projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: clean up foreground GC flow
[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:
63c52d78
JK		 226 /* we don't need to invalidate this in the sccessful status */
227 if (drop || recover)
228 ClearPageUptodate(page);
28bc106b 229 set_page_private(page, 0);
c81ced05 230 ClearPagePrivate(page);
28bc106b 231 f2fs_put_page(page, 1);
29b96b54
CY		 232
233 list_del(&cur->list);
234 kmem_cache_free(inmem_entry_slab, cur);
235 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
236 }
28bc106b 237 return err;
29b96b54
CY		 238}
239
240void drop_inmem_pages(struct inode *inode)
241{
242 struct f2fs_inode_info *fi = F2FS_I(inode);
243
91942321 244 clear_inode_flag(inode, FI_ATOMIC_FILE);
26dc3d44 245
29b96b54 246 mutex_lock(&fi->inmem_lock);
28bc106b 247 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54
CY		 248 mutex_unlock(&fi->inmem_lock);
249}
250
28bc106b
CY		 251static int __commit_inmem_pages(struct inode *inode,
252 struct list_head *revoke_list)
88b88a66
JK		 253{
254 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct inmem_pages *cur, *tmp;
88b88a66 257 struct f2fs_io_info fio = {
05ca3632 258 .sbi = sbi,
88b88a66 259 .type = DATA,
04d328de
MC		 260 .op = REQ_OP_WRITE,
261 .op_flags = WRITE_SYNC | REQ_PRIO,
4375a336 262 .encrypted_page = NULL,
88b88a66 263 };
29b96b54 264 bool submit_bio = false;
edb27dee 265 int err = 0;
88b88a66 266
88b88a66 267 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 268 struct page *page = cur->page;
269
270 lock_page(page);
271 if (page->mapping == inode->i_mapping) {
272 trace_f2fs_commit_inmem_page(page, INMEM);
273
274 set_page_dirty(page);
275 f2fs_wait_on_page_writeback(page, DATA, true);
276 if (clear_page_dirty_for_io(page))
29b96b54 277 inode_dec_dirty_pages(inode);
28bc106b
CY		 278
279 fio.page = page;
29b96b54
CY		 280 err = do_write_data_page(&fio);
281 if (err) {
28bc106b 282 unlock_page(page);
29b96b54 283 break;
70c640b1 284 }
29b96b54 285
28bc106b
CY		 286 /* record old blkaddr for revoking */
287 cur->old_addr = fio.old_blkaddr;
decd36b6 288
28bc106b
CY		 289 clear_cold_data(page);
290 submit_bio = true;
291 }
292 unlock_page(page);
293 list_move_tail(&cur->list, revoke_list);
88b88a66 294 }
29b96b54
CY		 295
296 if (submit_bio)
297 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
28bc106b
CY		 298
299 if (!err)
300 __revoke_inmem_pages(inode, revoke_list, false, false);
301
29b96b54
CY		 302 return err;
303}
304
305int commit_inmem_pages(struct inode *inode)
306{
307 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
308 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 309 struct list_head revoke_list;
310 int err;
29b96b54 311
28bc106b 312 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 313 f2fs_balance_fs(sbi, true);
314 f2fs_lock_op(sbi);
315
316 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 317 err = __commit_inmem_pages(inode, &revoke_list);
318 if (err) {
319 int ret;
320 /*
321 * try to revoke all committed pages, but still we could fail
322 * due to no memory or other reason, if that happened, EAGAIN
323 * will be returned, which means in such case, transaction is
324 * already not integrity, caller should use journal to do the
325 * recovery or rewrite & commit last transaction. For other
326 * error number, revoking was done by filesystem itself.
327 */
328 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
329 if (ret)
330 err = ret;
331
332 /* drop all uncommitted pages */
333 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
334 }
88b88a66
JK		 335 mutex_unlock(&fi->inmem_lock);
336
29b96b54 337 f2fs_unlock_op(sbi);
edb27dee 338 return err;
88b88a66
JK		 339}
340
0a8165d7 341/*
351df4b2
JK		 342 * This function balances dirty node and dentry pages.
343 * In addition, it controls garbage collection.
344 */
2c4db1a6 345void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 346{
2c4db1a6
JK		 347 if (!need)
348 return;
e589c2c4
JK		 349
350 /* balance_fs_bg is able to be pending */
351 if (excess_cached_nats(sbi))
352 f2fs_balance_fs_bg(sbi);
353
351df4b2 354 /*
029cd28c
JK		 355 * We should do GC or end up with checkpoint, if there are so many dirty
356 * dir/node pages without enough free segments.
351df4b2 357 */
43727527 358 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 359 mutex_lock(&sbi->gc_mutex);
d530d4d8 360 f2fs_gc(sbi, false);
351df4b2
JK		 361 }
362}
363
4660f9c0
JK		 364void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
365{
1dcc336b 366 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 367 if (!available_free_memory(sbi, EXTENT_CACHE))
368 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 369
1b38dc8e
JK		 370 /* check the # of cached NAT entries */
371 if (!available_free_memory(sbi, NAT_ENTRIES))
372 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
373
31696580 374 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 375 try_to_free_nids(sbi, MAX_FREE_NIDS);
376 else
377 build_free_nids(sbi);
31696580 378
1b38dc8e
JK		 379 /* checkpoint is the only way to shrink partial cached entries */
380 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 381 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 382 excess_prefree_segs(sbi) ||
383 excess_dirty_nats(sbi) ||
d0239e1b 384 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
e9f5b8b8
CY		 385 if (test_opt(sbi, DATA_FLUSH)) {
386 struct blk_plug plug;
387
388 blk_start_plug(&plug);
36b35a0d 389 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 390 blk_finish_plug(&plug);
391 }
4660f9c0 392 f2fs_sync_fs(sbi->sb, true);
42190d2a 393 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 394 }
4660f9c0
JK		 395}
396
2163d198 397static int issue_flush_thread(void *data)
6b4afdd7
JK		 398{
399 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 400 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
401 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 402repeat:
403 if (kthread_should_stop())
404 return 0;
405
721bd4d5 406 if (!llist_empty(&fcc->issue_list)) {
740432f8 407 struct bio *bio;
6b4afdd7
JK		 408 struct flush_cmd *cmd, *next;
409 int ret;
410
740432f8
JK		 411 bio = f2fs_bio_alloc(0);
412
721bd4d5
GZ		 413 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
414 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
415
6b4afdd7 416 bio->bi_bdev = sbi->sb->s_bdev;
04d328de 417 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
4e49ea4a 418 ret = submit_bio_wait(bio);
6b4afdd7 419
721bd4d5
GZ		 420 llist_for_each_entry_safe(cmd, next,
421 fcc->dispatch_list, llnode) {
6b4afdd7 422 cmd->ret = ret;
6b4afdd7
JK		 423 complete(&cmd->wait);
424 }
a4ed23f2 425 bio_put(bio);
a688b9d9 426 fcc->dispatch_list = NULL;
6b4afdd7
JK		 427 }
428
a688b9d9 429 wait_event_interruptible(*q,
721bd4d5 430 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 431 goto repeat;
432}
433
434int f2fs_issue_flush(struct f2fs_sb_info *sbi)
435{
a688b9d9 436 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 437 struct flush_cmd cmd;
6b4afdd7 438
24a9ee0f
JK		 439 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
440 test_opt(sbi, FLUSH_MERGE));
441
0f7b2abd
JK		 442 if (test_opt(sbi, NOBARRIER))
443 return 0;
444
0a87f664 445 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
740432f8
JK		 446 struct bio *bio = f2fs_bio_alloc(0);
447 int ret;
448
0a87f664 449 atomic_inc(&fcc->submit_flush);
740432f8 450 bio->bi_bdev = sbi->sb->s_bdev;
04d328de 451 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
4e49ea4a 452 ret = submit_bio_wait(bio);
0a87f664 453 atomic_dec(&fcc->submit_flush);
740432f8
JK		 454 bio_put(bio);
455 return ret;
456 }
6b4afdd7 457
adf8d90b 458 init_completion(&cmd.wait);
6b4afdd7 459
0a87f664 460 atomic_inc(&fcc->submit_flush);
721bd4d5 461 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 462
a688b9d9
GZ		 463 if (!fcc->dispatch_list)
464 wake_up(&fcc->flush_wait_queue);
6b4afdd7 465
adf8d90b 466 wait_for_completion(&cmd.wait);
0a87f664 467 atomic_dec(&fcc->submit_flush);
adf8d90b
CY		 468
469 return cmd.ret;
6b4afdd7
JK		 470}
471
2163d198
GZ		 472int create_flush_cmd_control(struct f2fs_sb_info *sbi)
473{
474 dev_t dev = sbi->sb->s_bdev->bd_dev;
475 struct flush_cmd_control *fcc;
476 int err = 0;
477
478 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
479 if (!fcc)
480 return -ENOMEM;
0a87f664 481 atomic_set(&fcc->submit_flush, 0);
2163d198 482 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 483 init_llist_head(&fcc->issue_list);
6b2920a5 484 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 485 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
486 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
487 if (IS_ERR(fcc->f2fs_issue_flush)) {
488 err = PTR_ERR(fcc->f2fs_issue_flush);
489 kfree(fcc);
6b2920a5 490 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 491 return err;
492 }
2163d198
GZ		 493
494 return err;
495}
496
497void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
498{
6b2920a5 499 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 500
501 if (fcc && fcc->f2fs_issue_flush)
502 kthread_stop(fcc->f2fs_issue_flush);
503 kfree(fcc);
6b2920a5 504 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 505}
506
351df4b2
JK		 507static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
508 enum dirty_type dirty_type)
509{
510 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
511
512 /* need not be added */
513 if (IS_CURSEG(sbi, segno))
514 return;
515
516 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
517 dirty_i->nr_dirty[dirty_type]++;
518
519 if (dirty_type == DIRTY) {
520 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 521 enum dirty_type t = sentry->type;
b2f2c390 522
ec325b52
JK		 523 if (unlikely(t >= DIRTY)) {
524 f2fs_bug_on(sbi, 1);
525 return;
526 }
4625d6aa
CL		 527 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
528 dirty_i->nr_dirty[t]++;
351df4b2
JK		 529 }
530}
531
532static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
533 enum dirty_type dirty_type)
534{
535 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
536
537 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
538 dirty_i->nr_dirty[dirty_type]--;
539
540 if (dirty_type == DIRTY) {
4625d6aa
CL		 541 struct seg_entry *sentry = get_seg_entry(sbi, segno);
542 enum dirty_type t = sentry->type;
543
544 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
545 dirty_i->nr_dirty[t]--;
b2f2c390 546
5ec4e49f
JK		 547 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
548 clear_bit(GET_SECNO(sbi, segno),
549 dirty_i->victim_secmap);
351df4b2
JK		 550 }
551}
552
0a8165d7 553/*
351df4b2
JK		 554 * Should not occur error such as -ENOMEM.
