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/sched/mm.h> 11 #include <linux/mpage.h> 12 #include <linux/writeback.h> 13 #include <linux/pagevec.h> 14 #include <linux/blkdev.h> 15 #include <linux/bio.h> 16 #include <linux/blk-crypto.h> 17 #include <linux/swap.h> 18 #include <linux/prefetch.h> 19 #include <linux/uio.h> 20 #include <linux/sched/signal.h> 21 #include <linux/fiemap.h> 22 #include <linux/iomap.h> 23 24 #include "f2fs.h" 25 #include "node.h" 26 #include "segment.h" 27 #include "iostat.h" 28 #include <trace/events/f2fs.h> 29 30 #define NUM_PREALLOC_POST_READ_CTXS 128 31 32 static struct kmem_cache *bio_post_read_ctx_cache; 33 static struct kmem_cache *bio_entry_slab; 34 static mempool_t *bio_post_read_ctx_pool; 35 static struct bio_set f2fs_bioset; 36 37 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE 38 39 int __init f2fs_init_bioset(void) 40 { 41 return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE, 42 0, BIOSET_NEED_BVECS); 43 } 44 45 void f2fs_destroy_bioset(void) 46 { 47 bioset_exit(&f2fs_bioset); 48 } 49 50 bool f2fs_is_cp_guaranteed(struct page *page) 51 { 52 struct address_space *mapping = page->mapping; 53 struct inode *inode; 54 struct f2fs_sb_info *sbi; 55 56 if (!mapping) 57 return false; 58 59 inode = mapping->host; 60 sbi = F2FS_I_SB(inode); 61 62 if (inode->i_ino == F2FS_META_INO(sbi) || 63 inode->i_ino == F2FS_NODE_INO(sbi) || 64 S_ISDIR(inode->i_mode)) 65 return true; 66 67 if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) || 68 page_private_gcing(page)) 69 return true; 70 return false; 71 } 72 73 static enum count_type __read_io_type(struct page *page) 74 { 75 struct address_space *mapping = page_file_mapping(page); 76 77 if (mapping) { 78 struct inode *inode = mapping->host; 79 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 80 81 if (inode->i_ino == F2FS_META_INO(sbi)) 82 return F2FS_RD_META; 83 84 if (inode->i_ino == F2FS_NODE_INO(sbi)) 85 return F2FS_RD_NODE; 86 } 87 return F2FS_RD_DATA; 88 } 89 90 /* postprocessing steps for read bios */ 91 enum bio_post_read_step { 92 #ifdef CONFIG_FS_ENCRYPTION 93 STEP_DECRYPT = BIT(0), 94 #else 95 STEP_DECRYPT = 0, /* compile out the decryption-related code */ 96 #endif 97 #ifdef CONFIG_F2FS_FS_COMPRESSION 98 STEP_DECOMPRESS = BIT(1), 99 #else 100 STEP_DECOMPRESS = 0, /* compile out the decompression-related code */ 101 #endif 102 #ifdef CONFIG_FS_VERITY 103 STEP_VERITY = BIT(2), 104 #else 105 STEP_VERITY = 0, /* compile out the verity-related code */ 106 #endif 107 }; 108 109 struct bio_post_read_ctx { 110 struct bio *bio; 111 struct f2fs_sb_info *sbi; 112 struct work_struct work; 113 unsigned int enabled_steps; 114 /* 115 * decompression_attempted keeps track of whether 116 * f2fs_end_read_compressed_page() has been called on the pages in the 117 * bio that belong to a compressed cluster yet. 118 */ 119 bool decompression_attempted; 120 block_t fs_blkaddr; 121 }; 122 123 /* 124 * Update and unlock a bio's pages, and free the bio. 125 * 126 * This marks pages up-to-date only if there was no error in the bio (I/O error, 127 * decryption error, or verity error), as indicated by bio->bi_status. 128 * 129 * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk) 130 * aren't marked up-to-date here, as decompression is done on a per-compression- 131 * cluster basis rather than a per-bio basis. Instead, we only must do two 132 * things for each compressed page here: call f2fs_end_read_compressed_page() 133 * with failed=true if an error occurred before it would have normally gotten 134 * called (i.e., I/O error or decryption error, but *not* verity error), and 135 * release the bio's reference to the decompress_io_ctx of the page's cluster. 136 */ 137 static void f2fs_finish_read_bio(struct bio *bio, bool in_task) 138 { 139 struct bio_vec *bv; 140 struct bvec_iter_all iter_all; 141 struct bio_post_read_ctx *ctx = bio->bi_private; 142 143 bio_for_each_segment_all(bv, bio, iter_all) { 144 struct page *page = bv->bv_page; 145 146 if (f2fs_is_compressed_page(page)) { 147 if (ctx && !ctx->decompression_attempted) 148 f2fs_end_read_compressed_page(page, true, 0, 149 in_task); 150 f2fs_put_page_dic(page, in_task); 151 continue; 152 } 153 154 if (bio->bi_status) 155 ClearPageUptodate(page); 156 else 157 SetPageUptodate(page); 158 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 159 unlock_page(page); 160 } 161 162 if (ctx) 163 mempool_free(ctx, bio_post_read_ctx_pool); 164 bio_put(bio); 165 } 166 167 static void f2fs_verify_bio(struct work_struct *work) 168 { 169 struct bio_post_read_ctx *ctx = 170 container_of(work, struct bio_post_read_ctx, work); 171 struct bio *bio = ctx->bio; 172 bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS); 173 174 /* 175 * fsverity_verify_bio() may call readahead() again, and while verity 176 * will be disabled for this, decryption and/or decompression may still 177 * be needed, resulting in another bio_post_read_ctx being allocated. 178 * So to prevent deadlocks we need to release the current ctx to the 179 * mempool first. This assumes that verity is the last post-read step. 180 */ 181 mempool_free(ctx, bio_post_read_ctx_pool); 182 bio->bi_private = NULL; 183 184 /* 185 * Verify the bio's pages with fs-verity. Exclude compressed pages, 186 * as those were handled separately by f2fs_end_read_compressed_page(). 187 */ 188 if (may_have_compressed_pages) { 189 struct bio_vec *bv; 190 struct bvec_iter_all iter_all; 191 192 bio_for_each_segment_all(bv, bio, iter_all) { 193 struct page *page = bv->bv_page; 194 195 if (!f2fs_is_compressed_page(page) && 196 !fsverity_verify_page(page)) { 197 bio->bi_status = BLK_STS_IOERR; 198 break; 199 } 200 } 201 } else { 202 fsverity_verify_bio(bio); 203 } 204 205 f2fs_finish_read_bio(bio, true); 206 } 207 208 /* 209 * If the bio's data needs to be verified with fs-verity, then enqueue the 210 * verity work for the bio. Otherwise finish the bio now. 211 * 212 * Note that to avoid deadlocks, the verity work can't be done on the 213 * decryption/decompression workqueue. This is because verifying the data pages 214 * can involve reading verity metadata pages from the file, and these verity 215 * metadata pages may be encrypted and/or compressed. 216 */ 217 static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task) 218 { 219 struct bio_post_read_ctx *ctx = bio->bi_private; 220 221 if (ctx && (ctx->enabled_steps & STEP_VERITY)) { 222 INIT_WORK(&ctx->work, f2fs_verify_bio); 223 fsverity_enqueue_verify_work(&ctx->work); 224 } else { 225 f2fs_finish_read_bio(bio, in_task); 226 } 227 } 228 229 /* 230 * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last 231 * remaining page was read by @ctx->bio. 232 * 233 * Note that a bio may span clusters (even a mix of compressed and uncompressed 234 * clusters) or be for just part of a cluster. STEP_DECOMPRESS just indicates 235 * that the bio includes at least one compressed page. The actual decompression 236 * is done on a per-cluster basis, not a per-bio basis. 237 */ 238 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx, 239 bool in_task) 240 { 241 struct bio_vec *bv; 242 struct bvec_iter_all iter_all; 243 bool all_compressed = true; 244 block_t blkaddr = ctx->fs_blkaddr; 245 246 bio_for_each_segment_all(bv, ctx->bio, iter_all) { 247 struct page *page = bv->bv_page; 248 249 if (f2fs_is_compressed_page(page)) 250 f2fs_end_read_compressed_page(page, false, blkaddr, 251 in_task); 252 else 253 all_compressed = false; 254 255 blkaddr++; 256 } 257 258 ctx->decompression_attempted = true; 259 260 /* 261 * Optimization: if all the bio's pages are compressed, then scheduling 262 * the per-bio verity work is unnecessary, as verity will be fully 263 * handled at the compression cluster level. 264 */ 265 if (all_compressed) 266 ctx->enabled_steps &= ~STEP_VERITY; 267 } 268 269 static void f2fs_post_read_work(struct work_struct *work) 270 { 271 struct bio_post_read_ctx *ctx = 272 container_of(work, struct bio_post_read_ctx, work); 273 struct bio *bio = ctx->bio; 274 275 if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) { 276 f2fs_finish_read_bio(bio, true); 277 return; 278 } 279 280 if (ctx->enabled_steps & STEP_DECOMPRESS) 281 f2fs_handle_step_decompress(ctx, true); 282 283 f2fs_verify_and_finish_bio(bio, true); 284 } 285 286 static void f2fs_read_end_io(struct bio *bio) 287 { 288 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio)); 289 struct bio_post_read_ctx *ctx; 290 bool intask = in_task(); 291 292 iostat_update_and_unbind_ctx(bio); 293 ctx = bio->bi_private; 294 295 if (time_to_inject(sbi, FAULT_READ_IO)) 296 bio->bi_status = BLK_STS_IOERR; 297 298 if (bio->bi_status) { 299 f2fs_finish_read_bio(bio, intask); 300 return; 301 } 302 303 if (ctx) { 304 unsigned int enabled_steps = ctx->enabled_steps & 305 (STEP_DECRYPT | STEP_DECOMPRESS); 306 307 /* 308 * If we have only decompression step between decompression and 309 * decrypt, we don't need post processing for this. 310 */ 311 if (enabled_steps == STEP_DECOMPRESS && 312 !f2fs_low_mem_mode(sbi)) { 313 f2fs_handle_step_decompress(ctx, intask); 314 } else if (enabled_steps) { 315 INIT_WORK(&ctx->work, f2fs_post_read_work); 316 queue_work(ctx->sbi->post_read_wq, &ctx->work); 317 return; 318 } 319 } 320 321 f2fs_verify_and_finish_bio(bio, intask); 322 } 323 324 static void f2fs_write_end_io(struct bio *bio) 325 { 326 struct f2fs_sb_info *sbi; 327 struct bio_vec *bvec; 328 struct bvec_iter_all iter_all; 329 330 iostat_update_and_unbind_ctx(bio); 331 sbi = bio->bi_private; 332 333 if (time_to_inject(sbi, FAULT_WRITE_IO)) 334 bio->bi_status = BLK_STS_IOERR; 335 336 bio_for_each_segment_all(bvec, bio, iter_all) { 337 struct page *page = bvec->bv_page; 338 enum count_type type = WB_DATA_TYPE(page, false); 339 340 fscrypt_finalize_bounce_page(&page); 341 342 #ifdef CONFIG_F2FS_FS_COMPRESSION 343 if (f2fs_is_compressed_page(page)) { 344 f2fs_compress_write_end_io(bio, page); 345 continue; 346 } 347 #endif 348 349 if (unlikely(bio->bi_status)) { 350 mapping_set_error(page->mapping, -EIO); 351 if (type == F2FS_WB_CP_DATA) 352 f2fs_stop_checkpoint(sbi, true, 353 STOP_CP_REASON_WRITE_FAIL); 354 } 355 356 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 357 page_folio(page)->index != nid_of_node(page)); 358 359 dec_page_count(sbi, type); 360 if (f2fs_in_warm_node_list(sbi, page)) 361 f2fs_del_fsync_node_entry(sbi, page); 362 clear_page_private_gcing(page); 363 end_page_writeback(page); 364 } 365 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 366 wq_has_sleeper(&sbi->cp_wait)) 367 wake_up(&sbi->cp_wait); 368 369 bio_put(bio); 370 } 371 372 #ifdef CONFIG_BLK_DEV_ZONED 373 static void f2fs_zone_write_end_io(struct bio *bio) 374 { 375 struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private; 376 377 bio->bi_private = io->bi_private; 378 complete(&io->zone_wait); 379 f2fs_write_end_io(bio); 380 } 381 #endif 382 383 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 384 block_t blk_addr, sector_t *sector) 385 { 386 struct block_device *bdev = sbi->sb->s_bdev; 387 int i; 388 389 if (f2fs_is_multi_device(sbi)) { 390 for (i = 0; i < sbi->s_ndevs; i++) { 391 if (FDEV(i).start_blk <= blk_addr && 392 FDEV(i).end_blk >= blk_addr) { 393 blk_addr -= FDEV(i).start_blk; 394 bdev = FDEV(i).bdev; 395 break; 396 } 397 } 398 } 399 400 if (sector) 401 *sector = SECTOR_FROM_BLOCK(blk_addr); 402 return bdev; 403 } 404 405 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 406 { 407 int i; 408 409 if (!f2fs_is_multi_device(sbi)) 410 return 0; 411 412 for (i = 0; i < sbi->s_ndevs; i++) 413 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 414 return i; 415 return 0; 416 } 417 418 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio) 419 { 420 unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0); 421 unsigned int fua_flag, meta_flag, io_flag; 422 blk_opf_t op_flags = 0; 423 424 if (fio->op != REQ_OP_WRITE) 425 return 0; 426 if (fio->type == DATA) 427 io_flag = fio->sbi->data_io_flag; 428 else if (fio->type == NODE) 429 io_flag = fio->sbi->node_io_flag; 430 else 431 return 0; 432 433 fua_flag = io_flag & temp_mask; 434 meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask; 435 436 /* 437 * data/node io flag bits per temp: 438 * REQ_META | REQ_FUA | 439 * 5 | 4 | 3 | 2 | 1 | 0 | 440 * Cold | Warm | Hot | Cold | Warm | Hot | 441 */ 442 if (BIT(fio->temp) & meta_flag) 443 op_flags |= REQ_META; 444 if (BIT(fio->temp) & fua_flag) 445 op_flags |= REQ_FUA; 446 return op_flags; 447 } 448 449 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) 450 { 451 struct f2fs_sb_info *sbi = fio->sbi; 452 struct block_device *bdev; 453 sector_t sector; 454 struct bio *bio; 455 456 bdev = f2fs_target_device(sbi, fio->new_blkaddr, §or); 457 bio = bio_alloc_bioset(bdev, npages, 458 fio->op | fio->op_flags | f2fs_io_flags(fio), 459 GFP_NOIO, &f2fs_bioset); 460 bio->bi_iter.bi_sector = sector; 461 if (is_read_io(fio->op)) { 462 bio->bi_end_io = f2fs_read_end_io; 463 bio->bi_private = NULL; 464 } else { 465 bio->bi_end_io = f2fs_write_end_io; 466 bio->bi_private = sbi; 467 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, 468 fio->type, fio->temp); 469 } 470 iostat_alloc_and_bind_ctx(sbi, bio, NULL); 471 472 if (fio->io_wbc) 473 wbc_init_bio(fio->io_wbc, bio); 474 475 return bio; 476 } 477 478 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode, 479 pgoff_t first_idx, 480 const struct f2fs_io_info *fio, 481 gfp_t gfp_mask) 482 { 483 /* 484 * The f2fs garbage collector sets ->encrypted_page when it wants to 485 * read/write raw data without encryption. 486 */ 487 if (!fio || !fio->encrypted_page) 488 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask); 489 } 490 491 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode, 492 pgoff_t next_idx, 493 const struct f2fs_io_info *fio) 494 { 495 /* 496 * The f2fs garbage collector sets ->encrypted_page when it wants to 497 * read/write raw data without encryption. 