projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: fix null pointer dereference when issuing flush in ->fsync
[linux-2.6-block.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
74de593a 17#include <linux/swap.h>
60b99b48 18#include <linux/timer.h>
351df4b2
JK		 19
20#include "f2fs.h"
21#include "segment.h"
22#include "node.h"
9e4ded3f 23#include "trace.h"
6ec178da 24#include <trace/events/f2fs.h>
351df4b2 25
9a7f143a
CL		 26#define __reverse_ffz(x) __reverse_ffs(~(x))
27
7fd9e544 28static struct kmem_cache *discard_entry_slab;
b01a9201 29static struct kmem_cache *discard_cmd_slab;
184a5cd2 30static struct kmem_cache *sit_entry_set_slab;
88b88a66 31static struct kmem_cache *inmem_entry_slab;
7fd9e544 32
f96999c3
JK		 33static unsigned long __reverse_ulong(unsigned char *str)
34{
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38#if BITS_PER_LONG == 64
39 shift = 56;
40#endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46}
47
9a7f143a
CL		 48/*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52static inline unsigned long __reverse_ffs(unsigned long word)
53{
54 int num = 0;
55
56#if BITS_PER_LONG == 64
f96999c3 57 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 58 num += 32;
f96999c3 59 else
9a7f143a 60 word >>= 32;
9a7f143a 61#endif
f96999c3 62 if ((word & 0xffff0000) == 0)
9a7f143a 63 num += 16;
f96999c3 64 else
9a7f143a 65 word >>= 16;
f96999c3
JK		 66
67 if ((word & 0xff00) == 0)
9a7f143a 68 num += 8;
f96999c3 69 else
9a7f143a 70 word >>= 8;
f96999c3 71
9a7f143a
CL		 72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
f96999c3 76
9a7f143a
CL		 77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
f96999c3 81
9a7f143a
CL		 82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85}
86
87/*
e1c42045 88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 90 * @size must be integral times of unsigned long.
9a7f143a 91 * Example:
f96999c3
JK		 92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 95 */
96static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98{
99 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 100 unsigned long result = size;
9a7f143a 101 unsigned long tmp;
9a7f143a
CL		 102
103 if (offset >= size)
104 return size;
105
692223d1 106 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 107 offset %= BITS_PER_LONG;
f96999c3 108
692223d1
FL		 109 while (1) {
110 if (*p == 0)
111 goto pass;
9a7f143a 112
f96999c3 113 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 118 if (tmp)
692223d1
FL		 119 goto found;
120pass:
121 if (size <= BITS_PER_LONG)
122 break;
9a7f143a 123 size -= BITS_PER_LONG;
692223d1 124 offset = 0;
f96999c3 125 p++;
9a7f143a 126 }
692223d1
FL		 127 return result;
128found:
129 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 130}
131
132static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134{
135 const unsigned long *p = addr + BIT_WORD(offset);
80609448 136 unsigned long result = size;
9a7f143a 137 unsigned long tmp;
9a7f143a
CL		 138
139 if (offset >= size)
140 return size;
141
80609448 142 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 143 offset %= BITS_PER_LONG;
80609448
JK		 144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
f96999c3 149 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
f96999c3 155 if (tmp != ~0UL)
80609448
JK		 156 goto found;
157pass:
158 if (size <= BITS_PER_LONG)
159 break;
9a7f143a 160 size -= BITS_PER_LONG;
80609448 161 offset = 0;
f96999c3 162 p++;
9a7f143a 163 }
80609448
JK		 164 return result;
165found:
166 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 167}
168
88b88a66
JK		 169void register_inmem_page(struct inode *inode, struct page *page)
170{
171 struct f2fs_inode_info *fi = F2FS_I(inode);
172 struct inmem_pages *new;
9be32d72 173
9e4ded3f 174 f2fs_trace_pid(page);
0722b101 175
decd36b6
CY		 176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
177 SetPagePrivate(page);
178
88b88a66
JK		 179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
180
181 /* add atomic page indices to the list */
182 new->page = page;
183 INIT_LIST_HEAD(&new->list);
decd36b6 184
88b88a66
JK		 185 /* increase reference count with clean state */
186 mutex_lock(&fi->inmem_lock);
187 get_page(page);
188 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 190 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 191
192 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 193}
194
28bc106b
CY		 195static int __revoke_inmem_pages(struct inode *inode,
196 struct list_head *head, bool drop, bool recover)
29b96b54 197{
28bc106b 198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 199 struct inmem_pages *cur, *tmp;
28bc106b 200 int err = 0;
29b96b54
CY		 201
202 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 203 struct page *page = cur->page;
204
205 if (drop)
206 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
207
208 lock_page(page);
29b96b54 209
28bc106b
CY		 210 if (recover) {
211 struct dnode_of_data dn;
212 struct node_info ni;
213
214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
215
216 set_new_dnode(&dn, inode, NULL, NULL, 0);
217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
218 err = -EAGAIN;
219 goto next;
220 }
221 get_node_info(sbi, dn.nid, &ni);
222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
223 cur->old_addr, ni.version, true, true);
224 f2fs_put_dnode(&dn);
225 }
226next:
63c52d78
JK		 227 /* we don't need to invalidate this in the sccessful status */
228 if (drop || recover)
229 ClearPageUptodate(page);
28bc106b 230 set_page_private(page, 0);
c81ced05 231 ClearPagePrivate(page);
28bc106b 232 f2fs_put_page(page, 1);
29b96b54
CY		 233
234 list_del(&cur->list);
235 kmem_cache_free(inmem_entry_slab, cur);
236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
237 }
28bc106b 238 return err;
29b96b54
CY		 239}
240
241void drop_inmem_pages(struct inode *inode)
242{
243 struct f2fs_inode_info *fi = F2FS_I(inode);
244
245 mutex_lock(&fi->inmem_lock);
28bc106b 246 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 247 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 248
249 clear_inode_flag(inode, FI_ATOMIC_FILE);
250 stat_dec_atomic_write(inode);
29b96b54
CY		 251}
252
28bc106b
CY		 253static int __commit_inmem_pages(struct inode *inode,
254 struct list_head *revoke_list)
88b88a66
JK		 255{
256 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
257 struct f2fs_inode_info *fi = F2FS_I(inode);
258 struct inmem_pages *cur, *tmp;
88b88a66 259 struct f2fs_io_info fio = {
05ca3632 260 .sbi = sbi,
88b88a66 261 .type = DATA,
04d328de 262 .op = REQ_OP_WRITE,
70fd7614 263 .op_flags = REQ_SYNC | REQ_PRIO,
4375a336 264 .encrypted_page = NULL,
88b88a66 265 };
29b96b54 266 bool submit_bio = false;
edb27dee 267 int err = 0;
88b88a66 268
88b88a66 269 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 270 struct page *page = cur->page;
271
272 lock_page(page);
273 if (page->mapping == inode->i_mapping) {
274 trace_f2fs_commit_inmem_page(page, INMEM);
275
276 set_page_dirty(page);
277 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 278 if (clear_page_dirty_for_io(page)) {
29b96b54 279 inode_dec_dirty_pages(inode);
933439c8
CY		 280 remove_dirty_inode(inode);
281 }
28bc106b
CY		 282
283 fio.page = page;
29b96b54
CY		 284 err = do_write_data_page(&fio);
285 if (err) {
28bc106b 286 unlock_page(page);
29b96b54 287 break;
70c640b1 288 }
29b96b54 289
28bc106b
CY		 290 /* record old blkaddr for revoking */
291 cur->old_addr = fio.old_blkaddr;
decd36b6 292
28bc106b
CY		 293 submit_bio = true;
294 }
295 unlock_page(page);
296 list_move_tail(&cur->list, revoke_list);
88b88a66 297 }
29b96b54
CY		 298
299 if (submit_bio)
300 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
28bc106b
CY		 301
302 if (!err)
303 __revoke_inmem_pages(inode, revoke_list, false, false);
304
29b96b54
CY		 305 return err;
306}
307
308int commit_inmem_pages(struct inode *inode)
309{
310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
311 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 312 struct list_head revoke_list;
313 int err;
29b96b54 314
28bc106b 315 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 316 f2fs_balance_fs(sbi, true);
317 f2fs_lock_op(sbi);
318
5fe45743
CY		 319 set_inode_flag(inode, FI_ATOMIC_COMMIT);
320
29b96b54 321 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 322 err = __commit_inmem_pages(inode, &revoke_list);
323 if (err) {
324 int ret;
325 /*
326 * try to revoke all committed pages, but still we could fail
327 * due to no memory or other reason, if that happened, EAGAIN
328 * will be returned, which means in such case, transaction is
329 * already not integrity, caller should use journal to do the
330 * recovery or rewrite & commit last transaction. For other
331 * error number, revoking was done by filesystem itself.
332 */
333 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
334 if (ret)
335 err = ret;
336
337 /* drop all uncommitted pages */
338 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
339 }
88b88a66
JK		 340 mutex_unlock(&fi->inmem_lock);
341
5fe45743
CY		 342 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
343
29b96b54 344 f2fs_unlock_op(sbi);
edb27dee 345 return err;
88b88a66
JK		 346}
347
0a8165d7 348/*
351df4b2
JK		 349 * This function balances dirty node and dentry pages.
350 * In addition, it controls garbage collection.
351 */
2c4db1a6 352void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 353{
0f348028
CY		 354#ifdef CONFIG_F2FS_FAULT_INJECTION
355 if (time_to_inject(sbi, FAULT_CHECKPOINT))
356 f2fs_stop_checkpoint(sbi, false);
357#endif
358
2c4db1a6
JK		 359 if (!need)
360 return;
e589c2c4
JK		 361
362 /* balance_fs_bg is able to be pending */
363 if (excess_cached_nats(sbi))
364 f2fs_balance_fs_bg(sbi);
365
351df4b2 366 /*
029cd28c
JK		 367 * We should do GC or end up with checkpoint, if there are so many dirty
368 * dir/node pages without enough free segments.
