projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'regulator-v4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
[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
26dc3d44
JK		 244 clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
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;
351df4b2 349 /*
029cd28c
JK		 350 * We should do GC or end up with checkpoint, if there are so many dirty
351 * dir/node pages without enough free segments.
351df4b2 352 */
43727527 353 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 354 mutex_lock(&sbi->gc_mutex);
d530d4d8 355 f2fs_gc(sbi, false);
351df4b2
JK		 356 }
357}
358
4660f9c0
JK		 359void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
360{
1dcc336b 361 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 362 if (!available_free_memory(sbi, EXTENT_CACHE))
363 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 364
1b38dc8e
JK		 365 /* check the # of cached NAT entries */
366 if (!available_free_memory(sbi, NAT_ENTRIES))
367 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
368
31696580
CY		 369 if (!available_free_memory(sbi, FREE_NIDS))
370 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
371
1b38dc8e
JK		 372 /* checkpoint is the only way to shrink partial cached entries */
373 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 374 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 375 excess_prefree_segs(sbi) ||
376 excess_dirty_nats(sbi) ||
d0239e1b 377 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
e9f5b8b8
CY		 378 if (test_opt(sbi, DATA_FLUSH)) {
379 struct blk_plug plug;
380
381 blk_start_plug(&plug);
36b35a0d 382 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 383 blk_finish_plug(&plug);
384 }
4660f9c0 385 f2fs_sync_fs(sbi->sb, true);
42190d2a 386 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 387 }
4660f9c0
JK		 388}
389
2163d198 390static int issue_flush_thread(void *data)
6b4afdd7
JK		 391{
392 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 393 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
394 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 395repeat:
396 if (kthread_should_stop())
397 return 0;
398
721bd4d5 399 if (!llist_empty(&fcc->issue_list)) {
740432f8 400 struct bio *bio;
6b4afdd7
JK		 401 struct flush_cmd *cmd, *next;
402 int ret;
403
740432f8
JK		 404 bio = f2fs_bio_alloc(0);
405
721bd4d5
GZ		 406 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
407 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
408
6b4afdd7 409 bio->bi_bdev = sbi->sb->s_bdev;
04d328de 410 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
4e49ea4a 411 ret = submit_bio_wait(bio);
6b4afdd7 412
721bd4d5
GZ		 413 llist_for_each_entry_safe(cmd, next,
414 fcc->dispatch_list, llnode) {
6b4afdd7 415 cmd->ret = ret;
6b4afdd7
JK		 416 complete(&cmd->wait);
417 }
a4ed23f2 418 bio_put(bio);
a688b9d9 419 fcc->dispatch_list = NULL;
6b4afdd7
JK		 420 }
421
a688b9d9 422 wait_event_interruptible(*q,
721bd4d5 423 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 424 goto repeat;
425}
426
427int f2fs_issue_flush(struct f2fs_sb_info *sbi)
428{
a688b9d9 429 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 430 struct flush_cmd cmd;
6b4afdd7 431
24a9ee0f
JK		 432 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
433 test_opt(sbi, FLUSH_MERGE));
434
0f7b2abd
JK		 435 if (test_opt(sbi, NOBARRIER))
436 return 0;
437
740432f8
JK		 438 if (!test_opt(sbi, FLUSH_MERGE)) {
439 struct bio *bio = f2fs_bio_alloc(0);
440 int ret;
441
442 bio->bi_bdev = sbi->sb->s_bdev;
04d328de 443 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
4e49ea4a 444 ret = submit_bio_wait(bio);
740432f8
JK		 445 bio_put(bio);
446 return ret;
447 }
6b4afdd7 448
adf8d90b 449 init_completion(&cmd.wait);
6b4afdd7 450
721bd4d5 451 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 452
a688b9d9
GZ		 453 if (!fcc->dispatch_list)
454 wake_up(&fcc->flush_wait_queue);
6b4afdd7 455
adf8d90b
CY		 456 wait_for_completion(&cmd.wait);
457
458 return cmd.ret;
6b4afdd7
JK		 459}
460
2163d198
GZ		 461int create_flush_cmd_control(struct f2fs_sb_info *sbi)
462{
463 dev_t dev = sbi->sb->s_bdev->bd_dev;
464 struct flush_cmd_control *fcc;
465 int err = 0;
466
467 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
468 if (!fcc)
469 return -ENOMEM;
2163d198 470 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 471 init_llist_head(&fcc->issue_list);
6b2920a5 472 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 473 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
474 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
475 if (IS_ERR(fcc->f2fs_issue_flush)) {
476 err = PTR_ERR(fcc->f2fs_issue_flush);
477 kfree(fcc);
6b2920a5 478 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 479 return err;
480 }
2163d198
GZ		 481
482 return err;
483}
484
485void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
486{
6b2920a5 487 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 488
489 if (fcc && fcc->f2fs_issue_flush)
490 kthread_stop(fcc->f2fs_issue_flush);
491 kfree(fcc);
6b2920a5 492 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 493}
494
351df4b2
JK		 495static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
496 enum dirty_type dirty_type)
497{
498 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
499
500 /* need not be added */
501 if (IS_CURSEG(sbi, segno))
502 return;
503
504 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
505 dirty_i->nr_dirty[dirty_type]++;
506
507 if (dirty_type == DIRTY) {
508 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 509 enum dirty_type t = sentry->type;
b2f2c390 510
ec325b52
JK		 511 if (unlikely(t >= DIRTY)) {
512 f2fs_bug_on(sbi, 1);
513 return;
514 }
4625d6aa
CL		 515 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
516 dirty_i->nr_dirty[t]++;
351df4b2
JK		 517 }
518}
519
520static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
521 enum dirty_type dirty_type)
522{
523 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
524
525 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
526 dirty_i->nr_dirty[dirty_type]--;
527
528 if (dirty_type == DIRTY) {
4625d6aa
CL		 529 struct seg_entry *sentry = get_seg_entry(sbi, segno);
530 enum dirty_type t = sentry->type;
531
532 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
533 dirty_i->nr_dirty[t]--;
b2f2c390 534
5ec4e49f
JK		 535 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
536 clear_bit(GET_SECNO(sbi, segno),
537 dirty_i->victim_secmap);
351df4b2
JK		 538 }
539}
540
0a8165d7 541/*
351df4b2
JK		 542 * Should not occur error such as -ENOMEM.
543 * Adding dirty entry into seglist is not critical operation.
544 * If a given segment is one of current working segments, it won't be added.