498 */ 499 if (fio && fio->encrypted_page) 500 return !bio_has_crypt_ctx(bio); 501 502 return fscrypt_mergeable_bio(bio, inode, next_idx); 503 } 504 505 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio, 506 enum page_type type) 507 { 508 WARN_ON_ONCE(!is_read_io(bio_op(bio))); 509 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 510 511 iostat_update_submit_ctx(bio, type); 512 submit_bio(bio); 513 } 514 515 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio, 516 enum page_type type) 517 { 518 WARN_ON_ONCE(is_read_io(bio_op(bio))); 519 520 if (f2fs_lfs_mode(sbi) && current->plug && PAGE_TYPE_ON_MAIN(type)) 521 blk_finish_plug(current->plug); 522 523 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 524 iostat_update_submit_ctx(bio, type); 525 submit_bio(bio); 526 } 527 528 static void __submit_merged_bio(struct f2fs_bio_info *io) 529 { 530 struct f2fs_io_info *fio = &io->fio; 531 532 if (!io->bio) 533 return; 534 535 if (is_read_io(fio->op)) { 536 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 537 f2fs_submit_read_bio(io->sbi, io->bio, fio->type); 538 } else { 539 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 540 f2fs_submit_write_bio(io->sbi, io->bio, fio->type); 541 } 542 io->bio = NULL; 543 } 544 545 static bool __has_merged_page(struct bio *bio, struct inode *inode, 546 struct page *page, nid_t ino) 547 { 548 struct bio_vec *bvec; 549 struct bvec_iter_all iter_all; 550 551 if (!bio) 552 return false; 553 554 if (!inode && !page && !ino) 555 return true; 556 557 bio_for_each_segment_all(bvec, bio, iter_all) { 558 struct page *target = bvec->bv_page; 559 560 if (fscrypt_is_bounce_page(target)) { 561 target = fscrypt_pagecache_page(target); 562 if (IS_ERR(target)) 563 continue; 564 } 565 if (f2fs_is_compressed_page(target)) { 566 target = f2fs_compress_control_page(target); 567 if (IS_ERR(target)) 568 continue; 569 } 570 571 if (inode && inode == target->mapping->host) 572 return true; 573 if (page && page == target) 574 return true; 575 if (ino && ino == ino_of_node(target)) 576 return true; 577 } 578 579 return false; 580 } 581 582 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi) 583 { 584 int i; 585 586 for (i = 0; i < NR_PAGE_TYPE; i++) { 587 int n = (i == META) ? 1 : NR_TEMP_TYPE; 588 int j; 589 590 sbi->write_io[i] = f2fs_kmalloc(sbi, 591 array_size(n, sizeof(struct f2fs_bio_info)), 592 GFP_KERNEL); 593 if (!sbi->write_io[i]) 594 return -ENOMEM; 595 596 for (j = HOT; j < n; j++) { 597 struct f2fs_bio_info *io = &sbi->write_io[i][j]; 598 599 init_f2fs_rwsem(&io->io_rwsem); 600 io->sbi = sbi; 601 io->bio = NULL; 602 io->last_block_in_bio = 0; 603 spin_lock_init(&io->io_lock); 604 INIT_LIST_HEAD(&io->io_list); 605 INIT_LIST_HEAD(&io->bio_list); 606 init_f2fs_rwsem(&io->bio_list_lock); 607 #ifdef CONFIG_BLK_DEV_ZONED 608 init_completion(&io->zone_wait); 609 io->zone_pending_bio = NULL; 610 io->bi_private = NULL; 611 #endif 612 } 613 } 614 615 return 0; 616 } 617 618 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 619 enum page_type type, enum temp_type temp) 620 { 621 enum page_type btype = PAGE_TYPE_OF_BIO(type); 622 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 623 624 f2fs_down_write(&io->io_rwsem); 625 626 if (!io->bio) 627 goto unlock_out; 628 629 /* change META to META_FLUSH in the checkpoint procedure */ 630 if (type >= META_FLUSH) { 631 io->fio.type = META_FLUSH; 632 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC; 633 if (!test_opt(sbi, NOBARRIER)) 634 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA; 635 } 636 __submit_merged_bio(io); 637 unlock_out: 638 f2fs_up_write(&io->io_rwsem); 639 } 640 641 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 642 struct inode *inode, struct page *page, 643 nid_t ino, enum page_type type, bool force) 644 { 645 enum temp_type temp; 646 bool ret = true; 647 648 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 649 if (!force) { 650 enum page_type btype = PAGE_TYPE_OF_BIO(type); 651 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 652 653 f2fs_down_read(&io->io_rwsem); 654 ret = __has_merged_page(io->bio, inode, page, ino); 655 f2fs_up_read(&io->io_rwsem); 656 } 657 if (ret) 658 __f2fs_submit_merged_write(sbi, type, temp); 659 660 /* TODO: use HOT temp only for meta pages now. */ 661 if (type >= META) 662 break; 663 } 664 } 665 666 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 667 { 668 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true); 669 } 670 671 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 672 struct inode *inode, struct page *page, 673 nid_t ino, enum page_type type) 674 { 675 __submit_merged_write_cond(sbi, inode, page, ino, type, false); 676 } 677 678 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 679 { 680 f2fs_submit_merged_write(sbi, DATA); 681 f2fs_submit_merged_write(sbi, NODE); 682 f2fs_submit_merged_write(sbi, META); 683 } 684 685 /* 686 * Fill the locked page with data located in the block address. 687 * A caller needs to unlock the page on failure. 688 */ 689 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 690 { 691 struct bio *bio; 692 struct page *page = fio->encrypted_page ? 693 fio->encrypted_page : fio->page; 694 695 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 696 fio->is_por ? META_POR : (__is_meta_io(fio) ? 697 META_GENERIC : DATA_GENERIC_ENHANCE))) 698 return -EFSCORRUPTED; 699 700 trace_f2fs_submit_page_bio(page, fio); 701 702 /* Allocate a new bio */ 703 bio = __bio_alloc(fio, 1); 704 705 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host, 706 page_folio(fio->page)->index, fio, GFP_NOIO); 707 708 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 709 bio_put(bio); 710 return -EFAULT; 711 } 712 713 if (fio->io_wbc && !is_read_io(fio->op)) 714 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE); 715 716 inc_page_count(fio->sbi, is_read_io(fio->op) ? 717 __read_io_type(page) : WB_DATA_TYPE(fio->page, false)); 718 719 if (is_read_io(bio_op(bio))) 720 f2fs_submit_read_bio(fio->sbi, bio, fio->type); 721 else 722 f2fs_submit_write_bio(fio->sbi, bio, fio->type); 723 return 0; 724 } 725 726 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 727 block_t last_blkaddr, block_t cur_blkaddr) 728 { 729 if (unlikely(sbi->max_io_bytes && 730 bio->bi_iter.bi_size >= sbi->max_io_bytes)) 731 return false; 732 if (last_blkaddr + 1 != cur_blkaddr) 733 return false; 734 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL); 735 } 736 737 static bool io_type_is_mergeable(struct f2fs_bio_info *io, 738 struct f2fs_io_info *fio) 739 { 740 if (io->fio.op != fio->op) 741 return false; 742 return io->fio.op_flags == fio->op_flags; 743 } 744 745 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 746 struct f2fs_bio_info *io, 747 struct f2fs_io_info *fio, 748 block_t last_blkaddr, 749 block_t cur_blkaddr) 750 { 751 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr)) 752 return false; 753 return io_type_is_mergeable(io, fio); 754 } 755 756 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio, 757 struct page *page, enum temp_type temp) 758 { 759 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 760 struct bio_entry *be; 761 762 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL); 763 be->bio = bio; 764 bio_get(bio); 765 766 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE) 767 f2fs_bug_on(sbi, 1); 768 769 f2fs_down_write(&io->bio_list_lock); 770 list_add_tail(&be->list, &io->bio_list); 771 f2fs_up_write(&io->bio_list_lock); 772 } 773 774 static void del_bio_entry(struct bio_entry *be) 775 { 776 list_del(&be->list); 777 kmem_cache_free(bio_entry_slab, be); 778 } 779 780 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio, 781 struct page *page) 782 { 783 struct f2fs_sb_info *sbi = fio->sbi; 784 enum temp_type temp; 785 bool found = false; 786 int ret = -EAGAIN; 787 788 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 789 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 790 struct list_head *head = &io->bio_list; 791 struct bio_entry *be; 792 793 f2fs_down_write(&io->bio_list_lock); 794 list_for_each_entry(be, head, list) { 795 if (be->bio != *bio) 796 continue; 797 798 found = true; 799 800 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio, 801 *fio->last_block, 802 fio->new_blkaddr)); 803 if (f2fs_crypt_mergeable_bio(*bio, 804 fio->page->mapping->host, 805 page_folio(fio->page)->index, fio) && 806 bio_add_page(*bio, page, PAGE_SIZE, 0) == 807 PAGE_SIZE) { 808 ret = 0; 809 break; 810 } 811 812 /* page can't be merged into bio; submit the bio */ 813 del_bio_entry(be); 814 f2fs_submit_write_bio(sbi, *bio, DATA); 815 break; 816 } 817 f2fs_up_write(&io->bio_list_lock); 818 } 819 820 if (ret) { 821 bio_put(*bio); 822 *bio = NULL; 823 } 824 825 return ret; 826 } 827 828 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi, 829 struct bio **bio, struct page *page) 830 { 831 enum temp_type temp; 832 bool found = false; 833 struct bio *target = bio ? *bio : NULL; 834 835 f2fs_bug_on(sbi, !target && !page); 836 837 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 838 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 839 struct list_head *head = &io->bio_list; 840 struct bio_entry *be; 841 842 if (list_empty(head)) 843 continue; 844 845 f2fs_down_read(&io->bio_list_lock); 846 list_for_each_entry(be, head, list) { 847 if (target) 848 found = (target == be->bio); 849 else 850 found = __has_merged_page(be->bio, NULL, 851 page, 0); 852 if (found) 853 break; 854 } 855 f2fs_up_read(&io->bio_list_lock); 856 857 if (!found) 858 continue; 859 860 found = false; 861 862 f2fs_down_write(&io->bio_list_lock); 863 list_for_each_entry(be, head, list) { 864 if (target) 865 found = (target == be->bio); 866 else 867 found = __has_merged_page(be->bio, NULL, 868 page, 0); 869 if (found) { 870 target = be->bio; 871 del_bio_entry(be); 872 break; 873 } 874 } 875 f2fs_up_write(&io->bio_list_lock); 876 } 877 878 if (found) 879 f2fs_submit_write_bio(sbi, target, DATA); 880 if (bio && *bio) { 881 bio_put(*bio); 882 *bio = NULL; 883 } 884 } 885 886 int f2fs_merge_page_bio(struct f2fs_io_info *fio) 887 { 888 struct bio *bio = *fio->bio; 889 struct page *page = fio->encrypted_page ? 890 fio->encrypted_page : fio->page; 891 892 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 893 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) 894 return -EFSCORRUPTED; 895 896 trace_f2fs_submit_page_bio(page, fio); 897 898 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block, 899 fio->new_blkaddr)) 900 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL); 901 alloc_new: 902 if (!bio) { 903 bio = __bio_alloc(fio, BIO_MAX_VECS); 904 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host, 905 page_folio(fio->page)->index, fio, GFP_NOIO); 906 907 add_bio_entry(fio->sbi, bio, page, fio->temp); 908 } else { 909 if (add_ipu_page(fio, &bio, page)) 910 goto alloc_new; 911 } 912 913 if (fio->io_wbc) 914 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE); 915 916 inc_page_count(fio->sbi, WB_DATA_TYPE(page, false)); 917 918 *fio->last_block = fio->new_blkaddr; 919 *fio->bio = bio; 920 921 return 0; 922 } 923 924 #ifdef CONFIG_BLK_DEV_ZONED 925 static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr) 926 { 927 struct block_device *bdev = sbi->sb->s_bdev; 928 int devi = 0; 929 930 if (f2fs_is_multi_device(sbi)) { 931 devi = f2fs_target_device_index(sbi, blkaddr); 932 if (blkaddr < FDEV(devi).start_blk || 933 blkaddr > FDEV(devi).end_blk) { 934 f2fs_err(sbi, "Invalid block %x", blkaddr); 935 return false; 936 } 937 blkaddr -= FDEV(devi).start_blk; 938 bdev = FDEV(devi).bdev; 939 } 940 return bdev_is_zoned(bdev) && 941 f2fs_blkz_is_seq(sbi, devi, blkaddr) && 942 (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1); 943 } 944 #endif 945 946 void f2fs_submit_page_write(struct f2fs_io_info *fio) 947 { 948 struct f2fs_sb_info *sbi = fio->sbi; 949 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 950 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 951 struct page *bio_page; 952 enum count_type type; 953 954 f2fs_bug_on(sbi, is_read_io(fio->op)); 955 956 f2fs_down_write(&io->io_rwsem); 957 next: 958 #ifdef CONFIG_BLK_DEV_ZONED 959 if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) { 960 wait_for_completion_io(&io->zone_wait); 961 bio_put(io->zone_pending_bio); 962 io->zone_pending_bio = NULL; 963 io->bi_private = NULL; 964 } 965 #endif 966 967 if (fio->in_list) { 968 spin_lock(&io->io_lock); 969 if (list_empty(&io->io_list)) { 970 spin_unlock(&io->io_lock); 971 goto out; 972 } 973 fio = list_first_entry(&io->io_list, 974 struct f2fs_io_info, list); 975 list_del(&fio->list); 976 spin_unlock(&io->io_lock); 977 } 978 979 verify_fio_blkaddr(fio); 980 981 if (fio->encrypted_page) 982 bio_page = fio->encrypted_page; 983 else if (fio->compressed_page) 984 bio_page = fio->compressed_page; 985 else 986 bio_page = fio->page; 987 988 /* set submitted = true as a return value */ 989 fio->submitted = 1; 990 991 type = WB_DATA_TYPE(bio_page, fio->compressed_page); 992 inc_page_count(sbi, type); 993 994 if (io->bio && 995 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio, 996 fio->new_blkaddr) || 997 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host, 998 page_folio(bio_page)->index, fio))) 999 __submit_merged_bio(io); 1000 alloc_new: 1001 if (io->bio == NULL) { 1002 io->bio = __bio_alloc(fio, BIO_MAX_VECS); 1003 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host, 1004 page_folio(bio_page)->index, fio, GFP_NOIO); 1005 io->fio = *fio; 1006 } 1007 1008 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 1009 __submit_merged_bio(io); 1010 goto alloc_new; 1011 } 1012 1013 if (fio->io_wbc) 1014 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE); 1015 1016 io->last_block_in_bio = fio->new_blkaddr; 1017 1018 trace_f2fs_submit_page_write(fio->page, fio); 1019 #ifdef CONFIG_BLK_DEV_ZONED 1020 if (f2fs_sb_has_blkzoned(sbi) && btype < META && 1021 is_end_zone_blkaddr(sbi, fio->new_blkaddr)) { 1022 bio_get(io->bio); 1023 reinit_completion(&io->zone_wait); 1024 io->bi_private = io->bio->bi_private; 1025 io->bio->bi_private = io; 1026 io->bio->bi_end_io = f2fs_zone_write_end_io; 1027 io->zone_pending_bio = io->bio; 1028 __submit_merged_bio(io); 1029 } 1030 #endif 1031 if (fio->in_list) 1032 goto next; 1033 out: 1034 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 1035 !f2fs_is_checkpoint_ready(sbi)) 1036 __submit_merged_bio(io); 1037 f2fs_up_write(&io->io_rwsem); 1038 } 1039 1040 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 1041 unsigned nr_pages, blk_opf_t op_flag, 1042 pgoff_t first_idx, bool for_write) 1043 { 1044 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1045 struct bio *bio; 1046 struct bio_post_read_ctx *ctx = NULL; 1047 unsigned int post_read_steps = 0; 1048 sector_t sector; 1049 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or); 1050 1051 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages), 1052 REQ_OP_READ | op_flag, 1053 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset); 1054 if (!bio) 1055 return ERR_PTR(-ENOMEM); 1056 bio->bi_iter.bi_sector = sector; 1057 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS); 1058 bio->bi_end_io = f2fs_read_end_io; 1059 1060 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 1061 post_read_steps |= STEP_DECRYPT; 1062 1063 if (f2fs_need_verity(inode, first_idx)) 1064 post_read_steps |= STEP_VERITY; 1065 1066 /* 1067 * STEP_DECOMPRESS is handled specially, since a compressed file might 1068 * contain both compressed and uncompressed clusters. We'll allocate a 1069 * bio_post_read_ctx if the file is compressed, but the caller is 1070 * responsible for enabling STEP_DECOMPRESS if it's actually needed. 