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/swap.h> 18 #include <linux/prefetch.h> 19 #include <linux/uio.h> 20 #include <linux/cleancache.h> 21 #include <linux/sched/signal.h> 22 #include <linux/fiemap.h> 23 24 #include "f2fs.h" 25 #include "node.h" 26 #include "segment.h" 27 #include "trace.h" 28 #include <trace/events/f2fs.h> 29 30 #define NUM_PREALLOC_POST_READ_CTXS 128 31 32 static struct kmem_cache *bio_post_read_ctx_cache; 33 static struct kmem_cache *bio_entry_slab; 34 static mempool_t *bio_post_read_ctx_pool; 35 static struct bio_set f2fs_bioset; 36 37 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE 38 39 int __init f2fs_init_bioset(void) 40 { 41 if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE, 42 0, BIOSET_NEED_BVECS)) 43 return -ENOMEM; 44 return 0; 45 } 46 47 void f2fs_destroy_bioset(void) 48 { 49 bioset_exit(&f2fs_bioset); 50 } 51 52 static inline struct bio *__f2fs_bio_alloc(gfp_t gfp_mask, 53 unsigned int nr_iovecs) 54 { 55 return bio_alloc_bioset(gfp_mask, nr_iovecs, &f2fs_bioset); 56 } 57 58 struct bio *f2fs_bio_alloc(struct f2fs_sb_info *sbi, int npages, bool noio) 59 { 60 if (noio) { 61 /* No failure on bio allocation */ 62 return __f2fs_bio_alloc(GFP_NOIO, npages); 63 } 64 65 if (time_to_inject(sbi, FAULT_ALLOC_BIO)) { 66 f2fs_show_injection_info(sbi, FAULT_ALLOC_BIO); 67 return NULL; 68 } 69 70 return __f2fs_bio_alloc(GFP_KERNEL, npages); 71 } 72 73 static bool __is_cp_guaranteed(struct page *page) 74 { 75 struct address_space *mapping = page->mapping; 76 struct inode *inode; 77 struct f2fs_sb_info *sbi; 78 79 if (!mapping) 80 return false; 81 82 if (f2fs_is_compressed_page(page)) 83 return false; 84 85 inode = mapping->host; 86 sbi = F2FS_I_SB(inode); 87 88 if (inode->i_ino == F2FS_META_INO(sbi) || 89 inode->i_ino == F2FS_NODE_INO(sbi) || 90 S_ISDIR(inode->i_mode) || 91 (S_ISREG(inode->i_mode) && 92 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) || 93 is_cold_data(page)) 94 return true; 95 return false; 96 } 97 98 static enum count_type __read_io_type(struct page *page) 99 { 100 struct address_space *mapping = page_file_mapping(page); 101 102 if (mapping) { 103 struct inode *inode = mapping->host; 104 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 105 106 if (inode->i_ino == F2FS_META_INO(sbi)) 107 return F2FS_RD_META; 108 109 if (inode->i_ino == F2FS_NODE_INO(sbi)) 110 return F2FS_RD_NODE; 111 } 112 return F2FS_RD_DATA; 113 } 114 115 /* postprocessing steps for read bios */ 116 enum bio_post_read_step { 117 STEP_DECRYPT, 118 STEP_DECOMPRESS, 119 STEP_VERITY, 120 }; 121 122 struct bio_post_read_ctx { 123 struct bio *bio; 124 struct f2fs_sb_info *sbi; 125 struct work_struct work; 126 unsigned int enabled_steps; 127 }; 128 129 static void __read_end_io(struct bio *bio, bool compr, bool verity) 130 { 131 struct page *page; 132 struct bio_vec *bv; 133 struct bvec_iter_all iter_all; 134 135 bio_for_each_segment_all(bv, bio, iter_all) { 136 page = bv->bv_page; 137 138 #ifdef CONFIG_F2FS_FS_COMPRESSION 139 if (compr && f2fs_is_compressed_page(page)) { 140 f2fs_decompress_pages(bio, page, verity); 141 continue; 142 } 143 if (verity) 144 continue; 145 #endif 146 147 /* PG_error was set if any post_read step failed */ 148 if (bio->bi_status || PageError(page)) { 149 ClearPageUptodate(page); 150 /* will re-read again later */ 151 ClearPageError(page); 152 } else { 153 SetPageUptodate(page); 154 } 155 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 156 unlock_page(page); 157 } 158 } 159 160 static void f2fs_release_read_bio(struct bio *bio); 161 static void __f2fs_read_end_io(struct bio *bio, bool compr, bool verity) 162 { 163 if (!compr) 164 __read_end_io(bio, false, verity); 165 f2fs_release_read_bio(bio); 166 } 167 168 static void f2fs_decompress_bio(struct bio *bio, bool verity) 169 { 170 __read_end_io(bio, true, verity); 171 } 172 173 static void bio_post_read_processing(struct bio_post_read_ctx *ctx); 174 175 static void f2fs_decrypt_work(struct bio_post_read_ctx *ctx) 176 { 177 fscrypt_decrypt_bio(ctx->bio); 178 } 179 180 static void f2fs_decompress_work(struct bio_post_read_ctx *ctx) 181 { 182 f2fs_decompress_bio(ctx->bio, ctx->enabled_steps & (1 << STEP_VERITY)); 183 } 184 185 #ifdef CONFIG_F2FS_FS_COMPRESSION 186 static void f2fs_verify_pages(struct page **rpages, unsigned int cluster_size) 187 { 188 f2fs_decompress_end_io(rpages, cluster_size, false, true); 189 } 190 191 static void f2fs_verify_bio(struct bio *bio) 192 { 193 struct bio_vec *bv; 194 struct bvec_iter_all iter_all; 195 196 bio_for_each_segment_all(bv, bio, iter_all) { 197 struct page *page = bv->bv_page; 198 struct decompress_io_ctx *dic; 199 200 dic = (struct decompress_io_ctx *)page_private(page); 201 202 if (dic) { 203 if (refcount_dec_not_one(&dic->ref)) 204 continue; 205 f2fs_verify_pages(dic->rpages, 206 dic->cluster_size); 207 f2fs_free_dic(dic); 208 continue; 209 } 210 211 if (bio->bi_status || PageError(page)) 212 goto clear_uptodate; 213 214 if (fsverity_verify_page(page)) { 215 SetPageUptodate(page); 216 goto unlock; 217 } 218 clear_uptodate: 219 ClearPageUptodate(page); 220 ClearPageError(page); 221 unlock: 222 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 223 unlock_page(page); 224 } 225 } 226 #endif 227 228 static void f2fs_verity_work(struct work_struct *work) 229 { 230 struct bio_post_read_ctx *ctx = 231 container_of(work, struct bio_post_read_ctx, work); 232 struct bio *bio = ctx->bio; 233 #ifdef CONFIG_F2FS_FS_COMPRESSION 234 unsigned int enabled_steps = ctx->enabled_steps; 235 #endif 236 237 /* 238 * fsverity_verify_bio() may call readpages() again, and while verity 239 * will be disabled for this, decryption may still be needed, resulting 240 * in another bio_post_read_ctx being allocated. So to prevent 241 * deadlocks we need to release the current ctx to the mempool first. 242 * This assumes that verity is the last post-read step. 243 */ 244 mempool_free(ctx, bio_post_read_ctx_pool); 245 bio->bi_private = NULL; 246 247 #ifdef CONFIG_F2FS_FS_COMPRESSION 248 /* previous step is decompression */ 249 if (enabled_steps & (1 << STEP_DECOMPRESS)) { 250 f2fs_verify_bio(bio); 251 f2fs_release_read_bio(bio); 252 return; 253 } 254 #endif 255 256 fsverity_verify_bio(bio); 257 __f2fs_read_end_io(bio, false, false); 258 } 259 260 static void f2fs_post_read_work(struct work_struct *work) 261 { 262 struct bio_post_read_ctx *ctx = 263 container_of(work, struct bio_post_read_ctx, work); 264 265 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) 266 f2fs_decrypt_work(ctx); 267 268 if (ctx->enabled_steps & (1 << STEP_DECOMPRESS)) 269 f2fs_decompress_work(ctx); 270 271 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 272 INIT_WORK(&ctx->work, f2fs_verity_work); 273 fsverity_enqueue_verify_work(&ctx->work); 274 return; 275 } 276 277 __f2fs_read_end_io(ctx->bio, 278 ctx->enabled_steps & (1 << STEP_DECOMPRESS), false); 279 } 280 281 static void f2fs_enqueue_post_read_work(struct f2fs_sb_info *sbi, 282 struct work_struct *work) 283 { 284 queue_work(sbi->post_read_wq, work); 285 } 286 287 static void bio_post_read_processing(struct bio_post_read_ctx *ctx) 288 { 289 /* 290 * We use different work queues for decryption and for verity because 291 * verity may require reading metadata pages that need decryption, and 292 * we shouldn't recurse to the same workqueue. 293 */ 294 295 if (ctx->enabled_steps & (1 << STEP_DECRYPT) || 296 ctx->enabled_steps & (1 << STEP_DECOMPRESS)) { 297 INIT_WORK(&ctx->work, f2fs_post_read_work); 298 f2fs_enqueue_post_read_work(ctx->sbi, &ctx->work); 299 return; 300 } 301 302 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 303 INIT_WORK(&ctx->work, f2fs_verity_work); 304 fsverity_enqueue_verify_work(&ctx->work); 305 return; 306 } 307 308 __f2fs_read_end_io(ctx->bio, false, false); 309 } 310 311 static bool f2fs_bio_post_read_required(struct bio *bio) 312 { 313 return bio->bi_private; 314 } 315 316 static void f2fs_read_end_io(struct bio *bio) 317 { 318 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio)); 319 320 if (time_to_inject(sbi, FAULT_READ_IO)) { 321 f2fs_show_injection_info(sbi, FAULT_READ_IO); 322 bio->bi_status = BLK_STS_IOERR; 323 } 324 325 if (f2fs_bio_post_read_required(bio)) { 326 struct bio_post_read_ctx *ctx = bio->bi_private; 327 328 bio_post_read_processing(ctx); 329 return; 330 } 331 332 __f2fs_read_end_io(bio, false, false); 333 } 334 335 static void f2fs_write_end_io(struct bio *bio) 336 { 337 struct f2fs_sb_info *sbi = bio->bi_private; 338 struct bio_vec *bvec; 339 struct bvec_iter_all iter_all; 340 341 if (time_to_inject(sbi, FAULT_WRITE_IO)) { 342 f2fs_show_injection_info(sbi, FAULT_WRITE_IO); 343 bio->bi_status = BLK_STS_IOERR; 344 } 345 346 bio_for_each_segment_all(bvec, bio, iter_all) { 347 struct page *page = bvec->bv_page; 348 enum count_type type = WB_DATA_TYPE(page); 349 350 if (IS_DUMMY_WRITTEN_PAGE(page)) { 351 set_page_private(page, (unsigned long)NULL); 352 ClearPagePrivate(page); 353 unlock_page(page); 354 mempool_free(page, sbi->write_io_dummy); 355 356 if (unlikely(bio->bi_status)) 357 f2fs_stop_checkpoint(sbi, true); 358 continue; 359 } 360 361 fscrypt_finalize_bounce_page(&page); 362 363 #ifdef CONFIG_F2FS_FS_COMPRESSION 364 if (f2fs_is_compressed_page(page)) { 365 f2fs_compress_write_end_io(bio, page); 366 continue; 367 } 368 #endif 369 370 if (unlikely(bio->bi_status)) { 371 mapping_set_error(page->mapping, -EIO); 372 if (type == F2FS_WB_CP_DATA) 373 f2fs_stop_checkpoint(sbi, true); 374 } 375 376 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 377 page->index != nid_of_node(page)); 378 379 dec_page_count(sbi, type); 380 if (f2fs_in_warm_node_list(sbi, page)) 381 f2fs_del_fsync_node_entry(sbi, page); 382 clear_cold_data(page); 383 end_page_writeback(page); 384 } 385 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 386 wq_has_sleeper(&sbi->cp_wait)) 387 wake_up(&sbi->cp_wait); 388 389 bio_put(bio); 390 } 391 392 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 393 block_t blk_addr, struct bio *bio) 394 { 395 struct block_device *bdev = sbi->sb->s_bdev; 396 int i; 397 398 if (f2fs_is_multi_device(sbi)) { 399 for (i = 0; i < sbi->s_ndevs; i++) { 400 if (FDEV(i).start_blk <= blk_addr && 401 FDEV(i).end_blk >= blk_addr) { 402 blk_addr -= FDEV(i).start_blk; 403 bdev = FDEV(i).bdev; 404 break; 405 } 406 } 407 } 408 if (bio) { 409 bio_set_dev(bio, bdev); 410 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); 411 } 412 return bdev; 413 } 414 415 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 416 { 417 int i; 418 419 if (!f2fs_is_multi_device(sbi)) 420 return 0; 421 422 for (i = 0; i < sbi->s_ndevs; i++) 423 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 424 return i; 425 return 0; 426 } 427 428 /* 429 * Return true, if pre_bio's bdev is same as its target device. 430 */ 431 static bool __same_bdev(struct f2fs_sb_info *sbi, 432 block_t blk_addr, struct bio *bio) 433 { 434 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL); 435 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno; 436 } 437 438 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) 439 { 440 struct f2fs_sb_info *sbi = fio->sbi; 441 struct bio *bio; 442 443 bio = f2fs_bio_alloc(sbi, npages, true); 444 445 f2fs_target_device(sbi, fio->new_blkaddr, bio); 446 if (is_read_io(fio->op)) { 447 bio->bi_end_io = f2fs_read_end_io; 448 bio->bi_private = NULL; 449 } else { 450 bio->bi_end_io = f2fs_write_end_io; 451 bio->bi_private = sbi; 452 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, 453 fio->type, fio->temp); 454 } 455 if (fio->io_wbc) 456 wbc_init_bio(fio->io_wbc, bio); 457 458 return bio; 459 } 460 461 static inline void __submit_bio(struct f2fs_sb_info *sbi, 462 struct bio *bio, enum page_type type) 463 { 464 if (!is_read_io(bio_op(bio))) { 465 unsigned int start; 466 467 if (type != DATA && type != NODE) 468 goto submit_io; 469 470 if (f2fs_lfs_mode(sbi) && current->plug) 471 blk_finish_plug(current->plug); 472 473 if (F2FS_IO_ALIGNED(sbi)) 474 goto submit_io; 475 476 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS; 477 start %= F2FS_IO_SIZE(sbi); 478 479 if (start == 0) 480 goto submit_io; 481 482 /* fill dummy pages */ 483 for (; start < F2FS_IO_SIZE(sbi); start++) { 484 struct page *page = 485 mempool_alloc(sbi->write_io_dummy, 486 GFP_NOIO | __GFP_NOFAIL); 487 f2fs_bug_on(sbi, !page); 488 489 zero_user_segment(page, 0, PAGE_SIZE); 490 SetPagePrivate(page); 491 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE); 492 lock_page(page); 493 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 494 f2fs_bug_on(sbi, 1); 495 } 496 /* 497 * In the NODE case, we lose next block address chain. So, we 498 * need to do checkpoint in f2fs_sync_file. 499 */ 500 if (type == NODE) 501 set_sbi_flag(sbi, SBI_NEED_CP); 502 } 503 submit_io: 504 if (is_read_io(bio_op(bio))) 505 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 506 else 507 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 508 submit_bio(bio); 509 } 510 511 void f2fs_submit_bio(struct f2fs_sb_info *sbi, 512 struct bio *bio, enum page_type type) 513 { 514 __submit_bio(sbi, bio, type); 515 } 516 517 static void __submit_merged_bio(struct f2fs_bio_info *io) 518 { 519 struct f2fs_io_info *fio = &io->fio; 520 521 if (!io->bio) 522 return; 523 524 bio_set_op_attrs(io->bio, fio->op, fio->op_flags); 525 526 if (is_read_io(fio->op)) 527 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 528 else 529 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 530 531 __submit_bio(io->sbi, io->bio, fio->type); 532 io->bio = NULL; 533 } 534 535 static bool __has_merged_page(struct bio *bio, struct inode *inode, 536 struct page *page, nid_t ino) 537 { 538 struct bio_vec *bvec; 539 struct bvec_iter_all iter_all; 540 541 if (!bio) 542 return false; 543 544 if (!inode && !page && !ino) 545 return true; 546 547 bio_for_each_segment_all(bvec, bio, iter_all) { 548 struct page *target = bvec->bv_page; 549 550 if (fscrypt_is_bounce_page(target)) { 551 target = fscrypt_pagecache_page(target); 552 if (IS_ERR(target)) 553 continue; 554 } 555 if (f2fs_is_compressed_page(target)) { 556 target = f2fs_compress_control_page(target); 557 if (IS_ERR(target)) 558 continue; 559 } 560 561 if (inode && inode == target->mapping->host) 562 return true; 563 if (page && page == target) 564 return true; 565 if (ino && ino == ino_of_node(target)) 566 return true; 567 } 568 569 return false; 570 } 571 572 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 573 enum page_type type, enum temp_type temp) 574 { 575 enum page_type btype = PAGE_TYPE_OF_BIO(type); 576 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 577 578 down_write(&io->io_rwsem); 579 580 /* change META to META_FLUSH in the checkpoint procedure */ 581 if (type >= META_FLUSH) { 582 io->fio.type = META_FLUSH; 583 io->fio.op = REQ_OP_WRITE; 584 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC; 585 if (!test_opt(sbi, NOBARRIER)) 586 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 587 } 588 __submit_merged_bio(io); 589 up_write(&io->io_rwsem); 590 } 591 592 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 593 struct inode *inode, struct page *page, 594 nid_t ino, enum page_type type, bool force) 595 { 596 enum temp_type temp; 597 bool ret = true; 598 599 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 600 if (!force) { 601 enum page_type btype = PAGE_TYPE_OF_BIO(type); 602 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 603 604 down_read(&io->io_rwsem); 605 ret = __has_merged_page(io->bio, inode, page, ino); 606 up_read(&io->io_rwsem); 607 } 608 if (ret) 609 __f2fs_submit_merged_write(sbi, type, temp); 610 611 /* TODO: use HOT temp only for meta pages now. */ 612 if (type >= META) 613 break; 614 } 615 } 616 617 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 618 { 619 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true); 620 } 621 622 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 623 struct inode *inode, struct page *page, 624 nid_t ino, enum page_type type) 625 { 626 __submit_merged_write_cond(sbi, inode, page, ino, type, false); 627 } 628 629 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 630 { 631 f2fs_submit_merged_write(sbi, DATA); 632 f2fs_submit_merged_write(sbi, NODE); 633 f2fs_submit_merged_write(sbi, META); 634 } 635 636 /* 637 * Fill the locked page with data located in the block address. 638 * A caller needs to unlock the page on failure. 639 */ 640 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 641 { 642 struct bio *bio; 643 struct page *page = fio->encrypted_page ? 644 fio->encrypted_page : fio->page; 645 646 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 647 fio->is_por ? META_POR : (__is_meta_io(fio) ? 648 META_GENERIC : DATA_GENERIC_ENHANCE))) 649 return -EFSCORRUPTED; 650 651 trace_f2fs_submit_page_bio(page, fio); 652 f2fs_trace_ios(fio, 0); 653 654 /* Allocate a new bio */ 655 bio = __bio_alloc(fio, 1); 656 657 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 658 bio_put(bio); 659 return -EFAULT; 660 } 661 662 if (fio->io_wbc && !is_read_io(fio->op)) 663 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 664 665 bio_set_op_attrs(bio, fio->op, fio->op_flags); 666 667 inc_page_count(fio->sbi, is_read_io(fio->op) ? 668 __read_io_type(page): WB_DATA_TYPE(fio->page)); 669 670 __submit_bio(fio->sbi, bio, fio->type); 671 return 0; 672 } 673 674 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 675 block_t last_blkaddr, block_t cur_blkaddr) 676 { 677 if (last_blkaddr + 1 != cur_blkaddr) 678 return false; 679 return __same_bdev(sbi, cur_blkaddr, bio); 680 } 681 682 static bool io_type_is_mergeable(struct f2fs_bio_info *io, 683 struct f2fs_io_info *fio) 684 { 685 if (io->fio.op != fio->op) 686 return false; 687 return io->fio.op_flags == fio->op_flags; 688 } 689 690 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 691 struct f2fs_bio_info *io, 692 struct f2fs_io_info *fio, 693 block_t last_blkaddr, 694 block_t cur_blkaddr) 695 { 696 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) { 697 unsigned int filled_blocks = 698 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size); 699 unsigned int io_size = F2FS_IO_SIZE(sbi); 700 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt; 701 702 /* IOs in bio is aligned and left space of vectors is not enough */ 703 if (!(filled_blocks % io_size) && left_vecs < io_size) 704 return false; 705 } 706 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr)) 707 return false; 708 return io_type_is_mergeable(io, fio); 709 } 710 711 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio, 712 struct page *page, enum temp_type temp) 713 { 714 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 715 struct bio_entry *be; 716 717 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS); 718 be->bio = bio; 719 bio_get(bio); 720 721 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE) 722 f2fs_bug_on(sbi, 1); 723 724 down_write(&io->bio_list_lock); 725 list_add_tail(&be->list, &io->bio_list); 726 up_write(&io->bio_list_lock); 727 } 728 729 static void del_bio_entry(struct bio_entry *be) 730 { 731 list_del(&be->list); 732 kmem_cache_free(bio_entry_slab, be); 733 } 734 735 static int add_ipu_page(struct f2fs_sb_info *sbi, struct bio **bio, 736 struct page *page) 737 { 738 enum temp_type temp; 739 bool found = false; 740 int ret = -EAGAIN; 741 742 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 743 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 744 struct list_head *head = &io->bio_list; 745 struct bio_entry *be; 746 747 down_write(&io->bio_list_lock); 748 list_for_each_entry(be, head, list) { 749 if (be->bio != *bio) 750 continue; 751 752 found = true; 753 754 if (bio_add_page(*bio, page, PAGE_SIZE, 0) == 755 PAGE_SIZE) { 756 ret = 0; 757 break; 758 } 759 760 /* bio is full */ 761 del_bio_entry(be); 762 __submit_bio(sbi, *bio, DATA); 763 break; 764 } 765 up_write(&io->bio_list_lock); 766 } 767 768 if (ret) { 769 bio_put(*bio); 770 *bio = NULL; 771 } 772 773 return ret; 774 } 775 776 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi, 777 struct bio **bio, struct page *page) 778 { 779 enum temp_type temp; 780 bool found = false; 781 struct bio *target = bio ? *bio : NULL; 782 783 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 784 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 785 struct list_head *head = &io->bio_list; 786 struct bio_entry *be; 787 788 if (list_empty(head)) 789 continue; 790 791 down_read(&io->bio_list_lock); 792 list_for_each_entry(be, head, list) { 793 if (target) 794 found = (target == be->bio); 795 else 796 found = __has_merged_page(be->bio, NULL, 797 page, 0); 798 if (found) 799 break; 800 } 801 up_read(&io->bio_list_lock); 802 803 if (!found) 804 continue; 805 806 found = false; 807 808 down_write(&io->bio_list_lock); 809 list_for_each_entry(be, head, list) { 810 if (target) 811 found = (target == be->bio); 812 else 813 found = __has_merged_page(be->bio, NULL, 814 page, 0); 815 if (found) { 816 target = be->bio; 817 del_bio_entry(be); 818 break; 819 } 820 } 821 up_write(&io->bio_list_lock); 822 } 823 824 if (found) 825 __submit_bio(sbi, target, DATA); 826 if (bio && *bio) { 827 bio_put(*bio); 828 *bio = NULL; 829 } 830 } 831 832 int f2fs_merge_page_bio(struct f2fs_io_info *fio) 833 { 834 struct bio *bio = *fio->bio; 835 struct page *page = fio->encrypted_page ? 836 fio->encrypted_page : fio->page; 837 838 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 839 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) 840 return -EFSCORRUPTED; 841 842 trace_f2fs_submit_page_bio(page, fio); 843 f2fs_trace_ios(fio, 0); 844 845 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block, 846 fio->new_blkaddr)) 847 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL); 848 alloc_new: 849 if (!bio) { 850 bio = __bio_alloc(fio, BIO_MAX_PAGES); 851 bio_set_op_attrs(bio, fio->op, fio->op_flags); 852 853 add_bio_entry(fio->sbi, bio, page, fio->temp); 854 } else { 855 if (add_ipu_page(fio->sbi, &bio, page)) 856 goto alloc_new; 857 } 858 859 if (fio->io_wbc) 860 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 861 862 inc_page_count(fio->sbi, WB_DATA_TYPE(page)); 863 864 *fio->last_block = fio->new_blkaddr; 865 *fio->bio = bio; 866 867 return 0; 868 } 869 870 void f2fs_submit_page_write(struct f2fs_io_info *fio) 871 { 872 struct f2fs_sb_info *sbi = fio->sbi; 873 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 874 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 875 struct page *bio_page; 876 877 f2fs_bug_on(sbi, is_read_io(fio->op)); 878 879 down_write(&io->io_rwsem); 880 next: 881 if (fio->in_list) { 882 spin_lock(&io->io_lock); 883 if (list_empty(&io->io_list)) { 884 spin_unlock(&io->io_lock); 885 goto out; 886 } 887 fio = list_first_entry(&io->io_list, 888 struct f2fs_io_info, list); 889 list_del(&fio->list); 890 spin_unlock(&io->io_lock); 891 } 892 893 verify_fio_blkaddr(fio); 894 895 if (fio->encrypted_page) 896 bio_page = fio->encrypted_page; 897 else if (fio->compressed_page) 898 bio_page = fio->compressed_page; 899 else 900 bio_page = fio->page; 901 902 /* set submitted = true as a return value */ 903 fio->submitted = true; 904 905 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 906 907 if (io->bio && !io_is_mergeable(sbi, io->bio, io, fio, 908 io->last_block_in_bio, fio->new_blkaddr)) 909 __submit_merged_bio(io); 910 alloc_new: 911 if (io->bio == NULL) { 912 if (F2FS_IO_ALIGNED(sbi) && 913 (fio->type == DATA || fio->type == NODE) && 914 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 915 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 916 fio->retry = true; 917 goto skip; 918 } 919 io->bio = __bio_alloc(fio, BIO_MAX_PAGES); 920 io->fio = *fio; 921 } 922 923 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 924 __submit_merged_bio(io); 925 goto alloc_new; 926 } 927 928 if (fio->io_wbc) 929 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE); 930 931 io->last_block_in_bio = fio->new_blkaddr; 932 f2fs_trace_ios(fio, 0); 933 934 trace_f2fs_submit_page_write(fio->page, fio); 935 skip: 936 if (fio->in_list) 937 goto next; 938 out: 939 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 940 !f2fs_is_checkpoint_ready(sbi)) 941 __submit_merged_bio(io); 942 up_write(&io->io_rwsem); 943 } 944 945 static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx) 946 { 947 return fsverity_active(inode) && 948 idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 949 } 950 951 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 952 unsigned nr_pages, unsigned op_flag, 953 pgoff_t first_idx, bool for_write) 954 { 955 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 956 struct bio *bio; 957 struct bio_post_read_ctx *ctx; 958 unsigned int post_read_steps = 0; 959 960 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), 961 for_write); 962 if (!bio) 963 return ERR_PTR(-ENOMEM); 964 f2fs_target_device(sbi, blkaddr, bio); 965 bio->bi_end_io = f2fs_read_end_io; 966 bio_set_op_attrs(bio, REQ_OP_READ, op_flag); 967 968 if (f2fs_encrypted_file(inode)) 969 post_read_steps |= 1 << STEP_DECRYPT; 970 if (f2fs_compressed_file(inode)) 971 post_read_steps |= 1 << STEP_DECOMPRESS; 972 if (f2fs_need_verity(inode, first_idx)) 973 post_read_steps |= 1 << STEP_VERITY; 974 975 if (post_read_steps) { 976 /* Due to the mempool, this never fails. */ 977 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 978 ctx->bio = bio; 979 ctx->sbi = sbi; 980 ctx->enabled_steps = post_read_steps; 981 bio->bi_private = ctx; 982 } 983 984 return bio; 985 } 986 987 static void f2fs_release_read_bio(struct bio *bio) 988 { 989 if (bio->bi_private) 990 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 991 bio_put(bio); 992 } 993 994 /* This can handle encryption stuffs */ 995 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 996 block_t blkaddr, bool for_write) 997 { 998 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 999 struct bio *bio; 1000 1001 bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0, page->index, for_write); 1002 if (IS_ERR(bio)) 1003 return PTR_ERR(bio); 1004 1005 /* wait for GCed page writeback via META_MAPPING */ 1006 f2fs_wait_on_block_writeback(inode, blkaddr); 1007 1008 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 