545 */
8d8451af 546static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 547{
548 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
549 unsigned short valid_blocks;
550
551 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
552 return;
553
554 mutex_lock(&dirty_i->seglist_lock);
555
556 valid_blocks = get_valid_blocks(sbi, segno, 0);
557
558 if (valid_blocks == 0) {
559 __locate_dirty_segment(sbi, segno, PRE);
560 __remove_dirty_segment(sbi, segno, DIRTY);
561 } else if (valid_blocks < sbi->blocks_per_seg) {
562 __locate_dirty_segment(sbi, segno, DIRTY);
563 } else {
564 /* Recovery routine with SSR needs this */
565 __remove_dirty_segment(sbi, segno, DIRTY);
566 }
567
568 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 569}
570
1e87a78d 571static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 572 block_t blkstart, block_t blklen)
573{
55cf9cb6
CY		 574 sector_t start = SECTOR_FROM_BLOCK(blkstart);
575 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 576 struct seg_entry *se;
577 unsigned int offset;
578 block_t i;
579
580 for (i = blkstart; i < blkstart + blklen; i++) {
581 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
582 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
583
584 if (!f2fs_test_and_set_bit(offset, se->discard_map))
585 sbi->discard_blks--;
586 }
1661d07c 587 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
1e87a78d
JK		 588 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
589}
590
e90c2d28 591bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 592{
60b286c4 593 int err = -EOPNOTSUPP;
40a02be1
JK		 594
595 if (test_opt(sbi, DISCARD)) {
596 struct seg_entry *se = get_seg_entry(sbi,
597 GET_SEGNO(sbi, blkaddr));
598 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
599
600 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 601 return false;
40a02be1
JK		 602
603 err = f2fs_issue_discard(sbi, blkaddr, 1);
604 }
605
e90c2d28 606 if (err) {
381722d2 607 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 608 return true;
609 }
610 return false;
37208879
JK		 611}
612
adf4983b 613static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 614 struct cp_control *cpc, struct seg_entry *se,
615 unsigned int start, unsigned int end)
b2955550
JK		 616{
617 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 618 struct discard_entry *new, *last;
619
620 if (!list_empty(head)) {
621 last = list_last_entry(head, struct discard_entry, list);
622 if (START_BLOCK(sbi, cpc->trim_start) + start ==
623 last->blkaddr + last->len) {
624 last->len += end - start;
625 goto done;
626 }
627 }
628
629 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
630 INIT_LIST_HEAD(&new->list);
631 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
632 new->len = end - start;
633 list_add_tail(&new->list, head);
634done:
635 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 636}
637
638static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
639{
b2955550
JK		 640 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
641 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 642 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 643 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
644 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 645 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 646 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 647 unsigned int start = 0, end = -1;
4b2fecc8 648 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 649 int i;
650
a66cdd98 651 if (se->valid_blocks == max_blocks)
b2955550
JK		 652 return;
653
a66cdd98
JK		 654 if (!force) {
655 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 656 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
657 return;
4b2fecc8
JK		 658 }
659
b2955550
JK		 660 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
661 for (i = 0; i < entries; i++)
a66cdd98 662 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 663 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 664
4b2fecc8 665 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 666 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
667 if (start >= max_blocks)
668 break;
669
670 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
a66cdd98 671 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 672 }
673}
674
4b2fecc8
JK		 675void release_discard_addrs(struct f2fs_sb_info *sbi)
676{
677 struct list_head *head = &(SM_I(sbi)->discard_list);
678 struct discard_entry *entry, *this;
679
680 /* drop caches */
681 list_for_each_entry_safe(entry, this, head, list) {
682 list_del(&entry->list);
683 kmem_cache_free(discard_entry_slab, entry);
684 }
685}
686
0a8165d7 687/*
351df4b2
JK		 688 * Should call clear_prefree_segments after checkpoint is done.
689 */
690static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
691{
692 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 693 unsigned int segno;
351df4b2
JK		 694
695 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 696 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 697 __set_test_and_free(sbi, segno);
351df4b2
JK		 698 mutex_unlock(&dirty_i->seglist_lock);
699}
700
836b5a63 701void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 702{
b2955550 703 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 704 struct discard_entry *entry, *this;
351df4b2 705 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 706 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 707 unsigned int start = 0, end = -1;
351df4b2
JK		 708
709 mutex_lock(&dirty_i->seglist_lock);
29e59c14 710
351df4b2 711 while (1) {
29e59c14 712 int i;
7cd8558b
JK		 713 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
714 if (start >= MAIN_SEGS(sbi))
351df4b2 715 break;
7cd8558b
JK		 716 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
717 start + 1);
29e59c14
CL		 718
719 for (i = start; i < end; i++)
720 clear_bit(i, prefree_map);
721
722 dirty_i->nr_dirty[PRE] -= end - start;
723
724 if (!test_opt(sbi, DISCARD))
725 continue;
351df4b2 726
37208879
JK		 727 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
728 (end - start) << sbi->log_blocks_per_seg);
351df4b2
JK		 729 }
730 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 731
732 /* send small discards */
2d7b822a 733 list_for_each_entry_safe(entry, this, head, list) {
836b5a63
JK		 734 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
735 goto skip;
37208879 736 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 737 cpc->trimmed += entry->len;
836b5a63 738skip:
b2955550
JK		 739 list_del(&entry->list);
740 SM_I(sbi)->nr_discards -= entry->len;
741 kmem_cache_free(discard_entry_slab, entry);
742 }
351df4b2
JK		 743}
744
184a5cd2 745static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 746{
747 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 748
749 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 750 sit_i->dirty_sentries++;
184a5cd2
CY		 751 return false;
752 }
753
754 return true;
351df4b2
JK		 755}
756
757static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
758 unsigned int segno, int modified)
759{
760 struct seg_entry *se = get_seg_entry(sbi, segno);
761 se->type = type;
762 if (modified)
763 __mark_sit_entry_dirty(sbi, segno);
764}
765
766static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
767{
768 struct seg_entry *se;
769 unsigned int segno, offset;
770 long int new_vblocks;
771
772 segno = GET_SEGNO(sbi, blkaddr);
773
774 se = get_seg_entry(sbi, segno);
775 new_vblocks = se->valid_blocks + del;
491c0854 776 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 777
9850cf4a 778 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 779 (new_vblocks > sbi->blocks_per_seg)));
780
781 se->valid_blocks = new_vblocks;
782 se->mtime = get_mtime(sbi);
783 SIT_I(sbi)->max_mtime = se->mtime;
784
785 /* Update valid block bitmap */
786 if (del > 0) {
52aca074 787 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 788 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 789 if (!f2fs_test_and_set_bit(offset, se->discard_map))
790 sbi->discard_blks--;
351df4b2 791 } else {
52aca074 792 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 793 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 794 if (f2fs_test_and_clear_bit(offset, se->discard_map))
795 sbi->discard_blks++;
351df4b2
JK		 796 }
797 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
798 se->ckpt_valid_blocks += del;
799
800 __mark_sit_entry_dirty(sbi, segno);
801
802 /* update total number of valid blocks to be written in ckpt area */
803 SIT_I(sbi)->written_valid_blocks += del;
804
805 if (sbi->segs_per_sec > 1)
806 get_sec_entry(sbi, segno)->valid_blocks += del;
807}
808
5e443818 809void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 810{
5e443818
JK		 811 update_sit_entry(sbi, new, 1);
812 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
813 update_sit_entry(sbi, old, -1);
814
815 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
816 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 817}
818
819void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