351df4b2 369 */
7f3037a5 370 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 371 mutex_lock(&sbi->gc_mutex);
7702bdbe 372 f2fs_gc(sbi, false, false);
351df4b2
JK		 373 }
374}
375
4660f9c0
JK		 376void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
377{
1dcc336b 378 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 379 if (!available_free_memory(sbi, EXTENT_CACHE))
380 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 381
1b38dc8e
JK		 382 /* check the # of cached NAT entries */
383 if (!available_free_memory(sbi, NAT_ENTRIES))
384 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
385
31696580 386 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 387 try_to_free_nids(sbi, MAX_FREE_NIDS);
388 else
3a2ad567 389 build_free_nids(sbi, false);
31696580 390
f455c8a5
JK		 391 if (!is_idle(sbi))
392 return;
31696580 393
1b38dc8e
JK		 394 /* checkpoint is the only way to shrink partial cached entries */
395 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 396 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 397 excess_prefree_segs(sbi) ||
398 excess_dirty_nats(sbi) ||
f455c8a5 399 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 400 if (test_opt(sbi, DATA_FLUSH)) {
401 struct blk_plug plug;
402
403 blk_start_plug(&plug);
36b35a0d 404 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 405 blk_finish_plug(&plug);
406 }
4660f9c0 407 f2fs_sync_fs(sbi->sb, true);
42190d2a 408 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 409 }
4660f9c0
JK		 410}
411
3c62be17
JK		 412static int __submit_flush_wait(struct block_device *bdev)
413{
414 struct bio *bio = f2fs_bio_alloc(0);
415 int ret;
416
09cb6464 417 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3c62be17
JK		 418 bio->bi_bdev = bdev;
419 ret = submit_bio_wait(bio);
420 bio_put(bio);
421 return ret;
422}
423
424static int submit_flush_wait(struct f2fs_sb_info *sbi)
425{
426 int ret = __submit_flush_wait(sbi->sb->s_bdev);
427 int i;
428
429 if (sbi->s_ndevs && !ret) {
430 for (i = 1; i < sbi->s_ndevs; i++) {
431 ret = __submit_flush_wait(FDEV(i).bdev);
432 if (ret)
433 break;
434 }
435 }
436 return ret;
437}
438
2163d198 439static int issue_flush_thread(void *data)
6b4afdd7
JK		 440{
441 struct f2fs_sb_info *sbi = data;
b01a9201 442 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 443 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 444repeat:
445 if (kthread_should_stop())
446 return 0;
447
721bd4d5 448 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 449 struct flush_cmd *cmd, *next;
450 int ret;
451
721bd4d5
GZ		 452 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
453 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
454
3c62be17 455 ret = submit_flush_wait(sbi);
721bd4d5
GZ		 456 llist_for_each_entry_safe(cmd, next,
457 fcc->dispatch_list, llnode) {
6b4afdd7 458 cmd->ret = ret;
6b4afdd7
JK		 459 complete(&cmd->wait);
460 }
a688b9d9 461 fcc->dispatch_list = NULL;
6b4afdd7
JK		 462 }
463
a688b9d9 464 wait_event_interruptible(*q,
721bd4d5 465 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 466 goto repeat;
467}
468
469int f2fs_issue_flush(struct f2fs_sb_info *sbi)
470{
b01a9201 471 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 472 struct flush_cmd cmd;
6b4afdd7 473
24a9ee0f
JK		 474 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
475 test_opt(sbi, FLUSH_MERGE));
476
0f7b2abd
JK		 477 if (test_opt(sbi, NOBARRIER))
478 return 0;
479
73545817
CY		 480 if (!test_opt(sbi, FLUSH_MERGE))
481 return submit_flush_wait(sbi);
482
483 if (!atomic_read(&fcc->submit_flush)) {
740432f8
JK		 484 int ret;
485
0a87f664 486 atomic_inc(&fcc->submit_flush);
3c62be17 487 ret = submit_flush_wait(sbi);
0a87f664 488 atomic_dec(&fcc->submit_flush);
740432f8
JK		 489 return ret;
490 }
6b4afdd7 491
adf8d90b 492 init_completion(&cmd.wait);
6b4afdd7 493
0a87f664 494 atomic_inc(&fcc->submit_flush);
721bd4d5 495 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 496
a688b9d9
GZ		 497 if (!fcc->dispatch_list)
498 wake_up(&fcc->flush_wait_queue);
6b4afdd7 499
5eba8c5d
JK		 500 if (fcc->f2fs_issue_flush) {
501 wait_for_completion(&cmd.wait);
502 atomic_dec(&fcc->submit_flush);
503 } else {
504 llist_del_all(&fcc->issue_list);
505 atomic_set(&fcc->submit_flush, 0);
506 }
adf8d90b
CY		 507
508 return cmd.ret;
6b4afdd7
JK		 509}
510
2163d198
GZ		 511int create_flush_cmd_control(struct f2fs_sb_info *sbi)
512{
513 dev_t dev = sbi->sb->s_bdev->bd_dev;
514 struct flush_cmd_control *fcc;
515 int err = 0;
516
b01a9201
JK		 517 if (SM_I(sbi)->fcc_info) {
518 fcc = SM_I(sbi)->fcc_info;
5eba8c5d
JK		 519 goto init_thread;
520 }
521
2163d198
GZ		 522 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
523 if (!fcc)
524 return -ENOMEM;
0a87f664 525 atomic_set(&fcc->submit_flush, 0);
2163d198 526 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 527 init_llist_head(&fcc->issue_list);
b01a9201 528 SM_I(sbi)->fcc_info = fcc;
5eba8c5d 529init_thread:
2163d198
GZ		 530 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
531 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
532 if (IS_ERR(fcc->f2fs_issue_flush)) {
533 err = PTR_ERR(fcc->f2fs_issue_flush);
534 kfree(fcc);
b01a9201 535 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 536 return err;
537 }
2163d198
GZ		 538
539 return err;
540}
541
5eba8c5d 542void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 543{
b01a9201 544 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 545
5eba8c5d
JK		 546 if (fcc && fcc->f2fs_issue_flush) {
547 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
548
549 fcc->f2fs_issue_flush = NULL;
550 kthread_stop(flush_thread);
551 }
552 if (free) {
553 kfree(fcc);
b01a9201 554 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 555 }
2163d198
GZ		 556}
557
351df4b2
JK		 558static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
559 enum dirty_type dirty_type)
560{
561 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
562
563 /* need not be added */
564 if (IS_CURSEG(sbi, segno))
565 return;
566
567 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
568 dirty_i->nr_dirty[dirty_type]++;
569
570 if (dirty_type == DIRTY) {
571 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 572 enum dirty_type t = sentry->type;
b2f2c390 573
ec325b52
JK		 574 if (unlikely(t >= DIRTY)) {
575 f2fs_bug_on(sbi, 1);
576 return;
577 }
4625d6aa
CL		 578 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
579 dirty_i->nr_dirty[t]++;
351df4b2
JK		 580 }
581}
582
583static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
584 enum dirty_type dirty_type)
585{
586 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
587
588 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
589 dirty_i->nr_dirty[dirty_type]--;
590
591 if (dirty_type == DIRTY) {
4625d6aa
CL		 592 struct seg_entry *sentry = get_seg_entry(sbi, segno);
593 enum dirty_type t = sentry->type;
594
595 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
596 dirty_i->nr_dirty[t]--;
b2f2c390 597
5ec4e49f
JK		 598 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
599 clear_bit(GET_SECNO(sbi, segno),
600 dirty_i->victim_secmap);
351df4b2
JK		 601 }
602}
603
0a8165d7 604/*
351df4b2
JK		 605 * Should not occur error such as -ENOMEM.
606 * Adding dirty entry into seglist is not critical operation.
607 * If a given segment is one of current working segments, it won't be added.
608 */
8d8451af 609static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 610{
611 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
612 unsigned short valid_blocks;
613
614 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
615 return;
616
617 mutex_lock(&dirty_i->seglist_lock);
618
619 valid_blocks = get_valid_blocks(sbi, segno, 0);
620
621 if (valid_blocks == 0) {
622 __locate_dirty_segment(sbi, segno, PRE);
623 __remove_dirty_segment(sbi, segno, DIRTY);
624 } else if (valid_blocks < sbi->blocks_per_seg) {
625 __locate_dirty_segment(sbi, segno, DIRTY);
626 } else {
627 /* Recovery routine with SSR needs this */
628 __remove_dirty_segment(sbi, segno, DIRTY);
629 }
630
631 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 632}
633
15469963 634static void __add_discard_cmd(struct f2fs_sb_info *sbi,
4e6a8d9b 635 struct bio *bio, block_t lstart, block_t len)
275b66b0 636{
0b54fb84
JK		 637 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
638 struct list_head *cmd_list = &(dcc->discard_cmd_list);
b01a9201 639 struct discard_cmd *dc;
275b66b0 640
b01a9201
JK		 641 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
642 INIT_LIST_HEAD(&dc->list);
643 dc->bio = bio;
15469963 644 bio->bi_private = dc;
b01a9201
JK		 645 dc->lstart = lstart;
646 dc->len = len;
15469963 647 dc->state = D_PREP;
b01a9201 648 init_completion(&dc->wait);
15469963
JK		 649
650 mutex_lock(&dcc->cmd_lock);
0b54fb84 651 list_add_tail(&dc->list, cmd_list);
15469963
JK		 652 mutex_unlock(&dcc->cmd_lock);
653}
654
655static void __remove_discard_cmd(struct f2fs_sb_info *sbi, struct discard_cmd *dc)
656{
657 int err = dc->bio->bi_error;
275b66b0 658
dcc9165d
JK		 659 if (dc->state == D_DONE)
660 atomic_dec(&(SM_I(sbi)->dcc_info->submit_discard));
661
15469963
JK		 662 if (err == -EOPNOTSUPP)
663 err = 0;
664
665 if (err)
666 f2fs_msg(sbi->sb, KERN_INFO,
667 "Issue discard failed, ret: %d", err);
668 bio_put(dc->bio);
669 list_del(&dc->list);
670 kmem_cache_free(discard_cmd_slab, dc);
275b66b0
CY		 671}
672
4e6a8d9b
JK		 673/* This should be covered by global mutex, &sit_i->sentry_lock */
674void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
275b66b0 675{
0b54fb84
JK		 676 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
677 struct list_head *wait_list = &(dcc->discard_cmd_list);
b01a9201 678 struct discard_cmd *dc, *tmp;
275b66b0 679
15469963 680 mutex_lock(&dcc->cmd_lock);
b01a9201 681 list_for_each_entry_safe(dc, tmp, wait_list, list) {
275b66b0 682
15469963
JK		 683 if (blkaddr == NULL_ADDR) {
684 if (dc->state == D_PREP) {
685 dc->state = D_SUBMIT;
686 submit_bio(dc->bio);
dcc9165d 687 atomic_inc(&dcc->submit_discard);
15469963
JK		 688 }
689 wait_for_completion_io(&dc->wait);
690
691 __remove_discard_cmd(sbi, dc);
692 continue;
693 }
694
695 if (dc->lstart <= blkaddr && blkaddr < dc->lstart + dc->len) {
696 if (dc->state == D_SUBMIT)
b01a9201 697 wait_for_completion_io(&dc->wait);
4e6a8d9b 698 else
15469963 699 __remove_discard_cmd(sbi, dc);
4e6a8d9b 700 }
275b66b0 701 }
15469963 702 mutex_unlock(&dcc->cmd_lock);
275b66b0
CY		 703}
704
b01a9201 705static void f2fs_submit_discard_endio(struct bio *bio)
275b66b0 706{
b01a9201 707 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
275b66b0 708
b01a9201 709 complete(&dc->wait);
15469963 710 dc->state = D_DONE;
275b66b0
CY		 711}
712
15469963
JK		 713static int issue_discard_thread(void *data)
714{
715 struct f2fs_sb_info *sbi = data;
716 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
717 wait_queue_head_t *q = &dcc->discard_wait_queue;
718 struct list_head *cmd_list = &dcc->discard_cmd_list;
719 struct discard_cmd *dc, *tmp;
720 struct blk_plug plug;
721 int iter = 0;
722repeat:
723 if (kthread_should_stop())
724 return 0;
725
726 blk_start_plug(&plug);
727
728 mutex_lock(&dcc->cmd_lock);
729 list_for_each_entry_safe(dc, tmp, cmd_list, list) {
730 if (dc->state == D_PREP) {
731 dc->state = D_SUBMIT;
732 submit_bio(dc->bio);
dcc9165d 733 atomic_inc(&dcc->submit_discard);
15469963
JK		 734 if (iter++ > DISCARD_ISSUE_RATE)
735 break;
736 } else if (dc->state == D_DONE) {
737 __remove_discard_cmd(sbi, dc);
738 }
739 }
740 mutex_unlock(&dcc->cmd_lock);
741
742 blk_finish_plug(&plug);
743
744 iter = 0;
745 congestion_wait(BLK_RW_SYNC, HZ/50);
746
747 wait_event_interruptible(*q,
748 kthread_should_stop() || !list_empty(&dcc->discard_cmd_list));
749 goto repeat;
750}
751
752
275b66b0 753/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
f46e8809 754static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
3c62be17 755 struct block_device *bdev, block_t blkstart, block_t blklen)
275b66b0 756{
275b66b0 757 struct bio *bio = NULL;
4e6a8d9b 758 block_t lblkstart = blkstart;
275b66b0
CY		 759 int err;
760
f46e8809
DLM		 761 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
762
3c62be17
JK		 763 if (sbi->s_ndevs) {
764 int devi = f2fs_target_device_index(sbi, blkstart);
765
766 blkstart -= FDEV(devi).start_blk;
767 }
f46e8809
DLM		 768 err = __blkdev_issue_discard(bdev,
769 SECTOR_FROM_BLOCK(blkstart),
770 SECTOR_FROM_BLOCK(blklen),
771 GFP_NOFS, 0, &bio);
275b66b0 772 if (!err && bio) {
b01a9201 773 bio->bi_end_io = f2fs_submit_discard_endio;
275b66b0 774 bio->bi_opf |= REQ_SYNC;
15469963
JK		 775
776 __add_discard_cmd(sbi, bio, lblkstart, blklen);
777 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
275b66b0 778 }
275b66b0
CY		 779 return err;
780}
781
f46e8809 782#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 783static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
784 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 785{
f46e8809 786 sector_t nr_sects = SECTOR_FROM_BLOCK(blklen);
3c62be17
JK		 787 sector_t sector;
788 int devi = 0;
789
790 if (sbi->s_ndevs) {
791 devi = f2fs_target_device_index(sbi, blkstart);
792 blkstart -= FDEV(devi).start_blk;
793 }
794 sector = SECTOR_FROM_BLOCK(blkstart);
f46e8809 795
f99e8648
DLM		 796 if (sector & (bdev_zone_sectors(bdev) - 1) ||
797 nr_sects != bdev_zone_sectors(bdev)) {
f46e8809 798 f2fs_msg(sbi->sb, KERN_INFO,
3c62be17
JK		 799 "(%d) %s: Unaligned discard attempted (block %x + %x)",
800 devi, sbi->s_ndevs ? FDEV(devi).path: "",
801 blkstart, blklen);
f46e8809
DLM		 802 return -EIO;
803 }
804
805 /*
806 * We need to know the type of the zone: for conventional zones,
807 * use regular discard if the drive supports it. For sequential
808 * zones, reset the zone write pointer.
