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/sched/mm.h> 12 #include <linux/mpage.h> 13 #include <linux/writeback.h> 14 #include <linux/pagevec.h> 15 #include <linux/blkdev.h> 16 #include <linux/bio.h> 17 #include <linux/blk-crypto.h> 18 #include <linux/swap.h> 19 #include <linux/prefetch.h> 20 #include <linux/uio.h> 21 #include <linux/sched/signal.h> 22 #include <linux/fiemap.h> 23 #include <linux/iomap.h> 24 25 #include "f2fs.h" 26 #include "node.h" 27 #include "segment.h" 28 #include "iostat.h" 29 #include <trace/events/f2fs.h> 30 31 #define NUM_PREALLOC_POST_READ_CTXS 128 32 33 static struct kmem_cache *bio_post_read_ctx_cache; 34 static struct kmem_cache *bio_entry_slab; 35 static mempool_t *bio_post_read_ctx_pool; 36 static struct bio_set f2fs_bioset; 37 38 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE 39 40 int __init f2fs_init_bioset(void) 41 { 42 if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE, 43 0, BIOSET_NEED_BVECS)) 44 return -ENOMEM; 45 return 0; 46 } 47 48 void f2fs_destroy_bioset(void) 49 { 50 bioset_exit(&f2fs_bioset); 51 } 52 53 static bool __is_cp_guaranteed(struct page *page) 54 { 55 struct address_space *mapping = page->mapping; 56 struct inode *inode; 57 struct f2fs_sb_info *sbi; 58 59 if (!mapping) 60 return false; 61 62 inode = mapping->host; 63 sbi = F2FS_I_SB(inode); 64 65 if (inode->i_ino == F2FS_META_INO(sbi) || 66 inode->i_ino == F2FS_NODE_INO(sbi) || 67 S_ISDIR(inode->i_mode)) 68 return true; 69 70 if (f2fs_is_compressed_page(page)) 71 return false; 72 if ((S_ISREG(inode->i_mode) && 73 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) || 74 page_private_gcing(page)) 75 return true; 76 return false; 77 } 78 79 static enum count_type __read_io_type(struct page *page) 80 { 81 struct address_space *mapping = page_file_mapping(page); 82 83 if (mapping) { 84 struct inode *inode = mapping->host; 85 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 86 87 if (inode->i_ino == F2FS_META_INO(sbi)) 88 return F2FS_RD_META; 89 90 if (inode->i_ino == F2FS_NODE_INO(sbi)) 91 return F2FS_RD_NODE; 92 } 93 return F2FS_RD_DATA; 94 } 95 96 /* postprocessing steps for read bios */ 97 enum bio_post_read_step { 98 #ifdef CONFIG_FS_ENCRYPTION 99 STEP_DECRYPT = 1 << 0, 100 #else 101 STEP_DECRYPT = 0, /* compile out the decryption-related code */ 102 #endif 103 #ifdef CONFIG_F2FS_FS_COMPRESSION 104 STEP_DECOMPRESS = 1 << 1, 105 #else 106 STEP_DECOMPRESS = 0, /* compile out the decompression-related code */ 107 #endif 108 #ifdef CONFIG_FS_VERITY 109 STEP_VERITY = 1 << 2, 110 #else 111 STEP_VERITY = 0, /* compile out the verity-related code */ 112 #endif 113 }; 114 115 struct bio_post_read_ctx { 116 struct bio *bio; 117 struct f2fs_sb_info *sbi; 118 struct work_struct work; 119 unsigned int enabled_steps; 120 block_t fs_blkaddr; 121 }; 122 123 static void f2fs_finish_read_bio(struct bio *bio) 124 { 125 struct bio_vec *bv; 126 struct bvec_iter_all iter_all; 127 128 /* 129 * Update and unlock the bio's pagecache pages, and put the 130 * decompression context for any compressed pages. 131 */ 132 bio_for_each_segment_all(bv, bio, iter_all) { 133 struct page *page = bv->bv_page; 134 135 if (f2fs_is_compressed_page(page)) { 136 if (bio->bi_status) 137 f2fs_end_read_compressed_page(page, true, 0); 138 f2fs_put_page_dic(page); 139 continue; 140 } 141 142 /* PG_error was set if decryption or verity failed. */ 143 if (bio->bi_status || PageError(page)) { 144 ClearPageUptodate(page); 145 /* will re-read again later */ 146 ClearPageError(page); 147 } else { 148 SetPageUptodate(page); 149 } 150 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 151 unlock_page(page); 152 } 153 154 if (bio->bi_private) 155 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 156 bio_put(bio); 157 } 158 159 static void f2fs_verify_bio(struct work_struct *work) 160 { 161 struct bio_post_read_ctx *ctx = 162 container_of(work, struct bio_post_read_ctx, work); 163 struct bio *bio = ctx->bio; 164 bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS); 165 166 /* 167 * fsverity_verify_bio() may call readahead() again, and while verity 168 * will be disabled for this, decryption and/or decompression may still 169 * be needed, resulting in another bio_post_read_ctx being allocated. 170 * So to prevent deadlocks we need to release the current ctx to the 171 * mempool first. This assumes that verity is the last post-read step. 172 */ 173 mempool_free(ctx, bio_post_read_ctx_pool); 174 bio->bi_private = NULL; 175 176 /* 177 * Verify the bio's pages with fs-verity. Exclude compressed pages, 178 * as those were handled separately by f2fs_end_read_compressed_page(). 179 */ 180 if (may_have_compressed_pages) { 181 struct bio_vec *bv; 182 struct bvec_iter_all iter_all; 183 184 bio_for_each_segment_all(bv, bio, iter_all) { 185 struct page *page = bv->bv_page; 186 187 if (!f2fs_is_compressed_page(page) && 188 !PageError(page) && !fsverity_verify_page(page)) 189 SetPageError(page); 190 } 191 } else { 192 fsverity_verify_bio(bio); 193 } 194 195 f2fs_finish_read_bio(bio); 196 } 197 198 /* 199 * If the bio's data needs to be verified with fs-verity, then enqueue the 200 * verity work for the bio. Otherwise finish the bio now. 201 * 202 * Note that to avoid deadlocks, the verity work can't be done on the 203 * decryption/decompression workqueue. This is because verifying the data pages 204 * can involve reading verity metadata pages from the file, and these verity 205 * metadata pages may be encrypted and/or compressed. 206 */ 207 static void f2fs_verify_and_finish_bio(struct bio *bio) 208 { 209 struct bio_post_read_ctx *ctx = bio->bi_private; 210 211 if (ctx && (ctx->enabled_steps & STEP_VERITY)) { 212 INIT_WORK(&ctx->work, f2fs_verify_bio); 213 fsverity_enqueue_verify_work(&ctx->work); 214 } else { 215 f2fs_finish_read_bio(bio); 216 } 217 } 218 219 /* 220 * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last 221 * remaining page was read by @ctx->bio. 222 * 223 * Note that a bio may span clusters (even a mix of compressed and uncompressed 224 * clusters) or be for just part of a cluster. STEP_DECOMPRESS just indicates 225 * that the bio includes at least one compressed page. The actual decompression 226 * is done on a per-cluster basis, not a per-bio basis. 227 */ 228 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx) 229 { 230 struct bio_vec *bv; 231 struct bvec_iter_all iter_all; 232 bool all_compressed = true; 233 block_t blkaddr = ctx->fs_blkaddr; 234 235 bio_for_each_segment_all(bv, ctx->bio, iter_all) { 236 struct page *page = bv->bv_page; 237 238 /* PG_error was set if decryption failed. */ 239 if (f2fs_is_compressed_page(page)) 240 f2fs_end_read_compressed_page(page, PageError(page), 241 blkaddr); 242 else 243 all_compressed = false; 244 245 blkaddr++; 246 } 247 248 /* 249 * Optimization: if all the bio's pages are compressed, then scheduling 250 * the per-bio verity work is unnecessary, as verity will be fully 251 * handled at the compression cluster level. 252 */ 253 if (all_compressed) 254 ctx->enabled_steps &= ~STEP_VERITY; 255 } 256 257 static void f2fs_post_read_work(struct work_struct *work) 258 { 259 struct bio_post_read_ctx *ctx = 260 container_of(work, struct bio_post_read_ctx, work); 261 262 if (ctx->enabled_steps & STEP_DECRYPT) 263 fscrypt_decrypt_bio(ctx->bio); 264 265 if (ctx->enabled_steps & STEP_DECOMPRESS) 266 f2fs_handle_step_decompress(ctx); 267 268 f2fs_verify_and_finish_bio(ctx->bio); 269 } 270 271 static void f2fs_read_end_io(struct bio *bio) 272 { 273 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio)); 274 struct bio_post_read_ctx *ctx; 275 276 iostat_update_and_unbind_ctx(bio, 0); 277 ctx = bio->bi_private; 278 279 if (time_to_inject(sbi, FAULT_READ_IO)) { 280 f2fs_show_injection_info(sbi, FAULT_READ_IO); 281 bio->bi_status = BLK_STS_IOERR; 282 } 283 284 if (bio->bi_status) { 285 f2fs_finish_read_bio(bio); 286 return; 287 } 288 289 if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) { 290 INIT_WORK(&ctx->work, f2fs_post_read_work); 291 queue_work(ctx->sbi->post_read_wq, &ctx->work); 292 } else { 293 f2fs_verify_and_finish_bio(bio); 294 } 295 } 296 297 static void f2fs_write_end_io(struct bio *bio) 298 { 299 struct f2fs_sb_info *sbi; 300 struct bio_vec *bvec; 301 struct bvec_iter_all iter_all; 302 303 iostat_update_and_unbind_ctx(bio, 1); 304 sbi = bio->bi_private; 305 306 if (time_to_inject(sbi, FAULT_WRITE_IO)) { 307 f2fs_show_injection_info(sbi, FAULT_WRITE_IO); 308 bio->bi_status = BLK_STS_IOERR; 309 } 310 311 bio_for_each_segment_all(bvec, bio, iter_all) { 312 struct page *page = bvec->bv_page; 313 enum count_type type = WB_DATA_TYPE(page); 314 315 if (page_private_dummy(page)) { 316 clear_page_private_dummy(page); 317 unlock_page(page); 318 mempool_free(page, sbi->write_io_dummy); 319 320 if (unlikely(bio->bi_status)) 321 f2fs_stop_checkpoint(sbi, true); 322 continue; 323 } 324 325 fscrypt_finalize_bounce_page(&page); 326 327 #ifdef CONFIG_F2FS_FS_COMPRESSION 328 if (f2fs_is_compressed_page(page)) { 329 f2fs_compress_write_end_io(bio, page); 330 continue; 331 } 332 #endif 333 334 if (unlikely(bio->bi_status)) { 335 mapping_set_error(page->mapping, -EIO); 336 if (type == F2FS_WB_CP_DATA) 337 f2fs_stop_checkpoint(sbi, true); 338 } 339 340 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 341 page->index != nid_of_node(page)); 342 343 dec_page_count(sbi, type); 344 if (f2fs_in_warm_node_list(sbi, page)) 345 f2fs_del_fsync_node_entry(sbi, page); 346 clear_page_private_gcing(page); 347 end_page_writeback(page); 348 } 349 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 350 wq_has_sleeper(&sbi->cp_wait)) 351 wake_up(&sbi->cp_wait); 352 353 bio_put(bio); 354 } 355 356 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 357 block_t blk_addr, sector_t *sector) 358 { 359 struct block_device *bdev = sbi->sb->s_bdev; 360 int i; 361 362 if (f2fs_is_multi_device(sbi)) { 363 for (i = 0; i < sbi->s_ndevs; i++) { 364 if (FDEV(i).start_blk <= blk_addr && 365 FDEV(i).end_blk >= blk_addr) { 366 blk_addr -= FDEV(i).start_blk; 367 bdev = FDEV(i).bdev; 368 break; 369 } 370 } 371 } 372 373 if (sector) 374 *sector = SECTOR_FROM_BLOCK(blk_addr); 375 return bdev; 376 } 377 378 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 379 { 380 int i; 381 382 if (!f2fs_is_multi_device(sbi)) 383 return 0; 384 385 for (i = 0; i < sbi->s_ndevs; i++) 386 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 387 return i; 388 return 0; 389 } 390 391 static unsigned int f2fs_io_flags(struct f2fs_io_info *fio) 392 { 393 unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1; 394 unsigned int fua_flag, meta_flag, io_flag; 395 unsigned int op_flags = 0; 396 397 if (fio->op != REQ_OP_WRITE) 398 return 0; 399 if (fio->type == DATA) 400 io_flag = fio->sbi->data_io_flag; 401 else if (fio->type == NODE) 402 io_flag = fio->sbi->node_io_flag; 403 else 404 return 0; 405 406 fua_flag = io_flag & temp_mask; 407 meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask; 408 409 /* 410 * data/node io flag bits per temp: 411 * REQ_META | REQ_FUA | 412 * 5 | 4 | 3 | 2 | 1 | 0 | 413 * Cold | Warm | Hot | Cold | Warm | Hot | 414 */ 415 if ((1 << fio->temp) & meta_flag) 416 op_flags |= REQ_META; 417 if ((1 << fio->temp) & fua_flag) 418 op_flags |= REQ_FUA; 419 return op_flags; 420 } 421 422 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) 423 { 424 struct f2fs_sb_info *sbi = fio->sbi; 425 struct block_device *bdev; 426 sector_t sector; 427 struct bio *bio; 428 429 bdev = f2fs_target_device(sbi, fio->new_blkaddr, §or); 430 bio = bio_alloc_bioset(bdev, npages, 431 fio->op | fio->op_flags | f2fs_io_flags(fio), 432 GFP_NOIO, &f2fs_bioset); 433 bio->bi_iter.bi_sector = sector; 434 if (is_read_io(fio->op)) { 435 bio->bi_end_io = f2fs_read_end_io; 436 bio->bi_private = NULL; 437 } else { 438 bio->bi_end_io = f2fs_write_end_io; 439 bio->bi_private = sbi; 440 } 441 iostat_alloc_and_bind_ctx(sbi, bio, NULL); 442 443 if (fio->io_wbc) 444 wbc_init_bio(fio->io_wbc, bio); 445 446 return bio; 447 } 448 449 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode, 450 pgoff_t first_idx, 451 const struct f2fs_io_info *fio, 452 gfp_t gfp_mask) 453 { 454 /* 455 * The f2fs garbage collector sets ->encrypted_page when it wants to 456 * read/write raw data without encryption. 457 */ 458 if (!fio || !fio->encrypted_page) 459 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask); 460 } 461 462 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode, 463 pgoff_t next_idx, 464 const struct f2fs_io_info *fio) 465 { 466 /* 467 * The f2fs garbage collector sets ->encrypted_page when it wants to 468 * read/write raw data without encryption. 