820{
821 unsigned int segno = GET_SEGNO(sbi, addr);
822 struct sit_info *sit_i = SIT_I(sbi);
823
9850cf4a 824 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 825 if (addr == NEW_ADDR)
826 return;
827
828 /* add it into sit main buffer */
829 mutex_lock(&sit_i->sentry_lock);
830
831 update_sit_entry(sbi, addr, -1);
832
833 /* add it into dirty seglist */
834 locate_dirty_segment(sbi, segno);
835
836 mutex_unlock(&sit_i->sentry_lock);
837}
838
6e2c64ad
JK		 839bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
840{
841 struct sit_info *sit_i = SIT_I(sbi);
842 unsigned int segno, offset;
843 struct seg_entry *se;
844 bool is_cp = false;
845
846 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
847 return true;
848
849 mutex_lock(&sit_i->sentry_lock);
850
851 segno = GET_SEGNO(sbi, blkaddr);
852 se = get_seg_entry(sbi, segno);
853 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
854
855 if (f2fs_test_bit(offset, se->ckpt_valid_map))
856 is_cp = true;
857
858 mutex_unlock(&sit_i->sentry_lock);
859
860 return is_cp;
861}
862
0a8165d7 863/*
351df4b2
JK		 864 * This function should be resided under the curseg_mutex lock
865 */
866static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 867 struct f2fs_summary *sum)
351df4b2
JK		 868{
869 struct curseg_info *curseg = CURSEG_I(sbi, type);
870 void *addr = curseg->sum_blk;
e79efe3b 871 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 872 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 873}
874
0a8165d7 875/*
351df4b2
JK		 876 * Calculate the number of current summary pages for writing
877 */
3fa06d7b 878int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 879{
351df4b2 880 int valid_sum_count = 0;
9a47938b 881 int i, sum_in_page;
351df4b2
JK		 882
883 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
884 if (sbi->ckpt->alloc_type[i] == SSR)
885 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 886 else {
887 if (for_ra)
888 valid_sum_count += le16_to_cpu(
889 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
890 else
891 valid_sum_count += curseg_blkoff(sbi, i);
892 }
351df4b2
JK		 893 }
894
09cbfeaf 895 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 896 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
897 if (valid_sum_count <= sum_in_page)
351df4b2 898 return 1;
9a47938b 899 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 900 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 901 return 2;
902 return 3;
903}
904
0a8165d7 905/*
351df4b2
JK		 906 * Caller should put this summary page
907 */
908struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
909{
910 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
911}
912
381722d2 913void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 914{
915 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 916 void *dst = page_address(page);
917
918 if (src)
09cbfeaf 919 memcpy(dst, src, PAGE_SIZE);
381722d2 920 else
09cbfeaf 921 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 922 set_page_dirty(page);
923 f2fs_put_page(page, 1);
924}
925
381722d2
CY		 926static void write_sum_page(struct f2fs_sb_info *sbi,
927 struct f2fs_summary_block *sum_blk, block_t blk_addr)
928{
929 update_meta_page(sbi, (void *)sum_blk, blk_addr);
930}
931
b7ad7512
CY		 932static void write_current_sum_page(struct f2fs_sb_info *sbi,
933 int type, block_t blk_addr)
934{
935 struct curseg_info *curseg = CURSEG_I(sbi, type);
936 struct page *page = grab_meta_page(sbi, blk_addr);
937 struct f2fs_summary_block *src = curseg->sum_blk;
938 struct f2fs_summary_block *dst;
939
940 dst = (struct f2fs_summary_block *)page_address(page);
941
942 mutex_lock(&curseg->curseg_mutex);
943
944 down_read(&curseg->journal_rwsem);
945 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
946 up_read(&curseg->journal_rwsem);
947
948 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
949 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
950
951 mutex_unlock(&curseg->curseg_mutex);
952
953 set_page_dirty(page);
954 f2fs_put_page(page, 1);
955}
956
60374688
JK		 957static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
958{
959 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 960 unsigned int segno = curseg->segno + 1;
60374688
JK		 961 struct free_segmap_info *free_i = FREE_I(sbi);
962
7cd8558b 963 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 964 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 965 return 0;
966}
967
0a8165d7 968/*
351df4b2
JK		 969 * Find a new segment from the free segments bitmap to right order
970 * This function should be returned with success, otherwise BUG
971 */
972static void get_new_segment(struct f2fs_sb_info *sbi,
973 unsigned int *newseg, bool new_sec, int dir)
974{
975 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 976 unsigned int segno, secno, zoneno;
7cd8558b 977 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 978 unsigned int hint = *newseg / sbi->segs_per_sec;
979 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
980 unsigned int left_start = hint;
981 bool init = true;
982 int go_left = 0;
983 int i;
984
1a118ccf 985 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 986
987 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
988 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 989 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
990 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 991 goto got_it;
992 }
993find_other_zone:
7cd8558b
JK		 994 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
995 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 996 if (dir == ALLOC_RIGHT) {
997 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 998 MAIN_SECS(sbi), 0);
999 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1000 } else {
1001 go_left = 1;
1002 left_start = hint - 1;
1003 }
1004 }
1005 if (go_left == 0)
1006 goto skip_left;
1007
1008 while (test_bit(left_start, free_i->free_secmap)) {
1009 if (left_start > 0) {
1010 left_start--;
1011 continue;
1012 }
1013 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1014 MAIN_SECS(sbi), 0);
1015 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1016 break;
1017 }
1018 secno = left_start;
1019skip_left:
1020 hint = secno;
1021 segno = secno * sbi->segs_per_sec;
1022 zoneno = secno / sbi->secs_per_zone;
1023
1024 /* give up on finding another zone */
1025 if (!init)
1026 goto got_it;
1027 if (sbi->secs_per_zone == 1)
1028 goto got_it;
1029 if (zoneno == old_zoneno)
1030 goto got_it;
1031 if (dir == ALLOC_LEFT) {
1032 if (!go_left && zoneno + 1 >= total_zones)
1033 goto got_it;
1034 if (go_left && zoneno == 0)
1035 goto got_it;
1036 }
1037 for (i = 0; i < NR_CURSEG_TYPE; i++)
1038 if (CURSEG_I(sbi, i)->zone == zoneno)
1039 break;
1040
1041 if (i < NR_CURSEG_TYPE) {
1042 /* zone is in user, try another */
1043 if (go_left)
1044 hint = zoneno * sbi->secs_per_zone - 1;
1045 else if (zoneno + 1 >= total_zones)
1046 hint = 0;
1047 else
1048 hint = (zoneno + 1) * sbi->secs_per_zone;
1049 init = false;
1050 goto find_other_zone;
1051 }
1052got_it:
1053 /* set it as dirty segment in free segmap */
9850cf4a 1054 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1055 __set_inuse(sbi, segno);
1056 *newseg = segno;
1a118ccf 1057 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1058}
1059
1060static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1061{
1062 struct curseg_info *curseg = CURSEG_I(sbi, type);
1063 struct summary_footer *sum_footer;
1064
1065 curseg->segno = curseg->next_segno;
1066 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1067 curseg->next_blkoff = 0;
1068 curseg->next_segno = NULL_SEGNO;
1069
1070 sum_footer = &(curseg->sum_blk->footer);
1071 memset(sum_footer, 0, sizeof(struct summary_footer));
1072 if (IS_DATASEG(type))
1073 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1074 if (IS_NODESEG(type))
1075 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1076 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1077}
1078
0a8165d7 1079/*
351df4b2
JK		 1080 * Allocate a current working segment.
1081 * This function always allocates a free segment in LFS manner.
1082 */
1083static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1084{
1085 struct curseg_info *curseg = CURSEG_I(sbi, type);
1086 unsigned int segno = curseg->segno;
1087 int dir = ALLOC_LEFT;
1088
1089 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1090 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1091 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1092 dir = ALLOC_RIGHT;
1093
1094 if (test_opt(sbi, NOHEAP))
1095 dir = ALLOC_RIGHT;
1096
1097 get_new_segment(sbi, &segno, new_sec, dir);
1098 curseg->next_segno = segno;
1099 reset_curseg(sbi, type, 1);
1100 curseg->alloc_type = LFS;
1101}
1102
1103static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1104 struct curseg_info *seg, block_t start)
1105{
1106 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1107 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1108 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1109 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1110 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1111 int i, pos;
