1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/data.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/buffer_head.h> 11 #include <linux/mpage.h> 12 #include <linux/writeback.h> 13 #include <linux/backing-dev.h> 14 #include <linux/pagevec.h> 15 #include <linux/blkdev.h> 16 #include <linux/bio.h> 17 #include <linux/swap.h> 18 #include <linux/prefetch.h> 19 #include <linux/uio.h> 20 #include <linux/cleancache.h> 21 #include <linux/sched/signal.h> 22 #include <linux/fiemap.h> 23 24 #include "f2fs.h" 25 #include "node.h" 26 #include "segment.h" 27 #include "trace.h" 28 #include <trace/events/f2fs.h> 29 30 #define NUM_PREALLOC_POST_READ_CTXS 128 31 32 static struct kmem_cache *bio_post_read_ctx_cache; 33 static struct kmem_cache *bio_entry_slab; 34 static mempool_t *bio_post_read_ctx_pool; 35 static struct bio_set f2fs_bioset; 36 37 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE 38 39 int __init f2fs_init_bioset(void) 40 { 41 if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE, 42 0, BIOSET_NEED_BVECS)) 43 return -ENOMEM; 44 return 0; 45 } 46 47 void f2fs_destroy_bioset(void) 48 { 49 bioset_exit(&f2fs_bioset); 50 } 51 52 static inline struct bio *__f2fs_bio_alloc(gfp_t gfp_mask, 53 unsigned int nr_iovecs) 54 { 55 return bio_alloc_bioset(gfp_mask, nr_iovecs, &f2fs_bioset); 56 } 57 58 struct bio *f2fs_bio_alloc(struct f2fs_sb_info *sbi, int npages, bool noio) 59 { 60 if (noio) { 61 /* No failure on bio allocation */ 62 return __f2fs_bio_alloc(GFP_NOIO, npages); 63 } 64 65 if (time_to_inject(sbi, FAULT_ALLOC_BIO)) { 66 f2fs_show_injection_info(sbi, FAULT_ALLOC_BIO); 67 return NULL; 68 } 69 70 return __f2fs_bio_alloc(GFP_KERNEL, npages); 71 } 72 73 static bool __is_cp_guaranteed(struct page *page) 74 { 75 struct address_space *mapping = page->mapping; 76 struct inode *inode; 77 struct f2fs_sb_info *sbi; 78 79 if (!mapping) 80 return false; 81 82 if (f2fs_is_compressed_page(page)) 83 return false; 84 85 inode = mapping->host; 86 sbi = F2FS_I_SB(inode); 87 88 if (inode->i_ino == F2FS_META_INO(sbi) || 89 inode->i_ino == F2FS_NODE_INO(sbi) || 90 S_ISDIR(inode->i_mode) || 91 (S_ISREG(inode->i_mode) && 92 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) || 93 is_cold_data(page)) 94 return true; 95 return false; 96 } 97 98 static enum count_type __read_io_type(struct page *page) 99 { 100 struct address_space *mapping = page_file_mapping(page); 101 102 if (mapping) { 103 struct inode *inode = mapping->host; 104 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 105 106 if (inode->i_ino == F2FS_META_INO(sbi)) 107 return F2FS_RD_META; 108 109 if (inode->i_ino == F2FS_NODE_INO(sbi)) 110 return F2FS_RD_NODE; 111 } 112 return F2FS_RD_DATA; 113 } 114 115 /* postprocessing steps for read bios */ 116 enum bio_post_read_step { 117 STEP_DECRYPT, 118 STEP_DECOMPRESS_NOWQ, /* handle normal cluster data inplace */ 119 STEP_DECOMPRESS, /* handle compressed cluster data in workqueue */ 120 STEP_VERITY, 121 }; 122 123 struct bio_post_read_ctx { 124 struct bio *bio; 125 struct f2fs_sb_info *sbi; 126 struct work_struct work; 127 unsigned int enabled_steps; 128 }; 129 130 static void __read_end_io(struct bio *bio, bool compr, bool verity) 131 { 132 struct page *page; 133 struct bio_vec *bv; 134 struct bvec_iter_all iter_all; 135 136 bio_for_each_segment_all(bv, bio, iter_all) { 137 page = bv->bv_page; 138 139 #ifdef CONFIG_F2FS_FS_COMPRESSION 140 if (compr && f2fs_is_compressed_page(page)) { 141 f2fs_decompress_pages(bio, page, verity); 142 continue; 143 } 144 if (verity) 145 continue; 146 #endif 147 148 /* PG_error was set if any post_read step failed */ 149 if (bio->bi_status || PageError(page)) { 150 ClearPageUptodate(page); 151 /* will re-read again later */ 152 ClearPageError(page); 153 } else { 154 SetPageUptodate(page); 155 } 156 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 157 unlock_page(page); 158 } 159 } 160 161 static void f2fs_release_read_bio(struct bio *bio); 162 static void __f2fs_read_end_io(struct bio *bio, bool compr, bool verity) 163 { 164 if (!compr) 165 __read_end_io(bio, false, verity); 166 f2fs_release_read_bio(bio); 167 } 168 169 static void f2fs_decompress_bio(struct bio *bio, bool verity) 170 { 171 __read_end_io(bio, true, verity); 172 } 173 174 static void bio_post_read_processing(struct bio_post_read_ctx *ctx); 175 176 static void f2fs_decrypt_work(struct bio_post_read_ctx *ctx) 177 { 178 fscrypt_decrypt_bio(ctx->bio); 179 } 180 181 static void f2fs_decompress_work(struct bio_post_read_ctx *ctx) 182 { 183 f2fs_decompress_bio(ctx->bio, ctx->enabled_steps & (1 << STEP_VERITY)); 184 } 185 186 #ifdef CONFIG_F2FS_FS_COMPRESSION 187 static void f2fs_verify_pages(struct page **rpages, unsigned int cluster_size) 188 { 189 f2fs_decompress_end_io(rpages, cluster_size, false, true); 190 } 191 192 static void f2fs_verify_bio(struct bio *bio) 193 { 194 struct bio_vec *bv; 195 struct bvec_iter_all iter_all; 196 197 bio_for_each_segment_all(bv, bio, iter_all) { 198 struct page *page = bv->bv_page; 199 struct decompress_io_ctx *dic; 200 201 dic = (struct decompress_io_ctx *)page_private(page); 202 203 if (dic) { 204 if (refcount_dec_not_one(&dic->ref)) 205 continue; 206 f2fs_verify_pages(dic->rpages, 207 dic->cluster_size); 208 f2fs_free_dic(dic); 209 continue; 210 } 211 212 if (bio->bi_status || PageError(page)) 213 goto clear_uptodate; 214 215 if (fsverity_verify_page(page)) { 216 SetPageUptodate(page); 217 goto unlock; 218 } 219 clear_uptodate: 220 ClearPageUptodate(page); 221 ClearPageError(page); 222 unlock: 223 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 224 unlock_page(page); 225 } 226 } 227 #endif 228 229 static void f2fs_verity_work(struct work_struct *work) 230 { 231 struct bio_post_read_ctx *ctx = 232 container_of(work, struct bio_post_read_ctx, work); 233 struct bio *bio = ctx->bio; 234 #ifdef CONFIG_F2FS_FS_COMPRESSION 235 unsigned int enabled_steps = ctx->enabled_steps; 236 #endif 237 238 /* 239 * fsverity_verify_bio() may call readpages() again, and while verity 240 * will be disabled for this, decryption may still be needed, resulting 241 * in another bio_post_read_ctx being allocated. So to prevent 242 * deadlocks we need to release the current ctx to the mempool first. 243 * This assumes that verity is the last post-read step. 244 */ 245 mempool_free(ctx, bio_post_read_ctx_pool); 246 bio->bi_private = NULL; 247 248 #ifdef CONFIG_F2FS_FS_COMPRESSION 249 /* previous step is decompression */ 250 if (enabled_steps & (1 << STEP_DECOMPRESS)) { 251 f2fs_verify_bio(bio); 252 f2fs_release_read_bio(bio); 253 return; 254 } 255 #endif 256 257 fsverity_verify_bio(bio); 258 __f2fs_read_end_io(bio, false, false); 259 } 260 261 static void f2fs_post_read_work(struct work_struct *work) 262 { 263 struct bio_post_read_ctx *ctx = 264 container_of(work, struct bio_post_read_ctx, work); 265 266 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) 267 f2fs_decrypt_work(ctx); 268 269 if (ctx->enabled_steps & (1 << STEP_DECOMPRESS)) 270 f2fs_decompress_work(ctx); 271 272 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 273 INIT_WORK(&ctx->work, f2fs_verity_work); 274 fsverity_enqueue_verify_work(&ctx->work); 275 return; 276 } 277 278 __f2fs_read_end_io(ctx->bio, 279 ctx->enabled_steps & (1 << STEP_DECOMPRESS), false); 280 } 281 282 static void f2fs_enqueue_post_read_work(struct f2fs_sb_info *sbi, 283 struct work_struct *work) 284 { 285 queue_work(sbi->post_read_wq, work); 286 } 287 288 static void bio_post_read_processing(struct bio_post_read_ctx *ctx) 289 { 290 /* 291 * We use different work queues for decryption and for verity because 292 * verity may require reading metadata pages that need decryption, and 293 * we shouldn't recurse to the same workqueue. 294 */ 295 296 if (ctx->enabled_steps & (1 << STEP_DECRYPT) || 297 ctx->enabled_steps & (1 << STEP_DECOMPRESS)) { 298 INIT_WORK(&ctx->work, f2fs_post_read_work); 299 f2fs_enqueue_post_read_work(ctx->sbi, &ctx->work); 300 return; 301 } 302 303 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 304 INIT_WORK(&ctx->work, f2fs_verity_work); 305 fsverity_enqueue_verify_work(&ctx->work); 306 return; 307 } 308 309 __f2fs_read_end_io(ctx->bio, false, false); 310 } 311 312 static bool f2fs_bio_post_read_required(struct bio *bio) 313 { 314 return bio->bi_private; 315 } 316 317 static void f2fs_read_end_io(struct bio *bio) 318 { 319 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio)); 320 321 if (time_to_inject(sbi, FAULT_READ_IO)) { 322 f2fs_show_injection_info(sbi, FAULT_READ_IO); 323 bio->bi_status = BLK_STS_IOERR; 324 } 325 326 if (f2fs_bio_post_read_required(bio)) { 327 struct bio_post_read_ctx *ctx = bio->bi_private; 328 329 bio_post_read_processing(ctx); 330 return; 331 } 332 333 __f2fs_read_end_io(bio, false, false); 334 } 335 336 static void f2fs_write_end_io(struct bio *bio) 337 { 338 struct f2fs_sb_info *sbi = bio->bi_private; 339 struct bio_vec *bvec; 340 struct bvec_iter_all iter_all; 341 342 if (time_to_inject(sbi, FAULT_WRITE_IO)) { 343 f2fs_show_injection_info(sbi, FAULT_WRITE_IO); 344 bio->bi_status = BLK_STS_IOERR; 345 } 346 347 bio_for_each_segment_all(bvec, bio, iter_all) { 348 struct page *page = bvec->bv_page; 349 enum count_type type = WB_DATA_TYPE(page); 350 351 if (IS_DUMMY_WRITTEN_PAGE(page)) { 352 set_page_private(page, (unsigned long)NULL); 353 ClearPagePrivate(page); 354 unlock_page(page); 355 mempool_free(page, sbi->write_io_dummy); 356 357 if (unlikely(bio->bi_status)) 358 f2fs_stop_checkpoint(sbi, true); 359 continue; 360 } 361 362 fscrypt_finalize_bounce_page(&page); 363 364 #ifdef CONFIG_F2FS_FS_COMPRESSION 365 if (f2fs_is_compressed_page(page)) { 366 f2fs_compress_write_end_io(bio, page); 367 continue; 368 } 369 #endif 370 371 if (unlikely(bio->bi_status)) { 372 mapping_set_error(page->mapping, -EIO); 373 if (type == F2FS_WB_CP_DATA) 374 f2fs_stop_checkpoint(sbi, true); 375 } 376 377 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 378 page->index != nid_of_node(page)); 379 380 dec_page_count(sbi, type); 381 if (f2fs_in_warm_node_list(sbi, page)) 382 f2fs_del_fsync_node_entry(sbi, page); 383 clear_cold_data(page); 384 end_page_writeback(page); 385 } 386 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 387 wq_has_sleeper(&sbi->cp_wait)) 388 wake_up(&sbi->cp_wait); 389 390 bio_put(bio); 391 } 392 393 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 394 block_t blk_addr, struct bio *bio) 395 { 396 struct block_device *bdev = sbi->sb->s_bdev; 397 int i; 398 399 if (f2fs_is_multi_device(sbi)) { 400 for (i = 0; i < sbi->s_ndevs; i++) { 401 if (FDEV(i).start_blk <= blk_addr && 402 FDEV(i).end_blk >= blk_addr) { 403 blk_addr -= FDEV(i).start_blk; 404 bdev = FDEV(i).bdev; 405 break; 406 } 407 } 408 } 409 if (bio) { 410 bio_set_dev(bio, bdev); 411 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); 412 } 413 return bdev; 414 } 415 416 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 417 { 418 int i; 419 420 if (!f2fs_is_multi_device(sbi)) 421 return 0; 422 423 for (i = 0; i < sbi->s_ndevs; i++) 424 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 425 return i; 426 return 0; 427 } 428 429 /* 430 * Return true, if pre_bio's bdev is same as its target device. 431 */ 432 static bool __same_bdev(struct f2fs_sb_info *sbi, 433 block_t blk_addr, struct bio *bio) 434 { 435 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL); 436 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno; 437 } 438 439 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) 440 { 441 struct f2fs_sb_info *sbi = fio->sbi; 442 struct bio *bio; 443 444 bio = f2fs_bio_alloc(sbi, npages, true); 445 446 f2fs_target_device(sbi, fio->new_blkaddr, bio); 447 if (is_read_io(fio->op)) { 448 bio->bi_end_io = f2fs_read_end_io; 449 bio->bi_private = NULL; 450 } else { 451 bio->bi_end_io = f2fs_write_end_io; 452 bio->bi_private = sbi; 453 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, 454 fio->type, fio->temp); 455 } 456 if (fio->io_wbc) 457 wbc_init_bio(fio->io_wbc, bio); 458 459 return bio; 460 } 461 462 static inline void __submit_bio(struct f2fs_sb_info *sbi, 463 struct bio *bio, enum page_type type) 464 { 465 if (!is_read_io(bio_op(bio))) { 466 unsigned int start; 467 468 if (type != DATA && type != NODE) 469 goto submit_io; 470 471 if (f2fs_lfs_mode(sbi) && current->plug) 472 blk_finish_plug(current->plug); 473 474 if (F2FS_IO_ALIGNED(sbi)) 475 goto submit_io; 476 477 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS; 478 start %= F2FS_IO_SIZE(sbi); 479 480 if (start == 0) 481 goto submit_io; 482 483 /* fill dummy pages */ 484 for (; start < F2FS_IO_SIZE(sbi); start++) { 485 struct page *page = 486 mempool_alloc(sbi->write_io_dummy, 487 GFP_NOIO | __GFP_NOFAIL); 488 f2fs_bug_on(sbi, !page); 489 490 zero_user_segment(page, 0, PAGE_SIZE); 491 SetPagePrivate(page); 492 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE); 493 lock_page(page); 494 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 495 f2fs_bug_on(sbi, 1); 496 } 497 /* 498 * In the NODE case, we lose next block address chain. So, we 499 * need to do checkpoint in f2fs_sync_file. 500 */ 501 if (type == NODE) 502 set_sbi_flag(sbi, SBI_NEED_CP); 503 } 504 submit_io: 505 if (is_read_io(bio_op(bio))) 506 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 507 else 508 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 509 submit_bio(bio); 510 } 511 512 void f2fs_submit_bio(struct f2fs_sb_info *sbi, 513 struct bio *bio, enum page_type type) 514 { 515 __submit_bio(sbi, bio, type); 516 } 517 518 static void __attach_io_flag(struct f2fs_io_info *fio) 519 { 520 struct f2fs_sb_info *sbi = fio->sbi; 521 unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1; 522 unsigned int io_flag, fua_flag, meta_flag; 523 524 if (fio->type == DATA) 525 io_flag = sbi->data_io_flag; 526 else if (fio->type == NODE) 527 io_flag = sbi->node_io_flag; 528 else 529 return; 530 531 fua_flag = io_flag & temp_mask; 532 meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask; 533 534 /* 535 * data/node io flag bits per temp: 536 * REQ_META | REQ_FUA | 537 * 5 | 4 | 3 | 2 | 1 | 0 | 538 * Cold | Warm | Hot | Cold | Warm | Hot | 539 */ 540 if ((1 << fio->temp) & meta_flag) 541 fio->op_flags |= REQ_META; 542 if ((1 << fio->temp) & fua_flag) 543 fio->op_flags |= REQ_FUA; 544 } 545 546 static void __submit_merged_bio(struct f2fs_bio_info *io) 547 { 548 struct f2fs_io_info *fio = &io->fio; 549 550 if (!io->bio) 551 return; 552 553 __attach_io_flag(fio); 554 bio_set_op_attrs(io->bio, fio->op, fio->op_flags); 555 556 if (is_read_io(fio->op)) 557 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 558 else 559 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 560 561 __submit_bio(io->sbi, io->bio, fio->type); 562 io->bio = NULL; 563 } 564 565 static bool __has_merged_page(struct bio *bio, struct inode *inode, 566 struct page *page, nid_t ino) 567 { 568 struct bio_vec *bvec; 569 struct bvec_iter_all iter_all; 570 571 if (!bio) 572 return false; 573 574 if (!inode && !page && !ino) 575 return true; 576 577 bio_for_each_segment_all(bvec, bio, iter_all) { 578 struct page *target = bvec->bv_page; 579 580 if (fscrypt_is_bounce_page(target)) { 581 target = fscrypt_pagecache_page(target); 582 if (IS_ERR(target)) 583 continue; 584 } 585 if (f2fs_is_compressed_page(target)) { 586 target = f2fs_compress_control_page(target); 587 if (IS_ERR(target)) 588 continue; 589 } 590 591 if (inode && inode == target->mapping->host) 592 return true; 593 if (page && page == target) 594 return true; 595 if (ino && ino == ino_of_node(target)) 596 return true; 597 } 598 599 return false; 600 } 601 602 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 603 enum page_type type, enum temp_type temp) 604 { 605 enum page_type btype = PAGE_TYPE_OF_BIO(type); 606 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 607 608 down_write(&io->io_rwsem); 609 610 /* change META to META_FLUSH in the checkpoint procedure */ 611 if (type >= META_FLUSH) { 612 io->fio.type = META_FLUSH; 613 io->fio.op = REQ_OP_WRITE; 614 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC; 615 if (!test_opt(sbi, NOBARRIER)) 616 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 617 } 618 __submit_merged_bio(io); 619 up_write(&io->io_rwsem); 620 } 621 622 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 623 struct inode *inode, struct page *page, 624 nid_t ino, enum page_type type, bool force) 625 { 626 enum temp_type temp; 627 bool ret = true; 628 629 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 630 if (!force) { 631 enum page_type btype = PAGE_TYPE_OF_BIO(type); 632 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 633 634 down_read(&io->io_rwsem); 635 ret = __has_merged_page(io->bio, inode, page, ino); 636 up_read(&io->io_rwsem); 637 } 638 if (ret) 639 __f2fs_submit_merged_write(sbi, type, temp); 640 641 /* TODO: use HOT temp only for meta pages now. */ 642 if (type >= META) 643 break; 644 } 645 } 646 647 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 648 { 649 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true); 650 } 651 652 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 653 struct inode *inode, struct page *page, 654 nid_t ino, enum page_type type) 655 { 656 __submit_merged_write_cond(sbi, inode, page, ino, type, false); 657 } 658 659 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 660 { 661 f2fs_submit_merged_write(sbi, DATA); 662 f2fs_submit_merged_write(sbi, NODE); 663 f2fs_submit_merged_write(sbi, META); 664 } 665 666 /* 667 * Fill the locked page with data located in the block address. 668 * A caller needs to unlock the page on failure. 669 */ 670 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 671 { 672 struct bio *bio; 673 struct page *page = fio->encrypted_page ? 674 fio->encrypted_page : fio->page; 675 676 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 677 fio->is_por ? META_POR : (__is_meta_io(fio) ? 678 META_GENERIC : DATA_GENERIC_ENHANCE))) 679 return -EFSCORRUPTED; 680 681 trace_f2fs_submit_page_bio(page, fio); 682 f2fs_trace_ios(fio, 0); 683 684 /* Allocate a new bio */ 685 bio = __bio_alloc(fio, 1); 686 687 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 688 bio_put(bio); 689 return -EFAULT; 690 } 691 692 if (fio->io_wbc && !is_read_io(fio->op)) 693 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 694 695 __attach_io_flag(fio); 696 bio_set_op_attrs(bio, fio->op, fio->op_flags); 697 698 inc_page_count(fio->sbi, is_read_io(fio->op) ? 699 __read_io_type(page): WB_DATA_TYPE(fio->page)); 700 701 __submit_bio(fio->sbi, bio, fio->type); 702 return 0; 703 } 704 705 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 706 block_t last_blkaddr, block_t cur_blkaddr) 707 { 708 if (last_blkaddr + 1 != cur_blkaddr) 709 return false; 710 return __same_bdev(sbi, cur_blkaddr, bio); 711 } 712 713 static bool io_type_is_mergeable(struct f2fs_bio_info *io, 714 struct f2fs_io_info *fio) 715 { 716 if (io->fio.op != fio->op) 717 return false; 718 return io->fio.op_flags == fio->op_flags; 719 } 720 721 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 722 struct f2fs_bio_info *io, 723 struct f2fs_io_info *fio, 724 block_t last_blkaddr, 725 block_t cur_blkaddr) 726 { 727 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) { 728 unsigned int filled_blocks = 729 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size); 730 unsigned int io_size = F2FS_IO_SIZE(sbi); 731 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt; 732 733 /* IOs in bio is aligned and left space of vectors is not enough */ 734 if (!(filled_blocks % io_size) && left_vecs < io_size) 735 return false; 736 } 737 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr)) 738 return false; 739 return io_type_is_mergeable(io, fio); 740 } 741 742 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio, 743 struct page *page, enum temp_type temp) 744 { 745 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 746 struct bio_entry *be; 747 748 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS); 749 be->bio = bio; 750 bio_get(bio); 751 752 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE) 753 f2fs_bug_on(sbi, 1); 754 755 down_write(&io->bio_list_lock); 756 list_add_tail(&be->list, &io->bio_list); 757 up_write(&io->bio_list_lock); 758 } 759 760 static void del_bio_entry(struct bio_entry *be) 761 { 762 list_del(&be->list); 763 kmem_cache_free(bio_entry_slab, be); 764 } 765 766 static int add_ipu_page(struct f2fs_sb_info *sbi, struct bio **bio, 767 struct page *page) 768 { 769 enum temp_type temp; 770 bool found = false; 771 int ret = -EAGAIN; 772 773 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 774 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 775 struct list_head *head = &io->bio_list; 776 struct bio_entry *be; 777 778 down_write(&io->bio_list_lock); 779 list_for_each_entry(be, head, list) { 780 if (be->bio != *bio) 781 continue; 782 783 found = true; 784 785 if (bio_add_page(*bio, page, PAGE_SIZE, 0) == 786 PAGE_SIZE) { 787 ret = 0; 788 break; 789 } 790 791 /* bio is full */ 792 del_bio_entry(be); 793 __submit_bio(sbi, *bio, DATA); 794 break; 795 } 796 up_write(&io->bio_list_lock); 797 } 798 799 if (ret) { 800 bio_put(*bio); 801 *bio = NULL; 802 } 803 804 return ret; 805 } 806 807 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi, 808 struct bio **bio, struct page *page) 809 { 810 enum temp_type temp; 811 bool found = false; 812 struct bio *target = bio ? *bio : NULL; 813 814 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 815 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 816 struct list_head *head = &io->bio_list; 817 struct bio_entry *be; 818 819 if (list_empty(head)) 820 continue; 821 822 down_read(&io->bio_list_lock); 823 list_for_each_entry(be, head, list) { 824 if (target) 825 found = (target == be->bio); 826 else 827 found = __has_merged_page(be->bio, NULL, 828 page, 0); 829 if (found) 830 break; 831 } 832 up_read(&io->bio_list_lock); 833 834 if (!found) 835 continue; 836 837 found = false; 838 839 down_write(&io->bio_list_lock); 840 list_for_each_entry(be, head, list) { 841 if (target) 842 found = (target == be->bio); 843 else 844 found = __has_merged_page(be->bio, NULL, 845 page, 0); 846 if (found) { 847 target = be->bio; 848 del_bio_entry(be); 849 break; 850 } 851 } 852 up_write(&io->bio_list_lock); 853 } 854 855 if (found) 856 __submit_bio(sbi, target, DATA); 857 if (bio && *bio) { 858 bio_put(*bio); 859 *bio = NULL; 860 } 861 } 862 863 int f2fs_merge_page_bio(struct f2fs_io_info *fio) 864 { 865 struct bio *bio = *fio->bio; 866 struct page *page = fio->encrypted_page ? 867 fio->encrypted_page : fio->page; 868 869 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 870 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) 871 return -EFSCORRUPTED; 872 873 trace_f2fs_submit_page_bio(page, fio); 874 f2fs_trace_ios(fio, 0); 875 876 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block, 877 fio->new_blkaddr)) 878 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL); 879 alloc_new: 880 if (!bio) { 881 bio = __bio_alloc(fio, BIO_MAX_PAGES); 882 __attach_io_flag(fio); 883 bio_set_op_attrs(bio, fio->op, fio->op_flags); 884 885 add_bio_entry(fio->sbi, bio, page, fio->temp); 886 } else { 887 if (add_ipu_page(fio->sbi, &bio, page)) 888 goto alloc_new; 889 } 890 891 if (fio->io_wbc) 892 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 893 894 inc_page_count(fio->sbi, WB_DATA_TYPE(page)); 895 896 *fio->last_block = fio->new_blkaddr; 897 *fio->bio = bio; 898 899 return 0; 900 } 901 902 void f2fs_submit_page_write(struct f2fs_io_info *fio) 903 { 904 struct f2fs_sb_info *sbi = fio->sbi; 905 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 906 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 907 struct page *bio_page; 908 909 f2fs_bug_on(sbi, is_read_io(fio->op)); 910 911 down_write(&io->io_rwsem); 912 next: 913 if (fio->in_list) { 914 spin_lock(&io->io_lock); 915 if (list_empty(&io->io_list)) { 916 spin_unlock(&io->io_lock); 917 goto out; 918 } 919 fio = list_first_entry(&io->io_list, 920 struct f2fs_io_info, list); 921 list_del(&fio->list); 922 spin_unlock(&io->io_lock); 923 } 924 925 verify_fio_blkaddr(fio); 926 927 if (fio->encrypted_page) 928 bio_page = fio->encrypted_page; 929 else if (fio->compressed_page) 930 bio_page = fio->compressed_page; 931 else 932 bio_page = fio->page; 933 934 /* set submitted = true as a return value */ 935 fio->submitted = true; 936 937 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 938 939 if (io->bio && !io_is_mergeable(sbi, io->bio, io, fio, 940 io->last_block_in_bio, fio->new_blkaddr)) 941 __submit_merged_bio(io); 942 alloc_new: 943 if (io->bio == NULL) { 944 if (F2FS_IO_ALIGNED(sbi) && 945 (fio->type == DATA || fio->type == NODE) && 946 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 947 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 948 fio->retry = true; 949 goto skip; 950 } 951 io->bio = __bio_alloc(fio, BIO_MAX_PAGES); 952 io->fio = *fio; 953 } 954 955 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 956 __submit_merged_bio(io); 957 goto alloc_new; 958 } 959 960 if (fio->io_wbc) 961 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE); 962 963 io->last_block_in_bio = fio->new_blkaddr; 964 f2fs_trace_ios(fio, 0); 965 966 trace_f2fs_submit_page_write(fio->page, fio); 967 skip: 968 if (fio->in_list) 969 goto next; 970 out: 971 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 972 !f2fs_is_checkpoint_ready(sbi)) 973 __submit_merged_bio(io); 974 up_write(&io->io_rwsem); 975 } 976 977 static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx) 978 { 979 return fsverity_active(inode) && 980 idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 981 } 982 983 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 984 unsigned nr_pages, unsigned op_flag, 985 pgoff_t first_idx, bool for_write) 986 { 987 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 988 struct bio *bio; 989 struct bio_post_read_ctx *ctx; 990 unsigned int post_read_steps = 0; 991 992 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), 993 for_write); 994 if (!bio) 995 return ERR_PTR(-ENOMEM); 996 f2fs_target_device(sbi, blkaddr, bio); 997 bio->bi_end_io = f2fs_read_end_io; 998 bio_set_op_attrs(bio, REQ_OP_READ, op_flag); 999 1000 if (f2fs_encrypted_file(inode)) 1001 post_read_steps |= 1 << STEP_DECRYPT; 1002 if (f2fs_compressed_file(inode)) 1003 post_read_steps |= 1 << STEP_DECOMPRESS_NOWQ; 1004 if (f2fs_need_verity(inode, first_idx)) 1005 post_read_steps |= 1 << STEP_VERITY; 1006 1007 if (post_read_steps) { 1008 /* Due to the mempool, this never fails. */ 1009 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 1010 ctx->bio = bio; 1011 ctx->sbi = sbi; 1012 ctx->enabled_steps = post_read_steps; 1013 bio->bi_private = ctx; 1014 } 1015 1016 return bio; 1017 } 1018 1019 static void f2fs_release_read_bio(struct bio *bio) 1020 { 1021 if (bio->bi_private) 1022 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 1023 bio_put(bio); 1024 } 1025 1026 /* This can handle encryption stuffs */ 1027 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 1028 block_t blkaddr, bool for_write) 1029 { 1030 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1031 struct bio *bio; 1032 1033 bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0, page->index, for_write); 