]> 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: clean up destroy_discard_cmd_control
[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
8c242db9
JK		 253void drop_inmem_page(struct inode *inode, struct page *page)
254{
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
257 struct list_head *head = &fi->inmem_pages;
258 struct inmem_pages *cur = NULL;
259
260 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
261
262 mutex_lock(&fi->inmem_lock);
263 list_for_each_entry(cur, head, list) {
264 if (cur->page == page)
265 break;
266 }
267
268 f2fs_bug_on(sbi, !cur || cur->page != page);
269 list_del(&cur->list);
270 mutex_unlock(&fi->inmem_lock);
271
272 dec_page_count(sbi, F2FS_INMEM_PAGES);
273 kmem_cache_free(inmem_entry_slab, cur);
274
275 ClearPageUptodate(page);
276 set_page_private(page, 0);
277 ClearPagePrivate(page);
278 f2fs_put_page(page, 0);
279
280 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
281}
282
28bc106b
CY		 283static int __commit_inmem_pages(struct inode *inode,
284 struct list_head *revoke_list)
88b88a66
JK		 285{
286 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
287 struct f2fs_inode_info *fi = F2FS_I(inode);
288 struct inmem_pages *cur, *tmp;
88b88a66 289 struct f2fs_io_info fio = {
05ca3632 290 .sbi = sbi,
88b88a66 291 .type = DATA,
04d328de 292 .op = REQ_OP_WRITE,
70fd7614 293 .op_flags = REQ_SYNC | REQ_PRIO,
4375a336 294 .encrypted_page = NULL,
88b88a66 295 };
942fd319 296 pgoff_t last_idx = ULONG_MAX;
edb27dee 297 int err = 0;
88b88a66 298
88b88a66 299 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 300 struct page *page = cur->page;
301
302 lock_page(page);
303 if (page->mapping == inode->i_mapping) {
304 trace_f2fs_commit_inmem_page(page, INMEM);
305
306 set_page_dirty(page);
307 f2fs_wait_on_page_writeback(page, DATA, true);
933439c8 308 if (clear_page_dirty_for_io(page)) {
29b96b54 309 inode_dec_dirty_pages(inode);
933439c8
CY		 310 remove_dirty_inode(inode);
311 }
28bc106b
CY		 312
313 fio.page = page;
29b96b54
CY		 314 err = do_write_data_page(&fio);
315 if (err) {
28bc106b 316 unlock_page(page);
29b96b54 317 break;
70c640b1 318 }
29b96b54 319
28bc106b
CY		 320 /* record old blkaddr for revoking */
321 cur->old_addr = fio.old_blkaddr;
942fd319 322 last_idx = page->index;
28bc106b
CY		 323 }
324 unlock_page(page);
325 list_move_tail(&cur->list, revoke_list);
88b88a66 326 }
29b96b54 327
942fd319
JK		 328 if (last_idx != ULONG_MAX)
329 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx,
330 DATA, WRITE);
28bc106b
CY		 331
332 if (!err)
333 __revoke_inmem_pages(inode, revoke_list, false, false);
334
29b96b54
CY		 335 return err;
336}
337
338int commit_inmem_pages(struct inode *inode)
339{
340 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
341 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 342 struct list_head revoke_list;
343 int err;
29b96b54 344
28bc106b 345 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 346 f2fs_balance_fs(sbi, true);
347 f2fs_lock_op(sbi);
348
5fe45743
CY		 349 set_inode_flag(inode, FI_ATOMIC_COMMIT);
350
29b96b54 351 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 352 err = __commit_inmem_pages(inode, &revoke_list);
353 if (err) {
354 int ret;
355 /*
356 * try to revoke all committed pages, but still we could fail
357 * due to no memory or other reason, if that happened, EAGAIN
358 * will be returned, which means in such case, transaction is
359 * already not integrity, caller should use journal to do the
360 * recovery or rewrite & commit last transaction. For other
361 * error number, revoking was done by filesystem itself.
362 */
363 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
364 if (ret)
365 err = ret;
366
367 /* drop all uncommitted pages */
368 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
369 }
88b88a66
JK		 370 mutex_unlock(&fi->inmem_lock);
371
5fe45743
CY		 372 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
373
29b96b54 374 f2fs_unlock_op(sbi);
edb27dee 375 return err;
88b88a66
JK		 376}
377
0a8165d7 378/*
351df4b2
JK		 379 * This function balances dirty node and dentry pages.
380 * In addition, it controls garbage collection.
381 */
2c4db1a6 382void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 383{
0f348028 384#ifdef CONFIG_F2FS_FAULT_INJECTION
55523519
CY		 385 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
386 f2fs_show_injection_info(FAULT_CHECKPOINT);
0f348028 387 f2fs_stop_checkpoint(sbi, false);
55523519 388 }
0f348028
CY		 389#endif
390
2c4db1a6
JK		 391 if (!need)
392 return;
e589c2c4
JK		 393
394 /* balance_fs_bg is able to be pending */
395 if (excess_cached_nats(sbi))
396 f2fs_balance_fs_bg(sbi);
397
351df4b2 398 /*
029cd28c
JK		 399 * We should do GC or end up with checkpoint, if there are so many dirty
400 * dir/node pages without enough free segments.
351df4b2 401 */
7f3037a5 402 if (has_not_enough_free_secs(sbi, 0, 0)) {
351df4b2 403 mutex_lock(&sbi->gc_mutex);
7702bdbe 404 f2fs_gc(sbi, false, false);
351df4b2
JK		 405 }
406}
407
4660f9c0
JK		 408void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
409{
1dcc336b 410 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 411 if (!available_free_memory(sbi, EXTENT_CACHE))
412 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 413
1b38dc8e
JK		 414 /* check the # of cached NAT entries */
415 if (!available_free_memory(sbi, NAT_ENTRIES))
416 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
417
31696580 418 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 419 try_to_free_nids(sbi, MAX_FREE_NIDS);
420 else
22ad0b6a 421 build_free_nids(sbi, false, false);
31696580 422
f455c8a5
JK		 423 if (!is_idle(sbi))
424 return;
31696580 425
1b38dc8e
JK		 426 /* checkpoint is the only way to shrink partial cached entries */
427 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 428 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 429 excess_prefree_segs(sbi) ||
430 excess_dirty_nats(sbi) ||
f455c8a5 431 f2fs_time_over(sbi, CP_TIME)) {
e9f5b8b8
CY		 432 if (test_opt(sbi, DATA_FLUSH)) {
433 struct blk_plug plug;
434
435 blk_start_plug(&plug);
36b35a0d 436 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 437 blk_finish_plug(&plug);
438 }
4660f9c0 439 f2fs_sync_fs(sbi->sb, true);
42190d2a 440 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 441 }
4660f9c0
JK		 442}
443
20fda56b
KM		 444static int __submit_flush_wait(struct f2fs_sb_info *sbi,
445 struct block_device *bdev)
3c62be17
JK		 446{
447 struct bio *bio = f2fs_bio_alloc(0);
448 int ret;
449
09cb6464 450 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3c62be17
JK		 451 bio->bi_bdev = bdev;
452 ret = submit_bio_wait(bio);
453 bio_put(bio);
20fda56b
KM		 454
455 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
456 test_opt(sbi, FLUSH_MERGE), ret);
3c62be17
JK		 457 return ret;
458}
459
460static int submit_flush_wait(struct f2fs_sb_info *sbi)
461{
20fda56b 462 int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
3c62be17
JK		 463 int i;
464
20fda56b
KM		 465 if (!sbi->s_ndevs || ret)
466 return ret;
467
468 for (i = 1; i < sbi->s_ndevs; i++) {
469 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
470 if (ret)
471 break;
3c62be17
JK		 472 }
473 return ret;
474}
475
2163d198 476static int issue_flush_thread(void *data)
6b4afdd7
JK		 477{
478 struct f2fs_sb_info *sbi = data;
b01a9201 479 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
a688b9d9 480 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 481repeat:
482 if (kthread_should_stop())
483 return 0;
484
721bd4d5 485 if (!llist_empty(&fcc->issue_list)) {
6b4afdd7
JK		 486 struct flush_cmd *cmd, *next;
487 int ret;
488
721bd4d5
GZ		 489 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
490 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
491
3c62be17 492 ret = submit_flush_wait(sbi);
8b8dd65f
CY		 493 atomic_inc(&fcc->issued_flush);
494
721bd4d5
GZ		 495 llist_for_each_entry_safe(cmd, next,
496 fcc->dispatch_list, llnode) {
6b4afdd7 497 cmd->ret = ret;
6b4afdd7
JK		 498 complete(&cmd->wait);
499 }
a688b9d9 500 fcc->dispatch_list = NULL;
6b4afdd7
JK		 501 }
502
a688b9d9 503 wait_event_interruptible(*q,
721bd4d5 504 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 505 goto repeat;
506}
507
508int f2fs_issue_flush(struct f2fs_sb_info *sbi)
509{
b01a9201 510 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
adf8d90b 511 struct flush_cmd cmd;
8b8dd65f 512 int ret;
6b4afdd7 513
0f7b2abd
JK		 514 if (test_opt(sbi, NOBARRIER))
515 return 0;
516
8b8dd65f
CY		 517 if (!test_opt(sbi, FLUSH_MERGE)) {
518 ret = submit_flush_wait(sbi);
519 atomic_inc(&fcc->issued_flush);
520 return ret;
521 }
740432f8 522
8b8dd65f
CY		 523 if (!atomic_read(&fcc->issing_flush)) {
524 atomic_inc(&fcc->issing_flush);
3c62be17 525 ret = submit_flush_wait(sbi);
8b8dd65f
CY		 526 atomic_dec(&fcc->issing_flush);
527
528 atomic_inc(&fcc->issued_flush);
740432f8
JK		 529 return ret;
530 }
6b4afdd7 531
adf8d90b 532 init_completion(&cmd.wait);
6b4afdd7 533
8b8dd65f 534 atomic_inc(&fcc->issing_flush);
721bd4d5 535 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 536
a688b9d9
GZ		 537 if (!fcc->dispatch_list)
538 wake_up(&fcc->flush_wait_queue);
6b4afdd7 539
5eba8c5d
JK		 540 if (fcc->f2fs_issue_flush) {
541 wait_for_completion(&cmd.wait);
8b8dd65f 542 atomic_dec(&fcc->issing_flush);
5eba8c5d
JK		 543 } else {
544 llist_del_all(&fcc->issue_list);
8b8dd65f 545 atomic_set(&fcc->issing_flush, 0);
5eba8c5d 546 }
adf8d90b
CY		 547
548 return cmd.ret;
6b4afdd7
JK		 549}
550
2163d198
GZ		 551int create_flush_cmd_control(struct f2fs_sb_info *sbi)
552{
553 dev_t dev = sbi->sb->s_bdev->bd_dev;
554 struct flush_cmd_control *fcc;
555 int err = 0;
556
b01a9201
JK		 557 if (SM_I(sbi)->fcc_info) {
558 fcc = SM_I(sbi)->fcc_info;
5eba8c5d
JK		 559 goto init_thread;
560 }
561
2163d198
GZ		 562 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
563 if (!fcc)
564 return -ENOMEM;
8b8dd65f
CY		 565 atomic_set(&fcc->issued_flush, 0);
566 atomic_set(&fcc->issing_flush, 0);
2163d198 567 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 568 init_llist_head(&fcc->issue_list);
b01a9201 569 SM_I(sbi)->fcc_info = fcc;
5eba8c5d 570init_thread:
2163d198
GZ		 571 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
572 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
573 if (IS_ERR(fcc->f2fs_issue_flush)) {
574 err = PTR_ERR(fcc->f2fs_issue_flush);
575 kfree(fcc);
b01a9201 576 SM_I(sbi)->fcc_info = NULL;
2163d198
GZ		 577 return err;
578 }
2163d198
GZ		 579
580 return err;
581}
582
5eba8c5d 583void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
2163d198 584{
b01a9201 585 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
2163d198 586
5eba8c5d
JK		 587 if (fcc && fcc->f2fs_issue_flush) {
588 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
589
590 fcc->f2fs_issue_flush = NULL;
591 kthread_stop(flush_thread);
592 }
593 if (free) {
594 kfree(fcc);
b01a9201 595 SM_I(sbi)->fcc_info = NULL;
5eba8c5d 596 }
2163d198
GZ		 597}
598
351df4b2
JK		 599static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
600 enum dirty_type dirty_type)
601{
602 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
603
604 /* need not be added */
605 if (IS_CURSEG(sbi, segno))
606 return;
607
608 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
609 dirty_i->nr_dirty[dirty_type]++;
610
611 if (dirty_type == DIRTY) {
612 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 613 enum dirty_type t = sentry->type;
b2f2c390 614
ec325b52
JK		 615 if (unlikely(t >= DIRTY)) {
616 f2fs_bug_on(sbi, 1);
617 return;
618 }
4625d6aa
CL		 619 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
620 dirty_i->nr_dirty[t]++;
351df4b2
JK		 621 }
622}
623
624static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
625 enum dirty_type dirty_type)
626{
627 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
628
629 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
630 dirty_i->nr_dirty[dirty_type]--;
631
632 if (dirty_type == DIRTY) {
4625d6aa
CL		 633 struct seg_entry *sentry = get_seg_entry(sbi, segno);
634 enum dirty_type t = sentry->type;
635
636 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
637 dirty_i->nr_dirty[t]--;
b2f2c390 638
5ec4e49f
JK		 639 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
640 clear_bit(GET_SECNO(sbi, segno),
641 dirty_i->victim_secmap);
351df4b2
JK		 642 }
643}
644
0a8165d7 645/*
351df4b2
JK		 646 * Should not occur error such as -ENOMEM.