1071 */ 1072 1073 if (post_read_steps || f2fs_compressed_file(inode)) { 1074 /* Due to the mempool, this never fails. */ 1075 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 1076 ctx->bio = bio; 1077 ctx->sbi = sbi; 1078 ctx->enabled_steps = post_read_steps; 1079 ctx->fs_blkaddr = blkaddr; 1080 ctx->decompression_attempted = false; 1081 bio->bi_private = ctx; 1082 } 1083 iostat_alloc_and_bind_ctx(sbi, bio, ctx); 1084 1085 return bio; 1086 } 1087 1088 /* This can handle encryption stuffs */ 1089 static int f2fs_submit_page_read(struct inode *inode, struct folio *folio, 1090 block_t blkaddr, blk_opf_t op_flags, 1091 bool for_write) 1092 { 1093 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1094 struct bio *bio; 1095 1096 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags, 1097 folio->index, for_write); 1098 if (IS_ERR(bio)) 1099 return PTR_ERR(bio); 1100 1101 /* wait for GCed page writeback via META_MAPPING */ 1102 f2fs_wait_on_block_writeback(inode, blkaddr); 1103 1104 if (!bio_add_folio(bio, folio, PAGE_SIZE, 0)) { 1105 iostat_update_and_unbind_ctx(bio); 1106 if (bio->bi_private) 1107 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 1108 bio_put(bio); 1109 return -EFAULT; 1110 } 1111 inc_page_count(sbi, F2FS_RD_DATA); 1112 f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE); 1113 f2fs_submit_read_bio(sbi, bio, DATA); 1114 return 0; 1115 } 1116 1117 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1118 { 1119 __le32 *addr = get_dnode_addr(dn->inode, dn->node_page); 1120 1121 dn->data_blkaddr = blkaddr; 1122 addr[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 1123 } 1124 1125 /* 1126 * Lock ordering for the change of data block address: 1127 * ->data_page 1128 * ->node_page 1129 * update block addresses in the node page 1130 */ 1131 void f2fs_set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1132 { 1133 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1134 __set_data_blkaddr(dn, blkaddr); 1135 if (set_page_dirty(dn->node_page)) 1136 dn->node_changed = true; 1137 } 1138 1139 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1140 { 1141 f2fs_set_data_blkaddr(dn, blkaddr); 1142 f2fs_update_read_extent_cache(dn); 1143 } 1144 1145 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 1146 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 1147 { 1148 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1149 int err; 1150 1151 if (!count) 1152 return 0; 1153 1154 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1155 return -EPERM; 1156 err = inc_valid_block_count(sbi, dn->inode, &count, true); 1157 if (unlikely(err)) 1158 return err; 1159 1160 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 1161 dn->ofs_in_node, count); 1162 1163 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1164 1165 for (; count > 0; dn->ofs_in_node++) { 1166 block_t blkaddr = f2fs_data_blkaddr(dn); 1167 1168 if (blkaddr == NULL_ADDR) { 1169 __set_data_blkaddr(dn, NEW_ADDR); 1170 count--; 1171 } 1172 } 1173 1174 if (set_page_dirty(dn->node_page)) 1175 dn->node_changed = true; 1176 return 0; 1177 } 1178 1179 /* Should keep dn->ofs_in_node unchanged */ 1180 int f2fs_reserve_new_block(struct dnode_of_data *dn) 1181 { 1182 unsigned int ofs_in_node = dn->ofs_in_node; 1183 int ret; 1184 1185 ret = f2fs_reserve_new_blocks(dn, 1); 1186 dn->ofs_in_node = ofs_in_node; 1187 return ret; 1188 } 1189 1190 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 1191 { 1192 bool need_put = dn->inode_page ? false : true; 1193 int err; 1194 1195 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE); 1196 if (err) 1197 return err; 1198 1199 if (dn->data_blkaddr == NULL_ADDR) 1200 err = f2fs_reserve_new_block(dn); 1201 if (err || need_put) 1202 f2fs_put_dnode(dn); 1203 return err; 1204 } 1205 1206 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, 1207 blk_opf_t op_flags, bool for_write, 1208 pgoff_t *next_pgofs) 1209 { 1210 struct address_space *mapping = inode->i_mapping; 1211 struct dnode_of_data dn; 1212 struct page *page; 1213 int err; 1214 1215 page = f2fs_grab_cache_page(mapping, index, for_write); 1216 if (!page) 1217 return ERR_PTR(-ENOMEM); 1218 1219 if (f2fs_lookup_read_extent_cache_block(inode, index, 1220 &dn.data_blkaddr)) { 1221 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr, 1222 DATA_GENERIC_ENHANCE_READ)) { 1223 err = -EFSCORRUPTED; 1224 goto put_err; 1225 } 1226 goto got_it; 1227 } 1228 1229 set_new_dnode(&dn, inode, NULL, NULL, 0); 1230 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1231 if (err) { 1232 if (err == -ENOENT && next_pgofs) 1233 *next_pgofs = f2fs_get_next_page_offset(&dn, index); 1234 goto put_err; 1235 } 1236 f2fs_put_dnode(&dn); 1237 1238 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1239 err = -ENOENT; 1240 if (next_pgofs) 1241 *next_pgofs = index + 1; 1242 goto put_err; 1243 } 1244 if (dn.data_blkaddr != NEW_ADDR && 1245 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode), 1246 dn.data_blkaddr, 1247 DATA_GENERIC_ENHANCE)) { 1248 err = -EFSCORRUPTED; 1249 goto put_err; 1250 } 1251 got_it: 1252 if (PageUptodate(page)) { 1253 unlock_page(page); 1254 return page; 1255 } 1256 1257 /* 1258 * A new dentry page is allocated but not able to be written, since its 1259 * new inode page couldn't be allocated due to -ENOSPC. 1260 * In such the case, its blkaddr can be remained as NEW_ADDR. 1261 * see, f2fs_add_link -> f2fs_get_new_data_page -> 1262 * f2fs_init_inode_metadata. 1263 */ 1264 if (dn.data_blkaddr == NEW_ADDR) { 1265 zero_user_segment(page, 0, PAGE_SIZE); 1266 if (!PageUptodate(page)) 1267 SetPageUptodate(page); 1268 unlock_page(page); 1269 return page; 1270 } 1271 1272 err = f2fs_submit_page_read(inode, page_folio(page), dn.data_blkaddr, 1273 op_flags, for_write); 1274 if (err) 1275 goto put_err; 1276 return page; 1277 1278 put_err: 1279 f2fs_put_page(page, 1); 1280 return ERR_PTR(err); 1281 } 1282 1283 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index, 1284 pgoff_t *next_pgofs) 1285 { 1286 struct address_space *mapping = inode->i_mapping; 1287 struct page *page; 1288 1289 page = find_get_page(mapping, index); 1290 if (page && PageUptodate(page)) 1291 return page; 1292 f2fs_put_page(page, 0); 1293 1294 page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs); 1295 if (IS_ERR(page)) 1296 return page; 1297 1298 if (PageUptodate(page)) 1299 return page; 1300 1301 wait_on_page_locked(page); 1302 if (unlikely(!PageUptodate(page))) { 1303 f2fs_put_page(page, 0); 1304 return ERR_PTR(-EIO); 1305 } 1306 return page; 1307 } 1308 1309 /* 1310 * If it tries to access a hole, return an error. 1311 * Because, the callers, functions in dir.c and GC, should be able to know 1312 * whether this page exists or not. 1313 */ 1314 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, 1315 bool for_write) 1316 { 1317 struct address_space *mapping = inode->i_mapping; 1318 struct page *page; 1319 1320 page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL); 1321 if (IS_ERR(page)) 1322 return page; 1323 1324 /* wait for read completion */ 1325 lock_page(page); 1326 if (unlikely(page->mapping != mapping || !PageUptodate(page))) { 1327 f2fs_put_page(page, 1); 1328 return ERR_PTR(-EIO); 1329 } 1330 return page; 1331 } 1332 1333 /* 1334 * Caller ensures that this data page is never allocated. 1335 * A new zero-filled data page is allocated in the page cache. 1336 * 1337 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 1338 * f2fs_unlock_op(). 1339 * Note that, ipage is set only by make_empty_dir, and if any error occur, 1340 * ipage should be released by this function. 1341 */ 1342 struct page *f2fs_get_new_data_page(struct inode *inode, 1343 struct page *ipage, pgoff_t index, bool new_i_size) 1344 { 1345 struct address_space *mapping = inode->i_mapping; 1346 struct page *page; 1347 struct dnode_of_data dn; 1348 int err; 1349 1350 page = f2fs_grab_cache_page(mapping, index, true); 1351 if (!page) { 1352 /* 1353 * before exiting, we should make sure ipage will be released 1354 * if any error occur. 1355 */ 1356 f2fs_put_page(ipage, 1); 1357 return ERR_PTR(-ENOMEM); 1358 } 1359 1360 set_new_dnode(&dn, inode, ipage, NULL, 0); 1361 err = f2fs_reserve_block(&dn, index); 1362 if (err) { 1363 f2fs_put_page(page, 1); 1364 return ERR_PTR(err); 1365 } 1366 if (!ipage) 1367 f2fs_put_dnode(&dn); 1368 1369 if (PageUptodate(page)) 1370 goto got_it; 1371 1372 if (dn.data_blkaddr == NEW_ADDR) { 1373 zero_user_segment(page, 0, PAGE_SIZE); 1374 if (!PageUptodate(page)) 1375 SetPageUptodate(page); 1376 } else { 1377 f2fs_put_page(page, 1); 1378 1379 /* if ipage exists, blkaddr should be NEW_ADDR */ 1380 f2fs_bug_on(F2FS_I_SB(inode), ipage); 1381 page = f2fs_get_lock_data_page(inode, index, true); 1382 if (IS_ERR(page)) 1383 return page; 1384 } 1385 got_it: 1386 if (new_i_size && i_size_read(inode) < 1387 ((loff_t)(index + 1) << PAGE_SHIFT)) 1388 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 1389 return page; 1390 } 1391 1392 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 1393 { 1394 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1395 struct f2fs_summary sum; 1396 struct node_info ni; 1397 block_t old_blkaddr; 1398 blkcnt_t count = 1; 1399 int err; 1400 1401 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1402 return -EPERM; 1403 1404 err = f2fs_get_node_info(sbi, dn->nid, &ni, false); 1405 if (err) 1406 return err; 1407 1408 dn->data_blkaddr = f2fs_data_blkaddr(dn); 1409 if (dn->data_blkaddr == NULL_ADDR) { 1410 err = inc_valid_block_count(sbi, dn->inode, &count, true); 1411 if (unlikely(err)) 1412 return err; 1413 } 1414 1415 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1416 old_blkaddr = dn->data_blkaddr; 1417 err = f2fs_allocate_data_block(sbi, NULL, old_blkaddr, 1418 &dn->data_blkaddr, &sum, seg_type, NULL); 1419 if (err) 1420 return err; 1421 1422 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 1423 f2fs_invalidate_internal_cache(sbi, old_blkaddr); 1424 1425 f2fs_update_data_blkaddr(dn, dn->data_blkaddr); 1426 return 0; 1427 } 1428 1429 static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag) 1430 { 1431 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1432 f2fs_down_read(&sbi->node_change); 1433 else 1434 f2fs_lock_op(sbi); 1435 } 1436 1437 static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag) 1438 { 1439 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1440 f2fs_up_read(&sbi->node_change); 1441 else 1442 f2fs_unlock_op(sbi); 1443 } 1444 1445 int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index) 1446 { 1447 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1448 int err = 0; 1449 1450 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO); 1451 if (!f2fs_lookup_read_extent_cache_block(dn->inode, index, 1452 &dn->data_blkaddr)) 1453 err = f2fs_reserve_block(dn, index); 1454 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO); 1455 1456 return err; 1457 } 1458 1459 static int f2fs_map_no_dnode(struct inode *inode, 1460 struct f2fs_map_blocks *map, struct dnode_of_data *dn, 1461 pgoff_t pgoff) 1462 { 1463 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1464 1465 /* 1466 * There is one exceptional case that read_node_page() may return 1467 * -ENOENT due to filesystem has been shutdown or cp_error, return 1468 * -EIO in that case. 1469 */ 1470 if (map->m_may_create && 1471 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi))) 1472 return -EIO; 1473 1474 if (map->m_next_pgofs) 1475 *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff); 1476 if (map->m_next_extent) 1477 *map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff); 1478 return 0; 1479 } 1480 1481 static bool f2fs_map_blocks_cached(struct inode *inode, 1482 struct f2fs_map_blocks *map, int flag) 1483 { 1484 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1485 unsigned int maxblocks = map->m_len; 1486 pgoff_t pgoff = (pgoff_t)map->m_lblk; 1487 struct extent_info ei = {}; 1488 1489 if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei)) 1490 return false; 1491 1492 map->m_pblk = ei.blk + pgoff - ei.fofs; 1493 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff); 1494 map->m_flags = F2FS_MAP_MAPPED; 1495 if (map->m_next_extent) 1496 *map->m_next_extent = pgoff + map->m_len; 1497 1498 /* for hardware encryption, but to avoid potential issue in future */ 1499 if (flag == F2FS_GET_BLOCK_DIO) 1500 f2fs_wait_on_block_writeback_range(inode, 1501 map->m_pblk, map->m_len); 1502 1503 if (f2fs_allow_multi_device_dio(sbi, flag)) { 1504 int bidx = f2fs_target_device_index(sbi, map->m_pblk); 1505 struct f2fs_dev_info *dev = &sbi->devs[bidx]; 1506 1507 map->m_bdev = dev->bdev; 1508 map->m_pblk -= dev->start_blk; 1509 map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk); 1510 } else { 1511 map->m_bdev = inode->i_sb->s_bdev; 1512 } 1513 return true; 1514 } 1515 1516 static bool map_is_mergeable(struct f2fs_sb_info *sbi, 1517 struct f2fs_map_blocks *map, 1518 block_t blkaddr, int flag, int bidx, 1519 int ofs) 1520 { 1521 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev) 1522 return false; 1523 if (map->m_pblk != NEW_ADDR && blkaddr == (map->m_pblk + ofs)) 1524 return true; 1525 if (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) 1526 return true; 1527 if (flag == F2FS_GET_BLOCK_PRE_DIO) 1528 return true; 1529 if (flag == F2FS_GET_BLOCK_DIO && 1530 map->m_pblk == NULL_ADDR && blkaddr == NULL_ADDR) 1531 return true; 1532 return false; 1533 } 1534 1535 /* 1536 * f2fs_map_blocks() tries to find or build mapping relationship which 1537 * maps continuous logical blocks to physical blocks, and return such 1538 * info via f2fs_map_blocks structure. 1539 */ 1540 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag) 1541 { 1542 unsigned int maxblocks = map->m_len; 1543 struct dnode_of_data dn; 1544 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1545 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE; 1546 pgoff_t pgofs, end_offset, end; 1547 int err = 0, ofs = 1; 1548 unsigned int ofs_in_node, last_ofs_in_node; 1549 blkcnt_t prealloc; 1550 block_t blkaddr; 1551 unsigned int start_pgofs; 1552 int bidx = 0; 1553 bool is_hole; 1554 1555 if (!maxblocks) 1556 return 0; 1557 1558 if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag)) 1559 goto out; 1560 1561 map->m_bdev = inode->i_sb->s_bdev; 1562 map->m_multidev_dio = 1563 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag); 1564 1565 map->m_len = 0; 1566 map->m_flags = 0; 1567 1568 /* it only supports block size == page size */ 1569 pgofs = (pgoff_t)map->m_lblk; 1570 end = pgofs + maxblocks; 1571 1572 next_dnode: 1573 if (map->m_may_create) 1574 f2fs_map_lock(sbi, flag); 1575 1576 /* When reading holes, we need its node page */ 1577 set_new_dnode(&dn, inode, NULL, NULL, 0); 1578 err = f2fs_get_dnode_of_data(&dn, pgofs, mode); 1579 if (err) { 1580 if (flag == F2FS_GET_BLOCK_BMAP) 1581 map->m_pblk = 0; 1582 if (err == -ENOENT) 1583 err = f2fs_map_no_dnode(inode, map, &dn, pgofs); 1584 goto unlock_out; 1585 } 1586 1587 start_pgofs = pgofs; 1588 prealloc = 0; 1589 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 1590 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 1591 1592 next_block: 1593 blkaddr = f2fs_data_blkaddr(&dn); 1594 is_hole = !