555 * Adding dirty entry into seglist is not critical operation.
556 * If a given segment is one of current working segments, it won't be added.
557 */
8d8451af 558static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 559{
560 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
561 unsigned short valid_blocks;
562
563 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
564 return;
565
566 mutex_lock(&dirty_i->seglist_lock);
567
568 valid_blocks = get_valid_blocks(sbi, segno, 0);
569
570 if (valid_blocks == 0) {
571 __locate_dirty_segment(sbi, segno, PRE);
572 __remove_dirty_segment(sbi, segno, DIRTY);
573 } else if (valid_blocks < sbi->blocks_per_seg) {
574 __locate_dirty_segment(sbi, segno, DIRTY);
575 } else {
576 /* Recovery routine with SSR needs this */
577 __remove_dirty_segment(sbi, segno, DIRTY);
578 }
579
580 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 581}
582
1e87a78d 583static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 584 block_t blkstart, block_t blklen)
585{
55cf9cb6
CY		 586 sector_t start = SECTOR_FROM_BLOCK(blkstart);
587 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 588 struct seg_entry *se;
589 unsigned int offset;
590 block_t i;
591
592 for (i = blkstart; i < blkstart + blklen; i++) {
593 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
594 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
595
596 if (!f2fs_test_and_set_bit(offset, se->discard_map))
597 sbi->discard_blks--;
598 }
1661d07c 599 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 600 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
601}
602
e90c2d28 603bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 604{
60b286c4 605 int err = -EOPNOTSUPP;
40a02be1
JK		 606
607 if (test_opt(sbi, DISCARD)) {
608 struct seg_entry *se = get_seg_entry(sbi,
609 GET_SEGNO(sbi, blkaddr));
610 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
611
612 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 613 return false;
40a02be1
JK		 614
615 err = f2fs_issue_discard(sbi, blkaddr, 1);
616 }
617
e90c2d28 618 if (err) {
381722d2 619 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 620 return true;
621 }
622 return false;
37208879
JK		 623}
624
adf4983b 625static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 626 struct cp_control *cpc, struct seg_entry *se,
627 unsigned int start, unsigned int end)
b2955550
JK		 628{
629 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 630 struct discard_entry *new, *last;
631
632 if (!list_empty(head)) {
633 last = list_last_entry(head, struct discard_entry, list);
634 if (START_BLOCK(sbi, cpc->trim_start) + start ==
635 last->blkaddr + last->len) {
636 last->len += end - start;
637 goto done;
638 }
639 }
640
641 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
642 INIT_LIST_HEAD(&new->list);
643 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
644 new->len = end - start;
645 list_add_tail(&new->list, head);
646done:
647 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 648}
649
650static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
651{
b2955550
JK		 652 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
653 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 654 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 655 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
656 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 657 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 658 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 659 unsigned int start = 0, end = -1;
4b2fecc8 660 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 661 int i;
662
3e025740 663 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
b2955550
JK		 664 return;
665
a66cdd98
JK		 666 if (!force) {
667 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 668 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
669 return;
4b2fecc8
JK		 670 }
671
b2955550
JK		 672 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
673 for (i = 0; i < entries; i++)
a66cdd98 674 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 675 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 676
4b2fecc8 677 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 678 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
679 if (start >= max_blocks)
680 break;
681
682 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 683 if (force && start && end != max_blocks
684 && (end - start) < cpc->trim_minlen)
685 continue;
686
a66cdd98 687 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 688 }
689}
690
4b2fecc8
JK		 691void release_discard_addrs(struct f2fs_sb_info *sbi)
692{
693 struct list_head *head = &(SM_I(sbi)->discard_list);
694 struct discard_entry *entry, *this;
695
696 /* drop caches */
697 list_for_each_entry_safe(entry, this, head, list) {
698 list_del(&entry->list);
699 kmem_cache_free(discard_entry_slab, entry);
700 }
701}
702
0a8165d7 703/*
351df4b2
JK		 704 * Should call clear_prefree_segments after checkpoint is done.
705 */
706static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
707{
708 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 709 unsigned int segno;
351df4b2
JK		 710
711 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 712 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 713 __set_test_and_free(sbi, segno);
351df4b2
JK		 714 mutex_unlock(&dirty_i->seglist_lock);
715}
716
836b5a63 717void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 718{
b2955550 719 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 720 struct discard_entry *entry, *this;
351df4b2 721 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 722 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 723 unsigned int start = 0, end = -1;
36abef4e 724 unsigned int secno, start_segno;
c24a0fd6 725 bool force = (cpc->reason == CP_DISCARD);
351df4b2
JK		 726
727 mutex_lock(&dirty_i->seglist_lock);
29e59c14 728
351df4b2 729 while (1) {
29e59c14 730 int i;
7cd8558b
JK		 731 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
732 if (start >= MAIN_SEGS(sbi))
351df4b2 733 break;
7cd8558b
JK		 734 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
735 start + 1);
29e59c14
CL		 736
737 for (i = start; i < end; i++)
738 clear_bit(i, prefree_map);
739
740 dirty_i->nr_dirty[PRE] -= end - start;
741
c24a0fd6 742 if (force || !test_opt(sbi, DISCARD))
29e59c14 743 continue;
351df4b2 744
36abef4e
JK		 745 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
746 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 747 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 748 continue;
749 }
750next:
751 secno = GET_SECNO(sbi, start);
752 start_segno = secno * sbi->segs_per_sec;
753 if (!IS_CURSEC(sbi, secno) &&
754 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
755 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
756 sbi->segs_per_sec << sbi->log_blocks_per_seg);
757
758 start = start_segno + sbi->segs_per_sec;
759 if (start < end)
760 goto next;
351df4b2
JK		 761 }
762 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 763
764 /* send small discards */
2d7b822a 765 list_for_each_entry_safe(entry, this, head, list) {
c24a0fd6 766 if (force && entry->len < cpc->trim_minlen)
836b5a63 767 goto skip;
37208879 768 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 769 cpc->trimmed += entry->len;
836b5a63 770skip:
b2955550
JK		 771 list_del(&entry->list);
772 SM_I(sbi)->nr_discards -= entry->len;
773 kmem_cache_free(discard_entry_slab, entry);
774 }
351df4b2
JK		 775}
776
184a5cd2 777static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 778{
779 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 780
781 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 782 sit_i->dirty_sentries++;
184a5cd2
CY		 783 return false;
784 }
785
786 return true;
351df4b2
JK		 787}
788
789static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
790 unsigned int segno, int modified)
791{
792 struct seg_entry *se = get_seg_entry(sbi, segno);
793 se->type = type;
794 if (modified)
795 __mark_sit_entry_dirty(sbi, segno);
796}
797
798static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
799{
800 struct seg_entry *se;
801 unsigned int segno, offset;
802 long int new_vblocks;
803
804 segno = GET_SEGNO(sbi, blkaddr);
805
806 se = get_seg_entry(sbi, segno);
807 new_vblocks = se->valid_blocks + del;
491c0854 808 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 809
9850cf4a 810 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 811 (new_vblocks > sbi->blocks_per_seg)));
812
813 se->valid_blocks = new_vblocks;
814 se->mtime = get_mtime(sbi);
815 SIT_I(sbi)->max_mtime = se->mtime;
816
817 /* Update valid block bitmap */
818 if (del > 0) {
52aca074 819 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 820 f2fs_bug_on(sbi, 1);
3e025740
JK		 821 if (f2fs_discard_en(sbi) &&
822 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 823 sbi->discard_blks--;
351df4b2 824 } else {
52aca074 825 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 826 f2fs_bug_on(sbi, 1);
3e025740
