1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/data.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/buffer_head.h> 11 #include <linux/mpage.h> 12 #include <linux/writeback.h> 13 #include <linux/backing-dev.h> 14 #include <linux/pagevec.h> 15 #include <linux/blkdev.h> 16 #include <linux/bio.h> 17 #include <linux/prefetch.h> 18 #include <linux/uio.h> 19 #include <linux/cleancache.h> 20 #include <linux/sched/signal.h> 21 22 #include "f2fs.h" 23 #include "node.h" 24 #include "segment.h" 25 #include "trace.h" 26 #include <trace/events/f2fs.h> 27 28 #define NUM_PREALLOC_POST_READ_CTXS 128 29 30 static struct kmem_cache *bio_post_read_ctx_cache; 31 static mempool_t *bio_post_read_ctx_pool; 32 33 static bool __is_cp_guaranteed(struct page *page) 34 { 35 struct address_space *mapping = page->mapping; 36 struct inode *inode; 37 struct f2fs_sb_info *sbi; 38 39 if (!mapping) 40 return false; 41 42 inode = mapping->host; 43 sbi = F2FS_I_SB(inode); 44 45 if (inode->i_ino == F2FS_META_INO(sbi) || 46 inode->i_ino == F2FS_NODE_INO(sbi) || 47 S_ISDIR(inode->i_mode) || 48 (S_ISREG(inode->i_mode) && 49 is_inode_flag_set(inode, FI_ATOMIC_FILE)) || 50 is_cold_data(page)) 51 return true; 52 return false; 53 } 54 55 static enum count_type __read_io_type(struct page *page) 56 { 57 struct address_space *mapping = page->mapping; 58 59 if (mapping) { 60 struct inode *inode = mapping->host; 61 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 62 63 if (inode->i_ino == F2FS_META_INO(sbi)) 64 return F2FS_RD_META; 65 66 if (inode->i_ino == F2FS_NODE_INO(sbi)) 67 return F2FS_RD_NODE; 68 } 69 return F2FS_RD_DATA; 70 } 71 72 /* postprocessing steps for read bios */ 73 enum bio_post_read_step { 74 STEP_INITIAL = 0, 75 STEP_DECRYPT, 76 }; 77 78 struct bio_post_read_ctx { 79 struct bio *bio; 80 struct work_struct work; 81 unsigned int cur_step; 82 unsigned int enabled_steps; 83 }; 84 85 static void __read_end_io(struct bio *bio) 86 { 87 struct page *page; 88 struct bio_vec *bv; 89 int i; 90 91 bio_for_each_segment_all(bv, bio, i) { 92 page = bv->bv_page; 93 94 /* PG_error was set if any post_read step failed */ 95 if (bio->bi_status || PageError(page)) { 96 ClearPageUptodate(page); 97 /* will re-read again later */ 98 ClearPageError(page); 99 } else { 100 SetPageUptodate(page); 101 } 102 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 103 unlock_page(page); 104 } 105 if (bio->bi_private) 106 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 107 bio_put(bio); 108 } 109 110 static void bio_post_read_processing(struct bio_post_read_ctx *ctx); 111 112 static void decrypt_work(struct work_struct *work) 113 { 114 struct bio_post_read_ctx *ctx = 115 container_of(work, struct bio_post_read_ctx, work); 116 117 fscrypt_decrypt_bio(ctx->bio); 118 119 bio_post_read_processing(ctx); 120 } 121 122 static void bio_post_read_processing(struct bio_post_read_ctx *ctx) 123 { 124 switch (++ctx->cur_step) { 125 case STEP_DECRYPT: 126 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) { 127 INIT_WORK(&ctx->work, decrypt_work); 128 fscrypt_enqueue_decrypt_work(&ctx->work); 129 return; 130 } 131 ctx->cur_step++; 132 /* fall-through */ 133 default: 134 __read_end_io(ctx->bio); 135 } 136 } 137 138 static bool f2fs_bio_post_read_required(struct bio *bio) 139 { 140 return bio->bi_private && !bio->bi_status; 141 } 142 143 static void f2fs_read_end_io(struct bio *bio) 144 { 145 if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)), 146 FAULT_READ_IO)) { 147 f2fs_show_injection_info(FAULT_READ_IO); 148 bio->bi_status = BLK_STS_IOERR; 149 } 150 151 if (f2fs_bio_post_read_required(bio)) { 152 struct bio_post_read_ctx *ctx = bio->bi_private; 153 154 ctx->cur_step = STEP_INITIAL; 155 bio_post_read_processing(ctx); 156 return; 157 } 158 159 __read_end_io(bio); 160 } 161 162 static void f2fs_write_end_io(struct bio *bio) 163 { 164 struct f2fs_sb_info *sbi = bio->bi_private; 165 struct bio_vec *bvec; 166 int i; 167 168 if (time_to_inject(sbi, FAULT_WRITE_IO)) { 169 f2fs_show_injection_info(FAULT_WRITE_IO); 170 bio->bi_status = BLK_STS_IOERR; 171 } 172 173 bio_for_each_segment_all(bvec, bio, i) { 174 struct page *page = bvec->bv_page; 175 enum count_type type = WB_DATA_TYPE(page); 176 177 if (IS_DUMMY_WRITTEN_PAGE(page)) { 178 set_page_private(page, (unsigned long)NULL); 179 ClearPagePrivate(page); 180 unlock_page(page); 181 mempool_free(page, sbi->write_io_dummy); 182 183 if (unlikely(bio->bi_status)) 184 f2fs_stop_checkpoint(sbi, true); 185 continue; 186 } 187 188 fscrypt_pullback_bio_page(&page, true); 189 190 if (unlikely(bio->bi_status)) { 191 mapping_set_error(page->mapping, -EIO); 192 if (type == F2FS_WB_CP_DATA) 193 f2fs_stop_checkpoint(sbi, true); 194 } 195 196 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 197 page->index != nid_of_node(page)); 198 199 dec_page_count(sbi, type); 200 if (f2fs_in_warm_node_list(sbi, page)) 201 f2fs_del_fsync_node_entry(sbi, page); 202 clear_cold_data(page); 203 end_page_writeback(page); 204 } 205 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 206 wq_has_sleeper(&sbi->cp_wait)) 207 wake_up(&sbi->cp_wait); 208 209 bio_put(bio); 210 } 211 212 /* 213 * Return true, if pre_bio's bdev is same as its target device. 214 */ 215 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 216 block_t blk_addr, struct bio *bio) 217 { 218 struct block_device *bdev = sbi->sb->s_bdev; 219 int i; 220 221 for (i = 0; i < sbi->s_ndevs; i++) { 222 if (FDEV(i).start_blk <= blk_addr && 223 FDEV(i).end_blk >= blk_addr) { 224 blk_addr -= FDEV(i).start_blk; 225 bdev = FDEV(i).bdev; 226 break; 227 } 228 } 229 if (bio) { 230 bio_set_dev(bio, bdev); 231 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); 232 } 233 return bdev; 234 } 235 236 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 237 { 238 int i; 239 240 for (i = 0; i < sbi->s_ndevs; i++) 241 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 242 return i; 243 return 0; 244 } 245 246 static bool __same_bdev(struct f2fs_sb_info *sbi, 247 block_t blk_addr, struct bio *bio) 248 { 249 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL); 250 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno; 251 } 252 253 /* 254 * Low-level block read/write IO operations. 255 */ 256 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr, 257 struct writeback_control *wbc, 258 int npages, bool is_read, 259 enum page_type type, enum temp_type temp) 260 { 261 struct bio *bio; 262 263 bio = f2fs_bio_alloc(sbi, npages, true); 264 265 f2fs_target_device(sbi, blk_addr, bio); 266 if (is_read) { 267 bio->bi_end_io = f2fs_read_end_io; 268 bio->bi_private = NULL; 269 } else { 270 bio->bi_end_io = f2fs_write_end_io; 271 bio->bi_private = sbi; 272 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp); 273 } 274 if (wbc) 275 wbc_init_bio(wbc, bio); 276 277 return bio; 278 } 279 280 static inline void __submit_bio(struct f2fs_sb_info *sbi, 281 struct bio *bio, enum page_type type) 282 { 283 if (!is_read_io(bio_op(bio))) { 284 unsigned int start; 285 286 if (type != DATA && type != NODE) 287 goto submit_io; 288 289 if (test_opt(sbi, LFS) && current->plug) 290 blk_finish_plug(current->plug); 291 292 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS; 293 start %= F2FS_IO_SIZE(sbi); 294 295 if (start == 0) 296 goto submit_io; 297 298 /* fill dummy pages */ 299 for (; start < F2FS_IO_SIZE(sbi); start++) { 300 struct page *page = 301 mempool_alloc(sbi->write_io_dummy, 302 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL); 303 f2fs_bug_on(sbi, !page); 304 305 SetPagePrivate(page); 306 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE); 307 lock_page(page); 308 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 309 f2fs_bug_on(sbi, 1); 310 } 311 /* 312 * In the NODE case, we lose next block address chain. So, we 313 * need to do checkpoint in f2fs_sync_file. 314 */ 315 if (type == NODE) 316 set_sbi_flag(sbi, SBI_NEED_CP); 317 } 318 submit_io: 319 if (is_read_io(bio_op(bio))) 320 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 321 else 322 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 323 submit_bio(bio); 324 } 325 326 static void __submit_merged_bio(struct f2fs_bio_info *io) 327 { 328 struct f2fs_io_info *fio = &io->fio; 329 330 if (!io->bio) 331 return; 332 333 bio_set_op_attrs(io->bio, fio->op, fio->op_flags); 334 335 if (is_read_io(fio->op)) 336 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 337 else 338 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 339 340 __submit_bio(io->sbi, io->bio, fio->type); 341 io->bio = NULL; 342 } 343 344 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode, 345 struct page *page, nid_t ino) 346 { 347 struct bio_vec *bvec; 348 struct page *target; 349 int i; 350 351 if (!io->bio) 352 return false; 353 354 if (!inode && !page && !ino) 355 return true; 356 357 bio_for_each_segment_all(bvec, io->bio, i) { 358 359 if (bvec->bv_page->mapping) 360 target = bvec->bv_page; 361 else 362 target = fscrypt_control_page(bvec->bv_page); 363 364 if (inode && inode == target->mapping->host) 365 return true; 366 if (page && page == target) 367 return true; 368 if (ino && ino == ino_of_node(target)) 369 return true; 370 } 371 372 return false; 373 } 374 375 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode, 376 struct page *page, nid_t ino, 377 enum page_type type) 378 { 379 enum page_type btype = PAGE_TYPE_OF_BIO(type); 380 enum temp_type temp; 381 struct f2fs_bio_info *io; 382 bool ret = false; 383 384 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 385 io = sbi->write_io[btype] + temp; 