]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-zesty-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: remove inmem radix tree
[mirror_ubuntu-zesty-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
351df4b2 17#include <linux/vmalloc.h>
74de593a 18#include <linux/swap.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;
184a5cd2 29static struct kmem_cache *sit_entry_set_slab;
88b88a66 30static struct kmem_cache *inmem_entry_slab;
7fd9e544 31
9a7f143a
CL		 32/*
33 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
34 * MSB and LSB are reversed in a byte by f2fs_set_bit.
35 */
36static inline unsigned long __reverse_ffs(unsigned long word)
37{
38 int num = 0;
39
40#if BITS_PER_LONG == 64
41 if ((word & 0xffffffff) == 0) {
42 num += 32;
43 word >>= 32;
44 }
45#endif
46 if ((word & 0xffff) == 0) {
47 num += 16;
48 word >>= 16;
49 }
50 if ((word & 0xff) == 0) {
51 num += 8;
52 word >>= 8;
53 }
54 if ((word & 0xf0) == 0)
55 num += 4;
56 else
57 word >>= 4;
58 if ((word & 0xc) == 0)
59 num += 2;
60 else
61 word >>= 2;
62 if ((word & 0x2) == 0)
63 num += 1;
64 return num;
65}
66
67/*
e1c42045 68 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a
CL		 69 * f2fs_set_bit makes MSB and LSB reversed in a byte.
70 * Example:
71 * LSB <--> MSB
72 * f2fs_set_bit(0, bitmap) => 0000 0001
73 * f2fs_set_bit(7, bitmap) => 1000 0000
74 */
75static unsigned long __find_rev_next_bit(const unsigned long *addr,
76 unsigned long size, unsigned long offset)
77{
e19ef527
JK		 78 while (!f2fs_test_bit(offset, (unsigned char *)addr))
79 offset++;
80
81 if (offset > size)
82 offset = size;
83
84 return offset;
85#if 0
9a7f143a
CL		 86 const unsigned long *p = addr + BIT_WORD(offset);
87 unsigned long result = offset & ~(BITS_PER_LONG - 1);
88 unsigned long tmp;
89 unsigned long mask, submask;
90 unsigned long quot, rest;
91
92 if (offset >= size)
93 return size;
94
95 size -= result;
96 offset %= BITS_PER_LONG;
97 if (!offset)
98 goto aligned;
99
100 tmp = *(p++);
101 quot = (offset >> 3) << 3;
102 rest = offset & 0x7;
103 mask = ~0UL << quot;
104 submask = (unsigned char)(0xff << rest) >> rest;
105 submask <<= quot;
106 mask &= submask;
107 tmp &= mask;
108 if (size < BITS_PER_LONG)
109 goto found_first;
110 if (tmp)
111 goto found_middle;
112
113 size -= BITS_PER_LONG;
114 result += BITS_PER_LONG;
115aligned:
116 while (size & ~(BITS_PER_LONG-1)) {
117 tmp = *(p++);
118 if (tmp)
119 goto found_middle;
120 result += BITS_PER_LONG;
121 size -= BITS_PER_LONG;
122 }
123 if (!size)
124 return result;
125 tmp = *p;
126found_first:
127 tmp &= (~0UL >> (BITS_PER_LONG - size));
128 if (tmp == 0UL) /* Are any bits set? */
129 return result + size; /* Nope. */
130found_middle:
131 return result + __reverse_ffs(tmp);
e19ef527 132#endif
9a7f143a
CL		 133}
134
135static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137{
e19ef527
JK		 138 while (f2fs_test_bit(offset, (unsigned char *)addr))
139 offset++;
140
141 if (offset > size)
142 offset = size;
143
144 return offset;
145#if 0
9a7f143a
CL		 146 const unsigned long *p = addr + BIT_WORD(offset);
147 unsigned long result = offset & ~(BITS_PER_LONG - 1);
148 unsigned long tmp;
149 unsigned long mask, submask;
150 unsigned long quot, rest;
151
152 if (offset >= size)
153 return size;
154
155 size -= result;
156 offset %= BITS_PER_LONG;
157 if (!offset)
158 goto aligned;
159
160 tmp = *(p++);
161 quot = (offset >> 3) << 3;
162 rest = offset & 0x7;
163 mask = ~(~0UL << quot);
164 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
165 submask <<= quot;
166 mask += submask;
167 tmp |= mask;
168 if (size < BITS_PER_LONG)
169 goto found_first;
170 if (~tmp)
171 goto found_middle;
172
173 size -= BITS_PER_LONG;
174 result += BITS_PER_LONG;
175aligned:
176 while (size & ~(BITS_PER_LONG - 1)) {
177 tmp = *(p++);
178 if (~tmp)
179 goto found_middle;
180 result += BITS_PER_LONG;
181 size -= BITS_PER_LONG;
182 }
183 if (!size)
184 return result;
185 tmp = *p;
186
187found_first:
188 tmp |= ~0UL << size;
189 if (tmp == ~0UL) /* Are any bits zero? */
190 return result + size; /* Nope. */
191found_middle:
192 return result + __reverse_ffz(tmp);
e19ef527 193#endif
9a7f143a
CL		 194}
195
88b88a66
JK		 196void register_inmem_page(struct inode *inode, struct page *page)
197{
198 struct f2fs_inode_info *fi = F2FS_I(inode);
199 struct inmem_pages *new;
9be32d72 200
9e4ded3f 201 f2fs_trace_pid(page);
0722b101 202
decd36b6
CY		 203 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
204 SetPagePrivate(page);
205
88b88a66
JK		 206 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
207
208 /* add atomic page indices to the list */
209 new->page = page;
210 INIT_LIST_HEAD(&new->list);
decd36b6 211
88b88a66
JK		 212 /* increase reference count with clean state */
213 mutex_lock(&fi->inmem_lock);
214 get_page(page);
215 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 216 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 217 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 218
219 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 220}
221
edb27dee 222int commit_inmem_pages(struct inode *inode, bool abort)
88b88a66
JK		 223{
224 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
225 struct f2fs_inode_info *fi = F2FS_I(inode);
226 struct inmem_pages *cur, *tmp;
227 bool submit_bio = false;
228 struct f2fs_io_info fio = {
05ca3632 229 .sbi = sbi,
88b88a66 230 .type = DATA,
1e84371f 231 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 232 .encrypted_page = NULL,
88b88a66 233 };
edb27dee 234 int err = 0;
88b88a66 235
0341845e
JK		 236 /*
237 * The abort is true only when f2fs_evict_inode is called.
238 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
239 * that we don't need to call f2fs_balance_fs.
240 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
241 * inode becomes free by iget_locked in f2fs_iget.
242 */
70c640b1 243 if (!abort) {
0341845e 244 f2fs_balance_fs(sbi);
70c640b1
JK		 245 f2fs_lock_op(sbi);
246 }
88b88a66
JK		 247
248 mutex_lock(&fi->inmem_lock);
249 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
decd36b6 250 lock_page(cur->page);
70c640b1 251 if (!abort) {
70c640b1 252 if (cur->page->mapping == inode->i_mapping) {
6282adbf 253 set_page_dirty(cur->page);
70c640b1
JK		 254 f2fs_wait_on_page_writeback(cur->page, DATA);
255 if (clear_page_dirty_for_io(cur->page))
256 inode_dec_dirty_pages(inode);
8ce67cb0 257 trace_f2fs_commit_inmem_page(cur->page, INMEM);
05ca3632 258 fio.page = cur->page;
edb27dee 259 err = do_write_data_page(&fio);
70c640b1 260 submit_bio = true;
edb27dee
JK		 261 if (err) {
262 unlock_page(cur->page);
263 break;
264 }
70c640b1 265 }
70c640b1 266 } else {
8ce67cb0 267 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
88b88a66 268 }
decd36b6
CY		 269 set_page_private(cur->page, 0);
270 ClearPagePrivate(cur->page);
271 f2fs_put_page(cur->page, 1);
272
88b88a66
JK		 273 list_del(&cur->list);
274 kmem_cache_free(inmem_entry_slab, cur);
8dcf2ff7 275 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 276 }
88b88a66
JK		 277 mutex_unlock(&fi->inmem_lock);
278
70c640b1
JK		 279 if (!abort) {
280 f2fs_unlock_op(sbi);
281 if (submit_bio)
282 f2fs_submit_merged_bio(sbi, DATA, WRITE);
283 }
edb27dee 284 return err;
88b88a66
JK		 285}
286
0a8165d7 287/*
351df4b2
JK		 288 * This function balances dirty node and dentry pages.
289 * In addition, it controls garbage collection.
290 */
291void f2fs_balance_fs(struct f2fs_sb_info *sbi)
292{
351df4b2 293 /*
029cd28c
JK		 294 * We should do GC or end up with checkpoint, if there are so many dirty
295 * dir/node pages without enough free segments.
