]> 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: support configuring fault injection per superblock
[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;
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 */
7f3037a5 359 if (has_not_enough_free_secs(sbi, 0, 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
adf4983b 672static void __add_discard_entry(struct f2fs_sb_info *sbi,
a66cdd98
JK		 673 struct cp_control *cpc, struct seg_entry *se,
674 unsigned int start, unsigned int end)
b2955550
JK		 675{
676 struct list_head *head = &SM_I(sbi)->discard_list;
adf4983b
JK		 677 struct discard_entry *new, *last;
678
679 if (!list_empty(head)) {
680 last = list_last_entry(head, struct discard_entry, list);
681 if (START_BLOCK(sbi, cpc->trim_start) + start ==
682 last->blkaddr + last->len) {
683 last->len += end - start;
684 goto done;
685 }
686 }
687
688 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
689 INIT_LIST_HEAD(&new->list);
690 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
691 new->len = end - start;
692 list_add_tail(&new->list, head);
693done:
694 SM_I(sbi)->nr_discards += end - start;
adf4983b
JK		 695}
696
697static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
698{
b2955550
JK		 699 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
700 int max_blocks = sbi->blocks_per_seg;
4b2fecc8 701 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
b2955550
JK		 702 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
703 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
a66cdd98 704 unsigned long *discard_map = (unsigned long *)se->discard_map;
60a3b782 705 unsigned long *dmap = SIT_I(sbi)->tmp_map;
b2955550 706 unsigned int start = 0, end = -1;
4b2fecc8 707 bool force = (cpc->reason == CP_DISCARD);
b2955550
JK		 708 int i;
709
3e025740 710 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
b2955550
JK		 711 return;
712
a66cdd98
JK		 713 if (!force) {
714 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
912a83b5
DC		 715 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
716 return;
4b2fecc8
JK		 717 }
718
b2955550
JK		 719 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
720 for (i = 0; i < entries; i++)
a66cdd98 721 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
d7bc2484 722 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
b2955550 723
4b2fecc8 724 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
b2955550
JK		 725 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
726 if (start >= max_blocks)
727 break;
728
729 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
c7b41e16
YH		 730 if (force && start && end != max_blocks
731 && (end - start) < cpc->trim_minlen)
732 continue;
733
a66cdd98 734 __add_discard_entry(sbi, cpc, se, start, end);
b2955550
JK		 735 }
736}
737
4b2fecc8
JK		 738void release_discard_addrs(struct f2fs_sb_info *sbi)
739{
740 struct list_head *head = &(SM_I(sbi)->discard_list);
741 struct discard_entry *entry, *this;
742
743 /* drop caches */
744 list_for_each_entry_safe(entry, this, head, list) {
745 list_del(&entry->list);
746 kmem_cache_free(discard_entry_slab, entry);
747 }
748}
749
0a8165d7 750/*
351df4b2
JK		 751 * Should call clear_prefree_segments after checkpoint is done.
752 */
753static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
754{
755 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
b65ee148 756 unsigned int segno;
351df4b2
JK		 757
758 mutex_lock(&dirty_i->seglist_lock);
7cd8558b 759 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
351df4b2 760 __set_test_and_free(sbi, segno);
351df4b2
JK		 761 mutex_unlock(&dirty_i->seglist_lock);
762}
763
836b5a63 764void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2 765{
b2955550 766 struct list_head *head = &(SM_I(sbi)->discard_list);
2d7b822a 767 struct discard_entry *entry, *this;
351df4b2 768 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
275b66b0 769 struct blk_plug plug;
29e59c14 770 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
29e59c14 771 unsigned int start = 0, end = -1;
36abef4e 772 unsigned int secno, start_segno;
c24a0fd6 773 bool force = (cpc->reason == CP_DISCARD);
351df4b2 774
275b66b0
CY		 775 blk_start_plug(&plug);
776
351df4b2 777 mutex_lock(&dirty_i->seglist_lock);
29e59c14 778
351df4b2 779 while (1) {
29e59c14 780 int i;
7cd8558b
JK		 781 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
782 if (start >= MAIN_SEGS(sbi))
351df4b2 783 break;
7cd8558b
JK		 784 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
785 start + 1);
29e59c14
CL		 786
787 for (i = start; i < end; i++)
788 clear_bit(i, prefree_map);
789
790 dirty_i->nr_dirty[PRE] -= end - start;
791
c24a0fd6 792 if (force || !test_opt(sbi, DISCARD))
29e59c14 793 continue;
351df4b2 794
36abef4e
JK		 795 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
796 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
37208879 797 (end - start) << sbi->log_blocks_per_seg);
36abef4e
JK		 798 continue;
799 }
800next:
801 secno = GET_SECNO(sbi, start);
802 start_segno = secno * sbi->segs_per_sec;
803 if (!IS_CURSEC(sbi, secno) &&
804 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
805 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
806 sbi->segs_per_sec << sbi->log_blocks_per_seg);
807
808 start = start_segno + sbi->segs_per_sec;
809 if (start < end)
810 goto next;
351df4b2
JK		 811 }
812 mutex_unlock(&dirty_i->seglist_lock);
b2955550
JK		 813
814 /* send small discards */
2d7b822a 815 list_for_each_entry_safe(entry, this, head, list) {
c24a0fd6 816 if (force && entry->len < cpc->trim_minlen)
836b5a63 817 goto skip;
37208879 818 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
f56aa1c5 819 cpc->trimmed += entry->len;
836b5a63 820skip:
b2955550
JK		 821 list_del(&entry->list);
822 SM_I(sbi)->nr_discards -= entry->len;
823 kmem_cache_free(discard_entry_slab, entry);
824 }
275b66b0
CY		 825
826 blk_finish_plug(&plug);
351df4b2
JK		 827}
828
184a5cd2 829static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
351df4b2
JK		 830{
831 struct sit_info *sit_i = SIT_I(sbi);
184a5cd2
CY		 832
833 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
351df4b2 834 sit_i->dirty_sentries++;
184a5cd2
CY		 835 return false;
836 }
837
838 return true;
351df4b2
JK		 839}
840
841static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
842 unsigned int segno, int modified)
843{
844 struct seg_entry *se = get_seg_entry(sbi, segno);
845 se->type = type;
846 if (modified)
847 __mark_sit_entry_dirty(sbi, segno);
848}
849
850static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
851{
852 struct seg_entry *se;
853 unsigned int segno, offset;
854 long int new_vblocks;
855
856 segno = GET_SEGNO(sbi, blkaddr);
857
858 se = get_seg_entry(sbi, segno);
859 new_vblocks = se->valid_blocks + del;
491c0854 860 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
351df4b2 861
9850cf4a 862 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
351df4b2
JK		 863 (new_vblocks > sbi->blocks_per_seg)));
864
865 se->valid_blocks = new_vblocks;
866 se->mtime = get_mtime(sbi);
867 SIT_I(sbi)->max_mtime = se->mtime;
868
869 /* Update valid block bitmap */
870 if (del > 0) {
52aca074 871 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
05796763 872 f2fs_bug_on(sbi, 1);
3e025740
JK		 873 if (f2fs_discard_en(sbi) &&
874 !f2fs_test_and_set_bit(offset, se->discard_map))
a66cdd98 875 sbi->discard_blks--;
351df4b2 876 } else {
52aca074 877 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
05796763 878 f2fs_bug_on(sbi, 1);
3e025740
JK		 879 if (f2fs_discard_en(sbi) &&
880 f2fs_test_and_clear_bit(offset, se->discard_map))
a66cdd98 881 sbi->discard_blks++;
351df4b2
JK		 882 }
883 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
884 se->ckpt_valid_blocks += del;
885
886 __mark_sit_entry_dirty(sbi, segno);
887
888 /* update total number of valid blocks to be written in ckpt area */
889 SIT_I(sbi)->written_valid_blocks += del;
890
891 if (sbi->segs_per_sec > 1)
892 get_sec_entry(sbi, segno)->valid_blocks += del;
893}
894
5e443818 895void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
351df4b2 896{
5e443818
JK		 897 update_sit_entry(sbi, new, 1);
898 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
899 update_sit_entry(sbi, old, -1);
900
901 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
902 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
351df4b2
JK		 903}
904