469 */ 470 if (fio && fio->encrypted_page) 471 return !bio_has_crypt_ctx(bio); 472 473 return fscrypt_mergeable_bio(bio, inode, next_idx); 474 } 475 476 static inline void __submit_bio(struct f2fs_sb_info *sbi, 477 struct bio *bio, enum page_type type) 478 { 479 if (!is_read_io(bio_op(bio))) { 480 unsigned int start; 481 482 if (type != DATA && type != NODE) 483 goto submit_io; 484 485 if (f2fs_lfs_mode(sbi) && current->plug) 486 blk_finish_plug(current->plug); 487 488 if (!F2FS_IO_ALIGNED(sbi)) 489 goto submit_io; 490 491 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS; 492 start %= F2FS_IO_SIZE(sbi); 493 494 if (start == 0) 495 goto submit_io; 496 497 /* fill dummy pages */ 498 for (; start < F2FS_IO_SIZE(sbi); start++) { 499 struct page *page = 500 mempool_alloc(sbi->write_io_dummy, 501 GFP_NOIO | __GFP_NOFAIL); 502 f2fs_bug_on(sbi, !page); 503 504 lock_page(page); 505 506 zero_user_segment(page, 0, PAGE_SIZE); 507 set_page_private_dummy(page); 508 509 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 510 f2fs_bug_on(sbi, 1); 511 } 512 /* 513 * In the NODE case, we lose next block address chain. So, we 514 * need to do checkpoint in f2fs_sync_file. 515 */ 516 if (type == NODE) 517 set_sbi_flag(sbi, SBI_NEED_CP); 518 } 519 submit_io: 520 if (is_read_io(bio_op(bio))) 521 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 522 else 523 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 524 525 iostat_update_submit_ctx(bio, type); 526 submit_bio(bio); 527 } 528 529 void f2fs_submit_bio(struct f2fs_sb_info *sbi, 530 struct bio *bio, enum page_type type) 531 { 532 __submit_bio(sbi, bio, type); 533 } 534 535 static void __submit_merged_bio(struct f2fs_bio_info *io) 536 { 537 struct f2fs_io_info *fio = &io->fio; 538 539 if (!io->bio) 540 return; 541 542 if (is_read_io(fio->op)) 543 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 544 else 545 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 546 547 __submit_bio(io->sbi, io->bio, fio->type); 548 io->bio = NULL; 549 } 550 551 static bool __has_merged_page(struct bio *bio, struct inode *inode, 552 struct page *page, nid_t ino) 553 { 554 struct bio_vec *bvec; 555 struct bvec_iter_all iter_all; 556 557 if (!bio) 558 return false; 559 560 if (!inode && !page && !ino) 561 return true; 562 563 bio_for_each_segment_all(bvec, bio, iter_all) { 564 struct page *target = bvec->bv_page; 565 566 if (fscrypt_is_bounce_page(target)) { 567 target = fscrypt_pagecache_page(target); 568 if (IS_ERR(target)) 569 continue; 570 } 571 if (f2fs_is_compressed_page(target)) { 572 target = f2fs_compress_control_page(target); 573 if (IS_ERR(target)) 574 continue; 575 } 576 577 if (inode && inode == target->mapping->host) 578 return true; 579 if (page && page == target) 580 return true; 581 if (ino && ino == ino_of_node(target)) 582 return true; 583 } 584 585 return false; 586 } 587 588 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 589 enum page_type type, enum temp_type temp) 590 { 591 enum page_type btype = PAGE_TYPE_OF_BIO(type); 592 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 593 594 f2fs_down_write(&io->io_rwsem); 595 596 /* change META to META_FLUSH in the checkpoint procedure */ 597 if (type >= META_FLUSH) { 598 io->fio.type = META_FLUSH; 599 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC; 600 if (!test_opt(sbi, NOBARRIER)) 601 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA; 602 } 603 __submit_merged_bio(io); 604 f2fs_up_write(&io->io_rwsem); 605 } 606 607 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 608 struct inode *inode, struct page *page, 609 nid_t ino, enum page_type type, bool force) 610 { 611 enum temp_type temp; 612 bool ret = true; 613 614 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 615 if (!force) { 616 enum page_type btype = PAGE_TYPE_OF_BIO(type); 617 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 618 619 f2fs_down_read(&io->io_rwsem); 620 ret = __has_merged_page(io->bio, inode, page, ino); 621 f2fs_up_read(&io->io_rwsem); 622 } 623 if (ret) 624 __f2fs_submit_merged_write(sbi, type, temp); 625 626 /* TODO: use HOT temp only for meta pages now. */ 627 if (type >= META) 628 break; 629 } 630 } 631 632 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 633 { 634 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true); 635 } 636 637 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 638 struct inode *inode, struct page *page, 639 nid_t ino, enum page_type type) 640 { 641 __submit_merged_write_cond(sbi, inode, page, ino, type, false); 642 } 643 644 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 645 { 646 f2fs_submit_merged_write(sbi, DATA); 647 f2fs_submit_merged_write(sbi, NODE); 648 f2fs_submit_merged_write(sbi, META); 649 } 650 651 /* 652 * Fill the locked page with data located in the block address. 653 * A caller needs to unlock the page on failure. 654 */ 655 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 656 { 657 struct bio *bio; 658 struct page *page = fio->encrypted_page ? 659 fio->encrypted_page : fio->page; 660 661 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 662 fio->is_por ? META_POR : (__is_meta_io(fio) ? 663 META_GENERIC : DATA_GENERIC_ENHANCE))) 664 return -EFSCORRUPTED; 665 666 trace_f2fs_submit_page_bio(page, fio); 667 668 /* Allocate a new bio */ 669 bio = __bio_alloc(fio, 1); 670 671 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host, 672 fio->page->index, fio, GFP_NOIO); 673 674 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 675 bio_put(bio); 676 return -EFAULT; 677 } 678 679 if (fio->io_wbc && !is_read_io(fio->op)) 680 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 681 682 inc_page_count(fio->sbi, is_read_io(fio->op) ? 683 __read_io_type(page): WB_DATA_TYPE(fio->page)); 684 685 __submit_bio(fio->sbi, bio, fio->type); 686 return 0; 687 } 688 689 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 690 block_t last_blkaddr, block_t cur_blkaddr) 691 { 692 if (unlikely(sbi->max_io_bytes && 693 bio->bi_iter.bi_size >= sbi->max_io_bytes)) 694 return false; 695 if (last_blkaddr + 1 != cur_blkaddr) 696 return false; 697 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL); 698 } 699 700 static bool io_type_is_mergeable(struct f2fs_bio_info *io, 701 struct f2fs_io_info *fio) 702 { 703 if (io->fio.op != fio->op) 704 return false; 705 return io->fio.op_flags == fio->op_flags; 706 } 707 708 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 709 struct f2fs_bio_info *io, 710 struct f2fs_io_info *fio, 711 block_t last_blkaddr, 712 block_t cur_blkaddr) 713 { 714 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) { 715 unsigned int filled_blocks = 716 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size); 717 unsigned int io_size = F2FS_IO_SIZE(sbi); 718 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt; 719 720 /* IOs in bio is aligned and left space of vectors is not enough */ 721 if (!(filled_blocks % io_size) && left_vecs < io_size) 722 return false; 723 } 724 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr)) 725 return false; 726 return io_type_is_mergeable(io, fio); 727 } 728 729 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio, 730 struct page *page, enum temp_type temp) 731 { 732 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 733 struct bio_entry *be; 734 735 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL); 736 be->bio = bio; 737 bio_get(bio); 738 739 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE) 740 f2fs_bug_on(sbi, 1); 741 742 f2fs_down_write(&io->bio_list_lock); 743 list_add_tail(&be->list, &io->bio_list); 744 f2fs_up_write(&io->bio_list_lock); 745 } 746 747 static void del_bio_entry(struct bio_entry *be) 748 { 749 list_del(&be->list); 750 kmem_cache_free(bio_entry_slab, be); 751 } 752 753 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio, 754 struct page *page) 755 { 756 struct f2fs_sb_info *sbi = fio->sbi; 757 enum temp_type temp; 758 bool found = false; 759 int ret = -EAGAIN; 760 761 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 762 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 763 struct list_head *head = &io->bio_list; 764 struct bio_entry *be; 765 766 f2fs_down_write(&io->bio_list_lock); 767 list_for_each_entry(be, head, list) { 768 if (be->bio != *bio) 769 continue; 770 771 found = true; 772 773 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio, 774 *fio->last_block, 775 fio->new_blkaddr)); 776 if (f2fs_crypt_mergeable_bio(*bio, 777 fio->page->mapping->host, 778 fio->page->index, fio) && 779 bio_add_page(*bio, page, PAGE_SIZE, 0) == 780 PAGE_SIZE) { 781 ret = 0; 782 break; 783 } 784 785 /* page can't be merged into bio; submit the bio */ 786 del_bio_entry(be); 787 __submit_bio(sbi, *bio, DATA); 788 break; 789 } 790 f2fs_up_write(&io->bio_list_lock); 791 } 792 793 if (ret) { 794 bio_put(*bio); 795 *bio = NULL; 796 } 797 798 return ret; 799 } 800 801 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi, 802 struct bio **bio, struct page *page) 803 { 804 enum temp_type temp; 805 bool found = false; 806 struct bio *target = bio ? *bio : NULL; 807 808 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 809 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 810 struct list_head *head = &io->bio_list; 811 struct bio_entry *be; 812 813 if (list_empty(head)) 814 continue; 815 816 f2fs_down_read(&io->bio_list_lock); 817 list_for_each_entry(be, head, list) { 818 if (target) 819 found = (target == be->bio); 820 else 821 found = __has_merged_page(be->bio, NULL, 822 page, 0); 823 if (found) 824 break; 825 } 826 f2fs_up_read(&io->bio_list_lock); 827 828 if (!found) 829 continue; 830 831 found = false; 832 833 f2fs_down_write(&io->bio_list_lock); 834 list_for_each_entry(be, head, list) { 835 if (target) 836 found = (target == be->bio); 837 else 838 found = __has_merged_page(be->bio, NULL, 839 page, 0); 840 if (found) { 841 target = be->bio; 842 del_bio_entry(be); 843 break; 844 } 845 } 846 f2fs_up_write(&io->bio_list_lock); 847 } 848 849 if (found) 850 __submit_bio(sbi, target, DATA); 851 if (bio && *bio) { 852 bio_put(*bio); 853 *bio = NULL; 854 } 855 } 856 857 int f2fs_merge_page_bio(struct f2fs_io_info *fio) 858 { 859 struct bio *bio = *fio->bio; 860 struct page *page = fio->encrypted_page ? 861 fio->encrypted_page : fio->page; 862 863 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 864 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) 865 return -EFSCORRUPTED; 866 867 trace_f2fs_submit_page_bio(page, fio); 868 869 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block, 870 fio->new_blkaddr)) 871 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL); 872 alloc_new: 873 if (!bio) { 874 bio = __bio_alloc(fio, BIO_MAX_VECS); 875 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host, 876 fio->page->index, fio, GFP_NOIO); 877 878 add_bio_entry(fio->sbi, bio, page, fio->temp); 879 } else { 880 if (add_ipu_page(fio, &bio, page)) 881 goto alloc_new; 882 } 883 884 if (fio->io_wbc) 885 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 886 887 inc_page_count(fio->sbi, WB_DATA_TYPE(page)); 888 889 *fio->last_block = fio->new_blkaddr; 890 *fio->bio = bio; 891 892 return 0; 893 } 894 895 void f2fs_submit_page_write(struct f2fs_io_info *fio) 896 { 897 struct f2fs_sb_info *sbi = fio->sbi; 898 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 899 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 900 struct page *bio_page; 901 902 f2fs_bug_on(sbi, is_read_io(fio->op)); 903 904 f2fs_down_write(&io->io_rwsem); 905 next: 906 if (fio->in_list) { 907 spin_lock(&io->io_lock); 908 if (list_empty(&io->io_list)) { 909 spin_unlock(&io->io_lock); 910 goto out; 911 } 912 fio = list_first_entry(&io->io_list, 913 struct f2fs_io_info, list); 914 list_del(&fio->list); 915 spin_unlock(&io->io_lock); 916 } 917 918 verify_fio_blkaddr(fio); 919 920 if (fio->encrypted_page) 921 bio_page = fio->encrypted_page; 922 else if (fio->compressed_page) 923 bio_page = fio->compressed_page; 924 else 925 bio_page = fio->page; 926 927 /* set submitted = true as a return value */ 928 fio->submitted = true; 929 930 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 931 932 if (io->bio && 933 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio, 934 fio->new_blkaddr) || 935 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host, 936 bio_page->index, fio))) 937 __submit_merged_bio(io); 938 alloc_new: 939 if (io->bio == NULL) { 940 if (F2FS_IO_ALIGNED(sbi) && 941 (fio->type == DATA || fio->type == NODE) && 942 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 943 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 944 fio->retry = true; 945 goto skip; 946 } 947 io->bio = __bio_alloc(fio, BIO_MAX_VECS); 948 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host, 949 bio_page->index, fio, GFP_NOIO); 950 io->fio = *fio; 951 } 952 953 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 954 __submit_merged_bio(io); 955 goto alloc_new; 956 } 957 958 if (fio->io_wbc) 959 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE); 960 961 io->last_block_in_bio = fio->new_blkaddr; 962 963 trace_f2fs_submit_page_write(fio->page, fio); 964 skip: 965 if (fio->in_list) 966 goto next; 967 out: 968 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 969 !f2fs_is_checkpoint_ready(sbi)) 970 __submit_merged_bio(io); 971 f2fs_up_write(&io->io_rwsem); 972 } 973 974 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 975 unsigned nr_pages, unsigned op_flag, 976 pgoff_t first_idx, bool for_write) 977 { 978 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 979 struct bio *bio; 980 struct bio_post_read_ctx *ctx = NULL; 981 unsigned int post_read_steps = 0; 982 sector_t sector; 983 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or); 984 985 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages), 986 REQ_OP_READ | op_flag, 987 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset); 988 if (!bio) 989 return ERR_PTR(-ENOMEM); 990 bio->bi_iter.bi_sector = sector; 991 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS); 992 bio->bi_end_io = f2fs_read_end_io; 993 994 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 995 post_read_steps |= STEP_DECRYPT; 996 997 if (f2fs_need_verity(inode, first_idx)) 998 post_read_steps |= STEP_VERITY; 999 1000 /* 1001 * STEP_DECOMPRESS is handled specially, since a compressed file might 1002 * contain both compressed and uncompressed clusters. We'll allocate a 1003 * bio_post_read_ctx if the file is compressed, but the caller is 1004 * responsible for enabling STEP_DECOMPRESS if it's actually needed. 