386 387 down_read(&io->io_rwsem); 388 ret = __has_merged_page(io, inode, page, ino); 389 up_read(&io->io_rwsem); 390 391 /* TODO: use HOT temp only for meta pages now. */ 392 if (ret || btype == META) 393 break; 394 } 395 return ret; 396 } 397 398 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 399 enum page_type type, enum temp_type temp) 400 { 401 enum page_type btype = PAGE_TYPE_OF_BIO(type); 402 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 403 404 down_write(&io->io_rwsem); 405 406 /* change META to META_FLUSH in the checkpoint procedure */ 407 if (type >= META_FLUSH) { 408 io->fio.type = META_FLUSH; 409 io->fio.op = REQ_OP_WRITE; 410 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC; 411 if (!test_opt(sbi, NOBARRIER)) 412 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 413 } 414 __submit_merged_bio(io); 415 up_write(&io->io_rwsem); 416 } 417 418 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 419 struct inode *inode, struct page *page, 420 nid_t ino, enum page_type type, bool force) 421 { 422 enum temp_type temp; 423 424 if (!force && !has_merged_page(sbi, inode, page, ino, type)) 425 return; 426 427 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 428 429 __f2fs_submit_merged_write(sbi, type, temp); 430 431 /* TODO: use HOT temp only for meta pages now. */ 432 if (type >= META) 433 break; 434 } 435 } 436 437 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 438 { 439 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true); 440 } 441 442 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 443 struct inode *inode, struct page *page, 444 nid_t ino, enum page_type type) 445 { 446 __submit_merged_write_cond(sbi, inode, page, ino, type, false); 447 } 448 449 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 450 { 451 f2fs_submit_merged_write(sbi, DATA); 452 f2fs_submit_merged_write(sbi, NODE); 453 f2fs_submit_merged_write(sbi, META); 454 } 455 456 /* 457 * Fill the locked page with data located in the block address. 458 * A caller needs to unlock the page on failure. 459 */ 460 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 461 { 462 struct bio *bio; 463 struct page *page = fio->encrypted_page ? 464 fio->encrypted_page : fio->page; 465 466 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 467 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) 468 return -EFAULT; 469 470 trace_f2fs_submit_page_bio(page, fio); 471 f2fs_trace_ios(fio, 0); 472 473 /* Allocate a new bio */ 474 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc, 475 1, is_read_io(fio->op), fio->type, fio->temp); 476 477 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 478 bio_put(bio); 479 return -EFAULT; 480 } 481 482 if (fio->io_wbc && !is_read_io(fio->op)) 483 wbc_account_io(fio->io_wbc, page, PAGE_SIZE); 484 485 bio_set_op_attrs(bio, fio->op, fio->op_flags); 486 487 inc_page_count(fio->sbi, is_read_io(fio->op) ? 488 __read_io_type(page): WB_DATA_TYPE(fio->page)); 489 490 __submit_bio(fio->sbi, bio, fio->type); 491 return 0; 492 } 493 494 void f2fs_submit_page_write(struct f2fs_io_info *fio) 495 { 496 struct f2fs_sb_info *sbi = fio->sbi; 497 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 498 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 499 struct page *bio_page; 500 501 f2fs_bug_on(sbi, is_read_io(fio->op)); 502 503 down_write(&io->io_rwsem); 504 next: 505 if (fio->in_list) { 506 spin_lock(&io->io_lock); 507 if (list_empty(&io->io_list)) { 508 spin_unlock(&io->io_lock); 509 goto out; 510 } 511 fio = list_first_entry(&io->io_list, 512 struct f2fs_io_info, list); 513 list_del(&fio->list); 514 spin_unlock(&io->io_lock); 515 } 516 517 if (__is_valid_data_blkaddr(fio->old_blkaddr)) 518 verify_block_addr(fio, fio->old_blkaddr); 519 verify_block_addr(fio, fio->new_blkaddr); 520 521 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page; 522 523 /* set submitted = true as a return value */ 524 fio->submitted = true; 525 526 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 527 528 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 || 529 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) || 530 !__same_bdev(sbi, fio->new_blkaddr, io->bio))) 531 __submit_merged_bio(io); 532 alloc_new: 533 if (io->bio == NULL) { 534 if ((fio->type == DATA || fio->type == NODE) && 535 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 536 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 537 fio->retry = true; 538 goto skip; 539 } 540 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc, 541 BIO_MAX_PAGES, false, 542 fio->type, fio->temp); 543 io->fio = *fio; 544 } 545 546 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 547 __submit_merged_bio(io); 548 goto alloc_new; 549 } 550 551 if (fio->io_wbc) 552 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE); 553 554 io->last_block_in_bio = fio->new_blkaddr; 555 f2fs_trace_ios(fio, 0); 556 557 trace_f2fs_submit_page_write(fio->page, fio); 558 skip: 559 if (fio->in_list) 560 goto next; 561 out: 562 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 563 f2fs_is_checkpoint_ready(sbi)) 564 __submit_merged_bio(io); 565 up_write(&io->io_rwsem); 566 } 567 568 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 569 unsigned nr_pages, unsigned op_flag) 570 { 571 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 572 struct bio *bio; 573 struct bio_post_read_ctx *ctx; 574 unsigned int post_read_steps = 0; 575 576 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) 577 return ERR_PTR(-EFAULT); 578 579 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false); 580 if (!bio) 581 return ERR_PTR(-ENOMEM); 582 f2fs_target_device(sbi, blkaddr, bio); 583 bio->bi_end_io = f2fs_read_end_io; 584 bio_set_op_attrs(bio, REQ_OP_READ, op_flag); 585 586 if (f2fs_encrypted_file(inode)) 587 post_read_steps |= 1 << STEP_DECRYPT; 588 if (post_read_steps) { 589 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 590 if (!ctx) { 591 bio_put(bio); 592 return ERR_PTR(-ENOMEM); 593 } 594 ctx->bio = bio; 595 ctx->enabled_steps = post_read_steps; 596 bio->bi_private = ctx; 597 } 598 599 return bio; 600 } 601 602 /* This can handle encryption stuffs */ 603 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 604 block_t blkaddr) 605 { 606 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0); 607 608 if (IS_ERR(bio)) 609 return PTR_ERR(bio); 610 611 /* wait for GCed page writeback via META_MAPPING */ 612 f2fs_wait_on_block_writeback(inode, blkaddr); 613 614 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 615 bio_put(bio); 616 return -EFAULT; 617 } 618 ClearPageError(page); 619 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 620 __submit_bio(F2FS_I_SB(inode), bio, DATA); 621 return 0; 622 } 623 624 static void __set_data_blkaddr(struct dnode_of_data *dn) 625 { 626 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 627 __le32 *addr_array; 628 int base = 0; 629 630 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 631 base = get_extra_isize(dn->inode); 632 633 /* Get physical address of data block */ 634 addr_array = blkaddr_in_node(rn); 635 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 636 } 637 638 /* 639 * Lock ordering for the change of data block address: 640 * ->data_page 641 * ->node_page 642 * update block addresses in the node page 643 */ 644 void f2fs_set_data_blkaddr(struct dnode_of_data *dn) 645 { 646 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 647 __set_data_blkaddr(dn); 648 if (set_page_dirty(dn->node_page)) 649 dn->node_changed = true; 650 } 651 652 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 653 { 654 dn->data_blkaddr = blkaddr; 655 f2fs_set_data_blkaddr(dn); 656 f2fs_update_extent_cache(dn); 657 } 658 659 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 660 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 661 { 662 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 663 int err; 664 665 if (!count) 666 return 0; 667 668 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 669 return -EPERM; 670 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 671 return err; 672 673 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 674 dn->ofs_in_node, count); 675 676 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 677 678 for (; count > 0; dn->ofs_in_node++) { 679 block_t blkaddr = datablock_addr(dn->inode, 680 dn->node_page, dn->ofs_in_node); 681 if (blkaddr == NULL_ADDR) { 682 dn->data_blkaddr = NEW_ADDR; 683 __set_data_blkaddr(dn); 684 count--; 685 } 686 } 687 688 if (set_page_dirty(dn->node_page)) 689 dn->node_changed = true; 690 return 0; 691 } 692 693 /* Should keep dn->ofs_in_node unchanged */ 694 int f2fs_reserve_new_block(struct dnode_of_data *dn) 695 { 696 unsigned int ofs_in_node = dn->ofs_in_node; 697 int ret; 698 699 ret = f2fs_reserve_new_blocks(dn, 1); 700 dn->ofs_in_node = ofs_in_node; 701 return ret; 702 } 703 704 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 705 { 706 bool need_put = dn->inode_page ? false : true; 707 int err; 708 709 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE); 710 if (err) 711 return err; 712 713 if (dn->data_blkaddr == NULL_ADDR) 714 err = f2fs_reserve_new_block(dn); 715 if (err || need_put) 716 f2fs_put_dnode(dn); 717 return err; 718 } 719 720 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 721 { 722 struct extent_info ei = {0,0,0}; 723 struct inode *inode = dn->inode; 724 725 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 726 dn->data_blkaddr = ei.blk + index - ei.fofs; 727 return 0; 728 } 729 730 return f2fs_reserve_block(dn, index); 731 } 732 733 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, 734 int op_flags, bool for_write) 735 { 736 struct address_space *mapping = inode->i_mapping; 737 struct dnode_of_data dn; 738 struct page *page; 739 struct extent_info ei = {0,0,0}; 740 int err; 741 742 page = f2fs_grab_cache_page(mapping, index, for_write); 743 if (!page) 744 return ERR_PTR(-ENOMEM); 745 746 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 747 dn.data_blkaddr = ei.blk + index - ei.fofs; 748 goto got_it; 749 } 750 751 set_new_dnode(&dn, inode, NULL, NULL, 0); 752 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 753 if (err) 754 goto put_err; 755 f2fs_put_dnode(&dn); 756 757 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 758 err = -ENOENT; 759 goto put_err; 760 } 761 got_it: 762 if (PageUptodate(page)) { 763 unlock_page(page); 764 return page; 765 } 766 767 /* 768 * A new dentry page is allocated but not able to be written, since its 769 * new inode page couldn't be allocated due to -ENOSPC. 