905void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
906{
907 unsigned int segno = GET_SEGNO(sbi, addr);
908 struct sit_info *sit_i = SIT_I(sbi);
909
9850cf4a 910 f2fs_bug_on(sbi, addr == NULL_ADDR);
351df4b2
JK		 911 if (addr == NEW_ADDR)
912 return;
913
914 /* add it into sit main buffer */
915 mutex_lock(&sit_i->sentry_lock);
916
917 update_sit_entry(sbi, addr, -1);
918
919 /* add it into dirty seglist */
920 locate_dirty_segment(sbi, segno);
921
922 mutex_unlock(&sit_i->sentry_lock);
923}
924
6e2c64ad
JK		 925bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
926{
927 struct sit_info *sit_i = SIT_I(sbi);
928 unsigned int segno, offset;
929 struct seg_entry *se;
930 bool is_cp = false;
931
932 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
933 return true;
934
935 mutex_lock(&sit_i->sentry_lock);
936
937 segno = GET_SEGNO(sbi, blkaddr);
938 se = get_seg_entry(sbi, segno);
939 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
940
941 if (f2fs_test_bit(offset, se->ckpt_valid_map))
942 is_cp = true;
943
944 mutex_unlock(&sit_i->sentry_lock);
945
946 return is_cp;
947}
948
0a8165d7 949/*
351df4b2
JK		 950 * This function should be resided under the curseg_mutex lock
951 */
952static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
e79efe3b 953 struct f2fs_summary *sum)
351df4b2
JK		 954{
955 struct curseg_info *curseg = CURSEG_I(sbi, type);
956 void *addr = curseg->sum_blk;
e79efe3b 957 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
351df4b2 958 memcpy(addr, sum, sizeof(struct f2fs_summary));
351df4b2
JK		 959}
960
0a8165d7 961/*
351df4b2
JK		 962 * Calculate the number of current summary pages for writing
963 */
3fa06d7b 964int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
351df4b2 965{
351df4b2 966 int valid_sum_count = 0;
9a47938b 967 int i, sum_in_page;
351df4b2
JK		 968
969 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
970 if (sbi->ckpt->alloc_type[i] == SSR)
971 valid_sum_count += sbi->blocks_per_seg;
3fa06d7b
CY		 972 else {
973 if (for_ra)
974 valid_sum_count += le16_to_cpu(
975 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
976 else
977 valid_sum_count += curseg_blkoff(sbi, i);
978 }
351df4b2
JK		 979 }
980
09cbfeaf 981 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
9a47938b
FL		 982 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
983 if (valid_sum_count <= sum_in_page)
351df4b2 984 return 1;
9a47938b 985 else if ((valid_sum_count - sum_in_page) <=
09cbfeaf 986 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
351df4b2
JK		 987 return 2;
988 return 3;
989}
990
0a8165d7 991/*
351df4b2
JK		 992 * Caller should put this summary page
993 */
994struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
995{
996 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
997}
998
381722d2 999void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
351df4b2
JK		 1000{
1001 struct page *page = grab_meta_page(sbi, blk_addr);
381722d2
CY		 1002 void *dst = page_address(page);
1003
1004 if (src)
09cbfeaf 1005 memcpy(dst, src, PAGE_SIZE);
381722d2 1006 else
09cbfeaf 1007 memset(dst, 0, PAGE_SIZE);
351df4b2
JK		 1008 set_page_dirty(page);
1009 f2fs_put_page(page, 1);
1010}
1011
381722d2
CY		 1012static void write_sum_page(struct f2fs_sb_info *sbi,
1013 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1014{
1015 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1016}
1017
b7ad7512
CY		 1018static void write_current_sum_page(struct f2fs_sb_info *sbi,
1019 int type, block_t blk_addr)
1020{
1021 struct curseg_info *curseg = CURSEG_I(sbi, type);
1022 struct page *page = grab_meta_page(sbi, blk_addr);
1023 struct f2fs_summary_block *src = curseg->sum_blk;
1024 struct f2fs_summary_block *dst;
1025
1026 dst = (struct f2fs_summary_block *)page_address(page);
1027
1028 mutex_lock(&curseg->curseg_mutex);
1029
1030 down_read(&curseg->journal_rwsem);
1031 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1032 up_read(&curseg->journal_rwsem);
1033
1034 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1035 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1036
1037 mutex_unlock(&curseg->curseg_mutex);
1038
1039 set_page_dirty(page);
1040 f2fs_put_page(page, 1);
1041}
1042
60374688
JK		 1043static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1044{
1045 struct curseg_info *curseg = CURSEG_I(sbi, type);
81fb5e87 1046 unsigned int segno = curseg->segno + 1;
60374688
JK		 1047 struct free_segmap_info *free_i = FREE_I(sbi);
1048
7cd8558b 1049 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
81fb5e87 1050 return !test_bit(segno, free_i->free_segmap);
60374688
JK		 1051 return 0;
1052}
1053
0a8165d7 1054/*
351df4b2
JK		 1055 * Find a new segment from the free segments bitmap to right order
1056 * This function should be returned with success, otherwise BUG
1057 */
1058static void get_new_segment(struct f2fs_sb_info *sbi,
1059 unsigned int *newseg, bool new_sec, int dir)
1060{
1061 struct free_segmap_info *free_i = FREE_I(sbi);
351df4b2 1062 unsigned int segno, secno, zoneno;
7cd8558b 1063 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
351df4b2
JK		 1064 unsigned int hint = *newseg / sbi->segs_per_sec;
1065 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1066 unsigned int left_start = hint;
1067 bool init = true;
1068 int go_left = 0;
1069 int i;
1070
1a118ccf 1071 spin_lock(&free_i->segmap_lock);
351df4b2
JK		 1072
1073 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1074 segno = find_next_zero_bit(free_i->free_segmap,
0ab14356
CY		 1075 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1076 if (segno < (hint + 1) * sbi->segs_per_sec)
351df4b2
JK		 1077 goto got_it;
1078 }
1079find_other_zone:
7cd8558b
JK		 1080 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1081 if (secno >= MAIN_SECS(sbi)) {
351df4b2
JK		 1082 if (dir == ALLOC_RIGHT) {
1083 secno = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1084 MAIN_SECS(sbi), 0);
1085 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
351df4b2
JK		 1086 } else {
1087 go_left = 1;
1088 left_start = hint - 1;
1089 }
1090 }
1091 if (go_left == 0)
1092 goto skip_left;
1093
1094 while (test_bit(left_start, free_i->free_secmap)) {
1095 if (left_start > 0) {
1096 left_start--;
1097 continue;
1098 }
1099 left_start = find_next_zero_bit(free_i->free_secmap,
7cd8558b
JK		 1100 MAIN_SECS(sbi), 0);
1101 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
351df4b2
JK		 1102 break;
1103 }
1104 secno = left_start;
1105skip_left:
1106 hint = secno;
1107 segno = secno * sbi->segs_per_sec;
1108 zoneno = secno / sbi->secs_per_zone;
1109
1110 /* give up on finding another zone */
1111 if (!init)
1112 goto got_it;
1113 if (sbi->secs_per_zone == 1)
1114 goto got_it;
1115 if (zoneno == old_zoneno)
1116 goto got_it;
1117 if (dir == ALLOC_LEFT) {
1118 if (!go_left && zoneno + 1 >= total_zones)
1119 goto got_it;
1120 if (go_left && zoneno == 0)
1121 goto got_it;
1122 }
1123 for (i = 0; i < NR_CURSEG_TYPE; i++)
1124 if (CURSEG_I(sbi, i)->zone == zoneno)
1125 break;
1126
1127 if (i < NR_CURSEG_TYPE) {
1128 /* zone is in user, try another */
1129 if (go_left)
1130 hint = zoneno * sbi->secs_per_zone - 1;
1131 else if (zoneno + 1 >= total_zones)
1132 hint = 0;
1133 else
1134 hint = (zoneno + 1) * sbi->secs_per_zone;
1135 init = false;
1136 goto find_other_zone;
1137 }
1138got_it:
1139 /* set it as dirty segment in free segmap */
9850cf4a 1140 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
351df4b2
JK		 1141 __set_inuse(sbi, segno);
1142 *newseg = segno;
1a118ccf 1143 spin_unlock(&free_i->segmap_lock);
351df4b2
JK		 1144}
1145
1146static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1147{
1148 struct curseg_info *curseg = CURSEG_I(sbi, type);
1149 struct summary_footer *sum_footer;
1150
1151 curseg->segno = curseg->next_segno;
1152 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1153 curseg->next_blkoff = 0;
1154 curseg->next_segno = NULL_SEGNO;
1155
1156 sum_footer = &(curseg->sum_blk->footer);
1157 memset(sum_footer, 0, sizeof(struct summary_footer));
1158 if (IS_DATASEG(type))
1159 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1160 if (IS_NODESEG(type))
1161 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1162 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1163}
1164
0a8165d7 1165/*
351df4b2
JK		 1166 * Allocate a current working segment.
1167 * This function always allocates a free segment in LFS manner.