351df4b2 296 */
43727527 297 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 298 mutex_lock(&sbi->gc_mutex);
408e9375 299 f2fs_gc(sbi);
351df4b2
JK		 300 }
301}
302
4660f9c0
JK		 303void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
304{
1dcc336b 305 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 306 if (!available_free_memory(sbi, EXTENT_CACHE))
307 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 308
1b38dc8e
JK		 309 /* check the # of cached NAT entries */
310 if (!available_free_memory(sbi, NAT_ENTRIES))
311 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
312
313 /* checkpoint is the only way to shrink partial cached entries */
314 if (!available_free_memory(sbi, NAT_ENTRIES) ||
e5e7ea3c 315 excess_prefree_segs(sbi) ||
88a70a69 316 !available_free_memory(sbi, INO_ENTRIES))
4660f9c0
JK		 317 f2fs_sync_fs(sbi->sb, true);
318}
319
2163d198 320static int issue_flush_thread(void *data)
6b4afdd7
JK		 321{
322 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 323 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
324 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 325repeat:
326 if (kthread_should_stop())
327 return 0;
328
721bd4d5 329 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 330 struct bio *bio = bio_alloc(GFP_NOIO, 0);
331 struct flush_cmd *cmd, *next;
332 int ret;
333
721bd4d5
GZ		 334 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
335 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
336
6b4afdd7
JK		 337 bio->bi_bdev = sbi->sb->s_bdev;
338 ret = submit_bio_wait(WRITE_FLUSH, bio);
339
721bd4d5
GZ		 340 llist_for_each_entry_safe(cmd, next,
341 fcc->dispatch_list, llnode) {
6b4afdd7 342 cmd->ret = ret;
6b4afdd7
JK		 343 complete(&cmd->wait);
344 }
a4ed23f2 345 bio_put(bio);
a688b9d9 346 fcc->dispatch_list = NULL;
6b4afdd7
JK		 347 }
348
a688b9d9 349 wait_event_interruptible(*q,
721bd4d5 350 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 351 goto repeat;
352}
353
354int f2fs_issue_flush(struct f2fs_sb_info *sbi)
355{
a688b9d9 356 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 357 struct flush_cmd cmd;
6b4afdd7 358
24a9ee0f
JK		 359 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
360 test_opt(sbi, FLUSH_MERGE));
361
0f7b2abd
JK		 362 if (test_opt(sbi, NOBARRIER))
363 return 0;
364
6b4afdd7
JK		 365 if (!test_opt(sbi, FLUSH_MERGE))
366 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
367
adf8d90b 368 init_completion(&cmd.wait);
6b4afdd7 369
721bd4d5 370 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 371
a688b9d9
GZ		 372 if (!fcc->dispatch_list)
373 wake_up(&fcc->flush_wait_queue);
6b4afdd7 374
adf8d90b
CY		 375 wait_for_completion(&cmd.wait);
376
377 return cmd.ret;
6b4afdd7
JK		 378}
379
2163d198
GZ		 380int create_flush_cmd_control(struct f2fs_sb_info *sbi)
381{
382 dev_t dev = sbi->sb->s_bdev->bd_dev;
383 struct flush_cmd_control *fcc;
384 int err = 0;
385
386 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
387 if (!fcc)
388 return -ENOMEM;
2163d198 389 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 390 init_llist_head(&fcc->issue_list);
6b2920a5 391 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 392 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
393 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
394 if (IS_ERR(fcc->f2fs_issue_flush)) {
395 err = PTR_ERR(fcc->f2fs_issue_flush);
396 kfree(fcc);
6b2920a5 397 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 398 return err;
399 }
2163d198
GZ		 400
401 return err;
402}
403
404void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
405{
6b2920a5 406 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 407
408 if (fcc && fcc->f2fs_issue_flush)
409 kthread_stop(fcc->f2fs_issue_flush);
410 kfree(fcc);
6b2920a5 411 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 412}
413
351df4b2
JK		 414static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
415 enum dirty_type dirty_type)
416{
417 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
418
419 /* need not be added */
420 if (IS_CURSEG(sbi, segno))
421 return;
422
423 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
424 dirty_i->nr_dirty[dirty_type]++;
425
426 if (dirty_type == DIRTY) {
427 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 428 enum dirty_type t = sentry->type;
b2f2c390 429
ec325b52
JK		 430 if (unlikely(t >= DIRTY)) {
431 f2fs_bug_on(sbi, 1);
432 return;
433 }
4625d6aa
CL		 434 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
435 dirty_i->nr_dirty[t]++;
351df4b2
JK		 436 }
437}
438
439static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
440 enum dirty_type dirty_type)
441{
442 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
443
444 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
445 dirty_i->nr_dirty[dirty_type]--;
446
447 if (dirty_type == DIRTY) {
4625d6aa
CL		 448 struct seg_entry *sentry = get_seg_entry(sbi, segno);
449 enum dirty_type t = sentry->type;
450
451 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
452 dirty_i->nr_dirty[t]--;
b2f2c390 453
5ec4e49f
JK		 454 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
455 clear_bit(GET_SECNO(sbi, segno),
456 dirty_i->victim_secmap);
351df4b2
JK		 457 }
458}
459
0a8165d7 460/*
351df4b2
JK		 461 * Should not occur error such as -ENOMEM.
462 * Adding dirty entry into seglist is not critical operation.
463 * If a given segment is one of current working segments, it won't be added.
464 */
8d8451af 465static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 466{
467 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
468 unsigned short valid_blocks;
469
470 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
471 return;
472
473 mutex_lock(&dirty_i->seglist_lock);
474
475 valid_blocks = get_valid_blocks(sbi, segno, 0);
476
477 if (valid_blocks == 0) {
478 __locate_dirty_segment(sbi, segno, PRE);
479 __remove_dirty_segment(sbi, segno, DIRTY);
480 } else if (valid_blocks < sbi->blocks_per_seg) {
481 __locate_dirty_segment(sbi, segno, DIRTY);
482 } else {
483 /* Recovery routine with SSR needs this */
484 __remove_dirty_segment(sbi, segno, DIRTY);
485 }
486
487 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 488}
489
1e87a78d 490static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 491 block_t blkstart, block_t blklen)
492{
55cf9cb6
CY		 493 sector_t start = SECTOR_FROM_BLOCK(blkstart);
494 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 495 struct seg_entry *se;
496 unsigned int offset;
497 block_t i;
498
499 for (i = blkstart; i < blkstart + blklen; i++) {
500 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
501 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
502
503 if (!f2fs_test_and_set_bit(offset, se->discard_map))
504 sbi->discard_blks--;
505 }
1661d07c 506 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 507 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
508}
509
e90c2d28 510bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 511{
40a02be1
JK		 512 int err = -ENOTSUPP;
513
514 if (test_opt(sbi, DISCARD)) {
515 struct seg_entry *se = get_seg_entry(sbi,
516 GET_SEGNO(sbi, blkaddr));
517 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
518
519 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 520 return false;
40a02be1
JK		 521
522 err = f2fs_issue_discard(sbi, blkaddr, 1);
523 }
524
e90c2d28 525 if (err) {
381722d2 526 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 527 return true;
528 }
529 return false;
37208879
JK		 530}
531
adf4983b 532static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 533 struct cp_control *cpc, struct seg_entry *se,
534 unsigned int start, unsigned int end)
b2955550
JK		 535{
536 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 537 struct discard_entry *new, *last;
538
539 if (!list_empty(head)) {
540 last = list_last_entry(head, struct discard_entry, list);
541 if (START_BLOCK(sbi, cpc->trim_start) + start ==
542 last->blkaddr + last->len) {
543 last->len += end - start;
544 goto done;
545 }
546 }
547
548 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
549 INIT_LIST_HEAD(&new->list);
550 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
551 new->len = end - start;
552 list_add_tail(&new->list, head);
553done:
554 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 555}
556
557static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
558{
b2955550
JK		 559 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
560 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 561 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 562 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
563 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 564 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 565 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 566 unsigned int start = 0, end = -1;
4b2fecc8 567 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 568 int i;
569
a66cdd98 570 if (se->valid_blocks == max_blocks)
b2955550
JK		 571 return;
572
a66cdd98
JK		 573 if (!force) {
574 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 575 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
576 return;
4b2fecc8
JK		 577 }
578
b2955550
JK		 579 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
580 for (i = 0; i < entries; i++)
a66cdd98 581 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 582 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 583
4b2fecc8 584 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 585 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
586 if (start >= max_blocks)
587 break;
588
589 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 590 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 591 }
592}
593
4b2fecc8
JK		 594void release_discard_addrs(struct f2fs_sb_info *sbi)
595{
596 struct list_head *head = &(SM_I(sbi)->discard_list);
597 struct discard_entry *entry, *this;
598
599 /* drop caches */
600 list_for_each_entry_safe(entry, this, head, list) {
601 list_del(&entry->list);
602 kmem_cache_free(discard_entry_slab, entry);
603 }
604}
605
0a8165d7 606/*
351df4b2
JK		 607 * Should call clear_prefree_segments after checkpoint is done.