647 * Adding dirty entry into seglist is not critical operation.
648 * If a given segment is one of current working segments, it won't be added.
649 */
8d8451af 650static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 651{
652 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
653 unsigned short valid_blocks;
654
655 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
656 return;
657
658 mutex_lock(&dirty_i->seglist_lock);
659
660 valid_blocks = get_valid_blocks(sbi, segno, 0);
661
662 if (valid_blocks == 0) {
663 __locate_dirty_segment(sbi, segno, PRE);
664 __remove_dirty_segment(sbi, segno, DIRTY);
665 } else if (valid_blocks < sbi->blocks_per_seg) {
666 __locate_dirty_segment(sbi, segno, DIRTY);
667 } else {
668 /* Recovery routine with SSR needs this */
669 __remove_dirty_segment(sbi, segno, DIRTY);
670 }
671
672 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 673}
674
15469963 675static void __add_discard_cmd(struct f2fs_sb_info *sbi,
c81abe34
JK		 676 struct block_device *bdev, block_t lstart,
677 block_t start, block_t len)
275b66b0 678{
0b54fb84
JK		 679 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
680 struct list_head *cmd_list = &(dcc->discard_cmd_list);
b01a9201 681 struct discard_cmd *dc;
275b66b0 682
b01a9201
JK		 683 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
684 INIT_LIST_HEAD(&dc->list);
c81abe34 685 dc->bdev = bdev;
b01a9201 686 dc->lstart = lstart;
c81abe34 687 dc->start = start;
b01a9201 688 dc->len = len;
15469963 689 dc->state = D_PREP;
c81abe34 690 dc->error = 0;
b01a9201 691 init_completion(&dc->wait);
15469963
JK		 692
693 mutex_lock(&dcc->cmd_lock);
0b54fb84 694 list_add_tail(&dc->list, cmd_list);
15469963 695 mutex_unlock(&dcc->cmd_lock);
5f32366a
CY		 696
697 atomic_inc(&dcc->discard_cmd_cnt);
15469963
JK		 698}
699
700static void __remove_discard_cmd(struct f2fs_sb_info *sbi, struct discard_cmd *dc)
701{
dcc9165d 702 if (dc->state == D_DONE)
8b8dd65f 703 atomic_dec(&(SM_I(sbi)->dcc_info->issing_discard));
dcc9165d 704
c81abe34
JK		 705 if (dc->error == -EOPNOTSUPP)
706 dc->error = 0;
15469963 707
c81abe34 708 if (dc->error)
15469963 709 f2fs_msg(sbi->sb, KERN_INFO,
c81abe34 710 "Issue discard failed, ret: %d", dc->error);
15469963
JK		 711 list_del(&dc->list);
712 kmem_cache_free(discard_cmd_slab, dc);
5f32366a 713 atomic_dec(&SM_I(sbi)->dcc_info->discard_cmd_cnt);
275b66b0
CY		 714}
715
c81abe34
JK		 716static void f2fs_submit_discard_endio(struct bio *bio)
717{
718 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
719
720 complete(&dc->wait);
721 dc->error = bio->bi_error;
722 dc->state = D_DONE;
723 bio_put(bio);
724}
725
726/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
727static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
728 struct discard_cmd *dc)
729{
730 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
731 struct bio *bio = NULL;
732
733 if (dc->state != D_PREP)
734 return;
735
736 dc->error = __blkdev_issue_discard(dc->bdev,
737 SECTOR_FROM_BLOCK(dc->start),
738 SECTOR_FROM_BLOCK(dc->len),
739 GFP_NOFS, 0, &bio);
740 if (!dc->error) {
741 /* should keep before submission to avoid D_DONE right away */
742 dc->state = D_SUBMIT;
8b8dd65f
CY		 743 atomic_inc(&dcc->issued_discard);
744 atomic_inc(&dcc->issing_discard);
c81abe34
JK		 745 if (bio) {
746 bio->bi_private = dc;
747 bio->bi_end_io = f2fs_submit_discard_endio;
748 bio->bi_opf |= REQ_SYNC;
749 submit_bio(bio);
750 }
751 } else {
752 __remove_discard_cmd(sbi, dc);
753 }
754}
755
756static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
757 struct block_device *bdev, block_t blkstart, block_t blklen)
758{
759 block_t lblkstart = blkstart;
760
761 trace_f2fs_issue_discard(bdev, blkstart, blklen);
762
763 if (sbi->s_ndevs) {
764 int devi = f2fs_target_device_index(sbi, blkstart);
765
766 blkstart -= FDEV(devi).start_blk;
767 }
768 __add_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
769 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
770 return 0;
771}
772
3d6a650f
YH		 773static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
774 struct discard_cmd *dc, block_t blkaddr)
775{
776 block_t end_block = START_BLOCK(sbi, GET_SEGNO(sbi, blkaddr) + 1);
777
778 if (dc->state == D_DONE || dc->lstart + dc->len <= end_block) {
779 __remove_discard_cmd(sbi, dc);
780 return;
781 }
782
783 if (blkaddr - dc->lstart < dc->lstart + dc->len - end_block) {
784 dc->start += (end_block - dc->lstart);
785 dc->len -= (end_block - dc->lstart);
786 dc->lstart = end_block;
787 } else {
788 dc->len = blkaddr - dc->lstart;
789 }
790}
791
4e6a8d9b
JK		 792/* This should be covered by global mutex, &sit_i->sentry_lock */
793void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
275b66b0 794{
0b54fb84
JK		 795 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
796 struct list_head *wait_list = &(dcc->discard_cmd_list);
b01a9201 797 struct discard_cmd *dc, *tmp;
40465257 798 struct blk_plug plug;
275b66b0 799
15469963 800 mutex_lock(&dcc->cmd_lock);
40465257
JK		 801
802 blk_start_plug(&plug);
803
b01a9201 804 list_for_each_entry_safe(dc, tmp, wait_list, list) {
275b66b0 805
15469963 806 if (blkaddr == NULL_ADDR) {
c81abe34 807 __submit_discard_cmd(sbi, dc);
15469963
JK		 808 continue;
809 }
810
811 if (dc->lstart <= blkaddr && blkaddr < dc->lstart + dc->len) {
812 if (dc->state == D_SUBMIT)
b01a9201 813 wait_for_completion_io(&dc->wait);
3d6a650f 814 __punch_discard_cmd(sbi, dc, blkaddr);
4e6a8d9b 815 }
275b66b0 816 }
40465257
JK		 817 blk_finish_plug(&plug);
818
819 /* this comes from f2fs_put_super */
820 if (blkaddr == NULL_ADDR) {
821 list_for_each_entry_safe(dc, tmp, wait_list, list) {
822 wait_for_completion_io(&dc->wait);
823 __remove_discard_cmd(sbi, dc);
824 }
825 }
15469963 826 mutex_unlock(&dcc->cmd_lock);
275b66b0
CY		 827}
828
15469963
JK		 829static int issue_discard_thread(void *data)
830{
831 struct f2fs_sb_info *sbi = data;
832 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
833 wait_queue_head_t *q = &dcc->discard_wait_queue;
834 struct list_head *cmd_list = &dcc->discard_cmd_list;
835 struct discard_cmd *dc, *tmp;
836 struct blk_plug plug;
837 int iter = 0;
838repeat:
839 if (kthread_should_stop())
840 return 0;
841
842 blk_start_plug(&plug);
843
844 mutex_lock(&dcc->cmd_lock);
845 list_for_each_entry_safe(dc, tmp, cmd_list, list) {
c81abe34
JK		 846
847 if (is_idle(sbi))
848 __submit_discard_cmd(sbi, dc);
849
850 if (dc->state == D_PREP && iter++ > DISCARD_ISSUE_RATE)
851 break;
852 if (dc->state == D_DONE)
15469963 853 __remove_discard_cmd(sbi, dc);
15469963
JK		 854 }
855 mutex_unlock(&dcc->cmd_lock);
856
857 blk_finish_plug(&plug);
858
859 iter = 0;
860 congestion_wait(BLK_RW_SYNC, HZ/50);
861
862 wait_event_interruptible(*q,
863 kthread_should_stop() || !list_empty(&dcc->discard_cmd_list));
864 goto repeat;
865}
866
f46e8809 867#ifdef CONFIG_BLK_DEV_ZONED
3c62be17
JK		 868static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
869 struct block_device *bdev, block_t blkstart, block_t blklen)
f46e8809 870{
92592285 871 sector_t sector, nr_sects;
10a875f8 872 block_t lblkstart = blkstart;
3c62be17
JK		 873 int devi = 0;
874
875 if (sbi->s_ndevs) {
876 devi = f2fs_target_device_index(sbi, blkstart);
877 blkstart -= FDEV(devi).start_blk;
878 }
f46e8809
DLM		 879
880 /*
881 * We need to know the type of the zone: for conventional zones,
882 * use regular discard if the drive supports it. For sequential
883 * zones, reset the zone write pointer.
