]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: not allow to write illegal blkaddr
[mirror_ubuntu-artful-kernel.git] / fs / f2fs / segment.c
CommitLineData
0a8165d7 1/*
351df4b2
JK		 2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/bio.h>
14#include <linux/blkdev.h>
690e4a3e 15#include <linux/prefetch.h>
6b4afdd7 16#include <linux/kthread.h>
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
a66cdd98 663 if (se->valid_blocks == max_blocks)
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);
a66cdd98
JK		 821 if (!f2fs_test_and_set_bit(offset, se->discard_map))
822 sbi->discard_blks--;
351df4b2 823 } else {
52aca074 824 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 825 f2fs_bug_on(sbi, 1);
a66cdd98
JK		 826 if (f2fs_test_and_clear_bit(offset, se->discard_map))
827 sbi->discard_blks++;
351df4b2
JK		 828 }
829 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
830 se->ckpt_valid_blocks += del;
831
832 __mark_sit_entry_dirty(sbi, segno);
833
834 /* update total number of valid blocks to be written in ckpt area */
835 SIT_I(sbi)->written_valid_blocks += del;
836
837 if (sbi->segs_per_sec > 1)
838 get_sec_entry(sbi, segno)->valid_blocks += del;
839}
840
5e443818 841void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 842{
5e443818
JK		 843 update_sit_entry(sbi, new, 1);
844 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
845 update_sit_entry(sbi, old, -1);
846
847 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
848 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 849}
850
851void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
852{
853 unsigned int segno = GET_SEGNO(sbi, addr);
854 struct sit_info *sit_i = SIT_I(sbi);
855
9850cf4a 856 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 857 if (addr == NEW_ADDR)
858 return;
859
860 /* add it into sit main buffer */
861 mutex_lock(&sit_i->sentry_lock);
862
863 update_sit_entry(sbi, addr, -1);
864
865 /* add it into dirty seglist */
866 locate_dirty_segment(sbi, segno);
867
868 mutex_unlock(&sit_i->sentry_lock);
869}
870
6e2c64ad
JK		 871bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
872{
873 struct sit_info *sit_i = SIT_I(sbi);
874 unsigned int segno, offset;
875 struct seg_entry *se;
876 bool is_cp = false;
877
878 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
879 return true;
880
881 mutex_lock(&sit_i->sentry_lock);
882
883 segno = GET_SEGNO(sbi, blkaddr);
884 se = get_seg_entry(sbi, segno);
885 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
886
887 if (f2fs_test_bit(offset, se->ckpt_valid_map))
888 is_cp = true;
889
890 mutex_unlock(&sit_i->sentry_lock);
891
892 return is_cp;
893}
894
0a8165d7 895/*
351df4b2
JK		 896 * This function should be resided under the curseg_mutex lock
897 */
898static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 899 struct f2fs_summary *sum)
351df4b2
JK		 900{
901 struct curseg_info *curseg = CURSEG_I(sbi, type);
902 void *addr = curseg->sum_blk;
e79efe3b 903 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 904 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 905}
906
0a8165d7 907/*
351df4b2
JK		 908 * Calculate the number of current summary pages for writing
909 */
3fa06d7b 910int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 911{
351df4b2 912 int valid_sum_count = 0;
9a47938b 913 int i, sum_in_page;
351df4b2
JK		 914
915 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
916 if (sbi->ckpt->alloc_type[i] == SSR)
917 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 918 else {
919 if (for_ra)
920 valid_sum_count += le16_to_cpu(
921 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
922 else
923 valid_sum_count += curseg_blkoff(sbi, i);
924 }
351df4b2
JK		 925 }
926
09cbfeaf 927 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 928 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
929 if (valid_sum_count <= sum_in_page)
351df4b2 930 return 1;
9a47938b 931 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 932 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 933 return 2;
934 return 3;
935}
936
0a8165d7 937/*
351df4b2
JK		 938 * Caller should put this summary page
939 */
940struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
941{
942 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
943}
944
381722d2 945void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 946{
947 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 948 void *dst = page_address(page);
949
950 if (src)
09cbfeaf 951 memcpy(dst, src, PAGE_SIZE);
381722d2 952 else
09cbfeaf 953 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 954 set_page_dirty(page);
955 f2fs_put_page(page, 1);
956}
957
381722d2
CY		 958static void write_sum_page(struct f2fs_sb_info *sbi,
959 struct f2fs_summary_block *sum_blk, block_t blk_addr)
960{
961 update_meta_page(sbi, (void *)sum_blk, blk_addr);
962}
963
b7ad7512
CY		 964static void write_current_sum_page(struct f2fs_sb_info *sbi,
965 int type, block_t blk_addr)
966{
967 struct curseg_info *curseg = CURSEG_I(sbi, type);
968 struct page *page = grab_meta_page(sbi, blk_addr);
969 struct f2fs_summary_block *src = curseg->sum_blk;
970 struct f2fs_summary_block *dst;
971
972 dst = (struct f2fs_summary_block *)page_address(page);
973
974 mutex_lock(&curseg->curseg_mutex);
975
976 down_read(&curseg->journal_rwsem);
977 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
978 up_read(&curseg->journal_rwsem);
979
980 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
981 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
982
983 mutex_unlock(&curseg->curseg_mutex);
984
985 set_page_dirty(page);
986 f2fs_put_page(page, 1);
987}
988
60374688
JK		 989static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
990{
991 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 992 unsigned int segno = curseg->segno + 1;
60374688
JK		 993 struct free_segmap_info *free_i = FREE_I(sbi);
994
7cd8558b 995 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 996 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 997 return 0;
998}
999
0a8165d7 1000/*
351df4b2
JK		 1001 * Find a new segment from the free segments bitmap to right order
1002 * This function should be returned with success, otherwise BUG
1003 */
1004static void get_new_segment(struct f2fs_sb_info *sbi,
1005 unsigned int *newseg, bool new_sec, int dir)
1006{
1007 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1008 unsigned int segno, secno, zoneno;
7cd8558b 1009 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1010 unsigned int hint = *newseg / sbi->segs_per_sec;
