431 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 432 mutex_lock(&sbi->gc_mutex);
e066b83c 433 f2fs_gc(sbi, false, false, NULL_SEGNO);
351df4b2
JK		 434 }
435}
436
4660f9c0
JK		 437void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
438{
1dcc336b 439 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 440 if (!available_free_memory(sbi, EXTENT_CACHE))
441 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 442
1b38dc8e
JK		 443 /* check the # of cached NAT entries */
444 if (!available_free_memory(sbi, NAT_ENTRIES))
445 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
446
31696580 447 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 448 try_to_free_nids(sbi, MAX_FREE_NIDS);
449 else
22ad0b6a 450 build_free_nids(sbi, false, false);
31696580 451
1c0f4bf5 452 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
f455c8a5 453 return;
31696580 454
1b38dc8e
JK		 455 /* checkpoint is the only way to shrink partial cached entries */
456 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 457 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 458 excess_prefree_segs(sbi) ||
459 excess_dirty_nats(sbi) ||
f455c8a5 460 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 461 if (test_opt(sbi, DATA_FLUSH)) {
462 struct blk_plug plug;
463
464 blk_start_plug(&plug);
36b35a0d 465 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 466 blk_finish_plug(&plug);
467 }
4660f9c0 468 f2fs_sync_fs(sbi->sb, true);
42190d2a 469 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 470 }
4660f9c0
JK		 471}
472
20fda56b
KM		 473static int __submit_flush_wait(struct f2fs_sb_info *sbi,
474 struct block_device *bdev)
3c62be17
JK		 475{
476 struct bio *bio = f2fs_bio_alloc(0);
477 int ret;
478
3adc5fcb 479 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
74d46992 480 bio_set_dev(bio, bdev);
3c62be17
JK		 481 ret = submit_bio_wait(bio);
482 bio_put(bio);
20fda56b
KM		 483
484 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
485 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 486 return ret;
487}
488
39d787be 489static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
3c62be17 490{
39d787be 491 int ret = 0;
3c62be17
JK		 492 int i;
493
39d787be
CY		 494 if (!sbi->s_ndevs)
495 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
20fda56b 496
39d787be
CY		 497 for (i = 0; i < sbi->s_ndevs; i++) {
498 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
499 continue;
20fda56b
KM		 500 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
501 if (ret)
502 break;
3c62be17
JK		 503 }
504 return ret;
505}
506
2163d198 507static int issue_flush_thread(void *data)
6b4afdd7
JK		 508{
509 struct f2fs_sb_info *sbi = data;
b01a9201 510 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 511 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 512repeat:
513 if (kthread_should_stop())
514 return 0;
515
dc6febb6
CY		 516 sb_start_intwrite(sbi->sb);
517
721bd4d5 518 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 519 struct flush_cmd *cmd, *next;
520 int ret;
521
721bd4d5
GZ		 522 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
523 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
524
39d787be
CY		 525 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
526
527 ret = submit_flush_wait(sbi, cmd->ino);
8b8dd65f
CY		 528 atomic_inc(&fcc->issued_flush);
529
721bd4d5
GZ		 530 llist_for_each_entry_safe(cmd, next,
531 fcc->dispatch_list, llnode) {
6b4afdd7 532 cmd->ret = ret;
6b4afdd7
JK		 533 complete(&cmd->wait);
534 }
a688b9d9 535 fcc->dispatch_list = NULL;
6b4afdd7
JK		 536 }
537
dc6febb6
CY		 538 sb_end_intwrite(sbi->sb);
539
a688b9d9 540 wait_event_interruptible(*q,
721bd4d5 541 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 542 goto repeat;
543}
544
39d787be 545int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
6b4afdd7 546{
b01a9201 547 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 548 struct flush_cmd cmd;
8b8dd65f 549 int ret;
6b4afdd7 550
0f7b2abd
JK		 551 if (test_opt(sbi, NOBARRIER))
552 return 0;
553
8b8dd65f 554 if (!test_opt(sbi, FLUSH_MERGE)) {
39d787be 555 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 556 atomic_inc(&fcc->issued_flush);
557 return ret;
558 }
740432f8 559
39d787be
CY		 560 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
561 ret = submit_flush_wait(sbi, ino);
8b8dd65f
CY		 562 atomic_dec(&fcc->issing_flush);
563
564 atomic_inc(&fcc->issued_flush);
740432f8
JK		 565 return ret;
566 }
6b4afdd7 567
39d787be 568 cmd.ino = ino;
adf8d90b 569 init_completion(&cmd.wait);
6b4afdd7 570
721bd4d5 571 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 572
6f890df0
CY		 573 /* update issue_list before we wake up issue_flush thread */
574 smp_mb();
575
576 if (waitqueue_active(&fcc->flush_wait_queue))
a688b9d9 577 wake_up(&fcc->flush_wait_queue);
6b4afdd7 578
5eba8c5d
JK		 579 if (fcc->f2fs_issue_flush) {
580 wait_for_completion(&cmd.wait);
8b8dd65f 581 atomic_dec(&fcc->issing_flush);
5eba8c5d 582 } else {
d3238691
CY		 583 struct llist_node *list;
584
585 list = llist_del_all(&fcc->issue_list);
586 if (!list) {
587 wait_for_completion(&cmd.wait);
588 atomic_dec(&fcc->issing_flush);
589 } else {
590 struct flush_cmd *tmp, *next;
591
39d787be 592 ret = submit_flush_wait(sbi, ino);
d3238691
CY		 593
594 llist_for_each_entry_safe(tmp, next, list, llnode) {
595 if (tmp == &cmd) {
596 cmd.ret = ret;
597 atomic_dec(&fcc->issing_flush);
598 continue;
599 }
600 tmp->ret = ret;
601 complete(&tmp->wait);
602 }
603 }
5eba8c5d 604 }
adf8d90b
CY		 605
606 return cmd.ret;
6b4afdd7
JK		 607}
608
2163d198
GZ		 609int create_flush_cmd_control(struct f2fs_sb_info *sbi)
610{
611 dev_t dev = sbi->sb->s_bdev->bd_dev;
612 struct flush_cmd_control *fcc;
613 int err = 0;
614
b01a9201
JK		 615 if (SM_I(sbi)->fcc_info) {
616 fcc = SM_I(sbi)->fcc_info;
d871cd04
YS		 617 if (fcc->f2fs_issue_flush)
618 return err;
5eba8c5d
JK		 619 goto init_thread;
620 }
621
2163d198
GZ		 622 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
623 if (!fcc)
624 return -ENOMEM;
8b8dd65f
CY		 625 atomic_set(&fcc->issued_flush, 0);
626 atomic_set(&fcc->issing_flush, 0);
2163d198 627 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 628 init_llist_head(&fcc->issue_list);
b01a9201 629 SM_I(sbi)->fcc_info = fcc;
d4fdf8ba
YH		 630 if (!test_opt(sbi, FLUSH_MERGE))
631 return err;
632
5eba8c5d 633init_thread:
2163d198
GZ		 634 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
635 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
636 if (IS_ERR(fcc->f2fs_issue_flush)) {
637 err = PTR_ERR(fcc->f2fs_issue_flush);
638 kfree(fcc);
b01a9201 639 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 640 return err;
641 }
2163d198
GZ		 642
643 return err;
644}
645
5eba8c5d 646void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 647{
b01a9201 648 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 649
5eba8c5d
JK		 650 if (fcc && fcc->f2fs_issue_flush) {
651 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
652
653 fcc->f2fs_issue_flush = NULL;
654 kthread_stop(flush_thread);
655 }
656 if (free) {
657 kfree(fcc);
b01a9201 658 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 659 }
2163d198
GZ		 660}
661
1228b482
CY		 662int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
663{
664 int ret = 0, i;
665
666 if (!sbi->s_ndevs)
667 return 0;
668
669 for (i = 1; i < sbi->s_ndevs; i++) {
670 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
671 continue;
672 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
673 if (ret)
674 break;
675
676 spin_lock(&sbi->dev_lock);
677 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
678 spin_unlock(&sbi->dev_lock);
679 }
680
681 return ret;
682}
683
351df4b2
JK		 684static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
685 enum dirty_type dirty_type)
686{
687 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
688
689 /* need not be added */
690 if (IS_CURSEG(sbi, segno))
691 return;
692
693 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
694 dirty_i->nr_dirty[dirty_type]++;
695
696 if (dirty_type == DIRTY) {
697 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 698 enum dirty_type t = sentry->type;
b2f2c390 699
ec325b52
JK		 700 if (unlikely(t >= DIRTY)) {
701 f2fs_bug_on(sbi, 1);
702 return;
703 }
4625d6aa
CL		 704 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
705 dirty_i->nr_dirty[t]++;
351df4b2
JK		 706 }
707}
708
709static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
710 enum dirty_type dirty_type)
711{
712 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
713
714 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
715 dirty_i->nr_dirty[dirty_type]--;
716
717 if (dirty_type == DIRTY) {
4625d6aa
CL		 718 struct seg_entry *sentry = get_seg_entry(sbi, segno);
719 enum dirty_type t = sentry->type;
720
721 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
722 dirty_i->nr_dirty[t]--;
b2f2c390 723
302bd348 724 if (get_valid_blocks(sbi, segno, true) == 0)
4ddb1a4d 725 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
5ec4e49f 726 dirty_i->victim_secmap);
351df4b2
JK		 727 }
728}
729
0a8165d7 730/*
351df4b2
JK		 731 * Should not occur error such as -ENOMEM.
732 * Adding dirty entry into seglist is not critical operation.
733 * If a given segment is one of current working segments, it won't be added.
734 */
8d8451af 735static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 736{
737 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
738 unsigned short valid_blocks;
739
740 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
741 return;
742
743 mutex_lock(&dirty_i->seglist_lock);
744
302bd348 745 valid_blocks = get_valid_blocks(sbi, segno, false);
351df4b2
JK		 746
747 if (valid_blocks == 0) {
748 __locate_dirty_segment(sbi, segno, PRE);
749 __remove_dirty_segment(sbi, segno, DIRTY);
750 } else if (valid_blocks < sbi->blocks_per_seg) {
751 __locate_dirty_segment(sbi, segno, DIRTY);
752 } else {
753 /* Recovery routine with SSR needs this */
754 __remove_dirty_segment(sbi, segno, DIRTY);
755 }
756
757 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 758}
759
004b6862 760static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 761 struct block_device *bdev, block_t lstart,
762 block_t start, block_t len)
275b66b0 763{
0b54fb84 764 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
ba48a33e 765 struct list_head *pend_list;
b01a9201 766 struct discard_cmd *dc;
275b66b0 767
ba48a33e
CY		 768 f2fs_bug_on(sbi, !len);
769
770 pend_list = &dcc->pend_list[plist_idx(len)];
771
b01a9201
JK		 772 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
773 INIT_LIST_HEAD(&dc->list);
c81abe34 774 dc->bdev = bdev;
b01a9201 775 dc->lstart = lstart;
c81abe34 776 dc->start = start;
b01a9201 777 dc->len = len;
ec9895ad 778 dc->ref = 0;
15469963 779 dc->state = D_PREP;
c81abe34 780 dc->error = 0;
b01a9201 781 init_completion(&dc->wait);
22d375dd 782 list_add_tail(&dc->list, pend_list);
5f32366a 783 atomic_inc(&dcc->discard_cmd_cnt);
d84d1cbd 784 dcc->undiscard_blks += len;
004b6862
CY		 785
786 return dc;
787}
788
789static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
790 struct block_device *bdev, block_t lstart,
791 block_t start, block_t len,
792 struct rb_node *parent, struct rb_node **p)
793{
794 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
795 struct discard_cmd *dc;
796
797 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
798
799 rb_link_node(&dc->rb_node, parent, p);
800 rb_insert_color(&dc->rb_node, &dcc->root);
801
802 return dc;
15469963
JK		 803}
804
004b6862
CY		 805static void __detach_discard_cmd(struct discard_cmd_control *dcc,
806 struct discard_cmd *dc)
15469963 807{
dcc9165d 808 if (dc->state == D_DONE)
004b6862
CY		 809 atomic_dec(&dcc->issing_discard);
810
811 list_del(&dc->list);
812 rb_erase(&dc->rb_node, &dcc->root);
d84d1cbd 813 dcc->undiscard_blks -= dc->len;
004b6862
CY		 814
815 kmem_cache_free(discard_cmd_slab, dc);
816
817 atomic_dec(&dcc->discard_cmd_cnt);
818}
819
820static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
821 struct discard_cmd *dc)
822{
823 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
dcc9165d 824
d9703d90
CY		 825 f2fs_bug_on(sbi, dc->ref);
826
c81abe34
JK		 827 if (dc->error == -EOPNOTSUPP)
828 dc->error = 0;
15469963 829
c81abe34 830 if (dc->error)
15469963 831 f2fs_msg(sbi->sb, KERN_INFO,
04dfc230
CY		 832 "Issue discard(%u, %u, %u) failed, ret: %d",
833 dc->lstart, dc->start, dc->len, dc->error);
004b6862 834 __detach_discard_cmd(dcc, dc);
275b66b0
CY		 835}
836
c81abe34
JK		 837static void f2fs_submit_discard_endio(struct bio *bio)
838{
839 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
840
4e4cbee9 841 dc->error = blk_status_to_errno(bio->bi_status);
c81abe34 842 dc->state = D_DONE;
e31b9821 843 complete_all(&dc->wait);
c81abe34
JK		 844 bio_put(bio);
845}
846
6915ea9d
CY		 847void __check_sit_bitmap(struct f2fs_sb_info *sbi,
848 block_t start, block_t end)
849{
850#ifdef CONFIG_F2FS_CHECK_FS
851 struct seg_entry *sentry;
852 unsigned int segno;
853 block_t blk = start;
854 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
855 unsigned long *map;
856
857 while (blk < end) {
858 segno = GET_SEGNO(sbi, blk);
859 sentry = get_seg_entry(sbi, segno);
860 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
