1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/data.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/buffer_head.h> 11 #include <linux/mpage.h> 12 #include <linux/writeback.h> 13 #include <linux/backing-dev.h> 14 #include <linux/pagevec.h> 15 #include <linux/blkdev.h> 16 #include <linux/bio.h> 17 #include <linux/swap.h> 18 #include <linux/prefetch.h> 19 #include <linux/uio.h> 20 #include <linux/cleancache.h> 21 #include <linux/sched/signal.h> 22 23 #include "f2fs.h" 24 #include "node.h" 25 #include "segment.h" 26 #include "trace.h" 27 #include <trace/events/f2fs.h> 28 29 #define NUM_PREALLOC_POST_READ_CTXS 128 30 31 static struct kmem_cache *bio_post_read_ctx_cache; 32 static struct kmem_cache *bio_entry_slab; 33 static mempool_t *bio_post_read_ctx_pool; 34 static struct bio_set f2fs_bioset; 35 36 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE 37 38 int __init f2fs_init_bioset(void) 39 { 40 if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE, 41 0, BIOSET_NEED_BVECS)) 42 return -ENOMEM; 43 return 0; 44 } 45 46 void f2fs_destroy_bioset(void) 47 { 48 bioset_exit(&f2fs_bioset); 49 } 50 51 static inline struct bio *__f2fs_bio_alloc(gfp_t gfp_mask, 52 unsigned int nr_iovecs) 53 { 54 return bio_alloc_bioset(gfp_mask, nr_iovecs, &f2fs_bioset); 55 } 56 57 struct bio *f2fs_bio_alloc(struct f2fs_sb_info *sbi, int npages, bool no_fail) 58 { 59 struct bio *bio; 60 61 if (no_fail) { 62 /* No failure on bio allocation */ 63 bio = __f2fs_bio_alloc(GFP_NOIO, npages); 64 if (!bio) 65 bio = __f2fs_bio_alloc(GFP_NOIO | __GFP_NOFAIL, npages); 66 return bio; 67 } 68 if (time_to_inject(sbi, FAULT_ALLOC_BIO)) { 69 f2fs_show_injection_info(sbi, FAULT_ALLOC_BIO); 70 return NULL; 71 } 72 73 return __f2fs_bio_alloc(GFP_KERNEL, npages); 74 } 75 76 static bool __is_cp_guaranteed(struct page *page) 77 { 78 struct address_space *mapping = page->mapping; 79 struct inode *inode; 80 struct f2fs_sb_info *sbi; 81 82 if (!mapping) 83 return false; 84 85 if (f2fs_is_compressed_page(page)) 86 return false; 87 88 inode = mapping->host; 89 sbi = F2FS_I_SB(inode); 90 91 if (inode->i_ino == F2FS_META_INO(sbi) || 92 inode->i_ino == F2FS_NODE_INO(sbi) || 93 S_ISDIR(inode->i_mode) || 94 (S_ISREG(inode->i_mode) && 95 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) || 96 is_cold_data(page)) 97 return true; 98 return false; 99 } 100 101 static enum count_type __read_io_type(struct page *page) 102 { 103 struct address_space *mapping = page_file_mapping(page); 104 105 if (mapping) { 106 struct inode *inode = mapping->host; 107 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 108 109 if (inode->i_ino == F2FS_META_INO(sbi)) 110 return F2FS_RD_META; 111 112 if (inode->i_ino == F2FS_NODE_INO(sbi)) 113 return F2FS_RD_NODE; 114 } 115 return F2FS_RD_DATA; 116 } 117 118 /* postprocessing steps for read bios */ 119 enum bio_post_read_step { 120 STEP_DECRYPT, 121 STEP_DECOMPRESS, 122 STEP_VERITY, 123 }; 124 125 struct bio_post_read_ctx { 126 struct bio *bio; 127 struct f2fs_sb_info *sbi; 128 struct work_struct work; 129 unsigned int enabled_steps; 130 }; 131 132 static void __read_end_io(struct bio *bio, bool compr, bool verity) 133 { 134 struct page *page; 135 struct bio_vec *bv; 136 struct bvec_iter_all iter_all; 137 138 bio_for_each_segment_all(bv, bio, iter_all) { 139 page = bv->bv_page; 140 141 #ifdef CONFIG_F2FS_FS_COMPRESSION 142 if (compr && f2fs_is_compressed_page(page)) { 143 f2fs_decompress_pages(bio, page, verity); 144 continue; 145 } 146 #endif 147 148 /* PG_error was set if any post_read step failed */ 149 if (bio->bi_status || PageError(page)) { 150 ClearPageUptodate(page); 151 /* will re-read again later */ 152 ClearPageError(page); 153 } else { 154 SetPageUptodate(page); 155 } 156 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 157 unlock_page(page); 158 } 159 } 160 161 static void f2fs_release_read_bio(struct bio *bio); 162 static void __f2fs_read_end_io(struct bio *bio, bool compr, bool verity) 163 { 164 if (!compr) 165 __read_end_io(bio, false, verity); 166 f2fs_release_read_bio(bio); 167 } 168 169 static void f2fs_decompress_bio(struct bio *bio, bool verity) 170 { 171 __read_end_io(bio, true, verity); 172 } 173 174 static void bio_post_read_processing(struct bio_post_read_ctx *ctx); 175 176 static void f2fs_decrypt_work(struct bio_post_read_ctx *ctx) 177 { 178 fscrypt_decrypt_bio(ctx->bio); 179 } 180 181 static void f2fs_decompress_work(struct bio_post_read_ctx *ctx) 182 { 183 f2fs_decompress_bio(ctx->bio, ctx->enabled_steps & (1 << STEP_VERITY)); 184 } 185 186 #ifdef CONFIG_F2FS_FS_COMPRESSION 187 static void f2fs_verify_pages(struct page **rpages, unsigned int cluster_size) 188 { 189 f2fs_decompress_end_io(rpages, cluster_size, false, true); 190 } 191 192 static void f2fs_verify_bio(struct bio *bio) 193 { 194 struct page *page = bio_first_page_all(bio); 195 struct decompress_io_ctx *dic = 196 (struct decompress_io_ctx *)page_private(page); 197 198 f2fs_verify_pages(dic->rpages, dic->cluster_size); 199 f2fs_free_dic(dic); 200 } 201 #endif 202 203 static void f2fs_verity_work(struct work_struct *work) 204 { 205 struct bio_post_read_ctx *ctx = 206 container_of(work, struct bio_post_read_ctx, work); 207 struct bio *bio = ctx->bio; 208 #ifdef CONFIG_F2FS_FS_COMPRESSION 209 unsigned int enabled_steps = ctx->enabled_steps; 210 #endif 211 212 /* 213 * fsverity_verify_bio() may call readpages() again, and while verity 214 * will be disabled for this, decryption may still be needed, resulting 215 * in another bio_post_read_ctx being allocated. So to prevent 216 * deadlocks we need to release the current ctx to the mempool first. 217 * This assumes that verity is the last post-read step. 218 */ 219 mempool_free(ctx, bio_post_read_ctx_pool); 220 bio->bi_private = NULL; 221 222 #ifdef CONFIG_F2FS_FS_COMPRESSION 223 /* previous step is decompression */ 224 if (enabled_steps & (1 << STEP_DECOMPRESS)) { 225 f2fs_verify_bio(bio); 226 f2fs_release_read_bio(bio); 227 return; 228 } 229 #endif 230 231 fsverity_verify_bio(bio); 232 __f2fs_read_end_io(bio, false, false); 233 } 234 235 static void f2fs_post_read_work(struct work_struct *work) 236 { 237 struct bio_post_read_ctx *ctx = 238 container_of(work, struct bio_post_read_ctx, work); 239 240 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) 241 f2fs_decrypt_work(ctx); 242 243 if (ctx->enabled_steps & (1 << STEP_DECOMPRESS)) 244 f2fs_decompress_work(ctx); 245 246 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 247 INIT_WORK(&ctx->work, f2fs_verity_work); 248 fsverity_enqueue_verify_work(&ctx->work); 249 return; 250 } 251 252 __f2fs_read_end_io(ctx->bio, 253 ctx->enabled_steps & (1 << STEP_DECOMPRESS), false); 254 } 255 256 static void f2fs_enqueue_post_read_work(struct f2fs_sb_info *sbi, 257 struct work_struct *work) 258 { 259 queue_work(sbi->post_read_wq, work); 260 } 261 262 static void bio_post_read_processing(struct bio_post_read_ctx *ctx) 263 { 264 /* 265 * We use different work queues for decryption and for verity because 266 * verity may require reading metadata pages that need decryption, and 267 * we shouldn't recurse to the same workqueue. 268 */ 269 270 if (ctx->enabled_steps & (1 << STEP_DECRYPT) || 271 ctx->enabled_steps & (1 << STEP_DECOMPRESS)) { 272 INIT_WORK(&ctx->work, f2fs_post_read_work); 273 f2fs_enqueue_post_read_work(ctx->sbi, &ctx->work); 274 return; 275 } 276 277 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 278 INIT_WORK(&ctx->work, f2fs_verity_work); 279 fsverity_enqueue_verify_work(&ctx->work); 280 return; 281 } 282 283 __f2fs_read_end_io(ctx->bio, false, false); 284 } 285 286 static bool f2fs_bio_post_read_required(struct bio *bio) 287 { 288 return bio->bi_private; 289 } 290 291 static void f2fs_read_end_io(struct bio *bio) 292 { 293 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio)); 294 295 if (time_to_inject(sbi, FAULT_READ_IO)) { 296 f2fs_show_injection_info(sbi, FAULT_READ_IO); 297 bio->bi_status = BLK_STS_IOERR; 298 } 299 300 if (f2fs_bio_post_read_required(bio)) { 301 struct bio_post_read_ctx *ctx = bio->bi_private; 302 303 bio_post_read_processing(ctx); 304 return; 305 } 306 307 __f2fs_read_end_io(bio, false, false); 308 } 309 310 static void f2fs_write_end_io(struct bio *bio) 311 { 312 struct f2fs_sb_info *sbi = bio->bi_private; 313 struct bio_vec *bvec; 314 struct bvec_iter_all iter_all; 315 316 if (time_to_inject(sbi, FAULT_WRITE_IO)) { 317 f2fs_show_injection_info(sbi, FAULT_WRITE_IO); 318 bio->bi_status = BLK_STS_IOERR; 319 } 320 321 bio_for_each_segment_all(bvec, bio, iter_all) { 322 struct page *page = bvec->bv_page; 323 enum count_type type = WB_DATA_TYPE(page); 324 325 if (IS_DUMMY_WRITTEN_PAGE(page)) { 326 set_page_private(page, (unsigned long)NULL); 327 ClearPagePrivate(page); 328 unlock_page(page); 329 mempool_free(page, sbi->write_io_dummy); 330 331 if (unlikely(bio->bi_status)) 332 f2fs_stop_checkpoint(sbi, true); 333 continue; 334 } 335 336 fscrypt_finalize_bounce_page(&page); 337 338 #ifdef CONFIG_F2FS_FS_COMPRESSION 339 if (f2fs_is_compressed_page(page)) { 340 f2fs_compress_write_end_io(bio, page); 341 continue; 342 } 343 #endif 344 345 if (unlikely(bio->bi_status)) { 346 mapping_set_error(page->mapping, -EIO); 347 if (type == F2FS_WB_CP_DATA) 348 f2fs_stop_checkpoint(sbi, true); 349 } 350 351 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 352 page->index != nid_of_node(page)); 353 354 dec_page_count(sbi, type); 355 if (f2fs_in_warm_node_list(sbi, page)) 356 f2fs_del_fsync_node_entry(sbi, page); 357 clear_cold_data(page); 358 end_page_writeback(page); 359 } 360 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 361 wq_has_sleeper(&sbi->cp_wait)) 362 wake_up(&sbi->cp_wait); 363 364 bio_put(bio); 365 } 366 367 /* 368 * Return true, if pre_bio's bdev is same as its target device. 369 */ 370 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 371 block_t blk_addr, struct bio *bio) 372 { 373 struct block_device *bdev = sbi->sb->s_bdev; 374 int i; 375 376 if (f2fs_is_multi_device(sbi)) { 377 for (i = 0; i < sbi->s_ndevs; i++) { 378 if (FDEV(i).start_blk <= blk_addr && 379 FDEV(i).end_blk >= blk_addr) { 380 blk_addr -= FDEV(i).start_blk; 381 bdev = FDEV(i).bdev; 382 break; 383 } 384 } 385 } 386 if (bio) { 387 bio_set_dev(bio, bdev); 388 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); 389 } 390 return bdev; 391 } 392 393 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 394 { 395 int i; 396 397 if (!f2fs_is_multi_device(sbi)) 398 return 0; 399 400 for (i = 0; i < sbi->s_ndevs; i++) 401 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 402 return i; 403 return 0; 404 } 405 406 static bool __same_bdev(struct f2fs_sb_info *sbi, 407 block_t blk_addr, struct bio *bio) 408 { 409 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL); 410 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno; 411 } 412 413 /* 414 * Low-level block read/write IO operations. 415 */ 416 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) 417 { 418 struct f2fs_sb_info *sbi = fio->sbi; 419 struct bio *bio; 420 421 bio = f2fs_bio_alloc(sbi, npages, true); 422 423 f2fs_target_device(sbi, fio->new_blkaddr, bio); 424 if (is_read_io(fio->op)) { 425 bio->bi_end_io = f2fs_read_end_io; 426 bio->bi_private = NULL; 427 } else { 428 bio->bi_end_io = f2fs_write_end_io; 429 bio->bi_private = sbi; 430 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, 431 fio->type, fio->temp); 432 } 433 if (fio->io_wbc) 434 wbc_init_bio(fio->io_wbc, bio); 435 436 return bio; 437 } 438 439 static inline void __submit_bio(struct f2fs_sb_info *sbi, 440 struct bio *bio, enum page_type type) 441 { 442 if (!is_read_io(bio_op(bio))) { 443 unsigned int start; 444 445 if (type != DATA && type != NODE) 446 goto submit_io; 447 448 if (test_opt(sbi, LFS) && current->plug) 449 blk_finish_plug(current->plug); 450 451 if (F2FS_IO_ALIGNED(sbi)) 452 goto submit_io; 453 454 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS; 455 start %= F2FS_IO_SIZE(sbi); 456 457 if (start == 0) 458 goto submit_io; 459 460 /* fill dummy pages */ 461 for (; start < F2FS_IO_SIZE(sbi); start++) { 462 struct page *page = 463 mempool_alloc(sbi->write_io_dummy, 464 GFP_NOIO | __GFP_NOFAIL); 465 f2fs_bug_on(sbi, !page); 466 467 zero_user_segment(page, 0, PAGE_SIZE); 468 SetPagePrivate(page); 469 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE); 470 lock_page(page); 471 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 472 f2fs_bug_on(sbi, 1); 473 } 474 /* 475 * In the NODE case, we lose next block address chain. So, we 476 * need to do checkpoint in f2fs_sync_file. 477 */ 478 if (type == NODE) 479 set_sbi_flag(sbi, SBI_NEED_CP); 480 } 481 submit_io: 482 if (is_read_io(bio_op(bio))) 483 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 484 else 485 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 486 submit_bio(bio); 487 } 488 489 void f2fs_submit_bio(struct f2fs_sb_info *sbi, 490 struct bio *bio, enum page_type type) 491 { 492 __submit_bio(sbi, bio, type); 493 } 494 495 static void __submit_merged_bio(struct f2fs_bio_info *io) 496 { 497 struct f2fs_io_info *fio = &io->fio; 498 499 if (!io->bio) 500 return; 501 502 