608 */
609static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
610{
611 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 612 unsigned int segno;
351df4b2
JK		 613
614 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 615 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 616 __set_test_and_free(sbi, segno);
351df4b2
JK		 617 mutex_unlock(&dirty_i->seglist_lock);
618}
619
836b5a63 620void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 621{
b2955550 622 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 623 struct discard_entry *entry, *this;
351df4b2 624 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 625 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 626 unsigned int start = 0, end = -1;
351df4b2
JK		 627
628 mutex_lock(&dirty_i->seglist_lock);
29e59c14 629
351df4b2 630 while (1) {
29e59c14 631 int i;
7cd8558b
JK		 632 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
633 if (start >= MAIN_SEGS(sbi))
351df4b2 634 break;
7cd8558b
JK		 635 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
636 start + 1);
29e59c14
CL		 637
638 for (i = start; i < end; i++)
639 clear_bit(i, prefree_map);
640
641 dirty_i->nr_dirty[PRE] -= end - start;
642
643 if (!test_opt(sbi, DISCARD))
644 continue;
351df4b2 645
37208879
JK		 646 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
647 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 648 }
649 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 650
651 /* send small discards */
2d7b822a 652 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 653 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
654 goto skip;
37208879 655 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 656 cpc->trimmed += entry->len;
836b5a63 657skip:
b2955550
JK		 658 list_del(&entry->list);
659 SM_I(sbi)->nr_discards -= entry->len;
660 kmem_cache_free(discard_entry_slab, entry);
661 }
351df4b2
JK		 662}
663
184a5cd2 664static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 665{
666 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 667
668 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 669 sit_i->dirty_sentries++;
184a5cd2
CY		 670 return false;
671 }
672
673 return true;
351df4b2
JK		 674}
675
676static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
677 unsigned int segno, int modified)
678{
679 struct seg_entry *se = get_seg_entry(sbi, segno);
680 se->type = type;
681 if (modified)
682 __mark_sit_entry_dirty(sbi, segno);
683}
684
685static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
686{
687 struct seg_entry *se;
688 unsigned int segno, offset;
689 long int new_vblocks;
690
691 segno = GET_SEGNO(sbi, blkaddr);
692
693 se = get_seg_entry(sbi, segno);
694 new_vblocks = se->valid_blocks + del;
491c0854 695 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 696
9850cf4a 697 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 698 (new_vblocks > sbi->blocks_per_seg)));
699
700 se->valid_blocks = new_vblocks;
701 se->mtime = get_mtime(sbi);
702 SIT_I(sbi)->max_mtime = se->mtime;
703
704 /* Update valid block bitmap */
705 if (del > 0) {
52aca074 706 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 707 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 708 if (!f2fs_test_and_set_bit(offset, se->discard_map))
709 sbi->discard_blks--;
351df4b2 710 } else {
52aca074 711 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 712 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 713 if (f2fs_test_and_clear_bit(offset, se->discard_map))
714 sbi->discard_blks++;
351df4b2
JK		 715 }
716 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
717 se->ckpt_valid_blocks += del;
718
719 __mark_sit_entry_dirty(sbi, segno);
720
721 /* update total number of valid blocks to be written in ckpt area */
722 SIT_I(sbi)->written_valid_blocks += del;
723
724 if (sbi->segs_per_sec > 1)
725 get_sec_entry(sbi, segno)->valid_blocks += del;
726}
727
5e443818 728void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 729{
5e443818
JK		 730 update_sit_entry(sbi, new, 1);
731 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
732 update_sit_entry(sbi, old, -1);
733
734 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
735 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 736}
737
738void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
739{
740 unsigned int segno = GET_SEGNO(sbi, addr);
741 struct sit_info *sit_i = SIT_I(sbi);
742
9850cf4a 743 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 744 if (addr == NEW_ADDR)
745 return;
746
747 /* add it into sit main buffer */
748 mutex_lock(&sit_i->sentry_lock);
749
750 update_sit_entry(sbi, addr, -1);
751
752 /* add it into dirty seglist */
753 locate_dirty_segment(sbi, segno);
754
755 mutex_unlock(&sit_i->sentry_lock);
756}
757
0a8165d7 758/*
351df4b2
JK		 759 * This function should be resided under the curseg_mutex lock
760 */
761static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 762 struct f2fs_summary *sum)
351df4b2
JK		 763{
764 struct curseg_info *curseg = CURSEG_I(sbi, type);
765 void *addr = curseg->sum_blk;
e79efe3b 766 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 767 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 768}
769
0a8165d7 770/*
351df4b2
JK		 771 * Calculate the number of current summary pages for writing
772 */
3fa06d7b 773int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 774{
351df4b2 775 int valid_sum_count = 0;
9a47938b 776 int i, sum_in_page;
351df4b2
JK		 777
778 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
779 if (sbi->ckpt->alloc_type[i] == SSR)
780 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 781 else {
782 if (for_ra)
783 valid_sum_count += le16_to_cpu(
784 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
785 else
786 valid_sum_count += curseg_blkoff(sbi, i);
787 }
351df4b2
JK		 788 }
789
9a47938b
FL		 790 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
791 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
792 if (valid_sum_count <= sum_in_page)
351df4b2 793 return 1;
9a47938b
FL		 794 else if ((valid_sum_count - sum_in_page) <=
795 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 796 return 2;
797 return 3;
798}
799
0a8165d7 800/*
351df4b2
JK		 801 * Caller should put this summary page
802 */
803struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
804{
805 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
806}
807
381722d2 808void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 809{
810 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 811 void *dst = page_address(page);
812
813 if (src)
814 memcpy(dst, src, PAGE_CACHE_SIZE);
815 else
816 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 817 set_page_dirty(page);
818 f2fs_put_page(page, 1);
819}
820
381722d2
CY		 821static void write_sum_page(struct f2fs_sb_info *sbi,
822 struct f2fs_summary_block *sum_blk, block_t blk_addr)
823{
824 update_meta_page(sbi, (void *)sum_blk, blk_addr);
825}
826
60374688
JK		 827static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
828{
829 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 830 unsigned int segno = curseg->segno + 1;
60374688
JK		 831 struct free_segmap_info *free_i = FREE_I(sbi);
832
7cd8558b 833 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 834 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 835 return 0;
836}
837
0a8165d7 838/*
351df4b2
JK		 839 * Find a new segment from the free segments bitmap to right order
840 * This function should be returned with success, otherwise BUG
841 */
842static void get_new_segment(struct f2fs_sb_info *sbi,
843 unsigned int *newseg, bool new_sec, int dir)
844{
845 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 846 unsigned int segno, secno, zoneno;
7cd8558b 847 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 848 unsigned int hint = *newseg / sbi->segs_per_sec;
849 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
850 unsigned int left_start = hint;
851 bool init = true;
852 int go_left = 0;
853 int i;
854
1a118ccf 855 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 856
857 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
858 segno = find_next_zero_bit(free_i->free_segmap,
7cd8558b 859 MAIN_SEGS(sbi), *newseg + 1);
33afa7fd
JK		 860 if (segno - *newseg < sbi->segs_per_sec -
861 (*newseg % sbi->segs_per_sec))
351df4b2
JK		 862 goto got_it;
863 }
864find_other_zone:
7cd8558b
JK		 865 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
866 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 867 if (dir == ALLOC_RIGHT) {
868 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 869 MAIN_SECS(sbi), 0);
870 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 871 } else {
872 go_left = 1;
873 left_start = hint - 1;
874 }
875 }
876 if (go_left == 0)
877 goto skip_left;
878
879 while (test_bit(left_start, free_i->free_secmap)) {
880 if (left_start > 0) {
881 left_start--;
882 continue;
883 }
884 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 885 MAIN_SECS(sbi), 0);
886 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 887 break;
888 }
889 secno = left_start;
890skip_left:
891 hint = secno;
892 segno = secno * sbi->segs_per_sec;
893 zoneno = secno / sbi->secs_per_zone;
894
895 /* give up on finding another zone */
896 if (!init)
897 goto got_it;
898 if (sbi->secs_per_zone == 1)
899 goto got_it;
900 if (zoneno == old_zoneno)
901 goto got_it;
902 if (dir == ALLOC_LEFT) {
903 if (!go_left && zoneno + 1 >= total_zones)
904 goto got_it;
905 if (go_left && zoneno == 0)
906 goto got_it;
907 }
908 for (i = 0; i < NR_CURSEG_TYPE; i++)
909 if (CURSEG_I(sbi, i)->zone == zoneno)
910 break;
911
912 if (i < NR_CURSEG_TYPE) {
913 /* zone is in user, try another */
914 if (go_left)
915 hint = zoneno * sbi->secs_per_zone - 1;
916 else if (zoneno + 1 >= total_zones)
917 hint = 0;
918 else
919 hint = (zoneno + 1) * sbi->secs_per_zone;
920 init = false;
921 goto find_other_zone;
922 }
923got_it:
924 /* set it as dirty segment in free segmap */
9850cf4a 925 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 926 __set_inuse(sbi, segno);
927 *newseg = segno;
1a118ccf 928 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 929}
930
931static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
932{
933 struct curseg_info *curseg = CURSEG_I(sbi, type);
934 struct summary_footer *sum_footer;
935
936 curseg->segno = curseg->next_segno;
937 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
938 curseg->next_blkoff = 0;
939 curseg->next_segno = NULL_SEGNO;
940
941 sum_footer = &(curseg->sum_blk->footer);
942 memset(sum_footer, 0, sizeof(struct summary_footer));
943 if (IS_DATASEG(type))
944 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
945 if (IS_NODESEG(type))
946 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
947 __set_sit_entry_type(sbi, type, curseg->segno, modified);
948}
949
0a8165d7 950/*
351df4b2
JK		 951 * Allocate a current working segment.
952 * This function always allocates a free segment in LFS manner.