JK		 827 if (f2fs_discard_en(sbi) &&
828 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 829 sbi->discard_blks++;
351df4b2
JK		 830 }
831 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
832 se->ckpt_valid_blocks += del;
833
834 __mark_sit_entry_dirty(sbi, segno);
835
836 /* update total number of valid blocks to be written in ckpt area */
837 SIT_I(sbi)->written_valid_blocks += del;
838
839 if (sbi->segs_per_sec > 1)
840 get_sec_entry(sbi, segno)->valid_blocks += del;
841}
842
5e443818 843void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 844{
5e443818
JK		 845 update_sit_entry(sbi, new, 1);
846 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
847 update_sit_entry(sbi, old, -1);
848
849 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
850 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 851}
852
853void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
854{
855 unsigned int segno = GET_SEGNO(sbi, addr);
856 struct sit_info *sit_i = SIT_I(sbi);
857
9850cf4a 858 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 859 if (addr == NEW_ADDR)
860 return;
861
862 /* add it into sit main buffer */
863 mutex_lock(&sit_i->sentry_lock);
864
865 update_sit_entry(sbi, addr, -1);
866
867 /* add it into dirty seglist */
868 locate_dirty_segment(sbi, segno);
869
870 mutex_unlock(&sit_i->sentry_lock);
871}
872
6e2c64ad
JK		 873bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
874{
875 struct sit_info *sit_i = SIT_I(sbi);
876 unsigned int segno, offset;
877 struct seg_entry *se;
878 bool is_cp = false;
879
880 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
881 return true;
882
883 mutex_lock(&sit_i->sentry_lock);
884
885 segno = GET_SEGNO(sbi, blkaddr);
886 se = get_seg_entry(sbi, segno);
887 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
888
889 if (f2fs_test_bit(offset, se->ckpt_valid_map))
890 is_cp = true;
891
892 mutex_unlock(&sit_i->sentry_lock);
893
894 return is_cp;
895}
896
0a8165d7 897/*
351df4b2
JK		 898 * This function should be resided under the curseg_mutex lock
899 */
900static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 901 struct f2fs_summary *sum)
351df4b2
JK		 902{
903 struct curseg_info *curseg = CURSEG_I(sbi, type);
904 void *addr = curseg->sum_blk;
e79efe3b 905 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 906 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 907}
908
0a8165d7 909/*
351df4b2
JK		 910 * Calculate the number of current summary pages for writing
911 */
3fa06d7b 912int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 913{
351df4b2 914 int valid_sum_count = 0;
9a47938b 915 int i, sum_in_page;
351df4b2
JK		 916
917 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
918 if (sbi->ckpt->alloc_type[i] == SSR)
919 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 920 else {
921 if (for_ra)
922 valid_sum_count += le16_to_cpu(
923 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
924 else
925 valid_sum_count += curseg_blkoff(sbi, i);
926 }
351df4b2
JK		 927 }
928
09cbfeaf 929 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 930 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
931 if (valid_sum_count <= sum_in_page)
351df4b2 932 return 1;
9a47938b 933 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 934 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 935 return 2;
936 return 3;
937}
938
0a8165d7 939/*
351df4b2
JK		 940 * Caller should put this summary page
941 */
942struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
943{
944 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
945}
946
381722d2 947void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 948{
949 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 950 void *dst = page_address(page);
951
952 if (src)
09cbfeaf 953 memcpy(dst, src, PAGE_SIZE);
381722d2 954 else
09cbfeaf 955 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 956 set_page_dirty(page);
957 f2fs_put_page(page, 1);
958}
959
381722d2
CY		 960static void write_sum_page(struct f2fs_sb_info *sbi,
961 struct f2fs_summary_block *sum_blk, block_t blk_addr)
962{
963 update_meta_page(sbi, (void *)sum_blk, blk_addr);
964}
965
b7ad7512
CY		 966static void write_current_sum_page(struct f2fs_sb_info *sbi,
967 int type, block_t blk_addr)
968{
969 struct curseg_info *curseg = CURSEG_I(sbi, type);
970 struct page *page = grab_meta_page(sbi, blk_addr);
971 struct f2fs_summary_block *src = curseg->sum_blk;
972 struct f2fs_summary_block *dst;
973
974 dst = (struct f2fs_summary_block *)page_address(page);
975
976 mutex_lock(&curseg->curseg_mutex);
977
978 down_read(&curseg->journal_rwsem);
979 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
980 up_read(&curseg->journal_rwsem);
981
982 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
983 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
984
985 mutex_unlock(&curseg->curseg_mutex);
986
987 set_page_dirty(page);
988 f2fs_put_page(page, 1);
989}
990
60374688
JK		 991static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
992{
993 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 994 unsigned int segno = curseg->segno + 1;
60374688
JK		 995 struct free_segmap_info *free_i = FREE_I(sbi);
996
7cd8558b 997 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 998 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 999 return 0;
1000}
1001
0a8165d7 1002/*
351df4b2
JK		 1003 * Find a new segment from the free segments bitmap to right order
1004 * This function should be returned with success, otherwise BUG
1005 */
1006static void get_new_segment(struct f2fs_sb_info *sbi,
1007 unsigned int *newseg, bool new_sec, int dir)
1008{
1009 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1010 unsigned int segno, secno, zoneno;
7cd8558b 1011 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1012 unsigned int hint = *newseg / sbi->segs_per_sec;
1013 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1014 unsigned int left_start = hint;
1015 bool init = true;
1016 int go_left = 0;
1017 int i;
1018
1a118ccf 1019 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1020
1021 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1022 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1023 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1024 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1025 goto got_it;
1026 }
1027find_other_zone:
7cd8558b
JK		 1028 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1029 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1030 if (dir == ALLOC_RIGHT) {
1031 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1032 MAIN_SECS(sbi), 0);
1033 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1034 } else {
1035 go_left = 1;
1036 left_start = hint - 1;
1037 }
1038 }
1039 if (go_left == 0)
1040 goto skip_left;
1041
1042 while (test_bit(left_start, free_i->free_secmap)) {
1043 if (left_start > 0) {
1044 left_start--;
1045 continue;
1046 }
1047 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1048 MAIN_SECS(sbi), 0);
1049 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1050 break;
1051 }
1052 secno = left_start;
1053skip_left:
1054 hint = secno;
1055 segno = secno * sbi->segs_per_sec;
1056 zoneno = secno / sbi->secs_per_zone;
1057
1058 /* give up on finding another zone */
1059 if (!init)
1060 goto got_it;
1061 if (sbi->secs_per_zone == 1)
1062 goto got_it;
1063 if (zoneno == old_zoneno)
1064 goto got_it;
1065 if (dir == ALLOC_LEFT) {
1066 if (!go_left && zoneno + 1 >= total_zones)
1067 goto got_it;
1068 if (go_left && zoneno == 0)
1069 goto got_it;
1070 }
1071 for (i = 0; i < NR_CURSEG_TYPE; i++)
1072 if (CURSEG_I(sbi, i)->zone == zoneno)
1073 break;
1074
1075 if (i < NR_CURSEG_TYPE) {
1076 /* zone is in user, try another */
1077 if (go_left)
1078 hint = zoneno * sbi->secs_per_zone - 1;
1079 else if (zoneno + 1 >= total_zones)
1080 hint = 0;
1081 else
1082 hint = (zoneno + 1) * sbi->secs_per_zone;
1083 init = false;
1084 goto find_other_zone;
1085 }
1086got_it:
1087 /* set it as dirty segment in free segmap */
9850cf4a 1088 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1089 __set_inuse(sbi, segno);
1090 *newseg = segno;
1a118ccf 1091 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1092}
1093
1094static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1095{
1096 struct curseg_info *curseg = CURSEG_I(sbi, type);
1097 struct summary_footer *sum_footer;
1098
1099 curseg->segno = curseg->next_segno;
1100 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1101 curseg->next_blkoff = 0;
1102 curseg->next_segno = NULL_SEGNO;
1103
1104 sum_footer = &(curseg->sum_blk->footer);
1105 memset(sum_footer, 0, sizeof(struct summary_footer));
1106 if (IS_DATASEG(type))
1107 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1108 if (IS_NODESEG(type))
1109 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1110 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1111}
1112
0a8165d7 1113/*
351df4b2
JK		 1114 * Allocate a current working segment.
1115 * This function always allocates a free segment in LFS manner.