1112
1113 for (i = 0; i < entries; i++)
1114 target_map[i] = ckpt_map[i] | cur_map[i];
1115
1116 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1117
1118 seg->next_blkoff = pos;
351df4b2
JK		 1119}
1120
0a8165d7 1121/*
351df4b2
JK		 1122 * If a segment is written by LFS manner, next block offset is just obtained
1123 * by increasing the current block offset. However, if a segment is written by
1124 * SSR manner, next block offset obtained by calling __next_free_blkoff
1125 */
1126static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1127 struct curseg_info *seg)
1128{
1129 if (seg->alloc_type == SSR)
1130 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1131 else
1132 seg->next_blkoff++;
1133}
1134
0a8165d7 1135/*
e1c42045 1136 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1137 * manner, so it should recover the existing segment information of valid blocks
1138 */
1139static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1140{
1141 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1142 struct curseg_info *curseg = CURSEG_I(sbi, type);
1143 unsigned int new_segno = curseg->next_segno;
1144 struct f2fs_summary_block *sum_node;
1145 struct page *sum_page;
1146
1147 write_sum_page(sbi, curseg->sum_blk,
1148 GET_SUM_BLOCK(sbi, curseg->segno));
1149 __set_test_and_inuse(sbi, new_segno);
1150
1151 mutex_lock(&dirty_i->seglist_lock);
1152 __remove_dirty_segment(sbi, new_segno, PRE);
1153 __remove_dirty_segment(sbi, new_segno, DIRTY);
1154 mutex_unlock(&dirty_i->seglist_lock);
1155
1156 reset_curseg(sbi, type, 1);
1157 curseg->alloc_type = SSR;
1158 __next_free_blkoff(sbi, curseg, 0);
1159
1160 if (reuse) {
1161 sum_page = get_sum_page(sbi, new_segno);
1162 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1163 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1164 f2fs_put_page(sum_page, 1);
1165 }
1166}
1167
43727527
JK		 1168static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1169{
1170 struct curseg_info *curseg = CURSEG_I(sbi, type);
1171 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1172
1173 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1174 return v_ops->get_victim(sbi,
1175 &(curseg)->next_segno, BG_GC, type, SSR);
1176
1177 /* For data segments, let's do SSR more intensively */
1178 for (; type >= CURSEG_HOT_DATA; type--)
1179 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1180 BG_GC, type, SSR))
1181 return 1;
1182 return 0;
1183}
1184
351df4b2
JK		 1185/*
1186 * flush out current segment and replace it with new segment
1187 * This function should be returned with success, otherwise BUG
1188 */
1189static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1190 int type, bool force)
1191{
1192 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1193
7b405275 1194 if (force)
351df4b2 1195 new_curseg(sbi, type, true);
7b405275 1196 else if (type == CURSEG_WARM_NODE)
351df4b2 1197 new_curseg(sbi, type, false);
60374688
JK		 1198 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1199 new_curseg(sbi, type, false);
351df4b2
JK		 1200 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1201 change_curseg(sbi, type, true);
1202 else
1203 new_curseg(sbi, type, false);
dcdfff65
JK		 1204
1205 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1206}
1207
38aa0889
JK		 1208static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1209{
1210 struct curseg_info *curseg = CURSEG_I(sbi, type);
1211 unsigned int old_segno;
1212
1213 old_segno = curseg->segno;
1214 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1215 locate_dirty_segment(sbi, old_segno);
1216}
1217
351df4b2
JK		 1218void allocate_new_segments(struct f2fs_sb_info *sbi)
1219{
351df4b2
JK		 1220 int i;
1221
38aa0889
JK		 1222 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1223 __allocate_new_segments(sbi, i);
351df4b2
JK		 1224}
1225
1226static const struct segment_allocation default_salloc_ops = {
1227 .allocate_segment = allocate_segment_by_default,
1228};
1229
4b2fecc8
JK		 1230int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1231{
f7ef9b83
JK		 1232 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1233 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1234 unsigned int start_segno, end_segno;
1235 struct cp_control cpc;
c34f42e2 1236 int err = 0;
4b2fecc8 1237
836b5a63 1238 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1239 return -EINVAL;
1240
9bd27ae4 1241 cpc.trimmed = 0;
7cd8558b 1242 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1243 goto out;
1244
1245 /* start/end segment number in main_area */
7cd8558b
JK		 1246 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1247 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1248 GET_SEGNO(sbi, end);
4b2fecc8 1249 cpc.reason = CP_DISCARD;
836b5a63 1250 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1251
1252 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1253 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1254 cpc.trim_start = start_segno;
a66cdd98
JK		 1255
1256 if (sbi->discard_blks == 0)
1257 break;
1258 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1259 cpc.trim_end = end_segno;
1260 else
1261 cpc.trim_end = min_t(unsigned int,
1262 rounddown(start_segno +
bba681cb
JK		 1263 BATCHED_TRIM_SEGMENTS(sbi),
1264 sbi->segs_per_sec) - 1, end_segno);
1265
1266 mutex_lock(&sbi->gc_mutex);
c34f42e2 1267 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1268 mutex_unlock(&sbi->gc_mutex);
1269 }
4b2fecc8 1270out:
f7ef9b83 1271 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1272 return err;
4b2fecc8
JK		 1273}
1274
351df4b2
JK		 1275static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1276{
1277 struct curseg_info *curseg = CURSEG_I(sbi, type);
1278 if (curseg->next_blkoff < sbi->blocks_per_seg)
1279 return true;
1280 return false;
1281}
1282
1283static int __get_segment_type_2(struct page *page, enum page_type p_type)
1284{
1285 if (p_type == DATA)
1286 return CURSEG_HOT_DATA;
1287 else
1288 return CURSEG_HOT_NODE;
1289}
1290
1291static int __get_segment_type_4(struct page *page, enum page_type p_type)
1292{
1293 if (p_type == DATA) {
1294 struct inode *inode = page->mapping->host;
1295
1296 if (S_ISDIR(inode->i_mode))
1297 return CURSEG_HOT_DATA;
1298 else
1299 return CURSEG_COLD_DATA;
1300 } else {
a344b9fd
JK		 1301 if (IS_DNODE(page) && is_cold_node(page))
1302 return CURSEG_WARM_NODE;
351df4b2
JK		 1303 else
1304 return CURSEG_COLD_NODE;
1305 }
1306}
1307
1308static int __get_segment_type_6(struct page *page, enum page_type p_type)
1309{
1310 if (p_type == DATA) {
1311 struct inode *inode = page->mapping->host;
1312
1313 if (S_ISDIR(inode->i_mode))
1314 return CURSEG_HOT_DATA;
354a3399 1315 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1316 return CURSEG_COLD_DATA;
1317 else
1318 return CURSEG_WARM_DATA;
1319 } else {
1320 if (IS_DNODE(page))
1321 return is_cold_node(page) ? CURSEG_WARM_NODE :
1322 CURSEG_HOT_NODE;
1323 else
1324 return CURSEG_COLD_NODE;
1325 }
1326}
1327
1328static int __get_segment_type(struct page *page, enum page_type p_type)
1329{
4081363f 1330 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1331 case 2:
1332 return __get_segment_type_2(page, p_type);
1333 case 4:
1334 return __get_segment_type_4(page, p_type);
351df4b2 1335 }
12a67146 1336 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1337 f2fs_bug_on(F2FS_P_SB(page),
1338 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1339 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1340}
1341
bfad7c2d
JK		 1342void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1343 block_t old_blkaddr, block_t *new_blkaddr,
1344 struct f2fs_summary *sum, int type)
351df4b2
JK		 1345{
1346 struct sit_info *sit_i = SIT_I(sbi);
1347 struct curseg_info *curseg;
38aa0889
JK		 1348 bool direct_io = (type == CURSEG_DIRECT_IO);
1349
1350 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1351
351df4b2
JK		 1352 curseg = CURSEG_I(sbi, type);
1353
1354 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1355 mutex_lock(&sit_i->sentry_lock);
351df4b2 1356
38aa0889 1357 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1358 if (direct_io && curseg->next_blkoff &&
1359 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1360 __allocate_new_segments(sbi, type);
1361
351df4b2 1362 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1363
1364 /*
1365 * __add_sum_entry should be resided under the curseg_mutex
1366 * because, this function updates a summary entry in the
1367 * current summary block.
1368 */
e79efe3b 1369 __add_sum_entry(sbi, type, sum);
351df4b2 1370
351df4b2 1371 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1372
1373 stat_inc_block_count(sbi, curseg);
351df4b2 1374
5e443818
JK		 1375 if (!__has_curseg_space(sbi, type))
1376 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1377 /*
1378 * SIT information should be updated before segment allocation,
1379 * since SSR needs latest valid block information.