1011 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1012 unsigned int left_start = hint;
1013 bool init = true;
1014 int go_left = 0;
1015 int i;
1016
1a118ccf 1017 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1018
1019 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1020 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1021 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1022 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1023 goto got_it;
1024 }
1025find_other_zone:
7cd8558b
JK		 1026 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1027 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1028 if (dir == ALLOC_RIGHT) {
1029 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1030 MAIN_SECS(sbi), 0);
1031 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1032 } else {
1033 go_left = 1;
1034 left_start = hint - 1;
1035 }
1036 }
1037 if (go_left == 0)
1038 goto skip_left;
1039
1040 while (test_bit(left_start, free_i->free_secmap)) {
1041 if (left_start > 0) {
1042 left_start--;
1043 continue;
1044 }
1045 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1046 MAIN_SECS(sbi), 0);
1047 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1048 break;
1049 }
1050 secno = left_start;
1051skip_left:
1052 hint = secno;
1053 segno = secno * sbi->segs_per_sec;
1054 zoneno = secno / sbi->secs_per_zone;
1055
1056 /* give up on finding another zone */
1057 if (!init)
1058 goto got_it;
1059 if (sbi->secs_per_zone == 1)
1060 goto got_it;
1061 if (zoneno == old_zoneno)
1062 goto got_it;
1063 if (dir == ALLOC_LEFT) {
1064 if (!go_left && zoneno + 1 >= total_zones)
1065 goto got_it;
1066 if (go_left && zoneno == 0)
1067 goto got_it;
1068 }
1069 for (i = 0; i < NR_CURSEG_TYPE; i++)
1070 if (CURSEG_I(sbi, i)->zone == zoneno)
1071 break;
1072
1073 if (i < NR_CURSEG_TYPE) {
1074 /* zone is in user, try another */
1075 if (go_left)
1076 hint = zoneno * sbi->secs_per_zone - 1;
1077 else if (zoneno + 1 >= total_zones)
1078 hint = 0;
1079 else
1080 hint = (zoneno + 1) * sbi->secs_per_zone;
1081 init = false;
1082 goto find_other_zone;
1083 }
1084got_it:
1085 /* set it as dirty segment in free segmap */
9850cf4a 1086 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1087 __set_inuse(sbi, segno);
1088 *newseg = segno;
1a118ccf 1089 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1090}
1091
1092static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1093{
1094 struct curseg_info *curseg = CURSEG_I(sbi, type);
1095 struct summary_footer *sum_footer;
1096
1097 curseg->segno = curseg->next_segno;
1098 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1099 curseg->next_blkoff = 0;
1100 curseg->next_segno = NULL_SEGNO;
1101
1102 sum_footer = &(curseg->sum_blk->footer);
1103 memset(sum_footer, 0, sizeof(struct summary_footer));
1104 if (IS_DATASEG(type))
1105 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1106 if (IS_NODESEG(type))
1107 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1108 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1109}
1110
0a8165d7 1111/*
351df4b2
JK		 1112 * Allocate a current working segment.
1113 * This function always allocates a free segment in LFS manner.
1114 */
1115static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1116{
1117 struct curseg_info *curseg = CURSEG_I(sbi, type);
1118 unsigned int segno = curseg->segno;
1119 int dir = ALLOC_LEFT;
1120
1121 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1122 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1123 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1124 dir = ALLOC_RIGHT;
1125
1126 if (test_opt(sbi, NOHEAP))
1127 dir = ALLOC_RIGHT;
1128
1129 get_new_segment(sbi, &segno, new_sec, dir);
1130 curseg->next_segno = segno;
1131 reset_curseg(sbi, type, 1);
1132 curseg->alloc_type = LFS;
1133}
1134
1135static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1136 struct curseg_info *seg, block_t start)
1137{
1138 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1139 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1140 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1141 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1142 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1143 int i, pos;
1144
1145 for (i = 0; i < entries; i++)
1146 target_map[i] = ckpt_map[i] | cur_map[i];
1147
1148 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1149
1150 seg->next_blkoff = pos;
351df4b2
JK		 1151}
1152
0a8165d7 1153/*
351df4b2
JK		 1154 * If a segment is written by LFS manner, next block offset is just obtained
1155 * by increasing the current block offset. However, if a segment is written by
1156 * SSR manner, next block offset obtained by calling __next_free_blkoff
1157 */
1158static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1159 struct curseg_info *seg)
1160{
1161 if (seg->alloc_type == SSR)
1162 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1163 else
1164 seg->next_blkoff++;
1165}
1166
0a8165d7 1167/*
e1c42045 1168 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1169 * manner, so it should recover the existing segment information of valid blocks
1170 */
1171static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1172{
1173 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1174 struct curseg_info *curseg = CURSEG_I(sbi, type);
1175 unsigned int new_segno = curseg->next_segno;
1176 struct f2fs_summary_block *sum_node;
1177 struct page *sum_page;
1178
1179 write_sum_page(sbi, curseg->sum_blk,
1180 GET_SUM_BLOCK(sbi, curseg->segno));
1181 __set_test_and_inuse(sbi, new_segno);
1182
1183 mutex_lock(&dirty_i->seglist_lock);
1184 __remove_dirty_segment(sbi, new_segno, PRE);
1185 __remove_dirty_segment(sbi, new_segno, DIRTY);
1186 mutex_unlock(&dirty_i->seglist_lock);
1187
1188 reset_curseg(sbi, type, 1);
1189 curseg->alloc_type = SSR;
1190 __next_free_blkoff(sbi, curseg, 0);
1191
1192 if (reuse) {
1193 sum_page = get_sum_page(sbi, new_segno);
1194 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1195 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1196 f2fs_put_page(sum_page, 1);
1197 }
1198}
1199
43727527
JK		 1200static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1201{
1202 struct curseg_info *curseg = CURSEG_I(sbi, type);
1203 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1204
1205 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1206 return v_ops->get_victim(sbi,
1207 &(curseg)->next_segno, BG_GC, type, SSR);
1208
1209 /* For data segments, let's do SSR more intensively */
1210 for (; type >= CURSEG_HOT_DATA; type--)
1211 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1212 BG_GC, type, SSR))