1009 bio_put(bio); 1010 return -EFAULT; 1011 } 1012 ClearPageError(page); 1013 inc_page_count(sbi, F2FS_RD_DATA); 1014 __submit_bio(sbi, bio, DATA); 1015 return 0; 1016 } 1017 1018 static void __set_data_blkaddr(struct dnode_of_data *dn) 1019 { 1020 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 1021 __le32 *addr_array; 1022 int base = 0; 1023 1024 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 1025 base = get_extra_isize(dn->inode); 1026 1027 /* Get physical address of data block */ 1028 addr_array = blkaddr_in_node(rn); 1029 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 1030 } 1031 1032 /* 1033 * Lock ordering for the change of data block address: 1034 * ->data_page 1035 * ->node_page 1036 * update block addresses in the node page 1037 */ 1038 void f2fs_set_data_blkaddr(struct dnode_of_data *dn) 1039 { 1040 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1041 __set_data_blkaddr(dn); 1042 if (set_page_dirty(dn->node_page)) 1043 dn->node_changed = true; 1044 } 1045 1046 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1047 { 1048 dn->data_blkaddr = blkaddr; 1049 f2fs_set_data_blkaddr(dn); 1050 f2fs_update_extent_cache(dn); 1051 } 1052 1053 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 1054 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 1055 { 1056 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1057 int err; 1058 1059 if (!count) 1060 return 0; 1061 1062 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1063 return -EPERM; 1064 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1065 return err; 1066 1067 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 1068 dn->ofs_in_node, count); 1069 1070 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1071 1072 for (; count > 0; dn->ofs_in_node++) { 1073 block_t blkaddr = f2fs_data_blkaddr(dn); 1074 if (blkaddr == NULL_ADDR) { 1075 dn->data_blkaddr = NEW_ADDR; 1076 __set_data_blkaddr(dn); 1077 count--; 1078 } 1079 } 1080 1081 if (set_page_dirty(dn->node_page)) 1082 dn->node_changed = true; 1083 return 0; 1084 } 1085 1086 /* Should keep dn->ofs_in_node unchanged */ 1087 int f2fs_reserve_new_block(struct dnode_of_data *dn) 1088 { 1089 unsigned int ofs_in_node = dn->ofs_in_node; 1090 int ret; 1091 1092 ret = f2fs_reserve_new_blocks(dn, 1); 1093 dn->ofs_in_node = ofs_in_node; 1094 return ret; 1095 } 1096 1097 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 1098 { 1099 bool need_put = dn->inode_page ? false : true; 1100 int err; 1101 1102 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE); 1103 if (err) 1104 return err; 1105 1106 if (dn->data_blkaddr == NULL_ADDR) 1107 err = f2fs_reserve_new_block(dn); 1108 if (err || need_put) 1109 f2fs_put_dnode(dn); 1110 return err; 1111 } 1112 1113 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 1114 { 1115 struct extent_info ei = {0,0,0}; 1116 struct inode *inode = dn->inode; 1117 1118 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1119 dn->data_blkaddr = ei.blk + index - ei.fofs; 1120 return 0; 1121 } 1122 1123 return f2fs_reserve_block(dn, index); 1124 } 1125 1126 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, 1127 int op_flags, bool for_write) 1128 { 1129 struct address_space *mapping = inode->i_mapping; 1130 struct dnode_of_data dn; 1131 struct page *page; 1132 struct extent_info ei = {0,0,0}; 1133 int err; 1134 1135 page = f2fs_grab_cache_page(mapping, index, for_write); 1136 if (!page) 1137 return ERR_PTR(-ENOMEM); 1138 1139 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1140 dn.data_blkaddr = ei.blk + index - ei.fofs; 1141 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr, 1142 DATA_GENERIC_ENHANCE_READ)) { 1143 err = -EFSCORRUPTED; 1144 goto put_err; 1145 } 1146 goto got_it; 1147 } 1148 1149 set_new_dnode(&dn, inode, NULL, NULL, 0); 1150 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1151 if (err) 1152 goto put_err; 1153 f2fs_put_dnode(&dn); 1154 1155 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1156 err = -ENOENT; 1157 goto put_err; 1158 } 1159 if (dn.data_blkaddr != NEW_ADDR && 1160 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode), 1161 dn.data_blkaddr, 1162 DATA_GENERIC_ENHANCE)) { 1163 err = -EFSCORRUPTED; 1164 goto put_err; 1165 } 1166 got_it: 1167 if (PageUptodate(page)) { 1168 unlock_page(page); 1169 return page; 1170 } 1171 1172 /* 1173 * A new dentry page is allocated but not able to be written, since its 1174 * new inode page couldn't be allocated due to -ENOSPC. 1175 * In such the case, its blkaddr can be remained as NEW_ADDR. 1176 * see, f2fs_add_link -> f2fs_get_new_data_page -> 1177 * f2fs_init_inode_metadata. 1178 */ 1179 if (dn.data_blkaddr == NEW_ADDR) { 1180 zero_user_segment(page, 0, PAGE_SIZE); 1181 if (!PageUptodate(page)) 1182 SetPageUptodate(page); 1183 unlock_page(page); 1184 return page; 1185 } 1186 1187 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr, for_write); 1188 if (err) 1189 goto put_err; 1190 return page; 1191 1192 put_err: 1193 f2fs_put_page(page, 1); 1194 return ERR_PTR(err); 1195 } 1196 1197 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index) 1198 { 1199 struct address_space *mapping = inode->i_mapping; 1200 struct page *page; 1201 1202 page = find_get_page(mapping, index); 1203 if (page && PageUptodate(page)) 1204 return page; 1205 f2fs_put_page(page, 0); 1206 1207 page = f2fs_get_read_data_page(inode, index, 0, false); 1208 if (IS_ERR(page)) 1209 return page; 1210 1211 if (PageUptodate(page)) 1212 return page; 1213 1214 wait_on_page_locked(page); 1215 if (unlikely(!PageUptodate(page))) { 1216 f2fs_put_page(page, 0); 1217 return ERR_PTR(-EIO); 1218 } 1219 return page; 1220 } 1221 1222 /* 1223 * If it tries to access a hole, return an error. 1224 * Because, the callers, functions in dir.c and GC, should be able to know 1225 * whether this page exists or not. 1226 */ 1227 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, 1228 bool for_write) 1229 { 1230 struct address_space *mapping = inode->i_mapping; 1231 struct page *page; 1232 repeat: 1233 page = f2fs_get_read_data_page(inode, index, 0, for_write); 1234 if (IS_ERR(page)) 1235 return page; 1236 1237 /* wait for read completion */ 1238 lock_page(page); 1239 if (unlikely(page->mapping != mapping)) { 1240 f2fs_put_page(page, 1); 1241 goto repeat; 1242 } 1243 if (unlikely(!PageUptodate(page))) { 1244 f2fs_put_page(page, 1); 1245 return ERR_PTR(-EIO); 1246 } 1247 return page; 1248 } 1249 1250 /* 1251 * Caller ensures that this data page is never allocated. 1252 * A new zero-filled data page is allocated in the page cache. 1253 * 1254 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 1255 * f2fs_unlock_op(). 1256 * Note that, ipage is set only by make_empty_dir, and if any error occur, 1257 * ipage should be released by this function. 1258 */ 1259 struct page *f2fs_get_new_data_page(struct inode *inode, 1260 struct page *ipage, pgoff_t index, bool new_i_size) 1261 { 1262 struct address_space *mapping = inode->i_mapping; 1263 struct page *page; 1264 struct dnode_of_data dn; 1265 int err; 1266 1267 page = f2fs_grab_cache_page(mapping, index, true); 1268 if (!page) { 1269 /* 1270 * before exiting, we should make sure ipage will be released 1271 * if any error occur. 1272 */ 1273 f2fs_put_page(ipage, 1); 1274 return ERR_PTR(-ENOMEM); 1275 } 1276 1277 set_new_dnode(&dn, inode, ipage, NULL, 0); 1278 err = f2fs_reserve_block(&dn, index); 1279 if (err) { 1280 f2fs_put_page(page, 1); 1281 return ERR_PTR(err); 1282 } 1283 if (!ipage) 1284 f2fs_put_dnode(&dn); 1285 1286 if (PageUptodate(page)) 1287 goto got_it; 1288 1289 if (dn.data_blkaddr == NEW_ADDR) { 1290 zero_user_segment(page, 0, PAGE_SIZE); 1291 if (!PageUptodate(page)) 1292 SetPageUptodate(page); 1293 } else { 1294 f2fs_put_page(page, 1); 1295 1296 /* if ipage exists, blkaddr should be NEW_ADDR */ 1297 f2fs_bug_on(F2FS_I_SB(inode), ipage); 1298 page = f2fs_get_lock_data_page(inode, index, true); 1299 if (IS_ERR(page)) 1300 return page; 1301 } 1302 got_it: 1303 if (new_i_size && i_size_read(inode) < 1304 ((loff_t)(index + 1) << PAGE_SHIFT)) 1305 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 1306 return page; 1307 } 1308 1309 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 1310 { 1311 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1312 struct f2fs_summary sum; 1313 struct node_info ni; 1314 block_t old_blkaddr; 1315 blkcnt_t count = 1; 1316 int err; 1317 1318 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1319 return -EPERM; 1320 1321 err = f2fs_get_node_info(sbi, dn->nid, &ni); 1322 if (err) 1323 return err; 1324 1325 dn->data_blkaddr = f2fs_data_blkaddr(dn); 1326 if (dn->data_blkaddr != NULL_ADDR) 1327 goto alloc; 1328 1329 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1330 return err; 1331 1332 alloc: 1333 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1334 old_blkaddr = dn->data_blkaddr; 1335 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr, 1336 &sum, seg_type, NULL, false); 1337 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 1338 invalidate_mapping_pages(META_MAPPING(sbi), 1339 old_blkaddr, old_blkaddr); 1340 f2fs_update_data_blkaddr(dn, dn->data_blkaddr); 1341 1342 /* 1343 * i_size will be updated by direct_IO. Otherwise, we'll get stale 1344 * data from unwritten block via dio_read. 1345 */ 1346 return 0; 1347 } 1348 1349 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from) 1350 { 1351 struct inode *inode = file_inode(iocb->ki_filp); 1352 struct f2fs_map_blocks map; 1353 int flag; 1354 int err = 0; 1355 bool direct_io = iocb->ki_flags & IOCB_DIRECT; 1356 1357 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos); 1358 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from)); 1359 if (map.m_len > map.m_lblk) 1360 map.m_len -= map.m_lblk; 1361 else 1362 map.m_len = 0; 1363 1364 map.m_next_pgofs = NULL; 1365 map.m_next_extent = NULL; 1366 map.m_seg_type = NO_CHECK_TYPE; 1367 map.m_may_create = true; 1368 1369 if (direct_io) { 1370 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint); 1371 flag = f2fs_force_buffered_io(inode, iocb, from) ? 1372 F2FS_GET_BLOCK_PRE_AIO : 1373 F2FS_GET_BLOCK_PRE_DIO; 1374 goto map_blocks; 1375 } 1376 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) { 1377 err = f2fs_convert_inline_inode(inode); 1378 if (err) 1379 return err; 1380 } 1381 if (f2fs_has_inline_data(inode)) 1382 return err; 1383 1384 flag = F2FS_GET_BLOCK_PRE_AIO; 1385 1386 map_blocks: 1387 err = f2fs_map_blocks(inode, &map, 1, flag); 1388 if (map.m_len > 0 && err == -ENOSPC) { 1389 if (!direct_io) 1390 set_inode_flag(inode, FI_NO_PREALLOC); 1391 err = 0; 1392 } 1393 return err; 1394 } 1395 1396 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 1397 { 1398 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1399 if (lock) 1400 down_read(&sbi->node_change); 1401 else 1402 up_read(&sbi->node_change); 1403 } else { 1404 if (lock) 1405 f2fs_lock_op(sbi); 1406 else 1407 f2fs_unlock_op(sbi); 1408 } 1409 } 1410 1411 /* 1412 * f2fs_map_blocks() tries to find or build mapping relationship which 1413 * maps continuous logical blocks to physical blocks, and return such 1414 * info via f2fs_map_blocks structure. 1415 */ 1416 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 1417 int create, int flag) 1418 { 1419 unsigned int maxblocks = map->m_len; 1420 struct dnode_of_data dn; 1421 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1422 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE; 1423 pgoff_t pgofs, end_offset, end; 1424 int err = 0, ofs = 1; 1425 unsigned int ofs_in_node, last_ofs_in_node; 1426 blkcnt_t prealloc; 1427 struct extent_info ei = {0,0,0}; 1428 block_t blkaddr; 1429 unsigned int start_pgofs; 1430 1431 if (!maxblocks) 1432 return 0; 1433 1434 map->m_len = 0; 1435 map->m_flags = 0; 1436 1437 /* it only supports block size == page size */ 1438 pgofs = (pgoff_t)map->m_lblk; 1439 end = pgofs + maxblocks; 1440 1441 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 1442 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1443 map->m_may_create) 1444 goto next_dnode; 1445 1446 map->m_pblk = ei.blk + pgofs - ei.fofs; 1447 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 1448 map->m_flags = F2FS_MAP_MAPPED; 1449 if (map->m_next_extent) 1450 *map->m_next_extent = pgofs + map->m_len; 1451 1452 /* for hardware encryption, but to avoid potential issue in future */ 1453 if (flag == F2FS_GET_BLOCK_DIO) 1454 f2fs_wait_on_block_writeback_range(inode, 1455 map->m_pblk, map->m_len); 1456 goto out; 1457 } 1458 1459 next_dnode: 1460 if (map->m_may_create) 1461 __do_map_lock(sbi, flag, true); 1462 1463 /* When reading holes, we need its node page */ 1464 set_new_dnode(&dn, inode, NULL, NULL, 0); 1465 err = f2fs_get_dnode_of_data(&dn, pgofs, mode); 1466 if (err) { 1467 if (flag == F2FS_GET_BLOCK_BMAP) 1468 map->m_pblk = 0; 1469 if (err == -ENOENT) { 1470 err = 0; 1471 if (map->m_next_pgofs) 1472 *map->m_next_pgofs = 1473 f2fs_get_next_page_offset(&dn, pgofs); 1474 if (map->m_next_extent) 1475 *map->m_next_extent = 1476 f2fs_get_next_page_offset(&dn, pgofs); 1477 } 1478 goto unlock_out; 1479 } 1480 1481 start_pgofs = pgofs; 1482 prealloc = 0; 1483 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 1484 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 1485 1486 next_block: 1487 blkaddr = f2fs_data_blkaddr(&dn); 1488 1489 if (__is_valid_data_blkaddr(blkaddr) && 1490 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) { 1491 err = -EFSCORRUPTED; 1492 goto sync_out; 1493 } 1494 1495 if (__is_valid_data_blkaddr(blkaddr)) { 1496 /* use out-place-update for driect IO under LFS mode */ 1497 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1498 map->m_may_create) { 1499 err = __allocate_data_block(&dn, map->m_seg_type); 1500 if (err) 1501 goto sync_out; 1502 blkaddr = dn.data_blkaddr; 1503 set_inode_flag(inode, FI_APPEND_WRITE); 1504 } 1505 } else { 1506 if (create) { 1507 if (unlikely(f2fs_cp_error(sbi))) { 1508 err = -EIO; 1509 goto sync_out; 1510 } 1511 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1512 if (blkaddr == NULL_ADDR) { 1513 prealloc++; 1514 last_ofs_in_node = dn.ofs_in_node; 1515 } 1516 } else { 1517 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO && 1518 flag != F2FS_GET_BLOCK_DIO); 1519 err = __allocate_data_block(&dn, 1520 map->m_seg_type); 1521 if (!err) 1522 set_inode_flag(inode, FI_APPEND_WRITE); 1523 } 1524 if (err) 1525 goto sync_out; 1526 map->m_flags |= F2FS_MAP_NEW; 1527 blkaddr = dn.data_blkaddr; 1528 } else { 1529 if (flag == F2FS_GET_BLOCK_BMAP) { 1530 map->m_pblk = 0; 1531 goto sync_out; 1532 } 1533 if (flag == F2FS_GET_BLOCK_PRECACHE) 1534 goto sync_out; 1535 if (flag == F2FS_GET_BLOCK_FIEMAP && 1536 blkaddr == NULL_ADDR) { 1537 if (map->m_next_pgofs) 1538 *map->m_next_pgofs = pgofs + 1; 1539 goto sync_out; 1540 } 1541 if (flag != F2FS_GET_BLOCK_FIEMAP) { 1542 /* for defragment case */ 1543 if (map->m_next_pgofs) 1544 *map->m_next_pgofs = pgofs + 1; 1545 goto sync_out; 1546 } 1547 } 1548 } 1549 1550 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1551 goto skip; 1552 1553 if (map->m_len == 0) { 1554 /* preallocated unwritten block should be mapped for fiemap. */ 1555 if (blkaddr == NEW_ADDR) 1556 map->m_flags |= F2FS_MAP_UNWRITTEN; 1557 map->m_flags |= F2FS_MAP_MAPPED; 1558 1559 map->m_pblk = blkaddr; 1560 map->m_len = 1; 1561 } else if ((map->m_pblk != NEW_ADDR && 1562 blkaddr == (map->m_pblk + ofs)) || 1563 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1564 flag == F2FS_GET_BLOCK_PRE_DIO) { 1565 ofs++; 1566 map->m_len++; 1567 } else { 1568 goto sync_out; 1569 } 1570 1571 skip: 1572 dn.ofs_in_node++; 1573 pgofs++; 1574 1575 /* preallocate blocks in batch for one dnode page */ 1576 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1577 (pgofs == end || dn.ofs_in_node == end_offset)) { 1578 1579 dn.ofs_in_node = ofs_in_node; 1580 err = f2fs_reserve_new_blocks(&dn, prealloc); 1581 if (err) 1582 goto sync_out; 1583 1584 map->m_len += dn.ofs_in_node - ofs_in_node; 1585 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1586 err = -ENOSPC; 1587 goto sync_out; 1588 } 1589 dn.ofs_in_node = end_offset; 1590 } 1591 1592 if (pgofs >= end) 1593 goto sync_out; 1594 else if (dn.ofs_in_node < end_offset) 1595 goto next_block; 1596 1597 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1598 if (map->m_flags & F2FS_MAP_MAPPED) { 1599 unsigned int ofs = start_pgofs - map->m_lblk; 1600 1601 f2fs_update_extent_cache_range(&dn, 1602 start_pgofs, map->m_pblk + ofs, 1603 map->m_len - ofs); 1604 } 1605 } 1606 1607 f2fs_put_dnode(&dn); 1608 1609 if (map->m_may_create) { 1610 __do_map_lock(sbi, flag, false); 1611 f2fs_balance_fs(sbi, dn.node_changed); 1612 } 1613 goto next_dnode; 1614 1615 sync_out: 1616 1617 /* for hardware encryption, but to avoid potential issue in future */ 1618 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) 1619 f2fs_wait_on_block_writeback_range(inode, 1620 map->m_pblk, map->m_len); 1621 1622 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1623 if (map->m_flags & F2FS_MAP_MAPPED) { 1624 unsigned int ofs = start_pgofs - map->m_lblk; 1625 1626 f2fs_update_extent_cache_range(&dn, 1627 start_pgofs, map->m_pblk + ofs, 1628 map->m_len - ofs); 1629 } 1630 if (map->m_next_extent) 1631 *map->m_next_extent = pgofs + 1; 1632 } 1633 f2fs_put_dnode(&dn); 1634 unlock_out: 1635 if (map->m_may_create) { 1636 __do_map_lock(sbi, flag, false); 1637 f2fs_balance_fs(sbi, dn.node_changed); 1638 } 1639 out: 1640 trace_f2fs_map_blocks(inode, map, err); 1641 return err; 1642 } 1643 1644 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) 1645 { 1646 struct f2fs_map_blocks map; 1647 block_t last_lblk; 1648 int err; 1649 1650 if (pos + len > i_size_read(inode)) 1651 return false; 1652 1653 map.m_lblk = F2FS_BYTES_TO_BLK(pos); 1654 map.m_next_pgofs = NULL; 1655 map.m_next_extent = NULL; 1656 map.m_seg_type = NO_CHECK_TYPE; 1657 map.m_may_create = false; 1658 last_lblk = F2FS_BLK_ALIGN(pos + len); 1659 1660 while (map.m_lblk < last_lblk) { 1661 map.m_len = last_lblk - map.m_lblk; 1662 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT); 1663 if (err || map.m_len == 0) 1664 return false; 1665 map.m_lblk += map.m_len; 1666 } 1667 return true; 1668 } 1669 1670 static int __get_data_block(struct inode *inode, sector_t iblock, 1671 struct buffer_head *bh, int create, int flag, 1672 pgoff_t *next_pgofs, int seg_type, bool may_write) 1673 { 1674 struct f2fs_map_blocks map; 1675 int err; 1676 1677 map.m_lblk = iblock; 1678 map.m_len = bh->b_size >> inode->i_blkbits; 1679 map.m_next_pgofs = next_pgofs; 1680 map.m_next_extent = NULL; 1681 map.m_seg_type = seg_type; 1682 map.m_may_create = may_write; 1683 1684 err = f2fs_map_blocks(inode, &map, create, flag); 1685 if (!err) { 1686 map_bh(bh, inode->i_sb, map.m_pblk); 1687 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags; 1688 bh->b_size = (u64)map.m_len << inode->i_blkbits; 1689 } 1690 return err; 1691 } 1692 1693 static int get_data_block(struct inode *inode, sector_t iblock, 1694 struct buffer_head *bh_result, int create, int flag, 1695 pgoff_t *next_pgofs) 1696 { 1697 return __get_data_block(inode, iblock, bh_result, create, 1698 flag, next_pgofs, 1699 NO_CHECK_TYPE, create); 1700 } 1701 1702 static int get_data_block_dio_write(struct inode *inode, sector_t iblock, 1703 struct buffer_head *bh_result, int create) 1704 { 1705 return __get_data_block(inode, iblock, bh_result, create, 1706 F2FS_GET_BLOCK_DIO, NULL, 1707 f2fs_rw_hint_to_seg_type(inode->i_write_hint), 1708 IS_SWAPFILE(inode) ? false : true); 1709 } 1710 1711 static int get_data_block_dio(struct inode *inode, sector_t iblock, 1712 struct buffer_head *bh_result, int create) 1713 { 1714 return __get_data_block(inode, iblock, bh_result, create, 1715 F2FS_GET_BLOCK_DIO, NULL, 1716 f2fs_rw_hint_to_seg_type(inode->i_write_hint), 1717 false); 1718 } 1719 1720 static int get_data_block_bmap(struct inode *inode, sector_t iblock, 1721 struct buffer_head *bh_result, int create) 1722 { 1723 /* Block number less than F2FS MAX BLOCKS */ 1724 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks)) 1725 return -EFBIG; 1726 1727 return __get_data_block(inode, iblock, bh_result, create, 1728 F2FS_GET_BLOCK_BMAP, NULL, 1729 NO_CHECK_TYPE, create); 1730 } 1731 1732 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset) 1733 { 1734 return (offset >> inode->i_blkbits); 1735 } 1736 1737 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk) 1738 { 1739 return (blk << inode->i_blkbits); 1740 } 1741 1742 static int f2fs_xattr_fiemap(struct inode *inode, 1743 struct fiemap_extent_info *fieinfo) 1744 { 1745 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1746 struct page *page; 1747 struct node_info ni; 1748 __u64 phys = 0, len; 1749 __u32 flags; 1750 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1751 int err = 0; 1752 1753 if (f2fs_has_inline_xattr(inode)) { 1754 int offset; 1755 1756 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1757 inode->i_ino, false); 1758 if (!page) 1759 return -ENOMEM; 1760 1761 err = f2fs_get_node_info(sbi, inode->i_ino, &ni); 1762 if (err) { 1763 f2fs_put_page(page, 1); 1764 return err; 1765 } 1766 1767 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1768 offset = offsetof(struct f2fs_inode, i_addr) + 1769 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1770 get_inline_xattr_addrs(inode)); 1771 1772 phys += offset; 1773 len = inline_xattr_size(inode); 1774 1775 f2fs_put_page(page, 1); 1776 1777 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1778 1779 if (!xnid) 1780 flags |= FIEMAP_EXTENT_LAST; 1781 1782 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1783 if (err || err == 1) 1784 return err; 1785 } 1786 1787 if (xnid) { 1788 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1789 if (!page) 1790 return -ENOMEM; 1791 1792 err = f2fs_get_node_info(sbi, xnid, &ni); 1793 if (err) { 1794 f2fs_put_page(page, 1); 1795 return err; 1796 } 1797 1798 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1799 len = inode->i_sb->s_blocksize; 1800 1801 f2fs_put_page(page, 1); 1802 1803 flags = FIEMAP_EXTENT_LAST; 1804 } 1805 1806 if (phys) 1807 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1808 1809 return (err < 0 ? err : 0); 1810 } 1811 1812 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1813 u64 start, u64 len) 1814 { 1815 struct buffer_head map_bh; 1816 sector_t start_blk, last_blk; 1817 pgoff_t next_pgofs; 1818 u64 logical = 0, phys = 0, size = 0; 1819 u32 flags = 0; 1820 int ret = 0; 1821 1822 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { 1823 ret = f2fs_precache_extents(inode); 1824 if (ret) 1825 return ret; 1826 } 1827 1828 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR); 1829 if (ret) 1830 return ret; 1831 1832 inode_lock(inode); 1833 1834 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1835 ret = f2fs_xattr_fiemap(inode, fieinfo); 1836 goto out; 1837 } 1838 1839 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) { 1840 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1841 if (ret != -EAGAIN) 1842 goto out; 1843 } 1844 1845 if (logical_to_blk(inode, len) == 0) 1846 len = blk_to_logical(inode, 1); 1847 1848 start_blk = logical_to_blk(inode, start); 1849 last_blk = logical_to_blk(inode, start + len - 1); 1850 1851 next: 1852 memset(&map_bh, 0, sizeof(struct buffer_head)); 1853 map_bh.b_size = len; 1854 1855 ret = get_data_block(inode, start_blk, &map_bh, 0, 1856 F2FS_GET_BLOCK_FIEMAP, &next_pgofs); 1857 if (ret) 1858 goto out; 1859 1860 /* HOLE */ 1861 if (!buffer_mapped(&map_bh)) { 1862 start_blk = next_pgofs; 1863 1864 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode, 1865 F2FS_I_SB(inode)->max_file_blocks)) 1866 goto prep_next; 1867 1868 flags |= FIEMAP_EXTENT_LAST; 1869 } 1870 1871 if (size) { 1872 if (IS_ENCRYPTED(inode)) 1873 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1874 1875 ret = fiemap_fill_next_extent(fieinfo, logical, 1876 phys, size, flags); 1877 } 1878 1879 if (start_blk > last_blk || ret) 1880 goto out; 1881 1882 logical = blk_to_logical(inode, start_blk); 1883 phys = blk_to_logical(inode, map_bh.b_blocknr); 1884 size = map_bh.b_size; 1885 flags = 0; 1886 if (buffer_unwritten(&map_bh)) 1887 flags = FIEMAP_EXTENT_UNWRITTEN; 1888 1889 start_blk += logical_to_blk(inode, size); 1890 1891 prep_next: 1892 cond_resched(); 1893 if (fatal_signal_pending(current)) 1894 ret = -EINTR; 1895 else 1896 goto next; 1897 out: 1898 if (ret == 1) 1899 ret = 0; 1900 1901 inode_unlock(inode); 1902 return ret; 1903 } 1904 1905 static inline loff_t f2fs_readpage_limit(struct inode *inode) 1906 { 1907 if (IS_ENABLED(CONFIG_FS_VERITY) && 1908 (IS_VERITY(inode) || f2fs_verity_in_progress(inode))) 1909 return inode->i_sb->s_maxbytes; 1910 1911 return i_size_read(inode); 1912 } 1913 1914 static int f2fs_read_single_page(struct inode *inode, struct page *page, 1915 unsigned nr_pages, 1916 struct f2fs_map_blocks *map, 1917 struct bio **bio_ret, 1918 sector_t *last_block_in_bio, 1919 bool is_readahead) 1920 { 1921 struct bio *bio = *bio_ret; 1922 const unsigned blkbits = inode->i_blkbits; 1923 const unsigned blocksize = 1 << blkbits; 1924 sector_t block_in_file; 1925 sector_t last_block; 1926 sector_t last_block_in_file; 1927 sector_t block_nr; 1928 int ret = 0; 1929 1930 block_in_file = (sector_t)page_index(page); 1931 last_block = block_in_file + nr_pages; 1932 last_block_in_file = (f2fs_readpage_limit(inode) + blocksize - 1) >> 1933 blkbits; 1934 if (last_block > last_block_in_file) 1935 last_block = last_block_in_file; 1936 1937 /* just zeroing out page which is beyond EOF */ 1938 if (block_in_file >= last_block) 1939 goto zero_out; 1940 /* 1941 * Map blocks using the previous result first. 