]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: avoid page allocation for truncating partial inline_data
[mirror_ubuntu-jammy-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;
275b66b0 29static struct kmem_cache *bio_entry_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
91942321 245 clear_inode_flag(inode, FI_ATOMIC_FILE);
26dc3d44 246
29b96b54 247 mutex_lock(&fi->inmem_lock);
28bc106b 248 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
29b96b54
CY		 249 mutex_unlock(&fi->inmem_lock);
250}
251
28bc106b
CY		 252static int __commit_inmem_pages(struct inode *inode,
253 struct list_head *revoke_list)
88b88a66
JK		 254{
255 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
256 struct f2fs_inode_info *fi = F2FS_I(inode);
257 struct inmem_pages *cur, *tmp;
88b88a66 258 struct f2fs_io_info fio = {
05ca3632 259 .sbi = sbi,
88b88a66 260 .type = DATA,
04d328de
MC		 261 .op = REQ_OP_WRITE,
262 .op_flags = WRITE_SYNC | REQ_PRIO,
4375a336 263 .encrypted_page = NULL,
88b88a66 264 };
29b96b54 265 bool submit_bio = false;
edb27dee 266 int err = 0;
88b88a66 267
88b88a66 268 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
28bc106b
CY		 269 struct page *page = cur->page;
270
271 lock_page(page);
272 if (page->mapping == inode->i_mapping) {
273 trace_f2fs_commit_inmem_page(page, INMEM);
274
275 set_page_dirty(page);
276 f2fs_wait_on_page_writeback(page, DATA, true);
277 if (clear_page_dirty_for_io(page))
29b96b54 278 inode_dec_dirty_pages(inode);
28bc106b
CY		 279
280 fio.page = page;
29b96b54
CY		 281 err = do_write_data_page(&fio);
282 if (err) {
28bc106b 283 unlock_page(page);
29b96b54 284 break;
70c640b1 285 }
29b96b54 286
28bc106b
CY		 287 /* record old blkaddr for revoking */
288 cur->old_addr = fio.old_blkaddr;
decd36b6 289
28bc106b
CY		 290 clear_cold_data(page);
291 submit_bio = true;
292 }
293 unlock_page(page);
294 list_move_tail(&cur->list, revoke_list);
88b88a66 295 }
29b96b54
CY		 296
297 if (submit_bio)
298 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
28bc106b
CY		 299
300 if (!err)
301 __revoke_inmem_pages(inode, revoke_list, false, false);
302
29b96b54
CY		 303 return err;
304}
305
306int commit_inmem_pages(struct inode *inode)
307{
308 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
309 struct f2fs_inode_info *fi = F2FS_I(inode);
28bc106b
CY		 310 struct list_head revoke_list;
311 int err;
29b96b54 312
28bc106b 313 INIT_LIST_HEAD(&revoke_list);
29b96b54
CY		 314 f2fs_balance_fs(sbi, true);
315 f2fs_lock_op(sbi);
316
317 mutex_lock(&fi->inmem_lock);
28bc106b
CY		 318 err = __commit_inmem_pages(inode, &revoke_list);
319 if (err) {
320 int ret;
321 /*
322 * try to revoke all committed pages, but still we could fail
323 * due to no memory or other reason, if that happened, EAGAIN
324 * will be returned, which means in such case, transaction is
325 * already not integrity, caller should use journal to do the
326 * recovery or rewrite & commit last transaction. For other
327 * error number, revoking was done by filesystem itself.
328 */
329 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
330 if (ret)
331 err = ret;
332
333 /* drop all uncommitted pages */
334 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
335 }
88b88a66
JK		 336 mutex_unlock(&fi->inmem_lock);
337
29b96b54 338 f2fs_unlock_op(sbi);
edb27dee 339 return err;
88b88a66
JK		 340}
341
0a8165d7 342/*
351df4b2
JK		 343 * This function balances dirty node and dentry pages.
344 * In addition, it controls garbage collection.
345 */
2c4db1a6 346void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351df4b2 347{
2c4db1a6
JK		 348 if (!need)
349 return;
e589c2c4
JK		 350
351 /* balance_fs_bg is able to be pending */
352 if (excess_cached_nats(sbi))
353 f2fs_balance_fs_bg(sbi);
354
351df4b2 355 /*
029cd28c
JK		 356 * We should do GC or end up with checkpoint, if there are so many dirty
357 * dir/node pages without enough free segments.
351df4b2 358 */
43727527 359 if (has_not_enough_free_secs(sbi, 0)) {
351df4b2 360 mutex_lock(&sbi->gc_mutex);
d530d4d8 361 f2fs_gc(sbi, false);
351df4b2
JK		 362 }
363}
364
4660f9c0
JK		 365void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
366{
1dcc336b 367 /* try to shrink extent cache when there is no enough memory */
554df79e
JK		 368 if (!available_free_memory(sbi, EXTENT_CACHE))
369 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
1dcc336b 370
1b38dc8e
JK		 371 /* check the # of cached NAT entries */
372 if (!available_free_memory(sbi, NAT_ENTRIES))
373 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
374
31696580 375 if (!available_free_memory(sbi, FREE_NIDS))
ad4edb83
JK		 376 try_to_free_nids(sbi, MAX_FREE_NIDS);
377 else
378 build_free_nids(sbi);
31696580 379
1b38dc8e
JK		 380 /* checkpoint is the only way to shrink partial cached entries */
381 if (!available_free_memory(sbi, NAT_ENTRIES) ||
60b99b48 382 !available_free_memory(sbi, INO_ENTRIES) ||
7d768d2c
CY		 383 excess_prefree_segs(sbi) ||
384 excess_dirty_nats(sbi) ||
d0239e1b 385 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
e9f5b8b8
CY		 386 if (test_opt(sbi, DATA_FLUSH)) {
387 struct blk_plug plug;
388
389 blk_start_plug(&plug);
36b35a0d 390 sync_dirty_inodes(sbi, FILE_INODE);
e9f5b8b8
CY		 391 blk_finish_plug(&plug);
392 }
4660f9c0 393 f2fs_sync_fs(sbi->sb, true);
42190d2a 394 stat_inc_bg_cp_count(sbi->stat_info);
36b35a0d 395 }
4660f9c0
JK		 396}
397
2163d198 398static int issue_flush_thread(void *data)
6b4afdd7
JK		 399{
400 struct f2fs_sb_info *sbi = data;
a688b9d9
GZ		 401 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
402 wait_queue_head_t *q = &fcc->flush_wait_queue;
6b4afdd7
JK		 403repeat:
404 if (kthread_should_stop())
405 return 0;
406
721bd4d5 407 if (!llist_empty(&fcc->issue_list)) {
740432f8 408 struct bio *bio;
6b4afdd7
JK		 409 struct flush_cmd *cmd, *next;
410 int ret;
411
740432f8
JK		 412 bio = f2fs_bio_alloc(0);
413
721bd4d5
GZ		 414 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
415 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
416
6b4afdd7 417 bio->bi_bdev = sbi->sb->s_bdev;
04d328de 418 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
4e49ea4a 419 ret = submit_bio_wait(bio);
6b4afdd7 420
721bd4d5
GZ		 421 llist_for_each_entry_safe(cmd, next,
422 fcc->dispatch_list, llnode) {
6b4afdd7 423 cmd->ret = ret;
6b4afdd7
JK		 424 complete(&cmd->wait);
425 }
a4ed23f2 426 bio_put(bio);
a688b9d9 427 fcc->dispatch_list = NULL;
6b4afdd7
JK		 428 }
429
a688b9d9 430 wait_event_interruptible(*q,
721bd4d5 431 kthread_should_stop() || !llist_empty(&fcc->issue_list));
6b4afdd7
JK		 432 goto repeat;
433}
434
435int f2fs_issue_flush(struct f2fs_sb_info *sbi)
436{
a688b9d9 437 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
adf8d90b 438 struct flush_cmd cmd;
6b4afdd7 439
24a9ee0f
JK		 440 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
441 test_opt(sbi, FLUSH_MERGE));
442
0f7b2abd
JK		 443 if (test_opt(sbi, NOBARRIER))
444 return 0;
445
0a87f664 446 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
740432f8
JK		 447 struct bio *bio = f2fs_bio_alloc(0);
448 int ret;
449
0a87f664 450 atomic_inc(&fcc->submit_flush);
740432f8 451 bio->bi_bdev = sbi->sb->s_bdev;
04d328de 452 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
4e49ea4a 453 ret = submit_bio_wait(bio);
0a87f664 454 atomic_dec(&fcc->submit_flush);
740432f8
JK		 455 bio_put(bio);
456 return ret;
457 }
6b4afdd7 458
adf8d90b 459 init_completion(&cmd.wait);
6b4afdd7 460
0a87f664 461 atomic_inc(&fcc->submit_flush);
721bd4d5 462 llist_add(&cmd.llnode, &fcc->issue_list);
6b4afdd7 463
a688b9d9
GZ		 464 if (!fcc->dispatch_list)
465 wake_up(&fcc->flush_wait_queue);
6b4afdd7 466
adf8d90b 467 wait_for_completion(&cmd.wait);
0a87f664 468 atomic_dec(&fcc->submit_flush);
adf8d90b
CY		 469
470 return cmd.ret;
6b4afdd7
JK		 471}
472
2163d198
GZ		 473int create_flush_cmd_control(struct f2fs_sb_info *sbi)
474{
475 dev_t dev = sbi->sb->s_bdev->bd_dev;
476 struct flush_cmd_control *fcc;
477 int err = 0;
478
479 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
480 if (!fcc)
481 return -ENOMEM;
0a87f664 482 atomic_set(&fcc->submit_flush, 0);
2163d198 483 init_waitqueue_head(&fcc->flush_wait_queue);
721bd4d5 484 init_llist_head(&fcc->issue_list);
6b2920a5 485 SM_I(sbi)->cmd_control_info = fcc;
2163d198
GZ		 486 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
487 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
488 if (IS_ERR(fcc->f2fs_issue_flush)) {
489 err = PTR_ERR(fcc->f2fs_issue_flush);
490 kfree(fcc);
6b2920a5 491 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 492 return err;
493 }
2163d198
GZ		 494
495 return err;
496}
497
498void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
499{
6b2920a5 500 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
2163d198
GZ		 501
502 if (fcc && fcc->f2fs_issue_flush)
503 kthread_stop(fcc->f2fs_issue_flush);
504 kfree(fcc);
6b2920a5 505 SM_I(sbi)->cmd_control_info = NULL;
2163d198
GZ		 506}
507
351df4b2
JK		 508static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
509 enum dirty_type dirty_type)
510{
511 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
512
513 /* need not be added */
514 if (IS_CURSEG(sbi, segno))
515 return;
516
517 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
518 dirty_i->nr_dirty[dirty_type]++;
519
520 if (dirty_type == DIRTY) {
521 struct seg_entry *sentry = get_seg_entry(sbi, segno);
4625d6aa 522 enum dirty_type t = sentry->type;
b2f2c390 523
ec325b52
JK		 524 if (unlikely(t >= DIRTY)) {
525 f2fs_bug_on(sbi, 1);
526 return;
527 }
4625d6aa
CL		 528 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
529 dirty_i->nr_dirty[t]++;
351df4b2
JK		 530 }
531}
532
533static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
534 enum dirty_type dirty_type)
535{
536 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
537
538 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
539 dirty_i->nr_dirty[dirty_type]--;
540
541 if (dirty_type == DIRTY) {
4625d6aa
CL		 542 struct seg_entry *sentry = get_seg_entry(sbi, segno);
543 enum dirty_type t = sentry->type;
544
545 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
546 dirty_i->nr_dirty[t]--;
b2f2c390 547
5ec4e49f
JK		 548 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
549 clear_bit(GET_SECNO(sbi, segno),
550 dirty_i->victim_secmap);
351df4b2
JK		 551 }
552}
553
0a8165d7 554/*
351df4b2
JK		 555 * Should not occur error such as -ENOMEM.
