]> 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: build stat_info before orphan inode recovery
[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;
b01a9201 29static struct kmem_cache *discard_cmd_slab;
184a5cd2 30static struct kmem_cache *sit_entry_set_slab;
88b88a66 31static struct kmem_cache *inmem_entry_slab;
7fd9e544 32
f96999c3
JK		 33static unsigned long __reverse_ulong(unsigned char *str)
34{
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38#if BITS_PER_LONG == 64
39 shift = 56;
40#endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46}
47
9a7f143a
CL		 48/*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52static inline unsigned long __reverse_ffs(unsigned long word)
53{
54 int num = 0;
55
56#if BITS_PER_LONG == 64
f96999c3 57 if ((word & 0xffffffff00000000UL) == 0)
9a7f143a 58 num += 32;
f96999c3 59 else
9a7f143a 60 word >>= 32;
9a7f143a 61#endif
f96999c3 62 if ((word & 0xffff0000) == 0)
9a7f143a 63 num += 16;
f96999c3 64 else
9a7f143a 65 word >>= 16;
f96999c3
JK		 66
67 if ((word & 0xff00) == 0)
9a7f143a 68 num += 8;
f96999c3 69 else
9a7f143a 70 word >>= 8;
f96999c3 71
9a7f143a
CL		 72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
f96999c3 76
9a7f143a
CL		 77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
f96999c3 81
9a7f143a
CL		 82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85}
86
87/*
e1c42045 88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
9a7f143a 89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
692223d1 90 * @size must be integral times of unsigned long.
9a7f143a 91 * Example:
f96999c3
JK		 92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
9a7f143a
CL		 95 */
96static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98{
99 const unsigned long *p = addr + BIT_WORD(offset);
692223d1 100 unsigned long result = size;
9a7f143a 101 unsigned long tmp;
9a7f143a
CL		 102
103 if (offset >= size)
104 return size;
105
692223d1 106 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 107 offset %= BITS_PER_LONG;
f96999c3 108
692223d1
FL		 109 while (1) {
110 if (*p == 0)
111 goto pass;
9a7f143a 112
f96999c3 113 tmp = __reverse_ulong((unsigned char *)p);
692223d1
FL		 114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
9a7f143a 118 if (tmp)
692223d1
FL		 119 goto found;
120pass:
121 if (size <= BITS_PER_LONG)
122 break;
9a7f143a 123 size -= BITS_PER_LONG;
692223d1 124 offset = 0;
f96999c3 125 p++;
9a7f143a 126 }
692223d1
FL		 127 return result;
128found:
129 return result - size + __reverse_ffs(tmp);
9a7f143a
CL		 130}
131
132static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134{
135 const unsigned long *p = addr + BIT_WORD(offset);
80609448 136 unsigned long result = size;
9a7f143a 137 unsigned long tmp;
9a7f143a
CL		 138
139 if (offset >= size)
140 return size;
141
80609448 142 size -= (offset & ~(BITS_PER_LONG - 1));
9a7f143a 143 offset %= BITS_PER_LONG;
80609448
JK		 144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
f96999c3 149 tmp = __reverse_ulong((unsigned char *)p);
80609448
JK		 150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
f96999c3 155 if (tmp != ~0UL)
80609448
JK		 156 goto found;
157pass:
158 if (size <= BITS_PER_LONG)
159 break;
9a7f143a 160 size -= BITS_PER_LONG;
80609448 161 offset = 0;
f96999c3 162 p++;
9a7f143a 163 }
80609448
JK		 164 return result;
165found:
166 return result - size + __reverse_ffz(tmp);
9a7f143a
CL		 167}
168
88b88a66
JK		 169void register_inmem_page(struct inode *inode, struct page *page)
170{
171 struct f2fs_inode_info *fi = F2FS_I(inode);
172 struct inmem_pages *new;
9be32d72 173
9e4ded3f 174 f2fs_trace_pid(page);
0722b101 175
decd36b6
CY		 176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
177 SetPagePrivate(page);
178
88b88a66
JK		 179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
180
181 /* add atomic page indices to the list */
182 new->page = page;
183 INIT_LIST_HEAD(&new->list);
decd36b6 184
88b88a66
JK		 185 /* increase reference count with clean state */
186 mutex_lock(&fi->inmem_lock);
187 get_page(page);
188 list_add_tail(&new->list, &fi->inmem_pages);
8dcf2ff7 189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
88b88a66 190 mutex_unlock(&fi->inmem_lock);
8ce67cb0
JK		 191
192 trace_f2fs_register_inmem_page(page, INMEM);
88b88a66
JK		 193}
194
28bc106b
CY		 195static int __revoke_inmem_pages(struct inode *inode,
196 struct list_head *head, bool drop, bool recover)
29b96b54 197{
28bc106b 198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29b96b54 199 struct inmem_pages *cur, *tmp;
28bc106b 200 int err = 0;
29b96b54
CY		 201
202 list_for_each_entry_safe(cur, tmp, head, list) {
28bc106b
CY		 203 struct page *page = cur->page;
204
205 if (drop)
206 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
207
208 lock_page(page);
29b96b54 209
28bc106b
CY		 210 if (recover) {
211 struct dnode_of_data dn;
212 struct node_info ni;
213
214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
215
216 set_new_dnode(&dn, inode, NULL, NULL, 0);
217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
218 err = -EAGAIN;
219 goto next;
220 }
221 get_node_info(sbi, dn.nid, &ni);
222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
223 cur->old_addr, ni.version, true, true);
224 f2fs_put_dnode(&dn);
225 }
226next:
63c52d78
JK		 227 /* we don't need to invalidate this in the sccessful status */
228 if (drop || recover)
229 ClearPageUptodate(page);
28bc106b 230 set_page_private(page, 0);
c81ced05 231 ClearPagePrivate(page);
28bc106b 232 f2fs_put_page(page, 1);
29b96b54
CY		 233
234 list_del(&cur->list);
235 kmem_cache_free(inmem_entry_slab, cur);
236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
237 }
28bc106b 238 return err;
29b96b54
CY		 239}
240
241void drop_inmem_pages(struct inode *inode)
242{
243 struct f2fs_inode_info *fi = F2FS_I(inode);
244
245 mutex_lock(&fi->inmem_lock);
28bc106b 246 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54 247 mutex_unlock(&fi->inmem_lock);
5fe45743
CY		 248
249 clear_inode_flag(inode, FI_ATOMIC_FILE);
250 stat_dec_atomic_write(inode);
29b96b54
CY		 251}
252
28bc106b
CY		 253static int __commit_inmem_pages(struct inode *inode,
254 struct list_head *revoke_list)
88b88a66
JK		 255{
256 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
257 struct f2fs_inode_info *fi = F2FS_I(inode);
258 struct inmem_pages *cur, *tmp;
88b88a66 259 struct f2fs_io_info fio = {
05ca3632 260 .sbi = sbi,
88b88a66 261 .type = DATA,
04d328de 262 .op = REQ_OP_WRITE,
70fd7614 263 .op_flags = REQ_SYNC | REQ_PRIO,
4375a336 264 .encrypted_page = NULL,
88b88a66 265 };
942fd319 266 pgoff_t last_idx = ULONG_MAX;
edb27dee 267 int err = 0;
88b88a66 268
88b88a66 269 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 270 struct page *page = cur->page;
271
272 lock_page(page);
273 if (page->mapping == inode->i_mapping) {
274 trace_f2fs_commit_inmem_page(page, INMEM);
275
276 set_page_dirty(page);
277 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 278 if (clear_page_dirty_for_io(page)) {
29b96b54 279 inode_dec_dirty_pages(inode);
933439c8
CY		 280 remove_dirty_inode(inode);
281 }
28bc106b
CY		 282
283 fio.page = page;
29b96b54
CY		 284 err = do_write_data_page(&fio);
285 if (err) {
28bc106b 286 unlock_page(page);
29b96b54 287 break;
70c640b1 288 }
29b96b54 289
28bc106b
CY		 290 /* record old blkaddr for revoking */
291 cur->old_addr = fio.old_blkaddr;
942fd319 292 last_idx = page->index;
28bc106b
CY		 293 }
294 unlock_page(page);
295 list_move_tail(&cur->list, revoke_list);
88b88a66 296 }
29b96b54 297
942fd319
JK		 298 if (last_idx != ULONG_MAX)
299 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx,
300 DATA, WRITE);
28bc106b
CY		 301
302 if (!err)
303 __revoke_inmem_pages(inode, revoke_list, false, false);
304
29b96b54
CY		 305 return err;
306}
307
308int commit_inmem_pages(struct inode *inode)
309{
310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
311 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 312 struct list_head revoke_list;
313 int err;
29b96b54 314
28bc106b 315 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 316 f2fs_balance_fs(sbi, true);
317 f2fs_lock_op(sbi);
318
5fe45743
CY		 319 set_inode_flag(inode, FI_ATOMIC_COMMIT);
320
29b96b54 321 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 322 err = __commit_inmem_pages(inode, &revoke_list);
323 if (err) {
324 int ret;
325 /*
326 * try to revoke all committed pages, but still we could fail
327 * due to no memory or other reason, if that happened, EAGAIN
328 * will be returned, which means in such case, transaction is
329 * already not integrity, caller should use journal to do the
330 * recovery or rewrite & commit last transaction. For other
331 * error number, revoking was done by filesystem itself.
332 */
333 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
334 if (ret)
335 err = ret;
336
337 /* drop all uncommitted pages */
338 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
339 }
88b88a66
JK		 340 mutex_unlock(&fi->inmem_lock);
341
5fe45743
CY		 342 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
343
29b96b54 344 f2fs_unlock_op(sbi);
edb27dee 345 return err;
88b88a66
JK		 346}
347
0a8165d7 348/*
351df4b2
JK		 349 * This function balances dirty node and dentry pages.
350 * In addition, it controls garbage collection.
351 */
2c4db1a6 352void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 353{
0f348028 354#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 355 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
356 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 357 f2fs_stop_checkpoint(sbi, false);
55523519 358 }
0f348028
CY		 359#endif
360
2c4db1a6
JK		 361 if (!need)
362 return;
e589c2c4
JK		 363
364 /* balance_fs_bg is able to be pending */
365 if (excess_cached_nats(sbi))
366 f2fs_balance_fs_bg(sbi);
367
351df4b2 368 /*
029cd28c
JK		 369 * We should do GC or end up with checkpoint, if there are so many dirty
370 * dir/node pages without enough free segments.