1213 return 1;
1214 return 0;
1215}
1216
351df4b2
JK		 1217/*
1218 * flush out current segment and replace it with new segment
1219 * This function should be returned with success, otherwise BUG
1220 */
1221static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1222 int type, bool force)
1223{
1224 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1225
7b405275 1226 if (force)
351df4b2 1227 new_curseg(sbi, type, true);
7b405275 1228 else if (type == CURSEG_WARM_NODE)
351df4b2 1229 new_curseg(sbi, type, false);
60374688
JK		 1230 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1231 new_curseg(sbi, type, false);
351df4b2
JK		 1232 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1233 change_curseg(sbi, type, true);
1234 else
1235 new_curseg(sbi, type, false);
dcdfff65
JK		 1236
1237 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1238}
1239
38aa0889
JK		 1240static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1241{
1242 struct curseg_info *curseg = CURSEG_I(sbi, type);
1243 unsigned int old_segno;
1244
1245 old_segno = curseg->segno;
1246 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1247 locate_dirty_segment(sbi, old_segno);
1248}
1249
351df4b2
JK		 1250void allocate_new_segments(struct f2fs_sb_info *sbi)
1251{
351df4b2
JK		 1252 int i;
1253
36abef4e
JK		 1254 if (test_opt(sbi, LFS))
1255 return;
1256
38aa0889
JK		 1257 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1258 __allocate_new_segments(sbi, i);
351df4b2
JK		 1259}
1260
1261static const struct segment_allocation default_salloc_ops = {
1262 .allocate_segment = allocate_segment_by_default,
1263};
1264
4b2fecc8
JK		 1265int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1266{
f7ef9b83
JK		 1267 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1268 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1269 unsigned int start_segno, end_segno;
1270 struct cp_control cpc;
c34f42e2 1271 int err = 0;
4b2fecc8 1272
836b5a63 1273 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1274 return -EINVAL;
1275
9bd27ae4 1276 cpc.trimmed = 0;
7cd8558b 1277 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1278 goto out;
1279
1280 /* start/end segment number in main_area */
7cd8558b
JK		 1281 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1282 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1283 GET_SEGNO(sbi, end);
4b2fecc8 1284 cpc.reason = CP_DISCARD;
836b5a63 1285 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1286
1287 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1288 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1289 cpc.trim_start = start_segno;
a66cdd98
JK		 1290
1291 if (sbi->discard_blks == 0)
1292 break;
1293 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1294 cpc.trim_end = end_segno;
1295 else
1296 cpc.trim_end = min_t(unsigned int,
1297 rounddown(start_segno +
bba681cb
JK		 1298 BATCHED_TRIM_SEGMENTS(sbi),
1299 sbi->segs_per_sec) - 1, end_segno);
1300
1301 mutex_lock(&sbi->gc_mutex);
c34f42e2 1302 err = write_checkpoint(sbi, &cpc);
bba681cb
JK		 1303 mutex_unlock(&sbi->gc_mutex);
1304 }
4b2fecc8 1305out:
f7ef9b83 1306 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1307 return err;
4b2fecc8
JK		 1308}
1309
351df4b2
JK		 1310static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1311{
1312 struct curseg_info *curseg = CURSEG_I(sbi, type);
1313 if (curseg->next_blkoff < sbi->blocks_per_seg)
1314 return true;
1315 return false;
1316}
1317
1318static int __get_segment_type_2(struct page *page, enum page_type p_type)
1319{
1320 if (p_type == DATA)
1321 return CURSEG_HOT_DATA;
1322 else
1323 return CURSEG_HOT_NODE;
1324}
1325
1326static int __get_segment_type_4(struct page *page, enum page_type p_type)
1327{
1328 if (p_type == DATA) {
1329 struct inode *inode = page->mapping->host;
1330
1331 if (S_ISDIR(inode->i_mode))
1332 return CURSEG_HOT_DATA;
1333 else
1334 return CURSEG_COLD_DATA;
1335 } else {
a344b9fd
JK		 1336 if (IS_DNODE(page) && is_cold_node(page))
1337 return CURSEG_WARM_NODE;
351df4b2
JK		 1338 else
1339 return CURSEG_COLD_NODE;
1340 }
1341}
1342
1343static int __get_segment_type_6(struct page *page, enum page_type p_type)
1344{
1345 if (p_type == DATA) {
1346 struct inode *inode = page->mapping->host;
1347
1348 if (S_ISDIR(inode->i_mode))
1349 return CURSEG_HOT_DATA;
354a3399 1350 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1351 return CURSEG_COLD_DATA;
1352 else
1353 return CURSEG_WARM_DATA;
1354 } else {
1355 if (IS_DNODE(page))
1356 return is_cold_node(page) ? CURSEG_WARM_NODE :
1357 CURSEG_HOT_NODE;
1358 else
1359 return CURSEG_COLD_NODE;
1360 }
1361}
1362
1363static int __get_segment_type(struct page *page, enum page_type p_type)
1364{
4081363f 1365 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1366 case 2:
1367 return __get_segment_type_2(page, p_type);
1368 case 4:
1369 return __get_segment_type_4(page, p_type);
351df4b2 1370 }
12a67146 1371 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1372 f2fs_bug_on(F2FS_P_SB(page),
1373 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1374 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1375}
1376
bfad7c2d
JK		 1377void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1378 block_t old_blkaddr, block_t *new_blkaddr,
1379 struct f2fs_summary *sum, int type)
351df4b2
JK		 1380{
1381 struct sit_info *sit_i = SIT_I(sbi);
1382 struct curseg_info *curseg;
38aa0889
JK		 1383 bool direct_io = (type == CURSEG_DIRECT_IO);
1384
1385 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1386
351df4b2
JK		 1387 curseg = CURSEG_I(sbi, type);
1388
1389 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1390 mutex_lock(&sit_i->sentry_lock);
351df4b2 1391
38aa0889 1392 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1393 if (direct_io && curseg->next_blkoff &&
1394 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1395 __allocate_new_segments(sbi, type);
1396
351df4b2 1397 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1398
1399 /*
1400 * __add_sum_entry should be resided under the curseg_mutex
1401 * because, this function updates a summary entry in the
1402 * current summary block.
1403 */
e79efe3b 1404 __add_sum_entry(sbi, type, sum);
351df4b2 1405
351df4b2 1406 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1407
1408 stat_inc_block_count(sbi, curseg);
351df4b2 1409
5e443818
JK		 1410 if (!__has_curseg_space(sbi, type))
1411 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1412 /*
1413 * SIT information should be updated before segment allocation,
1414 * since SSR needs latest valid block information.