1116 */
1117static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1118{
1119 struct curseg_info *curseg = CURSEG_I(sbi, type);
1120 unsigned int segno = curseg->segno;
1121 int dir = ALLOC_LEFT;
1122
1123 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1124 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1125 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1126 dir = ALLOC_RIGHT;
1127
1128 if (test_opt(sbi, NOHEAP))
1129 dir = ALLOC_RIGHT;
1130
1131 get_new_segment(sbi, &segno, new_sec, dir);
1132 curseg->next_segno = segno;
1133 reset_curseg(sbi, type, 1);
1134 curseg->alloc_type = LFS;
1135}
1136
1137static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1138 struct curseg_info *seg, block_t start)
1139{
1140 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1141 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1142 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1143 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1144 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1145 int i, pos;
1146
1147 for (i = 0; i < entries; i++)
1148 target_map[i] = ckpt_map[i] | cur_map[i];
1149
1150 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1151
1152 seg->next_blkoff = pos;
351df4b2
JK		 1153}
1154
0a8165d7 1155/*
351df4b2
JK		 1156 * If a segment is written by LFS manner, next block offset is just obtained
1157 * by increasing the current block offset. However, if a segment is written by
1158 * SSR manner, next block offset obtained by calling __next_free_blkoff
1159 */
1160static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1161 struct curseg_info *seg)
1162{
1163 if (seg->alloc_type == SSR)
1164 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1165 else
1166 seg->next_blkoff++;
1167}
1168
0a8165d7 1169/*
e1c42045 1170 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1171 * manner, so it should recover the existing segment information of valid blocks
1172 */
1173static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1174{
1175 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1176 struct curseg_info *curseg = CURSEG_I(sbi, type);
1177 unsigned int new_segno = curseg->next_segno;
1178 struct f2fs_summary_block *sum_node;
1179 struct page *sum_page;
1180
1181 write_sum_page(sbi, curseg->sum_blk,
1182 GET_SUM_BLOCK(sbi, curseg->segno));
1183 __set_test_and_inuse(sbi, new_segno);
1184
1185 mutex_lock(&dirty_i->seglist_lock);
1186 __remove_dirty_segment(sbi, new_segno, PRE);
1187 __remove_dirty_segment(sbi, new_segno, DIRTY);
1188 mutex_unlock(&dirty_i->seglist_lock);
1189
1190 reset_curseg(sbi, type, 1);
1191 curseg->alloc_type = SSR;
1192 __next_free_blkoff(sbi, curseg, 0);
1193
1194 if (reuse) {
1195 sum_page = get_sum_page(sbi, new_segno);
1196 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1197 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1198 f2fs_put_page(sum_page, 1);
1199 }
1200}
1201
43727527
JK		 1202static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1203{
1204 struct curseg_info *curseg = CURSEG_I(sbi, type);
1205 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1206
1207 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1208 return v_ops->get_victim(sbi,
1209 &(curseg)->next_segno, BG_GC, type, SSR);
1210
1211 /* For data segments, let's do SSR more intensively */
1212 for (; type >= CURSEG_HOT_DATA; type--)
1213 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1214 BG_GC, type, SSR))
1215 return 1;
1216 return 0;
1217}
1218
351df4b2
JK		 1219/*
1220 * flush out current segment and replace it with new segment
1221 * This function should be returned with success, otherwise BUG
1222 */
1223static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1224 int type, bool force)
1225{
1226 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1227
7b405275 1228 if (force)
351df4b2 1229 new_curseg(sbi, type, true);
7b405275 1230 else if (type == CURSEG_WARM_NODE)
351df4b2 1231 new_curseg(sbi, type, false);
60374688
JK		 1232 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1233 new_curseg(sbi, type, false);
351df4b2
JK		 1234 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1235 change_curseg(sbi, type, true);
1236 else
1237 new_curseg(sbi, type, false);
dcdfff65
JK		 1238
1239 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1240}
1241
38aa0889
JK		 1242static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1243{
1244 struct curseg_info *curseg = CURSEG_I(sbi, type);
1245 unsigned int old_segno;
1246
1247 old_segno = curseg->segno;
1248 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1249 locate_dirty_segment(sbi, old_segno);
1250}
1251
351df4b2
JK		 1252void allocate_new_segments(struct f2fs_sb_info *sbi)
1253{
351df4b2
JK		 1254 int i;
1255
36abef4e
JK		 1256 if (test_opt(sbi, LFS))
1257 return;
1258
38aa0889
JK		 1259 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1260 __allocate_new_segments(sbi, i);
351df4b2
JK		 1261}
1262
1263static const struct segment_allocation default_salloc_ops = {
1264 .allocate_segment = allocate_segment_by_default,
1265};
1266
4b2fecc8
JK		 1267int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1268{
f7ef9b83
JK		 1269 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1270 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1271 unsigned int start_segno, end_segno;
1272 struct cp_control cpc;
c34f42e2 1273 int err = 0;
4b2fecc8 1274
836b5a63 1275 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1276 return -EINVAL;
1277
9bd27ae4 1278 cpc.trimmed = 0;
7cd8558b 1279 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1280 goto out;
1281
1282 /* start/end segment number in main_area */
7cd8558b
JK		 1283 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1284 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1285 GET_SEGNO(sbi, end);
4b2fecc8 1286 cpc.reason = CP_DISCARD;
836b5a63 1287 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1288
1289 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1290 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1291 cpc.trim_start = start_segno;
a66cdd98
JK		 1292
1293 if (sbi->discard_blks == 0)
1294 break;
1295 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1296 cpc.trim_end = end_segno;
1297 else
1298 cpc.trim_end = min_t(unsigned int,
1299 rounddown(start_segno +
bba681cb
JK		 1300 BATCHED_TRIM_SEGMENTS(sbi),
1301 sbi->segs_per_sec) - 1, end_segno);
1302
1303 mutex_lock(&sbi->gc_mutex);
c34f42e2 1304 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1305 mutex_unlock(&sbi->gc_mutex);
1306 }
4b2fecc8 1307out:
f7ef9b83 1308 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1309 return err;
4b2fecc8
JK		 1310}
1311
351df4b2
JK		 1312static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1313{
1314 struct curseg_info *curseg = CURSEG_I(sbi, type);
1315 if (curseg->next_blkoff < sbi->blocks_per_seg)
1316 return true;
1317 return false;
1318}
1319
1320static int __get_segment_type_2(struct page *page, enum page_type p_type)
1321{
1322 if (p_type == DATA)
1323 return CURSEG_HOT_DATA;
1324 else
1325 return CURSEG_HOT_NODE;
1326}
1327
1328static int __get_segment_type_4(struct page *page, enum page_type p_type)
1329{
1330 if (p_type == DATA) {
1331 struct inode *inode = page->mapping->host;
1332
1333 if (S_ISDIR(inode->i_mode))
1334 return CURSEG_HOT_DATA;
1335 else
1336 return CURSEG_COLD_DATA;
1337 } else {
a344b9fd
JK		 1338 if (IS_DNODE(page) && is_cold_node(page))
1339 return CURSEG_WARM_NODE;
351df4b2
JK		 1340 else
1341 return CURSEG_COLD_NODE;
1342 }
1343}
1344
1345static int __get_segment_type_6(struct page *page, enum page_type p_type)
1346{
1347 if (p_type == DATA) {
1348 struct inode *inode = page->mapping->host;
1349
1350 if (S_ISDIR(inode->i_mode))
1351 return CURSEG_HOT_DATA;
354a3399 1352 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1353 return CURSEG_COLD_DATA;
1354 else
1355 return CURSEG_WARM_DATA;
1356 } else {
1357 if (IS_DNODE(page))
1358 return is_cold_node(page) ? CURSEG_WARM_NODE :
1359 CURSEG_HOT_NODE;
1360 else
1361 return CURSEG_COLD_NODE;
1362 }
1363}
1364
1365static int __get_segment_type(struct page *page, enum page_type p_type)
1366{
4081363f 1367 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1368 case 2:
1369 return __get_segment_type_2(page, p_type);
1370 case 4:
1371 return __get_segment_type_4(page, p_type);
351df4b2 1372 }
12a67146 1373 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1374 f2fs_bug_on(F2FS_P_SB(page),
1375 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1376 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1377}
1378
bfad7c2d
JK		 1379void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1380 block_t old_blkaddr, block_t *new_blkaddr,
1381 struct f2fs_summary *sum, int type)
351df4b2
JK		 1382{
1383 struct sit_info *sit_i = SIT_I(sbi);
1384 struct curseg_info *curseg;
38aa0889
JK		 1385 bool direct_io = (type == CURSEG_DIRECT_IO);
1386
1387 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1388
351df4b2
JK		 1389 curseg = CURSEG_I(sbi, type);
1390
1391 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1392 mutex_lock(&sit_i->sentry_lock);
351df4b2 1393
38aa0889 1394 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1395 if (direct_io && curseg->next_blkoff &&
1396 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1397 __allocate_new_segments(sbi, type);
1398
351df4b2 1399 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1400
1401 /*
1402 * __add_sum_entry should be resided under the curseg_mutex
1403 * because, this function updates a summary entry in the
1404 * current summary block.
