1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/data.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/buffer_head.h> 11 #include <linux/sched/mm.h> 12 #include <linux/mpage.h> 13 #include <linux/writeback.h> 14 #include <linux/pagevec.h> 15 #include <linux/blkdev.h> 16 #include <linux/bio.h> 17 #include <linux/blk-crypto.h> 18 #include <linux/swap.h> 19 #include <linux/prefetch.h> 20 #include <linux/uio.h> 21 #include <linux/sched/signal.h> 22 #include <linux/fiemap.h> 23 #include <linux/iomap.h> 24 25 #include "f2fs.h" 26 #include "node.h" 27 #include "segment.h" 28 #include "iostat.h" 29 #include <trace/events/f2fs.h> 30 31 #define NUM_PREALLOC_POST_READ_CTXS 128 32 33 static struct kmem_cache *bio_post_read_ctx_cache; 34 static struct kmem_cache *bio_entry_slab; 35 static mempool_t *bio_post_read_ctx_pool; 36 static struct bio_set f2fs_bioset; 37 38 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE 39 40 int __init f2fs_init_bioset(void) 41 { 42 return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE, 43 0, BIOSET_NEED_BVECS); 44 } 45 46 void f2fs_destroy_bioset(void) 47 { 48 bioset_exit(&f2fs_bioset); 49 } 50 51 static bool __is_cp_guaranteed(struct page *page) 52 { 53 struct address_space *mapping = page->mapping; 54 struct inode *inode; 55 struct f2fs_sb_info *sbi; 56 57 if (!mapping) 58 return false; 59 60 inode = mapping->host; 61 sbi = F2FS_I_SB(inode); 62 63 if (inode->i_ino == F2FS_META_INO(sbi) || 64 inode->i_ino == F2FS_NODE_INO(sbi) || 65 S_ISDIR(inode->i_mode)) 66 return true; 67 68 if (f2fs_is_compressed_page(page)) 69 return false; 70 if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) || 71 page_private_gcing(page)) 72 return true; 73 return false; 74 } 75 76 static enum count_type __read_io_type(struct page *page) 77 { 78 struct address_space *mapping = page_file_mapping(page); 79 80 if (mapping) { 81 struct inode *inode = mapping->host; 82 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 83 84 if (inode->i_ino == F2FS_META_INO(sbi)) 85 return F2FS_RD_META; 86 87 if (inode->i_ino == F2FS_NODE_INO(sbi)) 88 return F2FS_RD_NODE; 89 } 90 return F2FS_RD_DATA; 91 } 92 93 /* postprocessing steps for read bios */ 94 enum bio_post_read_step { 95 #ifdef CONFIG_FS_ENCRYPTION 96 STEP_DECRYPT = 1 << 0, 97 #else 98 STEP_DECRYPT = 0, /* compile out the decryption-related code */ 99 #endif 100 #ifdef CONFIG_F2FS_FS_COMPRESSION 101 STEP_DECOMPRESS = 1 << 1, 102 #else 103 STEP_DECOMPRESS = 0, /* compile out the decompression-related code */ 104 #endif 105 #ifdef CONFIG_FS_VERITY 106 STEP_VERITY = 1 << 2, 107 #else 108 STEP_VERITY = 0, /* compile out the verity-related code */ 109 #endif 110 }; 111 112 struct bio_post_read_ctx { 113 struct bio *bio; 114 struct f2fs_sb_info *sbi; 115 struct work_struct work; 116 unsigned int enabled_steps; 117 /* 118 * decompression_attempted keeps track of whether 119 * f2fs_end_read_compressed_page() has been called on the pages in the 120 * bio that belong to a compressed cluster yet. 121 */ 122 bool decompression_attempted; 123 block_t fs_blkaddr; 124 }; 125 126 /* 127 * Update and unlock a bio's pages, and free the bio. 128 * 129 * This marks pages up-to-date only if there was no error in the bio (I/O error, 130 * decryption error, or verity error), as indicated by bio->bi_status. 131 * 132 * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk) 133 * aren't marked up-to-date here, as decompression is done on a per-compression- 134 * cluster basis rather than a per-bio basis. Instead, we only must do two 135 * things for each compressed page here: call f2fs_end_read_compressed_page() 136 * with failed=true if an error occurred before it would have normally gotten 137 * called (i.e., I/O error or decryption error, but *not* verity error), and 138 * release the bio's reference to the decompress_io_ctx of the page's cluster. 139 */ 140 static void f2fs_finish_read_bio(struct bio *bio, bool in_task) 141 { 142 struct bio_vec *bv; 143 struct bvec_iter_all iter_all; 144 struct bio_post_read_ctx *ctx = bio->bi_private; 145 146 bio_for_each_segment_all(bv, bio, iter_all) { 147 struct page *page = bv->bv_page; 148 149 if (f2fs_is_compressed_page(page)) { 150 if (ctx && !ctx->decompression_attempted) 151 f2fs_end_read_compressed_page(page, true, 0, 152 in_task); 153 f2fs_put_page_dic(page, in_task); 154 continue; 155 } 156 157 if (bio->bi_status) 158 ClearPageUptodate(page); 159 else 160 SetPageUptodate(page); 161 dec_page_count(F2FS_P_SB(page), __read_io_type(page)); 162 unlock_page(page); 163 } 164 165 if (ctx) 166 mempool_free(ctx, bio_post_read_ctx_pool); 167 bio_put(bio); 168 } 169 170 static void f2fs_verify_bio(struct work_struct *work) 171 { 172 struct bio_post_read_ctx *ctx = 173 container_of(work, struct bio_post_read_ctx, work); 174 struct bio *bio = ctx->bio; 175 bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS); 176 177 /* 178 * fsverity_verify_bio() may call readahead() again, and while verity 179 * will be disabled for this, decryption and/or decompression may still 180 * be needed, resulting in another bio_post_read_ctx being allocated. 