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