__is_valid_data_blkaddr(blkaddr); 1595 if (!is_hole && 1596 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) { 1597 err = -EFSCORRUPTED; 1598 goto sync_out; 1599 } 1600 1601 /* use out-place-update for direct IO under LFS mode */ 1602 if (map->m_may_create && (is_hole || 1603 (flag == F2FS_GET_BLOCK_DIO && f2fs_lfs_mode(sbi) && 1604 !f2fs_is_pinned_file(inode)))) { 1605 if (unlikely(f2fs_cp_error(sbi))) { 1606 err = -EIO; 1607 goto sync_out; 1608 } 1609 1610 switch (flag) { 1611 case F2FS_GET_BLOCK_PRE_AIO: 1612 if (blkaddr == NULL_ADDR) { 1613 prealloc++; 1614 last_ofs_in_node = dn.ofs_in_node; 1615 } 1616 break; 1617 case F2FS_GET_BLOCK_PRE_DIO: 1618 case F2FS_GET_BLOCK_DIO: 1619 err = __allocate_data_block(&dn, map->m_seg_type); 1620 if (err) 1621 goto sync_out; 1622 if (flag == F2FS_GET_BLOCK_PRE_DIO) 1623 file_need_truncate(inode); 1624 set_inode_flag(inode, FI_APPEND_WRITE); 1625 break; 1626 default: 1627 WARN_ON_ONCE(1); 1628 err = -EIO; 1629 goto sync_out; 1630 } 1631 1632 blkaddr = dn.data_blkaddr; 1633 if (is_hole) 1634 map->m_flags |= F2FS_MAP_NEW; 1635 } else if (is_hole) { 1636 if (f2fs_compressed_file(inode) && 1637 f2fs_sanity_check_cluster(&dn)) { 1638 err = -EFSCORRUPTED; 1639 f2fs_handle_error(sbi, 1640 ERROR_CORRUPTED_CLUSTER); 1641 goto sync_out; 1642 } 1643 1644 switch (flag) { 1645 case F2FS_GET_BLOCK_PRECACHE: 1646 goto sync_out; 1647 case F2FS_GET_BLOCK_BMAP: 1648 map->m_pblk = 0; 1649 goto sync_out; 1650 case F2FS_GET_BLOCK_FIEMAP: 1651 if (blkaddr == NULL_ADDR) { 1652 if (map->m_next_pgofs) 1653 *map->m_next_pgofs = pgofs + 1; 1654 goto sync_out; 1655 } 1656 break; 1657 case F2FS_GET_BLOCK_DIO: 1658 if (map->m_next_pgofs) 1659 *map->m_next_pgofs = pgofs + 1; 1660 break; 1661 default: 1662 /* for defragment case */ 1663 if (map->m_next_pgofs) 1664 *map->m_next_pgofs = pgofs + 1; 1665 goto sync_out; 1666 } 1667 } 1668 1669 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1670 goto skip; 1671 1672 if (map->m_multidev_dio) 1673 bidx = f2fs_target_device_index(sbi, blkaddr); 1674 1675 if (map->m_len == 0) { 1676 /* reserved delalloc block should be mapped for fiemap. */ 1677 if (blkaddr == NEW_ADDR) 1678 map->m_flags |= F2FS_MAP_DELALLOC; 1679 if (flag != F2FS_GET_BLOCK_DIO || !is_hole) 1680 map->m_flags |= F2FS_MAP_MAPPED; 1681 1682 map->m_pblk = blkaddr; 1683 map->m_len = 1; 1684 1685 if (map->m_multidev_dio) 1686 map->m_bdev = FDEV(bidx).bdev; 1687 } else if (map_is_mergeable(sbi, map, blkaddr, flag, bidx, ofs)) { 1688 ofs++; 1689 map->m_len++; 1690 } else { 1691 goto sync_out; 1692 } 1693 1694 skip: 1695 dn.ofs_in_node++; 1696 pgofs++; 1697 1698 /* preallocate blocks in batch for one dnode page */ 1699 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1700 (pgofs == end || dn.ofs_in_node == end_offset)) { 1701 1702 dn.ofs_in_node = ofs_in_node; 1703 err = f2fs_reserve_new_blocks(&dn, prealloc); 1704 if (err) 1705 goto sync_out; 1706 1707 map->m_len += dn.ofs_in_node - ofs_in_node; 1708 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1709 err = -ENOSPC; 1710 goto sync_out; 1711 } 1712 dn.ofs_in_node = end_offset; 1713 } 1714 1715 if (flag == F2FS_GET_BLOCK_DIO && f2fs_lfs_mode(sbi) && 1716 map->m_may_create) { 1717 /* the next block to be allocated may not be contiguous. */ 1718 if (GET_SEGOFF_FROM_SEG0(sbi, blkaddr) % BLKS_PER_SEC(sbi) == 1719 CAP_BLKS_PER_SEC(sbi) - 1) 1720 goto sync_out; 1721 } 1722 1723 if (pgofs >= end) 1724 goto sync_out; 1725 else if (dn.ofs_in_node < end_offset) 1726 goto next_block; 1727 1728 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1729 if (map->m_flags & F2FS_MAP_MAPPED) { 1730 unsigned int ofs = start_pgofs - map->m_lblk; 1731 1732 f2fs_update_read_extent_cache_range(&dn, 1733 start_pgofs, map->m_pblk + ofs, 1734 map->m_len - ofs); 1735 } 1736 } 1737 1738 f2fs_put_dnode(&dn); 1739 1740 if (map->m_may_create) { 1741 f2fs_map_unlock(sbi, flag); 1742 f2fs_balance_fs(sbi, dn.node_changed); 1743 } 1744 goto next_dnode; 1745 1746 sync_out: 1747 1748 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) { 1749 /* 1750 * for hardware encryption, but to avoid potential issue 1751 * in future 1752 */ 1753 f2fs_wait_on_block_writeback_range(inode, 1754 map->m_pblk, map->m_len); 1755 1756 if (map->m_multidev_dio) { 1757 block_t blk_addr = map->m_pblk; 1758 1759 bidx = f2fs_target_device_index(sbi, map->m_pblk); 1760 1761 map->m_bdev = FDEV(bidx).bdev; 1762 map->m_pblk -= FDEV(bidx).start_blk; 1763 1764 if (map->m_may_create) 1765 f2fs_update_device_state(sbi, inode->i_ino, 1766 blk_addr, map->m_len); 1767 1768 f2fs_bug_on(sbi, blk_addr + map->m_len > 1769 FDEV(bidx).end_blk + 1); 1770 } 1771 } 1772 1773 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1774 if (map->m_flags & F2FS_MAP_MAPPED) { 1775 unsigned int ofs = start_pgofs - map->m_lblk; 1776 1777 f2fs_update_read_extent_cache_range(&dn, 1778 start_pgofs, map->m_pblk + ofs, 1779 map->m_len - ofs); 1780 } 1781 if (map->m_next_extent) 1782 *map->m_next_extent = pgofs + 1; 1783 } 1784 f2fs_put_dnode(&dn); 1785 unlock_out: 1786 if (map->m_may_create) { 1787 f2fs_map_unlock(sbi, flag); 1788 f2fs_balance_fs(sbi, dn.node_changed); 1789 } 1790 out: 1791 trace_f2fs_map_blocks(inode, map, flag, err); 1792 return err; 1793 } 1794 1795 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) 1796 { 1797 struct f2fs_map_blocks map; 1798 block_t last_lblk; 1799 int err; 1800 1801 if (pos + len > i_size_read(inode)) 1802 return false; 1803 1804 map.m_lblk = F2FS_BYTES_TO_BLK(pos); 1805 map.m_next_pgofs = NULL; 1806 map.m_next_extent = NULL; 1807 map.m_seg_type = NO_CHECK_TYPE; 1808 map.m_may_create = false; 1809 last_lblk = F2FS_BLK_ALIGN(pos + len); 1810 1811 while (map.m_lblk < last_lblk) { 1812 map.m_len = last_lblk - map.m_lblk; 1813 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT); 1814 if (err || map.m_len == 0) 1815 return false; 1816 map.m_lblk += map.m_len; 1817 } 1818 return true; 1819 } 1820 1821 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes) 1822 { 1823 return (bytes >> inode->i_blkbits); 1824 } 1825 1826 static inline u64 blks_to_bytes(struct inode *inode, u64 blks) 1827 { 1828 return (blks << inode->i_blkbits); 1829 } 1830 1831 static int f2fs_xattr_fiemap(struct inode *inode, 1832 struct fiemap_extent_info *fieinfo) 1833 { 1834 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1835 struct page *page; 1836 struct node_info ni; 1837 __u64 phys = 0, len; 1838 __u32 flags; 1839 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1840 int err = 0; 1841 1842 if (f2fs_has_inline_xattr(inode)) { 1843 int offset; 1844 1845 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1846 inode->i_ino, false); 1847 if (!page) 1848 return -ENOMEM; 1849 1850 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false); 1851 if (err) { 1852 f2fs_put_page(page, 1); 1853 return err; 1854 } 1855 1856 phys = blks_to_bytes(inode, ni.blk_addr); 1857 offset = offsetof(struct f2fs_inode, i_addr) + 1858 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1859 get_inline_xattr_addrs(inode)); 1860 1861 phys += offset; 1862 len = inline_xattr_size(inode); 1863 1864 f2fs_put_page(page, 1); 1865 1866 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1867 1868 if (!xnid) 1869 flags |= FIEMAP_EXTENT_LAST; 1870 1871 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1872 trace_f2fs_fiemap(inode, 0, phys, len, flags, err); 1873 if (err) 1874 return err; 1875 } 1876 1877 if (xnid) { 1878 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1879 if (!page) 1880 return -ENOMEM; 1881 1882 err = f2fs_get_node_info(sbi, xnid, &ni, false); 1883 if (err) { 1884 f2fs_put_page(page, 1); 1885 return err; 1886 } 1887 1888 phys = blks_to_bytes(inode, ni.blk_addr); 1889 len = inode->i_sb->s_blocksize; 1890 1891 f2fs_put_page(page, 1); 1892 1893 flags = FIEMAP_EXTENT_LAST; 1894 } 1895 1896 if (phys) { 1897 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1898 trace_f2fs_fiemap(inode, 0, phys, len, flags, err); 1899 } 1900 1901 return (err < 0 ? err : 0); 1902 } 1903 1904 static loff_t max_inode_blocks(struct inode *inode) 1905 { 1906 loff_t result = ADDRS_PER_INODE(inode); 1907 loff_t leaf_count = ADDRS_PER_BLOCK(inode); 1908 1909 /* two direct node blocks */ 1910 result += (leaf_count * 2); 1911 1912 /* two indirect node blocks */ 1913 leaf_count *= NIDS_PER_BLOCK; 1914 result += (leaf_count * 2); 1915 1916 /* one double indirect node block */ 1917 leaf_count *= NIDS_PER_BLOCK; 1918 result += leaf_count; 1919 1920 return result; 1921 } 1922 1923 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1924 u64 start, u64 len) 1925 { 1926 struct f2fs_map_blocks map; 1927 sector_t start_blk, last_blk; 1928 pgoff_t next_pgofs; 1929 u64 logical = 0, phys = 0, size = 0; 1930 u32 flags = 0; 1931 int ret = 0; 1932 bool compr_cluster = false, compr_appended; 1933 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size; 1934 unsigned int count_in_cluster = 0; 1935 loff_t maxbytes; 1936 1937 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { 1938 ret = f2fs_precache_extents(inode); 1939 if (ret) 1940 return ret; 1941 } 1942 1943 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR); 1944 if (ret) 1945 return ret; 1946 1947 inode_lock_shared(inode); 1948 1949 maxbytes = F2FS_BLK_TO_BYTES(max_file_blocks(inode)); 1950 if (start > maxbytes) { 1951 ret = -EFBIG; 1952 goto out; 1953 } 1954 1955 if (len > maxbytes || (maxbytes - len) < start) 1956 len = maxbytes - start; 1957 1958 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1959 ret = f2fs_xattr_fiemap(inode, fieinfo); 1960 goto out; 1961 } 1962 1963 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) { 1964 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1965 if (ret != -EAGAIN) 1966 goto out; 1967 } 1968 1969 if (bytes_to_blks(inode, len) == 0) 1970 len = blks_to_bytes(inode, 1); 1971 1972 start_blk = bytes_to_blks(inode, start); 1973 last_blk = bytes_to_blks(inode, start + len - 1); 1974 1975 next: 1976 memset(&map, 0, sizeof(map)); 1977 map.m_lblk = start_blk; 1978 map.m_len = bytes_to_blks(inode, len); 1979 map.m_next_pgofs = &next_pgofs; 1980 map.m_seg_type = NO_CHECK_TYPE; 1981 1982 if (compr_cluster) { 1983 map.m_lblk += 1; 1984 map.m_len = cluster_size - count_in_cluster; 1985 } 1986 1987 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP); 1988 if (ret) 1989 goto out; 1990 1991 /* HOLE */ 1992 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) { 1993 start_blk = next_pgofs; 1994 1995 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode, 1996 max_inode_blocks(inode))) 1997 goto prep_next; 1998 1999 flags |= FIEMAP_EXTENT_LAST; 2000 } 2001 2002 compr_appended = false; 2003 /* In a case of compressed cluster, append this to the last extent */ 2004 if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) || 2005 !(map.m_flags & F2FS_MAP_FLAGS))) { 2006 compr_appended = true; 2007 goto skip_fill; 2008 } 2009 2010 if (size) { 2011 flags |= FIEMAP_EXTENT_MERGED; 2012 if (IS_ENCRYPTED(inode)) 2013 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 2014 2015 ret = fiemap_fill_next_extent(fieinfo, logical, 2016 phys, size, flags); 2017 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret); 2018 if (ret) 2019 goto out; 2020 size = 0; 2021 } 2022 2023 if (start_blk > last_blk) 2024 goto out; 2025 2026 skip_fill: 2027 if (map.m_pblk == COMPRESS_ADDR) { 2028 compr_cluster = true; 2029 count_in_cluster = 1; 2030 } else if (compr_appended) { 2031 unsigned int appended_blks = cluster_size - 2032 count_in_cluster + 1; 2033 size += blks_to_bytes(inode, appended_blks); 2034 start_blk += appended_blks; 2035 compr_cluster = false; 2036 } else { 2037 logical = blks_to_bytes(inode, start_blk); 2038 phys = __is_valid_data_blkaddr(map.m_pblk) ? 2039 blks_to_bytes(inode, map.m_pblk) : 0; 2040 size = blks_to_bytes(inode, map.m_len); 2041 flags = 0; 2042 2043 if (compr_cluster) { 2044 flags = FIEMAP_EXTENT_ENCODED; 2045 count_in_cluster += map.m_len; 2046 if (count_in_cluster == cluster_size) { 2047 compr_cluster = false; 2048 size += blks_to_bytes(inode, 1); 2049 } 2050 } else if (map.m_flags & F2FS_MAP_DELALLOC) { 2051 flags = FIEMAP_EXTENT_UNWRITTEN; 2052 } 2053 2054 start_blk += bytes_to_blks(inode, size); 2055 } 2056 2057 prep_next: 2058 cond_resched(); 2059 if (fatal_signal_pending(current)) 2060 ret = -EINTR; 2061 else 2062 goto next; 2063 out: 2064 if (ret == 1) 2065 ret = 0; 2066 2067 inode_unlock_shared(inode); 2068 return ret; 2069 } 2070 2071 static inline loff_t f2fs_readpage_limit(struct inode *inode) 2072 { 2073 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) 2074 return F2FS_BLK_TO_BYTES(max_file_blocks(inode)); 2075 2076 return i_size_read(inode); 2077 } 2078 2079 static inline blk_opf_t f2fs_ra_op_flags(struct readahead_control *rac) 2080 { 2081 return rac ? REQ_RAHEAD : 0; 2082 } 2083 2084 static int f2fs_read_single_page(struct inode *inode, struct folio *folio, 2085 unsigned nr_pages, 2086 struct f2fs_map_blocks *map, 2087 struct bio **bio_ret, 2088 sector_t *last_block_in_bio, 2089 struct readahead_control *rac) 2090 { 2091 struct bio *bio = *bio_ret; 2092 const unsigned blocksize = blks_to_bytes(inode, 1); 2093 sector_t block_in_file; 2094 sector_t last_block; 2095 sector_t last_block_in_file; 2096 sector_t block_nr; 2097 pgoff_t index = folio_index(folio); 2098 int ret = 0; 2099 2100 block_in_file = (sector_t)index; 2101 last_block = block_in_file + nr_pages; 2102 last_block_in_file = bytes_to_blks(inode, 2103 f2fs_readpage_limit(inode) + blocksize - 1); 2104 if (last_block > last_block_in_file) 2105 last_block = last_block_in_file; 2106 2107 /* just zeroing out page which is beyond EOF */ 2108 if (block_in_file >= last_block) 2109 goto zero_out; 2110 /* 2111 * Map blocks using the previous result first. 2112 */ 2113 if ((map->m_flags & F2FS_MAP_MAPPED) && 2114 block_in_file > map->m_lblk && 2115 block_in_file < (map->m_lblk + map->m_len)) 2116 goto got_it; 2117 2118 /* 2119 * Then do more f2fs_map_blocks() calls until we are 2120 * done with this page. 2121 */ 2122 map->m_lblk = block_in_file; 2123 map->m_len = last_block - block_in_file; 2124 2125 ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT); 2126 if (ret) 2127 goto out; 2128 got_it: 2129 if ((map->m_flags & F2FS_MAP_MAPPED)) { 2130 block_nr = map->m_pblk + block_in_file - map->m_lblk; 2131 folio_set_mappedtodisk(folio); 2132 2133 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr, 2134 DATA_GENERIC_ENHANCE_READ)) { 2135 ret = -EFSCORRUPTED; 2136 goto out; 2137 } 2138 } else { 2139 zero_out: 2140 folio_zero_segment(folio, 0, folio_size(folio)); 2141 if (f2fs_need_verity(inode, index) && 2142 !fsverity_verify_folio(folio)) { 2143 ret = -EIO; 2144 goto out; 2145 } 2146 if (!folio_test_uptodate(folio)) 2147 folio_mark_uptodate(folio); 2148 folio_unlock(folio); 2149 goto out; 2150 } 2151 2152 /* 2153 * This page will go to BIO. Do we need to send this 2154 * BIO off first? 