1415 */
1416 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1417
351df4b2
JK		 1418 mutex_unlock(&sit_i->sentry_lock);
1419
bfad7c2d 1420 if (page && IS_NODESEG(type))
351df4b2
JK		 1421 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1422
bfad7c2d
JK		 1423 mutex_unlock(&curseg->curseg_mutex);
1424}
1425
05ca3632 1426static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1427{
05ca3632 1428 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1429
7dfeaa32
JK		 1430 if (fio->type == NODE || fio->type == DATA)
1431 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1432
7a9d7548
CY		 1433 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1434 &fio->new_blkaddr, sum, type);
bfad7c2d 1435
351df4b2 1436 /* writeout dirty page into bdev */
05ca3632 1437 f2fs_submit_page_mbio(fio);
7dfeaa32
JK		 1438
1439 if (fio->type == NODE || fio->type == DATA)
1440 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1441}
1442
577e3495 1443void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1444{
458e6197 1445 struct f2fs_io_info fio = {
05ca3632 1446 .sbi = sbi,
458e6197 1447 .type = META,
04d328de
MC		 1448 .op = REQ_OP_WRITE,
1449 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1450 .old_blkaddr = page->index,
1451 .new_blkaddr = page->index,
05ca3632 1452 .page = page,
4375a336 1453 .encrypted_page = NULL,
458e6197
JK		 1454 };
1455
2b947003 1456 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1457 fio.op_flags &= ~REQ_META;
2b947003 1458
351df4b2 1459 set_page_writeback(page);
05ca3632 1460 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1461}
1462
05ca3632 1463void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1464{
1465 struct f2fs_summary sum;
05ca3632 1466
351df4b2 1467 set_summary(&sum, nid, 0, 0);
05ca3632 1468 do_write_page(&sum, fio);
351df4b2
JK		 1469}
1470
05ca3632 1471void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1472{
05ca3632 1473 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1474 struct f2fs_summary sum;
1475 struct node_info ni;
1476
9850cf4a 1477 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1478 get_node_info(sbi, dn->nid, &ni);
1479 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1480 do_write_page(&sum, fio);
f28b3434 1481 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1482}
1483
05ca3632 1484void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1485{
7a9d7548 1486 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1487 stat_inc_inplace_blocks(fio->sbi);
1488 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1489}
1490
4356e48e 1491void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1492 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1493 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1494{
1495 struct sit_info *sit_i = SIT_I(sbi);
1496 struct curseg_info *curseg;
1497 unsigned int segno, old_cursegno;
1498 struct seg_entry *se;
1499 int type;
19f106bc 1500 unsigned short old_blkoff;
351df4b2
JK		 1501
1502 segno = GET_SEGNO(sbi, new_blkaddr);
1503 se = get_seg_entry(sbi, segno);
1504 type = se->type;
1505
19f106bc
CY		 1506 if (!recover_curseg) {
1507 /* for recovery flow */
1508 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1509 if (old_blkaddr == NULL_ADDR)
1510 type = CURSEG_COLD_DATA;
1511 else
1512 type = CURSEG_WARM_DATA;
1513 }
1514 } else {
1515 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1516 type = CURSEG_WARM_DATA;
1517 }
19f106bc 1518
351df4b2
JK		 1519 curseg = CURSEG_I(sbi, type);
1520
1521 mutex_lock(&curseg->curseg_mutex);
1522 mutex_lock(&sit_i->sentry_lock);
1523
1524 old_cursegno = curseg->segno;
19f106bc 1525 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1526
1527 /* change the current segment */
1528 if (segno != curseg->segno) {
1529 curseg->next_segno = segno;
1530 change_curseg(sbi, type, true);
1531 }
1532
491c0854 1533 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1534 __add_sum_entry(sbi, type, sum);
351df4b2 1535
28bc106b 1536 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1537 update_sit_entry(sbi, new_blkaddr, 1);
1538 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1539 update_sit_entry(sbi, old_blkaddr, -1);
1540
1541 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1542 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1543
351df4b2 1544 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1545
19f106bc
CY		 1546 if (recover_curseg) {
1547 if (old_cursegno != curseg->segno) {
1548 curseg->next_segno = old_cursegno;
1549 change_curseg(sbi, type, true);
1550 }
1551 curseg->next_blkoff = old_blkoff;
1552 }
1553
351df4b2
JK		 1554 mutex_unlock(&sit_i->sentry_lock);
1555 mutex_unlock(&curseg->curseg_mutex);
1556}
1557
528e3459
CY		 1558void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1559 block_t old_addr, block_t new_addr,
28bc106b
CY		 1560 unsigned char version, bool recover_curseg,
1561 bool recover_newaddr)
528e3459
CY		 1562{
1563 struct f2fs_summary sum;
1564
1565 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1566
28bc106b
CY		 1567 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1568 recover_curseg, recover_newaddr);
528e3459 1569
f28b3434 1570 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1571}
1572
93dfe2ac 1573void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1574 enum page_type type, bool ordered)
93dfe2ac 1575{
93dfe2ac 1576 if (PageWriteback(page)) {
4081363f
JK		 1577 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1578
0c3a5797 1579 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1580 if (ordered)
1581 wait_on_page_writeback(page);
1582 else
1583 wait_for_stable_page(page);
93dfe2ac
JK		 1584 }
1585}
1586
08b39fbd
CY		 1587void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1588 block_t blkaddr)
1589{
1590 struct page *cpage;
1591
1592 if (blkaddr == NEW_ADDR)
1593 return;
1594
1595 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1596
1597 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1598 if (cpage) {
fec1d657 1599 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1600 f2fs_put_page(cpage, 1);
1601 }
1602}
1603
351df4b2
JK		 1604static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1605{
1606 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1607 struct curseg_info *seg_i;
1608 unsigned char *kaddr;
1609 struct page *page;
1610 block_t start;
1611 int i, j, offset;
1612
1613 start = start_sum_block(sbi);
1614
1615 page = get_meta_page(sbi, start++);
1616 kaddr = (unsigned char *)page_address(page);
1617
1618 /* Step 1: restore nat cache */
1619 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1620 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1621
1622 /* Step 2: restore sit cache */
1623 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1624 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1625 offset = 2 * SUM_JOURNAL_SIZE;
1626
1627 /* Step 3: restore summary entries */
1628 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1629 unsigned short blk_off;
1630 unsigned int segno;
1631
1632 seg_i = CURSEG_I(sbi, i);
1633 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1634 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1635 seg_i->next_segno = segno;
1636 reset_curseg(sbi, i, 0);
1637 seg_i->alloc_type = ckpt->alloc_type[i];
1638 seg_i->next_blkoff = blk_off;
1639
1640 if (seg_i->alloc_type == SSR)
1641 blk_off = sbi->blocks_per_seg;
1642
1643 for (j = 0; j < blk_off; j++) {
1644 struct f2fs_summary *s;
1645 s = (struct f2fs_summary *)(kaddr + offset);
1646 seg_i->sum_blk->entries[j] = *s;
1647 offset += SUMMARY_SIZE;
09cbfeaf 1648 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1649 SUM_FOOTER_SIZE)
1650 continue;
1651
1652 f2fs_put_page(page, 1);
1653 page = NULL;
1654
1655 page = get_meta_page(sbi, start++);
1656 kaddr = (unsigned char *)page_address(page);
1657 offset = 0;
1658 }
1659 }
1660 f2fs_put_page(page, 1);
1661 return 0;
1662}
1663
1664static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1665{
1666 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1667 struct f2fs_summary_block *sum;
1668 struct curseg_info *curseg;
1669 struct page *new;
1670 unsigned short blk_off;
1671 unsigned int segno = 0;
1672 block_t blk_addr = 0;
1673
1674 /* get segment number and block addr */
1675 if (IS_DATASEG(type)) {
1676 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1677 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1678 CURSEG_HOT_DATA]);
119ee914 1679 if (__exist_node_summaries(sbi))
351df4b2
JK		 1680 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1681 else
1682 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1683 } else {
1684 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1685 CURSEG_HOT_NODE]);
1686 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1687 CURSEG_HOT_NODE]);
119ee914 1688 if (__exist_node_summaries(sbi))
351df4b2
JK		 1689 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1690 type - CURSEG_HOT_NODE);
1691 else
1692 blk_addr = GET_SUM_BLOCK(sbi, segno);
1693 }
1694
