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