2155 */ 2156 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio, 2157 *last_block_in_bio, block_nr) || 2158 !f2fs_crypt_mergeable_bio(bio, inode, index, NULL))) { 2159 submit_and_realloc: 2160 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA); 2161 bio = NULL; 2162 } 2163 if (bio == NULL) { 2164 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages, 2165 f2fs_ra_op_flags(rac), index, 2166 false); 2167 if (IS_ERR(bio)) { 2168 ret = PTR_ERR(bio); 2169 bio = NULL; 2170 goto out; 2171 } 2172 } 2173 2174 /* 2175 * If the page is under writeback, we need to wait for 2176 * its completion to see the correct decrypted data. 2177 */ 2178 f2fs_wait_on_block_writeback(inode, block_nr); 2179 2180 if (!bio_add_folio(bio, folio, blocksize, 0)) 2181 goto submit_and_realloc; 2182 2183 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 2184 f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO, 2185 F2FS_BLKSIZE); 2186 *last_block_in_bio = block_nr; 2187 out: 2188 *bio_ret = bio; 2189 return ret; 2190 } 2191 2192 #ifdef CONFIG_F2FS_FS_COMPRESSION 2193 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret, 2194 unsigned nr_pages, sector_t *last_block_in_bio, 2195 struct readahead_control *rac, bool for_write) 2196 { 2197 struct dnode_of_data dn; 2198 struct inode *inode = cc->inode; 2199 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2200 struct bio *bio = *bio_ret; 2201 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size; 2202 sector_t last_block_in_file; 2203 const unsigned blocksize = blks_to_bytes(inode, 1); 2204 struct decompress_io_ctx *dic = NULL; 2205 struct extent_info ei = {}; 2206 bool from_dnode = true; 2207 int i; 2208 int ret = 0; 2209 2210 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc)); 2211 2212 last_block_in_file = bytes_to_blks(inode, 2213 f2fs_readpage_limit(inode) + blocksize - 1); 2214 2215 /* get rid of pages beyond EOF */ 2216 for (i = 0; i < cc->cluster_size; i++) { 2217 struct page *page = cc->rpages[i]; 2218 struct folio *folio; 2219 2220 if (!page) 2221 continue; 2222 2223 folio = page_folio(page); 2224 if ((sector_t)folio->index >= last_block_in_file) { 2225 folio_zero_segment(folio, 0, folio_size(folio)); 2226 if (!folio_test_uptodate(folio)) 2227 folio_mark_uptodate(folio); 2228 } else if (!folio_test_uptodate(folio)) { 2229 continue; 2230 } 2231 folio_unlock(folio); 2232 if (for_write) 2233 folio_put(folio); 2234 cc->rpages[i] = NULL; 2235 cc->nr_rpages--; 2236 } 2237 2238 /* we are done since all pages are beyond EOF */ 2239 if (f2fs_cluster_is_empty(cc)) 2240 goto out; 2241 2242 if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei)) 2243 from_dnode = false; 2244 2245 if (!from_dnode) 2246 goto skip_reading_dnode; 2247 2248 set_new_dnode(&dn, inode, NULL, NULL, 0); 2249 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 2250 if (ret) 2251 goto out; 2252 2253 if (unlikely(f2fs_cp_error(sbi))) { 2254 ret = -EIO; 2255 goto out_put_dnode; 2256 } 2257 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR); 2258 2259 skip_reading_dnode: 2260 for (i = 1; i < cc->cluster_size; i++) { 2261 block_t blkaddr; 2262 2263 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page, 2264 dn.ofs_in_node + i) : 2265 ei.blk + i - 1; 2266 2267 if (!__is_valid_data_blkaddr(blkaddr)) 2268 break; 2269 2270 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) { 2271 ret = -EFAULT; 2272 goto out_put_dnode; 2273 } 2274 cc->nr_cpages++; 2275 2276 if (!from_dnode && i >= ei.c_len) 2277 break; 2278 } 2279 2280 /* nothing to decompress */ 2281 if (cc->nr_cpages == 0) { 2282 ret = 0; 2283 goto out_put_dnode; 2284 } 2285 2286 dic = f2fs_alloc_dic(cc); 2287 if (IS_ERR(dic)) { 2288 ret = PTR_ERR(dic); 2289 goto out_put_dnode; 2290 } 2291 2292 for (i = 0; i < cc->nr_cpages; i++) { 2293 struct folio *folio = page_folio(dic->cpages[i]); 2294 block_t blkaddr; 2295 struct bio_post_read_ctx *ctx; 2296 2297 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page, 2298 dn.ofs_in_node + i + 1) : 2299 ei.blk + i; 2300 2301 f2fs_wait_on_block_writeback(inode, blkaddr); 2302 2303 if (f2fs_load_compressed_page(sbi, folio_page(folio, 0), 2304 blkaddr)) { 2305 if (atomic_dec_and_test(&dic->remaining_pages)) { 2306 f2fs_decompress_cluster(dic, true); 2307 break; 2308 } 2309 continue; 2310 } 2311 2312 if (bio && (!page_is_mergeable(sbi, bio, 2313 *last_block_in_bio, blkaddr) || 2314 !f2fs_crypt_mergeable_bio(bio, inode, folio->index, NULL))) { 2315 submit_and_realloc: 2316 f2fs_submit_read_bio(sbi, bio, DATA); 2317 bio = NULL; 2318 } 2319 2320 if (!bio) { 2321 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages, 2322 f2fs_ra_op_flags(rac), 2323 folio->index, for_write); 2324 if (IS_ERR(bio)) { 2325 ret = PTR_ERR(bio); 2326 f2fs_decompress_end_io(dic, ret, true); 2327 f2fs_put_dnode(&dn); 2328 *bio_ret = NULL; 2329 return ret; 2330 } 2331 } 2332 2333 if (!bio_add_folio(bio, folio, blocksize, 0)) 2334 goto submit_and_realloc; 2335 2336 ctx = get_post_read_ctx(bio); 2337 ctx->enabled_steps |= STEP_DECOMPRESS; 2338 refcount_inc(&dic->refcnt); 2339 2340 inc_page_count(sbi, F2FS_RD_DATA); 2341 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE); 2342 *last_block_in_bio = blkaddr; 2343 } 2344 2345 if (from_dnode) 2346 f2fs_put_dnode(&dn); 2347 2348 *bio_ret = bio; 2349 return 0; 2350 2351 out_put_dnode: 2352 if (from_dnode) 2353 f2fs_put_dnode(&dn); 2354 out: 2355 for (i = 0; i < cc->cluster_size; i++) { 2356 if (cc->rpages[i]) { 2357 ClearPageUptodate(cc->rpages[i]); 2358 unlock_page(cc->rpages[i]); 2359 } 2360 } 2361 *bio_ret = bio; 2362 return ret; 2363 } 2364 #endif 2365 2366 /* 2367 * This function was originally taken from fs/mpage.c, and customized for f2fs. 2368 * Major change was from block_size == page_size in f2fs by default. 2369 */ 2370 static int f2fs_mpage_readpages(struct inode *inode, 2371 struct readahead_control *rac, struct folio *folio) 2372 { 2373 struct bio *bio = NULL; 2374 sector_t last_block_in_bio = 0; 2375 struct f2fs_map_blocks map; 2376 #ifdef CONFIG_F2FS_FS_COMPRESSION 2377 struct compress_ctx cc = { 2378 .inode = inode, 2379 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2380 .cluster_size = F2FS_I(inode)->i_cluster_size, 2381 .cluster_idx = NULL_CLUSTER, 2382 .rpages = NULL, 2383 .cpages = NULL, 2384 .nr_rpages = 0, 2385 .nr_cpages = 0, 2386 }; 2387 pgoff_t nc_cluster_idx = NULL_CLUSTER; 2388 #endif 2389 unsigned nr_pages = rac ? readahead_count(rac) : 1; 2390 unsigned max_nr_pages = nr_pages; 2391 pgoff_t index; 2392 int ret = 0; 2393 2394 map.m_pblk = 0; 2395 map.m_lblk = 0; 2396 map.m_len = 0; 2397 map.m_flags = 0; 2398 map.m_next_pgofs = NULL; 2399 map.m_next_extent = NULL; 2400 map.m_seg_type = NO_CHECK_TYPE; 2401 map.m_may_create = false; 2402 2403 for (; nr_pages; nr_pages--) { 2404 if (rac) { 2405 folio = readahead_folio(rac); 2406 prefetchw(&folio->flags); 2407 } 2408 2409 index = folio_index(folio); 2410 2411 #ifdef CONFIG_F2FS_FS_COMPRESSION 2412 if (!f2fs_compressed_file(inode)) 2413 goto read_single_page; 2414 2415 /* there are remained compressed pages, submit them */ 2416 if (!f2fs_cluster_can_merge_page(&cc, index)) { 2417 ret = f2fs_read_multi_pages(&cc, &bio, 2418 max_nr_pages, 2419 &last_block_in_bio, 2420 rac, false); 2421 f2fs_destroy_compress_ctx(&cc, false); 2422 if (ret) 2423 goto set_error_page; 2424 } 2425 if (cc.cluster_idx == NULL_CLUSTER) { 2426 if (nc_cluster_idx == index >> cc.log_cluster_size) 2427 goto read_single_page; 2428 2429 ret = f2fs_is_compressed_cluster(inode, index); 2430 if (ret < 0) 2431 goto set_error_page; 2432 else if (!ret) { 2433 nc_cluster_idx = 2434 index >> cc.log_cluster_size; 2435 goto read_single_page; 2436 } 2437 2438 nc_cluster_idx = NULL_CLUSTER; 2439 } 2440 ret = f2fs_init_compress_ctx(&cc); 2441 if (ret) 2442 goto set_error_page; 2443 2444 f2fs_compress_ctx_add_page(&cc, folio); 2445 2446 goto next_page; 2447 read_single_page: 2448 #endif 2449 2450 ret = f2fs_read_single_page(inode, folio, max_nr_pages, &map, 2451 &bio, &last_block_in_bio, rac); 2452 if (ret) { 2453 #ifdef CONFIG_F2FS_FS_COMPRESSION 2454 set_error_page: 2455 #endif 2456 folio_zero_segment(folio, 0, folio_size(folio)); 2457 folio_unlock(folio); 2458 } 2459 #ifdef CONFIG_F2FS_FS_COMPRESSION 2460 next_page: 2461 #endif 2462 2463 #ifdef CONFIG_F2FS_FS_COMPRESSION 2464 if (f2fs_compressed_file(inode)) { 2465 /* last page */ 2466 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) { 2467 ret = f2fs_read_multi_pages(&cc, &bio, 2468 max_nr_pages, 2469 &last_block_in_bio, 2470 rac, false); 2471 f2fs_destroy_compress_ctx(&cc, false); 2472 } 2473 } 2474 #endif 2475 } 2476 if (bio) 2477 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA); 2478 return ret; 2479 } 2480 2481 static int f2fs_read_data_folio(struct file *file, struct folio *folio) 2482 { 2483 struct inode *inode = folio_file_mapping(folio)->host; 2484 int ret = -EAGAIN; 2485 2486 trace_f2fs_readpage(folio, DATA); 2487 2488 if (!f2fs_is_compress_backend_ready(inode)) { 2489 folio_unlock(folio); 2490 return -EOPNOTSUPP; 2491 } 2492 2493 /* If the file has inline data, try to read it directly */ 2494 if (f2fs_has_inline_data(inode)) 2495 ret = f2fs_read_inline_data(inode, folio); 2496 if (ret == -EAGAIN) 2497 ret = f2fs_mpage_readpages(inode, NULL, folio); 2498 return ret; 2499 } 2500 2501 static void f2fs_readahead(struct readahead_control *rac) 2502 { 2503 struct inode *inode = rac->mapping->host; 2504 2505 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac)); 2506 2507 if (!f2fs_is_compress_backend_ready(inode)) 2508 return; 2509 2510 /* If the file has inline data, skip readahead */ 2511 if (f2fs_has_inline_data(inode)) 2512 return; 2513 2514 f2fs_mpage_readpages(inode, rac, NULL); 2515 } 2516 2517 int f2fs_encrypt_one_page(struct f2fs_io_info *fio) 2518 { 2519 struct inode *inode = fio->page->mapping->host; 2520 struct page *mpage, *page; 2521 gfp_t gfp_flags = GFP_NOFS; 2522 2523 if (!f2fs_encrypted_file(inode)) 2524 return 0; 2525 2526 page = fio->compressed_page ? fio->compressed_page : fio->page; 2527 2528 if (fscrypt_inode_uses_inline_crypto(inode)) 2529 return 0; 2530 2531 retry_encrypt: 2532 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page, 2533 PAGE_SIZE, 0, gfp_flags); 2534 if (IS_ERR(fio->encrypted_page)) { 2535 /* flush pending IOs and wait for a while in the ENOMEM case */ 2536 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 2537 f2fs_flush_merged_writes(fio->sbi); 2538 memalloc_retry_wait(GFP_NOFS); 2539 gfp_flags |= __GFP_NOFAIL; 2540 goto retry_encrypt; 2541 } 2542 return PTR_ERR(fio->encrypted_page); 2543 } 2544 2545 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr); 2546 if (mpage) { 2547 if (PageUptodate(mpage)) 2548 memcpy(page_address(mpage), 2549 page_address(fio->encrypted_page), PAGE_SIZE); 2550 f2fs_put_page(mpage, 1); 2551 } 2552 return 0; 2553 } 2554 2555 static inline bool check_inplace_update_policy(struct inode *inode, 2556 struct f2fs_io_info *fio) 2557 { 2558 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2559 2560 if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) && 2561 is_inode_flag_set(inode, FI_OPU_WRITE)) 2562 return false; 2563 if (IS_F2FS_IPU_FORCE(sbi)) 2564 return true; 2565 if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi)) 2566 return true; 2567 if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util) 2568 return true; 2569 if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) && 2570 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2571 return true; 2572 2573 /* 2574 * IPU for rewrite async pages 2575 */ 2576 if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE && 2577 !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode)) 2578 return true; 2579 2580 /* this is only set during fdatasync */ 2581 if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU)) 2582 return true; 2583 2584 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2585 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2586 return true; 2587 2588 return false; 2589 } 2590 2591 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) 2592 { 2593 /* swap file is migrating in aligned write mode */ 2594 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 2595 return false; 2596 2597 if (f2fs_is_pinned_file(inode)) 2598 return true; 2599 2600 /* if this is cold file, we should overwrite to avoid fragmentation */ 2601 if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE)) 2602 return true; 2603 2604 return check_inplace_update_policy(inode, fio); 2605 } 2606 2607 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) 2608 { 2609 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2610 2611 /* The below cases were checked when setting it. */ 2612 if (f2fs_is_pinned_file(inode)) 2613 return false; 2614 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 2615 return true; 2616 if (f2fs_lfs_mode(sbi)) 2617 return true; 2618 if (S_ISDIR(inode->i_mode)) 2619 return true; 2620 if (IS_NOQUOTA(inode)) 2621 return true; 2622 if (f2fs_used_in_atomic_write(inode)) 2623 return true; 2624 /* rewrite low ratio compress data w/ OPU mode to avoid fragmentation */ 2625 if (f2fs_compressed_file(inode) && 2626 F2FS_OPTION(sbi).compress_mode == COMPR_MODE_USER && 2627 is_inode_flag_set(inode, FI_ENABLE_COMPRESS)) 2628 return true; 2629 2630 /* swap file is migrating in aligned write mode */ 2631 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 2632 return true; 2633 2634 if (is_inode_flag_set(inode, FI_OPU_WRITE)) 2635 return true; 2636 2637 if (fio) { 2638 if (page_private_gcing(fio->page)) 2639 return true; 2640 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2641 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2642 return true; 2643 } 2644 return false; 2645 } 2646 2647 static inline bool need_inplace_update(struct f2fs_io_info *fio) 2648 { 2649 struct inode *inode = fio->page->mapping->host; 2650 2651 if (f2fs_should_update_outplace(inode, fio)) 2652 return false; 2653 2654 return f2fs_should_update_inplace(inode, fio); 2655 } 2656 2657 int f2fs_do_write_data_page(struct f2fs_io_info *fio) 2658 { 2659 struct folio *folio = page_folio(fio->page); 2660 struct inode *inode = folio->mapping->host; 2661 struct dnode_of_data dn; 2662 struct node_info ni; 2663 bool ipu_force = false; 2664 bool atomic_commit; 2665 int err = 0; 2666 2667 /* Use COW inode to make dnode_of_data for atomic write */ 2668 atomic_commit = f2fs_is_atomic_file(inode) && 2669 page_private_atomic(folio_page(folio, 0)); 2670 if (atomic_commit) 2671 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0); 2672 else 2673 set_new_dnode(&dn, inode, NULL, NULL, 0); 2674 2675 if (need_inplace_update(fio) && 2676 f2fs_lookup_read_extent_cache_block(inode, folio->index, 