351df4b2 371 */
7f3037a5 372 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 373 mutex_lock(&sbi->gc_mutex);
7702bdbe 374 f2fs_gc(sbi, false, false);
351df4b2
JK		 375 }
376}
377
4660f9c0
JK		 378void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
379{
1dcc336b 380 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 381 if (!available_free_memory(sbi, EXTENT_CACHE))
382 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 383
1b38dc8e
JK		 384 /* check the # of cached NAT entries */
385 if (!available_free_memory(sbi, NAT_ENTRIES))
386 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
387
31696580 388 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 389 try_to_free_nids(sbi, MAX_FREE_NIDS);
390 else
22ad0b6a 391 build_free_nids(sbi, false, false);
31696580 392
f455c8a5
JK		 393 if (!is_idle(sbi))
394 return;
31696580 395
1b38dc8e
JK		 396 /* checkpoint is the only way to shrink partial cached entries */
397 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 398 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 399 excess_prefree_segs(sbi) ||
400 excess_dirty_nats(sbi) ||
f455c8a5 401 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 402 if (test_opt(sbi, DATA_FLUSH)) {
403 struct blk_plug plug;
404
405 blk_start_plug(&plug);
36b35a0d 406 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 407 blk_finish_plug(&plug);
408 }
4660f9c0 409 f2fs_sync_fs(sbi->sb, true);
42190d2a 410 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 411 }
4660f9c0
JK		 412}
413
3c62be17
JK		 414static int __submit_flush_wait(struct block_device *bdev)
415{
416 struct bio *bio = f2fs_bio_alloc(0);
417 int ret;
418
09cb6464 419 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3c62be17
JK		 420 bio->bi_bdev = bdev;
421 ret = submit_bio_wait(bio);
422 bio_put(bio);
423 return ret;
424}
425
426static int submit_flush_wait(struct f2fs_sb_info *sbi)
427{
428 int ret = __submit_flush_wait(sbi->sb->s_bdev);
429 int i;
430
431 if (sbi->s_ndevs && !ret) {
432 for (i = 1; i < sbi->s_ndevs; i++) {
d50aaeec
JK		 433 trace_f2fs_issue_flush(FDEV(i).bdev,
434 test_opt(sbi, NOBARRIER),
435 test_opt(sbi, FLUSH_MERGE));
3c62be17
JK		 436 ret = __submit_flush_wait(FDEV(i).bdev);
437 if (ret)
438 break;
439 }
440 }
441 return ret;
442}
443
2163d198 444static int issue_flush_thread(void *data)
6b4afdd7
JK		 445{
446 struct f2fs_sb_info *sbi = data;
b01a9201 447 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 448 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 449repeat:
450 if (kthread_should_stop())
451 return 0;
452
721bd4d5 453 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 454 struct flush_cmd *cmd, *next;
455 int ret;
456
721bd4d5
GZ		 457 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
458 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
459
3c62be17 460 ret = submit_flush_wait(sbi);
721bd4d5
GZ		 461 llist_for_each_entry_safe(cmd, next,
462 fcc->dispatch_list, llnode) {
6b4afdd7 463 cmd->ret = ret;
6b4afdd7
JK		 464 complete(&cmd->wait);
465 }
a688b9d9 466 fcc->dispatch_list = NULL;
6b4afdd7
JK		 467 }
468
a688b9d9 469 wait_event_interruptible(*q,
721bd4d5 470 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 471 goto repeat;
472}
473
474int f2fs_issue_flush(struct f2fs_sb_info *sbi)
475{
b01a9201 476 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 477 struct flush_cmd cmd;
6b4afdd7 478
0f7b2abd
JK		 479 if (test_opt(sbi, NOBARRIER))
480 return 0;
481
73545817
CY		 482 if (!test_opt(sbi, FLUSH_MERGE))
483 return submit_flush_wait(sbi);
484
485 if (!atomic_read(&fcc->submit_flush)) {
740432f8
JK		 486 int ret;
487
0a87f664 488 atomic_inc(&fcc->submit_flush);
3c62be17 489 ret = submit_flush_wait(sbi);
0a87f664 490 atomic_dec(&fcc->submit_flush);
740432f8
JK		 491 return ret;
492 }
6b4afdd7 493
adf8d90b 494 init_completion(&cmd.wait);
6b4afdd7 495
0a87f664 496 atomic_inc(&fcc->submit_flush);
721bd4d5 497 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 498
a688b9d9
GZ		 499 if (!fcc->dispatch_list)
500 wake_up(&fcc->flush_wait_queue);
6b4afdd7 501
5eba8c5d
JK		 502 if (fcc->f2fs_issue_flush) {
503 wait_for_completion(&cmd.wait);
504 atomic_dec(&fcc->submit_flush);
505 } else {
506 llist_del_all(&fcc->issue_list);
507 atomic_set(&fcc->submit_flush, 0);
508 }
adf8d90b
CY		 509
510 return cmd.ret;
6b4afdd7
JK		 511}
512
2163d198
GZ		 513int create_flush_cmd_control(struct f2fs_sb_info *sbi)
514{
515 dev_t dev = sbi->sb->s_bdev->bd_dev;
516 struct flush_cmd_control *fcc;
517 int err = 0;
518
b01a9201
JK		 519 if (SM_I(sbi)->fcc_info) {
520 fcc = SM_I(sbi)->fcc_info;
5eba8c5d
JK		 521 goto init_thread;
522 }
523
2163d198
GZ		 524 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
525 if (!fcc)
526 return -ENOMEM;
0a87f664 527 atomic_set(&fcc->submit_flush, 0);
2163d198 528 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 529 init_llist_head(&fcc->issue_list);
b01a9201 530 SM_I(sbi)->fcc_info = fcc;
5eba8c5d 531init_thread:
2163d198
GZ		 532 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
533 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
534 if (IS_ERR(fcc->f2fs_issue_flush)) {
535 err = PTR_ERR(fcc->f2fs_issue_flush);
536 kfree(fcc);
b01a9201 537 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 538 return err;
539 }
2163d198
GZ		 540
541 return err;
542}
543
5eba8c5d 544void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 545{
b01a9201 546 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 547
5eba8c5d
JK		 548 if (fcc && fcc->f2fs_issue_flush) {
549 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
550
551 fcc->f2fs_issue_flush = NULL;
552 kthread_stop(flush_thread);
553 }
554 if (free) {
555 kfree(fcc);
b01a9201 556 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 557 }
2163d198
GZ		 558}
559
351df4b2
JK		 560static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
561 enum dirty_type dirty_type)
562{
563 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
564
565 /* need not be added */
566 if (IS_CURSEG(sbi, segno))
567 return;
568
569 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
570 dirty_i->nr_dirty[dirty_type]++;
571
572 if (dirty_type == DIRTY) {
573 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 574 enum dirty_type t = sentry->type;
b2f2c390 575
ec325b52
JK		 576 if (unlikely(t >= DIRTY)) {
577 f2fs_bug_on(sbi, 1);
578 return;
579 }
4625d6aa
CL		 580 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
581 dirty_i->nr_dirty[t]++;
351df4b2
JK		 582 }
583}
584
585static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
586 enum dirty_type dirty_type)
587{
588 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
589
590 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
591 dirty_i->nr_dirty[dirty_type]--;
592
593 if (dirty_type == DIRTY) {
4625d6aa
CL		 594 struct seg_entry *sentry = get_seg_entry(sbi, segno);
595 enum dirty_type t = sentry->type;
596
597 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
598 dirty_i->nr_dirty[t]--;
b2f2c390 599
5ec4e49f
JK		 600 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
601 clear_bit(GET_SECNO(sbi, segno),
602 dirty_i->victim_secmap);
351df4b2
JK		 603 }
604}
605
0a8165d7 606/*
351df4b2
JK		 607 * Should not occur error such as -ENOMEM.
608 * Adding dirty entry into seglist is not critical operation.
609 * If a given segment is one of current working segments, it won't be added.
610 */
8d8451af 611static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 612{
613 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
614 unsigned short valid_blocks;
615
616 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
617 return;
618
619 mutex_lock(&dirty_i->seglist_lock);
620
621 valid_blocks = get_valid_blocks(sbi, segno, 0);
622
623 if (valid_blocks == 0) {
624 __locate_dirty_segment(sbi, segno, PRE);
625 __remove_dirty_segment(sbi, segno, DIRTY);
626 } else if (valid_blocks < sbi->blocks_per_seg) {
627 __locate_dirty_segment(sbi, segno, DIRTY);
628 } else {
629 /* Recovery routine with SSR needs this */
630 __remove_dirty_segment(sbi, segno, DIRTY);
631 }
632
633 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 634}
635
15469963 636static void __add_discard_cmd(struct f2fs_sb_info *sbi,
4e6a8d9b 637 struct bio *bio, block_t lstart, block_t len)
275b66b0 638{
0b54fb84
JK		 639 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
640 struct list_head *cmd_list = &(dcc->discard_cmd_list);
b01a9201 641 struct discard_cmd *dc;
275b66b0 642
b01a9201
JK		 643 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
644 INIT_LIST_HEAD(&dc->list);
645 dc->bio = bio;
15469963 646 bio->bi_private = dc;
b01a9201
JK		 647 dc->lstart = lstart;
648 dc->len = len;
15469963 649 dc->state = D_PREP;
b01a9201 650 init_completion(&dc->wait);
15469963
JK		 651
652 mutex_lock(&dcc->cmd_lock);
0b54fb84 653 list_add_tail(&dc->list, cmd_list);
15469963
JK		 654 mutex_unlock(&dcc->cmd_lock);
655}
656
657static void __remove_discard_cmd(struct f2fs_sb_info *sbi, struct discard_cmd *dc)
658{
659 int err = dc->bio->bi_error;
275b66b0 660
dcc9165d
JK		 661 if (dc->state == D_DONE)
662 atomic_dec(&(SM_I(sbi)->dcc_info->submit_discard));
663
15469963
JK		 664 if (err == -EOPNOTSUPP)
665 err = 0;
666
667 if (err)
668 f2fs_msg(sbi->sb, KERN_INFO,
669 "Issue discard failed, ret: %d", err);
670 bio_put(dc->bio);
671 list_del(&dc->list);
672 kmem_cache_free(discard_cmd_slab, dc);
275b66b0
CY		 673}
674
4e6a8d9b
JK		 675/* This should be covered by global mutex, &sit_i->sentry_lock */
676void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
275b66b0 677{
0b54fb84
JK		 678 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
679 struct list_head *wait_list = &(dcc->discard_cmd_list);
b01a9201 680 struct discard_cmd *dc, *tmp;
40465257 681 struct blk_plug plug;
275b66b0 682
15469963 683 mutex_lock(&dcc->cmd_lock);
40465257
JK		 684
685 blk_start_plug(&plug);
686
b01a9201 687 list_for_each_entry_safe(dc, tmp, wait_list, list) {
275b66b0 688
15469963
JK		 689 if (blkaddr == NULL_ADDR) {
690 if (dc->state == D_PREP) {
691 dc->state = D_SUBMIT;
692 submit_bio(dc->bio);
dcc9165d 693 atomic_inc(&dcc->submit_discard);
15469963 694 }
15469963
JK		 695 continue;
696 }
697
698 if (dc->lstart <= blkaddr && blkaddr < dc->lstart + dc->len) {
699 if (dc->state == D_SUBMIT)
b01a9201 700 wait_for_completion_io(&dc->wait);
4e6a8d9b 701 else
15469963 702 __remove_discard_cmd(sbi, dc);
4e6a8d9b 703 }
275b66b0 704 }
40465257
JK		 705 blk_finish_plug(&plug);
706
707 /* this comes from f2fs_put_super */
708 if (blkaddr == NULL_ADDR) {
709 list_for_each_entry_safe(dc, tmp, wait_list, list) {
710 wait_for_completion_io(&dc->wait);
711 __remove_discard_cmd(sbi, dc);
712 }
713 }
15469963 714 mutex_unlock(&dcc->cmd_lock);
275b66b0
CY		 715}
716
b01a9201 717static void f2fs_submit_discard_endio(struct bio *bio)
275b66b0 718{
b01a9201 719 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
275b66b0 720
b01a9201 721 complete(&dc->wait);
15469963 722 dc->state = D_DONE;
275b66b0
CY		 723}
724
15469963
JK		 725static int issue_discard_thread(void *data)
726{
727 struct f2fs_sb_info *sbi = data;
728 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
729 wait_queue_head_t *q = &dcc->discard_wait_queue;
730 struct list_head *cmd_list = &dcc->discard_cmd_list;
731 struct discard_cmd *dc, *tmp;
732 struct blk_plug plug;
733 int iter = 0;
734repeat:
735 if (kthread_should_stop())
736 return 0;
737
738 blk_start_plug(&plug);
739
740 mutex_lock(&dcc->cmd_lock);
741 list_for_each_entry_safe(dc, tmp, cmd_list, list) {
742 if (dc->state == D_PREP) {
743 dc->state = D_SUBMIT;
744 submit_bio(dc->bio);
dcc9165d 745 atomic_inc(&dcc->submit_discard);
15469963
JK		 746 if (iter++ > DISCARD_ISSUE_RATE)
747 break;
748 } else if (dc->state == D_DONE) {
749 __remove_discard_cmd(sbi, dc);
750 }
751 }
752 mutex_unlock(&dcc->cmd_lock);
753
754 blk_finish_plug(&plug);
755
756 iter = 0;
757 congestion_wait(BLK_RW_SYNC, HZ/50);
758
759 wait_event_interruptible(*q,
760 kthread_should_stop() || !list_empty(&dcc->discard_cmd_list));
761 goto repeat;
762}
763
764
275b66b0 765/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
f46e8809 766static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
3c62be17 767 struct block_device *bdev, block_t blkstart, block_t blklen)
275b66b0 768{
275b66b0 769 struct bio *bio = NULL;
4e6a8d9b 770 block_t lblkstart = blkstart;
275b66b0
CY		 771 int err;
772
d50aaeec 773 trace_f2fs_issue_discard(bdev, blkstart, blklen);
f46e8809 774
3c62be17
JK		 775 if (sbi->s_ndevs) {
776 int devi = f2fs_target_device_index(sbi, blkstart);
777
778 blkstart -= FDEV(devi).start_blk;
779 }
f46e8809
DLM		 780 err = __blkdev_issue_discard(bdev,
781 SECTOR_FROM_BLOCK(blkstart),
782 SECTOR_FROM_BLOCK(blklen),
783 GFP_NOFS, 0, &bio);
275b66b0 784 if (!err && bio) {
b01a9201 785 bio->bi_end_io = f2fs_submit_discard_endio;
275b66b0 786 bio->bi_opf |= REQ_SYNC;
15469963
JK		 787
788 __add_discard_cmd(sbi, bio, lblkstart, blklen);
789 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
275b66b0 790 }
275b66b0
CY		 791 return err;
792}
793
f46e8809 794#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 795static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
796 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 797{
92592285 798 sector_t sector, nr_sects;
10a875f8 799 block_t lblkstart = blkstart;
3c62be17
JK		 800 int devi = 0;
801
802 if (sbi->s_ndevs) {
803 devi = f2fs_target_device_index(sbi, blkstart);
804 blkstart -= FDEV(devi).start_blk;
805 }
f46e8809
DLM		 806
807 /*
808 * We need to know the type of the zone: for conventional zones,
809 * use regular discard if the drive supports it. For sequential
810 * zones, reset the zone write pointer.