1034 if (IS_ERR(bio)) 1035 return PTR_ERR(bio); 1036 1037 /* wait for GCed page writeback via META_MAPPING */ 1038 f2fs_wait_on_block_writeback(inode, blkaddr); 1039 1040 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 1041 bio_put(bio); 1042 return -EFAULT; 1043 } 1044 ClearPageError(page); 1045 inc_page_count(sbi, F2FS_RD_DATA); 1046 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); 1047 __submit_bio(sbi, bio, DATA); 1048 return 0; 1049 } 1050 1051 static void __set_data_blkaddr(struct dnode_of_data *dn) 1052 { 1053 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 1054 __le32 *addr_array; 1055 int base = 0; 1056 1057 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 1058 base = get_extra_isize(dn->inode); 1059 1060 /* Get physical address of data block */ 1061 addr_array = blkaddr_in_node(rn); 1062 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 1063 } 1064 1065 /* 1066 * Lock ordering for the change of data block address: 1067 * ->data_page 1068 * ->node_page 1069 * update block addresses in the node page 1070 */ 1071 void f2fs_set_data_blkaddr(struct dnode_of_data *dn) 1072 { 1073 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1074 __set_data_blkaddr(dn); 1075 if (set_page_dirty(dn->node_page)) 1076 dn->node_changed = true; 1077 } 1078 1079 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1080 { 1081 dn->data_blkaddr = blkaddr; 1082 f2fs_set_data_blkaddr(dn); 1083 f2fs_update_extent_cache(dn); 1084 } 1085 1086 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 1087 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 1088 { 1089 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1090 int err; 1091 1092 if (!count) 1093 return 0; 1094 1095 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1096 return -EPERM; 1097 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1098 return err; 1099 1100 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 1101 dn->ofs_in_node, count); 1102 1103 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1104 1105 for (; count > 0; dn->ofs_in_node++) { 1106 block_t blkaddr = f2fs_data_blkaddr(dn); 1107 if (blkaddr == NULL_ADDR) { 1108 dn->data_blkaddr = NEW_ADDR; 1109 __set_data_blkaddr(dn); 1110 count--; 1111 } 1112 } 1113 1114 if (set_page_dirty(dn->node_page)) 1115 dn->node_changed = true; 1116 return 0; 1117 } 1118 1119 /* Should keep dn->ofs_in_node unchanged */ 1120 int f2fs_reserve_new_block(struct dnode_of_data *dn) 1121 { 1122 unsigned int ofs_in_node = dn->ofs_in_node; 1123 int ret; 1124 1125 ret = f2fs_reserve_new_blocks(dn, 1); 1126 dn->ofs_in_node = ofs_in_node; 1127 return ret; 1128 } 1129 1130 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 1131 { 1132 bool need_put = dn->inode_page ? false : true; 1133 int err; 1134 1135 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE); 1136 if (err) 1137 return err; 1138 1139 if (dn->data_blkaddr == NULL_ADDR) 1140 err = f2fs_reserve_new_block(dn); 1141 if (err || need_put) 1142 f2fs_put_dnode(dn); 1143 return err; 1144 } 1145 1146 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 1147 { 1148 struct extent_info ei = {0,0,0}; 1149 struct inode *inode = dn->inode; 1150 1151 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1152 dn->data_blkaddr = ei.blk + index - ei.fofs; 1153 return 0; 1154 } 1155 1156 return f2fs_reserve_block(dn, index); 1157 } 1158 1159 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, 1160 int op_flags, bool for_write) 1161 { 1162 struct address_space *mapping = inode->i_mapping; 1163 struct dnode_of_data dn; 1164 struct page *page; 1165 struct extent_info ei = {0,0,0}; 1166 int err; 1167 1168 page = f2fs_grab_cache_page(mapping, index, for_write); 1169 if (!page) 1170 return ERR_PTR(-ENOMEM); 1171 1172 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1173 dn.data_blkaddr = ei.blk + index - ei.fofs; 1174 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr, 1175 DATA_GENERIC_ENHANCE_READ)) { 1176 err = -EFSCORRUPTED; 1177 goto put_err; 1178 } 1179 goto got_it; 1180 } 1181 1182 set_new_dnode(&dn, inode, NULL, NULL, 0); 1183 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1184 if (err) 1185 goto put_err; 1186 f2fs_put_dnode(&dn); 1187 1188 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1189 err = -ENOENT; 1190 goto put_err; 1191 } 1192 if (dn.data_blkaddr != NEW_ADDR && 1193 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode), 1194 dn.data_blkaddr, 1195 DATA_GENERIC_ENHANCE)) { 1196 err = -EFSCORRUPTED; 1197 goto put_err; 1198 } 1199 got_it: 1200 if (PageUptodate(page)) { 1201 unlock_page(page); 1202 return page; 1203 } 1204 1205 /* 1206 * A new dentry page is allocated but not able to be written, since its 1207 * new inode page couldn't be allocated due to -ENOSPC. 1208 * In such the case, its blkaddr can be remained as NEW_ADDR. 1209 * see, f2fs_add_link -> f2fs_get_new_data_page -> 1210 * f2fs_init_inode_metadata. 1211 */ 1212 if (dn.data_blkaddr == NEW_ADDR) { 1213 zero_user_segment(page, 0, PAGE_SIZE); 1214 if (!PageUptodate(page)) 1215 SetPageUptodate(page); 1216 unlock_page(page); 1217 return page; 1218 } 1219 1220 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr, for_write); 1221 if (err) 1222 goto put_err; 1223 return page; 1224 1225 put_err: 1226 f2fs_put_page(page, 1); 1227 return ERR_PTR(err); 1228 } 1229 1230 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index) 1231 { 1232 struct address_space *mapping = inode->i_mapping; 1233 struct page *page; 1234 1235 page = find_get_page(mapping, index); 1236 if (page && PageUptodate(page)) 1237 return page; 1238 f2fs_put_page(page, 0); 1239 1240 page = f2fs_get_read_data_page(inode, index, 0, false); 1241 if (IS_ERR(page)) 1242 return page; 1243 1244 if (PageUptodate(page)) 1245 return page; 1246 1247 wait_on_page_locked(page); 1248 if (unlikely(!PageUptodate(page))) { 1249 f2fs_put_page(page, 0); 1250 return ERR_PTR(-EIO); 1251 } 1252 return page; 1253 } 1254 1255 /* 1256 * If it tries to access a hole, return an error. 1257 * Because, the callers, functions in dir.c and GC, should be able to know 1258 * whether this page exists or not. 1259 */ 1260 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, 1261 bool for_write) 1262 { 1263 struct address_space *mapping = inode->i_mapping; 1264 struct page *page; 1265 repeat: 1266 page = f2fs_get_read_data_page(inode, index, 0, for_write); 1267 if (IS_ERR(page)) 1268 return page; 1269 1270 /* wait for read completion */ 1271 lock_page(page); 1272 if (unlikely(page->mapping != mapping)) { 1273 f2fs_put_page(page, 1); 1274 goto repeat; 1275 } 1276 if (unlikely(!PageUptodate(page))) { 1277 f2fs_put_page(page, 1); 1278 return ERR_PTR(-EIO); 1279 } 1280 return page; 1281 } 1282 1283 /* 1284 * Caller ensures that this data page is never allocated. 1285 * A new zero-filled data page is allocated in the page cache. 1286 * 1287 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 1288 * f2fs_unlock_op(). 1289 * Note that, ipage is set only by make_empty_dir, and if any error occur, 1290 * ipage should be released by this function. 1291 */ 1292 struct page *f2fs_get_new_data_page(struct inode *inode, 1293 struct page *ipage, pgoff_t index, bool new_i_size) 1294 { 1295 struct address_space *mapping = inode->i_mapping; 1296 struct page *page; 1297 struct dnode_of_data dn; 1298 int err; 1299 1300 page = f2fs_grab_cache_page(mapping, index, true); 1301 if (!page) { 1302 /* 1303 * before exiting, we should make sure ipage will be released 1304 * if any error occur. 1305 */ 1306 f2fs_put_page(ipage, 1); 1307 return ERR_PTR(-ENOMEM); 1308 } 1309 1310 set_new_dnode(&dn, inode, ipage, NULL, 0); 1311 err = f2fs_reserve_block(&dn, index); 1312 if (err) { 1313 f2fs_put_page(page, 1); 1314 return ERR_PTR(err); 1315 } 1316 if (!ipage) 1317 f2fs_put_dnode(&dn); 1318 1319 if (PageUptodate(page)) 1320 goto got_it; 1321 1322 if (dn.data_blkaddr == NEW_ADDR) { 1323 zero_user_segment(page, 0, PAGE_SIZE); 1324 if (!PageUptodate(page)) 1325 SetPageUptodate(page); 1326 } else { 1327 f2fs_put_page(page, 1); 1328 1329 /* if ipage exists, blkaddr should be NEW_ADDR */ 1330 f2fs_bug_on(F2FS_I_SB(inode), ipage); 1331 page = f2fs_get_lock_data_page(inode, index, true); 1332 if (IS_ERR(page)) 1333 return page; 1334 } 1335 got_it: 1336 if (new_i_size && i_size_read(inode) < 1337 ((loff_t)(index + 1) << PAGE_SHIFT)) 1338 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 1339 return page; 1340 } 1341 1342 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 1343 { 1344 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1345 struct f2fs_summary sum; 1346 struct node_info ni; 1347 block_t old_blkaddr; 1348 blkcnt_t count = 1; 1349 int err; 1350 1351 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1352 return -EPERM; 1353 1354 err = f2fs_get_node_info(sbi, dn->nid, &ni); 1355 if (err) 1356 return err; 1357 1358 dn->data_blkaddr = f2fs_data_blkaddr(dn); 1359 if (dn->data_blkaddr != NULL_ADDR) 1360 goto alloc; 1361 1362 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1363 return err; 1364 1365 alloc: 1366 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1367 old_blkaddr = dn->data_blkaddr; 1368 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr, 1369 &sum, seg_type, NULL, false); 1370 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 1371 invalidate_mapping_pages(META_MAPPING(sbi), 1372 old_blkaddr, old_blkaddr); 1373 f2fs_update_data_blkaddr(dn, dn->data_blkaddr); 1374 1375 /* 1376 * i_size will be updated by direct_IO. Otherwise, we'll get stale 1377 * data from unwritten block via dio_read. 1378 */ 1379 return 0; 1380 } 1381 1382 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from) 1383 { 1384 struct inode *inode = file_inode(iocb->ki_filp); 1385 struct f2fs_map_blocks map; 1386 int flag; 1387 int err = 0; 1388 bool direct_io = iocb->ki_flags & IOCB_DIRECT; 1389 1390 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos); 1391 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from)); 1392 if (map.m_len > map.m_lblk) 1393 map.m_len -= map.m_lblk; 1394 else 1395 map.m_len = 0; 1396 1397 map.m_next_pgofs = NULL; 1398 map.m_next_extent = NULL; 1399 map.m_seg_type = NO_CHECK_TYPE; 1400 map.m_may_create = true; 1401 1402 if (direct_io) { 1403 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint); 1404 flag = f2fs_force_buffered_io(inode, iocb, from) ? 1405 F2FS_GET_BLOCK_PRE_AIO : 1406 F2FS_GET_BLOCK_PRE_DIO; 1407 goto map_blocks; 1408 } 1409 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) { 1410 err = f2fs_convert_inline_inode(inode); 1411 if (err) 1412 return err; 1413 } 1414 if (f2fs_has_inline_data(inode)) 1415 return err; 1416 1417 flag = F2FS_GET_BLOCK_PRE_AIO; 1418 1419 map_blocks: 1420 err = f2fs_map_blocks(inode, &map, 1, flag); 1421 if (map.m_len > 0 && err == -ENOSPC) { 1422 if (!direct_io) 1423 set_inode_flag(inode, FI_NO_PREALLOC); 1424 err = 0; 1425 } 1426 return err; 1427 } 1428 1429 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 1430 { 1431 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1432 if (lock) 1433 down_read(&sbi->node_change); 1434 else 1435 up_read(&sbi->node_change); 1436 } else { 1437 if (lock) 1438 f2fs_lock_op(sbi); 1439 else 1440 f2fs_unlock_op(sbi); 1441 } 1442 } 1443 1444 /* 1445 * f2fs_map_blocks() tries to find or build mapping relationship which 1446 * maps continuous logical blocks to physical blocks, and return such 1447 * info via f2fs_map_blocks structure. 1448 */ 1449 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 1450 int create, int flag) 1451 { 1452 unsigned int maxblocks = map->m_len; 1453 struct dnode_of_data dn; 1454 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1455 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE; 1456 pgoff_t pgofs, end_offset, end; 1457 int err = 0, ofs = 1; 1458 unsigned int ofs_in_node, last_ofs_in_node; 1459 blkcnt_t prealloc; 1460 struct extent_info ei = {0,0,0}; 1461 block_t blkaddr; 1462 unsigned int start_pgofs; 1463 1464 if (!maxblocks) 1465 return 0; 1466 1467 map->m_len = 0; 1468 map->m_flags = 0; 1469 1470 /* it only supports block size == page size */ 1471 pgofs = (pgoff_t)map->m_lblk; 1472 end = pgofs + maxblocks; 1473 1474 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 1475 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1476 map->m_may_create) 1477 goto next_dnode; 1478 1479 map->m_pblk = ei.blk + pgofs - ei.fofs; 1480 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 1481 map->m_flags = F2FS_MAP_MAPPED; 1482 if (map->m_next_extent) 1483 *map->m_next_extent = pgofs + map->m_len; 1484 1485 /* for hardware encryption, but to avoid potential issue in future */ 1486 if (flag == F2FS_GET_BLOCK_DIO) 1487 f2fs_wait_on_block_writeback_range(inode, 1488 map->m_pblk, map->m_len); 1489 goto out; 1490 } 1491 1492 next_dnode: 1493 if (map->m_may_create) 1494 f2fs_do_map_lock(sbi, flag, true); 1495 1496 /* When reading holes, we need its node page */ 1497 set_new_dnode(&dn, inode, NULL, NULL, 0); 1498 err = f2fs_get_dnode_of_data(&dn, pgofs, mode); 1499 if (err) { 1500 if (flag == F2FS_GET_BLOCK_BMAP) 1501 map->m_pblk = 0; 1502 if (err == -ENOENT) { 1503 err = 0; 1504 if (map->m_next_pgofs) 1505 *map->m_next_pgofs = 1506 f2fs_get_next_page_offset(&dn, pgofs); 1507 if (map->m_next_extent) 1508 *map->m_next_extent = 1509 f2fs_get_next_page_offset(&dn, pgofs); 1510 } 1511 goto unlock_out; 1512 } 1513 1514 start_pgofs = pgofs; 1515 prealloc = 0; 1516 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 1517 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 1518 1519 next_block: 1520 blkaddr = f2fs_data_blkaddr(&dn); 1521 1522 if (__is_valid_data_blkaddr(blkaddr) && 1523 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) { 1524 err = -EFSCORRUPTED; 1525 goto sync_out; 1526 } 1527 1528 if (__is_valid_data_blkaddr(blkaddr)) { 1529 /* use out-place-update for driect IO under LFS mode */ 1530 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1531 map->m_may_create) { 1532 err = __allocate_data_block(&dn, map->m_seg_type); 1533 if (err) 1534 goto sync_out; 1535 blkaddr = dn.data_blkaddr; 1536 set_inode_flag(inode, FI_APPEND_WRITE); 1537 } 1538 } else { 1539 if (create) { 1540 if (unlikely(f2fs_cp_error(sbi))) { 1541 err = -EIO; 1542 goto sync_out; 1543 } 1544 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1545 if (blkaddr == NULL_ADDR) { 1546 prealloc++; 1547 last_ofs_in_node = dn.ofs_in_node; 1548 } 1549 } else { 1550 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO && 1551 flag != F2FS_GET_BLOCK_DIO); 1552 err = __allocate_data_block(&dn, 1553 map->m_seg_type); 1554 if (!err) 1555 set_inode_flag(inode, FI_APPEND_WRITE); 1556 } 1557 if (err) 1558 goto sync_out; 1559 map->m_flags |= F2FS_MAP_NEW; 1560 blkaddr = dn.data_blkaddr; 1561 } else { 1562 if (flag == F2FS_GET_BLOCK_BMAP) { 1563 map->m_pblk = 0; 1564 goto sync_out; 1565 } 1566 if (flag == F2FS_GET_BLOCK_PRECACHE) 1567 goto sync_out; 1568 if (flag == F2FS_GET_BLOCK_FIEMAP && 1569 blkaddr == NULL_ADDR) { 1570 if (map->m_next_pgofs) 1571 *map->m_next_pgofs = pgofs + 1; 1572 goto sync_out; 1573 } 1574 if (flag != F2FS_GET_BLOCK_FIEMAP) { 1575 /* for defragment case */ 1576 if (map->m_next_pgofs) 1577 *map->m_next_pgofs = pgofs + 1; 1578 goto sync_out; 1579 } 1580 } 1581 } 1582 1583 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1584 goto skip; 1585 1586 if (map->m_len == 0) { 1587 /* preallocated unwritten block should be mapped for fiemap. */ 1588 if (blkaddr == NEW_ADDR) 1589 map->m_flags |= F2FS_MAP_UNWRITTEN; 1590 map->m_flags |= F2FS_MAP_MAPPED; 1591 1592 map->m_pblk = blkaddr; 1593 map->m_len = 1; 1594 } else if ((map->m_pblk != NEW_ADDR && 1595 blkaddr == (map->m_pblk + ofs)) || 1596 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1597 flag == F2FS_GET_BLOCK_PRE_DIO) { 1598 ofs++; 1599 map->m_len++; 1600 } else { 1601 goto sync_out; 1602 } 1603 1604 skip: 1605 dn.ofs_in_node++; 1606 pgofs++; 1607 1608 /* preallocate blocks in batch for one dnode page */ 1609 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1610 (pgofs == end || dn.ofs_in_node == end_offset)) { 1611 1612 dn.ofs_in_node = ofs_in_node; 1613 err = f2fs_reserve_new_blocks(&dn, prealloc); 1614 if (err) 1615 goto sync_out; 1616 1617 map->m_len += dn.ofs_in_node - ofs_in_node; 1618 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1619 err = -ENOSPC; 1620 goto sync_out; 1621 } 1622 dn.ofs_in_node = end_offset; 1623 } 1624 1625 if (pgofs >= end) 1626 goto sync_out; 1627 else if (dn.ofs_in_node < end_offset) 1628 goto next_block; 1629 1630 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1631 if (map->m_flags & F2FS_MAP_MAPPED) { 1632 unsigned int ofs = start_pgofs - map->m_lblk; 1633 1634 f2fs_update_extent_cache_range(&dn, 1635 start_pgofs, map->m_pblk + ofs, 1636 map->m_len - ofs); 1637 } 1638 } 1639 1640 f2fs_put_dnode(&dn); 1641 1642 if (map->m_may_create) { 1643 f2fs_do_map_lock(sbi, flag, false); 1644 f2fs_balance_fs(sbi, dn.node_changed); 1645 } 1646 goto next_dnode; 1647 1648 sync_out: 1649 1650 /* for hardware encryption, but to avoid potential issue in future */ 1651 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) 1652 f2fs_wait_on_block_writeback_range(inode, 1653 map->m_pblk, map->m_len); 1654 1655 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1656 if (map->m_flags & F2FS_MAP_MAPPED) { 1657 unsigned int ofs = start_pgofs - map->m_lblk; 1658 1659 f2fs_update_extent_cache_range(&dn, 1660 start_pgofs, map->m_pblk + ofs, 1661 map->m_len - ofs); 1662 } 1663 if (map->m_next_extent) 1664 *map->m_next_extent = pgofs + 1; 1665 } 1666 f2fs_put_dnode(&dn); 1667 unlock_out: 1668 if (map->m_may_create) { 1669 f2fs_do_map_lock(sbi, flag, false); 1670 f2fs_balance_fs(sbi, dn.node_changed); 1671 } 1672 out: 1673 trace_f2fs_map_blocks(inode, map, err); 1674 return err; 1675 } 1676 1677 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) 1678 { 1679 struct f2fs_map_blocks map; 1680 block_t last_lblk; 1681 int err; 1682 1683 if (pos + len > i_size_read(inode)) 1684 return false; 1685 1686 map.m_lblk = F2FS_BYTES_TO_BLK(pos); 1687 map.m_next_pgofs = NULL; 1688 map.m_next_extent = NULL; 1689 map.m_seg_type = NO_CHECK_TYPE; 1690 map.m_may_create = false; 1691 last_lblk = F2FS_BLK_ALIGN(pos + len); 1692 1693 while (map.m_lblk < last_lblk) { 1694 map.m_len = last_lblk - map.m_lblk; 1695 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT); 1696 if (err || map.m_len == 0) 1697 return false; 1698 map.m_lblk += map.m_len; 1699 } 1700 return true; 1701 } 1702 1703 static int __get_data_block(struct inode *inode, sector_t iblock, 1704 struct buffer_head *bh, int create, int flag, 1705 pgoff_t *next_pgofs, int seg_type, bool may_write) 1706 { 1707 struct f2fs_map_blocks map; 1708 int err; 1709 1710 map.m_lblk = iblock; 1711 map.m_len = bh->b_size >> inode->i_blkbits; 1712 map.m_next_pgofs = next_pgofs; 1713 map.m_next_extent = NULL; 1714 map.m_seg_type = seg_type; 1715 map.m_may_create = may_write; 1716 1717 err = f2fs_map_blocks(inode, &map, create, flag); 1718 if (!err) { 1719 map_bh(bh, inode->i_sb, map.m_pblk); 1720 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags; 1721 bh->b_size = (u64)map.m_len << inode->i_blkbits; 1722 } 1723 return err; 1724 } 1725 1726 static int get_data_block(struct inode *inode, sector_t iblock, 1727 struct buffer_head *bh_result, int create, int flag, 1728 pgoff_t *next_pgofs) 1729 { 1730 return __get_data_block(inode, iblock, bh_result, create, 1731 flag, next_pgofs, 1732 NO_CHECK_TYPE, create); 1733 } 1734 1735 static int get_data_block_dio_write(struct inode *inode, sector_t iblock, 1736 struct buffer_head *bh_result, int create) 1737 { 1738 return __get_data_block(inode, iblock, bh_result, create, 1739 F2FS_GET_BLOCK_DIO, NULL, 1740 f2fs_rw_hint_to_seg_type(inode->i_write_hint), 1741 IS_SWAPFILE(inode) ? false : true); 1742 } 1743 1744 static int get_data_block_dio(struct inode *inode, sector_t iblock, 1745 struct buffer_head *bh_result, int create) 1746 { 1747 return __get_data_block(inode, iblock, bh_result, create, 1748 F2FS_GET_BLOCK_DIO, NULL, 1749 f2fs_rw_hint_to_seg_type(inode->i_write_hint), 1750 false); 1751 } 1752 1753 static int get_data_block_bmap(struct inode *inode, sector_t iblock, 1754 struct buffer_head *bh_result, int create) 1755 { 1756 /* Block number less than F2FS MAX BLOCKS */ 1757 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks)) 1758 return -EFBIG; 1759 1760 return __get_data_block(inode, iblock, bh_result, create, 1761 F2FS_GET_BLOCK_BMAP, NULL, 1762 NO_CHECK_TYPE, create); 1763 } 1764 1765 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset) 1766 { 1767 return (offset >> inode->i_blkbits); 1768 } 1769 1770 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk) 1771 { 1772 return (blk << inode->i_blkbits); 1773 } 1774 1775 static int f2fs_xattr_fiemap(struct inode *inode, 1776 struct fiemap_extent_info *fieinfo) 1777 { 1778 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1779 struct page *page; 1780 struct node_info ni; 1781 __u64 phys = 0, len; 1782 __u32 flags; 1783 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1784 int err = 0; 1785 1786 if (f2fs_has_inline_xattr(inode)) { 1787 int offset; 1788 1789 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1790 inode->i_ino, false); 1791 if (!page) 1792 return -ENOMEM; 1793 1794 err = f2fs_get_node_info(sbi, inode->i_ino, &ni); 1795 if (err) { 1796 f2fs_put_page(page, 1); 1797 return err; 1798 } 1799 1800 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1801 offset = offsetof(struct f2fs_inode, i_addr) + 1802 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1803 get_inline_xattr_addrs(inode)); 1804 1805 phys += offset; 1806 len = inline_xattr_size(inode); 1807 1808 f2fs_put_page(page, 1); 1809 1810 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1811 1812 if (!xnid) 1813 flags |= FIEMAP_EXTENT_LAST; 1814 1815 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1816 if (err || err == 1) 1817 return err; 1818 } 1819 1820 if (xnid) { 1821 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1822 if (!page) 1823 return -ENOMEM; 1824 1825 err = f2fs_get_node_info(sbi, xnid, &ni); 1826 if (err) { 1827 f2fs_put_page(page, 1); 1828 return err; 1829 } 1830 1831 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1832 len = inode->i_sb->s_blocksize; 1833 1834 f2fs_put_page(page, 1); 1835 1836 flags = FIEMAP_EXTENT_LAST; 1837 } 1838 1839 if (phys) 1840 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1841 1842 return (err < 0 ? err : 0); 1843 } 1844 1845 static loff_t max_inode_blocks(struct inode *inode) 1846 { 1847 loff_t result = ADDRS_PER_INODE(inode); 1848 loff_t leaf_count = ADDRS_PER_BLOCK(inode); 1849 1850 /* two direct node blocks */ 1851 result += (leaf_count * 2); 1852 1853 /* two indirect node blocks */ 1854 leaf_count *= NIDS_PER_BLOCK; 1855 result += (leaf_count * 2); 1856 1857 /* one double indirect node block */ 1858 leaf_count *= NIDS_PER_BLOCK; 1859 result += leaf_count; 1860 1861 return result; 1862 } 1863 1864 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1865 u64 start, u64 len) 1866 { 1867 struct buffer_head map_bh; 1868 sector_t start_blk, last_blk; 1869 pgoff_t next_pgofs; 1870 u64 logical = 0, phys = 0, size = 0; 1871 u32 flags = 0; 1872 int ret = 0; 1873 bool compr_cluster = false; 1874 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size; 1875 1876 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { 1877 ret = f2fs_precache_extents(inode); 1878 if (ret) 1879 return ret; 1880 } 1881 1882 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR); 1883 if (ret) 1884 return ret; 1885 1886 inode_lock(inode); 1887 1888 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1889 ret = f2fs_xattr_fiemap(inode, fieinfo); 1890 goto out; 1891 } 1892 1893 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) { 1894 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1895 if (ret != -EAGAIN) 1896 goto out; 1897 } 1898 1899 if (logical_to_blk(inode, len) == 0) 1900 len = blk_to_logical(inode, 1); 1901 1902 start_blk = logical_to_blk(inode, start); 1903 last_blk = logical_to_blk(inode, start + len - 1); 1904 1905 next: 1906 memset(&map_bh, 0, sizeof(struct buffer_head)); 1907 map_bh.b_size = len; 1908 1909 if (compr_cluster) 1910 map_bh.b_size = blk_to_logical(inode, cluster_size - 1); 1911 1912 ret = get_data_block(inode, start_blk, &map_bh, 0, 1913 F2FS_GET_BLOCK_FIEMAP, &next_pgofs); 1914 if (ret) 1915 goto out; 1916 1917 /* HOLE */ 1918 if (!buffer_mapped(&map_bh)) { 1919 start_blk = next_pgofs; 1920 1921 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode, 1922 max_inode_blocks(inode))) 1923 goto prep_next; 1924 1925 flags |= FIEMAP_EXTENT_LAST; 1926 } 1927 1928 if (size) { 1929 if (IS_ENCRYPTED(inode)) 1930 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1931 1932 ret = fiemap_fill_next_extent(fieinfo, logical, 1933 phys, size, flags); 1934 if (ret) 1935 goto out; 1936 size = 0; 1937 } 1938 1939 if (start_blk > last_blk) 1940 goto out; 1941 1942 if (compr_cluster) { 1943 compr_cluster = false; 1944 1945 1946 logical = blk_to_logical(inode, start_blk - 1); 1947 phys = blk_to_logical(inode, map_bh.b_blocknr); 1948 size = blk_to_logical(inode, cluster_size); 1949 1950 flags |= FIEMAP_EXTENT_ENCODED; 1951 1952 start_blk += cluster_size - 1; 1953 1954 if (start_blk > last_blk) 1955 goto out; 1956 1957 goto prep_next; 1958 } 1959 1960 if (map_bh.b_blocknr == COMPRESS_ADDR) { 1961 compr_cluster = true; 1962 start_blk++; 1963 goto prep_next; 1964 } 1965 1966 logical = blk_to_logical(inode, start_blk); 1967 phys = blk_to_logical(inode, map_bh.b_blocknr); 1968 size = map_bh.b_size; 1969 flags = 0; 1970 if (buffer_unwritten(&map_bh)) 1971 flags = FIEMAP_EXTENT_UNWRITTEN; 1972 1973 start_blk += logical_to_blk(inode, size); 1974 1975 prep_next: 1976 cond_resched(); 1977 if (fatal_signal_pending(current)) 1978 ret = -EINTR; 1979 else 1980 goto next; 1981 out: 1982 if (ret == 1) 1983 ret = 0; 1984 1985 inode_unlock(inode); 1986 return ret; 1987 } 1988 1989 static inline loff_t f2fs_readpage_limit(struct inode *inode) 1990 { 1991 if (IS_ENABLED(CONFIG_FS_VERITY) && 1992 (IS_VERITY(inode) || f2fs_verity_in_progress(inode))) 1993 return inode->i_sb->s_maxbytes; 1994 1995 return i_size_read(inode); 1996 } 1997 1998 static int f2fs_read_single_page(struct inode *inode, struct page *page, 1999 unsigned nr_pages, 2000 struct f2fs_map_blocks *map, 2001 struct bio **bio_ret, 2002 sector_t *last_block_in_bio, 2003 bool is_readahead) 2004 { 2005 struct bio *bio = *bio_ret; 2006 const unsigned blkbits = inode->i_blkbits; 2007 const unsigned blocksize = 1 << blkbits; 2008 sector_t block_in_file; 2009 sector_t last_block; 2010 sector_t last_block_in_file; 2011 sector_t block_nr; 2012 int ret = 0; 2013 2014 block_in_file = (sector_t)page_index(page); 2015 last_block = block_in_file + nr_pages; 2016 last_block_in_file = (f2fs_readpage_limit(inode) + blocksize - 1) >> 2017 blkbits; 2018 if (last_block > last_block_in_file) 2019 last_block = last_block_in_file; 2020 2021 /* just zeroing out page which is beyond EOF */ 2022 if (block_in_file >= last_block) 2023 goto zero_out; 2024 /* 2025 * Map blocks using the previous result first. 