884 */
3c62be17 885 switch (get_blkz_type(sbi, bdev, blkstart)) {
f46e8809
DLM		 886
887 case BLK_ZONE_TYPE_CONVENTIONAL:
888 if (!blk_queue_discard(bdev_get_queue(bdev)))
889 return 0;
c81abe34 890 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
f46e8809
DLM		 891 case BLK_ZONE_TYPE_SEQWRITE_REQ:
892 case BLK_ZONE_TYPE_SEQWRITE_PREF:
92592285
JK		 893 sector = SECTOR_FROM_BLOCK(blkstart);
894 nr_sects = SECTOR_FROM_BLOCK(blklen);
895
896 if (sector & (bdev_zone_sectors(bdev) - 1) ||
897 nr_sects != bdev_zone_sectors(bdev)) {
898 f2fs_msg(sbi->sb, KERN_INFO,
899 "(%d) %s: Unaligned discard attempted (block %x + %x)",
900 devi, sbi->s_ndevs ? FDEV(devi).path: "",
901 blkstart, blklen);
902 return -EIO;
903 }
d50aaeec 904 trace_f2fs_issue_reset_zone(bdev, blkstart);
f46e8809
DLM		 905 return blkdev_reset_zones(bdev, sector,
906 nr_sects, GFP_NOFS);
907 default:
908 /* Unknown zone type: broken device ? */
909 return -EIO;
910 }
911}
912#endif
913
3c62be17
JK		 914static int __issue_discard_async(struct f2fs_sb_info *sbi,
915 struct block_device *bdev, block_t blkstart, block_t blklen)
916{
917#ifdef CONFIG_BLK_DEV_ZONED
918 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
919 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
920 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
921#endif
c81abe34 922 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
3c62be17
JK		 923}
924
1e87a78d 925static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 926 block_t blkstart, block_t blklen)
927{
3c62be17
JK		 928 sector_t start = blkstart, len = 0;
929 struct block_device *bdev;
a66cdd98
JK		 930 struct seg_entry *se;
931 unsigned int offset;
932 block_t i;
3c62be17
JK		 933 int err = 0;
934
935 bdev = f2fs_target_device(sbi, blkstart, NULL);
936
937 for (i = blkstart; i < blkstart + blklen; i++, len++) {
938 if (i != start) {
939 struct block_device *bdev2 =
940 f2fs_target_device(sbi, i, NULL);
941
942 if (bdev2 != bdev) {
943 err = __issue_discard_async(sbi, bdev,
944 start, len);
945 if (err)
946 return err;
947 bdev = bdev2;
948 start = i;
949 len = 0;
950 }
951 }
a66cdd98 952
a66cdd98
JK		 953 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
954 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
955
956 if (!f2fs_test_and_set_bit(offset, se->discard_map))
957 sbi->discard_blks--;
958 }
f46e8809 959
3c62be17
JK		 960 if (len)
961 err = __issue_discard_async(sbi, bdev, start, len);
962 return err;
1e87a78d
JK		 963}
964
adf4983b 965static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 966 struct cp_control *cpc, struct seg_entry *se,
967 unsigned int start, unsigned int end)
b2955550 968{
0b54fb84 969 struct list_head *head = &SM_I(sbi)->dcc_info->discard_entry_list;
adf4983b
JK		 970 struct discard_entry *new, *last;
971
972 if (!list_empty(head)) {
973 last = list_last_entry(head, struct discard_entry, list);
974 if (START_BLOCK(sbi, cpc->trim_start) + start ==
ad4d307f
JK		 975 last->blkaddr + last->len &&
976 last->len < MAX_DISCARD_BLOCKS(sbi)) {
adf4983b
JK		 977 last->len += end - start;
978 goto done;
979 }
980 }
981
982 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
983 INIT_LIST_HEAD(&new->list);
984 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
985 new->len = end - start;
986 list_add_tail(&new->list, head);
987done:
0b54fb84 988 SM_I(sbi)->dcc_info->nr_discards += end - start;
adf4983b
JK		 989}
990
25290fa5
JK		 991static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
992 bool check_only)
adf4983b 993{
b2955550
JK		 994 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
995 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 996 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 997 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
998 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 999 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 1000 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 1001 unsigned int start = 0, end = -1;
4b2fecc8 1002 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 1003 int i;
1004
3e025740 1005 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
25290fa5 1006 return false;
b2955550 1007
a66cdd98
JK		 1008 if (!force) {
1009 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
0b54fb84
JK		 1010 SM_I(sbi)->dcc_info->nr_discards >=
1011 SM_I(sbi)->dcc_info->max_discards)
25290fa5 1012 return false;
4b2fecc8
JK		 1013 }
1014
b2955550
JK		 1015 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1016 for (i = 0; i < entries; i++)
a66cdd98 1017 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 1018 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 1019
0b54fb84
JK		 1020 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1021 SM_I(sbi)->dcc_info->max_discards) {
b2955550
JK		 1022 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1023 if (start >= max_blocks)
1024 break;
1025
1026 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 1027 if (force && start && end != max_blocks
1028 && (end - start) < cpc->trim_minlen)
1029 continue;
1030
25290fa5
JK		 1031 if (check_only)
1032 return true;
1033
a66cdd98 1034 __add_discard_entry(sbi, cpc, se, start, end);
b2955550 1035 }
25290fa5 1036 return false;
b2955550
JK		 1037}
1038
4b2fecc8
JK		 1039void release_discard_addrs(struct f2fs_sb_info *sbi)
1040{
0b54fb84 1041 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
4b2fecc8
JK		 1042 struct discard_entry *entry, *this;
1043
1044 /* drop caches */
1045 list_for_each_entry_safe(entry, this, head, list) {
1046 list_del(&entry->list);
1047 kmem_cache_free(discard_entry_slab, entry);
1048 }
1049}
1050
0a8165d7 1051/*
351df4b2
JK		 1052 * Should call clear_prefree_segments after checkpoint is done.
1053 */
1054static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1055{
1056 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 1057 unsigned int segno;
351df4b2
JK		 1058
1059 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 1060 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 1061 __set_test_and_free(sbi, segno);
351df4b2
JK		 1062 mutex_unlock(&dirty_i->seglist_lock);
1063}
1064
836b5a63 1065void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 1066{
0b54fb84 1067 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
2d7b822a 1068 struct discard_entry *entry, *this;
351df4b2 1069 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
29e59c14 1070 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 1071 unsigned int start = 0, end = -1;
36abef4e 1072 unsigned int secno, start_segno;
c24a0fd6 1073 bool force = (cpc->reason == CP_DISCARD);
351df4b2
JK		 1074
1075 mutex_lock(&dirty_i->seglist_lock);
29e59c14 1076
351df4b2 1077 while (1) {
29e59c14 1078 int i;
7cd8558b
JK		 1079 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1080 if (start >= MAIN_SEGS(sbi))
351df4b2 1081 break;
7cd8558b
JK		 1082 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1083 start + 1);
29e59c14
CL		 1084
1085 for (i = start; i < end; i++)
1086 clear_bit(i, prefree_map);
1087
1088 dirty_i->nr_dirty[PRE] -= end - start;
1089
650d3c4e 1090 if (!test_opt(sbi, DISCARD))
29e59c14 1091 continue;
351df4b2 1092
650d3c4e
YH		 1093 if (force && start >= cpc->trim_start &&
1094 (end - 1) <= cpc->trim_end)
1095 continue;
1096
36abef4e
JK		 1097 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1098 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 1099 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 1100 continue;
1101 }
1102next:
1103 secno = GET_SECNO(sbi, start);
1104 start_segno = secno * sbi->segs_per_sec;
1105 if (!IS_CURSEC(sbi, secno) &&
1106 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
1107 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1108 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1109
1110 start = start_segno + sbi->segs_per_sec;
1111 if (start < end)
1112 goto next;
8b107f5b
JK		 1113 else
1114 end = start - 1;
351df4b2
JK		 1115 }
1116 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 1117
1118 /* send small discards */
2d7b822a 1119 list_for_each_entry_safe(entry, this, head, list) {
c24a0fd6 1120 if (force && entry->len < cpc->trim_minlen)
836b5a63 1121 goto skip;
37208879 1122 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 1123 cpc->trimmed += entry->len;
836b5a63 1124skip:
b2955550 1125 list_del(&entry->list);
0b54fb84 1126 SM_I(sbi)->dcc_info->nr_discards -= entry->len;
b2955550
JK		 1127 kmem_cache_free(discard_entry_slab, entry);
1128 }
351df4b2
JK		 1129}
1130
8ed59745 1131static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84 1132{
15469963 1133 dev_t dev = sbi->sb->s_bdev->bd_dev;
0b54fb84
JK		 1134 struct discard_cmd_control *dcc;
1135 int err = 0;
1136
1137 if (SM_I(sbi)->dcc_info) {
1138 dcc = SM_I(sbi)->dcc_info;
1139 goto init_thread;
1140 }
1141
1142 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1143 if (!dcc)
1144 return -ENOMEM;
1145
1146 INIT_LIST_HEAD(&dcc->discard_entry_list);
1147 INIT_LIST_HEAD(&dcc->discard_cmd_list);
15469963 1148 mutex_init(&dcc->cmd_lock);
8b8dd65f
CY		 1149 atomic_set(&dcc->issued_discard, 0);
1150 atomic_set(&dcc->issing_discard, 0);
5f32366a 1151 atomic_set(&dcc->discard_cmd_cnt, 0);
0b54fb84
JK		 1152 dcc->nr_discards = 0;
1153 dcc->max_discards = 0;
1154
15469963 1155 init_waitqueue_head(&dcc->discard_wait_queue);
0b54fb84
JK		 1156 SM_I(sbi)->dcc_info = dcc;
1157init_thread:
15469963
JK		 1158 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1159 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1160 if (IS_ERR(dcc->f2fs_issue_discard)) {
1161 err = PTR_ERR(dcc->f2fs_issue_discard);
1162 kfree(dcc);
1163 SM_I(sbi)->dcc_info = NULL;
1164 return err;
1165 }
1166
0b54fb84
JK		 1167 return err;
1168}
1169
f099405f 1170static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
0b54fb84
JK		 1171{
1172 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1173
f099405f
CY		 1174 if (!dcc)
1175 return;
1176
1177 if (dcc->f2fs_issue_discard) {
15469963
JK		 1178 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1179
1180 dcc->f2fs_issue_discard = NULL;
1181 kthread_stop(discard_thread);
1182 }
f099405f
CY		 1183
1184 kfree(dcc);
1185 SM_I(sbi)->dcc_info = NULL;
0b54fb84
JK		 1186}
1187
184a5cd2 1188static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 1189{
1190 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 1191
1192 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 1193 sit_i->dirty_sentries++;
184a5cd2
CY		 1194 return false;
1195 }
1196
1197 return true;
351df4b2
JK		 1198}
1199
1200static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1201 unsigned int segno, int modified)
1202{
1203 struct seg_entry *se = get_seg_entry(sbi, segno);