1168 */
1169static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1170{
1171 struct curseg_info *curseg = CURSEG_I(sbi, type);
1172 unsigned int segno = curseg->segno;
1173 int dir = ALLOC_LEFT;
1174
1175 write_sum_page(sbi, curseg->sum_blk,
81fb5e87 1176 GET_SUM_BLOCK(sbi, segno));
351df4b2
JK		 1177 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1178 dir = ALLOC_RIGHT;
1179
1180 if (test_opt(sbi, NOHEAP))
1181 dir = ALLOC_RIGHT;
1182
1183 get_new_segment(sbi, &segno, new_sec, dir);
1184 curseg->next_segno = segno;
1185 reset_curseg(sbi, type, 1);
1186 curseg->alloc_type = LFS;
1187}
1188
1189static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1190 struct curseg_info *seg, block_t start)
1191{
1192 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
e81c93cf 1193 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
60a3b782 1194 unsigned long *target_map = SIT_I(sbi)->tmp_map;
e81c93cf
CL		 1195 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1196 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1197 int i, pos;
1198
1199 for (i = 0; i < entries; i++)
1200 target_map[i] = ckpt_map[i] | cur_map[i];
1201
1202 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1203
1204 seg->next_blkoff = pos;
351df4b2
JK		 1205}
1206
0a8165d7 1207/*
351df4b2
JK		 1208 * If a segment is written by LFS manner, next block offset is just obtained
1209 * by increasing the current block offset. However, if a segment is written by
1210 * SSR manner, next block offset obtained by calling __next_free_blkoff
1211 */
1212static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1213 struct curseg_info *seg)
1214{
1215 if (seg->alloc_type == SSR)
1216 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1217 else
1218 seg->next_blkoff++;
1219}
1220
0a8165d7 1221/*
e1c42045 1222 * This function always allocates a used segment(from dirty seglist) by SSR
351df4b2
JK		 1223 * manner, so it should recover the existing segment information of valid blocks
1224 */
1225static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1226{
1227 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1228 struct curseg_info *curseg = CURSEG_I(sbi, type);
1229 unsigned int new_segno = curseg->next_segno;
1230 struct f2fs_summary_block *sum_node;
1231 struct page *sum_page;
1232
1233 write_sum_page(sbi, curseg->sum_blk,
1234 GET_SUM_BLOCK(sbi, curseg->segno));
1235 __set_test_and_inuse(sbi, new_segno);
1236
1237 mutex_lock(&dirty_i->seglist_lock);
1238 __remove_dirty_segment(sbi, new_segno, PRE);
1239 __remove_dirty_segment(sbi, new_segno, DIRTY);
1240 mutex_unlock(&dirty_i->seglist_lock);
1241
1242 reset_curseg(sbi, type, 1);
1243 curseg->alloc_type = SSR;
1244 __next_free_blkoff(sbi, curseg, 0);
1245
1246 if (reuse) {
1247 sum_page = get_sum_page(sbi, new_segno);
1248 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1249 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1250 f2fs_put_page(sum_page, 1);
1251 }
1252}
1253
43727527
JK		 1254static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1255{
1256 struct curseg_info *curseg = CURSEG_I(sbi, type);
1257 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1258
7f3037a5 1259 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
43727527
JK		 1260 return v_ops->get_victim(sbi,
1261 &(curseg)->next_segno, BG_GC, type, SSR);
1262
1263 /* For data segments, let's do SSR more intensively */
1264 for (; type >= CURSEG_HOT_DATA; type--)
1265 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1266 BG_GC, type, SSR))
1267 return 1;
1268 return 0;
1269}
1270
351df4b2
JK		 1271/*
1272 * flush out current segment and replace it with new segment
1273 * This function should be returned with success, otherwise BUG
1274 */
1275static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1276 int type, bool force)
1277{
1278 struct curseg_info *curseg = CURSEG_I(sbi, type);
351df4b2 1279
7b405275 1280 if (force)
351df4b2 1281 new_curseg(sbi, type, true);
7b405275 1282 else if (type == CURSEG_WARM_NODE)
351df4b2 1283 new_curseg(sbi, type, false);
60374688
JK		 1284 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1285 new_curseg(sbi, type, false);
351df4b2
JK		 1286 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1287 change_curseg(sbi, type, true);
1288 else
1289 new_curseg(sbi, type, false);
dcdfff65
JK		 1290
1291 stat_inc_seg_type(sbi, curseg);
351df4b2
JK		 1292}
1293
38aa0889
JK		 1294static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1295{
1296 struct curseg_info *curseg = CURSEG_I(sbi, type);
1297 unsigned int old_segno;
1298
1299 old_segno = curseg->segno;
1300 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1301 locate_dirty_segment(sbi, old_segno);
1302}
1303
351df4b2
JK		 1304void allocate_new_segments(struct f2fs_sb_info *sbi)
1305{
351df4b2
JK		 1306 int i;
1307
36abef4e
JK		 1308 if (test_opt(sbi, LFS))
1309 return;
1310
38aa0889
JK		 1311 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1312 __allocate_new_segments(sbi, i);
351df4b2
JK		 1313}
1314
1315static const struct segment_allocation default_salloc_ops = {
1316 .allocate_segment = allocate_segment_by_default,
1317};
1318
4b2fecc8
JK		 1319int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1320{
f7ef9b83
JK		 1321 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1322 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
4b2fecc8
JK		 1323 unsigned int start_segno, end_segno;
1324 struct cp_control cpc;
c34f42e2 1325 int err = 0;
4b2fecc8 1326
836b5a63 1327 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
4b2fecc8
JK		 1328 return -EINVAL;
1329
9bd27ae4 1330 cpc.trimmed = 0;
7cd8558b 1331 if (end <= MAIN_BLKADDR(sbi))
4b2fecc8
JK		 1332 goto out;
1333
ed214a11
YH		 1334 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1335 f2fs_msg(sbi->sb, KERN_WARNING,
1336 "Found FS corruption, run fsck to fix.");
1337 goto out;
1338 }
1339
4b2fecc8 1340 /* start/end segment number in main_area */
7cd8558b
JK		 1341 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1342 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1343 GET_SEGNO(sbi, end);
4b2fecc8 1344 cpc.reason = CP_DISCARD;
836b5a63 1345 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
4b2fecc8
JK		 1346
1347 /* do checkpoint to issue discard commands safely */
bba681cb
JK		 1348 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1349 cpc.trim_start = start_segno;
a66cdd98
JK		 1350
1351 if (sbi->discard_blks == 0)
1352 break;
1353 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1354 cpc.trim_end = end_segno;
1355 else
1356 cpc.trim_end = min_t(unsigned int,
1357 rounddown(start_segno +
bba681cb
JK		 1358 BATCHED_TRIM_SEGMENTS(sbi),
1359 sbi->segs_per_sec) - 1, end_segno);
1360
1361 mutex_lock(&sbi->gc_mutex);
c34f42e2 1362 err = write_checkpoint(sbi, &cpc);
bba681cb 1363 mutex_unlock(&sbi->gc_mutex);
e9328353
CY		 1364 if (err)
1365 break;
74fa5f3d
CY		 1366
1367 schedule();
bba681cb 1368 }
4b2fecc8 1369out:
f7ef9b83 1370 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
c34f42e2 1371 return err;
4b2fecc8
JK		 1372}
1373
351df4b2
JK		 1374static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1375{
1376 struct curseg_info *curseg = CURSEG_I(sbi, type);
1377 if (curseg->next_blkoff < sbi->blocks_per_seg)
1378 return true;
1379 return false;
1380}
1381
1382static int __get_segment_type_2(struct page *page, enum page_type p_type)
1383{
1384 if (p_type == DATA)
1385 return CURSEG_HOT_DATA;
1386 else
1387 return CURSEG_HOT_NODE;
1388}
1389
1390static int __get_segment_type_4(struct page *page, enum page_type p_type)
1391{
1392 if (p_type == DATA) {
1393 struct inode *inode = page->mapping->host;
1394
1395 if (S_ISDIR(inode->i_mode))
1396 return CURSEG_HOT_DATA;
1397 else
1398 return CURSEG_COLD_DATA;
1399 } else {
a344b9fd
JK		 1400 if (IS_DNODE(page) && is_cold_node(page))
1401 return CURSEG_WARM_NODE;
351df4b2
JK		 1402 else
1403 return CURSEG_COLD_NODE;
1404 }
1405}
1406
1407static int __get_segment_type_6(struct page *page, enum page_type p_type)
1408{
1409 if (p_type == DATA) {
1410 struct inode *inode = page->mapping->host;
1411
1412 if (S_ISDIR(inode->i_mode))
1413 return CURSEG_HOT_DATA;
354a3399 1414 else if (is_cold_data(page) || file_is_cold(inode))
351df4b2
JK		 1415 return CURSEG_COLD_DATA;
1416 else
1417 return CURSEG_WARM_DATA;
1418 } else {
1419 if (IS_DNODE(page))
1420 return is_cold_node(page) ? CURSEG_WARM_NODE :
1421 CURSEG_HOT_NODE;
1422 else
1423 return CURSEG_COLD_NODE;
1424 }
1425}
1426
1427static int __get_segment_type(struct page *page, enum page_type p_type)
1428{
4081363f 1429 switch (F2FS_P_SB(page)->active_logs) {
351df4b2
JK		 1430 case 2:
1431 return __get_segment_type_2(page, p_type);
1432 case 4:
1433 return __get_segment_type_4(page, p_type);
351df4b2 1434 }
12a67146 1435 /* NR_CURSEG_TYPE(6) logs by default */
9850cf4a
JK		 1436 f2fs_bug_on(F2FS_P_SB(page),
1437 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
12a67146 1438 return __get_segment_type_6(page, p_type);
351df4b2
JK		 1439}
1440
bfad7c2d
JK		 1441void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1442 block_t old_blkaddr, block_t *new_blkaddr,
1443 struct f2fs_summary *sum, int type)
351df4b2
JK		 1444{
1445 struct sit_info *sit_i = SIT_I(sbi);
1446 struct curseg_info *curseg;
38aa0889
JK		 1447 bool direct_io = (type == CURSEG_DIRECT_IO);
1448
1449 type = direct_io ? CURSEG_WARM_DATA : type;
351df4b2 1450
351df4b2
JK		 1451 curseg = CURSEG_I(sbi, type);
1452
1453 mutex_lock(&curseg->curseg_mutex);
21cb1d99 1454 mutex_lock(&sit_i->sentry_lock);
351df4b2 1455
38aa0889 1456 /* direct_io'ed data is aligned to the segment for better performance */
47e70ca4 1457 if (direct_io && curseg->next_blkoff &&
7f3037a5 1458 !has_not_enough_free_secs(sbi, 0, 0))
38aa0889
JK		 1459 __allocate_new_segments(sbi, type);
1460
351df4b2 1461 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
351df4b2
JK		 1462
1463 /*
1464 * __add_sum_entry should be resided under the curseg_mutex
1465 * because, this function updates a summary entry in the
1466 * current summary block.