bio_set_op_attrs(io->bio, fio->op, fio->op_flags); 503 504 if (is_read_io(fio->op)) 505 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 506 else 507 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 508 509 __submit_bio(io->sbi, io->bio, fio->type); 510 io->bio = NULL; 511 } 512 513 static bool __has_merged_page(struct bio *bio, struct inode *inode, 514 struct page *page, nid_t ino) 515 { 516 struct bio_vec *bvec; 517 struct bvec_iter_all iter_all; 518 519 if (!bio) 520 return false; 521 522 if (!inode && !page && !ino) 523 return true; 524 525 bio_for_each_segment_all(bvec, bio, iter_all) { 526 struct page *target = bvec->bv_page; 527 528 if (fscrypt_is_bounce_page(target)) { 529 target = fscrypt_pagecache_page(target); 530 if (IS_ERR(target)) 531 continue; 532 } 533 if (f2fs_is_compressed_page(target)) { 534 target = f2fs_compress_control_page(target); 535 if (IS_ERR(target)) 536 continue; 537 } 538 539 if (inode && inode == target->mapping->host) 540 return true; 541 if (page && page == target) 542 return true; 543 if (ino && ino == ino_of_node(target)) 544 return true; 545 } 546 547 return false; 548 } 549 550 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 551 enum page_type type, enum temp_type temp) 552 { 553 enum page_type btype = PAGE_TYPE_OF_BIO(type); 554 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 555 556 down_write(&io->io_rwsem); 557 558 /* change META to META_FLUSH in the checkpoint procedure */ 559 if (type >= META_FLUSH) { 560 io->fio.type = META_FLUSH; 561 io->fio.op = REQ_OP_WRITE; 562 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC; 563 if (!test_opt(sbi, NOBARRIER)) 564 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 565 } 566 __submit_merged_bio(io); 567 up_write(&io->io_rwsem); 568 } 569 570 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 571 struct inode *inode, struct page *page, 572 nid_t ino, enum page_type type, bool force) 573 { 574 enum temp_type temp; 575 bool ret = true; 576 577 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 578 if (!force) { 579 enum page_type btype = PAGE_TYPE_OF_BIO(type); 580 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 581 582 down_read(&io->io_rwsem); 583 ret = __has_merged_page(io->bio, inode, page, ino); 584 up_read(&io->io_rwsem); 585 } 586 if (ret) 587 __f2fs_submit_merged_write(sbi, type, temp); 588 589 /* TODO: use HOT temp only for meta pages now. */ 590 if (type >= META) 591 break; 592 } 593 } 594 595 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 596 { 597 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true); 598 } 599 600 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 601 struct inode *inode, struct page *page, 602 nid_t ino, enum page_type type) 603 { 604 __submit_merged_write_cond(sbi, inode, page, ino, type, false); 605 } 606 607 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 608 { 609 f2fs_submit_merged_write(sbi, DATA); 610 f2fs_submit_merged_write(sbi, NODE); 611 f2fs_submit_merged_write(sbi, META); 612 } 613 614 /* 615 * Fill the locked page with data located in the block address. 616 * A caller needs to unlock the page on failure. 617 */ 618 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 619 { 620 struct bio *bio; 621 struct page *page = fio->encrypted_page ? 622 fio->encrypted_page : fio->page; 623 624 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 625 fio->is_por ? META_POR : (__is_meta_io(fio) ? 626 META_GENERIC : DATA_GENERIC_ENHANCE))) 627 return -EFSCORRUPTED; 628 629 trace_f2fs_submit_page_bio(page, fio); 630 f2fs_trace_ios(fio, 0); 631 632 /* Allocate a new bio */ 633 bio = __bio_alloc(fio, 1); 634 635 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 636 bio_put(bio); 637 return -EFAULT; 638 } 639 640 if (fio->io_wbc && !is_read_io(fio->op)) 641 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 642 643 bio_set_op_attrs(bio, fio->op, fio->op_flags); 644 645 inc_page_count(fio->sbi, is_read_io(fio->op) ? 646 __read_io_type(page): WB_DATA_TYPE(fio->page)); 647 648 __submit_bio(fio->sbi, bio, fio->type); 649 return 0; 650 } 651 652 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 653 block_t last_blkaddr, block_t cur_blkaddr) 654 { 655 if (last_blkaddr + 1 != cur_blkaddr) 656 return false; 657 return __same_bdev(sbi, cur_blkaddr, bio); 658 } 659 660 static bool io_type_is_mergeable(struct f2fs_bio_info *io, 661 struct f2fs_io_info *fio) 662 { 663 if (io->fio.op != fio->op) 664 return false; 665 return io->fio.op_flags == fio->op_flags; 666 } 667 668 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 669 struct f2fs_bio_info *io, 670 struct f2fs_io_info *fio, 671 block_t last_blkaddr, 672 block_t cur_blkaddr) 673 { 674 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) { 675 unsigned int filled_blocks = 676 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size); 677 unsigned int io_size = F2FS_IO_SIZE(sbi); 678 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt; 679 680 /* IOs in bio is aligned and left space of vectors is not enough */ 681 if (!(filled_blocks % io_size) && left_vecs < io_size) 682 return false; 683 } 684 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr)) 685 return false; 686 return io_type_is_mergeable(io, fio); 687 } 688 689 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio, 690 struct page *page, enum temp_type temp) 691 { 692 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 693 struct bio_entry *be; 694 695 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS); 696 be->bio = bio; 697 bio_get(bio); 698 699 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE) 700 f2fs_bug_on(sbi, 1); 701 702 down_write(&io->bio_list_lock); 703 list_add_tail(&be->list, &io->bio_list); 704 up_write(&io->bio_list_lock); 705 } 706 707 static void del_bio_entry(struct bio_entry *be) 708 { 709 list_del(&be->list); 710 kmem_cache_free(bio_entry_slab, be); 711 } 712 713 static int add_ipu_page(struct f2fs_sb_info *sbi, struct bio **bio, 714 struct page *page) 715 { 716 enum temp_type temp; 717 bool found = false; 718 int ret = -EAGAIN; 719 720 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 721 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 722 struct list_head *head = &io->bio_list; 723 struct bio_entry *be; 724 725 down_write(&io->bio_list_lock); 726 list_for_each_entry(be, head, list) { 727 if (be->bio != *bio) 728 continue; 729 730 found = true; 731 732 if (bio_add_page(*bio, page, PAGE_SIZE, 0) == 733 PAGE_SIZE) { 734 ret = 0; 735 break; 736 } 737 738 /* bio is full */ 739 del_bio_entry(be); 740 __submit_bio(sbi, *bio, DATA); 741 break; 742 } 743 up_write(&io->bio_list_lock); 744 } 745 746 if (ret) { 747 bio_put(*bio); 748 *bio = NULL; 749 } 750 751 return ret; 752 } 753 754 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi, 755 struct bio **bio, struct page *page) 756 { 757 enum temp_type temp; 758 bool found = false; 759 struct bio *target = bio ? *bio : NULL; 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 if (list_empty(head)) 767 continue; 768 769 down_read(&io->bio_list_lock); 770 list_for_each_entry(be, head, list) { 771 if (target) 772 found = (target == be->bio); 773 else 774 found = __has_merged_page(be->bio, NULL, 775 page, 0); 776 if (found) 777 break; 778 } 779 up_read(&io->bio_list_lock); 780 781 if (!found) 782 continue; 783 784 found = false; 785 786 down_write(&io->bio_list_lock); 787 list_for_each_entry(be, head, list) { 788 if (target) 789 found = (target == be->bio); 790 else 791 found = __has_merged_page(be->bio, NULL, 792 page, 0); 793 if (found) { 794 target = be->bio; 795 del_bio_entry(be); 796 break; 797 } 798 } 799 up_write(&io->bio_list_lock); 800 } 801 802 if (found) 803 __submit_bio(sbi, target, DATA); 804 if (bio && *bio) { 805 bio_put(*bio); 806 *bio = NULL; 807 } 808 } 809 810 int f2fs_merge_page_bio(struct f2fs_io_info *fio) 811 { 812 struct bio *bio = *fio->bio; 813 struct page *page = fio->encrypted_page ? 814 fio->encrypted_page : fio->page; 815 816 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 817 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) 818 return -EFSCORRUPTED; 819 820 trace_f2fs_submit_page_bio(page, fio); 821 f2fs_trace_ios(fio, 0); 822 823 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block, 824 fio->new_blkaddr)) 825 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL); 826 alloc_new: 827 if (!bio) { 828 bio = __bio_alloc(fio, BIO_MAX_PAGES); 829 bio_set_op_attrs(bio, fio->op, fio->op_flags); 830 831 add_bio_entry(fio->sbi, bio, page, fio->temp); 832 } else { 833 if (add_ipu_page(fio->sbi, &bio, page)) 834 goto alloc_new; 835 } 836 837 if (fio->io_wbc) 838 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 839 840 inc_page_count(fio->sbi, WB_DATA_TYPE(page)); 841 842 *fio->last_block = fio->new_blkaddr; 843 *fio->bio = bio; 844 845 return 0; 846 } 847 848 void f2fs_submit_page_write(struct f2fs_io_info *fio) 849 { 850 struct f2fs_sb_info *sbi = fio->sbi; 851 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 852 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 853 struct page *bio_page; 854 855 f2fs_bug_on(sbi, is_read_io(fio->op)); 856 857 down_write(&io->io_rwsem); 858 next: 859 if (fio->in_list) { 860 spin_lock(&io->io_lock); 861 if (list_empty(&io->io_list)) { 862 spin_unlock(&io->io_lock); 863 goto out; 864 } 865 fio = list_first_entry(&io->io_list, 866 struct f2fs_io_info, list); 867 list_del(&fio->list); 868 spin_unlock(&io->io_lock); 869 } 870 871 verify_fio_blkaddr(fio); 872 873 if (fio->encrypted_page) 874 bio_page = fio->encrypted_page; 875 else if (fio->compressed_page) 876 bio_page = fio->compressed_page; 877 else 878 bio_page = fio->page; 879 880 /* set submitted = true as a return value */ 881 fio->submitted = true; 882 883 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 884 885 if (io->bio && !io_is_mergeable(sbi, io->bio, io, fio, 886 io->last_block_in_bio, fio->new_blkaddr)) 887 __submit_merged_bio(io); 888 alloc_new: 889 if (io->bio == NULL) { 890 if (F2FS_IO_ALIGNED(sbi) && 891 (fio->type == DATA || fio->type == NODE) && 892 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 893 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 894 fio->retry = true; 895 goto skip; 896 } 897 io->bio = __bio_alloc(fio, BIO_MAX_PAGES); 898 io->fio = *fio; 899 } 900 901 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 902 __submit_merged_bio(io); 903 goto alloc_new; 904 } 905 906 if (fio->io_wbc) 907 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE); 908 909 io->last_block_in_bio = fio->new_blkaddr; 910 f2fs_trace_ios(fio, 0); 911 912 trace_f2fs_submit_page_write(fio->page, fio); 913 skip: 914 if (fio->in_list) 915 goto next; 916 out: 917 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 918 !f2fs_is_checkpoint_ready(sbi)) 919 __submit_merged_bio(io); 920 up_write(&io->io_rwsem); 921 } 922 923 static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx) 924 { 925 return fsverity_active(inode) && 926 idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 927 } 928 929 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 930 unsigned nr_pages, unsigned op_flag, 931 pgoff_t first_idx) 932 { 933 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 934 struct bio *bio; 935 struct bio_post_read_ctx *ctx; 936 unsigned int post_read_steps = 0; 937 938 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false); 939 if (!bio) 940 return ERR_PTR(-ENOMEM); 941 f2fs_target_device(sbi, blkaddr, bio); 942 bio->bi_end_io = f2fs_read_end_io; 943 bio_set_op_attrs(bio, REQ_OP_READ, op_flag); 944 945 if (f2fs_encrypted_file(inode)) 946 post_read_steps |= 1 << STEP_DECRYPT; 947 if (f2fs_compressed_file(inode)) 948 post_read_steps |= 1 << STEP_DECOMPRESS; 949 if (f2fs_need_verity(inode, first_idx)) 950 post_read_steps |= 1 << STEP_VERITY; 951 952 if (post_read_steps) { 953 /* Due to the mempool, this never fails. */ 954 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 955 ctx->bio = bio; 956 ctx->sbi = sbi; 957 ctx->enabled_steps = post_read_steps; 958 bio->bi_private = ctx; 959 } 960 961 return bio; 962 } 963 964 static void f2fs_release_read_bio(struct bio *bio) 965 { 966 if (bio->bi_private) 967 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 968 bio_put(bio); 969 } 970 971 /* This can handle encryption stuffs */ 972 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 973 block_t blkaddr) 974 { 975 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 976 struct bio *bio; 977 978 bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0, page->index); 979 if (IS_ERR(bio)) 980 return PTR_ERR(bio); 981 982 /* wait for GCed page writeback via META_MAPPING */ 983 f2fs_wait_on_block_writeback(inode, blkaddr); 984 985 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 986 bio_put(bio); 987 return -EFAULT; 988 } 989 ClearPageError(page); 990 inc_page_count(sbi, F2FS_RD_DATA); 991 __submit_bio(sbi, bio, DATA); 992 return 0; 993 } 994 995 static void __set_data_blkaddr(struct dnode_of_data *dn) 996 { 997 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 998 __le32 *addr_array; 999 int base = 0; 1000 1001 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 