861
008396e1
YS		 862 if (end < START_BLOCK(sbi, segno + 1))
863 size = GET_BLKOFF_FROM_SEG0(sbi, end);
864 else
865 size = max_blocks;
6915ea9d
CY		 866 map = (unsigned long *)(sentry->cur_valid_map);
867 offset = __find_rev_next_bit(map, size, offset);
868 f2fs_bug_on(sbi, offset != size);
008396e1 869 blk = START_BLOCK(sbi, segno + 1);
6915ea9d
CY		 870 }
871#endif
872}
873
c81abe34
JK		 874/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
875static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
8412663d 876 struct discard_cmd *dc, bool fstrim)
c81abe34
JK		 877{
878 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
8412663d
CY		 879 struct list_head *wait_list = fstrim ? &(dcc->fstrim_list) :
880 &(dcc->wait_list);
c81abe34
JK		 881 struct bio *bio = NULL;
882
883 if (dc->state != D_PREP)
884 return;
885
0243a5f9
CY		 886 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
887
c81abe34
JK		 888 dc->error = __blkdev_issue_discard(dc->bdev,
889 SECTOR_FROM_BLOCK(dc->start),
890 SECTOR_FROM_BLOCK(dc->len),
891 GFP_NOFS, 0, &bio);
892 if (!dc->error) {
893 /* should keep before submission to avoid D_DONE right away */
894 dc->state = D_SUBMIT;
8b8dd65f
CY		 895 atomic_inc(&dcc->issued_discard);
896 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 897 if (bio) {
898 bio->bi_private = dc;
899 bio->bi_end_io = f2fs_submit_discard_endio;
900 bio->bi_opf |= REQ_SYNC;
901 submit_bio(bio);
8412663d 902 list_move_tail(&dc->list, wait_list);
6915ea9d 903 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
b0af6d49
CY		 904
905 f2fs_update_iostat(sbi, FS_DISCARD, 1);
c81abe34
JK		 906 }
907 } else {
908 __remove_discard_cmd(sbi, dc);
909 }
910}
911
004b6862
CY		 912static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
913 struct block_device *bdev, block_t lstart,
914 block_t start, block_t len,
915 struct rb_node **insert_p,
916 struct rb_node *insert_parent)
c81abe34 917{
004b6862
CY		 918 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
919 struct rb_node **p = &dcc->root.rb_node;
920 struct rb_node *parent = NULL;
921 struct discard_cmd *dc = NULL;
922
923 if (insert_p && insert_parent) {
924 parent = insert_parent;
925 p = insert_p;
926 goto do_insert;
927 }
c81abe34 928
004b6862
CY		 929 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
930do_insert:
931 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
932 if (!dc)
933 return NULL;
c81abe34 934
004b6862 935 return dc;
c81abe34
JK		 936}
937
ba48a33e
CY		 938static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
939 struct discard_cmd *dc)
940{
941 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
942}
943
3d6a650f
YH		 944static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
945 struct discard_cmd *dc, block_t blkaddr)
946{
ba48a33e 947 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 948 struct discard_info di = dc->di;
949 bool modified = false;
3d6a650f 950
004b6862 951 if (dc->state == D_DONE || dc->len == 1) {
3d6a650f
YH		 952 __remove_discard_cmd(sbi, dc);
953 return;
954 }
955
d84d1cbd
CY		 956 dcc->undiscard_blks -= di.len;
957
004b6862 958 if (blkaddr > di.lstart) {
3d6a650f 959 dc->len = blkaddr - dc->lstart;
d84d1cbd 960 dcc->undiscard_blks += dc->len;
ba48a33e 961 __relocate_discard_cmd(dcc, dc);
004b6862
CY		 962 modified = true;
963 }
964
965 if (blkaddr < di.lstart + di.len - 1) {
966 if (modified) {
967 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
968 di.start + blkaddr + 1 - di.lstart,
969 di.lstart + di.len - 1 - blkaddr,
970 NULL, NULL);
971 } else {
972 dc->lstart++;
973 dc->len--;
974 dc->start++;
d84d1cbd 975 dcc->undiscard_blks += dc->len;
ba48a33e 976 __relocate_discard_cmd(dcc, dc);
004b6862 977 }
3d6a650f
YH		 978 }
979}
980
004b6862
CY		 981static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
982 struct block_device *bdev, block_t lstart,
983 block_t start, block_t len)
275b66b0 984{
0b54fb84 985 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
004b6862
CY		 986 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
987 struct discard_cmd *dc;
988 struct discard_info di = {0};
989 struct rb_node **insert_p = NULL, *insert_parent = NULL;
990 block_t end = lstart + len;
275b66b0 991
15469963 992 mutex_lock(&dcc->cmd_lock);
40465257 993
004b6862
CY		 994 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
995 NULL, lstart,
996 (struct rb_entry **)&prev_dc,
997 (struct rb_entry **)&next_dc,
998 &insert_p, &insert_parent, true);
999 if (dc)
1000 prev_dc = dc;
1001
1002 if (!prev_dc) {
1003 di.lstart = lstart;
1004 di.len = next_dc ? next_dc->lstart - lstart : len;
1005 di.len = min(di.len, len);
1006 di.start = start;
22d375dd 1007 }
15469963 1008
004b6862
CY		 1009 while (1) {
1010 struct rb_node *node;
1011 bool merged = false;
1012 struct discard_cmd *tdc = NULL;
1013
1014 if (prev_dc) {
1015 di.lstart = prev_dc->lstart + prev_dc->len;
1016 if (di.lstart < lstart)
1017 di.lstart = lstart;
1018 if (di.lstart >= end)
1019 break;
1020
1021 if (!next_dc || next_dc->lstart > end)
1022 di.len = end - di.lstart;
1023 else
1024 di.len = next_dc->lstart - di.lstart;
1025 di.start = start + di.lstart - lstart;
1026 }
1027
1028 if (!di.len)
1029 goto next;
1030
1031 if (prev_dc && prev_dc->state == D_PREP &&
1032 prev_dc->bdev == bdev &&
1033 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1034 prev_dc->di.len += di.len;
d84d1cbd 1035 dcc->undiscard_blks += di.len;
ba48a33e 1036 __relocate_discard_cmd(dcc, prev_dc);
004b6862
CY		 1037 di = prev_dc->di;
1038 tdc = prev_dc;
1039 merged = true;
1040 }
1041
1042 if (next_dc && next_dc->state == D_PREP &&
1043 next_dc->bdev == bdev &&
1044 __is_discard_front_mergeable(&di, &next_dc->di)) {
1045 next_dc->di.lstart = di.lstart;
1046 next_dc->di.len += di.len;
1047 next_dc->di.start = di.start;
d84d1cbd 1048 dcc->undiscard_blks += di.len;
ba48a33e 1049 __relocate_discard_cmd(dcc, next_dc);
004b6862
CY		 1050 if (tdc)
1051 __remove_discard_cmd(sbi, tdc);
004b6862 1052 merged = true;
4e6a8d9b 1053 }
004b6862 1054
df0f6b44 1055 if (!merged) {
004b6862
CY		 1056 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1057 di.len, NULL, NULL);
df0f6b44 1058 }
004b6862
CY		 1059 next:
1060 prev_dc = next_dc;
1061 if (!prev_dc)
1062 break;
1063
1064 node = rb_next(&prev_dc->rb_node);
1065 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1066 }
1067
1068 mutex_unlock(&dcc->cmd_lock);
1069}
1070
1071static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1072 struct block_device *bdev, block_t blkstart, block_t blklen)
1073{
1074 block_t lblkstart = blkstart;
1075
0243a5f9 1076 trace_f2fs_queue_discard(bdev, blkstart, blklen);
004b6862
CY		 1077
1078 if (sbi->s_ndevs) {
1079 int devi = f2fs_target_device_index(sbi, blkstart);
1080
1081 blkstart -= FDEV(devi).start_blk;
1082 }
1083 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
004b6862
CY		 1084 return 0;
1085}
1086
8412663d
CY		 1087static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
1088 unsigned int start, unsigned int end,
1089 unsigned int granularity)
1090{
1091 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1092 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1093 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1094 struct discard_cmd *dc;
1095 struct blk_plug plug;
1096 int issued;
1097
1098next:
1099 issued = 0;
1100
1101 mutex_lock(&dcc->cmd_lock);
1102 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1103
1104 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1105 NULL, start,
1106 (struct rb_entry **)&prev_dc,
1107 (struct rb_entry **)&next_dc,
1108 &insert_p, &insert_parent, true);
1109 if (!dc)
1110 dc = next_dc;
1111
1112 blk_start_plug(&plug);
1113
1114 while (dc && dc->lstart <= end) {
1115 struct rb_node *node;
1116
1117 if (dc->len < granularity)
1118 goto skip;
1119
1120 if (dc->state != D_PREP) {
1121 list_move_tail(&dc->list, &dcc->fstrim_list);
1122 goto skip;
1123 }
1124
1125 __submit_discard_cmd(sbi, dc, true);
1126
1127 if (++issued >= DISCARD_ISSUE_RATE) {
1128 start = dc->lstart + dc->len;
1129
1130 blk_finish_plug(&plug);
1131 mutex_unlock(&dcc->cmd_lock);
1132
1133 schedule();
1134
1135 goto next;
1136 }
1137skip:
1138 node = rb_next(&dc->rb_node);
1139 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1140
1141 if (fatal_signal_pending(current))
1142 break;
1143 }
1144
1145 blk_finish_plug(&plug);
1146 mutex_unlock(&dcc->cmd_lock);
1147}
1148
969d1b18 1149static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
bd5b0738
CY		 1150{
1151 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1152 struct list_head *pend_list;
1153 struct discard_cmd *dc, *tmp;
1154 struct blk_plug plug;
969d1b18
CY		 1155 int iter = 0, issued = 0;
1156 int i;
e6c6de18 1157 bool io_interrupted = false;
bd5b0738
CY		 1158
1159 mutex_lock(&dcc->cmd_lock);
963932a9
YH		 1160 f2fs_bug_on(sbi,
1161 !__check_rb_tree_consistence(sbi, &dcc->root));
bd5b0738 1162 blk_start_plug(&plug);
969d1b18
CY		 1163 for (i = MAX_PLIST_NUM - 1;
1164 i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
bd5b0738
CY		 1165 pend_list = &dcc->pend_list[i];
1166 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1167 f2fs_bug_on(sbi, dc->state != D_PREP);
1168
969d1b18
CY		 1169 /* Hurry up to finish fstrim */
1170 if (dcc->pend_list_tag[i] & P_TRIM) {
8412663d 1171 __submit_discard_cmd(sbi, dc, false);
969d1b18
CY		 1172 issued++;
1173 continue;
1174 }
1175
e6c6de18 1176 if (!issue_cond) {
8412663d 1177 __submit_discard_cmd(sbi, dc, false);
969d1b18 1178 issued++;
e6c6de18
CY		 1179 continue;
1180 }
1181
1182 if (is_idle(sbi)) {
8412663d 1183 __submit_discard_cmd(sbi, dc, false);
e6c6de18
CY		 1184 issued++;
1185 } else {
1186 io_interrupted = true;
969d1b18 1187 }
e6c6de18
CY		 1188
1189 if (++iter >= DISCARD_ISSUE_RATE)
bd5b0738
CY		 1190 goto out;
1191 }
969d1b18
CY		 1192 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1193 dcc->pend_list_tag[i] &= (~P_TRIM);
bd5b0738
CY		 1194 }
1195out:
1196 blk_finish_plug(&plug);
1197 mutex_unlock(&dcc->cmd_lock);
969d1b18 1198
e6c6de18
CY		 1199 if (!issued && io_interrupted)
1200 issued = -1;
1201
969d1b18
CY		 1202 return issued;
1203}
1204
1205static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1206{
1207 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1208 struct list_head *pend_list;
1209 struct discard_cmd *dc, *tmp;
1210 int i;
1211
1212 mutex_lock(&dcc->cmd_lock);
1213 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1214 pend_list = &dcc->pend_list[i];
1215 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1216 f2fs_bug_on(sbi, dc->state != D_PREP);
1217 __remove_discard_cmd(sbi, dc);
1218 }
1219 }
1220 mutex_unlock(&dcc->cmd_lock);
bd5b0738
CY		 1221}
1222
2a510c00
CY		 1223static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1224 struct discard_cmd *dc)
1225{
1226 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1227
1228 wait_for_completion_io(&dc->wait);
1229 mutex_lock(&dcc->cmd_lock);
1230 f2fs_bug_on(sbi, dc->state != D_DONE);
1231 dc->ref--;
1232 if (!dc->ref)
1233 __remove_discard_cmd(sbi, dc);
1234 mutex_unlock(&dcc->cmd_lock);
1235}
1236
8412663d
CY		 1237static void __wait_discard_cmd_range(struct f2fs_sb_info *sbi, bool wait_cond,
1238 block_t start, block_t end,
1239 unsigned int granularity,
1240 bool fstrim)
63a94fa1
CY		 1241{
1242 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
8412663d
CY		 1243 struct list_head *wait_list = fstrim ? &(dcc->fstrim_list) :
1244 &(dcc->wait_list);
63a94fa1 1245 struct discard_cmd *dc, *tmp;
6afae633
CY		 1246 bool need_wait;
1247
1248next:
1249 need_wait = false;
63a94fa1
CY		 1250
1251 mutex_lock(&dcc->cmd_lock);
1252 list_for_each_entry_safe(dc, tmp, wait_list, list) {
8412663d
CY		 1253 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1254 continue;
1255 if (dc->len < granularity)
1256 continue;
6afae633 1257 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
63a94fa1
CY		 1258 wait_for_completion_io(&dc->wait);
1259 __remove_discard_cmd(sbi, dc);
6afae633
CY		 1260 } else {
1261 dc->ref++;
1262 need_wait = true;
1263 break;
63a94fa1
CY		 1264 }
1265 }
1266 mutex_unlock(&dcc->cmd_lock);
6afae633
CY		 1267
1268 if (need_wait) {
2a510c00 1269 __wait_one_discard_bio(sbi, dc);
6afae633
CY		 1270 goto next;
1271 }
63a94fa1
CY		 1272}
1273
8412663d
CY		 1274static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1275{
1276 __wait_discard_cmd_range(sbi, wait_cond, 0, UINT_MAX, 1, false);
1277}
1278
004b6862
CY		 1279/* This should be covered by global mutex, &sit_i->sentry_lock */
1280void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1281{
1282 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1283 struct discard_cmd *dc;
ec9895ad 1284 bool need_wait = false;
004b6862
CY		 1285
1286 mutex_lock(&dcc->cmd_lock);
004b6862
CY		 1287 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1288 if (dc) {
ec9895ad
CY		 1289 if (dc->state == D_PREP) {
1290 __punch_discard_cmd(sbi, dc, blkaddr);
1291 } else {
1292 dc->ref++;
1293 need_wait = true;
1294 }
275b66b0 1295 }
d431413f 1296 mutex_unlock(&dcc->cmd_lock);
ec9895ad 1297
2a510c00
CY		 1298 if (need_wait)
1299 __wait_one_discard_bio(sbi, dc);
d431413f
CY		 1300}
1301
cce13252
CY		 1302void stop_discard_thread(struct f2fs_sb_info *sbi)
1303{