1695 new = get_meta_page(sbi, blk_addr);
1696 sum = (struct f2fs_summary_block *)page_address(new);
1697
1698 if (IS_NODESEG(type)) {
119ee914 1699 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1700 struct f2fs_summary *ns = &sum->entries[0];
1701 int i;
1702 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1703 ns->version = 0;
1704 ns->ofs_in_node = 0;
1705 }
1706 } else {
d653788a
GZ		 1707 int err;
1708
1709 err = restore_node_summary(sbi, segno, sum);
1710 if (err) {
351df4b2 1711 f2fs_put_page(new, 1);
d653788a 1712 return err;
351df4b2
JK		 1713 }
1714 }
1715 }
1716
1717 /* set uncompleted segment to curseg */
1718 curseg = CURSEG_I(sbi, type);
1719 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1720
1721 /* update journal info */
1722 down_write(&curseg->journal_rwsem);
1723 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1724 up_write(&curseg->journal_rwsem);
1725
1726 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1727 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1728 curseg->next_segno = segno;
1729 reset_curseg(sbi, type, 0);
1730 curseg->alloc_type = ckpt->alloc_type[type];
1731 curseg->next_blkoff = blk_off;
1732 mutex_unlock(&curseg->curseg_mutex);
1733 f2fs_put_page(new, 1);
1734 return 0;
1735}
1736
1737static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1738{
1739 int type = CURSEG_HOT_DATA;
e4fc5fbf 1740 int err;
351df4b2 1741
25ca923b 1742 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1743 int npages = npages_for_summary_flush(sbi, true);
1744
1745 if (npages >= 2)
1746 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1747 META_CP, true);
3fa06d7b 1748
351df4b2
JK		 1749 /* restore for compacted data summary */
1750 if (read_compacted_summaries(sbi))
1751 return -EINVAL;
1752 type = CURSEG_HOT_NODE;
1753 }
1754
119ee914 1755 if (__exist_node_summaries(sbi))
3fa06d7b 1756 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1757 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1758
e4fc5fbf
CY		 1759 for (; type <= CURSEG_COLD_NODE; type++) {
1760 err = read_normal_summaries(sbi, type);
1761 if (err)
1762 return err;
1763 }
1764
351df4b2
JK		 1765 return 0;
1766}
1767
1768static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1769{
1770 struct page *page;
1771 unsigned char *kaddr;
1772 struct f2fs_summary *summary;
1773 struct curseg_info *seg_i;
1774 int written_size = 0;
1775 int i, j;
1776
1777 page = grab_meta_page(sbi, blkaddr++);
1778 kaddr = (unsigned char *)page_address(page);
1779
1780 /* Step 1: write nat cache */
1781 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1782 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1783 written_size += SUM_JOURNAL_SIZE;
1784
1785 /* Step 2: write sit cache */
1786 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1787 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1788 written_size += SUM_JOURNAL_SIZE;
1789
351df4b2
JK		 1790 /* Step 3: write summary entries */
1791 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1792 unsigned short blkoff;
1793 seg_i = CURSEG_I(sbi, i);
1794 if (sbi->ckpt->alloc_type[i] == SSR)
1795 blkoff = sbi->blocks_per_seg;
1796 else
1797 blkoff = curseg_blkoff(sbi, i);
1798
1799 for (j = 0; j < blkoff; j++) {
1800 if (!page) {
1801 page = grab_meta_page(sbi, blkaddr++);
1802 kaddr = (unsigned char *)page_address(page);
1803 written_size = 0;
1804 }
1805 summary = (struct f2fs_summary *)(kaddr + written_size);
1806 *summary = seg_i->sum_blk->entries[j];
1807 written_size += SUMMARY_SIZE;
351df4b2 1808
09cbfeaf 1809 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1810 SUM_FOOTER_SIZE)
1811 continue;
1812
e8d61a74 1813 set_page_dirty(page);
351df4b2
JK		 1814 f2fs_put_page(page, 1);
1815 page = NULL;
1816 }
1817 }
e8d61a74
CY		 1818 if (page) {
1819 set_page_dirty(page);
351df4b2 1820 f2fs_put_page(page, 1);
e8d61a74 1821 }
351df4b2
JK		 1822}
1823
1824static void write_normal_summaries(struct f2fs_sb_info *sbi,
1825 block_t blkaddr, int type)
1826{
1827 int i, end;
1828 if (IS_DATASEG(type))
1829 end = type + NR_CURSEG_DATA_TYPE;
1830 else
1831 end = type + NR_CURSEG_NODE_TYPE;
1832
b7ad7512
CY		 1833 for (i = type; i < end; i++)
1834 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1835}
1836
1837void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1838{
25ca923b 1839 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1840 write_compacted_summaries(sbi, start_blk);
1841 else
1842 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1843}
1844
1845void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1846{
119ee914 1847 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1848}
1849
dfc08a12 1850int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1851 unsigned int val, int alloc)
1852{
1853 int i;
1854
1855 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1856 for (i = 0; i < nats_in_cursum(journal); i++) {
1857 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1858 return i;
1859 }
dfc08a12
CY		 1860 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1861 return update_nats_in_cursum(journal, 1);
351df4b2 1862 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1863 for (i = 0; i < sits_in_cursum(journal); i++)
1864 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1865 return i;
dfc08a12
CY		 1866 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1867 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1868 }
1869 return -1;
1870}
1871
1872static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1873 unsigned int segno)
1874{
2cc22186 1875 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1876}
1877
1878static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1879 unsigned int start)
1880{
1881 struct sit_info *sit_i = SIT_I(sbi);
1882 struct page *src_page, *dst_page;
1883 pgoff_t src_off, dst_off;
1884 void *src_addr, *dst_addr;
1885
1886 src_off = current_sit_addr(sbi, start);
1887 dst_off = next_sit_addr(sbi, src_off);
1888
1889 /* get current sit block page without lock */
1890 src_page = get_meta_page(sbi, src_off);
1891 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1892 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1893
1894 src_addr = page_address(src_page);
1895 dst_addr = page_address(dst_page);
09cbfeaf 1896 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1897
1898 set_page_dirty(dst_page);
1899 f2fs_put_page(src_page, 1);
1900
1901 set_to_next_sit(sit_i, start);
1902
1903 return dst_page;
1904}
1905
184a5cd2
CY		 1906static struct sit_entry_set *grab_sit_entry_set(void)
1907{
1908 struct sit_entry_set *ses =
80c54505 1909 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1910
1911 ses->entry_cnt = 0;
1912 INIT_LIST_HEAD(&ses->set_list);
1913 return ses;
1914}
1915
1916static void release_sit_entry_set(struct sit_entry_set *ses)
1917{
1918 list_del(&ses->set_list);
1919 kmem_cache_free(sit_entry_set_slab, ses);
1920}
1921
1922static void adjust_sit_entry_set(struct sit_entry_set *ses,
1923 struct list_head *head)
1924{
1925 struct sit_entry_set *next = ses;
1926
1927 if (list_is_last(&ses->set_list, head))
1928 return;
1929
1930 list_for_each_entry_continue(next, head, set_list)
1931 if (ses->entry_cnt <= next->entry_cnt)
1932 break;
1933
1934 list_move_tail(&ses->set_list, &next->set_list);
1935}
1936
1937static void add_sit_entry(unsigned int segno, struct list_head *head)
1938{
1939 struct sit_entry_set *ses;
1940 unsigned int start_segno = START_SEGNO(segno);
1941
1942 list_for_each_entry(ses, head, set_list) {
1943 if (ses->start_segno == start_segno) {
1944 ses->entry_cnt++;
1945 adjust_sit_entry_set(ses, head);
1946 return;
1947 }
1948 }
1949
1950 ses = grab_sit_entry_set();
1951
1952 ses->start_segno = start_segno;
1953 ses->entry_cnt++;
1954 list_add(&ses->set_list, head);
1955}
1956
1957static void add_sits_in_set(struct f2fs_sb_info *sbi)
1958{
1959 struct f2fs_sm_info *sm_info = SM_I(sbi);
1960 struct list_head *set_list = &sm_info->sit_entry_set;
1961 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 1962 unsigned int segno;
1963
7cd8558b 1964 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 1965 add_sit_entry(segno, set_list);
1966}
1967
1968static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 1969{
1970 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1971 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 1972 int i;
1973
b7ad7512 1974 down_write(&curseg->journal_rwsem);
dfc08a12 1975 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 1976 unsigned int segno;
1977 bool dirtied;
1978
dfc08a12 1979 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 1980 dirtied = __mark_sit_entry_dirty(sbi, segno);
1981
1982 if (!dirtied)
1983 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 1984 }
dfc08a12 1985 update_sits_in_cursum(journal, -i);
b7ad7512 1986 up_write(&curseg->journal_rwsem);
351df4b2
JK		 1987}
1988
0a8165d7 1989/*
351df4b2
JK		 1990 * CP calls this function, which flushes SIT entries including sit_journal,
1991 * and moves prefree segs to free segs.