1002 base = get_extra_isize(dn->inode); 1003 1004 /* Get physical address of data block */ 1005 addr_array = blkaddr_in_node(rn); 1006 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 1007 } 1008 1009 /* 1010 * Lock ordering for the change of data block address: 1011 * ->data_page 1012 * ->node_page 1013 * update block addresses in the node page 1014 */ 1015 void f2fs_set_data_blkaddr(struct dnode_of_data *dn) 1016 { 1017 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1018 __set_data_blkaddr(dn); 1019 if (set_page_dirty(dn->node_page)) 1020 dn->node_changed = true; 1021 } 1022 1023 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1024 { 1025 dn->data_blkaddr = blkaddr; 1026 f2fs_set_data_blkaddr(dn); 1027 f2fs_update_extent_cache(dn); 1028 } 1029 1030 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 1031 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 1032 { 1033 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1034 int err; 1035 1036 if (!count) 1037 return 0; 1038 1039 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1040 return -EPERM; 1041 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1042 return err; 1043 1044 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 1045 dn->ofs_in_node, count); 1046 1047 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1048 1049 for (; count > 0; dn->ofs_in_node++) { 1050 block_t blkaddr = datablock_addr(dn->inode, 1051 dn->node_page, dn->ofs_in_node); 1052 if (blkaddr == NULL_ADDR) { 1053 dn->data_blkaddr = NEW_ADDR; 1054 __set_data_blkaddr(dn); 1055 count--; 1056 } 1057 } 1058 1059 if (set_page_dirty(dn->node_page)) 1060 dn->node_changed = true; 1061 return 0; 1062 } 1063 1064 /* Should keep dn->ofs_in_node unchanged */ 1065 int f2fs_reserve_new_block(struct dnode_of_data *dn) 1066 { 1067 unsigned int ofs_in_node = dn->ofs_in_node; 1068 int ret; 1069 1070 ret = f2fs_reserve_new_blocks(dn, 1); 1071 dn->ofs_in_node = ofs_in_node; 1072 return ret; 1073 } 1074 1075 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 1076 { 1077 bool need_put = dn->inode_page ? false : true; 1078 int err; 1079 1080 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE); 1081 if (err) 1082 return err; 1083 1084 if (dn->data_blkaddr == NULL_ADDR) 1085 err = f2fs_reserve_new_block(dn); 1086 if (err || need_put) 1087 f2fs_put_dnode(dn); 1088 return err; 1089 } 1090 1091 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 1092 { 1093 struct extent_info ei = {0,0,0}; 1094 struct inode *inode = dn->inode; 1095 1096 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1097 dn->data_blkaddr = ei.blk + index - ei.fofs; 1098 return 0; 1099 } 1100 1101 return f2fs_reserve_block(dn, index); 1102 } 1103 1104 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, 1105 int op_flags, bool for_write) 1106 { 1107 struct address_space *mapping = inode->i_mapping; 1108 struct dnode_of_data dn; 1109 struct page *page; 1110 struct extent_info ei = {0,0,0}; 1111 int err; 1112 1113 page = f2fs_grab_cache_page(mapping, index, for_write); 1114 if (!page) 1115 return ERR_PTR(-ENOMEM); 1116 1117 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1118 dn.data_blkaddr = ei.blk + index - ei.fofs; 1119 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr, 1120 DATA_GENERIC_ENHANCE_READ)) { 1121 err = -EFSCORRUPTED; 1122 goto put_err; 1123 } 1124 goto got_it; 1125 } 1126 1127 set_new_dnode(&dn, inode, NULL, NULL, 0); 1128 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1129 if (err) 1130 goto put_err; 1131 f2fs_put_dnode(&dn); 1132 1133 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1134 err = -ENOENT; 1135 goto put_err; 1136 } 1137 if (dn.data_blkaddr != NEW_ADDR && 1138 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode), 1139 dn.data_blkaddr, 1140 DATA_GENERIC_ENHANCE)) { 1141 err = -EFSCORRUPTED; 1142 goto put_err; 1143 } 1144 got_it: 1145 if (PageUptodate(page)) { 1146 unlock_page(page); 1147 return page; 1148 } 1149 1150 /* 1151 * A new dentry page is allocated but not able to be written, since its 1152 * new inode page couldn't be allocated due to -ENOSPC. 1153 * In such the case, its blkaddr can be remained as NEW_ADDR. 1154 * see, f2fs_add_link -> f2fs_get_new_data_page -> 1155 * f2fs_init_inode_metadata. 1156 */ 1157 if (dn.data_blkaddr == NEW_ADDR) { 1158 zero_user_segment(page, 0, PAGE_SIZE); 1159 if (!PageUptodate(page)) 1160 SetPageUptodate(page); 1161 unlock_page(page); 1162 return page; 1163 } 1164 1165 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr); 1166 if (err) 1167 goto put_err; 1168 return page; 1169 1170 put_err: 1171 f2fs_put_page(page, 1); 1172 return ERR_PTR(err); 1173 } 1174 1175 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index) 1176 { 1177 struct address_space *mapping = inode->i_mapping; 1178 struct page *page; 1179 1180 page = find_get_page(mapping, index); 1181 if (page && PageUptodate(page)) 1182 return page; 1183 f2fs_put_page(page, 0); 1184 1185 page = f2fs_get_read_data_page(inode, index, 0, false); 1186 if (IS_ERR(page)) 1187 return page; 1188 1189 if (PageUptodate(page)) 1190 return page; 1191 1192 wait_on_page_locked(page); 1193 if (unlikely(!PageUptodate(page))) { 1194 f2fs_put_page(page, 0); 1195 return ERR_PTR(-EIO); 1196 } 1197 return page; 1198 } 1199 1200 /* 1201 * If it tries to access a hole, return an error. 1202 * Because, the callers, functions in dir.c and GC, should be able to know 1203 * whether this page exists or not. 1204 */ 1205 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, 1206 bool for_write) 1207 { 1208 struct address_space *mapping = inode->i_mapping; 1209 struct page *page; 1210 repeat: 1211 page = f2fs_get_read_data_page(inode, index, 0, for_write); 1212 if (IS_ERR(page)) 1213 return page; 1214 1215 /* wait for read completion */ 1216 lock_page(page); 1217 if (unlikely(page->mapping != mapping)) { 1218 f2fs_put_page(page, 1); 1219 goto repeat; 1220 } 1221 if (unlikely(!PageUptodate(page))) { 1222 f2fs_put_page(page, 1); 1223 return ERR_PTR(-EIO); 1224 } 1225 return page; 1226 } 1227 1228 /* 1229 * Caller ensures that this data page is never allocated. 1230 * A new zero-filled data page is allocated in the page cache. 1231 * 1232 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 1233 * f2fs_unlock_op(). 1234 * Note that, ipage is set only by make_empty_dir, and if any error occur, 1235 * ipage should be released by this function. 1236 */ 1237 struct page *f2fs_get_new_data_page(struct inode *inode, 1238 struct page *ipage, pgoff_t index, bool new_i_size) 1239 { 1240 struct address_space *mapping = inode->i_mapping; 1241 struct page *page; 1242 struct dnode_of_data dn; 1243 int err; 1244 1245 page = f2fs_grab_cache_page(mapping, index, true); 1246 if (!page) { 1247 /* 1248 * before exiting, we should make sure ipage will be released 1249 * if any error occur. 1250 */ 1251 f2fs_put_page(ipage, 1); 1252 return ERR_PTR(-ENOMEM); 1253 } 1254 1255 set_new_dnode(&dn, inode, ipage, NULL, 0); 1256 err = f2fs_reserve_block(&dn, index); 1257 if (err) { 1258 f2fs_put_page(page, 1); 1259 return ERR_PTR(err); 1260 } 1261 if (!ipage) 1262 f2fs_put_dnode(&dn); 1263 1264 if (PageUptodate(page)) 1265 goto got_it; 1266 1267 if (dn.data_blkaddr == NEW_ADDR) { 1268 zero_user_segment(page, 0, PAGE_SIZE); 1269 if (!PageUptodate(page)) 1270 SetPageUptodate(page); 1271 } else { 1272 f2fs_put_page(page, 1); 1273 1274 /* if ipage exists, blkaddr should be NEW_ADDR */ 1275 f2fs_bug_on(F2FS_I_SB(inode), ipage); 1276 page = f2fs_get_lock_data_page(inode, index, true); 1277 if (IS_ERR(page)) 1278 return page; 1279 } 1280 got_it: 1281 if (new_i_size && i_size_read(inode) < 1282 ((loff_t)(index + 1) << PAGE_SHIFT)) 1283 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 1284 return page; 1285 } 1286 1287 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 1288 { 1289 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1290 struct f2fs_summary sum; 1291 struct node_info ni; 1292 block_t old_blkaddr; 1293 blkcnt_t count = 1; 1294 int err; 1295 1296 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1297 return -EPERM; 1298 1299 err = f2fs_get_node_info(sbi, dn->nid, &ni); 1300 if (err) 1301 return err; 1302 1303 dn->data_blkaddr = datablock_addr(dn->inode, 1304 dn->node_page, dn->ofs_in_node); 1305 if (dn->data_blkaddr != NULL_ADDR) 1306 goto alloc; 1307 1308 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1309 return err; 1310 1311 alloc: 1312 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1313 old_blkaddr = dn->data_blkaddr; 1314 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr, 1315 &sum, seg_type, NULL, false); 1316 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 1317 invalidate_mapping_pages(META_MAPPING(sbi), 1318 old_blkaddr, old_blkaddr); 1319 f2fs_update_data_blkaddr(dn, dn->data_blkaddr); 1320 1321 /* 1322 * i_size will be updated by direct_IO. Otherwise, we'll get stale 1323 * data from unwritten block via dio_read. 1324 */ 1325 return 0; 1326 } 1327 1328 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from) 1329 { 1330 struct inode *inode = file_inode(iocb->ki_filp); 1331 struct f2fs_map_blocks map; 1332 int flag; 1333 int err = 0; 1334 bool direct_io = iocb->ki_flags & IOCB_DIRECT; 1335 1336 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos); 1337 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from)); 1338 if (map.m_len > map.m_lblk) 1339 map.m_len -= map.m_lblk; 1340 else 1341 map.m_len = 0; 1342 1343 map.m_next_pgofs = NULL; 1344 map.m_next_extent = NULL; 1345 map.m_seg_type = NO_CHECK_TYPE; 1346 map.m_may_create = true; 1347 1348 if (direct_io) { 1349 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint); 1350 flag = f2fs_force_buffered_io(inode, iocb, from) ? 1351 F2FS_GET_BLOCK_PRE_AIO : 1352 F2FS_GET_BLOCK_PRE_DIO; 1353 goto map_blocks; 1354 } 1355 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) { 1356 err = f2fs_convert_inline_inode(inode); 1357 if (err) 1358 return err; 1359 } 1360 if (f2fs_has_inline_data(inode)) 1361 return err; 1362 1363 flag = F2FS_GET_BLOCK_PRE_AIO; 1364 1365 map_blocks: 1366 err = f2fs_map_blocks(inode, &map, 1, flag); 1367 if (map.m_len > 0 && err == -ENOSPC) { 1368 if (!direct_io) 1369 set_inode_flag(inode, FI_NO_PREALLOC); 1370 err = 0; 1371 } 1372 return err; 1373 } 1374 1375 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 1376 { 1377 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1378 if (lock) 1379 down_read(&sbi->node_change); 1380 else 1381 up_read(&sbi->node_change); 1382 } else { 1383 if (lock) 1384 f2fs_lock_op(sbi); 1385 else 1386 f2fs_unlock_op(sbi); 1387 } 1388 } 1389 1390 /* 1391 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with 1392 * f2fs_map_blocks structure. 1393 * If original data blocks are allocated, then give them to blockdev. 1394 * Otherwise, 1395 * a. preallocate requested block addresses 1396 * b. do not use extent cache for better performance 1397 * c. give the block addresses to blockdev 1398 */ 1399 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 1400 int create, int flag) 1401 { 1402 unsigned int maxblocks = map->m_len; 1403 struct dnode_of_data dn; 1404 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1405 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE; 1406 pgoff_t pgofs, end_offset, end; 1407 int err = 0, ofs = 1; 1408 unsigned int ofs_in_node, last_ofs_in_node; 1409 blkcnt_t prealloc; 1410 struct extent_info ei = {0,0,0}; 1411 block_t blkaddr; 1412 unsigned int start_pgofs; 1413 1414 if (!maxblocks) 1415 return 0; 1416 1417 map->m_len = 0; 1418 map->m_flags = 0; 1419 1420 /* it only supports block size == page size */ 1421 pgofs = (pgoff_t)map->m_lblk; 1422 end = pgofs + maxblocks; 1423 1424 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 1425 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO && 1426 map->m_may_create) 1427 goto next_dnode; 1428 1429 map->m_pblk = ei.blk + pgofs - ei.fofs; 1430 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 1431 map->m_flags = F2FS_MAP_MAPPED; 1432 if (map->m_next_extent) 1433 *map->m_next_extent = pgofs + map->m_len; 1434 1435 /* for hardware encryption, but to avoid potential issue in future */ 1436 if (flag == F2FS_GET_BLOCK_DIO) 1437 f2fs_wait_on_block_writeback_range(inode, 1438 map->m_pblk, map->m_len); 1439 goto out; 1440 } 1441 1442 next_dnode: 1443 if (map->m_may_create) 1444 __do_map_lock(sbi, flag, true); 1445 1446 /* When reading holes, we need its node page */ 1447 set_new_dnode(&dn, inode, NULL, NULL, 0); 1448 err = f2fs_get_dnode_of_data(&dn, pgofs, mode); 1449 if (err) { 1450 if (flag == F2FS_GET_BLOCK_BMAP) 1451 map->m_pblk = 0; 1452 if (err == -ENOENT) { 1453 err = 0; 1454 if (map->m_next_pgofs) 1455 *map->m_next_pgofs = 1456 f2fs_get_next_page_offset(&dn, pgofs); 1457 if (map->m_next_extent) 1458 *map->m_next_extent = 1459 f2fs_get_next_page_offset(&dn, pgofs); 1460 } 1461 goto unlock_out; 1462 } 1463 1464 start_pgofs = pgofs; 1465 prealloc = 0; 1466 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 1467 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 1468 1469 next_block: 1470 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node); 