1304 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1305
1306 if (dcc && dcc->f2fs_issue_discard) {
1307 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1308
1309 dcc->f2fs_issue_discard = NULL;
1310 kthread_stop(discard_thread);
ec9895ad 1311 }
d431413f
CY		 1312}
1313
8412663d
CY		 1314/* This comes from f2fs_put_super */
1315void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
d431413f 1316{
bd5b0738 1317 __issue_discard_cmd(sbi, false);
969d1b18 1318 __drop_discard_cmd(sbi);
8412663d 1319 __wait_all_discard_cmd(sbi, false);
275b66b0
CY		 1320}
1321
969d1b18
CY		 1322static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1323{
1324 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1325 int i;
1326
1327 mutex_lock(&dcc->cmd_lock);
1328 for (i = 0; i < MAX_PLIST_NUM; i++)
1329 dcc->pend_list_tag[i] |= P_TRIM;
1330 mutex_unlock(&dcc->cmd_lock);
1331}
1332
15469963
JK		 1333static int issue_discard_thread(void *data)
1334{
1335 struct f2fs_sb_info *sbi = data;
1336 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1337 wait_queue_head_t *q = &dcc->discard_wait_queue;
969d1b18
CY		 1338 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1339 int issued;
15469963 1340
1d7be270 1341 set_freezable();
15469963 1342
1d7be270 1343 do {
969d1b18
CY		 1344 wait_event_interruptible_timeout(*q,
1345 kthread_should_stop() || freezing(current) ||
1346 dcc->discard_wake,
1347 msecs_to_jiffies(wait_ms));
1d7be270
JK		 1348 if (try_to_freeze())
1349 continue;
1350 if (kthread_should_stop())
1351 return 0;
15469963 1352
5f656541 1353 if (dcc->discard_wake) {
969d1b18 1354 dcc->discard_wake = 0;
5f656541
JK		 1355 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1356 mark_discard_range_all(sbi);
1357 }
969d1b18 1358
dc6febb6
CY		 1359 sb_start_intwrite(sbi->sb);
1360
969d1b18
CY		 1361 issued = __issue_discard_cmd(sbi, true);
1362 if (issued) {
8412663d 1363 __wait_all_discard_cmd(sbi, true);
969d1b18
CY		 1364 wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1365 } else {
1366 wait_ms = DEF_MAX_DISCARD_ISSUE_TIME;
1367 }
1d7be270 1368
dc6febb6 1369 sb_end_intwrite(sbi->sb);
1d7be270 1370
1d7be270
JK		 1371 } while (!kthread_should_stop());
1372 return 0;
15469963
JK		 1373}
1374
f46e8809 1375#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 1376static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1377 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 1378{
92592285 1379 sector_t sector, nr_sects;
10a875f8 1380 block_t lblkstart = blkstart;
3c62be17
JK		 1381 int devi = 0;
1382
1383 if (sbi->s_ndevs) {
1384 devi = f2fs_target_device_index(sbi, blkstart);
1385 blkstart -= FDEV(devi).start_blk;
1386 }
f46e8809
DLM		 1387
1388 /*
1389 * We need to know the type of the zone: for conventional zones,
1390 * use regular discard if the drive supports it. For sequential
1391 * zones, reset the zone write pointer.
1392 */
3c62be17 1393 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 1394
1395 case BLK_ZONE_TYPE_CONVENTIONAL:
1396 if (!blk_queue_discard(bdev_get_queue(bdev)))
1397 return 0;
c81abe34 1398 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 1399 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1400 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 1401 sector = SECTOR_FROM_BLOCK(blkstart);
1402 nr_sects = SECTOR_FROM_BLOCK(blklen);
1403
1404 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1405 nr_sects != bdev_zone_sectors(bdev)) {
1406 f2fs_msg(sbi->sb, KERN_INFO,
1407 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1408 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1409 blkstart, blklen);
1410 return -EIO;
1411 }
d50aaeec 1412 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 1413 return blkdev_reset_zones(bdev, sector,
1414 nr_sects, GFP_NOFS);
1415 default:
1416 /* Unknown zone type: broken device ? */
1417 return -EIO;
1418 }
1419}
1420#endif
1421
3c62be17
JK		 1422static int __issue_discard_async(struct f2fs_sb_info *sbi,
1423 struct block_device *bdev, block_t blkstart, block_t blklen)
1424{
1425#ifdef CONFIG_BLK_DEV_ZONED
1426 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1427 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1428 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1429#endif
c81abe34 1430 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 1431}
1432
1e87a78d 1433static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 1434 block_t blkstart, block_t blklen)
1435{
3c62be17
JK		 1436 sector_t start = blkstart, len = 0;
1437 struct block_device *bdev;
a66cdd98
JK		 1438 struct seg_entry *se;
1439 unsigned int offset;
1440 block_t i;
3c62be17
JK		 1441 int err = 0;
1442
1443 bdev = f2fs_target_device(sbi, blkstart, NULL);
1444
1445 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1446 if (i != start) {
1447 struct block_device *bdev2 =
1448 f2fs_target_device(sbi, i, NULL);
1449
1450 if (bdev2 != bdev) {
1451 err = __issue_discard_async(sbi, bdev,
1452 start, len);
1453 if (err)
1454 return err;
1455 bdev = bdev2;
1456 start = i;
1457 len = 0;
1458 }
1459 }
a66cdd98 1460
a66cdd98
JK		 1461 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1462 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1463
1464 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1465 sbi->discard_blks--;
1466 }
f46e8809 1467
3c62be17
JK		 1468 if (len)
1469 err = __issue_discard_async(sbi, bdev, start, len);
1470 return err;
1e87a78d
JK		 1471}
1472
25290fa5
JK		 1473static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1474 bool check_only)
adf4983b 1475{
b2955550
JK		 1476 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1477 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 1478 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 1479 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1480 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 1481 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1482 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1483 unsigned int start = 0, end = -1;
c473f1a9 1484 bool force = (cpc->reason & CP_DISCARD);
a7eeb823 1485 struct discard_entry *de = NULL;
46f84c2c 1486 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
b2955550
JK		 1487 int i;
1488
3e025740 1489 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1490 return false;
b2955550 1491
a66cdd98
JK		 1492 if (!force) {
1493 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1494 SM_I(sbi)->dcc_info->nr_discards >=
1495 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1496 return false;
4b2fecc8
JK		 1497 }
1498
b2955550
JK		 1499 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1500 for (i = 0; i < entries; i++)
a66cdd98 1501 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1502 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1503
0b54fb84
JK		 1504 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1505 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1506 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1507 if (start >= max_blocks)
1508 break;
1509
1510 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1511 if (force && start && end != max_blocks
1512 && (end - start) < cpc->trim_minlen)
1513 continue;
1514
25290fa5
JK		 1515 if (check_only)
1516 return true;
1517
a7eeb823
CY		 1518 if (!de) {
1519 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1520 GFP_F2FS_ZERO);
1521 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1522 list_add_tail(&de->list, head);
1523 }
1524
1525 for (i = start; i < end; i++)
1526 __set_bit_le(i, (void *)de->discard_map);
1527
1528 SM_I(sbi)->dcc_info->nr_discards += end - start;
b2955550 1529 }
25290fa5 1530 return false;
b2955550
JK		 1531}
1532
4b2fecc8
JK		 1533void release_discard_addrs(struct f2fs_sb_info *sbi)
1534{
46f84c2c 1535 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
4b2fecc8
JK		 1536 struct discard_entry *entry, *this;
1537
1538 /* drop caches */
1539 list_for_each_entry_safe(entry, this, head, list) {
1540 list_del(&entry->list);
1541 kmem_cache_free(discard_entry_slab, entry);
1542 }
1543}
1544
0a8165d7 1545/*
351df4b2
JK		 1546 * Should call clear_prefree_segments after checkpoint is done.
1547 */
1548static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1549{
1550 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1551 unsigned int segno;
351df4b2
JK		 1552
1553 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1554 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1555 __set_test_and_free(sbi, segno);
351df4b2
JK		 1556 mutex_unlock(&dirty_i->seglist_lock);
1557}
1558
836b5a63 1559void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1560{
969d1b18
CY		 1561 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1562 struct list_head *head = &dcc->entry_list;
2d7b822a 1563 struct discard_entry *entry, *this;
351df4b2 1564 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1565 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1566 unsigned int start = 0, end = -1;
36abef4e 1567 unsigned int secno, start_segno;
c473f1a9 1568 bool force = (cpc->reason & CP_DISCARD);
351df4b2
JK		 1569
1570 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1571
351df4b2 1572 while (1) {
29e59c14 1573 int i;
7cd8558b
JK		 1574 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1575 if (start >= MAIN_SEGS(sbi))
351df4b2 1576 break;
7cd8558b
JK		 1577 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1578 start + 1);
29e59c14
CL		 1579
1580 for (i = start; i < end; i++)
1581 clear_bit(i, prefree_map);
1582
1583 dirty_i->nr_dirty[PRE] -= end - start;
1584
650d3c4e 1585 if (!test_opt(sbi, DISCARD))
29e59c14 1586 continue;
351df4b2 1587
650d3c4e
YH		 1588 if (force && start >= cpc->trim_start &&
1589 (end - 1) <= cpc->trim_end)
1590 continue;
1591
36abef4e
JK		 1592 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1593 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1594 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1595 continue;
1596 }
1597next:
4ddb1a4d
JK		 1598 secno = GET_SEC_FROM_SEG(sbi, start);
1599 start_segno = GET_SEG_FROM_SEC(sbi, secno);
36abef4e 1600 if (!IS_CURSEC(sbi, secno) &&
302bd348 1601 !get_valid_blocks(sbi, start, true))
36abef4e
JK		 1602 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1603 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1604
1605 start = start_segno + sbi->segs_per_sec;
1606 if (start < end)
1607 goto next;
8b107f5b
JK		 1608 else
1609 end = start - 1;
351df4b2
JK		 1610 }
1611 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1612
1613 /* send small discards */
2d7b822a 1614 list_for_each_entry_safe(entry, this, head, list) {
a7eeb823
CY		 1615 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1616 bool is_valid = test_bit_le(0, entry->discard_map);
1617
1618find_next:
1619 if (is_valid) {
1620 next_pos = find_next_zero_bit_le(entry->discard_map,
1621 sbi->blocks_per_seg, cur_pos);
1622 len = next_pos - cur_pos;
1623
acfd2810
DLM		 1624 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1625 (force && len < cpc->trim_minlen))
a7eeb823
CY		 1626 goto skip;
1627
1628 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1629 len);
1630 cpc->trimmed += len;
1631 total_len += len;
1632 } else {
1633 next_pos = find_next_bit_le(entry->discard_map,
1634 sbi->blocks_per_seg, cur_pos);
1635 }
836b5a63 1636skip:
a7eeb823
CY		 1637 cur_pos = next_pos;
1638 is_valid = !is_valid;
1639
1640 if (cur_pos < sbi->blocks_per_seg)
1641 goto find_next;
1642
b2955550 1643 list_del(&entry->list);
969d1b18 1644 dcc->nr_discards -= total_len;
b2955550
JK		 1645 kmem_cache_free(discard_entry_slab, entry);
1646 }
34e159da 1647
01983c71 1648 wake_up_discard_thread(sbi, false);
351df4b2
JK		 1649}
1650
8ed59745 1651static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1652{
15469963 1653 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84 1654 struct discard_cmd_control *dcc;
ba48a33e 1655 int err = 0, i;
0b54fb84
JK		 1656
1657 if (SM_I(sbi)->dcc_info) {
1658 dcc = SM_I(sbi)->dcc_info;
1659 goto init_thread;
1660 }
1661
1662 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1663 if (!dcc)
1664 return -ENOMEM;
1665
969d1b18 1666 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
46f84c2c 1667 INIT_LIST_HEAD(&dcc->entry_list);
969d1b18 1668 for (i = 0; i < MAX_PLIST_NUM; i++) {
ba48a33e 1669 INIT_LIST_HEAD(&dcc->pend_list[i]);
969d1b18
CY		 1670 if (i >= dcc->discard_granularity - 1)
1671 dcc->pend_list_tag[i] |= P_ACTIVE;
1672 }
46f84c2c 1673 INIT_LIST_HEAD(&dcc->wait_list);
8412663d 1674 INIT_LIST_HEAD(&dcc->fstrim_list);
15469963 1675 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1676 atomic_set(&dcc->issued_discard, 0);
1677 atomic_set(&dcc->issing_discard, 0);
5f32366a 1678 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84 1679 dcc->nr_discards = 0;
d618ebaf 1680 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
d84d1cbd 1681 dcc->undiscard_blks = 0;
004b6862 1682 dcc->root = RB_ROOT;
0b54fb84 1683
15469963 1684 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1685 SM_I(sbi)->dcc_info = dcc;
1686init_thread:
15469963
JK		 1687 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1688 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1689 if (IS_ERR(dcc->f2fs_issue_discard)) {
1690 err = PTR_ERR(dcc->f2fs_issue_discard);
1691 kfree(dcc);