811 */
3c62be17 812 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 813
814 case BLK_ZONE_TYPE_CONVENTIONAL:
815 if (!blk_queue_discard(bdev_get_queue(bdev)))
816 return 0;
10a875f8 817 return __f2fs_issue_discard_async(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 818 case BLK_ZONE_TYPE_SEQWRITE_REQ:
819 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 820 sector = SECTOR_FROM_BLOCK(blkstart);
821 nr_sects = SECTOR_FROM_BLOCK(blklen);
822
823 if (sector & (bdev_zone_sectors(bdev) - 1) ||
824 nr_sects != bdev_zone_sectors(bdev)) {
825 f2fs_msg(sbi->sb, KERN_INFO,
826 "(%d) %s: Unaligned discard attempted (block %x + %x)",
827 devi, sbi->s_ndevs ? FDEV(devi).path: "",
828 blkstart, blklen);
829 return -EIO;
830 }
d50aaeec 831 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 832 return blkdev_reset_zones(bdev, sector,
833 nr_sects, GFP_NOFS);
834 default:
835 /* Unknown zone type: broken device ? */
836 return -EIO;
837 }
838}
839#endif
840
3c62be17
JK		 841static int __issue_discard_async(struct f2fs_sb_info *sbi,
842 struct block_device *bdev, block_t blkstart, block_t blklen)
843{
844#ifdef CONFIG_BLK_DEV_ZONED
845 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
846 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
847 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
848#endif
849 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
850}
851
1e87a78d 852static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 853 block_t blkstart, block_t blklen)
854{
3c62be17
JK		 855 sector_t start = blkstart, len = 0;
856 struct block_device *bdev;
a66cdd98
JK		 857 struct seg_entry *se;
858 unsigned int offset;
859 block_t i;
3c62be17
JK		 860 int err = 0;
861
862 bdev = f2fs_target_device(sbi, blkstart, NULL);
863
864 for (i = blkstart; i < blkstart + blklen; i++, len++) {
865 if (i != start) {
866 struct block_device *bdev2 =
867 f2fs_target_device(sbi, i, NULL);
868
869 if (bdev2 != bdev) {
870 err = __issue_discard_async(sbi, bdev,
871 start, len);
872 if (err)
873 return err;
874 bdev = bdev2;
875 start = i;
876 len = 0;
877 }
878 }
a66cdd98 879
a66cdd98
JK		 880 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
881 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
882
883 if (!f2fs_test_and_set_bit(offset, se->discard_map))
884 sbi->discard_blks--;
885 }
f46e8809 886
3c62be17
JK		 887 if (len)
888 err = __issue_discard_async(sbi, bdev, start, len);
889 return err;
1e87a78d
JK		 890}
891
adf4983b 892static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 893 struct cp_control *cpc, struct seg_entry *se,
894 unsigned int start, unsigned int end)
b2955550 895{
0b54fb84 896 struct list_head *head = &SM_I(sbi)->dcc_info->discard_entry_list;
adf4983b
JK		 897 struct discard_entry *new, *last;
898
899 if (!list_empty(head)) {
900 last = list_last_entry(head, struct discard_entry, list);
901 if (START_BLOCK(sbi, cpc->trim_start) + start ==
ad4d307f
JK		 902 last->blkaddr + last->len &&
903 last->len < MAX_DISCARD_BLOCKS(sbi)) {
adf4983b
JK		 904 last->len += end - start;
905 goto done;
906 }
907 }
908
909 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
910 INIT_LIST_HEAD(&new->list);
911 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
912 new->len = end - start;
913 list_add_tail(&new->list, head);
914done:
0b54fb84 915 SM_I(sbi)->dcc_info->nr_discards += end - start;
adf4983b
JK		 916}
917
25290fa5
JK		 918static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
919 bool check_only)
adf4983b 920{
b2955550
JK		 921 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
922 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 923 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 924 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
925 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 926 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 927 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 928 unsigned int start = 0, end = -1;
4b2fecc8 929 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 930 int i;
931
3e025740 932 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 933 return false;
b2955550 934
a66cdd98
JK		 935 if (!force) {
936 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 937 SM_I(sbi)->dcc_info->nr_discards >=
938 SM_I(sbi)->dcc_info->max_discards)
25290fa5 939 return false;
4b2fecc8
JK		 940 }
941
b2955550
JK		 942 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
943 for (i = 0; i < entries; i++)
a66cdd98 944 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 945 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 946
0b54fb84
JK		 947 while (force || SM_I(sbi)->dcc_info->nr_discards <=
948 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 949 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
950 if (start >= max_blocks)
951 break;
952
953 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 954 if (force && start && end != max_blocks
955 && (end - start) < cpc->trim_minlen)
956 continue;
957
25290fa5
JK		 958 if (check_only)
959 return true;
960
a66cdd98 961 __add_discard_entry(sbi, cpc, se, start, end);
b2955550 962 }
25290fa5 963 return false;
b2955550
JK		 964}
965
4b2fecc8
JK		 966void release_discard_addrs(struct f2fs_sb_info *sbi)
967{
0b54fb84 968 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
4b2fecc8
JK		 969 struct discard_entry *entry, *this;
970
971 /* drop caches */
972 list_for_each_entry_safe(entry, this, head, list) {
973 list_del(&entry->list);
974 kmem_cache_free(discard_entry_slab, entry);
975 }
976}
977
0a8165d7 978/*
351df4b2
JK		 979 * Should call clear_prefree_segments after checkpoint is done.
980 */
981static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
982{
983 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 984 unsigned int segno;
351df4b2
JK		 985
986 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 987 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 988 __set_test_and_free(sbi, segno);
351df4b2
JK		 989 mutex_unlock(&dirty_i->seglist_lock);
990}
991
836b5a63 992void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 993{
0b54fb84 994 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
2d7b822a 995 struct discard_entry *entry, *this;
351df4b2 996 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 997 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 998 unsigned int start = 0, end = -1;
36abef4e 999 unsigned int secno, start_segno;
c24a0fd6 1000 bool force = (cpc->reason == CP_DISCARD);
351df4b2
JK		 1001
1002 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1003
351df4b2 1004 while (1) {
29e59c14 1005 int i;
7cd8558b
JK		 1006 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1007 if (start >= MAIN_SEGS(sbi))
351df4b2 1008 break;
7cd8558b
JK		 1009 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1010 start + 1);
29e59c14
CL		 1011
1012 for (i = start; i < end; i++)
1013 clear_bit(i, prefree_map);
1014
1015 dirty_i->nr_dirty[PRE] -= end - start;
1016
650d3c4e 1017 if (!test_opt(sbi, DISCARD))
29e59c14 1018 continue;
351df4b2 1019
650d3c4e
YH		 1020 if (force && start >= cpc->trim_start &&
1021 (end - 1) <= cpc->trim_end)
1022 continue;
1023
36abef4e
JK		 1024 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1025 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1026 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1027 continue;
1028 }
1029next:
1030 secno = GET_SECNO(sbi, start);
1031 start_segno = secno * sbi->segs_per_sec;
1032 if (!IS_CURSEC(sbi, secno) &&
1033 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
1034 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1035 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1036
1037 start = start_segno + sbi->segs_per_sec;
1038 if (start < end)
1039 goto next;
8b107f5b
JK		 1040 else
1041 end = start - 1;
351df4b2
JK		 1042 }
1043 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1044
1045 /* send small discards */
2d7b822a 1046 list_for_each_entry_safe(entry, this, head, list) {
c24a0fd6 1047 if (force && entry->len < cpc->trim_minlen)
836b5a63 1048 goto skip;
37208879 1049 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 1050 cpc->trimmed += entry->len;
836b5a63 1051skip:
b2955550 1052 list_del(&entry->list);
0b54fb84 1053 SM_I(sbi)->dcc_info->nr_discards -= entry->len;
b2955550
JK		 1054 kmem_cache_free(discard_entry_slab, entry);
1055 }
351df4b2
JK		 1056}
1057
8ed59745 1058static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1059{
15469963 1060 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84
JK		 1061 struct discard_cmd_control *dcc;
1062 int err = 0;
1063
1064 if (SM_I(sbi)->dcc_info) {
1065 dcc = SM_I(sbi)->dcc_info;
1066 goto init_thread;
1067 }
1068
1069 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1070 if (!dcc)
1071 return -ENOMEM;
1072
1073 INIT_LIST_HEAD(&dcc->discard_entry_list);
1074 INIT_LIST_HEAD(&dcc->discard_cmd_list);
15469963 1075 mutex_init(&dcc->cmd_lock);
dcc9165d 1076 atomic_set(&dcc->submit_discard, 0);
0b54fb84
JK		 1077 dcc->nr_discards = 0;
1078 dcc->max_discards = 0;
1079
15469963 1080 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1081 SM_I(sbi)->dcc_info = dcc;
1082init_thread:
15469963
JK		 1083 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1084 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1085 if (IS_ERR(dcc->f2fs_issue_discard)) {
1086 err = PTR_ERR(dcc->f2fs_issue_discard);
1087 kfree(dcc);
1088 SM_I(sbi)->dcc_info = NULL;
1089 return err;
1090 }
1091
0b54fb84
JK		 1092 return err;
1093}
1094
8ed59745 1095static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi, bool free)
0b54fb84
JK		 1096{
1097 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1098
15469963
JK		 1099 if (dcc && dcc->f2fs_issue_discard) {
1100 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1101
1102 dcc->f2fs_issue_discard = NULL;
1103 kthread_stop(discard_thread);
1104 }
0b54fb84
JK		 1105 if (free) {
1106 kfree(dcc);
1107 SM_I(sbi)->dcc_info = NULL;
1108 }
1109}
1110
184a5cd2 1111static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1112{
1113 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1114
1115 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1116 sit_i->dirty_sentries++;
184a5cd2
CY		 1117 return false;
1118 }
1119
1120 return true;
351df4b2
JK		 1121}
1122
1123static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1124 unsigned int segno, int modified)
1125{
1126 struct seg_entry *se = get_seg_entry(sbi, segno);