770 * In such the case, its blkaddr can be remained as NEW_ADDR. 771 * see, f2fs_add_link -> f2fs_get_new_data_page -> 772 * f2fs_init_inode_metadata. 773 */ 774 if (dn.data_blkaddr == NEW_ADDR) { 775 zero_user_segment(page, 0, PAGE_SIZE); 776 if (!PageUptodate(page)) 777 SetPageUptodate(page); 778 unlock_page(page); 779 return page; 780 } 781 782 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr); 783 if (err) 784 goto put_err; 785 return page; 786 787 put_err: 788 f2fs_put_page(page, 1); 789 return ERR_PTR(err); 790 } 791 792 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index) 793 { 794 struct address_space *mapping = inode->i_mapping; 795 struct page *page; 796 797 page = find_get_page(mapping, index); 798 if (page && PageUptodate(page)) 799 return page; 800 f2fs_put_page(page, 0); 801 802 page = f2fs_get_read_data_page(inode, index, 0, false); 803 if (IS_ERR(page)) 804 return page; 805 806 if (PageUptodate(page)) 807 return page; 808 809 wait_on_page_locked(page); 810 if (unlikely(!PageUptodate(page))) { 811 f2fs_put_page(page, 0); 812 return ERR_PTR(-EIO); 813 } 814 return page; 815 } 816 817 /* 818 * If it tries to access a hole, return an error. 819 * Because, the callers, functions in dir.c and GC, should be able to know 820 * whether this page exists or not. 821 */ 822 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, 823 bool for_write) 824 { 825 struct address_space *mapping = inode->i_mapping; 826 struct page *page; 827 repeat: 828 page = f2fs_get_read_data_page(inode, index, 0, for_write); 829 if (IS_ERR(page)) 830 return page; 831 832 /* wait for read completion */ 833 lock_page(page); 834 if (unlikely(page->mapping != mapping)) { 835 f2fs_put_page(page, 1); 836 goto repeat; 837 } 838 if (unlikely(!PageUptodate(page))) { 839 f2fs_put_page(page, 1); 840 return ERR_PTR(-EIO); 841 } 842 return page; 843 } 844 845 /* 846 * Caller ensures that this data page is never allocated. 847 * A new zero-filled data page is allocated in the page cache. 848 * 849 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 850 * f2fs_unlock_op(). 851 * Note that, ipage is set only by make_empty_dir, and if any error occur, 852 * ipage should be released by this function. 853 */ 854 struct page *f2fs_get_new_data_page(struct inode *inode, 855 struct page *ipage, pgoff_t index, bool new_i_size) 856 { 857 struct address_space *mapping = inode->i_mapping; 858 struct page *page; 859 struct dnode_of_data dn; 860 int err; 861 862 page = f2fs_grab_cache_page(mapping, index, true); 863 if (!page) { 864 /* 865 * before exiting, we should make sure ipage will be released 866 * if any error occur. 867 */ 868 f2fs_put_page(ipage, 1); 869 return ERR_PTR(-ENOMEM); 870 } 871 872 set_new_dnode(&dn, inode, ipage, NULL, 0); 873 err = f2fs_reserve_block(&dn, index); 874 if (err) { 875 f2fs_put_page(page, 1); 876 return ERR_PTR(err); 877 } 878 if (!ipage) 879 f2fs_put_dnode(&dn); 880 881 if (PageUptodate(page)) 882 goto got_it; 883 884 if (dn.data_blkaddr == NEW_ADDR) { 885 zero_user_segment(page, 0, PAGE_SIZE); 886 if (!PageUptodate(page)) 887 SetPageUptodate(page); 888 } else { 889 f2fs_put_page(page, 1); 890 891 /* if ipage exists, blkaddr should be NEW_ADDR */ 892 f2fs_bug_on(F2FS_I_SB(inode), ipage); 893 page = f2fs_get_lock_data_page(inode, index, true); 894 if (IS_ERR(page)) 895 return page; 896 } 897 got_it: 898 if (new_i_size && i_size_read(inode) < 899 ((loff_t)(index + 1) << PAGE_SHIFT)) 900 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 901 return page; 902 } 903 904 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 905 { 906 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 907 struct f2fs_summary sum; 908 struct node_info ni; 909 block_t old_blkaddr; 910 blkcnt_t count = 1; 911 int err; 912 913 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 914 return -EPERM; 915 916 err = f2fs_get_node_info(sbi, dn->nid, &ni); 917 if (err) 918 return err; 919 920 dn->data_blkaddr = datablock_addr(dn->inode, 921 dn->node_page, dn->ofs_in_node); 922 if (dn->data_blkaddr != NULL_ADDR) 923 goto alloc; 924 925 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 926 return err; 927 928 alloc: 929 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 930 old_blkaddr = dn->data_blkaddr; 931 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr, 932 &sum, seg_type, NULL, false); 933 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 934 invalidate_mapping_pages(META_MAPPING(sbi), 935 old_blkaddr, old_blkaddr); 936 f2fs_set_data_blkaddr(dn); 937 938 /* 939 * i_size will be updated by direct_IO. Otherwise, we'll get stale 940 * data from unwritten block via dio_read. 941 */ 942 return 0; 943 } 944 945 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from) 946 { 947 struct inode *inode = file_inode(iocb->ki_filp); 948 struct f2fs_map_blocks map; 949 int flag; 950 int err = 0; 951 bool direct_io = iocb->ki_flags & IOCB_DIRECT; 952 953 /* convert inline data for Direct I/O*/ 954 if (direct_io) { 955 err = f2fs_convert_inline_inode(inode); 956 if (err) 957 return err; 958 } 959 960 if (is_inode_flag_set(inode, FI_NO_PREALLOC)) 961 return 0; 962 963 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos); 964 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from)); 965 if (map.m_len > map.m_lblk) 966 map.m_len -= map.m_lblk; 967 else 968 map.m_len = 0; 969 970 map.m_next_pgofs = NULL; 971 map.m_next_extent = NULL; 972 map.m_seg_type = NO_CHECK_TYPE; 973 974 if (direct_io) { 975 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint); 976 flag = f2fs_force_buffered_io(inode, iocb, from) ? 977 F2FS_GET_BLOCK_PRE_AIO : 978 F2FS_GET_BLOCK_PRE_DIO; 979 goto map_blocks; 980 } 981 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) { 982 err = f2fs_convert_inline_inode(inode); 983 if (err) 984 return err; 985 } 986 if (f2fs_has_inline_data(inode)) 987 return err; 988 989 flag = F2FS_GET_BLOCK_PRE_AIO; 990 991 map_blocks: 992 err = f2fs_map_blocks(inode, &map, 1, flag); 993 if (map.m_len > 0 && err == -ENOSPC) { 994 if (!direct_io) 995 set_inode_flag(inode, FI_NO_PREALLOC); 996 err = 0; 997 } 998 return err; 999 } 1000 1001 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 1002 { 1003 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1004 if (lock) 1005 down_read(&sbi->node_change); 1006 else 1007 up_read(&sbi->node_change); 1008 } else { 1009 if (lock) 1010 f2fs_lock_op(sbi); 1011 else 1012 f2fs_unlock_op(sbi); 1013 } 1014 } 1015 1016 /* 1017 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with 1018 * f2fs_map_blocks structure. 1019 * If original data blocks are allocated, then give them to blockdev. 1020 * Otherwise, 1021 * a. preallocate requested block addresses 1022 * b. do not use extent cache for better performance 1023 * c. give the block addresses to blockdev 1024 */ 1025 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 1026 int create, int flag) 1027 { 1028 unsigned int maxblocks = map->m_len; 1029 struct dnode_of_data dn; 1030 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1031 int mode = create ? ALLOC_NODE : LOOKUP_NODE; 1032 pgoff_t pgofs, end_offset, end; 1033 int err = 0, ofs = 1; 1034 unsigned int ofs_in_node, last_ofs_in_node; 1035 blkcnt_t prealloc; 1036 struct extent_info ei = {0,0,0}; 1037 block_t blkaddr; 1038 unsigned int start_pgofs; 1039 1040 if (!maxblocks) 1041 return 0; 1042 1043 map->m_len = 0; 1044 map->m_flags = 0; 1045 1046 /* it only supports block size == page size */ 1047 pgofs = (pgoff_t)map->m_lblk; 1048 end = pgofs + maxblocks; 1049 1050 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 1051 map->m_pblk = ei.blk + pgofs - ei.fofs; 1052 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 1053 map->m_flags = F2FS_MAP_MAPPED; 1054 if (map->m_next_extent) 1055 *map->m_next_extent = pgofs + map->m_len; 1056 goto out; 1057 } 1058 1059 next_dnode: 1060 if (create) 1061 __do_map_lock(sbi, flag, true); 1062 1063 /* When reading holes, we need its node page */ 1064 set_new_dnode(&dn, inode, NULL, NULL, 0); 1065 err = f2fs_get_dnode_of_data(&dn, pgofs, mode); 1066 if (err) { 1067 if (flag == F2FS_GET_BLOCK_BMAP) 1068 map->m_pblk = 0; 1069 if (err == -ENOENT) { 1070 err = 0; 1071 if (map->m_next_pgofs) 1072 *map->m_next_pgofs = 1073 f2fs_get_next_page_offset(&dn, pgofs); 1074 if (map->m_next_extent) 1075 *map->m_next_extent = 1076 f2fs_get_next_page_offset(&dn, pgofs); 1077 } 1078 goto unlock_out; 1079 } 1080 1081 start_pgofs = pgofs; 1082 prealloc = 0; 1083 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 1084 