1204 se->type = type;
1205 if (modified)
1206 __mark_sit_entry_dirty(sbi, segno);
1207}
1208
1209static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1210{
1211 struct seg_entry *se;
1212 unsigned int segno, offset;
1213 long int new_vblocks;
1214
1215 segno = GET_SEGNO(sbi, blkaddr);
1216
1217 se = get_seg_entry(sbi, segno);
1218 new_vblocks = se->valid_blocks + del;
491c0854 1219 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 1220
9850cf4a 1221 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 1222 (new_vblocks > sbi->blocks_per_seg)));
1223
1224 se->valid_blocks = new_vblocks;
1225 se->mtime = get_mtime(sbi);
1226 SIT_I(sbi)->max_mtime = se->mtime;
1227
1228 /* Update valid block bitmap */
1229 if (del > 0) {
355e7891
CY		 1230 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1231#ifdef CONFIG_F2FS_CHECK_FS
1232 if (f2fs_test_and_set_bit(offset,
1233 se->cur_valid_map_mir))
1234 f2fs_bug_on(sbi, 1);
1235 else
1236 WARN_ON(1);
1237#else
05796763 1238 f2fs_bug_on(sbi, 1);
355e7891
CY		 1239#endif
1240 }
3e025740
JK		 1241 if (f2fs_discard_en(sbi) &&
1242 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 1243 sbi->discard_blks--;
720037f9
JK		 1244
1245 /* don't overwrite by SSR to keep node chain */
1246 if (se->type == CURSEG_WARM_NODE) {
1247 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1248 se->ckpt_valid_blocks++;
1249 }
351df4b2 1250 } else {
355e7891
CY		 1251 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1252#ifdef CONFIG_F2FS_CHECK_FS
1253 if (!f2fs_test_and_clear_bit(offset,
1254 se->cur_valid_map_mir))
1255 f2fs_bug_on(sbi, 1);
1256 else
1257 WARN_ON(1);
1258#else
05796763 1259 f2fs_bug_on(sbi, 1);
355e7891
CY		 1260#endif
1261 }
3e025740
JK		 1262 if (f2fs_discard_en(sbi) &&
1263 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 1264 sbi->discard_blks++;
351df4b2
JK		 1265 }
1266 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1267 se->ckpt_valid_blocks += del;
1268
1269 __mark_sit_entry_dirty(sbi, segno);
1270
1271 /* update total number of valid blocks to be written in ckpt area */
1272 SIT_I(sbi)->written_valid_blocks += del;
1273
1274 if (sbi->segs_per_sec > 1)
1275 get_sec_entry(sbi, segno)->valid_blocks += del;
1276}
1277
5e443818 1278void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 1279{
5e443818
JK		 1280 update_sit_entry(sbi, new, 1);
1281 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1282 update_sit_entry(sbi, old, -1);
1283
1284 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1285 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 1286}
1287
1288void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1289{
1290 unsigned int segno = GET_SEGNO(sbi, addr);
1291 struct sit_info *sit_i = SIT_I(sbi);
1292
9850cf4a 1293 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 1294 if (addr == NEW_ADDR)
1295 return;
1296
1297 /* add it into sit main buffer */
1298 mutex_lock(&sit_i->sentry_lock);
1299
1300 update_sit_entry(sbi, addr, -1);
1301
1302 /* add it into dirty seglist */
1303 locate_dirty_segment(sbi, segno);
1304
1305 mutex_unlock(&sit_i->sentry_lock);
1306}
1307
6e2c64ad
JK		 1308bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1309{
1310 struct sit_info *sit_i = SIT_I(sbi);
1311 unsigned int segno, offset;
1312 struct seg_entry *se;
1313 bool is_cp = false;
1314
1315 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1316 return true;
1317
1318 mutex_lock(&sit_i->sentry_lock);
1319
1320 segno = GET_SEGNO(sbi, blkaddr);
1321 se = get_seg_entry(sbi, segno);
1322 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1323
1324 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1325 is_cp = true;
1326
1327 mutex_unlock(&sit_i->sentry_lock);
1328
1329 return is_cp;
1330}
1331
0a8165d7 1332/*
351df4b2
JK		 1333 * This function should be resided under the curseg_mutex lock
1334 */
1335static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 1336 struct f2fs_summary *sum)
351df4b2
JK		 1337{
1338 struct curseg_info *curseg = CURSEG_I(sbi, type);
1339 void *addr = curseg->sum_blk;
e79efe3b 1340 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 1341 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 1342}
1343
0a8165d7 1344/*
351df4b2
JK		 1345 * Calculate the number of current summary pages for writing
1346 */
3fa06d7b 1347int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 1348{
351df4b2 1349 int valid_sum_count = 0;
9a47938b 1350 int i, sum_in_page;
351df4b2
JK		 1351
1352 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1353 if (sbi->ckpt->alloc_type[i] == SSR)
1354 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 1355 else {
1356 if (for_ra)
1357 valid_sum_count += le16_to_cpu(
1358 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1359 else
1360 valid_sum_count += curseg_blkoff(sbi, i);
1361 }
351df4b2
JK		 1362 }
1363
09cbfeaf 1364 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1365 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1366 if (valid_sum_count <= sum_in_page)
351df4b2 1367 return 1;
9a47938b 1368 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1369 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1370 return 2;
1371 return 3;
1372}
1373
0a8165d7 1374/*
351df4b2
JK		 1375 * Caller should put this summary page
1376 */
1377struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1378{
1379 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1380}
1381
381722d2 1382void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1383{
1384 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1385 void *dst = page_address(page);
1386
1387 if (src)
09cbfeaf 1388 memcpy(dst, src, PAGE_SIZE);
381722d2 1389 else
09cbfeaf 1390 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1391 set_page_dirty(page);
1392 f2fs_put_page(page, 1);
1393}
1394
381722d2
CY		 1395static void write_sum_page(struct f2fs_sb_info *sbi,
1396 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1397{
1398 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1399}
1400
b7ad7512
CY		 1401static void write_current_sum_page(struct f2fs_sb_info *sbi,
1402 int type, block_t blk_addr)
1403{
1404 struct curseg_info *curseg = CURSEG_I(sbi, type);
1405 struct page *page = grab_meta_page(sbi, blk_addr);
1406 struct f2fs_summary_block *src = curseg->sum_blk;
1407 struct f2fs_summary_block *dst;
1408
1409 dst = (struct f2fs_summary_block *)page_address(page);
1410
1411 mutex_lock(&curseg->curseg_mutex);
1412
1413 down_read(&curseg->journal_rwsem);
1414 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1415 up_read(&curseg->journal_rwsem);
1416
1417 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1418 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1419
1420 mutex_unlock(&curseg->curseg_mutex);
1421
1422 set_page_dirty(page);
1423 f2fs_put_page(page, 1);
1424}
1425
0a8165d7 1426/*
351df4b2
JK		 1427 * Find a new segment from the free segments bitmap to right order
1428 * This function should be returned with success, otherwise BUG
1429 */
1430static void get_new_segment(struct f2fs_sb_info *sbi,
1431 unsigned int *newseg, bool new_sec, int dir)
1432{
1433 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1434 unsigned int segno, secno, zoneno;
7cd8558b 1435 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1436 unsigned int hint = *newseg / sbi->segs_per_sec;
1437 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1438 unsigned int left_start = hint;
1439 bool init = true;
1440 int go_left = 0;
1441 int i;
1442
1a118ccf 1443 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1444
1445 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1446 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1447 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1448 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1449 goto got_it;
1450 }
1451find_other_zone:
7cd8558b
JK		 1452 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1453 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1454 if (dir == ALLOC_RIGHT) {
1455 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1456 MAIN_SECS(sbi), 0);
1457 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1458 } else {
1459 go_left = 1;
1460 left_start = hint - 1;
1461 }
1462 }
1463 if (go_left == 0)
1464 goto skip_left;
1465
1466 while (test_bit(left_start, free_i->free_secmap)) {
1467 if (left_start > 0) {
1468 left_start--;
1469 continue;
1470 }
1471 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1472 MAIN_SECS(sbi), 0);
1473 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1474 break;
1475 }
1476 secno = left_start;
1477skip_left:
1478 hint = secno;
1479 segno = secno * sbi->segs_per_sec;
1480 zoneno = secno / sbi->secs_per_zone;
1481
1482 /* give up on finding another zone */
1483 if (!init)
1484 goto got_it;
1485 if (sbi->secs_per_zone == 1)
1486 goto got_it;
1487 if (zoneno == old_zoneno)
1488 goto got_it;
1489 if (dir == ALLOC_LEFT) {
1490 if (!go_left && zoneno + 1 >= total_zones)
1491 goto got_it;
1492 if (go_left && zoneno == 0)
1493 goto got_it;
1494 }
1495 for (i = 0; i < NR_CURSEG_TYPE; i++)
1496 if (CURSEG_I(sbi, i)->zone == zoneno)
1497 break;
1498
1499 if (i < NR_CURSEG_TYPE) {
1500 /* zone is in user, try another */
1501 if (go_left)
1502 hint = zoneno * sbi->secs_per_zone - 1;
1503 else if (zoneno + 1 >= total_zones)
1504 hint = 0;
1505 else
1506 hint = (zoneno + 1) * sbi->secs_per_zone;
1507 init = false;
1508 goto find_other_zone;
1509 }
1510got_it:
1511 /* set it as dirty segment in free segmap */
9850cf4a 1512 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1513 __set_inuse(sbi, segno);
1514 *newseg = segno;
1a118ccf 1515 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1516}
1517
1518static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1519{
1520 struct curseg_info *curseg = CURSEG_I(sbi, type);
1521 struct summary_footer *sum_footer;
1522
1523 curseg->segno = curseg->next_segno;
1524 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1525 curseg->next_blkoff = 0;
1526 curseg->next_segno = NULL_SEGNO;
1527
1528 sum_footer = &(curseg->sum_blk->footer);
1529 memset(sum_footer, 0, sizeof(struct summary_footer));
1530 if (IS_DATASEG(type))
1531 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1532 if (IS_NODESEG(type))
1533 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1534 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1535}
1536
7a20b8a6
JK		 1537static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1538{
1539 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1540 return 0;
1541
1542 return CURSEG_I(sbi, type)->segno;
1543}
1544
0a8165d7 1545/*
351df4b2
JK		 1546 * Allocate a current working segment.
1547 * This function always allocates a free segment in LFS manner.