809 */
3c62be17 810 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 811
812 case BLK_ZONE_TYPE_CONVENTIONAL:
813 if (!blk_queue_discard(bdev_get_queue(bdev)))
814 return 0;
3c62be17 815 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
f46e8809
DLM		 816 case BLK_ZONE_TYPE_SEQWRITE_REQ:
817 case BLK_ZONE_TYPE_SEQWRITE_PREF:
126606c7 818 trace_f2fs_issue_reset_zone(sbi->sb, blkstart);
f46e8809
DLM		 819 return blkdev_reset_zones(bdev, sector,
820 nr_sects, GFP_NOFS);
821 default:
822 /* Unknown zone type: broken device ? */
823 return -EIO;
824 }
825}
826#endif
827
3c62be17
JK		 828static int __issue_discard_async(struct f2fs_sb_info *sbi,
829 struct block_device *bdev, block_t blkstart, block_t blklen)
830{
831#ifdef CONFIG_BLK_DEV_ZONED
832 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
833 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
834 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
835#endif
836 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
837}
838
1e87a78d 839static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 840 block_t blkstart, block_t blklen)
841{
3c62be17
JK		 842 sector_t start = blkstart, len = 0;
843 struct block_device *bdev;
a66cdd98
JK		 844 struct seg_entry *se;
845 unsigned int offset;
846 block_t i;
3c62be17
JK		 847 int err = 0;
848
849 bdev = f2fs_target_device(sbi, blkstart, NULL);
850
851 for (i = blkstart; i < blkstart + blklen; i++, len++) {
852 if (i != start) {
853 struct block_device *bdev2 =
854 f2fs_target_device(sbi, i, NULL);
855
856 if (bdev2 != bdev) {
857 err = __issue_discard_async(sbi, bdev,
858 start, len);
859 if (err)
860 return err;
861 bdev = bdev2;
862 start = i;
863 len = 0;
864 }
865 }
a66cdd98 866
a66cdd98
JK		 867 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
868 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
869
870 if (!f2fs_test_and_set_bit(offset, se->discard_map))
871 sbi->discard_blks--;
872 }
f46e8809 873
3c62be17
JK		 874 if (len)
875 err = __issue_discard_async(sbi, bdev, start, len);
876 return err;
1e87a78d
JK		 877}
878
adf4983b 879static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 880 struct cp_control *cpc, struct seg_entry *se,
881 unsigned int start, unsigned int end)
b2955550 882{
0b54fb84 883 struct list_head *head = &SM_I(sbi)->dcc_info->discard_entry_list;
adf4983b
JK		 884 struct discard_entry *new, *last;
885
886 if (!list_empty(head)) {
887 last = list_last_entry(head, struct discard_entry, list);
888 if (START_BLOCK(sbi, cpc->trim_start) + start ==
889 last->blkaddr + last->len) {
890 last->len += end - start;
891 goto done;
892 }
893 }
894
895 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
896 INIT_LIST_HEAD(&new->list);
897 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
898 new->len = end - start;
899 list_add_tail(&new->list, head);
900done:
0b54fb84 901 SM_I(sbi)->dcc_info->nr_discards += end - start;
adf4983b
JK		 902}
903
25290fa5
JK		 904static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
905 bool check_only)
adf4983b 906{
b2955550
JK		 907 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
908 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 909 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 910 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
911 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 912 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 913 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 914 unsigned int start = 0, end = -1;
4b2fecc8 915 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 916 int i;
917
3e025740 918 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 919 return false;
b2955550 920
a66cdd98
JK		 921 if (!force) {
922 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 923 SM_I(sbi)->dcc_info->nr_discards >=
924 SM_I(sbi)->dcc_info->max_discards)
25290fa5 925 return false;
4b2fecc8
JK		 926 }
927
b2955550
JK		 928 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
929 for (i = 0; i < entries; i++)
a66cdd98 930 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 931 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 932
0b54fb84
JK		 933 while (force || SM_I(sbi)->dcc_info->nr_discards <=
934 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 935 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
936 if (start >= max_blocks)
937 break;
938
939 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 940 if (force && start && end != max_blocks
941 && (end - start) < cpc->trim_minlen)
942 continue;
943
25290fa5
JK		 944 if (check_only)
945 return true;
946
a66cdd98 947 __add_discard_entry(sbi, cpc, se, start, end);
b2955550 948 }
25290fa5 949 return false;
b2955550
JK		 950}
951
4b2fecc8
JK		 952void release_discard_addrs(struct f2fs_sb_info *sbi)
953{
0b54fb84 954 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
4b2fecc8
JK		 955 struct discard_entry *entry, *this;
956
957 /* drop caches */
958 list_for_each_entry_safe(entry, this, head, list) {
959 list_del(&entry->list);
960 kmem_cache_free(discard_entry_slab, entry);
961 }
962}
963
0a8165d7 964/*
351df4b2
JK		 965 * Should call clear_prefree_segments after checkpoint is done.
966 */
967static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
968{
969 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 970 unsigned int segno;
351df4b2
JK		 971
972 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 973 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 974 __set_test_and_free(sbi, segno);
351df4b2
JK		 975 mutex_unlock(&dirty_i->seglist_lock);
976}
977
836b5a63 978void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 979{
0b54fb84 980 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
2d7b822a 981 struct discard_entry *entry, *this;
351df4b2 982 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 983 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 984 unsigned int start = 0, end = -1;
36abef4e 985 unsigned int secno, start_segno;
c24a0fd6 986 bool force = (cpc->reason == CP_DISCARD);
351df4b2
JK		 987
988 mutex_lock(&dirty_i->seglist_lock);
29e59c14 989
351df4b2 990 while (1) {
29e59c14 991 int i;
7cd8558b
JK		 992 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
993 if (start >= MAIN_SEGS(sbi))
351df4b2 994 break;
7cd8558b
JK		 995 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
996 start + 1);
29e59c14
CL		 997
998 for (i = start; i < end; i++)
999 clear_bit(i, prefree_map);
1000
1001 dirty_i->nr_dirty[PRE] -= end - start;
1002
650d3c4e 1003 if (!test_opt(sbi, DISCARD))
29e59c14 1004 continue;
351df4b2 1005
650d3c4e
YH		 1006 if (force && start >= cpc->trim_start &&
1007 (end - 1) <= cpc->trim_end)
1008 continue;
1009
36abef4e
JK		 1010 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1011 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1012 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1013 continue;
1014 }
1015next:
1016 secno = GET_SECNO(sbi, start);
1017 start_segno = secno * sbi->segs_per_sec;
1018 if (!IS_CURSEC(sbi, secno) &&
1019 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
1020 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1021 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1022
1023 start = start_segno + sbi->segs_per_sec;
1024 if (start < end)
1025 goto next;
351df4b2
JK		 1026 }
1027 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1028
1029 /* send small discards */
2d7b822a 1030 list_for_each_entry_safe(entry, this, head, list) {
c24a0fd6 1031 if (force && entry->len < cpc->trim_minlen)
836b5a63 1032 goto skip;
37208879 1033 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 1034 cpc->trimmed += entry->len;
836b5a63 1035skip:
b2955550 1036 list_del(&entry->list);
0b54fb84 1037 SM_I(sbi)->dcc_info->nr_discards -= entry->len;
b2955550
JK		 1038 kmem_cache_free(discard_entry_slab, entry);
1039 }
351df4b2
JK		 1040}
1041
0b54fb84
JK		 1042int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1043{
15469963 1044 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84
JK		 1045 struct discard_cmd_control *dcc;
1046 int err = 0;
1047
1048 if (SM_I(sbi)->dcc_info) {
1049 dcc = SM_I(sbi)->dcc_info;
1050 goto init_thread;
1051 }
1052
1053 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1054 if (!dcc)
1055 return -ENOMEM;
1056
1057 INIT_LIST_HEAD(&dcc->discard_entry_list);
1058 INIT_LIST_HEAD(&dcc->discard_cmd_list);
15469963 1059 mutex_init(&dcc->cmd_lock);
dcc9165d 1060 atomic_set(&dcc->submit_discard, 0);
0b54fb84
JK		 1061 dcc->nr_discards = 0;
1062 dcc->max_discards = 0;
1063
15469963 1064 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1065 SM_I(sbi)->dcc_info = dcc;
1066init_thread:
15469963
JK		 1067 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1068 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1069 if (IS_ERR(dcc->f2fs_issue_discard)) {
1070 err = PTR_ERR(dcc->f2fs_issue_discard);
1071 kfree(dcc);
1072 SM_I(sbi)->dcc_info = NULL;
1073 return err;
1074 }
1075
0b54fb84
JK		 1076 return err;
1077}
1078
1079void destroy_discard_cmd_control(struct f2fs_sb_info *sbi, bool free)
1080{
1081 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1082
15469963
JK		 1083 if (dcc && dcc->f2fs_issue_discard) {
1084 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1085
1086 dcc->f2fs_issue_discard = NULL;
1087 kthread_stop(discard_thread);
1088 }
0b54fb84
JK		 1089 if (free) {
1090 kfree(dcc);
1091 SM_I(sbi)->dcc_info = NULL;
1092 }
1093}
1094
184a5cd2 1095static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1096{
1097 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1098
1099 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1100 sit_i->dirty_sentries++;
184a5cd2
CY		 1101 return false;
1102 }
1103
1104 return true;
351df4b2
JK		 1105}
1106
1107static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1108 unsigned int segno, int modified)
1109{
1110 struct seg_entry *se = get_seg_entry(sbi, segno);
1111 se->type = type;
1112 if (modified)
1113 __mark_sit_entry_dirty(sbi, segno);
1114}
1115
1116static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1117{
1118 struct seg_entry *se;
1119 unsigned int segno, offset;
1120 long int new_vblocks;
1121
1122 segno = GET_SEGNO(sbi, blkaddr);
1123
1124 se = get_seg_entry(sbi, segno);
1125 new_vblocks = se->valid_blocks + del;