556 * Adding dirty entry into seglist is not critical operation.
557 * If a given segment is one of current working segments, it won't be added.
558 */
8d8451af 559static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 560{
561 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
562 unsigned short valid_blocks;
563
564 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
565 return;
566
567 mutex_lock(&dirty_i->seglist_lock);
568
569 valid_blocks = get_valid_blocks(sbi, segno, 0);
570
571 if (valid_blocks == 0) {
572 __locate_dirty_segment(sbi, segno, PRE);
573 __remove_dirty_segment(sbi, segno, DIRTY);
574 } else if (valid_blocks < sbi->blocks_per_seg) {
575 __locate_dirty_segment(sbi, segno, DIRTY);
576 } else {
577 /* Recovery routine with SSR needs this */
578 __remove_dirty_segment(sbi, segno, DIRTY);
579 }
580
581 mutex_unlock(&dirty_i->seglist_lock);
351df4b2
JK		 582}
583
275b66b0
CY		 584static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
585 struct bio *bio)
586{
587 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
588 struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
589
590 INIT_LIST_HEAD(&be->list);
591 be->bio = bio;
592 init_completion(&be->event);
593 list_add_tail(&be->list, wait_list);
594
595 return be;
596}
597
598void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
599{
600 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
601 struct bio_entry *be, *tmp;
602
603 list_for_each_entry_safe(be, tmp, wait_list, list) {
604 struct bio *bio = be->bio;
605 int err;
606
607 wait_for_completion_io(&be->event);
608 err = be->error;
609 if (err == -EOPNOTSUPP)
610 err = 0;
611
612 if (err)
613 f2fs_msg(sbi->sb, KERN_INFO,
614 "Issue discard failed, ret: %d", err);
615
616 bio_put(bio);
617 list_del(&be->list);
618 kmem_cache_free(bio_entry_slab, be);
619 }
620}
621
622static void f2fs_submit_bio_wait_endio(struct bio *bio)
623{
624 struct bio_entry *be = (struct bio_entry *)bio->bi_private;
625
626 be->error = bio->bi_error;
627 complete(&be->event);
628}
629
630/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
631int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi, sector_t sector,
632 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
633{
634 struct block_device *bdev = sbi->sb->s_bdev;
635 struct bio *bio = NULL;
636 int err;
637
638 err = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
639 &bio);
640 if (!err && bio) {
641 struct bio_entry *be = __add_bio_entry(sbi, bio);
642
643 bio->bi_private = be;
644 bio->bi_end_io = f2fs_submit_bio_wait_endio;
645 bio->bi_opf |= REQ_SYNC;
646 submit_bio(bio);
647 }
648
649 return err;
650}
651
1e87a78d 652static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
37208879
JK		 653 block_t blkstart, block_t blklen)
654{
55cf9cb6
CY		 655 sector_t start = SECTOR_FROM_BLOCK(blkstart);
656 sector_t len = SECTOR_FROM_BLOCK(blklen);
a66cdd98
JK		 657 struct seg_entry *se;
658 unsigned int offset;
659 block_t i;
660
661 for (i = blkstart; i < blkstart + blklen; i++) {
662 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
663 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
664
665 if (!f2fs_test_and_set_bit(offset, se->discard_map))
666 sbi->discard_blks--;
667 }
1661d07c 668 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
275b66b0 669 return __f2fs_issue_discard_async(sbi, start, len, GFP_NOFS, 0);
1e87a78d
JK		 670}
671
e90c2d28 672bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
1e87a78d 673{
60b286c4 674 int err = -EOPNOTSUPP;
40a02be1
JK		 675
676 if (test_opt(sbi, DISCARD)) {
677 struct seg_entry *se = get_seg_entry(sbi,
678 GET_SEGNO(sbi, blkaddr));
679 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
680
681 if (f2fs_test_bit(offset, se->discard_map))
e90c2d28 682 return false;
40a02be1
JK		 683
684 err = f2fs_issue_discard(sbi, blkaddr, 1);
685 }
686
e90c2d28 687 if (err) {
381722d2 688 update_meta_page(sbi, NULL, blkaddr);
e90c2d28
CY		 689 return true;
690 }
691 return false;
37208879
JK		 692}
693
adf4983b 694static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 695 struct cp_control *cpc, struct seg_entry *se,
696 unsigned int start, unsigned int end)
b2955550
JK		 697{
698 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 699 struct discard_entry *new, *last;
700
701 if (!list_empty(head)) {
702 last = list_last_entry(head, struct discard_entry, list);
703 if (START_BLOCK(sbi, cpc->trim_start) + start ==
704 last->blkaddr + last->len) {
705 last->len += end - start;
706 goto done;
707 }
708 }
709
710 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
711 INIT_LIST_HEAD(&new->list);
712 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
713 new->len = end - start;
714 list_add_tail(&new->list, head);
715done:
716 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 717}
718
719static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
720{
b2955550
JK		 721 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
722 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 723 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 724 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
725 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 726 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 727 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 728 unsigned int start = 0, end = -1;
4b2fecc8 729 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 730 int i;
731
3e025740 732 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
b2955550
JK		 733 return;
734
a66cdd98
JK		 735 if (!force) {
736 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 737 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
738 return;
4b2fecc8
JK		 739 }
740
b2955550
JK		 741 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
742 for (i = 0; i < entries; i++)
a66cdd98 743 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 744 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 745
4b2fecc8 746 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 747 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
748 if (start >= max_blocks)
749 break;
750
751 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 752 if (force && start && end != max_blocks
753 && (end - start) < cpc->trim_minlen)
754 continue;
755
a66cdd98 756 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 757 }
758}
759
4b2fecc8
JK		 760void release_discard_addrs(struct f2fs_sb_info *sbi)
761{
762 struct list_head *head = &(SM_I(sbi)->discard_list);
763 struct discard_entry *entry, *this;
764
765 /* drop caches */
766 list_for_each_entry_safe(entry, this, head, list) {
767 list_del(&entry->list);
768 kmem_cache_free(discard_entry_slab, entry);
769 }
770}
771
0a8165d7 772/*
351df4b2
JK		 773 * Should call clear_prefree_segments after checkpoint is done.
