projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: enhance IO path with block plug
[deliverable/linux.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;
184a5cd2 29static struct kmem_cache *sit_entry_set_slab;
88b88a66 30static struct kmem_cache *inmem_entry_slab;
7fd9e544 31
f96999c3
JK		 32static unsigned long __reverse_ulong(unsigned char *str)
33{
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
36
37#if BITS_PER_LONG == 64
38 shift = 56;
39#endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
43 }
44 return tmp;
45}
46
9a7f143a
CL		 47/*
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
50 */
51static inline unsigned long __reverse_ffs(unsigned long word)
52{
53 int num = 0;
54
55#if BITS_PER_LONG == 64
f96999c3 56 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 57 num += 32;
f96999c3 58 else
9a7f143a 59 word >>= 32;
9a7f143a 60#endif
f96999c3 61 if ((word & 0xffff0000) == 0)
9a7f143a 62 num += 16;
f96999c3 63 else
9a7f143a 64 word >>= 16;
f96999c3
JK		 65
66 if ((word & 0xff00) == 0)
9a7f143a 67 num += 8;
f96999c3 68 else
9a7f143a 69 word >>= 8;
f96999c3 70
9a7f143a
CL		 71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
f96999c3 75
9a7f143a
CL		 76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
f96999c3 80
9a7f143a
CL		 81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
84}
85
86/*
e1c42045 87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 89 * @size must be integral times of unsigned long.
9a7f143a 90 * Example:
f96999c3
JK		 91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 94 */
95static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
97{
98 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 99 unsigned long result = size;
9a7f143a 100 unsigned long tmp;
9a7f143a
CL		 101
102 if (offset >= size)
103 return size;
104
692223d1 105 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 106 offset %= BITS_PER_LONG;
f96999c3 107
692223d1
FL		 108 while (1) {
109 if (*p == 0)
110 goto pass;
9a7f143a 111
f96999c3 112 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 113
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 117 if (tmp)
692223d1
FL		 118 goto found;
119pass:
120 if (size <= BITS_PER_LONG)
121 break;
9a7f143a 122 size -= BITS_PER_LONG;
692223d1 123 offset = 0;
f96999c3 124 p++;
9a7f143a 125 }
692223d1
FL		 126 return result;
127found:
128 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 129}
130
131static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
133{
134 const unsigned long *p = addr + BIT_WORD(offset);
80609448 135 unsigned long result = size;
9a7f143a 136 unsigned long tmp;
9a7f143a
CL		 137
138 if (offset >= size)
139 return size;
140
80609448 141 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 142 offset %= BITS_PER_LONG;
80609448
JK		 143
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
147
f96999c3 148 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 149
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
f96999c3 154 if (tmp != ~0UL)
80609448
JK		 155 goto found;
156pass:
157 if (size <= BITS_PER_LONG)
158 break;
9a7f143a 159 size -= BITS_PER_LONG;
80609448 160 offset = 0;
f96999c3 161 p++;
9a7f143a 162 }
80609448
JK		 163 return result;
164found:
165 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 166}
167
88b88a66
JK		 168void register_inmem_page(struct inode *inode, struct page *page)
169{
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
9be32d72 172
9e4ded3f 173 f2fs_trace_pid(page);
0722b101 174
decd36b6
CY		 175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
177
88b88a66
JK		 178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
179
180 /* add atomic page indices to the list */
181 new->page = page;
182 INIT_LIST_HEAD(&new->list);
decd36b6 183
88b88a66
JK		 184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
186 get_page(page);
187 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 189 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 190
191 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 192}
193
28bc106b
CY		 194static int __revoke_inmem_pages(struct inode *inode,
195 struct list_head *head, bool drop, bool recover)
29b96b54 196{
28bc106b 197 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 198 struct inmem_pages *cur, *tmp;
28bc106b 199 int err = 0;
29b96b54
CY		 200
201 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 202 struct page *page = cur->page;
203
204 if (drop)
205 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
206
207 lock_page(page);
29b96b54 208
28bc106b
CY		 209 if (recover) {
210 struct dnode_of_data dn;
211 struct node_info ni;
212
213 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
214
215 set_new_dnode(&dn, inode, NULL, NULL, 0);
216 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
217 err = -EAGAIN;
218 goto next;
219 }
220 get_node_info(sbi, dn.nid, &ni);
221 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
222 cur->old_addr, ni.version, true, true);
223 f2fs_put_dnode(&dn);
224 }
225next:
226 ClearPageUptodate(page);
227 set_page_private(page, 0);
228 ClearPageUptodate(page);
229 f2fs_put_page(page, 1);
29b96b54
CY		 230
231 list_del(&cur->list);
232 kmem_cache_free(inmem_entry_slab, cur);
233 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
234 }
28bc106b 235 return err;
29b96b54
CY		 236}
237
238void drop_inmem_pages(struct inode *inode)
239{
240 struct f2fs_inode_info *fi = F2FS_I(inode);
241
242 mutex_lock(&fi->inmem_lock);
28bc106b 243 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54
CY		 244 mutex_unlock(&fi->inmem_lock);
245}
246
28bc106b
CY		 247static int __commit_inmem_pages(struct inode *inode,
248 struct list_head *revoke_list)
88b88a66
JK		 249{
250 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
251 struct f2fs_inode_info *fi = F2FS_I(inode);
252 struct inmem_pages *cur, *tmp;
88b88a66 253 struct f2fs_io_info fio = {
05ca3632 254 .sbi = sbi,
88b88a66 255 .type = DATA,
1e84371f 256 .rw = WRITE_SYNC | REQ_PRIO,
4375a336 257 .encrypted_page = NULL,
88b88a66 258 };
29b96b54 259 bool submit_bio = false;
edb27dee 260 int err = 0;
88b88a66 261
88b88a66 262 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 263 struct page *page = cur->page;
264
265 lock_page(page);
266 if (page->mapping == inode->i_mapping) {
267 trace_f2fs_commit_inmem_page(page, INMEM);
268
269 set_page_dirty(page);
270 f2fs_wait_on_page_writeback(page, DATA, true);
271 if (clear_page_dirty_for_io(page))
29b96b54 272 inode_dec_dirty_pages(inode);
28bc106b
CY		 273
274 fio.page = page;
29b96b54
CY		 275 err = do_write_data_page(&fio);
276 if (err) {
28bc106b 277 unlock_page(page);
29b96b54 278 break;
70c640b1 279 }
29b96b54 280
28bc106b
CY		 281 /* record old blkaddr for revoking */
282 cur->old_addr = fio.old_blkaddr;
decd36b6 283
28bc106b
CY		 284 clear_cold_data(page);
285 submit_bio = true;
286 }
287 unlock_page(page);
288 list_move_tail(&cur->list, revoke_list);
88b88a66 289 }
29b96b54
CY		 290
291 if (submit_bio)
292 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
28bc106b
CY		 293
294 if (!err)
295 __revoke_inmem_pages(inode, revoke_list, false, false);
296
29b96b54
CY		 297 return err;
298}
299
300int commit_inmem_pages(struct inode *inode)
301{
302 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
303 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 304 struct list_head revoke_list;
305 int err;
29b96b54 306
28bc106b 307 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 308 f2fs_balance_fs(sbi, true);
309 f2fs_lock_op(sbi);
310
311 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 312 err = __commit_inmem_pages(inode, &revoke_list);
313 if (err) {
314 int ret;
315 /*
316 * try to revoke all committed pages, but still we could fail
317 * due to no memory or other reason, if that happened, EAGAIN
318 * will be returned, which means in such case, transaction is
319 * already not integrity, caller should use journal to do the
320 * recovery or rewrite & commit last transaction. For other
321 * error number, revoking was done by filesystem itself.
322 */
323 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
324 if (ret)
325 err = ret;
326
327 /* drop all uncommitted pages */
328 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
329 }
88b88a66
JK		 330 mutex_unlock(&fi->inmem_lock);
331
29b96b54 332 f2fs_unlock_op(sbi);
edb27dee 333 return err;
88b88a66
JK		 334}
335
0a8165d7 336/*
351df4b2
JK		 337 * This function balances dirty node and dentry pages.
338 * In addition, it controls garbage collection.
339 */
2c4db1a6 340void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 341{
2c4db1a6
JK		 342 if (!need)
343 return;
351df4b2 344 /*
029cd28c
JK		 345 * We should do GC or end up with checkpoint, if there are so many dirty
346 * dir/node pages without enough free segments.