1005 */ 1006 1007 if (post_read_steps || f2fs_compressed_file(inode)) { 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 ctx->fs_blkaddr = blkaddr; 1014 bio->bi_private = ctx; 1015 } 1016 iostat_alloc_and_bind_ctx(sbi, bio, ctx); 1017 1018 return bio; 1019 } 1020 1021 /* This can handle encryption stuffs */ 1022 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 1023 block_t blkaddr, int op_flags, bool for_write) 1024 { 1025 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1026 struct bio *bio; 1027 1028 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags, 1029 page->index, for_write); 1030 if (IS_ERR(bio)) 1031 return PTR_ERR(bio); 1032 1033 /* wait for GCed page writeback via META_MAPPING */ 1034 f2fs_wait_on_block_writeback(inode, blkaddr); 1035 1036 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 1037 bio_put(bio); 1038 return -EFAULT; 1039 } 1040 ClearPageError(page); 1041 inc_page_count(sbi, F2FS_RD_DATA); 1042 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); 1043 __submit_bio(sbi, bio, DATA); 1044 return 0; 1045 } 1046 1047 static void __set_data_blkaddr(struct dnode_of_data *dn) 1048 { 1049 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 1050 __le32 *addr_array; 1051 int base = 0; 1052 1053 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 1054 base = get_extra_isize(dn->inode); 1055 1056 /* Get physical address of data block */ 1057 addr_array = blkaddr_in_node(rn); 1058 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 1059 } 1060 1061 /* 1062 * Lock ordering for the change of data block address: 1063 * ->data_page 1064 * ->node_page 1065 * update block addresses in the node page 1066 */ 1067 void f2fs_set_data_blkaddr(struct dnode_of_data *dn) 1068 { 1069 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1070 __set_data_blkaddr(dn); 1071 if (set_page_dirty(dn->node_page)) 1072 dn->node_changed = true; 1073 } 1074 1075 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1076 { 1077 dn->data_blkaddr = blkaddr; 1078 f2fs_set_data_blkaddr(dn); 1079 f2fs_update_extent_cache(dn); 1080 } 1081 1082 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 1083 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 1084 { 1085 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1086 int err; 1087 1088 if (!count) 1089 return 0; 1090 1091 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1092 return -EPERM; 1093 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1094 return err; 1095 1096 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 1097 dn->ofs_in_node, count); 1098 1099 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1100 1101 for (; count > 0; dn->ofs_in_node++) { 1102 block_t blkaddr = f2fs_data_blkaddr(dn); 1103 1104 if (blkaddr == NULL_ADDR) { 1105 dn->data_blkaddr = NEW_ADDR; 1106 __set_data_blkaddr(dn); 1107 count--; 1108 } 1109 } 1110 1111 if (set_page_dirty(dn->node_page)) 1112 dn->node_changed = true; 1113 return 0; 1114 } 1115 1116 /* Should keep dn->ofs_in_node unchanged */ 1117 int f2fs_reserve_new_block(struct dnode_of_data *dn) 1118 { 1119 unsigned int ofs_in_node = dn->ofs_in_node; 1120 int ret; 1121 1122 ret = f2fs_reserve_new_blocks(dn, 1); 1123 dn->ofs_in_node = ofs_in_node; 1124 return ret; 1125 } 1126 1127 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 1128 { 1129 bool need_put = dn->inode_page ? false : true; 1130 int err; 1131 1132 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE); 1133 if (err) 1134 return err; 1135 1136 if (dn->data_blkaddr == NULL_ADDR) 1137 err = f2fs_reserve_new_block(dn); 1138 if (err || need_put) 1139 f2fs_put_dnode(dn); 1140 return err; 1141 } 1142 1143 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 1144 { 1145 struct extent_info ei = {0, }; 1146 struct inode *inode = dn->inode; 1147 1148 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1149 dn->data_blkaddr = ei.blk + index - ei.fofs; 1150 return 0; 1151 } 1152 1153 return f2fs_reserve_block(dn, index); 1154 } 1155 1156 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, 1157 int op_flags, bool for_write) 1158 { 1159 struct address_space *mapping = inode->i_mapping; 1160 struct dnode_of_data dn; 1161 struct page *page; 1162 struct extent_info ei = {0, }; 1163 int err; 1164 1165 page = f2fs_grab_cache_page(mapping, index, for_write); 1166 if (!page) 1167 return ERR_PTR(-ENOMEM); 1168 1169 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1170 dn.data_blkaddr = ei.blk + index - ei.fofs; 1171 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr, 1172 DATA_GENERIC_ENHANCE_READ)) { 1173 err = -EFSCORRUPTED; 1174 goto put_err; 1175 } 1176 goto got_it; 1177 } 1178 1179 set_new_dnode(&dn, inode, NULL, NULL, 0); 1180 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1181 if (err) 1182 goto put_err; 1183 f2fs_put_dnode(&dn); 1184 1185 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1186 err = -ENOENT; 1187 goto put_err; 1188 } 1189 if (dn.data_blkaddr != NEW_ADDR && 1190 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode), 1191 dn.data_blkaddr, 1192 DATA_GENERIC_ENHANCE)) { 1193 err = -EFSCORRUPTED; 1194 goto put_err; 1195 } 1196 got_it: 1197 if (PageUptodate(page)) { 1198 unlock_page(page); 1199 return page; 1200 } 1201 1202 /* 1203 * A new dentry page is allocated but not able to be written, since its 1204 * new inode page couldn't be allocated due to -ENOSPC. 1205 * In such the case, its blkaddr can be remained as NEW_ADDR. 1206 * see, f2fs_add_link -> f2fs_get_new_data_page -> 1207 * f2fs_init_inode_metadata. 1208 */ 1209 if (dn.data_blkaddr == NEW_ADDR) { 1210 zero_user_segment(page, 0, PAGE_SIZE); 1211 if (!PageUptodate(page)) 1212 SetPageUptodate(page); 1213 unlock_page(page); 1214 return page; 1215 } 1216 1217 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr, 1218 op_flags, for_write); 1219 if (err) 1220 goto put_err; 1221 return page; 1222 1223 put_err: 1224 f2fs_put_page(page, 1); 1225 return ERR_PTR(err); 1226 } 1227 1228 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index) 1229 { 1230 struct address_space *mapping = inode->i_mapping; 1231 struct page *page; 1232 1233 page = find_get_page(mapping, index); 1234 if (page && PageUptodate(page)) 1235 return page; 1236 f2fs_put_page(page, 0); 1237 1238 page = f2fs_get_read_data_page(inode, index, 0, false); 1239 if (IS_ERR(page)) 1240 return page; 1241 1242 if (PageUptodate(page)) 1243 return page; 1244 1245 wait_on_page_locked(page); 1246 if (unlikely(!PageUptodate(page))) { 1247 f2fs_put_page(page, 0); 1248 return ERR_PTR(-EIO); 1249 } 1250 return page; 1251 } 1252 1253 /* 1254 * If it tries to access a hole, return an error. 1255 * Because, the callers, functions in dir.c and GC, should be able to know 1256 * whether this page exists or not. 1257 */ 1258 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, 1259 bool for_write) 1260 { 1261 struct address_space *mapping = inode->i_mapping; 1262 struct page *page; 1263 repeat: 1264 page = f2fs_get_read_data_page(inode, index, 0, for_write); 1265 if (IS_ERR(page)) 1266 return page; 1267 1268 /* wait for read completion */ 1269 lock_page(page); 1270 if (unlikely(page->mapping != mapping)) { 1271 f2fs_put_page(page, 1); 1272 goto repeat; 1273 } 1274 if (unlikely(!PageUptodate(page))) { 1275 f2fs_put_page(page, 1); 1276 return ERR_PTR(-EIO); 1277 } 1278 return page; 1279 } 1280 1281 /* 1282 * Caller ensures that this data page is never allocated. 1283 * A new zero-filled data page is allocated in the page cache. 1284 * 1285 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 1286 * f2fs_unlock_op(). 1287 * Note that, ipage is set only by make_empty_dir, and if any error occur, 1288 * ipage should be released by this function. 1289 */ 1290 struct page *f2fs_get_new_data_page(struct inode *inode, 1291 struct page *ipage, pgoff_t index, bool new_i_size) 1292 { 1293 struct address_space *mapping = inode->i_mapping; 1294 struct page *page; 1295 struct dnode_of_data dn; 1296 int err; 1297 1298 page = f2fs_grab_cache_page(mapping, index, true); 1299 if (!page) { 1300 /* 1301 * before exiting, we should make sure ipage will be released 1302 * if any error occur. 1303 */ 1304 f2fs_put_page(ipage, 1); 1305 return ERR_PTR(-ENOMEM); 1306 } 1307 1308 set_new_dnode(&dn, inode, ipage, NULL, 0); 1309 err = f2fs_reserve_block(&dn, index); 1310 if (err) { 1311 f2fs_put_page(page, 1); 1312 return ERR_PTR(err); 1313 } 1314 if (!ipage) 1315 f2fs_put_dnode(&dn); 1316 1317 if (PageUptodate(page)) 1318 goto got_it; 1319 1320 if (dn.data_blkaddr == NEW_ADDR) { 1321 zero_user_segment(page, 0, PAGE_SIZE); 1322 if (!PageUptodate(page)) 1323 SetPageUptodate(page); 1324 } else { 1325 f2fs_put_page(page, 1); 1326 1327 /* if ipage exists, blkaddr should be NEW_ADDR */ 1328 f2fs_bug_on(F2FS_I_SB(inode), ipage); 1329 page = f2fs_get_lock_data_page(inode, index, true); 1330 if (IS_ERR(page)) 1331 return page; 1332 } 1333 got_it: 1334 if (new_i_size && i_size_read(inode) < 1335 ((loff_t)(index + 1) << PAGE_SHIFT)) 1336 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 1337 return page; 1338 } 1339 1340 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 1341 { 1342 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1343 struct f2fs_summary sum; 1344 struct node_info ni; 1345 block_t old_blkaddr; 1346 blkcnt_t count = 1; 1347 int err; 1348 1349 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1350 return -EPERM; 1351 1352 err = f2fs_get_node_info(sbi, dn->nid, &ni, false); 1353 if (err) 1354 return err; 1355 1356 dn->data_blkaddr = f2fs_data_blkaddr(dn); 1357 if (dn->data_blkaddr != NULL_ADDR) 1358 goto alloc; 1359 1360 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1361 return err; 1362 1363 alloc: 1364 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1365 old_blkaddr = dn->data_blkaddr; 1366 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr, 1367 &sum, seg_type, NULL); 1368 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 1369 invalidate_mapping_pages(META_MAPPING(sbi), 1370 old_blkaddr, old_blkaddr); 1371 f2fs_invalidate_compress_page(sbi, old_blkaddr); 1372 } 1373 f2fs_update_data_blkaddr(dn, dn->data_blkaddr); 1374 return 0; 1375 } 1376 1377 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 1378 { 1379 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1380 if (lock) 1381 f2fs_down_read(&sbi->node_change); 1382 else 1383 f2fs_up_read(&sbi->node_change); 1384 } else { 1385 if (lock) 1386 f2fs_lock_op(sbi); 1387 else 1388 f2fs_unlock_op(sbi); 1389 } 1390 } 1391 1392 /* 1393 * f2fs_map_blocks() tries to find or build mapping relationship which 1394 * maps continuous logical blocks to physical blocks, and return such 1395 * info via f2fs_map_blocks structure. 1396 */ 1397 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 1398 int create, int flag) 1399 { 1400 unsigned int maxblocks = map->m_len; 1401 struct dnode_of_data dn; 1402 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1403 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE; 1404 pgoff_t pgofs, end_offset, end; 1405 int err = 0, ofs = 1; 1406 unsigned int ofs_in_node, last_ofs_in_node; 1407 blkcnt_t prealloc; 1408 struct extent_info ei = {0, }; 1409 block_t blkaddr; 1410 unsigned int start_pgofs; 1411 int bidx = 0; 1412 1413 if (!maxblocks) 1414 return 0; 1415 1416 map->m_bdev = inode->i_sb->s_bdev; 1417 map->m_multidev_dio = 1418 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag); 1419 1420 map->m_len = 0; 1421 map->m_flags = 0; 1422 1423 /* it only supports block size == page size */ 1424 pgofs = (pgoff_t)map->m_lblk; 1425 end = pgofs + maxblocks; 1426 1427 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 1428 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1429 map->m_may_create) 1430 goto next_dnode; 1431 1432 map->m_pblk = ei.blk + pgofs - ei.fofs; 1433 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 1434 map->m_flags = F2FS_MAP_MAPPED; 1435 if (map->m_next_extent) 1436 *map->m_next_extent = pgofs + map->m_len; 1437 1438 /* for hardware encryption, but to avoid potential issue in future */ 1439 if (flag == F2FS_GET_BLOCK_DIO) 1440 f2fs_wait_on_block_writeback_range(inode, 1441 map->m_pblk, map->m_len); 1442 1443 if (map->m_multidev_dio) { 1444 block_t blk_addr = map->m_pblk; 1445 1446 bidx = f2fs_target_device_index(sbi, map->m_pblk); 1447 1448 map->m_bdev = FDEV(bidx).bdev; 1449 map->m_pblk -= FDEV(bidx).start_blk; 1450 map->m_len = min(map->m_len, 1451 FDEV(bidx).end_blk + 1 - map->m_pblk); 1452 1453 if (map->m_may_create) 1454 f2fs_update_device_state(sbi, inode->i_ino, 1455 blk_addr, map->m_len); 1456 } 1457 goto out; 1458 } 1459 1460 next_dnode: 1461 if (map->m_may_create) 1462 f2fs_do_map_lock(sbi, flag, true); 1463 1464 /* When reading holes, we need its node page */ 1465 set_new_dnode(&dn, inode, NULL, NULL, 0); 1466 err = f2fs_get_dnode_of_data(&dn, pgofs, mode); 1467 if (err) { 1468 if (flag == F2FS_GET_BLOCK_BMAP) 1469 map->m_pblk = 0; 1470 1471 if (err == -ENOENT) { 1472 /* 1473 * There is one exceptional case that read_node_page() 1474 * may return -ENOENT due to filesystem has been 1475 * shutdown or cp_error, so force to convert error 1476 * number to EIO for such case. 1477 */ 1478 if (map->m_may_create && 1479 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 1480 f2fs_cp_error(sbi))) { 1481 err = -EIO; 1482 goto unlock_out; 1483 } 1484 1485 err = 0; 1486 if (map->m_next_pgofs) 1487 *map->m_next_pgofs = 1488 f2fs_get_next_page_offset(&dn, pgofs); 1489 if (map->m_next_extent) 1490 *map->m_next_extent = 1491 f2fs_get_next_page_offset(&dn, pgofs); 1492 } 1493 goto unlock_out; 1494 } 1495 1496 start_pgofs = pgofs; 1497 prealloc = 0; 1498 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 1499 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 1500 1501 next_block: 1502 blkaddr = f2fs_data_blkaddr(&dn); 1503 1504 if (__is_valid_data_blkaddr(blkaddr) && 1505 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) { 1506 err = -EFSCORRUPTED; 1507 goto sync_out; 1508 } 1509 1510 if (__is_valid_data_blkaddr(blkaddr)) { 1511 /* use out-place-update for driect IO under LFS mode */ 1512 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1513 map->m_may_create) { 1514 err = __allocate_data_block(&dn, map->m_seg_type); 1515 if (err) 1516 goto sync_out; 1517 blkaddr = dn.data_blkaddr; 1518 set_inode_flag(inode, FI_APPEND_WRITE); 1519 } 1520 } else { 1521 if (create) { 1522 if (unlikely(f2fs_cp_error(sbi))) { 