774 */
775static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
776{
777 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 778 unsigned int segno;
351df4b2
JK		 779
780 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 781 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 782 __set_test_and_free(sbi, segno);
351df4b2
JK		 783 mutex_unlock(&dirty_i->seglist_lock);
784}
785
836b5a63 786void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 787{
b2955550 788 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 789 struct discard_entry *entry, *this;
351df4b2 790 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
275b66b0 791 struct blk_plug plug;
29e59c14 792 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 793 unsigned int start = 0, end = -1;
36abef4e 794 unsigned int secno, start_segno;
c24a0fd6 795 bool force = (cpc->reason == CP_DISCARD);
351df4b2 796
275b66b0
CY		 797 blk_start_plug(&plug);
798
351df4b2 799 mutex_lock(&dirty_i->seglist_lock);
29e59c14 800
351df4b2 801 while (1) {
29e59c14 802 int i;
7cd8558b
JK		 803 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
804 if (start >= MAIN_SEGS(sbi))
351df4b2 805 break;
7cd8558b
JK		 806 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
807 start + 1);
29e59c14
CL		 808
809 for (i = start; i < end; i++)
810 clear_bit(i, prefree_map);
811
812 dirty_i->nr_dirty[PRE] -= end - start;
813
c24a0fd6 814 if (force || !test_opt(sbi, DISCARD))
29e59c14 815 continue;
351df4b2 816
36abef4e
JK		 817 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
818 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 819 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 820 continue;
821 }
822next:
823 secno = GET_SECNO(sbi, start);
824 start_segno = secno * sbi->segs_per_sec;
825 if (!IS_CURSEC(sbi, secno) &&
826 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
827 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
828 sbi->segs_per_sec << sbi->log_blocks_per_seg);
829
830 start = start_segno + sbi->segs_per_sec;
831 if (start < end)
832 goto next;
351df4b2
JK		 833 }
834 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 835
836 /* send small discards */
2d7b822a 837 list_for_each_entry_safe(entry, this, head, list) {
c24a0fd6 838 if (force && entry->len < cpc->trim_minlen)
836b5a63 839 goto skip;
37208879 840 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 841 cpc->trimmed += entry->len;
836b5a63 842skip:
b2955550
JK		 843 list_del(&entry->list);
844 SM_I(sbi)->nr_discards -= entry->len;
845 kmem_cache_free(discard_entry_slab, entry);
846 }
275b66b0
CY		 847
848 blk_finish_plug(&plug);
351df4b2
JK		 849}
850
184a5cd2 851static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 852{
853 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 854
855 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 856 sit_i->dirty_sentries++;
184a5cd2
CY		 857 return false;
858 }
859
860 return true;
351df4b2
JK		 861}
862
863static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
864 unsigned int segno, int modified)
865{
866 struct seg_entry *se = get_seg_entry(sbi, segno);
867 se->type = type;
868 if (modified)
869 __mark_sit_entry_dirty(sbi, segno);
870}
871
872static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
873{
874 struct seg_entry *se;
875 unsigned int segno, offset;
876 long int new_vblocks;
877
878 segno = GET_SEGNO(sbi, blkaddr);
879
880 se = get_seg_entry(sbi, segno);
881 new_vblocks = se->valid_blocks + del;
491c0854 882 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 883
9850cf4a 884 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 885 (new_vblocks > sbi->blocks_per_seg)));
886
887 se->valid_blocks = new_vblocks;
888 se->mtime = get_mtime(sbi);
889 SIT_I(sbi)->max_mtime = se->mtime;
890
891 /* Update valid block bitmap */
892 if (del > 0) {
52aca074 893 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 894 f2fs_bug_on(sbi, 1);
3e025740
JK		 895 if (f2fs_discard_en(sbi) &&
896 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 897 sbi->discard_blks--;
351df4b2 898 } else {
52aca074 899 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 900 f2fs_bug_on(sbi, 1);
3e025740
JK		 901 if (f2fs_discard_en(sbi) &&
902 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 903 sbi->discard_blks++;
351df4b2
JK		 904 }
905 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
906 se->ckpt_valid_blocks += del;
907
908 __mark_sit_entry_dirty(sbi, segno);
909
910 /* update total number of valid blocks to be written in ckpt area */
911 SIT_I(sbi)->written_valid_blocks += del;
912
913 if (sbi->segs_per_sec > 1)
914 get_sec_entry(sbi, segno)->valid_blocks += del;
915}
916
5e443818 917void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 918{
5e443818
JK		 919 update_sit_entry(sbi, new, 1);
920 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
921 update_sit_entry(sbi, old, -1);
922
923 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
924 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 925}
926
927void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
928{
929 unsigned int segno = GET_SEGNO(sbi, addr);
930 struct sit_info *sit_i = SIT_I(sbi);
931
9850cf4a 932 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 933 if (addr == NEW_ADDR)
934 return;
935
936 /* add it into sit main buffer */
937 mutex_lock(&sit_i->sentry_lock);
938
939 update_sit_entry(sbi, addr, -1);
940
941 /* add it into dirty seglist */
942 locate_dirty_segment(sbi, segno);
943
944 mutex_unlock(&sit_i->sentry_lock);
945}
946
6e2c64ad
JK		 947bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
948{
949 struct sit_info *sit_i = SIT_I(sbi);
950 unsigned int segno, offset;
951 struct seg_entry *se;
952 bool is_cp = false;
953
954 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
955 return true;
956
957 mutex_lock(&sit_i->sentry_lock);
958
959 segno = GET_SEGNO(sbi, blkaddr);
960 se = get_seg_entry(sbi, segno);
961 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
962
963 if (f2fs_test_bit(offset, se->ckpt_valid_map))
964 is_cp = true;
965
966 mutex_unlock(&sit_i->sentry_lock);
967
968 return is_cp;
969}
970
0a8165d7 971/*
351df4b2
JK		 972 * This function should be resided under the curseg_mutex lock
973 */
974static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 975 struct f2fs_summary *sum)
351df4b2
JK		 976{
977 struct curseg_info *curseg = CURSEG_I(sbi, type);
978 void *addr = curseg->sum_blk;
e79efe3b 979 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 980 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 981}
982
0a8165d7 983/*
351df4b2
JK		 984 * Calculate the number of current summary pages for writing
985 */
3fa06d7b 986int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 987{
351df4b2 988 int valid_sum_count = 0;
9a47938b 989 int i, sum_in_page;
351df4b2
JK		 990
991 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
992 if (sbi->ckpt->alloc_type[i] == SSR)
993 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 994 else {
995 if (for_ra)
996 valid_sum_count += le16_to_cpu(
997 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
998 else
999 valid_sum_count += curseg_blkoff(sbi, i);
1000 }
351df4b2
JK		 1001 }
1002
09cbfeaf 1003 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 1004 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1005 if (valid_sum_count <= sum_in_page)
351df4b2 1006 return 1;
9a47938b 1007 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 1008 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 1009 return 2;
1010 return 3;
1011}
1012
0a8165d7 1013/*
351df4b2
JK		 1014 * Caller should put this summary page
1015 */
1016struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1017{
1018 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1019}
1020
381722d2 1021void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1022{
1023 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1024 void *dst = page_address(page);
1025
1026 if (src)
09cbfeaf 1027 memcpy(dst, src, PAGE_SIZE);
381722d2 1028 else
09cbfeaf 1029 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1030 set_page_dirty(page);
1031 f2fs_put_page(page, 1);
1032}
1033
381722d2
CY		 1034static void write_sum_page(struct f2fs_sb_info *sbi,
1035 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1036{
1037 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1038}
1039
b7ad7512
CY		 1040static void write_current_sum_page(struct f2fs_sb_info *sbi,
1041 int type, block_t blk_addr)
1042{
1043 struct curseg_info *curseg = CURSEG_I(sbi, type);
1044 struct page *page = grab_meta_page(sbi, blk_addr);
1045 struct f2fs_summary_block *src = curseg->sum_blk;
1046 struct f2fs_summary_block *dst;
1047
1048 dst = (struct f2fs_summary_block *)page_address(page);
1049
1050 mutex_lock(&curseg->curseg_mutex);
1051
1052 down_read(&curseg->journal_rwsem);
1053 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1054 up_read(&curseg->journal_rwsem);
1055
1056 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1057 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1058
1059 mutex_unlock(&curseg->curseg_mutex);
1060
1061 set_page_dirty(page);
1062 f2fs_put_page(page, 1);
1063}
1064
60374688
JK		 1065static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1066{
1067 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 1068 unsigned int segno = curseg->segno + 1;
60374688
JK		 1069 struct free_segmap_info *free_i = FREE_I(sbi);
1070
7cd8558b 1071 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 1072 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 1073 return 0;
1074}
1075
0a8165d7 1076/*
351df4b2
JK		 1077 * Find a new segment from the free segments bitmap to right order
1078 * This function should be returned with success, otherwise BUG
1079 */
1080static void get_new_segment(struct f2fs_sb_info *sbi,
1081 unsigned int *newseg, bool new_sec, int dir)
1082{
1083 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1084 unsigned int segno, secno, zoneno;
7cd8558b 1085 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1086 unsigned int hint = *newseg / sbi->segs_per_sec;
1087 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1088 unsigned int left_start = hint;
1089 bool init = true;
1090 int go_left = 0;
1091 int i;
1092
1a118ccf 1093 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1094
1095 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1096 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1097 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1098 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1099 goto got_it;
1100 }
1101find_other_zone:
7cd8558b
JK		 1102 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1103 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1104 if (dir == ALLOC_RIGHT) {
1105 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1106 MAIN_SECS(sbi), 0);
1107 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1108 } else {
1109 go_left = 1;
1110 left_start = hint - 1;
1111 }
1112 }
1113 if (go_left == 0)
1114 goto skip_left;
1115
1116 while (test_bit(left_start, free_i->free_secmap)) {
1117 if (left_start > 0) {
1118 left_start--;
1119 continue;
1120 }
1121 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1122 MAIN_SECS(sbi), 0);
1123 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1124 break;
1125 }
1126 secno = left_start;
1127skip_left:
1128 hint = secno;
1129 segno = secno * sbi->segs_per_sec;
1130 zoneno = secno / sbi->secs_per_zone;
1131
1132 /* give up on finding another zone */
1133 if (!init)
1134 goto got_it;
1135 if (sbi->secs_per_zone == 1)
1136 goto got_it;
1137 if (zoneno == old_zoneno)
1138 goto got_it;
1139 if (dir == ALLOC_LEFT) {
1140 if (!go_left && zoneno + 1 >= total_zones)
1141 goto got_it;
1142 if (go_left && zoneno == 0)
1143 goto got_it;
1144 }
1145 for (i = 0; i < NR_CURSEG_TYPE; i++)
1146 if (CURSEG_I(sbi, i)->zone == zoneno)
1147 break;
1148
1149 if (i < NR_CURSEG_TYPE) {
1150 /* zone is in user, try another */
1151 if (go_left)
1152 hint = zoneno * sbi->secs_per_zone - 1;
1153 else if (zoneno + 1 >= total_zones)
1154 hint = 0;
1155 else
1156 hint = (zoneno + 1) * sbi->secs_per_zone;
1157 init = false;
1158 goto find_other_zone;
1159 }
1160got_it:
1161 /* set it as dirty segment in free segmap */
9850cf4a 1162 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1163 __set_inuse(sbi, segno);
1164 *newseg = segno;
1a118ccf 1165 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1166}
1167
1168static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1169{
1170 struct curseg_info *curseg = CURSEG_I(sbi, type);
1171 struct summary_footer *sum_footer;
1172
1173 curseg->segno = curseg->next_segno;
1174 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1175 curseg->next_blkoff = 0;
1176 curseg->next_segno = NULL_SEGNO;
1177
1178 sum_footer = &(curseg->sum_blk->footer);
1179 memset(sum_footer, 0, sizeof(struct summary_footer));
1180 if (IS_DATASEG(type))
1181 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1182 if (IS_NODESEG(type))
1183 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1184 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1185}
1186
0a8165d7 1187/*
351df4b2
JK		 1188 * Allocate a current working segment.