491c0854 1126 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1127
9850cf4a 1128 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1129 (new_vblocks > sbi->blocks_per_seg)));
1130
1131 se->valid_blocks = new_vblocks;
1132 se->mtime = get_mtime(sbi);
1133 SIT_I(sbi)->max_mtime = se->mtime;
1134
1135 /* Update valid block bitmap */
1136 if (del > 0) {
355e7891
CY		 1137 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1138#ifdef CONFIG_F2FS_CHECK_FS
1139 if (f2fs_test_and_set_bit(offset,
1140 se->cur_valid_map_mir))
1141 f2fs_bug_on(sbi, 1);
1142 else
1143 WARN_ON(1);
1144#else
05796763 1145 f2fs_bug_on(sbi, 1);
355e7891
CY		 1146#endif
1147 }
3e025740
JK		 1148 if (f2fs_discard_en(sbi) &&
1149 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1150 sbi->discard_blks--;
351df4b2 1151 } else {
355e7891
CY		 1152 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1153#ifdef CONFIG_F2FS_CHECK_FS
1154 if (!f2fs_test_and_clear_bit(offset,
1155 se->cur_valid_map_mir))
1156 f2fs_bug_on(sbi, 1);
1157 else
1158 WARN_ON(1);
1159#else
05796763 1160 f2fs_bug_on(sbi, 1);
355e7891
CY		 1161#endif
1162 }
3e025740
JK		 1163 if (f2fs_discard_en(sbi) &&
1164 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1165 sbi->discard_blks++;
351df4b2
JK		 1166 }
1167 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1168 se->ckpt_valid_blocks += del;
1169
1170 __mark_sit_entry_dirty(sbi, segno);
1171
1172 /* update total number of valid blocks to be written in ckpt area */
1173 SIT_I(sbi)->written_valid_blocks += del;
1174
1175 if (sbi->segs_per_sec > 1)
1176 get_sec_entry(sbi, segno)->valid_blocks += del;
1177}
1178
5e443818 1179void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1180{
5e443818
JK		 1181 update_sit_entry(sbi, new, 1);
1182 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1183 update_sit_entry(sbi, old, -1);
1184
1185 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1186 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1187}
1188
1189void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1190{
1191 unsigned int segno = GET_SEGNO(sbi, addr);
1192 struct sit_info *sit_i = SIT_I(sbi);
1193
9850cf4a 1194 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1195 if (addr == NEW_ADDR)
1196 return;
1197
1198 /* add it into sit main buffer */
1199 mutex_lock(&sit_i->sentry_lock);
1200
1201 update_sit_entry(sbi, addr, -1);
1202
1203 /* add it into dirty seglist */
1204 locate_dirty_segment(sbi, segno);
1205
1206 mutex_unlock(&sit_i->sentry_lock);
1207}
1208
6e2c64ad
JK		 1209bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1210{
1211 struct sit_info *sit_i = SIT_I(sbi);
1212 unsigned int segno, offset;
1213 struct seg_entry *se;
1214 bool is_cp = false;
1215
1216 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1217 return true;
1218
1219 mutex_lock(&sit_i->sentry_lock);
1220
1221 segno = GET_SEGNO(sbi, blkaddr);
1222 se = get_seg_entry(sbi, segno);
1223 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1224
1225 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1226 is_cp = true;
1227
1228 mutex_unlock(&sit_i->sentry_lock);
1229
1230 return is_cp;
1231}
1232
0a8165d7 1233/*
351df4b2
JK		 1234 * This function should be resided under the curseg_mutex lock
1235 */
1236static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1237 struct f2fs_summary *sum)
351df4b2
JK		 1238{
1239 struct curseg_info *curseg = CURSEG_I(sbi, type);
1240 void *addr = curseg->sum_blk;
e79efe3b 1241 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1242 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1243}
1244
0a8165d7 1245/*
351df4b2
JK		 1246 * Calculate the number of current summary pages for writing
1247 */
3fa06d7b 1248int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1249{
351df4b2 1250 int valid_sum_count = 0;
9a47938b 1251 int i, sum_in_page;
351df4b2
JK		 1252
1253 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1254 if (sbi->ckpt->alloc_type[i] == SSR)
1255 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1256 else {
1257 if (for_ra)
1258 valid_sum_count += le16_to_cpu(
1259 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1260 else
1261 valid_sum_count += curseg_blkoff(sbi, i);
1262 }
351df4b2
JK		 1263 }
1264
09cbfeaf 1265 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1266 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1267 if (valid_sum_count <= sum_in_page)
351df4b2 1268 return 1;
9a47938b 1269 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1270 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1271 return 2;
1272 return 3;
1273}
1274
0a8165d7 1275/*
351df4b2
JK		 1276 * Caller should put this summary page
1277 */
1278struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1279{
1280 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1281}
1282
381722d2 1283void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1284{
1285 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1286 void *dst = page_address(page);
1287
1288 if (src)
09cbfeaf 1289 memcpy(dst, src, PAGE_SIZE);
381722d2 1290 else
09cbfeaf 1291 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1292 set_page_dirty(page);
1293 f2fs_put_page(page, 1);
1294}
1295
381722d2
CY		 1296static void write_sum_page(struct f2fs_sb_info *sbi,
1297 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1298{
1299 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1300}
1301
b7ad7512
CY		 1302static void write_current_sum_page(struct f2fs_sb_info *sbi,
1303 int type, block_t blk_addr)
1304{
1305 struct curseg_info *curseg = CURSEG_I(sbi, type);
1306 struct page *page = grab_meta_page(sbi, blk_addr);
1307 struct f2fs_summary_block *src = curseg->sum_blk;
1308 struct f2fs_summary_block *dst;
1309
1310 dst = (struct f2fs_summary_block *)page_address(page);
1311
1312 mutex_lock(&curseg->curseg_mutex);
1313
1314 down_read(&curseg->journal_rwsem);
1315 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1316 up_read(&curseg->journal_rwsem);
1317
1318 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1319 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1320
1321 mutex_unlock(&curseg->curseg_mutex);
1322
1323 set_page_dirty(page);
1324 f2fs_put_page(page, 1);
1325}
1326
60374688
JK		 1327static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1328{
1329 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 1330 unsigned int segno = curseg->segno + 1;
60374688
JK		 1331 struct free_segmap_info *free_i = FREE_I(sbi);
1332
7cd8558b 1333 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 1334 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 1335 return 0;
1336}
1337
0a8165d7 1338/*
351df4b2
JK		 1339 * Find a new segment from the free segments bitmap to right order
1340 * This function should be returned with success, otherwise BUG
1341 */
1342static void get_new_segment(struct f2fs_sb_info *sbi,
1343 unsigned int *newseg, bool new_sec, int dir)
1344{
1345 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1346 unsigned int segno, secno, zoneno;
7cd8558b 1347 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1348 unsigned int hint = *newseg / sbi->segs_per_sec;
1349 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1350 unsigned int left_start = hint;
1351 bool init = true;
1352 int go_left = 0;
1353 int i;
1354
1a118ccf 1355 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1356
1357 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1358 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1359 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1360 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1361 goto got_it;
1362 }
1363find_other_zone:
7cd8558b
JK		 1364 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1365 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1366 if (dir == ALLOC_RIGHT) {
1367 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1368 MAIN_SECS(sbi), 0);
1369 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1370 } else {
1371 go_left = 1;
1372 left_start = hint - 1;
1373 }
1374 }
1375 if (go_left == 0)
1376 goto skip_left;
1377
1378 while (test_bit(left_start, free_i->free_secmap)) {
1379 if (left_start > 0) {
1380 left_start--;
1381 continue;
1382 }
1383 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1384 MAIN_SECS(sbi), 0);
1385 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1386 break;
1387 }
1388 secno = left_start;
1389skip_left:
1390 hint = secno;
1391 segno = secno * sbi->segs_per_sec;
1392 zoneno = secno / sbi->secs_per_zone;
1393
1394 /* give up on finding another zone */
1395 if (!init)
1396 goto got_it;
1397 if (sbi->secs_per_zone == 1)
1398 goto got_it;
1399 if (zoneno == old_zoneno)
1400 goto got_it;
1401 if (dir == ALLOC_LEFT) {
1402 if (!go_left && zoneno + 1 >= total_zones)
1403 goto got_it;
1404 if (go_left && zoneno == 0)
1405 goto got_it;
1406 }
1407 for (i = 0; i < NR_CURSEG_TYPE; i++)
1408 if (CURSEG_I(sbi, i)->zone == zoneno)
1409 break;
1410
1411 if (i < NR_CURSEG_TYPE) {
1412 /* zone is in user, try another */
1413 if (go_left)
1414 hint = zoneno * sbi->secs_per_zone - 1;
1415 else if (zoneno + 1 >= total_zones)
1416 hint = 0;
1417 else
1418 hint = (zoneno + 1) * sbi->secs_per_zone;
1419 init = false;
1420 goto find_other_zone;
1421 }
1422got_it:
1423 /* set it as dirty segment in free segmap */
9850cf4a 1424 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1425 __set_inuse(sbi, segno);
1426 *newseg = segno;
1a118ccf 1427 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1428}
1429
1430static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1431{
1432 struct curseg_info *curseg = CURSEG_I(sbi, type);
1433 struct summary_footer *sum_footer;
1434
1435 curseg->segno = curseg->next_segno;
1436 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1437 curseg->next_blkoff = 0;
1438 curseg->next_segno = NULL_SEGNO;
1439
1440 sum_footer = &(curseg->sum_blk->footer);
1441 memset(sum_footer, 0, sizeof(struct summary_footer));
1442 if (IS_DATASEG(type))
1443 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1444 if (IS_NODESEG(type))
1445 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1446 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1447}
1448
0a8165d7 1449/*
351df4b2
JK		 1450 * Allocate a current working segment.
1451 * This function always allocates a free segment in LFS manner.