1467 */
e79efe3b 1468 __add_sum_entry(sbi, type, sum);
351df4b2 1469
351df4b2 1470 __refresh_next_blkoff(sbi, curseg);
dcdfff65
JK		 1471
1472 stat_inc_block_count(sbi, curseg);
351df4b2 1473
5e443818
JK		 1474 if (!__has_curseg_space(sbi, type))
1475 sit_i->s_ops->allocate_segment(sbi, type, false);
351df4b2
JK		 1476 /*
1477 * SIT information should be updated before segment allocation,
1478 * since SSR needs latest valid block information.
1479 */
1480 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
5e443818 1481
351df4b2
JK		 1482 mutex_unlock(&sit_i->sentry_lock);
1483
bfad7c2d 1484 if (page && IS_NODESEG(type))
351df4b2
JK		 1485 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1486
bfad7c2d
JK		 1487 mutex_unlock(&curseg->curseg_mutex);
1488}
1489
05ca3632 1490static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
bfad7c2d 1491{
05ca3632 1492 int type = __get_segment_type(fio->page, fio->type);
bfad7c2d 1493
7dfeaa32
JK		 1494 if (fio->type == NODE || fio->type == DATA)
1495 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1496
7a9d7548
CY		 1497 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1498 &fio->new_blkaddr, sum, type);
bfad7c2d 1499
351df4b2 1500 /* writeout dirty page into bdev */
05ca3632 1501 f2fs_submit_page_mbio(fio);
7dfeaa32
JK		 1502
1503 if (fio->type == NODE || fio->type == DATA)
1504 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
351df4b2
JK		 1505}
1506
577e3495 1507void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
351df4b2 1508{
458e6197 1509 struct f2fs_io_info fio = {
05ca3632 1510 .sbi = sbi,
458e6197 1511 .type = META,
04d328de
MC		 1512 .op = REQ_OP_WRITE,
1513 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
7a9d7548
CY		 1514 .old_blkaddr = page->index,
1515 .new_blkaddr = page->index,
05ca3632 1516 .page = page,
4375a336 1517 .encrypted_page = NULL,
458e6197
JK		 1518 };
1519
2b947003 1520 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
04d328de 1521 fio.op_flags &= ~REQ_META;
2b947003 1522
351df4b2 1523 set_page_writeback(page);
05ca3632 1524 f2fs_submit_page_mbio(&fio);
351df4b2
JK		 1525}
1526
05ca3632 1527void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
351df4b2
JK		 1528{
1529 struct f2fs_summary sum;
05ca3632 1530
351df4b2 1531 set_summary(&sum, nid, 0, 0);
05ca3632 1532 do_write_page(&sum, fio);
351df4b2
JK		 1533}
1534
05ca3632 1535void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
351df4b2 1536{
05ca3632 1537 struct f2fs_sb_info *sbi = fio->sbi;
351df4b2
JK		 1538 struct f2fs_summary sum;
1539 struct node_info ni;
1540
9850cf4a 1541 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
351df4b2
JK		 1542 get_node_info(sbi, dn->nid, &ni);
1543 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
05ca3632 1544 do_write_page(&sum, fio);
f28b3434 1545 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
351df4b2
JK		 1546}
1547
05ca3632 1548void rewrite_data_page(struct f2fs_io_info *fio)
351df4b2 1549{
7a9d7548 1550 fio->new_blkaddr = fio->old_blkaddr;
05ca3632
JK		 1551 stat_inc_inplace_blocks(fio->sbi);
1552 f2fs_submit_page_mbio(fio);
351df4b2
JK		 1553}
1554
4356e48e 1555void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
19f106bc 1556 block_t old_blkaddr, block_t new_blkaddr,
28bc106b 1557 bool recover_curseg, bool recover_newaddr)
351df4b2
JK		 1558{
1559 struct sit_info *sit_i = SIT_I(sbi);
1560 struct curseg_info *curseg;
1561 unsigned int segno, old_cursegno;
1562 struct seg_entry *se;
1563 int type;
19f106bc 1564 unsigned short old_blkoff;
351df4b2
JK		 1565
1566 segno = GET_SEGNO(sbi, new_blkaddr);
1567 se = get_seg_entry(sbi, segno);
1568 type = se->type;
1569
19f106bc
CY		 1570 if (!recover_curseg) {
1571 /* for recovery flow */
1572 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1573 if (old_blkaddr == NULL_ADDR)
1574 type = CURSEG_COLD_DATA;
1575 else
1576 type = CURSEG_WARM_DATA;
1577 }
1578 } else {
1579 if (!IS_CURSEG(sbi, segno))
351df4b2
JK		 1580 type = CURSEG_WARM_DATA;
1581 }
19f106bc 1582
351df4b2
JK		 1583 curseg = CURSEG_I(sbi, type);
1584
1585 mutex_lock(&curseg->curseg_mutex);
1586 mutex_lock(&sit_i->sentry_lock);
1587
1588 old_cursegno = curseg->segno;
19f106bc 1589 old_blkoff = curseg->next_blkoff;
351df4b2
JK		 1590
1591 /* change the current segment */
1592 if (segno != curseg->segno) {
1593 curseg->next_segno = segno;
1594 change_curseg(sbi, type, true);
1595 }
1596
491c0854 1597 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
e79efe3b 1598 __add_sum_entry(sbi, type, sum);
351df4b2 1599
28bc106b 1600 if (!recover_curseg || recover_newaddr)
6e2c64ad
JK		 1601 update_sit_entry(sbi, new_blkaddr, 1);
1602 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1603 update_sit_entry(sbi, old_blkaddr, -1);
1604
1605 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1606 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1607
351df4b2 1608 locate_dirty_segment(sbi, old_cursegno);
351df4b2 1609
19f106bc
CY		 1610 if (recover_curseg) {
1611 if (old_cursegno != curseg->segno) {
1612 curseg->next_segno = old_cursegno;
1613 change_curseg(sbi, type, true);
1614 }
1615 curseg->next_blkoff = old_blkoff;
1616 }
1617
351df4b2
JK		 1618 mutex_unlock(&sit_i->sentry_lock);
1619 mutex_unlock(&curseg->curseg_mutex);
1620}
1621
528e3459
CY		 1622void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1623 block_t old_addr, block_t new_addr,
28bc106b
CY		 1624 unsigned char version, bool recover_curseg,
1625 bool recover_newaddr)
528e3459
CY		 1626{
1627 struct f2fs_summary sum;
1628
1629 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1630
28bc106b
CY		 1631 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1632 recover_curseg, recover_newaddr);
528e3459 1633
f28b3434 1634 f2fs_update_data_blkaddr(dn, new_addr);
528e3459
CY		 1635}
1636
93dfe2ac 1637void f2fs_wait_on_page_writeback(struct page *page,
fec1d657 1638 enum page_type type, bool ordered)
93dfe2ac 1639{
93dfe2ac 1640 if (PageWriteback(page)) {
4081363f
JK		 1641 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1642
0c3a5797 1643 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
fec1d657
JK		 1644 if (ordered)
1645 wait_on_page_writeback(page);
1646 else
1647 wait_for_stable_page(page);
93dfe2ac
JK		 1648 }
1649}
1650
08b39fbd
CY		 1651void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1652 block_t blkaddr)
1653{
1654 struct page *cpage;
1655
5d4c0af4 1656 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
08b39fbd
CY		 1657 return;
1658
08b39fbd
CY		 1659 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1660 if (cpage) {
fec1d657 1661 f2fs_wait_on_page_writeback(cpage, DATA, true);
08b39fbd
CY		 1662 f2fs_put_page(cpage, 1);
1663 }
1664}
1665
351df4b2
JK		 1666static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1667{
1668 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1669 struct curseg_info *seg_i;
1670 unsigned char *kaddr;
1671 struct page *page;
1672 block_t start;
1673 int i, j, offset;
1674
1675 start = start_sum_block(sbi);
1676
1677 page = get_meta_page(sbi, start++);
1678 kaddr = (unsigned char *)page_address(page);
1679
1680 /* Step 1: restore nat cache */
1681 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1682 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
351df4b2
JK		 1683
1684 /* Step 2: restore sit cache */
1685 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1686 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
351df4b2
JK		 1687 offset = 2 * SUM_JOURNAL_SIZE;
1688
1689 /* Step 3: restore summary entries */
1690 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1691 unsigned short blk_off;
1692 unsigned int segno;
1693
1694 seg_i = CURSEG_I(sbi, i);
1695 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1696 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1697 seg_i->next_segno = segno;
1698 reset_curseg(sbi, i, 0);
1699 seg_i->alloc_type = ckpt->alloc_type[i];
1700 seg_i->next_blkoff = blk_off;
1701
1702 if (seg_i->alloc_type == SSR)
1703 blk_off = sbi->blocks_per_seg;
1704
1705 for (j = 0; j < blk_off; j++) {