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 1085 1086 next_block: 1087 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node); 1088 1089 if (__is_valid_data_blkaddr(blkaddr) && 1090 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) { 1091 err = -EFAULT; 1092 goto sync_out; 1093 } 1094 1095 if (is_valid_data_blkaddr(sbi, blkaddr)) { 1096 /* use out-place-update for driect IO under LFS mode */ 1097 if (test_opt(sbi, LFS) && create && 1098 flag == F2FS_GET_BLOCK_DIO) { 1099 err = __allocate_data_block(&dn, map->m_seg_type); 1100 if (!err) 1101 set_inode_flag(inode, FI_APPEND_WRITE); 1102 } 1103 } else { 1104 if (create) { 1105 if (unlikely(f2fs_cp_error(sbi))) { 1106 err = -EIO; 1107 goto sync_out; 1108 } 1109 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1110 if (blkaddr == NULL_ADDR) { 1111 prealloc++; 1112 last_ofs_in_node = dn.ofs_in_node; 1113 } 1114 } else { 1115 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO && 1116 flag != F2FS_GET_BLOCK_DIO); 1117 err = __allocate_data_block(&dn, 1118 map->m_seg_type); 1119 if (!err) 1120 set_inode_flag(inode, FI_APPEND_WRITE); 1121 } 1122 if (err) 1123 goto sync_out; 1124 map->m_flags |= F2FS_MAP_NEW; 1125 blkaddr = dn.data_blkaddr; 1126 } else { 1127 if (flag == F2FS_GET_BLOCK_BMAP) { 1128 map->m_pblk = 0; 1129 goto sync_out; 1130 } 1131 if (flag == F2FS_GET_BLOCK_PRECACHE) 1132 goto sync_out; 1133 if (flag == F2FS_GET_BLOCK_FIEMAP && 1134 blkaddr == NULL_ADDR) { 1135 if (map->m_next_pgofs) 1136 *map->m_next_pgofs = pgofs + 1; 1137 goto sync_out; 1138 } 1139 if (flag != F2FS_GET_BLOCK_FIEMAP) { 1140 /* for defragment case */ 1141 if (map->m_next_pgofs) 1142 *map->m_next_pgofs = pgofs + 1; 1143 goto sync_out; 1144 } 1145 } 1146 } 1147 1148 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1149 goto skip; 1150 1151 if (map->m_len == 0) { 1152 /* preallocated unwritten block should be mapped for fiemap. */ 1153 if (blkaddr == NEW_ADDR) 1154 map->m_flags |= F2FS_MAP_UNWRITTEN; 1155 map->m_flags |= F2FS_MAP_MAPPED; 1156 1157 map->m_pblk = blkaddr; 1158 map->m_len = 1; 1159 } else if ((map->m_pblk != NEW_ADDR && 1160 blkaddr == (map->m_pblk + ofs)) || 1161 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1162 flag == F2FS_GET_BLOCK_PRE_DIO) { 1163 ofs++; 1164 map->m_len++; 1165 } else { 1166 goto sync_out; 1167 } 1168 1169 skip: 1170 dn.ofs_in_node++; 1171 pgofs++; 1172 1173 /* preallocate blocks in batch for one dnode page */ 1174 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1175 (pgofs == end || dn.ofs_in_node == end_offset)) { 1176 1177 dn.ofs_in_node = ofs_in_node; 1178 err = f2fs_reserve_new_blocks(&dn, prealloc); 1179 if (err) 1180 goto sync_out; 1181 1182 map->m_len += dn.ofs_in_node - ofs_in_node; 1183 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1184 err = -ENOSPC; 1185 goto sync_out; 1186 } 1187 dn.ofs_in_node = end_offset; 1188 } 1189 1190 if (pgofs >= end) 1191 goto sync_out; 1192 else if (dn.ofs_in_node < end_offset) 1193 goto next_block; 1194 1195 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1196 if (map->m_flags & F2FS_MAP_MAPPED) { 1197 unsigned int ofs = start_pgofs - map->m_lblk; 1198 1199 f2fs_update_extent_cache_range(&dn, 1200 start_pgofs, map->m_pblk + ofs, 1201 map->m_len - ofs); 1202 } 1203 } 1204 1205 f2fs_put_dnode(&dn); 1206 1207 if (create) { 1208 __do_map_lock(sbi, flag, false); 1209 f2fs_balance_fs(sbi, dn.node_changed); 1210 } 1211 goto next_dnode; 1212 1213 sync_out: 1214 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1215 if (map->m_flags & F2FS_MAP_MAPPED) { 1216 unsigned int ofs = start_pgofs - map->m_lblk; 1217 1218 f2fs_update_extent_cache_range(&dn, 1219 start_pgofs, map->m_pblk + ofs, 1220 map->m_len - ofs); 1221 } 1222 if (map->m_next_extent) 1223 *map->m_next_extent = pgofs + 1; 1224 } 1225 f2fs_put_dnode(&dn); 1226 unlock_out: 1227 if (create) { 1228 __do_map_lock(sbi, flag, false); 1229 f2fs_balance_fs(sbi, dn.node_changed); 1230 } 1231 out: 1232 trace_f2fs_map_blocks(inode, map, err); 1233 return err; 1234 } 1235 1236 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) 1237 { 1238 struct f2fs_map_blocks map; 1239 block_t last_lblk; 1240 int err; 1241 1242 if (pos + len > i_size_read(inode)) 1243 return false; 1244 1245 map.m_lblk = F2FS_BYTES_TO_BLK(pos); 1246 map.m_next_pgofs = NULL; 1247 map.m_next_extent = NULL; 1248 map.m_seg_type = NO_CHECK_TYPE; 1249 last_lblk = F2FS_BLK_ALIGN(pos + len); 1250 1251 while (map.m_lblk < last_lblk) { 1252 map.m_len = last_lblk - map.m_lblk; 1253 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT); 1254 if (err || map.m_len == 0) 1255 return false; 1256 map.m_lblk += map.m_len; 1257 } 1258 return true; 1259 } 1260 1261 static int __get_data_block(struct inode *inode, sector_t iblock, 1262 struct buffer_head *bh, int create, int flag, 1263 pgoff_t *next_pgofs, int seg_type) 1264 { 1265 struct f2fs_map_blocks map; 1266 int err; 1267 1268 map.m_lblk = iblock; 1269 map.m_len = bh->b_size >> inode->i_blkbits; 1270 map.m_next_pgofs = next_pgofs; 1271 map.m_next_extent = NULL; 1272 map.m_seg_type = seg_type; 1273 1274 err = f2fs_map_blocks(inode, &map, create, flag); 1275 if (!err) { 1276 map_bh(bh, inode->i_sb, map.m_pblk); 1277 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags; 1278 bh->b_size = (u64)map.m_len << inode->i_blkbits; 1279 } 1280 return err; 1281 } 1282 1283 static int get_data_block(struct inode *inode, sector_t iblock, 1284 struct buffer_head *bh_result, int create, int flag, 1285 pgoff_t *next_pgofs) 1286 { 1287 return __get_data_block(inode, iblock, bh_result, create, 1288 flag, next_pgofs, 1289 NO_CHECK_TYPE); 1290 } 1291 1292 static int get_data_block_dio(struct inode *inode, sector_t iblock, 1293 struct buffer_head *bh_result, int create) 1294 { 1295 return __get_data_block(inode, iblock, bh_result, create, 1296 F2FS_GET_BLOCK_DIO, NULL, 1297 f2fs_rw_hint_to_seg_type( 1298 inode->i_write_hint)); 1299 } 1300 1301 static int get_data_block_bmap(struct inode *inode, sector_t iblock, 1302 struct buffer_head *bh_result, int create) 1303 { 1304 /* Block number less than F2FS MAX BLOCKS */ 1305 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks)) 1306 return -EFBIG; 1307 1308 return __get_data_block(inode, iblock, bh_result, create, 1309 F2FS_GET_BLOCK_BMAP, NULL, 1310 NO_CHECK_TYPE); 1311 } 1312 1313 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset) 1314 { 1315 return (offset >> inode->i_blkbits); 1316 } 1317 1318 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk) 1319 { 1320 return (blk << inode->i_blkbits); 1321 } 1322 1323 static int f2fs_xattr_fiemap(struct inode *inode, 1324 struct fiemap_extent_info *fieinfo) 1325 { 1326 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1327 struct page *page; 1328 struct node_info ni; 1329 __u64 phys = 0, len; 1330 __u32 flags; 1331 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1332 int err = 0; 1333 1334 if (f2fs_has_inline_xattr(inode)) { 1335 int offset; 1336 1337 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1338 inode->i_ino, false); 1339 if (!page) 1340 return -ENOMEM; 1341 1342 err = f2fs_get_node_info(sbi, inode->i_ino, &ni); 1343 if (err) { 1344 f2fs_put_page(page, 1); 1345 return err; 1346 } 1347 1348 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1349 offset = offsetof(struct f2fs_inode, i_addr) + 1350 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1351 get_inline_xattr_addrs(inode)); 1352 1353 phys += offset; 1354 len = inline_xattr_size(inode); 1355 1356 f2fs_put_page(page, 1); 1357 1358 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1359 1360 if (!xnid) 1361 flags |= FIEMAP_EXTENT_LAST; 1362 1363 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1364 if (err || err == 1) 1365 return err; 1366 } 1367 1368 if (xnid) { 1369 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1370 if (!page) 1371 return -ENOMEM; 1372 1373 err = f2fs_get_node_info(sbi, xnid, &ni); 1374 if (err) { 1375 f2fs_put_page(page, 1); 1376 return err; 1377 } 1378 1379 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1380 len = inode->i_sb->s_blocksize; 1381 1382 f2fs_put_page(page, 1); 1383 1384 flags = FIEMAP_EXTENT_LAST; 1385 } 1386 1387 if (phys) 1388 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1389 1390 return (err < 0 ? err : 0); 1391 } 1392 1393 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1394 u64 start, u64 len) 1395 { 1396 struct buffer_head map_bh; 1397 sector_t start_blk, last_blk; 1398 pgoff_t next_pgofs; 1399 u64 logical = 0, phys = 0, size = 0; 1400 u32 flags = 0; 1401 int ret = 0; 1402 1403 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { 1404 ret = f2fs_precache_extents(inode); 1405 if (ret) 1406 return ret; 1407 } 1408 1409 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR); 1410 if (ret) 1411 return ret; 1412 1413 inode_lock(inode); 1414 1415 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1416 ret = f2fs_xattr_fiemap(inode, fieinfo); 1417 goto out; 1418 } 1419 1420 if (f2fs_has_inline_data(inode)) { 1421 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1422 if (ret != -EAGAIN) 1423 goto out; 1424 } 1425 1426 if (logical_to_blk(inode, len) == 0) 1427 len = blk_to_logical(inode, 1); 1428 1429 start_blk = logical_to_blk(inode, start); 1430 last_blk = logical_to_blk(inode, start + len - 1); 1431 1432 next: 1433 memset(&map_bh, 0, sizeof(struct buffer_head)); 1434 map_bh.b_size = len; 1435 1436 ret = get_data_block(inode, start_blk, &map_bh, 0, 1437 F2FS_GET_BLOCK_FIEMAP, &next_pgofs); 1438 if (ret) 1439 goto out; 1440 1441 /* HOLE */ 1442 if (!buffer_mapped(&map_bh)) { 1443 start_blk = next_pgofs; 1444 1445 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode, 1446 F2FS_I_SB(inode)->max_file_blocks)) 1447 goto prep_next; 1448 1449 flags |= FIEMAP_EXTENT_LAST; 1450 } 1451 1452 if (size) { 1453 if (f2fs_encrypted_inode(inode)) 1454 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1455 1456 ret = fiemap_fill_next_extent(fieinfo, logical, 1457 phys, size, flags); 1458 } 1459 1460 if (start_blk > last_blk || ret) 1461 goto out; 1462 1463 logical = blk_to_logical(inode, start_blk); 1464 phys = blk_to_logical(inode, map_bh.b_blocknr); 1465 size = map_bh.b_size; 1466 flags = 0; 1467 if (buffer_unwritten(&map_bh)) 1468 flags = FIEMAP_EXTENT_UNWRITTEN; 1469 1470 start_blk += logical_to_blk(inode, size); 1471 1472 prep_next: 1473 cond_resched(); 1474 if (fatal_signal_pending(current)) 1475 ret = -EINTR; 1476 else 1477 goto next; 1478 out: 1479 if (ret == 1) 1480 ret = 0; 1481 1482 inode_unlock(inode); 1483 return ret; 1484 } 1485 1486 /* 1487 * This function was originally taken from fs/mpage.c, and customized for f2fs. 1488 * Major change was from block_size == page_size in f2fs by default. 