1452 */
1453static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1454{
1455 struct curseg_info *curseg = CURSEG_I(sbi, type);
1456 unsigned int segno = curseg->segno;
1457 int dir = ALLOC_LEFT;
1458
1459 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1460 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1461 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1462 dir = ALLOC_RIGHT;
1463
1464 if (test_opt(sbi, NOHEAP))
1465 dir = ALLOC_RIGHT;
1466
1467 get_new_segment(sbi, &segno, new_sec, dir);
1468 curseg->next_segno = segno;
1469 reset_curseg(sbi, type, 1);
1470 curseg->alloc_type = LFS;
1471}
1472
1473static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1474 struct curseg_info *seg, block_t start)
1475{
1476 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1477 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1478 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1479 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1480 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1481 int i, pos;
1482
1483 for (i = 0; i < entries; i++)
1484 target_map[i] = ckpt_map[i] | cur_map[i];
1485
1486 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1487
1488 seg->next_blkoff = pos;
351df4b2
JK		 1489}
1490
0a8165d7 1491/*
351df4b2
JK		 1492 * If a segment is written by LFS manner, next block offset is just obtained
1493 * by increasing the current block offset. However, if a segment is written by
1494 * SSR manner, next block offset obtained by calling __next_free_blkoff
1495 */
1496static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1497 struct curseg_info *seg)
1498{
1499 if (seg->alloc_type == SSR)
1500 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1501 else
1502 seg->next_blkoff++;
1503}
1504
0a8165d7 1505/*
e1c42045 1506 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1507 * manner, so it should recover the existing segment information of valid blocks
1508 */
1509static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1510{
1511 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1512 struct curseg_info *curseg = CURSEG_I(sbi, type);
1513 unsigned int new_segno = curseg->next_segno;
1514 struct f2fs_summary_block *sum_node;
1515 struct page *sum_page;
1516
1517 write_sum_page(sbi, curseg->sum_blk,
1518 GET_SUM_BLOCK(sbi, curseg->segno));
1519 __set_test_and_inuse(sbi, new_segno);
1520
1521 mutex_lock(&dirty_i->seglist_lock);
1522 __remove_dirty_segment(sbi, new_segno, PRE);
1523 __remove_dirty_segment(sbi, new_segno, DIRTY);
1524 mutex_unlock(&dirty_i->seglist_lock);
1525
1526 reset_curseg(sbi, type, 1);
1527 curseg->alloc_type = SSR;
1528 __next_free_blkoff(sbi, curseg, 0);
1529
1530 if (reuse) {
1531 sum_page = get_sum_page(sbi, new_segno);
1532 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1533 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1534 f2fs_put_page(sum_page, 1);
1535 }
1536}
1537
43727527
JK		 1538static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1539{
1540 struct curseg_info *curseg = CURSEG_I(sbi, type);
1541 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1542
7f3037a5 1543 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
43727527
JK		 1544 return v_ops->get_victim(sbi,
1545 &(curseg)->next_segno, BG_GC, type, SSR);
1546
1547 /* For data segments, let's do SSR more intensively */
1548 for (; type >= CURSEG_HOT_DATA; type--)
1549 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1550 BG_GC, type, SSR))
1551 return 1;
1552 return 0;
1553}
1554
351df4b2
JK		 1555/*
1556 * flush out current segment and replace it with new segment
1557 * This function should be returned with success, otherwise BUG
1558 */
1559static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1560 int type, bool force)
1561{
1562 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1563
7b405275 1564 if (force)
351df4b2 1565 new_curseg(sbi, type, true);
7b405275 1566 else if (type == CURSEG_WARM_NODE)
351df4b2 1567 new_curseg(sbi, type, false);
60374688
JK		 1568 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1569 new_curseg(sbi, type, false);
351df4b2
JK		 1570 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1571 change_curseg(sbi, type, true);
1572 else
1573 new_curseg(sbi, type, false);
dcdfff65
JK		 1574
1575 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1576}
1577
1578void allocate_new_segments(struct f2fs_sb_info *sbi)
1579{
6ae1be13
JK		 1580 struct curseg_info *curseg;
1581 unsigned int old_segno;
351df4b2
JK		 1582 int i;
1583
6ae1be13
JK		 1584 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1585 curseg = CURSEG_I(sbi, i);
1586 old_segno = curseg->segno;
1587 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1588 locate_dirty_segment(sbi, old_segno);
1589 }
351df4b2
JK		 1590}
1591
1592static const struct segment_allocation default_salloc_ops = {
1593 .allocate_segment = allocate_segment_by_default,
1594};
1595
25290fa5
JK		 1596bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1597{
1598 __u64 trim_start = cpc->trim_start;
1599 bool has_candidate = false;
1600
1601 mutex_lock(&SIT_I(sbi)->sentry_lock);
1602 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1603 if (add_discard_addrs(sbi, cpc, true)) {
1604 has_candidate = true;
1605 break;
1606 }
1607 }
1608 mutex_unlock(&SIT_I(sbi)->sentry_lock);
1609
1610 cpc->trim_start = trim_start;
1611 return has_candidate;
1612}
1613
4b2fecc8
JK		 1614int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1615{
f7ef9b83
JK		 1616 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1617 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1618 unsigned int start_segno, end_segno;
1619 struct cp_control cpc;
c34f42e2 1620 int err = 0;
4b2fecc8 1621
836b5a63 1622 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1623 return -EINVAL;
1624
9bd27ae4 1625 cpc.trimmed = 0;
7cd8558b 1626 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1627 goto out;
1628
ed214a11
YH		 1629 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1630 f2fs_msg(sbi->sb, KERN_WARNING,
1631 "Found FS corruption, run fsck to fix.");
1632 goto out;
1633 }
1634
4b2fecc8 1635 /* start/end segment number in main_area */
7cd8558b
JK		 1636 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1637 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1638 GET_SEGNO(sbi, end);
4b2fecc8 1639 cpc.reason = CP_DISCARD;
836b5a63 1640 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1641
1642 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1643 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1644 cpc.trim_start = start_segno;
a66cdd98
JK		 1645
1646 if (sbi->discard_blks == 0)
1647 break;
1648 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1649 cpc.trim_end = end_segno;
1650 else
1651 cpc.trim_end = min_t(unsigned int,
1652 rounddown(start_segno +
bba681cb
JK		 1653 BATCHED_TRIM_SEGMENTS(sbi),
1654 sbi->segs_per_sec) - 1, end_segno);
1655
1656 mutex_lock(&sbi->gc_mutex);
c34f42e2 1657 err = write_checkpoint(sbi, &cpc);
bba681cb 1658 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 1659 if (err)
1660 break;
74fa5f3d
CY		 1661
1662 schedule();
bba681cb 1663 }
4b2fecc8 1664out:
f7ef9b83 1665 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1666 return err;
4b2fecc8
JK		 1667}
1668
351df4b2
JK		 1669static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1670{
1671 struct curseg_info *curseg = CURSEG_I(sbi, type);
1672 if (curseg->next_blkoff < sbi->blocks_per_seg)
1673 return true;
1674 return false;
1675}
1676
1677static int __get_segment_type_2(struct page *page, enum page_type p_type)
1678{
1679 if (p_type == DATA)
1680 return CURSEG_HOT_DATA;
1681 else
1682 return CURSEG_HOT_NODE;
1683}
1684
1685static int __get_segment_type_4(struct page *page, enum page_type p_type)
1686{
1687 if (p_type == DATA) {
1688 struct inode *inode = page->mapping->host;
1689
1690 if (S_ISDIR(inode->i_mode))
1691 return CURSEG_HOT_DATA;
1692 else
1693 return CURSEG_COLD_DATA;
1694 } else {
a344b9fd
JK		 1695 if (IS_DNODE(page) && is_cold_node(page))
1696 return CURSEG_WARM_NODE;
351df4b2
JK		 1697 else
1698 return CURSEG_COLD_NODE;
1699 }
1700}
1701
1702static int __get_segment_type_6(struct page *page, enum page_type p_type)
1703{
1704 if (p_type == DATA) {
1705 struct inode *inode = page->mapping->host;
1706
1707 if (S_ISDIR(inode->i_mode))
1708 return CURSEG_HOT_DATA;
354a3399 1709 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1710 return CURSEG_COLD_DATA;
1711 else
1712 return CURSEG_WARM_DATA;
1713 } else {
1714 if (IS_DNODE(page))
1715 return is_cold_node(page) ? CURSEG_WARM_NODE :
1716 CURSEG_HOT_NODE;
1717 else
1718 return CURSEG_COLD_NODE;
1719 }
1720}
1721
1722static int __get_segment_type(struct page *page, enum page_type p_type)
1723{
4081363f 1724 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1725 case 2:
1726 return __get_segment_type_2(page, p_type);
1727 case 4:
1728 return __get_segment_type_4(page, p_type);
351df4b2 1729 }
12a67146 1730 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1731 f2fs_bug_on(F2FS_P_SB(page),
1732 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1733 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1734}
1735
bfad7c2d
JK		 1736void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1737 block_t old_blkaddr, block_t *new_blkaddr,
1738 struct f2fs_summary *sum, int type)
351df4b2
JK		 1739{
1740 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 1741 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 1742
1743 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1744 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 1745
1746 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 1747
4e6a8d9b
JK		 1748 f2fs_wait_discard_bio(sbi, *new_blkaddr);
1749
351df4b2
JK		 1750 /*
1751 * __add_sum_entry should be resided under the curseg_mutex
1752 * because, this function updates a summary entry in the
1753 * current summary block.
1754 */
e79efe3b 1755 __add_sum_entry(sbi, type, sum);
351df4b2 1756
351df4b2 1757 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1758
1759 stat_inc_block_count(sbi, curseg);
351df4b2 1760
5e443818
JK		 1761 if (!__has_curseg_space(sbi, type))
1762 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1763 /*
1764 * SIT information should be updated before segment allocation,
1765 * since SSR needs latest valid block information.