1471 1472 if (__is_valid_data_blkaddr(blkaddr) && 1473 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) { 1474 err = -EFSCORRUPTED; 1475 goto sync_out; 1476 } 1477 1478 if (__is_valid_data_blkaddr(blkaddr)) { 1479 /* use out-place-update for driect IO under LFS mode */ 1480 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO && 1481 map->m_may_create) { 1482 err = __allocate_data_block(&dn, map->m_seg_type); 1483 if (err) 1484 goto sync_out; 1485 blkaddr = dn.data_blkaddr; 1486 set_inode_flag(inode, FI_APPEND_WRITE); 1487 } 1488 } else { 1489 if (create) { 1490 if (unlikely(f2fs_cp_error(sbi))) { 1491 err = -EIO; 1492 goto sync_out; 1493 } 1494 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1495 if (blkaddr == NULL_ADDR) { 1496 prealloc++; 1497 last_ofs_in_node = dn.ofs_in_node; 1498 } 1499 } else { 1500 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO && 1501 flag != F2FS_GET_BLOCK_DIO); 1502 err = __allocate_data_block(&dn, 1503 map->m_seg_type); 1504 if (!err) 1505 set_inode_flag(inode, FI_APPEND_WRITE); 1506 } 1507 if (err) 1508 goto sync_out; 1509 map->m_flags |= F2FS_MAP_NEW; 1510 blkaddr = dn.data_blkaddr; 1511 } else { 1512 if (flag == F2FS_GET_BLOCK_BMAP) { 1513 map->m_pblk = 0; 1514 goto sync_out; 1515 } 1516 if (flag == F2FS_GET_BLOCK_PRECACHE) 1517 goto sync_out; 1518 if (flag == F2FS_GET_BLOCK_FIEMAP && 1519 blkaddr == NULL_ADDR) { 1520 if (map->m_next_pgofs) 1521 *map->m_next_pgofs = pgofs + 1; 1522 goto sync_out; 1523 } 1524 if (flag != F2FS_GET_BLOCK_FIEMAP) { 1525 /* for defragment case */ 1526 if (map->m_next_pgofs) 1527 *map->m_next_pgofs = pgofs + 1; 1528 goto sync_out; 1529 } 1530 } 1531 } 1532 1533 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1534 goto skip; 1535 1536 if (map->m_len == 0) { 1537 /* preallocated unwritten block should be mapped for fiemap. */ 1538 if (blkaddr == NEW_ADDR) 1539 map->m_flags |= F2FS_MAP_UNWRITTEN; 1540 map->m_flags |= F2FS_MAP_MAPPED; 1541 1542 map->m_pblk = blkaddr; 1543 map->m_len = 1; 1544 } else if ((map->m_pblk != NEW_ADDR && 1545 blkaddr == (map->m_pblk + ofs)) || 1546 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1547 flag == F2FS_GET_BLOCK_PRE_DIO) { 1548 ofs++; 1549 map->m_len++; 1550 } else { 1551 goto sync_out; 1552 } 1553 1554 skip: 1555 dn.ofs_in_node++; 1556 pgofs++; 1557 1558 /* preallocate blocks in batch for one dnode page */ 1559 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1560 (pgofs == end || dn.ofs_in_node == end_offset)) { 1561 1562 dn.ofs_in_node = ofs_in_node; 1563 err = f2fs_reserve_new_blocks(&dn, prealloc); 1564 if (err) 1565 goto sync_out; 1566 1567 map->m_len += dn.ofs_in_node - ofs_in_node; 1568 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1569 err = -ENOSPC; 1570 goto sync_out; 1571 } 1572 dn.ofs_in_node = end_offset; 1573 } 1574 1575 if (pgofs >= end) 1576 goto sync_out; 1577 else if (dn.ofs_in_node < end_offset) 1578 goto next_block; 1579 1580 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1581 if (map->m_flags & F2FS_MAP_MAPPED) { 1582 unsigned int ofs = start_pgofs - map->m_lblk; 1583 1584 f2fs_update_extent_cache_range(&dn, 1585 start_pgofs, map->m_pblk + ofs, 1586 map->m_len - ofs); 1587 } 1588 } 1589 1590 f2fs_put_dnode(&dn); 1591 1592 if (map->m_may_create) { 1593 __do_map_lock(sbi, flag, false); 1594 f2fs_balance_fs(sbi, dn.node_changed); 1595 } 1596 goto next_dnode; 1597 1598 sync_out: 1599 1600 /* for hardware encryption, but to avoid potential issue in future */ 1601 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) 1602 f2fs_wait_on_block_writeback_range(inode, 1603 map->m_pblk, map->m_len); 1604 1605 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1606 if (map->m_flags & F2FS_MAP_MAPPED) { 1607 unsigned int ofs = start_pgofs - map->m_lblk; 1608 1609 f2fs_update_extent_cache_range(&dn, 1610 start_pgofs, map->m_pblk + ofs, 1611 map->m_len - ofs); 1612 } 1613 if (map->m_next_extent) 1614 *map->m_next_extent = pgofs + 1; 1615 } 1616 f2fs_put_dnode(&dn); 1617 unlock_out: 1618 if (map->m_may_create) { 1619 __do_map_lock(sbi, flag, false); 1620 f2fs_balance_fs(sbi, dn.node_changed); 1621 } 1622 out: 1623 trace_f2fs_map_blocks(inode, map, err); 1624 return err; 1625 } 1626 1627 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) 1628 { 1629 struct f2fs_map_blocks map; 1630 block_t last_lblk; 1631 int err; 1632 1633 if (pos + len > i_size_read(inode)) 1634 return false; 1635 1636 map.m_lblk = F2FS_BYTES_TO_BLK(pos); 1637 map.m_next_pgofs = NULL; 1638 map.m_next_extent = NULL; 1639 map.m_seg_type = NO_CHECK_TYPE; 1640 map.m_may_create = false; 1641 last_lblk = F2FS_BLK_ALIGN(pos + len); 1642 1643 while (map.m_lblk < last_lblk) { 1644 map.m_len = last_lblk - map.m_lblk; 1645 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT); 1646 if (err || map.m_len == 0) 1647 return false; 1648 map.m_lblk += map.m_len; 1649 } 1650 return true; 1651 } 1652 1653 static int __get_data_block(struct inode *inode, sector_t iblock, 1654 struct buffer_head *bh, int create, int flag, 1655 pgoff_t *next_pgofs, int seg_type, bool may_write) 1656 { 1657 struct f2fs_map_blocks map; 1658 int err; 1659 1660 map.m_lblk = iblock; 1661 map.m_len = bh->b_size >> inode->i_blkbits; 1662 map.m_next_pgofs = next_pgofs; 1663 map.m_next_extent = NULL; 1664 map.m_seg_type = seg_type; 1665 map.m_may_create = may_write; 1666 1667 err = f2fs_map_blocks(inode, &map, create, flag); 1668 if (!err) { 1669 map_bh(bh, inode->i_sb, map.m_pblk); 1670 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags; 1671 bh->b_size = (u64)map.m_len << inode->i_blkbits; 1672 } 1673 return err; 1674 } 1675 1676 static int get_data_block(struct inode *inode, sector_t iblock, 1677 struct buffer_head *bh_result, int create, int flag, 1678 pgoff_t *next_pgofs) 1679 { 1680 return __get_data_block(inode, iblock, bh_result, create, 1681 flag, next_pgofs, 1682 NO_CHECK_TYPE, create); 1683 } 1684 1685 static int get_data_block_dio_write(struct inode *inode, sector_t iblock, 1686 struct buffer_head *bh_result, int create) 1687 { 1688 return __get_data_block(inode, iblock, bh_result, create, 1689 F2FS_GET_BLOCK_DIO, NULL, 1690 f2fs_rw_hint_to_seg_type(inode->i_write_hint), 1691 IS_SWAPFILE(inode) ? false : true); 1692 } 1693 1694 static int get_data_block_dio(struct inode *inode, sector_t iblock, 1695 struct buffer_head *bh_result, int create) 1696 { 1697 return __get_data_block(inode, iblock, bh_result, create, 1698 F2FS_GET_BLOCK_DIO, NULL, 1699 f2fs_rw_hint_to_seg_type(inode->i_write_hint), 1700 false); 1701 } 1702 1703 static int get_data_block_bmap(struct inode *inode, sector_t iblock, 1704 struct buffer_head *bh_result, int create) 1705 { 1706 /* Block number less than F2FS MAX BLOCKS */ 1707 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks)) 1708 return -EFBIG; 1709 1710 return __get_data_block(inode, iblock, bh_result, create, 1711 F2FS_GET_BLOCK_BMAP, NULL, 1712 NO_CHECK_TYPE, create); 1713 } 1714 1715 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset) 1716 { 1717 return (offset >> inode->i_blkbits); 1718 } 1719 1720 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk) 1721 { 1722 return (blk << inode->i_blkbits); 1723 } 1724 1725 static int f2fs_xattr_fiemap(struct inode *inode, 1726 struct fiemap_extent_info *fieinfo) 1727 { 1728 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1729 struct page *page; 1730 struct node_info ni; 1731 __u64 phys = 0, len; 1732 __u32 flags; 1733 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1734 int err = 0; 1735 1736 if (f2fs_has_inline_xattr(inode)) { 1737 int offset; 1738 1739 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1740 inode->i_ino, false); 1741 if (!page) 1742 return -ENOMEM; 1743 1744 err = f2fs_get_node_info(sbi, inode->i_ino, &ni); 1745 if (err) { 1746 f2fs_put_page(page, 1); 1747 return err; 1748 } 1749 1750 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1751 offset = offsetof(struct f2fs_inode, i_addr) + 1752 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1753 get_inline_xattr_addrs(inode)); 1754 1755 phys += offset; 1756 len = inline_xattr_size(inode); 1757 1758 f2fs_put_page(page, 1); 1759 1760 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1761 1762 if (!xnid) 1763 flags |= FIEMAP_EXTENT_LAST; 1764 1765 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1766 if (err || err == 1) 1767 return err; 1768 } 1769 1770 if (xnid) { 1771 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1772 if (!page) 1773 return -ENOMEM; 1774 1775 err = f2fs_get_node_info(sbi, xnid, &ni); 1776 if (err) { 1777 f2fs_put_page(page, 1); 1778 return err; 1779 } 1780 1781 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1782 len = inode->i_sb->s_blocksize; 1783 1784 f2fs_put_page(page, 1); 1785 1786 flags = FIEMAP_EXTENT_LAST; 1787 } 1788 1789 if (phys) 1790 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1791 1792 return (err < 0 ? err : 0); 1793 } 1794 1795 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1796 u64 start, u64 len) 1797 { 1798 struct buffer_head map_bh; 1799 sector_t start_blk, last_blk; 1800 pgoff_t next_pgofs; 1801 u64 logical = 0, phys = 0, size = 0; 1802 u32 flags = 0; 1803 int ret = 0; 1804 1805 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { 1806 ret = f2fs_precache_extents(inode); 1807 if (ret) 1808 return ret; 1809 } 1810 1811 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR); 1812 if (ret) 1813 return ret; 1814 1815 inode_lock(inode); 1816 1817 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1818 ret = f2fs_xattr_fiemap(inode, fieinfo); 1819 goto out; 1820 } 1821 1822 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) { 1823 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1824 if (ret != -EAGAIN) 1825 goto out; 1826 } 1827 1828 if (logical_to_blk(inode, len) == 0) 1829 len = blk_to_logical(inode, 1); 1830 1831 start_blk = logical_to_blk(inode, start); 1832 last_blk = logical_to_blk(inode, start + len - 1); 1833 1834 next: 1835 memset(&map_bh, 0, sizeof(struct buffer_head)); 1836 map_bh.b_size = len; 1837 1838 ret = get_data_block(inode, start_blk, &map_bh, 0, 1839 F2FS_GET_BLOCK_FIEMAP, &next_pgofs); 1840 if (ret) 1841 goto out; 1842 1843 /* HOLE */ 1844 if (!buffer_mapped(&map_bh)) { 1845 start_blk = next_pgofs; 1846 1847 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode, 1848 F2FS_I_SB(inode)->max_file_blocks)) 1849 goto prep_next; 1850 1851 flags |= FIEMAP_EXTENT_LAST; 1852 } 1853 1854 if (size) { 1855 if (IS_ENCRYPTED(inode)) 1856 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1857 1858 ret = fiemap_fill_next_extent(fieinfo, logical, 1859 phys, size, flags); 1860 } 1861 1862 if (start_blk > last_blk || ret) 1863 goto out; 1864 1865 logical = blk_to_logical(inode, start_blk); 1866 phys = blk_to_logical(inode, map_bh.b_blocknr); 1867 size = map_bh.b_size; 1868 flags = 0; 1869 if (buffer_unwritten(&map_bh)) 1870 flags = FIEMAP_EXTENT_UNWRITTEN; 1871 1872 start_blk += logical_to_blk(inode, size); 1873 1874 prep_next: 1875 cond_resched(); 1876 if (fatal_signal_pending(current)) 1877 ret = -EINTR; 1878 else 1879 goto next; 1880 out: 1881 if (ret == 1) 1882 ret = 0; 1883 1884 inode_unlock(inode); 1885 return ret; 1886 } 1887 1888 static inline loff_t f2fs_readpage_limit(struct inode *inode) 1889 { 1890 if (IS_ENABLED(CONFIG_FS_VERITY) && 1891 (IS_VERITY(inode) || f2fs_verity_in_progress(inode))) 1892 return inode->i_sb->s_maxbytes; 1893 1894 return i_size_read(inode); 1895 } 1896 1897 static int f2fs_read_single_page(struct inode *inode, struct page *page, 1898 unsigned nr_pages, 1899 struct f2fs_map_blocks *map, 1900 struct bio **bio_ret, 1901 sector_t *last_block_in_bio, 1902 bool is_readahead) 1903 { 1904 struct bio *bio = *bio_ret; 1905 const unsigned blkbits = inode->i_blkbits; 1906 const unsigned blocksize = 1 << blkbits; 1907 sector_t block_in_file; 1908 sector_t last_block; 1909 sector_t last_block_in_file; 1910 sector_t block_nr; 1911 int ret = 0; 1912 1913 block_in_file = (sector_t)page_index(page); 1914 last_block = block_in_file + nr_pages; 1915 last_block_in_file = (f2fs_readpage_limit(inode) + blocksize - 1) >> 1916 blkbits; 1917 if (last_block > last_block_in_file) 1918 last_block = last_block_in_file; 1919 1920 /* just zeroing out page which is beyond EOF */ 1921 if (block_in_file >= last_block) 1922 goto zero_out; 1923 /* 1924 * Map blocks using the previous result first. 1925 */ 1926 if ((map->m_flags & F2FS_MAP_MAPPED) && 1927 block_in_file > map->m_lblk && 1928 block_in_file < (map->m_lblk + map->m_len)) 1929 goto got_it; 1930 1931 /* 1932 * Then do more f2fs_map_blocks() calls until we are 1933 * done with this page. 1934 */ 1935 map->m_lblk = block_in_file; 1936 map->m_len = last_block - block_in_file; 1937 1938 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT); 1939 if (ret) 1940 goto out; 1941 got_it: 1942 if ((map->m_flags & F2FS_MAP_MAPPED)) { 1943 block_nr = map->m_pblk + block_in_file - map->m_lblk; 1944 SetPageMappedToDisk(page); 1945 1946 if (!PageUptodate(page) && (!PageSwapCache(page) && 1947 !cleancache_get_page(page))) { 1948 SetPageUptodate(page); 1949 goto confused; 1950 } 1951 1952 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr, 1953 DATA_GENERIC_ENHANCE_READ)) { 1954 ret = -EFSCORRUPTED; 1955 goto out; 1956 } 1957 } else { 1958 zero_out: 1959 zero_user_segment(page, 0, PAGE_SIZE); 1960 if (f2fs_need_verity(inode, page->index) && 1961 !fsverity_verify_page(page)) { 1962 ret = -EIO; 1963 goto out; 1964 } 1965 if (!PageUptodate(page)) 1966 SetPageUptodate(page); 1967 unlock_page(page); 1968 goto out; 1969 } 1970 1971 /* 1972 * This page will go to BIO. Do we need to send this 1973 * BIO off first? 