1380 */
1381 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1382
351df4b2
JK		 1383 mutex_unlock(&sit_i->sentry_lock);
1384
bfad7c2d 1385 if (page && IS_NODESEG(type))
351df4b2
JK		 1386 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1387
bfad7c2d
JK		 1388 mutex_unlock(&curseg->curseg_mutex);
1389}
1390
05ca3632 1391static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1392{
05ca3632 1393 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1394
7a9d7548
CY		 1395 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1396 &fio->new_blkaddr, sum, type);
bfad7c2d 1397
351df4b2 1398 /* writeout dirty page into bdev */
05ca3632 1399 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1400}
1401
577e3495 1402void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1403{
458e6197 1404 struct f2fs_io_info fio = {
05ca3632 1405 .sbi = sbi,
458e6197 1406 .type = META,
04d328de
MC		 1407 .op = REQ_OP_WRITE,
1408 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1409 .old_blkaddr = page->index,
1410 .new_blkaddr = page->index,
05ca3632 1411 .page = page,
4375a336 1412 .encrypted_page = NULL,
458e6197
JK		 1413 };
1414
2b947003 1415 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1416 fio.op_flags &= ~REQ_META;
2b947003 1417
351df4b2 1418 set_page_writeback(page);
05ca3632 1419 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1420}
1421
05ca3632 1422void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1423{
1424 struct f2fs_summary sum;
05ca3632 1425
351df4b2 1426 set_summary(&sum, nid, 0, 0);
05ca3632 1427 do_write_page(&sum, fio);
351df4b2
JK		 1428}
1429
05ca3632 1430void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1431{
05ca3632 1432 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1433 struct f2fs_summary sum;
1434 struct node_info ni;
1435
9850cf4a 1436 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1437 get_node_info(sbi, dn->nid, &ni);
1438 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1439 do_write_page(&sum, fio);
f28b3434 1440 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1441}
1442
05ca3632 1443void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1444{
7a9d7548 1445 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1446 stat_inc_inplace_blocks(fio->sbi);
1447 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1448}
1449
4356e48e 1450void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1451 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1452 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1453{
1454 struct sit_info *sit_i = SIT_I(sbi);
1455 struct curseg_info *curseg;
1456 unsigned int segno, old_cursegno;
1457 struct seg_entry *se;
1458 int type;
19f106bc 1459 unsigned short old_blkoff;
351df4b2
JK		 1460
1461 segno = GET_SEGNO(sbi, new_blkaddr);
1462 se = get_seg_entry(sbi, segno);
1463 type = se->type;
1464
19f106bc
CY		 1465 if (!recover_curseg) {
1466 /* for recovery flow */
1467 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1468 if (old_blkaddr == NULL_ADDR)
1469 type = CURSEG_COLD_DATA;
1470 else
1471 type = CURSEG_WARM_DATA;
1472 }
1473 } else {
1474 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1475 type = CURSEG_WARM_DATA;
1476 }
19f106bc 1477
351df4b2
JK		 1478 curseg = CURSEG_I(sbi, type);
1479
1480 mutex_lock(&curseg->curseg_mutex);
1481 mutex_lock(&sit_i->sentry_lock);
1482
1483 old_cursegno = curseg->segno;
19f106bc 1484 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1485
1486 /* change the current segment */
1487 if (segno != curseg->segno) {
1488 curseg->next_segno = segno;
1489 change_curseg(sbi, type, true);
1490 }
1491
491c0854 1492 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1493 __add_sum_entry(sbi, type, sum);
351df4b2 1494
28bc106b 1495 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1496 update_sit_entry(sbi, new_blkaddr, 1);
1497 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1498 update_sit_entry(sbi, old_blkaddr, -1);
1499
1500 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1501 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1502
351df4b2 1503 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1504
19f106bc
CY		 1505 if (recover_curseg) {
1506 if (old_cursegno != curseg->segno) {
1507 curseg->next_segno = old_cursegno;
1508 change_curseg(sbi, type, true);
1509 }
1510 curseg->next_blkoff = old_blkoff;
1511 }
1512
351df4b2
JK		 1513 mutex_unlock(&sit_i->sentry_lock);
1514 mutex_unlock(&curseg->curseg_mutex);
1515}
1516
528e3459
CY		 1517void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1518 block_t old_addr, block_t new_addr,
28bc106b
CY		 1519 unsigned char version, bool recover_curseg,
1520 bool recover_newaddr)
528e3459
CY		 1521{
1522 struct f2fs_summary sum;
1523
1524 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1525
28bc106b
CY		 1526 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1527 recover_curseg, recover_newaddr);
528e3459 1528
f28b3434 1529 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1530}
1531
93dfe2ac 1532void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1533 enum page_type type, bool ordered)
93dfe2ac 1534{
93dfe2ac 1535 if (PageWriteback(page)) {
4081363f
JK		 1536 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1537
0c3a5797 1538 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1539 if (ordered)
1540 wait_on_page_writeback(page);
1541 else
1542 wait_for_stable_page(page);
93dfe2ac
JK		 1543 }
1544}
1545
08b39fbd
CY		 1546void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1547 block_t blkaddr)
1548{
1549 struct page *cpage;
1550
1551 if (blkaddr == NEW_ADDR)
1552 return;
1553
1554 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1555
1556 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1557 if (cpage) {
fec1d657 1558 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1559 f2fs_put_page(cpage, 1);
1560 }
1561}
1562
351df4b2
JK		 1563static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1564{
1565 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1566 struct curseg_info *seg_i;
1567 unsigned char *kaddr;
1568 struct page *page;
1569 block_t start;
1570 int i, j, offset;
1571
1572 start = start_sum_block(sbi);
1573
1574 page = get_meta_page(sbi, start++);
1575 kaddr = (unsigned char *)page_address(page);
1576
1577 /* Step 1: restore nat cache */
1578 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1579 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1580
1581 /* Step 2: restore sit cache */
1582 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1583 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1584 offset = 2 * SUM_JOURNAL_SIZE;
1585
1586 /* Step 3: restore summary entries */
1587 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1588 unsigned short blk_off;
1589 unsigned int segno;
1590
1591 seg_i = CURSEG_I(sbi, i);
1592 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1593 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1594 seg_i->next_segno = segno;
1595 reset_curseg(sbi, i, 0);
1596 seg_i->alloc_type = ckpt->alloc_type[i];
1597 seg_i->next_blkoff = blk_off;
1598
1599 if (seg_i->alloc_type == SSR)
1600 blk_off = sbi->blocks_per_seg;
1601
1602 for (j = 0; j < blk_off; j++) {
1603 struct f2fs_summary *s;
1604 s = (struct f2fs_summary *)(kaddr + offset);
1605 seg_i->sum_blk->entries[j] = *s;
1606 offset += SUMMARY_SIZE;
09cbfeaf 1607 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1608 SUM_FOOTER_SIZE)
1609 continue;
1610
1611 f2fs_put_page(page, 1);
1612 page = NULL;
1613
1614 page = get_meta_page(sbi, start++);
1615 kaddr = (unsigned char *)page_address(page);
1616 offset = 0;
1617 }
1618 }
1619 f2fs_put_page(page, 1);
1620 return 0;
1621}
1622
1623static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1624{
1625 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1626 struct f2fs_summary_block *sum;
1627 struct curseg_info *curseg;
1628 struct page *new;
1629 unsigned short blk_off;
1630 unsigned int segno = 0;
1631 block_t blk_addr = 0;
1632
1633 /* get segment number and block addr */
1634 if (IS_DATASEG(type)) {
1635 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1636 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1637 CURSEG_HOT_DATA]);
119ee914 1638 if (__exist_node_summaries(sbi))
351df4b2
JK		 1639 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1640 else
1641 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1642 } else {
1643 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1644 CURSEG_HOT_NODE]);
1645 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1646 CURSEG_HOT_NODE]);
119ee914 1647 if (__exist_node_summaries(sbi))
351df4b2
JK		 1648 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1649 type - CURSEG_HOT_NODE);
1650 else
1651 blk_addr = GET_SUM_BLOCK(sbi, segno);
1652 }
1653
1654 new = get_meta_page(sbi, blk_addr);
1655 sum = (struct f2fs_summary_block *)page_address(new);
1656
1657 if (IS_NODESEG(type)) {
119ee914 1658 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1659 struct f2fs_summary *ns = &sum->entries[0];
1660 int i;
1661 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1662 ns->version = 0;
1663 ns->ofs_in_node = 0;
1664 }
1665 } else {
d653788a
GZ		 1666 int err;
1667
1668 err = restore_node_summary(sbi, segno, sum);
1669 if (err) {
351df4b2 1670 f2fs_put_page(new, 1);
d653788a 1671 return err;
351df4b2
JK		 1672 }
1673 }
1674 }
1675
1676 /* set uncompleted segment to curseg */
1677 curseg = CURSEG_I(sbi, type);
1678 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1679
1680 /* update journal info */