1127 se->type = type;
1128 if (modified)
1129 __mark_sit_entry_dirty(sbi, segno);
1130}
1131
1132static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1133{
1134 struct seg_entry *se;
1135 unsigned int segno, offset;
1136 long int new_vblocks;
1137
1138 segno = GET_SEGNO(sbi, blkaddr);
1139
1140 se = get_seg_entry(sbi, segno);
1141 new_vblocks = se->valid_blocks + del;
491c0854 1142 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1143
9850cf4a 1144 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1145 (new_vblocks > sbi->blocks_per_seg)));
1146
1147 se->valid_blocks = new_vblocks;
1148 se->mtime = get_mtime(sbi);
1149 SIT_I(sbi)->max_mtime = se->mtime;
1150
1151 /* Update valid block bitmap */
1152 if (del > 0) {
355e7891
CY		 1153 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1154#ifdef CONFIG_F2FS_CHECK_FS
1155 if (f2fs_test_and_set_bit(offset,
1156 se->cur_valid_map_mir))
1157 f2fs_bug_on(sbi, 1);
1158 else
1159 WARN_ON(1);
1160#else
05796763 1161 f2fs_bug_on(sbi, 1);
355e7891
CY		 1162#endif
1163 }
3e025740
JK		 1164 if (f2fs_discard_en(sbi) &&
1165 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1166 sbi->discard_blks--;
720037f9
JK		 1167
1168 /* don't overwrite by SSR to keep node chain */
1169 if (se->type == CURSEG_WARM_NODE) {
1170 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1171 se->ckpt_valid_blocks++;
1172 }
351df4b2 1173 } else {
355e7891
CY		 1174 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1175#ifdef CONFIG_F2FS_CHECK_FS
1176 if (!f2fs_test_and_clear_bit(offset,
1177 se->cur_valid_map_mir))
1178 f2fs_bug_on(sbi, 1);
1179 else
1180 WARN_ON(1);
1181#else
05796763 1182 f2fs_bug_on(sbi, 1);
355e7891
CY		 1183#endif
1184 }
3e025740
JK		 1185 if (f2fs_discard_en(sbi) &&
1186 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1187 sbi->discard_blks++;
351df4b2
JK		 1188 }
1189 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1190 se->ckpt_valid_blocks += del;
1191
1192 __mark_sit_entry_dirty(sbi, segno);
1193
1194 /* update total number of valid blocks to be written in ckpt area */
1195 SIT_I(sbi)->written_valid_blocks += del;
1196
1197 if (sbi->segs_per_sec > 1)
1198 get_sec_entry(sbi, segno)->valid_blocks += del;
1199}
1200
5e443818 1201void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1202{
5e443818
JK		 1203 update_sit_entry(sbi, new, 1);
1204 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1205 update_sit_entry(sbi, old, -1);
1206
1207 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1208 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1209}
1210
1211void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1212{
1213 unsigned int segno = GET_SEGNO(sbi, addr);
1214 struct sit_info *sit_i = SIT_I(sbi);
1215
9850cf4a 1216 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1217 if (addr == NEW_ADDR)
1218 return;
1219
1220 /* add it into sit main buffer */
1221 mutex_lock(&sit_i->sentry_lock);
1222
1223 update_sit_entry(sbi, addr, -1);
1224
1225 /* add it into dirty seglist */
1226 locate_dirty_segment(sbi, segno);
1227
1228 mutex_unlock(&sit_i->sentry_lock);
1229}
1230
6e2c64ad
JK		 1231bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1232{
1233 struct sit_info *sit_i = SIT_I(sbi);
1234 unsigned int segno, offset;
1235 struct seg_entry *se;
1236 bool is_cp = false;
1237
1238 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1239 return true;
1240
1241 mutex_lock(&sit_i->sentry_lock);
1242
1243 segno = GET_SEGNO(sbi, blkaddr);
1244 se = get_seg_entry(sbi, segno);
1245 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1246
1247 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1248 is_cp = true;
1249
1250 mutex_unlock(&sit_i->sentry_lock);
1251
1252 return is_cp;
1253}
1254
0a8165d7 1255/*
351df4b2
JK		 1256 * This function should be resided under the curseg_mutex lock
1257 */
1258static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1259 struct f2fs_summary *sum)
351df4b2
JK		 1260{
1261 struct curseg_info *curseg = CURSEG_I(sbi, type);
1262 void *addr = curseg->sum_blk;
e79efe3b 1263 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1264 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1265}
1266
0a8165d7 1267/*
351df4b2
JK		 1268 * Calculate the number of current summary pages for writing
1269 */
3fa06d7b 1270int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1271{
351df4b2 1272 int valid_sum_count = 0;
9a47938b 1273 int i, sum_in_page;
351df4b2
JK		 1274
1275 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1276 if (sbi->ckpt->alloc_type[i] == SSR)
1277 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1278 else {
1279 if (for_ra)
1280 valid_sum_count += le16_to_cpu(
1281 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1282 else
1283 valid_sum_count += curseg_blkoff(sbi, i);
1284 }
351df4b2
JK		 1285 }
1286
09cbfeaf 1287 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1288 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1289 if (valid_sum_count <= sum_in_page)
351df4b2 1290 return 1;
9a47938b 1291 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1292 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1293 return 2;
1294 return 3;
1295}
1296
0a8165d7 1297/*
351df4b2
JK		 1298 * Caller should put this summary page
1299 */
1300struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1301{
1302 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1303}
1304
381722d2 1305void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1306{
1307 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1308 void *dst = page_address(page);
1309
1310 if (src)
09cbfeaf 1311 memcpy(dst, src, PAGE_SIZE);
381722d2 1312 else
09cbfeaf 1313 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1314 set_page_dirty(page);
1315 f2fs_put_page(page, 1);
1316}
1317
381722d2
CY		 1318static void write_sum_page(struct f2fs_sb_info *sbi,
1319 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1320{
1321 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1322}
1323
b7ad7512
CY		 1324static void write_current_sum_page(struct f2fs_sb_info *sbi,
1325 int type, block_t blk_addr)
1326{
1327 struct curseg_info *curseg = CURSEG_I(sbi, type);
1328 struct page *page = grab_meta_page(sbi, blk_addr);
1329 struct f2fs_summary_block *src = curseg->sum_blk;
1330 struct f2fs_summary_block *dst;
1331
1332 dst = (struct f2fs_summary_block *)page_address(page);
1333
1334 mutex_lock(&curseg->curseg_mutex);
1335
1336 down_read(&curseg->journal_rwsem);
1337 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1338 up_read(&curseg->journal_rwsem);
1339
1340 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1341 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1342
1343 mutex_unlock(&curseg->curseg_mutex);
1344
1345 set_page_dirty(page);
1346 f2fs_put_page(page, 1);
1347}
1348
0a8165d7 1349/*
351df4b2
JK		 1350 * Find a new segment from the free segments bitmap to right order
1351 * This function should be returned with success, otherwise BUG
1352 */
1353static void get_new_segment(struct f2fs_sb_info *sbi,
1354 unsigned int *newseg, bool new_sec, int dir)
1355{
1356 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1357 unsigned int segno, secno, zoneno;
7cd8558b 1358 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1359 unsigned int hint = *newseg / sbi->segs_per_sec;
1360 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1361 unsigned int left_start = hint;
1362 bool init = true;
1363 int go_left = 0;
1364 int i;
1365
1a118ccf 1366 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1367
1368 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1369 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1370 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1371 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1372 goto got_it;
1373 }
1374find_other_zone:
7cd8558b
JK		 1375 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1376 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1377 if (dir == ALLOC_RIGHT) {
1378 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1379 MAIN_SECS(sbi), 0);
1380 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1381 } else {
1382 go_left = 1;
1383 left_start = hint - 1;
1384 }
1385 }
1386 if (go_left == 0)
1387 goto skip_left;
1388
1389 while (test_bit(left_start, free_i->free_secmap)) {
1390 if (left_start > 0) {
1391 left_start--;
1392 continue;
1393 }
1394 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1395 MAIN_SECS(sbi), 0);
1396 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1397 break;
1398 }
1399 secno = left_start;
1400skip_left:
1401 hint = secno;
1402 segno = secno * sbi->segs_per_sec;
1403 zoneno = secno / sbi->secs_per_zone;
1404
1405 /* give up on finding another zone */
1406 if (!init)
1407 goto got_it;
1408 if (sbi->secs_per_zone == 1)
1409 goto got_it;
1410 if (zoneno == old_zoneno)
1411 goto got_it;
1412 if (dir == ALLOC_LEFT) {
1413 if (!go_left && zoneno + 1 >= total_zones)
1414 goto got_it;
1415 if (go_left && zoneno == 0)
1416 goto got_it;
1417 }
1418 for (i = 0; i < NR_CURSEG_TYPE; i++)
1419 if (CURSEG_I(sbi, i)->zone == zoneno)
1420 break;
1421
1422 if (i < NR_CURSEG_TYPE) {
1423 /* zone is in user, try another */
1424 if (go_left)
1425 hint = zoneno * sbi->secs_per_zone - 1;
1426 else if (zoneno + 1 >= total_zones)
1427 hint = 0;
1428 else
1429 hint = (zoneno + 1) * sbi->secs_per_zone;
1430 init = false;
1431 goto find_other_zone;
1432 }
1433got_it:
1434 /* set it as dirty segment in free segmap */
9850cf4a 1435 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1436 __set_inuse(sbi, segno);
1437 *newseg = segno;
1a118ccf 1438 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1439}
1440
1441static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1442{
1443 struct curseg_info *curseg = CURSEG_I(sbi, type);
1444 struct summary_footer *sum_footer;
1445
1446 curseg->segno = curseg->next_segno;
1447 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1448 curseg->next_blkoff = 0;
1449 curseg->next_segno = NULL_SEGNO;
1450
1451 sum_footer = &(curseg->sum_blk->footer);
1452 memset(sum_footer, 0, sizeof(struct summary_footer));
1453 if (IS_DATASEG(type))
1454 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1455 if (IS_NODESEG(type))
1456 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1457 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1458}
1459
0a8165d7 1460/*
351df4b2
JK		 1461 * Allocate a current working segment.
1462 * This function always allocates a free segment in LFS manner.