1523 err = -EIO; 1524 goto sync_out; 1525 } 1526 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1527 if (blkaddr == NULL_ADDR) { 1528 prealloc++; 1529 last_ofs_in_node = dn.ofs_in_node; 1530 } 1531 } else { 1532 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO && 1533 flag != F2FS_GET_BLOCK_DIO); 1534 err = __allocate_data_block(&dn, 1535 map->m_seg_type); 1536 if (!err) { 1537 if (flag == F2FS_GET_BLOCK_PRE_DIO) 1538 file_need_truncate(inode); 1539 set_inode_flag(inode, FI_APPEND_WRITE); 1540 } 1541 } 1542 if (err) 1543 goto sync_out; 1544 map->m_flags |= F2FS_MAP_NEW; 1545 blkaddr = dn.data_blkaddr; 1546 } else { 1547 if (f2fs_compressed_file(inode) && 1548 f2fs_sanity_check_cluster(&dn) && 1549 (flag != F2FS_GET_BLOCK_FIEMAP || 1550 IS_ENABLED(CONFIG_F2FS_CHECK_FS))) { 1551 err = -EFSCORRUPTED; 1552 goto sync_out; 1553 } 1554 if (flag == F2FS_GET_BLOCK_BMAP) { 1555 map->m_pblk = 0; 1556 goto sync_out; 1557 } 1558 if (flag == F2FS_GET_BLOCK_PRECACHE) 1559 goto sync_out; 1560 if (flag == F2FS_GET_BLOCK_FIEMAP && 1561 blkaddr == NULL_ADDR) { 1562 if (map->m_next_pgofs) 1563 *map->m_next_pgofs = pgofs + 1; 1564 goto sync_out; 1565 } 1566 if (flag != F2FS_GET_BLOCK_FIEMAP) { 1567 /* for defragment case */ 1568 if (map->m_next_pgofs) 1569 *map->m_next_pgofs = pgofs + 1; 1570 goto sync_out; 1571 } 1572 } 1573 } 1574 1575 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1576 goto skip; 1577 1578 if (map->m_multidev_dio) 1579 bidx = f2fs_target_device_index(sbi, blkaddr); 1580 1581 if (map->m_len == 0) { 1582 /* preallocated unwritten block should be mapped for fiemap. */ 1583 if (blkaddr == NEW_ADDR) 1584 map->m_flags |= F2FS_MAP_UNWRITTEN; 1585 map->m_flags |= F2FS_MAP_MAPPED; 1586 1587 map->m_pblk = blkaddr; 1588 map->m_len = 1; 1589 1590 if (map->m_multidev_dio) 1591 map->m_bdev = FDEV(bidx).bdev; 1592 } else if ((map->m_pblk != NEW_ADDR && 1593 blkaddr == (map->m_pblk + ofs)) || 1594 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1595 flag == F2FS_GET_BLOCK_PRE_DIO) { 1596 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev) 1597 goto sync_out; 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 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) { 1651 /* 1652 * for hardware encryption, but to avoid potential issue 1653 * in future 1654 */ 1655 f2fs_wait_on_block_writeback_range(inode, 1656 map->m_pblk, map->m_len); 1657 invalidate_mapping_pages(META_MAPPING(sbi), 1658 map->m_pblk, map->m_pblk); 1659 1660 if (map->m_multidev_dio) { 1661 block_t blk_addr = map->m_pblk; 1662 1663 bidx = f2fs_target_device_index(sbi, map->m_pblk); 1664 1665 map->m_bdev = FDEV(bidx).bdev; 1666 map->m_pblk -= FDEV(bidx).start_blk; 1667 1668 if (map->m_may_create) 1669 f2fs_update_device_state(sbi, inode->i_ino, 1670 blk_addr, map->m_len); 1671 1672 f2fs_bug_on(sbi, blk_addr + map->m_len > 1673 FDEV(bidx).end_blk + 1); 1674 } 1675 } 1676 1677 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1678 if (map->m_flags & F2FS_MAP_MAPPED) { 1679 unsigned int ofs = start_pgofs - map->m_lblk; 1680 1681 f2fs_update_extent_cache_range(&dn, 1682 start_pgofs, map->m_pblk + ofs, 1683 map->m_len - ofs); 1684 } 1685 if (map->m_next_extent) 1686 *map->m_next_extent = pgofs + 1; 1687 } 1688 f2fs_put_dnode(&dn); 1689 unlock_out: 1690 if (map->m_may_create) { 1691 f2fs_do_map_lock(sbi, flag, false); 1692 f2fs_balance_fs(sbi, dn.node_changed); 1693 } 1694 out: 1695 trace_f2fs_map_blocks(inode, map, create, flag, err); 1696 return err; 1697 } 1698 1699 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) 1700 { 1701 struct f2fs_map_blocks map; 1702 block_t last_lblk; 1703 int err; 1704 1705 if (pos + len > i_size_read(inode)) 1706 return false; 1707 1708 map.m_lblk = F2FS_BYTES_TO_BLK(pos); 1709 map.m_next_pgofs = NULL; 1710 map.m_next_extent = NULL; 1711 map.m_seg_type = NO_CHECK_TYPE; 1712 map.m_may_create = false; 1713 last_lblk = F2FS_BLK_ALIGN(pos + len); 1714 1715 while (map.m_lblk < last_lblk) { 1716 map.m_len = last_lblk - map.m_lblk; 1717 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT); 1718 if (err || map.m_len == 0) 1719 return false; 1720 map.m_lblk += map.m_len; 1721 } 1722 return true; 1723 } 1724 1725 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes) 1726 { 1727 return (bytes >> inode->i_blkbits); 1728 } 1729 1730 static inline u64 blks_to_bytes(struct inode *inode, u64 blks) 1731 { 1732 return (blks << inode->i_blkbits); 1733 } 1734 1735 static int f2fs_xattr_fiemap(struct inode *inode, 1736 struct fiemap_extent_info *fieinfo) 1737 { 1738 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1739 struct page *page; 1740 struct node_info ni; 1741 __u64 phys = 0, len; 1742 __u32 flags; 1743 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1744 int err = 0; 1745 1746 if (f2fs_has_inline_xattr(inode)) { 1747 int offset; 1748 1749 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1750 inode->i_ino, false); 1751 if (!page) 1752 return -ENOMEM; 1753 1754 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false); 1755 if (err) { 1756 f2fs_put_page(page, 1); 1757 return err; 1758 } 1759 1760 phys = blks_to_bytes(inode, ni.blk_addr); 1761 offset = offsetof(struct f2fs_inode, i_addr) + 1762 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1763 get_inline_xattr_addrs(inode)); 1764 1765 phys += offset; 1766 len = inline_xattr_size(inode); 1767 1768 f2fs_put_page(page, 1); 1769 1770 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1771 1772 if (!xnid) 1773 flags |= FIEMAP_EXTENT_LAST; 1774 1775 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1776 trace_f2fs_fiemap(inode, 0, phys, len, flags, err); 1777 if (err || err == 1) 1778 return err; 1779 } 1780 1781 if (xnid) { 1782 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1783 if (!page) 1784 return -ENOMEM; 1785 1786 err = f2fs_get_node_info(sbi, xnid, &ni, false); 1787 if (err) { 1788 f2fs_put_page(page, 1); 1789 return err; 1790 } 1791 1792 phys = blks_to_bytes(inode, ni.blk_addr); 1793 len = inode->i_sb->s_blocksize; 1794 1795 f2fs_put_page(page, 1); 1796 1797 flags = FIEMAP_EXTENT_LAST; 1798 } 1799 1800 if (phys) { 1801 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1802 trace_f2fs_fiemap(inode, 0, phys, len, flags, err); 1803 } 1804 1805 return (err < 0 ? err : 0); 1806 } 1807 1808 static loff_t max_inode_blocks(struct inode *inode) 1809 { 1810 loff_t result = ADDRS_PER_INODE(inode); 1811 loff_t leaf_count = ADDRS_PER_BLOCK(inode); 1812 1813 /* two direct node blocks */ 1814 result += (leaf_count * 2); 1815 1816 /* two indirect node blocks */ 1817 leaf_count *= NIDS_PER_BLOCK; 1818 result += (leaf_count * 2); 1819 1820 /* one double indirect node block */ 1821 leaf_count *= NIDS_PER_BLOCK; 1822 result += leaf_count; 1823 1824 return result; 1825 } 1826 1827 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1828 u64 start, u64 len) 1829 { 1830 struct f2fs_map_blocks map; 1831 sector_t start_blk, last_blk; 1832 pgoff_t next_pgofs; 1833 u64 logical = 0, phys = 0, size = 0; 1834 u32 flags = 0; 1835 int ret = 0; 1836 bool compr_cluster = false, compr_appended; 1837 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size; 1838 unsigned int count_in_cluster = 0; 1839 loff_t maxbytes; 1840 1841 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { 1842 ret = f2fs_precache_extents(inode); 1843 if (ret) 1844 return ret; 1845 } 1846 1847 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR); 1848 if (ret) 1849 return ret; 1850 1851 inode_lock(inode); 1852 1853 maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS; 1854 if (start > maxbytes) { 1855 ret = -EFBIG; 1856 goto out; 1857 } 1858 1859 if (len > maxbytes || (maxbytes - len) < start) 1860 len = maxbytes - start; 1861 1862 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1863 ret = f2fs_xattr_fiemap(inode, fieinfo); 1864 goto out; 1865 } 1866 1867 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) { 1868 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1869 if (ret != -EAGAIN) 1870 goto out; 1871 } 1872 1873 if (bytes_to_blks(inode, len) == 0) 1874 len = blks_to_bytes(inode, 1); 1875 1876 start_blk = bytes_to_blks(inode, start); 1877 last_blk = bytes_to_blks(inode, start + len - 1); 1878 1879 next: 1880 memset(&map, 0, sizeof(map)); 1881 map.m_lblk = start_blk; 1882 map.m_len = bytes_to_blks(inode, len); 1883 map.m_next_pgofs = &next_pgofs; 1884 map.m_seg_type = NO_CHECK_TYPE; 1885 1886 if (compr_cluster) { 1887 map.m_lblk += 1; 1888 map.m_len = cluster_size - count_in_cluster; 1889 } 1890 1891 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP); 1892 if (ret) 1893 goto out; 1894 1895 /* HOLE */ 1896 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) { 1897 start_blk = next_pgofs; 1898 1899 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode, 1900 max_inode_blocks(inode))) 1901 goto prep_next; 1902 1903 flags |= FIEMAP_EXTENT_LAST; 1904 } 1905 1906 compr_appended = false; 1907 /* In a case of compressed cluster, append this to the last extent */ 1908 if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) || 1909 !(map.m_flags & F2FS_MAP_FLAGS))) { 1910 compr_appended = true; 1911 goto skip_fill; 1912 } 1913 1914 if (size) { 1915 flags |= FIEMAP_EXTENT_MERGED; 1916 if (IS_ENCRYPTED(inode)) 1917 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1918 1919 ret = fiemap_fill_next_extent(fieinfo, logical, 1920 phys, size, flags); 1921 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret); 1922 if (ret) 1923 goto out; 1924 size = 0; 1925 } 1926 1927 if (start_blk > last_blk) 1928 goto out; 1929 1930 skip_fill: 1931 if (map.m_pblk == COMPRESS_ADDR) { 1932 compr_cluster = true; 1933 count_in_cluster = 1; 1934 } else if (compr_appended) { 1935 unsigned int appended_blks = cluster_size - 1936 count_in_cluster + 1; 1937 size += blks_to_bytes(inode, appended_blks); 1938 start_blk += appended_blks; 1939 compr_cluster = false; 1940 } else { 1941 logical = blks_to_bytes(inode, start_blk); 1942 phys = __is_valid_data_blkaddr(map.m_pblk) ? 1943 blks_to_bytes(inode, map.m_pblk) : 0; 1944 size = blks_to_bytes(inode, map.m_len); 1945 flags = 0; 1946 1947 if (compr_cluster) { 1948 flags = FIEMAP_EXTENT_ENCODED; 1949 count_in_cluster += map.m_len; 1950 if (count_in_cluster == cluster_size) { 1951 compr_cluster = false; 1952 size += blks_to_bytes(inode, 1); 1953 } 1954 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) { 1955 flags = FIEMAP_EXTENT_UNWRITTEN; 1956 } 1957 1958 start_blk += bytes_to_blks(inode, size); 1959 } 1960 1961 prep_next: 1962 cond_resched(); 1963 if (fatal_signal_pending(current)) 1964 ret = -EINTR; 1965 else 1966 goto next; 1967 out: 1968 if (ret == 1) 1969 ret = 0; 1970 1971 inode_unlock(inode); 1972 return ret; 1973 } 1974 1975 static inline loff_t f2fs_readpage_limit(struct inode *inode) 1976 { 1977 if (IS_ENABLED(CONFIG_FS_VERITY) && 1978 (IS_VERITY(inode) || f2fs_verity_in_progress(inode))) 1979 return inode->i_sb->s_maxbytes; 1980 1981 return i_size_read(inode); 1982 } 1983 1984 static int f2fs_read_single_page(struct inode *inode, struct page *page, 1985 unsigned nr_pages, 1986 struct f2fs_map_blocks *map, 1987 struct bio **bio_ret, 1988 sector_t *last_block_in_bio, 1989 bool is_readahead) 1990 { 1991 struct bio *bio = *bio_ret; 1992 const unsigned blocksize = blks_to_bytes(inode, 1); 1993 sector_t block_in_file; 1994 sector_t last_block; 1995 sector_t last_block_in_file; 1996 sector_t block_nr; 1997 int ret = 0; 1998 1999 block_in_file = (sector_t)page_index(page); 2000 last_block = block_in_file + nr_pages; 2001 last_block_in_file = bytes_to_blks(inode, 2002 f2fs_readpage_limit(inode) + blocksize - 1); 2003 if (last_block > last_block_in_file) 2004 last_block = last_block_in_file; 2005 2006 /* just zeroing out page which is beyond EOF */ 2007 if (block_in_file >= last_block) 2008 goto zero_out; 2009 /* 2010 * Map blocks using the previous result first. 2011 */ 2012 if ((map->m_flags & F2FS_MAP_MAPPED) && 2013 block_in_file > map->m_lblk && 2014 block_in_file < (map->m_lblk + map->m_len)) 2015 goto got_it; 2016 2017 /* 2018 * Then do more f2fs_map_blocks() calls until we are 2019 * done with this page. 2020 */ 2021 map->m_lblk = block_in_file; 2022 map->m_len = last_block - block_in_file; 2023 2024 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT); 2025 if (ret) 2026 goto out; 2027 got_it: 2028 if ((map->m_flags & F2FS_MAP_MAPPED)) { 2029 block_nr = map->m_pblk + block_in_file - map->m_lblk; 2030 SetPageMappedToDisk(page); 2031 2032 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr, 2033 DATA_GENERIC_ENHANCE_READ)) { 2034 ret = -EFSCORRUPTED; 2035 goto out; 2036 } 2037 } else { 2038 zero_out: 2039 zero_user_segment(page, 0, PAGE_SIZE); 2040 if (f2fs_need_verity(inode, page->index) && 2041 !fsverity_verify_page(page)) { 2042 ret = -EIO; 2043 goto out; 2044 } 2045 if (!PageUptodate(page)) 2046 SetPageUptodate(page); 2047 unlock_page(page); 2048 goto out; 2049 } 2050 2051 /* 2052 * This page will go to BIO. Do we need to send this 2053 * BIO off first? 