2677 &fio->old_blkaddr)) { 2678 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2679 DATA_GENERIC_ENHANCE)) 2680 return -EFSCORRUPTED; 2681 2682 ipu_force = true; 2683 fio->need_lock = LOCK_DONE; 2684 goto got_it; 2685 } 2686 2687 /* Deadlock due to between page->lock and f2fs_lock_op */ 2688 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 2689 return -EAGAIN; 2690 2691 err = f2fs_get_dnode_of_data(&dn, folio->index, LOOKUP_NODE); 2692 if (err) 2693 goto out; 2694 2695 fio->old_blkaddr = dn.data_blkaddr; 2696 2697 /* This page is already truncated */ 2698 if (fio->old_blkaddr == NULL_ADDR) { 2699 folio_clear_uptodate(folio); 2700 clear_page_private_gcing(folio_page(folio, 0)); 2701 goto out_writepage; 2702 } 2703 got_it: 2704 if (__is_valid_data_blkaddr(fio->old_blkaddr) && 2705 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2706 DATA_GENERIC_ENHANCE)) { 2707 err = -EFSCORRUPTED; 2708 goto out_writepage; 2709 } 2710 2711 /* wait for GCed page writeback via META_MAPPING */ 2712 if (fio->meta_gc) 2713 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr); 2714 2715 /* 2716 * If current allocation needs SSR, 2717 * it had better in-place writes for updated data. 2718 */ 2719 if (ipu_force || 2720 (__is_valid_data_blkaddr(fio->old_blkaddr) && 2721 need_inplace_update(fio))) { 2722 err = f2fs_encrypt_one_page(fio); 2723 if (err) 2724 goto out_writepage; 2725 2726 folio_start_writeback(folio); 2727 f2fs_put_dnode(&dn); 2728 if (fio->need_lock == LOCK_REQ) 2729 f2fs_unlock_op(fio->sbi); 2730 err = f2fs_inplace_write_data(fio); 2731 if (err) { 2732 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 2733 fscrypt_finalize_bounce_page(&fio->encrypted_page); 2734 folio_end_writeback(folio); 2735 } else { 2736 set_inode_flag(inode, FI_UPDATE_WRITE); 2737 } 2738 trace_f2fs_do_write_data_page(folio, IPU); 2739 return err; 2740 } 2741 2742 if (fio->need_lock == LOCK_RETRY) { 2743 if (!f2fs_trylock_op(fio->sbi)) { 2744 err = -EAGAIN; 2745 goto out_writepage; 2746 } 2747 fio->need_lock = LOCK_REQ; 2748 } 2749 2750 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false); 2751 if (err) 2752 goto out_writepage; 2753 2754 fio->version = ni.version; 2755 2756 err = f2fs_encrypt_one_page(fio); 2757 if (err) 2758 goto out_writepage; 2759 2760 folio_start_writeback(folio); 2761 2762 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR) 2763 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false); 2764 2765 /* LFS mode write path */ 2766 f2fs_outplace_write_data(&dn, fio); 2767 trace_f2fs_do_write_data_page(folio, OPU); 2768 set_inode_flag(inode, FI_APPEND_WRITE); 2769 if (atomic_commit) 2770 clear_page_private_atomic(folio_page(folio, 0)); 2771 out_writepage: 2772 f2fs_put_dnode(&dn); 2773 out: 2774 if (fio->need_lock == LOCK_REQ) 2775 f2fs_unlock_op(fio->sbi); 2776 return err; 2777 } 2778 2779 int f2fs_write_single_data_page(struct folio *folio, int *submitted, 2780 struct bio **bio, 2781 sector_t *last_block, 2782 struct writeback_control *wbc, 2783 enum iostat_type io_type, 2784 int compr_blocks, 2785 bool allow_balance) 2786 { 2787 struct inode *inode = folio->mapping->host; 2788 struct page *page = folio_page(folio, 0); 2789 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2790 loff_t i_size = i_size_read(inode); 2791 const pgoff_t end_index = ((unsigned long long)i_size) 2792 >> PAGE_SHIFT; 2793 loff_t psize = (loff_t)(folio->index + 1) << PAGE_SHIFT; 2794 unsigned offset = 0; 2795 bool need_balance_fs = false; 2796 bool quota_inode = IS_NOQUOTA(inode); 2797 int err = 0; 2798 struct f2fs_io_info fio = { 2799 .sbi = sbi, 2800 .ino = inode->i_ino, 2801 .type = DATA, 2802 .op = REQ_OP_WRITE, 2803 .op_flags = wbc_to_write_flags(wbc), 2804 .old_blkaddr = NULL_ADDR, 2805 .page = page, 2806 .encrypted_page = NULL, 2807 .submitted = 0, 2808 .compr_blocks = compr_blocks, 2809 .need_lock = compr_blocks ? LOCK_DONE : LOCK_RETRY, 2810 .meta_gc = f2fs_meta_inode_gc_required(inode) ? 1 : 0, 2811 .io_type = io_type, 2812 .io_wbc = wbc, 2813 .bio = bio, 2814 .last_block = last_block, 2815 }; 2816 2817 trace_f2fs_writepage(folio, DATA); 2818 2819 /* we should bypass data pages to proceed the kworker jobs */ 2820 if (unlikely(f2fs_cp_error(sbi))) { 2821 mapping_set_error(folio->mapping, -EIO); 2822 /* 2823 * don't drop any dirty dentry pages for keeping lastest 2824 * directory structure. 2825 */ 2826 if (S_ISDIR(inode->i_mode) && 2827 !is_sbi_flag_set(sbi, SBI_IS_CLOSE)) 2828 goto redirty_out; 2829 2830 /* keep data pages in remount-ro mode */ 2831 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY) 2832 goto redirty_out; 2833 goto out; 2834 } 2835 2836 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2837 goto redirty_out; 2838 2839 if (folio->index < end_index || 2840 f2fs_verity_in_progress(inode) || 2841 compr_blocks) 2842 goto write; 2843 2844 /* 2845 * If the offset is out-of-range of file size, 2846 * this page does not have to be written to disk. 2847 */ 2848 offset = i_size & (PAGE_SIZE - 1); 2849 if ((folio->index >= end_index + 1) || !offset) 2850 goto out; 2851 2852 folio_zero_segment(folio, offset, folio_size(folio)); 2853 write: 2854 /* Dentry/quota blocks are controlled by checkpoint */ 2855 if (S_ISDIR(inode->i_mode) || quota_inode) { 2856 /* 2857 * We need to wait for node_write to avoid block allocation during 2858 * checkpoint. This can only happen to quota writes which can cause 2859 * the below discard race condition. 2860 */ 2861 if (quota_inode) 2862 f2fs_down_read(&sbi->node_write); 2863 2864 fio.need_lock = LOCK_DONE; 2865 err = f2fs_do_write_data_page(&fio); 2866 2867 if (quota_inode) 2868 f2fs_up_read(&sbi->node_write); 2869 2870 goto done; 2871 } 2872 2873 if (!wbc->for_reclaim) 2874 need_balance_fs = true; 2875 else if (has_not_enough_free_secs(sbi, 0, 0)) 2876 goto redirty_out; 2877 else 2878 set_inode_flag(inode, FI_HOT_DATA); 2879 2880 err = -EAGAIN; 2881 if (f2fs_has_inline_data(inode)) { 2882 err = f2fs_write_inline_data(inode, folio); 2883 if (!err) 2884 goto out; 2885 } 2886 2887 if (err == -EAGAIN) { 2888 err = f2fs_do_write_data_page(&fio); 2889 if (err == -EAGAIN) { 2890 f2fs_bug_on(sbi, compr_blocks); 2891 fio.need_lock = LOCK_REQ; 2892 err = f2fs_do_write_data_page(&fio); 2893 } 2894 } 2895 2896 if (err) { 2897 file_set_keep_isize(inode); 2898 } else { 2899 spin_lock(&F2FS_I(inode)->i_size_lock); 2900 if (F2FS_I(inode)->last_disk_size < psize) 2901 F2FS_I(inode)->last_disk_size = psize; 2902 spin_unlock(&F2FS_I(inode)->i_size_lock); 2903 } 2904 2905 done: 2906 if (err && err != -ENOENT) 2907 goto redirty_out; 2908 2909 out: 2910 inode_dec_dirty_pages(inode); 2911 if (err) { 2912 folio_clear_uptodate(folio); 2913 clear_page_private_gcing(page); 2914 } 2915 2916 if (wbc->for_reclaim) { 2917 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA); 2918 clear_inode_flag(inode, FI_HOT_DATA); 2919 f2fs_remove_dirty_inode(inode); 2920 submitted = NULL; 2921 } 2922 folio_unlock(folio); 2923 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) && 2924 !F2FS_I(inode)->wb_task && allow_balance) 2925 f2fs_balance_fs(sbi, need_balance_fs); 2926 2927 if (unlikely(f2fs_cp_error(sbi))) { 2928 f2fs_submit_merged_write(sbi, DATA); 2929 if (bio && *bio) 2930 f2fs_submit_merged_ipu_write(sbi, bio, NULL); 2931 submitted = NULL; 2932 } 2933 2934 if (submitted) 2935 *submitted = fio.submitted; 2936 2937 return 0; 2938 2939 redirty_out: 2940 folio_redirty_for_writepage(wbc, folio); 2941 /* 2942 * pageout() in MM translates EAGAIN, so calls handle_write_error() 2943 * -> mapping_set_error() -> set_bit(AS_EIO, ...). 2944 * file_write_and_wait_range() will see EIO error, which is critical 2945 * to return value of fsync() followed by atomic_write failure to user. 2946 */ 2947 if (!err || wbc->for_reclaim) 2948 return AOP_WRITEPAGE_ACTIVATE; 2949 folio_unlock(folio); 2950 return err; 2951 } 2952 2953 static int f2fs_write_data_page(struct page *page, 2954 struct writeback_control *wbc) 2955 { 2956 struct folio *folio = page_folio(page); 2957 #ifdef CONFIG_F2FS_FS_COMPRESSION 2958 struct inode *inode = folio->mapping->host; 2959 2960 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 2961 goto out; 2962 2963 if (f2fs_compressed_file(inode)) { 2964 if (f2fs_is_compressed_cluster(inode, folio->index)) { 2965 folio_redirty_for_writepage(wbc, folio); 2966 return AOP_WRITEPAGE_ACTIVATE; 2967 } 2968 } 2969 out: 2970 #endif 2971 2972 return f2fs_write_single_data_page(folio, NULL, NULL, NULL, 2973 wbc, FS_DATA_IO, 0, true); 2974 } 2975 2976 /* 2977 * This function was copied from write_cache_pages from mm/page-writeback.c. 2978 * The major change is making write step of cold data page separately from 2979 * warm/hot data page. 2980 */ 2981 static int f2fs_write_cache_pages(struct address_space *mapping, 2982 struct writeback_control *wbc, 2983 enum iostat_type io_type) 2984 { 2985 int ret = 0; 2986 int done = 0, retry = 0; 2987 struct page *pages_local[F2FS_ONSTACK_PAGES]; 2988 struct page **pages = pages_local; 2989 struct folio_batch fbatch; 2990 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2991 struct bio *bio = NULL; 2992 sector_t last_block; 2993 #ifdef CONFIG_F2FS_FS_COMPRESSION 2994 struct inode *inode = mapping->host; 2995 struct compress_ctx cc = { 2996 .inode = inode, 2997 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2998 .cluster_size = F2FS_I(inode)->i_cluster_size, 2999 .cluster_idx = NULL_CLUSTER, 3000 .rpages = NULL, 3001 .nr_rpages = 0, 3002 .cpages = NULL, 3003 .valid_nr_cpages = 0, 3004 .rbuf = NULL, 3005 .cbuf = NULL, 3006 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size, 3007 .private = NULL, 3008 }; 3009 #endif 3010 int nr_folios, p, idx; 3011 int nr_pages; 3012 unsigned int max_pages = F2FS_ONSTACK_PAGES; 3013 pgoff_t index; 3014 pgoff_t end; /* Inclusive */ 3015 pgoff_t done_index; 3016 int range_whole = 0; 3017 xa_mark_t tag; 3018 int nwritten = 0; 3019 int submitted = 0; 3020 int i; 3021 3022 #ifdef CONFIG_F2FS_FS_COMPRESSION 3023 if (f2fs_compressed_file(inode) && 3024 1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) { 3025 pages = f2fs_kzalloc(sbi, sizeof(struct page *) << 3026 cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL); 3027 max_pages = 1 << cc.log_cluster_size; 3028 } 3029 #endif 3030 3031 folio_batch_init(&fbatch); 3032 3033 if (get_dirty_pages(mapping->host) <= 3034 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 3035 set_inode_flag(mapping->host, FI_HOT_DATA); 3036 else 3037 clear_inode_flag(mapping->host, FI_HOT_DATA); 3038 3039 if (wbc->range_cyclic) { 3040 index = mapping->writeback_index; /* prev offset */ 3041 end = -1; 3042 } else { 3043 index = wbc->range_start >> PAGE_SHIFT; 3044 end = wbc->range_end >> PAGE_SHIFT; 3045 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 3046 range_whole = 1; 3047 } 3048 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 3049 tag = PAGECACHE_TAG_TOWRITE; 3050 else 3051 tag = PAGECACHE_TAG_DIRTY; 3052 retry: 3053 retry = 0; 3054 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 3055 tag_pages_for_writeback(mapping, index, end); 3056 done_index = index; 3057 while (!done && !retry && (index <= end)) { 3058 nr_pages = 0; 3059 again: 3060 nr_folios = filemap_get_folios_tag(mapping, &index, end, 3061 tag, &fbatch); 3062 if (nr_folios == 0) { 3063 if (nr_pages) 3064 goto write; 3065 break; 3066 } 3067 3068 for (i = 0; i < nr_folios; i++) { 3069 struct folio *folio = fbatch.folios[i]; 3070 3071 idx = 0; 3072 p = folio_nr_pages(folio); 3073 add_more: 3074 pages[nr_pages] = folio_page(folio, idx); 3075 folio_get(folio); 3076 if (++nr_pages == max_pages) { 3077 index = folio->index + idx + 1; 3078 folio_batch_release(&fbatch); 3079 goto write; 3080 } 3081 if (++idx < p) 3082 goto add_more; 3083 } 3084 folio_batch_release(&fbatch); 3085 goto again; 3086 write: 3087 for (i = 0; i < nr_pages; i++) { 3088 struct page *page = pages[i]; 3089 struct folio *folio = page_folio(page); 3090 bool need_readd; 3091 readd: 3092 need_readd = false; 3093 #ifdef CONFIG_F2FS_FS_COMPRESSION 3094 if (f2fs_compressed_file(inode)) { 3095 void *fsdata = NULL; 3096 struct page *pagep; 3097 int ret2; 3098 3099 ret = f2fs_init_compress_ctx(&cc); 3100 if (ret) { 3101 done = 1; 3102 break; 3103 } 3104 3105 if (!f2fs_cluster_can_merge_page(&cc, 3106 folio->index)) { 3107 ret = f2fs_write_multi_pages(&cc, 3108 &submitted, wbc, io_type); 3109 if (!ret) 3110 need_readd = true; 3111 goto result; 3112 } 3113 3114 if (unlikely(f2fs_cp_error(sbi))) 3115 goto lock_folio; 3116 3117 if (!f2fs_cluster_is_empty(&cc)) 3118 goto lock_folio; 3119 3120 if (f2fs_all_cluster_page_ready(&cc, 3121 pages, i, nr_pages, true)) 3122 goto lock_folio; 3123 3124 ret2 = f2fs_prepare_compress_overwrite( 3125 inode, &pagep, 3126 folio->index, &fsdata); 3127 if (ret2 < 0) { 3128 ret = ret2; 3129 done = 1; 3130 break; 3131 } else if (ret2 && 3132 (!f2fs_compress_write_end(inode, 3133 fsdata, folio->index, 1) || 3134 !f2fs_all_cluster_page_ready(&cc, 3135 pages, i, nr_pages, 3136 false))) { 3137 retry = 1; 3138 break; 3139 } 3140 } 3141 #endif 3142 /* give a priority to WB_SYNC threads */ 3143 if (atomic_read(&sbi->wb_sync_req[DATA]) && 3144 wbc->sync_mode == WB_SYNC_NONE) { 3145 done = 1; 3146 break; 3147 } 3148 #ifdef CONFIG_F2FS_FS_COMPRESSION 3149 lock_folio: 3150 #endif 3151 done_index = folio->index; 3152 retry_write: 3153 folio_lock(folio); 3154 3155 if (unlikely(folio->mapping != mapping)) { 3156 continue_unlock: 3157 folio_unlock(folio); 3158 continue; 3159 } 3160 3161 if (!folio_test_dirty(folio)) { 3162 /* someone wrote it for us */ 3163 goto continue_unlock; 3164 } 3165 3166 if (folio_test_writeback(folio)) { 3167 if (wbc->sync_mode == WB_SYNC_NONE) 3168 goto continue_unlock; 3169 f2fs_wait_on_page_writeback(&folio->page, DATA, true, true); 3170 } 3171 3172 if (!folio_clear_dirty_for_io(folio)) 3173 goto continue_unlock; 3174 3175 #ifdef CONFIG_F2FS_FS_COMPRESSION 3176 if (f2fs_compressed_file(inode)) { 3177 folio_get(folio); 3178 f2fs_compress_ctx_add_page(&cc, folio); 3179 continue; 3180 } 3181 #endif 3182 ret = f2fs_write_single_data_page(folio, 3183 &submitted, &bio, &last_block, 3184 wbc, io_type, 0, true); 3185 if (ret == AOP_WRITEPAGE_ACTIVATE) 3186 folio_unlock(folio); 3187 #ifdef CONFIG_F2FS_FS_COMPRESSION 3188 result: 3189 #endif 3190 nwritten += submitted; 3191 wbc->nr_to_write -= submitted; 3192 3193 if (unlikely(ret)) { 3194 /* 3195 * keep nr_to_write, since vfs uses this to 3196 * get # of written pages. 