1548 */
1549static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1550{
1551 struct curseg_info *curseg = CURSEG_I(sbi, type);
1552 unsigned int segno = curseg->segno;
1553 int dir = ALLOC_LEFT;
1554
1555 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1556 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1557 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1558 dir = ALLOC_RIGHT;
1559
1560 if (test_opt(sbi, NOHEAP))
1561 dir = ALLOC_RIGHT;
1562
7a20b8a6 1563 segno = __get_next_segno(sbi, type);
351df4b2
JK		 1564 get_new_segment(sbi, &segno, new_sec, dir);
1565 curseg->next_segno = segno;
1566 reset_curseg(sbi, type, 1);
1567 curseg->alloc_type = LFS;
1568}
1569
1570static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1571 struct curseg_info *seg, block_t start)
1572{
1573 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1574 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1575 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1576 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1577 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1578 int i, pos;
1579
1580 for (i = 0; i < entries; i++)
1581 target_map[i] = ckpt_map[i] | cur_map[i];
1582
1583 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1584
1585 seg->next_blkoff = pos;
351df4b2
JK		 1586}
1587
0a8165d7 1588/*
351df4b2
JK		 1589 * If a segment is written by LFS manner, next block offset is just obtained
1590 * by increasing the current block offset. However, if a segment is written by
1591 * SSR manner, next block offset obtained by calling __next_free_blkoff
1592 */
1593static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1594 struct curseg_info *seg)
1595{
1596 if (seg->alloc_type == SSR)
1597 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1598 else
1599 seg->next_blkoff++;
1600}
1601
0a8165d7 1602/*
e1c42045 1603 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1604 * manner, so it should recover the existing segment information of valid blocks
1605 */
1606static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1607{
1608 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1609 struct curseg_info *curseg = CURSEG_I(sbi, type);
1610 unsigned int new_segno = curseg->next_segno;
1611 struct f2fs_summary_block *sum_node;
1612 struct page *sum_page;
1613
1614 write_sum_page(sbi, curseg->sum_blk,
1615 GET_SUM_BLOCK(sbi, curseg->segno));
1616 __set_test_and_inuse(sbi, new_segno);
1617
1618 mutex_lock(&dirty_i->seglist_lock);
1619 __remove_dirty_segment(sbi, new_segno, PRE);
1620 __remove_dirty_segment(sbi, new_segno, DIRTY);
1621 mutex_unlock(&dirty_i->seglist_lock);
1622
1623 reset_curseg(sbi, type, 1);
1624 curseg->alloc_type = SSR;
1625 __next_free_blkoff(sbi, curseg, 0);
1626
1627 if (reuse) {
1628 sum_page = get_sum_page(sbi, new_segno);
1629 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1630 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1631 f2fs_put_page(sum_page, 1);
1632 }
1633}
1634
43727527
JK		 1635static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1636{
1637 struct curseg_info *curseg = CURSEG_I(sbi, type);
1638 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
d27c3d89
CY		 1639 int i, cnt;
1640 bool reversed = false;
c192f7a4
JK		 1641
1642 /* need_SSR() already forces to do this */
1643 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
1644 return 1;
43727527 1645
70d625cb
JK		 1646 /* For node segments, let's do SSR more intensively */
1647 if (IS_NODESEG(type)) {
d27c3d89
CY		 1648 if (type >= CURSEG_WARM_NODE) {
1649 reversed = true;
1650 i = CURSEG_COLD_NODE;
1651 } else {
1652 i = CURSEG_HOT_NODE;
1653 }
1654 cnt = NR_CURSEG_NODE_TYPE;
70d625cb 1655 } else {
d27c3d89
CY		 1656 if (type >= CURSEG_WARM_DATA) {
1657 reversed = true;
1658 i = CURSEG_COLD_DATA;
1659 } else {
1660 i = CURSEG_HOT_DATA;
1661 }
1662 cnt = NR_CURSEG_DATA_TYPE;
70d625cb 1663 }
43727527 1664
d27c3d89 1665 for (; cnt-- > 0; reversed ? i-- : i++) {
c192f7a4
JK		 1666 if (i == type)
1667 continue;
43727527 1668 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
c192f7a4 1669 BG_GC, i, SSR))
43727527 1670 return 1;
c192f7a4 1671 }
43727527
JK		 1672 return 0;
1673}
1674
351df4b2
JK		 1675/*
1676 * flush out current segment and replace it with new segment
1677 * This function should be returned with success, otherwise BUG
1678 */
1679static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1680 int type, bool force)
1681{
7b405275 1682 if (force)
351df4b2 1683 new_curseg(sbi, type, true);
5b6c6be2
JK		 1684 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1685 type == CURSEG_WARM_NODE)
351df4b2
JK		 1686 new_curseg(sbi, type, false);
1687 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1688 change_curseg(sbi, type, true);
1689 else
1690 new_curseg(sbi, type, false);
dcdfff65 1691
d0db7703 1692 stat_inc_seg_type(sbi, CURSEG_I(sbi, type));
351df4b2
JK		 1693}
1694
1695void allocate_new_segments(struct f2fs_sb_info *sbi)
1696{
6ae1be13
JK		 1697 struct curseg_info *curseg;
1698 unsigned int old_segno;
351df4b2
JK		 1699 int i;
1700
6ae1be13
JK		 1701 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1702 curseg = CURSEG_I(sbi, i);
1703 old_segno = curseg->segno;
1704 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1705 locate_dirty_segment(sbi, old_segno);
1706 }
351df4b2
JK		 1707}
1708
1709static const struct segment_allocation default_salloc_ops = {
1710 .allocate_segment = allocate_segment_by_default,
1711};
1712
25290fa5
JK		 1713bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1714{
1715 __u64 trim_start = cpc->trim_start;
1716 bool has_candidate = false;
1717
1718 mutex_lock(&SIT_I(sbi)->sentry_lock);
1719 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1720 if (add_discard_addrs(sbi, cpc, true)) {
1721 has_candidate = true;
1722 break;
1723 }
1724 }
1725 mutex_unlock(&SIT_I(sbi)->sentry_lock);
1726
1727 cpc->trim_start = trim_start;
1728 return has_candidate;
1729}
1730
4b2fecc8
JK		 1731int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1732{
f7ef9b83
JK		 1733 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1734 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1735 unsigned int start_segno, end_segno;
1736 struct cp_control cpc;
c34f42e2 1737 int err = 0;
4b2fecc8 1738
836b5a63 1739 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1740 return -EINVAL;
1741
9bd27ae4 1742 cpc.trimmed = 0;
7cd8558b 1743 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1744 goto out;
1745
ed214a11
YH		 1746 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1747 f2fs_msg(sbi->sb, KERN_WARNING,
1748 "Found FS corruption, run fsck to fix.");
1749 goto out;
1750 }
1751
4b2fecc8 1752 /* start/end segment number in main_area */
7cd8558b
JK		 1753 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1754 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1755 GET_SEGNO(sbi, end);
4b2fecc8 1756 cpc.reason = CP_DISCARD;
836b5a63 1757 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1758
1759 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1760 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1761 cpc.trim_start = start_segno;
a66cdd98
JK		 1762
1763 if (sbi->discard_blks == 0)
1764 break;
1765 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1766 cpc.trim_end = end_segno;
1767 else
1768 cpc.trim_end = min_t(unsigned int,
1769 rounddown(start_segno +
bba681cb
JK		 1770 BATCHED_TRIM_SEGMENTS(sbi),
1771 sbi->segs_per_sec) - 1, end_segno);
1772
1773 mutex_lock(&sbi->gc_mutex);
c34f42e2 1774 err = write_checkpoint(sbi, &cpc);
bba681cb 1775 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 1776 if (err)
1777 break;
74fa5f3d
CY		 1778
1779 schedule();
bba681cb 1780 }
4b2fecc8 1781out:
f7ef9b83 1782 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1783 return err;
4b2fecc8
JK		 1784}
1785
351df4b2
JK		 1786static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1787{
1788 struct curseg_info *curseg = CURSEG_I(sbi, type);
1789 if (curseg->next_blkoff < sbi->blocks_per_seg)
1790 return true;
1791 return false;
1792}
1793
1794static int __get_segment_type_2(struct page *page, enum page_type p_type)
1795{
1796 if (p_type == DATA)
1797 return CURSEG_HOT_DATA;
1798 else
1799 return CURSEG_HOT_NODE;
1800}
1801
1802static int __get_segment_type_4(struct page *page, enum page_type p_type)
1803{
1804 if (p_type == DATA) {
1805 struct inode *inode = page->mapping->host;
1806
1807 if (S_ISDIR(inode->i_mode))
1808 return CURSEG_HOT_DATA;
1809 else
1810 return CURSEG_COLD_DATA;
1811 } else {
a344b9fd
JK		 1812 if (IS_DNODE(page) && is_cold_node(page))
1813 return CURSEG_WARM_NODE;
351df4b2
JK		 1814 else
1815 return CURSEG_COLD_NODE;
1816 }
1817}
1818
1819static int __get_segment_type_6(struct page *page, enum page_type p_type)
1820{
1821 if (p_type == DATA) {
1822 struct inode *inode = page->mapping->host;
1823
1824 if (S_ISDIR(inode->i_mode))
1825 return CURSEG_HOT_DATA;
354a3399 1826 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1827 return CURSEG_COLD_DATA;
1828 else
1829 return CURSEG_WARM_DATA;
1830 } else {
1831 if (IS_DNODE(page))
1832 return is_cold_node(page) ? CURSEG_WARM_NODE :
1833 CURSEG_HOT_NODE;
1834 else
1835 return CURSEG_COLD_NODE;
1836 }
1837}
1838
1839static int __get_segment_type(struct page *page, enum page_type p_type)
1840{
4081363f 1841 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1842 case 2:
1843 return __get_segment_type_2(page, p_type);
1844 case 4:
1845 return __get_segment_type_4(page, p_type);
351df4b2 1846 }
12a67146 1847 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1848 f2fs_bug_on(F2FS_P_SB(page),
1849 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1850 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1851}
1852
bfad7c2d
JK		 1853void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1854 block_t old_blkaddr, block_t *new_blkaddr,
1855 struct f2fs_summary *sum, int type)
351df4b2
JK		 1856{
1857 struct sit_info *sit_i = SIT_I(sbi);
6ae1be13 1858 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2
JK		 1859
1860 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1861 mutex_lock(&sit_i->sentry_lock);
351df4b2
JK		 1862
1863 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2 1864
4e6a8d9b
JK		 1865 f2fs_wait_discard_bio(sbi, *new_blkaddr);
1866
351df4b2
JK		 1867 /*
1868 * __add_sum_entry should be resided under the curseg_mutex
1869 * because, this function updates a summary entry in the
1870 * current summary block.
1871 */
e79efe3b 1872 __add_sum_entry(sbi, type, sum);
351df4b2 1873
351df4b2 1874 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1875
1876 stat_inc_block_count(sbi, curseg);
351df4b2
JK		 1877
1878 /*
1879 * SIT information should be updated before segment allocation,
1880 * since SSR needs latest valid block information.
1881 */
1882 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1883
3436c4bd
YS		 1884 if (!__has_curseg_space(sbi, type))
1885 sit_i->s_ops->allocate_segment(sbi, type, false);
1886
351df4b2
JK		 1887 mutex_unlock(&sit_i->sentry_lock);
1888
bfad7c2d 1889 if (page && IS_NODESEG(type))
351df4b2
JK		 1890 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1891
bfad7c2d
JK		 1892 mutex_unlock(&curseg->curseg_mutex);
1893}
1894
05ca3632 1895static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1896{
05ca3632 1897 int type = __get_segment_type(fio->page, fio->type);
0a595eba 1898 int err;
bfad7c2d 1899
7dfeaa32
JK		 1900 if (fio->type == NODE || fio->type == DATA)
1901 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
0a595eba 1902reallocate:
7a9d7548
CY		 1903 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1904 &fio->new_blkaddr, sum, type);
bfad7c2d 1905
351df4b2 1906 /* writeout dirty page into bdev */
0a595eba
JK		 1907 err = f2fs_submit_page_mbio(fio);
1908 if (err == -EAGAIN) {
1909 fio->old_blkaddr = fio->new_blkaddr;
1910 goto reallocate;
1911 }
7dfeaa32
JK		 1912
1913 if (fio->type == NODE || fio->type == DATA)
1914 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1915}
1916
577e3495 1917void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1918{
458e6197 1919 struct f2fs_io_info fio = {
05ca3632 1920 .sbi = sbi,
458e6197 1921 .type = META,
04d328de 1922 .op = REQ_OP_WRITE,
70fd7614 1923 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1924 .old_blkaddr = page->index,
1925 .new_blkaddr = page->index,
05ca3632 1926 .page = page,
4375a336 1927 .encrypted_page = NULL,
458e6197
JK		 1928 };
1929
2b947003 1930 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1931 fio.op_flags &= ~REQ_META;
2b947003 1932
351df4b2 1933 set_page_writeback(page);
05ca3632 1934 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1935}
1936
05ca3632 1937void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1938{
1939 struct f2fs_summary sum;
05ca3632 1940
351df4b2 1941 set_summary(&sum, nid, 0, 0);
05ca3632 1942 do_write_page(&sum, fio);
351df4b2
JK		 1943}
1944
05ca3632 1945void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1946{
05ca3632 1947 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1948 struct f2fs_summary sum;
1949 struct node_info ni;
1950
9850cf4a 1951 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1952 get_node_info(sbi, dn->nid, &ni);
1953 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1954 do_write_page(&sum, fio);
f28b3434 1955 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1956}
1957
05ca3632 1958void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1959{
7a9d7548 1960 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1961 stat_inc_inplace_blocks(fio->sbi);
1962 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1963}
1964
4356e48e 1965void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1966 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1967 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1968{
1969 struct sit_info *sit_i = SIT_I(sbi);
1970 struct curseg_info *curseg;
1971 unsigned int segno, old_cursegno;
1972 struct seg_entry *se;
1973 int type;
19f106bc 1974 unsigned short old_blkoff;
351df4b2
JK		 1975
1976 segno = GET_SEGNO(sbi, new_blkaddr);
1977 se = get_seg_entry(sbi, segno);
1978 type = se->type;
1979
19f106bc
CY		 1980 if (!recover_curseg) {
1981 /* for recovery flow */
1982 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1983 if (old_blkaddr == NULL_ADDR)
1984 type = CURSEG_COLD_DATA;
1985 else
1986 type = CURSEG_WARM_DATA;
1987 }
1988 } else {
1989 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1990 type = CURSEG_WARM_DATA;
1991 }
19f106bc 1992
351df4b2
JK		 1993 curseg = CURSEG_I(sbi, type);
1994