1942 */ 1943 if ((map->m_flags & F2FS_MAP_MAPPED) && 1944 block_in_file > map->m_lblk && 1945 block_in_file < (map->m_lblk + map->m_len)) 1946 goto got_it; 1947 1948 /* 1949 * Then do more f2fs_map_blocks() calls until we are 1950 * done with this page. 1951 */ 1952 map->m_lblk = block_in_file; 1953 map->m_len = last_block - block_in_file; 1954 1955 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT); 1956 if (ret) 1957 goto out; 1958 got_it: 1959 if ((map->m_flags & F2FS_MAP_MAPPED)) { 1960 block_nr = map->m_pblk + block_in_file - map->m_lblk; 1961 SetPageMappedToDisk(page); 1962 1963 if (!PageUptodate(page) && (!PageSwapCache(page) && 1964 !cleancache_get_page(page))) { 1965 SetPageUptodate(page); 1966 goto confused; 1967 } 1968 1969 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr, 1970 DATA_GENERIC_ENHANCE_READ)) { 1971 ret = -EFSCORRUPTED; 1972 goto out; 1973 } 1974 } else { 1975 zero_out: 1976 zero_user_segment(page, 0, PAGE_SIZE); 1977 if (f2fs_need_verity(inode, page->index) && 1978 !fsverity_verify_page(page)) { 1979 ret = -EIO; 1980 goto out; 1981 } 1982 if (!PageUptodate(page)) 1983 SetPageUptodate(page); 1984 unlock_page(page); 1985 goto out; 1986 } 1987 1988 /* 1989 * This page will go to BIO. Do we need to send this 1990 * BIO off first? 1991 */ 1992 if (bio && !page_is_mergeable(F2FS_I_SB(inode), bio, 1993 *last_block_in_bio, block_nr)) { 1994 submit_and_realloc: 1995 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1996 bio = NULL; 1997 } 1998 if (bio == NULL) { 1999 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages, 2000 is_readahead ? REQ_RAHEAD : 0, page->index, 2001 false); 2002 if (IS_ERR(bio)) { 2003 ret = PTR_ERR(bio); 2004 bio = NULL; 2005 goto out; 2006 } 2007 } 2008 2009 /* 2010 * If the page is under writeback, we need to wait for 2011 * its completion to see the correct decrypted data. 2012 */ 2013 f2fs_wait_on_block_writeback(inode, block_nr); 2014 2015 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2016 goto submit_and_realloc; 2017 2018 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 2019 ClearPageError(page); 2020 *last_block_in_bio = block_nr; 2021 goto out; 2022 confused: 2023 if (bio) { 2024 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2025 bio = NULL; 2026 } 2027 unlock_page(page); 2028 out: 2029 *bio_ret = bio; 2030 return ret; 2031 } 2032 2033 #ifdef CONFIG_F2FS_FS_COMPRESSION 2034 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret, 2035 unsigned nr_pages, sector_t *last_block_in_bio, 2036 bool is_readahead, bool for_write) 2037 { 2038 struct dnode_of_data dn; 2039 struct inode *inode = cc->inode; 2040 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2041 struct bio *bio = *bio_ret; 2042 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size; 2043 sector_t last_block_in_file; 2044 const unsigned blkbits = inode->i_blkbits; 2045 const unsigned blocksize = 1 << blkbits; 2046 struct decompress_io_ctx *dic = NULL; 2047 int i; 2048 int ret = 0; 2049 2050 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc)); 2051 2052 last_block_in_file = (f2fs_readpage_limit(inode) + 2053 blocksize - 1) >> blkbits; 2054 2055 /* get rid of pages beyond EOF */ 2056 for (i = 0; i < cc->cluster_size; i++) { 2057 struct page *page = cc->rpages[i]; 2058 2059 if (!page) 2060 continue; 2061 if ((sector_t)page->index >= last_block_in_file) { 2062 zero_user_segment(page, 0, PAGE_SIZE); 2063 if (!PageUptodate(page)) 2064 SetPageUptodate(page); 2065 } else if (!PageUptodate(page)) { 2066 continue; 2067 } 2068 unlock_page(page); 2069 cc->rpages[i] = NULL; 2070 cc->nr_rpages--; 2071 } 2072 2073 /* we are done since all pages are beyond EOF */ 2074 if (f2fs_cluster_is_empty(cc)) 2075 goto out; 2076 2077 set_new_dnode(&dn, inode, NULL, NULL, 0); 2078 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 2079 if (ret) 2080 goto out; 2081 2082 /* cluster was overwritten as normal cluster */ 2083 if (dn.data_blkaddr != COMPRESS_ADDR) 2084 goto out; 2085 2086 for (i = 1; i < cc->cluster_size; i++) { 2087 block_t blkaddr; 2088 2089 blkaddr = data_blkaddr(dn.inode, dn.node_page, 2090 dn.ofs_in_node + i); 2091 2092 if (!__is_valid_data_blkaddr(blkaddr)) 2093 break; 2094 2095 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) { 2096 ret = -EFAULT; 2097 goto out_put_dnode; 2098 } 2099 cc->nr_cpages++; 2100 } 2101 2102 /* nothing to decompress */ 2103 if (cc->nr_cpages == 0) { 2104 ret = 0; 2105 goto out_put_dnode; 2106 } 2107 2108 dic = f2fs_alloc_dic(cc); 2109 if (IS_ERR(dic)) { 2110 ret = PTR_ERR(dic); 2111 goto out_put_dnode; 2112 } 2113 2114 for (i = 0; i < dic->nr_cpages; i++) { 2115 struct page *page = dic->cpages[i]; 2116 block_t blkaddr; 2117 2118 blkaddr = data_blkaddr(dn.inode, dn.node_page, 2119 dn.ofs_in_node + i + 1); 2120 2121 if (bio && !page_is_mergeable(sbi, bio, 2122 *last_block_in_bio, blkaddr)) { 2123 submit_and_realloc: 2124 __submit_bio(sbi, bio, DATA); 2125 bio = NULL; 2126 } 2127 2128 if (!bio) { 2129 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages, 2130 is_readahead ? REQ_RAHEAD : 0, 2131 page->index, for_write); 2132 if (IS_ERR(bio)) { 2133 ret = PTR_ERR(bio); 2134 bio = NULL; 2135 dic->failed = true; 2136 if (refcount_sub_and_test(dic->nr_cpages - i, 2137 &dic->ref)) 2138 f2fs_decompress_end_io(dic->rpages, 2139 cc->cluster_size, true, 2140 false); 2141 f2fs_free_dic(dic); 2142 f2fs_put_dnode(&dn); 2143 *bio_ret = bio; 2144 return ret; 2145 } 2146 } 2147 2148 f2fs_wait_on_block_writeback(inode, blkaddr); 2149 2150 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2151 goto submit_and_realloc; 2152 2153 inc_page_count(sbi, F2FS_RD_DATA); 2154 ClearPageError(page); 2155 *last_block_in_bio = blkaddr; 2156 } 2157 2158 f2fs_put_dnode(&dn); 2159 2160 *bio_ret = bio; 2161 return 0; 2162 2163 out_put_dnode: 2164 f2fs_put_dnode(&dn); 2165 out: 2166 f2fs_decompress_end_io(cc->rpages, cc->cluster_size, true, false); 2167 *bio_ret = bio; 2168 return ret; 2169 } 2170 #endif 2171 2172 /* 2173 * This function was originally taken from fs/mpage.c, and customized for f2fs. 2174 * Major change was from block_size == page_size in f2fs by default. 2175 * 2176 * Note that the aops->readpages() function is ONLY used for read-ahead. If 2177 * this function ever deviates from doing just read-ahead, it should either 2178 * use ->readpage() or do the necessary surgery to decouple ->readpages() 2179 * from read-ahead. 2180 */ 2181 static int f2fs_mpage_readpages(struct inode *inode, 2182 struct readahead_control *rac, struct page *page) 2183 { 2184 struct bio *bio = NULL; 2185 sector_t last_block_in_bio = 0; 2186 struct f2fs_map_blocks map; 2187 #ifdef CONFIG_F2FS_FS_COMPRESSION 2188 struct compress_ctx cc = { 2189 .inode = inode, 2190 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2191 .cluster_size = F2FS_I(inode)->i_cluster_size, 2192 .cluster_idx = NULL_CLUSTER, 2193 .rpages = NULL, 2194 .cpages = NULL, 2195 .nr_rpages = 0, 2196 .nr_cpages = 0, 2197 }; 2198 #endif 2199 unsigned nr_pages = rac ? readahead_count(rac) : 1; 2200 unsigned max_nr_pages = nr_pages; 2201 int ret = 0; 2202 2203 map.m_pblk = 0; 2204 map.m_lblk = 0; 2205 map.m_len = 0; 2206 map.m_flags = 0; 2207 map.m_next_pgofs = NULL; 2208 map.m_next_extent = NULL; 2209 map.m_seg_type = NO_CHECK_TYPE; 2210 map.m_may_create = false; 2211 2212 for (; nr_pages; nr_pages--) { 2213 if (rac) { 2214 page = readahead_page(rac); 2215 prefetchw(&page->flags); 2216 } 2217 2218 #ifdef CONFIG_F2FS_FS_COMPRESSION 2219 if (f2fs_compressed_file(inode)) { 2220 /* there are remained comressed pages, submit them */ 2221 if (!f2fs_cluster_can_merge_page(&cc, page->index)) { 2222 ret = f2fs_read_multi_pages(&cc, &bio, 2223 max_nr_pages, 2224 &last_block_in_bio, 2225 rac != NULL, false); 2226 f2fs_destroy_compress_ctx(&cc); 2227 if (ret) 2228 goto set_error_page; 2229 } 2230 ret = f2fs_is_compressed_cluster(inode, page->index); 2231 if (ret < 0) 2232 goto set_error_page; 2233 else if (!ret) 2234 goto read_single_page; 2235 2236 ret = f2fs_init_compress_ctx(&cc); 2237 if (ret) 2238 goto set_error_page; 2239 2240 f2fs_compress_ctx_add_page(&cc, page); 2241 2242 goto next_page; 2243 } 2244 read_single_page: 2245 #endif 2246 2247 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map, 2248 &bio, &last_block_in_bio, rac); 2249 if (ret) { 2250 #ifdef CONFIG_F2FS_FS_COMPRESSION 2251 set_error_page: 2252 #endif 2253 SetPageError(page); 2254 zero_user_segment(page, 0, PAGE_SIZE); 2255 unlock_page(page); 2256 } 2257 #ifdef CONFIG_F2FS_FS_COMPRESSION 2258 next_page: 2259 #endif 2260 if (rac) 2261 put_page(page); 2262 2263 #ifdef CONFIG_F2FS_FS_COMPRESSION 2264 if (f2fs_compressed_file(inode)) { 2265 /* last page */ 2266 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) { 2267 ret = f2fs_read_multi_pages(&cc, &bio, 2268 max_nr_pages, 2269 &last_block_in_bio, 2270 rac != NULL, false); 2271 f2fs_destroy_compress_ctx(&cc); 2272 } 2273 } 2274 #endif 2275 } 2276 if (bio) 2277 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2278 return ret; 2279 } 2280 2281 static int f2fs_read_data_page(struct file *file, struct page *page) 2282 { 2283 struct inode *inode = page_file_mapping(page)->host; 2284 int ret = -EAGAIN; 2285 2286 trace_f2fs_readpage(page, DATA); 2287 2288 if (!f2fs_is_compress_backend_ready(inode)) { 2289 unlock_page(page); 2290 return -EOPNOTSUPP; 2291 } 2292 2293 /* If the file has inline data, try to read it directly */ 2294 if (f2fs_has_inline_data(inode)) 2295 ret = f2fs_read_inline_data(inode, page); 2296 if (ret == -EAGAIN) 2297 ret = f2fs_mpage_readpages(inode, NULL, page); 2298 return ret; 2299 } 2300 2301 static void f2fs_readahead(struct readahead_control *rac) 2302 { 2303 struct inode *inode = rac->mapping->host; 2304 2305 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac)); 2306 2307 if (!f2fs_is_compress_backend_ready(inode)) 2308 return; 2309 2310 /* If the file has inline data, skip readpages */ 2311 if (f2fs_has_inline_data(inode)) 2312 return; 2313 2314 f2fs_mpage_readpages(inode, rac, NULL); 2315 } 2316 2317 int f2fs_encrypt_one_page(struct f2fs_io_info *fio) 2318 { 2319 struct inode *inode = fio->page->mapping->host; 2320 struct page *mpage, *page; 2321 gfp_t gfp_flags = GFP_NOFS; 2322 2323 if (!f2fs_encrypted_file(inode)) 2324 return 0; 2325 2326 page = fio->compressed_page ? fio->compressed_page : fio->page; 2327 2328 /* wait for GCed page writeback via META_MAPPING */ 2329 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr); 2330 2331 retry_encrypt: 2332 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page, 2333 PAGE_SIZE, 0, gfp_flags); 2334 if (IS_ERR(fio->encrypted_page)) { 2335 /* flush pending IOs and wait for a while in the ENOMEM case */ 2336 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 2337 f2fs_flush_merged_writes(fio->sbi); 2338 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 2339 gfp_flags |= __GFP_NOFAIL; 2340 goto retry_encrypt; 2341 } 2342 return PTR_ERR(fio->encrypted_page); 2343 } 2344 2345 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr); 2346 if (mpage) { 2347 if (PageUptodate(mpage)) 2348 memcpy(page_address(mpage), 2349 page_address(fio->encrypted_page), PAGE_SIZE); 2350 f2fs_put_page(mpage, 1); 2351 } 2352 return 0; 2353 } 2354 2355 static inline bool check_inplace_update_policy(struct inode *inode, 2356 struct f2fs_io_info *fio) 2357 { 2358 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2359 unsigned int policy = SM_I(sbi)->ipu_policy; 2360 2361 if (policy & (0x1 << F2FS_IPU_FORCE)) 2362 return true; 2363 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi)) 2364 return true; 2365 if (policy & (0x1 << F2FS_IPU_UTIL) && 2366 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2367 return true; 2368 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) && 2369 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2370 return true; 2371 2372 /* 2373 * IPU for rewrite async pages 2374 */ 2375 if (policy & (0x1 << F2FS_IPU_ASYNC) && 2376 fio && fio->op == REQ_OP_WRITE && 2377 !