1766 */
1767 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1768
351df4b2
JK		 1769 mutex_unlock(&sit_i->sentry_lock);
1770
bfad7c2d 1771 if (page && IS_NODESEG(type))
351df4b2
JK		 1772 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1773
bfad7c2d
JK		 1774 mutex_unlock(&curseg->curseg_mutex);
1775}
1776
05ca3632 1777static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1778{
05ca3632 1779 int type = __get_segment_type(fio->page, fio->type);
0a595eba 1780 int err;
bfad7c2d 1781
7dfeaa32
JK		 1782 if (fio->type == NODE || fio->type == DATA)
1783 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
0a595eba 1784reallocate:
7a9d7548
CY		 1785 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1786 &fio->new_blkaddr, sum, type);
bfad7c2d 1787
351df4b2 1788 /* writeout dirty page into bdev */
0a595eba
JK		 1789 err = f2fs_submit_page_mbio(fio);
1790 if (err == -EAGAIN) {
1791 fio->old_blkaddr = fio->new_blkaddr;
1792 goto reallocate;
1793 }
7dfeaa32
JK		 1794
1795 if (fio->type == NODE || fio->type == DATA)
1796 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1797}
1798
577e3495 1799void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1800{
458e6197 1801 struct f2fs_io_info fio = {
05ca3632 1802 .sbi = sbi,
458e6197 1803 .type = META,
04d328de 1804 .op = REQ_OP_WRITE,
70fd7614 1805 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1806 .old_blkaddr = page->index,
1807 .new_blkaddr = page->index,
05ca3632 1808 .page = page,
4375a336 1809 .encrypted_page = NULL,
458e6197
JK		 1810 };
1811
2b947003 1812 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1813 fio.op_flags &= ~REQ_META;
2b947003 1814
351df4b2 1815 set_page_writeback(page);
05ca3632 1816 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1817}
1818
05ca3632 1819void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1820{
1821 struct f2fs_summary sum;
05ca3632 1822
351df4b2 1823 set_summary(&sum, nid, 0, 0);
05ca3632 1824 do_write_page(&sum, fio);
351df4b2
JK		 1825}
1826
05ca3632 1827void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1828{
05ca3632 1829 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1830 struct f2fs_summary sum;
1831 struct node_info ni;
1832
9850cf4a 1833 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1834 get_node_info(sbi, dn->nid, &ni);
1835 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1836 do_write_page(&sum, fio);
f28b3434 1837 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1838}
1839
05ca3632 1840void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1841{
7a9d7548 1842 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1843 stat_inc_inplace_blocks(fio->sbi);
1844 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1845}
1846
4356e48e 1847void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1848 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1849 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1850{
1851 struct sit_info *sit_i = SIT_I(sbi);
1852 struct curseg_info *curseg;
1853 unsigned int segno, old_cursegno;
1854 struct seg_entry *se;
1855 int type;
19f106bc 1856 unsigned short old_blkoff;
351df4b2
JK		 1857
1858 segno = GET_SEGNO(sbi, new_blkaddr);
1859 se = get_seg_entry(sbi, segno);
1860 type = se->type;
1861
19f106bc
CY		 1862 if (!recover_curseg) {
1863 /* for recovery flow */
1864 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1865 if (old_blkaddr == NULL_ADDR)
1866 type = CURSEG_COLD_DATA;
1867 else
1868 type = CURSEG_WARM_DATA;
1869 }
1870 } else {
1871 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1872 type = CURSEG_WARM_DATA;
1873 }
19f106bc 1874
351df4b2
JK		 1875 curseg = CURSEG_I(sbi, type);
1876
1877 mutex_lock(&curseg->curseg_mutex);
1878 mutex_lock(&sit_i->sentry_lock);
1879
1880 old_cursegno = curseg->segno;
19f106bc 1881 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1882
1883 /* change the current segment */
1884 if (segno != curseg->segno) {
1885 curseg->next_segno = segno;
1886 change_curseg(sbi, type, true);
1887 }
1888
491c0854 1889 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1890 __add_sum_entry(sbi, type, sum);
351df4b2 1891
28bc106b 1892 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1893 update_sit_entry(sbi, new_blkaddr, 1);
1894 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1895 update_sit_entry(sbi, old_blkaddr, -1);
1896
1897 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1898 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1899
351df4b2 1900 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1901
19f106bc
CY		 1902 if (recover_curseg) {
1903 if (old_cursegno != curseg->segno) {
1904 curseg->next_segno = old_cursegno;
1905 change_curseg(sbi, type, true);
1906 }
1907 curseg->next_blkoff = old_blkoff;
1908 }
1909
351df4b2
JK		 1910 mutex_unlock(&sit_i->sentry_lock);
1911 mutex_unlock(&curseg->curseg_mutex);
1912}
1913
528e3459
CY		 1914void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1915 block_t old_addr, block_t new_addr,
28bc106b
CY		 1916 unsigned char version, bool recover_curseg,
1917 bool recover_newaddr)
528e3459
CY		 1918{
1919 struct f2fs_summary sum;
1920
1921 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1922
28bc106b
CY		 1923 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1924 recover_curseg, recover_newaddr);
528e3459 1925
f28b3434 1926 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1927}
1928
93dfe2ac 1929void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1930 enum page_type type, bool ordered)
93dfe2ac 1931{
93dfe2ac 1932 if (PageWriteback(page)) {
4081363f
JK		 1933 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1934
0c3a5797 1935 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1936 if (ordered)
1937 wait_on_page_writeback(page);
1938 else
1939 wait_for_stable_page(page);
93dfe2ac
JK		 1940 }
1941}
1942
08b39fbd
CY		 1943void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1944 block_t blkaddr)
1945{
1946 struct page *cpage;
1947
5d4c0af4 1948 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 1949 return;
1950
08b39fbd
CY		 1951 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1952 if (cpage) {
fec1d657 1953 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1954 f2fs_put_page(cpage, 1);
1955 }
1956}
1957
351df4b2
JK		 1958static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1959{
1960 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1961 struct curseg_info *seg_i;
1962 unsigned char *kaddr;
1963 struct page *page;
1964 block_t start;
1965 int i, j, offset;
1966
1967 start = start_sum_block(sbi);
1968
1969 page = get_meta_page(sbi, start++);
1970 kaddr = (unsigned char *)page_address(page);
1971
1972 /* Step 1: restore nat cache */
1973 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1974 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1975
1976 /* Step 2: restore sit cache */
1977 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1978 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1979 offset = 2 * SUM_JOURNAL_SIZE;
1980
1981 /* Step 3: restore summary entries */
1982 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1983 unsigned short blk_off;
1984 unsigned int segno;
1985
1986 seg_i = CURSEG_I(sbi, i);
1987 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1988 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1989 seg_i->next_segno = segno;
1990 reset_curseg(sbi, i, 0);
1991 seg_i->alloc_type = ckpt->alloc_type[i];
1992 seg_i->next_blkoff = blk_off;
1993
1994 if (seg_i->alloc_type == SSR)
1995 blk_off = sbi->blocks_per_seg;
1996
1997 for (j = 0; j < blk_off; j++) {
1998 struct f2fs_summary *s;
1999 s = (struct f2fs_summary *)(kaddr + offset);
2000 seg_i->sum_blk->entries[j] = *s;
2001 offset += SUMMARY_SIZE;
09cbfeaf 2002 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2003 SUM_FOOTER_SIZE)
2004 continue;
2005
2006 f2fs_put_page(page, 1);
2007 page = NULL;
2008
2009 page = get_meta_page(sbi, start++);
2010 kaddr = (unsigned char *)page_address(page);
2011 offset = 0;
2012 }
2013 }
2014 f2fs_put_page(page, 1);
2015 return 0;
2016}
2017
2018static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2019{
2020 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2021 struct f2fs_summary_block *sum;
2022 struct curseg_info *curseg;
2023 struct page *new;
2024 unsigned short blk_off;
2025 unsigned int segno = 0;
2026 block_t blk_addr = 0;
2027
2028 /* get segment number and block addr */
2029 if (IS_DATASEG(type)) {
2030 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2031 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2032 CURSEG_HOT_DATA]);
119ee914 2033 if (__exist_node_summaries(sbi))
351df4b2
JK		 2034 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2035 else
2036 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2037 } else {
2038 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2039 CURSEG_HOT_NODE]);
2040 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2041 CURSEG_HOT_NODE]);
119ee914 2042 if (__exist_node_summaries(sbi))
351df4b2
JK		 2043 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2044 type - CURSEG_HOT_NODE);
2045 else
2046 blk_addr = GET_SUM_BLOCK(sbi, segno);
2047 }
2048
2049 new = get_meta_page(sbi, blk_addr);
2050 sum = (struct f2fs_summary_block *)page_address(new);
2051
2052 if (IS_NODESEG(type)) {
119ee914 2053 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2054 struct f2fs_summary *ns = &sum->entries[0];
2055 int i;
2056 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2057 ns->version = 0;
2058 ns->ofs_in_node = 0;
2059 }
2060 } else {
d653788a
GZ		 2061 int err;
2062
2063 err = restore_node_summary(sbi, segno, sum);
2064 if (err) {
351df4b2 2065 f2fs_put_page(new, 1);
d653788a 2066 return err;
351df4b2
JK		 2067 }
2068 }
2069 }
2070
2071 /* set uncompleted segment to curseg */
2072 curseg = CURSEG_I(sbi, type);
2073 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2074
2075 /* update journal info */
2076 down_write(&curseg->journal_rwsem);
2077 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2078 up_write(&curseg->journal_rwsem);
2079
2080 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2081 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2082 curseg->next_segno = segno;
2083 reset_curseg(sbi, type, 0);
2084 curseg->alloc_type = ckpt->alloc_type[type];
2085 curseg->next_blkoff = blk_off;
2086 mutex_unlock(&curseg->curseg_mutex);
2087 f2fs_put_page(new, 1);
2088 return 0;
2089}
2090
2091static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2092{
2093 int type = CURSEG_HOT_DATA;
e4fc5fbf 2094 int err;
351df4b2 2095
aaec2b1d 2096 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2097 int npages = npages_for_summary_flush(sbi, true);
2098
2099 if (npages >= 2)
2100 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2101 META_CP, true);
3fa06d7b 2102
351df4b2
JK		 2103 /* restore for compacted data summary */
2104 if (read_compacted_summaries(sbi))
2105 return -EINVAL;
2106 type = CURSEG_HOT_NODE;
2107 }
2108
119ee914 2109 if (__exist_node_summaries(sbi))
3fa06d7b 2110 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2111 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2112
e4fc5fbf
CY		 2113 for (; type <= CURSEG_COLD_NODE; type++) {
2114 err = read_normal_summaries(sbi, type);
2115 if (err)
2116 return err;
2117 }
2118
351df4b2
JK		 2119 return 0;
2120}
2121
2122static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2123{
2124 struct page *page;
2125 unsigned char *kaddr;
2126 struct f2fs_summary *summary;
2127 struct curseg_info *seg_i;
2128 int written_size = 0;
2129 int i, j;
2130
2131 page = grab_meta_page(sbi, blkaddr++);
2132 kaddr = (unsigned char *)page_address(page);
2133
2134 /* Step 1: write nat cache */
2135 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2136 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2137 written_size += SUM_JOURNAL_SIZE;
2138
2139 /* Step 2: write sit cache */
2140 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2141 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2142 written_size += SUM_JOURNAL_SIZE;
2143
351df4b2
JK		 2144 /* Step 3: write summary entries */
2145 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2146 unsigned short blkoff;
2147 seg_i = CURSEG_I(sbi, i);
2148 if (sbi->ckpt->alloc_type[i] == SSR)
2149 blkoff = sbi->blocks_per_seg;
2150 else
2151 blkoff = curseg_blkoff(sbi, i);
2152
2153 for (j = 0; j < blkoff; j++) {
2154 if (!page) {
2155 page = grab_meta_page(sbi, blkaddr++);
2156 kaddr = (unsigned char *)page_address(page);
2157 written_size = 0;
2158 }
2159 summary = (struct f2fs_summary *)(kaddr + written_size);
2160 *summary = seg_i->sum_blk->entries[j];
2161 written_size += SUMMARY_SIZE;
351df4b2 2162
09cbfeaf 2163 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2164 SUM_FOOTER_SIZE)
2165 continue;
2166
e8d61a74 2167 set_page_dirty(page);
351df4b2
JK		 2168 f2fs_put_page(page, 1);
2169 page = NULL;
2170 }
2171 }
e8d61a74
CY		 2172 if (page) {
2173 set_page_dirty(page);
351df4b2 2174 f2fs_put_page(page, 1);
e8d61a74 2175 }
351df4b2
JK		 2176}
2177
2178static void write_normal_summaries(struct f2fs_sb_info *sbi,
2179 block_t blkaddr, int type)
2180{
2181 int i, end;
2182 if (IS_DATASEG(type))
2183 end = type + NR_CURSEG_DATA_TYPE;
2184 else
2185 end = type + NR_CURSEG_NODE_TYPE;
2186
b7ad7512
CY		 2187 for (i = type; i < end; i++)
2188 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2189}
2190
2191void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2192{
aaec2b1d 2193 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2194 write_compacted_summaries(sbi, start_blk);
2195 else
2196 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2197}
2198
2199void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2200{
119ee914 2201 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2202}
2203
dfc08a12 2204int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2205 unsigned int val, int alloc)
2206{
2207 int i;
2208
2209 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2210 for (i = 0; i < nats_in_cursum(journal); i++) {
2211 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2212 return i;
2213 }
dfc08a12
CY		 2214 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2215 return update_nats_in_cursum(journal, 1);
351df4b2 2216 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2217 for (i = 0; i < sits_in_cursum(journal); i++)
2218 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2219 return i;
dfc08a12
CY		 2220 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2221 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2222 }
2223 return -1;
2224}
2225
2226static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2227 unsigned int segno)
2228{
2cc22186 2229 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2230}
2231
2232static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2233 unsigned int start)
2234{
2235 struct sit_info *sit_i = SIT_I(sbi);
2236 struct page *src_page, *dst_page;
2237 pgoff_t src_off, dst_off;
2238 void *src_addr, *dst_addr;
2239
2240 src_off = current_sit_addr(sbi, start);
2241 dst_off = next_sit_addr(sbi, src_off);
2242
2243 /* get current sit block page without lock */
2244 src_page = get_meta_page(sbi, src_off);
2245 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2246 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2247
2248 src_addr = page_address(src_page);
2249 dst_addr = page_address(dst_page);
09cbfeaf 2250 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2251
2252 set_page_dirty(dst_page);
2253 f2fs_put_page(src_page, 1);
2254
2255 set_to_next_sit(sit_i, start);
2256
2257 return dst_page;
2258}
2259
184a5cd2
CY		 2260static struct sit_entry_set *grab_sit_entry_set(void)
2261{
2262 struct sit_entry_set *ses =
80c54505 2263 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2264
2265 ses->entry_cnt = 0;
2266 INIT_LIST_HEAD(&ses->set_list);
2267 return ses;
2268}
2269
2270static void release_sit_entry_set(struct sit_entry_set *ses)
2271{
2272 list_del(&ses->set_list);
2273 kmem_cache_free(sit_entry_set_slab, ses);
2274}
2275
2276static void adjust_sit_entry_set(struct sit_entry_set *ses,
2277 struct list_head *head)
2278{
2279 struct sit_entry_set *next = ses;
2280
2281 if (list_is_last(&ses->set_list, head))
2282 return;
2283
2284 list_for_each_entry_continue(next, head, set_list)
2285 if (ses->entry_cnt <= next->entry_cnt)
2286 break;
2287
2288 list_move_tail(&ses->set_list, &next->set_list);
2289}
2290
2291static void add_sit_entry(unsigned int segno, struct list_head *head)
2292{
2293 struct sit_entry_set *ses;
2294 unsigned int start_segno = START_SEGNO(segno);
2295
2296 list_for_each_entry(ses, head, set_list) {
2297 if (ses->start_segno == start_segno) {
2298 ses->entry_cnt++;
2299 adjust_sit_entry_set(ses, head);
2300 return;
2301 }
2302 }
2303
2304 ses = grab_sit_entry_set();
2305
2306 ses->start_segno = start_segno;
2307 ses->entry_cnt++;
2308 list_add(&ses->set_list, head);
2309}
2310
2311static void add_sits_in_set(struct f2fs_sb_info *sbi)
2312{
2313 struct f2fs_sm_info *sm_info = SM_I(sbi);
2314 struct list_head *set_list = &sm_info->sit_entry_set;
2315 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2316 unsigned int segno;
2317
7cd8558b 2318 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2319 add_sit_entry(segno, set_list);
2320}
2321
2322static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2323{
2324 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2325 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2326 int i;
2327
b7ad7512 2328 down_write(&curseg->journal_rwsem);
dfc08a12 2329 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2330 unsigned int segno;
2331 bool dirtied;
2332
dfc08a12 2333 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2334 dirtied = __mark_sit_entry_dirty(sbi, segno);
2335
2336 if (!dirtied)
2337 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2338 }
dfc08a12 2339 update_sits_in_cursum(journal, -i);
b7ad7512 2340 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2341}
2342
0a8165d7 2343/*
351df4b2
JK		 2344 * CP calls this function, which flushes SIT entries including sit_journal,
2345 * and moves prefree segs to free segs.