1463 */
1464static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1465{
1466 struct curseg_info *curseg = CURSEG_I(sbi, type);
1467 unsigned int segno = curseg->segno;
1468 int dir = ALLOC_LEFT;
1469
1470 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1471 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1472 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1473 dir = ALLOC_RIGHT;
1474
1475 if (test_opt(sbi, NOHEAP))
1476 dir = ALLOC_RIGHT;
1477
1478 get_new_segment(sbi, &segno, new_sec, dir);
1479 curseg->next_segno = segno;
1480 reset_curseg(sbi, type, 1);
1481 curseg->alloc_type = LFS;
1482}
1483
1484static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1485 struct curseg_info *seg, block_t start)
1486{
1487 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1488 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1489 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1490 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1491 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1492 int i, pos;
1493
1494 for (i = 0; i < entries; i++)
1495 target_map[i] = ckpt_map[i] | cur_map[i];
1496
1497 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1498
1499 seg->next_blkoff = pos;
351df4b2
JK		 1500}
1501
0a8165d7 1502/*
351df4b2
JK		 1503 * If a segment is written by LFS manner, next block offset is just obtained
1504 * by increasing the current block offset. However, if a segment is written by
1505 * SSR manner, next block offset obtained by calling __next_free_blkoff
1506 */
1507static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1508 struct curseg_info *seg)
1509{
1510 if (seg->alloc_type == SSR)
1511 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1512 else
1513 seg->next_blkoff++;
1514}
1515
0a8165d7 1516/*
e1c42045 1517 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1518 * manner, so it should recover the existing segment information of valid blocks
1519 */
1520static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1521{
1522 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1523 struct curseg_info *curseg = CURSEG_I(sbi, type);
1524 unsigned int new_segno = curseg->next_segno;
1525 struct f2fs_summary_block *sum_node;
1526 struct page *sum_page;
1527
1528 write_sum_page(sbi, curseg->sum_blk,
1529 GET_SUM_BLOCK(sbi, curseg->segno));
1530 __set_test_and_inuse(sbi, new_segno);
1531
1532 mutex_lock(&dirty_i->seglist_lock);
1533 __remove_dirty_segment(sbi, new_segno, PRE);
1534 __remove_dirty_segment(sbi, new_segno, DIRTY);
1535 mutex_unlock(&dirty_i->seglist_lock);
1536
1537 reset_curseg(sbi, type, 1);
1538 curseg->alloc_type = SSR;
1539 __next_free_blkoff(sbi, curseg, 0);
1540
1541 if (reuse) {
1542 sum_page = get_sum_page(sbi, new_segno);
1543 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1544 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1545 f2fs_put_page(sum_page, 1);
1546 }
1547}
1548
43727527
JK		 1549static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1550{
1551 struct curseg_info *curseg = CURSEG_I(sbi, type);
1552 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
d27c3d89
CY		 1553 int i, cnt;
1554 bool reversed = false;
c192f7a4
JK		 1555
1556 /* need_SSR() already forces to do this */
1557 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
1558 return 1;
43727527 1559
70d625cb
JK		 1560 /* For node segments, let's do SSR more intensively */
1561 if (IS_NODESEG(type)) {
d27c3d89
CY		 1562 if (type >= CURSEG_WARM_NODE) {
1563 reversed = true;
1564 i = CURSEG_COLD_NODE;
1565 } else {
1566 i = CURSEG_HOT_NODE;
1567 }
1568 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 1569 } else {
d27c3d89
CY		 1570 if (type >= CURSEG_WARM_DATA) {
1571 reversed = true;
1572 i = CURSEG_COLD_DATA;
1573 } else {
1574 i = CURSEG_HOT_DATA;
1575 }
1576 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 1577 }
43727527 1578
d27c3d89 1579 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 1580 if (i == type)
1581 continue;
43727527 1582 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
c192f7a4 1583 BG_GC, i, SSR))
43727527 1584 return 1;
c192f7a4 1585 }
43727527
JK		 1586 return 0;
1587}
1588
351df4b2
JK		 1589/*
1590 * flush out current segment and replace it with new segment
1591 * This function should be returned with success, otherwise BUG
1592 */
1593static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1594 int type, bool force)
1595{
7b405275 1596 if (force)
351df4b2 1597 new_curseg(sbi, type, true);
5b6c6be2
JK		 1598 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1599 type == CURSEG_WARM_NODE)
351df4b2
JK		 1600 new_curseg(sbi, type, false);
1601 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1602 change_curseg(sbi, type, true);
1603 else
1604 new_curseg(sbi, type, false);
dcdfff65 1605
d0db7703 1606 stat_inc_seg_type(sbi, CURSEG_I(sbi, type));
351df4b2
JK		 1607}
1608
1609void allocate_new_segments(struct f2fs_sb_info *sbi)
1610{
6ae1be13
JK		 1611 struct curseg_info *curseg;
1612 unsigned int old_segno;
351df4b2
JK		 1613 int i;
1614
6ae1be13
JK		 1615 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1616 curseg = CURSEG_I(sbi, i);
1617 old_segno = curseg->segno;
1618 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1619 locate_dirty_segment(sbi, old_segno);
1620 }
351df4b2
JK		 1621}
1622
1623static const struct segment_allocation default_salloc_ops = {
1624 .allocate_segment = allocate_segment_by_default,
1625};
1626
25290fa5
JK		 1627bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1628{
1629 __u64 trim_start = cpc->trim_start;
1630 bool has_candidate = false;
1631
1632 mutex_lock(&SIT_I(sbi)->sentry_lock);
1633 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1634 if (add_discard_addrs(sbi, cpc, true)) {
1635 has_candidate = true;
1636 break;
1637 }
1638 }
1639 mutex_unlock(&SIT_I(sbi)->sentry_lock);
1640
1641 cpc->trim_start = trim_start;
1642 return has_candidate;
1643}
1644
4b2fecc8
JK		 1645int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1646{
f7ef9b83
JK		 1647 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1648 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1649 unsigned int start_segno, end_segno;
1650 struct cp_control cpc;
c34f42e2 1651 int err = 0;
4b2fecc8 1652
836b5a63 1653 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1654 return -EINVAL;
1655
9bd27ae4 1656 cpc.trimmed = 0;
7cd8558b 1657 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1658 goto out;
1659
ed214a11
YH		 1660 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1661 f2fs_msg(sbi->sb, KERN_WARNING,
1662 "Found FS corruption, run fsck to fix.");
1663 goto out;
1664 }
1665
4b2fecc8 1666 /* start/end segment number in main_area */
7cd8558b
JK		 1667 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1668 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1669 GET_SEGNO(sbi, end);
4b2fecc8 1670 cpc.reason = CP_DISCARD;
836b5a63 1671 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1672
1673 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1674 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1675 cpc.trim_start = start_segno;
a66cdd98
JK		 1676
1677 if (sbi->discard_blks == 0)
1678 break;
1679 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1680 cpc.trim_end = end_segno;
1681 else
1682 cpc.trim_end = min_t(unsigned int,
1683 rounddown(start_segno +
bba681cb
JK		 1684 BATCHED_TRIM_SEGMENTS(sbi),
1685 sbi->segs_per_sec) - 1, end_segno);
1686
1687 mutex_lock(&sbi->gc_mutex);
c34f42e2 1688 err = write_checkpoint(sbi, &cpc);
bba681cb 1689 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 1690 if (err)
1691 break;
74fa5f3d
CY		 1692
1693 schedule();
bba681cb 1694 }
4b2fecc8 1695out:
f7ef9b83 1696 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1697 return err;
4b2fecc8
JK		 1698}
1699
351df4b2
JK		 1700static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1701{
1702 struct curseg_info *curseg = CURSEG_I(sbi, type);
1703 if (curseg->next_blkoff < sbi->blocks_per_seg)
1704 return true;
1705 return false;
1706}
1707
1708static int __get_segment_type_2(struct page *page, enum page_type p_type)
1709{
1710 if (p_type == DATA)
1711 return CURSEG_HOT_DATA;
1712 else
1713 return CURSEG_HOT_NODE;
1714}
1715
1716static int __get_segment_type_4(struct page *page, enum page_type p_type)
1717{
1718 if (p_type == DATA) {
1719 struct inode *inode = page->mapping->host;
1720
1721 if (S_ISDIR(inode->i_mode))
1722 return CURSEG_HOT_DATA;
1723 else
1724 return CURSEG_COLD_DATA;
1725 } else {
a344b9fd
JK		 1726 if (IS_DNODE(page) && is_cold_node(page))
1727 return CURSEG_WARM_NODE;
351df4b2
JK		 1728 else
1729 return CURSEG_COLD_NODE;
1730 }
1731}
1732
1733static int __get_segment_type_6(struct page *page, enum page_type p_type)
1734{
1735 if (p_type == DATA) {
1736 struct inode *inode = page->mapping->host;
1737
1738 if (S_ISDIR(inode->i_mode))
1739 return CURSEG_HOT_DATA;
354a3399 1740 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1741 return CURSEG_COLD_DATA;
1742 else
1743 return CURSEG_WARM_DATA;
1744 } else {
1745 if (IS_DNODE(page))
1746 return is_cold_node(page) ? CURSEG_WARM_NODE :
1747 CURSEG_HOT_NODE;
1748 else
1749 return CURSEG_COLD_NODE;
1750 }
1751}
1752
1753static int __get_segment_type(struct page *page, enum page_type p_type)
1754{
4081363f 1755 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1756 case 2:
1757 return __get_segment_type_2(page, p_type);
1758 case 4:
1759 return __get_segment_type_4(page, p_type);
351df4b2 1760 }
12a67146 1761 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1762 f2fs_bug_on(F2FS_P_SB(page),
1763 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1764 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1765}
1766
bfad7c2d
JK		 1767void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1768 block_t old_blkaddr, block_t *new_blkaddr,
1769 struct f2fs_summary *sum, int type)
351df4b2
JK		 1770{
1771 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 1772 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 1773
1774 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1775 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 1776
1777 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 1778
4e6a8d9b
JK		 1779 f2fs_wait_discard_bio(sbi, *new_blkaddr);
1780
351df4b2
JK		 1781 /*
1782 * __add_sum_entry should be resided under the curseg_mutex
1783 * because, this function updates a summary entry in the
1784 * current summary block.
1785 */
e79efe3b 1786 __add_sum_entry(sbi, type, sum);
351df4b2 1787
351df4b2 1788 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1789
1790 stat_inc_block_count(sbi, curseg);
351df4b2
JK		 1791
1792 /*
1793 * SIT information should be updated before segment allocation,
1794 * since SSR needs latest valid block information.