(fio->op_flags & REQ_SYNC) && 2378 !IS_ENCRYPTED(inode)) 2379 return true; 2380 2381 /* this is only set during fdatasync */ 2382 if (policy & (0x1 << F2FS_IPU_FSYNC) && 2383 is_inode_flag_set(inode, FI_NEED_IPU)) 2384 return true; 2385 2386 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2387 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2388 return true; 2389 2390 return false; 2391 } 2392 2393 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) 2394 { 2395 if (f2fs_is_pinned_file(inode)) 2396 return true; 2397 2398 /* if this is cold file, we should overwrite to avoid fragmentation */ 2399 if (file_is_cold(inode)) 2400 return true; 2401 2402 return check_inplace_update_policy(inode, fio); 2403 } 2404 2405 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) 2406 { 2407 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2408 2409 if (f2fs_lfs_mode(sbi)) 2410 return true; 2411 if (S_ISDIR(inode->i_mode)) 2412 return true; 2413 if (IS_NOQUOTA(inode)) 2414 return true; 2415 if (f2fs_is_atomic_file(inode)) 2416 return true; 2417 if (fio) { 2418 if (is_cold_data(fio->page)) 2419 return true; 2420 if (IS_ATOMIC_WRITTEN_PAGE(fio->page)) 2421 return true; 2422 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2423 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2424 return true; 2425 } 2426 return false; 2427 } 2428 2429 static inline bool need_inplace_update(struct f2fs_io_info *fio) 2430 { 2431 struct inode *inode = fio->page->mapping->host; 2432 2433 if (f2fs_should_update_outplace(inode, fio)) 2434 return false; 2435 2436 return f2fs_should_update_inplace(inode, fio); 2437 } 2438 2439 int f2fs_do_write_data_page(struct f2fs_io_info *fio) 2440 { 2441 struct page *page = fio->page; 2442 struct inode *inode = page->mapping->host; 2443 struct dnode_of_data dn; 2444 struct extent_info ei = {0,0,0}; 2445 struct node_info ni; 2446 bool ipu_force = false; 2447 int err = 0; 2448 2449 set_new_dnode(&dn, inode, NULL, NULL, 0); 2450 if (need_inplace_update(fio) && 2451 f2fs_lookup_extent_cache(inode, page->index, &ei)) { 2452 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 2453 2454 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2455 DATA_GENERIC_ENHANCE)) 2456 return -EFSCORRUPTED; 2457 2458 ipu_force = true; 2459 fio->need_lock = LOCK_DONE; 2460 goto got_it; 2461 } 2462 2463 /* Deadlock due to between page->lock and f2fs_lock_op */ 2464 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 2465 return -EAGAIN; 2466 2467 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 2468 if (err) 2469 goto out; 2470 2471 fio->old_blkaddr = dn.data_blkaddr; 2472 2473 /* This page is already truncated */ 2474 if (fio->old_blkaddr == NULL_ADDR) { 2475 ClearPageUptodate(page); 2476 clear_cold_data(page); 2477 goto out_writepage; 2478 } 2479 got_it: 2480 if (__is_valid_data_blkaddr(fio->old_blkaddr) && 2481 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2482 DATA_GENERIC_ENHANCE)) { 2483 err = -EFSCORRUPTED; 2484 goto out_writepage; 2485 } 2486 /* 2487 * If current allocation needs SSR, 2488 * it had better in-place writes for updated data. 2489 */ 2490 if (ipu_force || 2491 (__is_valid_data_blkaddr(fio->old_blkaddr) && 2492 need_inplace_update(fio))) { 2493 err = f2fs_encrypt_one_page(fio); 2494 if (err) 2495 goto out_writepage; 2496 2497 set_page_writeback(page); 2498 ClearPageError(page); 2499 f2fs_put_dnode(&dn); 2500 if (fio->need_lock == LOCK_REQ) 2501 f2fs_unlock_op(fio->sbi); 2502 err = f2fs_inplace_write_data(fio); 2503 if (err) { 2504 if (f2fs_encrypted_file(inode)) 2505 fscrypt_finalize_bounce_page(&fio->encrypted_page); 2506 if (PageWriteback(page)) 2507 end_page_writeback(page); 2508 } else { 2509 set_inode_flag(inode, FI_UPDATE_WRITE); 2510 } 2511 trace_f2fs_do_write_data_page(fio->page, IPU); 2512 return err; 2513 } 2514 2515 if (fio->need_lock == LOCK_RETRY) { 2516 if (!f2fs_trylock_op(fio->sbi)) { 2517 err = -EAGAIN; 2518 goto out_writepage; 2519 } 2520 fio->need_lock = LOCK_REQ; 2521 } 2522 2523 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni); 2524 if (err) 2525 goto out_writepage; 2526 2527 fio->version = ni.version; 2528 2529 err = f2fs_encrypt_one_page(fio); 2530 if (err) 2531 goto out_writepage; 2532 2533 set_page_writeback(page); 2534 ClearPageError(page); 2535 2536 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR) 2537 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false); 2538 2539 /* LFS mode write path */ 2540 f2fs_outplace_write_data(&dn, fio); 2541 trace_f2fs_do_write_data_page(page, OPU); 2542 set_inode_flag(inode, FI_APPEND_WRITE); 2543 if (page->index == 0) 2544 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 2545 out_writepage: 2546 f2fs_put_dnode(&dn); 2547 out: 2548 if (fio->need_lock == LOCK_REQ) 2549 f2fs_unlock_op(fio->sbi); 2550 return err; 2551 } 2552 2553 int f2fs_write_single_data_page(struct page *page, int *submitted, 2554 struct bio **bio, 2555 sector_t *last_block, 2556 struct writeback_control *wbc, 2557 enum iostat_type io_type, 2558 int compr_blocks) 2559 { 2560 struct inode *inode = page->mapping->host; 2561 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2562 loff_t i_size = i_size_read(inode); 2563 const pgoff_t end_index = ((unsigned long long)i_size) 2564 >> PAGE_SHIFT; 2565 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT; 2566 unsigned offset = 0; 2567 bool need_balance_fs = false; 2568 int err = 0; 2569 struct f2fs_io_info fio = { 2570 .sbi = sbi, 2571 .ino = inode->i_ino, 2572 .type = DATA, 2573 .op = REQ_OP_WRITE, 2574 .op_flags = wbc_to_write_flags(wbc), 2575 .old_blkaddr = NULL_ADDR, 2576 .page = page, 2577 .encrypted_page = NULL, 2578 .submitted = false, 2579 .compr_blocks = compr_blocks, 2580 .need_lock = LOCK_RETRY, 2581 .io_type = io_type, 2582 .io_wbc = wbc, 2583 .bio = bio, 2584 .last_block = last_block, 2585 }; 2586 2587 trace_f2fs_writepage(page, DATA); 2588 2589 /* we should bypass data pages to proceed the kworkder jobs */ 2590 if (unlikely(f2fs_cp_error(sbi))) { 2591 mapping_set_error(page->mapping, -EIO); 2592 /* 2593 * don't drop any dirty dentry pages for keeping lastest 2594 * directory structure. 2595 */ 2596 if (S_ISDIR(inode->i_mode)) 2597 goto redirty_out; 2598 goto out; 2599 } 2600 2601 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2602 goto redirty_out; 2603 2604 if (page->index < end_index || 2605 f2fs_verity_in_progress(inode) || 2606 compr_blocks) 2607 goto write; 2608 2609 /* 2610 * If the offset is out-of-range of file size, 2611 * this page does not have to be written to disk. 2612 */ 2613 offset = i_size & (PAGE_SIZE - 1); 2614 if ((page->index >= end_index + 1) || !offset) 2615 goto out; 2616 2617 zero_user_segment(page, offset, PAGE_SIZE); 2618 write: 2619 if (f2fs_is_drop_cache(inode)) 2620 goto out; 2621 /* we should not write 0'th page having journal header */ 2622 if (f2fs_is_volatile_file(inode) && (!page->index || 2623 (!wbc->for_reclaim && 2624 f2fs_available_free_memory(sbi, BASE_CHECK)))) 2625 goto redirty_out; 2626 2627 /* Dentry blocks are controlled by checkpoint */ 2628 if (S_ISDIR(inode->i_mode)) { 2629 fio.need_lock = LOCK_DONE; 2630 err = f2fs_do_write_data_page(&fio); 2631 goto done; 2632 } 2633 2634 if (!wbc->for_reclaim) 2635 need_balance_fs = true; 2636 else if (has_not_enough_free_secs(sbi, 0, 0)) 2637 goto redirty_out; 2638 else 2639 set_inode_flag(inode, FI_HOT_DATA); 2640 2641 err = -EAGAIN; 2642 if (f2fs_has_inline_data(inode)) { 2643 err = f2fs_write_inline_data(inode, page); 2644 if (!err) 2645 goto out; 2646 } 2647 2648 if (err == -EAGAIN) { 2649 err = f2fs_do_write_data_page(&fio); 2650 if (err == -EAGAIN) { 2651 fio.need_lock = LOCK_REQ; 2652 err = f2fs_do_write_data_page(&fio); 2653 } 2654 } 2655 2656 if (err) { 2657 file_set_keep_isize(inode); 2658 } else { 2659 spin_lock(&F2FS_I(inode)->i_size_lock); 2660 if (F2FS_I(inode)->last_disk_size < psize) 2661 F2FS_I(inode)->last_disk_size = psize; 2662 spin_unlock(&F2FS_I(inode)->i_size_lock); 2663 } 2664 2665 done: 2666 if (err && err != -ENOENT) 2667 goto redirty_out; 2668 2669 out: 2670 inode_dec_dirty_pages(inode); 2671 if (err) { 2672 ClearPageUptodate(page); 2673 clear_cold_data(page); 2674 } 2675 2676 if (wbc->for_reclaim) { 2677 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA); 2678 clear_inode_flag(inode, FI_HOT_DATA); 2679 f2fs_remove_dirty_inode(inode); 2680 submitted = NULL; 2681 } 2682 unlock_page(page); 2683 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) && 2684 !F2FS_I(inode)->cp_task) 2685 f2fs_balance_fs(sbi, need_balance_fs); 2686 2687 if (unlikely(f2fs_cp_error(sbi))) { 2688 f2fs_submit_merged_write(sbi, DATA); 2689 f2fs_submit_merged_ipu_write(sbi, bio, NULL); 2690 submitted = NULL; 2691 } 2692 2693 if (submitted) 2694 *submitted = fio.submitted ? 1 : 0; 2695 2696 return 0; 2697 2698 redirty_out: 2699 redirty_page_for_writepage(wbc, page); 2700 /* 2701 * pageout() in MM traslates EAGAIN, so calls handle_write_error() 2702 * -> mapping_set_error() -> set_bit(AS_EIO, ...). 2703 * file_write_and_wait_range() will see EIO error, which is critical 2704 * to return value of fsync() followed by atomic_write failure to user. 2705 */ 2706 if (!err || wbc->for_reclaim) 2707 return AOP_WRITEPAGE_ACTIVATE; 2708 unlock_page(page); 2709 return err; 2710 } 2711 2712 static int f2fs_write_data_page(struct page *page, 2713 struct writeback_control *wbc) 2714 { 2715 #ifdef CONFIG_F2FS_FS_COMPRESSION 2716 struct inode *inode = page->mapping->host; 2717 2718 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 2719 goto out; 2720 2721 if (f2fs_compressed_file(inode)) { 2722 if (f2fs_is_compressed_cluster(inode, page->index)) { 2723 redirty_page_for_writepage(wbc, page); 2724 return AOP_WRITEPAGE_ACTIVATE; 2725 } 2726 } 2727 out: 2728 #endif 2729 2730 return f2fs_write_single_data_page(page, NULL, NULL, NULL, 2731 wbc, FS_DATA_IO, 0); 2732 } 2733 2734 /* 2735 * This function was copied from write_cche_pages from mm/page-writeback.c. 2736 * The major change is making write step of cold data page separately from 2737 * warm/hot data page. 2738 */ 2739 static int f2fs_write_cache_pages(struct address_space *mapping, 2740 struct writeback_control *wbc, 2741 enum iostat_type io_type) 2742 { 2743 int ret = 0; 2744 int done = 0, retry = 0; 2745 struct pagevec pvec; 2746 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2747 struct bio *bio = NULL; 2748 sector_t last_block; 2749 #ifdef CONFIG_F2FS_FS_COMPRESSION 2750 struct inode *inode = mapping->host; 2751 struct compress_ctx cc = { 2752 .inode = inode, 2753 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2754 .cluster_size = F2FS_I(inode)->i_cluster_size, 2755 .cluster_idx = NULL_CLUSTER, 2756 .rpages = NULL, 2757 .nr_rpages = 0, 2758 .cpages = NULL, 2759 .rbuf = NULL, 2760 .cbuf = NULL, 2761 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size, 2762 .private = NULL, 2763 }; 2764 #endif 2765 int nr_pages; 2766 pgoff_t uninitialized_var(writeback_index); 2767 pgoff_t index; 2768 pgoff_t end; /* Inclusive */ 2769 pgoff_t done_index; 2770 int cycled; 2771 int range_whole = 0; 2772 xa_mark_t tag; 2773 int nwritten = 0; 2774 int submitted = 0; 2775 int i; 2776 2777 pagevec_init(&pvec); 2778 2779 if (get_dirty_pages(mapping->host) <= 2780 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 2781 set_inode_flag(mapping->host, FI_HOT_DATA); 2782 else 2783 clear_inode_flag(mapping->host, FI_HOT_DATA); 2784 2785 if (wbc->range_cyclic) { 2786 writeback_index = mapping->writeback_index; /* prev offset */ 2787 index = writeback_index; 2788 if (index == 0) 2789 cycled = 1; 2790 else 2791 cycled = 0; 2792 end = -1; 2793 } else { 2794 index = wbc->range_start >> PAGE_SHIFT; 2795 end = wbc->range_end >> PAGE_SHIFT; 2796 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2797 range_whole = 1; 2798 cycled = 1; /* ignore range_cyclic tests */ 2799 } 2800 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2801 tag = PAGECACHE_TAG_TOWRITE; 2802 else 2803 tag = PAGECACHE_TAG_DIRTY; 2804 retry: 2805 retry = 0; 2806 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2807 tag_pages_for_writeback(mapping, index, end); 2808 done_index = index; 2809 while (!done && !retry && (index <= end)) { 2810 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 2811 tag); 2812 if (nr_pages == 0) 2813 break; 2814 2815 for (i = 0; i < nr_pages; i++) { 2816 struct page *page = pvec.pages[i]; 2817 bool need_readd; 2818 readd: 2819 need_readd = false; 2820 #ifdef CONFIG_F2FS_FS_COMPRESSION 2821 if (f2fs_compressed_file(inode)) { 2822 ret = f2fs_init_compress_ctx(&cc); 2823 if (ret) { 2824 done = 1; 2825 break; 2826 } 2827 2828 if (!f2fs_cluster_can_merge_page(&cc, 2829 page->index)) { 2830 ret = f2fs_write_multi_pages(&cc, 2831 &submitted, wbc, io_type); 2832 if (!ret) 2833 need_readd = true; 2834 goto result; 2835 } 2836 2837 if (unlikely(f2fs_cp_error(sbi))) 2838 goto lock_page; 2839 2840 if (f2fs_cluster_is_empty(&cc)) { 2841 void *fsdata = NULL; 2842 struct page *pagep; 2843 int ret2; 2844 2845 ret2 = f2fs_prepare_compress_overwrite( 2846 inode, &pagep, 2847 page->index, &fsdata); 2848 if (ret2 < 0) { 2849 ret = ret2; 2850 done = 1; 2851 break; 2852 } else if (ret2 && 2853 !f2fs_compress_write_end(inode, 2854 fsdata, page->index, 2855 1)) { 2856 retry = 1; 2857 break; 2858 } 2859 } else { 2860 goto lock_page; 2861 } 2862 } 2863 #endif 2864 /* give a priority to WB_SYNC threads */ 2865 if (atomic_read(&sbi->wb_sync_req[DATA]) && 2866 wbc->sync_mode == WB_SYNC_NONE) { 2867 done = 1; 2868 break; 2869 } 2870 #ifdef CONFIG_F2FS_FS_COMPRESSION 2871 lock_page: 2872 #endif 2873 done_index = page->index; 2874 retry_write: 2875 lock_page(page); 2876 2877 if (unlikely(page->mapping != mapping)) { 2878 continue_unlock: 2879 unlock_page(page); 2880 continue; 2881 } 2882 2883 if (!PageDirty(page)) { 2884 /* someone wrote it for us */ 2885 goto continue_unlock; 2886 } 2887 2888 if (PageWriteback(page)) { 2889 if (wbc->sync_mode != WB_SYNC_NONE) 2890 f2fs_wait_on_page_writeback(page, 2891 DATA, true, true); 2892 else 2893 goto continue_unlock; 2894 } 2895 2896 if (!clear_page_dirty_for_io(page)) 2897 goto continue_unlock; 2898 2899 #ifdef CONFIG_F2FS_FS_COMPRESSION 2900 if (f2fs_compressed_file(inode)) { 2901 get_page(page); 2902 f2fs_compress_ctx_add_page(&cc, page); 2903 continue; 2904 } 2905 #endif 2906 ret = f2fs_write_single_data_page(page, &submitted, 2907 &bio, &last_block, wbc, io_type, 0); 2908 if (ret == AOP_WRITEPAGE_ACTIVATE) 2909 unlock_page(page); 2910 #ifdef CONFIG_F2FS_FS_COMPRESSION 2911 result: 2912 #endif 2913 nwritten += submitted; 2914 wbc->nr_to_write -= submitted; 2915 2916 if (unlikely(ret)) { 2917 /* 2918 * keep nr_to_write, since vfs uses this to 2919 * get # of written pages. 