2346 */
4b2fecc8 2347void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2348{
2349 struct sit_info *sit_i = SIT_I(sbi);
2350 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2351 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2352 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2353 struct sit_entry_set *ses, *tmp;
2354 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2355 bool to_journal = true;
4b2fecc8 2356 struct seg_entry *se;
351df4b2 2357
351df4b2
JK		 2358 mutex_lock(&sit_i->sentry_lock);
2359
2b11a74b
WL		 2360 if (!sit_i->dirty_sentries)
2361 goto out;
2362
351df4b2 2363 /*
184a5cd2
CY		 2364 * add and account sit entries of dirty bitmap in sit entry
2365 * set temporarily
351df4b2 2366 */
184a5cd2 2367 add_sits_in_set(sbi);
351df4b2 2368
184a5cd2
CY		 2369 /*
2370 * if there are no enough space in journal to store dirty sit
2371 * entries, remove all entries from journal and add and account
2372 * them in sit entry set.
2373 */
dfc08a12 2374 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2375 remove_sits_in_journal(sbi);
b2955550 2376
184a5cd2
CY		 2377 /*
2378 * there are two steps to flush sit entries:
2379 * #1, flush sit entries to journal in current cold data summary block.
2380 * #2, flush sit entries to sit page.
2381 */
2382 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2383 struct page *page = NULL;
184a5cd2
CY		 2384 struct f2fs_sit_block *raw_sit = NULL;
2385 unsigned int start_segno = ses->start_segno;
2386 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2387 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2388 unsigned int segno = start_segno;
2389
2390 if (to_journal &&
dfc08a12 2391 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2392 to_journal = false;
2393
b7ad7512
CY		 2394 if (to_journal) {
2395 down_write(&curseg->journal_rwsem);
2396 } else {
184a5cd2
CY		 2397 page = get_next_sit_page(sbi, start_segno);
2398 raw_sit = page_address(page);
351df4b2 2399 }
351df4b2 2400
184a5cd2
CY		 2401 /* flush dirty sit entries in region of current sit set */
2402 for_each_set_bit_from(segno, bitmap, end) {
2403 int offset, sit_offset;
4b2fecc8
JK		 2404
2405 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2406
2407 /* add discard candidates */
d7bc2484 2408 if (cpc->reason != CP_DISCARD) {
4b2fecc8 2409 cpc->trim_start = segno;
25290fa5 2410 add_discard_addrs(sbi, cpc, false);
4b2fecc8 2411 }
184a5cd2
CY		 2412
2413 if (to_journal) {
dfc08a12 2414 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2415 SIT_JOURNAL, segno, 1);
2416 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2417 segno_in_journal(journal, offset) =
184a5cd2
CY		 2418 cpu_to_le32(segno);
2419 seg_info_to_raw_sit(se,
dfc08a12 2420 &sit_in_journal(journal, offset));
184a5cd2
CY		 2421 } else {
2422 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2423 seg_info_to_raw_sit(se,
2424 &raw_sit->entries[sit_offset]);
2425 }
351df4b2 2426
184a5cd2
CY		 2427 __clear_bit(segno, bitmap);
2428 sit_i->dirty_sentries--;
2429 ses->entry_cnt--;
351df4b2
JK		 2430 }
2431
b7ad7512
CY		 2432 if (to_journal)
2433 up_write(&curseg->journal_rwsem);
2434 else
184a5cd2
CY		 2435 f2fs_put_page(page, 1);
2436
2437 f2fs_bug_on(sbi, ses->entry_cnt);
2438 release_sit_entry_set(ses);
351df4b2 2439 }
184a5cd2
CY		 2440
2441 f2fs_bug_on(sbi, !list_empty(head));
2442 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2443out:
4b2fecc8 2444 if (cpc->reason == CP_DISCARD) {
650d3c4e
YH		 2445 __u64 trim_start = cpc->trim_start;
2446
4b2fecc8 2447 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 2448 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 2449
2450 cpc->trim_start = trim_start;
4b2fecc8 2451 }
351df4b2 2452 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2453
351df4b2
JK		 2454 set_prefree_as_free_segments(sbi);
2455}
2456
2457static int build_sit_info(struct f2fs_sb_info *sbi)
2458{
2459 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 2460 struct sit_info *sit_i;
2461 unsigned int sit_segs, start;
ae27d62e 2462 char *src_bitmap;
351df4b2
JK		 2463 unsigned int bitmap_size;
2464
2465 /* allocate memory for SIT information */
2466 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2467 if (!sit_i)
2468 return -ENOMEM;
2469
2470 SM_I(sbi)->sit_info = sit_i;
2471
39307a8e
JK		 2472 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2473 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2474 if (!sit_i->sentries)
2475 return -ENOMEM;
2476
7cd8558b 2477 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2478 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2479 if (!sit_i->dirty_sentries_bitmap)
2480 return -ENOMEM;
2481
7cd8558b 2482 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2483 sit_i->sentries[start].cur_valid_map
2484 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2485 sit_i->sentries[start].ckpt_valid_map
2486 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 2487 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 2488 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 2489 return -ENOMEM;
3e025740 2490
355e7891
CY		 2491#ifdef CONFIG_F2FS_CHECK_FS
2492 sit_i->sentries[start].cur_valid_map_mir
2493 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2494 if (!sit_i->sentries[start].cur_valid_map_mir)
2495 return -ENOMEM;
2496#endif
2497
3e025740
JK		 2498 if (f2fs_discard_en(sbi)) {
2499 sit_i->sentries[start].discard_map
2500 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2501 if (!sit_i->sentries[start].discard_map)
2502 return -ENOMEM;
2503 }
351df4b2
JK		 2504 }
2505
60a3b782
JK		 2506 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2507 if (!sit_i->tmp_map)
2508 return -ENOMEM;
2509
351df4b2 2510 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2511 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2512 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2513 if (!sit_i->sec_entries)
2514 return -ENOMEM;
2515 }
2516
2517 /* get information related with SIT */
2518 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2519
2520 /* setup SIT bitmap from ckeckpoint pack */
2521 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2522 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2523
ae27d62e
CY		 2524 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2525 if (!sit_i->sit_bitmap)
351df4b2 2526 return -ENOMEM;
351df4b2 2527
ae27d62e
CY		 2528#ifdef CONFIG_F2FS_CHECK_FS
2529 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2530 if (!sit_i->sit_bitmap_mir)
2531 return -ENOMEM;
2532#endif
2533
351df4b2
JK		 2534 /* init SIT information */
2535 sit_i->s_ops = &default_salloc_ops;
2536
2537 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2538 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 2539 sit_i->written_valid_blocks = 0;
351df4b2
JK		 2540 sit_i->bitmap_size = bitmap_size;
2541 sit_i->dirty_sentries = 0;
2542 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2543 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2544 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2545 mutex_init(&sit_i->sentry_lock);
2546 return 0;
2547}
2548
2549static int build_free_segmap(struct f2fs_sb_info *sbi)
2550{
351df4b2
JK		 2551 struct free_segmap_info *free_i;
2552 unsigned int bitmap_size, sec_bitmap_size;
2553
2554 /* allocate memory for free segmap information */
2555 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2556 if (!free_i)
2557 return -ENOMEM;
2558
2559 SM_I(sbi)->free_info = free_i;
2560
7cd8558b 2561 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2562 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2563 if (!free_i->free_segmap)
2564 return -ENOMEM;
2565
7cd8558b 2566 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2567 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2568 if (!free_i->free_secmap)
2569 return -ENOMEM;
2570
2571 /* set all segments as dirty temporarily */
2572 memset(free_i->free_segmap, 0xff, bitmap_size);
2573 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2574
2575 /* init free segmap information */
7cd8558b 2576 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2577 free_i->free_segments = 0;
2578 free_i->free_sections = 0;
1a118ccf 2579 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2580 return 0;
2581}
2582
2583static int build_curseg(struct f2fs_sb_info *sbi)
2584{
1042d60f 2585 struct curseg_info *array;
351df4b2
JK		 2586 int i;
2587
b434babf 2588 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2589 if (!array)
2590 return -ENOMEM;
2591
2592 SM_I(sbi)->curseg_array = array;
2593
2594 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2595 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2596 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2597 if (!array[i].sum_blk)
2598 return -ENOMEM;
b7ad7512
CY		 2599 init_rwsem(&array[i].journal_rwsem);
2600 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2601 GFP_KERNEL);
2602 if (!array[i].journal)
2603 return -ENOMEM;
351df4b2
JK		 2604 array[i].segno = NULL_SEGNO;
2605 array[i].next_blkoff = 0;
2606 }
2607 return restore_curseg_summaries(sbi);
2608}
2609
2610static void build_sit_entries(struct f2fs_sb_info *sbi)
2611{
2612 struct sit_info *sit_i = SIT_I(sbi);
2613 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2614 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 2615 struct seg_entry *se;
2616 struct f2fs_sit_entry sit;
74de593a
CY		 2617 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2618 unsigned int i, start, end;
2619 unsigned int readed, start_blk = 0;
351df4b2 2620
74de593a 2621 do {
664ba972
JK		 2622 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2623 META_SIT, true);
74de593a
CY		 2624
2625 start = start_blk * sit_i->sents_per_block;
2626 end = (start_blk + readed) * sit_i->sents_per_block;
2627
7cd8558b 2628 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 2629 struct f2fs_sit_block *sit_blk;
74de593a
CY		 2630 struct page *page;
2631
9c094040 2632 se = &sit_i->sentries[start];
74de593a
CY		 2633 page = get_current_sit_page(sbi, start);
2634 sit_blk = (struct f2fs_sit_block *)page_address(page);
2635 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2636 f2fs_put_page(page, 1);
d600af23 2637
74de593a
CY		 2638 check_block_count(sbi, start, &sit);
2639 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2640
2641 /* build discard map only one time */
3e025740
JK		 2642 if (f2fs_discard_en(sbi)) {
2643 memcpy(se->discard_map, se->cur_valid_map,
2644 SIT_VBLOCK_MAP_SIZE);
2645 sbi->discard_blks += sbi->blocks_per_seg -
2646 se->valid_blocks;
2647 }
a66cdd98 2648
d600af23
CY		 2649 if (sbi->segs_per_sec > 1)
2650 get_sec_entry(sbi, start)->valid_blocks +=
2651 se->valid_blocks;
351df4b2 2652 }
74de593a
CY		 2653 start_blk += readed;
2654 } while (start_blk < sit_blk_cnt);
d600af23
CY		 2655
2656 down_read(&curseg->journal_rwsem);
2657 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 2658 unsigned int old_valid_blocks;
2659