1795 */
1796 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1797
3436c4bd
YS		 1798 if (!__has_curseg_space(sbi, type))
1799 sit_i->s_ops->allocate_segment(sbi, type, false);
1800
351df4b2
JK		 1801 mutex_unlock(&sit_i->sentry_lock);
1802
bfad7c2d 1803 if (page && IS_NODESEG(type))
351df4b2
JK		 1804 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1805
bfad7c2d
JK		 1806 mutex_unlock(&curseg->curseg_mutex);
1807}
1808
05ca3632 1809static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1810{
05ca3632 1811 int type = __get_segment_type(fio->page, fio->type);
0a595eba 1812 int err;
bfad7c2d 1813
7dfeaa32
JK		 1814 if (fio->type == NODE || fio->type == DATA)
1815 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
0a595eba 1816reallocate:
7a9d7548
CY		 1817 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1818 &fio->new_blkaddr, sum, type);
bfad7c2d 1819
351df4b2 1820 /* writeout dirty page into bdev */
0a595eba
JK		 1821 err = f2fs_submit_page_mbio(fio);
1822 if (err == -EAGAIN) {
1823 fio->old_blkaddr = fio->new_blkaddr;
1824 goto reallocate;
1825 }
7dfeaa32
JK		 1826
1827 if (fio->type == NODE || fio->type == DATA)
1828 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1829}
1830
577e3495 1831void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1832{
458e6197 1833 struct f2fs_io_info fio = {
05ca3632 1834 .sbi = sbi,
458e6197 1835 .type = META,
04d328de 1836 .op = REQ_OP_WRITE,
70fd7614 1837 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1838 .old_blkaddr = page->index,
1839 .new_blkaddr = page->index,
05ca3632 1840 .page = page,
4375a336 1841 .encrypted_page = NULL,
458e6197
JK		 1842 };
1843
2b947003 1844 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1845 fio.op_flags &= ~REQ_META;
2b947003 1846
351df4b2 1847 set_page_writeback(page);
05ca3632 1848 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1849}
1850
05ca3632 1851void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1852{
1853 struct f2fs_summary sum;
05ca3632 1854
351df4b2 1855 set_summary(&sum, nid, 0, 0);
05ca3632 1856 do_write_page(&sum, fio);
351df4b2
JK		 1857}
1858
05ca3632 1859void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1860{
05ca3632 1861 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1862 struct f2fs_summary sum;
1863 struct node_info ni;
1864
9850cf4a 1865 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1866 get_node_info(sbi, dn->nid, &ni);
1867 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1868 do_write_page(&sum, fio);
f28b3434 1869 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1870}
1871
05ca3632 1872void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1873{
7a9d7548 1874 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1875 stat_inc_inplace_blocks(fio->sbi);
1876 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1877}
1878
4356e48e 1879void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1880 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1881 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1882{
1883 struct sit_info *sit_i = SIT_I(sbi);
1884 struct curseg_info *curseg;
1885 unsigned int segno, old_cursegno;
1886 struct seg_entry *se;
1887 int type;
19f106bc 1888 unsigned short old_blkoff;
351df4b2
JK		 1889
1890 segno = GET_SEGNO(sbi, new_blkaddr);
1891 se = get_seg_entry(sbi, segno);
1892 type = se->type;
1893
19f106bc
CY		 1894 if (!recover_curseg) {
1895 /* for recovery flow */
1896 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1897 if (old_blkaddr == NULL_ADDR)
1898 type = CURSEG_COLD_DATA;
1899 else
1900 type = CURSEG_WARM_DATA;
1901 }
1902 } else {
1903 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1904 type = CURSEG_WARM_DATA;
1905 }
19f106bc 1906
351df4b2
JK		 1907 curseg = CURSEG_I(sbi, type);
1908
1909 mutex_lock(&curseg->curseg_mutex);
1910 mutex_lock(&sit_i->sentry_lock);
1911
1912 old_cursegno = curseg->segno;
19f106bc 1913 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1914
1915 /* change the current segment */
1916 if (segno != curseg->segno) {
1917 curseg->next_segno = segno;
1918 change_curseg(sbi, type, true);
1919 }
1920
491c0854 1921 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1922 __add_sum_entry(sbi, type, sum);
351df4b2 1923
28bc106b 1924 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1925 update_sit_entry(sbi, new_blkaddr, 1);
1926 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1927 update_sit_entry(sbi, old_blkaddr, -1);
1928
1929 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1930 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1931
351df4b2 1932 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1933
19f106bc
CY		 1934 if (recover_curseg) {
1935 if (old_cursegno != curseg->segno) {
1936 curseg->next_segno = old_cursegno;
1937 change_curseg(sbi, type, true);
1938 }
1939 curseg->next_blkoff = old_blkoff;
1940 }
1941
351df4b2
JK		 1942 mutex_unlock(&sit_i->sentry_lock);
1943 mutex_unlock(&curseg->curseg_mutex);
1944}
1945
528e3459
CY		 1946void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1947 block_t old_addr, block_t new_addr,
28bc106b
CY		 1948 unsigned char version, bool recover_curseg,
1949 bool recover_newaddr)
528e3459
CY		 1950{
1951 struct f2fs_summary sum;
1952
1953 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1954
28bc106b
CY		 1955 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1956 recover_curseg, recover_newaddr);
528e3459 1957
f28b3434 1958 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1959}
1960
93dfe2ac 1961void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1962 enum page_type type, bool ordered)
93dfe2ac 1963{
93dfe2ac 1964 if (PageWriteback(page)) {
4081363f
JK		 1965 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1966
942fd319
JK		 1967 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
1968 0, page->index, type, WRITE);
fec1d657
JK		 1969 if (ordered)
1970 wait_on_page_writeback(page);
1971 else
1972 wait_for_stable_page(page);
93dfe2ac
JK		 1973 }
1974}
1975
08b39fbd
CY		 1976void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1977 block_t blkaddr)
1978{
1979 struct page *cpage;
1980
5d4c0af4 1981 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 1982 return;
1983
08b39fbd
CY		 1984 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1985 if (cpage) {
fec1d657 1986 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1987 f2fs_put_page(cpage, 1);
1988 }
1989}
1990
351df4b2
JK		 1991static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1992{
1993 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1994 struct curseg_info *seg_i;
1995 unsigned char *kaddr;
1996 struct page *page;
1997 block_t start;
1998 int i, j, offset;
1999
2000 start = start_sum_block(sbi);
2001
2002 page = get_meta_page(sbi, start++);
2003 kaddr = (unsigned char *)page_address(page);
2004
2005 /* Step 1: restore nat cache */
2006 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2007 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2008
2009 /* Step 2: restore sit cache */
2010 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2011 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2012 offset = 2 * SUM_JOURNAL_SIZE;
2013
2014 /* Step 3: restore summary entries */
2015 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2016 unsigned short blk_off;
2017 unsigned int segno;
2018
2019 seg_i = CURSEG_I(sbi, i);
2020 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2021 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2022 seg_i->next_segno = segno;
2023 reset_curseg(sbi, i, 0);
2024 seg_i->alloc_type = ckpt->alloc_type[i];
2025 seg_i->next_blkoff = blk_off;
2026
2027 if (seg_i->alloc_type == SSR)
2028 blk_off = sbi->blocks_per_seg;
2029
2030 for (j = 0; j < blk_off; j++) {
2031 struct f2fs_summary *s;
2032 s = (struct f2fs_summary *)(kaddr + offset);
2033 seg_i->sum_blk->entries[j] = *s;
2034 offset += SUMMARY_SIZE;
09cbfeaf 2035 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2036 SUM_FOOTER_SIZE)
2037 continue;
2038
2039 f2fs_put_page(page, 1);
2040 page = NULL;
2041
2042 page = get_meta_page(sbi, start++);
2043 kaddr = (unsigned char *)page_address(page);
2044 offset = 0;
2045 }
2046 }
2047 f2fs_put_page(page, 1);
2048 return 0;
2049}
2050
2051static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2052{
2053 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2054 struct f2fs_summary_block *sum;
2055 struct curseg_info *curseg;
2056 struct page *new;
2057 unsigned short blk_off;
2058 unsigned int segno = 0;
2059 block_t blk_addr = 0;
2060
2061 /* get segment number and block addr */
2062 if (IS_DATASEG(type)) {
2063 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2064 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2065 CURSEG_HOT_DATA]);
119ee914 2066 if (__exist_node_summaries(sbi))
351df4b2
JK		 2067 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2068 else
2069 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2070 } else {
2071 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2072 CURSEG_HOT_NODE]);
2073 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2074 CURSEG_HOT_NODE]);
119ee914 2075 if (__exist_node_summaries(sbi))
351df4b2
JK		 2076 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2077 type - CURSEG_HOT_NODE);
2078 else
2079 blk_addr = GET_SUM_BLOCK(sbi, segno);
2080 }
2081
2082 new = get_meta_page(sbi, blk_addr);
2083 sum = (struct f2fs_summary_block *)page_address(new);
2084
2085 if (IS_NODESEG(type)) {
119ee914 2086 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2087 struct f2fs_summary *ns = &sum->entries[0];
2088 int i;
2089 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2090 ns->version = 0;
2091 ns->ofs_in_node = 0;
2092 }
2093 } else {
d653788a
GZ		 2094 int err;
2095
2096 err = restore_node_summary(sbi, segno, sum);
2097 if (err) {
351df4b2 2098 f2fs_put_page(new, 1);
d653788a 2099 return err;
351df4b2
JK		 2100 }
2101 }
2102 }
2103
2104 /* set uncompleted segment to curseg */
2105 curseg = CURSEG_I(sbi, type);
2106 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2107
2108 /* update journal info */
2109 down_write(&curseg->journal_rwsem);
2110 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2111 up_write(&curseg->journal_rwsem);
2112
2113 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2114 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2115 curseg->next_segno = segno;
2116 reset_curseg(sbi, type, 0);
2117 curseg->alloc_type = ckpt->alloc_type[type];
2118 curseg->next_blkoff = blk_off;
2119 mutex_unlock(&curseg->curseg_mutex);
2120 f2fs_put_page(new, 1);
2121 return 0;
2122}
2123
2124static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2125{
2126 int type = CURSEG_HOT_DATA;
e4fc5fbf 2127 int err;
351df4b2 2128
aaec2b1d 2129 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2130 int npages = npages_for_summary_flush(sbi, true);
2131
2132 if (npages >= 2)
2133 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2134 META_CP, true);
3fa06d7b 2135
351df4b2
JK		 2136 /* restore for compacted data summary */
2137 if (read_compacted_summaries(sbi))
2138 return -EINVAL;
2139 type = CURSEG_HOT_NODE;
2140 }
2141
119ee914 2142 if (__exist_node_summaries(sbi))
3fa06d7b 2143 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2144 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2145
e4fc5fbf
CY		 2146 for (; type <= CURSEG_COLD_NODE; type++) {
2147 err = read_normal_summaries(sbi, type);
2148 if (err)
2149 return err;
2150 }
2151
351df4b2
JK		 2152 return 0;
2153}
2154
2155static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2156{
2157 struct page *page;
2158 unsigned char *kaddr;
2159 struct f2fs_summary *summary;
2160 struct curseg_info *seg_i;
2161 int written_size = 0;
2162 int i, j;
2163
2164 page = grab_meta_page(sbi, blkaddr++);
2165 kaddr = (unsigned char *)page_address(page);
2166
2167 /* Step 1: write nat cache */
2168 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2169 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2170 written_size += SUM_JOURNAL_SIZE;
2171
2172 /* Step 2: write sit cache */
2173 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2174 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2175 written_size += SUM_JOURNAL_SIZE;
2176
351df4b2
JK		 2177 /* Step 3: write summary entries */
2178 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2179 unsigned short blkoff;
2180 seg_i = CURSEG_I(sbi, i);
2181 if (sbi->ckpt->alloc_type[i] == SSR)
2182 blkoff = sbi->blocks_per_seg;
2183 else
2184 blkoff = curseg_blkoff(sbi, i);
2185
2186 for (j = 0; j < blkoff; j++) {
2187 if (!page) {
2188 page = grab_meta_page(sbi, blkaddr++);
2189 kaddr = (unsigned char *)page_address(page);
2190 written_size = 0;
2191 }
2192 summary = (struct f2fs_summary *)(kaddr + written_size);
2193 *summary = seg_i->sum_blk->entries[j];
2194 written_size += SUMMARY_SIZE;
351df4b2 2195
09cbfeaf 2196 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2197 SUM_FOOTER_SIZE)
2198 continue;
2199
e8d61a74 2200 set_page_dirty(page);
351df4b2
JK		 2201 f2fs_put_page(page, 1);
2202 page = NULL;
2203 }
2204 }
e8d61a74
CY		 2205 if (page) {
2206 set_page_dirty(page);
351df4b2 2207 f2fs_put_page(page, 1);
e8d61a74 2208 }
351df4b2
JK		 2209}
2210
2211static void write_normal_summaries(struct f2fs_sb_info *sbi,
2212 block_t blkaddr, int type)
2213{
2214 int i, end;
2215 if (IS_DATASEG(type))
2216 end = type + NR_CURSEG_DATA_TYPE;
2217 else
2218 end = type + NR_CURSEG_NODE_TYPE;
2219
b7ad7512
CY		 2220 for (i = type; i < end; i++)
2221 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2222}
2223
2224void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2225{
aaec2b1d 2226 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2227 write_compacted_summaries(sbi, start_blk);
2228 else
2229 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2230}
2231
2232void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2233{
119ee914 2234 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2235}
2236
dfc08a12 2237int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2238 unsigned int val, int alloc)
2239{
2240 int i;
2241
2242 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2243 for (i = 0; i < nats_in_cursum(journal); i++) {
2244 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2245 return i;
2246 }
dfc08a12
CY		 2247 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2248 return update_nats_in_cursum(journal, 1);
351df4b2 2249 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2250 for (i = 0; i < sits_in_cursum(journal); i++)
2251 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2252 return i;
dfc08a12
CY		 2253 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2254 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2255 }
2256 return -1;
2257}
2258
2259static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2260 unsigned int segno)
2261{
2cc22186 2262 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2263}
2264
2265static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2266 unsigned int start)
2267{
2268 struct sit_info *sit_i = SIT_I(sbi);
2269 struct page *src_page, *dst_page;
2270 pgoff_t src_off, dst_off;
2271 void *src_addr, *dst_addr;
2272
2273 src_off = current_sit_addr(sbi, start);
2274 dst_off = next_sit_addr(sbi, src_off);
2275
2276 /* get current sit block page without lock */
2277 src_page = get_meta_page(sbi, src_off);
2278 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2279 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2280
2281 src_addr = page_address(src_page);
2282 dst_addr = page_address(dst_page);
09cbfeaf 2283 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2284
2285 set_page_dirty(dst_page);
2286 f2fs_put_page(src_page, 1);
2287
2288 set_to_next_sit(sit_i, start);
2289
2290 return dst_page;
2291}
2292
184a5cd2
CY		 2293static struct sit_entry_set *grab_sit_entry_set(void)
2294{
2295 struct sit_entry_set *ses =
80c54505 2296 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2297
2298 ses->entry_cnt = 0;
2299 INIT_LIST_HEAD(&ses->set_list);
2300 return ses;
2301}
2302
2303static void release_sit_entry_set(struct sit_entry_set *ses)
2304{
2305 list_del(&ses->set_list);
2306 kmem_cache_free(sit_entry_set_slab, ses);
2307}
2308
2309static void adjust_sit_entry_set(struct sit_entry_set *ses,
2310 struct list_head *head)
2311{
2312 struct sit_entry_set *next = ses;
2313
2314 if (list_is_last(&ses->set_list, head))
2315 return;
2316
2317 list_for_each_entry_continue(next, head, set_list)
2318 if (ses->entry_cnt <= next->entry_cnt)
2319 break;
2320
2321 list_move_tail(&ses->set_list, &next->set_list);
2322}
2323
2324static void add_sit_entry(unsigned int segno, struct list_head *head)
2325{
2326 struct sit_entry_set *ses;
2327 unsigned int start_segno = START_SEGNO(segno);
2328
2329 list_for_each_entry(ses, head, set_list) {
2330 if (ses->start_segno == start_segno) {
2331 ses->entry_cnt++;
2332 adjust_sit_entry_set(ses, head);
2333 return;
2334 }
2335 }
2336
2337 ses = grab_sit_entry_set();
2338
2339 ses->start_segno = start_segno;
2340 ses->entry_cnt++;
2341 list_add(&ses->set_list, head);
2342}
2343
2344static void add_sits_in_set(struct f2fs_sb_info *sbi)
2345{
2346 struct f2fs_sm_info *sm_info = SM_I(sbi);
2347 struct list_head *set_list = &sm_info->sit_entry_set;
2348 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2349 unsigned int segno;
2350
7cd8558b 2351 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2352 add_sit_entry(segno, set_list);
2353}
2354
2355static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2356{
2357 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2358 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2359 int i;
2360
b7ad7512 2361 down_write(&curseg->journal_rwsem);
dfc08a12 2362 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2363 unsigned int segno;
2364 bool dirtied;
2365
dfc08a12 2366 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2367 dirtied = __mark_sit_entry_dirty(sbi, segno);
2368
2369 if (!dirtied)
2370 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2371 }
dfc08a12 2372 update_sits_in_cursum(journal, -i);
b7ad7512 2373 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2374}
2375
0a8165d7 2376/*
351df4b2
JK		 2377 * CP calls this function, which flushes SIT entries including sit_journal,
2378 * and moves prefree segs to free segs.