1706 struct f2fs_summary *s;
1707 s = (struct f2fs_summary *)(kaddr + offset);
1708 seg_i->sum_blk->entries[j] = *s;
1709 offset += SUMMARY_SIZE;
09cbfeaf 1710 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1711 SUM_FOOTER_SIZE)
1712 continue;
1713
1714 f2fs_put_page(page, 1);
1715 page = NULL;
1716
1717 page = get_meta_page(sbi, start++);
1718 kaddr = (unsigned char *)page_address(page);
1719 offset = 0;
1720 }
1721 }
1722 f2fs_put_page(page, 1);
1723 return 0;
1724}
1725
1726static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1727{
1728 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1729 struct f2fs_summary_block *sum;
1730 struct curseg_info *curseg;
1731 struct page *new;
1732 unsigned short blk_off;
1733 unsigned int segno = 0;
1734 block_t blk_addr = 0;
1735
1736 /* get segment number and block addr */
1737 if (IS_DATASEG(type)) {
1738 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1739 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1740 CURSEG_HOT_DATA]);
119ee914 1741 if (__exist_node_summaries(sbi))
351df4b2
JK		 1742 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1743 else
1744 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1745 } else {
1746 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1747 CURSEG_HOT_NODE]);
1748 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1749 CURSEG_HOT_NODE]);
119ee914 1750 if (__exist_node_summaries(sbi))
351df4b2
JK		 1751 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1752 type - CURSEG_HOT_NODE);
1753 else
1754 blk_addr = GET_SUM_BLOCK(sbi, segno);
1755 }
1756
1757 new = get_meta_page(sbi, blk_addr);
1758 sum = (struct f2fs_summary_block *)page_address(new);
1759
1760 if (IS_NODESEG(type)) {
119ee914 1761 if (__exist_node_summaries(sbi)) {
351df4b2
JK		 1762 struct f2fs_summary *ns = &sum->entries[0];
1763 int i;
1764 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1765 ns->version = 0;
1766 ns->ofs_in_node = 0;
1767 }
1768 } else {
d653788a
GZ		 1769 int err;
1770
1771 err = restore_node_summary(sbi, segno, sum);
1772 if (err) {
351df4b2 1773 f2fs_put_page(new, 1);
d653788a 1774 return err;
351df4b2
JK		 1775 }
1776 }
1777 }
1778
1779 /* set uncompleted segment to curseg */
1780 curseg = CURSEG_I(sbi, type);
1781 mutex_lock(&curseg->curseg_mutex);
b7ad7512
CY		 1782
1783 /* update journal info */
1784 down_write(&curseg->journal_rwsem);
1785 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1786 up_write(&curseg->journal_rwsem);
1787
1788 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1789 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
351df4b2
JK		 1790 curseg->next_segno = segno;
1791 reset_curseg(sbi, type, 0);
1792 curseg->alloc_type = ckpt->alloc_type[type];
1793 curseg->next_blkoff = blk_off;
1794 mutex_unlock(&curseg->curseg_mutex);
1795 f2fs_put_page(new, 1);
1796 return 0;
1797}
1798
1799static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1800{
1801 int type = CURSEG_HOT_DATA;
e4fc5fbf 1802 int err;
351df4b2 1803
aaec2b1d 1804 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3fa06d7b
CY		 1805 int npages = npages_for_summary_flush(sbi, true);
1806
1807 if (npages >= 2)
1808 ra_meta_pages(sbi, start_sum_block(sbi), npages,
26879fb1 1809 META_CP, true);
3fa06d7b 1810
351df4b2
JK		 1811 /* restore for compacted data summary */
1812 if (read_compacted_summaries(sbi))
1813 return -EINVAL;
1814 type = CURSEG_HOT_NODE;
1815 }
1816
119ee914 1817 if (__exist_node_summaries(sbi))
3fa06d7b 1818 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
26879fb1 1819 NR_CURSEG_TYPE - type, META_CP, true);
3fa06d7b 1820
e4fc5fbf
CY		 1821 for (; type <= CURSEG_COLD_NODE; type++) {
1822 err = read_normal_summaries(sbi, type);
1823 if (err)
1824 return err;
1825 }
1826
351df4b2
JK		 1827 return 0;
1828}
1829
1830static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1831{
1832 struct page *page;
1833 unsigned char *kaddr;
1834 struct f2fs_summary *summary;
1835 struct curseg_info *seg_i;
1836 int written_size = 0;
1837 int i, j;
1838
1839 page = grab_meta_page(sbi, blkaddr++);
1840 kaddr = (unsigned char *)page_address(page);
1841
1842 /* Step 1: write nat cache */
1843 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
b7ad7512 1844 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1845 written_size += SUM_JOURNAL_SIZE;
1846
1847 /* Step 2: write sit cache */
1848 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 1849 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
351df4b2
JK		 1850 written_size += SUM_JOURNAL_SIZE;
1851
351df4b2
JK		 1852 /* Step 3: write summary entries */
1853 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1854 unsigned short blkoff;
1855 seg_i = CURSEG_I(sbi, i);
1856 if (sbi->ckpt->alloc_type[i] == SSR)
1857 blkoff = sbi->blocks_per_seg;
1858 else
1859 blkoff = curseg_blkoff(sbi, i);
1860
1861 for (j = 0; j < blkoff; j++) {
1862 if (!page) {
1863 page = grab_meta_page(sbi, blkaddr++);
1864 kaddr = (unsigned char *)page_address(page);
1865 written_size = 0;
1866 }
1867 summary = (struct f2fs_summary *)(kaddr + written_size);
1868 *summary = seg_i->sum_blk->entries[j];
1869 written_size += SUMMARY_SIZE;
351df4b2 1870
09cbfeaf 1871 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
351df4b2
JK		 1872 SUM_FOOTER_SIZE)
1873 continue;
1874
e8d61a74 1875 set_page_dirty(page);
351df4b2
JK		 1876 f2fs_put_page(page, 1);
1877 page = NULL;
1878 }
1879 }
e8d61a74
CY		 1880 if (page) {
1881 set_page_dirty(page);
351df4b2 1882 f2fs_put_page(page, 1);
e8d61a74 1883 }
351df4b2
JK		 1884}
1885
1886static void write_normal_summaries(struct f2fs_sb_info *sbi,
1887 block_t blkaddr, int type)
1888{
1889 int i, end;
1890 if (IS_DATASEG(type))
1891 end = type + NR_CURSEG_DATA_TYPE;
1892 else
1893 end = type + NR_CURSEG_NODE_TYPE;
1894
b7ad7512
CY		 1895 for (i = type; i < end; i++)
1896 write_current_sum_page(sbi, i, blkaddr + (i - type));
351df4b2
JK		 1897}
1898
1899void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1900{
aaec2b1d 1901 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
351df4b2
JK		 1902 write_compacted_summaries(sbi, start_blk);
1903 else
1904 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1905}
1906
1907void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1908{
119ee914 1909 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
351df4b2
JK		 1910}
1911
dfc08a12 1912int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
351df4b2
JK		 1913 unsigned int val, int alloc)
1914{
1915 int i;
1916
1917 if (type == NAT_JOURNAL) {
dfc08a12
CY		 1918 for (i = 0; i < nats_in_cursum(journal); i++) {
1919 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
351df4b2
JK		 1920 return i;
1921 }
dfc08a12
CY		 1922 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1923 return update_nats_in_cursum(journal, 1);
351df4b2 1924 } else if (type == SIT_JOURNAL) {
dfc08a12
CY		 1925 for (i = 0; i < sits_in_cursum(journal); i++)
1926 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
351df4b2 1927 return i;
dfc08a12
CY		 1928 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1929 return update_sits_in_cursum(journal, 1);
351df4b2
JK		 1930 }
1931 return -1;
1932}
1933
1934static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1935 unsigned int segno)
1936{
2cc22186 1937 return get_meta_page(sbi, current_sit_addr(sbi, segno));
351df4b2
JK		 1938}
1939
1940static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1941 unsigned int start)
1942{
1943 struct sit_info *sit_i = SIT_I(sbi);
1944 struct page *src_page, *dst_page;
1945 pgoff_t src_off, dst_off;
1946 void *src_addr, *dst_addr;
1947
1948 src_off = current_sit_addr(sbi, start);
1949 dst_off = next_sit_addr(sbi, src_off);
1950
1951 /* get current sit block page without lock */
1952 src_page = get_meta_page(sbi, src_off);
1953 dst_page = grab_meta_page(sbi, dst_off);
9850cf4a 1954 f2fs_bug_on(sbi, PageDirty(src_page));
351df4b2
JK		 1955
1956 src_addr = page_address(src_page);
1957 dst_addr = page_address(dst_page);
09cbfeaf 1958 memcpy(dst_addr, src_addr, PAGE_SIZE);
351df4b2