1189 * This function always allocates a free segment in LFS manner.
1190 */
1191static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1192{
1193 struct curseg_info *curseg = CURSEG_I(sbi, type);
1194 unsigned int segno = curseg->segno;
1195 int dir = ALLOC_LEFT;
1196
1197 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1198 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1199 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1200 dir = ALLOC_RIGHT;
1201
1202 if (test_opt(sbi, NOHEAP))
1203 dir = ALLOC_RIGHT;
1204
1205 get_new_segment(sbi, &segno, new_sec, dir);
1206 curseg->next_segno = segno;
1207 reset_curseg(sbi, type, 1);
1208 curseg->alloc_type = LFS;
1209}
1210
1211static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1212 struct curseg_info *seg, block_t start)
1213{
1214 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1215 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1216 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1217 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1218 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1219 int i, pos;
1220
1221 for (i = 0; i < entries; i++)
1222 target_map[i] = ckpt_map[i] | cur_map[i];
1223
1224 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1225
1226 seg->next_blkoff = pos;
351df4b2
JK		 1227}
1228
0a8165d7 1229/*
351df4b2
JK		 1230 * If a segment is written by LFS manner, next block offset is just obtained
1231 * by increasing the current block offset. However, if a segment is written by
1232 * SSR manner, next block offset obtained by calling __next_free_blkoff
1233 */
1234static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1235 struct curseg_info *seg)
1236{
1237 if (seg->alloc_type == SSR)
1238 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1239 else
1240 seg->next_blkoff++;
1241}
1242
0a8165d7 1243/*
e1c42045 1244 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1245 * manner, so it should recover the existing segment information of valid blocks
1246 */
1247static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1248{
1249 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1250 struct curseg_info *curseg = CURSEG_I(sbi, type);
1251 unsigned int new_segno = curseg->next_segno;
1252 struct f2fs_summary_block *sum_node;
1253 struct page *sum_page;
1254
1255 write_sum_page(sbi, curseg->sum_blk,
1256 GET_SUM_BLOCK(sbi, curseg->segno));
1257 __set_test_and_inuse(sbi, new_segno);
1258
1259 mutex_lock(&dirty_i->seglist_lock);
1260 __remove_dirty_segment(sbi, new_segno, PRE);
1261 __remove_dirty_segment(sbi, new_segno, DIRTY);
1262 mutex_unlock(&dirty_i->seglist_lock);
1263
1264 reset_curseg(sbi, type, 1);
1265 curseg->alloc_type = SSR;
1266 __next_free_blkoff(sbi, curseg, 0);
1267
1268 if (reuse) {
1269 sum_page = get_sum_page(sbi, new_segno);
1270 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1271 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1272 f2fs_put_page(sum_page, 1);
1273 }
1274}
1275
43727527
JK		 1276static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1277{
1278 struct curseg_info *curseg = CURSEG_I(sbi, type);
1279 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1280
1281 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1282 return v_ops->get_victim(sbi,
1283 &(curseg)->next_segno, BG_GC, type, SSR);
1284
1285 /* For data segments, let's do SSR more intensively */
1286 for (; type >= CURSEG_HOT_DATA; type--)
1287 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1288 BG_GC, type, SSR))
1289 return 1;
1290 return 0;
1291}
1292
351df4b2
JK		 1293/*
1294 * flush out current segment and replace it with new segment
1295 * This function should be returned with success, otherwise BUG
1296 */
1297static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1298 int type, bool force)
1299{
1300 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1301
7b405275 1302 if (force)
351df4b2 1303 new_curseg(sbi, type, true);
7b405275 1304 else if (type == CURSEG_WARM_NODE)
351df4b2 1305 new_curseg(sbi, type, false);
60374688
JK		 1306 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1307 new_curseg(sbi, type, false);
351df4b2
JK		 1308 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1309 change_curseg(sbi, type, true);
1310 else
1311 new_curseg(sbi, type, false);
dcdfff65
JK		 1312
1313 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1314}
1315
38aa0889
JK		 1316static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1317{
1318 struct curseg_info *curseg = CURSEG_I(sbi, type);
1319 unsigned int old_segno;
1320
1321 old_segno = curseg->segno;
1322 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1323 locate_dirty_segment(sbi, old_segno);
1324}
1325
351df4b2
JK		 1326void allocate_new_segments(struct f2fs_sb_info *sbi)
1327{
351df4b2
JK		 1328 int i;
1329
36abef4e
JK		 1330 if (test_opt(sbi, LFS))
1331 return;
1332
38aa0889
JK		 1333 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1334 __allocate_new_segments(sbi, i);
351df4b2
JK		 1335}
1336
1337static const struct segment_allocation default_salloc_ops = {
1338 .allocate_segment = allocate_segment_by_default,
1339};
1340
4b2fecc8
JK		 1341int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1342{
f7ef9b83
JK		 1343 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1344 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1345 unsigned int start_segno, end_segno;
1346 struct cp_control cpc;
c34f42e2 1347 int err = 0;
4b2fecc8 1348
836b5a63 1349 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1350 return -EINVAL;
1351
9bd27ae4 1352 cpc.trimmed = 0;
7cd8558b 1353 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1354 goto out;
1355
1356 /* start/end segment number in main_area */
7cd8558b
JK		 1357 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1358 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1359 GET_SEGNO(sbi, end);
4b2fecc8 1360 cpc.reason = CP_DISCARD;
836b5a63 1361 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1362
1363 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1364 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1365 cpc.trim_start = start_segno;
a66cdd98
JK		 1366
1367 if (sbi->discard_blks == 0)
1368 break;
1369 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1370 cpc.trim_end = end_segno;
1371 else
1372 cpc.trim_end = min_t(unsigned int,
1373 rounddown(start_segno +
bba681cb
JK		 1374 BATCHED_TRIM_SEGMENTS(sbi),
1375 sbi->segs_per_sec) - 1, end_segno);
1376
1377 mutex_lock(&sbi->gc_mutex);
c34f42e2 1378 err = write_checkpoint(sbi, &cpc);
bba681cb 1379 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 1380 if (err)
1381 break;
74fa5f3d
CY		 1382
1383 schedule();
bba681cb 1384 }
4b2fecc8 1385out:
f7ef9b83 1386 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1387 return err;
4b2fecc8
JK		 1388}
1389
351df4b2
JK		 1390static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1391{
1392 struct curseg_info *curseg = CURSEG_I(sbi, type);
1393 if (curseg->next_blkoff < sbi->blocks_per_seg)
1394 return true;
1395 return false;
1396}
1397
1398static int __get_segment_type_2(struct page *page, enum page_type p_type)
1399{
1400 if (p_type == DATA)
1401 return CURSEG_HOT_DATA;
1402 else
1403 return CURSEG_HOT_NODE;
1404}
1405
1406static int __get_segment_type_4(struct page *page, enum page_type p_type)
1407{
1408 if (p_type == DATA) {
1409 struct inode *inode = page->mapping->host;
1410
1411 if (S_ISDIR(inode->i_mode))
1412 return CURSEG_HOT_DATA;
1413 else
1414 return CURSEG_COLD_DATA;
1415 } else {
a344b9fd
JK		 1416 if (IS_DNODE(page) && is_cold_node(page))
1417 return CURSEG_WARM_NODE;
351df4b2
JK		 1418 else
1419 return CURSEG_COLD_NODE;
1420 }
1421}
1422
1423static int __get_segment_type_6(struct page *page, enum page_type p_type)
1424{
1425 if (p_type == DATA) {
1426 struct inode *inode = page->mapping->host;
1427
1428 if (S_ISDIR(inode->i_mode))
1429 return CURSEG_HOT_DATA;
354a3399 1430 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1431 return CURSEG_COLD_DATA;
1432 else
1433 return CURSEG_WARM_DATA;
1434 } else {
1435 if (IS_DNODE(page))
1436 return is_cold_node(page) ? CURSEG_WARM_NODE :
1437 CURSEG_HOT_NODE;
1438 else
1439 return CURSEG_COLD_NODE;
1440 }
1441}
1442
1443static int __get_segment_type(struct page *page, enum page_type p_type)
1444{
4081363f 1445 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1446 case 2:
1447 return __get_segment_type_2(page, p_type);
1448 case 4:
1449 return __get_segment_type_4(page, p_type);
351df4b2 1450 }
12a67146 1451 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1452 f2fs_bug_on(F2FS_P_SB(page),
1453 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1454 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1455}
1456
bfad7c2d
JK		 1457void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1458 block_t old_blkaddr, block_t *new_blkaddr,
1459 struct f2fs_summary *sum, int type)
351df4b2
JK		 1460{
1461 struct sit_info *sit_i = SIT_I(sbi);
1462 struct curseg_info *curseg;
38aa0889
JK		 1463 bool direct_io = (type == CURSEG_DIRECT_IO);
1464
1465 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1466
351df4b2
JK		 1467 curseg = CURSEG_I(sbi, type);
1468
1469 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1470 mutex_lock(&sit_i->sentry_lock);
351df4b2 1471
38aa0889 1472 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4
JK		 1473 if (direct_io && curseg->next_blkoff &&
1474 !has_not_enough_free_secs(sbi, 0))
38aa0889
JK		 1475 __allocate_new_segments(sbi, type);
1476
351df4b2 1477 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1478
1479 /*
1480 * __add_sum_entry should be resided under the curseg_mutex
1481 * because, this function updates a summary entry in the
1482 * current summary block.