2026 */ 2027 if ((map->m_flags & F2FS_MAP_MAPPED) && 2028 block_in_file > map->m_lblk && 2029 block_in_file < (map->m_lblk + map->m_len)) 2030 goto got_it; 2031 2032 /* 2033 * Then do more f2fs_map_blocks() calls until we are 2034 * done with this page. 2035 */ 2036 map->m_lblk = block_in_file; 2037 map->m_len = last_block - block_in_file; 2038 2039 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT); 2040 if (ret) 2041 goto out; 2042 got_it: 2043 if ((map->m_flags & F2FS_MAP_MAPPED)) { 2044 block_nr = map->m_pblk + block_in_file - map->m_lblk; 2045 SetPageMappedToDisk(page); 2046 2047 if (!PageUptodate(page) && (!PageSwapCache(page) && 2048 !cleancache_get_page(page))) { 2049 SetPageUptodate(page); 2050 goto confused; 2051 } 2052 2053 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr, 2054 DATA_GENERIC_ENHANCE_READ)) { 2055 ret = -EFSCORRUPTED; 2056 goto out; 2057 } 2058 } else { 2059 zero_out: 2060 zero_user_segment(page, 0, PAGE_SIZE); 2061 if (f2fs_need_verity(inode, page->index) && 2062 !fsverity_verify_page(page)) { 2063 ret = -EIO; 2064 goto out; 2065 } 2066 if (!PageUptodate(page)) 2067 SetPageUptodate(page); 2068 unlock_page(page); 2069 goto out; 2070 } 2071 2072 /* 2073 * This page will go to BIO. Do we need to send this 2074 * BIO off first? 2075 */ 2076 if (bio && !page_is_mergeable(F2FS_I_SB(inode), bio, 2077 *last_block_in_bio, block_nr)) { 2078 submit_and_realloc: 2079 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2080 bio = NULL; 2081 } 2082 if (bio == NULL) { 2083 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages, 2084 is_readahead ? REQ_RAHEAD : 0, page->index, 2085 false); 2086 if (IS_ERR(bio)) { 2087 ret = PTR_ERR(bio); 2088 bio = NULL; 2089 goto out; 2090 } 2091 } 2092 2093 /* 2094 * If the page is under writeback, we need to wait for 2095 * its completion to see the correct decrypted data. 2096 */ 2097 f2fs_wait_on_block_writeback(inode, block_nr); 2098 2099 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2100 goto submit_and_realloc; 2101 2102 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 2103 f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE); 2104 ClearPageError(page); 2105 *last_block_in_bio = block_nr; 2106 goto out; 2107 confused: 2108 if (bio) { 2109 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2110 bio = NULL; 2111 } 2112 unlock_page(page); 2113 out: 2114 *bio_ret = bio; 2115 return ret; 2116 } 2117 2118 #ifdef CONFIG_F2FS_FS_COMPRESSION 2119 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret, 2120 unsigned nr_pages, sector_t *last_block_in_bio, 2121 bool is_readahead, bool for_write) 2122 { 2123 struct dnode_of_data dn; 2124 struct inode *inode = cc->inode; 2125 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2126 struct bio *bio = *bio_ret; 2127 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size; 2128 sector_t last_block_in_file; 2129 const unsigned blkbits = inode->i_blkbits; 2130 const unsigned blocksize = 1 << blkbits; 2131 struct decompress_io_ctx *dic = NULL; 2132 int i; 2133 int ret = 0; 2134 2135 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc)); 2136 2137 last_block_in_file = (f2fs_readpage_limit(inode) + 2138 blocksize - 1) >> blkbits; 2139 2140 /* get rid of pages beyond EOF */ 2141 for (i = 0; i < cc->cluster_size; i++) { 2142 struct page *page = cc->rpages[i]; 2143 2144 if (!page) 2145 continue; 2146 if ((sector_t)page->index >= last_block_in_file) { 2147 zero_user_segment(page, 0, PAGE_SIZE); 2148 if (!PageUptodate(page)) 2149 SetPageUptodate(page); 2150 } else if (!PageUptodate(page)) { 2151 continue; 2152 } 2153 unlock_page(page); 2154 cc->rpages[i] = NULL; 2155 cc->nr_rpages--; 2156 } 2157 2158 /* we are done since all pages are beyond EOF */ 2159 if (f2fs_cluster_is_empty(cc)) 2160 goto out; 2161 2162 set_new_dnode(&dn, inode, NULL, NULL, 0); 2163 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 2164 if (ret) 2165 goto out; 2166 2167 /* cluster was overwritten as normal cluster */ 2168 if (dn.data_blkaddr != COMPRESS_ADDR) 2169 goto out; 2170 2171 for (i = 1; i < cc->cluster_size; i++) { 2172 block_t blkaddr; 2173 2174 blkaddr = data_blkaddr(dn.inode, dn.node_page, 2175 dn.ofs_in_node + i); 2176 2177 if (!__is_valid_data_blkaddr(blkaddr)) 2178 break; 2179 2180 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) { 2181 ret = -EFAULT; 2182 goto out_put_dnode; 2183 } 2184 cc->nr_cpages++; 2185 } 2186 2187 /* nothing to decompress */ 2188 if (cc->nr_cpages == 0) { 2189 ret = 0; 2190 goto out_put_dnode; 2191 } 2192 2193 dic = f2fs_alloc_dic(cc); 2194 if (IS_ERR(dic)) { 2195 ret = PTR_ERR(dic); 2196 goto out_put_dnode; 2197 } 2198 2199 for (i = 0; i < dic->nr_cpages; i++) { 2200 struct page *page = dic->cpages[i]; 2201 block_t blkaddr; 2202 struct bio_post_read_ctx *ctx; 2203 2204 blkaddr = data_blkaddr(dn.inode, dn.node_page, 2205 dn.ofs_in_node + i + 1); 2206 2207 if (bio && !page_is_mergeable(sbi, bio, 2208 *last_block_in_bio, blkaddr)) { 2209 submit_and_realloc: 2210 __submit_bio(sbi, bio, DATA); 2211 bio = NULL; 2212 } 2213 2214 if (!bio) { 2215 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages, 2216 is_readahead ? REQ_RAHEAD : 0, 2217 page->index, for_write); 2218 if (IS_ERR(bio)) { 2219 ret = PTR_ERR(bio); 2220 dic->failed = true; 2221 if (refcount_sub_and_test(dic->nr_cpages - i, 2222 &dic->ref)) { 2223 f2fs_decompress_end_io(dic->rpages, 2224 cc->cluster_size, true, 2225 false); 2226 f2fs_free_dic(dic); 2227 } 2228 f2fs_put_dnode(&dn); 2229 *bio_ret = NULL; 2230 return ret; 2231 } 2232 } 2233 2234 f2fs_wait_on_block_writeback(inode, blkaddr); 2235 2236 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2237 goto submit_and_realloc; 2238 2239 /* tag STEP_DECOMPRESS to handle IO in wq */ 2240 ctx = bio->bi_private; 2241 if (!(ctx->enabled_steps & (1 << STEP_DECOMPRESS))) 2242 ctx->enabled_steps |= 1 << STEP_DECOMPRESS; 2243 2244 inc_page_count(sbi, F2FS_RD_DATA); 2245 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); 2246 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE); 2247 ClearPageError(page); 2248 *last_block_in_bio = blkaddr; 2249 } 2250 2251 f2fs_put_dnode(&dn); 2252 2253 *bio_ret = bio; 2254 return 0; 2255 2256 out_put_dnode: 2257 f2fs_put_dnode(&dn); 2258 out: 2259 f2fs_decompress_end_io(cc->rpages, cc->cluster_size, true, false); 2260 *bio_ret = bio; 2261 return ret; 2262 } 2263 #endif 2264 2265 /* 2266 * This function was originally taken from fs/mpage.c, and customized for f2fs. 2267 * Major change was from block_size == page_size in f2fs by default. 2268 * 2269 * Note that the aops->readpages() function is ONLY used for read-ahead. If 2270 * this function ever deviates from doing just read-ahead, it should either 2271 * use ->readpage() or do the necessary surgery to decouple ->readpages() 2272 * from read-ahead. 2273 */ 2274 static int f2fs_mpage_readpages(struct inode *inode, 2275 struct readahead_control *rac, struct page *page) 2276 { 2277 struct bio *bio = NULL; 2278 sector_t last_block_in_bio = 0; 2279 struct f2fs_map_blocks map; 2280 #ifdef CONFIG_F2FS_FS_COMPRESSION 2281 struct compress_ctx cc = { 2282 .inode = inode, 2283 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2284 .cluster_size = F2FS_I(inode)->i_cluster_size, 2285 .cluster_idx = NULL_CLUSTER, 2286 .rpages = NULL, 2287 .cpages = NULL, 2288 .nr_rpages = 0, 2289 .nr_cpages = 0, 2290 }; 2291 #endif 2292 unsigned nr_pages = rac ? readahead_count(rac) : 1; 2293 unsigned max_nr_pages = nr_pages; 2294 int ret = 0; 2295 2296 map.m_pblk = 0; 2297 map.m_lblk = 0; 2298 map.m_len = 0; 2299 map.m_flags = 0; 2300 map.m_next_pgofs = NULL; 2301 map.m_next_extent = NULL; 2302 map.m_seg_type = NO_CHECK_TYPE; 2303 map.m_may_create = false; 2304 2305 for (; nr_pages; nr_pages--) { 2306 if (rac) { 2307 page = readahead_page(rac); 2308 prefetchw(&page->flags); 2309 } 2310 2311 #ifdef CONFIG_F2FS_FS_COMPRESSION 2312 if (f2fs_compressed_file(inode)) { 2313 /* there are remained comressed pages, submit them */ 2314 if (!f2fs_cluster_can_merge_page(&cc, page->index)) { 2315 ret = f2fs_read_multi_pages(&cc, &bio, 2316 max_nr_pages, 2317 &last_block_in_bio, 2318 rac != NULL, false); 2319 f2fs_destroy_compress_ctx(&cc); 2320 if (ret) 2321 goto set_error_page; 2322 } 2323 ret = f2fs_is_compressed_cluster(inode, page->index); 2324 if (ret < 0) 2325 goto set_error_page; 2326 else if (!ret) 2327 goto read_single_page; 2328 2329 ret = f2fs_init_compress_ctx(&cc); 2330 if (ret) 2331 goto set_error_page; 2332 2333 f2fs_compress_ctx_add_page(&cc, page); 2334 2335 goto next_page; 2336 } 2337 read_single_page: 2338 #endif 2339 2340 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map, 2341 &bio, &last_block_in_bio, rac); 2342 if (ret) { 2343 #ifdef CONFIG_F2FS_FS_COMPRESSION 2344 set_error_page: 2345 #endif 2346 SetPageError(page); 2347 zero_user_segment(page, 0, PAGE_SIZE); 2348 unlock_page(page); 2349 } 2350 #ifdef CONFIG_F2FS_FS_COMPRESSION 2351 next_page: 2352 #endif 2353 if (rac) 2354 put_page(page); 2355 2356 #ifdef CONFIG_F2FS_FS_COMPRESSION 2357 if (f2fs_compressed_file(inode)) { 2358 /* last page */ 2359 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) { 2360 ret = f2fs_read_multi_pages(&cc, &bio, 2361 max_nr_pages, 2362 &last_block_in_bio, 2363 rac != NULL, false); 2364 f2fs_destroy_compress_ctx(&cc); 2365 } 2366 } 2367 #endif 2368 } 2369 if (bio) 2370 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2371 return ret; 2372 } 2373 2374 static int f2fs_read_data_page(struct file *file, struct page *page) 2375 { 2376 struct inode *inode = page_file_mapping(page)->host; 2377 int ret = -EAGAIN; 2378 2379 trace_f2fs_readpage(page, DATA); 2380 2381 if (!f2fs_is_compress_backend_ready(inode)) { 2382 unlock_page(page); 2383 return -EOPNOTSUPP; 2384 } 2385 2386 /* If the file has inline data, try to read it directly */ 2387 if (f2fs_has_inline_data(inode)) 2388 ret = f2fs_read_inline_data(inode, page); 2389 if (ret == -EAGAIN) 2390 ret = f2fs_mpage_readpages(inode, NULL, page); 2391 return ret; 2392 } 2393 2394 static void f2fs_readahead(struct readahead_control *rac) 2395 { 2396 struct inode *inode = rac->mapping->host; 2397 2398 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac)); 2399 2400 if (!f2fs_is_compress_backend_ready(inode)) 2401 return; 2402 2403 /* If the file has inline data, skip readpages */ 2404 if (f2fs_has_inline_data(inode)) 2405 return; 2406 2407 f2fs_mpage_readpages(inode, rac, NULL); 2408 } 2409 2410 int f2fs_encrypt_one_page(struct f2fs_io_info *fio) 2411 { 2412 struct inode *inode = fio->page->mapping->host; 2413 struct page *mpage, *page; 2414 gfp_t gfp_flags = GFP_NOFS; 2415 2416 if (!f2fs_encrypted_file(inode)) 2417 return 0; 2418 2419 page = fio->compressed_page ? fio->compressed_page : fio->page; 2420 2421 /* wait for GCed page writeback via META_MAPPING */ 2422 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr); 2423 2424 retry_encrypt: 2425 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page, 2426 PAGE_SIZE, 0, gfp_flags); 2427 if (IS_ERR(fio->encrypted_page)) { 2428 /* flush pending IOs and wait for a while in the ENOMEM case */ 2429 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 2430 f2fs_flush_merged_writes(fio->sbi); 2431 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 2432 gfp_flags |= __GFP_NOFAIL; 2433 goto retry_encrypt; 2434 } 2435 return PTR_ERR(fio->encrypted_page); 2436 } 2437 2438 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr); 2439 if (mpage) { 2440 if (PageUptodate(mpage)) 2441 memcpy(page_address(mpage), 2442 page_address(fio->encrypted_page), PAGE_SIZE); 2443 f2fs_put_page(mpage, 1); 2444 } 2445 return 0; 2446 } 2447 2448 static inline bool check_inplace_update_policy(struct inode *inode, 2449 struct f2fs_io_info *fio) 2450 { 2451 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2452 unsigned int policy = SM_I(sbi)->ipu_policy; 2453 2454 if (policy & (0x1 << F2FS_IPU_FORCE)) 2455 return true; 2456 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi)) 2457 return true; 2458 if (policy & (0x1 << F2FS_IPU_UTIL) && 2459 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2460 return true; 2461 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) && 2462 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2463 return true; 2464 2465 /* 2466 * IPU for rewrite async pages 2467 */ 2468 if (policy & (0x1 << F2FS_IPU_ASYNC) && 2469 fio && fio->op == REQ_OP_WRITE && 2470 !