JK		 1959
1960 set_page_dirty(dst_page);
1961 f2fs_put_page(src_page, 1);
1962
1963 set_to_next_sit(sit_i, start);
1964
1965 return dst_page;
1966}
1967
184a5cd2
CY		 1968static struct sit_entry_set *grab_sit_entry_set(void)
1969{
1970 struct sit_entry_set *ses =
80c54505 1971 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
184a5cd2
CY		 1972
1973 ses->entry_cnt = 0;
1974 INIT_LIST_HEAD(&ses->set_list);
1975 return ses;
1976}
1977
1978static void release_sit_entry_set(struct sit_entry_set *ses)
1979{
1980 list_del(&ses->set_list);
1981 kmem_cache_free(sit_entry_set_slab, ses);
1982}
1983
1984static void adjust_sit_entry_set(struct sit_entry_set *ses,
1985 struct list_head *head)
1986{
1987 struct sit_entry_set *next = ses;
1988
1989 if (list_is_last(&ses->set_list, head))
1990 return;
1991
1992 list_for_each_entry_continue(next, head, set_list)
1993 if (ses->entry_cnt <= next->entry_cnt)
1994 break;
1995
1996 list_move_tail(&ses->set_list, &next->set_list);
1997}
1998
1999static void add_sit_entry(unsigned int segno, struct list_head *head)
2000{
2001 struct sit_entry_set *ses;
2002 unsigned int start_segno = START_SEGNO(segno);
2003
2004 list_for_each_entry(ses, head, set_list) {
2005 if (ses->start_segno == start_segno) {
2006 ses->entry_cnt++;
2007 adjust_sit_entry_set(ses, head);
2008 return;
2009 }
2010 }
2011
2012 ses = grab_sit_entry_set();
2013
2014 ses->start_segno = start_segno;
2015 ses->entry_cnt++;
2016 list_add(&ses->set_list, head);
2017}
2018
2019static void add_sits_in_set(struct f2fs_sb_info *sbi)
2020{
2021 struct f2fs_sm_info *sm_info = SM_I(sbi);
2022 struct list_head *set_list = &sm_info->sit_entry_set;
2023 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
184a5cd2
CY		 2024 unsigned int segno;
2025
7cd8558b 2026 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
184a5cd2
CY		 2027 add_sit_entry(segno, set_list);
2028}
2029
2030static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
351df4b2
JK		 2031{
2032 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2033 struct f2fs_journal *journal = curseg->journal;
351df4b2
JK		 2034 int i;
2035
b7ad7512 2036 down_write(&curseg->journal_rwsem);
dfc08a12 2037 for (i = 0; i < sits_in_cursum(journal); i++) {
184a5cd2
CY		 2038 unsigned int segno;
2039 bool dirtied;
2040
dfc08a12 2041 segno = le32_to_cpu(segno_in_journal(journal, i));
184a5cd2
CY		 2042 dirtied = __mark_sit_entry_dirty(sbi, segno);
2043
2044 if (!dirtied)
2045 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
351df4b2 2046 }
dfc08a12 2047 update_sits_in_cursum(journal, -i);
b7ad7512 2048 up_write(&curseg->journal_rwsem);
351df4b2
JK		 2049}
2050
0a8165d7 2051/*
351df4b2
JK		 2052 * CP calls this function, which flushes SIT entries including sit_journal,
2053 * and moves prefree segs to free segs.
2054 */
4b2fecc8 2055void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
351df4b2
JK		 2056{
2057 struct sit_info *sit_i = SIT_I(sbi);
2058 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2059 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2060 struct f2fs_journal *journal = curseg->journal;
184a5cd2
CY		 2061 struct sit_entry_set *ses, *tmp;
2062 struct list_head *head = &SM_I(sbi)->sit_entry_set;
184a5cd2 2063 bool to_journal = true;
4b2fecc8 2064 struct seg_entry *se;
351df4b2 2065
351df4b2
JK		 2066 mutex_lock(&sit_i->sentry_lock);
2067
2b11a74b
WL		 2068 if (!sit_i->dirty_sentries)
2069 goto out;
2070
351df4b2 2071 /*
184a5cd2
CY		 2072 * add and account sit entries of dirty bitmap in sit entry
2073 * set temporarily
351df4b2 2074 */
184a5cd2 2075 add_sits_in_set(sbi);
351df4b2 2076
184a5cd2
CY		 2077 /*
2078 * if there are no enough space in journal to store dirty sit
2079 * entries, remove all entries from journal and add and account
2080 * them in sit entry set.
2081 */
dfc08a12 2082 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
184a5cd2 2083 remove_sits_in_journal(sbi);
b2955550 2084
184a5cd2
CY		 2085 /*
2086 * there are two steps to flush sit entries:
2087 * #1, flush sit entries to journal in current cold data summary block.
2088 * #2, flush sit entries to sit page.
2089 */
2090 list_for_each_entry_safe(ses, tmp, head, set_list) {
4a257ed6 2091 struct page *page = NULL;
184a5cd2
CY		 2092 struct f2fs_sit_block *raw_sit = NULL;
2093 unsigned int start_segno = ses->start_segno;
2094 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
7cd8558b 2095 (unsigned long)MAIN_SEGS(sbi));
184a5cd2
CY		 2096 unsigned int segno = start_segno;
2097
2098 if (to_journal &&
dfc08a12 2099 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
184a5cd2
CY		 2100 to_journal = false;
2101
b7ad7512
CY		 2102 if (to_journal) {
2103 down_write(&curseg->journal_rwsem);
2104 } else {
184a5cd2
CY		 2105 page = get_next_sit_page(sbi, start_segno);
2106 raw_sit = page_address(page);
351df4b2 2107 }
351df4b2 2108
184a5cd2
CY		 2109 /* flush dirty sit entries in region of current sit set */
2110 for_each_set_bit_from(segno, bitmap, end) {
2111 int offset, sit_offset;
4b2fecc8
JK		 2112
2113 se = get_seg_entry(sbi, segno);
184a5cd2
CY		 2114
2115 /* add discard candidates */
d7bc2484 2116 if (cpc->reason != CP_DISCARD) {
4b2fecc8
JK		 2117 cpc->trim_start = segno;
2118 add_discard_addrs(sbi, cpc);
2119 }
184a5cd2
CY		 2120
2121 if (to_journal) {
dfc08a12 2122 offset = lookup_journal_in_cursum(journal,
184a5cd2
CY		 2123 SIT_JOURNAL, segno, 1);
2124 f2fs_bug_on(sbi, offset < 0);
dfc08a12 2125 segno_in_journal(journal, offset) =
184a5cd2
CY		 2126 cpu_to_le32(segno);
2127 seg_info_to_raw_sit(se,
dfc08a12 2128 &sit_in_journal(journal, offset));
184a5cd2
CY		 2129 } else {
2130 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2131 seg_info_to_raw_sit(se,
2132 &raw_sit->entries[sit_offset]);
2133 }
351df4b2 2134
184a5cd2
CY		 2135 __clear_bit(segno, bitmap);
2136 sit_i->dirty_sentries--;
2137 ses->entry_cnt--;
351df4b2
JK		 2138 }
2139
b7ad7512
CY		 2140 if (to_journal)
2141 up_write(&curseg->journal_rwsem);
2142 else
184a5cd2
CY		 2143 f2fs_put_page(page, 1);
2144
2145 f2fs_bug_on(sbi, ses->entry_cnt);
2146 release_sit_entry_set(ses);
351df4b2 2147 }
184a5cd2
CY		 2148
2149 f2fs_bug_on(sbi, !list_empty(head));
2150 f2fs_bug_on(sbi, sit_i->dirty_sentries);
184a5cd2 2151out:
4b2fecc8
JK		 2152 if (cpc->reason == CP_DISCARD) {
2153 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2154 add_discard_addrs(sbi, cpc);
2155 }
351df4b2 2156 mutex_unlock(&sit_i->sentry_lock);
351df4b2 2157
351df4b2
JK		 2158 set_prefree_as_free_segments(sbi);
2159}
2160
2161static int build_sit_info(struct f2fs_sb_info *sbi)
2162{
2163 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2164 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2165 struct sit_info *sit_i;
2166 unsigned int sit_segs, start;
2167 char *src_bitmap, *dst_bitmap;
2168 unsigned int bitmap_size;
2169
2170 /* allocate memory for SIT information */
2171 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2172 if (!sit_i)
2173 return -ENOMEM;
2174
2175 SM_I(sbi)->sit_info = sit_i;
2176
39307a8e
JK		 2177 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2178 sizeof(struct seg_entry), GFP_KERNEL);
351df4b2
JK		 2179 if (!sit_i->sentries)
2180 return -ENOMEM;
2181
7cd8558b 2182 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2183 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2184 if (!sit_i->dirty_sentries_bitmap)
2185 return -ENOMEM;
2186
7cd8558b 2187 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2188 sit_i->sentries[start].cur_valid_map
2189 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2190 sit_i->sentries[start].ckpt_valid_map
2191 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
a66cdd98 2192 if (!sit_i->sentries[start].cur_valid_map ||
3e025740 2193 !sit_i->sentries[start].ckpt_valid_map)
351df4b2 2194 return -ENOMEM;
3e025740
JK		 2195
2196 if (f2fs_discard_en(sbi)) {
2197 sit_i->sentries[start].discard_map
2198 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2199 if (!sit_i->sentries[start].discard_map)
2200 return -ENOMEM;
2201 }
351df4b2
JK		 2202 }
2203
60a3b782