2054 */ 2055 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio, 2056 *last_block_in_bio, block_nr) || 2057 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) { 2058 submit_and_realloc: 2059 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2060 bio = NULL; 2061 } 2062 if (bio == NULL) { 2063 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages, 2064 is_readahead ? REQ_RAHEAD : 0, page->index, 2065 false); 2066 if (IS_ERR(bio)) { 2067 ret = PTR_ERR(bio); 2068 bio = NULL; 2069 goto out; 2070 } 2071 } 2072 2073 /* 2074 * If the page is under writeback, we need to wait for 2075 * its completion to see the correct decrypted data. 2076 */ 2077 f2fs_wait_on_block_writeback(inode, block_nr); 2078 2079 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2080 goto submit_and_realloc; 2081 2082 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 2083 f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE); 2084 ClearPageError(page); 2085 *last_block_in_bio = block_nr; 2086 goto out; 2087 out: 2088 *bio_ret = bio; 2089 return ret; 2090 } 2091 2092 #ifdef CONFIG_F2FS_FS_COMPRESSION 2093 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret, 2094 unsigned nr_pages, sector_t *last_block_in_bio, 2095 bool is_readahead, bool for_write) 2096 { 2097 struct dnode_of_data dn; 2098 struct inode *inode = cc->inode; 2099 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2100 struct bio *bio = *bio_ret; 2101 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size; 2102 sector_t last_block_in_file; 2103 const unsigned blocksize = blks_to_bytes(inode, 1); 2104 struct decompress_io_ctx *dic = NULL; 2105 struct extent_info ei = {0, }; 2106 bool from_dnode = true; 2107 int i; 2108 int ret = 0; 2109 2110 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc)); 2111 2112 last_block_in_file = bytes_to_blks(inode, 2113 f2fs_readpage_limit(inode) + blocksize - 1); 2114 2115 /* get rid of pages beyond EOF */ 2116 for (i = 0; i < cc->cluster_size; i++) { 2117 struct page *page = cc->rpages[i]; 2118 2119 if (!page) 2120 continue; 2121 if ((sector_t)page->index >= last_block_in_file) { 2122 zero_user_segment(page, 0, PAGE_SIZE); 2123 if (!PageUptodate(page)) 2124 SetPageUptodate(page); 2125 } else if (!PageUptodate(page)) { 2126 continue; 2127 } 2128 unlock_page(page); 2129 if (for_write) 2130 put_page(page); 2131 cc->rpages[i] = NULL; 2132 cc->nr_rpages--; 2133 } 2134 2135 /* we are done since all pages are beyond EOF */ 2136 if (f2fs_cluster_is_empty(cc)) 2137 goto out; 2138 2139 if (f2fs_lookup_extent_cache(inode, start_idx, &ei)) 2140 from_dnode = false; 2141 2142 if (!from_dnode) 2143 goto skip_reading_dnode; 2144 2145 set_new_dnode(&dn, inode, NULL, NULL, 0); 2146 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 2147 if (ret) 2148 goto out; 2149 2150 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR); 2151 2152 skip_reading_dnode: 2153 for (i = 1; i < cc->cluster_size; i++) { 2154 block_t blkaddr; 2155 2156 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page, 2157 dn.ofs_in_node + i) : 2158 ei.blk + i - 1; 2159 2160 if (!__is_valid_data_blkaddr(blkaddr)) 2161 break; 2162 2163 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) { 2164 ret = -EFAULT; 2165 goto out_put_dnode; 2166 } 2167 cc->nr_cpages++; 2168 2169 if (!from_dnode && i >= ei.c_len) 2170 break; 2171 } 2172 2173 /* nothing to decompress */ 2174 if (cc->nr_cpages == 0) { 2175 ret = 0; 2176 goto out_put_dnode; 2177 } 2178 2179 dic = f2fs_alloc_dic(cc); 2180 if (IS_ERR(dic)) { 2181 ret = PTR_ERR(dic); 2182 goto out_put_dnode; 2183 } 2184 2185 for (i = 0; i < cc->nr_cpages; i++) { 2186 struct page *page = dic->cpages[i]; 2187 block_t blkaddr; 2188 struct bio_post_read_ctx *ctx; 2189 2190 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page, 2191 dn.ofs_in_node + i + 1) : 2192 ei.blk + i; 2193 2194 f2fs_wait_on_block_writeback(inode, blkaddr); 2195 2196 if (f2fs_load_compressed_page(sbi, page, blkaddr)) { 2197 if (atomic_dec_and_test(&dic->remaining_pages)) 2198 f2fs_decompress_cluster(dic); 2199 continue; 2200 } 2201 2202 if (bio && (!page_is_mergeable(sbi, bio, 2203 *last_block_in_bio, blkaddr) || 2204 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) { 2205 submit_and_realloc: 2206 __submit_bio(sbi, bio, DATA); 2207 bio = NULL; 2208 } 2209 2210 if (!bio) { 2211 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages, 2212 is_readahead ? REQ_RAHEAD : 0, 2213 page->index, for_write); 2214 if (IS_ERR(bio)) { 2215 ret = PTR_ERR(bio); 2216 f2fs_decompress_end_io(dic, ret); 2217 f2fs_put_dnode(&dn); 2218 *bio_ret = NULL; 2219 return ret; 2220 } 2221 } 2222 2223 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2224 goto submit_and_realloc; 2225 2226 ctx = get_post_read_ctx(bio); 2227 ctx->enabled_steps |= STEP_DECOMPRESS; 2228 refcount_inc(&dic->refcnt); 2229 2230 inc_page_count(sbi, F2FS_RD_DATA); 2231 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); 2232 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE); 2233 ClearPageError(page); 2234 *last_block_in_bio = blkaddr; 2235 } 2236 2237 if (from_dnode) 2238 f2fs_put_dnode(&dn); 2239 2240 *bio_ret = bio; 2241 return 0; 2242 2243 out_put_dnode: 2244 if (from_dnode) 2245 f2fs_put_dnode(&dn); 2246 out: 2247 for (i = 0; i < cc->cluster_size; i++) { 2248 if (cc->rpages[i]) { 2249 ClearPageUptodate(cc->rpages[i]); 2250 ClearPageError(cc->rpages[i]); 2251 unlock_page(cc->rpages[i]); 2252 } 2253 } 2254 *bio_ret = bio; 2255 return ret; 2256 } 2257 #endif 2258 2259 /* 2260 * This function was originally taken from fs/mpage.c, and customized for f2fs. 2261 * Major change was from block_size == page_size in f2fs by default. 2262 */ 2263 static int f2fs_mpage_readpages(struct inode *inode, 2264 struct readahead_control *rac, struct page *page) 2265 { 2266 struct bio *bio = NULL; 2267 sector_t last_block_in_bio = 0; 2268 struct f2fs_map_blocks map; 2269 #ifdef CONFIG_F2FS_FS_COMPRESSION 2270 struct compress_ctx cc = { 2271 .inode = inode, 2272 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2273 .cluster_size = F2FS_I(inode)->i_cluster_size, 2274 .cluster_idx = NULL_CLUSTER, 2275 .rpages = NULL, 2276 .cpages = NULL, 2277 .nr_rpages = 0, 2278 .nr_cpages = 0, 2279 }; 2280 pgoff_t nc_cluster_idx = NULL_CLUSTER; 2281 #endif 2282 unsigned nr_pages = rac ? readahead_count(rac) : 1; 2283 unsigned max_nr_pages = nr_pages; 2284 int ret = 0; 2285 2286 map.m_pblk = 0; 2287 map.m_lblk = 0; 2288 map.m_len = 0; 2289 map.m_flags = 0; 2290 map.m_next_pgofs = NULL; 2291 map.m_next_extent = NULL; 2292 map.m_seg_type = NO_CHECK_TYPE; 2293 map.m_may_create = false; 2294 2295 for (; nr_pages; nr_pages--) { 2296 if (rac) { 2297 page = readahead_page(rac); 2298 prefetchw(&page->flags); 2299 } 2300 2301 #ifdef CONFIG_F2FS_FS_COMPRESSION 2302 if (f2fs_compressed_file(inode)) { 2303 /* there are remained comressed pages, submit them */ 2304 if (!f2fs_cluster_can_merge_page(&cc, page->index)) { 2305 ret = f2fs_read_multi_pages(&cc, &bio, 2306 max_nr_pages, 2307 &last_block_in_bio, 2308 rac != NULL, false); 2309 f2fs_destroy_compress_ctx(&cc, false); 2310 if (ret) 2311 goto set_error_page; 2312 } 2313 if (cc.cluster_idx == NULL_CLUSTER) { 2314 if (nc_cluster_idx == 2315 page->index >> cc.log_cluster_size) { 2316 goto read_single_page; 2317 } 2318 2319 ret = f2fs_is_compressed_cluster(inode, page->index); 2320 if (ret < 0) 2321 goto set_error_page; 2322 else if (!ret) { 2323 nc_cluster_idx = 2324 page->index >> cc.log_cluster_size; 2325 goto read_single_page; 2326 } 2327 2328 nc_cluster_idx = NULL_CLUSTER; 2329 } 2330 ret = f2fs_init_compress_ctx(&cc); 2331 if (ret) 2332 goto set_error_page; 2333 2334 f2fs_compress_ctx_add_page(&cc, page); 2335 2336 goto next_page; 2337 } 2338 read_single_page: 2339 #endif 2340 2341 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map, 2342 &bio, &last_block_in_bio, rac); 2343 if (ret) { 2344 #ifdef CONFIG_F2FS_FS_COMPRESSION 2345 set_error_page: 2346 #endif 2347 SetPageError(page); 2348 zero_user_segment(page, 0, PAGE_SIZE); 2349 unlock_page(page); 2350 } 2351 #ifdef CONFIG_F2FS_FS_COMPRESSION 2352 next_page: 2353 #endif 2354 if (rac) 2355 put_page(page); 2356 2357 #ifdef CONFIG_F2FS_FS_COMPRESSION 2358 if (f2fs_compressed_file(inode)) { 2359 /* last page */ 2360 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) { 2361 ret = f2fs_read_multi_pages(&cc, &bio, 2362 max_nr_pages, 2363 &last_block_in_bio, 2364 rac != NULL, false); 2365 f2fs_destroy_compress_ctx(&cc, false); 2366 } 2367 } 2368 #endif 2369 } 2370 if (bio) 2371 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2372 return ret; 2373 } 2374 2375 static int f2fs_read_data_folio(struct file *file, struct folio *folio) 2376 { 2377 struct page *page = &folio->page; 2378 struct inode *inode = page_file_mapping(page)->host; 2379 int ret = -EAGAIN; 2380 2381 trace_f2fs_readpage(page, DATA); 2382 2383 if (!f2fs_is_compress_backend_ready(inode)) { 2384 unlock_page(page); 2385 return -EOPNOTSUPP; 2386 } 2387 2388 /* If the file has inline data, try to read it directly */ 2389 if (f2fs_has_inline_data(inode)) 2390 ret = f2fs_read_inline_data(inode, page); 2391 if (ret == -EAGAIN) 2392 ret = f2fs_mpage_readpages(inode, NULL, page); 2393 return ret; 2394 } 2395 2396 static void f2fs_readahead(struct readahead_control *rac) 2397 { 2398 struct inode *inode = rac->mapping->host; 2399 2400 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac)); 2401 2402 if (!f2fs_is_compress_backend_ready(inode)) 2403 return; 2404 2405 /* If the file has inline data, skip readahead */ 2406 if (f2fs_has_inline_data(inode)) 2407 return; 2408 2409 f2fs_mpage_readpages(inode, rac, NULL); 2410 } 2411 2412 int f2fs_encrypt_one_page(struct f2fs_io_info *fio) 2413 { 2414 struct inode *inode = fio->page->mapping->host; 2415 struct page *mpage, *page; 2416 gfp_t gfp_flags = GFP_NOFS; 2417 2418 if (!f2fs_encrypted_file(inode)) 2419 return 0; 2420 2421 page = fio->compressed_page ? fio->compressed_page : fio->page; 2422 2423 /* wait for GCed page writeback via META_MAPPING */ 2424 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr); 2425 2426 if (fscrypt_inode_uses_inline_crypto(inode)) 2427 return 0; 2428 2429 retry_encrypt: 2430 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page, 2431 PAGE_SIZE, 0, gfp_flags); 2432 if (IS_ERR(fio->encrypted_page)) { 2433 /* flush pending IOs and wait for a while in the ENOMEM case */ 2434 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 2435 f2fs_flush_merged_writes(fio->sbi); 2436 memalloc_retry_wait(GFP_NOFS); 2437 gfp_flags |= __GFP_NOFAIL; 2438 goto retry_encrypt; 2439 } 2440 return PTR_ERR(fio->encrypted_page); 2441 } 2442 2443 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr); 2444 if (mpage) { 2445 if (PageUptodate(mpage)) 2446 memcpy(page_address(mpage), 2447 page_address(fio->encrypted_page), PAGE_SIZE); 2448 f2fs_put_page(mpage, 1); 2449 } 2450 return 0; 2451 } 2452 2453 static inline bool check_inplace_update_policy(struct inode *inode, 2454 struct f2fs_io_info *fio) 2455 { 2456 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2457 unsigned int policy = SM_I(sbi)->ipu_policy; 2458 2459 if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) && 2460 is_inode_flag_set(inode, FI_OPU_WRITE)) 2461 return false; 2462 if (policy & (0x1 << F2FS_IPU_FORCE)) 2463 return true; 2464 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi)) 2465 return true; 2466 if (policy & (0x1 << F2FS_IPU_UTIL) && 2467 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2468 return true; 2469 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) && 2470 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2471 return true; 2472 2473 /* 2474 * IPU for rewrite async pages 2475 */ 2476 if (policy & (0x1 << F2FS_IPU_ASYNC) && 2477 fio && fio->op == REQ_OP_WRITE && 2478 !(fio->op_flags & REQ_SYNC) && 2479 !IS_ENCRYPTED(inode)) 2480 return true; 2481 2482 /* this is only set during fdatasync */ 2483 if (policy & (0x1 << F2FS_IPU_FSYNC) && 2484 is_inode_flag_set(inode, FI_NEED_IPU)) 2485 return true; 2486 2487 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2488 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2489 return true; 2490 2491 return false; 2492 } 2493 2494 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) 2495 { 2496 /* swap file is migrating in aligned write mode */ 2497 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 2498 return false; 2499 2500 if (f2fs_is_pinned_file(inode)) 2501 return true; 2502 2503 /* if this is cold file, we should overwrite to avoid fragmentation */ 2504 if (file_is_cold(inode)) 2505 return true; 2506 2507 return check_inplace_update_policy(inode, fio); 2508 } 2509 2510 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) 2511 { 2512 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2513 2514 /* The below cases were checked when setting it. */ 2515 if (f2fs_is_pinned_file(inode)) 2516 return false; 2517 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 2518 return true; 2519 if (f2fs_lfs_mode(sbi)) 2520 return true; 2521 if (S_ISDIR(inode->i_mode)) 2522 return true; 2523 if (IS_NOQUOTA(inode)) 2524 return true; 2525 if (f2fs_is_atomic_file(inode)) 2526 return true; 2527 2528 /* swap file is migrating in aligned write mode */ 2529 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 2530 return true; 2531 2532 if (is_inode_flag_set(inode, FI_OPU_WRITE)) 2533 return true; 2534 2535 if (fio) { 2536 if (page_private_gcing(fio->page)) 2537 return true; 2538 if (page_private_dummy(fio->page)) 2539 return true; 2540 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2541 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2542 return true; 2543 } 2544 return false; 2545 } 2546 2547 static inline bool need_inplace_update(struct f2fs_io_info *fio) 2548 { 2549 struct inode *inode = fio->page->mapping->host; 2550 2551 if (f2fs_should_update_outplace(inode, fio)) 2552 return false; 2553 2554 return f2fs_should_update_inplace(inode, fio); 2555 } 2556 2557 int f2fs_do_write_data_page(struct f2fs_io_info *fio) 2558 { 2559 struct page *page = fio->page; 2560 struct inode *inode = page->mapping->host; 2561 struct dnode_of_data dn; 2562 struct extent_info ei = {0, }; 2563 struct node_info ni; 2564 bool ipu_force = false; 2565 int err = 0; 2566 2567 set_new_dnode(&dn, inode, NULL, NULL, 0); 2568 if (need_inplace_update(fio) && 2569 f2fs_lookup_extent_cache(inode, page->index, &ei)) { 2570 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 2571 2572 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2573 DATA_GENERIC_ENHANCE)) 2574 return -EFSCORRUPTED; 2575 2576 ipu_force = true; 2577 fio->need_lock = LOCK_DONE; 2578 goto got_it; 2579 } 2580 2581 /* Deadlock due to between page->lock and f2fs_lock_op */ 2582 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 2583 return -EAGAIN; 2584 2585 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 2586 if (err) 2587 goto out; 2588 2589 fio->old_blkaddr = dn.data_blkaddr; 2590 2591 /* This page is already truncated */ 2592 if (fio->old_blkaddr == NULL_ADDR) { 2593 ClearPageUptodate(page); 2594 clear_page_private_gcing(page); 2595 goto out_writepage; 2596 } 2597 got_it: 2598 if (__is_valid_data_blkaddr(fio->old_blkaddr) && 2599 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2600 DATA_GENERIC_ENHANCE)) { 2601 err = -EFSCORRUPTED; 2602 goto out_writepage; 2603 } 2604 /* 2605 * If current allocation needs SSR, 2606 * it had better in-place writes for updated data. 2607 */ 2608 if (ipu_force || 2609 (__is_valid_data_blkaddr(fio->old_blkaddr) && 2610 need_inplace_update(fio))) { 2611 err = f2fs_encrypt_one_page(fio); 2612 if (err) 2613 goto out_writepage; 2614 2615 set_page_writeback(page); 2616 ClearPageError(page); 2617 f2fs_put_dnode(&dn); 2618 if (fio->need_lock == LOCK_REQ) 2619 f2fs_unlock_op(fio->sbi); 2620 err = f2fs_inplace_write_data(fio); 2621 if (err) { 2622 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 2623 fscrypt_finalize_bounce_page(&fio->encrypted_page); 2624 if (PageWriteback(page)) 2625 end_page_writeback(page); 2626 } else { 2627 set_inode_flag(inode, FI_UPDATE_WRITE); 2628 } 2629 trace_f2fs_do_write_data_page(fio->page, IPU); 2630 return err; 2631 } 2632 2633 if (fio->need_lock == LOCK_RETRY) { 2634 if (!f2fs_trylock_op(fio->sbi)) { 2635 err = -EAGAIN; 2636 goto out_writepage; 2637 } 2638 fio->need_lock = LOCK_REQ; 2639 } 2640 2641 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false); 2642 if (err) 2643 goto out_writepage; 2644 2645 fio->version = ni.version; 2646 2647 err = f2fs_encrypt_one_page(fio); 2648 if (err) 2649 goto out_writepage; 2650 2651 set_page_writeback(page); 2652 ClearPageError(page); 2653 2654 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR) 2655 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false); 2656 2657 /* LFS mode write path */ 2658 f2fs_outplace_write_data(&dn, fio); 2659 trace_f2fs_do_write_data_page(page, OPU); 2660 set_inode_flag(inode, FI_APPEND_WRITE); 2661 if (page->index == 0) 2662 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 2663 out_writepage: 2664 f2fs_put_dnode(&dn); 2665 out: 2666 if (fio->need_lock == LOCK_REQ) 2667 f2fs_unlock_op(fio->sbi); 2668 return err; 2669 } 2670 2671 int f2fs_write_single_data_page(struct page *page, int *submitted, 2672 struct bio **bio, 2673 sector_t *last_block, 2674 struct writeback_control *wbc, 2675 enum iostat_type io_type, 2676 int compr_blocks, 2677 bool allow_balance) 2678 { 2679 struct inode *inode = page->mapping->host; 2680 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2681 loff_t i_size = i_size_read(inode); 2682 const pgoff_t end_index = ((unsigned long long)i_size) 2683 >> PAGE_SHIFT; 2684 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT; 2685 unsigned offset = 0; 2686 bool need_balance_fs = false; 2687 int err = 0; 2688 struct f2fs_io_info fio = { 2689 .sbi = sbi, 2690 .ino = inode->i_ino, 2691 .type = DATA, 2692 .op = REQ_OP_WRITE, 2693 .op_flags = wbc_to_write_flags(wbc), 2694 .old_blkaddr = NULL_ADDR, 2695 .page = page, 2696 .encrypted_page = NULL, 2697 .submitted = false, 2698 .compr_blocks = compr_blocks, 2699 .need_lock = LOCK_RETRY, 2700 .io_type = io_type, 2701 .io_wbc = wbc, 2702 .bio = bio, 2703 .last_block = last_block, 2704 }; 2705 2706 trace_f2fs_writepage(page, DATA); 2707 2708 /* we should bypass data pages to proceed the kworkder jobs */ 2709 if (unlikely(f2fs_cp_error(sbi))) { 2710 mapping_set_error(page->mapping, -EIO); 2711 /* 2712 * don't drop any dirty dentry pages for keeping lastest 2713 * directory structure. 2714 */ 2715 if (S_ISDIR(inode->i_mode)) 2716 goto redirty_out; 2717 goto out; 2718 } 2719 2720 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2721 goto redirty_out; 2722 2723 if (page->index < end_index || 2724 f2fs_verity_in_progress(inode) || 2725 compr_blocks) 2726 goto write; 2727 2728 /* 2729 * If the offset is out-of-range of file size, 2730 * this page does not have to be written to disk. 2731 */ 2732 offset = i_size & (PAGE_SIZE - 1); 2733 if ((page->index >= end_index + 1) || !offset) 2734 goto out; 2735 2736 zero_user_segment(page, offset, PAGE_SIZE); 2737 write: 2738 if (f2fs_is_drop_cache(inode)) 2739 goto out; 2740 /* we should not write 0'th page having journal header */ 2741 if (f2fs_is_volatile_file(inode) && (!page->index || 2742 (!wbc->for_reclaim && 2743 f2fs_available_free_memory(sbi, BASE_CHECK)))) 2744 goto redirty_out; 2745 2746 /* Dentry/quota blocks are controlled by checkpoint */ 2747 if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) { 2748 /* 2749 * We need to wait for node_write to avoid block allocation during 2750 * checkpoint. This can only happen to quota writes which can cause 2751 * the below discard race condition. 2752 */ 2753 if (IS_NOQUOTA(inode)) 2754 f2fs_down_read(&sbi->node_write); 2755 2756 fio.need_lock = LOCK_DONE; 2757 err = f2fs_do_write_data_page(&fio); 2758 2759 if (IS_NOQUOTA(inode)) 2760 f2fs_up_read(&sbi->node_write); 2761 2762 goto done; 2763 } 2764 2765 if (!wbc->for_reclaim) 2766 need_balance_fs = true; 2767 else if (has_not_enough_free_secs(sbi, 0, 0)) 2768 goto redirty_out; 2769 else 2770 set_inode_flag(inode, FI_HOT_DATA); 2771 2772 err = -EAGAIN; 2773 if (f2fs_has_inline_data(inode)) { 2774 err = f2fs_write_inline_data(inode, page); 2775 if (!err) 2776 goto out; 2777 } 2778 2779 if (err == -EAGAIN) { 2780 err = f2fs_do_write_data_page(&fio); 2781 if (err == -EAGAIN) { 2782 fio.need_lock = LOCK_REQ; 2783 err = f2fs_do_write_data_page(&fio); 2784 } 2785 } 2786 2787 if (err) { 2788 file_set_keep_isize(inode); 2789 } else { 2790 spin_lock(&F2FS_I(inode)->i_size_lock); 2791 if (F2FS_I(inode)->last_disk_size < psize) 2792 F2FS_I(inode)->last_disk_size = psize; 2793 spin_unlock(&F2FS_I(inode)->i_size_lock); 2794 } 2795 2796 done: 2797 if (err && err != -ENOENT) 2798 goto redirty_out; 2799 2800 out: 2801 inode_dec_dirty_pages(inode); 2802 if (err) { 2803 ClearPageUptodate(page); 2804 clear_page_private_gcing(page); 2805 } 2806 2807 if (wbc->for_reclaim) { 2808 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA); 2809 clear_inode_flag(inode, FI_HOT_DATA); 2810 f2fs_remove_dirty_inode(inode); 2811 submitted = NULL; 2812 } 2813 unlock_page(page); 2814 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) && 2815 !F2FS_I(inode)->cp_task && allow_balance) 2816 f2fs_balance_fs(sbi, need_balance_fs); 2817 2818 if (unlikely(f2fs_cp_error(sbi))) { 2819 f2fs_submit_merged_write(sbi, DATA); 2820 f2fs_submit_merged_ipu_write(sbi, bio, NULL); 2821 submitted = NULL; 2822 } 2823 2824 if (submitted) 2825 *submitted = fio.submitted ? 1 : 0; 2826 2827 return 0; 2828 2829 redirty_out: 2830 redirty_page_for_writepage(wbc, page); 2831 /* 2832 * pageout() in MM traslates EAGAIN, so calls handle_write_error() 2833 * -> mapping_set_error() -> set_bit(AS_EIO, ...). 2834 * file_write_and_wait_range() will see EIO error, which is critical 2835 * to return value of fsync() followed by atomic_write failure to user. 2836 */ 2837 if (!err || wbc->for_reclaim) 2838 return AOP_WRITEPAGE_ACTIVATE; 2839 unlock_page(page); 2840 return err; 2841 } 2842 2843 static int f2fs_write_data_page(struct page *page, 2844 struct writeback_control *wbc) 2845 { 2846 #ifdef CONFIG_F2FS_FS_COMPRESSION 2847 struct inode *inode = page->mapping->host; 2848 2849 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 2850 goto out; 2851 2852 if (f2fs_compressed_file(inode)) { 2853 if (f2fs_is_compressed_cluster(inode, page->index)) { 2854 redirty_page_for_writepage(wbc, page); 2855 return AOP_WRITEPAGE_ACTIVATE; 2856 } 2857 } 2858 out: 2859 #endif 2860 2861 return f2fs_write_single_data_page(page, NULL, NULL, NULL, 2862 wbc, FS_DATA_IO, 0, true); 2863 } 2864 2865 /* 2866 * This function was copied from write_cche_pages from mm/page-writeback.c. 2867 * The major change is making write step of cold data page separately from 2868 * warm/hot data page. 2869 */ 2870 static int f2fs_write_cache_pages(struct address_space *mapping, 2871 struct writeback_control *wbc, 2872 enum iostat_type io_type) 2873 { 2874 int ret = 0; 2875 int done = 0, retry = 0; 2876 struct pagevec pvec; 2877 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2878 struct bio *bio = NULL; 2879 sector_t last_block; 2880 #ifdef CONFIG_F2FS_FS_COMPRESSION 2881 struct inode *inode = mapping->host; 2882 struct compress_ctx cc = { 2883 .inode = inode, 2884 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2885 .cluster_size = F2FS_I(inode)->i_cluster_size, 2886 .cluster_idx = NULL_CLUSTER, 2887 .rpages = NULL, 2888 .nr_rpages = 0, 2889 .cpages = NULL, 2890 .valid_nr_cpages = 0, 2891 .rbuf = NULL, 2892 .cbuf = NULL, 2893 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size, 2894 .private = NULL, 2895 }; 2896 #endif 2897 int nr_pages; 2898 pgoff_t index; 2899 pgoff_t end; /* Inclusive */ 2900 pgoff_t done_index; 2901 int range_whole = 0; 2902 xa_mark_t tag; 2903 int nwritten = 0; 2904 int submitted = 0; 2905 int i; 2906 2907 pagevec_init(&pvec); 2908 2909 if (get_dirty_pages(mapping->host) <= 2910 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 2911 set_inode_flag(mapping->host, FI_HOT_DATA); 2912 else 2913 clear_inode_flag(mapping->host, FI_HOT_DATA); 2914 2915 if (wbc->range_cyclic) { 2916 index = mapping->writeback_index; /* prev offset */ 2917 end = -1; 2918 } else { 2919 index = wbc->range_start >> PAGE_SHIFT; 2920 end = wbc->range_end >> PAGE_SHIFT; 2921 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2922 range_whole = 1; 2923 } 2924 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2925 tag = PAGECACHE_TAG_TOWRITE; 2926 else 2927 tag = PAGECACHE_TAG_DIRTY; 2928 retry: 2929 retry = 0; 2930 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2931 tag_pages_for_writeback(mapping, index, end); 2932 done_index = index; 2933 while (!done && !retry && (index <= end)) { 2934 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 2935 tag); 2936 if (nr_pages == 0) 2937 break; 2938 2939 for (i = 0; i < nr_pages; i++) { 2940 struct page *page = pvec.pages[i]; 2941 bool need_readd; 2942 readd: 2943 need_readd = false; 2944 #ifdef CONFIG_F2FS_FS_COMPRESSION 2945 if (f2fs_compressed_file(inode)) { 2946 void *fsdata = NULL; 2947 struct page *pagep; 2948 int ret2; 2949 2950 ret = f2fs_init_compress_ctx(&cc); 2951 if (ret) { 2952 done = 1; 2953 break; 2954 } 2955 2956 if (!f2fs_cluster_can_merge_page(&cc, 2957 page->index)) { 2958 ret = f2fs_write_multi_pages(&cc, 2959 &submitted, wbc, io_type); 2960 if (!ret) 2961 need_readd = true; 2962 goto result; 2963 } 2964 2965 if (unlikely(f2fs_cp_error(sbi))) 2966 goto lock_page; 2967 2968 if (!f2fs_cluster_is_empty(&cc)) 2969 goto lock_page; 2970 2971 ret2 = f2fs_prepare_compress_overwrite( 2972 inode, &pagep, 2973 page->index, &fsdata); 2974 if (ret2 < 0) { 2975 ret = ret2; 2976 done = 1; 2977 break; 2978 } else if (ret2 && 2979 (!f2fs_compress_write_end(inode, 2980 fsdata, page->index, 1) || 2981 !f2fs_all_cluster_page_loaded(&cc, 2982 &pvec, i, nr_pages))) { 2983 retry = 1; 2984 break; 2985 } 2986 } 2987 #endif 2988 /* give a priority to WB_SYNC threads */ 2989 if (atomic_read(&sbi->wb_sync_req[DATA]) && 2990 wbc->sync_mode == WB_SYNC_NONE) { 2991 done = 1; 2992 break; 2993 } 2994 #ifdef CONFIG_F2FS_FS_COMPRESSION 2995 lock_page: 2996 #endif 2997 done_index = page->index; 2998 retry_write: 2999 lock_page(page); 3000 3001 if (unlikely(page->mapping != mapping)) { 3002 continue_unlock: 3003 unlock_page(page); 3004 continue; 3005 } 3006 3007 if (!PageDirty(page)) { 3008 /* someone wrote it for us */ 3009 goto continue_unlock; 3010 } 3011 3012 if (PageWriteback(page)) { 3013 if (wbc->sync_mode != WB_SYNC_NONE) 3014 f2fs_wait_on_page_writeback(page, 3015 DATA, true, true); 3016 else 3017 goto continue_unlock; 3018 } 3019 3020 if (!clear_page_dirty_for_io(page)) 3021 goto continue_unlock; 3022 3023 #ifdef CONFIG_F2FS_FS_COMPRESSION 3024 if (f2fs_compressed_file(inode)) { 3025 get_page(page); 3026 f2fs_compress_ctx_add_page(&cc, page); 3027 continue; 3028 } 3029 #endif 3030 ret = f2fs_write_single_data_page(page, &submitted, 3031 &bio, &last_block, wbc, io_type, 3032 0, true); 3033 if (ret == AOP_WRITEPAGE_ACTIVATE) 3034 unlock_page(page); 3035 #ifdef CONFIG_F2FS_FS_COMPRESSION 3036 result: 3037 #endif 3038 nwritten += submitted; 3039 wbc->nr_to_write -= submitted; 3040 3041 if (unlikely(ret)) { 3042 /* 3043 * keep nr_to_write, since vfs uses this to 3044 * get # of written pages. 