351df4b2 347 */
43727527 348 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 349 mutex_lock(&sbi->gc_mutex);
d530d4d8 350 f2fs_gc(sbi, false);
351df4b2
JK		 351 }
352}
353
4660f9c0
JK		 354void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
355{
1dcc336b 356 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 357 if (!available_free_memory(sbi, EXTENT_CACHE))
358 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 359
1b38dc8e
JK		 360 /* check the # of cached NAT entries */
361 if (!available_free_memory(sbi, NAT_ENTRIES))
362 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
363
31696580
CY		 364 if (!available_free_memory(sbi, FREE_NIDS))
365 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
366
1b38dc8e
JK		 367 /* checkpoint is the only way to shrink partial cached entries */
368 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 369 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 370 excess_prefree_segs(sbi) ||
371 excess_dirty_nats(sbi) ||
d0239e1b 372 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
e9f5b8b8
CY		 373 if (test_opt(sbi, DATA_FLUSH)) {
374 struct blk_plug plug;
375
376 blk_start_plug(&plug);
36b35a0d 377 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 378 blk_finish_plug(&plug);
379 }
4660f9c0 380 f2fs_sync_fs(sbi->sb, true);
42190d2a 381 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 382 }
4660f9c0
JK		 383}
384
2163d198 385static int issue_flush_thread(void *data)
6b4afdd7
JK		 386{
387 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 388 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
389 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 390repeat:
391 if (kthread_should_stop())
392 return 0;
393
721bd4d5 394 if (!llist_empty(&fcc->issue_list)) {
740432f8 395 struct bio *bio;
6b4afdd7
JK		 396 struct flush_cmd *cmd, *next;
397 int ret;
398
740432f8
JK		 399 bio = f2fs_bio_alloc(0);
400
721bd4d5
GZ		 401 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
402 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
403
6b4afdd7
JK		 404 bio->bi_bdev = sbi->sb->s_bdev;
405 ret = submit_bio_wait(WRITE_FLUSH, bio);
406
721bd4d5
GZ		 407 llist_for_each_entry_safe(cmd, next,
408 fcc->dispatch_list, llnode) {
6b4afdd7 409 cmd->ret = ret;
6b4afdd7
JK		 410 complete(&cmd->wait);
411 }
a4ed23f2 412 bio_put(bio);
a688b9d9 413 fcc->dispatch_list = NULL;
6b4afdd7
JK		 414 }
415
a688b9d9 416 wait_event_interruptible(*q,
721bd4d5 417 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 418 goto repeat;
419}
420
421int f2fs_issue_flush(struct f2fs_sb_info *sbi)
422{
a688b9d9 423 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 424 struct flush_cmd cmd;
6b4afdd7 425
24a9ee0f
JK		 426 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
427 test_opt(sbi, FLUSH_MERGE));
428
0f7b2abd
JK		 429 if (test_opt(sbi, NOBARRIER))
430 return 0;
431
740432f8
JK		 432 if (!test_opt(sbi, FLUSH_MERGE)) {
433 struct bio *bio = f2fs_bio_alloc(0);
434 int ret;
435
436 bio->bi_bdev = sbi->sb->s_bdev;
437 ret = submit_bio_wait(WRITE_FLUSH, bio);
438 bio_put(bio);
439 return ret;
440 }
6b4afdd7 441
adf8d90b 442 init_completion(&cmd.wait);
6b4afdd7 443
721bd4d5 444 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 445
a688b9d9
GZ		 446 if (!fcc->dispatch_list)
447 wake_up(&fcc->flush_wait_queue);
6b4afdd7 448
adf8d90b
CY		 449 wait_for_completion(&cmd.wait);
450
451 return cmd.ret;
6b4afdd7
JK		 452}
453
2163d198
GZ		 454int create_flush_cmd_control(struct f2fs_sb_info *sbi)
455{
456 dev_t dev = sbi->sb->s_bdev->bd_dev;
457 struct flush_cmd_control *fcc;
458 int err = 0;
459
460 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
461 if (!fcc)
462 return -ENOMEM;
2163d198 463 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 464 init_llist_head(&fcc->issue_list);
6b2920a5 465 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 466 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
467 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
468 if (IS_ERR(fcc->f2fs_issue_flush)) {
469 err = PTR_ERR(fcc->f2fs_issue_flush);
470 kfree(fcc);
6b2920a5 471 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 472 return err;
473 }
2163d198
GZ		 474
475 return err;
476}
477
478void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
479{
6b2920a5 480 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 481
482 if (fcc && fcc->f2fs_issue_flush)
483 kthread_stop(fcc->f2fs_issue_flush);
484 kfree(fcc);
6b2920a5 485 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 486}
487
351df4b2
JK		 488static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
489 enum dirty_type dirty_type)
490{
491 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
492
493 /* need not be added */
494 if (IS_CURSEG(sbi, segno))
495 return;
496
497 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
498 dirty_i->nr_dirty[dirty_type]++;
499
500 if (dirty_type == DIRTY) {
501 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 502 enum dirty_type t = sentry->type;
b2f2c390 503
ec325b52
JK		 504 if (unlikely(t >= DIRTY)) {
505 f2fs_bug_on(sbi, 1);
506 return;
507 }
4625d6aa
CL		 508 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
509 dirty_i->nr_dirty[t]++;
351df4b2
JK		 510 }
511}
512
513static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
514 enum dirty_type dirty_type)
515{
516 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
517
518 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
519 dirty_i->nr_dirty[dirty_type]--;
520
521 if (dirty_type == DIRTY) {
4625d6aa
CL		 522 struct seg_entry *sentry = get_seg_entry(sbi, segno);
523 enum dirty_type t = sentry->type;
524
525 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
526 dirty_i->nr_dirty[t]--;
b2f2c390 527
5ec4e49f
JK		 528 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
529 clear_bit(GET_SECNO(sbi, segno),
530 dirty_i->victim_secmap);
351df4b2
JK		 531 }
532}
533
0a8165d7 534/*
351df4b2
JK		 535 * Should not occur error such as -ENOMEM.
536 * Adding dirty entry into seglist is not critical operation.
537 * If a given segment is one of current working segments, it won't be added.
538 */
8d8451af 539static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 540{
541 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
542 unsigned short valid_blocks;
543
544 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
545 return;
546
547 mutex_lock(&dirty_i->seglist_lock);
548
549 valid_blocks = get_valid_blocks(sbi, segno, 0);
550
551 if (valid_blocks == 0) {
552 __locate_dirty_segment(sbi, segno, PRE);
553 __remove_dirty_segment(sbi, segno, DIRTY);
554 } else if (valid_blocks < sbi->blocks_per_seg) {
555 __locate_dirty_segment(sbi, segno, DIRTY);
556 } else {
557 /* Recovery routine with SSR needs this */
558 __remove_dirty_segment(sbi, segno, DIRTY);
559 }
560
561 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 562}
563
1e87a78d 564static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 565 block_t blkstart, block_t blklen)
566{
55cf9cb6
CY		 567 sector_t start = SECTOR_FROM_BLOCK(blkstart);
568 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 569 struct seg_entry *se;
570 unsigned int offset;
571 block_t i;
572
573 for (i = blkstart; i < blkstart + blklen; i++) {
574 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
575 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
576
577 if (!f2fs_test_and_set_bit(offset, se->discard_map))
578 sbi->discard_blks--;
579 }
1661d07c 580 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 581 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
582}
583
e90c2d28 584bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 585{
60b286c4 586 int err = -EOPNOTSUPP;
40a02be1
JK		 587
588 if (test_opt(sbi, DISCARD)) {
589 struct seg_entry *se = get_seg_entry(sbi,
590 GET_SEGNO(sbi, blkaddr));
591 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
592
593 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 594 return false;
40a02be1
JK		 595
596 err = f2fs_issue_discard(sbi, blkaddr, 1);
597 }
598
e90c2d28 599 if (err) {
381722d2 600 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 601 return true;
602 }
603 return false;
37208879
JK		 604}
605
adf4983b 606static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 607 struct cp_control *cpc, struct seg_entry *se,
608 unsigned int start, unsigned int end)
b2955550
JK		 609{
610 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 611 struct discard_entry *new, *last;
612
613 if (!list_empty(head)) {
614 last = list_last_entry(head, struct discard_entry, list);
615 if (START_BLOCK(sbi, cpc->trim_start) + start ==
616 last->blkaddr + last->len) {
617 last->len += end - start;
618 goto done;
619 }
620 }
621
622 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
623 INIT_LIST_HEAD(&new->list);
624 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
625 new->len = end - start;
626 list_add_tail(&new->list, head);
627done:
628 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 629}
630
631static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
632{
b2955550
JK		 633 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
634 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 635 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 636 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
637 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 638 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 639 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 640 unsigned int start = 0, end = -1;
4b2fecc8 641 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 642 int i;
643
a66cdd98 644 if (se->valid_blocks == max_blocks)
b2955550
JK		 645 return;
646
a66cdd98
JK		 647 if (!force) {
648 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 649 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
650 return;
4b2fecc8
JK		 651 }
652
b2955550
JK		 653 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
654 for (i = 0; i < entries; i++)
a66cdd98 655 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 656 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 657
4b2fecc8 658 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 659 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
660 if (start >= max_blocks)
661 break;
662
663 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 664 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 665 }
666}
667
4b2fecc8
JK		 668void release_discard_addrs(struct f2fs_sb_info *sbi)
669{
670 struct list_head *head = &(SM_I(sbi)->discard_list);
671 struct discard_entry *entry, *this;
672
673 /* drop caches */
674 list_for_each_entry_safe(entry, this, head, list) {
675 list_del(&entry->list);
676 kmem_cache_free(discard_entry_slab, entry);
677 }
678}
679
0a8165d7 680/*
351df4b2
JK		 681 * Should call clear_prefree_segments after checkpoint is done.