953 */
954static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
955{
956 struct curseg_info *curseg = CURSEG_I(sbi, type);
957 unsigned int segno = curseg->segno;
958 int dir = ALLOC_LEFT;
959
960 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 961 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 962 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
963 dir = ALLOC_RIGHT;
964
965 if (test_opt(sbi, NOHEAP))
966 dir = ALLOC_RIGHT;
967
968 get_new_segment(sbi, &segno, new_sec, dir);
969 curseg->next_segno = segno;
970 reset_curseg(sbi, type, 1);
971 curseg->alloc_type = LFS;
972}
973
974static void __next_free_blkoff(struct f2fs_sb_info *sbi,
975 struct curseg_info *seg, block_t start)
976{
977 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 978 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 979 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 980 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
981 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
982 int i, pos;
983
984 for (i = 0; i < entries; i++)
985 target_map[i] = ckpt_map[i] | cur_map[i];
986
987 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
988
989 seg->next_blkoff = pos;
351df4b2
JK		 990}
991
0a8165d7 992/*
351df4b2
JK		 993 * If a segment is written by LFS manner, next block offset is just obtained
994 * by increasing the current block offset. However, if a segment is written by
995 * SSR manner, next block offset obtained by calling __next_free_blkoff
996 */
997static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
998 struct curseg_info *seg)
999{
1000 if (seg->alloc_type == SSR)
1001 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1002 else
1003 seg->next_blkoff++;
1004}
1005
0a8165d7 1006/*
e1c42045 1007 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1008 * manner, so it should recover the existing segment information of valid blocks
1009 */
1010static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1011{
1012 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1013 struct curseg_info *curseg = CURSEG_I(sbi, type);
1014 unsigned int new_segno = curseg->next_segno;
1015 struct f2fs_summary_block *sum_node;
1016 struct page *sum_page;
1017
1018 write_sum_page(sbi, curseg->sum_blk,
1019 GET_SUM_BLOCK(sbi, curseg->segno));
1020 __set_test_and_inuse(sbi, new_segno);
1021
1022 mutex_lock(&dirty_i->seglist_lock);
1023 __remove_dirty_segment(sbi, new_segno, PRE);
1024 __remove_dirty_segment(sbi, new_segno, DIRTY);
1025 mutex_unlock(&dirty_i->seglist_lock);
1026
1027 reset_curseg(sbi, type, 1);
1028 curseg->alloc_type = SSR;
1029 __next_free_blkoff(sbi, curseg, 0);
1030
1031 if (reuse) {
1032 sum_page = get_sum_page(sbi, new_segno);
1033 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1034 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1035 f2fs_put_page(sum_page, 1);
1036 }
1037}
1038
43727527
JK		 1039static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1040{
1041 struct curseg_info *curseg = CURSEG_I(sbi, type);
1042 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1043
1044 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1045 return v_ops->get_victim(sbi,
1046 &(curseg)->next_segno, BG_GC, type, SSR);
1047
1048 /* For data segments, let's do SSR more intensively */
1049 for (; type >= CURSEG_HOT_DATA; type--)
1050 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1051 BG_GC, type, SSR))
1052 return 1;
1053 return 0;
1054}
1055
351df4b2
JK		 1056/*
1057 * flush out current segment and replace it with new segment
1058 * This function should be returned with success, otherwise BUG
1059 */
1060static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1061 int type, bool force)
1062{
1063 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1064
7b405275 1065 if (force)
351df4b2 1066 new_curseg(sbi, type, true);
7b405275 1067 else if (type == CURSEG_WARM_NODE)
351df4b2 1068 new_curseg(sbi, type, false);
60374688
JK		 1069 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1070 new_curseg(sbi, type, false);
351df4b2
JK		 1071 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1072 change_curseg(sbi, type, true);
1073 else
1074 new_curseg(sbi, type, false);
dcdfff65
JK		 1075
1076 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1077}
1078
38aa0889
JK		 1079static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1080{
1081 struct curseg_info *curseg = CURSEG_I(sbi, type);
1082 unsigned int old_segno;
1083
1084 old_segno = curseg->segno;
1085 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1086 locate_dirty_segment(sbi, old_segno);
1087}
1088
351df4b2
JK		 1089void allocate_new_segments(struct f2fs_sb_info *sbi)
1090{
351df4b2
JK		 1091 int i;
1092
38aa0889
JK		 1093 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1094 __allocate_new_segments(sbi, i);
351df4b2
JK		 1095}
1096
1097static const struct segment_allocation default_salloc_ops = {
1098 .allocate_segment = allocate_segment_by_default,
1099};
1100
4b2fecc8
JK		 1101int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1102{
f7ef9b83
JK		 1103 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1104 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1105 unsigned int start_segno, end_segno;
1106 struct cp_control cpc;
1107
836b5a63 1108 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1109 return -EINVAL;
1110
9bd27ae4 1111 cpc.trimmed = 0;
7cd8558b 1112 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1113 goto out;
1114
1115 /* start/end segment number in main_area */
7cd8558b
JK		 1116 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1117 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1118 GET_SEGNO(sbi, end);
4b2fecc8 1119 cpc.reason = CP_DISCARD;
836b5a63 1120 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1121
1122 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1123 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1124 cpc.trim_start = start_segno;
a66cdd98
JK		 1125
1126 if (sbi->discard_blks == 0)
1127 break;
1128 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1129 cpc.trim_end = end_segno;
1130 else
1131 cpc.trim_end = min_t(unsigned int,
1132 rounddown(start_segno +
bba681cb
JK		 1133 BATCHED_TRIM_SEGMENTS(sbi),
1134 sbi->segs_per_sec) - 1, end_segno);
1135
1136 mutex_lock(&sbi->gc_mutex);
1137 write_checkpoint(sbi, &cpc);
1138 mutex_unlock(&sbi->gc_mutex);
1139 }
4b2fecc8 1140out:
f7ef9b83 1141 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
4b2fecc8
JK		 1142 return 0;
1143}
1144
351df4b2
JK		 1145static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1146{
1147 struct curseg_info *curseg = CURSEG_I(sbi, type);
1148 if (curseg->next_blkoff < sbi->blocks_per_seg)
1149 return true;
1150 return false;
1151}
1152
1153static int __get_segment_type_2(struct page *page, enum page_type p_type)
1154{
1155 if (p_type == DATA)
1156 return CURSEG_HOT_DATA;
1157 else
1158 return CURSEG_HOT_NODE;
1159}
1160
1161static int __get_segment_type_4(struct page *page, enum page_type p_type)
1162{
1163 if (p_type == DATA) {
1164 struct inode *inode = page->mapping->host;
1165
1166 if (S_ISDIR(inode->i_mode))
1167 return CURSEG_HOT_DATA;
1168 else
1169 return CURSEG_COLD_DATA;
1170 } else {
a344b9fd
JK		 1171 if (IS_DNODE(page) && is_cold_node(page))
1172 return CURSEG_WARM_NODE;
351df4b2
JK		 1173 else
1174 return CURSEG_COLD_NODE;
1175 }
1176}
1177
1178static int __get_segment_type_6(struct page *page, enum page_type p_type)
1179{
1180 if (p_type == DATA) {
1181 struct inode *inode = page->mapping->host;
1182
1183 if (S_ISDIR(inode->i_mode))
1184 return CURSEG_HOT_DATA;
354a3399 1185 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1186 return CURSEG_COLD_DATA;
1187 else
1188 return CURSEG_WARM_DATA;
1189 } else {
1190 if (IS_DNODE(page))
1191 return is_cold_node(page) ? CURSEG_WARM_NODE :
1192 CURSEG_HOT_NODE;
1193 else
1194 return CURSEG_COLD_NODE;
1195 }
1196}
1197
1198static int __get_segment_type(struct page *page, enum page_type p_type)
1199{
4081363f 1200 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1201 case 2:
1202 return __get_segment_type_2(page, p_type);
1203 case 4:
1204 return __get_segment_type_4(page, p_type);
351df4b2 1205 }
12a67146 1206 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1207 f2fs_bug_on(F2FS_P_SB(page),
1208 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1209 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1210}
1211
bfad7c2d
JK		 1212void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1213 block_t old_blkaddr, block_t *new_blkaddr,
1214 struct f2fs_summary *sum, int type)
351df4b2
JK		 1215{
1216 struct sit_info *sit_i = SIT_I(sbi);
1217 struct curseg_info *curseg;
38aa0889
JK		 1218 bool direct_io = (type == CURSEG_DIRECT_IO);
1219
1220 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1221
351df4b2
JK		 1222 curseg = CURSEG_I(sbi, type);
1223
1224 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1225 mutex_lock(&sit_i->sentry_lock);
351df4b2 1226
38aa0889
JK		 1227 /* direct_io'ed data is aligned to the segment for better performance */
1228 if (direct_io && curseg->next_blkoff)
1229 __allocate_new_segments(sbi, type);
1230
351df4b2 1231 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1232
1233 /*
1234 * __add_sum_entry should be resided under the curseg_mutex
1235 * because, this function updates a summary entry in the
1236 * current summary block.