1692 SM_I(sbi)->dcc_info = NULL;
1693 return err;
1694 }
1695
0b54fb84
JK		 1696 return err;
1697}
1698
f099405f 1699static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1700{
1701 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1702
f099405f
CY		 1703 if (!dcc)
1704 return;
1705
cce13252 1706 stop_discard_thread(sbi);
f099405f
CY		 1707
1708 kfree(dcc);
1709 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1710}
1711
184a5cd2 1712static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1713{
1714 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1715
1716 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1717 sit_i->dirty_sentries++;
184a5cd2
CY		 1718 return false;
1719 }
1720
1721 return true;
351df4b2
JK		 1722}
1723
1724static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1725 unsigned int segno, int modified)
1726{
1727 struct seg_entry *se = get_seg_entry(sbi, segno);
1728 se->type = type;
1729 if (modified)
1730 __mark_sit_entry_dirty(sbi, segno);
1731}
1732
1733static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1734{
1735 struct seg_entry *se;
1736 unsigned int segno, offset;
1737 long int new_vblocks;
6415fedc
YS		 1738 bool exist;
1739#ifdef CONFIG_F2FS_CHECK_FS
1740 bool mir_exist;
1741#endif
351df4b2
JK		 1742
1743 segno = GET_SEGNO(sbi, blkaddr);
1744
1745 se = get_seg_entry(sbi, segno);
1746 new_vblocks = se->valid_blocks + del;
491c0854 1747 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1748
9850cf4a 1749 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1750 (new_vblocks > sbi->blocks_per_seg)));
1751
1752 se->valid_blocks = new_vblocks;
1753 se->mtime = get_mtime(sbi);
1754 SIT_I(sbi)->max_mtime = se->mtime;
1755
1756 /* Update valid block bitmap */
1757 if (del > 0) {
6415fedc 1758 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
355e7891 1759#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1760 mir_exist = f2fs_test_and_set_bit(offset,
1761 se->cur_valid_map_mir);
1762 if (unlikely(exist != mir_exist)) {
1763 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1764 "when setting bitmap, blk:%u, old bit:%d",
1765 blkaddr, exist);
05796763 1766 f2fs_bug_on(sbi, 1);
6415fedc 1767 }
355e7891 1768#endif
6415fedc
YS		 1769 if (unlikely(exist)) {
1770 f2fs_msg(sbi->sb, KERN_ERR,
1771 "Bitmap was wrongly set, blk:%u", blkaddr);
1772 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1773 se->valid_blocks--;
1774 del = 0;
355e7891 1775 }
6415fedc 1776
3e025740
JK		 1777 if (f2fs_discard_en(sbi) &&
1778 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1779 sbi->discard_blks--;
720037f9
JK		 1780
1781 /* don't overwrite by SSR to keep node chain */
1782 if (se->type == CURSEG_WARM_NODE) {
1783 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1784 se->ckpt_valid_blocks++;
1785 }
351df4b2 1786 } else {
6415fedc 1787 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
355e7891 1788#ifdef CONFIG_F2FS_CHECK_FS
6415fedc
YS		 1789 mir_exist = f2fs_test_and_clear_bit(offset,
1790 se->cur_valid_map_mir);
1791 if (unlikely(exist != mir_exist)) {
1792 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1793 "when clearing bitmap, blk:%u, old bit:%d",
1794 blkaddr, exist);
05796763 1795 f2fs_bug_on(sbi, 1);
6415fedc 1796 }
355e7891 1797#endif
6415fedc
YS		 1798 if (unlikely(!exist)) {
1799 f2fs_msg(sbi->sb, KERN_ERR,
1800 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1801 f2fs_bug_on(sbi, 1);
35ee82ca
YS		 1802 se->valid_blocks++;
1803 del = 0;
355e7891 1804 }
6415fedc 1805
3e025740
JK		 1806 if (f2fs_discard_en(sbi) &&
1807 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1808 sbi->discard_blks++;
351df4b2
JK		 1809 }
1810 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1811 se->ckpt_valid_blocks += del;
1812
1813 __mark_sit_entry_dirty(sbi, segno);
1814
1815 /* update total number of valid blocks to be written in ckpt area */
1816 SIT_I(sbi)->written_valid_blocks += del;
1817
1818 if (sbi->segs_per_sec > 1)
1819 get_sec_entry(sbi, segno)->valid_blocks += del;
1820}
1821
5e443818 1822void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1823{
5e443818
JK		 1824 update_sit_entry(sbi, new, 1);
1825 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1826 update_sit_entry(sbi, old, -1);
1827
1828 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1829 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1830}
1831
1832void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1833{
1834 unsigned int segno = GET_SEGNO(sbi, addr);
1835 struct sit_info *sit_i = SIT_I(sbi);
1836
9850cf4a 1837 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1838 if (addr == NEW_ADDR)
1839 return;
1840
1841 /* add it into sit main buffer */
1842 mutex_lock(&sit_i->sentry_lock);
1843
1844 update_sit_entry(sbi, addr, -1);
1845
1846 /* add it into dirty seglist */
1847 locate_dirty_segment(sbi, segno);
1848
1849 mutex_unlock(&sit_i->sentry_lock);
1850}
1851
6e2c64ad
JK		 1852bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1853{
1854 struct sit_info *sit_i = SIT_I(sbi);
1855 unsigned int segno, offset;
1856 struct seg_entry *se;
1857 bool is_cp = false;
1858
1859 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1860 return true;
1861
1862 mutex_lock(&sit_i->sentry_lock);
1863
1864 segno = GET_SEGNO(sbi, blkaddr);
1865 se = get_seg_entry(sbi, segno);
1866 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1867
1868 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1869 is_cp = true;
1870
1871 mutex_unlock(&sit_i->sentry_lock);
1872
1873 return is_cp;
1874}
1875
0a8165d7 1876/*
351df4b2
JK		 1877 * This function should be resided under the curseg_mutex lock
1878 */
1879static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1880 struct f2fs_summary *sum)
351df4b2
JK		 1881{
1882 struct curseg_info *curseg = CURSEG_I(sbi, type);
1883 void *addr = curseg->sum_blk;
e79efe3b 1884 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1885 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1886}
1887
0a8165d7 1888/*
351df4b2
JK		 1889 * Calculate the number of current summary pages for writing
1890 */
3fa06d7b 1891int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1892{
351df4b2 1893 int valid_sum_count = 0;
9a47938b 1894 int i, sum_in_page;
351df4b2
JK		 1895
1896 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1897 if (sbi->ckpt->alloc_type[i] == SSR)
1898 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1899 else {
1900 if (for_ra)
1901 valid_sum_count += le16_to_cpu(
1902 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1903 else
1904 valid_sum_count += curseg_blkoff(sbi, i);
1905 }
351df4b2
JK		 1906 }
1907
09cbfeaf 1908 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1909 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1910 if (valid_sum_count <= sum_in_page)
351df4b2 1911 return 1;
9a47938b 1912 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1913 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1914 return 2;
1915 return 3;
1916}
1917
0a8165d7 1918/*
351df4b2
JK		 1919 * Caller should put this summary page
1920 */
1921struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1922{
1923 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1924}
1925
381722d2 1926void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1927{
1928 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1929 void *dst = page_address(page);
1930
1931 if (src)
09cbfeaf 1932 memcpy(dst, src, PAGE_SIZE);
381722d2 1933 else
09cbfeaf 1934 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1935 set_page_dirty(page);
1936 f2fs_put_page(page, 1);
1937}
1938
381722d2
CY		 1939static void write_sum_page(struct f2fs_sb_info *sbi,
1940 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1941{
1942 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1943}
1944
b7ad7512
CY		 1945static void write_current_sum_page(struct f2fs_sb_info *sbi,
1946 int type, block_t blk_addr)
1947{
1948 struct curseg_info *curseg = CURSEG_I(sbi, type);
1949 struct page *page = grab_meta_page(sbi, blk_addr);
1950 struct f2fs_summary_block *src = curseg->sum_blk;
1951 struct f2fs_summary_block *dst;
1952
1953 dst = (struct f2fs_summary_block *)page_address(page);
1954
1955 mutex_lock(&curseg->curseg_mutex);
1956
1957 down_read(&curseg->journal_rwsem);
1958 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1959 up_read(&curseg->journal_rwsem);
1960
1961 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1962 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1963
1964 mutex_unlock(&curseg->curseg_mutex);
1965
1966 set_page_dirty(page);
1967 f2fs_put_page(page, 1);
1968}
1969
a7881893
JK		 1970static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1971{
1972 struct curseg_info *curseg = CURSEG_I(sbi, type);
1973 unsigned int segno = curseg->segno + 1;
1974 struct free_segmap_info *free_i = FREE_I(sbi);
1975
1976 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1977 return !test_bit(segno, free_i->free_segmap);
1978 return 0;
1979}
1980
0a8165d7 1981/*
351df4b2
JK		 1982 * Find a new segment from the free segments bitmap to right order
1983 * This function should be returned with success, otherwise BUG
1984 */
1985static void get_new_segment(struct f2fs_sb_info *sbi,
1986 unsigned int *newseg, bool new_sec, int dir)
1987{
1988 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1989 unsigned int segno, secno, zoneno;
7cd8558b 1990 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
4ddb1a4d
JK		 1991 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1992 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
351df4b2
JK		 1993 unsigned int left_start = hint;
1994 bool init = true;
1995 int go_left = 0;
1996 int i;
1997
1a118ccf 1998 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1999
2000 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2001 segno = find_next_zero_bit(free_i->free_segmap,
4ddb1a4d
JK		 2002 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2003 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
351df4b2
JK		 2004 goto got_it;
2005 }
2006find_other_zone:
7cd8558b
JK		 2007 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2008 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 2009 if (dir == ALLOC_RIGHT) {
2010 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2011 MAIN_SECS(sbi), 0);
2012 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 2013 } else {
2014 go_left = 1;
2015 left_start = hint - 1;
2016 }
2017 }
2018 if (go_left == 0)
2019 goto skip_left;
2020
2021 while (test_bit(left_start, free_i->free_secmap)) {
2022 if (left_start > 0) {
2023 left_start--;
2024 continue;
2025 }
2026 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 2027 MAIN_SECS(sbi), 0);
2028 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 2029 break;
2030 }
2031 secno = left_start;
2032skip_left:
2033 hint = secno;
4ddb1a4d
JK		 2034 segno = GET_SEG_FROM_SEC(sbi, secno);
2035 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
351df4b2
JK		 2036
2037 /* give up on finding another zone */
2038 if (!init)
2039 goto got_it;
2040 if (sbi->secs_per_zone == 1)
2041 goto got_it;
2042 if (zoneno == old_zoneno)
2043 goto got_it;
2044 if (dir == ALLOC_LEFT) {
2045 if (!go_left && zoneno + 1 >= total_zones)
2046 goto got_it;
2047 if (go_left && zoneno == 0)
2048 goto got_it;
2049 }
2050 for (i = 0; i < NR_CURSEG_TYPE; i++)
2051 if (CURSEG_I(sbi, i)->zone == zoneno)
2052 break;
2053
2054 if (i < NR_CURSEG_TYPE) {
2055 /* zone is in user, try another */
2056 if (go_left)
2057 hint = zoneno * sbi->secs_per_zone - 1;
2058 else if (zoneno + 1 >= total_zones)
2059 hint = 0;
2060 else
2061 hint = (zoneno + 1) * sbi->secs_per_zone;
2062 init = false;
2063 goto find_other_zone;
2064 }
2065got_it:
2066 /* set it as dirty segment in free segmap */
9850cf4a 2067 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 2068 __set_inuse(sbi, segno);
2069 *newseg = segno;
1a118ccf 2070 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 2071}
2072
2073static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2074{
2075 struct curseg_info *curseg = CURSEG_I(sbi, type);
2076 struct summary_footer *sum_footer;
2077
2078 curseg->segno = curseg->next_segno;
4ddb1a4d 2079 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
351df4b2
JK		 2080 curseg->next_blkoff = 0;
2081 curseg->next_segno = NULL_SEGNO;
2082
2083 sum_footer = &(curseg->sum_blk->footer);
2084 memset(sum_footer, 0, sizeof(struct summary_footer));
2085 if (IS_DATASEG(type))
2086 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2087 if (IS_NODESEG(type))
2088 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2089 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2090}
2091
7a20b8a6
JK		 2092static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2093{
a7881893
JK		 2094 /* if segs_per_sec is large than 1, we need to keep original policy. */
2095 if (sbi->segs_per_sec != 1)
2096 return CURSEG_I(sbi, type)->segno;
2097
7a20b8a6
JK		 2098 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
2099 return 0;
2100
e066b83c
JK		 2101 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2102 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
7a20b8a6
JK		 2103 return CURSEG_I(sbi, type)->segno;
2104}
2105
0a8165d7 2106/*
351df4b2
JK		 2107 * Allocate a current working segment.