682 */
683static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
684{
685 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 686 unsigned int segno;
351df4b2
JK		 687
688 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 689 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 690 __set_test_and_free(sbi, segno);
351df4b2
JK		 691 mutex_unlock(&dirty_i->seglist_lock);
692}
693
836b5a63 694void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 695{
b2955550 696 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 697 struct discard_entry *entry, *this;
351df4b2 698 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 699 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 700 unsigned int start = 0, end = -1;
351df4b2
JK		 701
702 mutex_lock(&dirty_i->seglist_lock);
29e59c14 703
351df4b2 704 while (1) {
29e59c14 705 int i;
7cd8558b
JK		 706 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
707 if (start >= MAIN_SEGS(sbi))
351df4b2 708 break;
7cd8558b
JK		 709 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
710 start + 1);
29e59c14
CL		 711
712 for (i = start; i < end; i++)
713 clear_bit(i, prefree_map);
714
715 dirty_i->nr_dirty[PRE] -= end - start;
716
717 if (!test_opt(sbi, DISCARD))
718 continue;
351df4b2 719
37208879
JK		 720 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
721 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 722 }
723 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 724
725 /* send small discards */
2d7b822a 726 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 727 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
728 goto skip;
37208879 729 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 730 cpc->trimmed += entry->len;
836b5a63 731skip:
b2955550
JK		 732 list_del(&entry->list);
733 SM_I(sbi)->nr_discards -= entry->len;
734 kmem_cache_free(discard_entry_slab, entry);
735 }
351df4b2
JK		 736}
737
184a5cd2 738static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 739{
740 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 741
742 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 743 sit_i->dirty_sentries++;
184a5cd2
CY		 744 return false;
745 }
746
747 return true;
351df4b2
JK		 748}
749
750static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
751 unsigned int segno, int modified)
752{
753 struct seg_entry *se = get_seg_entry(sbi, segno);
754 se->type = type;
755 if (modified)
756 __mark_sit_entry_dirty(sbi, segno);
757}
758
759static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
760{
761 struct seg_entry *se;
762 unsigned int segno, offset;
763 long int new_vblocks;
764
765 segno = GET_SEGNO(sbi, blkaddr);
766
767 se = get_seg_entry(sbi, segno);
768 new_vblocks = se->valid_blocks + del;
491c0854 769 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 770
9850cf4a 771 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 772 (new_vblocks > sbi->blocks_per_seg)));
773
774 se->valid_blocks = new_vblocks;
775 se->mtime = get_mtime(sbi);
776 SIT_I(sbi)->max_mtime = se->mtime;
777
778 /* Update valid block bitmap */
779 if (del > 0) {
52aca074 780 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 781 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 782 if (!f2fs_test_and_set_bit(offset, se->discard_map))
783 sbi->discard_blks--;
351df4b2 784 } else {
52aca074 785 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 786 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 787 if (f2fs_test_and_clear_bit(offset, se->discard_map))
788 sbi->discard_blks++;
351df4b2
JK		 789 }
790 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
791 se->ckpt_valid_blocks += del;
792
793 __mark_sit_entry_dirty(sbi, segno);
794
795 /* update total number of valid blocks to be written in ckpt area */
796 SIT_I(sbi)->written_valid_blocks += del;
797
798 if (sbi->segs_per_sec > 1)
799 get_sec_entry(sbi, segno)->valid_blocks += del;
800}
801
5e443818 802void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 803{
5e443818
JK		 804 update_sit_entry(sbi, new, 1);
805 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
806 update_sit_entry(sbi, old, -1);
807
808 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
809 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 810}
811
812void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
813{
814 unsigned int segno = GET_SEGNO(sbi, addr);
815 struct sit_info *sit_i = SIT_I(sbi);
816
9850cf4a 817 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 818 if (addr == NEW_ADDR)
819 return;
820
821 /* add it into sit main buffer */
822 mutex_lock(&sit_i->sentry_lock);
823
824 update_sit_entry(sbi, addr, -1);
825
826 /* add it into dirty seglist */
827 locate_dirty_segment(sbi, segno);
828
829 mutex_unlock(&sit_i->sentry_lock);
830}
831
6e2c64ad
JK		 832bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
833{
834 struct sit_info *sit_i = SIT_I(sbi);
835 unsigned int segno, offset;
836 struct seg_entry *se;
837 bool is_cp = false;
838
839 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
840 return true;
841
842 mutex_lock(&sit_i->sentry_lock);
843
844 segno = GET_SEGNO(sbi, blkaddr);
845 se = get_seg_entry(sbi, segno);
846 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
847
848 if (f2fs_test_bit(offset, se->ckpt_valid_map))
849 is_cp = true;
850
851 mutex_unlock(&sit_i->sentry_lock);
852
853 return is_cp;
854}
855
0a8165d7 856/*
351df4b2
JK		 857 * This function should be resided under the curseg_mutex lock
858 */
859static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 860 struct f2fs_summary *sum)
351df4b2
JK		 861{
862 struct curseg_info *curseg = CURSEG_I(sbi, type);
863 void *addr = curseg->sum_blk;
e79efe3b 864 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 865 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 866}
867
0a8165d7 868/*
351df4b2
JK		 869 * Calculate the number of current summary pages for writing
870 */
3fa06d7b 871int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 872{
351df4b2 873 int valid_sum_count = 0;
9a47938b 874 int i, sum_in_page;
351df4b2
JK		 875
876 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
877 if (sbi->ckpt->alloc_type[i] == SSR)
878 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 879 else {
880 if (for_ra)
881 valid_sum_count += le16_to_cpu(
882 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
883 else
884 valid_sum_count += curseg_blkoff(sbi, i);
885 }
351df4b2
JK		 886 }
887
9a47938b
FL		 888 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
889 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
890 if (valid_sum_count <= sum_in_page)
351df4b2 891 return 1;
9a47938b
FL		 892 else if ((valid_sum_count - sum_in_page) <=
893 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 894 return 2;
895 return 3;
896}
897
0a8165d7 898/*
351df4b2
JK		 899 * Caller should put this summary page
900 */
901struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
902{
903 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
904}
905
381722d2 906void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 907{
908 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 909 void *dst = page_address(page);
910
911 if (src)
912 memcpy(dst, src, PAGE_CACHE_SIZE);
913 else
914 memset(dst, 0, PAGE_CACHE_SIZE);
351df4b2
JK		 915 set_page_dirty(page);
916 f2fs_put_page(page, 1);
917}
918
381722d2
CY		 919static void write_sum_page(struct f2fs_sb_info *sbi,
920 struct f2fs_summary_block *sum_blk, block_t blk_addr)
921{
922 update_meta_page(sbi, (void *)sum_blk, blk_addr);
923}
924
60374688
JK		 925static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
926{
927 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 928 unsigned int segno = curseg->segno + 1;
60374688
JK		 929 struct free_segmap_info *free_i = FREE_I(sbi);
930
7cd8558b 931 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 932 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 933 return 0;
934}
935
0a8165d7 936/*
351df4b2
JK		 937 * Find a new segment from the free segments bitmap to right order
938 * This function should be returned with success, otherwise BUG
939 */
940static void get_new_segment(struct f2fs_sb_info *sbi,
941 unsigned int *newseg, bool new_sec, int dir)
942{
943 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 944 unsigned int segno, secno, zoneno;
7cd8558b 945 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 946 unsigned int hint = *newseg / sbi->segs_per_sec;
947 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
948 unsigned int left_start = hint;
949 bool init = true;
950 int go_left = 0;
951 int i;
952
1a118ccf 953 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 954
955 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
956 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 957 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
958 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 959 goto got_it;
960 }
961find_other_zone:
7cd8558b
JK		 962 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
963 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 964 if (dir == ALLOC_RIGHT) {
965 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 966 MAIN_SECS(sbi), 0);
967 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 968 } else {
969 go_left = 1;
970 left_start = hint - 1;
971 }
972 }
973 if (go_left == 0)
974 goto skip_left;
975
976 while (test_bit(left_start, free_i->free_secmap)) {
977 if (left_start > 0) {
978 left_start--;
979 continue;
980 }
981 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 982 MAIN_SECS(sbi), 0);
983 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 984 break;
985 }
986 secno = left_start;
987skip_left:
988 hint = secno;
989 segno = secno * sbi->segs_per_sec;
990 zoneno = secno / sbi->secs_per_zone;
991
992 /* give up on finding another zone */
993 if (!init)
994 goto got_it;
995 if (sbi->secs_per_zone == 1)
996 goto got_it;
997 if (zoneno == old_zoneno)
998 goto got_it;
999 if (dir == ALLOC_LEFT) {
1000 if (!go_left && zoneno + 1 >= total_zones)
1001 goto got_it;
1002 if (go_left && zoneno == 0)
1003 goto got_it;
1004 }
1005 for (i = 0; i < NR_CURSEG_TYPE; i++)
1006 if (CURSEG_I(sbi, i)->zone == zoneno)
1007 break;
1008
1009 if (i < NR_CURSEG_TYPE) {
1010 /* zone is in user, try another */
1011 if (go_left)
1012 hint = zoneno * sbi->secs_per_zone - 1;
1013 else if (zoneno + 1 >= total_zones)
1014 hint = 0;
1015 else
1016 hint = (zoneno + 1) * sbi->secs_per_zone;
1017 init = false;
1018 goto find_other_zone;
1019 }
1020got_it:
1021 /* set it as dirty segment in free segmap */
9850cf4a 1022 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1023 __set_inuse(sbi, segno);
1024 *newseg = segno;
1a118ccf 1025 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1026}
1027
1028static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1029{
1030 struct curseg_info *curseg = CURSEG_I(sbi, type);
1031 struct summary_footer *sum_footer;
1032
1033 curseg->segno = curseg->next_segno;
1034 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1035 curseg->next_blkoff = 0;
1036 curseg->next_segno = NULL_SEGNO;
1037
1038 sum_footer = &(curseg->sum_blk->footer);
1039 memset(sum_footer, 0, sizeof(struct summary_footer));
1040 if (IS_DATASEG(type))
1041 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1042 if (IS_NODESEG(type))
1043 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1044 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1045}
1046
0a8165d7 1047/*
351df4b2
JK		 1048 * Allocate a current working segment.