1681 down_write(&curseg->journal_rwsem);
1682 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1683 up_write(&curseg->journal_rwsem);
1684
1685 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1686 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1687 curseg->next_segno = segno;
1688 reset_curseg(sbi, type, 0);
1689 curseg->alloc_type = ckpt->alloc_type[type];
1690 curseg->next_blkoff = blk_off;
1691 mutex_unlock(&curseg->curseg_mutex);
1692 f2fs_put_page(new, 1);
1693 return 0;
1694}
1695
1696static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1697{
1698 int type = CURSEG_HOT_DATA;
e4fc5fbf 1699 int err;
351df4b2 1700
25ca923b 1701 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1702 int npages = npages_for_summary_flush(sbi, true);
1703
1704 if (npages >= 2)
1705 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1706 META_CP, true);
3fa06d7b 1707
351df4b2
JK		 1708 /* restore for compacted data summary */
1709 if (read_compacted_summaries(sbi))
1710 return -EINVAL;
1711 type = CURSEG_HOT_NODE;
1712 }
1713
119ee914 1714 if (__exist_node_summaries(sbi))
3fa06d7b 1715 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1716 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1717
e4fc5fbf
CY		 1718 for (; type <= CURSEG_COLD_NODE; type++) {
1719 err = read_normal_summaries(sbi, type);
1720 if (err)
1721 return err;
1722 }
1723
351df4b2
JK		 1724 return 0;
1725}
1726
1727static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1728{
1729 struct page *page;
1730 unsigned char *kaddr;
1731 struct f2fs_summary *summary;
1732 struct curseg_info *seg_i;
1733 int written_size = 0;
1734 int i, j;
1735
1736 page = grab_meta_page(sbi, blkaddr++);
1737 kaddr = (unsigned char *)page_address(page);
1738
1739 /* Step 1: write nat cache */
1740 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1741 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1742 written_size += SUM_JOURNAL_SIZE;
1743
1744 /* Step 2: write sit cache */
1745 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1746 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1747 written_size += SUM_JOURNAL_SIZE;
1748
351df4b2
JK		 1749 /* Step 3: write summary entries */
1750 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1751 unsigned short blkoff;
1752 seg_i = CURSEG_I(sbi, i);
1753 if (sbi->ckpt->alloc_type[i] == SSR)
1754 blkoff = sbi->blocks_per_seg;
1755 else
1756 blkoff = curseg_blkoff(sbi, i);
1757
1758 for (j = 0; j < blkoff; j++) {
1759 if (!page) {
1760 page = grab_meta_page(sbi, blkaddr++);
1761 kaddr = (unsigned char *)page_address(page);
1762 written_size = 0;
1763 }
1764 summary = (struct f2fs_summary *)(kaddr + written_size);
1765 *summary = seg_i->sum_blk->entries[j];
1766 written_size += SUMMARY_SIZE;
351df4b2 1767
09cbfeaf 1768 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1769 SUM_FOOTER_SIZE)
1770 continue;
1771
e8d61a74 1772 set_page_dirty(page);
351df4b2
JK		 1773 f2fs_put_page(page, 1);
1774 page = NULL;
1775 }
1776 }
e8d61a74
CY		 1777 if (page) {
1778 set_page_dirty(page);
351df4b2 1779 f2fs_put_page(page, 1);
e8d61a74 1780 }
351df4b2
JK		 1781}
1782
1783static void write_normal_summaries(struct f2fs_sb_info *sbi,
1784 block_t blkaddr, int type)
1785{
1786 int i, end;
1787 if (IS_DATASEG(type))
1788 end = type + NR_CURSEG_DATA_TYPE;
1789 else
1790 end = type + NR_CURSEG_NODE_TYPE;
1791
b7ad7512
CY		 1792 for (i = type; i < end; i++)
1793 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1794}
1795
1796void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1797{
25ca923b 1798 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1799 write_compacted_summaries(sbi, start_blk);
1800 else
1801 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1802}
1803
1804void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1805{
119ee914 1806 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1807}
1808
dfc08a12 1809int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1810 unsigned int val, int alloc)
1811{
1812 int i;
1813
1814 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1815 for (i = 0; i < nats_in_cursum(journal); i++) {
1816 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1817 return i;
1818 }
dfc08a12
CY		 1819 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1820 return update_nats_in_cursum(journal, 1);
351df4b2 1821 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1822 for (i = 0; i < sits_in_cursum(journal); i++)
1823 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1824 return i;
dfc08a12
CY		 1825 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1826 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1827 }
1828 return -1;
1829}
1830
1831static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1832 unsigned int segno)
1833{
2cc22186 1834 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1835}
1836
1837static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1838 unsigned int start)
1839{
1840 struct sit_info *sit_i = SIT_I(sbi);
1841 struct page *src_page, *dst_page;
1842 pgoff_t src_off, dst_off;
1843 void *src_addr, *dst_addr;
1844
1845 src_off = current_sit_addr(sbi, start);
1846 dst_off = next_sit_addr(sbi, src_off);
1847
1848 /* get current sit block page without lock */
1849 src_page = get_meta_page(sbi, src_off);
1850 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1851 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1852
1853 src_addr = page_address(src_page);
1854 dst_addr = page_address(dst_page);
09cbfeaf 1855 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1856
1857 set_page_dirty(dst_page);
1858 f2fs_put_page(src_page, 1);
1859
1860 set_to_next_sit(sit_i, start);
1861
1862 return dst_page;
1863}
1864
184a5cd2
CY		 1865static struct sit_entry_set *grab_sit_entry_set(void)
1866{
1867 struct sit_entry_set *ses =
80c54505 1868 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1869
1870 ses->entry_cnt = 0;
1871 INIT_LIST_HEAD(&ses->set_list);
1872 return ses;
1873}
1874
1875static void release_sit_entry_set(struct sit_entry_set *ses)
1876{
1877 list_del(&ses->set_list);
1878 kmem_cache_free(sit_entry_set_slab, ses);
1879}
1880
1881static void adjust_sit_entry_set(struct sit_entry_set *ses,
1882 struct list_head *head)
1883{
1884 struct sit_entry_set *next = ses;
1885
1886 if (list_is_last(&ses->set_list, head))
1887 return;
1888
1889 list_for_each_entry_continue(next, head, set_list)
1890 if (ses->entry_cnt <= next->entry_cnt)
1891 break;
1892
1893 list_move_tail(&ses->set_list, &next->set_list);
1894}
1895
1896static void add_sit_entry(unsigned int segno, struct list_head *head)
1897{
1898 struct sit_entry_set *ses;
1899 unsigned int start_segno = START_SEGNO(segno);
1900
1901 list_for_each_entry(ses, head, set_list) {
1902 if (ses->start_segno == start_segno) {
1903 ses->entry_cnt++;
1904 adjust_sit_entry_set(ses, head);
1905 return;
1906 }
1907 }
1908
1909 ses = grab_sit_entry_set();
1910
1911 ses->start_segno = start_segno;
1912 ses->entry_cnt++;
1913 list_add(&ses->set_list, head);
1914}
1915
1916static void add_sits_in_set(struct f2fs_sb_info *sbi)
1917{
1918 struct f2fs_sm_info *sm_info = SM_I(sbi);
1919 struct list_head *set_list = &sm_info->sit_entry_set;
1920 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1921 unsigned int segno;
1922
7cd8558b 1923 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1924 add_sit_entry(segno, set_list);
1925}
1926
1927static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1928{
1929 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1930 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1931 int i;
1932
b7ad7512 1933 down_write(&curseg->journal_rwsem);
dfc08a12 1934 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1935 unsigned int segno;
1936 bool dirtied;
1937
dfc08a12 1938 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1939 dirtied = __mark_sit_entry_dirty(sbi, segno);
1940
1941 if (!dirtied)
1942 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1943 }
dfc08a12 1944 update_sits_in_cursum(journal, -i);
b7ad7512 1945 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1946}
1947
0a8165d7 1948/*
351df4b2
JK		 1949 * CP calls this function, which flushes SIT entries including sit_journal,
1950 * and moves prefree segs to free segs.
1951 */
4b2fecc8 1952void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1953{
1954 struct sit_info *sit_i = SIT_I(sbi);
1955 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1956 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1957 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 1958 struct sit_entry_set *ses, *tmp;
1959 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 1960 bool to_journal = true;
4b2fecc8 1961 struct seg_entry *se;
351df4b2 1962
351df4b2
JK		 1963 mutex_lock(&sit_i->sentry_lock);
1964
2b11a74b
WL		 1965 if (!sit_i->dirty_sentries)
1966 goto out;
1967
351df4b2 1968 /*
184a5cd2
CY		 1969 * add and account sit entries of dirty bitmap in sit entry
1970 * set temporarily
351df4b2 1971 */
184a5cd2 1972 add_sits_in_set(sbi);
351df4b2 1973
184a5cd2
CY		 1974 /*
1975 * if there are no enough space in journal to store dirty sit
1976 * entries, remove all entries from journal and add and account
1977 * them in sit entry set.
1978 */
dfc08a12 1979 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 1980 remove_sits_in_journal(sbi);
b2955550 1981
184a5cd2
CY		 1982 /*
1983 * there are two steps to flush sit entries:
1984 * #1, flush sit entries to journal in current cold data summary block.
1985 * #2, flush sit entries to sit page.