1483 */
e79efe3b 1484 __add_sum_entry(sbi, type, sum);
351df4b2 1485
351df4b2 1486 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1487
1488 stat_inc_block_count(sbi, curseg);
351df4b2 1489
5e443818
JK		 1490 if (!__has_curseg_space(sbi, type))
1491 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1492 /*
1493 * SIT information should be updated before segment allocation,
1494 * since SSR needs latest valid block information.
1495 */
1496 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1497
351df4b2
JK		 1498 mutex_unlock(&sit_i->sentry_lock);
1499
bfad7c2d 1500 if (page && IS_NODESEG(type))
351df4b2
JK		 1501 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1502
bfad7c2d
JK		 1503 mutex_unlock(&curseg->curseg_mutex);
1504}
1505
05ca3632 1506static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1507{
05ca3632 1508 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1509
7dfeaa32
JK		 1510 if (fio->type == NODE || fio->type == DATA)
1511 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1512
7a9d7548
CY		 1513 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1514 &fio->new_blkaddr, sum, type);
bfad7c2d 1515
351df4b2 1516 /* writeout dirty page into bdev */
05ca3632 1517 f2fs_submit_page_mbio(fio);
7dfeaa32
JK		 1518
1519 if (fio->type == NODE || fio->type == DATA)
1520 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1521}
1522
577e3495 1523void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1524{
458e6197 1525 struct f2fs_io_info fio = {
05ca3632 1526 .sbi = sbi,
458e6197 1527 .type = META,
04d328de
MC		 1528 .op = REQ_OP_WRITE,
1529 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1530 .old_blkaddr = page->index,
1531 .new_blkaddr = page->index,
05ca3632 1532 .page = page,
4375a336 1533 .encrypted_page = NULL,
458e6197
JK		 1534 };
1535
2b947003 1536 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1537 fio.op_flags &= ~REQ_META;
2b947003 1538
351df4b2 1539 set_page_writeback(page);
05ca3632 1540 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1541}
1542
05ca3632 1543void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1544{
1545 struct f2fs_summary sum;
05ca3632 1546
351df4b2 1547 set_summary(&sum, nid, 0, 0);
05ca3632 1548 do_write_page(&sum, fio);
351df4b2
JK		 1549}
1550
05ca3632 1551void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1552{
05ca3632 1553 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1554 struct f2fs_summary sum;
1555 struct node_info ni;
1556
9850cf4a 1557 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1558 get_node_info(sbi, dn->nid, &ni);
1559 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1560 do_write_page(&sum, fio);
f28b3434 1561 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1562}
1563
05ca3632 1564void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1565{
7a9d7548 1566 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1567 stat_inc_inplace_blocks(fio->sbi);
1568 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1569}
1570
4356e48e 1571void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1572 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1573 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1574{
1575 struct sit_info *sit_i = SIT_I(sbi);
1576 struct curseg_info *curseg;
1577 unsigned int segno, old_cursegno;
1578 struct seg_entry *se;
1579 int type;
19f106bc 1580 unsigned short old_blkoff;
351df4b2
JK		 1581
1582 segno = GET_SEGNO(sbi, new_blkaddr);
1583 se = get_seg_entry(sbi, segno);
1584 type = se->type;
1585
19f106bc
CY		 1586 if (!recover_curseg) {
1587 /* for recovery flow */
1588 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1589 if (old_blkaddr == NULL_ADDR)
1590 type = CURSEG_COLD_DATA;
1591 else
1592 type = CURSEG_WARM_DATA;
1593 }
1594 } else {
1595 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1596 type = CURSEG_WARM_DATA;
1597 }
19f106bc 1598
351df4b2
JK		 1599 curseg = CURSEG_I(sbi, type);
1600
1601 mutex_lock(&curseg->curseg_mutex);
1602 mutex_lock(&sit_i->sentry_lock);
1603
1604 old_cursegno = curseg->segno;
19f106bc 1605 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1606
1607 /* change the current segment */
1608 if (segno != curseg->segno) {
1609 curseg->next_segno = segno;
1610 change_curseg(sbi, type, true);
1611 }
1612
491c0854 1613 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1614 __add_sum_entry(sbi, type, sum);
351df4b2 1615
28bc106b 1616 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1617 update_sit_entry(sbi, new_blkaddr, 1);
1618 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1619 update_sit_entry(sbi, old_blkaddr, -1);
1620
1621 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1622 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1623
351df4b2 1624 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1625
19f106bc
CY		 1626 if (recover_curseg) {
1627 if (old_cursegno != curseg->segno) {
1628 curseg->next_segno = old_cursegno;
1629 change_curseg(sbi, type, true);
1630 }
1631 curseg->next_blkoff = old_blkoff;
1632 }
1633
351df4b2
JK		 1634 mutex_unlock(&sit_i->sentry_lock);
1635 mutex_unlock(&curseg->curseg_mutex);
1636}
1637
528e3459
CY		 1638void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1639 block_t old_addr, block_t new_addr,
28bc106b
CY		 1640 unsigned char version, bool recover_curseg,
1641 bool recover_newaddr)
528e3459
CY		 1642{
1643 struct f2fs_summary sum;
1644
1645 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1646
28bc106b
CY		 1647 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1648 recover_curseg, recover_newaddr);
528e3459 1649
f28b3434 1650 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1651}
1652
93dfe2ac 1653void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1654 enum page_type type, bool ordered)
93dfe2ac 1655{
93dfe2ac 1656 if (PageWriteback(page)) {
4081363f
JK		 1657 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1658
0c3a5797 1659 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1660 if (ordered)
1661 wait_on_page_writeback(page);
1662 else
1663 wait_for_stable_page(page);
93dfe2ac
JK		 1664 }
1665}
1666
08b39fbd
CY		 1667void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1668 block_t blkaddr)
1669{
1670 struct page *cpage;
1671
1672 if (blkaddr == NEW_ADDR)
1673 return;
1674
1675 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1676
1677 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1678 if (cpage) {
fec1d657 1679 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1680 f2fs_put_page(cpage, 1);
1681 }
1682}
1683
351df4b2
JK		 1684static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1685{
1686 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1687 struct curseg_info *seg_i;
1688 unsigned char *kaddr;
1689 struct page *page;
1690 block_t start;
1691 int i, j, offset;
1692
1693 start = start_sum_block(sbi);
1694
1695 page = get_meta_page(sbi, start++);
1696 kaddr = (unsigned char *)page_address(page);
1697
1698 /* Step 1: restore nat cache */
1699 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1700 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1701
1702 /* Step 2: restore sit cache */
1703 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1704 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1705 offset = 2 * SUM_JOURNAL_SIZE;
1706
1707 /* Step 3: restore summary entries */
1708 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1709 unsigned short blk_off;
1710 unsigned int segno;
1711
1712 seg_i = CURSEG_I(sbi, i);
1713 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1714 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1715 seg_i->next_segno = segno;
1716 reset_curseg(sbi, i, 0);
1717 seg_i->alloc_type = ckpt->alloc_type[i];
1718 seg_i->next_blkoff = blk_off;
1719
1720 if (seg_i->alloc_type == SSR)
1721 blk_off = sbi->blocks_per_seg;
1722
1723 for (j = 0; j < blk_off; j++) {
1724 struct f2fs_summary *s;
1725 s = (struct f2fs_summary *)(kaddr + offset);
1726 seg_i->sum_blk->entries[j] = *s;
1727 offset += SUMMARY_SIZE;
09cbfeaf 1728 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1729 SUM_FOOTER_SIZE)
1730 continue;
1731
1732 f2fs_put_page(page, 1);
1733 page = NULL;
1734
1735 page = get_meta_page(sbi, start++);
1736 kaddr = (unsigned char *)page_address(page);
1737 offset = 0;
1738 }
1739 }
1740 f2fs_put_page(page, 1);
1741 return 0;
1742}
1743
1744static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1745{
1746 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1747 struct f2fs_summary_block *sum;
1748 struct curseg_info *curseg;
1749 struct page *new;
1750 unsigned short blk_off;
1751 unsigned int segno = 0;
1752 block_t blk_addr = 0;
1753
1754 /* get segment number and block addr */
1755 if (IS_DATASEG(type)) {
1756 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1757 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1758 CURSEG_HOT_DATA]);
119ee914 1759 if (__exist_node_summaries(sbi))
351df4b2
JK		 1760 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1761 else
1762 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1763 } else {
1764 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1765 CURSEG_HOT_NODE]);
1766 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1767 CURSEG_HOT_NODE]);
119ee914 1768 if (__exist_node_summaries(sbi))
351df4b2
JK		 1769 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1770 type - CURSEG_HOT_NODE);
1771 else
1772 blk_addr = GET_SUM_BLOCK(sbi, segno);
1773 }
1774
1775 new = get_meta_page(sbi, blk_addr);
1776 sum = (struct f2fs_summary_block *)page_address(new);
1777
1778 if (IS_NODESEG(type)) {
119ee914 1779 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1780 struct f2fs_summary *ns = &sum->entries[0];
1781 int i;