1049 * This function always allocates a free segment in LFS manner.
1050 */
1051static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1052{
1053 struct curseg_info *curseg = CURSEG_I(sbi, type);
1054 unsigned int segno = curseg->segno;
1055 int dir = ALLOC_LEFT;
1056
1057 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1058 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1059 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1060 dir = ALLOC_RIGHT;
1061
1062 if (test_opt(sbi, NOHEAP))
1063 dir = ALLOC_RIGHT;
1064
1065 get_new_segment(sbi, &segno, new_sec, dir);
1066 curseg->next_segno = segno;
1067 reset_curseg(sbi, type, 1);
1068 curseg->alloc_type = LFS;
1069}
1070
1071static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1072 struct curseg_info *seg, block_t start)
1073{
1074 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1075 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1076 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1077 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1078 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1079 int i, pos;
1080
1081 for (i = 0; i < entries; i++)
1082 target_map[i] = ckpt_map[i] | cur_map[i];
1083
1084 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1085
1086 seg->next_blkoff = pos;
351df4b2
JK		 1087}
1088
0a8165d7 1089/*
351df4b2
JK		 1090 * If a segment is written by LFS manner, next block offset is just obtained
1091 * by increasing the current block offset. However, if a segment is written by
1092 * SSR manner, next block offset obtained by calling __next_free_blkoff
1093 */
1094static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1095 struct curseg_info *seg)
1096{
1097 if (seg->alloc_type == SSR)
1098 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1099 else
1100 seg->next_blkoff++;
1101}
1102
0a8165d7 1103/*
e1c42045 1104 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1105 * manner, so it should recover the existing segment information of valid blocks
1106 */
1107static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1108{
1109 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1110 struct curseg_info *curseg = CURSEG_I(sbi, type);
1111 unsigned int new_segno = curseg->next_segno;
1112 struct f2fs_summary_block *sum_node;
1113 struct page *sum_page;
1114
1115 write_sum_page(sbi, curseg->sum_blk,
1116 GET_SUM_BLOCK(sbi, curseg->segno));
1117 __set_test_and_inuse(sbi, new_segno);
1118
1119 mutex_lock(&dirty_i->seglist_lock);
1120 __remove_dirty_segment(sbi, new_segno, PRE);
1121 __remove_dirty_segment(sbi, new_segno, DIRTY);
1122 mutex_unlock(&dirty_i->seglist_lock);
1123
1124 reset_curseg(sbi, type, 1);
1125 curseg->alloc_type = SSR;
1126 __next_free_blkoff(sbi, curseg, 0);
1127
1128 if (reuse) {
1129 sum_page = get_sum_page(sbi, new_segno);
1130 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1131 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1132 f2fs_put_page(sum_page, 1);
1133 }
1134}
1135
43727527
JK		 1136static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1137{
1138 struct curseg_info *curseg = CURSEG_I(sbi, type);
1139 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1140
1141 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1142 return v_ops->get_victim(sbi,
1143 &(curseg)->next_segno, BG_GC, type, SSR);
1144
1145 /* For data segments, let's do SSR more intensively */
1146 for (; type >= CURSEG_HOT_DATA; type--)
1147 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1148 BG_GC, type, SSR))
1149 return 1;
1150 return 0;
1151}
1152
351df4b2
JK		 1153/*
1154 * flush out current segment and replace it with new segment
1155 * This function should be returned with success, otherwise BUG
1156 */
1157static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1158 int type, bool force)
1159{
1160 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1161
7b405275 1162 if (force)
351df4b2 1163 new_curseg(sbi, type, true);
7b405275 1164 else if (type == CURSEG_WARM_NODE)
351df4b2 1165 new_curseg(sbi, type, false);
60374688
JK		 1166 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1167 new_curseg(sbi, type, false);
351df4b2
JK		 1168 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1169 change_curseg(sbi, type, true);
1170 else
1171 new_curseg(sbi, type, false);
dcdfff65
JK		 1172
1173 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1174}
1175
38aa0889
JK		 1176static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1177{
1178 struct curseg_info *curseg = CURSEG_I(sbi, type);
1179 unsigned int old_segno;
1180
1181 old_segno = curseg->segno;
1182 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1183 locate_dirty_segment(sbi, old_segno);
1184}
1185
351df4b2
JK		 1186void allocate_new_segments(struct f2fs_sb_info *sbi)
1187{
351df4b2
JK		 1188 int i;
1189
38aa0889
JK		 1190 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1191 __allocate_new_segments(sbi, i);
351df4b2
JK		 1192}
1193
1194static const struct segment_allocation default_salloc_ops = {
1195 .allocate_segment = allocate_segment_by_default,
1196};
1197
4b2fecc8
JK		 1198int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1199{
f7ef9b83
JK		 1200 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1201 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1202 unsigned int start_segno, end_segno;
1203 struct cp_control cpc;
c34f42e2 1204 int err = 0;
4b2fecc8 1205
836b5a63 1206 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1207 return -EINVAL;
1208
9bd27ae4 1209 cpc.trimmed = 0;
7cd8558b 1210 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1211 goto out;
1212
1213 /* start/end segment number in main_area */
7cd8558b
JK		 1214 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1215 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1216 GET_SEGNO(sbi, end);
4b2fecc8 1217 cpc.reason = CP_DISCARD;
836b5a63 1218 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1219
1220 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1221 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1222 cpc.trim_start = start_segno;
a66cdd98
JK		 1223
1224 if (sbi->discard_blks == 0)
1225 break;
1226 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1227 cpc.trim_end = end_segno;
1228 else
1229 cpc.trim_end = min_t(unsigned int,
1230 rounddown(start_segno +
bba681cb
JK		 1231 BATCHED_TRIM_SEGMENTS(sbi),
1232 sbi->segs_per_sec) - 1, end_segno);
1233
1234 mutex_lock(&sbi->gc_mutex);
c34f42e2 1235 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1236 mutex_unlock(&sbi->gc_mutex);
1237 }
4b2fecc8 1238out:
f7ef9b83 1239 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1240 return err;
4b2fecc8
JK		 1241}
1242
351df4b2
JK		 1243static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1244{
1245 struct curseg_info *curseg = CURSEG_I(sbi, type);
1246 if (curseg->next_blkoff < sbi->blocks_per_seg)
1247 return true;
1248 return false;
1249}
1250
1251static int __get_segment_type_2(struct page *page, enum page_type p_type)
1252{
1253 if (p_type == DATA)
1254 return CURSEG_HOT_DATA;
1255 else
1256 return CURSEG_HOT_NODE;
1257}
1258
1259static int __get_segment_type_4(struct page *page, enum page_type p_type)
1260{
1261 if (p_type == DATA) {
1262 struct inode *inode = page->mapping->host;
1263
1264 if (S_ISDIR(inode->i_mode))
1265 return CURSEG_HOT_DATA;
1266 else
1267 return CURSEG_COLD_DATA;
1268 } else {
a344b9fd
JK		 1269 if (IS_DNODE(page) && is_cold_node(page))
1270 return CURSEG_WARM_NODE;
351df4b2
JK		 1271 else
1272 return CURSEG_COLD_NODE;
1273 }
1274}
1275
1276static int __get_segment_type_6(struct page *page, enum page_type p_type)
1277{
1278 if (p_type == DATA) {
1279 struct inode *inode = page->mapping->host;
1280
1281 if (S_ISDIR(inode->i_mode))
1282 return CURSEG_HOT_DATA;
354a3399 1283 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1284 return CURSEG_COLD_DATA;
1285 else
1286 return CURSEG_WARM_DATA;
1287 } else {
1288 if (IS_DNODE(page))
1289 return is_cold_node(page) ? CURSEG_WARM_NODE :
1290 CURSEG_HOT_NODE;
1291 else
1292 return CURSEG_COLD_NODE;
1293 }
1294}
1295
1296static int __get_segment_type(struct page *page, enum page_type p_type)
1297{
4081363f 1298 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1299 case 2:
1300 return __get_segment_type_2(page, p_type);
1301 case 4:
1302 return __get_segment_type_4(page, p_type);
351df4b2 1303 }
12a67146 1304 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1305 f2fs_bug_on(F2FS_P_SB(page),
1306 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1307 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1308}
1309
bfad7c2d
JK		 1310void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1311 block_t old_blkaddr, block_t *new_blkaddr,
1312 struct f2fs_summary *sum, int type)
351df4b2
JK		 1313{
1314 struct sit_info *sit_i = SIT_I(sbi);
1315 struct curseg_info *curseg;
38aa0889
JK		 1316 bool direct_io = (type == CURSEG_DIRECT_IO);
1317
1318 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1319
351df4b2
JK		 1320 curseg = CURSEG_I(sbi, type);
1321
1322 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1323 mutex_lock(&sit_i->sentry_lock);
351df4b2 1324
38aa0889 1325 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1326 if (direct_io && curseg->next_blkoff &&
1327 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1328 __allocate_new_segments(sbi, type);
1329
351df4b2 1330 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1331
1332 /*
1333 * __add_sum_entry should be resided under the curseg_mutex
1334 * because, this function updates a summary entry in the
1335 * current summary block.