2379 */
4b2fecc8 2380void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2381{
2382 struct sit_info *sit_i = SIT_I(sbi);
2383 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2384 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2385 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2386 struct sit_entry_set *ses, *tmp;
2387 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2388 bool to_journal = true;
4b2fecc8 2389 struct seg_entry *se;
351df4b2 2390
351df4b2
JK		 2391 mutex_lock(&sit_i->sentry_lock);
2392
2b11a74b
WL		 2393 if (!sit_i->dirty_sentries)
2394 goto out;
2395
351df4b2 2396 /*
184a5cd2
CY		 2397 * add and account sit entries of dirty bitmap in sit entry
2398 * set temporarily
351df4b2 2399 */
184a5cd2 2400 add_sits_in_set(sbi);
351df4b2 2401
184a5cd2
CY		 2402 /*
2403 * if there are no enough space in journal to store dirty sit
2404 * entries, remove all entries from journal and add and account
2405 * them in sit entry set.
2406 */
dfc08a12 2407 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2408 remove_sits_in_journal(sbi);
b2955550 2409
184a5cd2
CY		 2410 /*
2411 * there are two steps to flush sit entries:
2412 * #1, flush sit entries to journal in current cold data summary block.
2413 * #2, flush sit entries to sit page.
2414 */
2415 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2416 struct page *page = NULL;
184a5cd2
CY		 2417 struct f2fs_sit_block *raw_sit = NULL;
2418 unsigned int start_segno = ses->start_segno;
2419 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2420 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2421 unsigned int segno = start_segno;
2422
2423 if (to_journal &&
dfc08a12 2424 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2425 to_journal = false;
2426
b7ad7512
CY		 2427 if (to_journal) {
2428 down_write(&curseg->journal_rwsem);
2429 } else {
184a5cd2
CY		 2430 page = get_next_sit_page(sbi, start_segno);
2431 raw_sit = page_address(page);
351df4b2 2432 }
351df4b2 2433
184a5cd2
CY		 2434 /* flush dirty sit entries in region of current sit set */
2435 for_each_set_bit_from(segno, bitmap, end) {
2436 int offset, sit_offset;
4b2fecc8
JK		 2437
2438 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2439
2440 /* add discard candidates */
d7bc2484 2441 if (cpc->reason != CP_DISCARD) {
4b2fecc8 2442 cpc->trim_start = segno;
25290fa5 2443 add_discard_addrs(sbi, cpc, false);
4b2fecc8 2444 }
184a5cd2
CY		 2445
2446 if (to_journal) {
dfc08a12 2447 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2448 SIT_JOURNAL, segno, 1);
2449 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2450 segno_in_journal(journal, offset) =
184a5cd2
CY		 2451 cpu_to_le32(segno);
2452 seg_info_to_raw_sit(se,
dfc08a12 2453 &sit_in_journal(journal, offset));
184a5cd2
CY		 2454 } else {
2455 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2456 seg_info_to_raw_sit(se,
2457 &raw_sit->entries[sit_offset]);
2458 }
351df4b2 2459
184a5cd2
CY		 2460 __clear_bit(segno, bitmap);
2461 sit_i->dirty_sentries--;
2462 ses->entry_cnt--;
351df4b2
JK		 2463 }
2464
b7ad7512
CY		 2465 if (to_journal)
2466 up_write(&curseg->journal_rwsem);
2467 else
184a5cd2
CY		 2468 f2fs_put_page(page, 1);
2469
2470 f2fs_bug_on(sbi, ses->entry_cnt);
2471 release_sit_entry_set(ses);
351df4b2 2472 }
184a5cd2
CY		 2473
2474 f2fs_bug_on(sbi, !list_empty(head));
2475 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2476out:
4b2fecc8 2477 if (cpc->reason == CP_DISCARD) {
650d3c4e
YH		 2478 __u64 trim_start = cpc->trim_start;
2479
4b2fecc8 2480 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 2481 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 2482
2483 cpc->trim_start = trim_start;
4b2fecc8 2484 }
351df4b2 2485 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2486
351df4b2
JK		 2487 set_prefree_as_free_segments(sbi);
2488}
2489
2490static int build_sit_info(struct f2fs_sb_info *sbi)
2491{
2492 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 2493 struct sit_info *sit_i;
2494 unsigned int sit_segs, start;
ae27d62e 2495 char *src_bitmap;
351df4b2
JK		 2496 unsigned int bitmap_size;
2497
2498 /* allocate memory for SIT information */
2499 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2500 if (!sit_i)
2501 return -ENOMEM;
2502
2503 SM_I(sbi)->sit_info = sit_i;
2504
39307a8e
JK		 2505 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2506 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2507 if (!sit_i->sentries)
2508 return -ENOMEM;
2509
7cd8558b 2510 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2511 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2512 if (!sit_i->dirty_sentries_bitmap)
2513 return -ENOMEM;
2514
7cd8558b 2515 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2516 sit_i->sentries[start].cur_valid_map
2517 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2518 sit_i->sentries[start].ckpt_valid_map
2519 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 2520 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 2521 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 2522 return -ENOMEM;
3e025740 2523
355e7891
CY		 2524#ifdef CONFIG_F2FS_CHECK_FS
2525 sit_i->sentries[start].cur_valid_map_mir
2526 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2527 if (!sit_i->sentries[start].cur_valid_map_mir)
2528 return -ENOMEM;
2529#endif
2530
3e025740
JK		 2531 if (f2fs_discard_en(sbi)) {
2532 sit_i->sentries[start].discard_map
2533 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2534 if (!sit_i->sentries[start].discard_map)
2535 return -ENOMEM;
2536 }
351df4b2
JK		 2537 }
2538
60a3b782
JK		 2539 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2540 if (!sit_i->tmp_map)
2541 return -ENOMEM;
2542
351df4b2 2543 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2544 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2545 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2546 if (!sit_i->sec_entries)
2547 return -ENOMEM;
2548 }
2549
2550 /* get information related with SIT */
2551 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2552
2553 /* setup SIT bitmap from ckeckpoint pack */
2554 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2555 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2556
ae27d62e
CY		 2557 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2558 if (!sit_i->sit_bitmap)
351df4b2 2559 return -ENOMEM;
351df4b2 2560
ae27d62e
CY		 2561#ifdef CONFIG_F2FS_CHECK_FS
2562 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2563 if (!sit_i->sit_bitmap_mir)
2564 return -ENOMEM;
2565#endif
2566
351df4b2
JK		 2567 /* init SIT information */
2568 sit_i->s_ops = &default_salloc_ops;
2569
2570 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2571 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 2572 sit_i->written_valid_blocks = 0;
351df4b2
JK		 2573 sit_i->bitmap_size = bitmap_size;
2574 sit_i->dirty_sentries = 0;
2575 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2576 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2577 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2578 mutex_init(&sit_i->sentry_lock);
2579 return 0;
2580}
2581
2582static int build_free_segmap(struct f2fs_sb_info *sbi)
2583{
351df4b2
JK		 2584 struct free_segmap_info *free_i;
2585 unsigned int bitmap_size, sec_bitmap_size;
2586
2587 /* allocate memory for free segmap information */
2588 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2589 if (!free_i)
2590 return -ENOMEM;
2591
2592 SM_I(sbi)->free_info = free_i;
2593
7cd8558b 2594 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2595 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2596 if (!free_i->free_segmap)
2597 return -ENOMEM;
2598
7cd8558b 2599 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2600 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2601 if (!free_i->free_secmap)
2602 return -ENOMEM;
2603
2604 /* set all segments as dirty temporarily */
2605 memset(free_i->free_segmap, 0xff, bitmap_size);
2606 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2607
2608 /* init free segmap information */
7cd8558b 2609 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2610 free_i->free_segments = 0;
2611 free_i->free_sections = 0;
1a118ccf 2612 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2613 return 0;
2614}
2615
2616static int build_curseg(struct f2fs_sb_info *sbi)
2617{
1042d60f 2618 struct curseg_info *array;
351df4b2
JK		 2619 int i;
2620
b434babf 2621 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2622 if (!array)
2623 return -ENOMEM;
2624
2625 SM_I(sbi)->curseg_array = array;
2626
2627 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2628 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2629 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2630 if (!array[i].sum_blk)
2631 return -ENOMEM;
b7ad7512
CY		 2632 init_rwsem(&array[i].journal_rwsem);
2633 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2634 GFP_KERNEL);
2635 if (!array[i].journal)
2636 return -ENOMEM;
351df4b2
JK		 2637 array[i].segno = NULL_SEGNO;
2638 array[i].next_blkoff = 0;
2639 }
2640 return restore_curseg_summaries(sbi);
2641}
2642
2643static void build_sit_entries(struct f2fs_sb_info *sbi)
2644{
2645 struct sit_info *sit_i = SIT_I(sbi);
2646 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2647 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 2648 struct seg_entry *se;
2649 struct f2fs_sit_entry sit;
74de593a
CY		 2650 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2651 unsigned int i, start, end;
2652 unsigned int readed, start_blk = 0;
351df4b2 2653
74de593a 2654 do {
664ba972
JK		 2655 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2656 META_SIT, true);
74de593a
CY		 2657
2658 start = start_blk * sit_i->sents_per_block;
2659 end = (start_blk + readed) * sit_i->sents_per_block;
2660
7cd8558b 2661 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 2662 struct f2fs_sit_block *sit_blk;
74de593a
CY		 2663 struct page *page;
2664
9c094040 2665 se = &sit_i->sentries[start];
74de593a
CY		 2666 page = get_current_sit_page(sbi, start);
2667 sit_blk = (struct f2fs_sit_block *)page_address(page);
2668 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2669 f2fs_put_page(page, 1);
d600af23 2670
74de593a
CY		 2671 check_block_count(sbi, start, &sit);
2672 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2673
2674 /* build discard map only one time */
3e025740
JK		 2675 if (f2fs_discard_en(sbi)) {
2676 memcpy(se->discard_map, se->cur_valid_map,
2677 SIT_VBLOCK_MAP_SIZE);
2678 sbi->discard_blks += sbi->blocks_per_seg -
2679 se->valid_blocks;
2680 }
a66cdd98 2681
d600af23
CY		 2682 if (sbi->segs_per_sec > 1)
2683 get_sec_entry(sbi, start)->valid_blocks +=
2684 se->valid_blocks;
351df4b2 2685 }
74de593a
CY		 2686 start_blk += readed;
2687 } while (start_blk < sit_blk_cnt);
d600af23
CY		 2688
2689 down_read(&curseg->journal_rwsem);
2690 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 2691 unsigned int old_valid_blocks;