1782 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1783 ns->version = 0;
1784 ns->ofs_in_node = 0;
1785 }
1786 } else {
d653788a
GZ		 1787 int err;
1788
1789 err = restore_node_summary(sbi, segno, sum);
1790 if (err) {
351df4b2 1791 f2fs_put_page(new, 1);
d653788a 1792 return err;
351df4b2
JK		 1793 }
1794 }
1795 }
1796
1797 /* set uncompleted segment to curseg */
1798 curseg = CURSEG_I(sbi, type);
1799 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1800
1801 /* update journal info */
1802 down_write(&curseg->journal_rwsem);
1803 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1804 up_write(&curseg->journal_rwsem);
1805
1806 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1807 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1808 curseg->next_segno = segno;
1809 reset_curseg(sbi, type, 0);
1810 curseg->alloc_type = ckpt->alloc_type[type];
1811 curseg->next_blkoff = blk_off;
1812 mutex_unlock(&curseg->curseg_mutex);
1813 f2fs_put_page(new, 1);
1814 return 0;
1815}
1816
1817static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1818{
1819 int type = CURSEG_HOT_DATA;
e4fc5fbf 1820 int err;
351df4b2 1821
25ca923b 1822 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1823 int npages = npages_for_summary_flush(sbi, true);
1824
1825 if (npages >= 2)
1826 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1827 META_CP, true);
3fa06d7b 1828
351df4b2
JK		 1829 /* restore for compacted data summary */
1830 if (read_compacted_summaries(sbi))
1831 return -EINVAL;
1832 type = CURSEG_HOT_NODE;
1833 }
1834
119ee914 1835 if (__exist_node_summaries(sbi))
3fa06d7b 1836 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1837 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1838
e4fc5fbf
CY		 1839 for (; type <= CURSEG_COLD_NODE; type++) {
1840 err = read_normal_summaries(sbi, type);
1841 if (err)
1842 return err;
1843 }
1844
351df4b2
JK		 1845 return 0;
1846}
1847
1848static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1849{
1850 struct page *page;
1851 unsigned char *kaddr;
1852 struct f2fs_summary *summary;
1853 struct curseg_info *seg_i;
1854 int written_size = 0;
1855 int i, j;
1856
1857 page = grab_meta_page(sbi, blkaddr++);
1858 kaddr = (unsigned char *)page_address(page);
1859
1860 /* Step 1: write nat cache */
1861 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1862 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1863 written_size += SUM_JOURNAL_SIZE;
1864
1865 /* Step 2: write sit cache */
1866 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1867 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1868 written_size += SUM_JOURNAL_SIZE;
1869
351df4b2
JK		 1870 /* Step 3: write summary entries */
1871 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1872 unsigned short blkoff;
1873 seg_i = CURSEG_I(sbi, i);
1874 if (sbi->ckpt->alloc_type[i] == SSR)
1875 blkoff = sbi->blocks_per_seg;
1876 else
1877 blkoff = curseg_blkoff(sbi, i);
1878
1879 for (j = 0; j < blkoff; j++) {
1880 if (!page) {
1881 page = grab_meta_page(sbi, blkaddr++);
1882 kaddr = (unsigned char *)page_address(page);
1883 written_size = 0;
1884 }
1885 summary = (struct f2fs_summary *)(kaddr + written_size);
1886 *summary = seg_i->sum_blk->entries[j];
1887 written_size += SUMMARY_SIZE;
351df4b2 1888
09cbfeaf 1889 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1890 SUM_FOOTER_SIZE)
1891 continue;
1892
e8d61a74 1893 set_page_dirty(page);
351df4b2
JK		 1894 f2fs_put_page(page, 1);
1895 page = NULL;
1896 }
1897 }
e8d61a74
CY		 1898 if (page) {
1899 set_page_dirty(page);
351df4b2 1900 f2fs_put_page(page, 1);
e8d61a74 1901 }
351df4b2
JK		 1902}
1903
1904static void write_normal_summaries(struct f2fs_sb_info *sbi,
1905 block_t blkaddr, int type)
1906{
1907 int i, end;
1908 if (IS_DATASEG(type))
1909 end = type + NR_CURSEG_DATA_TYPE;
1910 else
1911 end = type + NR_CURSEG_NODE_TYPE;
1912
b7ad7512
CY		 1913 for (i = type; i < end; i++)
1914 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1915}
1916
1917void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1918{
25ca923b 1919 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1920 write_compacted_summaries(sbi, start_blk);
1921 else
1922 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1923}
1924
1925void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1926{
119ee914 1927 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1928}
1929
dfc08a12 1930int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1931 unsigned int val, int alloc)
1932{
1933 int i;
1934
1935 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1936 for (i = 0; i < nats_in_cursum(journal); i++) {
1937 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1938 return i;
1939 }
dfc08a12
CY		 1940 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1941 return update_nats_in_cursum(journal, 1);
351df4b2 1942 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1943 for (i = 0; i < sits_in_cursum(journal); i++)
1944 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1945 return i;
dfc08a12
CY		 1946 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1947 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1948 }
1949 return -1;
1950}
1951
1952static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1953 unsigned int segno)
1954{
2cc22186 1955 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1956}
1957
1958static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1959 unsigned int start)
1960{
1961 struct sit_info *sit_i = SIT_I(sbi);
1962 struct page *src_page, *dst_page;
1963 pgoff_t src_off, dst_off;
1964 void *src_addr, *dst_addr;
1965
1966 src_off = current_sit_addr(sbi, start);
1967 dst_off = next_sit_addr(sbi, src_off);
1968
1969 /* get current sit block page without lock */
1970 src_page = get_meta_page(sbi, src_off);
1971 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1972 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1973
1974 src_addr = page_address(src_page);
1975 dst_addr = page_address(dst_page);
09cbfeaf 1976 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1977
1978 set_page_dirty(dst_page);
1979 f2fs_put_page(src_page, 1);
1980
1981 set_to_next_sit(sit_i, start);
1982
1983 return dst_page;
1984}
1985
184a5cd2
CY		 1986static struct sit_entry_set *grab_sit_entry_set(void)
1987{
1988 struct sit_entry_set *ses =
80c54505 1989 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1990
1991 ses->entry_cnt = 0;
1992 INIT_LIST_HEAD(&ses->set_list);
1993 return ses;
1994}
1995
1996static void release_sit_entry_set(struct sit_entry_set *ses)
1997{
1998 list_del(&ses->set_list);
1999 kmem_cache_free(sit_entry_set_slab, ses);
2000}
2001
2002static void adjust_sit_entry_set(struct sit_entry_set *ses,
2003 struct list_head *head)
2004{
2005 struct sit_entry_set *next = ses;
2006
2007 if (list_is_last(&ses->set_list, head))
2008 return;
2009
2010 list_for_each_entry_continue(next, head, set_list)
2011 if (ses->entry_cnt <= next->entry_cnt)
2012 break;
2013
2014 list_move_tail(&ses->set_list, &next->set_list);
2015}
2016
2017static void add_sit_entry(unsigned int segno, struct list_head *head)
2018{
2019 struct sit_entry_set *ses;
2020 unsigned int start_segno = START_SEGNO(segno);
2021
2022 list_for_each_entry(ses, head, set_list) {
2023 if (ses->start_segno == start_segno) {
2024 ses->entry_cnt++;
2025 adjust_sit_entry_set(ses, head);
2026 return;
2027 }
2028 }
2029
2030 ses = grab_sit_entry_set();
2031
2032 ses->start_segno = start_segno;
2033 ses->entry_cnt++;
2034 list_add(&ses->set_list, head);
2035}
2036
2037static void add_sits_in_set(struct f2fs_sb_info *sbi)
2038{
2039 struct f2fs_sm_info *sm_info = SM_I(sbi);
2040 struct list_head *set_list = &sm_info->sit_entry_set;
2041 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2042 unsigned int segno;
2043
7cd8558b 2044 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2045 add_sit_entry(segno, set_list);
2046}
2047
2048static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2049{
2050 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2051 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2052 int i;
2053
b7ad7512 2054 down_write(&curseg->journal_rwsem);
dfc08a12 2055 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2056 unsigned int segno;
2057 bool dirtied;
2058
dfc08a12 2059 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2060 dirtied = __mark_sit_entry_dirty(sbi, segno);
2061
2062 if (!dirtied)
2063 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2064 }
dfc08a12 2065 update_sits_in_cursum(journal, -i);
b7ad7512 2066 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2067}
2068
0a8165d7 2069/*
351df4b2
JK		 2070 * CP calls this function, which flushes SIT entries including sit_journal,
2071 * and moves prefree segs to free segs.
2072 */
4b2fecc8 2073void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2074{
2075 struct sit_info *sit_i = SIT_I(sbi);
2076 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2077 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2078 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2079 struct sit_entry_set *ses, *tmp;
2080 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2081 bool to_journal = true;
4b2fecc8 2082 struct seg_entry *se;
351df4b2 2083
351df4b2
JK		 2084 mutex_lock(&sit_i->sentry_lock);
2085
2b11a74b
WL		 2086 if (!sit_i->dirty_sentries)
2087 goto out;
2088
351df4b2 2089 /*
184a5cd2
CY		 2090 * add and account sit entries of dirty bitmap in sit entry
2091 * set temporarily
351df4b2 2092 */
184a5cd2 2093 add_sits_in_set(sbi);
351df4b2 2094
184a5cd2
CY		 2095 /*
2096 * if there are no enough space in journal to store dirty sit
2097 * entries, remove all entries from journal and add and account
2098 * them in sit entry set.