1336 */
e79efe3b 1337 __add_sum_entry(sbi, type, sum);
351df4b2 1338
351df4b2 1339 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1340
1341 stat_inc_block_count(sbi, curseg);
351df4b2 1342
5e443818
JK		 1343 if (!__has_curseg_space(sbi, type))
1344 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1345 /*
1346 * SIT information should be updated before segment allocation,
1347 * since SSR needs latest valid block information.
1348 */
1349 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1350
351df4b2
JK		 1351 mutex_unlock(&sit_i->sentry_lock);
1352
bfad7c2d 1353 if (page && IS_NODESEG(type))
351df4b2
JK		 1354 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1355
bfad7c2d
JK		 1356 mutex_unlock(&curseg->curseg_mutex);
1357}
1358
05ca3632 1359static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1360{
05ca3632 1361 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1362
05ca3632
JK		 1363 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1364 &fio->blk_addr, sum, type);
bfad7c2d 1365
351df4b2 1366 /* writeout dirty page into bdev */
05ca3632 1367 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1368}
1369
577e3495 1370void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1371{
458e6197 1372 struct f2fs_io_info fio = {
05ca3632 1373 .sbi = sbi,
458e6197 1374 .type = META,
cf04e8eb
JK		 1375 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1376 .blk_addr = page->index,
05ca3632 1377 .page = page,
4375a336 1378 .encrypted_page = NULL,
458e6197
JK		 1379 };
1380
2b947003
CY		 1381 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1382 fio.rw &= ~REQ_META;
1383
351df4b2 1384 set_page_writeback(page);
05ca3632 1385 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1386}
1387
05ca3632 1388void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1389{
1390 struct f2fs_summary sum;
05ca3632 1391
351df4b2 1392 set_summary(&sum, nid, 0, 0);
05ca3632 1393 do_write_page(&sum, fio);
351df4b2
JK		 1394}
1395
05ca3632 1396void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1397{
05ca3632 1398 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1399 struct f2fs_summary sum;
1400 struct node_info ni;
1401
9850cf4a 1402 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1403 get_node_info(sbi, dn->nid, &ni);
1404 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1405 do_write_page(&sum, fio);
e1509cf2 1406 dn->data_blkaddr = fio->blk_addr;
351df4b2
JK		 1407}
1408
05ca3632 1409void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1410{
05ca3632
JK		 1411 stat_inc_inplace_blocks(fio->sbi);
1412 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1413}
1414
528e3459
CY		 1415static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1416 struct f2fs_summary *sum,
19f106bc 1417 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1418 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1419{
1420 struct sit_info *sit_i = SIT_I(sbi);
1421 struct curseg_info *curseg;
1422 unsigned int segno, old_cursegno;
1423 struct seg_entry *se;
1424 int type;
19f106bc 1425 unsigned short old_blkoff;
351df4b2
JK		 1426
1427 segno = GET_SEGNO(sbi, new_blkaddr);
1428 se = get_seg_entry(sbi, segno);
1429 type = se->type;
1430
19f106bc
CY		 1431 if (!recover_curseg) {
1432 /* for recovery flow */
1433 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1434 if (old_blkaddr == NULL_ADDR)
1435 type = CURSEG_COLD_DATA;
1436 else
1437 type = CURSEG_WARM_DATA;
1438 }
1439 } else {
1440 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1441 type = CURSEG_WARM_DATA;
1442 }
19f106bc 1443
351df4b2
JK		 1444 curseg = CURSEG_I(sbi, type);
1445
1446 mutex_lock(&curseg->curseg_mutex);
1447 mutex_lock(&sit_i->sentry_lock);
1448
1449 old_cursegno = curseg->segno;
19f106bc 1450 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1451
1452 /* change the current segment */
1453 if (segno != curseg->segno) {
1454 curseg->next_segno = segno;
1455 change_curseg(sbi, type, true);
1456 }
1457
491c0854 1458 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1459 __add_sum_entry(sbi, type, sum);
351df4b2 1460
28bc106b 1461 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1462 update_sit_entry(sbi, new_blkaddr, 1);
1463 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1464 update_sit_entry(sbi, old_blkaddr, -1);
1465
1466 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1467 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1468
351df4b2 1469 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1470
19f106bc
CY		 1471 if (recover_curseg) {
1472 if (old_cursegno != curseg->segno) {
1473 curseg->next_segno = old_cursegno;
1474 change_curseg(sbi, type, true);
1475 }
1476 curseg->next_blkoff = old_blkoff;
1477 }
1478
351df4b2
JK		 1479 mutex_unlock(&sit_i->sentry_lock);
1480 mutex_unlock(&curseg->curseg_mutex);
1481}
1482
528e3459
CY		 1483void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1484 block_t old_addr, block_t new_addr,
28bc106b
CY		 1485 unsigned char version, bool recover_curseg,
1486 bool recover_newaddr)
528e3459
CY		 1487{
1488 struct f2fs_summary sum;
1489
1490 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1491
28bc106b
CY		 1492 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1493 recover_curseg, recover_newaddr);
528e3459
CY		 1494
1495 dn->data_blkaddr = new_addr;
1496 set_data_blkaddr(dn);
1497 f2fs_update_extent_cache(dn);
1498}
1499
93dfe2ac 1500void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1501 enum page_type type, bool ordered)
93dfe2ac 1502{
93dfe2ac 1503 if (PageWriteback(page)) {
4081363f
JK		 1504 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1505
0c3a5797 1506 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1507 if (ordered)
1508 wait_on_page_writeback(page);
1509 else
1510 wait_for_stable_page(page);
93dfe2ac
JK		 1511 }
1512}
1513
08b39fbd
CY		 1514void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1515 block_t blkaddr)
1516{
1517 struct page *cpage;
1518
1519 if (blkaddr == NEW_ADDR)
1520 return;
1521
1522 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1523
1524 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1525 if (cpage) {
fec1d657 1526 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1527 f2fs_put_page(cpage, 1);
1528 }
1529}
1530
351df4b2
JK		 1531static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1532{
1533 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1534 struct curseg_info *seg_i;
1535 unsigned char *kaddr;
1536 struct page *page;
1537 block_t start;
1538 int i, j, offset;
1539
1540 start = start_sum_block(sbi);
1541
1542 page = get_meta_page(sbi, start++);
1543 kaddr = (unsigned char *)page_address(page);
1544
1545 /* Step 1: restore nat cache */
1546 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
dfc08a12 1547 memcpy(&seg_i->sum_blk->journal.n_nats, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1548
1549 /* Step 2: restore sit cache */
1550 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
dfc08a12 1551 memcpy(&seg_i->sum_blk->journal.n_sits, kaddr + SUM_JOURNAL_SIZE,
351df4b2
JK		 1552 SUM_JOURNAL_SIZE);
1553 offset = 2 * SUM_JOURNAL_SIZE;
1554
1555 /* Step 3: restore summary entries */
1556 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1557 unsigned short blk_off;
1558 unsigned int segno;
1559
1560 seg_i = CURSEG_I(sbi, i);
1561 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1562 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1563 seg_i->next_segno = segno;
1564 reset_curseg(sbi, i, 0);
1565 seg_i->alloc_type = ckpt->alloc_type[i];
1566 seg_i->next_blkoff = blk_off;
1567
1568 if (seg_i->alloc_type == SSR)
1569 blk_off = sbi->blocks_per_seg;
1570
1571 for (j = 0; j < blk_off; j++) {
1572 struct f2fs_summary *s;
1573 s = (struct f2fs_summary *)(kaddr + offset);
1574 seg_i->sum_blk->entries[j] = *s;
1575 offset += SUMMARY_SIZE;
1576 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1577 SUM_FOOTER_SIZE)
1578 continue;
1579
1580 f2fs_put_page(page, 1);
1581 page = NULL;
1582
1583 page = get_meta_page(sbi, start++);
1584 kaddr = (unsigned char *)page_address(page);
1585 offset = 0;
1586 }
1587 }
1588 f2fs_put_page(page, 1);
1589 return 0;
1590}
1591
1592static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1593{
1594 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1595 struct f2fs_summary_block *sum;
1596 struct curseg_info *curseg;
1597 struct page *new;
1598 unsigned short blk_off;
1599 unsigned int segno = 0;
1600 block_t blk_addr = 0;
1601
1602 /* get segment number and block addr */
1603 if (IS_DATASEG(type)) {
1604 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1605 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1606 CURSEG_HOT_DATA]);
119ee914 1607 if (__exist_node_summaries(sbi))
351df4b2
JK		 1608 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1609 else
1610 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1611 } else {
1612 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1613 CURSEG_HOT_NODE]);
1614 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1615 CURSEG_HOT_NODE]);
119ee914 1616 if (__exist_node_summaries(sbi))