2692
2693 start = le32_to_cpu(segno_in_journal(journal, i));
2694 se = &sit_i->sentries[start];
2695 sit = sit_in_journal(journal, i);
2696
2697 old_valid_blocks = se->valid_blocks;
2698
2699 check_block_count(sbi, start, &sit);
2700 seg_info_from_raw_sit(se, &sit);
2701
2702 if (f2fs_discard_en(sbi)) {
2703 memcpy(se->discard_map, se->cur_valid_map,
2704 SIT_VBLOCK_MAP_SIZE);
2705 sbi->discard_blks += old_valid_blocks -
2706 se->valid_blocks;
2707 }
2708
2709 if (sbi->segs_per_sec > 1)
2710 get_sec_entry(sbi, start)->valid_blocks +=
2711 se->valid_blocks - old_valid_blocks;
2712 }
2713 up_read(&curseg->journal_rwsem);
351df4b2
JK		 2714}
2715
2716static void init_free_segmap(struct f2fs_sb_info *sbi)
2717{
2718 unsigned int start;
2719 int type;
2720
7cd8558b 2721 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2722 struct seg_entry *sentry = get_seg_entry(sbi, start);
2723 if (!sentry->valid_blocks)
2724 __set_free(sbi, start);
c79b7ff1
JK		 2725 else
2726 SIT_I(sbi)->written_valid_blocks +=
2727 sentry->valid_blocks;
351df4b2
JK		 2728 }
2729
2730 /* set use the current segments */
2731 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2732 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2733 __set_test_and_inuse(sbi, curseg_t->segno);
2734 }
2735}
2736
2737static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2738{
2739 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2740 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2741 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2742 unsigned short valid_blocks;
2743
8736fbf0 2744 while (1) {
351df4b2 2745 /* find dirty segment based on free segmap */
7cd8558b
JK		 2746 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2747 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2748 break;
2749 offset = segno + 1;
2750 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2751 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2752 continue;
ec325b52
JK		 2753 if (valid_blocks > sbi->blocks_per_seg) {
2754 f2fs_bug_on(sbi, 1);
2755 continue;
2756 }
351df4b2
JK		 2757 mutex_lock(&dirty_i->seglist_lock);
2758 __locate_dirty_segment(sbi, segno, DIRTY);
2759 mutex_unlock(&dirty_i->seglist_lock);
2760 }
2761}
2762
5ec4e49f 2763static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2764{
2765 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2766 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2767
39307a8e 2768 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2769 if (!dirty_i->victim_secmap)
351df4b2
JK		 2770 return -ENOMEM;
2771 return 0;
2772}
2773
2774static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2775{
2776 struct dirty_seglist_info *dirty_i;
2777 unsigned int bitmap_size, i;
2778
2779 /* allocate memory for dirty segments list information */
2780 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2781 if (!dirty_i)
2782 return -ENOMEM;
2783
2784 SM_I(sbi)->dirty_info = dirty_i;
2785 mutex_init(&dirty_i->seglist_lock);
2786
7cd8558b 2787 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2788
2789 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2790 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2791 if (!dirty_i->dirty_segmap[i])
2792 return -ENOMEM;
2793 }
2794
2795 init_dirty_segmap(sbi);
5ec4e49f 2796 return init_victim_secmap(sbi);
351df4b2
JK		 2797}
2798
0a8165d7 2799/*
351df4b2
JK		 2800 * Update min, max modified time for cost-benefit GC algorithm
2801 */
2802static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2803{
2804 struct sit_info *sit_i = SIT_I(sbi);
2805 unsigned int segno;
2806
2807 mutex_lock(&sit_i->sentry_lock);
2808
2809 sit_i->min_mtime = LLONG_MAX;
2810
7cd8558b 2811 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2812 unsigned int i;
2813 unsigned long long mtime = 0;
2814
2815 for (i = 0; i < sbi->segs_per_sec; i++)
2816 mtime += get_seg_entry(sbi, segno + i)->mtime;
2817
2818 mtime = div_u64(mtime, sbi->segs_per_sec);
2819
2820 if (sit_i->min_mtime > mtime)
2821 sit_i->min_mtime = mtime;
2822 }
2823 sit_i->max_mtime = get_mtime(sbi);
2824 mutex_unlock(&sit_i->sentry_lock);
2825}
2826
2827int build_segment_manager(struct f2fs_sb_info *sbi)
2828{
2829 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2830 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2831 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2832 int err;
2833
2834 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2835 if (!sm_info)
2836 return -ENOMEM;
2837
2838 /* init sm info */
2839 sbi->sm_info = sm_info;
351df4b2
JK		 2840 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2841 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2842 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2843 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2844 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2845 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2846 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2847 sm_info->rec_prefree_segments = sm_info->main_segments *
2848 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2849 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2850 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2851
52763a4b
JK		 2852 if (!test_opt(sbi, LFS))
2853 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2854 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2855 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2856
bba681cb
JK		 2857 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2858
184a5cd2
CY		 2859 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2860
b270ad6f 2861 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2862 err = create_flush_cmd_control(sbi);
2863 if (err)
a688b9d9 2864 return err;
6b4afdd7
JK		 2865 }
2866
0b54fb84
JK		 2867 err = create_discard_cmd_control(sbi);
2868 if (err)
2869 return err;
2870
351df4b2
JK		 2871 err = build_sit_info(sbi);
2872 if (err)
2873 return err;
2874 err = build_free_segmap(sbi);
2875 if (err)
2876 return err;
2877 err = build_curseg(sbi);
2878 if (err)
2879 return err;
2880
2881 /* reinit free segmap based on SIT */
2882 build_sit_entries(sbi);
2883
2884 init_free_segmap(sbi);
2885 err = build_dirty_segmap(sbi);
2886 if (err)
2887 return err;
2888
2889 init_min_max_mtime(sbi);
2890 return 0;
2891}
2892
2893static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2894 enum dirty_type dirty_type)
2895{
2896 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2897
2898 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2899 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2900 dirty_i->nr_dirty[dirty_type] = 0;
2901 mutex_unlock(&dirty_i->seglist_lock);
2902}
2903
5ec4e49f 2904static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2905{
2906 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2907 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2908}
2909
2910static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2911{
2912 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2913 int i;
2914
2915 if (!dirty_i)
2916 return;
2917
2918 /* discard pre-free/dirty segments list */
2919 for (i = 0; i < NR_DIRTY_TYPE; i++)
2920 discard_dirty_segmap(sbi, i);
2921
5ec4e49f 2922 destroy_victim_secmap(sbi);
351df4b2
JK		 2923 SM_I(sbi)->dirty_info = NULL;
2924 kfree(dirty_i);
2925}
2926
2927static void destroy_curseg(struct f2fs_sb_info *sbi)
2928{
2929 struct curseg_info *array = SM_I(sbi)->curseg_array;
2930 int i;
2931
2932 if (!array)
2933 return;
2934 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2935 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2936 kfree(array[i].sum_blk);
b7ad7512
CY		 2937 kfree(array[i].journal);
2938 }
351df4b2
JK		 2939 kfree(array);
2940}
2941
2942static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2943{
2944 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2945 if (!free_i)
2946 return;
2947 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2948 kvfree(free_i->free_segmap);
2949 kvfree(free_i->free_secmap);
351df4b2
JK		 2950 kfree(free_i);
2951}
2952
2953static void destroy_sit_info(struct f2fs_sb_info *sbi)
2954{
2955 struct sit_info *sit_i = SIT_I(sbi);
2956 unsigned int start;
2957
2958 if (!sit_i)
2959 return;
2960
2961 if (sit_i->sentries) {
7cd8558b 2962 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 2963 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 2964#ifdef CONFIG_F2FS_CHECK_FS
2965 kfree(sit_i->sentries[start].cur_valid_map_mir);
2966#endif
351df4b2 2967 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2968 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2969 }
2970 }
60a3b782
JK		 2971 kfree(sit_i->tmp_map);
2972
39307a8e
JK		 2973 kvfree(sit_i->sentries);
2974 kvfree(sit_i->sec_entries);
2975 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2976
2977 SM_I(sbi)->sit_info = NULL;
2978 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 2979#ifdef CONFIG_F2FS_CHECK_FS
2980 kfree(sit_i->sit_bitmap_mir);
2981#endif
351df4b2
JK		 2982 kfree(sit_i);
2983}
2984
2985void destroy_segment_manager(struct f2fs_sb_info *sbi)
2986{
2987 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2988
3b03f724
CY		 2989 if (!sm_info)
2990 return;
5eba8c5d 2991 destroy_flush_cmd_control(sbi, true);
0b54fb84 2992 destroy_discard_cmd_control(sbi, true);
351df4b2
JK		 2993 destroy_dirty_segmap(sbi);
2994 destroy_curseg(sbi);
2995 destroy_free_segmap(sbi);
2996 destroy_sit_info(sbi);
2997 sbi->sm_info = NULL;
2998 kfree(sm_info);
2999}
7fd9e544
JK		 3000
3001int __init create_segment_manager_caches(void)
3002{
3003 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3004 sizeof(struct discard_entry));
7fd9e544 3005 if (!discard_entry_slab)
184a5cd2
CY		 3006 goto fail;
3007
b01a9201
JK		 3008 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3009 sizeof(struct discard_cmd));
3010 if (!discard_cmd_slab)
6ab2a308 3011 goto destroy_discard_entry;
275b66b0 3012
184a5cd2 3013 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3014 sizeof(struct sit_entry_set));
184a5cd2 3015 if (!sit_entry_set_slab)
b01a9201 3016 goto destroy_discard_cmd;
88b88a66
JK		 3017
3018 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3019 sizeof(struct inmem_pages));
3020 if (!inmem_entry_slab)
3021 goto destroy_sit_entry_set;
7fd9e544 3022 return 0;
184a5cd2 3023
88b88a66
JK		 3024destroy_sit_entry_set:
3025 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3026destroy_discard_cmd:
3027 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3028destroy_discard_entry:
184a5cd2
CY		 3029 kmem_cache_destroy(discard_entry_slab);
3030fail:
3031 return -ENOMEM;
7fd9e544
JK		 3032}
3033
3034void destroy_segment_manager_caches(void)
3035{
184a5cd2 3036 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3037 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3038 kmem_cache_destroy(discard_entry_slab);
88b88a66 3039 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3040}
Ubuntu Artful kernel mirror