1489 * 1490 * Note that the aops->readpages() function is ONLY used for read-ahead. If 1491 * this function ever deviates from doing just read-ahead, it should either 1492 * use ->readpage() or do the necessary surgery to decouple ->readpages() 1493 * from read-ahead. 1494 */ 1495 static int f2fs_mpage_readpages(struct address_space *mapping, 1496 struct list_head *pages, struct page *page, 1497 unsigned nr_pages, bool is_readahead) 1498 { 1499 struct bio *bio = NULL; 1500 sector_t last_block_in_bio = 0; 1501 struct inode *inode = mapping->host; 1502 const unsigned blkbits = inode->i_blkbits; 1503 const unsigned blocksize = 1 << blkbits; 1504 sector_t block_in_file; 1505 sector_t last_block; 1506 sector_t last_block_in_file; 1507 sector_t block_nr; 1508 struct f2fs_map_blocks map; 1509 1510 map.m_pblk = 0; 1511 map.m_lblk = 0; 1512 map.m_len = 0; 1513 map.m_flags = 0; 1514 map.m_next_pgofs = NULL; 1515 map.m_next_extent = NULL; 1516 map.m_seg_type = NO_CHECK_TYPE; 1517 1518 for (; nr_pages; nr_pages--) { 1519 if (pages) { 1520 page = list_last_entry(pages, struct page, lru); 1521 1522 prefetchw(&page->flags); 1523 list_del(&page->lru); 1524 if (add_to_page_cache_lru(page, mapping, 1525 page->index, 1526 readahead_gfp_mask(mapping))) 1527 goto next_page; 1528 } 1529 1530 block_in_file = (sector_t)page->index; 1531 last_block = block_in_file + nr_pages; 1532 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> 1533 blkbits; 1534 if (last_block > last_block_in_file) 1535 last_block = last_block_in_file; 1536 1537 /* 1538 * Map blocks using the previous result first. 1539 */ 1540 if ((map.m_flags & F2FS_MAP_MAPPED) && 1541 block_in_file > map.m_lblk && 1542 block_in_file < (map.m_lblk + map.m_len)) 1543 goto got_it; 1544 1545 /* 1546 * Then do more f2fs_map_blocks() calls until we are 1547 * done with this page. 1548 */ 1549 map.m_flags = 0; 1550 1551 if (block_in_file < last_block) { 1552 map.m_lblk = block_in_file; 1553 map.m_len = last_block - block_in_file; 1554 1555 if (f2fs_map_blocks(inode, &map, 0, 1556 F2FS_GET_BLOCK_DEFAULT)) 1557 goto set_error_page; 1558 } 1559 got_it: 1560 if ((map.m_flags & F2FS_MAP_MAPPED)) { 1561 block_nr = map.m_pblk + block_in_file - map.m_lblk; 1562 SetPageMappedToDisk(page); 1563 1564 if (!PageUptodate(page) && !cleancache_get_page(page)) { 1565 SetPageUptodate(page); 1566 goto confused; 1567 } 1568 1569 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr, 1570 DATA_GENERIC)) 1571 goto set_error_page; 1572 } else { 1573 zero_user_segment(page, 0, PAGE_SIZE); 1574 if (!PageUptodate(page)) 1575 SetPageUptodate(page); 1576 unlock_page(page); 1577 goto next_page; 1578 } 1579 1580 /* 1581 * This page will go to BIO. Do we need to send this 1582 * BIO off first? 1583 */ 1584 if (bio && (last_block_in_bio != block_nr - 1 || 1585 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) { 1586 submit_and_realloc: 1587 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1588 bio = NULL; 1589 } 1590 if (bio == NULL) { 1591 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages, 1592 is_readahead ? REQ_RAHEAD : 0); 1593 if (IS_ERR(bio)) { 1594 bio = NULL; 1595 goto set_error_page; 1596 } 1597 } 1598 1599 /* 1600 * If the page is under writeback, we need to wait for 1601 * its completion to see the correct decrypted data. 1602 */ 1603 f2fs_wait_on_block_writeback(inode, block_nr); 1604 1605 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 1606 goto submit_and_realloc; 1607 1608 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 1609 ClearPageError(page); 1610 last_block_in_bio = block_nr; 1611 goto next_page; 1612 set_error_page: 1613 SetPageError(page); 1614 zero_user_segment(page, 0, PAGE_SIZE); 1615 unlock_page(page); 1616 goto next_page; 1617 confused: 1618 if (bio) { 1619 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1620 bio = NULL; 1621 } 1622 unlock_page(page); 1623 next_page: 1624 if (pages) 1625 put_page(page); 1626 } 1627 BUG_ON(pages && !list_empty(pages)); 1628 if (bio) 1629 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1630 return 0; 1631 } 1632 1633 static int f2fs_read_data_page(struct file *file, struct page *page) 1634 { 1635 struct inode *inode = page->mapping->host; 1636 int ret = -EAGAIN; 1637 1638 trace_f2fs_readpage(page, DATA); 1639 1640 /* If the file has inline data, try to read it directly */ 1641 if (f2fs_has_inline_data(inode)) 1642 ret = f2fs_read_inline_data(inode, page); 1643 if (ret == -EAGAIN) 1644 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false); 1645 return ret; 1646 } 1647 1648 static int f2fs_read_data_pages(struct file *file, 1649 struct address_space *mapping, 1650 struct list_head *pages, unsigned nr_pages) 1651 { 1652 struct inode *inode = mapping->host; 1653 struct page *page = list_last_entry(pages, struct page, lru); 1654 1655 trace_f2fs_readpages(inode, page, nr_pages); 1656 1657 /* If the file has inline data, skip readpages */ 1658 if (f2fs_has_inline_data(inode)) 1659 return 0; 1660 1661 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true); 1662 } 1663 1664 static int encrypt_one_page(struct f2fs_io_info *fio) 1665 { 1666 struct inode *inode = fio->page->mapping->host; 1667 struct page *mpage; 1668 gfp_t gfp_flags = GFP_NOFS; 1669 1670 if (!f2fs_encrypted_file(inode)) 1671 return 0; 1672 1673 /* wait for GCed page writeback via META_MAPPING */ 1674 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr); 1675 1676 retry_encrypt: 1677 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page, 1678 PAGE_SIZE, 0, fio->page->index, gfp_flags); 1679 if (IS_ERR(fio->encrypted_page)) { 1680 /* flush pending IOs and wait for a while in the ENOMEM case */ 1681 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 1682 f2fs_flush_merged_writes(fio->sbi); 1683 congestion_wait(BLK_RW_ASYNC, HZ/50); 1684 gfp_flags |= __GFP_NOFAIL; 1685 goto retry_encrypt; 1686 } 1687 return PTR_ERR(fio->encrypted_page); 1688 } 1689 1690 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr); 1691 if (mpage) { 1692 if (PageUptodate(mpage)) 1693 memcpy(page_address(mpage), 1694 page_address(fio->encrypted_page), PAGE_SIZE); 1695 f2fs_put_page(mpage, 1); 1696 } 1697 return 0; 1698 } 1699 1700 static inline bool check_inplace_update_policy(struct inode *inode, 1701 struct f2fs_io_info *fio) 1702 { 1703 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1704 unsigned int policy = SM_I(sbi)->ipu_policy; 1705 1706 if (policy & (0x1 << F2FS_IPU_FORCE)) 1707 return true; 1708 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi)) 1709 return true; 1710 if (policy & (0x1 << F2FS_IPU_UTIL) && 1711 utilization(sbi) > SM_I(sbi)->min_ipu_util) 1712 return true; 1713 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) && 1714 utilization(sbi) > SM_I(sbi)->min_ipu_util) 1715 return true; 1716 1717 /* 1718 * IPU for rewrite async pages 1719 */ 1720 if (policy & (0x1 << F2FS_IPU_ASYNC) && 1721 fio && fio->op == REQ_OP_WRITE && 1722 !(fio->op_flags & REQ_SYNC) && 1723 !f2fs_encrypted_inode(inode)) 1724 return true; 1725 1726 /* this is only set during fdatasync */ 1727 if (policy & (0x1 << F2FS_IPU_FSYNC) && 1728 is_inode_flag_set(inode, FI_NEED_IPU)) 1729 return true; 1730 1731 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1732 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 1733 return true; 1734 1735 return false; 1736 } 1737 1738 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) 1739 { 1740 if (f2fs_is_pinned_file(inode)) 1741 return true; 1742 1743 /* if this is cold file, we should overwrite to avoid fragmentation */ 1744 if (file_is_cold(inode)) 1745 return true; 1746 1747 return check_inplace_update_policy(inode, fio); 1748 } 1749 1750 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) 1751 { 1752 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1753 1754 if (test_opt(sbi, LFS)) 1755 return true; 1756 if (S_ISDIR(inode->i_mode)) 1757 return true; 1758 if (f2fs_is_atomic_file(inode)) 1759 return true; 1760 if (fio) { 1761 if (is_cold_data(fio->page)) 1762 return true; 1763 if (IS_ATOMIC_WRITTEN_PAGE(fio->page)) 1764 return true; 1765 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 1766 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 1767 return true; 1768 } 1769 return false; 1770 } 1771 1772 static inline bool need_inplace_update(struct f2fs_io_info *fio) 1773 { 1774 struct inode *inode = fio->page->mapping->host; 