2108 * This function always allocates a free segment in LFS manner.
2109 */
2110static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2111{
2112 struct curseg_info *curseg = CURSEG_I(sbi, type);
2113 unsigned int segno = curseg->segno;
2114 int dir = ALLOC_LEFT;
2115
2116 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 2117 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 2118 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2119 dir = ALLOC_RIGHT;
2120
2121 if (test_opt(sbi, NOHEAP))
2122 dir = ALLOC_RIGHT;
2123
7a20b8a6 2124 segno = __get_next_segno(sbi, type);
351df4b2
JK		 2125 get_new_segment(sbi, &segno, new_sec, dir);
2126 curseg->next_segno = segno;
2127 reset_curseg(sbi, type, 1);
2128 curseg->alloc_type = LFS;
2129}
2130
2131static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2132 struct curseg_info *seg, block_t start)
2133{
2134 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 2135 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 2136 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 2137 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2138 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2139 int i, pos;
2140
2141 for (i = 0; i < entries; i++)
2142 target_map[i] = ckpt_map[i] | cur_map[i];
2143
2144 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2145
2146 seg->next_blkoff = pos;
351df4b2
JK		 2147}
2148
0a8165d7 2149/*
351df4b2
JK		 2150 * If a segment is written by LFS manner, next block offset is just obtained
2151 * by increasing the current block offset. However, if a segment is written by
2152 * SSR manner, next block offset obtained by calling __next_free_blkoff
2153 */
2154static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2155 struct curseg_info *seg)
2156{
2157 if (seg->alloc_type == SSR)
2158 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2159 else
2160 seg->next_blkoff++;
2161}
2162
0a8165d7 2163/*
e1c42045 2164 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 2165 * manner, so it should recover the existing segment information of valid blocks
2166 */
025d63a4 2167static void change_curseg(struct f2fs_sb_info *sbi, int type)
351df4b2
JK		 2168{
2169 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2170 struct curseg_info *curseg = CURSEG_I(sbi, type);
2171 unsigned int new_segno = curseg->next_segno;
2172 struct f2fs_summary_block *sum_node;
2173 struct page *sum_page;
2174
2175 write_sum_page(sbi, curseg->sum_blk,
2176 GET_SUM_BLOCK(sbi, curseg->segno));
2177 __set_test_and_inuse(sbi, new_segno);
2178
2179 mutex_lock(&dirty_i->seglist_lock);
2180 __remove_dirty_segment(sbi, new_segno, PRE);
2181 __remove_dirty_segment(sbi, new_segno, DIRTY);
2182 mutex_unlock(&dirty_i->seglist_lock);
2183
2184 reset_curseg(sbi, type, 1);
2185 curseg->alloc_type = SSR;
2186 __next_free_blkoff(sbi, curseg, 0);
2187
025d63a4
CY		 2188 sum_page = get_sum_page(sbi, new_segno);
2189 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2190 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2191 f2fs_put_page(sum_page, 1);
351df4b2
JK		 2192}
2193
43727527
JK		 2194static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2195{
2196 struct curseg_info *curseg = CURSEG_I(sbi, type);
2197 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
e066b83c 2198 unsigned segno = NULL_SEGNO;
d27c3d89
CY		 2199 int i, cnt;
2200 bool reversed = false;
c192f7a4
JK		 2201
2202 /* need_SSR() already forces to do this */
e066b83c
JK		 2203 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2204 curseg->next_segno = segno;
c192f7a4 2205 return 1;
e066b83c 2206 }
43727527 2207
70d625cb
JK		 2208 /* For node segments, let's do SSR more intensively */
2209 if (IS_NODESEG(type)) {
d27c3d89
CY		 2210 if (type >= CURSEG_WARM_NODE) {
2211 reversed = true;
2212 i = CURSEG_COLD_NODE;
2213 } else {
2214 i = CURSEG_HOT_NODE;
2215 }
2216 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 2217 } else {
d27c3d89
CY		 2218 if (type >= CURSEG_WARM_DATA) {
2219 reversed = true;
2220 i = CURSEG_COLD_DATA;
2221 } else {
2222 i = CURSEG_HOT_DATA;
2223 }
2224 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 2225 }
43727527 2226
d27c3d89 2227 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 2228 if (i == type)
2229 continue;
e066b83c
JK		 2230 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2231 curseg->next_segno = segno;
43727527 2232 return 1;
e066b83c 2233 }
c192f7a4 2234 }
43727527
JK		 2235 return 0;
2236}
2237
351df4b2
JK		 2238/*
2239 * flush out current segment and replace it with new segment
2240 * This function should be returned with success, otherwise BUG
2241 */
2242static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2243 int type, bool force)
2244{
a7881893
JK		 2245 struct curseg_info *curseg = CURSEG_I(sbi, type);
2246
7b405275 2247 if (force)
351df4b2 2248 new_curseg(sbi, type, true);
5b6c6be2
JK		 2249 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2250 type == CURSEG_WARM_NODE)
351df4b2 2251 new_curseg(sbi, type, false);
a7881893
JK		 2252 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2253 new_curseg(sbi, type, false);
351df4b2 2254 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
025d63a4 2255 change_curseg(sbi, type);
351df4b2
JK		 2256 else
2257 new_curseg(sbi, type, false);
dcdfff65 2258
a7881893 2259 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 2260}
2261
2262void allocate_new_segments(struct f2fs_sb_info *sbi)
2263{
6ae1be13
JK		 2264 struct curseg_info *curseg;
2265 unsigned int old_segno;
351df4b2
JK		 2266 int i;
2267
6ae1be13
JK		 2268 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2269 curseg = CURSEG_I(sbi, i);
2270 old_segno = curseg->segno;
2271 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2272 locate_dirty_segment(sbi, old_segno);
2273 }
351df4b2
JK		 2274}
2275
2276static const struct segment_allocation default_salloc_ops = {
2277 .allocate_segment = allocate_segment_by_default,
2278};
2279
25290fa5
JK		 2280bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2281{
2282 __u64 trim_start = cpc->trim_start;
2283 bool has_candidate = false;
2284
2285 mutex_lock(&SIT_I(sbi)->sentry_lock);
2286 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2287 if (add_discard_addrs(sbi, cpc, true)) {
2288 has_candidate = true;
2289 break;
2290 }
2291 }
2292 mutex_unlock(&SIT_I(sbi)->sentry_lock);
2293
2294 cpc->trim_start = trim_start;
2295 return has_candidate;
2296}
2297
4b2fecc8
JK		 2298int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2299{
f7ef9b83
JK		 2300 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2301 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
8412663d
CY		 2302 unsigned int start_segno, end_segno, cur_segno;
2303 block_t start_block, end_block;
4b2fecc8 2304 struct cp_control cpc;
c34f42e2 2305 int err = 0;
4b2fecc8 2306
836b5a63 2307 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 2308 return -EINVAL;
2309
9bd27ae4 2310 cpc.trimmed = 0;
7cd8558b 2311 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 2312 goto out;
2313
ed214a11
YH		 2314 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2315 f2fs_msg(sbi->sb, KERN_WARNING,
2316 "Found FS corruption, run fsck to fix.");
2317 goto out;
2318 }
2319
4b2fecc8 2320 /* start/end segment number in main_area */
7cd8558b
JK		 2321 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2322 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2323 GET_SEGNO(sbi, end);
8412663d
CY		 2324
2325 start_block = START_BLOCK(sbi, start_segno);
2326 end_block = START_BLOCK(sbi, end_segno + 1);
2327
4b2fecc8 2328 cpc.reason = CP_DISCARD;
836b5a63 2329 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 2330
2331 /* do checkpoint to issue discard commands safely */
8412663d
CY		 2332 for (cur_segno = start_segno; cur_segno <= end_segno;
2333 cur_segno = cpc.trim_end + 1) {
2334 cpc.trim_start = cur_segno;
a66cdd98
JK		 2335
2336 if (sbi->discard_blks == 0)
2337 break;
2338 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2339 cpc.trim_end = end_segno;
2340 else
2341 cpc.trim_end = min_t(unsigned int,
8412663d 2342 rounddown(cur_segno +
bba681cb
JK		 2343 BATCHED_TRIM_SEGMENTS(sbi),
2344 sbi->segs_per_sec) - 1, end_segno);
2345
2346 mutex_lock(&sbi->gc_mutex);
c34f42e2 2347 err = write_checkpoint(sbi, &cpc);
bba681cb 2348 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 2349 if (err)
2350 break;
74fa5f3d
CY		 2351
2352 schedule();
bba681cb 2353 }
8412663d
CY		 2354
2355 start_block = START_BLOCK(sbi, start_segno);
2356 end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2357
2358 __issue_discard_cmd_range(sbi, start_block, end_block, cpc.trim_minlen);
2359 __wait_discard_cmd_range(sbi, true, start_block, end_block,
2360 cpc.trim_minlen, true);
4b2fecc8 2361out:
f7ef9b83 2362 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 2363 return err;
4b2fecc8
JK		 2364}
2365
351df4b2
JK		 2366static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2367{
2368 struct curseg_info *curseg = CURSEG_I(sbi, type);
2369 if (curseg->next_blkoff < sbi->blocks_per_seg)
2370 return true;
2371 return false;
2372}
2373
81377bd6 2374static int __get_segment_type_2(struct f2fs_io_info *fio)
351df4b2 2375{
81377bd6 2376 if (fio->type == DATA)
351df4b2
JK		 2377 return CURSEG_HOT_DATA;
2378 else
2379 return CURSEG_HOT_NODE;
2380}
2381
81377bd6 2382static int __get_segment_type_4(struct f2fs_io_info *fio)
351df4b2 2383{
81377bd6
JK		 2384 if (fio->type == DATA) {
2385 struct inode *inode = fio->page->mapping->host;
351df4b2
JK		 2386
2387 if (S_ISDIR(inode->i_mode))
2388 return CURSEG_HOT_DATA;
2389 else
2390 return CURSEG_COLD_DATA;
2391 } else {
81377bd6 2392 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
a344b9fd 2393 return CURSEG_WARM_NODE;
351df4b2
JK		 2394 else
2395 return CURSEG_COLD_NODE;
2396 }
2397}
2398
81377bd6 2399static int __get_segment_type_6(struct f2fs_io_info *fio)
351df4b2 2400{
81377bd6
JK		 2401 if (fio->type == DATA) {
2402 struct inode *inode = fio->page->mapping->host;
351df4b2 2403
81377bd6 2404 if (is_cold_data(fio->page) || file_is_cold(inode))
351df4b2 2405 return CURSEG_COLD_DATA;
ef095d19
JK		 2406 if (is_inode_flag_set(inode, FI_HOT_DATA))
2407 return CURSEG_HOT_DATA;
2408 return CURSEG_WARM_DATA;
351df4b2 2409 } else {
81377bd6
JK		 2410 if (IS_DNODE(fio->page))
2411 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
351df4b2 2412 CURSEG_HOT_NODE;
ef095d19 2413 return CURSEG_COLD_NODE;
351df4b2
JK		 2414 }
2415}
2416
81377bd6 2417static int __get_segment_type(struct f2fs_io_info *fio)
351df4b2 2418{
a912b54d
JK		 2419 int type = 0;
2420
81377bd6 2421 switch (fio->sbi->active_logs) {
351df4b2 2422 case 2:
a912b54d
JK		 2423 type = __get_segment_type_2(fio);
2424 break;
351df4b2 2425 case 4:
a912b54d
JK		 2426 type = __get_segment_type_4(fio);
2427 break;
2428 case 6:
2429 type = __get_segment_type_6(fio);
2430 break;
2431 default:
2432 f2fs_bug_on(fio->sbi, true);
351df4b2 2433 }
81377bd6 2434
a912b54d
JK		 2435 if (IS_HOT(type))
2436 fio->temp = HOT;
2437 else if (IS_WARM(type))
2438 fio->temp = WARM;
2439 else
2440 fio->temp = COLD;
2441 return type;
351df4b2
JK		 2442}
2443
bfad7c2d
JK		 2444void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2445 block_t old_blkaddr, block_t *new_blkaddr,
fb830fc5
CY		 2446 struct f2fs_summary *sum, int type,
2447 struct f2fs_io_info *fio, bool add_list)
351df4b2
JK		 2448{
2449 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 2450 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 2451
2452 mutex_lock(&curseg->curseg_mutex);
21cb1d99 2453 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 2454
2455 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 2456
4e6a8d9b
JK		 2457 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2458
351df4b2
JK		 2459 /*
2460 * __add_sum_entry should be resided under the curseg_mutex
2461 * because, this function updates a summary entry in the
2462 * current summary block.
2463 */
e79efe3b 2464 __add_sum_entry(sbi, type, sum);
351df4b2 2465
351df4b2 2466 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 2467
2468 stat_inc_block_count(sbi, curseg);
351df4b2 2469
c6f82fe9
JK		 2470 if (!__has_curseg_space(sbi, type))
2471 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2 2472 /*
c6f82fe9
JK		 2473 * SIT information should be updated after segment allocation,
2474 * since we need to keep dirty segments precisely under SSR.