2920 */ 2921 if (ret == AOP_WRITEPAGE_ACTIVATE) { 2922 ret = 0; 2923 goto next; 2924 } else if (ret == -EAGAIN) { 2925 ret = 0; 2926 if (wbc->sync_mode == WB_SYNC_ALL) { 2927 cond_resched(); 2928 congestion_wait(BLK_RW_ASYNC, 2929 DEFAULT_IO_TIMEOUT); 2930 goto retry_write; 2931 } 2932 goto next; 2933 } 2934 done_index = page->index + 1; 2935 done = 1; 2936 break; 2937 } 2938 2939 if (wbc->nr_to_write <= 0 && 2940 wbc->sync_mode == WB_SYNC_NONE) { 2941 done = 1; 2942 break; 2943 } 2944 next: 2945 if (need_readd) 2946 goto readd; 2947 } 2948 pagevec_release(&pvec); 2949 cond_resched(); 2950 } 2951 #ifdef CONFIG_F2FS_FS_COMPRESSION 2952 /* flush remained pages in compress cluster */ 2953 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) { 2954 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type); 2955 nwritten += submitted; 2956 wbc->nr_to_write -= submitted; 2957 if (ret) { 2958 done = 1; 2959 retry = 0; 2960 } 2961 } 2962 #endif 2963 if ((!cycled && !done) || retry) { 2964 cycled = 1; 2965 index = 0; 2966 end = writeback_index - 1; 2967 goto retry; 2968 } 2969 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 2970 mapping->writeback_index = done_index; 2971 2972 if (nwritten) 2973 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 2974 NULL, 0, DATA); 2975 /* submit cached bio of IPU write */ 2976 if (bio) 2977 f2fs_submit_merged_ipu_write(sbi, &bio, NULL); 2978 2979 return ret; 2980 } 2981 2982 static inline bool __should_serialize_io(struct inode *inode, 2983 struct writeback_control *wbc) 2984 { 2985 /* to avoid deadlock in path of data flush */ 2986 if (F2FS_I(inode)->cp_task) 2987 return false; 2988 2989 if (!S_ISREG(inode->i_mode)) 2990 return false; 2991 if (IS_NOQUOTA(inode)) 2992 return false; 2993 2994 if (f2fs_compressed_file(inode)) 2995 return true; 2996 if (wbc->sync_mode != WB_SYNC_ALL) 2997 return true; 2998 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks) 2999 return true; 3000 return false; 3001 } 3002 3003 static int __f2fs_write_data_pages(struct address_space *mapping, 3004 struct writeback_control *wbc, 3005 enum iostat_type io_type) 3006 { 3007 struct inode *inode = mapping->host; 3008 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3009 struct blk_plug plug; 3010 int ret; 3011 bool locked = false; 3012 3013 /* deal with chardevs and other special file */ 3014 if (!mapping->a_ops->writepage) 3015 return 0; 3016 3017 /* skip writing if there is no dirty page in this inode */ 3018 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 3019 return 0; 3020 3021 /* during POR, we don't need to trigger writepage at all. */ 3022 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 3023 goto skip_write; 3024 3025 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) && 3026 wbc->sync_mode == WB_SYNC_NONE && 3027 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 3028 f2fs_available_free_memory(sbi, DIRTY_DENTS)) 3029 goto skip_write; 3030 3031 /* skip writing during file defragment */ 3032 if (is_inode_flag_set(inode, FI_DO_DEFRAG)) 3033 goto skip_write; 3034 3035 trace_f2fs_writepages(mapping->host, wbc, DATA); 3036 3037 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 3038 if (wbc->sync_mode == WB_SYNC_ALL) 3039 atomic_inc(&sbi->wb_sync_req[DATA]); 3040 else if (atomic_read(&sbi->wb_sync_req[DATA])) 3041 goto skip_write; 3042 3043 if (__should_serialize_io(inode, wbc)) { 3044 mutex_lock(&sbi->writepages); 3045 locked = true; 3046 } 3047 3048 blk_start_plug(&plug); 3049 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 3050 blk_finish_plug(&plug); 3051 3052 if (locked) 3053 mutex_unlock(&sbi->writepages); 3054 3055 if (wbc->sync_mode == WB_SYNC_ALL) 3056 atomic_dec(&sbi->wb_sync_req[DATA]); 3057 /* 3058 * if some pages were truncated, we cannot guarantee its mapping->host 3059 * to detect pending bios. 3060 */ 3061 3062 f2fs_remove_dirty_inode(inode); 3063 return ret; 3064 3065 skip_write: 3066 wbc->pages_skipped += get_dirty_pages(inode); 3067 trace_f2fs_writepages(mapping->host, wbc, DATA); 3068 return 0; 3069 } 3070 3071 static int f2fs_write_data_pages(struct address_space *mapping, 3072 struct writeback_control *wbc) 3073 { 3074 struct inode *inode = mapping->host; 3075 3076 return __f2fs_write_data_pages(mapping, wbc, 3077 F2FS_I(inode)->cp_task == current ? 3078 FS_CP_DATA_IO : FS_DATA_IO); 3079 } 3080 3081 static void f2fs_write_failed(struct address_space *mapping, loff_t to) 3082 { 3083 struct inode *inode = mapping->host; 3084 loff_t i_size = i_size_read(inode); 3085 3086 if (IS_NOQUOTA(inode)) 3087 return; 3088 3089 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */ 3090 if (to > i_size && !f2fs_verity_in_progress(inode)) { 3091 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3092 down_write(&F2FS_I(inode)->i_mmap_sem); 3093 3094 truncate_pagecache(inode, i_size); 3095 f2fs_truncate_blocks(inode, i_size, true); 3096 3097 up_write(&F2FS_I(inode)->i_mmap_sem); 3098 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3099 } 3100 } 3101 3102 static int prepare_write_begin(struct f2fs_sb_info *sbi, 3103 struct page *page, loff_t pos, unsigned len, 3104 block_t *blk_addr, bool *node_changed) 3105 { 3106 struct inode *inode = page->mapping->host; 3107 pgoff_t index = page->index; 3108 struct dnode_of_data dn; 3109 struct page *ipage; 3110 bool locked = false; 3111 struct extent_info ei = {0,0,0}; 3112 int err = 0; 3113 int flag; 3114 3115 /* 3116 * we already allocated all the blocks, so we don't need to get 3117 * the block addresses when there is no need to fill the page. 3118 */ 3119 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE && 3120 !is_inode_flag_set(inode, FI_NO_PREALLOC) && 3121 !f2fs_verity_in_progress(inode)) 3122 return 0; 3123 3124 /* f2fs_lock_op avoids race between write CP and convert_inline_page */ 3125 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode)) 3126 flag = F2FS_GET_BLOCK_DEFAULT; 3127 else 3128 flag = F2FS_GET_BLOCK_PRE_AIO; 3129 3130 if (f2fs_has_inline_data(inode) || 3131 (pos & PAGE_MASK) >= i_size_read(inode)) { 3132 __do_map_lock(sbi, flag, true); 3133 locked = true; 3134 } 3135 3136 restart: 3137 /* check inline_data */ 3138 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3139 if (IS_ERR(ipage)) { 3140 err = PTR_ERR(ipage); 3141 goto unlock_out; 3142 } 3143 3144 set_new_dnode(&dn, inode, ipage, ipage, 0); 3145 3146 if (f2fs_has_inline_data(inode)) { 3147 if (pos + len <= MAX_INLINE_DATA(inode)) { 3148 f2fs_do_read_inline_data(page, ipage); 3149 set_inode_flag(inode, FI_DATA_EXIST); 3150 if (inode->i_nlink) 3151 set_inline_node(ipage); 3152 } else { 3153 err = f2fs_convert_inline_page(&dn, page); 3154 if (err) 3155 goto out; 3156 if (dn.data_blkaddr == NULL_ADDR) 3157 err = f2fs_get_block(&dn, index); 3158 } 3159 } else if (locked) { 3160 err = f2fs_get_block(&dn, index); 3161 } else { 3162 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 3163 dn.data_blkaddr = ei.blk + index - ei.fofs; 3164 } else { 3165 /* hole case */ 3166 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3167 if (err || dn.data_blkaddr == NULL_ADDR) { 3168 f2fs_put_dnode(&dn); 3169 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 3170 true); 3171 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO); 3172 locked = true; 3173 goto restart; 3174 } 3175 } 3176 } 3177 3178 /* convert_inline_page can make node_changed */ 3179 *blk_addr = dn.data_blkaddr; 3180 *node_changed = dn.node_changed; 3181 out: 3182 f2fs_put_dnode(&dn); 3183 unlock_out: 3184 if (locked) 3185 __do_map_lock(sbi, flag, false); 3186 return err; 3187 } 3188 3189 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 3190 loff_t pos, unsigned len, unsigned flags, 3191 struct page **pagep, void **fsdata) 3192 { 3193 struct inode *inode = mapping->host; 3194 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3195 struct page *page = NULL; 3196 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 3197 bool need_balance = false, drop_atomic = false; 3198 block_t blkaddr = NULL_ADDR; 3199 int err = 0; 3200 3201 trace_f2fs_write_begin(inode, pos, len, flags); 3202 3203 if (!f2fs_is_checkpoint_ready(sbi)) { 3204 err = -ENOSPC; 3205 goto fail; 3206 } 3207 3208 if ((f2fs_is_atomic_file(inode) && 3209 !f2fs_available_free_memory(sbi, INMEM_PAGES)) || 3210 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) { 3211 err = -ENOMEM; 3212 drop_atomic = true; 3213 goto fail; 3214 } 3215 3216 /* 3217 * We should check this at this moment to avoid deadlock on inode page 3218 * and #0 page. The locking rule for inline_data conversion should be: 3219 * lock_page(page #0) -> lock_page(inode_page) 3220 */ 3221 if (index != 0) { 3222 err = f2fs_convert_inline_inode(inode); 3223 if (err) 3224 goto fail; 3225 } 3226 3227 #ifdef CONFIG_F2FS_FS_COMPRESSION 3228 if (f2fs_compressed_file(inode)) { 3229 int ret; 3230 3231 *fsdata = NULL; 3232 3233 ret = f2fs_prepare_compress_overwrite(inode, pagep, 3234 index, fsdata); 3235 if (ret < 0) { 3236 err = ret; 3237 goto fail; 3238 } else if (ret) { 3239 return 0; 3240 } 3241 } 3242 #endif 3243 3244 repeat: 3245 /* 3246 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 3247 * wait_for_stable_page. Will wait that below with our IO control. 3248 */ 3249 page = f2fs_pagecache_get_page(mapping, index, 3250 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 3251 if (!page) { 3252 err = -ENOMEM; 3253 goto fail; 3254 } 3255 3256 /* TODO: cluster can be compressed due to race with .writepage */ 3257 3258 *pagep = page; 3259 3260 err = prepare_write_begin(sbi, page, pos, len, 3261 &blkaddr, &need_balance); 3262 if (err) 3263 goto fail; 3264 3265 if (need_balance && !IS_NOQUOTA(inode) && 3266 has_not_enough_free_secs(sbi, 0, 0)) { 3267 unlock_page(page); 3268 f2fs_balance_fs(sbi, true); 3269 lock_page(page); 3270 if (page->mapping != mapping) { 3271 /* The page got truncated from under us */ 3272 f2fs_put_page(page, 1); 3273 goto repeat; 3274 } 3275 } 3276 3277 f2fs_wait_on_page_writeback(page, DATA, false, true); 3278 3279 if (len == PAGE_SIZE || PageUptodate(page)) 3280 return 0; 3281 3282 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) && 3283 !f2fs_verity_in_progress(inode)) { 3284 zero_user_segment(page, len, PAGE_SIZE); 3285 return 0; 3286 } 3287 3288 if (blkaddr == NEW_ADDR) { 3289 zero_user_segment(page, 0, PAGE_SIZE); 3290 SetPageUptodate(page); 3291 } else { 3292 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 3293 DATA_GENERIC_ENHANCE_READ)) { 3294 err = -EFSCORRUPTED; 3295 goto fail; 3296 } 3297 err = f2fs_submit_page_read(inode, page, blkaddr, true); 3298 if (err) 3299 goto fail; 3300 3301 lock_page(page); 3302 if (unlikely(page->mapping != mapping)) { 3303 f2fs_put_page(page, 1); 3304 goto repeat; 3305 } 3306 if (unlikely(!PageUptodate(page))) { 3307 err = -EIO; 3308 goto fail; 3309 } 3310 } 3311 return 0; 3312 3313 fail: 3314 f2fs_put_page(page, 1); 3315 f2fs_write_failed(mapping, pos + len); 3316 if (drop_atomic) 3317 f2fs_drop_inmem_pages_all(sbi, false); 3318 return err; 3319 } 3320 3321 static int f2fs_write_end(struct file *file, 3322 struct address_space *mapping, 3323 loff_t pos, unsigned len, unsigned copied, 3324 struct page *page, void *fsdata) 3325 { 3326 struct inode *inode = page->mapping->host; 3327 3328 trace_f2fs_write_end(inode, pos, len, copied); 3329 3330 /* 3331 * This should be come from len == PAGE_SIZE, and we expect copied 3332 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 3333 * let generic_perform_write() try to copy data again through copied=0. 