1775 1776 if (f2fs_should_update_outplace(inode, fio)) 1777 return false; 1778 1779 return f2fs_should_update_inplace(inode, fio); 1780 } 1781 1782 int f2fs_do_write_data_page(struct f2fs_io_info *fio) 1783 { 1784 struct page *page = fio->page; 1785 struct inode *inode = page->mapping->host; 1786 struct dnode_of_data dn; 1787 struct extent_info ei = {0,0,0}; 1788 struct node_info ni; 1789 bool ipu_force = false; 1790 int err = 0; 1791 1792 set_new_dnode(&dn, inode, NULL, NULL, 0); 1793 if (need_inplace_update(fio) && 1794 f2fs_lookup_extent_cache(inode, page->index, &ei)) { 1795 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 1796 1797 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 1798 DATA_GENERIC)) 1799 return -EFAULT; 1800 1801 ipu_force = true; 1802 fio->need_lock = LOCK_DONE; 1803 goto got_it; 1804 } 1805 1806 /* Deadlock due to between page->lock and f2fs_lock_op */ 1807 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 1808 return -EAGAIN; 1809 1810 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 1811 if (err) 1812 goto out; 1813 1814 fio->old_blkaddr = dn.data_blkaddr; 1815 1816 /* This page is already truncated */ 1817 if (fio->old_blkaddr == NULL_ADDR) { 1818 ClearPageUptodate(page); 1819 clear_cold_data(page); 1820 goto out_writepage; 1821 } 1822 got_it: 1823 if (__is_valid_data_blkaddr(fio->old_blkaddr) && 1824 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 1825 DATA_GENERIC)) { 1826 err = -EFAULT; 1827 goto out_writepage; 1828 } 1829 /* 1830 * If current allocation needs SSR, 1831 * it had better in-place writes for updated data. 1832 */ 1833 if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) && 1834 need_inplace_update(fio))) { 1835 err = encrypt_one_page(fio); 1836 if (err) 1837 goto out_writepage; 1838 1839 set_page_writeback(page); 1840 ClearPageError(page); 1841 f2fs_put_dnode(&dn); 1842 if (fio->need_lock == LOCK_REQ) 1843 f2fs_unlock_op(fio->sbi); 1844 err = f2fs_inplace_write_data(fio); 1845 trace_f2fs_do_write_data_page(fio->page, IPU); 1846 set_inode_flag(inode, FI_UPDATE_WRITE); 1847 return err; 1848 } 1849 1850 if (fio->need_lock == LOCK_RETRY) { 1851 if (!f2fs_trylock_op(fio->sbi)) { 1852 err = -EAGAIN; 1853 goto out_writepage; 1854 } 1855 fio->need_lock = LOCK_REQ; 1856 } 1857 1858 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni); 1859 if (err) 1860 goto out_writepage; 1861 1862 fio->version = ni.version; 1863 1864 err = encrypt_one_page(fio); 1865 if (err) 1866 goto out_writepage; 1867 1868 set_page_writeback(page); 1869 ClearPageError(page); 1870 1871 /* LFS mode write path */ 1872 f2fs_outplace_write_data(&dn, fio); 1873 trace_f2fs_do_write_data_page(page, OPU); 1874 set_inode_flag(inode, FI_APPEND_WRITE); 1875 if (page->index == 0) 1876 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 1877 out_writepage: 1878 f2fs_put_dnode(&dn); 1879 out: 1880 if (fio->need_lock == LOCK_REQ) 1881 f2fs_unlock_op(fio->sbi); 1882 return err; 1883 } 1884 1885 static int __write_data_page(struct page *page, bool *submitted, 1886 struct writeback_control *wbc, 1887 enum iostat_type io_type) 1888 { 1889 struct inode *inode = page->mapping->host; 1890 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1891 loff_t i_size = i_size_read(inode); 1892 const pgoff_t end_index = ((unsigned long long) i_size) 1893 >> PAGE_SHIFT; 1894 loff_t psize = (page->index + 1) << PAGE_SHIFT; 1895 unsigned offset = 0; 1896 bool need_balance_fs = false; 1897 int err = 0; 1898 struct f2fs_io_info fio = { 1899 .sbi = sbi, 1900 .ino = inode->i_ino, 1901 .type = DATA, 1902 .op = REQ_OP_WRITE, 1903 .op_flags = wbc_to_write_flags(wbc), 1904 .old_blkaddr = NULL_ADDR, 1905 .page = page, 1906 .encrypted_page = NULL, 1907 .submitted = false, 1908 .need_lock = LOCK_RETRY, 1909 .io_type = io_type, 1910 .io_wbc = wbc, 1911 }; 1912 1913 trace_f2fs_writepage(page, DATA); 1914 1915 /* we should bypass data pages to proceed the kworkder jobs */ 1916 if (unlikely(f2fs_cp_error(sbi))) { 1917 mapping_set_error(page->mapping, -EIO); 1918 /* 1919 * don't drop any dirty dentry pages for keeping lastest 1920 * directory structure. 1921 */ 1922 if (S_ISDIR(inode->i_mode)) 1923 goto redirty_out; 1924 goto out; 1925 } 1926 1927 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1928 goto redirty_out; 1929 1930 if (page->index < end_index) 1931 goto write; 1932 1933 /* 1934 * If the offset is out-of-range of file size, 1935 * this page does not have to be written to disk. 1936 */ 1937 offset = i_size & (PAGE_SIZE - 1); 1938 if ((page->index >= end_index + 1) || !offset) 1939 goto out; 1940 1941 zero_user_segment(page, offset, PAGE_SIZE); 1942 write: 1943 if (f2fs_is_drop_cache(inode)) 1944 goto out; 1945 /* we should not write 0'th page having journal header */ 1946 if (f2fs_is_volatile_file(inode) && (!page->index || 1947 (!wbc->for_reclaim && 1948 f2fs_available_free_memory(sbi, BASE_CHECK)))) 1949 goto redirty_out; 1950 1951 /* Dentry blocks are controlled by checkpoint */ 1952 if (S_ISDIR(inode->i_mode)) { 1953 fio.need_lock = LOCK_DONE; 1954 err = f2fs_do_write_data_page(&fio); 1955 goto done; 1956 } 1957 1958 if (!wbc->for_reclaim) 1959 need_balance_fs = true; 1960 else if (has_not_enough_free_secs(sbi, 0, 0)) 1961 goto redirty_out; 1962 else 1963 set_inode_flag(inode, FI_HOT_DATA); 1964 1965 err = -EAGAIN; 1966 if (f2fs_has_inline_data(inode)) { 1967 err = f2fs_write_inline_data(inode, page); 1968 if (!err) 1969 goto out; 1970 } 1971 1972 if (err == -EAGAIN) { 1973 err = f2fs_do_write_data_page(&fio); 1974 if (err == -EAGAIN) { 1975 fio.need_lock = LOCK_REQ; 1976 err = f2fs_do_write_data_page(&fio); 1977 } 1978 } 1979 1980 if (err) { 1981 file_set_keep_isize(inode); 1982 } else { 1983 down_write(&F2FS_I(inode)->i_sem); 1984 if (F2FS_I(inode)->last_disk_size < psize) 1985 F2FS_I(inode)->last_disk_size = psize; 1986 up_write(&F2FS_I(inode)->i_sem); 1987 } 1988 1989 done: 1990 if (err && err != -ENOENT) 1991 goto redirty_out; 1992 1993 out: 1994 inode_dec_dirty_pages(inode); 1995 if (err) { 1996 ClearPageUptodate(page); 1997 clear_cold_data(page); 1998 } 1999 2000 if (wbc->for_reclaim) { 2001 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA); 2002 clear_inode_flag(inode, FI_HOT_DATA); 2003 f2fs_remove_dirty_inode(inode); 2004 submitted = NULL; 2005 } 2006 2007 unlock_page(page); 2008 if (!S_ISDIR(inode->i_mode)) 2009 f2fs_balance_fs(sbi, need_balance_fs); 2010 2011 if (unlikely(f2fs_cp_error(sbi))) { 2012 f2fs_submit_merged_write(sbi, DATA); 2013 submitted = NULL; 2014 } 2015 2016 if (submitted) 2017 *submitted = fio.submitted; 2018 2019 return 0; 2020 2021 redirty_out: 2022 redirty_page_for_writepage(wbc, page); 2023 /* 2024 * pageout() in MM traslates EAGAIN, so calls handle_write_error() 2025 * -> mapping_set_error() -> set_bit(AS_EIO, ...). 2026 * file_write_and_wait_range() will see EIO error, which is critical 2027 * to return value of fsync() followed by atomic_write failure to user. 2028 */ 2029 if (!err || wbc->for_reclaim) 2030 return AOP_WRITEPAGE_ACTIVATE; 2031 unlock_page(page); 2032 return err; 2033 } 2034 2035 static int f2fs_write_data_page(struct page *page, 2036 struct writeback_control *wbc) 2037 { 2038 return __write_data_page(page, NULL, wbc, FS_DATA_IO); 2039 } 2040 2041 /* 2042 * This function was copied from write_cche_pages from mm/page-writeback.c. 2043 * The major change is making write step of cold data page separately from 2044 * warm/hot data page. 2045 */ 2046 static int f2fs_write_cache_pages(struct address_space *mapping, 2047 struct writeback_control *wbc, 2048 enum iostat_type io_type) 2049 { 2050 int ret = 0; 2051 int done = 0; 2052 struct pagevec pvec; 2053 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2054 int nr_pages; 2055 pgoff_t uninitialized_var(writeback_index); 2056 pgoff_t index; 2057 pgoff_t end; /* Inclusive */ 2058 pgoff_t done_index; 2059 int cycled; 2060 int range_whole = 0; 2061 int tag; 2062 int nwritten = 0; 2063 2064 pagevec_init(&pvec); 2065 2066 if (get_dirty_pages(mapping->host) <= 2067 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 2068 set_inode_flag(mapping->host, FI_HOT_DATA); 2069 else 2070 clear_inode_flag(mapping->host, FI_HOT_DATA); 2071 2072 if (wbc->range_cyclic) { 2073 writeback_index = mapping->writeback_index; /* prev offset */ 2074 index = writeback_index; 2075 if (index == 0) 2076 cycled = 1; 2077 else 2078 cycled = 0; 2079 end = -1; 2080 } else { 2081 index = wbc->range_start >> PAGE_SHIFT; 2082 end = wbc->range_end >> PAGE_SHIFT; 2083 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2084 range_whole = 1; 2085 cycled = 1; /* ignore range_cyclic tests */ 2086 } 2087 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2088 tag = PAGECACHE_TAG_TOWRITE; 2089 else 2090 tag = PAGECACHE_TAG_DIRTY; 2091 retry: 2092 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2093 tag_pages_for_writeback(mapping, index, end); 2094 done_index = index; 2095 while (!done && (index <= end)) { 2096 int i; 2097 2098 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 2099 tag); 2100 if (nr_pages == 0) 2101 break; 2102 2103 for (i = 0; i < nr_pages; i++) { 2104 struct page *page = pvec.pages[i]; 2105 bool submitted = false; 2106 2107 /* give a priority to WB_SYNC threads */ 2108 if (atomic_read(&sbi->wb_sync_req[DATA]) && 2109 wbc->sync_mode == WB_SYNC_NONE) { 2110 done = 1; 2111 break; 2112 } 2113 2114 done_index = page->index; 2115 retry_write: 2116 lock_page(page); 2117 2118 if (unlikely(page->mapping != mapping)) { 2119 continue_unlock: 2120 unlock_page(page); 2121 continue; 2122 } 2123 2124 if (!PageDirty(page)) { 2125 /* someone wrote it for us */ 2126 goto continue_unlock; 2127 } 2128 2129 if (PageWriteback(page)) { 2130 if (wbc->sync_mode != WB_SYNC_NONE) 2131 f2fs_wait_on_page_writeback(page, 2132 DATA, true); 2133 else 2134 goto continue_unlock; 2135 } 2136 2137 BUG_ON(PageWriteback(page)); 2138 if (!clear_page_dirty_for_io(page)) 2139 goto continue_unlock; 2140 2141 ret = __write_data_page(page, &submitted, wbc, io_type); 2142 if (unlikely(ret)) { 2143 /* 2144 * keep nr_to_write, since vfs uses this to 2145 * get # of written pages. 