3045 */ 3046 if (ret == AOP_WRITEPAGE_ACTIVATE) { 3047 ret = 0; 3048 goto next; 3049 } else if (ret == -EAGAIN) { 3050 ret = 0; 3051 if (wbc->sync_mode == WB_SYNC_ALL) { 3052 f2fs_io_schedule_timeout( 3053 DEFAULT_IO_TIMEOUT); 3054 goto retry_write; 3055 } 3056 goto next; 3057 } 3058 done_index = page->index + 1; 3059 done = 1; 3060 break; 3061 } 3062 3063 if (wbc->nr_to_write <= 0 && 3064 wbc->sync_mode == WB_SYNC_NONE) { 3065 done = 1; 3066 break; 3067 } 3068 next: 3069 if (need_readd) 3070 goto readd; 3071 } 3072 pagevec_release(&pvec); 3073 cond_resched(); 3074 } 3075 #ifdef CONFIG_F2FS_FS_COMPRESSION 3076 /* flush remained pages in compress cluster */ 3077 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) { 3078 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type); 3079 nwritten += submitted; 3080 wbc->nr_to_write -= submitted; 3081 if (ret) { 3082 done = 1; 3083 retry = 0; 3084 } 3085 } 3086 if (f2fs_compressed_file(inode)) 3087 f2fs_destroy_compress_ctx(&cc, false); 3088 #endif 3089 if (retry) { 3090 index = 0; 3091 end = -1; 3092 goto retry; 3093 } 3094 if (wbc->range_cyclic && !done) 3095 done_index = 0; 3096 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 3097 mapping->writeback_index = done_index; 3098 3099 if (nwritten) 3100 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 3101 NULL, 0, DATA); 3102 /* submit cached bio of IPU write */ 3103 if (bio) 3104 f2fs_submit_merged_ipu_write(sbi, &bio, NULL); 3105 3106 return ret; 3107 } 3108 3109 static inline bool __should_serialize_io(struct inode *inode, 3110 struct writeback_control *wbc) 3111 { 3112 /* to avoid deadlock in path of data flush */ 3113 if (F2FS_I(inode)->cp_task) 3114 return false; 3115 3116 if (!S_ISREG(inode->i_mode)) 3117 return false; 3118 if (IS_NOQUOTA(inode)) 3119 return false; 3120 3121 if (f2fs_need_compress_data(inode)) 3122 return true; 3123 if (wbc->sync_mode != WB_SYNC_ALL) 3124 return true; 3125 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks) 3126 return true; 3127 return false; 3128 } 3129 3130 static int __f2fs_write_data_pages(struct address_space *mapping, 3131 struct writeback_control *wbc, 3132 enum iostat_type io_type) 3133 { 3134 struct inode *inode = mapping->host; 3135 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3136 struct blk_plug plug; 3137 int ret; 3138 bool locked = false; 3139 3140 /* deal with chardevs and other special file */ 3141 if (!mapping->a_ops->writepage) 3142 return 0; 3143 3144 /* skip writing if there is no dirty page in this inode */ 3145 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 3146 return 0; 3147 3148 /* during POR, we don't need to trigger writepage at all. */ 3149 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 3150 goto skip_write; 3151 3152 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) && 3153 wbc->sync_mode == WB_SYNC_NONE && 3154 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 3155 f2fs_available_free_memory(sbi, DIRTY_DENTS)) 3156 goto skip_write; 3157 3158 /* skip writing in file defragment preparing stage */ 3159 if (is_inode_flag_set(inode, FI_SKIP_WRITES)) 3160 goto skip_write; 3161 3162 trace_f2fs_writepages(mapping->host, wbc, DATA); 3163 3164 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 3165 if (wbc->sync_mode == WB_SYNC_ALL) 3166 atomic_inc(&sbi->wb_sync_req[DATA]); 3167 else if (atomic_read(&sbi->wb_sync_req[DATA])) { 3168 /* to avoid potential deadlock */ 3169 if (current->plug) 3170 blk_finish_plug(current->plug); 3171 goto skip_write; 3172 } 3173 3174 if (__should_serialize_io(inode, wbc)) { 3175 mutex_lock(&sbi->writepages); 3176 locked = true; 3177 } 3178 3179 blk_start_plug(&plug); 3180 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 3181 blk_finish_plug(&plug); 3182 3183 if (locked) 3184 mutex_unlock(&sbi->writepages); 3185 3186 if (wbc->sync_mode == WB_SYNC_ALL) 3187 atomic_dec(&sbi->wb_sync_req[DATA]); 3188 /* 3189 * if some pages were truncated, we cannot guarantee its mapping->host 3190 * to detect pending bios. 3191 */ 3192 3193 f2fs_remove_dirty_inode(inode); 3194 return ret; 3195 3196 skip_write: 3197 wbc->pages_skipped += get_dirty_pages(inode); 3198 trace_f2fs_writepages(mapping->host, wbc, DATA); 3199 return 0; 3200 } 3201 3202 static int f2fs_write_data_pages(struct address_space *mapping, 3203 struct writeback_control *wbc) 3204 { 3205 struct inode *inode = mapping->host; 3206 3207 return __f2fs_write_data_pages(mapping, wbc, 3208 F2FS_I(inode)->cp_task == current ? 3209 FS_CP_DATA_IO : FS_DATA_IO); 3210 } 3211 3212 void f2fs_write_failed(struct inode *inode, loff_t to) 3213 { 3214 loff_t i_size = i_size_read(inode); 3215 3216 if (IS_NOQUOTA(inode)) 3217 return; 3218 3219 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */ 3220 if (to > i_size && !f2fs_verity_in_progress(inode)) { 3221 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3222 filemap_invalidate_lock(inode->i_mapping); 3223 3224 truncate_pagecache(inode, i_size); 3225 f2fs_truncate_blocks(inode, i_size, true); 3226 3227 filemap_invalidate_unlock(inode->i_mapping); 3228 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3229 } 3230 } 3231 3232 static int prepare_write_begin(struct f2fs_sb_info *sbi, 3233 struct page *page, loff_t pos, unsigned len, 3234 block_t *blk_addr, bool *node_changed) 3235 { 3236 struct inode *inode = page->mapping->host; 3237 pgoff_t index = page->index; 3238 struct dnode_of_data dn; 3239 struct page *ipage; 3240 bool locked = false; 3241 struct extent_info ei = {0, }; 3242 int err = 0; 3243 int flag; 3244 3245 /* 3246 * If a whole page is being written and we already preallocated all the 3247 * blocks, then there is no need to get a block address now. 3248 */ 3249 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL)) 3250 return 0; 3251 3252 /* f2fs_lock_op avoids race between write CP and convert_inline_page */ 3253 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode)) 3254 flag = F2FS_GET_BLOCK_DEFAULT; 3255 else 3256 flag = F2FS_GET_BLOCK_PRE_AIO; 3257 3258 if (f2fs_has_inline_data(inode) || 3259 (pos & PAGE_MASK) >= i_size_read(inode)) { 3260 f2fs_do_map_lock(sbi, flag, true); 3261 locked = true; 3262 } 3263 3264 restart: 3265 /* check inline_data */ 3266 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3267 if (IS_ERR(ipage)) { 3268 err = PTR_ERR(ipage); 3269 goto unlock_out; 3270 } 3271 3272 set_new_dnode(&dn, inode, ipage, ipage, 0); 3273 3274 if (f2fs_has_inline_data(inode)) { 3275 if (pos + len <= MAX_INLINE_DATA(inode)) { 3276 f2fs_do_read_inline_data(page, ipage); 3277 set_inode_flag(inode, FI_DATA_EXIST); 3278 if (inode->i_nlink) 3279 set_page_private_inline(ipage); 3280 } else { 3281 err = f2fs_convert_inline_page(&dn, page); 3282 if (err) 3283 goto out; 3284 if (dn.data_blkaddr == NULL_ADDR) 3285 err = f2fs_get_block(&dn, index); 3286 } 3287 } else if (locked) { 3288 err = f2fs_get_block(&dn, index); 3289 } else { 3290 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 3291 dn.data_blkaddr = ei.blk + index - ei.fofs; 3292 } else { 3293 /* hole case */ 3294 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3295 if (err || dn.data_blkaddr == NULL_ADDR) { 3296 f2fs_put_dnode(&dn); 3297 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 3298 true); 3299 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO); 3300 locked = true; 3301 goto restart; 3302 } 3303 } 3304 } 3305 3306 /* convert_inline_page can make node_changed */ 3307 *blk_addr = dn.data_blkaddr; 3308 *node_changed = dn.node_changed; 3309 out: 3310 f2fs_put_dnode(&dn); 3311 unlock_out: 3312 if (locked) 3313 f2fs_do_map_lock(sbi, flag, false); 3314 return err; 3315 } 3316 3317 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 3318 loff_t pos, unsigned len, struct page **pagep, void **fsdata) 3319 { 3320 struct inode *inode = mapping->host; 3321 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3322 struct page *page = NULL; 3323 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 3324 bool need_balance = false, drop_atomic = false; 3325 block_t blkaddr = NULL_ADDR; 3326 int err = 0; 3327 3328 trace_f2fs_write_begin(inode, pos, len); 3329 3330 if (!f2fs_is_checkpoint_ready(sbi)) { 3331 err = -ENOSPC; 3332 goto fail; 3333 } 3334 3335 if ((f2fs_is_atomic_file(inode) && 3336 !f2fs_available_free_memory(sbi, INMEM_PAGES)) || 3337 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) { 3338 err = -ENOMEM; 3339 drop_atomic = true; 3340 goto fail; 3341 } 3342 3343 /* 3344 * We should check this at this moment to avoid deadlock on inode page 3345 * and #0 page. The locking rule for inline_data conversion should be: 3346 * lock_page(page #0) -> lock_page(inode_page) 3347 */ 3348 if (index != 0) { 3349 err = f2fs_convert_inline_inode(inode); 3350 if (err) 3351 goto fail; 3352 } 3353 3354 #ifdef CONFIG_F2FS_FS_COMPRESSION 3355 if (f2fs_compressed_file(inode)) { 3356 int ret; 3357 3358 *fsdata = NULL; 3359 3360 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode))) 3361 goto repeat; 3362 3363 ret = f2fs_prepare_compress_overwrite(inode, pagep, 3364 index, fsdata); 3365 if (ret < 0) { 3366 err = ret; 3367 goto fail; 3368 } else if (ret) { 3369 return 0; 3370 } 3371 } 3372 #endif 3373 3374 repeat: 3375 /* 3376 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 3377 * wait_for_stable_page. Will wait that below with our IO control. 3378 */ 3379 page = f2fs_pagecache_get_page(mapping, index, 3380 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 3381 if (!page) { 3382 err = -ENOMEM; 3383 goto fail; 3384 } 3385 3386 /* TODO: cluster can be compressed due to race with .writepage */ 3387 3388 *pagep = page; 3389 3390 err = prepare_write_begin(sbi, page, pos, len, 3391 &blkaddr, &need_balance); 3392 if (err) 3393 goto fail; 3394 3395 if (need_balance && !IS_NOQUOTA(inode) && 3396 has_not_enough_free_secs(sbi, 0, 0)) { 3397 unlock_page(page); 3398 f2fs_balance_fs(sbi, true); 3399 lock_page(page); 3400 if (page->mapping != mapping) { 3401 /* The page got truncated from under us */ 3402 f2fs_put_page(page, 1); 3403 goto repeat; 3404 } 3405 } 3406 3407 f2fs_wait_on_page_writeback(page, DATA, false, true); 3408 3409 if (len == PAGE_SIZE || PageUptodate(page)) 3410 return 0; 3411 3412 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) && 3413 !f2fs_verity_in_progress(inode)) { 3414 zero_user_segment(page, len, PAGE_SIZE); 3415 return 0; 3416 } 3417 3418 if (blkaddr == NEW_ADDR) { 3419 zero_user_segment(page, 0, PAGE_SIZE); 3420 SetPageUptodate(page); 3421 } else { 3422 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 3423 DATA_GENERIC_ENHANCE_READ)) { 3424 err = -EFSCORRUPTED; 3425 goto fail; 3426 } 3427 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true); 3428 if (err) 3429 goto fail; 3430 3431 lock_page(page); 3432 if (unlikely(page->mapping != mapping)) { 3433 f2fs_put_page(page, 1); 3434 goto repeat; 3435 } 3436 if (unlikely(!PageUptodate(page))) { 3437 err = -EIO; 3438 goto fail; 3439 } 3440 } 3441 return 0; 3442 3443 fail: 3444 f2fs_put_page(page, 1); 3445 f2fs_write_failed(inode, pos + len); 3446 if (drop_atomic) 3447 f2fs_drop_inmem_pages_all(sbi, false); 3448 return err; 3449 } 3450 3451 static int f2fs_write_end(struct file *file, 3452 struct address_space *mapping, 3453 loff_t pos, unsigned len, unsigned copied, 3454 struct page *page, void *fsdata) 3455 { 3456 struct inode *inode = page->mapping->host; 3457 3458 trace_f2fs_write_end(inode, pos, len, copied); 3459 3460 /* 3461 * This should be come from len == PAGE_SIZE, and we expect copied 3462 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 3463 * let generic_perform_write() try to copy data again through copied=0. 3464 */ 3465 if (!PageUptodate(page)) { 3466 if (unlikely(copied != len)) 3467 copied = 0; 3468 else 3469 SetPageUptodate(page); 3470 } 3471 3472 #ifdef CONFIG_F2FS_FS_COMPRESSION 3473 /* overwrite compressed file */ 3474 if (f2fs_compressed_file(inode) && fsdata) { 3475 f2fs_compress_write_end(inode, fsdata, page->index, copied); 3476 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3477 3478 if (pos + copied > i_size_read(inode) && 3479 !f2fs_verity_in_progress(inode)) 3480 f2fs_i_size_write(inode, pos + copied); 3481 return copied; 3482 } 3483 #endif 3484 3485 if (!copied) 3486 goto unlock_out; 3487 3488 set_page_dirty(page); 3489 3490 if (pos + copied > i_size_read(inode) && 3491 !f2fs_verity_in_progress(inode)) 3492 f2fs_i_size_write(inode, pos + copied); 3493 unlock_out: 3494 f2fs_put_page(page, 1); 3495 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3496 return copied; 3497 } 3498 3499 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length) 3500 { 3501 struct inode *inode = folio->mapping->host; 3502 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3503 3504 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 3505 (offset || length != folio_size(folio))) 3506 return; 3507 3508 if (folio_test_dirty(folio)) { 3509 if (inode->i_ino == F2FS_META_INO(sbi)) { 3510 dec_page_count(sbi, F2FS_DIRTY_META); 3511 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 3512 dec_page_count(sbi, F2FS_DIRTY_NODES); 3513 } else { 3514 inode_dec_dirty_pages(inode); 3515 f2fs_remove_dirty_inode(inode); 3516 } 3517 } 3518 3519 clear_page_private_gcing(&folio->page); 3520 3521 if (test_opt(sbi, COMPRESS_CACHE) && 3522 inode->i_ino == F2FS_COMPRESS_INO(sbi)) 3523 clear_page_private_data(&folio->page); 3524 3525 if (page_private_atomic(&folio->page)) 3526 return f2fs_drop_inmem_page(inode, &folio->page); 3527 3528 folio_detach_private(folio); 3529 } 3530 3531 bool f2fs_release_folio(struct folio *folio, gfp_t wait) 3532 { 3533 struct f2fs_sb_info *sbi; 3534 3535 /* If this is dirty folio, keep private data */ 3536 if (folio_test_dirty(folio)) 3537 return false; 3538 3539 /* This is atomic written page, keep Private */ 3540 if (page_private_atomic(&folio->page)) 3541 return false; 3542 3543 sbi = F2FS_M_SB(folio->mapping); 3544 if (test_opt(sbi, COMPRESS_CACHE)) { 3545 struct inode *inode = folio->mapping->host; 3546 3547 if (inode->i_ino == F2FS_COMPRESS_INO(sbi)) 3548 clear_page_private_data(&folio->page); 3549 } 3550 3551 clear_page_private_gcing(&folio->page); 3552 3553 folio_detach_private(folio); 3554 return true; 3555 } 3556 3557 static bool f2fs_dirty_data_folio(struct address_space *mapping, 3558 struct folio *folio) 3559 { 3560 struct inode *inode = mapping->host; 3561 3562 trace_f2fs_set_page_dirty(&folio->page, DATA); 3563 3564 if (!folio_test_uptodate(folio)) 3565 folio_mark_uptodate(folio); 3566 BUG_ON(folio_test_swapcache(folio)); 3567 3568 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) { 3569 if (!page_private_atomic(&folio->page)) { 3570 f2fs_register_inmem_page(inode, &folio->page); 3571 return true; 3572 } 3573 /* 3574 * Previously, this page has been registered, we just 3575 * return here. 3576 */ 3577 return false; 3578 } 3579 3580 if (!folio_test_dirty(folio)) { 3581 filemap_dirty_folio(mapping, folio); 3582 f2fs_update_dirty_folio(inode, folio); 3583 return true; 3584 } 3585 return false; 3586 } 3587 3588 3589 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block) 3590 { 3591 #ifdef CONFIG_F2FS_FS_COMPRESSION 3592 struct dnode_of_data dn; 3593 sector_t start_idx, blknr = 0; 3594 int ret; 3595 3596 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size); 3597 3598 set_new_dnode(&dn, inode, NULL, NULL, 0); 3599 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 3600 if (ret) 3601 return 0; 3602 3603 if (dn.data_blkaddr != COMPRESS_ADDR) { 3604 dn.ofs_in_node += block - start_idx; 3605 blknr = f2fs_data_blkaddr(&dn); 3606 if (!