(fio->op_flags & REQ_SYNC) && 2471 !IS_ENCRYPTED(inode)) 2472 return true; 2473 2474 /* this is only set during fdatasync */ 2475 if (policy & (0x1 << F2FS_IPU_FSYNC) && 2476 is_inode_flag_set(inode, FI_NEED_IPU)) 2477 return true; 2478 2479 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2480 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2481 return true; 2482 2483 return false; 2484 } 2485 2486 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) 2487 { 2488 if (f2fs_is_pinned_file(inode)) 2489 return true; 2490 2491 /* if this is cold file, we should overwrite to avoid fragmentation */ 2492 if (file_is_cold(inode)) 2493 return true; 2494 2495 return check_inplace_update_policy(inode, fio); 2496 } 2497 2498 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) 2499 { 2500 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2501 2502 if (f2fs_lfs_mode(sbi)) 2503 return true; 2504 if (S_ISDIR(inode->i_mode)) 2505 return true; 2506 if (IS_NOQUOTA(inode)) 2507 return true; 2508 if (f2fs_is_atomic_file(inode)) 2509 return true; 2510 if (fio) { 2511 if (is_cold_data(fio->page)) 2512 return true; 2513 if (IS_ATOMIC_WRITTEN_PAGE(fio->page)) 2514 return true; 2515 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2516 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2517 return true; 2518 } 2519 return false; 2520 } 2521 2522 static inline bool need_inplace_update(struct f2fs_io_info *fio) 2523 { 2524 struct inode *inode = fio->page->mapping->host; 2525 2526 if (f2fs_should_update_outplace(inode, fio)) 2527 return false; 2528 2529 return f2fs_should_update_inplace(inode, fio); 2530 } 2531 2532 int f2fs_do_write_data_page(struct f2fs_io_info *fio) 2533 { 2534 struct page *page = fio->page; 2535 struct inode *inode = page->mapping->host; 2536 struct dnode_of_data dn; 2537 struct extent_info ei = {0,0,0}; 2538 struct node_info ni; 2539 bool ipu_force = false; 2540 int err = 0; 2541 2542 set_new_dnode(&dn, inode, NULL, NULL, 0); 2543 if (need_inplace_update(fio) && 2544 f2fs_lookup_extent_cache(inode, page->index, &ei)) { 2545 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 2546 2547 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2548 DATA_GENERIC_ENHANCE)) 2549 return -EFSCORRUPTED; 2550 2551 ipu_force = true; 2552 fio->need_lock = LOCK_DONE; 2553 goto got_it; 2554 } 2555 2556 /* Deadlock due to between page->lock and f2fs_lock_op */ 2557 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 2558 return -EAGAIN; 2559 2560 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 2561 if (err) 2562 goto out; 2563 2564 fio->old_blkaddr = dn.data_blkaddr; 2565 2566 /* This page is already truncated */ 2567 if (fio->old_blkaddr == NULL_ADDR) { 2568 ClearPageUptodate(page); 2569 clear_cold_data(page); 2570 goto out_writepage; 2571 } 2572 got_it: 2573 if (__is_valid_data_blkaddr(fio->old_blkaddr) && 2574 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2575 DATA_GENERIC_ENHANCE)) { 2576 err = -EFSCORRUPTED; 2577 goto out_writepage; 2578 } 2579 /* 2580 * If current allocation needs SSR, 2581 * it had better in-place writes for updated data. 2582 */ 2583 if (ipu_force || 2584 (__is_valid_data_blkaddr(fio->old_blkaddr) && 2585 need_inplace_update(fio))) { 2586 err = f2fs_encrypt_one_page(fio); 2587 if (err) 2588 goto out_writepage; 2589 2590 set_page_writeback(page); 2591 ClearPageError(page); 2592 f2fs_put_dnode(&dn); 2593 if (fio->need_lock == LOCK_REQ) 2594 f2fs_unlock_op(fio->sbi); 2595 err = f2fs_inplace_write_data(fio); 2596 if (err) { 2597 if (f2fs_encrypted_file(inode)) 2598 fscrypt_finalize_bounce_page(&fio->encrypted_page); 2599 if (PageWriteback(page)) 2600 end_page_writeback(page); 2601 } else { 2602 set_inode_flag(inode, FI_UPDATE_WRITE); 2603 } 2604 trace_f2fs_do_write_data_page(fio->page, IPU); 2605 return err; 2606 } 2607 2608 if (fio->need_lock == LOCK_RETRY) { 2609 if (!f2fs_trylock_op(fio->sbi)) { 2610 err = -EAGAIN; 2611 goto out_writepage; 2612 } 2613 fio->need_lock = LOCK_REQ; 2614 } 2615 2616 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni); 2617 if (err) 2618 goto out_writepage; 2619 2620 fio->version = ni.version; 2621 2622 err = f2fs_encrypt_one_page(fio); 2623 if (err) 2624 goto out_writepage; 2625 2626 set_page_writeback(page); 2627 ClearPageError(page); 2628 2629 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR) 2630 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false); 2631 2632 /* LFS mode write path */ 2633 f2fs_outplace_write_data(&dn, fio); 2634 trace_f2fs_do_write_data_page(page, OPU); 2635 set_inode_flag(inode, FI_APPEND_WRITE); 2636 if (page->index == 0) 2637 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 2638 out_writepage: 2639 f2fs_put_dnode(&dn); 2640 out: 2641 if (fio->need_lock == LOCK_REQ) 2642 f2fs_unlock_op(fio->sbi); 2643 return err; 2644 } 2645 2646 int f2fs_write_single_data_page(struct page *page, int *submitted, 2647 struct bio **bio, 2648 sector_t *last_block, 2649 struct writeback_control *wbc, 2650 enum iostat_type io_type, 2651 int compr_blocks) 2652 { 2653 struct inode *inode = page->mapping->host; 2654 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2655 loff_t i_size = i_size_read(inode); 2656 const pgoff_t end_index = ((unsigned long long)i_size) 2657 >> PAGE_SHIFT; 2658 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT; 2659 unsigned offset = 0; 2660 bool need_balance_fs = false; 2661 int err = 0; 2662 struct f2fs_io_info fio = { 2663 .sbi = sbi, 2664 .ino = inode->i_ino, 2665 .type = DATA, 2666 .op = REQ_OP_WRITE, 2667 .op_flags = wbc_to_write_flags(wbc), 2668 .old_blkaddr = NULL_ADDR, 2669 .page = page, 2670 .encrypted_page = NULL, 2671 .submitted = false, 2672 .compr_blocks = compr_blocks, 2673 .need_lock = LOCK_RETRY, 2674 .io_type = io_type, 2675 .io_wbc = wbc, 2676 .bio = bio, 2677 .last_block = last_block, 2678 }; 2679 2680 trace_f2fs_writepage(page, DATA); 2681 2682 /* we should bypass data pages to proceed the kworkder jobs */ 2683 if (unlikely(f2fs_cp_error(sbi))) { 2684 mapping_set_error(page->mapping, -EIO); 2685 /* 2686 * don't drop any dirty dentry pages for keeping lastest 2687 * directory structure. 2688 */ 2689 if (S_ISDIR(inode->i_mode)) 2690 goto redirty_out; 2691 goto out; 2692 } 2693 2694 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2695 goto redirty_out; 2696 2697 if (page->index < end_index || 2698 f2fs_verity_in_progress(inode) || 2699 compr_blocks) 2700 goto write; 2701 2702 /* 2703 * If the offset is out-of-range of file size, 2704 * this page does not have to be written to disk. 2705 */ 2706 offset = i_size & (PAGE_SIZE - 1); 2707 if ((page->index >= end_index + 1) || !offset) 2708 goto out; 2709 2710 zero_user_segment(page, offset, PAGE_SIZE); 2711 write: 2712 if (f2fs_is_drop_cache(inode)) 2713 goto out; 2714 /* we should not write 0'th page having journal header */ 2715 if (f2fs_is_volatile_file(inode) && (!page->index || 2716 (!wbc->for_reclaim && 2717 f2fs_available_free_memory(sbi, BASE_CHECK)))) 2718 goto redirty_out; 2719 2720 /* Dentry/quota blocks are controlled by checkpoint */ 2721 if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) { 2722 /* 2723 * We need to wait for node_write to avoid block allocation during 2724 * checkpoint. This can only happen to quota writes which can cause 2725 * the below discard race condition. 2726 */ 2727 if (IS_NOQUOTA(inode)) 2728 down_read(&sbi->node_write); 2729 2730 fio.need_lock = LOCK_DONE; 2731 err = f2fs_do_write_data_page(&fio); 2732 2733 if (IS_NOQUOTA(inode)) 2734 up_read(&sbi->node_write); 2735 2736 goto done; 2737 } 2738 2739 if (!wbc->for_reclaim) 2740 need_balance_fs = true; 2741 else if (has_not_enough_free_secs(sbi, 0, 0)) 2742 goto redirty_out; 2743 else 2744 set_inode_flag(inode, FI_HOT_DATA); 2745 2746 err = -EAGAIN; 2747 if (f2fs_has_inline_data(inode)) { 2748 err = f2fs_write_inline_data(inode, page); 2749 if (!err) 2750 goto out; 2751 } 2752 2753 if (err == -EAGAIN) { 2754 err = f2fs_do_write_data_page(&fio); 2755 if (err == -EAGAIN) { 2756 fio.need_lock = LOCK_REQ; 2757 err = f2fs_do_write_data_page(&fio); 2758 } 2759 } 2760 2761 if (err) { 2762 file_set_keep_isize(inode); 2763 } else { 2764 spin_lock(&F2FS_I(inode)->i_size_lock); 2765 if (F2FS_I(inode)->last_disk_size < psize) 2766 F2FS_I(inode)->last_disk_size = psize; 2767 spin_unlock(&F2FS_I(inode)->i_size_lock); 2768 } 2769 2770 done: 2771 if (err && err != -ENOENT) 2772 goto redirty_out; 2773 2774 out: 2775 inode_dec_dirty_pages(inode); 2776 if (err) { 2777 ClearPageUptodate(page); 2778 clear_cold_data(page); 2779 } 2780 2781 if (wbc->for_reclaim) { 2782 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA); 2783 clear_inode_flag(inode, FI_HOT_DATA); 2784 f2fs_remove_dirty_inode(inode); 2785 submitted = NULL; 2786 } 2787 unlock_page(page); 2788 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) && 2789 !F2FS_I(inode)->cp_task) 2790 f2fs_balance_fs(sbi, need_balance_fs); 2791 2792 if (unlikely(f2fs_cp_error(sbi))) { 2793 f2fs_submit_merged_write(sbi, DATA); 2794 f2fs_submit_merged_ipu_write(sbi, bio, NULL); 2795 submitted = NULL; 2796 } 2797 2798 if (submitted) 2799 *submitted = fio.submitted ? 1 : 0; 2800 2801 return 0; 2802 2803 redirty_out: 2804 redirty_page_for_writepage(wbc, page); 2805 /* 2806 * pageout() in MM traslates EAGAIN, so calls handle_write_error() 2807 * -> mapping_set_error() -> set_bit(AS_EIO, ...). 2808 * file_write_and_wait_range() will see EIO error, which is critical 2809 * to return value of fsync() followed by atomic_write failure to user. 2810 */ 2811 if (!err || wbc->for_reclaim) 2812 return AOP_WRITEPAGE_ACTIVATE; 2813 unlock_page(page); 2814 return err; 2815 } 2816 2817 static int f2fs_write_data_page(struct page *page, 2818 struct writeback_control *wbc) 2819 { 2820 #ifdef CONFIG_F2FS_FS_COMPRESSION 2821 struct inode *inode = page->mapping->host; 2822 2823 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 2824 goto out; 2825 2826 if (f2fs_compressed_file(inode)) { 2827 if (f2fs_is_compressed_cluster(inode, page->index)) { 2828 redirty_page_for_writepage(wbc, page); 2829 return AOP_WRITEPAGE_ACTIVATE; 2830 } 2831 } 2832 out: 2833 #endif 2834 2835 return f2fs_write_single_data_page(page, NULL, NULL, NULL, 2836 wbc, FS_DATA_IO, 0); 2837 } 2838 2839 /* 2840 * This function was copied from write_cche_pages from mm/page-writeback.c. 2841 * The major change is making write step of cold data page separately from 2842 * warm/hot data page. 2843 */ 2844 static int f2fs_write_cache_pages(struct address_space *mapping, 2845 struct writeback_control *wbc, 2846 enum iostat_type io_type) 2847 { 2848 int ret = 0; 2849 int done = 0, retry = 0; 2850 struct pagevec pvec; 2851 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2852 struct bio *bio = NULL; 2853 sector_t last_block; 2854 #ifdef CONFIG_F2FS_FS_COMPRESSION 2855 struct inode *inode = mapping->host; 2856 struct compress_ctx cc = { 2857 .inode = inode, 2858 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2859 .cluster_size = F2FS_I(inode)->i_cluster_size, 2860 .cluster_idx = NULL_CLUSTER, 2861 .rpages = NULL, 2862 .nr_rpages = 0, 2863 .cpages = NULL, 2864 .rbuf = NULL, 2865 .cbuf = NULL, 2866 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size, 2867 .private = NULL, 2868 }; 2869 #endif 2870 int nr_pages; 2871 pgoff_t uninitialized_var(writeback_index); 2872 pgoff_t index; 2873 pgoff_t end; /* Inclusive */ 2874 pgoff_t done_index; 2875 int range_whole = 0; 2876 xa_mark_t tag; 2877 int nwritten = 0; 2878 int submitted = 0; 2879 int i; 2880 2881 pagevec_init(&pvec); 2882 2883 if (get_dirty_pages(mapping->host) <= 2884 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 2885 set_inode_flag(mapping->host, FI_HOT_DATA); 2886 else 2887 clear_inode_flag(mapping->host, FI_HOT_DATA); 2888 2889 if (wbc->range_cyclic) { 2890 writeback_index = mapping->writeback_index; /* prev offset */ 2891 index = writeback_index; 2892 end = -1; 2893 } else { 2894 index = wbc->range_start >> PAGE_SHIFT; 2895 end = wbc->range_end >> PAGE_SHIFT; 2896 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2897 range_whole = 1; 2898 } 2899 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2900 tag = PAGECACHE_TAG_TOWRITE; 2901 else 2902 tag = PAGECACHE_TAG_DIRTY; 2903 retry: 2904 retry = 0; 2905 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2906 tag_pages_for_writeback(mapping, index, end); 2907 done_index = index; 2908 while (!done && !retry && (index <= end)) { 2909 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 2910 tag); 2911 if (nr_pages == 0) 2912 break; 2913 2914 for (i = 0; i < nr_pages; i++) { 2915 struct page *page = pvec.pages[i]; 2916 bool need_readd; 2917 readd: 2918 need_readd = false; 2919 #ifdef CONFIG_F2FS_FS_COMPRESSION 2920 if (f2fs_compressed_file(inode)) { 2921 ret = f2fs_init_compress_ctx(&cc); 2922 if (ret) { 2923 done = 1; 2924 break; 2925 } 2926 2927 if (!f2fs_cluster_can_merge_page(&cc, 2928 page->index)) { 2929 ret = f2fs_write_multi_pages(&cc, 2930 &submitted, wbc, io_type); 2931 if (!ret) 2932 need_readd = true; 2933 goto result; 2934 } 2935 2936 if (unlikely(f2fs_cp_error(sbi))) 2937 goto lock_page; 2938 2939 if (f2fs_cluster_is_empty(&cc)) { 2940 void *fsdata = NULL; 2941 struct page *pagep; 2942 int ret2; 2943 2944 ret2 = f2fs_prepare_compress_overwrite( 2945 inode, &pagep, 2946 page->index, &fsdata); 2947 if (ret2 < 0) { 2948 ret = ret2; 2949 done = 1; 2950 break; 2951 } else if (ret2 && 2952 !f2fs_compress_write_end(inode, 2953 fsdata, page->index, 2954 1)) { 2955 retry = 1; 2956 break; 2957 } 2958 } else { 2959 goto lock_page; 2960 } 2961 } 2962 #endif 2963 /* give a priority to WB_SYNC threads */ 2964 if (atomic_read(&sbi->wb_sync_req[DATA]) && 2965 wbc->sync_mode == WB_SYNC_NONE) { 2966 done = 1; 2967 break; 2968 } 2969 #ifdef CONFIG_F2FS_FS_COMPRESSION 2970 lock_page: 2971 #endif 2972 done_index = page->index; 2973 retry_write: 2974 lock_page(page); 2975 2976 if (unlikely(page->mapping != mapping)) { 2977 continue_unlock: 2978 unlock_page(page); 2979 continue; 2980 } 2981 2982 if (!PageDirty(page)) { 2983 /* someone wrote it for us */ 2984 goto continue_unlock; 2985 } 2986 2987 if (PageWriteback(page)) { 2988 if (wbc->sync_mode != WB_SYNC_NONE) 2989 f2fs_wait_on_page_writeback(page, 2990 DATA, true, true); 2991 else 2992 goto continue_unlock; 2993 } 2994 2995 if (!clear_page_dirty_for_io(page)) 2996 goto continue_unlock; 2997 2998 #ifdef CONFIG_F2FS_FS_COMPRESSION 2999 if (f2fs_compressed_file(inode)) { 3000 get_page(page); 3001 f2fs_compress_ctx_add_page(&cc, page); 3002 continue; 3003 } 3004 #endif 3005 ret = f2fs_write_single_data_page(page, &submitted, 3006 &bio, &last_block, wbc, io_type, 0); 3007 if (ret == AOP_WRITEPAGE_ACTIVATE) 3008 unlock_page(page); 3009 #ifdef CONFIG_F2FS_FS_COMPRESSION 3010 result: 3011 #endif 3012 nwritten += submitted; 3013 wbc->nr_to_write -= submitted; 3014 3015 if (unlikely(ret)) { 3016 /* 3017 * keep nr_to_write, since vfs uses this to 3018 * get # of written pages. 