2099 */
dfc08a12 2100 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2101 remove_sits_in_journal(sbi);
b2955550 2102
184a5cd2
CY		 2103 /*
2104 * there are two steps to flush sit entries:
2105 * #1, flush sit entries to journal in current cold data summary block.
2106 * #2, flush sit entries to sit page.
2107 */
2108 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2109 struct page *page = NULL;
184a5cd2
CY		 2110 struct f2fs_sit_block *raw_sit = NULL;
2111 unsigned int start_segno = ses->start_segno;
2112 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2113 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2114 unsigned int segno = start_segno;
2115
2116 if (to_journal &&
dfc08a12 2117 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2118 to_journal = false;
2119
b7ad7512
CY		 2120 if (to_journal) {
2121 down_write(&curseg->journal_rwsem);
2122 } else {
184a5cd2
CY		 2123 page = get_next_sit_page(sbi, start_segno);
2124 raw_sit = page_address(page);
351df4b2 2125 }
351df4b2 2126
184a5cd2
CY		 2127 /* flush dirty sit entries in region of current sit set */
2128 for_each_set_bit_from(segno, bitmap, end) {
2129 int offset, sit_offset;
4b2fecc8
JK		 2130
2131 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2132
2133 /* add discard candidates */
d7bc2484 2134 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2135 cpc->trim_start = segno;
2136 add_discard_addrs(sbi, cpc);
2137 }
184a5cd2
CY		 2138
2139 if (to_journal) {
dfc08a12 2140 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2141 SIT_JOURNAL, segno, 1);
2142 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2143 segno_in_journal(journal, offset) =
184a5cd2
CY		 2144 cpu_to_le32(segno);
2145 seg_info_to_raw_sit(se,
dfc08a12 2146 &sit_in_journal(journal, offset));
184a5cd2
CY		 2147 } else {
2148 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2149 seg_info_to_raw_sit(se,
2150 &raw_sit->entries[sit_offset]);
2151 }
351df4b2 2152
184a5cd2
CY		 2153 __clear_bit(segno, bitmap);
2154 sit_i->dirty_sentries--;
2155 ses->entry_cnt--;
351df4b2
JK		 2156 }
2157
b7ad7512
CY		 2158 if (to_journal)
2159 up_write(&curseg->journal_rwsem);
2160 else
184a5cd2
CY		 2161 f2fs_put_page(page, 1);
2162
2163 f2fs_bug_on(sbi, ses->entry_cnt);
2164 release_sit_entry_set(ses);
351df4b2 2165 }
184a5cd2
CY		 2166
2167 f2fs_bug_on(sbi, !list_empty(head));
2168 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2169out:
4b2fecc8
JK		 2170 if (cpc->reason == CP_DISCARD) {
2171 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2172 add_discard_addrs(sbi, cpc);
2173 }
351df4b2 2174 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2175
351df4b2
JK		 2176 set_prefree_as_free_segments(sbi);
2177}
2178
2179static int build_sit_info(struct f2fs_sb_info *sbi)
2180{
2181 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2182 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2183 struct sit_info *sit_i;
2184 unsigned int sit_segs, start;
2185 char *src_bitmap, *dst_bitmap;
2186 unsigned int bitmap_size;
2187
2188 /* allocate memory for SIT information */
2189 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2190 if (!sit_i)
2191 return -ENOMEM;
2192
2193 SM_I(sbi)->sit_info = sit_i;
2194
39307a8e
JK		 2195 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2196 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2197 if (!sit_i->sentries)
2198 return -ENOMEM;
2199
7cd8558b 2200 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2201 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2202 if (!sit_i->dirty_sentries_bitmap)
2203 return -ENOMEM;
2204
7cd8558b 2205 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2206 sit_i->sentries[start].cur_valid_map
2207 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2208 sit_i->sentries[start].ckpt_valid_map
2209 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 2210 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 2211 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 2212 return -ENOMEM;
3e025740
JK		 2213
2214 if (f2fs_discard_en(sbi)) {
2215 sit_i->sentries[start].discard_map
2216 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2217 if (!sit_i->sentries[start].discard_map)
2218 return -ENOMEM;
2219 }
351df4b2
JK		 2220 }
2221
60a3b782
JK		 2222 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2223 if (!sit_i->tmp_map)
2224 return -ENOMEM;
2225
351df4b2 2226 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2227 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2228 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2229 if (!sit_i->sec_entries)
2230 return -ENOMEM;
2231 }
2232
2233 /* get information related with SIT */
2234 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2235
2236 /* setup SIT bitmap from ckeckpoint pack */
2237 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2238 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2239
79b5793b 2240 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2241 if (!dst_bitmap)
2242 return -ENOMEM;
351df4b2
JK		 2243
2244 /* init SIT information */
2245 sit_i->s_ops = &default_salloc_ops;
2246
2247 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2248 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2249 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2250 sit_i->sit_bitmap = dst_bitmap;
2251 sit_i->bitmap_size = bitmap_size;
2252 sit_i->dirty_sentries = 0;
2253 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2254 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2255 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2256 mutex_init(&sit_i->sentry_lock);
2257 return 0;
2258}
2259
2260static int build_free_segmap(struct f2fs_sb_info *sbi)
2261{
351df4b2
JK		 2262 struct free_segmap_info *free_i;
2263 unsigned int bitmap_size, sec_bitmap_size;
2264
2265 /* allocate memory for free segmap information */
2266 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2267 if (!free_i)
2268 return -ENOMEM;
2269
2270 SM_I(sbi)->free_info = free_i;
2271
7cd8558b 2272 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2273 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2274 if (!free_i->free_segmap)
2275 return -ENOMEM;
2276
7cd8558b 2277 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2278 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2279 if (!free_i->free_secmap)
2280 return -ENOMEM;
2281
2282 /* set all segments as dirty temporarily */
2283 memset(free_i->free_segmap, 0xff, bitmap_size);
2284 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2285
2286 /* init free segmap information */
7cd8558b 2287 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2288 free_i->free_segments = 0;
2289 free_i->free_sections = 0;
1a118ccf 2290 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2291 return 0;
2292}
2293
2294static int build_curseg(struct f2fs_sb_info *sbi)
2295{
1042d60f 2296 struct curseg_info *array;
351df4b2
JK		 2297 int i;
2298
b434babf 2299 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2300 if (!array)
2301 return -ENOMEM;
2302
2303 SM_I(sbi)->curseg_array = array;
2304
2305 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2306 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2307 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2308 if (!array[i].sum_blk)
2309 return -ENOMEM;
b7ad7512
CY		 2310 init_rwsem(&array[i].journal_rwsem);
2311 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2312 GFP_KERNEL);
2313 if (!array[i].journal)
2314 return -ENOMEM;
351df4b2
JK		 2315 array[i].segno = NULL_SEGNO;
2316 array[i].next_blkoff = 0;
2317 }
2318 return restore_curseg_summaries(sbi);
2319}
2320
2321static void build_sit_entries(struct f2fs_sb_info *sbi)
2322{
2323 struct sit_info *sit_i = SIT_I(sbi);
2324 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2325 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2326 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2327 unsigned int i, start, end;
2328 unsigned int readed, start_blk = 0;
e9f5b8b8 2329 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2330
74de593a 2331 do {
26879fb1 2332 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2333
2334 start = start_blk * sit_i->sents_per_block;
2335 end = (start_blk + readed) * sit_i->sents_per_block;
2336
7cd8558b 2337 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2338 struct seg_entry *se = &sit_i->sentries[start];
2339 struct f2fs_sit_block *sit_blk;
2340 struct f2fs_sit_entry sit;
2341 struct page *page;
2342
74de593a
CY		 2343 page = get_current_sit_page(sbi, start);
2344 sit_blk = (struct f2fs_sit_block *)page_address(page);
2345 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2346 f2fs_put_page(page, 1);
d600af23 2347
74de593a
CY		 2348 check_block_count(sbi, start, &sit);
2349 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2350
2351 /* build discard map only one time */
3e025740
JK		 2352 if (f2fs_discard_en(sbi)) {
2353 memcpy(se->discard_map, se->cur_valid_map,
2354 SIT_VBLOCK_MAP_SIZE);
2355 sbi->discard_blks += sbi->blocks_per_seg -
2356 se->valid_blocks;
2357 }
a66cdd98 2358
d600af23
CY		 2359 if (sbi->segs_per_sec > 1)
2360 get_sec_entry(sbi, start)->valid_blocks +=
2361 se->valid_blocks;
351df4b2 2362 }
74de593a
CY		 2363 start_blk += readed;
2364 } while (start_blk < sit_blk_cnt);
d600af23
CY		 2365
2366 down_read(&curseg->journal_rwsem);
2367 for (i = 0; i < sits_in_cursum(journal); i++) {
2368 struct f2fs_sit_entry sit;
2369 struct seg_entry *se;
2370 unsigned int old_valid_blocks;
2371
2372 start = le32_to_cpu(segno_in_journal(journal, i));
2373 se = &sit_i->sentries[start];
2374 sit = sit_in_journal(journal, i);
2375
2376 old_valid_blocks = se->valid_blocks;
2377
2378 check_block_count(sbi, start, &sit);
2379 seg_info_from_raw_sit(se, &sit);
2380
2381 if (f2fs_discard_en(sbi)) {
2382 memcpy(se->discard_map, se->cur_valid_map,
2383 SIT_VBLOCK_MAP_SIZE);
2384 sbi->discard_blks += old_valid_blocks -
2385 se->valid_blocks;
2386 }
2387
2388 if (sbi->segs_per_sec > 1)
2389 get_sec_entry(sbi, start)->valid_blocks +=
2390 se->valid_blocks - old_valid_blocks;