1237 */
e79efe3b 1238 __add_sum_entry(sbi, type, sum);
351df4b2 1239
351df4b2 1240 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1241
1242 stat_inc_block_count(sbi, curseg);
351df4b2 1243
5e443818
JK		 1244 if (!__has_curseg_space(sbi, type))
1245 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1246 /*
1247 * SIT information should be updated before segment allocation,
1248 * since SSR needs latest valid block information.
1249 */
1250 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1251
351df4b2
JK		 1252 mutex_unlock(&sit_i->sentry_lock);
1253
bfad7c2d 1254 if (page && IS_NODESEG(type))
351df4b2
JK		 1255 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1256
bfad7c2d
JK		 1257 mutex_unlock(&curseg->curseg_mutex);
1258}
1259
05ca3632 1260static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1261{
05ca3632 1262 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1263
05ca3632
JK		 1264 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1265 &fio->blk_addr, sum, type);
bfad7c2d 1266
351df4b2 1267 /* writeout dirty page into bdev */
05ca3632 1268 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1269}
1270
577e3495 1271void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1272{
458e6197 1273 struct f2fs_io_info fio = {
05ca3632 1274 .sbi = sbi,
458e6197 1275 .type = META,
cf04e8eb
JK		 1276 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1277 .blk_addr = page->index,
05ca3632 1278 .page = page,
4375a336 1279 .encrypted_page = NULL,
458e6197
JK		 1280 };
1281
351df4b2 1282 set_page_writeback(page);
05ca3632 1283 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1284}
1285
05ca3632 1286void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1287{
1288 struct f2fs_summary sum;
05ca3632 1289
351df4b2 1290 set_summary(&sum, nid, 0, 0);
05ca3632 1291 do_write_page(&sum, fio);
351df4b2
JK		 1292}
1293
05ca3632 1294void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1295{
05ca3632 1296 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1297 struct f2fs_summary sum;
1298 struct node_info ni;
1299
9850cf4a 1300 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1301 get_node_info(sbi, dn->nid, &ni);
1302 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1303 do_write_page(&sum, fio);
e1509cf2 1304 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1305}
1306
05ca3632 1307void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1308{
05ca3632
JK		 1309 stat_inc_inplace_blocks(fio->sbi);
1310 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1311}
1312
528e3459
CY		 1313static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1314 struct f2fs_summary *sum,
19f106bc
CY		 1315 block_t old_blkaddr, block_t new_blkaddr,
1316 bool recover_curseg)
351df4b2
JK		 1317{
1318 struct sit_info *sit_i = SIT_I(sbi);
1319 struct curseg_info *curseg;
1320 unsigned int segno, old_cursegno;
1321 struct seg_entry *se;
1322 int type;
19f106bc 1323 unsigned short old_blkoff;
351df4b2
JK		 1324
1325 segno = GET_SEGNO(sbi, new_blkaddr);
1326 se = get_seg_entry(sbi, segno);
1327 type = se->type;
1328
19f106bc
CY		 1329 if (!recover_curseg) {
1330 /* for recovery flow */
1331 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1332 if (old_blkaddr == NULL_ADDR)
1333 type = CURSEG_COLD_DATA;
1334 else
1335 type = CURSEG_WARM_DATA;
1336 }
1337 } else {
1338 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1339 type = CURSEG_WARM_DATA;
1340 }
19f106bc 1341
351df4b2
JK		 1342 curseg = CURSEG_I(sbi, type);
1343
1344 mutex_lock(&curseg->curseg_mutex);
1345 mutex_lock(&sit_i->sentry_lock);
1346
1347 old_cursegno = curseg->segno;
19f106bc 1348 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1349
1350 /* change the current segment */
1351 if (segno != curseg->segno) {
1352 curseg->next_segno = segno;
1353 change_curseg(sbi, type, true);
1354 }
1355
491c0854 1356 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1357 __add_sum_entry(sbi, type, sum);
351df4b2
JK		 1358
1359 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
351df4b2 1360 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1361
19f106bc
CY		 1362 if (recover_curseg) {
1363 if (old_cursegno != curseg->segno) {
1364 curseg->next_segno = old_cursegno;
1365 change_curseg(sbi, type, true);
1366 }
1367 curseg->next_blkoff = old_blkoff;
1368 }
1369
351df4b2
JK		 1370 mutex_unlock(&sit_i->sentry_lock);
1371 mutex_unlock(&curseg->curseg_mutex);
1372}
1373
528e3459
CY		 1374void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1375 block_t old_addr, block_t new_addr,
1376 unsigned char version, bool recover_curseg)
1377{
1378 struct f2fs_summary sum;
1379
1380 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1381
1382 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1383
1384 dn->data_blkaddr = new_addr;
1385 set_data_blkaddr(dn);
1386 f2fs_update_extent_cache(dn);
1387}
1388
df0f8dc0
CY		 1389static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1390 struct page *page, enum page_type type)
1391{
1392 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1393 struct f2fs_bio_info *io = &sbi->write_io[btype];
df0f8dc0 1394 struct bio_vec *bvec;
4375a336 1395 struct page *target;
df0f8dc0
CY		 1396 int i;
1397
1398 down_read(&io->io_rwsem);
4375a336
JK		 1399 if (!io->bio) {
1400 up_read(&io->io_rwsem);
1401 return false;
1402 }
df0f8dc0 1403
ce23447f 1404 bio_for_each_segment_all(bvec, io->bio, i) {
4375a336
JK		 1405
1406 if (bvec->bv_page->mapping) {
1407 target = bvec->bv_page;
1408 } else {
1409 struct f2fs_crypto_ctx *ctx;
1410
1411 /* encrypted page */
1412 ctx = (struct f2fs_crypto_ctx *)page_private(
1413 bvec->bv_page);
ca40b030 1414 target = ctx->w.control_page;
4375a336
JK		 1415 }
1416
1417 if (page == target) {
df0f8dc0
CY		 1418 up_read(&io->io_rwsem);
1419 return true;
1420 }
1421 }
1422
df0f8dc0
CY		 1423 up_read(&io->io_rwsem);
1424 return false;
1425}
1426
93dfe2ac 1427void f2fs_wait_on_page_writeback(struct page *page,
5514f0aa 1428 enum page_type type)
93dfe2ac 1429{
93dfe2ac 1430 if (PageWriteback(page)) {
4081363f
JK		 1431 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1432
df0f8dc0
CY		 1433 if (is_merged_page(sbi, page, type))
1434 f2fs_submit_merged_bio(sbi, type, WRITE);
93dfe2ac
JK		 1435 wait_on_page_writeback(page);
1436 }
1437}
1438
351df4b2
JK		 1439static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1440{
1441 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1442 struct curseg_info *seg_i;
1443 unsigned char *kaddr;
1444 struct page *page;
1445 block_t start;
1446 int i, j, offset;
1447
1448 start = start_sum_block(sbi);
1449
1450 page = get_meta_page(sbi, start++);
1451 kaddr = (unsigned char *)page_address(page);
1452
1453 /* Step 1: restore nat cache */
1454 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1455 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1456
1457 /* Step 2: restore sit cache */
1458 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1459 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1460 SUM_JOURNAL_SIZE);
1461 offset = 2 * SUM_JOURNAL_SIZE;
1462
1463 /* Step 3: restore summary entries */
1464 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1465 unsigned short blk_off;
1466 unsigned int segno;
1467
1468 seg_i = CURSEG_I(sbi, i);
1469 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1470 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1471 seg_i->next_segno = segno;
1472 reset_curseg(sbi, i, 0);
1473 seg_i->alloc_type = ckpt->alloc_type[i];
1474 seg_i->next_blkoff = blk_off;
1475
1476 if (seg_i->alloc_type == SSR)
1477 blk_off = sbi->blocks_per_seg;
1478
1479 for (j = 0; j < blk_off; j++) {
1480 struct f2fs_summary *s;
1481 s = (struct f2fs_summary *)(kaddr + offset);
1482 seg_i->sum_blk->entries[j] = *s;
1483 offset += SUMMARY_SIZE;
1484 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1485 SUM_FOOTER_SIZE)
1486 continue;
1487
1488 f2fs_put_page(page, 1);
1489 page = NULL;
1490
1491 page = get_meta_page(sbi, start++);
1492 kaddr = (unsigned char *)page_address(page);
1493 offset = 0;
1494 }
1495 }
1496 f2fs_put_page(page, 1);
1497 return 0;
1498}
1499
1500static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1501{
1502 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1503 struct f2fs_summary_block *sum;
1504 struct curseg_info *curseg;
1505 struct page *new;
1506 unsigned short blk_off;
1507 unsigned int segno = 0;
1508 block_t blk_addr = 0;
1509
1510 /* get segment number and block addr */
1511 if (IS_DATASEG(type)) {
1512 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1513 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1514 CURSEG_HOT_DATA]);
119ee914 1515 if (__exist_node_summaries(sbi))
351df4b2
JK		 1516 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1517 else
1518 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1519 } else {
1520 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1521 CURSEG_HOT_NODE]);