1995 mutex_lock(&curseg->curseg_mutex);
1996 mutex_lock(&sit_i->sentry_lock);
1997
1998 old_cursegno = curseg->segno;
19f106bc 1999 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 2000
2001 /* change the current segment */
2002 if (segno != curseg->segno) {
2003 curseg->next_segno = segno;
2004 change_curseg(sbi, type, true);
2005 }
2006
491c0854 2007 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 2008 __add_sum_entry(sbi, type, sum);
351df4b2 2009
28bc106b 2010 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 2011 update_sit_entry(sbi, new_blkaddr, 1);
2012 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2013 update_sit_entry(sbi, old_blkaddr, -1);
2014
2015 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2016 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2017
351df4b2 2018 locate_dirty_segment(sbi, old_cursegno);
351df4b2 2019
19f106bc
CY		 2020 if (recover_curseg) {
2021 if (old_cursegno != curseg->segno) {
2022 curseg->next_segno = old_cursegno;
2023 change_curseg(sbi, type, true);
2024 }
2025 curseg->next_blkoff = old_blkoff;
2026 }
2027
351df4b2
JK		 2028 mutex_unlock(&sit_i->sentry_lock);
2029 mutex_unlock(&curseg->curseg_mutex);
2030}
2031
528e3459
CY		 2032void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2033 block_t old_addr, block_t new_addr,
28bc106b
CY		 2034 unsigned char version, bool recover_curseg,
2035 bool recover_newaddr)
528e3459
CY		 2036{
2037 struct f2fs_summary sum;
2038
2039 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2040
28bc106b
CY		 2041 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2042 recover_curseg, recover_newaddr);
528e3459 2043
f28b3434 2044 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 2045}
2046
93dfe2ac 2047void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 2048 enum page_type type, bool ordered)
93dfe2ac 2049{
93dfe2ac 2050 if (PageWriteback(page)) {
4081363f
JK		 2051 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2052
942fd319
JK		 2053 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
2054 0, page->index, type, WRITE);
fec1d657
JK		 2055 if (ordered)
2056 wait_on_page_writeback(page);
2057 else
2058 wait_for_stable_page(page);
93dfe2ac
JK		 2059 }
2060}
2061
08b39fbd
CY		 2062void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2063 block_t blkaddr)
2064{
2065 struct page *cpage;
2066
5d4c0af4 2067 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 2068 return;
2069
08b39fbd
CY		 2070 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2071 if (cpage) {
fec1d657 2072 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 2073 f2fs_put_page(cpage, 1);
2074 }
2075}
2076
351df4b2
JK		 2077static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2078{
2079 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2080 struct curseg_info *seg_i;
2081 unsigned char *kaddr;
2082 struct page *page;
2083 block_t start;
2084 int i, j, offset;
2085
2086 start = start_sum_block(sbi);
2087
2088 page = get_meta_page(sbi, start++);
2089 kaddr = (unsigned char *)page_address(page);
2090
2091 /* Step 1: restore nat cache */
2092 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2093 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 2094
2095 /* Step 2: restore sit cache */
2096 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2097 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 2098 offset = 2 * SUM_JOURNAL_SIZE;
2099
2100 /* Step 3: restore summary entries */
2101 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2102 unsigned short blk_off;
2103 unsigned int segno;
2104
2105 seg_i = CURSEG_I(sbi, i);
2106 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2107 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2108 seg_i->next_segno = segno;
2109 reset_curseg(sbi, i, 0);
2110 seg_i->alloc_type = ckpt->alloc_type[i];
2111 seg_i->next_blkoff = blk_off;
2112
2113 if (seg_i->alloc_type == SSR)
2114 blk_off = sbi->blocks_per_seg;
2115
2116 for (j = 0; j < blk_off; j++) {
2117 struct f2fs_summary *s;
2118 s = (struct f2fs_summary *)(kaddr + offset);
2119 seg_i->sum_blk->entries[j] = *s;
2120 offset += SUMMARY_SIZE;
09cbfeaf 2121 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2122 SUM_FOOTER_SIZE)
2123 continue;
2124
2125 f2fs_put_page(page, 1);
2126 page = NULL;
2127
2128 page = get_meta_page(sbi, start++);
2129 kaddr = (unsigned char *)page_address(page);
2130 offset = 0;
2131 }
2132 }
2133 f2fs_put_page(page, 1);
2134 return 0;
2135}
2136
2137static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2138{
2139 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2140 struct f2fs_summary_block *sum;
2141 struct curseg_info *curseg;
2142 struct page *new;
2143 unsigned short blk_off;
2144 unsigned int segno = 0;
2145 block_t blk_addr = 0;
2146
2147 /* get segment number and block addr */
2148 if (IS_DATASEG(type)) {
2149 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2150 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2151 CURSEG_HOT_DATA]);
119ee914 2152 if (__exist_node_summaries(sbi))
351df4b2
JK		 2153 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2154 else
2155 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2156 } else {
2157 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2158 CURSEG_HOT_NODE]);
2159 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2160 CURSEG_HOT_NODE]);
119ee914 2161 if (__exist_node_summaries(sbi))
351df4b2
JK		 2162 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2163 type - CURSEG_HOT_NODE);
2164 else
2165 blk_addr = GET_SUM_BLOCK(sbi, segno);
2166 }
2167
2168 new = get_meta_page(sbi, blk_addr);
2169 sum = (struct f2fs_summary_block *)page_address(new);
2170
2171 if (IS_NODESEG(type)) {
119ee914 2172 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 2173 struct f2fs_summary *ns = &sum->entries[0];
2174 int i;
2175 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2176 ns->version = 0;
2177 ns->ofs_in_node = 0;
2178 }
2179 } else {
d653788a
GZ		 2180 int err;
2181
2182 err = restore_node_summary(sbi, segno, sum);
2183 if (err) {
351df4b2 2184 f2fs_put_page(new, 1);
d653788a 2185 return err;
351df4b2
JK		 2186 }
2187 }
2188 }
2189
2190 /* set uncompleted segment to curseg */
2191 curseg = CURSEG_I(sbi, type);
2192 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 2193
2194 /* update journal info */
2195 down_write(&curseg->journal_rwsem);
2196 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2197 up_write(&curseg->journal_rwsem);
2198
2199 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2200 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 2201 curseg->next_segno = segno;
2202 reset_curseg(sbi, type, 0);
2203 curseg->alloc_type = ckpt->alloc_type[type];
2204 curseg->next_blkoff = blk_off;
2205 mutex_unlock(&curseg->curseg_mutex);
2206 f2fs_put_page(new, 1);
2207 return 0;
2208}
2209
2210static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2211{
2212 int type = CURSEG_HOT_DATA;
e4fc5fbf 2213 int err;
351df4b2 2214
aaec2b1d 2215 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 2216 int npages = npages_for_summary_flush(sbi, true);
2217
2218 if (npages >= 2)
2219 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 2220 META_CP, true);
3fa06d7b 2221
351df4b2
JK		 2222 /* restore for compacted data summary */
2223 if (read_compacted_summaries(sbi))
2224 return -EINVAL;
2225 type = CURSEG_HOT_NODE;
2226 }
2227
119ee914 2228 if (__exist_node_summaries(sbi))
3fa06d7b 2229 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 2230 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 2231
e4fc5fbf
CY		 2232 for (; type <= CURSEG_COLD_NODE; type++) {
2233 err = read_normal_summaries(sbi, type);
2234 if (err)
2235 return err;
2236 }
2237
351df4b2
JK		 2238 return 0;
2239}
2240
2241static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2242{
2243 struct page *page;
2244 unsigned char *kaddr;
2245 struct f2fs_summary *summary;
2246 struct curseg_info *seg_i;
2247 int written_size = 0;
2248 int i, j;
2249
2250 page = grab_meta_page(sbi, blkaddr++);
2251 kaddr = (unsigned char *)page_address(page);
2252
2253 /* Step 1: write nat cache */
2254 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 2255 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2256 written_size += SUM_JOURNAL_SIZE;
2257
2258 /* Step 2: write sit cache */
2259 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2260 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 2261 written_size += SUM_JOURNAL_SIZE;
2262
351df4b2
JK		 2263 /* Step 3: write summary entries */
2264 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2265 unsigned short blkoff;
2266 seg_i = CURSEG_I(sbi, i);
2267 if (sbi->ckpt->alloc_type[i] == SSR)
2268 blkoff = sbi->blocks_per_seg;
2269 else
2270 blkoff = curseg_blkoff(sbi, i);
2271
2272 for (j = 0; j < blkoff; j++) {
2273 if (!page) {
2274 page = grab_meta_page(sbi, blkaddr++);
2275 kaddr = (unsigned char *)page_address(page);
2276 written_size = 0;
2277 }
2278 summary = (struct f2fs_summary *)(kaddr + written_size);
2279 *summary = seg_i->sum_blk->entries[j];
2280 written_size += SUMMARY_SIZE;
351df4b2 2281
09cbfeaf 2282 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 2283 SUM_FOOTER_SIZE)
2284 continue;
2285
e8d61a74 2286 set_page_dirty(page);
351df4b2
JK		 2287 f2fs_put_page(page, 1);
2288 page = NULL;
2289 }
2290 }
e8d61a74
CY		 2291 if (page) {
2292 set_page_dirty(page);
351df4b2 2293 f2fs_put_page(page, 1);
e8d61a74 2294 }
351df4b2
JK		 2295}
2296
2297static void write_normal_summaries(struct f2fs_sb_info *sbi,
2298 block_t blkaddr, int type)
2299{
2300 int i, end;
2301 if (IS_DATASEG(type))
2302 end = type + NR_CURSEG_DATA_TYPE;
2303 else
2304 end = type + NR_CURSEG_NODE_TYPE;
2305
b7ad7512
CY		 2306 for (i = type; i < end; i++)
2307 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 2308}
2309
2310void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2311{
aaec2b1d 2312 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 2313 write_compacted_summaries(sbi, start_blk);
2314 else
2315 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2316}
2317
2318void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2319{
119ee914 2320 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 2321}
2322
dfc08a12 2323int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 2324 unsigned int val, int alloc)
2325{
2326 int i;
2327
2328 if (type == NAT_JOURNAL) {
dfc08a12
CY		 2329 for (i = 0; i < nats_in_cursum(journal); i++) {
2330 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 2331 return i;
2332 }
dfc08a12
CY		 2333 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2334 return update_nats_in_cursum(journal, 1);
351df4b2 2335 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 2336 for (i = 0; i < sits_in_cursum(journal); i++)
2337 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 2338 return i;
dfc08a12
CY		 2339 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2340 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 2341 }
2342 return -1;
2343}
2344
2345static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2346 unsigned int segno)
2347{
2cc22186 2348 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 2349}
2350
2351static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2352 unsigned int start)
2353{
2354 struct sit_info *sit_i = SIT_I(sbi);
2355 struct page *src_page, *dst_page;
2356 pgoff_t src_off, dst_off;
2357 void *src_addr, *dst_addr;
2358
2359 src_off = current_sit_addr(sbi, start);
2360 dst_off = next_sit_addr(sbi, src_off);
2361
2362 /* get current sit block page without lock */
2363 src_page = get_meta_page(sbi, src_off);
2364 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 2365 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 2366
2367 src_addr = page_address(src_page);
2368 dst_addr = page_address(dst_page);
09cbfeaf 2369 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 2370
2371 set_page_dirty(dst_page);
2372 f2fs_put_page(src_page, 1);
2373
2374 set_to_next_sit(sit_i, start);
2375
2376 return dst_page;
2377}
2378
184a5cd2
CY		 2379static struct sit_entry_set *grab_sit_entry_set(void)
2380{
2381 struct sit_entry_set *ses =
80c54505 2382 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 2383
2384 ses->entry_cnt = 0;
2385 INIT_LIST_HEAD(&ses->set_list);
2386 return ses;
2387}
2388
2389static void release_sit_entry_set(struct sit_entry_set *ses)
2390{
2391 list_del(&ses->set_list);
2392 kmem_cache_free(sit_entry_set_slab, ses);
2393}
2394
2395static void adjust_sit_entry_set(struct sit_entry_set *ses,
2396 struct list_head *head)
2397{
2398 struct sit_entry_set *next = ses;
2399
2400 if (list_is_last(&ses->set_list, head))
2401 return;
2402
2403 list_for_each_entry_continue(next, head, set_list)
2404 if (ses->entry_cnt <= next->entry_cnt)
2405 break;
2406
2407 list_move_tail(&ses->set_list, &next->set_list);
2408}
2409
2410static void add_sit_entry(unsigned int segno, struct list_head *head)
2411{
2412 struct sit_entry_set *ses;
2413 unsigned int start_segno = START_SEGNO(segno);
2414
2415 list_for_each_entry(ses, head, set_list) {
2416 if (ses->start_segno == start_segno) {
2417 ses->entry_cnt++;
2418 adjust_sit_entry_set(ses, head);
2419 return;
2420 }
2421 }
2422
2423 ses = grab_sit_entry_set();
2424
2425 ses->start_segno = start_segno;
2426 ses->entry_cnt++;
2427 list_add(&ses->set_list, head);
2428}
2429
2430static void add_sits_in_set(struct f2fs_sb_info *sbi)
2431{
2432 struct f2fs_sm_info *sm_info = SM_I(sbi);
2433 struct list_head *set_list = &sm_info->sit_entry_set;
2434 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2435 unsigned int segno;
2436
7cd8558b 2437 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2438 add_sit_entry(segno, set_list);
2439}
2440
2441static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2442{
2443 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2444 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2445 int i;
2446
b7ad7512 2447 down_write(&curseg->journal_rwsem);
dfc08a12 2448 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2449 unsigned int segno;
2450 bool dirtied;
2451
dfc08a12 2452 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2453 dirtied = __mark_sit_entry_dirty(sbi, segno);
2454
2455 if (!dirtied)
2456 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2457 }
dfc08a12 2458 update_sits_in_cursum(journal, -i);
b7ad7512 2459 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2460}
2461
0a8165d7 2462/*
351df4b2
JK		 2463 * CP calls this function, which flushes SIT entries including sit_journal,
2464 * and moves prefree segs to free segs.