351df4b2
JK		 1617 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1618 type - CURSEG_HOT_NODE);
1619 else
1620 blk_addr = GET_SUM_BLOCK(sbi, segno);
1621 }
1622
1623 new = get_meta_page(sbi, blk_addr);
1624 sum = (struct f2fs_summary_block *)page_address(new);
1625
1626 if (IS_NODESEG(type)) {
119ee914 1627 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1628 struct f2fs_summary *ns = &sum->entries[0];
1629 int i;
1630 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1631 ns->version = 0;
1632 ns->ofs_in_node = 0;
1633 }
1634 } else {
d653788a
GZ		 1635 int err;
1636
1637 err = restore_node_summary(sbi, segno, sum);
1638 if (err) {
351df4b2 1639 f2fs_put_page(new, 1);
d653788a 1640 return err;
351df4b2
JK		 1641 }
1642 }
1643 }
1644
1645 /* set uncompleted segment to curseg */
1646 curseg = CURSEG_I(sbi, type);
1647 mutex_lock(&curseg->curseg_mutex);
1648 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1649 curseg->next_segno = segno;
1650 reset_curseg(sbi, type, 0);
1651 curseg->alloc_type = ckpt->alloc_type[type];
1652 curseg->next_blkoff = blk_off;
1653 mutex_unlock(&curseg->curseg_mutex);
1654 f2fs_put_page(new, 1);
1655 return 0;
1656}
1657
1658static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1659{
1660 int type = CURSEG_HOT_DATA;
e4fc5fbf 1661 int err;
351df4b2 1662
25ca923b 1663 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1664 int npages = npages_for_summary_flush(sbi, true);
1665
1666 if (npages >= 2)
1667 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1668 META_CP, true);
3fa06d7b 1669
351df4b2
JK		 1670 /* restore for compacted data summary */
1671 if (read_compacted_summaries(sbi))
1672 return -EINVAL;
1673 type = CURSEG_HOT_NODE;
1674 }
1675
119ee914 1676 if (__exist_node_summaries(sbi))
3fa06d7b 1677 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1678 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1679
e4fc5fbf
CY		 1680 for (; type <= CURSEG_COLD_NODE; type++) {
1681 err = read_normal_summaries(sbi, type);
1682 if (err)
1683 return err;
1684 }
1685
351df4b2
JK		 1686 return 0;
1687}
1688
1689static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1690{
1691 struct page *page;
1692 unsigned char *kaddr;
1693 struct f2fs_summary *summary;
1694 struct curseg_info *seg_i;
1695 int written_size = 0;
1696 int i, j;
1697
1698 page = grab_meta_page(sbi, blkaddr++);
1699 kaddr = (unsigned char *)page_address(page);
1700
1701 /* Step 1: write nat cache */
1702 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
dfc08a12 1703 memcpy(kaddr, &seg_i->sum_blk->journal.n_nats, SUM_JOURNAL_SIZE);
351df4b2
JK		 1704 written_size += SUM_JOURNAL_SIZE;
1705
1706 /* Step 2: write sit cache */
1707 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
dfc08a12 1708 memcpy(kaddr + written_size, &seg_i->sum_blk->journal.n_sits,
351df4b2
JK		 1709 SUM_JOURNAL_SIZE);
1710 written_size += SUM_JOURNAL_SIZE;
1711
351df4b2
JK		 1712 /* Step 3: write summary entries */
1713 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1714 unsigned short blkoff;
1715 seg_i = CURSEG_I(sbi, i);
1716 if (sbi->ckpt->alloc_type[i] == SSR)
1717 blkoff = sbi->blocks_per_seg;
1718 else
1719 blkoff = curseg_blkoff(sbi, i);
1720
1721 for (j = 0; j < blkoff; j++) {
1722 if (!page) {
1723 page = grab_meta_page(sbi, blkaddr++);
1724 kaddr = (unsigned char *)page_address(page);
1725 written_size = 0;
1726 }
1727 summary = (struct f2fs_summary *)(kaddr + written_size);
1728 *summary = seg_i->sum_blk->entries[j];
1729 written_size += SUMMARY_SIZE;
351df4b2
JK		 1730
1731 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1732 SUM_FOOTER_SIZE)
1733 continue;
1734
e8d61a74 1735 set_page_dirty(page);
351df4b2
JK		 1736 f2fs_put_page(page, 1);
1737 page = NULL;
1738 }
1739 }
e8d61a74
CY		 1740 if (page) {
1741 set_page_dirty(page);
351df4b2 1742 f2fs_put_page(page, 1);
e8d61a74 1743 }
351df4b2
JK		 1744}
1745
1746static void write_normal_summaries(struct f2fs_sb_info *sbi,
1747 block_t blkaddr, int type)
1748{
1749 int i, end;
1750 if (IS_DATASEG(type))
1751 end = type + NR_CURSEG_DATA_TYPE;
1752 else
1753 end = type + NR_CURSEG_NODE_TYPE;
1754
1755 for (i = type; i < end; i++) {
1756 struct curseg_info *sum = CURSEG_I(sbi, i);
1757 mutex_lock(&sum->curseg_mutex);
1758 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1759 mutex_unlock(&sum->curseg_mutex);
1760 }
1761}
1762
1763void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1764{
25ca923b 1765 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1766 write_compacted_summaries(sbi, start_blk);
1767 else
1768 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1769}
1770
1771void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1772{
119ee914 1773 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1774}
1775
dfc08a12 1776int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1777 unsigned int val, int alloc)
1778{
1779 int i;
1780
1781 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1782 for (i = 0; i < nats_in_cursum(journal); i++) {
1783 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1784 return i;
1785 }
dfc08a12
CY		 1786 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1787 return update_nats_in_cursum(journal, 1);
351df4b2 1788 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1789 for (i = 0; i < sits_in_cursum(journal); i++)
1790 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1791 return i;
dfc08a12
CY		 1792 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1793 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1794 }
1795 return -1;
1796}
1797
1798static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1799 unsigned int segno)
1800{
2cc22186 1801 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1802}
1803
1804static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1805 unsigned int start)
1806{
1807 struct sit_info *sit_i = SIT_I(sbi);
1808 struct page *src_page, *dst_page;
1809 pgoff_t src_off, dst_off;
1810 void *src_addr, *dst_addr;
1811
1812 src_off = current_sit_addr(sbi, start);
1813 dst_off = next_sit_addr(sbi, src_off);
1814
1815 /* get current sit block page without lock */
1816 src_page = get_meta_page(sbi, src_off);
1817 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1818 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1819
1820 src_addr = page_address(src_page);
1821 dst_addr = page_address(dst_page);
1822 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1823
1824 set_page_dirty(dst_page);
1825 f2fs_put_page(src_page, 1);
1826
1827 set_to_next_sit(sit_i, start);
1828
1829 return dst_page;
1830}
1831
184a5cd2
CY		 1832static struct sit_entry_set *grab_sit_entry_set(void)
1833{
1834 struct sit_entry_set *ses =
80c54505 1835 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1836
1837 ses->entry_cnt = 0;
1838 INIT_LIST_HEAD(&ses->set_list);
1839 return ses;
1840}
1841
1842static void release_sit_entry_set(struct sit_entry_set *ses)
1843{
1844 list_del(&ses->set_list);
1845 kmem_cache_free(sit_entry_set_slab, ses);
1846}
1847
1848static void adjust_sit_entry_set(struct sit_entry_set *ses,
1849 struct list_head *head)
1850{
1851 struct sit_entry_set *next = ses;
1852
1853 if (list_is_last(&ses->set_list, head))
1854 return;
1855
1856 list_for_each_entry_continue(next, head, set_list)
1857 if (ses->entry_cnt <= next->entry_cnt)
1858 break;
1859
1860 list_move_tail(&ses->set_list, &next->set_list);
1861}
1862
1863static void add_sit_entry(unsigned int segno, struct list_head *head)
1864{
1865 struct sit_entry_set *ses;
1866 unsigned int start_segno = START_SEGNO(segno);
1867
1868 list_for_each_entry(ses, head, set_list) {
1869 if (ses->start_segno == start_segno) {
1870 ses->entry_cnt++;
1871 adjust_sit_entry_set(ses, head);
1872 return;
1873 }
1874 }
1875
1876 ses = grab_sit_entry_set();
1877
1878 ses->start_segno = start_segno;
1879 ses->entry_cnt++;
1880 list_add(&ses->set_list, head);
1881}
1882
1883static void add_sits_in_set(struct f2fs_sb_info *sbi)
1884{
1885 struct f2fs_sm_info *sm_info = SM_I(sbi);
1886 struct list_head *set_list = &sm_info->sit_entry_set;
1887 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1888 unsigned int segno;
1889
7cd8558b 1890 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1891 add_sit_entry(segno, set_list);
1892}
1893
1894static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1895{
1896 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
dfc08a12 1897 struct f2fs_journal *journal = &curseg->sum_blk->journal;
351df4b2
JK		 1898 int i;
1899
dfc08a12 1900 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1901 unsigned int segno;
1902 bool dirtied;
1903
dfc08a12 1904 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1905 dirtied = __mark_sit_entry_dirty(sbi, segno);
1906
1907 if (!dirtied)
1908 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1909 }
dfc08a12 1910 update_sits_in_cursum(journal, -i);
351df4b2
JK		 1911}
1912
0a8165d7 1913/*
351df4b2
JK		 1914 * CP calls this function, which flushes SIT entries including sit_journal,
1915 * and moves prefree segs to free segs.