2146 */ 2147 if (ret == AOP_WRITEPAGE_ACTIVATE) { 2148 unlock_page(page); 2149 ret = 0; 2150 continue; 2151 } else if (ret == -EAGAIN) { 2152 ret = 0; 2153 if (wbc->sync_mode == WB_SYNC_ALL) { 2154 cond_resched(); 2155 congestion_wait(BLK_RW_ASYNC, 2156 HZ/50); 2157 goto retry_write; 2158 } 2159 continue; 2160 } 2161 done_index = page->index + 1; 2162 done = 1; 2163 break; 2164 } else if (submitted) { 2165 nwritten++; 2166 } 2167 2168 if (--wbc->nr_to_write <= 0 && 2169 wbc->sync_mode == WB_SYNC_NONE) { 2170 done = 1; 2171 break; 2172 } 2173 } 2174 pagevec_release(&pvec); 2175 cond_resched(); 2176 } 2177 2178 if (!cycled && !done) { 2179 cycled = 1; 2180 index = 0; 2181 end = writeback_index - 1; 2182 goto retry; 2183 } 2184 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 2185 mapping->writeback_index = done_index; 2186 2187 if (nwritten) 2188 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 2189 NULL, 0, DATA); 2190 2191 return ret; 2192 } 2193 2194 static inline bool __should_serialize_io(struct inode *inode, 2195 struct writeback_control *wbc) 2196 { 2197 if (!S_ISREG(inode->i_mode)) 2198 return false; 2199 if (wbc->sync_mode != WB_SYNC_ALL) 2200 return true; 2201 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks) 2202 return true; 2203 return false; 2204 } 2205 2206 static int __f2fs_write_data_pages(struct address_space *mapping, 2207 struct writeback_control *wbc, 2208 enum iostat_type io_type) 2209 { 2210 struct inode *inode = mapping->host; 2211 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2212 struct blk_plug plug; 2213 int ret; 2214 bool locked = false; 2215 2216 /* deal with chardevs and other special file */ 2217 if (!mapping->a_ops->writepage) 2218 return 0; 2219 2220 /* skip writing if there is no dirty page in this inode */ 2221 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 2222 return 0; 2223 2224 /* during POR, we don't need to trigger writepage at all. */ 2225 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2226 goto skip_write; 2227 2228 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE && 2229 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 2230 f2fs_available_free_memory(sbi, DIRTY_DENTS)) 2231 goto skip_write; 2232 2233 /* skip writing during file defragment */ 2234 if (is_inode_flag_set(inode, FI_DO_DEFRAG)) 2235 goto skip_write; 2236 2237 trace_f2fs_writepages(mapping->host, wbc, DATA); 2238 2239 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 2240 if (wbc->sync_mode == WB_SYNC_ALL) 2241 atomic_inc(&sbi->wb_sync_req[DATA]); 2242 else if (atomic_read(&sbi->wb_sync_req[DATA])) 2243 goto skip_write; 2244 2245 if (__should_serialize_io(inode, wbc)) { 2246 mutex_lock(&sbi->writepages); 2247 locked = true; 2248 } 2249 2250 blk_start_plug(&plug); 2251 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 2252 blk_finish_plug(&plug); 2253 2254 if (locked) 2255 mutex_unlock(&sbi->writepages); 2256 2257 if (wbc->sync_mode == WB_SYNC_ALL) 2258 atomic_dec(&sbi->wb_sync_req[DATA]); 2259 /* 2260 * if some pages were truncated, we cannot guarantee its mapping->host 2261 * to detect pending bios. 2262 */ 2263 2264 f2fs_remove_dirty_inode(inode); 2265 return ret; 2266 2267 skip_write: 2268 wbc->pages_skipped += get_dirty_pages(inode); 2269 trace_f2fs_writepages(mapping->host, wbc, DATA); 2270 return 0; 2271 } 2272 2273 static int f2fs_write_data_pages(struct address_space *mapping, 2274 struct writeback_control *wbc) 2275 { 2276 struct inode *inode = mapping->host; 2277 2278 return __f2fs_write_data_pages(mapping, wbc, 2279 F2FS_I(inode)->cp_task == current ? 2280 FS_CP_DATA_IO : FS_DATA_IO); 2281 } 2282 2283 static void f2fs_write_failed(struct address_space *mapping, loff_t to) 2284 { 2285 struct inode *inode = mapping->host; 2286 loff_t i_size = i_size_read(inode); 2287 2288 if (to > i_size) { 2289 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 2290 down_write(&F2FS_I(inode)->i_mmap_sem); 2291 2292 truncate_pagecache(inode, i_size); 2293 f2fs_truncate_blocks(inode, i_size, true); 2294 2295 up_write(&F2FS_I(inode)->i_mmap_sem); 2296 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 2297 } 2298 } 2299 2300 static int prepare_write_begin(struct f2fs_sb_info *sbi, 2301 struct page *page, loff_t pos, unsigned len, 2302 block_t *blk_addr, bool *node_changed) 2303 { 2304 struct inode *inode = page->mapping->host; 2305 pgoff_t index = page->index; 2306 struct dnode_of_data dn; 2307 struct page *ipage; 2308 bool locked = false; 2309 struct extent_info ei = {0,0,0}; 2310 int err = 0; 2311 2312 /* 2313 * we already allocated all the blocks, so we don't need to get 2314 * the block addresses when there is no need to fill the page. 2315 */ 2316 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE && 2317 !is_inode_flag_set(inode, FI_NO_PREALLOC)) 2318 return 0; 2319 2320 if (f2fs_has_inline_data(inode) || 2321 (pos & PAGE_MASK) >= i_size_read(inode)) { 2322 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true); 2323 locked = true; 2324 } 2325 restart: 2326 /* check inline_data */ 2327 ipage = f2fs_get_node_page(sbi, inode->i_ino); 2328 if (IS_ERR(ipage)) { 2329 err = PTR_ERR(ipage); 2330 goto unlock_out; 2331 } 2332 2333 set_new_dnode(&dn, inode, ipage, ipage, 0); 2334 2335 if (f2fs_has_inline_data(inode)) { 2336 if (pos + len <= MAX_INLINE_DATA(inode)) { 2337 f2fs_do_read_inline_data(page, ipage); 2338 set_inode_flag(inode, FI_DATA_EXIST); 2339 if (inode->i_nlink) 2340 set_inline_node(ipage); 2341 } else { 2342 err = f2fs_convert_inline_page(&dn, page); 2343 if (err) 2344 goto out; 2345 if (dn.data_blkaddr == NULL_ADDR) 2346 err = f2fs_get_block(&dn, index); 2347 } 2348 } else if (locked) { 2349 err = f2fs_get_block(&dn, index); 2350 } else { 2351 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 2352 dn.data_blkaddr = ei.blk + index - ei.fofs; 2353 } else { 2354 /* hole case */ 2355 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 2356 if (err || dn.data_blkaddr == NULL_ADDR) { 2357 f2fs_put_dnode(&dn); 2358 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 2359 true); 2360 locked = true; 2361 goto restart; 2362 } 2363 } 2364 } 2365 2366 /* convert_inline_page can make node_changed */ 2367 *blk_addr = dn.data_blkaddr; 2368 *node_changed = dn.node_changed; 2369 out: 2370 f2fs_put_dnode(&dn); 2371 unlock_out: 2372 if (locked) 2373 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false); 2374 return err; 2375 } 2376 2377 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 2378 loff_t pos, unsigned len, unsigned flags, 2379 struct page **pagep, void **fsdata) 2380 { 2381 struct inode *inode = mapping->host; 2382 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2383 struct page *page = NULL; 2384 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 2385 bool need_balance = false, drop_atomic = false; 2386 block_t blkaddr = NULL_ADDR; 2387 int err = 0; 2388 2389 trace_f2fs_write_begin(inode, pos, len, flags); 2390 2391 err = f2fs_is_checkpoint_ready(sbi); 2392 if (err) 2393 goto fail; 2394 2395 if ((f2fs_is_atomic_file(inode) && 2396 !f2fs_available_free_memory(sbi, INMEM_PAGES)) || 2397 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) { 2398 err = -ENOMEM; 2399 drop_atomic = true; 2400 goto fail; 2401 } 2402 2403 /* 2404 * We should check this at this moment to avoid deadlock on inode page 2405 * and #0 page. The locking rule for inline_data conversion should be: 2406 * lock_page(page #0) -> lock_page(inode_page) 2407 */ 2408 if (index != 0) { 2409 err = f2fs_convert_inline_inode(inode); 2410 if (err) 2411 goto fail; 2412 } 2413 repeat: 2414 /* 2415 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 2416 * wait_for_stable_page. Will wait that below with our IO control. 2417 */ 2418 page = f2fs_pagecache_get_page(mapping, index, 2419 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 2420 if (!page) { 2421 err = -ENOMEM; 2422 goto fail; 2423 } 2424 2425 *pagep = page; 2426 2427 err = prepare_write_begin(sbi, page, pos, len, 2428 &blkaddr, &need_balance); 2429 if (err) 2430 goto fail; 2431 2432 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) { 2433 unlock_page(page); 2434 f2fs_balance_fs(sbi, true); 2435 lock_page(page); 2436 if (page->mapping != mapping) { 2437 /* The page got truncated from under us */ 2438 f2fs_put_page(page, 1); 2439 goto repeat; 2440 } 2441 } 2442 2443 f2fs_wait_on_page_writeback(page, DATA, false); 2444 2445 if (len == PAGE_SIZE || PageUptodate(page)) 2446 return 0; 2447 2448 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) { 2449 zero_user_segment(page, len, PAGE_SIZE); 2450 return 0; 2451 } 2452 2453 if (blkaddr == NEW_ADDR) { 2454 zero_user_segment(page, 0, PAGE_SIZE); 2455 SetPageUptodate(page); 2456 } else { 2457 err = f2fs_submit_page_read(inode, page, blkaddr); 2458 if (err) 2459 goto fail; 2460 2461 lock_page(page); 2462 if (unlikely(page->mapping != mapping)) { 2463 f2fs_put_page(page, 1); 2464 goto repeat; 2465 } 2466 if (unlikely(!PageUptodate(page))) { 2467 err = -EIO; 2468 goto fail; 2469 } 2470 } 2471 return 0; 2472 2473 fail: 2474 f2fs_put_page(page, 1); 2475 f2fs_write_failed(mapping, pos + len); 2476 if (drop_atomic) 2477 f2fs_drop_inmem_pages_all(sbi, false); 2478 return err; 2479 } 2480 2481 static int f2fs_write_end(struct file *file, 2482 struct address_space *mapping, 2483 loff_t pos, unsigned len, unsigned copied, 2484 struct page *page, void *fsdata) 2485 { 2486 struct inode *inode = page->mapping->host; 2487 2488 trace_f2fs_write_end(inode, pos, len, copied); 2489 2490 /* 2491 * This should be come from len == PAGE_SIZE, and we expect copied 2492 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 2493 * let generic_perform_write() try to copy data again through copied=0. 