1405 */
e79efe3b 1406 __add_sum_entry(sbi, type, sum);
351df4b2 1407
351df4b2 1408 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1409
1410 stat_inc_block_count(sbi, curseg);
351df4b2 1411
5e443818
JK		 1412 if (!__has_curseg_space(sbi, type))
1413 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1414 /*
1415 * SIT information should be updated before segment allocation,
1416 * since SSR needs latest valid block information.
1417 */
1418 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1419
351df4b2
JK		 1420 mutex_unlock(&sit_i->sentry_lock);
1421
bfad7c2d 1422 if (page && IS_NODESEG(type))
351df4b2
JK		 1423 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1424
bfad7c2d
JK		 1425 mutex_unlock(&curseg->curseg_mutex);
1426}
1427
05ca3632 1428static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1429{
05ca3632 1430 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1431
7dfeaa32
JK		 1432 if (fio->type == NODE || fio->type == DATA)
1433 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1434
7a9d7548
CY		 1435 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1436 &fio->new_blkaddr, sum, type);
bfad7c2d 1437
351df4b2 1438 /* writeout dirty page into bdev */
05ca3632 1439 f2fs_submit_page_mbio(fio);
7dfeaa32
JK		 1440
1441 if (fio->type == NODE || fio->type == DATA)
1442 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1443}
1444
577e3495 1445void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1446{
458e6197 1447 struct f2fs_io_info fio = {
05ca3632 1448 .sbi = sbi,
458e6197 1449 .type = META,
04d328de
MC		 1450 .op = REQ_OP_WRITE,
1451 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1452 .old_blkaddr = page->index,
1453 .new_blkaddr = page->index,
05ca3632 1454 .page = page,
4375a336 1455 .encrypted_page = NULL,
458e6197
JK		 1456 };
1457
2b947003 1458 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1459 fio.op_flags &= ~REQ_META;
2b947003 1460
351df4b2 1461 set_page_writeback(page);
05ca3632 1462 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1463}
1464
05ca3632 1465void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1466{
1467 struct f2fs_summary sum;
05ca3632 1468
351df4b2 1469 set_summary(&sum, nid, 0, 0);
05ca3632 1470 do_write_page(&sum, fio);
351df4b2
JK		 1471}
1472
05ca3632 1473void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1474{
05ca3632 1475 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1476 struct f2fs_summary sum;
1477 struct node_info ni;
1478
9850cf4a 1479 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1480 get_node_info(sbi, dn->nid, &ni);
1481 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1482 do_write_page(&sum, fio);
f28b3434 1483 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1484}
1485
05ca3632 1486void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1487{
7a9d7548 1488 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1489 stat_inc_inplace_blocks(fio->sbi);
1490 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1491}
1492
4356e48e 1493void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1494 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1495 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1496{
1497 struct sit_info *sit_i = SIT_I(sbi);
1498 struct curseg_info *curseg;
1499 unsigned int segno, old_cursegno;
1500 struct seg_entry *se;
1501 int type;
19f106bc 1502 unsigned short old_blkoff;
351df4b2
JK		 1503
1504 segno = GET_SEGNO(sbi, new_blkaddr);
1505 se = get_seg_entry(sbi, segno);
1506 type = se->type;
1507
19f106bc
CY		 1508 if (!recover_curseg) {
1509 /* for recovery flow */
1510 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1511 if (old_blkaddr == NULL_ADDR)
1512 type = CURSEG_COLD_DATA;
1513 else
1514 type = CURSEG_WARM_DATA;
1515 }
1516 } else {
1517 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1518 type = CURSEG_WARM_DATA;
1519 }
19f106bc 1520
351df4b2
JK		 1521 curseg = CURSEG_I(sbi, type);
1522
1523 mutex_lock(&curseg->curseg_mutex);
1524 mutex_lock(&sit_i->sentry_lock);
1525
1526 old_cursegno = curseg->segno;
19f106bc 1527 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1528
1529 /* change the current segment */
1530 if (segno != curseg->segno) {
1531 curseg->next_segno = segno;
1532 change_curseg(sbi, type, true);
1533 }
1534
491c0854 1535 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1536 __add_sum_entry(sbi, type, sum);
351df4b2 1537
28bc106b 1538 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1539 update_sit_entry(sbi, new_blkaddr, 1);
1540 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1541 update_sit_entry(sbi, old_blkaddr, -1);
1542
1543 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1544 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1545
351df4b2 1546 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1547
19f106bc
CY		 1548 if (recover_curseg) {
1549 if (old_cursegno != curseg->segno) {
1550 curseg->next_segno = old_cursegno;
1551 change_curseg(sbi, type, true);
1552 }
1553 curseg->next_blkoff = old_blkoff;
1554 }
1555
351df4b2
JK		 1556 mutex_unlock(&sit_i->sentry_lock);
1557 mutex_unlock(&curseg->curseg_mutex);
1558}
1559
528e3459
CY		 1560void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1561 block_t old_addr, block_t new_addr,
28bc106b
CY		 1562 unsigned char version, bool recover_curseg,
1563 bool recover_newaddr)
528e3459
CY		 1564{
1565 struct f2fs_summary sum;
1566
1567 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1568
28bc106b
CY		 1569 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1570 recover_curseg, recover_newaddr);
528e3459 1571
f28b3434 1572 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1573}
1574
93dfe2ac 1575void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1576 enum page_type type, bool ordered)
93dfe2ac 1577{
93dfe2ac 1578 if (PageWriteback(page)) {
4081363f
JK		 1579 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1580
0c3a5797 1581 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1582 if (ordered)
1583 wait_on_page_writeback(page);
1584 else
1585 wait_for_stable_page(page);
93dfe2ac
JK		 1586 }
1587}
1588
08b39fbd
CY		 1589void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1590 block_t blkaddr)
1591{
1592 struct page *cpage;
1593
1594 if (blkaddr == NEW_ADDR)
1595 return;
1596
1597 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1598
1599 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1600 if (cpage) {
fec1d657 1601 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1602 f2fs_put_page(cpage, 1);
1603 }
1604}
1605
351df4b2
JK		 1606static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1607{
1608 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1609 struct curseg_info *seg_i;
1610 unsigned char *kaddr;
1611 struct page *page;
1612 block_t start;
1613 int i, j, offset;
1614
1615 start = start_sum_block(sbi);
1616
1617 page = get_meta_page(sbi, start++);
1618 kaddr = (unsigned char *)page_address(page);
1619
1620 /* Step 1: restore nat cache */
1621 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1622 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1623
1624 /* Step 2: restore sit cache */
1625 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1626 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1627 offset = 2 * SUM_JOURNAL_SIZE;
1628
1629 /* Step 3: restore summary entries */
1630 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1631 unsigned short blk_off;
1632 unsigned int segno;
1633
1634 seg_i = CURSEG_I(sbi, i);
1635 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1636 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1637 seg_i->next_segno = segno;
1638 reset_curseg(sbi, i, 0);
1639 seg_i->alloc_type = ckpt->alloc_type[i];
1640 seg_i->next_blkoff = blk_off;
1641
1642 if (seg_i->alloc_type == SSR)
1643 blk_off = sbi->blocks_per_seg;
1644
1645 for (j = 0; j < blk_off; j++) {
1646 struct f2fs_summary *s;
1647 s = (struct f2fs_summary *)(kaddr + offset);
1648 seg_i->sum_blk->entries[j] = *s;
1649 offset += SUMMARY_SIZE;
09cbfeaf 1650 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1651 SUM_FOOTER_SIZE)
1652 continue;
1653
1654 f2fs_put_page(page, 1);
1655 page = NULL;
1656
1657 page = get_meta_page(sbi, start++);
1658 kaddr = (unsigned char *)page_address(page);
1659 offset = 0;
1660 }
1661 }
1662 f2fs_put_page(page, 1);
1663 return 0;
1664}
1665
1666static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1667{
1668 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1669 struct f2fs_summary_block *sum;
1670 struct curseg_info *curseg;
1671 struct page *new;
1672 unsigned short blk_off;
1673 unsigned int segno = 0;
1674 block_t blk_addr = 0;
1675
1676 /* get segment number and block addr */
1677 if (IS_DATASEG(type)) {
1678 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1679 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1680 CURSEG_HOT_DATA]);
119ee914 1681 if (__exist_node_summaries(sbi))
351df4b2
JK		 1682 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1683 else
1684 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1685 } else {
1686 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1687 CURSEG_HOT_NODE]);
1688 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1689 CURSEG_HOT_NODE]);
119ee914 1690 if (__exist_node_summaries(sbi))
351df4b2
JK		 1691 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1692 type - CURSEG_HOT_NODE);
1693 else
1694 blk_addr = GET_SUM_BLOCK(sbi, segno);
1695 }
1696
1697 new = get_meta_page(sbi, blk_addr);
1698 sum = (struct f2fs_summary_block *)page_address(new);
1699
1700 if (IS_NODESEG(type)) {
119ee914 1701 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1702 struct f2fs_summary *ns = &sum->entries[0];
1703 int i;