1916 */
4b2fecc8 1917void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1918{
1919 struct sit_info *sit_i = SIT_I(sbi);
1920 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1921 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
dfc08a12 1922 struct f2fs_journal *journal = &curseg->sum_blk->journal;
184a5cd2
CY		 1923 struct sit_entry_set *ses, *tmp;
1924 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1925 bool to_journal = true;
4b2fecc8 1926 struct seg_entry *se;
351df4b2
JK		 1927
1928 mutex_lock(&curseg->curseg_mutex);
1929 mutex_lock(&sit_i->sentry_lock);
1930
2b11a74b
WL		 1931 if (!sit_i->dirty_sentries)
1932 goto out;
1933
351df4b2 1934 /*
184a5cd2
CY		 1935 * add and account sit entries of dirty bitmap in sit entry
1936 * set temporarily
351df4b2 1937 */
184a5cd2 1938 add_sits_in_set(sbi);
351df4b2 1939
184a5cd2
CY		 1940 /*
1941 * if there are no enough space in journal to store dirty sit
1942 * entries, remove all entries from journal and add and account
1943 * them in sit entry set.
1944 */
dfc08a12 1945 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 1946 remove_sits_in_journal(sbi);
b2955550 1947
184a5cd2
CY		 1948 /*
1949 * there are two steps to flush sit entries:
1950 * #1, flush sit entries to journal in current cold data summary block.
1951 * #2, flush sit entries to sit page.
1952 */
1953 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1954 struct page *page = NULL;
184a5cd2
CY		 1955 struct f2fs_sit_block *raw_sit = NULL;
1956 unsigned int start_segno = ses->start_segno;
1957 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1958 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1959 unsigned int segno = start_segno;
1960
1961 if (to_journal &&
dfc08a12 1962 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 1963 to_journal = false;
1964
1965 if (!to_journal) {
1966 page = get_next_sit_page(sbi, start_segno);
1967 raw_sit = page_address(page);
351df4b2 1968 }
351df4b2 1969
184a5cd2
CY		 1970 /* flush dirty sit entries in region of current sit set */
1971 for_each_set_bit_from(segno, bitmap, end) {
1972 int offset, sit_offset;
4b2fecc8
JK		 1973
1974 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 1975
1976 /* add discard candidates */
d7bc2484 1977 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 1978 cpc->trim_start = segno;
1979 add_discard_addrs(sbi, cpc);
1980 }
184a5cd2
CY		 1981
1982 if (to_journal) {
dfc08a12 1983 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 1984 SIT_JOURNAL, segno, 1);
1985 f2fs_bug_on(sbi, offset < 0);
dfc08a12 1986 segno_in_journal(journal, offset) =
184a5cd2
CY		 1987 cpu_to_le32(segno);
1988 seg_info_to_raw_sit(se,
dfc08a12 1989 &sit_in_journal(journal, offset));
184a5cd2
CY		 1990 } else {
1991 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1992 seg_info_to_raw_sit(se,
1993 &raw_sit->entries[sit_offset]);
1994 }
351df4b2 1995
184a5cd2
CY		 1996 __clear_bit(segno, bitmap);
1997 sit_i->dirty_sentries--;
1998 ses->entry_cnt--;
351df4b2
JK		 1999 }
2000
184a5cd2
CY		 2001 if (!to_journal)
2002 f2fs_put_page(page, 1);
2003
2004 f2fs_bug_on(sbi, ses->entry_cnt);
2005 release_sit_entry_set(ses);
351df4b2 2006 }
184a5cd2
CY		 2007
2008 f2fs_bug_on(sbi, !list_empty(head));
2009 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2010out:
4b2fecc8
JK		 2011 if (cpc->reason == CP_DISCARD) {
2012 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2013 add_discard_addrs(sbi, cpc);
2014 }
351df4b2
JK		 2015 mutex_unlock(&sit_i->sentry_lock);
2016 mutex_unlock(&curseg->curseg_mutex);
2017
351df4b2
JK		 2018 set_prefree_as_free_segments(sbi);
2019}
2020
2021static int build_sit_info(struct f2fs_sb_info *sbi)
2022{
2023 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2024 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2025 struct sit_info *sit_i;
2026 unsigned int sit_segs, start;
2027 char *src_bitmap, *dst_bitmap;
2028 unsigned int bitmap_size;
2029
2030 /* allocate memory for SIT information */
2031 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2032 if (!sit_i)
2033 return -ENOMEM;
2034
2035 SM_I(sbi)->sit_info = sit_i;
2036
39307a8e
JK		 2037 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2038 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2039 if (!sit_i->sentries)
2040 return -ENOMEM;
2041
7cd8558b 2042 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2043 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2044 if (!sit_i->dirty_sentries_bitmap)
2045 return -ENOMEM;
2046
7cd8558b 2047 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2048 sit_i->sentries[start].cur_valid_map
2049 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2050 sit_i->sentries[start].ckpt_valid_map
2051 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2052 sit_i->sentries[start].discard_map
2053 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2054 if (!sit_i->sentries[start].cur_valid_map ||
2055 !sit_i->sentries[start].ckpt_valid_map ||
2056 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2057 return -ENOMEM;
2058 }
2059
60a3b782
JK		 2060 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2061 if (!sit_i->tmp_map)
2062 return -ENOMEM;
2063
351df4b2 2064 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2065 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2066 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2067 if (!sit_i->sec_entries)
2068 return -ENOMEM;
2069 }
2070
2071 /* get information related with SIT */
2072 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2073
2074 /* setup SIT bitmap from ckeckpoint pack */
2075 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2076 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2077
79b5793b 2078 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2079 if (!dst_bitmap)
2080 return -ENOMEM;
351df4b2
JK		 2081
2082 /* init SIT information */
2083 sit_i->s_ops = &default_salloc_ops;
2084
2085 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2086 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2087 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2088 sit_i->sit_bitmap = dst_bitmap;
2089 sit_i->bitmap_size = bitmap_size;
2090 sit_i->dirty_sentries = 0;
2091 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2092 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2093 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2094 mutex_init(&sit_i->sentry_lock);
2095 return 0;
2096}
2097
2098static int build_free_segmap(struct f2fs_sb_info *sbi)
2099{
351df4b2
JK		 2100 struct free_segmap_info *free_i;
2101 unsigned int bitmap_size, sec_bitmap_size;
2102
2103 /* allocate memory for free segmap information */
2104 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2105 if (!free_i)
2106 return -ENOMEM;
2107
2108 SM_I(sbi)->free_info = free_i;
2109
7cd8558b 2110 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2111 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2112 if (!free_i->free_segmap)
2113 return -ENOMEM;
2114
7cd8558b 2115 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2116 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2117 if (!free_i->free_secmap)
2118 return -ENOMEM;
2119
2120 /* set all segments as dirty temporarily */
2121 memset(free_i->free_segmap, 0xff, bitmap_size);
2122 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2123
2124 /* init free segmap information */
7cd8558b 2125 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2126 free_i->free_segments = 0;
2127 free_i->free_sections = 0;
1a118ccf 2128 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2129 return 0;
2130}
2131
2132static int build_curseg(struct f2fs_sb_info *sbi)
2133{
1042d60f 2134 struct curseg_info *array;
351df4b2
JK		 2135 int i;
2136
b434babf 2137 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2138 if (!array)
2139 return -ENOMEM;
2140
2141 SM_I(sbi)->curseg_array = array;
2142
2143 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2144 mutex_init(&array[i].curseg_mutex);
2145 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2146 if (!array[i].sum_blk)
2147 return -ENOMEM;
2148 array[i].segno = NULL_SEGNO;
2149 array[i].next_blkoff = 0;
2150 }
2151 return restore_curseg_summaries(sbi);
2152}
2153
2154static void build_sit_entries(struct f2fs_sb_info *sbi)
2155{
2156 struct sit_info *sit_i = SIT_I(sbi);
2157 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
dfc08a12 2158 struct f2fs_journal *journal = &curseg->sum_blk->journal;
74de593a
CY		 2159 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2160 unsigned int i, start, end;
2161 unsigned int readed, start_blk = 0;
e9f5b8b8 2162 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2163
74de593a 2164 do {
26879fb1 2165 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2166
2167 start = start_blk * sit_i->sents_per_block;
2168 end = (start_blk + readed) * sit_i->sents_per_block;
2169
7cd8558b 2170 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2171 struct seg_entry *se = &sit_i->sentries[start];
2172 struct f2fs_sit_block *sit_blk;
2173 struct f2fs_sit_entry sit;
2174 struct page *page;
2175
2176 mutex_lock(&curseg->curseg_mutex);
dfc08a12
CY		 2177 for (i = 0; i < sits_in_cursum(journal); i++) {
2178 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2179 == start) {
dfc08a12 2180 sit = sit_in_journal(journal, i);
74de593a
CY		 2181 mutex_unlock(&curseg->curseg_mutex);
2182 goto got_it;
2183 }
351df4b2 2184 }
74de593a
CY		 2185 mutex_unlock(&curseg->curseg_mutex);
2186
2187 page = get_current_sit_page(sbi, start);
2188 sit_blk = (struct f2fs_sit_block *)page_address(page);
2189 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2190 f2fs_put_page(page, 1);
351df4b2 2191got_it:
74de593a
CY		 2192 check_block_count(sbi, start, &sit);
2193 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2194