2465 */
4b2fecc8 2466void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2467{
2468 struct sit_info *sit_i = SIT_I(sbi);
2469 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2470 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2471 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2472 struct sit_entry_set *ses, *tmp;
2473 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2474 bool to_journal = true;
4b2fecc8 2475 struct seg_entry *se;
351df4b2 2476
351df4b2
JK		 2477 mutex_lock(&sit_i->sentry_lock);
2478
2b11a74b
WL		 2479 if (!sit_i->dirty_sentries)
2480 goto out;
2481
351df4b2 2482 /*
184a5cd2
CY		 2483 * add and account sit entries of dirty bitmap in sit entry
2484 * set temporarily
351df4b2 2485 */
184a5cd2 2486 add_sits_in_set(sbi);
351df4b2 2487
184a5cd2
CY		 2488 /*
2489 * if there are no enough space in journal to store dirty sit
2490 * entries, remove all entries from journal and add and account
2491 * them in sit entry set.
2492 */
dfc08a12 2493 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2494 remove_sits_in_journal(sbi);
b2955550 2495
184a5cd2
CY		 2496 /*
2497 * there are two steps to flush sit entries:
2498 * #1, flush sit entries to journal in current cold data summary block.
2499 * #2, flush sit entries to sit page.
2500 */
2501 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2502 struct page *page = NULL;
184a5cd2
CY		 2503 struct f2fs_sit_block *raw_sit = NULL;
2504 unsigned int start_segno = ses->start_segno;
2505 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2506 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2507 unsigned int segno = start_segno;
2508
2509 if (to_journal &&
dfc08a12 2510 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2511 to_journal = false;
2512
b7ad7512
CY		 2513 if (to_journal) {
2514 down_write(&curseg->journal_rwsem);
2515 } else {
184a5cd2
CY		 2516 page = get_next_sit_page(sbi, start_segno);
2517 raw_sit = page_address(page);
351df4b2 2518 }
351df4b2 2519
184a5cd2
CY		 2520 /* flush dirty sit entries in region of current sit set */
2521 for_each_set_bit_from(segno, bitmap, end) {
2522 int offset, sit_offset;
4b2fecc8
JK		 2523
2524 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2525
2526 /* add discard candidates */
d7bc2484 2527 if (cpc->reason != CP_DISCARD) {
4b2fecc8 2528 cpc->trim_start = segno;
25290fa5 2529 add_discard_addrs(sbi, cpc, false);
4b2fecc8 2530 }
184a5cd2
CY		 2531
2532 if (to_journal) {
dfc08a12 2533 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2534 SIT_JOURNAL, segno, 1);
2535 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2536 segno_in_journal(journal, offset) =
184a5cd2
CY		 2537 cpu_to_le32(segno);
2538 seg_info_to_raw_sit(se,
dfc08a12 2539 &sit_in_journal(journal, offset));
184a5cd2
CY		 2540 } else {
2541 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2542 seg_info_to_raw_sit(se,
2543 &raw_sit->entries[sit_offset]);
2544 }
351df4b2 2545
184a5cd2
CY		 2546 __clear_bit(segno, bitmap);
2547 sit_i->dirty_sentries--;
2548 ses->entry_cnt--;
351df4b2
JK		 2549 }
2550
b7ad7512
CY		 2551 if (to_journal)
2552 up_write(&curseg->journal_rwsem);
2553 else
184a5cd2
CY		 2554 f2fs_put_page(page, 1);
2555
2556 f2fs_bug_on(sbi, ses->entry_cnt);
2557 release_sit_entry_set(ses);
351df4b2 2558 }
184a5cd2
CY		 2559
2560 f2fs_bug_on(sbi, !list_empty(head));
2561 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2562out:
4b2fecc8 2563 if (cpc->reason == CP_DISCARD) {
650d3c4e
YH		 2564 __u64 trim_start = cpc->trim_start;
2565
4b2fecc8 2566 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
25290fa5 2567 add_discard_addrs(sbi, cpc, false);
650d3c4e
YH		 2568
2569 cpc->trim_start = trim_start;
4b2fecc8 2570 }
351df4b2 2571 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2572
351df4b2
JK		 2573 set_prefree_as_free_segments(sbi);
2574}
2575
2576static int build_sit_info(struct f2fs_sb_info *sbi)
2577{
2578 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
351df4b2
JK		 2579 struct sit_info *sit_i;
2580 unsigned int sit_segs, start;
ae27d62e 2581 char *src_bitmap;
351df4b2
JK		 2582 unsigned int bitmap_size;
2583
2584 /* allocate memory for SIT information */
2585 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2586 if (!sit_i)
2587 return -ENOMEM;
2588
2589 SM_I(sbi)->sit_info = sit_i;
2590
39307a8e
JK		 2591 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2592 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2593 if (!sit_i->sentries)
2594 return -ENOMEM;
2595
7cd8558b 2596 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2597 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2598 if (!sit_i->dirty_sentries_bitmap)
2599 return -ENOMEM;
2600
7cd8558b 2601 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2602 sit_i->sentries[start].cur_valid_map
2603 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2604 sit_i->sentries[start].ckpt_valid_map
2605 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 2606 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 2607 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 2608 return -ENOMEM;
3e025740 2609
355e7891
CY		 2610#ifdef CONFIG_F2FS_CHECK_FS
2611 sit_i->sentries[start].cur_valid_map_mir
2612 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2613 if (!sit_i->sentries[start].cur_valid_map_mir)
2614 return -ENOMEM;
2615#endif
2616
3e025740
JK		 2617 if (f2fs_discard_en(sbi)) {
2618 sit_i->sentries[start].discard_map
2619 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2620 if (!sit_i->sentries[start].discard_map)
2621 return -ENOMEM;
2622 }
351df4b2
JK		 2623 }
2624
60a3b782
JK		 2625 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2626 if (!sit_i->tmp_map)
2627 return -ENOMEM;
2628
351df4b2 2629 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2630 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2631 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2632 if (!sit_i->sec_entries)
2633 return -ENOMEM;
2634 }
2635
2636 /* get information related with SIT */
2637 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2638
2639 /* setup SIT bitmap from ckeckpoint pack */
2640 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2641 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2642
ae27d62e
CY		 2643 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2644 if (!sit_i->sit_bitmap)
351df4b2 2645 return -ENOMEM;
351df4b2 2646
ae27d62e
CY		 2647#ifdef CONFIG_F2FS_CHECK_FS
2648 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2649 if (!sit_i->sit_bitmap_mir)
2650 return -ENOMEM;
2651#endif
2652
351df4b2
JK		 2653 /* init SIT information */
2654 sit_i->s_ops = &default_salloc_ops;
2655
2656 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2657 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
c79b7ff1 2658 sit_i->written_valid_blocks = 0;
351df4b2
JK		 2659 sit_i->bitmap_size = bitmap_size;
2660 sit_i->dirty_sentries = 0;
2661 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2662 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2663 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2664 mutex_init(&sit_i->sentry_lock);
2665 return 0;
2666}
2667
2668static int build_free_segmap(struct f2fs_sb_info *sbi)
2669{
351df4b2
JK		 2670 struct free_segmap_info *free_i;
2671 unsigned int bitmap_size, sec_bitmap_size;
2672
2673 /* allocate memory for free segmap information */
2674 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2675 if (!free_i)
2676 return -ENOMEM;
2677
2678 SM_I(sbi)->free_info = free_i;
2679
7cd8558b 2680 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2681 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2682 if (!free_i->free_segmap)
2683 return -ENOMEM;
2684
7cd8558b 2685 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2686 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2687 if (!free_i->free_secmap)
2688 return -ENOMEM;
2689
2690 /* set all segments as dirty temporarily */
2691 memset(free_i->free_segmap, 0xff, bitmap_size);
2692 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2693
2694 /* init free segmap information */
7cd8558b 2695 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2696 free_i->free_segments = 0;
2697 free_i->free_sections = 0;
1a118ccf 2698 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2699 return 0;
2700}
2701
2702static int build_curseg(struct f2fs_sb_info *sbi)
2703{
1042d60f 2704 struct curseg_info *array;
351df4b2
JK		 2705 int i;
2706
b434babf 2707 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2708 if (!array)
2709 return -ENOMEM;
2710
2711 SM_I(sbi)->curseg_array = array;
2712
2713 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2714 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2715 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2716 if (!array[i].sum_blk)
2717 return -ENOMEM;
b7ad7512
CY		 2718 init_rwsem(&array[i].journal_rwsem);
2719 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2720 GFP_KERNEL);
2721 if (!array[i].journal)
2722 return -ENOMEM;
351df4b2
JK		 2723 array[i].segno = NULL_SEGNO;
2724 array[i].next_blkoff = 0;
2725 }
2726 return restore_curseg_summaries(sbi);
2727}
2728
2729static void build_sit_entries(struct f2fs_sb_info *sbi)
2730{
2731 struct sit_info *sit_i = SIT_I(sbi);
2732 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2733 struct f2fs_journal *journal = curseg->journal;
9c094040
YH		 2734 struct seg_entry *se;
2735 struct f2fs_sit_entry sit;
74de593a
CY		 2736 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2737 unsigned int i, start, end;
2738 unsigned int readed, start_blk = 0;
351df4b2 2739
74de593a 2740 do {
664ba972
JK		 2741 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2742 META_SIT, true);
74de593a
CY		 2743
2744 start = start_blk * sit_i->sents_per_block;
2745 end = (start_blk + readed) * sit_i->sents_per_block;
2746
7cd8558b 2747 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a 2748 struct f2fs_sit_block *sit_blk;
74de593a
CY		 2749 struct page *page;
2750
9c094040 2751 se = &sit_i->sentries[start];
74de593a
CY		 2752 page = get_current_sit_page(sbi, start);
2753 sit_blk = (struct f2fs_sit_block *)page_address(page);
2754 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2755 f2fs_put_page(page, 1);
d600af23 2756
74de593a
CY		 2757 check_block_count(sbi, start, &sit);
2758 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2759
2760 /* build discard map only one time */
3e025740
JK		 2761 if (f2fs_discard_en(sbi)) {
2762 memcpy(se->discard_map, se->cur_valid_map,
2763 SIT_VBLOCK_MAP_SIZE);
2764 sbi->discard_blks += sbi->blocks_per_seg -
2765 se->valid_blocks;
2766 }
a66cdd98 2767
d600af23
CY		 2768 if (sbi->segs_per_sec > 1)
2769 get_sec_entry(sbi, start)->valid_blocks +=
2770 se->valid_blocks;
351df4b2 2771 }
74de593a
CY		 2772 start_blk += readed;
2773 } while (start_blk < sit_blk_cnt);
d600af23
CY		 2774
2775 down_read(&curseg->journal_rwsem);
2776 for (i = 0; i < sits_in_cursum(journal); i++) {
d600af23