1704 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1705 ns->version = 0;
1706 ns->ofs_in_node = 0;
1707 }
1708 } else {
d653788a
GZ		 1709 int err;
1710
1711 err = restore_node_summary(sbi, segno, sum);
1712 if (err) {
351df4b2 1713 f2fs_put_page(new, 1);
d653788a 1714 return err;
351df4b2
JK		 1715 }
1716 }
1717 }
1718
1719 /* set uncompleted segment to curseg */
1720 curseg = CURSEG_I(sbi, type);
1721 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1722
1723 /* update journal info */
1724 down_write(&curseg->journal_rwsem);
1725 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1726 up_write(&curseg->journal_rwsem);
1727
1728 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1729 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1730 curseg->next_segno = segno;
1731 reset_curseg(sbi, type, 0);
1732 curseg->alloc_type = ckpt->alloc_type[type];
1733 curseg->next_blkoff = blk_off;
1734 mutex_unlock(&curseg->curseg_mutex);
1735 f2fs_put_page(new, 1);
1736 return 0;
1737}
1738
1739static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1740{
1741 int type = CURSEG_HOT_DATA;
e4fc5fbf 1742 int err;
351df4b2 1743
25ca923b 1744 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1745 int npages = npages_for_summary_flush(sbi, true);
1746
1747 if (npages >= 2)
1748 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1749 META_CP, true);
3fa06d7b 1750
351df4b2
JK		 1751 /* restore for compacted data summary */
1752 if (read_compacted_summaries(sbi))
1753 return -EINVAL;
1754 type = CURSEG_HOT_NODE;
1755 }
1756
119ee914 1757 if (__exist_node_summaries(sbi))
3fa06d7b 1758 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1759 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1760
e4fc5fbf
CY		 1761 for (; type <= CURSEG_COLD_NODE; type++) {
1762 err = read_normal_summaries(sbi, type);
1763 if (err)
1764 return err;
1765 }
1766
351df4b2
JK		 1767 return 0;
1768}
1769
1770static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1771{
1772 struct page *page;
1773 unsigned char *kaddr;
1774 struct f2fs_summary *summary;
1775 struct curseg_info *seg_i;
1776 int written_size = 0;
1777 int i, j;
1778
1779 page = grab_meta_page(sbi, blkaddr++);
1780 kaddr = (unsigned char *)page_address(page);
1781
1782 /* Step 1: write nat cache */
1783 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1784 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1785 written_size += SUM_JOURNAL_SIZE;
1786
1787 /* Step 2: write sit cache */
1788 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1789 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1790 written_size += SUM_JOURNAL_SIZE;
1791
351df4b2
JK		 1792 /* Step 3: write summary entries */
1793 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1794 unsigned short blkoff;
1795 seg_i = CURSEG_I(sbi, i);
1796 if (sbi->ckpt->alloc_type[i] == SSR)
1797 blkoff = sbi->blocks_per_seg;
1798 else
1799 blkoff = curseg_blkoff(sbi, i);
1800
1801 for (j = 0; j < blkoff; j++) {
1802 if (!page) {
1803 page = grab_meta_page(sbi, blkaddr++);
1804 kaddr = (unsigned char *)page_address(page);
1805 written_size = 0;
1806 }
1807 summary = (struct f2fs_summary *)(kaddr + written_size);
1808 *summary = seg_i->sum_blk->entries[j];
1809 written_size += SUMMARY_SIZE;
351df4b2 1810
09cbfeaf 1811 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1812 SUM_FOOTER_SIZE)
1813 continue;
1814
e8d61a74 1815 set_page_dirty(page);
351df4b2
JK		 1816 f2fs_put_page(page, 1);
1817 page = NULL;
1818 }
1819 }
e8d61a74
CY		 1820 if (page) {
1821 set_page_dirty(page);
351df4b2 1822 f2fs_put_page(page, 1);
e8d61a74 1823 }
351df4b2
JK		 1824}
1825
1826static void write_normal_summaries(struct f2fs_sb_info *sbi,
1827 block_t blkaddr, int type)
1828{
1829 int i, end;
1830 if (IS_DATASEG(type))
1831 end = type + NR_CURSEG_DATA_TYPE;
1832 else
1833 end = type + NR_CURSEG_NODE_TYPE;
1834
b7ad7512
CY		 1835 for (i = type; i < end; i++)
1836 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1837}
1838
1839void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1840{
25ca923b 1841 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1842 write_compacted_summaries(sbi, start_blk);
1843 else
1844 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1845}
1846
1847void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1848{
119ee914 1849 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1850}
1851
dfc08a12 1852int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1853 unsigned int val, int alloc)
1854{
1855 int i;
1856
1857 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1858 for (i = 0; i < nats_in_cursum(journal); i++) {
1859 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1860 return i;
1861 }
dfc08a12
CY		 1862 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1863 return update_nats_in_cursum(journal, 1);
351df4b2 1864 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1865 for (i = 0; i < sits_in_cursum(journal); i++)
1866 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1867 return i;
dfc08a12
CY		 1868 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1869 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1870 }
1871 return -1;
1872}
1873
1874static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1875 unsigned int segno)
1876{
2cc22186 1877 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1878}
1879
1880static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1881 unsigned int start)
1882{
1883 struct sit_info *sit_i = SIT_I(sbi);
1884 struct page *src_page, *dst_page;
1885 pgoff_t src_off, dst_off;
1886 void *src_addr, *dst_addr;
1887
1888 src_off = current_sit_addr(sbi, start);
1889 dst_off = next_sit_addr(sbi, src_off);
1890
1891 /* get current sit block page without lock */
1892 src_page = get_meta_page(sbi, src_off);
1893 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1894 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1895
1896 src_addr = page_address(src_page);
1897 dst_addr = page_address(dst_page);
09cbfeaf 1898 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1899
1900 set_page_dirty(dst_page);
1901 f2fs_put_page(src_page, 1);
1902
1903 set_to_next_sit(sit_i, start);
1904
1905 return dst_page;
1906}
1907
184a5cd2
CY		 1908static struct sit_entry_set *grab_sit_entry_set(void)
1909{
1910 struct sit_entry_set *ses =
80c54505 1911 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1912
1913 ses->entry_cnt = 0;
1914 INIT_LIST_HEAD(&ses->set_list);
1915 return ses;
1916}
1917
1918static void release_sit_entry_set(struct sit_entry_set *ses)
1919{
1920 list_del(&ses->set_list);
1921 kmem_cache_free(sit_entry_set_slab, ses);
1922}
1923
1924static void adjust_sit_entry_set(struct sit_entry_set *ses,
1925 struct list_head *head)
1926{
1927 struct sit_entry_set *next = ses;
1928
1929 if (list_is_last(&ses->set_list, head))
1930 return;
1931
1932 list_for_each_entry_continue(next, head, set_list)
1933 if (ses->entry_cnt <= next->entry_cnt)
1934 break;
1935
1936 list_move_tail(&ses->set_list, &next->set_list);
1937}
1938
1939static void add_sit_entry(unsigned int segno, struct list_head *head)
1940{
1941 struct sit_entry_set *ses;
1942 unsigned int start_segno = START_SEGNO(segno);
1943
1944 list_for_each_entry(ses, head, set_list) {
1945 if (ses->start_segno == start_segno) {
1946 ses->entry_cnt++;
1947 adjust_sit_entry_set(ses, head);
1948 return;
1949 }
1950 }
1951
1952 ses = grab_sit_entry_set();
1953
1954 ses->start_segno = start_segno;
1955 ses->entry_cnt++;
1956 list_add(&ses->set_list, head);
1957}
1958
1959static void add_sits_in_set(struct f2fs_sb_info *sbi)
1960{
1961 struct f2fs_sm_info *sm_info = SM_I(sbi);
1962 struct list_head *set_list = &sm_info->sit_entry_set;
1963 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1964 unsigned int segno;
1965
7cd8558b 1966 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1967 add_sit_entry(segno, set_list);
1968}
1969
1970static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1971{
1972 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1973 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1974 int i;
1975
b7ad7512 1976 down_write(&curseg->journal_rwsem);
dfc08a12 1977 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1978 unsigned int segno;
1979 bool dirtied;
1980
dfc08a12 1981 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1982 dirtied = __mark_sit_entry_dirty(sbi, segno);
1983
1984 if (!dirtied)
1985 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1986 }
dfc08a12 1987 update_sits_in_cursum(journal, -i);
b7ad7512 1988 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1989}
1990
0a8165d7 1991/*
351df4b2
JK		 1992 * CP calls this function, which flushes SIT entries including sit_journal,
1993 * and moves prefree segs to free segs.
1994 */
4b2fecc8 1995void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1996{
1997 struct sit_info *sit_i = SIT_I(sbi);
1998 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1999 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2000 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2001 struct sit_entry_set *ses, *tmp;
2002 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2003 bool to_journal = true;
4b2fecc8 2004 struct seg_entry *se;
351df4b2 2005
351df4b2
JK		 2006 mutex_lock(&sit_i->sentry_lock);
2007
2b11a74b
WL		 2008 if (!sit_i->dirty_sentries)
2009 goto out;
2010
351df4b2 2011 /*
184a5cd2
CY		 2012 * add and account sit entries of dirty bitmap in sit entry
2013 * set temporarily
351df4b2 2014 */
184a5cd2 2015 add_sits_in_set(sbi);
351df4b2 2016
184a5cd2
CY		 2017 /*
2018 * if there are no enough space in journal to store dirty sit
2019 * entries, remove all entries from journal and add and account
2020 * them in sit entry set.