1986 */
1987 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 1988 struct page *page = NULL;
184a5cd2
CY		 1989 struct f2fs_sit_block *raw_sit = NULL;
1990 unsigned int start_segno = ses->start_segno;
1991 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 1992 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 1993 unsigned int segno = start_segno;
1994
1995 if (to_journal &&
dfc08a12 1996 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 1997 to_journal = false;
1998
b7ad7512
CY		 1999 if (to_journal) {
2000 down_write(&curseg->journal_rwsem);
2001 } else {
184a5cd2
CY		 2002 page = get_next_sit_page(sbi, start_segno);
2003 raw_sit = page_address(page);
351df4b2 2004 }
351df4b2 2005
184a5cd2
CY		 2006 /* flush dirty sit entries in region of current sit set */
2007 for_each_set_bit_from(segno, bitmap, end) {
2008 int offset, sit_offset;
4b2fecc8
JK		 2009
2010 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2011
2012 /* add discard candidates */
d7bc2484 2013 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2014 cpc->trim_start = segno;
2015 add_discard_addrs(sbi, cpc);
2016 }
184a5cd2
CY		 2017
2018 if (to_journal) {
dfc08a12 2019 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2020 SIT_JOURNAL, segno, 1);
2021 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2022 segno_in_journal(journal, offset) =
184a5cd2
CY		 2023 cpu_to_le32(segno);
2024 seg_info_to_raw_sit(se,
dfc08a12 2025 &sit_in_journal(journal, offset));
184a5cd2
CY		 2026 } else {
2027 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2028 seg_info_to_raw_sit(se,
2029 &raw_sit->entries[sit_offset]);
2030 }
351df4b2 2031
184a5cd2
CY		 2032 __clear_bit(segno, bitmap);
2033 sit_i->dirty_sentries--;
2034 ses->entry_cnt--;
351df4b2
JK		 2035 }
2036
b7ad7512
CY		 2037 if (to_journal)
2038 up_write(&curseg->journal_rwsem);
2039 else
184a5cd2
CY		 2040 f2fs_put_page(page, 1);
2041
2042 f2fs_bug_on(sbi, ses->entry_cnt);
2043 release_sit_entry_set(ses);
351df4b2 2044 }
184a5cd2
CY		 2045
2046 f2fs_bug_on(sbi, !list_empty(head));
2047 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2048out:
4b2fecc8
JK		 2049 if (cpc->reason == CP_DISCARD) {
2050 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2051 add_discard_addrs(sbi, cpc);
2052 }
351df4b2 2053 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2054
351df4b2
JK		 2055 set_prefree_as_free_segments(sbi);
2056}
2057
2058static int build_sit_info(struct f2fs_sb_info *sbi)
2059{
2060 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2061 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2062 struct sit_info *sit_i;
2063 unsigned int sit_segs, start;
2064 char *src_bitmap, *dst_bitmap;
2065 unsigned int bitmap_size;
2066
2067 /* allocate memory for SIT information */
2068 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2069 if (!sit_i)
2070 return -ENOMEM;
2071
2072 SM_I(sbi)->sit_info = sit_i;
2073
39307a8e
JK		 2074 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2075 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2076 if (!sit_i->sentries)
2077 return -ENOMEM;
2078
7cd8558b 2079 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2080 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2081 if (!sit_i->dirty_sentries_bitmap)
2082 return -ENOMEM;
2083
7cd8558b 2084 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2085 sit_i->sentries[start].cur_valid_map
2086 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2087 sit_i->sentries[start].ckpt_valid_map
2088 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2089 sit_i->sentries[start].discard_map
2090 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2091 if (!sit_i->sentries[start].cur_valid_map ||
2092 !sit_i->sentries[start].ckpt_valid_map ||
2093 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2094 return -ENOMEM;
2095 }
2096
60a3b782
JK		 2097 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2098 if (!sit_i->tmp_map)
2099 return -ENOMEM;
2100
351df4b2 2101 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2102 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2103 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2104 if (!sit_i->sec_entries)
2105 return -ENOMEM;
2106 }
2107
2108 /* get information related with SIT */
2109 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2110
2111 /* setup SIT bitmap from ckeckpoint pack */
2112 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2113 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2114
79b5793b 2115 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2116 if (!dst_bitmap)
2117 return -ENOMEM;
351df4b2
JK		 2118
2119 /* init SIT information */
2120 sit_i->s_ops = &default_salloc_ops;
2121
2122 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2123 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2124 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2125 sit_i->sit_bitmap = dst_bitmap;
2126 sit_i->bitmap_size = bitmap_size;
2127 sit_i->dirty_sentries = 0;
2128 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2129 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2130 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2131 mutex_init(&sit_i->sentry_lock);
2132 return 0;
2133}
2134
2135static int build_free_segmap(struct f2fs_sb_info *sbi)
2136{
351df4b2
JK		 2137 struct free_segmap_info *free_i;
2138 unsigned int bitmap_size, sec_bitmap_size;
2139
2140 /* allocate memory for free segmap information */
2141 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2142 if (!free_i)
2143 return -ENOMEM;
2144
2145 SM_I(sbi)->free_info = free_i;
2146
7cd8558b 2147 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2148 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2149 if (!free_i->free_segmap)
2150 return -ENOMEM;
2151
7cd8558b 2152 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2153 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2154 if (!free_i->free_secmap)
2155 return -ENOMEM;
2156
2157 /* set all segments as dirty temporarily */
2158 memset(free_i->free_segmap, 0xff, bitmap_size);
2159 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2160
2161 /* init free segmap information */
7cd8558b 2162 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2163 free_i->free_segments = 0;
2164 free_i->free_sections = 0;
1a118ccf 2165 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2166 return 0;
2167}
2168
2169static int build_curseg(struct f2fs_sb_info *sbi)
2170{
1042d60f 2171 struct curseg_info *array;
351df4b2
JK		 2172 int i;
2173
b434babf 2174 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2175 if (!array)
2176 return -ENOMEM;
2177
2178 SM_I(sbi)->curseg_array = array;
2179
2180 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2181 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2182 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2183 if (!array[i].sum_blk)
2184 return -ENOMEM;
b7ad7512
CY		 2185 init_rwsem(&array[i].journal_rwsem);
2186 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2187 GFP_KERNEL);
2188 if (!array[i].journal)
2189 return -ENOMEM;
351df4b2
JK		 2190 array[i].segno = NULL_SEGNO;
2191 array[i].next_blkoff = 0;
2192 }
2193 return restore_curseg_summaries(sbi);
2194}
2195
2196static void build_sit_entries(struct f2fs_sb_info *sbi)
2197{
2198 struct sit_info *sit_i = SIT_I(sbi);
2199 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2200 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2201 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2202 unsigned int i, start, end;
2203 unsigned int readed, start_blk = 0;
e9f5b8b8 2204 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2205
74de593a 2206 do {
26879fb1 2207 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2208
2209 start = start_blk * sit_i->sents_per_block;
2210 end = (start_blk + readed) * sit_i->sents_per_block;
2211
7cd8558b 2212 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2213 struct seg_entry *se = &sit_i->sentries[start];
2214 struct f2fs_sit_block *sit_blk;
2215 struct f2fs_sit_entry sit;
2216 struct page *page;
2217
b7ad7512 2218 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2219 for (i = 0; i < sits_in_cursum(journal); i++) {
2220 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2221 == start) {
dfc08a12 2222 sit = sit_in_journal(journal, i);
b7ad7512 2223 up_read(&curseg->journal_rwsem);
74de593a
CY		 2224 goto got_it;
2225 }
351df4b2 2226 }
b7ad7512 2227 up_read(&curseg->journal_rwsem);
74de593a
CY		 2228
2229 page = get_current_sit_page(sbi, start);
2230 sit_blk = (struct f2fs_sit_block *)page_address(page);
2231 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2232 f2fs_put_page(page, 1);
351df4b2 2233got_it:
74de593a
CY		 2234 check_block_count(sbi, start, &sit);
2235 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2236
2237 /* build discard map only one time */
2238 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2239 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2240
74de593a
CY		 2241 if (sbi->segs_per_sec > 1) {
2242 struct sec_entry *e = get_sec_entry(sbi, start);
2243 e->valid_blocks += se->valid_blocks;
2244 }
351df4b2 2245 }
74de593a
CY		 2246 start_blk += readed;
2247 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2248}
2249
2250static void init_free_segmap(struct f2fs_sb_info *sbi)
2251{
2252 unsigned int start;
2253 int type;
2254