2660 start = le32_to_cpu(segno_in_journal(journal, i));
2661 se = &sit_i->sentries[start];
2662 sit = sit_in_journal(journal, i);
2663
2664 old_valid_blocks = se->valid_blocks;
2665
2666 check_block_count(sbi, start, &sit);
2667 seg_info_from_raw_sit(se, &sit);
2668
2669 if (f2fs_discard_en(sbi)) {
2670 memcpy(se->discard_map, se->cur_valid_map,
2671 SIT_VBLOCK_MAP_SIZE);
2672 sbi->discard_blks += old_valid_blocks -
2673 se->valid_blocks;
2674 }
2675
2676 if (sbi->segs_per_sec > 1)
2677 get_sec_entry(sbi, start)->valid_blocks +=
2678 se->valid_blocks - old_valid_blocks;
2679 }
2680 up_read(&curseg->journal_rwsem);
351df4b2
JK		 2681}
2682
2683static void init_free_segmap(struct f2fs_sb_info *sbi)
2684{
2685 unsigned int start;
2686 int type;
2687
7cd8558b 2688 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2689 struct seg_entry *sentry = get_seg_entry(sbi, start);
2690 if (!sentry->valid_blocks)
2691 __set_free(sbi, start);
c79b7ff1
JK		 2692 else
2693 SIT_I(sbi)->written_valid_blocks +=
2694 sentry->valid_blocks;
351df4b2
JK		 2695 }
2696
2697 /* set use the current segments */
2698 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2699 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2700 __set_test_and_inuse(sbi, curseg_t->segno);
2701 }
2702}
2703
2704static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2705{
2706 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2707 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2708 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2709 unsigned short valid_blocks;
2710
8736fbf0 2711 while (1) {
351df4b2 2712 /* find dirty segment based on free segmap */
7cd8558b
JK		 2713 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2714 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2715 break;
2716 offset = segno + 1;
2717 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2718 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2719 continue;
ec325b52
JK		 2720 if (valid_blocks > sbi->blocks_per_seg) {
2721 f2fs_bug_on(sbi, 1);
2722 continue;
2723 }
351df4b2
JK		 2724 mutex_lock(&dirty_i->seglist_lock);
2725 __locate_dirty_segment(sbi, segno, DIRTY);
2726 mutex_unlock(&dirty_i->seglist_lock);
2727 }
2728}
2729
5ec4e49f 2730static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2731{
2732 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2733 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2734
39307a8e 2735 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2736 if (!dirty_i->victim_secmap)
351df4b2
JK		 2737 return -ENOMEM;
2738 return 0;
2739}
2740
2741static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2742{
2743 struct dirty_seglist_info *dirty_i;
2744 unsigned int bitmap_size, i;
2745
2746 /* allocate memory for dirty segments list information */
2747 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2748 if (!dirty_i)
2749 return -ENOMEM;
2750
2751 SM_I(sbi)->dirty_info = dirty_i;
2752 mutex_init(&dirty_i->seglist_lock);
2753
7cd8558b 2754 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2755
2756 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2757 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2758 if (!dirty_i->dirty_segmap[i])
2759 return -ENOMEM;
2760 }
2761
2762 init_dirty_segmap(sbi);
5ec4e49f 2763 return init_victim_secmap(sbi);
351df4b2
JK		 2764}
2765
0a8165d7 2766/*
351df4b2
JK		 2767 * Update min, max modified time for cost-benefit GC algorithm
2768 */
2769static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2770{
2771 struct sit_info *sit_i = SIT_I(sbi);
2772 unsigned int segno;
2773
2774 mutex_lock(&sit_i->sentry_lock);
2775
2776 sit_i->min_mtime = LLONG_MAX;
2777
7cd8558b 2778 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2779 unsigned int i;
2780 unsigned long long mtime = 0;
2781
2782 for (i = 0; i < sbi->segs_per_sec; i++)
2783 mtime += get_seg_entry(sbi, segno + i)->mtime;
2784
2785 mtime = div_u64(mtime, sbi->segs_per_sec);
2786
2787 if (sit_i->min_mtime > mtime)
2788 sit_i->min_mtime = mtime;
2789 }
2790 sit_i->max_mtime = get_mtime(sbi);
2791 mutex_unlock(&sit_i->sentry_lock);
2792}
2793
2794int build_segment_manager(struct f2fs_sb_info *sbi)
2795{
2796 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2797 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2798 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2799 int err;
2800
2801 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2802 if (!sm_info)
2803 return -ENOMEM;
2804
2805 /* init sm info */
2806 sbi->sm_info = sm_info;
351df4b2
JK		 2807 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2808 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2809 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2810 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2811 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2812 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2813 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2814 sm_info->rec_prefree_segments = sm_info->main_segments *
2815 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2816 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2817 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2818
52763a4b
JK		 2819 if (!test_opt(sbi, LFS))
2820 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2821 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2822 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2823
bba681cb
JK		 2824 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2825
184a5cd2
CY		 2826 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2827
b270ad6f 2828 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2829 err = create_flush_cmd_control(sbi);
2830 if (err)
a688b9d9 2831 return err;
6b4afdd7
JK		 2832 }
2833
0b54fb84
JK		 2834 err = create_discard_cmd_control(sbi);
2835 if (err)
2836 return err;
2837
351df4b2
JK		 2838 err = build_sit_info(sbi);
2839 if (err)
2840 return err;
2841 err = build_free_segmap(sbi);
2842 if (err)
2843 return err;
2844 err = build_curseg(sbi);
2845 if (err)
2846 return err;
2847
2848 /* reinit free segmap based on SIT */
2849 build_sit_entries(sbi);
2850
2851 init_free_segmap(sbi);
2852 err = build_dirty_segmap(sbi);
2853 if (err)
2854 return err;
2855
2856 init_min_max_mtime(sbi);
2857 return 0;
2858}
2859
2860static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2861 enum dirty_type dirty_type)
2862{
2863 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2864
2865 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2866 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2867 dirty_i->nr_dirty[dirty_type] = 0;
2868 mutex_unlock(&dirty_i->seglist_lock);
2869}
2870
5ec4e49f 2871static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2872{
2873 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2874 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2875}
2876
2877static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2878{
2879 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2880 int i;
2881
2882 if (!dirty_i)
2883 return;
2884
2885 /* discard pre-free/dirty segments list */
2886 for (i = 0; i < NR_DIRTY_TYPE; i++)
2887 discard_dirty_segmap(sbi, i);
2888
5ec4e49f 2889 destroy_victim_secmap(sbi);
351df4b2
JK		 2890 SM_I(sbi)->dirty_info = NULL;
2891 kfree(dirty_i);
2892}
2893
2894static void destroy_curseg(struct f2fs_sb_info *sbi)
2895{
2896 struct curseg_info *array = SM_I(sbi)->curseg_array;
2897 int i;
2898
2899 if (!array)
2900 return;
2901 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2902 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2903 kfree(array[i].sum_blk);
b7ad7512
CY		 2904 kfree(array[i].journal);
2905 }
351df4b2
JK		 2906 kfree(array);
2907}
2908
2909static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2910{
2911 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2912 if (!free_i)
2913 return;
2914 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2915 kvfree(free_i->free_segmap);
2916 kvfree(free_i->free_secmap);
351df4b2
JK		 2917 kfree(free_i);
2918}
2919
2920static void destroy_sit_info(struct f2fs_sb_info *sbi)
2921{
2922 struct sit_info *sit_i = SIT_I(sbi);
2923 unsigned int start;
2924
2925 if (!sit_i)
2926 return;
2927
2928 if (sit_i->sentries) {
7cd8558b 2929 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 2930 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 2931#ifdef CONFIG_F2FS_CHECK_FS
2932 kfree(sit_i->sentries[start].cur_valid_map_mir);
2933#endif
351df4b2 2934 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2935 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2936 }
2937 }
60a3b782
JK		 2938 kfree(sit_i->tmp_map);
2939
39307a8e
JK		 2940 kvfree(sit_i->sentries);
2941 kvfree(sit_i->sec_entries);
2942 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2943
2944 SM_I(sbi)->sit_info = NULL;
2945 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 2946#ifdef CONFIG_F2FS_CHECK_FS
2947 kfree(sit_i->sit_bitmap_mir);
2948#endif
351df4b2
JK		 2949 kfree(sit_i);
2950}
2951
2952void destroy_segment_manager(struct f2fs_sb_info *sbi)
2953{
2954 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2955
3b03f724
CY		 2956 if (!sm_info)
2957 return;
5eba8c5d 2958 destroy_flush_cmd_control(sbi, true);
0b54fb84 2959 destroy_discard_cmd_control(sbi, true);
351df4b2
JK		 2960 destroy_dirty_segmap(sbi);
2961 destroy_curseg(sbi);
2962 destroy_free_segmap(sbi);
2963 destroy_sit_info(sbi);
2964 sbi->sm_info = NULL;
2965 kfree(sm_info);
2966}
7fd9e544
JK		 2967
2968int __init create_segment_manager_caches(void)
2969{
2970 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2971 sizeof(struct discard_entry));
7fd9e544 2972 if (!discard_entry_slab)
184a5cd2
CY		 2973 goto fail;
2974
b01a9201
JK		 2975 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
2976 sizeof(struct discard_cmd));
2977 if (!discard_cmd_slab)
6ab2a308 2978 goto destroy_discard_entry;
275b66b0 2979
184a5cd2 2980 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2981 sizeof(struct sit_entry_set));
184a5cd2 2982 if (!sit_entry_set_slab)
b01a9201 2983 goto destroy_discard_cmd;
88b88a66
JK		 2984
2985 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2986 sizeof(struct inmem_pages));
2987 if (!inmem_entry_slab)
2988 goto destroy_sit_entry_set;
7fd9e544 2989 return 0;
184a5cd2 2990
88b88a66
JK		 2991destroy_sit_entry_set:
2992 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 2993destroy_discard_cmd:
2994 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 2995destroy_discard_entry:
184a5cd2
CY		 2996 kmem_cache_destroy(discard_entry_slab);
2997fail:
2998 return -ENOMEM;
7fd9e544
JK		 2999}
3000
3001void destroy_segment_manager_caches(void)
3002{
184a5cd2 3003 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3004 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3005 kmem_cache_destroy(discard_entry_slab);
88b88a66 3006 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3007}
Linux 6.x block layer and io_uring tree(s)