2195 /* build discard map only one time */
2196 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2197 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2198
74de593a
CY		 2199 if (sbi->segs_per_sec > 1) {
2200 struct sec_entry *e = get_sec_entry(sbi, start);
2201 e->valid_blocks += se->valid_blocks;
2202 }
351df4b2 2203 }
74de593a
CY		 2204 start_blk += readed;
2205 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2206}
2207
2208static void init_free_segmap(struct f2fs_sb_info *sbi)
2209{
2210 unsigned int start;
2211 int type;
2212
7cd8558b 2213 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2214 struct seg_entry *sentry = get_seg_entry(sbi, start);
2215 if (!sentry->valid_blocks)
2216 __set_free(sbi, start);
2217 }
2218
2219 /* set use the current segments */
2220 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2221 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2222 __set_test_and_inuse(sbi, curseg_t->segno);
2223 }
2224}
2225
2226static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2227{
2228 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2229 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2230 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2231 unsigned short valid_blocks;
2232
8736fbf0 2233 while (1) {
351df4b2 2234 /* find dirty segment based on free segmap */
7cd8558b
JK		 2235 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2236 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2237 break;
2238 offset = segno + 1;
2239 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2240 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2241 continue;
ec325b52
JK		 2242 if (valid_blocks > sbi->blocks_per_seg) {
2243 f2fs_bug_on(sbi, 1);
2244 continue;
2245 }
351df4b2
JK		 2246 mutex_lock(&dirty_i->seglist_lock);
2247 __locate_dirty_segment(sbi, segno, DIRTY);
2248 mutex_unlock(&dirty_i->seglist_lock);
2249 }
2250}
2251
5ec4e49f 2252static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2253{
2254 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2255 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2256
39307a8e 2257 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2258 if (!dirty_i->victim_secmap)
351df4b2
JK		 2259 return -ENOMEM;
2260 return 0;
2261}
2262
2263static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2264{
2265 struct dirty_seglist_info *dirty_i;
2266 unsigned int bitmap_size, i;
2267
2268 /* allocate memory for dirty segments list information */
2269 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2270 if (!dirty_i)
2271 return -ENOMEM;
2272
2273 SM_I(sbi)->dirty_info = dirty_i;
2274 mutex_init(&dirty_i->seglist_lock);
2275
7cd8558b 2276 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2277
2278 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2279 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2280 if (!dirty_i->dirty_segmap[i])
2281 return -ENOMEM;
2282 }
2283
2284 init_dirty_segmap(sbi);
5ec4e49f 2285 return init_victim_secmap(sbi);
351df4b2
JK		 2286}
2287
0a8165d7 2288/*
351df4b2
JK		 2289 * Update min, max modified time for cost-benefit GC algorithm
2290 */
2291static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2292{
2293 struct sit_info *sit_i = SIT_I(sbi);
2294 unsigned int segno;
2295
2296 mutex_lock(&sit_i->sentry_lock);
2297
2298 sit_i->min_mtime = LLONG_MAX;
2299
7cd8558b 2300 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2301 unsigned int i;
2302 unsigned long long mtime = 0;
2303
2304 for (i = 0; i < sbi->segs_per_sec; i++)
2305 mtime += get_seg_entry(sbi, segno + i)->mtime;
2306
2307 mtime = div_u64(mtime, sbi->segs_per_sec);
2308
2309 if (sit_i->min_mtime > mtime)
2310 sit_i->min_mtime = mtime;
2311 }
2312 sit_i->max_mtime = get_mtime(sbi);
2313 mutex_unlock(&sit_i->sentry_lock);
2314}
2315
2316int build_segment_manager(struct f2fs_sb_info *sbi)
2317{
2318 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2319 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2320 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2321 int err;
2322
2323 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2324 if (!sm_info)
2325 return -ENOMEM;
2326
2327 /* init sm info */
2328 sbi->sm_info = sm_info;
351df4b2
JK		 2329 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2330 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2331 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2332 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2333 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2334 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2335 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2336 sm_info->rec_prefree_segments = sm_info->main_segments *
2337 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2338 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2339 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2340 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2341
7fd9e544
JK		 2342 INIT_LIST_HEAD(&sm_info->discard_list);
2343 sm_info->nr_discards = 0;
2344 sm_info->max_discards = 0;
2345
bba681cb
JK		 2346 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2347
184a5cd2
CY		 2348 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2349
b270ad6f 2350 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2351 err = create_flush_cmd_control(sbi);
2352 if (err)
a688b9d9 2353 return err;
6b4afdd7
JK		 2354 }
2355
351df4b2
JK		 2356 err = build_sit_info(sbi);
2357 if (err)
2358 return err;
2359 err = build_free_segmap(sbi);
2360 if (err)
2361 return err;
2362 err = build_curseg(sbi);
2363 if (err)
2364 return err;
2365
2366 /* reinit free segmap based on SIT */
2367 build_sit_entries(sbi);
2368
2369 init_free_segmap(sbi);
2370 err = build_dirty_segmap(sbi);
2371 if (err)
2372 return err;
2373
2374 init_min_max_mtime(sbi);
2375 return 0;
2376}
2377
2378static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2379 enum dirty_type dirty_type)
2380{
2381 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2382
2383 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2384 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2385 dirty_i->nr_dirty[dirty_type] = 0;
2386 mutex_unlock(&dirty_i->seglist_lock);
2387}
2388
5ec4e49f 2389static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2390{
2391 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2392 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2393}
2394
2395static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2396{
2397 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2398 int i;
2399
2400 if (!dirty_i)
2401 return;
2402
2403 /* discard pre-free/dirty segments list */
2404 for (i = 0; i < NR_DIRTY_TYPE; i++)
2405 discard_dirty_segmap(sbi, i);
2406
5ec4e49f 2407 destroy_victim_secmap(sbi);
351df4b2
JK		 2408 SM_I(sbi)->dirty_info = NULL;
2409 kfree(dirty_i);
2410}
2411
2412static void destroy_curseg(struct f2fs_sb_info *sbi)
2413{
2414 struct curseg_info *array = SM_I(sbi)->curseg_array;
2415 int i;
2416
2417 if (!array)
2418 return;
2419 SM_I(sbi)->curseg_array = NULL;
2420 for (i = 0; i < NR_CURSEG_TYPE; i++)
2421 kfree(array[i].sum_blk);
2422 kfree(array);
2423}
2424
2425static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2426{
2427 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2428 if (!free_i)
2429 return;
2430 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2431 kvfree(free_i->free_segmap);
2432 kvfree(free_i->free_secmap);
351df4b2
JK		 2433 kfree(free_i);
2434}
2435
2436static void destroy_sit_info(struct f2fs_sb_info *sbi)
2437{
2438 struct sit_info *sit_i = SIT_I(sbi);
2439 unsigned int start;
2440
2441 if (!sit_i)
2442 return;
2443
2444 if (sit_i->sentries) {
7cd8558b 2445 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2446 kfree(sit_i->sentries[start].cur_valid_map);
2447 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2448 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2449 }
2450 }
60a3b782
JK		 2451 kfree(sit_i->tmp_map);
2452
39307a8e
JK		 2453 kvfree(sit_i->sentries);
2454 kvfree(sit_i->sec_entries);
2455 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2456
2457 SM_I(sbi)->sit_info = NULL;
2458 kfree(sit_i->sit_bitmap);
2459 kfree(sit_i);
2460}
2461
2462void destroy_segment_manager(struct f2fs_sb_info *sbi)
2463{
2464 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2465
3b03f724
CY		 2466 if (!sm_info)
2467 return;
2163d198 2468 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2469 destroy_dirty_segmap(sbi);
2470 destroy_curseg(sbi);
2471 destroy_free_segmap(sbi);
2472 destroy_sit_info(sbi);
2473 sbi->sm_info = NULL;
2474 kfree(sm_info);
2475}
7fd9e544
JK		 2476
2477int __init create_segment_manager_caches(void)
2478{
2479 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2480 sizeof(struct discard_entry));
7fd9e544 2481 if (!discard_entry_slab)
184a5cd2
CY		 2482 goto fail;
2483
2484 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2485 sizeof(struct sit_entry_set));
184a5cd2
CY		 2486 if (!sit_entry_set_slab)
2487 goto destory_discard_entry;
88b88a66
JK		 2488
2489 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2490 sizeof(struct inmem_pages));
2491 if (!inmem_entry_slab)
2492 goto destroy_sit_entry_set;
7fd9e544 2493 return 0;
184a5cd2 2494
88b88a66
JK		 2495destroy_sit_entry_set:
2496 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2497destory_discard_entry:
2498 kmem_cache_destroy(discard_entry_slab);
2499fail:
2500 return -ENOMEM;
7fd9e544
JK		 2501}
2502
2503void destroy_segment_manager_caches(void)
2504{
184a5cd2 2505 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2506 kmem_cache_destroy(discard_entry_slab);
88b88a66 2507 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2508}
Linux kernel - Deliverables
This page took 0.260178 seconds and 5 git commands to generate.