2391 }
2392 up_read(&curseg->journal_rwsem);
351df4b2
JK		 2393}
2394
2395static void init_free_segmap(struct f2fs_sb_info *sbi)
2396{
2397 unsigned int start;
2398 int type;
2399
7cd8558b 2400 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2401 struct seg_entry *sentry = get_seg_entry(sbi, start);
2402 if (!sentry->valid_blocks)
2403 __set_free(sbi, start);
2404 }
2405
2406 /* set use the current segments */
2407 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2408 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2409 __set_test_and_inuse(sbi, curseg_t->segno);
2410 }
2411}
2412
2413static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2414{
2415 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2416 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2417 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2418 unsigned short valid_blocks;
2419
8736fbf0 2420 while (1) {
351df4b2 2421 /* find dirty segment based on free segmap */
7cd8558b
JK		 2422 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2423 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2424 break;
2425 offset = segno + 1;
2426 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2427 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2428 continue;
ec325b52
JK		 2429 if (valid_blocks > sbi->blocks_per_seg) {
2430 f2fs_bug_on(sbi, 1);
2431 continue;
2432 }
351df4b2
JK		 2433 mutex_lock(&dirty_i->seglist_lock);
2434 __locate_dirty_segment(sbi, segno, DIRTY);
2435 mutex_unlock(&dirty_i->seglist_lock);
2436 }
2437}
2438
5ec4e49f 2439static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2440{
2441 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2442 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2443
39307a8e 2444 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2445 if (!dirty_i->victim_secmap)
351df4b2
JK		 2446 return -ENOMEM;
2447 return 0;
2448}
2449
2450static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2451{
2452 struct dirty_seglist_info *dirty_i;
2453 unsigned int bitmap_size, i;
2454
2455 /* allocate memory for dirty segments list information */
2456 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2457 if (!dirty_i)
2458 return -ENOMEM;
2459
2460 SM_I(sbi)->dirty_info = dirty_i;
2461 mutex_init(&dirty_i->seglist_lock);
2462
7cd8558b 2463 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2464
2465 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2466 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2467 if (!dirty_i->dirty_segmap[i])
2468 return -ENOMEM;
2469 }
2470
2471 init_dirty_segmap(sbi);
5ec4e49f 2472 return init_victim_secmap(sbi);
351df4b2
JK		 2473}
2474
0a8165d7 2475/*
351df4b2
JK		 2476 * Update min, max modified time for cost-benefit GC algorithm
2477 */
2478static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2479{
2480 struct sit_info *sit_i = SIT_I(sbi);
2481 unsigned int segno;
2482
2483 mutex_lock(&sit_i->sentry_lock);
2484
2485 sit_i->min_mtime = LLONG_MAX;
2486
7cd8558b 2487 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2488 unsigned int i;
2489 unsigned long long mtime = 0;
2490
2491 for (i = 0; i < sbi->segs_per_sec; i++)
2492 mtime += get_seg_entry(sbi, segno + i)->mtime;
2493
2494 mtime = div_u64(mtime, sbi->segs_per_sec);
2495
2496 if (sit_i->min_mtime > mtime)
2497 sit_i->min_mtime = mtime;
2498 }
2499 sit_i->max_mtime = get_mtime(sbi);
2500 mutex_unlock(&sit_i->sentry_lock);
2501}
2502
2503int build_segment_manager(struct f2fs_sb_info *sbi)
2504{
2505 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2506 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2507 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2508 int err;
2509
2510 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2511 if (!sm_info)
2512 return -ENOMEM;
2513
2514 /* init sm info */
2515 sbi->sm_info = sm_info;
351df4b2
JK		 2516 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2517 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2518 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2519 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2520 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2521 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2522 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2523 sm_info->rec_prefree_segments = sm_info->main_segments *
2524 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2525 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2526 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2527
52763a4b
JK		 2528 if (!test_opt(sbi, LFS))
2529 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2530 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2531 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2532
7fd9e544 2533 INIT_LIST_HEAD(&sm_info->discard_list);
275b66b0 2534 INIT_LIST_HEAD(&sm_info->wait_list);
7fd9e544
JK		 2535 sm_info->nr_discards = 0;
2536 sm_info->max_discards = 0;
2537
bba681cb
JK		 2538 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2539
184a5cd2
CY		 2540 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2541
b270ad6f 2542 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2543 err = create_flush_cmd_control(sbi);
2544 if (err)
a688b9d9 2545 return err;
6b4afdd7
JK		 2546 }
2547
351df4b2
JK		 2548 err = build_sit_info(sbi);
2549 if (err)
2550 return err;
2551 err = build_free_segmap(sbi);
2552 if (err)
2553 return err;
2554 err = build_curseg(sbi);
2555 if (err)
2556 return err;
2557
2558 /* reinit free segmap based on SIT */
2559 build_sit_entries(sbi);
2560
2561 init_free_segmap(sbi);
2562 err = build_dirty_segmap(sbi);
2563 if (err)
2564 return err;
2565
2566 init_min_max_mtime(sbi);
2567 return 0;
2568}
2569
2570static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2571 enum dirty_type dirty_type)
2572{
2573 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2574
2575 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2576 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2577 dirty_i->nr_dirty[dirty_type] = 0;
2578 mutex_unlock(&dirty_i->seglist_lock);
2579}
2580
5ec4e49f 2581static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2582{
2583 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2584 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2585}
2586
2587static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2588{
2589 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2590 int i;
2591
2592 if (!dirty_i)
2593 return;
2594
2595 /* discard pre-free/dirty segments list */
2596 for (i = 0; i < NR_DIRTY_TYPE; i++)
2597 discard_dirty_segmap(sbi, i);
2598
5ec4e49f 2599 destroy_victim_secmap(sbi);
351df4b2
JK		 2600 SM_I(sbi)->dirty_info = NULL;
2601 kfree(dirty_i);
2602}
2603
2604static void destroy_curseg(struct f2fs_sb_info *sbi)
2605{
2606 struct curseg_info *array = SM_I(sbi)->curseg_array;
2607 int i;
2608
2609 if (!array)
2610 return;
2611 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2612 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2613 kfree(array[i].sum_blk);
b7ad7512
CY		 2614 kfree(array[i].journal);
2615 }
351df4b2
JK		 2616 kfree(array);
2617}
2618
2619static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2620{
2621 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2622 if (!free_i)
2623 return;
2624 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2625 kvfree(free_i->free_segmap);
2626 kvfree(free_i->free_secmap);
351df4b2
JK		 2627 kfree(free_i);
2628}
2629
2630static void destroy_sit_info(struct f2fs_sb_info *sbi)
2631{
2632 struct sit_info *sit_i = SIT_I(sbi);
2633 unsigned int start;
2634
2635 if (!sit_i)
2636 return;
2637
2638 if (sit_i->sentries) {
7cd8558b 2639 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2640 kfree(sit_i->sentries[start].cur_valid_map);
2641 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2642 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2643 }
2644 }
60a3b782
JK		 2645 kfree(sit_i->tmp_map);
2646
39307a8e
JK		 2647 kvfree(sit_i->sentries);
2648 kvfree(sit_i->sec_entries);
2649 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2650
2651 SM_I(sbi)->sit_info = NULL;
2652 kfree(sit_i->sit_bitmap);
2653 kfree(sit_i);
2654}
2655
2656void destroy_segment_manager(struct f2fs_sb_info *sbi)
2657{
2658 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2659
3b03f724
CY		 2660 if (!sm_info)
2661 return;
2163d198 2662 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2663 destroy_dirty_segmap(sbi);
2664 destroy_curseg(sbi);
2665 destroy_free_segmap(sbi);
2666 destroy_sit_info(sbi);
2667 sbi->sm_info = NULL;
2668 kfree(sm_info);
2669}
7fd9e544
JK		 2670
2671int __init create_segment_manager_caches(void)
2672{
2673 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2674 sizeof(struct discard_entry));
7fd9e544 2675 if (!discard_entry_slab)
184a5cd2
CY		 2676 goto fail;
2677
275b66b0
CY		 2678 bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2679 sizeof(struct bio_entry));
2680 if (!bio_entry_slab)
6ab2a308 2681 goto destroy_discard_entry;
275b66b0 2682
184a5cd2 2683 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2684 sizeof(struct sit_entry_set));
184a5cd2 2685 if (!sit_entry_set_slab)
275b66b0 2686 goto destroy_bio_entry;
88b88a66
JK		 2687
2688 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2689 sizeof(struct inmem_pages));
2690 if (!inmem_entry_slab)
2691 goto destroy_sit_entry_set;
7fd9e544 2692 return 0;
184a5cd2 2693
88b88a66
JK		 2694destroy_sit_entry_set:
2695 kmem_cache_destroy(sit_entry_set_slab);
275b66b0
CY		 2696destroy_bio_entry:
2697 kmem_cache_destroy(bio_entry_slab);
6ab2a308 2698destroy_discard_entry:
184a5cd2
CY		 2699 kmem_cache_destroy(discard_entry_slab);
2700fail:
2701 return -ENOMEM;
7fd9e544
JK		 2702}
2703
2704void destroy_segment_manager_caches(void)
2705{
184a5cd2 2706 kmem_cache_destroy(sit_entry_set_slab);
275b66b0 2707 kmem_cache_destroy(bio_entry_slab);
7fd9e544 2708 kmem_cache_destroy(discard_entry_slab);
88b88a66 2709 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2710}
Ubuntu Jammy Linux kernel mirror