3197 */ 3198 if (ret == AOP_WRITEPAGE_ACTIVATE) { 3199 ret = 0; 3200 goto next; 3201 } else if (ret == -EAGAIN) { 3202 ret = 0; 3203 if (wbc->sync_mode == WB_SYNC_ALL) { 3204 f2fs_io_schedule_timeout( 3205 DEFAULT_IO_TIMEOUT); 3206 goto retry_write; 3207 } 3208 goto next; 3209 } 3210 done_index = folio_next_index(folio); 3211 done = 1; 3212 break; 3213 } 3214 3215 if (wbc->nr_to_write <= 0 && 3216 wbc->sync_mode == WB_SYNC_NONE) { 3217 done = 1; 3218 break; 3219 } 3220 next: 3221 if (need_readd) 3222 goto readd; 3223 } 3224 release_pages(pages, nr_pages); 3225 cond_resched(); 3226 } 3227 #ifdef CONFIG_F2FS_FS_COMPRESSION 3228 /* flush remained pages in compress cluster */ 3229 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) { 3230 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type); 3231 nwritten += submitted; 3232 wbc->nr_to_write -= submitted; 3233 if (ret) { 3234 done = 1; 3235 retry = 0; 3236 } 3237 } 3238 if (f2fs_compressed_file(inode)) 3239 f2fs_destroy_compress_ctx(&cc, false); 3240 #endif 3241 if (retry) { 3242 index = 0; 3243 end = -1; 3244 goto retry; 3245 } 3246 if (wbc->range_cyclic && !done) 3247 done_index = 0; 3248 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 3249 mapping->writeback_index = done_index; 3250 3251 if (nwritten) 3252 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 3253 NULL, 0, DATA); 3254 /* submit cached bio of IPU write */ 3255 if (bio) 3256 f2fs_submit_merged_ipu_write(sbi, &bio, NULL); 3257 3258 #ifdef CONFIG_F2FS_FS_COMPRESSION 3259 if (pages != pages_local) 3260 kfree(pages); 3261 #endif 3262 3263 return ret; 3264 } 3265 3266 static inline bool __should_serialize_io(struct inode *inode, 3267 struct writeback_control *wbc) 3268 { 3269 /* to avoid deadlock in path of data flush */ 3270 if (F2FS_I(inode)->wb_task) 3271 return false; 3272 3273 if (!S_ISREG(inode->i_mode)) 3274 return false; 3275 if (IS_NOQUOTA(inode)) 3276 return false; 3277 3278 if (f2fs_need_compress_data(inode)) 3279 return true; 3280 if (wbc->sync_mode != WB_SYNC_ALL) 3281 return true; 3282 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks) 3283 return true; 3284 return false; 3285 } 3286 3287 static int __f2fs_write_data_pages(struct address_space *mapping, 3288 struct writeback_control *wbc, 3289 enum iostat_type io_type) 3290 { 3291 struct inode *inode = mapping->host; 3292 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3293 struct blk_plug plug; 3294 int ret; 3295 bool locked = false; 3296 3297 /* deal with chardevs and other special file */ 3298 if (!mapping->a_ops->writepage) 3299 return 0; 3300 3301 /* skip writing if there is no dirty page in this inode */ 3302 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 3303 return 0; 3304 3305 /* during POR, we don't need to trigger writepage at all. */ 3306 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 3307 goto skip_write; 3308 3309 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) && 3310 wbc->sync_mode == WB_SYNC_NONE && 3311 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 3312 f2fs_available_free_memory(sbi, DIRTY_DENTS)) 3313 goto skip_write; 3314 3315 /* skip writing in file defragment preparing stage */ 3316 if (is_inode_flag_set(inode, FI_SKIP_WRITES)) 3317 goto skip_write; 3318 3319 trace_f2fs_writepages(mapping->host, wbc, DATA); 3320 3321 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 3322 if (wbc->sync_mode == WB_SYNC_ALL) 3323 atomic_inc(&sbi->wb_sync_req[DATA]); 3324 else if (atomic_read(&sbi->wb_sync_req[DATA])) { 3325 /* to avoid potential deadlock */ 3326 if (current->plug) 3327 blk_finish_plug(current->plug); 3328 goto skip_write; 3329 } 3330 3331 if (__should_serialize_io(inode, wbc)) { 3332 mutex_lock(&sbi->writepages); 3333 locked = true; 3334 } 3335 3336 blk_start_plug(&plug); 3337 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 3338 blk_finish_plug(&plug); 3339 3340 if (locked) 3341 mutex_unlock(&sbi->writepages); 3342 3343 if (wbc->sync_mode == WB_SYNC_ALL) 3344 atomic_dec(&sbi->wb_sync_req[DATA]); 3345 /* 3346 * if some pages were truncated, we cannot guarantee its mapping->host 3347 * to detect pending bios. 3348 */ 3349 3350 f2fs_remove_dirty_inode(inode); 3351 return ret; 3352 3353 skip_write: 3354 wbc->pages_skipped += get_dirty_pages(inode); 3355 trace_f2fs_writepages(mapping->host, wbc, DATA); 3356 return 0; 3357 } 3358 3359 static int f2fs_write_data_pages(struct address_space *mapping, 3360 struct writeback_control *wbc) 3361 { 3362 struct inode *inode = mapping->host; 3363 3364 return __f2fs_write_data_pages(mapping, wbc, 3365 F2FS_I(inode)->cp_task == current ? 3366 FS_CP_DATA_IO : FS_DATA_IO); 3367 } 3368 3369 void f2fs_write_failed(struct inode *inode, loff_t to) 3370 { 3371 loff_t i_size = i_size_read(inode); 3372 3373 if (IS_NOQUOTA(inode)) 3374 return; 3375 3376 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */ 3377 if (to > i_size && !f2fs_verity_in_progress(inode)) { 3378 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3379 filemap_invalidate_lock(inode->i_mapping); 3380 3381 truncate_pagecache(inode, i_size); 3382 f2fs_truncate_blocks(inode, i_size, true); 3383 3384 filemap_invalidate_unlock(inode->i_mapping); 3385 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3386 } 3387 } 3388 3389 static int prepare_write_begin(struct f2fs_sb_info *sbi, 3390 struct folio *folio, loff_t pos, unsigned int len, 3391 block_t *blk_addr, bool *node_changed) 3392 { 3393 struct inode *inode = folio->mapping->host; 3394 pgoff_t index = folio->index; 3395 struct dnode_of_data dn; 3396 struct page *ipage; 3397 bool locked = false; 3398 int flag = F2FS_GET_BLOCK_PRE_AIO; 3399 int err = 0; 3400 3401 /* 3402 * If a whole page is being written and we already preallocated all the 3403 * blocks, then there is no need to get a block address now. 3404 */ 3405 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL)) 3406 return 0; 3407 3408 /* f2fs_lock_op avoids race between write CP and convert_inline_page */ 3409 if (f2fs_has_inline_data(inode)) { 3410 if (pos + len > MAX_INLINE_DATA(inode)) 3411 flag = F2FS_GET_BLOCK_DEFAULT; 3412 f2fs_map_lock(sbi, flag); 3413 locked = true; 3414 } else if ((pos & PAGE_MASK) >= i_size_read(inode)) { 3415 f2fs_map_lock(sbi, flag); 3416 locked = true; 3417 } 3418 3419 restart: 3420 /* check inline_data */ 3421 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3422 if (IS_ERR(ipage)) { 3423 err = PTR_ERR(ipage); 3424 goto unlock_out; 3425 } 3426 3427 set_new_dnode(&dn, inode, ipage, ipage, 0); 3428 3429 if (f2fs_has_inline_data(inode)) { 3430 if (pos + len <= MAX_INLINE_DATA(inode)) { 3431 f2fs_do_read_inline_data(folio, ipage); 3432 set_inode_flag(inode, FI_DATA_EXIST); 3433 if (inode->i_nlink) 3434 set_page_private_inline(ipage); 3435 goto out; 3436 } 3437 err = f2fs_convert_inline_page(&dn, folio_page(folio, 0)); 3438 if (err || dn.data_blkaddr != NULL_ADDR) 3439 goto out; 3440 } 3441 3442 if (!f2fs_lookup_read_extent_cache_block(inode, index, 3443 &dn.data_blkaddr)) { 3444 if (locked) { 3445 err = f2fs_reserve_block(&dn, index); 3446 goto out; 3447 } 3448 3449 /* hole case */ 3450 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3451 if (!err && dn.data_blkaddr != NULL_ADDR) 3452 goto out; 3453 f2fs_put_dnode(&dn); 3454 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO); 3455 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO); 3456 locked = true; 3457 goto restart; 3458 } 3459 out: 3460 if (!err) { 3461 /* convert_inline_page can make node_changed */ 3462 *blk_addr = dn.data_blkaddr; 3463 *node_changed = dn.node_changed; 3464 } 3465 f2fs_put_dnode(&dn); 3466 unlock_out: 3467 if (locked) 3468 f2fs_map_unlock(sbi, flag); 3469 return err; 3470 } 3471 3472 static int __find_data_block(struct inode *inode, pgoff_t index, 3473 block_t *blk_addr) 3474 { 3475 struct dnode_of_data dn; 3476 struct page *ipage; 3477 int err = 0; 3478 3479 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 3480 if (IS_ERR(ipage)) 3481 return PTR_ERR(ipage); 3482 3483 set_new_dnode(&dn, inode, ipage, ipage, 0); 3484 3485 if (!f2fs_lookup_read_extent_cache_block(inode, index, 3486 &dn.data_blkaddr)) { 3487 /* hole case */ 3488 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3489 if (err) { 3490 dn.data_blkaddr = NULL_ADDR; 3491 err = 0; 3492 } 3493 } 3494 *blk_addr = dn.data_blkaddr; 3495 f2fs_put_dnode(&dn); 3496 return err; 3497 } 3498 3499 static int __reserve_data_block(struct inode *inode, pgoff_t index, 3500 block_t *blk_addr, bool *node_changed) 3501 { 3502 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3503 struct dnode_of_data dn; 3504 struct page *ipage; 3505 int err = 0; 3506 3507 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO); 3508 3509 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3510 if (IS_ERR(ipage)) { 3511 err = PTR_ERR(ipage); 3512 goto unlock_out; 3513 } 3514 set_new_dnode(&dn, inode, ipage, ipage, 0); 3515 3516 if (!f2fs_lookup_read_extent_cache_block(dn.inode, index, 3517 &dn.data_blkaddr)) 3518 err = f2fs_reserve_block(&dn, index); 3519 3520 *blk_addr = dn.data_blkaddr; 3521 *node_changed = dn.node_changed; 3522 f2fs_put_dnode(&dn); 3523 3524 unlock_out: 3525 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO); 3526 return err; 3527 } 3528 3529 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi, 3530 struct folio *folio, loff_t pos, unsigned int len, 3531 block_t *blk_addr, bool *node_changed, bool *use_cow) 3532 { 3533 struct inode *inode = folio->mapping->host; 3534 struct inode *cow_inode = F2FS_I(inode)->cow_inode; 3535 pgoff_t index = folio->index; 3536 int err = 0; 3537 block_t ori_blk_addr = NULL_ADDR; 3538 3539 /* If pos is beyond the end of file, reserve a new block in COW inode */ 3540 if ((pos & PAGE_MASK) >= i_size_read(inode)) 3541 goto reserve_block; 3542 3543 /* Look for the block in COW inode first */ 3544 err = __find_data_block(cow_inode, index, blk_addr); 3545 if (err) { 3546 return err; 3547 } else if (*blk_addr != NULL_ADDR) { 3548 *use_cow = true; 3549 return 0; 3550 } 3551 3552 if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE)) 3553 goto reserve_block; 3554 3555 /* Look for the block in the original inode */ 3556 err = __find_data_block(inode, index, &ori_blk_addr); 3557 if (err) 3558 return err; 3559 3560 reserve_block: 3561 /* Finally, we should reserve a new block in COW inode for the update */ 3562 err = __reserve_data_block(cow_inode, index, blk_addr, node_changed); 3563 if (err) 3564 return err; 3565 inc_atomic_write_cnt(inode); 3566 3567 if (ori_blk_addr != NULL_ADDR) 3568 *blk_addr = ori_blk_addr; 3569 return 0; 3570 } 3571 3572 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 3573 loff_t pos, unsigned len, struct folio **foliop, void **fsdata) 3574 { 3575 struct inode *inode = mapping->host; 3576 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3577 struct folio *folio; 3578 pgoff_t index = pos >> PAGE_SHIFT; 3579 bool need_balance = false; 3580 bool use_cow = false; 3581 block_t blkaddr = NULL_ADDR; 3582 int err = 0; 3583 3584 trace_f2fs_write_begin(inode, pos, len); 3585 3586 if (!f2fs_is_checkpoint_ready(sbi)) { 3587 err = -ENOSPC; 3588 goto fail; 3589 } 3590 3591 /* 3592 * We should check this at this moment to avoid deadlock on inode page 3593 * and #0 page. The locking rule for inline_data conversion should be: 3594 * folio_lock(folio #0) -> folio_lock(inode_page) 3595 */ 3596 if (index != 0) { 3597 err = f2fs_convert_inline_inode(inode); 3598 if (err) 3599 goto fail; 3600 } 3601 3602 #ifdef CONFIG_F2FS_FS_COMPRESSION 3603 if (f2fs_compressed_file(inode)) { 3604 int ret; 3605 struct page *page; 3606 3607 *fsdata = NULL; 3608 3609 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode))) 3610 goto repeat; 3611 3612 ret = f2fs_prepare_compress_overwrite(inode, &page, 3613 index, fsdata); 3614 if (ret < 0) { 3615 err = ret; 3616 goto fail; 3617 } else if (ret) { 3618 *foliop = page_folio(page); 3619 return 0; 3620 } 3621 } 3622 #endif 3623 3624 repeat: 3625 /* 3626 * Do not use FGP_STABLE to avoid deadlock. 3627 * Will wait that below with our IO control. 3628 */ 3629 folio = __filemap_get_folio(mapping, index, 3630 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 3631 if (IS_ERR(folio)) { 3632 err = PTR_ERR(folio); 3633 goto fail; 3634 } 3635 3636 /* TODO: cluster can be compressed due to race with .writepage */ 3637 3638 *foliop = folio; 3639 3640 if (f2fs_is_atomic_file(inode)) 3641 err = prepare_atomic_write_begin(sbi, folio, pos, len, 3642 &blkaddr, &need_balance, &use_cow); 3643 else 3644 err = prepare_write_begin(sbi, folio, pos, len, 3645 &blkaddr, &need_balance); 3646 if (err) 3647 goto put_folio; 3648 3649 if (need_balance && !IS_NOQUOTA(inode) && 3650 has_not_enough_free_secs(sbi, 0, 0)) { 3651 folio_unlock(folio); 3652 f2fs_balance_fs(sbi, true); 3653 folio_lock(folio); 3654 if (folio->mapping != mapping) { 3655 /* The folio got truncated from under us */ 3656 folio_unlock(folio); 3657 folio_put(folio); 3658 goto repeat; 3659 } 3660 } 3661 3662 f2fs_wait_on_page_writeback(&folio->page, DATA, false, true); 3663 3664 if (len == folio_size(folio) || folio_test_uptodate(folio)) 3665 return 0; 3666 3667 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) && 3668 !f2fs_verity_in_progress(inode)) { 3669 folio_zero_segment(folio, len, folio_size(folio)); 3670 return 0; 3671 } 3672 3673 if (blkaddr == NEW_ADDR) { 3674 folio_zero_segment(folio, 0, folio_size(folio)); 3675 folio_mark_uptodate(folio); 3676 } else { 3677 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 3678 DATA_GENERIC_ENHANCE_READ)) { 3679 err = -EFSCORRUPTED; 3680 goto put_folio; 3681 } 3682 err = f2fs_submit_page_read(use_cow ? 3683 F2FS_I(inode)->cow_inode : inode, 3684 folio, blkaddr, 0, true); 3685 if (err) 3686 goto put_folio; 3687 3688 folio_lock(folio); 3689 if (unlikely(folio->mapping != mapping)) { 3690 folio_unlock(folio); 3691 folio_put(folio); 3692 goto repeat; 3693 } 3694 if (unlikely(!folio_test_uptodate(folio))) { 3695 err = -EIO; 3696 goto put_folio; 3697 } 3698 } 3699 return 0; 3700 3701 put_folio: 3702 folio_unlock(folio); 3703 folio_put(folio); 3704 fail: 3705 f2fs_write_failed(inode, pos + len); 3706 return err; 3707 } 3708 3709 static int f2fs_write_end(struct file *file, 3710 struct address_space *mapping, 3711 loff_t pos, unsigned len, unsigned copied, 3712 struct folio *folio, void *fsdata) 3713 { 3714 struct inode *inode = folio->mapping->host; 3715 3716 trace_f2fs_write_end(inode, pos, len, copied); 3717 3718 /* 3719 * This should be come from len == PAGE_SIZE, and we expect copied 3720 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 3721 * let generic_perform_write() try to copy data again through copied=0. 