2021 */
dfc08a12 2022 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2023 remove_sits_in_journal(sbi);
b2955550 2024
184a5cd2
CY		 2025 /*
2026 * there are two steps to flush sit entries:
2027 * #1, flush sit entries to journal in current cold data summary block.
2028 * #2, flush sit entries to sit page.
2029 */
2030 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2031 struct page *page = NULL;
184a5cd2
CY		 2032 struct f2fs_sit_block *raw_sit = NULL;
2033 unsigned int start_segno = ses->start_segno;
2034 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2035 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2036 unsigned int segno = start_segno;
2037
2038 if (to_journal &&
dfc08a12 2039 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2040 to_journal = false;
2041
b7ad7512
CY		 2042 if (to_journal) {
2043 down_write(&curseg->journal_rwsem);
2044 } else {
184a5cd2
CY		 2045 page = get_next_sit_page(sbi, start_segno);
2046 raw_sit = page_address(page);
351df4b2 2047 }
351df4b2 2048
184a5cd2
CY		 2049 /* flush dirty sit entries in region of current sit set */
2050 for_each_set_bit_from(segno, bitmap, end) {
2051 int offset, sit_offset;
4b2fecc8
JK		 2052
2053 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2054
2055 /* add discard candidates */
d7bc2484 2056 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2057 cpc->trim_start = segno;
2058 add_discard_addrs(sbi, cpc);
2059 }
184a5cd2
CY		 2060
2061 if (to_journal) {
dfc08a12 2062 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2063 SIT_JOURNAL, segno, 1);
2064 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2065 segno_in_journal(journal, offset) =
184a5cd2
CY		 2066 cpu_to_le32(segno);
2067 seg_info_to_raw_sit(se,
dfc08a12 2068 &sit_in_journal(journal, offset));
184a5cd2
CY		 2069 } else {
2070 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2071 seg_info_to_raw_sit(se,
2072 &raw_sit->entries[sit_offset]);
2073 }
351df4b2 2074
184a5cd2
CY		 2075 __clear_bit(segno, bitmap);
2076 sit_i->dirty_sentries--;
2077 ses->entry_cnt--;
351df4b2
JK		 2078 }
2079
b7ad7512
CY		 2080 if (to_journal)
2081 up_write(&curseg->journal_rwsem);
2082 else
184a5cd2
CY		 2083 f2fs_put_page(page, 1);
2084
2085 f2fs_bug_on(sbi, ses->entry_cnt);
2086 release_sit_entry_set(ses);
351df4b2 2087 }
184a5cd2
CY		 2088
2089 f2fs_bug_on(sbi, !list_empty(head));
2090 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2091out:
4b2fecc8
JK		 2092 if (cpc->reason == CP_DISCARD) {
2093 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2094 add_discard_addrs(sbi, cpc);
2095 }
351df4b2 2096 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2097
351df4b2
JK		 2098 set_prefree_as_free_segments(sbi);
2099}
2100
2101static int build_sit_info(struct f2fs_sb_info *sbi)
2102{
2103 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2104 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2105 struct sit_info *sit_i;
2106 unsigned int sit_segs, start;
2107 char *src_bitmap, *dst_bitmap;
2108 unsigned int bitmap_size;
2109
2110 /* allocate memory for SIT information */
2111 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2112 if (!sit_i)
2113 return -ENOMEM;
2114
2115 SM_I(sbi)->sit_info = sit_i;
2116
39307a8e
JK		 2117 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2118 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2119 if (!sit_i->sentries)
2120 return -ENOMEM;
2121
7cd8558b 2122 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2123 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2124 if (!sit_i->dirty_sentries_bitmap)
2125 return -ENOMEM;
2126
7cd8558b 2127 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2128 sit_i->sentries[start].cur_valid_map
2129 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2130 sit_i->sentries[start].ckpt_valid_map
2131 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 2132 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 2133 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 2134 return -ENOMEM;
3e025740
JK		 2135
2136 if (f2fs_discard_en(sbi)) {
2137 sit_i->sentries[start].discard_map
2138 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2139 if (!sit_i->sentries[start].discard_map)
2140 return -ENOMEM;
2141 }
351df4b2
JK		 2142 }
2143
60a3b782
JK		 2144 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2145 if (!sit_i->tmp_map)
2146 return -ENOMEM;
2147
351df4b2 2148 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2149 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2150 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2151 if (!sit_i->sec_entries)
2152 return -ENOMEM;
2153 }
2154
2155 /* get information related with SIT */
2156 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2157
2158 /* setup SIT bitmap from ckeckpoint pack */
2159 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2160 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2161
79b5793b 2162 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2163 if (!dst_bitmap)
2164 return -ENOMEM;
351df4b2
JK		 2165
2166 /* init SIT information */
2167 sit_i->s_ops = &default_salloc_ops;
2168
2169 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2170 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2171 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2172 sit_i->sit_bitmap = dst_bitmap;
2173 sit_i->bitmap_size = bitmap_size;
2174 sit_i->dirty_sentries = 0;
2175 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2176 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2177 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2178 mutex_init(&sit_i->sentry_lock);
2179 return 0;
2180}
2181
2182static int build_free_segmap(struct f2fs_sb_info *sbi)
2183{
351df4b2
JK		 2184 struct free_segmap_info *free_i;
2185 unsigned int bitmap_size, sec_bitmap_size;
2186
2187 /* allocate memory for free segmap information */
2188 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2189 if (!free_i)
2190 return -ENOMEM;
2191
2192 SM_I(sbi)->free_info = free_i;
2193
7cd8558b 2194 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2195 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2196 if (!free_i->free_segmap)
2197 return -ENOMEM;
2198
7cd8558b 2199 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2200 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2201 if (!free_i->free_secmap)
2202 return -ENOMEM;
2203
2204 /* set all segments as dirty temporarily */
2205 memset(free_i->free_segmap, 0xff, bitmap_size);
2206 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2207
2208 /* init free segmap information */
7cd8558b 2209 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2210 free_i->free_segments = 0;
2211 free_i->free_sections = 0;
1a118ccf 2212 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2213 return 0;
2214}
2215
2216static int build_curseg(struct f2fs_sb_info *sbi)
2217{
1042d60f 2218 struct curseg_info *array;
351df4b2
JK		 2219 int i;
2220
b434babf 2221 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2222 if (!array)
2223 return -ENOMEM;
2224
2225 SM_I(sbi)->curseg_array = array;
2226
2227 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2228 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2229 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2230 if (!array[i].sum_blk)
2231 return -ENOMEM;
b7ad7512
CY		 2232 init_rwsem(&array[i].journal_rwsem);
2233 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2234 GFP_KERNEL);
2235 if (!array[i].journal)
2236 return -ENOMEM;
351df4b2
JK		 2237 array[i].segno = NULL_SEGNO;
2238 array[i].next_blkoff = 0;
2239 }
2240 return restore_curseg_summaries(sbi);
2241}
2242
2243static void build_sit_entries(struct f2fs_sb_info *sbi)
2244{
2245 struct sit_info *sit_i = SIT_I(sbi);
2246 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2247 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2248 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2249 unsigned int i, start, end;
2250 unsigned int readed, start_blk = 0;
e9f5b8b8 2251 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2252
74de593a 2253 do {
26879fb1 2254 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2255
2256 start = start_blk * sit_i->sents_per_block;
2257 end = (start_blk + readed) * sit_i->sents_per_block;
2258
7cd8558b 2259 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2260 struct seg_entry *se = &sit_i->sentries[start];
2261 struct f2fs_sit_block *sit_blk;
2262 struct f2fs_sit_entry sit;
2263 struct page *page;
2264
b7ad7512 2265 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2266 for (i = 0; i < sits_in_cursum(journal); i++) {
2267 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2268 == start) {
dfc08a12 2269 sit = sit_in_journal(journal, i);
b7ad7512 2270 up_read(&curseg->journal_rwsem);
74de593a
CY		 2271 goto got_it;
2272 }
351df4b2 2273 }
b7ad7512 2274 up_read(&curseg->journal_rwsem);
74de593a
CY		 2275
2276 page = get_current_sit_page(sbi, start);
2277 sit_blk = (struct f2fs_sit_block *)page_address(page);
2278 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2279 f2fs_put_page(page, 1);
351df4b2 2280got_it:
74de593a
CY		 2281 check_block_count(sbi, start, &sit);
2282 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2283
2284 /* build discard map only one time */
3e025740
JK		 2285 if (f2fs_discard_en(sbi)) {
2286 memcpy(se->discard_map, se->cur_valid_map,
2287 SIT_VBLOCK_MAP_SIZE);
2288 sbi->discard_blks += sbi->blocks_per_seg -
2289 se->valid_blocks;
2290 }
a66cdd98 2291
74de593a
CY		 2292 if (sbi->segs_per_sec > 1) {
2293 struct sec_entry *e = get_sec_entry(sbi, start);
2294 e->valid_blocks += se->valid_blocks;
2295 }
351df4b2 2296 }
74de593a
CY		 2297 start_blk += readed;