CY		 2777 unsigned int old_valid_blocks;
2778
2779 start = le32_to_cpu(segno_in_journal(journal, i));
2780 se = &sit_i->sentries[start];
2781 sit = sit_in_journal(journal, i);
2782
2783 old_valid_blocks = se->valid_blocks;
2784
2785 check_block_count(sbi, start, &sit);
2786 seg_info_from_raw_sit(se, &sit);
2787
2788 if (f2fs_discard_en(sbi)) {
2789 memcpy(se->discard_map, se->cur_valid_map,
2790 SIT_VBLOCK_MAP_SIZE);
2791 sbi->discard_blks += old_valid_blocks -
2792 se->valid_blocks;
2793 }
2794
2795 if (sbi->segs_per_sec > 1)
2796 get_sec_entry(sbi, start)->valid_blocks +=
2797 se->valid_blocks - old_valid_blocks;
2798 }
2799 up_read(&curseg->journal_rwsem);
351df4b2
JK		 2800}
2801
2802static void init_free_segmap(struct f2fs_sb_info *sbi)
2803{
2804 unsigned int start;
2805 int type;
2806
7cd8558b 2807 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2808 struct seg_entry *sentry = get_seg_entry(sbi, start);
2809 if (!sentry->valid_blocks)
2810 __set_free(sbi, start);
c79b7ff1
JK		 2811 else
2812 SIT_I(sbi)->written_valid_blocks +=
2813 sentry->valid_blocks;
351df4b2
JK		 2814 }
2815
2816 /* set use the current segments */
2817 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2818 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2819 __set_test_and_inuse(sbi, curseg_t->segno);
2820 }
2821}
2822
2823static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2824{
2825 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2826 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2827 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2828 unsigned short valid_blocks;
2829
8736fbf0 2830 while (1) {
351df4b2 2831 /* find dirty segment based on free segmap */
7cd8558b
JK		 2832 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2833 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2834 break;
2835 offset = segno + 1;
2836 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2837 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2838 continue;
ec325b52
JK		 2839 if (valid_blocks > sbi->blocks_per_seg) {
2840 f2fs_bug_on(sbi, 1);
2841 continue;
2842 }
351df4b2
JK		 2843 mutex_lock(&dirty_i->seglist_lock);
2844 __locate_dirty_segment(sbi, segno, DIRTY);
2845 mutex_unlock(&dirty_i->seglist_lock);
2846 }
2847}
2848
5ec4e49f 2849static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2850{
2851 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2852 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2853
39307a8e 2854 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2855 if (!dirty_i->victim_secmap)
351df4b2
JK		 2856 return -ENOMEM;
2857 return 0;
2858}
2859
2860static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2861{
2862 struct dirty_seglist_info *dirty_i;
2863 unsigned int bitmap_size, i;
2864
2865 /* allocate memory for dirty segments list information */
2866 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2867 if (!dirty_i)
2868 return -ENOMEM;
2869
2870 SM_I(sbi)->dirty_info = dirty_i;
2871 mutex_init(&dirty_i->seglist_lock);
2872
7cd8558b 2873 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2874
2875 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2876 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2877 if (!dirty_i->dirty_segmap[i])
2878 return -ENOMEM;
2879 }
2880
2881 init_dirty_segmap(sbi);
5ec4e49f 2882 return init_victim_secmap(sbi);
351df4b2
JK		 2883}
2884
0a8165d7 2885/*
351df4b2
JK		 2886 * Update min, max modified time for cost-benefit GC algorithm
2887 */
2888static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2889{
2890 struct sit_info *sit_i = SIT_I(sbi);
2891 unsigned int segno;
2892
2893 mutex_lock(&sit_i->sentry_lock);
2894
2895 sit_i->min_mtime = LLONG_MAX;
2896
7cd8558b 2897 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2898 unsigned int i;
2899 unsigned long long mtime = 0;
2900
2901 for (i = 0; i < sbi->segs_per_sec; i++)
2902 mtime += get_seg_entry(sbi, segno + i)->mtime;
2903
2904 mtime = div_u64(mtime, sbi->segs_per_sec);
2905
2906 if (sit_i->min_mtime > mtime)
2907 sit_i->min_mtime = mtime;
2908 }
2909 sit_i->max_mtime = get_mtime(sbi);
2910 mutex_unlock(&sit_i->sentry_lock);
2911}
2912
2913int build_segment_manager(struct f2fs_sb_info *sbi)
2914{
2915 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2916 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2917 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2918 int err;
2919
2920 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2921 if (!sm_info)
2922 return -ENOMEM;
2923
2924 /* init sm info */
2925 sbi->sm_info = sm_info;
351df4b2
JK		 2926 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2927 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2928 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2929 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2930 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2931 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2932 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2933 sm_info->rec_prefree_segments = sm_info->main_segments *
2934 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2935 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2936 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2937
52763a4b
JK		 2938 if (!test_opt(sbi, LFS))
2939 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2940 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2941 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2942
bba681cb
JK		 2943 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2944
184a5cd2
CY		 2945 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2946
b270ad6f 2947 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2948 err = create_flush_cmd_control(sbi);
2949 if (err)
a688b9d9 2950 return err;
6b4afdd7
JK		 2951 }
2952
0b54fb84
JK		 2953 err = create_discard_cmd_control(sbi);
2954 if (err)
2955 return err;
2956
351df4b2
JK		 2957 err = build_sit_info(sbi);
2958 if (err)
2959 return err;
2960 err = build_free_segmap(sbi);
2961 if (err)
2962 return err;
2963 err = build_curseg(sbi);
2964 if (err)
2965 return err;
2966
2967 /* reinit free segmap based on SIT */
2968 build_sit_entries(sbi);
2969
2970 init_free_segmap(sbi);
2971 err = build_dirty_segmap(sbi);
2972 if (err)
2973 return err;
2974
2975 init_min_max_mtime(sbi);
2976 return 0;
2977}
2978
2979static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2980 enum dirty_type dirty_type)
2981{
2982 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2983
2984 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2985 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2986 dirty_i->nr_dirty[dirty_type] = 0;
2987 mutex_unlock(&dirty_i->seglist_lock);
2988}
2989
5ec4e49f 2990static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2991{
2992 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2993 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2994}
2995
2996static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2997{
2998 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2999 int i;
3000
3001 if (!dirty_i)
3002 return;
3003
3004 /* discard pre-free/dirty segments list */
3005 for (i = 0; i < NR_DIRTY_TYPE; i++)
3006 discard_dirty_segmap(sbi, i);
3007
5ec4e49f 3008 destroy_victim_secmap(sbi);
351df4b2
JK		 3009 SM_I(sbi)->dirty_info = NULL;
3010 kfree(dirty_i);
3011}
3012
3013static void destroy_curseg(struct f2fs_sb_info *sbi)
3014{
3015 struct curseg_info *array = SM_I(sbi)->curseg_array;
3016 int i;
3017
3018 if (!array)
3019 return;
3020 SM_I(sbi)->curseg_array = NULL;
b7ad7512 3021 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 3022 kfree(array[i].sum_blk);
b7ad7512
CY		 3023 kfree(array[i].journal);
3024 }
351df4b2
JK		 3025 kfree(array);
3026}
3027
3028static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3029{
3030 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3031 if (!free_i)
3032 return;
3033 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 3034 kvfree(free_i->free_segmap);
3035 kvfree(free_i->free_secmap);
351df4b2
JK		 3036 kfree(free_i);
3037}
3038
3039static void destroy_sit_info(struct f2fs_sb_info *sbi)
3040{
3041 struct sit_info *sit_i = SIT_I(sbi);
3042 unsigned int start;
3043
3044 if (!sit_i)
3045 return;
3046
3047 if (sit_i->sentries) {
7cd8558b 3048 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2 3049 kfree(sit_i->sentries[start].cur_valid_map);
355e7891
CY		 3050#ifdef CONFIG_F2FS_CHECK_FS
3051 kfree(sit_i->sentries[start].cur_valid_map_mir);
3052#endif
351df4b2 3053 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 3054 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 3055 }
3056 }
60a3b782
JK		 3057 kfree(sit_i->tmp_map);
3058
39307a8e
JK		 3059 kvfree(sit_i->sentries);
3060 kvfree(sit_i->sec_entries);
3061 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 3062
3063 SM_I(sbi)->sit_info = NULL;
3064 kfree(sit_i->sit_bitmap);
ae27d62e
CY		 3065#ifdef CONFIG_F2FS_CHECK_FS
3066 kfree(sit_i->sit_bitmap_mir);
3067#endif
351df4b2
JK		 3068 kfree(sit_i);
3069}
3070
3071void destroy_segment_manager(struct f2fs_sb_info *sbi)
3072{
3073 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 3074
3b03f724
CY		 3075 if (!sm_info)
3076 return;
5eba8c5d 3077 destroy_flush_cmd_control(sbi, true);
f099405f 3078 destroy_discard_cmd_control(sbi);
351df4b2
JK		 3079 destroy_dirty_segmap(sbi);
3080 destroy_curseg(sbi);
3081 destroy_free_segmap(sbi);
3082 destroy_sit_info(sbi);
3083 sbi->sm_info = NULL;
3084 kfree(sm_info);
3085}
7fd9e544
JK		 3086
3087int __init create_segment_manager_caches(void)
3088{
3089 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 3090 sizeof(struct discard_entry));
7fd9e544 3091 if (!discard_entry_slab)
184a5cd2
CY		 3092 goto fail;
3093
b01a9201
JK		 3094 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3095 sizeof(struct discard_cmd));
3096 if (!discard_cmd_slab)
6ab2a308 3097 goto destroy_discard_entry;
275b66b0 3098
184a5cd2 3099 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 3100 sizeof(struct sit_entry_set));
184a5cd2 3101 if (!sit_entry_set_slab)
b01a9201 3102 goto destroy_discard_cmd;
88b88a66
JK		 3103
3104 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3105 sizeof(struct inmem_pages));
3106 if (!inmem_entry_slab)
3107 goto destroy_sit_entry_set;
7fd9e544 3108 return 0;
184a5cd2 3109
88b88a66
JK		 3110destroy_sit_entry_set:
3111 kmem_cache_destroy(sit_entry_set_slab);
b01a9201
JK		 3112destroy_discard_cmd:
3113 kmem_cache_destroy(discard_cmd_slab);
6ab2a308 3114destroy_discard_entry:
184a5cd2
CY		 3115 kmem_cache_destroy(discard_entry_slab);
3116fail:
3117 return -ENOMEM;
7fd9e544
JK		 3118}
3119
3120void destroy_segment_manager_caches(void)
3121{
184a5cd2 3122 kmem_cache_destroy(sit_entry_set_slab);
b01a9201 3123 kmem_cache_destroy(discard_cmd_slab);
7fd9e544 3124 kmem_cache_destroy(discard_entry_slab);
88b88a66 3125 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 3126}
Ubuntu Artful kernel mirror