2494 */ 2495 if (!PageUptodate(page)) { 2496 if (unlikely(copied != len)) 2497 copied = 0; 2498 else 2499 SetPageUptodate(page); 2500 } 2501 if (!copied) 2502 goto unlock_out; 2503 2504 set_page_dirty(page); 2505 2506 if (pos + copied > i_size_read(inode)) 2507 f2fs_i_size_write(inode, pos + copied); 2508 unlock_out: 2509 f2fs_put_page(page, 1); 2510 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 2511 return copied; 2512 } 2513 2514 static int check_direct_IO(struct inode *inode, struct iov_iter *iter, 2515 loff_t offset) 2516 { 2517 unsigned i_blkbits = READ_ONCE(inode->i_blkbits); 2518 unsigned blkbits = i_blkbits; 2519 unsigned blocksize_mask = (1 << blkbits) - 1; 2520 unsigned long align = offset | iov_iter_alignment(iter); 2521 struct block_device *bdev = inode->i_sb->s_bdev; 2522 2523 if (align & blocksize_mask) { 2524 if (bdev) 2525 blkbits = blksize_bits(bdev_logical_block_size(bdev)); 2526 blocksize_mask = (1 << blkbits) - 1; 2527 if (align & blocksize_mask) 2528 return -EINVAL; 2529 return 1; 2530 } 2531 return 0; 2532 } 2533 2534 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 2535 { 2536 struct address_space *mapping = iocb->ki_filp->f_mapping; 2537 struct inode *inode = mapping->host; 2538 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2539 struct f2fs_inode_info *fi = F2FS_I(inode); 2540 size_t count = iov_iter_count(iter); 2541 loff_t offset = iocb->ki_pos; 2542 int rw = iov_iter_rw(iter); 2543 int err; 2544 enum rw_hint hint = iocb->ki_hint; 2545 int whint_mode = F2FS_OPTION(sbi).whint_mode; 2546 bool do_opu; 2547 2548 err = check_direct_IO(inode, iter, offset); 2549 if (err) 2550 return err < 0 ? err : 0; 2551 2552 if (f2fs_force_buffered_io(inode, iocb, iter)) 2553 return 0; 2554 2555 do_opu = allow_outplace_dio(inode, iocb, iter); 2556 2557 trace_f2fs_direct_IO_enter(inode, offset, count, rw); 2558 2559 if (rw == WRITE && whint_mode == WHINT_MODE_OFF) 2560 iocb->ki_hint = WRITE_LIFE_NOT_SET; 2561 2562 if (iocb->ki_flags & IOCB_NOWAIT) { 2563 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) { 2564 iocb->ki_hint = hint; 2565 err = -EAGAIN; 2566 goto out; 2567 } 2568 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) { 2569 up_read(&fi->i_gc_rwsem[rw]); 2570 iocb->ki_hint = hint; 2571 err = -EAGAIN; 2572 goto out; 2573 } 2574 } else { 2575 down_read(&fi->i_gc_rwsem[rw]); 2576 if (do_opu) 2577 down_read(&fi->i_gc_rwsem[READ]); 2578 } 2579 2580 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio); 2581 2582 if (do_opu) 2583 up_read(&fi->i_gc_rwsem[READ]); 2584 2585 up_read(&fi->i_gc_rwsem[rw]); 2586 2587 if (rw == WRITE) { 2588 if (whint_mode == WHINT_MODE_OFF) 2589 iocb->ki_hint = hint; 2590 if (err > 0) { 2591 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO, 2592 err); 2593 if (!do_opu) 2594 set_inode_flag(inode, FI_UPDATE_WRITE); 2595 } else if (err < 0) { 2596 f2fs_write_failed(mapping, offset + count); 2597 } 2598 } 2599 2600 out: 2601 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); 2602 2603 return err; 2604 } 2605 2606 void f2fs_invalidate_page(struct page *page, unsigned int offset, 2607 unsigned int length) 2608 { 2609 struct inode *inode = page->mapping->host; 2610 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2611 2612 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 2613 (offset % PAGE_SIZE || length != PAGE_SIZE)) 2614 return; 2615 2616 if (PageDirty(page)) { 2617 if (inode->i_ino == F2FS_META_INO(sbi)) { 2618 dec_page_count(sbi, F2FS_DIRTY_META); 2619 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 2620 dec_page_count(sbi, F2FS_DIRTY_NODES); 2621 } else { 2622 inode_dec_dirty_pages(inode); 2623 f2fs_remove_dirty_inode(inode); 2624 } 2625 } 2626 2627 clear_cold_data(page); 2628 2629 /* This is atomic written page, keep Private */ 2630 if (IS_ATOMIC_WRITTEN_PAGE(page)) 2631 return f2fs_drop_inmem_page(inode, page); 2632 2633 set_page_private(page, 0); 2634 ClearPagePrivate(page); 2635 } 2636 2637 int f2fs_release_page(struct page *page, gfp_t wait) 2638 { 2639 /* If this is dirty page, keep PagePrivate */ 2640 if (PageDirty(page)) 2641 return 0; 2642 2643 /* This is atomic written page, keep Private */ 2644 if (IS_ATOMIC_WRITTEN_PAGE(page)) 2645 return 0; 2646 2647 clear_cold_data(page); 2648 set_page_private(page, 0); 2649 ClearPagePrivate(page); 2650 return 1; 2651 } 2652 2653 static int f2fs_set_data_page_dirty(struct page *page) 2654 { 2655 struct address_space *mapping = page->mapping; 2656 struct inode *inode = mapping->host; 2657 2658 trace_f2fs_set_page_dirty(page, DATA); 2659 2660 if (!PageUptodate(page)) 2661 SetPageUptodate(page); 2662 2663 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) { 2664 if (!IS_ATOMIC_WRITTEN_PAGE(page)) { 2665 f2fs_register_inmem_page(inode, page); 2666 return 1; 2667 } 2668 /* 2669 * Previously, this page has been registered, we just 2670 * return here. 2671 */ 2672 return 0; 2673 } 2674 2675 if (!PageDirty(page)) { 2676 __set_page_dirty_nobuffers(page); 2677 f2fs_update_dirty_page(inode, page); 2678 return 1; 2679 } 2680 return 0; 2681 } 2682 2683 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 2684 { 2685 struct inode *inode = mapping->host; 2686 2687 if (f2fs_has_inline_data(inode)) 2688 return 0; 2689 2690 /* make sure allocating whole blocks */ 2691 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 2692 filemap_write_and_wait(mapping); 2693 2694 return generic_block_bmap(mapping, block, get_data_block_bmap); 2695 } 2696 2697 #ifdef CONFIG_MIGRATION 2698 #include <linux/migrate.h> 2699 2700 int f2fs_migrate_page(struct address_space *mapping, 2701 struct page *newpage, struct page *page, enum migrate_mode mode) 2702 { 2703 int rc, extra_count; 2704 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 2705 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page); 2706 2707 BUG_ON(PageWriteback(page)); 2708 2709 /* migrating an atomic written page is safe with the inmem_lock hold */ 2710 if (atomic_written) { 2711 if (mode != MIGRATE_SYNC) 2712 return -EBUSY; 2713 if (!mutex_trylock(&fi->inmem_lock)) 2714 return -EAGAIN; 2715 } 2716 2717 /* 2718 * A reference is expected if PagePrivate set when move mapping, 2719 * however F2FS breaks this for maintaining dirty page counts when 2720 * truncating pages. So here adjusting the 'extra_count' make it work. 2721 */ 2722 extra_count = (atomic_written ? 1 : 0) - page_has_private(page); 2723 rc = migrate_page_move_mapping(mapping, newpage, 2724 page, NULL, mode, extra_count); 2725 if (rc != MIGRATEPAGE_SUCCESS) { 2726 if (atomic_written) 2727 mutex_unlock(&fi->inmem_lock); 2728 return rc; 2729 } 2730 2731 if (atomic_written) { 2732 struct inmem_pages *cur; 2733 list_for_each_entry(cur, &fi->inmem_pages, list) 2734 if (cur->page == page) { 2735 cur->page = newpage; 2736 break; 2737 } 2738 mutex_unlock(&fi->inmem_lock); 2739 put_page(page); 2740 get_page(newpage); 2741 } 2742 2743 if (PagePrivate(page)) 2744 SetPagePrivate(newpage); 2745 set_page_private(newpage, page_private(page)); 2746 2747 if (mode != MIGRATE_SYNC_NO_COPY) 2748 migrate_page_copy(newpage, page); 2749 else 2750 migrate_page_states(newpage, page); 2751 2752 return MIGRATEPAGE_SUCCESS; 2753 } 2754 #endif 2755 2756 const struct address_space_operations f2fs_dblock_aops = { 2757 .readpage = f2fs_read_data_page, 2758 .readpages = f2fs_read_data_pages, 2759 .writepage = f2fs_write_data_page, 2760 .writepages = f2fs_write_data_pages, 2761 .write_begin = f2fs_write_begin, 2762 .write_end = f2fs_write_end, 2763 .set_page_dirty = f2fs_set_data_page_dirty, 2764 .invalidatepage = f2fs_invalidate_page, 2765 .releasepage = f2fs_release_page, 2766 .direct_IO = f2fs_direct_IO, 2767 .bmap = f2fs_bmap, 2768 #ifdef CONFIG_MIGRATION 2769 .migratepage = f2fs_migrate_page, 2770 #endif 2771 }; 2772 2773 void f2fs_clear_radix_tree_dirty_tag(struct page *page) 2774 { 2775 struct address_space *mapping = page_mapping(page); 2776 unsigned long flags; 2777 2778 xa_lock_irqsave(&mapping->i_pages, flags); 2779 radix_tree_tag_clear(&mapping->i_pages, page_index(page), 2780 PAGECACHE_TAG_DIRTY); 2781 xa_unlock_irqrestore(&mapping->i_pages, flags); 2782 } 2783 2784 int __init f2fs_init_post_read_processing(void) 2785 { 2786 bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0); 2787 if (!bio_post_read_ctx_cache) 2788 goto fail; 2789 bio_post_read_ctx_pool = 2790 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 2791 bio_post_read_ctx_cache); 2792 if (!bio_post_read_ctx_pool) 2793 goto fail_free_cache; 2794 return 0; 2795 2796 fail_free_cache: 2797 kmem_cache_destroy(bio_post_read_ctx_cache); 2798 fail: 2799 return -ENOMEM; 2800 } 2801 2802 void __exit f2fs_destroy_post_read_processing(void) 2803 { 2804 mempool_destroy(bio_post_read_ctx_pool); 2805 kmem_cache_destroy(bio_post_read_ctx_cache); 2806 } 2807