1974 */ 1975 if (bio && !page_is_mergeable(F2FS_I_SB(inode), bio, 1976 *last_block_in_bio, block_nr)) { 1977 submit_and_realloc: 1978 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1979 bio = NULL; 1980 } 1981 if (bio == NULL) { 1982 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages, 1983 is_readahead ? REQ_RAHEAD : 0, page->index); 1984 if (IS_ERR(bio)) { 1985 ret = PTR_ERR(bio); 1986 bio = NULL; 1987 goto out; 1988 } 1989 } 1990 1991 /* 1992 * If the page is under writeback, we need to wait for 1993 * its completion to see the correct decrypted data. 1994 */ 1995 f2fs_wait_on_block_writeback(inode, block_nr); 1996 1997 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 1998 goto submit_and_realloc; 1999 2000 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 2001 ClearPageError(page); 2002 *last_block_in_bio = block_nr; 2003 goto out; 2004 confused: 2005 if (bio) { 2006 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2007 bio = NULL; 2008 } 2009 unlock_page(page); 2010 out: 2011 *bio_ret = bio; 2012 return ret; 2013 } 2014 2015 #ifdef CONFIG_F2FS_FS_COMPRESSION 2016 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret, 2017 unsigned nr_pages, sector_t *last_block_in_bio, 2018 bool is_readahead) 2019 { 2020 struct dnode_of_data dn; 2021 struct inode *inode = cc->inode; 2022 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2023 struct bio *bio = *bio_ret; 2024 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size; 2025 sector_t last_block_in_file; 2026 const unsigned blkbits = inode->i_blkbits; 2027 const unsigned blocksize = 1 << blkbits; 2028 struct decompress_io_ctx *dic = NULL; 2029 int i; 2030 int ret = 0; 2031 2032 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc)); 2033 2034 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits; 2035 2036 /* get rid of pages beyond EOF */ 2037 for (i = 0; i < cc->cluster_size; i++) { 2038 struct page *page = cc->rpages[i]; 2039 2040 if (!page) 2041 continue; 2042 if ((sector_t)page->index >= last_block_in_file) { 2043 zero_user_segment(page, 0, PAGE_SIZE); 2044 if (!PageUptodate(page)) 2045 SetPageUptodate(page); 2046 } else if (!PageUptodate(page)) { 2047 continue; 2048 } 2049 unlock_page(page); 2050 cc->rpages[i] = NULL; 2051 cc->nr_rpages--; 2052 } 2053 2054 /* we are done since all pages are beyond EOF */ 2055 if (f2fs_cluster_is_empty(cc)) 2056 goto out; 2057 2058 set_new_dnode(&dn, inode, NULL, NULL, 0); 2059 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 2060 if (ret) 2061 goto out; 2062 2063 /* cluster was overwritten as normal cluster */ 2064 if (dn.data_blkaddr != COMPRESS_ADDR) 2065 goto out; 2066 2067 for (i = 1; i < cc->cluster_size; i++) { 2068 block_t blkaddr; 2069 2070 blkaddr = datablock_addr(dn.inode, dn.node_page, 2071 dn.ofs_in_node + i); 2072 2073 if (!__is_valid_data_blkaddr(blkaddr)) 2074 break; 2075 2076 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) { 2077 ret = -EFAULT; 2078 goto out_put_dnode; 2079 } 2080 cc->nr_cpages++; 2081 } 2082 2083 /* nothing to decompress */ 2084 if (cc->nr_cpages == 0) { 2085 ret = 0; 2086 goto out_put_dnode; 2087 } 2088 2089 dic = f2fs_alloc_dic(cc); 2090 if (IS_ERR(dic)) { 2091 ret = PTR_ERR(dic); 2092 goto out_put_dnode; 2093 } 2094 2095 for (i = 0; i < dic->nr_cpages; i++) { 2096 struct page *page = dic->cpages[i]; 2097 block_t blkaddr; 2098 2099 blkaddr = datablock_addr(dn.inode, dn.node_page, 2100 dn.ofs_in_node + i + 1); 2101 2102 if (bio && !page_is_mergeable(sbi, bio, 2103 *last_block_in_bio, blkaddr)) { 2104 submit_and_realloc: 2105 __submit_bio(sbi, bio, DATA); 2106 bio = NULL; 2107 } 2108 2109 if (!bio) { 2110 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages, 2111 is_readahead ? REQ_RAHEAD : 0, 2112 page->index); 2113 if (IS_ERR(bio)) { 2114 ret = PTR_ERR(bio); 2115 bio = NULL; 2116 dic->failed = true; 2117 if (refcount_sub_and_test(dic->nr_cpages - i, 2118 &dic->ref)) 2119 f2fs_decompress_end_io(dic->rpages, 2120 cc->cluster_size, true, 2121 false); 2122 f2fs_free_dic(dic); 2123 f2fs_put_dnode(&dn); 2124 *bio_ret = bio; 2125 return ret; 2126 } 2127 } 2128 2129 f2fs_wait_on_block_writeback(inode, blkaddr); 2130 2131 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2132 goto submit_and_realloc; 2133 2134 inc_page_count(sbi, F2FS_RD_DATA); 2135 ClearPageError(page); 2136 *last_block_in_bio = blkaddr; 2137 } 2138 2139 f2fs_put_dnode(&dn); 2140 2141 *bio_ret = bio; 2142 return 0; 2143 2144 out_put_dnode: 2145 f2fs_put_dnode(&dn); 2146 out: 2147 f2fs_decompress_end_io(cc->rpages, cc->cluster_size, true, false); 2148 *bio_ret = bio; 2149 return ret; 2150 } 2151 #endif 2152 2153 /* 2154 * This function was originally taken from fs/mpage.c, and customized for f2fs. 2155 * Major change was from block_size == page_size in f2fs by default. 2156 * 2157 * Note that the aops->readpages() function is ONLY used for read-ahead. If 2158 * this function ever deviates from doing just read-ahead, it should either 2159 * use ->readpage() or do the necessary surgery to decouple ->readpages() 2160 * from read-ahead. 2161 */ 2162 int f2fs_mpage_readpages(struct address_space *mapping, 2163 struct list_head *pages, struct page *page, 2164 unsigned nr_pages, bool is_readahead) 2165 { 2166 struct bio *bio = NULL; 2167 sector_t last_block_in_bio = 0; 2168 struct inode *inode = mapping->host; 2169 struct f2fs_map_blocks map; 2170 #ifdef CONFIG_F2FS_FS_COMPRESSION 2171 struct compress_ctx cc = { 2172 .inode = inode, 2173 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2174 .cluster_size = F2FS_I(inode)->i_cluster_size, 2175 .cluster_idx = NULL_CLUSTER, 2176 .rpages = NULL, 2177 .cpages = NULL, 2178 .nr_rpages = 0, 2179 .nr_cpages = 0, 2180 }; 2181 #endif 2182 unsigned max_nr_pages = nr_pages; 2183 int ret = 0; 2184 2185 map.m_pblk = 0; 2186 map.m_lblk = 0; 2187 map.m_len = 0; 2188 map.m_flags = 0; 2189 map.m_next_pgofs = NULL; 2190 map.m_next_extent = NULL; 2191 map.m_seg_type = NO_CHECK_TYPE; 2192 map.m_may_create = false; 2193 2194 for (; nr_pages; nr_pages--) { 2195 if (pages) { 2196 page = list_last_entry(pages, struct page, lru); 2197 2198 prefetchw(&page->flags); 2199 list_del(&page->lru); 2200 if (add_to_page_cache_lru(page, mapping, 2201 page_index(page), 2202 readahead_gfp_mask(mapping))) 2203 goto next_page; 2204 } 2205 2206 #ifdef CONFIG_F2FS_FS_COMPRESSION 2207 if (f2fs_compressed_file(inode)) { 2208 /* there are remained comressed pages, submit them */ 2209 if (!f2fs_cluster_can_merge_page(&cc, page->index)) { 2210 ret = f2fs_read_multi_pages(&cc, &bio, 2211 max_nr_pages, 2212 &last_block_in_bio, 2213 is_readahead); 2214 f2fs_destroy_compress_ctx(&cc); 2215 if (ret) 2216 goto set_error_page; 2217 } 2218 ret = f2fs_is_compressed_cluster(inode, page->index); 2219 if (ret < 0) 2220 goto set_error_page; 2221 else if (!ret) 2222 goto read_single_page; 2223 2224 ret = f2fs_init_compress_ctx(&cc); 2225 if (ret) 2226 goto set_error_page; 2227 2228 f2fs_compress_ctx_add_page(&cc, page); 2229 2230 goto next_page; 2231 } 2232 read_single_page: 2233 #endif 2234 2235 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map, 2236 &bio, &last_block_in_bio, is_readahead); 2237 if (ret) { 2238 #ifdef CONFIG_F2FS_FS_COMPRESSION 2239 set_error_page: 2240 #endif 2241 SetPageError(page); 2242 zero_user_segment(page, 0, PAGE_SIZE); 2243 unlock_page(page); 2244 } 2245 next_page: 2246 if (pages) 2247 put_page(page); 2248 2249 #ifdef CONFIG_F2FS_FS_COMPRESSION 2250 if (f2fs_compressed_file(inode)) { 2251 /* last page */ 2252 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) { 2253 ret = f2fs_read_multi_pages(&cc, &bio, 2254 max_nr_pages, 2255 &last_block_in_bio, 2256 is_readahead); 2257 f2fs_destroy_compress_ctx(&cc); 2258 } 2259 } 2260 #endif 2261 } 2262 BUG_ON(pages && !list_empty(pages)); 2263 if (bio) 2264 __submit_bio(F2FS_I_SB(inode), bio, DATA); 2265 return pages ? 0 : ret; 2266 } 2267 2268 static int f2fs_read_data_page(struct file *file, struct page *page) 2269 { 2270 struct inode *inode = page_file_mapping(page)->host; 2271 int ret = -EAGAIN; 2272 2273 trace_f2fs_readpage(page, DATA); 2274 2275 if (!f2fs_is_compress_backend_ready(inode)) { 2276 unlock_page(page); 2277 return -EOPNOTSUPP; 2278 } 2279 2280 /* If the file has inline data, try to read it directly */ 2281 if (f2fs_has_inline_data(inode)) 2282 ret = f2fs_read_inline_data(inode, page); 2283 if (ret == -EAGAIN) 2284 ret = f2fs_mpage_readpages(page_file_mapping(page), 2285 NULL, page, 1, false); 2286 return ret; 2287 } 2288 2289 static int f2fs_read_data_pages(struct file *file, 2290 struct address_space *mapping, 2291 struct list_head *pages, unsigned nr_pages) 2292 { 2293 struct inode *inode = mapping->host; 2294 struct page *page = list_last_entry(pages, struct page, lru); 2295 2296 trace_f2fs_readpages(inode, page, nr_pages); 2297 2298 if (!f2fs_is_compress_backend_ready(inode)) 2299 return 0; 2300 2301 /* If the file has inline data, skip readpages */ 2302 if (f2fs_has_inline_data(inode)) 2303 return 0; 2304 2305 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true); 2306 } 2307 2308 int f2fs_encrypt_one_page(struct f2fs_io_info *fio) 2309 { 2310 struct inode *inode = fio->page->mapping->host; 2311 struct page *mpage, *page; 2312 gfp_t gfp_flags = GFP_NOFS; 2313 2314 if (!f2fs_encrypted_file(inode)) 2315 return 0; 2316 2317 page = fio->compressed_page ? fio->compressed_page : fio->page; 2318 2319 /* wait for GCed page writeback via META_MAPPING */ 2320 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr); 2321 2322 retry_encrypt: 2323 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page, 2324 PAGE_SIZE, 0, gfp_flags); 2325 if (IS_ERR(fio->encrypted_page)) { 2326 /* flush pending IOs and wait for a while in the ENOMEM case */ 2327 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 2328 f2fs_flush_merged_writes(fio->sbi); 2329 congestion_wait(BLK_RW_ASYNC, HZ/50); 2330 gfp_flags |= __GFP_NOFAIL; 2331 goto retry_encrypt; 2332 } 2333 return PTR_ERR(fio->encrypted_page); 2334 } 2335 2336 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr); 2337 if (mpage) { 2338 if (PageUptodate(mpage)) 2339 memcpy(page_address(mpage), 2340 page_address(fio->encrypted_page), PAGE_SIZE); 2341 f2fs_put_page(mpage, 1); 2342 } 2343 return 0; 2344 } 2345 2346 static inline bool check_inplace_update_policy(struct inode *inode, 2347 struct f2fs_io_info *fio) 2348 { 2349 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2350 unsigned int policy = SM_I(sbi)->ipu_policy; 2351 2352 if (policy & (0x1 << F2FS_IPU_FORCE)) 2353 return true; 2354 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi)) 2355 return true; 2356 if (policy & (0x1 << F2FS_IPU_UTIL) && 2357 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2358 return true; 2359 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) && 2360 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2361 return true; 2362 2363 /* 2364 * IPU for rewrite async pages 2365 */ 2366 if (policy & (0x1 << F2FS_IPU_ASYNC) && 2367 fio && fio->op == REQ_OP_WRITE && 2368 !