351df4b2
JK		 2475 */
2476 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 2477
351df4b2
JK		 2478 mutex_unlock(&sit_i->sentry_lock);
2479
704956ec 2480 if (page && IS_NODESEG(type)) {
351df4b2
JK		 2481 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2482
704956ec
CY		 2483 f2fs_inode_chksum_set(sbi, page);
2484 }
2485
fb830fc5
CY		 2486 if (add_list) {
2487 struct f2fs_bio_info *io;
2488
2489 INIT_LIST_HEAD(&fio->list);
2490 fio->in_list = true;
2491 io = sbi->write_io[fio->type] + fio->temp;
2492 spin_lock(&io->io_lock);
2493 list_add_tail(&fio->list, &io->io_list);
2494 spin_unlock(&io->io_lock);
2495 }
2496
bfad7c2d
JK		 2497 mutex_unlock(&curseg->curseg_mutex);
2498}
2499
39d787be
CY		 2500static void update_device_state(struct f2fs_io_info *fio)
2501{
2502 struct f2fs_sb_info *sbi = fio->sbi;
2503 unsigned int devidx;
2504
2505 if (!sbi->s_ndevs)
2506 return;
2507
2508 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2509
2510 /* update device state for fsync */
2511 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
1228b482
CY		 2512
2513 /* update device state for checkpoint */
2514 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2515 spin_lock(&sbi->dev_lock);
2516 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2517 spin_unlock(&sbi->dev_lock);
2518 }
39d787be
CY		 2519}
2520
05ca3632 2521static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 2522{
81377bd6 2523 int type = __get_segment_type(fio);
0a595eba 2524 int err;
bfad7c2d 2525
0a595eba 2526reallocate:
7a9d7548 2527 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
fb830fc5 2528 &fio->new_blkaddr, sum, type, fio, true);
bfad7c2d 2529
351df4b2 2530 /* writeout dirty page into bdev */
b9109b0e 2531 err = f2fs_submit_page_write(fio);
0a595eba
JK		 2532 if (err == -EAGAIN) {
2533 fio->old_blkaddr = fio->new_blkaddr;
2534 goto reallocate;
39d787be
CY		 2535 } else if (!err) {
2536 update_device_state(fio);
0a595eba 2537 }
351df4b2
JK		 2538}
2539
b0af6d49
CY		 2540void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2541 enum iostat_type io_type)
351df4b2 2542{
458e6197 2543 struct f2fs_io_info fio = {
05ca3632 2544 .sbi = sbi,
458e6197 2545 .type = META,
04d328de 2546 .op = REQ_OP_WRITE,
70fd7614 2547 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 2548 .old_blkaddr = page->index,
2549 .new_blkaddr = page->index,
05ca3632 2550 .page = page,
4375a336 2551 .encrypted_page = NULL,
fb830fc5 2552 .in_list = false,
458e6197
JK		 2553 };
2554
2b947003 2555 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 2556 fio.op_flags &= ~REQ_META;
2b947003 2557
351df4b2 2558 set_page_writeback(page);
b9109b0e 2559 f2fs_submit_page_write(&fio);
b0af6d49
CY		 2560
2561 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
351df4b2
JK		 2562}
2563
05ca3632 2564void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 2565{
2566 struct f2fs_summary sum;
05ca3632 2567
351df4b2 2568 set_summary(&sum, nid, 0, 0);
05ca3632 2569 do_write_page(&sum, fio);
b0af6d49
CY		 2570
2571 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2572}
2573
05ca3632 2574void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 2575{
05ca3632 2576 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 2577 struct f2fs_summary sum;
2578 struct node_info ni;
2579
9850cf4a 2580 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 2581 get_node_info(sbi, dn->nid, &ni);
2582 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 2583 do_write_page(&sum, fio);
f28b3434 2584 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
b0af6d49
CY		 2585
2586 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
351df4b2
JK		 2587}
2588
d1b3e72d 2589int rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 2590{
b0af6d49
CY		 2591 int err;
2592
7a9d7548 2593 fio->new_blkaddr = fio->old_blkaddr;
05ca3632 2594 stat_inc_inplace_blocks(fio->sbi);
b0af6d49
CY		 2595
2596 err = f2fs_submit_page_bio(fio);
39d787be
CY		 2597 if (!err)
2598 update_device_state(fio);
b0af6d49
CY		 2599
2600 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2601
2602 return err;
351df4b2
JK		 2603}
2604
4356e48e 2605void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 2606 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 2607 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 2608{
2609 struct sit_info *sit_i = SIT_I(sbi);
2610 struct curseg_info *curseg;
2611 unsigned int segno, old_cursegno;
2612 struct seg_entry *se;
2613 int type;
19f106bc 2614 unsigned short old_blkoff;
351df4b2
JK		 2615
2616 segno = GET_SEGNO(sbi, new_blkaddr);
2617 se = get_seg_entry(sbi, segno);
2618 type = se->type;
2619
19f106bc
CY		 2620 if (!recover_curseg) {
2621 /* for recovery flow */
2622 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2623 if (old_blkaddr == NULL_ADDR)
2624 type = CURSEG_COLD_DATA;
2625 else
2626 type = CURSEG_WARM_DATA;
2627 }
2628 } else {
2629 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 2630 type = CURSEG_WARM_DATA;
2631 }
19f106bc 2632
351df4b2
JK		 2633 curseg = CURSEG_I(sbi, type);
2634
2635 mutex_lock(&curseg->curseg_mutex);
2636 mutex_lock(&sit_i->sentry_lock);
2637
2638 old_cursegno = curseg->segno;
19f106bc 2639 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2640
2641 /* change the current segment */
2642 if (segno != curseg->segno) {
2643 curseg->next_segno = segno;
025d63a4 2644 change_curseg(sbi, type);
351df4b2
JK		 2645 }
2646
491c0854 2647 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2648 __add_sum_entry(sbi, type, sum);
351df4b2 2649
28bc106b 2650 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2651 update_sit_entry(sbi, new_blkaddr, 1);
2652 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2653 update_sit_entry(sbi, old_blkaddr, -1);
2654
2655 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2656 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2657
351df4b2 2658 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2659
19f106bc
CY		 2660 if (recover_curseg) {
2661 if (old_cursegno != curseg->segno) {
2662 curseg->next_segno = old_cursegno;
025d63a4 2663 change_curseg(sbi, type);
19f106bc
CY		 2664 }
2665 curseg->next_blkoff = old_blkoff;
2666 }
2667
351df4b2
JK		 2668 mutex_unlock(&sit_i->sentry_lock);
2669 mutex_unlock(&curseg->curseg_mutex);
2670}
2671
528e3459
CY		 2672void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2673 block_t old_addr, block_t new_addr,
28bc106b
CY		 2674 unsigned char version, bool recover_curseg,
2675 bool recover_newaddr)
528e3459
CY		 2676{
2677 struct f2fs_summary sum;
2678
2679 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2680
28bc106b
CY		 2681 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2682 recover_curseg, recover_newaddr);
528e3459 2683
f28b3434 2684 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2685}
2686
93dfe2ac 2687void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2688 enum page_type type, bool ordered)
93dfe2ac 2689{
93dfe2ac 2690 if (PageWriteback(page)) {
4081363f
JK		 2691 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2692
b9109b0e
JK		 2693 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2694 0, page->index, type);
fec1d657
JK		 2695 if (ordered)
2696 wait_on_page_writeback(page);
2697 else
2698 wait_for_stable_page(page);
93dfe2ac
JK		 2699 }
2700}
2701
d4c759ee 2702void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
08b39fbd
CY		 2703{
2704 struct page *cpage;
2705
5d4c0af4 2706 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2707 return;
2708
08b39fbd
CY		 2709 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2710 if (cpage) {
fec1d657 2711 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2712 f2fs_put_page(cpage, 1);
2713 }
2714}
2715
351df4b2
JK		 2716static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2717{
2718 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2719 struct curseg_info *seg_i;
2720 unsigned char *kaddr;
2721 struct page *page;
2722 block_t start;
2723 int i, j, offset;
2724
2725 start = start_sum_block(sbi);
2726
2727 page = get_meta_page(sbi, start++);
2728 kaddr = (unsigned char *)page_address(page);
2729
2730 /* Step 1: restore nat cache */
2731 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2732 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2733
2734 /* Step 2: restore sit cache */
2735 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2736 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2737 offset = 2 * SUM_JOURNAL_SIZE;
2738
2739 /* Step 3: restore summary entries */
2740 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2741 unsigned short blk_off;
2742 unsigned int segno;
2743
2744 seg_i = CURSEG_I(sbi, i);
2745 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2746 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2747 seg_i->next_segno = segno;
2748 reset_curseg(sbi, i, 0);
2749 seg_i->alloc_type = ckpt->alloc_type[i];
2750 seg_i->next_blkoff = blk_off;
2751
2752 if (seg_i->alloc_type == SSR)
2753 blk_off = sbi->blocks_per_seg;
2754
2755 for (j = 0; j < blk_off; j++) {
2756 struct f2fs_summary *s;
2757 s = (struct f2fs_summary *)(kaddr + offset);
2758 seg_i->sum_blk->entries[j] = *s;
2759 offset += SUMMARY_SIZE;
09cbfeaf 2760 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2761 SUM_FOOTER_SIZE)
2762 continue;
2763
2764 f2fs_put_page(page, 1);
2765 page = NULL;
2766
2767 page = get_meta_page(sbi, start++);
2768 kaddr = (unsigned char *)page_address(page);
2769 offset = 0;
2770 }
2771 }
2772 f2fs_put_page(page, 1);
2773 return 0;
2774}
2775
2776static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2777{
2778 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2779 struct f2fs_summary_block *sum;
2780 struct curseg_info *curseg;
2781 struct page *new;
2782 unsigned short blk_off;
2783 unsigned int segno = 0;
2784 block_t blk_addr = 0;
2785
2786 /* get segment number and block addr */
2787 if (IS_DATASEG(type)) {
2788 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2789 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2790 CURSEG_HOT_DATA]);
119ee914 2791 if (__exist_node_summaries(sbi))
351df4b2
JK		 2792 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2793 else
2794 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2795 } else {
2796 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2797 CURSEG_HOT_NODE]);
2798 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2799 CURSEG_HOT_NODE]);
119ee914 2800 if (__exist_node_summaries(sbi))
351df4b2
JK		 2801 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2802 type - CURSEG_HOT_NODE);
2803 else
2804 blk_addr = GET_SUM_BLOCK(sbi, segno);
2805 }
2806
2807 new = get_meta_page(sbi, blk_addr);
2808 sum = (struct f2fs_summary_block *)page_address(new);
2809
2810 if (IS_NODESEG(type)) {
119ee914 2811 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2812 struct f2fs_summary *ns = &sum->entries[0];
2813 int i;
2814 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2815 ns->version = 0;
2816 ns->ofs_in_node = 0;
2817 }
2818 } else {
d653788a
GZ		 2819 int err;
2820
2821 err = restore_node_summary(sbi, segno, sum);
2822 if (err) {
351df4b2 2823 f2fs_put_page(new, 1);
d653788a 2824 return err;
351df4b2
JK		 2825 }
2826 }
2827 }
2828
2829 /* set uncompleted segment to curseg */
2830 curseg = CURSEG_I(sbi, type);
2831 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2832
2833 /* update journal info */
2834 down_write(&curseg->journal_rwsem);
2835 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2836 up_write(&curseg->journal_rwsem);
2837
2838 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2839 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2840 curseg->next_segno = segno;
2841 reset_curseg(sbi, type, 0);
2842 curseg->alloc_type = ckpt->alloc_type[type];
2843 curseg->next_blkoff = blk_off;
2844 mutex_unlock(&curseg->curseg_mutex);
2845 f2fs_put_page(new, 1);
2846 return 0;
2847}
2848
2849static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2850{
21d3f8e1
JQ		 2851 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2852 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
351df4b2 2853 int type = CURSEG_HOT_DATA;
e4fc5fbf 2854 int err;
351df4b2 2855
aaec2b1d 2856 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2857 int npages = npages_for_summary_flush(sbi, true);
2858
2859 if (npages >= 2)
2860 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2861 META_CP, true);
3fa06d7b 2862
351df4b2
JK		 2863 /* restore for compacted data summary */
2864 if (read_compacted_summaries(sbi))
2865 return -EINVAL;
2866 type = CURSEG_HOT_NODE;
2867 }
2868
119ee914 2869 if (__exist_node_summaries(sbi))
3fa06d7b 2870 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2871 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2872
e4fc5fbf
CY		 2873 for (; type <= CURSEG_COLD_NODE; type++) {
2874 err = read_normal_summaries(sbi, type);
2875 if (err)
2876 return err;
2877 }
2878
21d3f8e1
JQ		 2879 /* sanity check for summary blocks */
2880 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2881 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2882 return -EINVAL;
2883
351df4b2
JK		 2884 return 0;
2885}
2886
2887static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2888{
2889 struct page *page;
2890 unsigned char *kaddr;
2891 struct f2fs_summary *summary;
2892 struct curseg_info *seg_i;
2893 int written_size = 0;
2894 int i, j;
2895
2896 page = grab_meta_page(sbi, blkaddr++);
2897 kaddr = (unsigned char *)page_address(page);
2898
2899 /* Step 1: write nat cache */
2900 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2901 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2902 written_size += SUM_JOURNAL_SIZE;
2903
2904 /* Step 2: write sit cache */
2905 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2906 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2907 written_size += SUM_JOURNAL_SIZE;
2908
351df4b2
JK		 2909 /* Step 3: write summary entries */
2910 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2911 unsigned short blkoff;
2912 seg_i = CURSEG_I(sbi, i);
2913 if (sbi->ckpt->alloc_type[i] == SSR)
2914 blkoff = sbi->blocks_per_seg;