1522 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1523 CURSEG_HOT_NODE]);
119ee914 1524 if (__exist_node_summaries(sbi))
351df4b2
JK		 1525 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1526 type - CURSEG_HOT_NODE);
1527 else
1528 blk_addr = GET_SUM_BLOCK(sbi, segno);
1529 }
1530
1531 new = get_meta_page(sbi, blk_addr);
1532 sum = (struct f2fs_summary_block *)page_address(new);
1533
1534 if (IS_NODESEG(type)) {
119ee914 1535 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1536 struct f2fs_summary *ns = &sum->entries[0];
1537 int i;
1538 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1539 ns->version = 0;
1540 ns->ofs_in_node = 0;
1541 }
1542 } else {
d653788a
GZ		 1543 int err;
1544
1545 err = restore_node_summary(sbi, segno, sum);
1546 if (err) {
351df4b2 1547 f2fs_put_page(new, 1);
d653788a 1548 return err;
351df4b2
JK		 1549 }
1550 }
1551 }
1552
1553 /* set uncompleted segment to curseg */
1554 curseg = CURSEG_I(sbi, type);
1555 mutex_lock(&curseg->curseg_mutex);
1556 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1557 curseg->next_segno = segno;
1558 reset_curseg(sbi, type, 0);
1559 curseg->alloc_type = ckpt->alloc_type[type];
1560 curseg->next_blkoff = blk_off;
1561 mutex_unlock(&curseg->curseg_mutex);
1562 f2fs_put_page(new, 1);
1563 return 0;
1564}
1565
1566static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1567{
1568 int type = CURSEG_HOT_DATA;
e4fc5fbf 1569 int err;
351df4b2 1570
25ca923b 1571 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1572 int npages = npages_for_summary_flush(sbi, true);
1573
1574 if (npages >= 2)
1575 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1576 META_CP);
1577
351df4b2
JK		 1578 /* restore for compacted data summary */
1579 if (read_compacted_summaries(sbi))
1580 return -EINVAL;
1581 type = CURSEG_HOT_NODE;
1582 }
1583
119ee914 1584 if (__exist_node_summaries(sbi))
3fa06d7b
CY		 1585 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1586 NR_CURSEG_TYPE - type, META_CP);
1587
e4fc5fbf
CY		 1588 for (; type <= CURSEG_COLD_NODE; type++) {
1589 err = read_normal_summaries(sbi, type);
1590 if (err)
1591 return err;
1592 }
1593
351df4b2
JK		 1594 return 0;
1595}
1596
1597static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1598{
1599 struct page *page;
1600 unsigned char *kaddr;
1601 struct f2fs_summary *summary;
1602 struct curseg_info *seg_i;
1603 int written_size = 0;
1604 int i, j;
1605
1606 page = grab_meta_page(sbi, blkaddr++);
1607 kaddr = (unsigned char *)page_address(page);
1608
1609 /* Step 1: write nat cache */
1610 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1611 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1612 written_size += SUM_JOURNAL_SIZE;
1613
1614 /* Step 2: write sit cache */
1615 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1616 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1617 SUM_JOURNAL_SIZE);
1618 written_size += SUM_JOURNAL_SIZE;
1619
351df4b2
JK		 1620 /* Step 3: write summary entries */
1621 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1622 unsigned short blkoff;
1623 seg_i = CURSEG_I(sbi, i);
1624 if (sbi->ckpt->alloc_type[i] == SSR)
1625 blkoff = sbi->blocks_per_seg;
1626 else
1627 blkoff = curseg_blkoff(sbi, i);
1628
1629 for (j = 0; j < blkoff; j++) {
1630 if (!page) {
1631 page = grab_meta_page(sbi, blkaddr++);
1632 kaddr = (unsigned char *)page_address(page);
1633 written_size = 0;
1634 }
1635 summary = (struct f2fs_summary *)(kaddr + written_size);
1636 *summary = seg_i->sum_blk->entries[j];
1637 written_size += SUMMARY_SIZE;
351df4b2
JK		 1638
1639 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1640 SUM_FOOTER_SIZE)
1641 continue;
1642
e8d61a74 1643 set_page_dirty(page);
351df4b2
JK		 1644 f2fs_put_page(page, 1);
1645 page = NULL;
1646 }
1647 }
e8d61a74
CY		 1648 if (page) {
1649 set_page_dirty(page);
351df4b2 1650 f2fs_put_page(page, 1);
e8d61a74 1651 }
351df4b2
JK		 1652}
1653
1654static void write_normal_summaries(struct f2fs_sb_info *sbi,
1655 block_t blkaddr, int type)
1656{
1657 int i, end;
1658 if (IS_DATASEG(type))
1659 end = type + NR_CURSEG_DATA_TYPE;
1660 else
1661 end = type + NR_CURSEG_NODE_TYPE;
1662
1663 for (i = type; i < end; i++) {
1664 struct curseg_info *sum = CURSEG_I(sbi, i);
1665 mutex_lock(&sum->curseg_mutex);
1666 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1667 mutex_unlock(&sum->curseg_mutex);
1668 }
1669}
1670
1671void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1672{
25ca923b 1673 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1674 write_compacted_summaries(sbi, start_blk);
1675 else
1676 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1677}
1678
1679void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1680{
119ee914 1681 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1682}
1683
1684int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1685 unsigned int val, int alloc)
1686{
1687 int i;
1688
1689 if (type == NAT_JOURNAL) {
1690 for (i = 0; i < nats_in_cursum(sum); i++) {
1691 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1692 return i;
1693 }
1694 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1695 return update_nats_in_cursum(sum, 1);
1696 } else if (type == SIT_JOURNAL) {
1697 for (i = 0; i < sits_in_cursum(sum); i++)
1698 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1699 return i;
1700 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1701 return update_sits_in_cursum(sum, 1);
1702 }
1703 return -1;
1704}
1705
1706static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1707 unsigned int segno)
1708{
2cc22186 1709 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1710}
1711
1712static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1713 unsigned int start)
1714{
1715 struct sit_info *sit_i = SIT_I(sbi);
1716 struct page *src_page, *dst_page;
1717 pgoff_t src_off, dst_off;
1718 void *src_addr, *dst_addr;
1719
1720 src_off = current_sit_addr(sbi, start);
1721 dst_off = next_sit_addr(sbi, src_off);
1722
1723 /* get current sit block page without lock */
1724 src_page = get_meta_page(sbi, src_off);
1725 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1726 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1727
1728 src_addr = page_address(src_page);
1729 dst_addr = page_address(dst_page);
1730 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1731
1732 set_page_dirty(dst_page);
1733 f2fs_put_page(src_page, 1);
1734
1735 set_to_next_sit(sit_i, start);
1736
1737 return dst_page;
1738}
1739
184a5cd2
CY		 1740static struct sit_entry_set *grab_sit_entry_set(void)
1741{
1742 struct sit_entry_set *ses =
1743 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1744
1745 ses->entry_cnt = 0;
1746 INIT_LIST_HEAD(&ses->set_list);
1747 return ses;
1748}
1749
1750static void release_sit_entry_set(struct sit_entry_set *ses)
1751{
1752 list_del(&ses->set_list);
1753 kmem_cache_free(sit_entry_set_slab, ses);
1754}
1755
1756static void adjust_sit_entry_set(struct sit_entry_set *ses,
1757 struct list_head *head)
1758{
1759 struct sit_entry_set *next = ses;
1760
1761 if (list_is_last(&ses->set_list, head))
1762 return;
1763
1764 list_for_each_entry_continue(next, head, set_list)
1765 if (ses->entry_cnt <= next->entry_cnt)
1766 break;
1767
1768 list_move_tail(&ses->set_list, &next->set_list);
1769}
1770
1771static void add_sit_entry(unsigned int segno, struct list_head *head)
1772{
1773 struct sit_entry_set *ses;
1774 unsigned int start_segno = START_SEGNO(segno);
1775
1776 list_for_each_entry(ses, head, set_list) {
1777 if (ses->start_segno == start_segno) {
1778 ses->entry_cnt++;
1779 adjust_sit_entry_set(ses, head);
1780 return;
1781 }
1782 }
1783
1784 ses = grab_sit_entry_set();
1785
1786 ses->start_segno = start_segno;
1787 ses->entry_cnt++;
1788 list_add(&ses->set_list, head);
1789}
1790
1791static void add_sits_in_set(struct f2fs_sb_info *sbi)
1792{
1793 struct f2fs_sm_info *sm_info = SM_I(sbi);
1794 struct list_head *set_list = &sm_info->sit_entry_set;
1795 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1796 unsigned int segno;
1797
7cd8558b 1798 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1799 add_sit_entry(segno, set_list);
1800}
1801
1802static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1803{
1804 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1805 struct f2fs_summary_block *sum = curseg->sum_blk;
1806 int i;
1807
184a5cd2
CY		 1808 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1809 unsigned int segno;
1810 bool dirtied;
1811
1812 segno = le32_to_cpu(segno_in_journal(sum, i));
1813 dirtied = __mark_sit_entry_dirty(sbi, segno);
1814
1815 if (!dirtied)
1816 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1817 }
184a5cd2 1818 update_sits_in_cursum(sum, -sits_in_cursum(sum));
351df4b2
JK		 1819}
1820
0a8165d7 1821/*
351df4b2
JK		 1822 * CP calls this function, which flushes SIT entries including sit_journal,
1823 * and moves prefree segs to free segs.