3334 */ 3335 if (!PageUptodate(page)) { 3336 if (unlikely(copied != len)) 3337 copied = 0; 3338 else 3339 SetPageUptodate(page); 3340 } 3341 3342 #ifdef CONFIG_F2FS_FS_COMPRESSION 3343 /* overwrite compressed file */ 3344 if (f2fs_compressed_file(inode) && fsdata) { 3345 f2fs_compress_write_end(inode, fsdata, page->index, copied); 3346 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3347 return copied; 3348 } 3349 #endif 3350 3351 if (!copied) 3352 goto unlock_out; 3353 3354 set_page_dirty(page); 3355 3356 if (pos + copied > i_size_read(inode) && 3357 !f2fs_verity_in_progress(inode)) 3358 f2fs_i_size_write(inode, pos + copied); 3359 unlock_out: 3360 f2fs_put_page(page, 1); 3361 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3362 return copied; 3363 } 3364 3365 static int check_direct_IO(struct inode *inode, struct iov_iter *iter, 3366 loff_t offset) 3367 { 3368 unsigned i_blkbits = READ_ONCE(inode->i_blkbits); 3369 unsigned blkbits = i_blkbits; 3370 unsigned blocksize_mask = (1 << blkbits) - 1; 3371 unsigned long align = offset | iov_iter_alignment(iter); 3372 struct block_device *bdev = inode->i_sb->s_bdev; 3373 3374 if (align & blocksize_mask) { 3375 if (bdev) 3376 blkbits = blksize_bits(bdev_logical_block_size(bdev)); 3377 blocksize_mask = (1 << blkbits) - 1; 3378 if (align & blocksize_mask) 3379 return -EINVAL; 3380 return 1; 3381 } 3382 return 0; 3383 } 3384 3385 static void f2fs_dio_end_io(struct bio *bio) 3386 { 3387 struct f2fs_private_dio *dio = bio->bi_private; 3388 3389 dec_page_count(F2FS_I_SB(dio->inode), 3390 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ); 3391 3392 bio->bi_private = dio->orig_private; 3393 bio->bi_end_io = dio->orig_end_io; 3394 3395 kvfree(dio); 3396 3397 bio_endio(bio); 3398 } 3399 3400 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode, 3401 loff_t file_offset) 3402 { 3403 struct f2fs_private_dio *dio; 3404 bool write = (bio_op(bio) == REQ_OP_WRITE); 3405 3406 dio = f2fs_kzalloc(F2FS_I_SB(inode), 3407 sizeof(struct f2fs_private_dio), GFP_NOFS); 3408 if (!dio) 3409 goto out; 3410 3411 dio->inode = inode; 3412 dio->orig_end_io = bio->bi_end_io; 3413 dio->orig_private = bio->bi_private; 3414 dio->write = write; 3415 3416 bio->bi_end_io = f2fs_dio_end_io; 3417 bio->bi_private = dio; 3418 3419 inc_page_count(F2FS_I_SB(inode), 3420 write ? F2FS_DIO_WRITE : F2FS_DIO_READ); 3421 3422 submit_bio(bio); 3423 return; 3424 out: 3425 bio->bi_status = BLK_STS_IOERR; 3426 bio_endio(bio); 3427 } 3428 3429 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 3430 { 3431 struct address_space *mapping = iocb->ki_filp->f_mapping; 3432 struct inode *inode = mapping->host; 3433 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3434 struct f2fs_inode_info *fi = F2FS_I(inode); 3435 size_t count = iov_iter_count(iter); 3436 loff_t offset = iocb->ki_pos; 3437 int rw = iov_iter_rw(iter); 3438 int err; 3439 enum rw_hint hint = iocb->ki_hint; 3440 int whint_mode = F2FS_OPTION(sbi).whint_mode; 3441 bool do_opu; 3442 3443 err = check_direct_IO(inode, iter, offset); 3444 if (err) 3445 return err < 0 ? err : 0; 3446 3447 if (f2fs_force_buffered_io(inode, iocb, iter)) 3448 return 0; 3449 3450 do_opu = allow_outplace_dio(inode, iocb, iter); 3451 3452 trace_f2fs_direct_IO_enter(inode, offset, count, rw); 3453 3454 if (rw == WRITE && whint_mode == WHINT_MODE_OFF) 3455 iocb->ki_hint = WRITE_LIFE_NOT_SET; 3456 3457 if (iocb->ki_flags & IOCB_NOWAIT) { 3458 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) { 3459 iocb->ki_hint = hint; 3460 err = -EAGAIN; 3461 goto out; 3462 } 3463 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) { 3464 up_read(&fi->i_gc_rwsem[rw]); 3465 iocb->ki_hint = hint; 3466 err = -EAGAIN; 3467 goto out; 3468 } 3469 } else { 3470 down_read(&fi->i_gc_rwsem[rw]); 3471 if (do_opu) 3472 down_read(&fi->i_gc_rwsem[READ]); 3473 } 3474 3475 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, 3476 iter, rw == WRITE ? get_data_block_dio_write : 3477 get_data_block_dio, NULL, f2fs_dio_submit_bio, 3478 rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES : 3479 DIO_SKIP_HOLES); 3480 3481 if (do_opu) 3482 up_read(&fi->i_gc_rwsem[READ]); 3483 3484 up_read(&fi->i_gc_rwsem[rw]); 3485 3486 if (rw == WRITE) { 3487 if (whint_mode == WHINT_MODE_OFF) 3488 iocb->ki_hint = hint; 3489 if (err > 0) { 3490 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO, 3491 err); 3492 if (!do_opu) 3493 set_inode_flag(inode, FI_UPDATE_WRITE); 3494 } else if (err < 0) { 3495 f2fs_write_failed(mapping, offset + count); 3496 } 3497 } 3498 3499 out: 3500 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); 3501 3502 return err; 3503 } 3504 3505 void f2fs_invalidate_page(struct page *page, unsigned int offset, 3506 unsigned int length) 3507 { 3508 struct inode *inode = page->mapping->host; 3509 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3510 3511 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 3512 (offset % PAGE_SIZE || length != PAGE_SIZE)) 3513 return; 3514 3515 if (PageDirty(page)) { 3516 if (inode->i_ino == F2FS_META_INO(sbi)) { 3517 dec_page_count(sbi, F2FS_DIRTY_META); 3518 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 3519 dec_page_count(sbi, F2FS_DIRTY_NODES); 3520 } else { 3521 inode_dec_dirty_pages(inode); 3522 f2fs_remove_dirty_inode(inode); 3523 } 3524 } 3525 3526 clear_cold_data(page); 3527 3528 if (IS_ATOMIC_WRITTEN_PAGE(page)) 3529 return f2fs_drop_inmem_page(inode, page); 3530 3531 f2fs_clear_page_private(page); 3532 } 3533 3534 int f2fs_release_page(struct page *page, gfp_t wait) 3535 { 3536 /* If this is dirty page, keep PagePrivate */ 3537 if (PageDirty(page)) 3538 return 0; 3539 3540 /* This is atomic written page, keep Private */ 3541 if (IS_ATOMIC_WRITTEN_PAGE(page)) 3542 return 0; 3543 3544 clear_cold_data(page); 3545 f2fs_clear_page_private(page); 3546 return 1; 3547 } 3548 3549 static int f2fs_set_data_page_dirty(struct page *page) 3550 { 3551 struct inode *inode = page_file_mapping(page)->host; 3552 3553 trace_f2fs_set_page_dirty(page, DATA); 3554 3555 if (!PageUptodate(page)) 3556 SetPageUptodate(page); 3557 if (PageSwapCache(page)) 3558 return __set_page_dirty_nobuffers(page); 3559 3560 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) { 3561 if (!IS_ATOMIC_WRITTEN_PAGE(page)) { 3562 f2fs_register_inmem_page(inode, page); 3563 return 1; 3564 } 3565 /* 3566 * Previously, this page has been registered, we just 3567 * return here. 3568 */ 3569 return 0; 3570 } 3571 3572 if (!PageDirty(page)) { 3573 __set_page_dirty_nobuffers(page); 3574 f2fs_update_dirty_page(inode, page); 3575 return 1; 3576 } 3577 return 0; 3578 } 3579 3580 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 3581 { 3582 struct inode *inode = mapping->host; 3583 3584 if (f2fs_has_inline_data(inode)) 3585 return 0; 3586 3587 /* make sure allocating whole blocks */ 3588 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 3589 filemap_write_and_wait(mapping); 3590 3591 return generic_block_bmap(mapping, block, get_data_block_bmap); 3592 } 3593 3594 #ifdef CONFIG_MIGRATION 3595 #include <linux/migrate.h> 3596 3597 int f2fs_migrate_page(struct address_space *mapping, 3598 struct page *newpage, struct page *page, enum migrate_mode mode) 3599 { 3600 int rc, extra_count; 3601 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 3602 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page); 3603 3604 BUG_ON(PageWriteback(page)); 3605 3606 /* migrating an atomic written page is safe with the inmem_lock hold */ 3607 if (atomic_written) { 3608 if (mode != MIGRATE_SYNC) 3609 return -EBUSY; 3610 if (!mutex_trylock(&fi->inmem_lock)) 3611 return -EAGAIN; 3612 } 3613 3614 /* one extra reference was held for atomic_write page */ 3615 extra_count = atomic_written ? 1 : 0; 3616 rc = migrate_page_move_mapping(mapping, newpage, 3617 page, extra_count); 3618 if (rc != MIGRATEPAGE_SUCCESS) { 3619 if (atomic_written) 3620 mutex_unlock(&fi->inmem_lock); 3621 return rc; 3622 } 3623 3624 if (atomic_written) { 3625 struct inmem_pages *cur; 3626 list_for_each_entry(cur, &fi->inmem_pages, list) 3627 if (cur->page == page) { 3628 cur->page = newpage; 3629 break; 3630 } 3631 mutex_unlock(&fi->inmem_lock); 3632 put_page(page); 3633 get_page(newpage); 3634 } 3635 3636 if (PagePrivate(page)) { 3637 f2fs_set_page_private(newpage, page_private(page)); 3638 f2fs_clear_page_private(page); 3639 } 3640 3641 if (mode != MIGRATE_SYNC_NO_COPY) 3642 migrate_page_copy(newpage, page); 3643 else 3644 migrate_page_states(newpage, page); 3645 3646 return MIGRATEPAGE_SUCCESS; 3647 } 3648 #endif 3649 3650 #ifdef CONFIG_SWAP 3651 /* Copied from generic_swapfile_activate() to check any holes */ 3652 static int check_swap_activate(struct swap_info_struct *sis, 3653 struct file *swap_file, sector_t *span) 3654 { 3655 struct address_space *mapping = swap_file->f_mapping; 3656 struct inode *inode = mapping->host; 3657 unsigned blocks_per_page; 3658 unsigned long page_no; 3659 unsigned blkbits; 3660 sector_t probe_block; 3661 sector_t last_block; 3662 sector_t lowest_block = -1; 3663 sector_t highest_block = 0; 3664 int nr_extents = 0; 3665 int ret; 3666 3667 blkbits = inode->i_blkbits; 3668 blocks_per_page = PAGE_SIZE >> blkbits; 3669 3670 /* 3671 * Map all the blocks into the extent list. This code doesn't try 3672 * to be very smart. 3673 */ 3674 probe_block = 0; 3675 page_no = 0; 3676 last_block = i_size_read(inode) >> blkbits; 3677 while ((probe_block + blocks_per_page) <= last_block && 3678 page_no < sis->max) { 3679 unsigned block_in_page; 3680 sector_t first_block; 3681 sector_t block = 0; 3682 int err = 0; 3683 3684 cond_resched(); 3685 3686 block = probe_block; 3687 err = bmap(inode, &block); 3688 if (err || !block) 3689 goto bad_bmap; 3690 first_block = block; 3691 3692 /* 3693 * It must be PAGE_SIZE aligned on-disk 3694 */ 3695 if (first_block & (blocks_per_page - 1)) { 3696 probe_block++; 3697 goto reprobe; 3698 } 3699 3700 for (block_in_page = 1; block_in_page < blocks_per_page; 3701 block_in_page++) { 3702 3703 block = probe_block + block_in_page; 3704 err = bmap(inode, &block); 3705 3706 if (err || !block) 3707 goto bad_bmap; 3708 3709 if (block != first_block + block_in_page) { 3710 /* Discontiguity */ 3711 probe_block++; 3712 goto reprobe; 3713 } 3714 } 3715 3716 first_block >>= (PAGE_SHIFT - blkbits); 3717 if (page_no) { /* exclude the header page */ 3718 if (first_block < lowest_block) 3719 lowest_block = first_block; 3720 if (first_block > highest_block) 3721 highest_block = first_block; 3722 } 3723 3724 /* 3725 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks 3726 */ 3727 ret = add_swap_extent(sis, page_no, 1, first_block); 3728 if (ret < 0) 3729 goto out; 3730 nr_extents += ret; 3731 page_no++; 3732 probe_block += blocks_per_page; 3733 reprobe: 3734 continue; 3735 } 3736 ret = nr_extents; 3737 *span = 1 + highest_block - lowest_block; 3738 if (page_no == 0) 3739 page_no = 1; /* force Empty message */ 3740 sis->max = page_no; 3741 sis->pages = page_no - 1; 3742 sis->highest_bit = page_no - 1; 3743 out: 3744 return ret; 3745 bad_bmap: 3746 pr_err("swapon: swapfile has holes\n"); 3747 return -EINVAL; 3748 } 3749 3750 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3751 sector_t *span) 3752 { 3753 struct inode *inode = file_inode(file); 3754 int ret; 3755 3756 if (!S_ISREG(inode->i_mode)) 3757 return -EINVAL; 3758 3759 if (f2fs_readonly(F2FS_I_SB(inode)->sb)) 3760 return -EROFS; 3761 3762 ret = f2fs_convert_inline_inode(inode); 3763 if (ret) 3764 return ret; 3765 3766 if (f2fs_disable_compressed_file(inode)) 3767 return -EINVAL; 3768 3769 ret = check_swap_activate(sis, file, span); 3770 if (ret < 0) 3771 return ret; 3772 3773 set_inode_flag(inode, FI_PIN_FILE); 3774 f2fs_precache_extents(inode); 3775 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3776 return ret; 3777 } 3778 3779 static void f2fs_swap_deactivate(struct file *file) 3780 { 3781 struct inode *inode = file_inode(file); 3782 3783 clear_inode_flag(inode, FI_PIN_FILE); 3784 } 3785 #else 3786 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3787 sector_t *span) 3788 { 3789 return -EOPNOTSUPP; 3790 } 3791 3792 static void f2fs_swap_deactivate(struct file *file) 3793 { 3794 } 3795 #endif 3796 3797 const struct address_space_operations f2fs_dblock_aops = { 3798 .readpage = f2fs_read_data_page, 3799 .readahead = f2fs_readahead, 3800 .writepage = f2fs_write_data_page, 3801 .writepages = f2fs_write_data_pages, 3802 .write_begin = f2fs_write_begin, 3803 .write_end = f2fs_write_end, 3804 .set_page_dirty = f2fs_set_data_page_dirty, 3805 .invalidatepage = f2fs_invalidate_page, 3806 .releasepage = f2fs_release_page, 3807 .direct_IO = f2fs_direct_IO, 3808 .bmap = f2fs_bmap, 3809 .swap_activate = f2fs_swap_activate, 3810 .swap_deactivate = f2fs_swap_deactivate, 3811 #ifdef CONFIG_MIGRATION 3812 .migratepage = f2fs_migrate_page, 3813 #endif 3814 }; 3815 3816 void f2fs_clear_page_cache_dirty_tag(struct page *page) 3817 { 3818 struct address_space *mapping = page_mapping(page); 3819 unsigned long flags; 3820 3821 xa_lock_irqsave(&mapping->i_pages, flags); 3822 __xa_clear_mark(&mapping->i_pages, page_index(page), 3823 PAGECACHE_TAG_DIRTY); 3824 xa_unlock_irqrestore(&mapping->i_pages, flags); 3825 } 3826 3827 int __init f2fs_init_post_read_processing(void) 3828 { 3829 bio_post_read_ctx_cache = 3830 kmem_cache_create("f2fs_bio_post_read_ctx", 3831 sizeof(struct bio_post_read_ctx), 0, 0, NULL); 3832 if (!bio_post_read_ctx_cache) 3833 goto fail; 3834 bio_post_read_ctx_pool = 3835 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 3836 bio_post_read_ctx_cache); 3837 if (!bio_post_read_ctx_pool) 3838 goto fail_free_cache; 3839 return 0; 3840 3841 fail_free_cache: 3842 kmem_cache_destroy(bio_post_read_ctx_cache); 3843 fail: 3844 return -ENOMEM; 3845 } 3846 3847 void f2fs_destroy_post_read_processing(void) 3848 { 3849 mempool_destroy(bio_post_read_ctx_pool); 3850 kmem_cache_destroy(bio_post_read_ctx_cache); 3851 } 3852 3853 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) 3854 { 3855 if (!f2fs_sb_has_encrypt(sbi) && 3856 !f2fs_sb_has_verity(sbi) && 3857 !f2fs_sb_has_compression(sbi)) 3858 return 0; 3859 3860 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq", 3861 WQ_UNBOUND | WQ_HIGHPRI, 3862 num_online_cpus()); 3863 if (!sbi->post_read_wq) 3864 return -ENOMEM; 3865 return 0; 3866 } 3867 3868 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) 3869 { 3870 if (sbi->post_read_wq) 3871 destroy_workqueue(sbi->post_read_wq); 3872 } 3873 3874 int __init f2fs_init_bio_entry_cache(void) 3875 { 3876 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab", 3877 sizeof(struct bio_entry)); 3878 if (!bio_entry_slab) 3879 return -ENOMEM; 3880 return 0; 3881 } 3882 3883 void f2fs_destroy_bio_entry_cache(void) 3884 { 3885 kmem_cache_destroy(bio_entry_slab); 3886 } 3887