JK		 2204 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2205 if (!sit_i->tmp_map)
2206 return -ENOMEM;
2207
351df4b2 2208 if (sbi->segs_per_sec > 1) {
39307a8e
JK		 2209 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2210 sizeof(struct sec_entry), GFP_KERNEL);
351df4b2
JK		 2211 if (!sit_i->sec_entries)
2212 return -ENOMEM;
2213 }
2214
2215 /* get information related with SIT */
2216 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2217
2218 /* setup SIT bitmap from ckeckpoint pack */
2219 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2220 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2221
79b5793b 2222 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
351df4b2
JK		 2223 if (!dst_bitmap)
2224 return -ENOMEM;
351df4b2
JK		 2225
2226 /* init SIT information */
2227 sit_i->s_ops = &default_salloc_ops;
2228
2229 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2230 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2231 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2232 sit_i->sit_bitmap = dst_bitmap;
2233 sit_i->bitmap_size = bitmap_size;
2234 sit_i->dirty_sentries = 0;
2235 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2236 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2237 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2238 mutex_init(&sit_i->sentry_lock);
2239 return 0;
2240}
2241
2242static int build_free_segmap(struct f2fs_sb_info *sbi)
2243{
351df4b2
JK		 2244 struct free_segmap_info *free_i;
2245 unsigned int bitmap_size, sec_bitmap_size;
2246
2247 /* allocate memory for free segmap information */
2248 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2249 if (!free_i)
2250 return -ENOMEM;
2251
2252 SM_I(sbi)->free_info = free_i;
2253
7cd8558b 2254 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
39307a8e 2255 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2256 if (!free_i->free_segmap)
2257 return -ENOMEM;
2258
7cd8558b 2259 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
39307a8e 2260 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
351df4b2
JK		 2261 if (!free_i->free_secmap)
2262 return -ENOMEM;
2263
2264 /* set all segments as dirty temporarily */
2265 memset(free_i->free_segmap, 0xff, bitmap_size);
2266 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2267
2268 /* init free segmap information */
7cd8558b 2269 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
351df4b2
JK		 2270 free_i->free_segments = 0;
2271 free_i->free_sections = 0;
1a118ccf 2272 spin_lock_init(&free_i->segmap_lock);
351df4b2
JK		 2273 return 0;
2274}
2275
2276static int build_curseg(struct f2fs_sb_info *sbi)
2277{
1042d60f 2278 struct curseg_info *array;
351df4b2
JK		 2279 int i;
2280
b434babf 2281 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
351df4b2
JK		 2282 if (!array)
2283 return -ENOMEM;
2284
2285 SM_I(sbi)->curseg_array = array;
2286
2287 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2288 mutex_init(&array[i].curseg_mutex);
09cbfeaf 2289 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
351df4b2
JK		 2290 if (!array[i].sum_blk)
2291 return -ENOMEM;
b7ad7512
CY		 2292 init_rwsem(&array[i].journal_rwsem);
2293 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2294 GFP_KERNEL);
2295 if (!array[i].journal)
2296 return -ENOMEM;
351df4b2
JK		 2297 array[i].segno = NULL_SEGNO;
2298 array[i].next_blkoff = 0;
2299 }
2300 return restore_curseg_summaries(sbi);
2301}
2302
2303static void build_sit_entries(struct f2fs_sb_info *sbi)
2304{
2305 struct sit_info *sit_i = SIT_I(sbi);
2306 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
b7ad7512 2307 struct f2fs_journal *journal = curseg->journal;
74de593a
CY		 2308 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2309 unsigned int i, start, end;
2310 unsigned int readed, start_blk = 0;
e9f5b8b8 2311 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
351df4b2 2312
74de593a 2313 do {
26879fb1 2314 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
74de593a
CY		 2315
2316 start = start_blk * sit_i->sents_per_block;
2317 end = (start_blk + readed) * sit_i->sents_per_block;
2318
7cd8558b 2319 for (; start < end && start < MAIN_SEGS(sbi); start++) {
74de593a
CY		 2320 struct seg_entry *se = &sit_i->sentries[start];
2321 struct f2fs_sit_block *sit_blk;
2322 struct f2fs_sit_entry sit;
2323 struct page *page;
2324
74de593a
CY		 2325 page = get_current_sit_page(sbi, start);
2326 sit_blk = (struct f2fs_sit_block *)page_address(page);
2327 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2328 f2fs_put_page(page, 1);
d600af23 2329
74de593a
CY		 2330 check_block_count(sbi, start, &sit);
2331 seg_info_from_raw_sit(se, &sit);
a66cdd98
JK		 2332
2333 /* build discard map only one time */
3e025740
JK		 2334 if (f2fs_discard_en(sbi)) {
2335 memcpy(se->discard_map, se->cur_valid_map,
2336 SIT_VBLOCK_MAP_SIZE);
2337 sbi->discard_blks += sbi->blocks_per_seg -
2338 se->valid_blocks;
2339 }
a66cdd98 2340
d600af23
CY		 2341 if (sbi->segs_per_sec > 1)
2342 get_sec_entry(sbi, start)->valid_blocks +=
2343 se->valid_blocks;
351df4b2 2344 }
74de593a
CY		 2345 start_blk += readed;
2346 } while (start_blk < sit_blk_cnt);
d600af23
CY		 2347
2348 down_read(&curseg->journal_rwsem);
2349 for (i = 0; i < sits_in_cursum(journal); i++) {
2350 struct f2fs_sit_entry sit;
2351 struct seg_entry *se;
2352 unsigned int old_valid_blocks;
2353
2354 start = le32_to_cpu(segno_in_journal(journal, i));
2355 se = &sit_i->sentries[start];
2356 sit = sit_in_journal(journal, i);
2357
2358 old_valid_blocks = se->valid_blocks;
2359
2360 check_block_count(sbi, start, &sit);
2361 seg_info_from_raw_sit(se, &sit);
2362
2363 if (f2fs_discard_en(sbi)) {
2364 memcpy(se->discard_map, se->cur_valid_map,
2365 SIT_VBLOCK_MAP_SIZE);
2366 sbi->discard_blks += old_valid_blocks -
2367 se->valid_blocks;
2368 }
2369
2370 if (sbi->segs_per_sec > 1)
2371 get_sec_entry(sbi, start)->valid_blocks +=
2372 se->valid_blocks - old_valid_blocks;
2373 }
2374 up_read(&curseg->journal_rwsem);
351df4b2
JK		 2375}
2376
2377static void init_free_segmap(struct f2fs_sb_info *sbi)
2378{
2379 unsigned int start;
2380 int type;
2381
7cd8558b 2382 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2383 struct seg_entry *sentry = get_seg_entry(sbi, start);
2384 if (!sentry->valid_blocks)
2385 __set_free(sbi, start);
2386 }
2387
2388 /* set use the current segments */
2389 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2390 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2391 __set_test_and_inuse(sbi, curseg_t->segno);
2392 }
2393}
2394
2395static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2396{
2397 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2398 struct free_segmap_info *free_i = FREE_I(sbi);
7cd8558b 2399 unsigned int segno = 0, offset = 0;
351df4b2
JK		 2400 unsigned short valid_blocks;
2401
8736fbf0 2402 while (1) {
351df4b2 2403 /* find dirty segment based on free segmap */
7cd8558b
JK		 2404 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2405 if (segno >= MAIN_SEGS(sbi))
351df4b2
JK		 2406 break;
2407 offset = segno + 1;
2408 valid_blocks = get_valid_blocks(sbi, segno, 0);
ec325b52 2409 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
351df4b2 2410 continue;
ec325b52
JK		 2411 if (valid_blocks > sbi->blocks_per_seg) {
2412 f2fs_bug_on(sbi, 1);
2413 continue;
2414 }
351df4b2
JK		 2415 mutex_lock(&dirty_i->seglist_lock);
2416 __locate_dirty_segment(sbi, segno, DIRTY);
2417 mutex_unlock(&dirty_i->seglist_lock);
2418 }
2419}
2420
5ec4e49f 2421static int init_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2422{
2423 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
7cd8558b 2424 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
351df4b2 2425
39307a8e 2426 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
5ec4e49f 2427 if (!dirty_i->victim_secmap)
351df4b2
JK		 2428 return -ENOMEM;
2429 return 0;
2430}
2431
2432static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2433{
2434 struct dirty_seglist_info *dirty_i;
2435 unsigned int bitmap_size, i;
2436
2437 /* allocate memory for dirty segments list information */