(fio->op_flags & REQ_SYNC) && 2369 !IS_ENCRYPTED(inode)) 2370 return true; 2371 2372 /* this is only set during fdatasync */ 2373 if (policy & (0x1 << F2FS_IPU_FSYNC) && 2374 is_inode_flag_set(inode, FI_NEED_IPU)) 2375 return true; 2376 2377 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2378 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2379 return true; 2380 2381 return false; 2382 } 2383 2384 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) 2385 { 2386 if (f2fs_is_pinned_file(inode)) 2387 return true; 2388 2389 /* if this is cold file, we should overwrite to avoid fragmentation */ 2390 if (file_is_cold(inode)) 2391 return true; 2392 2393 return check_inplace_update_policy(inode, fio); 2394 } 2395 2396 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) 2397 { 2398 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2399 2400 if (test_opt(sbi, LFS)) 2401 return true; 2402 if (S_ISDIR(inode->i_mode)) 2403 return true; 2404 if (IS_NOQUOTA(inode)) 2405 return true; 2406 if (f2fs_is_atomic_file(inode)) 2407 return true; 2408 if (fio) { 2409 if (is_cold_data(fio->page)) 2410 return true; 2411 if (IS_ATOMIC_WRITTEN_PAGE(fio->page)) 2412 return true; 2413 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2414 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2415 return true; 2416 } 2417 return false; 2418 } 2419 2420 static inline bool need_inplace_update(struct f2fs_io_info *fio) 2421 { 2422 struct inode *inode = fio->page->mapping->host; 2423 2424 if (f2fs_should_update_outplace(inode, fio)) 2425 return false; 2426 2427 return f2fs_should_update_inplace(inode, fio); 2428 } 2429 2430 int f2fs_do_write_data_page(struct f2fs_io_info *fio) 2431 { 2432 struct page *page = fio->page; 2433 struct inode *inode = page->mapping->host; 2434 struct dnode_of_data dn; 2435 struct extent_info ei = {0,0,0}; 2436 struct node_info ni; 2437 bool ipu_force = false; 2438 int err = 0; 2439 2440 set_new_dnode(&dn, inode, NULL, NULL, 0); 2441 if (need_inplace_update(fio) && 2442 f2fs_lookup_extent_cache(inode, page->index, &ei)) { 2443 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 2444 2445 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2446 DATA_GENERIC_ENHANCE)) 2447 return -EFSCORRUPTED; 2448 2449 ipu_force = true; 2450 fio->need_lock = LOCK_DONE; 2451 goto got_it; 2452 } 2453 2454 /* Deadlock due to between page->lock and f2fs_lock_op */ 2455 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 2456 return -EAGAIN; 2457 2458 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 2459 if (err) 2460 goto out; 2461 2462 fio->old_blkaddr = dn.data_blkaddr; 2463 2464 /* This page is already truncated */ 2465 if (fio->old_blkaddr == NULL_ADDR) { 2466 ClearPageUptodate(page); 2467 clear_cold_data(page); 2468 goto out_writepage; 2469 } 2470 got_it: 2471 if (__is_valid_data_blkaddr(fio->old_blkaddr) && 2472 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2473 DATA_GENERIC_ENHANCE)) { 2474 err = -EFSCORRUPTED; 2475 goto out_writepage; 2476 } 2477 /* 2478 * If current allocation needs SSR, 2479 * it had better in-place writes for updated data. 2480 */ 2481 if (ipu_force || 2482 (__is_valid_data_blkaddr(fio->old_blkaddr) && 2483 need_inplace_update(fio))) { 2484 err = f2fs_encrypt_one_page(fio); 2485 if (err) 2486 goto out_writepage; 2487 2488 set_page_writeback(page); 2489 ClearPageError(page); 2490 f2fs_put_dnode(&dn); 2491 if (fio->need_lock == LOCK_REQ) 2492 f2fs_unlock_op(fio->sbi); 2493 err = f2fs_inplace_write_data(fio); 2494 if (err) { 2495 if (f2fs_encrypted_file(inode)) 2496 fscrypt_finalize_bounce_page(&fio->encrypted_page); 2497 if (PageWriteback(page)) 2498 end_page_writeback(page); 2499 } else { 2500 set_inode_flag(inode, FI_UPDATE_WRITE); 2501 } 2502 trace_f2fs_do_write_data_page(fio->page, IPU); 2503 return err; 2504 } 2505 2506 if (fio->need_lock == LOCK_RETRY) { 2507 if (!f2fs_trylock_op(fio->sbi)) { 2508 err = -EAGAIN; 2509 goto out_writepage; 2510 } 2511 fio->need_lock = LOCK_REQ; 2512 } 2513 2514 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni); 2515 if (err) 2516 goto out_writepage; 2517 2518 fio->version = ni.version; 2519 2520 err = f2fs_encrypt_one_page(fio); 2521 if (err) 2522 goto out_writepage; 2523 2524 set_page_writeback(page); 2525 ClearPageError(page); 2526 2527 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR) 2528 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false); 2529 2530 /* LFS mode write path */ 2531 f2fs_outplace_write_data(&dn, fio); 2532 trace_f2fs_do_write_data_page(page, OPU); 2533 set_inode_flag(inode, FI_APPEND_WRITE); 2534 if (page->index == 0) 2535 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 2536 out_writepage: 2537 f2fs_put_dnode(&dn); 2538 out: 2539 if (fio->need_lock == LOCK_REQ) 2540 f2fs_unlock_op(fio->sbi); 2541 return err; 2542 } 2543 2544 int f2fs_write_single_data_page(struct page *page, int *submitted, 2545 struct bio **bio, 2546 sector_t *last_block, 2547 struct writeback_control *wbc, 2548 enum iostat_type io_type, 2549 int compr_blocks) 2550 { 2551 struct inode *inode = page->mapping->host; 2552 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2553 loff_t i_size = i_size_read(inode); 2554 const pgoff_t end_index = ((unsigned long long)i_size) 2555 >> PAGE_SHIFT; 2556 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT; 2557 unsigned offset = 0; 2558 bool need_balance_fs = false; 2559 int err = 0; 2560 struct f2fs_io_info fio = { 2561 .sbi = sbi, 2562 .ino = inode->i_ino, 2563 .type = DATA, 2564 .op = REQ_OP_WRITE, 2565 .op_flags = wbc_to_write_flags(wbc), 2566 .old_blkaddr = NULL_ADDR, 2567 .page = page, 2568 .encrypted_page = NULL, 2569 .submitted = false, 2570 .compr_blocks = compr_blocks, 2571 .need_lock = LOCK_RETRY, 2572 .io_type = io_type, 2573 .io_wbc = wbc, 2574 .bio = bio, 2575 .last_block = last_block, 2576 }; 2577 2578 trace_f2fs_writepage(page, DATA); 2579 2580 /* we should bypass data pages to proceed the kworkder jobs */ 2581 if (unlikely(f2fs_cp_error(sbi))) { 2582 mapping_set_error(page->mapping, -EIO); 2583 /* 2584 * don't drop any dirty dentry pages for keeping lastest 2585 * directory structure. 2586 */ 2587 if (S_ISDIR(inode->i_mode)) 2588 goto redirty_out; 2589 goto out; 2590 } 2591 2592 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2593 goto redirty_out; 2594 2595 if (page->index < end_index || 2596 f2fs_verity_in_progress(inode) || 2597 compr_blocks) 2598 goto write; 2599 2600 /* 2601 * If the offset is out-of-range of file size, 2602 * this page does not have to be written to disk. 2603 */ 2604 offset = i_size & (PAGE_SIZE - 1); 2605 if ((page->index >= end_index + 1) || !offset) 2606 goto out; 2607 2608 zero_user_segment(page, offset, PAGE_SIZE); 2609 write: 2610 if (f2fs_is_drop_cache(inode)) 2611 goto out; 2612 /* we should not write 0'th page having journal header */ 2613 if (f2fs_is_volatile_file(inode) && (!page->index || 2614 (!wbc->for_reclaim && 2615 f2fs_available_free_memory(sbi, BASE_CHECK)))) 2616 goto redirty_out; 2617 2618 /* Dentry blocks are controlled by checkpoint */ 2619 if (S_ISDIR(inode->i_mode)) { 2620 fio.need_lock = LOCK_DONE; 2621 err = f2fs_do_write_data_page(&fio); 2622 goto done; 2623 } 2624 2625 if (!wbc->for_reclaim) 2626 need_balance_fs = true; 2627 else if (has_not_enough_free_secs(sbi, 0, 0)) 2628 goto redirty_out; 2629 else 2630 set_inode_flag(inode, FI_HOT_DATA); 2631 2632 err = -EAGAIN; 2633 if (f2fs_has_inline_data(inode)) { 2634 err = f2fs_write_inline_data(inode, page); 2635 if (!err) 2636 goto out; 2637 } 2638 2639 if (err == -EAGAIN) { 2640 err = f2fs_do_write_data_page(&fio); 2641 if (err == -EAGAIN) { 2642 fio.need_lock = LOCK_REQ; 2643 err = f2fs_do_write_data_page(&fio); 2644 } 2645 } 2646 2647 if (err) { 2648 file_set_keep_isize(inode); 2649 } else { 2650 down_write(&F2FS_I(inode)->i_sem); 2651 if (F2FS_I(inode)->last_disk_size < psize) 2652 F2FS_I(inode)->last_disk_size = psize; 2653 up_write(&F2FS_I(inode)->i_sem); 2654 } 2655 2656 done: 2657 if (err && err != -ENOENT) 2658 goto redirty_out; 2659 2660 out: 2661 inode_dec_dirty_pages(inode); 2662 if (err) { 2663 ClearPageUptodate(page); 2664 clear_cold_data(page); 2665 } 2666 2667 if (wbc->for_reclaim) { 2668 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA); 2669 clear_inode_flag(inode, FI_HOT_DATA); 2670 f2fs_remove_dirty_inode(inode); 2671 submitted = NULL; 2672 } 2673 unlock_page(page); 2674 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) && 2675 !F2FS_I(inode)->cp_task) 2676 f2fs_balance_fs(sbi, need_balance_fs); 2677 2678 if (unlikely(f2fs_cp_error(sbi))) { 2679 f2fs_submit_merged_write(sbi, DATA); 2680 f2fs_submit_merged_ipu_write(sbi, bio, NULL); 2681 submitted = NULL; 2682 } 2683 2684 if (submitted) 2685 *submitted = fio.submitted ? 1 : 0; 2686 2687 return 0; 2688 2689 redirty_out: 2690 redirty_page_for_writepage(wbc, page); 2691 /* 2692 * pageout() in MM traslates EAGAIN, so calls handle_write_error() 2693 * -> mapping_set_error() -> set_bit(AS_EIO, ...). 2694 * file_write_and_wait_range() will see EIO error, which is critical 2695 * to return value of fsync() followed by atomic_write failure to user. 2696 */ 2697 if (!err || wbc->for_reclaim) 2698 return AOP_WRITEPAGE_ACTIVATE; 2699 unlock_page(page); 2700 return err; 2701 } 2702 2703 static int f2fs_write_data_page(struct page *page, 2704 struct writeback_control *wbc) 2705 { 2706 #ifdef CONFIG_F2FS_FS_COMPRESSION 2707 struct inode *inode = page->mapping->host; 2708 2709 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 2710 goto out; 2711 2712 if (f2fs_compressed_file(inode)) { 2713 if (f2fs_is_compressed_cluster(inode, page->index)) { 2714 redirty_page_for_writepage(wbc, page); 2715 return AOP_WRITEPAGE_ACTIVATE; 2716 } 2717 } 2718 out: 2719 #endif 2720 2721 return f2fs_write_single_data_page(page, NULL, NULL, NULL, 2722 wbc, FS_DATA_IO, 0); 2723 } 2724 2725 /* 2726 * This function was copied from write_cche_pages from mm/page-writeback.c. 2727 * The major change is making write step of cold data page separately from 2728 * warm/hot data page. 2729 */ 2730 static int f2fs_write_cache_pages(struct address_space *mapping, 2731 struct writeback_control *wbc, 2732 enum iostat_type io_type) 2733 { 2734 int ret = 0; 2735 int done = 0, retry = 0; 2736 struct pagevec pvec; 2737 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2738 struct bio *bio = NULL; 2739 sector_t last_block; 2740 #ifdef CONFIG_F2FS_FS_COMPRESSION 2741 struct inode *inode = mapping->host; 2742 struct compress_ctx cc = { 2743 .inode = inode, 2744 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2745 .cluster_size = F2FS_I(inode)->i_cluster_size, 2746 .cluster_idx = NULL_CLUSTER, 2747 .rpages = NULL, 2748 .nr_rpages = 0, 2749 .cpages = NULL, 2750 .rbuf = NULL, 2751 .cbuf = NULL, 2752 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size, 2753 .private = NULL, 2754 }; 2755 #endif 2756 int nr_pages; 2757 pgoff_t uninitialized_var(writeback_index); 2758 pgoff_t index; 2759 pgoff_t end; /* Inclusive */ 2760 pgoff_t done_index; 2761 int cycled; 2762 int range_whole = 0; 2763 xa_mark_t tag; 2764 int nwritten = 0; 2765 int submitted = 0; 2766 int i; 2767 2768 pagevec_init(&pvec); 2769 2770 if (get_dirty_pages(mapping->host) <= 2771 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 2772 set_inode_flag(mapping->host, FI_HOT_DATA); 2773 else 2774 clear_inode_flag(mapping->host, FI_HOT_DATA); 2775 2776 if (wbc->range_cyclic) { 2777 writeback_index = mapping->writeback_index; /* prev offset */ 2778 index = writeback_index; 2779 if (index == 0) 2780 cycled = 1; 2781 else 2782 cycled = 0; 2783 end = -1; 2784 } else { 2785 index = wbc->range_start >> PAGE_SHIFT; 2786 end = wbc->range_end >> PAGE_SHIFT; 2787 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2788 range_whole = 1; 2789 cycled = 1; /* ignore range_cyclic tests */ 2790 } 2791 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2792 tag = PAGECACHE_TAG_TOWRITE; 2793 else 2794 tag = PAGECACHE_TAG_DIRTY; 2795 retry: 2796 retry = 0; 2797 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2798 tag_pages_for_writeback(mapping, index, end); 2799 done_index = index; 2800 while (!done && !retry && (index <= end)) { 2801 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 2802 tag); 2803 if (nr_pages == 0) 2804 break; 2805 2806 for (i = 0; i < nr_pages; i++) { 2807 struct page *page = pvec.pages[i]; 2808 bool need_readd; 2809 readd: 2810 need_readd = false; 2811 #ifdef CONFIG_F2FS_FS_COMPRESSION 2812 if (f2fs_compressed_file(inode)) { 2813 ret = f2fs_init_compress_ctx(&cc); 2814 if (ret) { 2815 done = 1; 2816 break; 2817 } 2818 2819 if (!f2fs_cluster_can_merge_page(&cc, 2820 page->index)) { 2821 ret = f2fs_write_multi_pages(&cc, 2822 &submitted, wbc, io_type); 2823 if (!ret) 2824 need_readd = true; 2825 goto result; 2826 } 2827 2828 if (unlikely(f2fs_cp_error(sbi))) 2829 goto lock_page; 2830 2831 if (f2fs_cluster_is_empty(&cc)) { 2832 void *fsdata = NULL; 2833 struct page *pagep; 2834 int ret2; 2835 2836 ret2 = f2fs_prepare_compress_overwrite( 2837 inode, &pagep, 2838 page->index, &fsdata); 2839 if (ret2 < 0) { 2840 ret = ret2; 2841 done = 1; 2842 break; 2843 } else if (ret2 && 2844 !f2fs_compress_write_end(inode, 2845 fsdata, page->index, 2846 1)) { 2847 retry = 1; 2848 break; 2849 } 2850 } else { 2851 goto lock_page; 2852 } 2853 } 2854 #endif 2855 /* give a priority to WB_SYNC threads */ 2856 if (atomic_read(&sbi->wb_sync_req[DATA]) && 2857 wbc->sync_mode == WB_SYNC_NONE) { 2858 done = 1; 2859 break; 2860 } 2861 #ifdef CONFIG_F2FS_FS_COMPRESSION 2862 lock_page: 2863 #endif 2864 done_index = page->index; 2865 retry_write: 2866 lock_page(page); 2867 2868 if (unlikely(page->mapping != mapping)) { 2869 continue_unlock: 2870 unlock_page(page); 2871 continue; 2872 } 2873 2874 if (!PageDirty(page)) { 2875 /* someone wrote it for us */ 2876 goto continue_unlock; 2877 } 2878 2879 if (PageWriteback(page)) { 2880 if (wbc->sync_mode != WB_SYNC_NONE) 2881 f2fs_wait_on_page_writeback(page, 2882 DATA, true, true); 2883 else 2884 goto continue_unlock; 2885 } 2886 2887 if (!clear_page_dirty_for_io(page)) 2888 goto continue_unlock; 2889 2890 #ifdef CONFIG_F2FS_FS_COMPRESSION 2891 if (f2fs_compressed_file(inode)) { 2892 get_page(page); 2893 f2fs_compress_ctx_add_page(&cc, page); 2894 continue; 2895 } 2896 #endif 2897 ret = f2fs_write_single_data_page(page, &submitted, 2898 &bio, &last_block, wbc, io_type, 0); 2899 if (ret == AOP_WRITEPAGE_ACTIVATE) 2900 unlock_page(page); 2901 #ifdef CONFIG_F2FS_FS_COMPRESSION 2902 result: 2903 #endif 2904 nwritten += submitted; 2905 wbc->nr_to_write -= submitted; 2906 2907 if (unlikely(ret)) { 2908 /* 2909 * keep nr_to_write, since vfs uses this to 2910 * get # of written pages. 