2298 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2299}
2300
2301static void init_free_segmap(struct f2fs_sb_info *sbi)
2302{
2303 unsigned int start;
2304 int type;
2305
7cd8558b 2306 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2307 struct seg_entry *sentry = get_seg_entry(sbi, start);
2308 if (!sentry->valid_blocks)
2309 __set_free(sbi, start);
2310 }
2311
2312 /* set use the current segments */
2313 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2314 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2315 __set_test_and_inuse(sbi, curseg_t->segno);
2316 }
2317}
2318
2319static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2320{
2321 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2322 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2323 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2324 unsigned short valid_blocks;
2325
8736fbf0 2326 while (1) {
351df4b2 2327 /* find dirty segment based on free segmap */
7cd8558b
JK		 2328 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2329 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2330 break;
2331 offset = segno + 1;
2332 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2333 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2334 continue;
ec325b52
JK		 2335 if (valid_blocks > sbi->blocks_per_seg) {
2336 f2fs_bug_on(sbi, 1);
2337 continue;
2338 }
351df4b2
JK		 2339 mutex_lock(&dirty_i->seglist_lock);
2340 __locate_dirty_segment(sbi, segno, DIRTY);
2341 mutex_unlock(&dirty_i->seglist_lock);
2342 }
2343}
2344
5ec4e49f 2345static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2346{
2347 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2348 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2349
39307a8e 2350 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2351 if (!dirty_i->victim_secmap)
351df4b2
JK		 2352 return -ENOMEM;
2353 return 0;
2354}
2355
2356static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2357{
2358 struct dirty_seglist_info *dirty_i;
2359 unsigned int bitmap_size, i;
2360
2361 /* allocate memory for dirty segments list information */
2362 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2363 if (!dirty_i)
2364 return -ENOMEM;
2365
2366 SM_I(sbi)->dirty_info = dirty_i;
2367 mutex_init(&dirty_i->seglist_lock);
2368
7cd8558b 2369 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2370
2371 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2372 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2373 if (!dirty_i->dirty_segmap[i])
2374 return -ENOMEM;
2375 }
2376
2377 init_dirty_segmap(sbi);
5ec4e49f 2378 return init_victim_secmap(sbi);
351df4b2
JK		 2379}
2380
0a8165d7 2381/*
351df4b2
JK		 2382 * Update min, max modified time for cost-benefit GC algorithm
2383 */
2384static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2385{
2386 struct sit_info *sit_i = SIT_I(sbi);
2387 unsigned int segno;
2388
2389 mutex_lock(&sit_i->sentry_lock);
2390
2391 sit_i->min_mtime = LLONG_MAX;
2392
7cd8558b 2393 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2394 unsigned int i;
2395 unsigned long long mtime = 0;
2396
2397 for (i = 0; i < sbi->segs_per_sec; i++)
2398 mtime += get_seg_entry(sbi, segno + i)->mtime;
2399
2400 mtime = div_u64(mtime, sbi->segs_per_sec);
2401
2402 if (sit_i->min_mtime > mtime)
2403 sit_i->min_mtime = mtime;
2404 }
2405 sit_i->max_mtime = get_mtime(sbi);
2406 mutex_unlock(&sit_i->sentry_lock);
2407}
2408
2409int build_segment_manager(struct f2fs_sb_info *sbi)
2410{
2411 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2412 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2413 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2414 int err;
2415
2416 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2417 if (!sm_info)
2418 return -ENOMEM;
2419
2420 /* init sm info */
2421 sbi->sm_info = sm_info;
351df4b2
JK		 2422 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2423 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2424 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2425 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2426 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2427 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2428 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2429 sm_info->rec_prefree_segments = sm_info->main_segments *
2430 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2431 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2432 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2433
52763a4b
JK		 2434 if (!test_opt(sbi, LFS))
2435 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2436 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2437 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2438
7fd9e544
JK		 2439 INIT_LIST_HEAD(&sm_info->discard_list);
2440 sm_info->nr_discards = 0;
2441 sm_info->max_discards = 0;
2442
bba681cb
JK		 2443 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2444
184a5cd2
CY		 2445 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2446
b270ad6f 2447 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2448 err = create_flush_cmd_control(sbi);
2449 if (err)
a688b9d9 2450 return err;
6b4afdd7
JK		 2451 }
2452
351df4b2
JK		 2453 err = build_sit_info(sbi);
2454 if (err)
2455 return err;
2456 err = build_free_segmap(sbi);
2457 if (err)
2458 return err;
2459 err = build_curseg(sbi);
2460 if (err)
2461 return err;
2462
2463 /* reinit free segmap based on SIT */
2464 build_sit_entries(sbi);
2465
2466 init_free_segmap(sbi);
2467 err = build_dirty_segmap(sbi);
2468 if (err)
2469 return err;
2470
2471 init_min_max_mtime(sbi);
2472 return 0;
2473}
2474
2475static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2476 enum dirty_type dirty_type)
2477{
2478 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2479
2480 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2481 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2482 dirty_i->nr_dirty[dirty_type] = 0;
2483 mutex_unlock(&dirty_i->seglist_lock);
2484}
2485
5ec4e49f 2486static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2487{
2488 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2489 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2490}
2491
2492static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2493{
2494 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2495 int i;
2496
2497 if (!dirty_i)
2498 return;
2499
2500 /* discard pre-free/dirty segments list */
2501 for (i = 0; i < NR_DIRTY_TYPE; i++)
2502 discard_dirty_segmap(sbi, i);
2503
5ec4e49f 2504 destroy_victim_secmap(sbi);
351df4b2
JK		 2505 SM_I(sbi)->dirty_info = NULL;
2506 kfree(dirty_i);
2507}
2508
2509static void destroy_curseg(struct f2fs_sb_info *sbi)
2510{
2511 struct curseg_info *array = SM_I(sbi)->curseg_array;
2512 int i;
2513
2514 if (!array)
2515 return;
2516 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2517 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2518 kfree(array[i].sum_blk);
b7ad7512
CY		 2519 kfree(array[i].journal);
2520 }
351df4b2
JK		 2521 kfree(array);
2522}
2523
2524static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2525{
2526 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2527 if (!free_i)
2528 return;
2529 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2530 kvfree(free_i->free_segmap);
2531 kvfree(free_i->free_secmap);
351df4b2
JK		 2532 kfree(free_i);
2533}
2534
2535static void destroy_sit_info(struct f2fs_sb_info *sbi)
2536{
2537 struct sit_info *sit_i = SIT_I(sbi);
2538 unsigned int start;
2539
2540 if (!sit_i)
2541 return;
2542
2543 if (sit_i->sentries) {
7cd8558b 2544 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2545 kfree(sit_i->sentries[start].cur_valid_map);
2546 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2547 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2548 }
2549 }
60a3b782
JK		 2550 kfree(sit_i->tmp_map);
2551
39307a8e
JK		 2552 kvfree(sit_i->sentries);
2553 kvfree(sit_i->sec_entries);
2554 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2555
2556 SM_I(sbi)->sit_info = NULL;
2557 kfree(sit_i->sit_bitmap);
2558 kfree(sit_i);
2559}
2560
2561void destroy_segment_manager(struct f2fs_sb_info *sbi)
2562{
2563 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2564
3b03f724
CY		 2565 if (!sm_info)
2566 return;
2163d198 2567 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2568 destroy_dirty_segmap(sbi);
2569 destroy_curseg(sbi);
2570 destroy_free_segmap(sbi);
2571 destroy_sit_info(sbi);
2572 sbi->sm_info = NULL;
2573 kfree(sm_info);
2574}
7fd9e544
JK		 2575
2576int __init create_segment_manager_caches(void)
2577{
2578 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2579 sizeof(struct discard_entry));
7fd9e544 2580 if (!discard_entry_slab)
184a5cd2
CY		 2581 goto fail;
2582
2583 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2584 sizeof(struct sit_entry_set));
184a5cd2
CY		 2585 if (!sit_entry_set_slab)
2586 goto destory_discard_entry;
88b88a66
JK		 2587
2588 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2589 sizeof(struct inmem_pages));
2590 if (!inmem_entry_slab)
2591 goto destroy_sit_entry_set;
7fd9e544 2592 return 0;
184a5cd2 2593
88b88a66
JK		 2594destroy_sit_entry_set:
2595 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2596destory_discard_entry:
2597 kmem_cache_destroy(discard_entry_slab);
2598fail:
2599 return -ENOMEM;
7fd9e544
JK		 2600}
2601
2602void destroy_segment_manager_caches(void)
2603{
184a5cd2 2604 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2605 kmem_cache_destroy(discard_entry_slab);
88b88a66 2606 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2607}
Linux kernel - Deliverables
This page took 0.28255 seconds and 5 git commands to generate.