2915 else
2916 blkoff = curseg_blkoff(sbi, i);
2917
2918 for (j = 0; j < blkoff; j++) {
2919 if (!page) {
2920 page = grab_meta_page(sbi, blkaddr++);
2921 kaddr = (unsigned char *)page_address(page);
2922 written_size = 0;
2923 }
2924 summary = (struct f2fs_summary *)(kaddr + written_size);
2925 *summary = seg_i->sum_blk->entries[j];
2926 written_size += SUMMARY_SIZE;
351df4b2 2927
09cbfeaf 2928 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2929 SUM_FOOTER_SIZE)
2930 continue;
2931
e8d61a74 2932 set_page_dirty(page);
351df4b2
JK		 2933 f2fs_put_page(page, 1);
2934 page = NULL;
2935 }
2936 }
e8d61a74
CY		 2937 if (page) {
2938 set_page_dirty(page);
351df4b2 2939 f2fs_put_page(page, 1);
e8d61a74 2940 }
351df4b2
JK		 2941}
2942
2943static void write_normal_summaries(struct f2fs_sb_info *sbi,
2944 block_t blkaddr, int type)
2945{
2946 int i, end;
2947 if (IS_DATASEG(type))
2948 end = type + NR_CURSEG_DATA_TYPE;
2949 else
2950 end = type + NR_CURSEG_NODE_TYPE;
2951
b7ad7512
CY		 2952 for (i = type; i < end; i++)
2953 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2954}
2955
2956void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2957{
aaec2b1d 2958 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2959 write_compacted_summaries(sbi, start_blk);
2960 else
2961 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2962}
2963
2964void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2965{
119ee914 2966 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2967}
2968
dfc08a12 2969int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2970 unsigned int val, int alloc)
2971{
2972 int i;
2973
2974 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2975 for (i = 0; i < nats_in_cursum(journal); i++) {
2976 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2977 return i;
2978 }
dfc08a12
CY		 2979 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2980 return update_nats_in_cursum(journal, 1);
351df4b2 2981 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2982 for (i = 0; i < sits_in_cursum(journal); i++)
2983 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2984 return i;
dfc08a12
CY		 2985 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2986 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2987 }
2988 return -1;
2989}
2990
2991static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2992 unsigned int segno)
2993{
2cc22186 2994 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2995}
2996
2997static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2998 unsigned int start)
2999{
3000 struct sit_info *sit_i = SIT_I(sbi);
3001 struct page *src_page, *dst_page;
3002 pgoff_t src_off, dst_off;
3003 void *src_addr, *dst_addr;
3004
3005 src_off = current_sit_addr(sbi, start);
3006 dst_off = next_sit_addr(sbi, src_off);
3007
3008 /* get current sit block page without lock */
3009 src_page = get_meta_page(sbi, src_off);
3010 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 3011 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 3012
3013 src_addr = page_address(src_page);
3014 dst_addr = page_address(dst_page);
09cbfeaf 3015 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 3016
3017 set_page_dirty(dst_page);
3018 f2fs_put_page(src_page, 1);
3019
3020 set_to_next_sit(sit_i, start);
3021
3022 return dst_page;
3023}
3024
184a5cd2
CY		 3025static struct sit_entry_set *grab_sit_entry_set(void)
3026{
3027 struct sit_entry_set *ses =
80c54505 3028 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 3029
3030 ses->entry_cnt = 0;
3031 INIT_LIST_HEAD(&ses->set_list);
3032 return ses;
3033}
3034
3035static void release_sit_entry_set(struct sit_entry_set *ses)
3036{
3037 list_del(&ses->set_list);
3038 kmem_cache_free(sit_entry_set_slab, ses);
3039}
3040
3041static void adjust_sit_entry_set(struct sit_entry_set *ses,
3042 struct list_head *head)
3043{
3044 struct sit_entry_set *next = ses;
3045
3046 if (list_is_last(&ses->set_list, head))
3047 return;
3048
3049 list_for_each_entry_continue(next, head, set_list)
3050 if (ses->entry_cnt <= next->entry_cnt)
3051 break;
3052
3053 list_move_tail(&ses->set_list, &next->set_list);
3054}
3055
3056static void add_sit_entry(unsigned int segno, struct list_head *head)
3057{
3058 struct sit_entry_set *ses;
3059 unsigned int start_segno = START_SEGNO(segno);
3060
3061 list_for_each_entry(ses, head, set_list) {
3062 if (ses->start_segno == start_segno) {
3063 ses->entry_cnt++;
3064 adjust_sit_entry_set(ses, head);
3065 return;
3066 }
3067 }
3068
3069 ses = grab_sit_entry_set();
3070
3071 ses->start_segno = start_segno;
3072 ses->entry_cnt++;
3073 list_add(&ses->set_list, head);
3074}
3075
3076static void add_sits_in_set(struct f2fs_sb_info *sbi)
3077{
3078 struct f2fs_sm_info *sm_info = SM_I(sbi);
3079 struct list_head *set_list = &sm_info->sit_entry_set;
3080 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 3081 unsigned int segno;
3082
7cd8558b 3083 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 3084 add_sit_entry(segno, set_list);
3085}
3086
3087static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 3088{
3089 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3090 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 3091 int i;
3092
b7ad7512 3093 down_write(&curseg->journal_rwsem);
dfc08a12 3094 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 3095 unsigned int segno;
3096 bool dirtied;
3097
dfc08a12 3098 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 3099 dirtied = __mark_sit_entry_dirty(sbi, segno);
3100
3101 if (!dirtied)
3102 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 3103 }
dfc08a12 3104 update_sits_in_cursum(journal, -i);
b7ad7512 3105 up_write(&curseg->journal_rwsem);
351df4b2
JK		 3106}
3107
0a8165d7 3108/*
351df4b2
JK		 3109 * CP calls this function, which flushes SIT entries including sit_journal,
3110 * and moves prefree segs to free segs.
3111 */
4b2fecc8 3112void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 3113{
3114 struct sit_info *sit_i = SIT_I(sbi);
3115 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3116 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3117 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 3118 struct sit_entry_set *ses, *tmp;
3119 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 3120 bool to_journal = true;
4b2fecc8 3121 struct seg_entry *se;
351df4b2 3122
351df4b2
JK		 3123 mutex_lock(&sit_i->sentry_lock);
3124
2b11a74b
WL		 3125 if (!sit_i->dirty_sentries)
3126 goto out;
3127
351df4b2 3128 /*
184a5cd2
CY		 3129 * add and account sit entries of dirty bitmap in sit entry
3130 * set temporarily
351df4b2 3131 */
184a5cd2 3132 add_sits_in_set(sbi);
351df4b2 3133
184a5cd2
CY		 3134 /*
3135 * if there are no enough space in journal to store dirty sit
3136 * entries, remove all entries from journal and add and account
3137 * them in sit entry set.
3138 */
dfc08a12 3139 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 3140 remove_sits_in_journal(sbi);
b2955550 3141
184a5cd2
CY		 3142 /*
3143 * there are two steps to flush sit entries:
3144 * #1, flush sit entries to journal in current cold data summary block.
3145 * #2, flush sit entries to sit page.
3146 */
3147 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 3148 struct page *page = NULL;
184a5cd2
CY		 3149 struct f2fs_sit_block *raw_sit = NULL;
3150 unsigned int start_segno = ses->start_segno;
3151 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 3152 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 3153 unsigned int segno = start_segno;
3154
3155 if (to_journal &&
dfc08a12 3156 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 3157 to_journal = false;
3158
b7ad7512
CY		 3159 if (to_journal) {
3160 down_write(&curseg->journal_rwsem);
3161 } else {
184a5cd2
CY		 3162 page = get_next_sit_page(sbi, start_segno);
3163 raw_sit = page_address(page);
351df4b2 3164 }
351df4b2 3165
184a5cd2
CY		 3166 /* flush dirty sit entries in region of current sit set */
3167 for_each_set_bit_from(segno, bitmap, end) {
3168 int offset, sit_offset;
4b2fecc8
JK		 3169
3170 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 3171
3172 /* add discard candidates */
c473f1a9 3173 if (!(cpc->reason & CP_DISCARD)) {
4b2fecc8 3174 cpc->trim_start = segno;
25290fa5 3175 add_discard_addrs(sbi, cpc, false);
4b2fecc8 3176 }
184a5cd2
CY		 3177
3178 if (to_journal) {
dfc08a12 3179 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 3180 SIT_JOURNAL, segno, 1);
3181 f2fs_bug_on(sbi, offset < 0);
dfc08a12 3182 segno_in_journal(journal, offset) =
184a5cd2
CY		 3183 cpu_to_le32(segno);
3184 seg_info_to_raw_sit(se,
dfc08a12 3185 &sit_in_journal(journal, offset));
184a5cd2
CY		 3186 } else {
3187 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3188 seg_info_to_raw_sit(se,
3189 &raw_sit->entries[sit_offset]);
3190 }
351df4b2 3191
184a5cd2
CY		 3192 __clear_bit(segno, bitmap);
3193 sit_i->dirty_sentries--;
3194 ses->entry_cnt--;
351df4b2
JK		 3195 }
3196
b7ad7512
CY		 3197 if (to_journal)
3198 up_write(&curseg->journal_rwsem);
3199 else
184a5cd2
CY		 3200 f2fs_put_page(page, 1);
3201
3202 f2fs_bug_on(sbi, ses->entry_cnt);
3203 release_sit_entry_set(ses);
351df4b2 3204 }
184a5cd2
CY		 3205
3206 f2fs_bug_on(sbi, !list_empty(head));
3207 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 3208out:
c473f1a9 3209 if (cpc->reason & CP_DISCARD) {
650d3c4e
YH		 3210 __u64 trim_start = cpc->trim_start;
3211
4b2fecc8 3212 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 3213 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 3214
3215 cpc->trim_start = trim_start;
4b2fecc8 3216 }
351df4b2 3217 mutex_unlock(&sit_i->sentry_lock);
351df4b2 3218
351df4b2
JK		 3219 set_prefree_as_free_segments(sbi);
3220}
3221
3222static int build_sit_info(struct f2fs_sb_info *sbi)
3223{
3224 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 3225 struct sit_info *sit_i;
3226 unsigned int sit_segs, start;
ae27d62e 3227 char *src_bitmap;
351df4b2
JK		 3228 unsigned int bitmap_size;
3229
3230 /* allocate memory for SIT information */
3231 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3232 if (!sit_i)
3233 return -ENOMEM;
3234
3235 SM_I(sbi)->sit_info = sit_i;
3236
a7c3e901 3237 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
39307a8e 3238 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 3239 if (!sit_i->sentries)
3240 return -ENOMEM;
3241
7cd8558b 3242 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3243 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3244 if (!sit_i->dirty_sentries_bitmap)
3245 return -ENOMEM;
3246
7cd8558b 3247 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3248 sit_i->sentries[start].cur_valid_map
3249 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3250 sit_i->sentries[start].ckpt_valid_map
3251 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 3252 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 3253 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 3254 return -ENOMEM;
3e025740 3255
355e7891
CY		 3256#ifdef CONFIG_F2FS_CHECK_FS
3257 sit_i->sentries[start].cur_valid_map_mir
3258 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3259 if (!sit_i->sentries[start].cur_valid_map_mir)
3260 return -ENOMEM;
3261#endif
3262
3e025740
JK		 3263 if (f2fs_discard_en(sbi)) {
3264 sit_i->sentries[start].discard_map
3265 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3266 if (!sit_i->sentries[start].discard_map)
3267 return -ENOMEM;
3268 }
351df4b2
JK		 3269 }
3270
60a3b782
JK		 3271 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3272 if (!sit_i->tmp_map)
3273 return -ENOMEM;
3274
351df4b2 3275 if (sbi->segs_per_sec > 1) {
a7c3e901 3276 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
39307a8e 3277 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 3278 if (!sit_i->sec_entries)
3279 return -ENOMEM;
3280 }
3281
3282 /* get information related with SIT */
3283 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3284
3285 /* setup SIT bitmap from ckeckpoint pack */
3286 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3287 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3288
ae27d62e
CY		 3289 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3290 if (!sit_i->sit_bitmap)
351df4b2 3291 return -ENOMEM;
351df4b2 3292
ae27d62e
CY		 3293#ifdef CONFIG_F2FS_CHECK_FS
3294 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3295 if (!sit_i->sit_bitmap_mir)
3296 return -ENOMEM;
3297#endif
3298
351df4b2
JK		 3299 /* init SIT information */
3300 sit_i->s_ops = &default_salloc_ops;
3301
3302 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3303 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 3304 sit_i->written_valid_blocks = 0;
351df4b2
JK		 3305 sit_i->bitmap_size = bitmap_size;
3306 sit_i->dirty_sentries = 0;
3307 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3308 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
48fbfe50 3309 sit_i->mounted_time = ktime_get_real_seconds();
351df4b2
JK		 3310 mutex_init(&sit_i->sentry_lock);
3311 return 0;
3312}
3313
3314static int build_free_segmap(struct f2fs_sb_info *sbi)
3315{
351df4b2
JK		 3316 struct free_segmap_info *free_i;
3317 unsigned int bitmap_size, sec_bitmap_size;
3318
3319 /* allocate memory for free segmap information */
3320 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3321 if (!free_i)
3322 return -ENOMEM;
3323
3324 SM_I(sbi)->free_info = free_i;
3325
7cd8558b 3326 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
a7c3e901 3327 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3328 if (!free_i->free_segmap)
3329 return -ENOMEM;
3330
7cd8558b 3331 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
a7c3e901 3332 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 3333 if (!free_i->free_secmap)
3334 return -ENOMEM;
3335
3336 /* set all segments as dirty temporarily */
3337 memset(free_i->free_segmap, 0xff, bitmap_size);
3338 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3339
3340 /* init free segmap information */
7cd8558b 3341 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 3342 free_i->free_segments = 0;
3343 free_i->free_sections = 0;