3722 */ 3723 if (!folio_test_uptodate(folio)) { 3724 if (unlikely(copied != len)) 3725 copied = 0; 3726 else 3727 folio_mark_uptodate(folio); 3728 } 3729 3730 #ifdef CONFIG_F2FS_FS_COMPRESSION 3731 /* overwrite compressed file */ 3732 if (f2fs_compressed_file(inode) && fsdata) { 3733 f2fs_compress_write_end(inode, fsdata, folio->index, copied); 3734 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3735 3736 if (pos + copied > i_size_read(inode) && 3737 !f2fs_verity_in_progress(inode)) 3738 f2fs_i_size_write(inode, pos + copied); 3739 return copied; 3740 } 3741 #endif 3742 3743 if (!copied) 3744 goto unlock_out; 3745 3746 folio_mark_dirty(folio); 3747 3748 if (f2fs_is_atomic_file(inode)) 3749 set_page_private_atomic(folio_page(folio, 0)); 3750 3751 if (pos + copied > i_size_read(inode) && 3752 !f2fs_verity_in_progress(inode)) { 3753 f2fs_i_size_write(inode, pos + copied); 3754 if (f2fs_is_atomic_file(inode)) 3755 f2fs_i_size_write(F2FS_I(inode)->cow_inode, 3756 pos + copied); 3757 } 3758 unlock_out: 3759 folio_unlock(folio); 3760 folio_put(folio); 3761 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3762 return copied; 3763 } 3764 3765 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length) 3766 { 3767 struct inode *inode = folio->mapping->host; 3768 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3769 3770 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 3771 (offset || length != folio_size(folio))) 3772 return; 3773 3774 if (folio_test_dirty(folio)) { 3775 if (inode->i_ino == F2FS_META_INO(sbi)) { 3776 dec_page_count(sbi, F2FS_DIRTY_META); 3777 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 3778 dec_page_count(sbi, F2FS_DIRTY_NODES); 3779 } else { 3780 inode_dec_dirty_pages(inode); 3781 f2fs_remove_dirty_inode(inode); 3782 } 3783 } 3784 clear_page_private_all(&folio->page); 3785 } 3786 3787 bool f2fs_release_folio(struct folio *folio, gfp_t wait) 3788 { 3789 /* If this is dirty folio, keep private data */ 3790 if (folio_test_dirty(folio)) 3791 return false; 3792 3793 clear_page_private_all(&folio->page); 3794 return true; 3795 } 3796 3797 static bool f2fs_dirty_data_folio(struct address_space *mapping, 3798 struct folio *folio) 3799 { 3800 struct inode *inode = mapping->host; 3801 3802 trace_f2fs_set_page_dirty(folio, DATA); 3803 3804 if (!folio_test_uptodate(folio)) 3805 folio_mark_uptodate(folio); 3806 BUG_ON(folio_test_swapcache(folio)); 3807 3808 if (filemap_dirty_folio(mapping, folio)) { 3809 f2fs_update_dirty_folio(inode, folio); 3810 return true; 3811 } 3812 return false; 3813 } 3814 3815 3816 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block) 3817 { 3818 #ifdef CONFIG_F2FS_FS_COMPRESSION 3819 struct dnode_of_data dn; 3820 sector_t start_idx, blknr = 0; 3821 int ret; 3822 3823 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size); 3824 3825 set_new_dnode(&dn, inode, NULL, NULL, 0); 3826 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 3827 if (ret) 3828 return 0; 3829 3830 if (dn.data_blkaddr != COMPRESS_ADDR) { 3831 dn.ofs_in_node += block - start_idx; 3832 blknr = f2fs_data_blkaddr(&dn); 3833 if (!__is_valid_data_blkaddr(blknr)) 3834 blknr = 0; 3835 } 3836 3837 f2fs_put_dnode(&dn); 3838 return blknr; 3839 #else 3840 return 0; 3841 #endif 3842 } 3843 3844 3845 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 3846 { 3847 struct inode *inode = mapping->host; 3848 sector_t blknr = 0; 3849 3850 if (f2fs_has_inline_data(inode)) 3851 goto out; 3852 3853 /* make sure allocating whole blocks */ 3854 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 3855 filemap_write_and_wait(mapping); 3856 3857 /* Block number less than F2FS MAX BLOCKS */ 3858 if (unlikely(block >= max_file_blocks(inode))) 3859 goto out; 3860 3861 if (f2fs_compressed_file(inode)) { 3862 blknr = f2fs_bmap_compress(inode, block); 3863 } else { 3864 struct f2fs_map_blocks map; 3865 3866 memset(&map, 0, sizeof(map)); 3867 map.m_lblk = block; 3868 map.m_len = 1; 3869 map.m_next_pgofs = NULL; 3870 map.m_seg_type = NO_CHECK_TYPE; 3871 3872 if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP)) 3873 blknr = map.m_pblk; 3874 } 3875 out: 3876 trace_f2fs_bmap(inode, block, blknr); 3877 return blknr; 3878 } 3879 3880 #ifdef CONFIG_SWAP 3881 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk, 3882 unsigned int blkcnt) 3883 { 3884 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3885 unsigned int blkofs; 3886 unsigned int blk_per_sec = BLKS_PER_SEC(sbi); 3887 unsigned int end_blk = start_blk + blkcnt - 1; 3888 unsigned int secidx = start_blk / blk_per_sec; 3889 unsigned int end_sec; 3890 int ret = 0; 3891 3892 if (!blkcnt) 3893 return 0; 3894 end_sec = end_blk / blk_per_sec; 3895 3896 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3897 filemap_invalidate_lock(inode->i_mapping); 3898 3899 set_inode_flag(inode, FI_ALIGNED_WRITE); 3900 set_inode_flag(inode, FI_OPU_WRITE); 3901 3902 for (; secidx <= end_sec; secidx++) { 3903 unsigned int blkofs_end = secidx == end_sec ? 3904 end_blk % blk_per_sec : blk_per_sec - 1; 3905 3906 f2fs_down_write(&sbi->pin_sem); 3907 3908 ret = f2fs_allocate_pinning_section(sbi); 3909 if (ret) { 3910 f2fs_up_write(&sbi->pin_sem); 3911 break; 3912 } 3913 3914 set_inode_flag(inode, FI_SKIP_WRITES); 3915 3916 for (blkofs = 0; blkofs <= blkofs_end; blkofs++) { 3917 struct page *page; 3918 unsigned int blkidx = secidx * blk_per_sec + blkofs; 3919 3920 page = f2fs_get_lock_data_page(inode, blkidx, true); 3921 if (IS_ERR(page)) { 3922 f2fs_up_write(&sbi->pin_sem); 3923 ret = PTR_ERR(page); 3924 goto done; 3925 } 3926 3927 set_page_dirty(page); 3928 f2fs_put_page(page, 1); 3929 } 3930 3931 clear_inode_flag(inode, FI_SKIP_WRITES); 3932 3933 ret = filemap_fdatawrite(inode->i_mapping); 3934 3935 f2fs_up_write(&sbi->pin_sem); 3936 3937 if (ret) 3938 break; 3939 } 3940 3941 done: 3942 clear_inode_flag(inode, FI_SKIP_WRITES); 3943 clear_inode_flag(inode, FI_OPU_WRITE); 3944 clear_inode_flag(inode, FI_ALIGNED_WRITE); 3945 3946 filemap_invalidate_unlock(inode->i_mapping); 3947 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3948 3949 return ret; 3950 } 3951 3952 static int check_swap_activate(struct swap_info_struct *sis, 3953 struct file *swap_file, sector_t *span) 3954 { 3955 struct address_space *mapping = swap_file->f_mapping; 3956 struct inode *inode = mapping->host; 3957 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3958 block_t cur_lblock; 3959 block_t last_lblock; 3960 block_t pblock; 3961 block_t lowest_pblock = -1; 3962 block_t highest_pblock = 0; 3963 int nr_extents = 0; 3964 unsigned int nr_pblocks; 3965 unsigned int blks_per_sec = BLKS_PER_SEC(sbi); 3966 unsigned int not_aligned = 0; 3967 int ret = 0; 3968 3969 /* 3970 * Map all the blocks into the extent list. This code doesn't try 3971 * to be very smart. 3972 */ 3973 cur_lblock = 0; 3974 last_lblock = bytes_to_blks(inode, i_size_read(inode)); 3975 3976 while (cur_lblock < last_lblock && cur_lblock < sis->max) { 3977 struct f2fs_map_blocks map; 3978 retry: 3979 cond_resched(); 3980 3981 memset(&map, 0, sizeof(map)); 3982 map.m_lblk = cur_lblock; 3983 map.m_len = last_lblock - cur_lblock; 3984 map.m_next_pgofs = NULL; 3985 map.m_next_extent = NULL; 3986 map.m_seg_type = NO_CHECK_TYPE; 3987 map.m_may_create = false; 3988 3989 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP); 3990 if (ret) 3991 goto out; 3992 3993 /* hole */ 3994 if (!(map.m_flags & F2FS_MAP_FLAGS)) { 3995 f2fs_err(sbi, "Swapfile has holes"); 3996 ret = -EINVAL; 3997 goto out; 3998 } 3999 4000 pblock = map.m_pblk; 4001 nr_pblocks = map.m_len; 4002 4003 if ((pblock - SM_I(sbi)->main_blkaddr) % blks_per_sec || 4004 nr_pblocks % blks_per_sec || 4005 !f2fs_valid_pinned_area(sbi, pblock)) { 4006 bool last_extent = false; 4007 4008 not_aligned++; 4009 4010 nr_pblocks = roundup(nr_pblocks, blks_per_sec); 4011 if (cur_lblock + nr_pblocks > sis->max) 4012 nr_pblocks -= blks_per_sec; 4013 4014 /* this extent is last one */ 4015 if (!nr_pblocks) { 4016 nr_pblocks = last_lblock - cur_lblock; 4017 last_extent = true; 4018 } 4019 4020 ret = f2fs_migrate_blocks(inode, cur_lblock, 4021 nr_pblocks); 4022 if (ret) { 4023 if (ret == -ENOENT) 4024 ret = -EINVAL; 4025 goto out; 4026 } 4027 4028 if (!last_extent) 4029 goto retry; 4030 } 4031 4032 if (cur_lblock + nr_pblocks >= sis->max) 4033 nr_pblocks = sis->max - cur_lblock; 4034 4035 if (cur_lblock) { /* exclude the header page */ 4036 if (pblock < lowest_pblock) 4037 lowest_pblock = pblock; 4038 if (pblock + nr_pblocks - 1 > highest_pblock) 4039 highest_pblock = pblock + nr_pblocks - 1; 4040 } 4041 4042 /* 4043 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks 4044 */ 4045 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock); 4046 if (ret < 0) 4047 goto out; 4048 nr_extents += ret; 4049 cur_lblock += nr_pblocks; 4050 } 4051 ret = nr_extents; 4052 *span = 1 + highest_pblock - lowest_pblock; 4053 if (cur_lblock == 0) 4054 cur_lblock = 1; /* force Empty message */ 4055 sis->max = cur_lblock; 4056 sis->pages = cur_lblock - 1; 4057 sis->highest_bit = cur_lblock - 1; 4058 out: 4059 if (not_aligned) 4060 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%lu * N)", 4061 not_aligned, blks_per_sec * F2FS_BLKSIZE); 4062 return ret; 4063 } 4064 4065 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 4066 sector_t *span) 4067 { 4068 struct inode *inode = file_inode(file); 4069 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4070 int ret; 4071 4072 if (!S_ISREG(inode->i_mode)) 4073 return -EINVAL; 4074 4075 if (f2fs_readonly(sbi->sb)) 4076 return -EROFS; 4077 4078 if (f2fs_lfs_mode(sbi) && !f2fs_sb_has_blkzoned(sbi)) { 4079 f2fs_err(sbi, "Swapfile not supported in LFS mode"); 4080 return -EINVAL; 4081 } 4082 4083 ret = f2fs_convert_inline_inode(inode); 4084 if (ret) 4085 return ret; 4086 4087 if (!f2fs_disable_compressed_file(inode)) 4088 return -EINVAL; 4089 4090 ret = filemap_fdatawrite(inode->i_mapping); 4091 if (ret < 0) 4092 return ret; 4093 4094 f2fs_precache_extents(inode); 4095 4096 ret = check_swap_activate(sis, file, span); 4097 if (ret < 0) 4098 return ret; 4099 4100 stat_inc_swapfile_inode(inode); 4101 set_inode_flag(inode, FI_PIN_FILE); 4102 f2fs_update_time(sbi, REQ_TIME); 4103 return ret; 4104 } 4105 4106 static void f2fs_swap_deactivate(struct file *file) 4107 { 4108 struct inode *inode = file_inode(file); 4109 4110 stat_dec_swapfile_inode(inode); 4111 clear_inode_flag(inode, FI_PIN_FILE); 4112 } 4113 #else 4114 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 4115 sector_t *span) 4116 { 4117 return -EOPNOTSUPP; 4118 } 4119 4120 static void f2fs_swap_deactivate(struct file *file) 4121 { 4122 } 4123 #endif 4124 4125 const struct address_space_operations f2fs_dblock_aops = { 4126 .read_folio = f2fs_read_data_folio, 4127 .readahead = f2fs_readahead, 4128 .writepage = f2fs_write_data_page, 4129 .writepages = f2fs_write_data_pages, 4130 .write_begin = f2fs_write_begin, 4131 .write_end = f2fs_write_end, 4132 .dirty_folio = f2fs_dirty_data_folio, 4133 .migrate_folio = filemap_migrate_folio, 4134 .invalidate_folio = f2fs_invalidate_folio, 4135 .release_folio = f2fs_release_folio, 4136 .bmap = f2fs_bmap, 4137 .swap_activate = f2fs_swap_activate, 4138 .swap_deactivate = f2fs_swap_deactivate, 4139 }; 4140 4141 void f2fs_clear_page_cache_dirty_tag(struct folio *folio) 4142 { 4143 struct address_space *mapping = folio->mapping; 4144 unsigned long flags; 4145 4146 xa_lock_irqsave(&mapping->i_pages, flags); 4147 __xa_clear_mark(&mapping->i_pages, folio->index, 4148 PAGECACHE_TAG_DIRTY); 4149 xa_unlock_irqrestore(&mapping->i_pages, flags); 4150 } 4151 4152 int __init f2fs_init_post_read_processing(void) 4153 { 4154 bio_post_read_ctx_cache = 4155 kmem_cache_create("f2fs_bio_post_read_ctx", 4156 sizeof(struct bio_post_read_ctx), 0, 0, NULL); 4157 if (!bio_post_read_ctx_cache) 4158 goto fail; 4159 bio_post_read_ctx_pool = 4160 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 4161 bio_post_read_ctx_cache); 4162 if (!bio_post_read_ctx_pool) 4163 goto fail_free_cache; 4164 return 0; 4165 4166 fail_free_cache: 4167 kmem_cache_destroy(bio_post_read_ctx_cache); 4168 fail: 4169 return -ENOMEM; 4170 } 4171 4172 void f2fs_destroy_post_read_processing(void) 4173 { 4174 mempool_destroy(bio_post_read_ctx_pool); 4175 kmem_cache_destroy(bio_post_read_ctx_cache); 4176 } 4177 4178 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) 4179 { 4180 if (!f2fs_sb_has_encrypt(sbi) && 4181 !f2fs_sb_has_verity(sbi) && 4182 !f2fs_sb_has_compression(sbi)) 4183 return 0; 4184 4185 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq", 4186 WQ_UNBOUND | WQ_HIGHPRI, 4187 num_online_cpus()); 4188 return sbi->post_read_wq ? 0 : -ENOMEM; 4189 } 4190 4191 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) 4192 { 4193 if (sbi->post_read_wq) 4194 destroy_workqueue(sbi->post_read_wq); 4195 } 4196 4197 int __init f2fs_init_bio_entry_cache(void) 4198 { 4199 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab", 4200 sizeof(struct bio_entry)); 4201 return bio_entry_slab ? 0 : -ENOMEM; 4202 } 4203 4204 void f2fs_destroy_bio_entry_cache(void) 4205 { 4206 kmem_cache_destroy(bio_entry_slab); 4207 } 4208 4209 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 4210 unsigned int flags, struct iomap *iomap, 4211 struct iomap *srcmap) 4212 { 4213 struct f2fs_map_blocks map = {}; 4214 pgoff_t next_pgofs = 0; 4215 int err; 4216 4217 map.m_lblk = bytes_to_blks(inode, offset); 4218 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1; 4219 map.m_next_pgofs = &next_pgofs; 4220 map.m_seg_type = f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode), 4221 inode->i_write_hint); 4222 if (flags & IOMAP_WRITE) 4223 map.m_may_create = true; 4224 4225 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO); 4226 if (err) 4227 return err; 4228 4229 iomap->offset = blks_to_bytes(inode, map.m_lblk); 4230 4231 /* 4232 * When inline encryption is enabled, sometimes I/O to an encrypted file 4233 * has to be broken up to guarantee DUN contiguity. Handle this by 4234 * limiting the length of the mapping returned. 4235 */ 4236 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len); 4237 4238 /* 4239 * We should never see delalloc or compressed extents here based on 4240 * prior flushing and checks. 4241 */ 4242 if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR)) 4243 return -EINVAL; 4244 4245 if (map.m_flags & F2FS_MAP_MAPPED) { 4246 if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR)) 4247 return -EINVAL; 4248 4249 iomap->length = blks_to_bytes(inode, map.m_len); 4250 iomap->type = IOMAP_MAPPED; 4251 iomap->flags |= IOMAP_F_MERGED; 4252 iomap->bdev = map.m_bdev; 4253 iomap->addr = blks_to_bytes(inode, map.m_pblk); 4254 } else { 4255 if (flags & IOMAP_WRITE) 4256 return -ENOTBLK; 4257 4258 if (map.m_pblk == NULL_ADDR) { 4259 iomap->length = blks_to_bytes(inode, next_pgofs) - 4260 iomap->offset; 4261 iomap->type = IOMAP_HOLE; 4262 } else if (map.m_pblk == NEW_ADDR) { 4263 iomap->length = blks_to_bytes(inode, map.m_len); 4264 iomap->type = IOMAP_UNWRITTEN; 4265 } else { 4266 f2fs_bug_on(F2FS_I_SB(inode), 1); 4267 } 4268 iomap->addr = IOMAP_NULL_ADDR; 4269 } 4270 4271 if (map.m_flags & F2FS_MAP_NEW) 4272 iomap->flags |= IOMAP_F_NEW; 4273 if ((inode->i_state & I_DIRTY_DATASYNC) || 4274 offset + length > i_size_read(inode)) 4275 iomap->flags |= IOMAP_F_DIRTY; 4276 4277 return 0; 4278 } 4279 4280 const struct iomap_ops f2fs_iomap_ops = { 4281 .iomap_begin = f2fs_iomap_begin, 4282 }; 4283