7cd8558b 2255 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2256 struct seg_entry *sentry = get_seg_entry(sbi, start);
2257 if (!sentry->valid_blocks)
2258 __set_free(sbi, start);
2259 }
2260
2261 /* set use the current segments */
2262 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2263 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2264 __set_test_and_inuse(sbi, curseg_t->segno);
2265 }
2266}
2267
2268static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2269{
2270 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2271 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2272 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2273 unsigned short valid_blocks;
2274
8736fbf0 2275 while (1) {
351df4b2 2276 /* find dirty segment based on free segmap */
7cd8558b
JK		 2277 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2278 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2279 break;
2280 offset = segno + 1;
2281 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2282 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2283 continue;
ec325b52
JK		 2284 if (valid_blocks > sbi->blocks_per_seg) {
2285 f2fs_bug_on(sbi, 1);
2286 continue;
2287 }
351df4b2
JK		 2288 mutex_lock(&dirty_i->seglist_lock);
2289 __locate_dirty_segment(sbi, segno, DIRTY);
2290 mutex_unlock(&dirty_i->seglist_lock);
2291 }
2292}
2293
5ec4e49f 2294static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2295{
2296 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2297 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2298
39307a8e 2299 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2300 if (!dirty_i->victim_secmap)
351df4b2
JK		 2301 return -ENOMEM;
2302 return 0;
2303}
2304
2305static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2306{
2307 struct dirty_seglist_info *dirty_i;
2308 unsigned int bitmap_size, i;
2309
2310 /* allocate memory for dirty segments list information */
2311 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2312 if (!dirty_i)
2313 return -ENOMEM;
2314
2315 SM_I(sbi)->dirty_info = dirty_i;
2316 mutex_init(&dirty_i->seglist_lock);
2317
7cd8558b 2318 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2319
2320 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2321 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2322 if (!dirty_i->dirty_segmap[i])
2323 return -ENOMEM;
2324 }
2325
2326 init_dirty_segmap(sbi);
5ec4e49f 2327 return init_victim_secmap(sbi);
351df4b2
JK		 2328}
2329
0a8165d7 2330/*
351df4b2
JK		 2331 * Update min, max modified time for cost-benefit GC algorithm
2332 */
2333static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2334{
2335 struct sit_info *sit_i = SIT_I(sbi);
2336 unsigned int segno;
2337
2338 mutex_lock(&sit_i->sentry_lock);
2339
2340 sit_i->min_mtime = LLONG_MAX;
2341
7cd8558b 2342 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2343 unsigned int i;
2344 unsigned long long mtime = 0;
2345
2346 for (i = 0; i < sbi->segs_per_sec; i++)
2347 mtime += get_seg_entry(sbi, segno + i)->mtime;
2348
2349 mtime = div_u64(mtime, sbi->segs_per_sec);
2350
2351 if (sit_i->min_mtime > mtime)
2352 sit_i->min_mtime = mtime;
2353 }
2354 sit_i->max_mtime = get_mtime(sbi);
2355 mutex_unlock(&sit_i->sentry_lock);
2356}
2357
2358int build_segment_manager(struct f2fs_sb_info *sbi)
2359{
2360 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2361 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2362 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2363 int err;
2364
2365 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2366 if (!sm_info)
2367 return -ENOMEM;
2368
2369 /* init sm info */
2370 sbi->sm_info = sm_info;
351df4b2
JK		 2371 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2372 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2373 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2374 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2375 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2376 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2377 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2378 sm_info->rec_prefree_segments = sm_info->main_segments *
2379 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
9b5f136f 2380 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2381 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2382 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2383
7fd9e544
JK		 2384 INIT_LIST_HEAD(&sm_info->discard_list);
2385 sm_info->nr_discards = 0;
2386 sm_info->max_discards = 0;
2387
bba681cb
JK		 2388 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2389
184a5cd2
CY		 2390 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2391
b270ad6f 2392 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2393 err = create_flush_cmd_control(sbi);
2394 if (err)
a688b9d9 2395 return err;
6b4afdd7
JK		 2396 }
2397
351df4b2
JK		 2398 err = build_sit_info(sbi);
2399 if (err)
2400 return err;
2401 err = build_free_segmap(sbi);
2402 if (err)
2403 return err;
2404 err = build_curseg(sbi);
2405 if (err)
2406 return err;
2407
2408 /* reinit free segmap based on SIT */
2409 build_sit_entries(sbi);
2410
2411 init_free_segmap(sbi);
2412 err = build_dirty_segmap(sbi);
2413 if (err)
2414 return err;
2415
2416 init_min_max_mtime(sbi);
2417 return 0;
2418}
2419
2420static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2421 enum dirty_type dirty_type)
2422{
2423 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2424
2425 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2426 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2427 dirty_i->nr_dirty[dirty_type] = 0;
2428 mutex_unlock(&dirty_i->seglist_lock);
2429}
2430
5ec4e49f 2431static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2432{
2433 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2434 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2435}
2436
2437static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2438{
2439 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2440 int i;
2441
2442 if (!dirty_i)
2443 return;
2444
2445 /* discard pre-free/dirty segments list */
2446 for (i = 0; i < NR_DIRTY_TYPE; i++)
2447 discard_dirty_segmap(sbi, i);
2448
5ec4e49f 2449 destroy_victim_secmap(sbi);
351df4b2
JK		 2450 SM_I(sbi)->dirty_info = NULL;
2451 kfree(dirty_i);
2452}
2453
2454static void destroy_curseg(struct f2fs_sb_info *sbi)
2455{
2456 struct curseg_info *array = SM_I(sbi)->curseg_array;
2457 int i;
2458
2459 if (!array)
2460 return;
2461 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2462 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2463 kfree(array[i].sum_blk);
b7ad7512
CY		 2464 kfree(array[i].journal);
2465 }
351df4b2
JK		 2466 kfree(array);
2467}
2468
2469static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2470{
2471 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2472 if (!free_i)
2473 return;
2474 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2475 kvfree(free_i->free_segmap);
2476 kvfree(free_i->free_secmap);
351df4b2
JK		 2477 kfree(free_i);
2478}
2479
2480static void destroy_sit_info(struct f2fs_sb_info *sbi)
2481{
2482 struct sit_info *sit_i = SIT_I(sbi);
2483 unsigned int start;
2484
2485 if (!sit_i)
2486 return;
2487
2488 if (sit_i->sentries) {
7cd8558b 2489 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2490 kfree(sit_i->sentries[start].cur_valid_map);
2491 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2492 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2493 }
2494 }
60a3b782
JK		 2495 kfree(sit_i->tmp_map);
2496
39307a8e
JK		 2497 kvfree(sit_i->sentries);
2498 kvfree(sit_i->sec_entries);
2499 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2500
2501 SM_I(sbi)->sit_info = NULL;
2502 kfree(sit_i->sit_bitmap);
2503 kfree(sit_i);
2504}
2505
2506void destroy_segment_manager(struct f2fs_sb_info *sbi)
2507{
2508 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2509
3b03f724
CY		 2510 if (!sm_info)
2511 return;
2163d198 2512 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2513 destroy_dirty_segmap(sbi);
2514 destroy_curseg(sbi);
2515 destroy_free_segmap(sbi);
2516 destroy_sit_info(sbi);
2517 sbi->sm_info = NULL;
2518 kfree(sm_info);
2519}
7fd9e544
JK		 2520
2521int __init create_segment_manager_caches(void)
2522{
2523 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2524 sizeof(struct discard_entry));
7fd9e544 2525 if (!discard_entry_slab)
184a5cd2
CY		 2526 goto fail;
2527
2528 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2529 sizeof(struct sit_entry_set));
184a5cd2
CY		 2530 if (!sit_entry_set_slab)
2531 goto destory_discard_entry;
88b88a66
JK		 2532
2533 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2534 sizeof(struct inmem_pages));
2535 if (!inmem_entry_slab)
2536 goto destroy_sit_entry_set;
7fd9e544 2537 return 0;
184a5cd2 2538
88b88a66
JK		 2539destroy_sit_entry_set:
2540 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2541destory_discard_entry:
2542 kmem_cache_destroy(discard_entry_slab);
2543fail:
2544 return -ENOMEM;
7fd9e544
JK		 2545}
2546
2547void destroy_segment_manager_caches(void)
2548{
184a5cd2 2549 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2550 kmem_cache_destroy(discard_entry_slab);
88b88a66 2551 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2552}
Linux kernel - Deliverables
This page took 0.269188 seconds and 5 git commands to generate.