1a118ccf 3344 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 3345 return 0;
3346}
3347
3348static int build_curseg(struct f2fs_sb_info *sbi)
3349{
1042d60f 3350 struct curseg_info *array;
351df4b2
JK		 3351 int i;
3352
b434babf 3353 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 3354 if (!array)
3355 return -ENOMEM;
3356
3357 SM_I(sbi)->curseg_array = array;
3358
3359 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3360 mutex_init(&array[i].curseg_mutex);
09cbfeaf 3361 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 3362 if (!array[i].sum_blk)
3363 return -ENOMEM;
b7ad7512
CY		 3364 init_rwsem(&array[i].journal_rwsem);
3365 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3366 GFP_KERNEL);
3367 if (!array[i].journal)
3368 return -ENOMEM;
351df4b2
JK		 3369 array[i].segno = NULL_SEGNO;
3370 array[i].next_blkoff = 0;
3371 }
3372 return restore_curseg_summaries(sbi);
3373}
3374
3375static void build_sit_entries(struct f2fs_sb_info *sbi)
3376{
3377 struct sit_info *sit_i = SIT_I(sbi);
3378 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 3379 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 3380 struct seg_entry *se;
3381 struct f2fs_sit_entry sit;
74de593a
CY		 3382 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3383 unsigned int i, start, end;
3384 unsigned int readed, start_blk = 0;
351df4b2 3385
74de593a 3386 do {
664ba972
JK		 3387 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3388 META_SIT, true);
74de593a
CY		 3389
3390 start = start_blk * sit_i->sents_per_block;
3391 end = (start_blk + readed) * sit_i->sents_per_block;
3392
7cd8558b 3393 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 3394 struct f2fs_sit_block *sit_blk;
74de593a
CY		 3395 struct page *page;
3396
9c094040 3397 se = &sit_i->sentries[start];
74de593a
CY		 3398 page = get_current_sit_page(sbi, start);
3399 sit_blk = (struct f2fs_sit_block *)page_address(page);
3400 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3401 f2fs_put_page(page, 1);
d600af23 3402
74de593a
CY		 3403 check_block_count(sbi, start, &sit);
3404 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 3405
3406 /* build discard map only one time */
3e025740 3407 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3408 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3409 memset(se->discard_map, 0xff,
3410 SIT_VBLOCK_MAP_SIZE);
3411 } else {
3412 memcpy(se->discard_map,
3413 se->cur_valid_map,
3414 SIT_VBLOCK_MAP_SIZE);
3415 sbi->discard_blks +=
3416 sbi->blocks_per_seg -
3417 se->valid_blocks;
3418 }
3e025740 3419 }
a66cdd98 3420
d600af23
CY		 3421 if (sbi->segs_per_sec > 1)
3422 get_sec_entry(sbi, start)->valid_blocks +=
3423 se->valid_blocks;
351df4b2 3424 }
74de593a
CY		 3425 start_blk += readed;
3426 } while (start_blk < sit_blk_cnt);
d600af23
CY		 3427
3428 down_read(&curseg->journal_rwsem);
3429 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 3430 unsigned int old_valid_blocks;
3431
3432 start = le32_to_cpu(segno_in_journal(journal, i));
3433 se = &sit_i->sentries[start];
3434 sit = sit_in_journal(journal, i);
3435
3436 old_valid_blocks = se->valid_blocks;
3437
3438 check_block_count(sbi, start, &sit);
3439 seg_info_from_raw_sit(se, &sit);
3440
3441 if (f2fs_discard_en(sbi)) {
1f43e2ad
CY		 3442 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3443 memset(se->discard_map, 0xff,
3444 SIT_VBLOCK_MAP_SIZE);
3445 } else {
3446 memcpy(se->discard_map, se->cur_valid_map,
3447 SIT_VBLOCK_MAP_SIZE);
3448 sbi->discard_blks += old_valid_blocks -
3449 se->valid_blocks;
3450 }
d600af23
CY		 3451 }
3452
3453 if (sbi->segs_per_sec > 1)
3454 get_sec_entry(sbi, start)->valid_blocks +=
3455 se->valid_blocks - old_valid_blocks;
3456 }
3457 up_read(&curseg->journal_rwsem);
351df4b2
JK		 3458}
3459
3460static void init_free_segmap(struct f2fs_sb_info *sbi)
3461{
3462 unsigned int start;
3463 int type;
3464
7cd8558b 3465 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 3466 struct seg_entry *sentry = get_seg_entry(sbi, start);
3467 if (!sentry->valid_blocks)
3468 __set_free(sbi, start);
c79b7ff1
JK		 3469 else
3470 SIT_I(sbi)->written_valid_blocks +=
3471 sentry->valid_blocks;
351df4b2
JK		 3472 }
3473
3474 /* set use the current segments */
3475 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3476 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3477 __set_test_and_inuse(sbi, curseg_t->segno);
3478 }
3479}
3480
3481static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3482{
3483 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3484 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 3485 unsigned int segno = 0, offset = 0;
351df4b2
JK		 3486 unsigned short valid_blocks;
3487
8736fbf0 3488 while (1) {
351df4b2 3489 /* find dirty segment based on free segmap */
7cd8558b
JK		 3490 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3491 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 3492 break;
3493 offset = segno + 1;
302bd348 3494 valid_blocks = get_valid_blocks(sbi, segno, false);
ec325b52 3495 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 3496 continue;
ec325b52
JK		 3497 if (valid_blocks > sbi->blocks_per_seg) {
3498 f2fs_bug_on(sbi, 1);
3499 continue;
3500 }
351df4b2
JK		 3501 mutex_lock(&dirty_i->seglist_lock);
3502 __locate_dirty_segment(sbi, segno, DIRTY);
3503 mutex_unlock(&dirty_i->seglist_lock);
3504 }
3505}
3506
5ec4e49f 3507static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3508{
3509 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 3510 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 3511
a7c3e901 3512 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 3513 if (!dirty_i->victim_secmap)
351df4b2
JK		 3514 return -ENOMEM;
3515 return 0;
3516}
3517
3518static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3519{
3520 struct dirty_seglist_info *dirty_i;
3521 unsigned int bitmap_size, i;
3522
3523 /* allocate memory for dirty segments list information */
3524 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3525 if (!dirty_i)
3526 return -ENOMEM;
3527
3528 SM_I(sbi)->dirty_info = dirty_i;
3529 mutex_init(&dirty_i->seglist_lock);
3530
7cd8558b 3531 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 3532
3533 for (i = 0; i < NR_DIRTY_TYPE; i++) {
a7c3e901 3534 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 3535 if (!dirty_i->dirty_segmap[i])
3536 return -ENOMEM;
3537 }
3538
3539 init_dirty_segmap(sbi);
5ec4e49f 3540 return init_victim_secmap(sbi);
351df4b2
JK		 3541}
3542
0a8165d7 3543/*
351df4b2
JK		 3544 * Update min, max modified time for cost-benefit GC algorithm
3545 */
3546static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3547{
3548 struct sit_info *sit_i = SIT_I(sbi);
3549 unsigned int segno;
3550
3551 mutex_lock(&sit_i->sentry_lock);
3552
3553 sit_i->min_mtime = LLONG_MAX;
3554
7cd8558b 3555 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 3556 unsigned int i;
3557 unsigned long long mtime = 0;
3558
3559 for (i = 0; i < sbi->segs_per_sec; i++)
3560 mtime += get_seg_entry(sbi, segno + i)->mtime;
3561
3562 mtime = div_u64(mtime, sbi->segs_per_sec);
3563
3564 if (sit_i->min_mtime > mtime)
3565 sit_i->min_mtime = mtime;
3566 }
3567 sit_i->max_mtime = get_mtime(sbi);
3568 mutex_unlock(&sit_i->sentry_lock);
3569}
3570
3571int build_segment_manager(struct f2fs_sb_info *sbi)
3572{
3573 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3574 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 3575 struct f2fs_sm_info *sm_info;
351df4b2
JK		 3576 int err;
3577
3578 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3579 if (!sm_info)
3580 return -ENOMEM;
3581
3582 /* init sm info */
3583 sbi->sm_info = sm_info;
351df4b2
JK		 3584 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3585 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3586 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3587 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3588 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3589 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3590 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 3591 sm_info->rec_prefree_segments = sm_info->main_segments *
3592 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 3593 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3594 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3595
52763a4b
JK		 3596 if (!test_opt(sbi, LFS))
3597 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 3598 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 3599 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
ef095d19 3600 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
351df4b2 3601
bba681cb
JK		 3602 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3603
184a5cd2
CY		 3604 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3605
d4fdf8ba 3606 if (!f2fs_readonly(sbi->sb)) {
2163d198
GZ		 3607 err = create_flush_cmd_control(sbi);
3608 if (err)
a688b9d9 3609 return err;
6b4afdd7
JK		 3610 }
3611
0b54fb84
JK		 3612 err = create_discard_cmd_control(sbi);
3613 if (err)
3614 return err;
3615
351df4b2
JK		 3616 err = build_sit_info(sbi);
3617 if (err)
3618 return err;
3619 err = build_free_segmap(sbi);
3620 if (err)
3621 return err;
3622 err = build_curseg(sbi);
3623 if (err)
3624 return err;
3625
3626 /* reinit free segmap based on SIT */
3627 build_sit_entries(sbi);
3628
3629 init_free_segmap(sbi);
3630 err = build_dirty_segmap(sbi);
3631 if (err)
3632 return err;
3633
3634 init_min_max_mtime(sbi);
3635 return 0;
3636}
3637
3638static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3639 enum dirty_type dirty_type)
3640{
3641 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3642
3643 mutex_lock(&dirty_i->seglist_lock);
39307a8e 3644 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 3645 dirty_i->nr_dirty[dirty_type] = 0;
3646 mutex_unlock(&dirty_i->seglist_lock);
3647}
3648
5ec4e49f 3649static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 3650{
3651 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 3652 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 3653}
3654
3655static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3656{
3657 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3658 int i;
3659
3660 if (!dirty_i)
3661 return;
3662
3663 /* discard pre-free/dirty segments list */
3664 for (i = 0; i < NR_DIRTY_TYPE; i++)
3665 discard_dirty_segmap(sbi, i);
3666
5ec4e49f 3667 destroy_victim_secmap(sbi);
351df4b2
JK		 3668 SM_I(sbi)->dirty_info = NULL;
3669 kfree(dirty_i);
3670}
3671
3672static void destroy_curseg(struct f2fs_sb_info *sbi)
3673{
3674 struct curseg_info *array = SM_I(sbi)->curseg_array;
3675 int i;
3676
3677 if (!array)
3678 return;
3679 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3680 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3681 kfree(array[i].sum_blk);
b7ad7512
CY		 3682 kfree(array[i].journal);
3683 }
351df4b2
JK		 3684 kfree(array);
3685}
3686
3687static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3688{
3689 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3690 if (!free_i)
3691 return;
3692 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3693 kvfree(free_i->free_segmap);
3694 kvfree(free_i->free_secmap);
351df4b2
JK		 3695 kfree(free_i);
3696}
3697
3698static void destroy_sit_info(struct f2fs_sb_info *sbi)
3699{
3700 struct sit_info *sit_i = SIT_I(sbi);
3701 unsigned int start;
3702
3703 if (!sit_i)
3704 return;
3705
3706 if (sit_i->sentries) {
7cd8558b 3707 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3708 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3709#ifdef CONFIG_F2FS_CHECK_FS
3710 kfree(sit_i->sentries[start].cur_valid_map_mir);
3711#endif
351df4b2 3712 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3713 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3714 }
3715 }
60a3b782
JK		 3716 kfree(sit_i->tmp_map);
3717
39307a8e
JK		 3718 kvfree(sit_i->sentries);
3719 kvfree(sit_i->sec_entries);
3720 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3721
3722 SM_I(sbi)->sit_info = NULL;
3723 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3724#ifdef CONFIG_F2FS_CHECK_FS
3725 kfree(sit_i->sit_bitmap_mir);
3726#endif
351df4b2
JK		 3727 kfree(sit_i);
3728}
3729
3730void destroy_segment_manager(struct f2fs_sb_info *sbi)
3731{
3732 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3733
3b03f724
CY		 3734 if (!sm_info)
3735 return;
5eba8c5d 3736 destroy_flush_cmd_control(sbi, true);
f099405f 3737 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3738 destroy_dirty_segmap(sbi);
3739 destroy_curseg(sbi);
3740 destroy_free_segmap(sbi);
3741 destroy_sit_info(sbi);
3742 sbi->sm_info = NULL;
3743 kfree(sm_info);
3744}
7fd9e544
JK		 3745
3746int __init create_segment_manager_caches(void)
3747{
3748 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3749 sizeof(struct discard_entry));
7fd9e544 3750 if (!discard_entry_slab)
184a5cd2
CY		 3751 goto fail;
3752
b01a9201
JK		 3753 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3754 sizeof(struct discard_cmd));
3755 if (!discard_cmd_slab)
6ab2a308 3756 goto destroy_discard_entry;
275b66b0 3757
184a5cd2 3758 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3759 sizeof(struct sit_entry_set));
184a5cd2 3760 if (!sit_entry_set_slab)
b01a9201 3761 goto destroy_discard_cmd;
88b88a66
JK		 3762
3763 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3764 sizeof(struct inmem_pages));
3765 if (!inmem_entry_slab)
3766 goto destroy_sit_entry_set;
7fd9e544 3767 return 0;
184a5cd2 3768
88b88a66
JK		 3769destroy_sit_entry_set:
3770 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3771destroy_discard_cmd:
3772 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3773destroy_discard_entry:
184a5cd2
CY		 3774 kmem_cache_destroy(discard_entry_slab);
3775fail:
3776 return -ENOMEM;
7fd9e544
JK		 3777}
3778
3779void destroy_segment_manager_caches(void)
3780{
184a5cd2 3781 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3782 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3783 kmem_cache_destroy(discard_entry_slab);
88b88a66 3784 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3785}
Ubuntu Jammy Linux kernel mirror