2911 */ 2912 if (ret == AOP_WRITEPAGE_ACTIVATE) { 2913 ret = 0; 2914 goto next; 2915 } else if (ret == -EAGAIN) { 2916 ret = 0; 2917 if (wbc->sync_mode == WB_SYNC_ALL) { 2918 cond_resched(); 2919 congestion_wait(BLK_RW_ASYNC, 2920 HZ/50); 2921 goto retry_write; 2922 } 2923 goto next; 2924 } 2925 done_index = page->index + 1; 2926 done = 1; 2927 break; 2928 } 2929 2930 if (wbc->nr_to_write <= 0 && 2931 wbc->sync_mode == WB_SYNC_NONE) { 2932 done = 1; 2933 break; 2934 } 2935 next: 2936 if (need_readd) 2937 goto readd; 2938 } 2939 pagevec_release(&pvec); 2940 cond_resched(); 2941 } 2942 #ifdef CONFIG_F2FS_FS_COMPRESSION 2943 /* flush remained pages in compress cluster */ 2944 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) { 2945 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type); 2946 nwritten += submitted; 2947 wbc->nr_to_write -= submitted; 2948 if (ret) { 2949 done = 1; 2950 retry = 0; 2951 } 2952 } 2953 #endif 2954 if ((!cycled && !done) || retry) { 2955 cycled = 1; 2956 index = 0; 2957 end = writeback_index - 1; 2958 goto retry; 2959 } 2960 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 2961 mapping->writeback_index = done_index; 2962 2963 if (nwritten) 2964 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 2965 NULL, 0, DATA); 2966 /* submit cached bio of IPU write */ 2967 if (bio) 2968 f2fs_submit_merged_ipu_write(sbi, &bio, NULL); 2969 2970 return ret; 2971 } 2972 2973 static inline bool __should_serialize_io(struct inode *inode, 2974 struct writeback_control *wbc) 2975 { 2976 if (!S_ISREG(inode->i_mode)) 2977 return false; 2978 if (f2fs_compressed_file(inode)) 2979 return true; 2980 if (IS_NOQUOTA(inode)) 2981 return false; 2982 /* to avoid deadlock in path of data flush */ 2983 if (F2FS_I(inode)->cp_task) 2984 return false; 2985 if (wbc->sync_mode != WB_SYNC_ALL) 2986 return true; 2987 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks) 2988 return true; 2989 return false; 2990 } 2991 2992 static int __f2fs_write_data_pages(struct address_space *mapping, 2993 struct writeback_control *wbc, 2994 enum iostat_type io_type) 2995 { 2996 struct inode *inode = mapping->host; 2997 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2998 struct blk_plug plug; 2999 int ret; 3000 bool locked = false; 3001 3002 /* deal with chardevs and other special file */ 3003 if (!mapping->a_ops->writepage) 3004 return 0; 3005 3006 /* skip writing if there is no dirty page in this inode */ 3007 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 3008 return 0; 3009 3010 /* during POR, we don't need to trigger writepage at all. */ 3011 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 3012 goto skip_write; 3013 3014 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) && 3015 wbc->sync_mode == WB_SYNC_NONE && 3016 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 3017 f2fs_available_free_memory(sbi, DIRTY_DENTS)) 3018 goto skip_write; 3019 3020 /* skip writing during file defragment */ 3021 if (is_inode_flag_set(inode, FI_DO_DEFRAG)) 3022 goto skip_write; 3023 3024 trace_f2fs_writepages(mapping->host, wbc, DATA); 3025 3026 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 3027 if (wbc->sync_mode == WB_SYNC_ALL) 3028 atomic_inc(&sbi->wb_sync_req[DATA]); 3029 else if (atomic_read(&sbi->wb_sync_req[DATA])) 3030 goto skip_write; 3031 3032 if (__should_serialize_io(inode, wbc)) { 3033 mutex_lock(&sbi->writepages); 3034 locked = true; 3035 } 3036 3037 blk_start_plug(&plug); 3038 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 3039 blk_finish_plug(&plug); 3040 3041 if (locked) 3042 mutex_unlock(&sbi->writepages); 3043 3044 if (wbc->sync_mode == WB_SYNC_ALL) 3045 atomic_dec(&sbi->wb_sync_req[DATA]); 3046 /* 3047 * if some pages were truncated, we cannot guarantee its mapping->host 3048 * to detect pending bios. 3049 */ 3050 3051 f2fs_remove_dirty_inode(inode); 3052 return ret; 3053 3054 skip_write: 3055 wbc->pages_skipped += get_dirty_pages(inode); 3056 trace_f2fs_writepages(mapping->host, wbc, DATA); 3057 return 0; 3058 } 3059 3060 static int f2fs_write_data_pages(struct address_space *mapping, 3061 struct writeback_control *wbc) 3062 { 3063 struct inode *inode = mapping->host; 3064 3065 return __f2fs_write_data_pages(mapping, wbc, 3066 F2FS_I(inode)->cp_task == current ? 3067 FS_CP_DATA_IO : FS_DATA_IO); 3068 } 3069 3070 static void f2fs_write_failed(struct address_space *mapping, loff_t to) 3071 { 3072 struct inode *inode = mapping->host; 3073 loff_t i_size = i_size_read(inode); 3074 3075 if (IS_NOQUOTA(inode)) 3076 return; 3077 3078 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */ 3079 if (to > i_size && !f2fs_verity_in_progress(inode)) { 3080 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3081 down_write(&F2FS_I(inode)->i_mmap_sem); 3082 3083 truncate_pagecache(inode, i_size); 3084 f2fs_truncate_blocks(inode, i_size, true); 3085 3086 up_write(&F2FS_I(inode)->i_mmap_sem); 3087 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3088 } 3089 } 3090 3091 static int prepare_write_begin(struct f2fs_sb_info *sbi, 3092 struct page *page, loff_t pos, unsigned len, 3093 block_t *blk_addr, bool *node_changed) 3094 { 3095 struct inode *inode = page->mapping->host; 3096 pgoff_t index = page->index; 3097 struct dnode_of_data dn; 3098 struct page *ipage; 3099 bool locked = false; 3100 struct extent_info ei = {0,0,0}; 3101 int err = 0; 3102 int flag; 3103 3104 /* 3105 * we already allocated all the blocks, so we don't need to get 3106 * the block addresses when there is no need to fill the page. 3107 */ 3108 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE && 3109 !is_inode_flag_set(inode, FI_NO_PREALLOC) && 3110 !f2fs_verity_in_progress(inode)) 3111 return 0; 3112 3113 /* f2fs_lock_op avoids race between write CP and convert_inline_page */ 3114 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode)) 3115 flag = F2FS_GET_BLOCK_DEFAULT; 3116 else 3117 flag = F2FS_GET_BLOCK_PRE_AIO; 3118 3119 if (f2fs_has_inline_data(inode) || 3120 (pos & PAGE_MASK) >= i_size_read(inode)) { 3121 __do_map_lock(sbi, flag, true); 3122 locked = true; 3123 } 3124 3125 restart: 3126 /* check inline_data */ 3127 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3128 if (IS_ERR(ipage)) { 3129 err = PTR_ERR(ipage); 3130 goto unlock_out; 3131 } 3132 3133 set_new_dnode(&dn, inode, ipage, ipage, 0); 3134 3135 if (f2fs_has_inline_data(inode)) { 3136 if (pos + len <= MAX_INLINE_DATA(inode)) { 3137 f2fs_do_read_inline_data(page, ipage); 3138 set_inode_flag(inode, FI_DATA_EXIST); 3139 if (inode->i_nlink) 3140 set_inline_node(ipage); 3141 } else { 3142 err = f2fs_convert_inline_page(&dn, page); 3143 if (err) 3144 goto out; 3145 if (dn.data_blkaddr == NULL_ADDR) 3146 err = f2fs_get_block(&dn, index); 3147 } 3148 } else if (locked) { 3149 err = f2fs_get_block(&dn, index); 3150 } else { 3151 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 3152 dn.data_blkaddr = ei.blk + index - ei.fofs; 3153 } else { 3154 /* hole case */ 3155 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3156 if (err || dn.data_blkaddr == NULL_ADDR) { 3157 f2fs_put_dnode(&dn); 3158 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 3159 true); 3160 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO); 3161 locked = true; 3162 goto restart; 3163 } 3164 } 3165 } 3166 3167 /* convert_inline_page can make node_changed */ 3168 *blk_addr = dn.data_blkaddr; 3169 *node_changed = dn.node_changed; 3170 out: 3171 f2fs_put_dnode(&dn); 3172 unlock_out: 3173 if (locked) 3174 __do_map_lock(sbi, flag, false); 3175 return err; 3176 } 3177 3178 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 3179 loff_t pos, unsigned len, unsigned flags, 3180 struct page **pagep, void **fsdata) 3181 { 3182 struct inode *inode = mapping->host; 3183 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3184 struct page *page = NULL; 3185 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 3186 bool need_balance = false, drop_atomic = false; 3187 block_t blkaddr = NULL_ADDR; 3188 int err = 0; 3189 3190 trace_f2fs_write_begin(inode, pos, len, flags); 3191 3192 if (!f2fs_is_checkpoint_ready(sbi)) { 3193 err = -ENOSPC; 3194 goto fail; 3195 } 3196 3197 if ((f2fs_is_atomic_file(inode) && 3198 !f2fs_available_free_memory(sbi, INMEM_PAGES)) || 3199 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) { 3200 err = -ENOMEM; 3201 drop_atomic = true; 3202 goto fail; 3203 } 3204 3205 /* 3206 * We should check this at this moment to avoid deadlock on inode page 3207 * and #0 page. The locking rule for inline_data conversion should be: 3208 * lock_page(page #0) -> lock_page(inode_page) 3209 */ 3210 if (index != 0) { 3211 err = f2fs_convert_inline_inode(inode); 3212 if (err) 3213 goto fail; 3214 } 3215 3216 #ifdef CONFIG_F2FS_FS_COMPRESSION 3217 if (f2fs_compressed_file(inode)) { 3218 int ret; 3219 3220 *fsdata = NULL; 3221 3222 ret = f2fs_prepare_compress_overwrite(inode, pagep, 3223 index, fsdata); 3224 if (ret < 0) { 3225 err = ret; 3226 goto fail; 3227 } else if (ret) { 3228 return 0; 3229 } 3230 } 3231 #endif 3232 3233 repeat: 3234 /* 3235 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 3236 * wait_for_stable_page. Will wait that below with our IO control. 3237 */ 3238 page = f2fs_pagecache_get_page(mapping, index, 3239 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 3240 if (!page) { 3241 err = -ENOMEM; 3242 goto fail; 3243 } 3244 3245 /* TODO: cluster can be compressed due to race with .writepage */ 3246 3247 *pagep = page; 3248 3249 err = prepare_write_begin(sbi, page, pos, len, 3250 &blkaddr, &need_balance); 3251 if (err) 3252 goto fail; 3253 3254 if (need_balance && !IS_NOQUOTA(inode) && 3255 has_not_enough_free_secs(sbi, 0, 0)) { 3256 unlock_page(page); 3257 f2fs_balance_fs(sbi, true); 3258 lock_page(page); 3259 if (page->mapping != mapping) { 3260 /* The page got truncated from under us */ 3261 f2fs_put_page(page, 1); 3262 goto repeat; 3263 } 3264 } 3265 3266 f2fs_wait_on_page_writeback(page, DATA, false, true); 3267 3268 if (len == PAGE_SIZE || PageUptodate(page)) 3269 return 0; 3270 3271 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) && 3272 !f2fs_verity_in_progress(inode)) { 3273 zero_user_segment(page, len, PAGE_SIZE); 3274 return 0; 3275 } 3276 3277 if (blkaddr == NEW_ADDR) { 3278 zero_user_segment(page, 0, PAGE_SIZE); 3279 SetPageUptodate(page); 3280 } else { 3281 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 3282 DATA_GENERIC_ENHANCE_READ)) { 3283 err = -EFSCORRUPTED; 3284 goto fail; 3285 } 3286 err = f2fs_submit_page_read(inode, page, blkaddr); 3287 if (err) 3288 goto fail; 3289 3290 lock_page(page); 3291 if (unlikely(page->mapping != mapping)) { 3292 f2fs_put_page(page, 1); 3293 goto repeat; 3294 } 3295 if (unlikely(!PageUptodate(page))) { 3296 err = -EIO; 3297 goto fail; 3298 } 3299 } 3300 return 0; 3301 3302 fail: 3303 f2fs_put_page(page, 1); 3304 f2fs_write_failed(mapping, pos + len); 3305 if (drop_atomic) 3306 f2fs_drop_inmem_pages_all(sbi, false); 3307 return err; 3308 } 3309 3310 static int f2fs_write_end(struct file *file, 3311 struct address_space *mapping, 3312 loff_t pos, unsigned len, unsigned copied, 3313 struct page *page, void *fsdata) 3314 { 3315 struct inode *inode = page->mapping->host; 3316 3317 trace_f2fs_write_end(inode, pos, len, copied); 3318 3319 /* 3320 * This should be come from len == PAGE_SIZE, and we expect copied 3321 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 3322 * let generic_perform_write() try to copy data again through copied=0. 