__is_valid_data_blkaddr(blknr)) 3607 blknr = 0; 3608 } 3609 3610 f2fs_put_dnode(&dn); 3611 return blknr; 3612 #else 3613 return 0; 3614 #endif 3615 } 3616 3617 3618 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 3619 { 3620 struct inode *inode = mapping->host; 3621 sector_t blknr = 0; 3622 3623 if (f2fs_has_inline_data(inode)) 3624 goto out; 3625 3626 /* make sure allocating whole blocks */ 3627 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 3628 filemap_write_and_wait(mapping); 3629 3630 /* Block number less than F2FS MAX BLOCKS */ 3631 if (unlikely(block >= max_file_blocks(inode))) 3632 goto out; 3633 3634 if (f2fs_compressed_file(inode)) { 3635 blknr = f2fs_bmap_compress(inode, block); 3636 } else { 3637 struct f2fs_map_blocks map; 3638 3639 memset(&map, 0, sizeof(map)); 3640 map.m_lblk = block; 3641 map.m_len = 1; 3642 map.m_next_pgofs = NULL; 3643 map.m_seg_type = NO_CHECK_TYPE; 3644 3645 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP)) 3646 blknr = map.m_pblk; 3647 } 3648 out: 3649 trace_f2fs_bmap(inode, block, blknr); 3650 return blknr; 3651 } 3652 3653 #ifdef CONFIG_MIGRATION 3654 #include <linux/migrate.h> 3655 3656 int f2fs_migrate_page(struct address_space *mapping, 3657 struct page *newpage, struct page *page, enum migrate_mode mode) 3658 { 3659 int rc, extra_count; 3660 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 3661 bool atomic_written = page_private_atomic(page); 3662 3663 BUG_ON(PageWriteback(page)); 3664 3665 /* migrating an atomic written page is safe with the inmem_lock hold */ 3666 if (atomic_written) { 3667 if (mode != MIGRATE_SYNC) 3668 return -EBUSY; 3669 if (!mutex_trylock(&fi->inmem_lock)) 3670 return -EAGAIN; 3671 } 3672 3673 /* one extra reference was held for atomic_write page */ 3674 extra_count = atomic_written ? 1 : 0; 3675 rc = migrate_page_move_mapping(mapping, newpage, 3676 page, extra_count); 3677 if (rc != MIGRATEPAGE_SUCCESS) { 3678 if (atomic_written) 3679 mutex_unlock(&fi->inmem_lock); 3680 return rc; 3681 } 3682 3683 if (atomic_written) { 3684 struct inmem_pages *cur; 3685 3686 list_for_each_entry(cur, &fi->inmem_pages, list) 3687 if (cur->page == page) { 3688 cur->page = newpage; 3689 break; 3690 } 3691 mutex_unlock(&fi->inmem_lock); 3692 put_page(page); 3693 get_page(newpage); 3694 } 3695 3696 /* guarantee to start from no stale private field */ 3697 set_page_private(newpage, 0); 3698 if (PagePrivate(page)) { 3699 set_page_private(newpage, page_private(page)); 3700 SetPagePrivate(newpage); 3701 get_page(newpage); 3702 3703 set_page_private(page, 0); 3704 ClearPagePrivate(page); 3705 put_page(page); 3706 } 3707 3708 if (mode != MIGRATE_SYNC_NO_COPY) 3709 migrate_page_copy(newpage, page); 3710 else 3711 migrate_page_states(newpage, page); 3712 3713 return MIGRATEPAGE_SUCCESS; 3714 } 3715 #endif 3716 3717 #ifdef CONFIG_SWAP 3718 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk, 3719 unsigned int blkcnt) 3720 { 3721 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3722 unsigned int blkofs; 3723 unsigned int blk_per_sec = BLKS_PER_SEC(sbi); 3724 unsigned int secidx = start_blk / blk_per_sec; 3725 unsigned int end_sec = secidx + blkcnt / blk_per_sec; 3726 int ret = 0; 3727 3728 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3729 filemap_invalidate_lock(inode->i_mapping); 3730 3731 set_inode_flag(inode, FI_ALIGNED_WRITE); 3732 set_inode_flag(inode, FI_OPU_WRITE); 3733 3734 for (; secidx < end_sec; secidx++) { 3735 f2fs_down_write(&sbi->pin_sem); 3736 3737 f2fs_lock_op(sbi); 3738 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false); 3739 f2fs_unlock_op(sbi); 3740 3741 set_inode_flag(inode, FI_SKIP_WRITES); 3742 3743 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) { 3744 struct page *page; 3745 unsigned int blkidx = secidx * blk_per_sec + blkofs; 3746 3747 page = f2fs_get_lock_data_page(inode, blkidx, true); 3748 if (IS_ERR(page)) { 3749 f2fs_up_write(&sbi->pin_sem); 3750 ret = PTR_ERR(page); 3751 goto done; 3752 } 3753 3754 set_page_dirty(page); 3755 f2fs_put_page(page, 1); 3756 } 3757 3758 clear_inode_flag(inode, FI_SKIP_WRITES); 3759 3760 ret = filemap_fdatawrite(inode->i_mapping); 3761 3762 f2fs_up_write(&sbi->pin_sem); 3763 3764 if (ret) 3765 break; 3766 } 3767 3768 done: 3769 clear_inode_flag(inode, FI_SKIP_WRITES); 3770 clear_inode_flag(inode, FI_OPU_WRITE); 3771 clear_inode_flag(inode, FI_ALIGNED_WRITE); 3772 3773 filemap_invalidate_unlock(inode->i_mapping); 3774 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3775 3776 return ret; 3777 } 3778 3779 static int check_swap_activate(struct swap_info_struct *sis, 3780 struct file *swap_file, sector_t *span) 3781 { 3782 struct address_space *mapping = swap_file->f_mapping; 3783 struct inode *inode = mapping->host; 3784 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3785 sector_t cur_lblock; 3786 sector_t last_lblock; 3787 sector_t pblock; 3788 sector_t lowest_pblock = -1; 3789 sector_t highest_pblock = 0; 3790 int nr_extents = 0; 3791 unsigned long nr_pblocks; 3792 unsigned int blks_per_sec = BLKS_PER_SEC(sbi); 3793 unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1; 3794 unsigned int not_aligned = 0; 3795 int ret = 0; 3796 3797 /* 3798 * Map all the blocks into the extent list. This code doesn't try 3799 * to be very smart. 3800 */ 3801 cur_lblock = 0; 3802 last_lblock = bytes_to_blks(inode, i_size_read(inode)); 3803 3804 while (cur_lblock < last_lblock && cur_lblock < sis->max) { 3805 struct f2fs_map_blocks map; 3806 retry: 3807 cond_resched(); 3808 3809 memset(&map, 0, sizeof(map)); 3810 map.m_lblk = cur_lblock; 3811 map.m_len = last_lblock - cur_lblock; 3812 map.m_next_pgofs = NULL; 3813 map.m_next_extent = NULL; 3814 map.m_seg_type = NO_CHECK_TYPE; 3815 map.m_may_create = false; 3816 3817 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP); 3818 if (ret) 3819 goto out; 3820 3821 /* hole */ 3822 if (!(map.m_flags & F2FS_MAP_FLAGS)) { 3823 f2fs_err(sbi, "Swapfile has holes"); 3824 ret = -EINVAL; 3825 goto out; 3826 } 3827 3828 pblock = map.m_pblk; 3829 nr_pblocks = map.m_len; 3830 3831 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask || 3832 nr_pblocks & sec_blks_mask) { 3833 not_aligned++; 3834 3835 nr_pblocks = roundup(nr_pblocks, blks_per_sec); 3836 if (cur_lblock + nr_pblocks > sis->max) 3837 nr_pblocks -= blks_per_sec; 3838 3839 if (!nr_pblocks) { 3840 /* this extent is last one */ 3841 nr_pblocks = map.m_len; 3842 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section"); 3843 goto next; 3844 } 3845 3846 ret = f2fs_migrate_blocks(inode, cur_lblock, 3847 nr_pblocks); 3848 if (ret) 3849 goto out; 3850 goto retry; 3851 } 3852 next: 3853 if (cur_lblock + nr_pblocks >= sis->max) 3854 nr_pblocks = sis->max - cur_lblock; 3855 3856 if (cur_lblock) { /* exclude the header page */ 3857 if (pblock < lowest_pblock) 3858 lowest_pblock = pblock; 3859 if (pblock + nr_pblocks - 1 > highest_pblock) 3860 highest_pblock = pblock + nr_pblocks - 1; 3861 } 3862 3863 /* 3864 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks 3865 */ 3866 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock); 3867 if (ret < 0) 3868 goto out; 3869 nr_extents += ret; 3870 cur_lblock += nr_pblocks; 3871 } 3872 ret = nr_extents; 3873 *span = 1 + highest_pblock - lowest_pblock; 3874 if (cur_lblock == 0) 3875 cur_lblock = 1; /* force Empty message */ 3876 sis->max = cur_lblock; 3877 sis->pages = cur_lblock - 1; 3878 sis->highest_bit = cur_lblock - 1; 3879 out: 3880 if (not_aligned) 3881 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)", 3882 not_aligned, blks_per_sec * F2FS_BLKSIZE); 3883 return ret; 3884 } 3885 3886 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3887 sector_t *span) 3888 { 3889 struct inode *inode = file_inode(file); 3890 int ret; 3891 3892 if (!S_ISREG(inode->i_mode)) 3893 return -EINVAL; 3894 3895 if (f2fs_readonly(F2FS_I_SB(inode)->sb)) 3896 return -EROFS; 3897 3898 if (f2fs_lfs_mode(F2FS_I_SB(inode))) { 3899 f2fs_err(F2FS_I_SB(inode), 3900 "Swapfile not supported in LFS mode"); 3901 return -EINVAL; 3902 } 3903 3904 ret = f2fs_convert_inline_inode(inode); 3905 if (ret) 3906 return ret; 3907 3908 if (!f2fs_disable_compressed_file(inode)) 3909 return -EINVAL; 3910 3911 f2fs_precache_extents(inode); 3912 3913 ret = check_swap_activate(sis, file, span); 3914 if (ret < 0) 3915 return ret; 3916 3917 set_inode_flag(inode, FI_PIN_FILE); 3918 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3919 return ret; 3920 } 3921 3922 static void f2fs_swap_deactivate(struct file *file) 3923 { 3924 struct inode *inode = file_inode(file); 3925 3926 clear_inode_flag(inode, FI_PIN_FILE); 3927 } 3928 #else 3929 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3930 sector_t *span) 3931 { 3932 return -EOPNOTSUPP; 3933 } 3934 3935 static void f2fs_swap_deactivate(struct file *file) 3936 { 3937 } 3938 #endif 3939 3940 const struct address_space_operations f2fs_dblock_aops = { 3941 .read_folio = f2fs_read_data_folio, 3942 .readahead = f2fs_readahead, 3943 .writepage = f2fs_write_data_page, 3944 .writepages = f2fs_write_data_pages, 3945 .write_begin = f2fs_write_begin, 3946 .write_end = f2fs_write_end, 3947 .dirty_folio = f2fs_dirty_data_folio, 3948 .invalidate_folio = f2fs_invalidate_folio, 3949 .release_folio = f2fs_release_folio, 3950 .direct_IO = noop_direct_IO, 3951 .bmap = f2fs_bmap, 3952 .swap_activate = f2fs_swap_activate, 3953 .swap_deactivate = f2fs_swap_deactivate, 3954 #ifdef CONFIG_MIGRATION 3955 .migratepage = f2fs_migrate_page, 3956 #endif 3957 }; 3958 3959 void f2fs_clear_page_cache_dirty_tag(struct page *page) 3960 { 3961 struct address_space *mapping = page_mapping(page); 3962 unsigned long flags; 3963 3964 xa_lock_irqsave(&mapping->i_pages, flags); 3965 __xa_clear_mark(&mapping->i_pages, page_index(page), 3966 PAGECACHE_TAG_DIRTY); 3967 xa_unlock_irqrestore(&mapping->i_pages, flags); 3968 } 3969 3970 int __init f2fs_init_post_read_processing(void) 3971 { 3972 bio_post_read_ctx_cache = 3973 kmem_cache_create("f2fs_bio_post_read_ctx", 3974 sizeof(struct bio_post_read_ctx), 0, 0, NULL); 3975 if (!bio_post_read_ctx_cache) 3976 goto fail; 3977 bio_post_read_ctx_pool = 3978 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 3979 bio_post_read_ctx_cache); 3980 if (!bio_post_read_ctx_pool) 3981 goto fail_free_cache; 3982 return 0; 3983 3984 fail_free_cache: 3985 kmem_cache_destroy(bio_post_read_ctx_cache); 3986 fail: 3987 return -ENOMEM; 3988 } 3989 3990 void f2fs_destroy_post_read_processing(void) 3991 { 3992 mempool_destroy(bio_post_read_ctx_pool); 3993 kmem_cache_destroy(bio_post_read_ctx_cache); 3994 } 3995 3996 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) 3997 { 3998 if (!f2fs_sb_has_encrypt(sbi) && 3999 !f2fs_sb_has_verity(sbi) && 4000 !f2fs_sb_has_compression(sbi)) 4001 return 0; 4002 4003 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq", 4004 WQ_UNBOUND | WQ_HIGHPRI, 4005 num_online_cpus()); 4006 if (!sbi->post_read_wq) 4007 return -ENOMEM; 4008 return 0; 4009 } 4010 4011 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) 4012 { 4013 if (sbi->post_read_wq) 4014 destroy_workqueue(sbi->post_read_wq); 4015 } 4016 4017 int __init f2fs_init_bio_entry_cache(void) 4018 { 4019 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab", 4020 sizeof(struct bio_entry)); 4021 if (!bio_entry_slab) 4022 return -ENOMEM; 4023 return 0; 4024 } 4025 4026 void f2fs_destroy_bio_entry_cache(void) 4027 { 4028 kmem_cache_destroy(bio_entry_slab); 4029 } 4030 4031 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 4032 unsigned int flags, struct iomap *iomap, 4033 struct iomap *srcmap) 4034 { 4035 struct f2fs_map_blocks map = {}; 4036 pgoff_t next_pgofs = 0; 4037 int err; 4038 4039 map.m_lblk = bytes_to_blks(inode, offset); 4040 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1; 4041 map.m_next_pgofs = &next_pgofs; 4042 map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint); 4043 if (flags & IOMAP_WRITE) 4044 map.m_may_create = true; 4045 4046 err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE, 4047 F2FS_GET_BLOCK_DIO); 4048 if (err) 4049 return err; 4050 4051 iomap->offset = blks_to_bytes(inode, map.m_lblk); 4052 4053 /* 4054 * When inline encryption is enabled, sometimes I/O to an encrypted file 4055 * has to be broken up to guarantee DUN contiguity. Handle this by 4056 * limiting the length of the mapping returned. 4057 */ 4058 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len); 4059 4060 if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) { 4061 iomap->length = blks_to_bytes(inode, map.m_len); 4062 if (map.m_flags & F2FS_MAP_MAPPED) { 4063 iomap->type = IOMAP_MAPPED; 4064 iomap->flags |= IOMAP_F_MERGED; 4065 } else { 4066 iomap->type = IOMAP_UNWRITTEN; 4067 } 4068 if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk))) 4069 return -EINVAL; 4070 4071 iomap->bdev = map.m_bdev; 4072 iomap->addr = blks_to_bytes(inode, map.m_pblk); 4073 } else { 4074 iomap->length = blks_to_bytes(inode, next_pgofs) - 4075 iomap->offset; 4076 iomap->type = IOMAP_HOLE; 4077 iomap->addr = IOMAP_NULL_ADDR; 4078 } 4079 4080 if (map.m_flags & F2FS_MAP_NEW) 4081 iomap->flags |= IOMAP_F_NEW; 4082 if ((inode->i_state & I_DIRTY_DATASYNC) || 4083 offset + length > i_size_read(inode)) 4084 iomap->flags |= IOMAP_F_DIRTY; 4085 4086 return 0; 4087 } 4088 4089 const struct iomap_ops f2fs_iomap_ops = { 4090 .iomap_begin = f2fs_iomap_begin, 4091 }; 4092