1992 */
4b2fecc8 1993void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 1994{
1995 struct sit_info *sit_i = SIT_I(sbi);
1996 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1997 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1998 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 1999 struct sit_entry_set *ses, *tmp;
2000 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2001 bool to_journal = true;
4b2fecc8 2002 struct seg_entry *se;
351df4b2 2003
351df4b2
JK		 2004 mutex_lock(&sit_i->sentry_lock);
2005
2b11a74b
WL		 2006 if (!sit_i->dirty_sentries)
2007 goto out;
2008
351df4b2 2009 /*
184a5cd2
CY		 2010 * add and account sit entries of dirty bitmap in sit entry
2011 * set temporarily
351df4b2 2012 */
184a5cd2 2013 add_sits_in_set(sbi);
351df4b2 2014
184a5cd2
CY		 2015 /*
2016 * if there are no enough space in journal to store dirty sit
2017 * entries, remove all entries from journal and add and account
2018 * them in sit entry set.
2019 */
dfc08a12 2020 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2021 remove_sits_in_journal(sbi);
b2955550 2022
184a5cd2
CY		 2023 /*
2024 * there are two steps to flush sit entries:
2025 * #1, flush sit entries to journal in current cold data summary block.
2026 * #2, flush sit entries to sit page.
2027 */
2028 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2029 struct page *page = NULL;
184a5cd2
CY		 2030 struct f2fs_sit_block *raw_sit = NULL;
2031 unsigned int start_segno = ses->start_segno;
2032 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2033 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2034 unsigned int segno = start_segno;
2035
2036 if (to_journal &&
dfc08a12 2037 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2038 to_journal = false;
2039
b7ad7512
CY		 2040 if (to_journal) {
2041 down_write(&curseg->journal_rwsem);
2042 } else {
184a5cd2
CY		 2043 page = get_next_sit_page(sbi, start_segno);
2044 raw_sit = page_address(page);
351df4b2 2045 }
351df4b2 2046
184a5cd2
CY		 2047 /* flush dirty sit entries in region of current sit set */
2048 for_each_set_bit_from(segno, bitmap, end) {
2049 int offset, sit_offset;
4b2fecc8
JK		 2050
2051 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2052
2053 /* add discard candidates */
d7bc2484 2054 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2055 cpc->trim_start = segno;
2056 add_discard_addrs(sbi, cpc);
2057 }
184a5cd2
CY		 2058
2059 if (to_journal) {
dfc08a12 2060 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2061 SIT_JOURNAL, segno, 1);
2062 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2063 segno_in_journal(journal, offset) =
184a5cd2
CY		 2064 cpu_to_le32(segno);
2065 seg_info_to_raw_sit(se,
dfc08a12 2066 &sit_in_journal(journal, offset));
184a5cd2
CY		 2067 } else {
2068 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2069 seg_info_to_raw_sit(se,
2070 &raw_sit->entries[sit_offset]);
2071 }
351df4b2 2072
184a5cd2
CY		 2073 __clear_bit(segno, bitmap);
2074 sit_i->dirty_sentries--;
2075 ses->entry_cnt--;
351df4b2
JK		 2076 }
2077
b7ad7512
CY		 2078 if (to_journal)
2079 up_write(&curseg->journal_rwsem);
2080 else
184a5cd2
CY		 2081 f2fs_put_page(page, 1);
2082
2083 f2fs_bug_on(sbi, ses->entry_cnt);
2084 release_sit_entry_set(ses);
351df4b2 2085 }
184a5cd2
CY		 2086
2087 f2fs_bug_on(sbi, !list_empty(head));
2088 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2089out:
4b2fecc8
JK		 2090 if (cpc->reason == CP_DISCARD) {
2091 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2092 add_discard_addrs(sbi, cpc);
2093 }
351df4b2 2094 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2095
351df4b2
JK		 2096 set_prefree_as_free_segments(sbi);
2097}
2098
2099static int build_sit_info(struct f2fs_sb_info *sbi)
2100{
2101 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2102 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2103 struct sit_info *sit_i;
2104 unsigned int sit_segs, start;
2105 char *src_bitmap, *dst_bitmap;
2106 unsigned int bitmap_size;
2107
2108 /* allocate memory for SIT information */
2109 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2110 if (!sit_i)
2111 return -ENOMEM;
2112
2113 SM_I(sbi)->sit_info = sit_i;
2114
39307a8e
JK		 2115 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2116 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2117 if (!sit_i->sentries)
2118 return -ENOMEM;
2119
7cd8558b 2120 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2121 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2122 if (!sit_i->dirty_sentries_bitmap)
2123 return -ENOMEM;
2124
7cd8558b 2125 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2126 sit_i->sentries[start].cur_valid_map
2127 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2128 sit_i->sentries[start].ckpt_valid_map
2129 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98
JK		 2130 sit_i->sentries[start].discard_map
2131 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2132 if (!sit_i->sentries[start].cur_valid_map ||
2133 !sit_i->sentries[start].ckpt_valid_map ||
2134 !sit_i->sentries[start].discard_map)
351df4b2
JK		 2135 return -ENOMEM;
2136 }
2137
60a3b782
JK		 2138 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2139 if (!sit_i->tmp_map)
2140 return -ENOMEM;
2141
351df4b2 2142 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2143 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2144 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2145 if (!sit_i->sec_entries)
2146 return -ENOMEM;
2147 }
2148
2149 /* get information related with SIT */
2150 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2151
2152 /* setup SIT bitmap from ckeckpoint pack */
2153 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2154 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2155
79b5793b 2156 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2157 if (!dst_bitmap)
2158 return -ENOMEM;
351df4b2
JK		 2159
2160 /* init SIT information */
2161 sit_i->s_ops = &default_salloc_ops;
2162
2163 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2164 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2165 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2166 sit_i->sit_bitmap = dst_bitmap;
2167 sit_i->bitmap_size = bitmap_size;
2168 sit_i->dirty_sentries = 0;
2169 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2170 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2171 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2172 mutex_init(&sit_i->sentry_lock);
2173 return 0;
2174}
2175
2176static int build_free_segmap(struct f2fs_sb_info *sbi)
2177{
351df4b2
JK		 2178 struct free_segmap_info *free_i;
2179 unsigned int bitmap_size, sec_bitmap_size;
2180