1824 */
4b2fecc8 1825void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1826{
1827 struct sit_info *sit_i = SIT_I(sbi);
1828 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1829 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1830 struct f2fs_summary_block *sum = curseg->sum_blk;
184a5cd2
CY		 1831 struct sit_entry_set *ses, *tmp;
1832 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1833 bool to_journal = true;
4b2fecc8 1834 struct seg_entry *se;
351df4b2
JK		 1835
1836 mutex_lock(&curseg->curseg_mutex);
1837 mutex_lock(&sit_i->sentry_lock);
1838
2b11a74b
WL		 1839 if (!sit_i->dirty_sentries)
1840 goto out;
1841
351df4b2 1842 /*
184a5cd2
CY		 1843 * add and account sit entries of dirty bitmap in sit entry
1844 * set temporarily
351df4b2 1845 */
184a5cd2 1846 add_sits_in_set(sbi);
351df4b2 1847
184a5cd2
CY		 1848 /*
1849 * if there are no enough space in journal to store dirty sit
1850 * entries, remove all entries from journal and add and account
1851 * them in sit entry set.
1852 */
1853 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1854 remove_sits_in_journal(sbi);
b2955550 1855
184a5cd2
CY		 1856 /*
1857 * there are two steps to flush sit entries:
1858 * #1, flush sit entries to journal in current cold data summary block.
1859 * #2, flush sit entries to sit page.
1860 */
1861 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1862 struct page *page = NULL;
184a5cd2
CY		 1863 struct f2fs_sit_block *raw_sit = NULL;
1864 unsigned int start_segno = ses->start_segno;
1865 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1866 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1867 unsigned int segno = start_segno;
1868
1869 if (to_journal &&
1870 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1871 to_journal = false;
1872
1873 if (!to_journal) {
1874 page = get_next_sit_page(sbi, start_segno);
1875 raw_sit = page_address(page);
351df4b2 1876 }
351df4b2 1877
184a5cd2
CY		 1878 /* flush dirty sit entries in region of current sit set */
1879 for_each_set_bit_from(segno, bitmap, end) {
1880 int offset, sit_offset;
4b2fecc8
JK		 1881
1882 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1883
1884 /* add discard candidates */
d7bc2484 1885 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1886 cpc->trim_start = segno;
1887 add_discard_addrs(sbi, cpc);
1888 }
184a5cd2
CY		 1889
1890 if (to_journal) {
1891 offset = lookup_journal_in_cursum(sum,
1892 SIT_JOURNAL, segno, 1);
1893 f2fs_bug_on(sbi, offset < 0);
1894 segno_in_journal(sum, offset) =
1895 cpu_to_le32(segno);
1896 seg_info_to_raw_sit(se,
1897 &sit_in_journal(sum, offset));
1898 } else {
1899 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1900 seg_info_to_raw_sit(se,
1901 &raw_sit->entries[sit_offset]);
1902 }
351df4b2 1903
184a5cd2
CY		 1904 __clear_bit(segno, bitmap);
1905 sit_i->dirty_sentries--;
1906 ses->entry_cnt--;
351df4b2
JK		 1907 }
1908
184a5cd2
CY		 1909 if (!to_journal)
1910 f2fs_put_page(page, 1);
1911
1912 f2fs_bug_on(sbi, ses->entry_cnt);
1913 release_sit_entry_set(ses);
351df4b2 1914 }
184a5cd2
CY		 1915
1916 f2fs_bug_on(sbi, !list_empty(head));
1917 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 1918out:
4b2fecc8
JK		 1919 if (cpc->reason == CP_DISCARD) {
1920 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1921 add_discard_addrs(sbi, cpc);
1922 }
351df4b2
JK		 1923 mutex_unlock(&sit_i->sentry_lock);
1924 mutex_unlock(&curseg->curseg_mutex);
1925
351df4b2
JK		 1926 set_prefree_as_free_segments(sbi);
1927}
1928
1929static int build_sit_info(struct f2fs_sb_info *sbi)
1930{
1931 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1932 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1933 struct sit_info *sit_i;
1934 unsigned int sit_segs, start;
1935 char *src_bitmap, *dst_bitmap;
1936 unsigned int bitmap_size;
1937
1938 /* allocate memory for SIT information */
1939 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1940 if (!sit_i)
1941 return -ENOMEM;
1942
1943 SM_I(sbi)->sit_info = sit_i;
1944
7cd8558b 1945 sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
351df4b2
JK		 1946 if (!sit_i->sentries)
1947 return -ENOMEM;
1948
7cd8558b 1949 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 1950 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1951 if (!sit_i->dirty_sentries_bitmap)
1952 return -ENOMEM;
1953
7cd8558b 1954 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 1955 sit_i->sentries[start].cur_valid_map
1956 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1957 sit_i->sentries[start].ckpt_valid_map
1958 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 1959 sit_i->sentries[start].discard_map
1960 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1961 if (!sit_i->sentries[start].cur_valid_map ||
1962 !sit_i->sentries[start].ckpt_valid_map ||
1963 !sit_i->sentries[start].discard_map)
351df4b2
JK		 1964 return -ENOMEM;
1965 }
1966
60a3b782
JK		 1967 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1968 if (!sit_i->tmp_map)
1969 return -ENOMEM;
1970
351df4b2 1971 if (sbi->segs_per_sec > 1) {
7cd8558b 1972 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
351df4b2
JK		 1973 sizeof(struct sec_entry));
1974 if (!sit_i->sec_entries)
1975 return -ENOMEM;
1976 }
1977
1978 /* get information related with SIT */
1979 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1980
1981 /* setup SIT bitmap from ckeckpoint pack */
1982 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1983 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1984
79b5793b 1985 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 1986 if (!dst_bitmap)
1987 return -ENOMEM;
351df4b2
JK		 1988
1989 /* init SIT information */
1990 sit_i->s_ops = &default_salloc_ops;
1991
1992 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1993 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1994 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1995 sit_i->sit_bitmap = dst_bitmap;
1996 sit_i->bitmap_size = bitmap_size;
1997 sit_i->dirty_sentries = 0;
1998 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1999 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2000 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2001 mutex_init(&sit_i->sentry_lock);
2002 return 0;
2003}
2004
2005static int build_free_segmap(struct f2fs_sb_info *sbi)
2006{
351df4b2
JK		 2007 struct free_segmap_info *free_i;
2008 unsigned int bitmap_size, sec_bitmap_size;
2009
2010 /* allocate memory for free segmap information */
2011 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2012 if (!free_i)
2013 return -ENOMEM;
2014
2015 SM_I(sbi)->free_info = free_i;
2016
7cd8558b 2017 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2018 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
2019 if (!free_i->free_segmap)
2020 return -ENOMEM;
2021
7cd8558b 2022 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2
JK		 2023 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
2024 if (!free_i->free_secmap)
2025 return -ENOMEM;
2026
2027 /* set all segments as dirty temporarily */
2028 memset(free_i->free_segmap, 0xff, bitmap_size);
2029 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2030
2031 /* init free segmap information */
7cd8558b 2032 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2033 free_i->free_segments = 0;
2034 free_i->free_sections = 0;
1a118ccf 2035 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2036 return 0;
2037}
2038
2039static int build_curseg(struct f2fs_sb_info *sbi)
2040{
1042d60f 2041 struct curseg_info *array;
351df4b2
JK		 2042 int i;
2043
b434babf 2044 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2045 if (!array)
2046 return -ENOMEM;
2047
2048 SM_I(sbi)->curseg_array = array;
2049
2050 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2051 mutex_init(&array[i].curseg_mutex);
2052 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2053 if (!array[i].sum_blk)
2054 return -ENOMEM;
2055 array[i].segno = NULL_SEGNO;
2056 array[i].next_blkoff = 0;
2057 }
2058 return restore_curseg_summaries(sbi);
2059}
2060
2061static void build_sit_entries(struct f2fs_sb_info *sbi)
2062{
2063 struct sit_info *sit_i = SIT_I(sbi);
2064 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2065 struct f2fs_summary_block *sum = curseg->sum_blk;
74de593a
CY		 2066 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2067 unsigned int i, start, end;
2068 unsigned int readed, start_blk = 0;
90a893c7 2069 int nrpages = MAX_BIO_BLOCKS(sbi);
351df4b2 2070
74de593a 2071 do {
662befda 2072 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
74de593a
CY		 2073
2074 start = start_blk * sit_i->sents_per_block;
2075 end = (start_blk + readed) * sit_i->sents_per_block;
2076
7cd8558b 2077 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2078 struct seg_entry *se = &sit_i->sentries[start];
2079 struct f2fs_sit_block *sit_blk;
2080 struct f2fs_sit_entry sit;
2081 struct page *page;
2082
2083 mutex_lock(&curseg->curseg_mutex);
2084 for (i = 0; i < sits_in_cursum(sum); i++) {
6c311ec6
CF		 2085 if (le32_to_cpu(segno_in_journal(sum, i))
2086 == start) {
74de593a
CY		 2087 sit = sit_in_journal(sum, i);
2088 mutex_unlock(&curseg->curseg_mutex);
2089 goto got_it;
2090 }
351df4b2 2091 }
74de593a
CY		 2092 mutex_unlock(&curseg->curseg_mutex);
2093
2094 page = get_current_sit_page(sbi, start);
2095 sit_blk = (struct f2fs_sit_block *)page_address(page);
2096 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2097 f2fs_put_page(page, 1);
351df4b2 2098got_it:
74de593a
CY		 2099 check_block_count(sbi, start, &sit);
2100 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2101
2102 /* build discard map only one time */