3019 */ 3020 if (ret == AOP_WRITEPAGE_ACTIVATE) { 3021 ret = 0; 3022 goto next; 3023 } else if (ret == -EAGAIN) { 3024 ret = 0; 3025 if (wbc->sync_mode == WB_SYNC_ALL) { 3026 cond_resched(); 3027 congestion_wait(BLK_RW_ASYNC, 3028 DEFAULT_IO_TIMEOUT); 3029 goto retry_write; 3030 } 3031 goto next; 3032 } 3033 done_index = page->index + 1; 3034 done = 1; 3035 break; 3036 } 3037 3038 if (wbc->nr_to_write <= 0 && 3039 wbc->sync_mode == WB_SYNC_NONE) { 3040 done = 1; 3041 break; 3042 } 3043 next: 3044 if (need_readd) 3045 goto readd; 3046 } 3047 pagevec_release(&pvec); 3048 cond_resched(); 3049 } 3050 #ifdef CONFIG_F2FS_FS_COMPRESSION 3051 /* flush remained pages in compress cluster */ 3052 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) { 3053 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type); 3054 nwritten += submitted; 3055 wbc->nr_to_write -= submitted; 3056 if (ret) { 3057 done = 1; 3058 retry = 0; 3059 } 3060 } 3061 #endif 3062 if (retry) { 3063 index = 0; 3064 end = -1; 3065 goto retry; 3066 } 3067 if (wbc->range_cyclic && !done) 3068 done_index = 0; 3069 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 3070 mapping->writeback_index = done_index; 3071 3072 if (nwritten) 3073 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 3074 NULL, 0, DATA); 3075 /* submit cached bio of IPU write */ 3076 if (bio) 3077 f2fs_submit_merged_ipu_write(sbi, &bio, NULL); 3078 3079 return ret; 3080 } 3081 3082 static inline bool __should_serialize_io(struct inode *inode, 3083 struct writeback_control *wbc) 3084 { 3085 /* to avoid deadlock in path of data flush */ 3086 if (F2FS_I(inode)->cp_task) 3087 return false; 3088 3089 if (!S_ISREG(inode->i_mode)) 3090 return false; 3091 if (IS_NOQUOTA(inode)) 3092 return false; 3093 3094 if (f2fs_compressed_file(inode)) 3095 return true; 3096 if (wbc->sync_mode != WB_SYNC_ALL) 3097 return true; 3098 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks) 3099 return true; 3100 return false; 3101 } 3102 3103 static int __f2fs_write_data_pages(struct address_space *mapping, 3104 struct writeback_control *wbc, 3105 enum iostat_type io_type) 3106 { 3107 struct inode *inode = mapping->host; 3108 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3109 struct blk_plug plug; 3110 int ret; 3111 bool locked = false; 3112 3113 /* deal with chardevs and other special file */ 3114 if (!mapping->a_ops->writepage) 3115 return 0; 3116 3117 /* skip writing if there is no dirty page in this inode */ 3118 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 3119 return 0; 3120 3121 /* during POR, we don't need to trigger writepage at all. */ 3122 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 3123 goto skip_write; 3124 3125 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) && 3126 wbc->sync_mode == WB_SYNC_NONE && 3127 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 3128 f2fs_available_free_memory(sbi, DIRTY_DENTS)) 3129 goto skip_write; 3130 3131 /* skip writing during file defragment */ 3132 if (is_inode_flag_set(inode, FI_DO_DEFRAG)) 3133 goto skip_write; 3134 3135 trace_f2fs_writepages(mapping->host, wbc, DATA); 3136 3137 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 3138 if (wbc->sync_mode == WB_SYNC_ALL) 3139 atomic_inc(&sbi->wb_sync_req[DATA]); 3140 else if (atomic_read(&sbi->wb_sync_req[DATA])) 3141 goto skip_write; 3142 3143 if (__should_serialize_io(inode, wbc)) { 3144 mutex_lock(&sbi->writepages); 3145 locked = true; 3146 } 3147 3148 blk_start_plug(&plug); 3149 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 3150 blk_finish_plug(&plug); 3151 3152 if (locked) 3153 mutex_unlock(&sbi->writepages); 3154 3155 if (wbc->sync_mode == WB_SYNC_ALL) 3156 atomic_dec(&sbi->wb_sync_req[DATA]); 3157 /* 3158 * if some pages were truncated, we cannot guarantee its mapping->host 3159 * to detect pending bios. 3160 */ 3161 3162 f2fs_remove_dirty_inode(inode); 3163 return ret; 3164 3165 skip_write: 3166 wbc->pages_skipped += get_dirty_pages(inode); 3167 trace_f2fs_writepages(mapping->host, wbc, DATA); 3168 return 0; 3169 } 3170 3171 static int f2fs_write_data_pages(struct address_space *mapping, 3172 struct writeback_control *wbc) 3173 { 3174 struct inode *inode = mapping->host; 3175 3176 return __f2fs_write_data_pages(mapping, wbc, 3177 F2FS_I(inode)->cp_task == current ? 3178 FS_CP_DATA_IO : FS_DATA_IO); 3179 } 3180 3181 static void f2fs_write_failed(struct address_space *mapping, loff_t to) 3182 { 3183 struct inode *inode = mapping->host; 3184 loff_t i_size = i_size_read(inode); 3185 3186 if (IS_NOQUOTA(inode)) 3187 return; 3188 3189 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */ 3190 if (to > i_size && !f2fs_verity_in_progress(inode)) { 3191 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3192 down_write(&F2FS_I(inode)->i_mmap_sem); 3193 3194 truncate_pagecache(inode, i_size); 3195 f2fs_truncate_blocks(inode, i_size, true); 3196 3197 up_write(&F2FS_I(inode)->i_mmap_sem); 3198 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3199 } 3200 } 3201 3202 static int prepare_write_begin(struct f2fs_sb_info *sbi, 3203 struct page *page, loff_t pos, unsigned len, 3204 block_t *blk_addr, bool *node_changed) 3205 { 3206 struct inode *inode = page->mapping->host; 3207 pgoff_t index = page->index; 3208 struct dnode_of_data dn; 3209 struct page *ipage; 3210 bool locked = false; 3211 struct extent_info ei = {0,0,0}; 3212 int err = 0; 3213 int flag; 3214 3215 /* 3216 * we already allocated all the blocks, so we don't need to get 3217 * the block addresses when there is no need to fill the page. 3218 */ 3219 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE && 3220 !is_inode_flag_set(inode, FI_NO_PREALLOC) && 3221 !f2fs_verity_in_progress(inode)) 3222 return 0; 3223 3224 /* f2fs_lock_op avoids race between write CP and convert_inline_page */ 3225 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode)) 3226 flag = F2FS_GET_BLOCK_DEFAULT; 3227 else 3228 flag = F2FS_GET_BLOCK_PRE_AIO; 3229 3230 if (f2fs_has_inline_data(inode) || 3231 (pos & PAGE_MASK) >= i_size_read(inode)) { 3232 f2fs_do_map_lock(sbi, flag, true); 3233 locked = true; 3234 } 3235 3236 restart: 3237 /* check inline_data */ 3238 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3239 if (IS_ERR(ipage)) { 3240 err = PTR_ERR(ipage); 3241 goto unlock_out; 3242 } 3243 3244 set_new_dnode(&dn, inode, ipage, ipage, 0); 3245 3246 if (f2fs_has_inline_data(inode)) { 3247 if (pos + len <= MAX_INLINE_DATA(inode)) { 3248 f2fs_do_read_inline_data(page, ipage); 3249 set_inode_flag(inode, FI_DATA_EXIST); 3250 if (inode->i_nlink) 3251 set_inline_node(ipage); 3252 } else { 3253 err = f2fs_convert_inline_page(&dn, page); 3254 if (err) 3255 goto out; 3256 if (dn.data_blkaddr == NULL_ADDR) 3257 err = f2fs_get_block(&dn, index); 3258 } 3259 } else if (locked) { 3260 err = f2fs_get_block(&dn, index); 3261 } else { 3262 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 3263 dn.data_blkaddr = ei.blk + index - ei.fofs; 3264 } else { 3265 /* hole case */ 3266 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3267 if (err || dn.data_blkaddr == NULL_ADDR) { 3268 f2fs_put_dnode(&dn); 3269 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 3270 true); 3271 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO); 3272 locked = true; 3273 goto restart; 3274 } 3275 } 3276 } 3277 3278 /* convert_inline_page can make node_changed */ 3279 *blk_addr = dn.data_blkaddr; 3280 *node_changed = dn.node_changed; 3281 out: 3282 f2fs_put_dnode(&dn); 3283 unlock_out: 3284 if (locked) 3285 f2fs_do_map_lock(sbi, flag, false); 3286 return err; 3287 } 3288 3289 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 3290 loff_t pos, unsigned len, unsigned flags, 3291 struct page **pagep, void **fsdata) 3292 { 3293 struct inode *inode = mapping->host; 3294 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3295 struct page *page = NULL; 3296 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 3297 bool need_balance = false, drop_atomic = false; 3298 block_t blkaddr = NULL_ADDR; 3299 int err = 0; 3300 3301 trace_f2fs_write_begin(inode, pos, len, flags); 3302 3303 if (!f2fs_is_checkpoint_ready(sbi)) { 3304 err = -ENOSPC; 3305 goto fail; 3306 } 3307 3308 if ((f2fs_is_atomic_file(inode) && 3309 !f2fs_available_free_memory(sbi, INMEM_PAGES)) || 3310 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) { 3311 err = -ENOMEM; 3312 drop_atomic = true; 3313 goto fail; 3314 } 3315 3316 /* 3317 * We should check this at this moment to avoid deadlock on inode page 3318 * and #0 page. The locking rule for inline_data conversion should be: 3319 * lock_page(page #0) -> lock_page(inode_page) 3320 */ 3321 if (index != 0) { 3322 err = f2fs_convert_inline_inode(inode); 3323 if (err) 3324 goto fail; 3325 } 3326 3327 #ifdef CONFIG_F2FS_FS_COMPRESSION 3328 if (f2fs_compressed_file(inode)) { 3329 int ret; 3330 3331 *fsdata = NULL; 3332 3333 ret = f2fs_prepare_compress_overwrite(inode, pagep, 3334 index, fsdata); 3335 if (ret < 0) { 3336 err = ret; 3337 goto fail; 3338 } else if (ret) { 3339 return 0; 3340 } 3341 } 3342 #endif 3343 3344 repeat: 3345 /* 3346 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 3347 * wait_for_stable_page. Will wait that below with our IO control. 3348 */ 3349 page = f2fs_pagecache_get_page(mapping, index, 3350 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 3351 if (!page) { 3352 err = -ENOMEM; 3353 goto fail; 3354 } 3355 3356 /* TODO: cluster can be compressed due to race with .writepage */ 3357 3358 *pagep = page; 3359 3360 err = prepare_write_begin(sbi, page, pos, len, 3361 &blkaddr, &need_balance); 3362 if (err) 3363 goto fail; 3364 3365 if (need_balance && !IS_NOQUOTA(inode) && 3366 has_not_enough_free_secs(sbi, 0, 0)) { 3367 unlock_page(page); 3368 f2fs_balance_fs(sbi, true); 3369 lock_page(page); 3370 if (page->mapping != mapping) { 3371 /* The page got truncated from under us */ 3372 f2fs_put_page(page, 1); 3373 goto repeat; 3374 } 3375 } 3376 3377 f2fs_wait_on_page_writeback(page, DATA, false, true); 3378 3379 if (len == PAGE_SIZE || PageUptodate(page)) 3380 return 0; 3381 3382 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) && 3383 !f2fs_verity_in_progress(inode)) { 3384 zero_user_segment(page, len, PAGE_SIZE); 3385 return 0; 3386 } 3387 3388 if (blkaddr == NEW_ADDR) { 3389 zero_user_segment(page, 0, PAGE_SIZE); 3390 SetPageUptodate(page); 3391 } else { 3392 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 3393 DATA_GENERIC_ENHANCE_READ)) { 3394 err = -EFSCORRUPTED; 3395 goto fail; 3396 } 3397 err = f2fs_submit_page_read(inode, page, blkaddr, true); 3398 if (err) 3399 goto fail; 3400 3401 lock_page(page); 3402 if (unlikely(page->mapping != mapping)) { 3403 f2fs_put_page(page, 1); 3404 goto repeat; 3405 } 3406 if (unlikely(!PageUptodate(page))) { 3407 err = -EIO; 3408 goto fail; 3409 } 3410 } 3411 return 0; 3412 3413 fail: 3414 f2fs_put_page(page, 1); 3415 f2fs_write_failed(mapping, pos + len); 3416 if (drop_atomic) 3417 f2fs_drop_inmem_pages_all(sbi, false); 3418 return err; 3419 } 3420 3421 static int f2fs_write_end(struct file *file, 3422 struct address_space *mapping, 3423 loff_t pos, unsigned len, unsigned copied, 3424 struct page *page, void *fsdata) 3425 { 3426 struct inode *inode = page->mapping->host; 3427 3428 trace_f2fs_write_end(inode, pos, len, copied); 3429 3430 /* 3431 * This should be come from len == PAGE_SIZE, and we expect copied 3432 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 3433 * let generic_perform_write() try to copy data again through copied=0. 3434 */ 3435 if (!PageUptodate(page)) { 3436 if (unlikely(copied != len)) 3437 copied = 0; 3438 else 3439 SetPageUptodate(page); 3440 } 3441 3442 #ifdef CONFIG_F2FS_FS_COMPRESSION 3443 /* overwrite compressed file */ 3444 if (f2fs_compressed_file(inode) && fsdata) { 3445 f2fs_compress_write_end(inode, fsdata, page->index, copied); 3446 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3447 return copied; 3448 } 3449 #endif 3450 3451 if (!copied) 3452 goto unlock_out; 3453 3454 set_page_dirty(page); 3455 3456 if (pos + copied > i_size_read(inode) && 3457 !f2fs_verity_in_progress(inode)) 3458 f2fs_i_size_write(inode, pos + copied); 3459 unlock_out: 3460 f2fs_put_page(page, 1); 3461 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3462 return copied; 3463 } 3464 3465 static int check_direct_IO(struct inode *inode, struct iov_iter *iter, 3466 loff_t offset) 3467 { 3468 unsigned i_blkbits = READ_ONCE(inode->i_blkbits); 3469 unsigned blkbits = i_blkbits; 3470 unsigned blocksize_mask = (1 << blkbits) - 1; 3471 unsigned long align = offset | iov_iter_alignment(iter); 3472 struct block_device *bdev = inode->i_sb->s_bdev; 3473 3474 if (align & blocksize_mask) { 3475 if (bdev) 3476 blkbits = blksize_bits(bdev_logical_block_size(bdev)); 3477 blocksize_mask = (1 << blkbits) - 1; 3478 if (align & blocksize_mask) 3479 return -EINVAL; 3480 return 1; 3481 } 3482 return 0; 3483 } 3484 3485 static void f2fs_dio_end_io(struct bio *bio) 3486 { 3487 struct f2fs_private_dio *dio = bio->bi_private; 3488 3489 dec_page_count(F2FS_I_SB(dio->inode), 3490 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ); 3491 3492 bio->bi_private = dio->orig_private; 3493 bio->bi_end_io = dio->orig_end_io; 3494 3495 kvfree(dio); 3496 3497 bio_endio(bio); 3498 } 3499 3500 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode, 3501 loff_t file_offset) 3502 { 3503 struct f2fs_private_dio *dio; 3504 bool write = (bio_op(bio) == REQ_OP_WRITE); 3505 3506 dio = f2fs_kzalloc(F2FS_I_SB(inode), 3507 sizeof(struct f2fs_private_dio), GFP_NOFS); 3508 if (!dio) 3509 goto out; 3510 3511 dio->inode = inode; 3512 dio->orig_end_io = bio->bi_end_io; 3513 dio->orig_private = bio->bi_private; 3514 dio->write = write; 3515 3516 bio->bi_end_io = f2fs_dio_end_io; 3517 bio->bi_private = dio; 3518 3519 inc_page_count(F2FS_I_SB(inode), 3520 write ? F2FS_DIO_WRITE : F2FS_DIO_READ); 3521 3522 submit_bio(bio); 3523 return; 3524 out: 3525 bio->bi_status = BLK_STS_IOERR; 3526 bio_endio(bio); 3527 } 3528 3529 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 3530 { 3531 struct address_space *mapping = iocb->ki_filp->f_mapping; 3532 struct inode *inode = mapping->host; 3533 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3534 struct f2fs_inode_info *fi = F2FS_I(inode); 3535 size_t count = iov_iter_count(iter); 3536 loff_t offset = iocb->ki_pos; 3537 int rw = iov_iter_rw(iter); 3538 int err; 3539 enum rw_hint hint = iocb->ki_hint; 3540 int whint_mode = F2FS_OPTION(sbi).whint_mode; 3541 bool do_opu; 3542 3543 err = check_direct_IO(inode, iter, offset); 3544 if (err) 3545 return err < 0 ? err : 0; 3546 3547 if (f2fs_force_buffered_io(inode, iocb, iter)) 3548 return 0; 3549 3550 do_opu = allow_outplace_dio(inode, iocb, iter); 3551 3552 trace_f2fs_direct_IO_enter(inode, offset, count, rw); 3553 3554 if (rw == WRITE && whint_mode == WHINT_MODE_OFF) 3555 iocb->ki_hint = WRITE_LIFE_NOT_SET; 3556 3557 if (iocb->ki_flags & IOCB_NOWAIT) { 3558 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) { 3559 iocb->ki_hint = hint; 3560 err = -EAGAIN; 3561 goto out; 3562 } 3563 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) { 3564 up_read(&fi->i_gc_rwsem[rw]); 3565 iocb->ki_hint = hint; 3566 err = -EAGAIN; 3567 goto out; 3568 } 3569 } else { 3570 down_read(&fi->i_gc_rwsem[rw]); 3571 if (do_opu) 3572 down_read(&fi->i_gc_rwsem[READ]); 3573 } 3574 3575 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, 3576 iter, rw == WRITE ? get_data_block_dio_write : 3577 get_data_block_dio, NULL, f2fs_dio_submit_bio, 3578 rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES : 3579 DIO_SKIP_HOLES); 3580 3581 if (do_opu) 3582 up_read(&fi->i_gc_rwsem[READ]); 3583 3584 up_read(&fi->i_gc_rwsem[rw]); 3585 3586 if (rw == WRITE) { 3587 if (whint_mode == WHINT_MODE_OFF) 3588 iocb->ki_hint = hint; 3589 if (err > 0) { 3590 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO, 3591 err); 3592 if (!do_opu) 3593 set_inode_flag(inode, FI_UPDATE_WRITE); 3594 } else if (err < 0) { 3595 f2fs_write_failed(mapping, offset + count); 3596 } 3597 } else { 3598 if (err > 0) 3599 f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err); 3600 } 3601 3602 out: 3603 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); 3604 3605 return err; 3606 } 3607 3608 void f2fs_invalidate_page(struct page *page, unsigned int offset, 3609 unsigned int length) 3610 { 3611 struct inode *inode = page->mapping->host; 3612 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3613 3614 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 3615 (offset % PAGE_SIZE || length != PAGE_SIZE)) 3616 return; 3617 3618 if (PageDirty(page)) { 3619 if (inode->i_ino == F2FS_META_INO(sbi)) { 3620 dec_page_count(sbi, F2FS_DIRTY_META); 3621 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 3622 dec_page_count(sbi, F2FS_DIRTY_NODES); 3623 } else { 3624 inode_dec_dirty_pages(inode); 3625 f2fs_remove_dirty_inode(inode); 3626 } 3627 } 3628 3629 clear_cold_data(page); 3630 3631 if (IS_ATOMIC_WRITTEN_PAGE(page)) 3632 return f2fs_drop_inmem_page(inode, page); 3633 3634 f2fs_clear_page_private(page); 3635 } 3636 3637 int f2fs_release_page(struct page *page, gfp_t wait) 3638 { 3639 /* If this is dirty page, keep PagePrivate */ 3640 if (PageDirty(page)) 3641 return 0; 3642 3643 /* This is atomic written page, keep Private */ 3644 if (IS_ATOMIC_WRITTEN_PAGE(page)) 3645 return 0; 3646 3647 clear_cold_data(page); 3648 f2fs_clear_page_private(page); 3649 return 1; 3650 } 3651 3652 static int f2fs_set_data_page_dirty(struct page *page) 3653 { 3654 struct inode *inode = page_file_mapping(page)->host; 3655 3656 trace_f2fs_set_page_dirty(page, DATA); 3657 3658 if (!PageUptodate(page)) 3659 SetPageUptodate(page); 3660 if (PageSwapCache(page)) 3661 return __set_page_dirty_nobuffers(page); 3662 3663 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) { 3664 if (!IS_ATOMIC_WRITTEN_PAGE(page)) { 3665 f2fs_register_inmem_page(inode, page); 3666 return 1; 3667 } 3668 /* 3669 * Previously, this page has been registered, we just 3670 * return here. 3671 */ 3672 return 0; 3673 } 3674 3675 if (!PageDirty(page)) { 3676 __set_page_dirty_nobuffers(page); 3677 f2fs_update_dirty_page(inode, page); 3678 return 1; 3679 } 3680 return 0; 3681 } 3682 3683 3684 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block) 3685 { 3686 #ifdef CONFIG_F2FS_FS_COMPRESSION 3687 struct dnode_of_data dn; 3688 sector_t start_idx, blknr = 0; 3689 int ret; 3690 3691 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size); 3692 3693 set_new_dnode(&dn, inode, NULL, NULL, 0); 3694 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 3695 if (ret) 3696 return 0; 3697 3698 if (dn.data_blkaddr != COMPRESS_ADDR) { 3699 dn.ofs_in_node += block - start_idx; 3700 blknr = f2fs_data_blkaddr(&dn); 3701 if (!__is_valid_data_blkaddr(blknr)) 3702 blknr = 0; 3703 } 3704 3705 f2fs_put_dnode(&dn); 3706 3707 return blknr; 3708 #else 3709 return -EOPNOTSUPP; 3710 #endif 3711 } 3712 3713 3714 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 3715 { 3716 struct inode *inode = mapping->host; 3717 3718 if (f2fs_has_inline_data(inode)) 3719 return 0; 3720 3721 /* make sure allocating whole blocks */ 3722 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 3723 filemap_write_and_wait(mapping); 3724 3725 if (f2fs_compressed_file(inode)) 3726 return f2fs_bmap_compress(inode, block); 3727 3728 return generic_block_bmap(mapping, block, get_data_block_bmap); 3729 } 3730 3731 #ifdef CONFIG_MIGRATION 3732 #include <linux/migrate.h> 3733 3734 int f2fs_migrate_page(struct address_space *mapping, 3735 struct page *newpage, struct page *page, enum migrate_mode mode) 3736 { 3737 int rc, extra_count; 3738 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 3739 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page); 3740 3741 BUG_ON(PageWriteback(page)); 3742 3743 /* migrating an atomic written page is safe with the inmem_lock hold */ 3744 if (atomic_written) { 3745 if (mode != MIGRATE_SYNC) 3746 return -EBUSY; 3747 if (!mutex_trylock(&fi->inmem_lock)) 3748 return -EAGAIN; 3749 } 3750 3751 /* one extra reference was held for atomic_write page */ 3752 extra_count = atomic_written ? 1 : 0; 3753 rc = migrate_page_move_mapping(mapping, newpage, 3754 page, extra_count); 3755 if (rc != MIGRATEPAGE_SUCCESS) { 3756 if (atomic_written) 3757 mutex_unlock(&fi->inmem_lock); 3758 return rc; 3759 } 3760 3761 if (atomic_written) { 3762 struct inmem_pages *cur; 3763 list_for_each_entry(cur, &fi->inmem_pages, list) 3764 if (cur->page == page) { 3765 cur->page = newpage; 3766 break; 3767 } 3768 mutex_unlock(&fi->inmem_lock); 3769 put_page(page); 3770 get_page(newpage); 3771 } 3772 3773 if (PagePrivate(page)) { 3774 f2fs_set_page_private(newpage, page_private(page)); 3775 f2fs_clear_page_private(page); 3776 } 3777 3778 if (mode != MIGRATE_SYNC_NO_COPY) 3779 migrate_page_copy(newpage, page); 3780 else 3781 migrate_page_states(newpage, page); 3782 3783 return MIGRATEPAGE_SUCCESS; 3784 } 3785 #endif 3786 3787 #ifdef CONFIG_SWAP 3788 /* Copied from generic_swapfile_activate() to check any holes */ 3789 static int check_swap_activate(struct swap_info_struct *sis, 3790 struct file *swap_file, sector_t *span) 3791 { 3792 struct address_space *mapping = swap_file->f_mapping; 3793 struct inode *inode = mapping->host; 3794 unsigned blocks_per_page; 3795 unsigned long page_no; 3796 unsigned blkbits; 3797 sector_t probe_block; 3798 sector_t last_block; 3799 sector_t lowest_block = -1; 3800 sector_t highest_block = 0; 3801 int nr_extents = 0; 3802 int ret; 3803 3804 blkbits = inode->i_blkbits; 3805 blocks_per_page = PAGE_SIZE >> blkbits; 3806 3807 /* 3808 * Map all the blocks into the extent list. This code doesn't try 3809 * to be very smart. 3810 */ 3811 probe_block = 0; 3812 page_no = 0; 3813 last_block = i_size_read(inode) >> blkbits; 3814 while ((probe_block + blocks_per_page) <= last_block && 3815 page_no < sis->max) { 3816 unsigned block_in_page; 3817 sector_t first_block; 3818 sector_t block = 0; 3819 int err = 0; 3820 3821 cond_resched(); 3822 3823 block = probe_block; 3824 err = bmap(inode, &block); 3825 if (err || !block) 3826 goto bad_bmap; 3827 first_block = block; 3828 3829 /* 3830 * It must be PAGE_SIZE aligned on-disk 3831 */ 3832 if (first_block & (blocks_per_page - 1)) { 3833 probe_block++; 3834 goto reprobe; 3835 } 3836 3837 for (block_in_page = 1; block_in_page < blocks_per_page; 3838 block_in_page++) { 3839 3840 block = probe_block + block_in_page; 3841 err = bmap(inode, &block); 3842 3843 if (err || !block) 3844 goto bad_bmap; 3845 3846 if (block != first_block + block_in_page) { 3847 /* Discontiguity */ 3848 probe_block++; 3849 goto reprobe; 3850 } 3851 } 3852 3853 first_block >>= (PAGE_SHIFT - blkbits); 3854 if (page_no) { /* exclude the header page */ 3855 if (first_block < lowest_block) 3856 lowest_block = first_block; 3857 if (first_block > highest_block) 3858 highest_block = first_block; 3859 } 3860 3861 /* 3862 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks 3863 */ 3864 ret = add_swap_extent(sis, page_no, 1, first_block); 3865 if (ret < 0) 3866 goto out; 3867 nr_extents += ret; 3868 page_no++; 3869 probe_block += blocks_per_page; 3870 reprobe: 3871 continue; 3872 } 3873 ret = nr_extents; 3874 *span = 1 + highest_block - lowest_block; 3875 if (page_no == 0) 3876 page_no = 1; /* force Empty message */ 3877 sis->max = page_no; 3878 sis->pages = page_no - 1; 3879 sis->highest_bit = page_no - 1; 3880 out: 3881 return ret; 3882 bad_bmap: 3883 pr_err("swapon: swapfile has holes\n"); 3884 return -EINVAL; 3885 } 3886 3887 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3888 sector_t *span) 3889 { 3890 struct inode *inode = file_inode(file); 3891 int ret; 3892 3893 if (!S_ISREG(inode->i_mode)) 3894 return -EINVAL; 3895 3896 if (f2fs_readonly(F2FS_I_SB(inode)->sb)) 3897 return -EROFS; 3898 3899 ret = f2fs_convert_inline_inode(inode); 3900 if (ret) 3901 return ret; 3902 3903 if (f2fs_disable_compressed_file(inode)) 3904 return -EINVAL; 3905 3906 ret = check_swap_activate(sis, file, span); 3907 if (ret < 0) 3908 return ret; 3909 3910 set_inode_flag(inode, FI_PIN_FILE); 3911 f2fs_precache_extents(inode); 3912 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3913 return ret; 3914 } 3915 3916 static void f2fs_swap_deactivate(struct file *file) 3917 { 3918 struct inode *inode = file_inode(file); 3919 3920 clear_inode_flag(inode, FI_PIN_FILE); 3921 } 3922 #else 3923 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3924 sector_t *span) 3925 { 3926 return -EOPNOTSUPP; 3927 } 3928 3929 static void f2fs_swap_deactivate(struct file *file) 3930 { 3931 } 3932 #endif 3933 3934 const struct address_space_operations f2fs_dblock_aops = { 3935 .readpage = f2fs_read_data_page, 3936 .readahead = f2fs_readahead, 3937 .writepage = f2fs_write_data_page, 3938 .writepages = f2fs_write_data_pages, 3939 .write_begin = f2fs_write_begin, 3940 .write_end = f2fs_write_end, 3941 .set_page_dirty = f2fs_set_data_page_dirty, 3942 .invalidatepage = f2fs_invalidate_page, 3943 .releasepage = f2fs_release_page, 3944 .direct_IO = f2fs_direct_IO, 3945 .bmap = f2fs_bmap, 3946 .swap_activate = f2fs_swap_activate, 3947 .swap_deactivate = f2fs_swap_deactivate, 3948 #ifdef CONFIG_MIGRATION 3949 .migratepage = f2fs_migrate_page, 3950 #endif 3951 }; 3952 3953 void f2fs_clear_page_cache_dirty_tag(struct page *page) 3954 { 3955 struct address_space *mapping = page_mapping(page); 3956 unsigned long flags; 3957 3958 xa_lock_irqsave(&mapping->i_pages, flags); 3959 __xa_clear_mark(&mapping->i_pages, page_index(page), 3960 PAGECACHE_TAG_DIRTY); 3961 xa_unlock_irqrestore(&mapping->i_pages, flags); 3962 } 3963 3964 int __init f2fs_init_post_read_processing(void) 3965 { 3966 bio_post_read_ctx_cache = 3967 kmem_cache_create("f2fs_bio_post_read_ctx", 3968 sizeof(struct bio_post_read_ctx), 0, 0, NULL); 3969 if (!bio_post_read_ctx_cache) 3970 goto fail; 3971 bio_post_read_ctx_pool = 3972 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 3973 bio_post_read_ctx_cache); 3974 if (!bio_post_read_ctx_pool) 3975 goto fail_free_cache; 3976 return 0; 3977 3978 fail_free_cache: 3979 kmem_cache_destroy(bio_post_read_ctx_cache); 3980 fail: 3981 return -ENOMEM; 3982 } 3983 3984 void f2fs_destroy_post_read_processing(void) 3985 { 3986 mempool_destroy(bio_post_read_ctx_pool); 3987 kmem_cache_destroy(bio_post_read_ctx_cache); 3988 } 3989 3990 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) 3991 { 3992 if (!f2fs_sb_has_encrypt(sbi) && 3993 !f2fs_sb_has_verity(sbi) && 3994 !f2fs_sb_has_compression(sbi)) 3995 return 0; 3996 3997 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq", 3998 WQ_UNBOUND | WQ_HIGHPRI, 3999 num_online_cpus()); 4000 if (!sbi->post_read_wq) 4001 return -ENOMEM; 4002 return 0; 4003 } 4004 4005 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) 4006 { 4007 if (sbi->post_read_wq) 4008 destroy_workqueue(sbi->post_read_wq); 4009 } 4010 4011 int __init f2fs_init_bio_entry_cache(void) 4012 { 4013 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab", 4014 sizeof(struct bio_entry)); 4015 if (!bio_entry_slab) 4016 return -ENOMEM; 4017 return 0; 4018 } 4019 4020 void f2fs_destroy_bio_entry_cache(void) 4021 { 4022 kmem_cache_destroy(bio_entry_slab); 4023 } 4024