3323 */ 3324 if (!PageUptodate(page)) { 3325 if (unlikely(copied != len)) 3326 copied = 0; 3327 else 3328 SetPageUptodate(page); 3329 } 3330 3331 #ifdef CONFIG_F2FS_FS_COMPRESSION 3332 /* overwrite compressed file */ 3333 if (f2fs_compressed_file(inode) && fsdata) { 3334 f2fs_compress_write_end(inode, fsdata, page->index, copied); 3335 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3336 return copied; 3337 } 3338 #endif 3339 3340 if (!copied) 3341 goto unlock_out; 3342 3343 set_page_dirty(page); 3344 3345 if (pos + copied > i_size_read(inode) && 3346 !f2fs_verity_in_progress(inode)) 3347 f2fs_i_size_write(inode, pos + copied); 3348 unlock_out: 3349 f2fs_put_page(page, 1); 3350 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3351 return copied; 3352 } 3353 3354 static int check_direct_IO(struct inode *inode, struct iov_iter *iter, 3355 loff_t offset) 3356 { 3357 unsigned i_blkbits = READ_ONCE(inode->i_blkbits); 3358 unsigned blkbits = i_blkbits; 3359 unsigned blocksize_mask = (1 << blkbits) - 1; 3360 unsigned long align = offset | iov_iter_alignment(iter); 3361 struct block_device *bdev = inode->i_sb->s_bdev; 3362 3363 if (align & blocksize_mask) { 3364 if (bdev) 3365 blkbits = blksize_bits(bdev_logical_block_size(bdev)); 3366 blocksize_mask = (1 << blkbits) - 1; 3367 if (align & blocksize_mask) 3368 return -EINVAL; 3369 return 1; 3370 } 3371 return 0; 3372 } 3373 3374 static void f2fs_dio_end_io(struct bio *bio) 3375 { 3376 struct f2fs_private_dio *dio = bio->bi_private; 3377 3378 dec_page_count(F2FS_I_SB(dio->inode), 3379 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ); 3380 3381 bio->bi_private = dio->orig_private; 3382 bio->bi_end_io = dio->orig_end_io; 3383 3384 kvfree(dio); 3385 3386 bio_endio(bio); 3387 } 3388 3389 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode, 3390 loff_t file_offset) 3391 { 3392 struct f2fs_private_dio *dio; 3393 bool write = (bio_op(bio) == REQ_OP_WRITE); 3394 3395 dio = f2fs_kzalloc(F2FS_I_SB(inode), 3396 sizeof(struct f2fs_private_dio), GFP_NOFS); 3397 if (!dio) 3398 goto out; 3399 3400 dio->inode = inode; 3401 dio->orig_end_io = bio->bi_end_io; 3402 dio->orig_private = bio->bi_private; 3403 dio->write = write; 3404 3405 bio->bi_end_io = f2fs_dio_end_io; 3406 bio->bi_private = dio; 3407 3408 inc_page_count(F2FS_I_SB(inode), 3409 write ? F2FS_DIO_WRITE : F2FS_DIO_READ); 3410 3411 submit_bio(bio); 3412 return; 3413 out: 3414 bio->bi_status = BLK_STS_IOERR; 3415 bio_endio(bio); 3416 } 3417 3418 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 3419 { 3420 struct address_space *mapping = iocb->ki_filp->f_mapping; 3421 struct inode *inode = mapping->host; 3422 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3423 struct f2fs_inode_info *fi = F2FS_I(inode); 3424 size_t count = iov_iter_count(iter); 3425 loff_t offset = iocb->ki_pos; 3426 int rw = iov_iter_rw(iter); 3427 int err; 3428 enum rw_hint hint = iocb->ki_hint; 3429 int whint_mode = F2FS_OPTION(sbi).whint_mode; 3430 bool do_opu; 3431 3432 err = check_direct_IO(inode, iter, offset); 3433 if (err) 3434 return err < 0 ? err : 0; 3435 3436 if (f2fs_force_buffered_io(inode, iocb, iter)) 3437 return 0; 3438 3439 do_opu = allow_outplace_dio(inode, iocb, iter); 3440 3441 trace_f2fs_direct_IO_enter(inode, offset, count, rw); 3442 3443 if (rw == WRITE && whint_mode == WHINT_MODE_OFF) 3444 iocb->ki_hint = WRITE_LIFE_NOT_SET; 3445 3446 if (iocb->ki_flags & IOCB_NOWAIT) { 3447 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) { 3448 iocb->ki_hint = hint; 3449 err = -EAGAIN; 3450 goto out; 3451 } 3452 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) { 3453 up_read(&fi->i_gc_rwsem[rw]); 3454 iocb->ki_hint = hint; 3455 err = -EAGAIN; 3456 goto out; 3457 } 3458 } else { 3459 down_read(&fi->i_gc_rwsem[rw]); 3460 if (do_opu) 3461 down_read(&fi->i_gc_rwsem[READ]); 3462 } 3463 3464 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, 3465 iter, rw == WRITE ? get_data_block_dio_write : 3466 get_data_block_dio, NULL, f2fs_dio_submit_bio, 3467 DIO_LOCKING | DIO_SKIP_HOLES); 3468 3469 if (do_opu) 3470 up_read(&fi->i_gc_rwsem[READ]); 3471 3472 up_read(&fi->i_gc_rwsem[rw]); 3473 3474 if (rw == WRITE) { 3475 if (whint_mode == WHINT_MODE_OFF) 3476 iocb->ki_hint = hint; 3477 if (err > 0) { 3478 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO, 3479 err); 3480 if (!do_opu) 3481 set_inode_flag(inode, FI_UPDATE_WRITE); 3482 } else if (err < 0) { 3483 f2fs_write_failed(mapping, offset + count); 3484 } 3485 } 3486 3487 out: 3488 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); 3489 3490 return err; 3491 } 3492 3493 void f2fs_invalidate_page(struct page *page, unsigned int offset, 3494 unsigned int length) 3495 { 3496 struct inode *inode = page->mapping->host; 3497 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3498 3499 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 3500 (offset % PAGE_SIZE || length != PAGE_SIZE)) 3501 return; 3502 3503 if (PageDirty(page)) { 3504 if (inode->i_ino == F2FS_META_INO(sbi)) { 3505 dec_page_count(sbi, F2FS_DIRTY_META); 3506 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 3507 dec_page_count(sbi, F2FS_DIRTY_NODES); 3508 } else { 3509 inode_dec_dirty_pages(inode); 3510 f2fs_remove_dirty_inode(inode); 3511 } 3512 } 3513 3514 clear_cold_data(page); 3515 3516 if (IS_ATOMIC_WRITTEN_PAGE(page)) 3517 return f2fs_drop_inmem_page(inode, page); 3518 3519 f2fs_clear_page_private(page); 3520 } 3521 3522 int f2fs_release_page(struct page *page, gfp_t wait) 3523 { 3524 /* If this is dirty page, keep PagePrivate */ 3525 if (PageDirty(page)) 3526 return 0; 3527 3528 /* This is atomic written page, keep Private */ 3529 if (IS_ATOMIC_WRITTEN_PAGE(page)) 3530 return 0; 3531 3532 clear_cold_data(page); 3533 f2fs_clear_page_private(page); 3534 return 1; 3535 } 3536 3537 static int f2fs_set_data_page_dirty(struct page *page) 3538 { 3539 struct inode *inode = page_file_mapping(page)->host; 3540 3541 trace_f2fs_set_page_dirty(page, DATA); 3542 3543 if (!PageUptodate(page)) 3544 SetPageUptodate(page); 3545 if (PageSwapCache(page)) 3546 return __set_page_dirty_nobuffers(page); 3547 3548 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) { 3549 if (!IS_ATOMIC_WRITTEN_PAGE(page)) { 3550 f2fs_register_inmem_page(inode, page); 3551 return 1; 3552 } 3553 /* 3554 * Previously, this page has been registered, we just 3555 * return here. 3556 */ 3557 return 0; 3558 } 3559 3560 if (!PageDirty(page)) { 3561 __set_page_dirty_nobuffers(page); 3562 f2fs_update_dirty_page(inode, page); 3563 return 1; 3564 } 3565 return 0; 3566 } 3567 3568 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 3569 { 3570 struct inode *inode = mapping->host; 3571 3572 if (f2fs_has_inline_data(inode)) 3573 return 0; 3574 3575 /* make sure allocating whole blocks */ 3576 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 3577 filemap_write_and_wait(mapping); 3578 3579 return generic_block_bmap(mapping, block, get_data_block_bmap); 3580 } 3581 3582 #ifdef CONFIG_MIGRATION 3583 #include <linux/migrate.h> 3584 3585 int f2fs_migrate_page(struct address_space *mapping, 3586 struct page *newpage, struct page *page, enum migrate_mode mode) 3587 { 3588 int rc, extra_count; 3589 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 3590 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page); 3591 3592 BUG_ON(PageWriteback(page)); 3593 3594 /* migrating an atomic written page is safe with the inmem_lock hold */ 3595 if (atomic_written) { 3596 if (mode != MIGRATE_SYNC) 3597 return -EBUSY; 3598 if (!mutex_trylock(&fi->inmem_lock)) 3599 return -EAGAIN; 3600 } 3601 3602 /* one extra reference was held for atomic_write page */ 3603 extra_count = atomic_written ? 1 : 0; 3604 rc = migrate_page_move_mapping(mapping, newpage, 3605 page, extra_count); 3606 if (rc != MIGRATEPAGE_SUCCESS) { 3607 if (atomic_written) 3608 mutex_unlock(&fi->inmem_lock); 3609 return rc; 3610 } 3611 3612 if (atomic_written) { 3613 struct inmem_pages *cur; 3614 list_for_each_entry(cur, &fi->inmem_pages, list) 3615 if (cur->page == page) { 3616 cur->page = newpage; 3617 break; 3618 } 3619 mutex_unlock(&fi->inmem_lock); 3620 put_page(page); 3621 get_page(newpage); 3622 } 3623 3624 if (PagePrivate(page)) { 3625 f2fs_set_page_private(newpage, page_private(page)); 3626 f2fs_clear_page_private(page); 3627 } 3628 3629 if (mode != MIGRATE_SYNC_NO_COPY) 3630 migrate_page_copy(newpage, page); 3631 else 3632 migrate_page_states(newpage, page); 3633 3634 return MIGRATEPAGE_SUCCESS; 3635 } 3636 #endif 3637 3638 #ifdef CONFIG_SWAP 3639 /* Copied from generic_swapfile_activate() to check any holes */ 3640 static int check_swap_activate(struct swap_info_struct *sis, 3641 struct file *swap_file, sector_t *span) 3642 { 3643 struct address_space *mapping = swap_file->f_mapping; 3644 struct inode *inode = mapping->host; 3645 unsigned blocks_per_page; 3646 unsigned long page_no; 3647 unsigned blkbits; 3648 sector_t probe_block; 3649 sector_t last_block; 3650 sector_t lowest_block = -1; 3651 sector_t highest_block = 0; 3652 int nr_extents = 0; 3653 int ret; 3654 3655 blkbits = inode->i_blkbits; 3656 blocks_per_page = PAGE_SIZE >> blkbits; 3657 3658 /* 3659 * Map all the blocks into the extent list. This code doesn't try 3660 * to be very smart. 3661 */ 3662 probe_block = 0; 3663 page_no = 0; 3664 last_block = i_size_read(inode) >> blkbits; 3665 while ((probe_block + blocks_per_page) <= last_block && 3666 page_no < sis->max) { 3667 unsigned block_in_page; 3668 sector_t first_block; 3669 sector_t block = 0; 3670 int err = 0; 3671 3672 cond_resched(); 3673 3674 block = probe_block; 3675 err = bmap(inode, &block); 3676 if (err || !block) 3677 goto bad_bmap; 3678 first_block = block; 3679 3680 /* 3681 * It must be PAGE_SIZE aligned on-disk 3682 */ 3683 if (first_block & (blocks_per_page - 1)) { 3684 probe_block++; 3685 goto reprobe; 3686 } 3687 3688 for (block_in_page = 1; block_in_page < blocks_per_page; 3689 block_in_page++) { 3690 3691 block = probe_block + block_in_page; 3692 err = bmap(inode, &block); 3693 3694 if (err || !block) 3695 goto bad_bmap; 3696 3697 if (block != first_block + block_in_page) { 3698 /* Discontiguity */ 3699 probe_block++; 3700 goto reprobe; 3701 } 3702 } 3703 3704 first_block >>= (PAGE_SHIFT - blkbits); 3705 if (page_no) { /* exclude the header page */ 3706 if (first_block < lowest_block) 3707 lowest_block = first_block; 3708 if (first_block > highest_block) 3709 highest_block = first_block; 3710 } 3711 3712 /* 3713 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks 3714 */ 3715 ret = add_swap_extent(sis, page_no, 1, first_block); 3716 if (ret < 0) 3717 goto out; 3718 nr_extents += ret; 3719 page_no++; 3720 probe_block += blocks_per_page; 3721 reprobe: 3722 continue; 3723 } 3724 ret = nr_extents; 3725 *span = 1 + highest_block - lowest_block; 3726 if (page_no == 0) 3727 page_no = 1; /* force Empty message */ 3728 sis->max = page_no; 3729 sis->pages = page_no - 1; 3730 sis->highest_bit = page_no - 1; 3731 out: 3732 return ret; 3733 bad_bmap: 3734 pr_err("swapon: swapfile has holes\n"); 3735 return -EINVAL; 3736 } 3737 3738 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3739 sector_t *span) 3740 { 3741 struct inode *inode = file_inode(file); 3742 int ret; 3743 3744 if (!S_ISREG(inode->i_mode)) 3745 return -EINVAL; 3746 3747 if (f2fs_readonly(F2FS_I_SB(inode)->sb)) 3748 return -EROFS; 3749 3750 ret = f2fs_convert_inline_inode(inode); 3751 if (ret) 3752 return ret; 3753 3754 if (f2fs_disable_compressed_file(inode)) 3755 return -EINVAL; 3756 3757 ret = check_swap_activate(sis, file, span); 3758 if (ret < 0) 3759 return ret; 3760 3761 set_inode_flag(inode, FI_PIN_FILE); 3762 f2fs_precache_extents(inode); 3763 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3764 return ret; 3765 } 3766 3767 static void f2fs_swap_deactivate(struct file *file) 3768 { 3769 struct inode *inode = file_inode(file); 3770 3771 clear_inode_flag(inode, FI_PIN_FILE); 3772 } 3773 #else 3774 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3775 sector_t *span) 3776 { 3777 return -EOPNOTSUPP; 3778 } 3779 3780 static void f2fs_swap_deactivate(struct file *file) 3781 { 3782 } 3783 #endif 3784 3785 const struct address_space_operations f2fs_dblock_aops = { 3786 .readpage = f2fs_read_data_page, 3787 .readpages = f2fs_read_data_pages, 3788 .writepage = f2fs_write_data_page, 3789 .writepages = f2fs_write_data_pages, 3790 .write_begin = f2fs_write_begin, 3791 .write_end = f2fs_write_end, 3792 .set_page_dirty = f2fs_set_data_page_dirty, 3793 .invalidatepage = f2fs_invalidate_page, 3794 .releasepage = f2fs_release_page, 3795 .direct_IO = f2fs_direct_IO, 3796 .bmap = f2fs_bmap, 3797 .swap_activate = f2fs_swap_activate, 3798 .swap_deactivate = f2fs_swap_deactivate, 3799 #ifdef CONFIG_MIGRATION 3800 .migratepage = f2fs_migrate_page, 3801 #endif 3802 }; 3803 3804 void f2fs_clear_page_cache_dirty_tag(struct page *page) 3805 { 3806 struct address_space *mapping = page_mapping(page); 3807 unsigned long flags; 3808 3809 xa_lock_irqsave(&mapping->i_pages, flags); 3810 __xa_clear_mark(&mapping->i_pages, page_index(page), 3811 PAGECACHE_TAG_DIRTY); 3812 xa_unlock_irqrestore(&mapping->i_pages, flags); 3813 } 3814 3815 int __init f2fs_init_post_read_processing(void) 3816 { 3817 bio_post_read_ctx_cache = 3818 kmem_cache_create("f2fs_bio_post_read_ctx", 3819 sizeof(struct bio_post_read_ctx), 0, 0, NULL); 3820 if (!bio_post_read_ctx_cache) 3821 goto fail; 3822 bio_post_read_ctx_pool = 3823 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 3824 bio_post_read_ctx_cache); 3825 if (!bio_post_read_ctx_pool) 3826 goto fail_free_cache; 3827 return 0; 3828 3829 fail_free_cache: 3830 kmem_cache_destroy(bio_post_read_ctx_cache); 3831 fail: 3832 return -ENOMEM; 3833 } 3834 3835 void f2fs_destroy_post_read_processing(void) 3836 { 3837 mempool_destroy(bio_post_read_ctx_pool); 3838 kmem_cache_destroy(bio_post_read_ctx_cache); 3839 } 3840 3841 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) 3842 { 3843 if (!f2fs_sb_has_encrypt(sbi) && 3844 !f2fs_sb_has_verity(sbi) && 3845 !f2fs_sb_has_compression(sbi)) 3846 return 0; 3847 3848 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq", 3849 WQ_UNBOUND | WQ_HIGHPRI, 3850 num_online_cpus()); 3851 if (!sbi->post_read_wq) 3852 return -ENOMEM; 3853 return 0; 3854 } 3855 3856 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) 3857 { 3858 if (sbi->post_read_wq) 3859 destroy_workqueue(sbi->post_read_wq); 3860 } 3861 3862 int __init f2fs_init_bio_entry_cache(void) 3863 { 3864 bio_entry_slab = f2fs_kmem_cache_create("bio_entry_slab", 3865 sizeof(struct bio_entry)); 3866 if (!bio_entry_slab) 3867 return -ENOMEM; 3868 return 0; 3869 } 3870 3871 void f2fs_destroy_bio_entry_cache(void) 3872 { 3873 kmem_cache_destroy(bio_entry_slab); 3874 } 3875