2181 /* allocate memory for free segmap information */
2182 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2183 if (!free_i)
2184 return -ENOMEM;
2185
2186 SM_I(sbi)->free_info = free_i;
2187
7cd8558b 2188 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2189 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2190 if (!free_i->free_segmap)
2191 return -ENOMEM;
2192
7cd8558b 2193 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2194 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2195 if (!free_i->free_secmap)
2196 return -ENOMEM;
2197
2198 /* set all segments as dirty temporarily */
2199 memset(free_i->free_segmap, 0xff, bitmap_size);
2200 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2201
2202 /* init free segmap information */
7cd8558b 2203 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2204 free_i->free_segments = 0;
2205 free_i->free_sections = 0;
1a118ccf 2206 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2207 return 0;
2208}
2209
2210static int build_curseg(struct f2fs_sb_info *sbi)
2211{
1042d60f 2212 struct curseg_info *array;
351df4b2
JK		 2213 int i;
2214
b434babf 2215 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2216 if (!array)
2217 return -ENOMEM;
2218
2219 SM_I(sbi)->curseg_array = array;
2220
2221 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2222 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2223 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2224 if (!array[i].sum_blk)
2225 return -ENOMEM;
b7ad7512
CY		 2226 init_rwsem(&array[i].journal_rwsem);
2227 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2228 GFP_KERNEL);
2229 if (!array[i].journal)
2230 return -ENOMEM;
351df4b2
JK		 2231 array[i].segno = NULL_SEGNO;
2232 array[i].next_blkoff = 0;
2233 }
2234 return restore_curseg_summaries(sbi);
2235}
2236
2237static void build_sit_entries(struct f2fs_sb_info *sbi)
2238{
2239 struct sit_info *sit_i = SIT_I(sbi);
2240 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2241 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2242 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2243 unsigned int i, start, end;
2244 unsigned int readed, start_blk = 0;
e9f5b8b8 2245 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2246
74de593a 2247 do {
26879fb1 2248 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2249
2250 start = start_blk * sit_i->sents_per_block;
2251 end = (start_blk + readed) * sit_i->sents_per_block;
2252
7cd8558b 2253 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2254 struct seg_entry *se = &sit_i->sentries[start];
2255 struct f2fs_sit_block *sit_blk;
2256 struct f2fs_sit_entry sit;
2257 struct page *page;
2258
b7ad7512 2259 down_read(&curseg->journal_rwsem);
dfc08a12
CY		 2260 for (i = 0; i < sits_in_cursum(journal); i++) {
2261 if (le32_to_cpu(segno_in_journal(journal, i))
6c311ec6 2262 == start) {
dfc08a12 2263 sit = sit_in_journal(journal, i);
b7ad7512 2264 up_read(&curseg->journal_rwsem);
74de593a
CY		 2265 goto got_it;
2266 }
351df4b2 2267 }
b7ad7512 2268 up_read(&curseg->journal_rwsem);
74de593a
CY		 2269
2270 page = get_current_sit_page(sbi, start);
2271 sit_blk = (struct f2fs_sit_block *)page_address(page);
2272 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2273 f2fs_put_page(page, 1);
351df4b2 2274got_it:
74de593a
CY		 2275 check_block_count(sbi, start, &sit);
2276 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2277
2278 /* build discard map only one time */
2279 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2280 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2281
74de593a
CY		 2282 if (sbi->segs_per_sec > 1) {
2283 struct sec_entry *e = get_sec_entry(sbi, start);
2284 e->valid_blocks += se->valid_blocks;
2285 }
351df4b2 2286 }
74de593a
CY		 2287 start_blk += readed;
2288 } while (start_blk < sit_blk_cnt);
351df4b2
JK		 2289}
2290
2291static void init_free_segmap(struct f2fs_sb_info *sbi)
2292{
2293 unsigned int start;
2294 int type;
2295
7cd8558b 2296 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2297 struct seg_entry *sentry = get_seg_entry(sbi, start);
2298 if (!sentry->valid_blocks)
2299 __set_free(sbi, start);
2300 }
2301
2302 /* set use the current segments */
2303 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2304 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2305 __set_test_and_inuse(sbi, curseg_t->segno);
2306 }
2307}
2308
2309static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2310{
2311 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2312 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2313 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2314 unsigned short valid_blocks;
2315
8736fbf0 2316 while (1) {
351df4b2 2317 /* find dirty segment based on free segmap */
7cd8558b
JK		 2318 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2319 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2320 break;
2321 offset = segno + 1;
2322 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2323 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2324 continue;
ec325b52
JK		 2325 if (valid_blocks > sbi->blocks_per_seg) {
2326 f2fs_bug_on(sbi, 1);
2327 continue;
2328 }
351df4b2
JK		 2329 mutex_lock(&dirty_i->seglist_lock);
2330 __locate_dirty_segment(sbi, segno, DIRTY);
2331 mutex_unlock(&dirty_i->seglist_lock);
2332 }
2333}
2334
5ec4e49f 2335static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2336{
2337 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2338 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2339
39307a8e 2340 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2341 if (!dirty_i->victim_secmap)
351df4b2
JK		 2342 return -ENOMEM;
2343 return 0;
2344}
2345
2346static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2347{
2348 struct dirty_seglist_info *dirty_i;
2349 unsigned int bitmap_size, i;
2350
2351 /* allocate memory for dirty segments list information */
2352 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2353 if (!dirty_i)
2354 return -ENOMEM;
2355
2356 SM_I(sbi)->dirty_info = dirty_i;
2357 mutex_init(&dirty_i->seglist_lock);
2358
7cd8558b 2359 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2360
2361 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2362 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2363 if (!dirty_i->dirty_segmap[i])
2364 return -ENOMEM;
2365 }
2366
2367 init_dirty_segmap(sbi);
5ec4e49f 2368 return init_victim_secmap(sbi);
351df4b2
JK		 2369}
2370
0a8165d7 2371/*
351df4b2
JK		 2372 * Update min, max modified time for cost-benefit GC algorithm
2373 */
2374static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2375{
2376 struct sit_info *sit_i = SIT_I(sbi);
2377 unsigned int segno;
2378
2379 mutex_lock(&sit_i->sentry_lock);