2438 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2439 if (!dirty_i)
2440 return -ENOMEM;
2441
2442 SM_I(sbi)->dirty_info = dirty_i;
2443 mutex_init(&dirty_i->seglist_lock);
2444
7cd8558b 2445 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
351df4b2
JK		 2446
2447 for (i = 0; i < NR_DIRTY_TYPE; i++) {
39307a8e 2448 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
351df4b2
JK		 2449 if (!dirty_i->dirty_segmap[i])
2450 return -ENOMEM;
2451 }
2452
2453 init_dirty_segmap(sbi);
5ec4e49f 2454 return init_victim_secmap(sbi);
351df4b2
JK		 2455}
2456
0a8165d7 2457/*
351df4b2
JK		 2458 * Update min, max modified time for cost-benefit GC algorithm
2459 */
2460static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2461{
2462 struct sit_info *sit_i = SIT_I(sbi);
2463 unsigned int segno;
2464
2465 mutex_lock(&sit_i->sentry_lock);
2466
2467 sit_i->min_mtime = LLONG_MAX;
2468
7cd8558b 2469 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
351df4b2
JK		 2470 unsigned int i;
2471 unsigned long long mtime = 0;
2472
2473 for (i = 0; i < sbi->segs_per_sec; i++)
2474 mtime += get_seg_entry(sbi, segno + i)->mtime;
2475
2476 mtime = div_u64(mtime, sbi->segs_per_sec);
2477
2478 if (sit_i->min_mtime > mtime)
2479 sit_i->min_mtime = mtime;
2480 }
2481 sit_i->max_mtime = get_mtime(sbi);
2482 mutex_unlock(&sit_i->sentry_lock);
2483}
2484
2485int build_segment_manager(struct f2fs_sb_info *sbi)
2486{
2487 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2488 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1042d60f 2489 struct f2fs_sm_info *sm_info;
351df4b2
JK		 2490 int err;
2491
2492 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2493 if (!sm_info)
2494 return -ENOMEM;
2495
2496 /* init sm info */
2497 sbi->sm_info = sm_info;
351df4b2
JK		 2498 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2499 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2500 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2501 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2502 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2503 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2504 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
58c41035
JK		 2505 sm_info->rec_prefree_segments = sm_info->main_segments *
2506 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
44a83499
JK		 2507 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2508 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2509
52763a4b
JK		 2510 if (!test_opt(sbi, LFS))
2511 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
216fbd64 2512 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
c1ce1b02 2513 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
351df4b2 2514
7fd9e544 2515 INIT_LIST_HEAD(&sm_info->discard_list);
275b66b0 2516 INIT_LIST_HEAD(&sm_info->wait_list);
7fd9e544
JK		 2517 sm_info->nr_discards = 0;
2518 sm_info->max_discards = 0;
2519
bba681cb
JK		 2520 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2521
184a5cd2
CY		 2522 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2523
b270ad6f 2524 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2163d198
GZ		 2525 err = create_flush_cmd_control(sbi);
2526 if (err)
a688b9d9 2527 return err;
6b4afdd7
JK		 2528 }
2529
351df4b2
JK		 2530 err = build_sit_info(sbi);
2531 if (err)
2532 return err;
2533 err = build_free_segmap(sbi);
2534 if (err)
2535 return err;
2536 err = build_curseg(sbi);
2537 if (err)
2538 return err;
2539
2540 /* reinit free segmap based on SIT */
2541 build_sit_entries(sbi);
2542
2543 init_free_segmap(sbi);
2544 err = build_dirty_segmap(sbi);
2545 if (err)
2546 return err;
2547
2548 init_min_max_mtime(sbi);
2549 return 0;
2550}
2551
2552static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2553 enum dirty_type dirty_type)
2554{
2555 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2556
2557 mutex_lock(&dirty_i->seglist_lock);
39307a8e 2558 kvfree(dirty_i->dirty_segmap[dirty_type]);
351df4b2
JK		 2559 dirty_i->nr_dirty[dirty_type] = 0;
2560 mutex_unlock(&dirty_i->seglist_lock);
2561}
2562
5ec4e49f 2563static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
351df4b2
JK		 2564{
2565 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
39307a8e 2566 kvfree(dirty_i->victim_secmap);
351df4b2
JK		 2567}
2568
2569static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2570{
2571 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2572 int i;
2573
2574 if (!dirty_i)
2575 return;
2576
2577 /* discard pre-free/dirty segments list */
2578 for (i = 0; i < NR_DIRTY_TYPE; i++)
2579 discard_dirty_segmap(sbi, i);
2580
5ec4e49f 2581 destroy_victim_secmap(sbi);
351df4b2
JK		 2582 SM_I(sbi)->dirty_info = NULL;
2583 kfree(dirty_i);
2584}
2585
2586static void destroy_curseg(struct f2fs_sb_info *sbi)
2587{
2588 struct curseg_info *array = SM_I(sbi)->curseg_array;
2589 int i;
2590
2591 if (!array)
2592 return;
2593 SM_I(sbi)->curseg_array = NULL;
b7ad7512 2594 for (i = 0; i < NR_CURSEG_TYPE; i++) {
351df4b2 2595 kfree(array[i].sum_blk);
b7ad7512
CY		 2596 kfree(array[i].journal);
2597 }
351df4b2
JK		 2598 kfree(array);
2599}
2600
2601static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2602{
2603 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2604 if (!free_i)
2605 return;
2606 SM_I(sbi)->free_info = NULL;
39307a8e
JK		 2607 kvfree(free_i->free_segmap);
2608 kvfree(free_i->free_secmap);
351df4b2
JK		 2609 kfree(free_i);
2610}
2611
2612static void destroy_sit_info(struct f2fs_sb_info *sbi)
2613{
2614 struct sit_info *sit_i = SIT_I(sbi);
2615 unsigned int start;
2616
2617 if (!sit_i)
2618 return;
2619
2620 if (sit_i->sentries) {
7cd8558b 2621 for (start = 0; start < MAIN_SEGS(sbi); start++) {
351df4b2
JK		 2622 kfree(sit_i->sentries[start].cur_valid_map);
2623 kfree(sit_i->sentries[start].ckpt_valid_map);
a66cdd98 2624 kfree(sit_i->sentries[start].discard_map);
351df4b2
JK		 2625 }
2626 }
60a3b782
JK		 2627 kfree(sit_i->tmp_map);
2628
39307a8e
JK		 2629 kvfree(sit_i->sentries);
2630 kvfree(sit_i->sec_entries);
2631 kvfree(sit_i->dirty_sentries_bitmap);
351df4b2
JK		 2632
2633 SM_I(sbi)->sit_info = NULL;
2634 kfree(sit_i->sit_bitmap);
2635 kfree(sit_i);
2636}
2637
2638void destroy_segment_manager(struct f2fs_sb_info *sbi)
2639{
2640 struct f2fs_sm_info *sm_info = SM_I(sbi);
a688b9d9 2641
3b03f724
CY		 2642 if (!sm_info)
2643 return;
2163d198 2644 destroy_flush_cmd_control(sbi);
351df4b2
JK		 2645 destroy_dirty_segmap(sbi);
2646 destroy_curseg(sbi);
2647 destroy_free_segmap(sbi);
2648 destroy_sit_info(sbi);
2649 sbi->sm_info = NULL;
2650 kfree(sm_info);
2651}
7fd9e544
JK		 2652
2653int __init create_segment_manager_caches(void)
2654{
2655 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
e8512d2e 2656 sizeof(struct discard_entry));
7fd9e544 2657 if (!discard_entry_slab)
184a5cd2
CY		 2658 goto fail;
2659
275b66b0
CY		 2660 bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2661 sizeof(struct bio_entry));
2662 if (!bio_entry_slab)
6ab2a308 2663 goto destroy_discard_entry;
275b66b0 2664
184a5cd2 2665 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
c9ee0085 2666 sizeof(struct sit_entry_set));
184a5cd2 2667 if (!sit_entry_set_slab)
275b66b0 2668 goto destroy_bio_entry;
88b88a66
JK		 2669
2670 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2671 sizeof(struct inmem_pages));
2672 if (!inmem_entry_slab)
2673 goto destroy_sit_entry_set;
7fd9e544 2674 return 0;
184a5cd2 2675
88b88a66
JK		 2676destroy_sit_entry_set:
2677 kmem_cache_destroy(sit_entry_set_slab);
275b66b0
CY		 2678destroy_bio_entry:
2679 kmem_cache_destroy(bio_entry_slab);
6ab2a308 2680destroy_discard_entry:
184a5cd2
CY		 2681 kmem_cache_destroy(discard_entry_slab);
2682fail:
2683 return -ENOMEM;
7fd9e544
JK		 2684}
2685
2686void destroy_segment_manager_caches(void)
2687{
184a5cd2 2688 kmem_cache_destroy(sit_entry_set_slab);
275b66b0 2689 kmem_cache_destroy(bio_entry_slab);
7fd9e544 2690 kmem_cache_destroy(discard_entry_slab);
88b88a66 2691 kmem_cache_destroy(inmem_entry_slab);
7fd9e544 2692}
Ubuntu Artful kernel mirror