2103 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2104 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2105
74de593a
CY		 2106 if (sbi->segs_per_sec > 1) {
2107 struct sec_entry *e = get_sec_entry(sbi, start);
2108 e->valid_blocks += se->valid_blocks;
2109 }
351df4b2 2110 }
74de593a
CY		 2111 start_blk += readed;
2112 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2113}
2114
2115static void init_free_segmap(struct f2fs_sb_info *sbi)
2116{
2117 unsigned int start;
2118 int type;
2119
7cd8558b 2120 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2121 struct seg_entry *sentry = get_seg_entry(sbi, start);
2122 if (!sentry->valid_blocks)
2123 __set_free(sbi, start);
2124 }
2125
2126 /* set use the current segments */
2127 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2128 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2129 __set_test_and_inuse(sbi, curseg_t->segno);
2130 }
2131}
2132
2133static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2134{
2135 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2136 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2137 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2138 unsigned short valid_blocks;
2139
8736fbf0 2140 while (1) {
351df4b2 2141 /* find dirty segment based on free segmap */
7cd8558b
JK		 2142 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2143 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2144 break;
2145 offset = segno + 1;
2146 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2147 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2148 continue;
ec325b52
JK		 2149 if (valid_blocks > sbi->blocks_per_seg) {
2150 f2fs_bug_on(sbi, 1);
2151 continue;
2152 }
351df4b2
JK		 2153 mutex_lock(&dirty_i->seglist_lock);
2154 __locate_dirty_segment(sbi, segno, DIRTY);
2155 mutex_unlock(&dirty_i->seglist_lock);
2156 }
2157}
2158
5ec4e49f 2159static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2160{
2161 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2162 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2163
5ec4e49f
JK		 2164 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2165 if (!dirty_i->victim_secmap)
351df4b2
JK		 2166 return -ENOMEM;
2167 return 0;
2168}
2169
2170static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2171{
2172 struct dirty_seglist_info *dirty_i;
2173 unsigned int bitmap_size, i;
2174
2175 /* allocate memory for dirty segments list information */
2176 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2177 if (!dirty_i)
2178 return -ENOMEM;
2179
2180 SM_I(sbi)->dirty_info = dirty_i;
2181 mutex_init(&dirty_i->seglist_lock);
2182
7cd8558b 2183 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2184
2185 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2186 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2187 if (!dirty_i->dirty_segmap[i])
2188 return -ENOMEM;
2189 }
2190
2191 init_dirty_segmap(sbi);
5ec4e49f 2192 return init_victim_secmap(sbi);
351df4b2
JK		 2193}
2194
0a8165d7 2195/*
351df4b2
JK		 2196 * Update min, max modified time for cost-benefit GC algorithm
2197 */
2198static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2199{
2200 struct sit_info *sit_i = SIT_I(sbi);
2201 unsigned int segno;
2202
2203 mutex_lock(&sit_i->sentry_lock);
2204
2205 sit_i->min_mtime = LLONG_MAX;
2206
7cd8558b 2207 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2208 unsigned int i;
2209 unsigned long long mtime = 0;
2210
2211 for (i = 0; i < sbi->segs_per_sec; i++)
2212 mtime += get_seg_entry(sbi, segno + i)->mtime;
2213
2214 mtime = div_u64(mtime, sbi->segs_per_sec);
2215
2216 if (sit_i->min_mtime > mtime)
2217 sit_i->min_mtime = mtime;
2218 }
2219 sit_i->max_mtime = get_mtime(sbi);
2220 mutex_unlock(&sit_i->sentry_lock);
2221}
2222
2223int build_segment_manager(struct f2fs_sb_info *sbi)
2224{
2225 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2226 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2227 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2228 int err;
2229
2230 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2231 if (!sm_info)
2232 return -ENOMEM;
2233
2234 /* init sm info */
2235 sbi->sm_info = sm_info;
351df4b2
JK		 2236 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2237 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2238 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2239 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2240 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2241 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2242 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2243 sm_info->rec_prefree_segments = sm_info->main_segments *
2244 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2245 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2246 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2247 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2248
7fd9e544
JK		 2249 INIT_LIST_HEAD(&sm_info->discard_list);
2250 sm_info->nr_discards = 0;
2251 sm_info->max_discards = 0;
2252
bba681cb
JK		 2253 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2254
184a5cd2
CY		 2255 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2256
b270ad6f 2257 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2258 err = create_flush_cmd_control(sbi);
2259 if (err)
a688b9d9 2260 return err;
6b4afdd7
JK		 2261 }
2262
351df4b2
JK		 2263 err = build_sit_info(sbi);
2264 if (err)
2265 return err;
2266 err = build_free_segmap(sbi);
2267 if (err)
2268 return err;
2269 err = build_curseg(sbi);
2270 if (err)
2271 return err;
2272
2273 /* reinit free segmap based on SIT */
2274 build_sit_entries(sbi);
2275
2276 init_free_segmap(sbi);
2277 err = build_dirty_segmap(sbi);
2278 if (err)
2279 return err;
2280
2281 init_min_max_mtime(sbi);
2282 return 0;
2283}
2284
2285static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2286 enum dirty_type dirty_type)
2287{
2288 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2289
2290 mutex_lock(&dirty_i->seglist_lock);
2291 kfree(dirty_i->dirty_segmap[dirty_type]);
2292 dirty_i->nr_dirty[dirty_type] = 0;
2293 mutex_unlock(&dirty_i->seglist_lock);
2294}
2295
5ec4e49f 2296static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2297{
2298 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5ec4e49f 2299 kfree(dirty_i->victim_secmap);
351df4b2
JK		 2300}
2301
2302static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2303{
2304 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2305 int i;
2306
2307 if (!dirty_i)
2308 return;
2309
2310 /* discard pre-free/dirty segments list */
2311 for (i = 0; i < NR_DIRTY_TYPE; i++)
2312 discard_dirty_segmap(sbi, i);
2313
5ec4e49f 2314 destroy_victim_secmap(sbi);
351df4b2
JK		 2315 SM_I(sbi)->dirty_info = NULL;
2316 kfree(dirty_i);
2317}
2318
2319static void destroy_curseg(struct f2fs_sb_info *sbi)
2320{
2321 struct curseg_info *array = SM_I(sbi)->curseg_array;
2322 int i;
2323
2324 if (!array)
2325 return;
2326 SM_I(sbi)->curseg_array = NULL;
2327 for (i = 0; i < NR_CURSEG_TYPE; i++)
2328 kfree(array[i].sum_blk);
2329 kfree(array);
2330}
2331
2332static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2333{
2334 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2335 if (!free_i)
2336 return;
2337 SM_I(sbi)->free_info = NULL;
2338 kfree(free_i->free_segmap);
2339 kfree(free_i->free_secmap);
2340 kfree(free_i);
2341}
2342
2343static void destroy_sit_info(struct f2fs_sb_info *sbi)
2344{
2345 struct sit_info *sit_i = SIT_I(sbi);
2346 unsigned int start;
2347
2348 if (!sit_i)
2349 return;
2350
2351 if (sit_i->sentries) {
7cd8558b 2352 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2353 kfree(sit_i->sentries[start].cur_valid_map);
2354 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2355 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2356 }
2357 }
60a3b782
JK		 2358 kfree(sit_i->tmp_map);
2359
351df4b2
JK		 2360 vfree(sit_i->sentries);
2361 vfree(sit_i->sec_entries);
2362 kfree(sit_i->dirty_sentries_bitmap);
2363
2364 SM_I(sbi)->sit_info = NULL;
2365 kfree(sit_i->sit_bitmap);
2366 kfree(sit_i);
2367}
2368
2369void destroy_segment_manager(struct f2fs_sb_info *sbi)
2370{
2371 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2372
3b03f724
CY		 2373 if (!sm_info)
2374 return;
2163d198 2375 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2376 destroy_dirty_segmap(sbi);
2377 destroy_curseg(sbi);
2378 destroy_free_segmap(sbi);
2379 destroy_sit_info(sbi);
2380 sbi->sm_info = NULL;
2381 kfree(sm_info);
2382}
7fd9e544
JK		 2383
2384int __init create_segment_manager_caches(void)
2385{
2386 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2387 sizeof(struct discard_entry));
7fd9e544 2388 if (!discard_entry_slab)
184a5cd2
CY		 2389 goto fail;
2390
2391 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2392 sizeof(struct sit_entry_set));
184a5cd2
CY		 2393 if (!sit_entry_set_slab)
2394 goto destory_discard_entry;
88b88a66
JK		 2395
2396 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2397 sizeof(struct inmem_pages));
2398 if (!inmem_entry_slab)
2399 goto destroy_sit_entry_set;
7fd9e544 2400 return 0;
184a5cd2 2401
88b88a66
JK		 2402destroy_sit_entry_set:
2403 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2404destory_discard_entry:
2405 kmem_cache_destroy(discard_entry_slab);
2406fail:
2407 return -ENOMEM;
7fd9e544
JK		 2408}
2409
2410void destroy_segment_manager_caches(void)
2411{
184a5cd2 2412 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2413 kmem_cache_destroy(discard_entry_slab);
88b88a66 2414 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2415}
ubuntu-zesty-kernel mirror