2380
2381 sit_i->min_mtime = LLONG_MAX;
2382
7cd8558b 2383 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2384 unsigned int i;
2385 unsigned long long mtime = 0;
2386
2387 for (i = 0; i < sbi->segs_per_sec; i++)
2388 mtime += get_seg_entry(sbi, segno + i)->mtime;
2389
2390 mtime = div_u64(mtime, sbi->segs_per_sec);
2391
2392 if (sit_i->min_mtime > mtime)
2393 sit_i->min_mtime = mtime;
2394 }
2395 sit_i->max_mtime = get_mtime(sbi);
2396 mutex_unlock(&sit_i->sentry_lock);
2397}
2398
2399int build_segment_manager(struct f2fs_sb_info *sbi)
2400{
2401 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2402 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2403 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2404 int err;
2405
2406 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2407 if (!sm_info)
2408 return -ENOMEM;
2409
2410 /* init sm info */
2411 sbi->sm_info = sm_info;
351df4b2
JK		 2412 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2413 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2414 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2415 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2416 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2417 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2418 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2419 sm_info->rec_prefree_segments = sm_info->main_segments *
2420 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2421 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2422 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2423
52763a4b
JK		 2424 if (!test_opt(sbi, LFS))
2425 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2426 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2427 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2428
7fd9e544
JK		 2429 INIT_LIST_HEAD(&sm_info->discard_list);
2430 sm_info->nr_discards = 0;
2431 sm_info->max_discards = 0;
2432
bba681cb
JK		 2433 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2434
184a5cd2
CY		 2435 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2436
b270ad6f 2437 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2438 err = create_flush_cmd_control(sbi);
2439 if (err)
a688b9d9 2440 return err;
6b4afdd7
JK		 2441 }
2442
351df4b2
JK		 2443 err = build_sit_info(sbi);
2444 if (err)
2445 return err;
2446 err = build_free_segmap(sbi);
2447 if (err)
2448 return err;
2449 err = build_curseg(sbi);
2450 if (err)
2451 return err;
2452
2453 /* reinit free segmap based on SIT */
2454 build_sit_entries(sbi);
2455
2456 init_free_segmap(sbi);
2457 err = build_dirty_segmap(sbi);
2458 if (err)
2459 return err;
2460
2461 init_min_max_mtime(sbi);
2462 return 0;
2463}
2464
2465static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2466 enum dirty_type dirty_type)
2467{
2468 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2469
2470 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2471 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2472 dirty_i->nr_dirty[dirty_type] = 0;
2473 mutex_unlock(&dirty_i->seglist_lock);
2474}
2475
5ec4e49f 2476static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2477{
2478 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2479 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2480}
2481
2482static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2483{
2484 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2485 int i;
2486
2487 if (!dirty_i)
2488 return;
2489
2490 /* discard pre-free/dirty segments list */
2491 for (i = 0; i < NR_DIRTY_TYPE; i++)
2492 discard_dirty_segmap(sbi, i);
2493
5ec4e49f 2494 destroy_victim_secmap(sbi);
351df4b2
JK		 2495 SM_I(sbi)->dirty_info = NULL;
2496 kfree(dirty_i);
2497}
2498
2499static void destroy_curseg(struct f2fs_sb_info *sbi)
2500{
2501 struct curseg_info *array = SM_I(sbi)->curseg_array;
2502 int i;
2503
2504 if (!array)
2505 return;
2506 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2507 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2508 kfree(array[i].sum_blk);
b7ad7512
CY		 2509 kfree(array[i].journal);
2510 }
351df4b2
JK		 2511 kfree(array);
2512}
2513
2514static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2515{
2516 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2517 if (!free_i)
2518 return;
2519 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2520 kvfree(free_i->free_segmap);
2521 kvfree(free_i->free_secmap);
351df4b2
JK		 2522 kfree(free_i);
2523}
2524
2525static void destroy_sit_info(struct f2fs_sb_info *sbi)
2526{
2527 struct sit_info *sit_i = SIT_I(sbi);
2528 unsigned int start;
2529
2530 if (!sit_i)
2531 return;
2532
2533 if (sit_i->sentries) {
7cd8558b 2534 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2535 kfree(sit_i->sentries[start].cur_valid_map);
2536 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2537 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2538 }
2539 }
60a3b782
JK		 2540 kfree(sit_i->tmp_map);
2541
39307a8e
JK		 2542 kvfree(sit_i->sentries);
2543 kvfree(sit_i->sec_entries);
2544 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2545
2546 SM_I(sbi)->sit_info = NULL;
2547 kfree(sit_i->sit_bitmap);
2548 kfree(sit_i);
2549}
2550
2551void destroy_segment_manager(struct f2fs_sb_info *sbi)
2552{
2553 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2554
3b03f724
CY		 2555 if (!sm_info)
2556 return;
2163d198 2557 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2558 destroy_dirty_segmap(sbi);
2559 destroy_curseg(sbi);
2560 destroy_free_segmap(sbi);
2561 destroy_sit_info(sbi);
2562 sbi->sm_info = NULL;
2563 kfree(sm_info);
2564}
7fd9e544
JK		 2565
2566int __init create_segment_manager_caches(void)
2567{
2568 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2569 sizeof(struct discard_entry));
7fd9e544 2570 if (!discard_entry_slab)
184a5cd2
CY		 2571 goto fail;
2572
2573 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2574 sizeof(struct sit_entry_set));
184a5cd2
CY		 2575 if (!sit_entry_set_slab)
2576 goto destory_discard_entry;
88b88a66
JK		 2577
2578 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2579 sizeof(struct inmem_pages));
2580 if (!inmem_entry_slab)
2581 goto destroy_sit_entry_set;
7fd9e544 2582 return 0;
184a5cd2 2583
88b88a66
JK		 2584destroy_sit_entry_set:
2585 kmem_cache_destroy(sit_entry_set_slab);
184a5cd2
CY		 2586destory_discard_entry:
2587 kmem_cache_destroy(discard_entry_slab);
2588fail:
2589 return -ENOMEM;
7fd9e544
JK		 2590}
2591
2592void destroy_segment_manager_caches(void)
2593{
184a5cd2 2594 kmem_cache_destroy(sit_entry_set_slab);
7fd9e544 2595 kmem_cache_destroy(discard_entry_slab);
88b88a66 2596 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2597}
Ubuntu Artful kernel mirror