181 * So to prevent deadlocks we need to release the current ctx to the 182 * mempool first. This assumes that verity is the last post-read step. 183 */ 184 mempool_free(ctx, bio_post_read_ctx_pool); 185 bio->bi_private = NULL; 186 187 /* 188 * Verify the bio's pages with fs-verity. Exclude compressed pages, 189 * as those were handled separately by f2fs_end_read_compressed_page(). 190 */ 191 if (may_have_compressed_pages) { 192 struct bio_vec *bv; 193 struct bvec_iter_all iter_all; 194 195 bio_for_each_segment_all(bv, bio, iter_all) { 196 struct page *page = bv->bv_page; 197 198 if (!f2fs_is_compressed_page(page) && 199 !fsverity_verify_page(page)) { 200 bio->bi_status = BLK_STS_IOERR; 201 break; 202 } 203 } 204 } else { 205 fsverity_verify_bio(bio); 206 } 207 208 f2fs_finish_read_bio(bio, true); 209 } 210 211 /* 212 * If the bio's data needs to be verified with fs-verity, then enqueue the 213 * verity work for the bio. Otherwise finish the bio now. 214 * 215 * Note that to avoid deadlocks, the verity work can't be done on the 216 * decryption/decompression workqueue. This is because verifying the data pages 217 * can involve reading verity metadata pages from the file, and these verity 218 * metadata pages may be encrypted and/or compressed. 219 */ 220 static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task) 221 { 222 struct bio_post_read_ctx *ctx = bio->bi_private; 223 224 if (ctx && (ctx->enabled_steps & STEP_VERITY)) { 225 INIT_WORK(&ctx->work, f2fs_verify_bio); 226 fsverity_enqueue_verify_work(&ctx->work); 227 } else { 228 f2fs_finish_read_bio(bio, in_task); 229 } 230 } 231 232 /* 233 * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last 234 * remaining page was read by @ctx->bio. 235 * 236 * Note that a bio may span clusters (even a mix of compressed and uncompressed 237 * clusters) or be for just part of a cluster. STEP_DECOMPRESS just indicates 238 * that the bio includes at least one compressed page. The actual decompression 239 * is done on a per-cluster basis, not a per-bio basis. 240 */ 241 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx, 242 bool in_task) 243 { 244 struct bio_vec *bv; 245 struct bvec_iter_all iter_all; 246 bool all_compressed = true; 247 block_t blkaddr = ctx->fs_blkaddr; 248 249 bio_for_each_segment_all(bv, ctx->bio, iter_all) { 250 struct page *page = bv->bv_page; 251 252 if (f2fs_is_compressed_page(page)) 253 f2fs_end_read_compressed_page(page, false, blkaddr, 254 in_task); 255 else 256 all_compressed = false; 257 258 blkaddr++; 259 } 260 261 ctx->decompression_attempted = true; 262 263 /* 264 * Optimization: if all the bio's pages are compressed, then scheduling 265 * the per-bio verity work is unnecessary, as verity will be fully 266 * handled at the compression cluster level. 267 */ 268 if (all_compressed) 269 ctx->enabled_steps &= ~STEP_VERITY; 270 } 271 272 static void f2fs_post_read_work(struct work_struct *work) 273 { 274 struct bio_post_read_ctx *ctx = 275 container_of(work, struct bio_post_read_ctx, work); 276 struct bio *bio = ctx->bio; 277 278 if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) { 279 f2fs_finish_read_bio(bio, true); 280 return; 281 } 282 283 if (ctx->enabled_steps & STEP_DECOMPRESS) 284 f2fs_handle_step_decompress(ctx, true); 285 286 f2fs_verify_and_finish_bio(bio, true); 287 } 288 289 static void f2fs_read_end_io(struct bio *bio) 290 { 291 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio)); 292 struct bio_post_read_ctx *ctx; 293 bool intask = in_task(); 294 295 iostat_update_and_unbind_ctx(bio, 0); 296 ctx = bio->bi_private; 297 298 if (time_to_inject(sbi, FAULT_READ_IO)) { 299 f2fs_show_injection_info(sbi, FAULT_READ_IO); 300 bio->bi_status = BLK_STS_IOERR; 301 } 302 303 if (bio->bi_status) { 304 f2fs_finish_read_bio(bio, intask); 305 return; 306 } 307 308 if (ctx) { 309 unsigned int enabled_steps = ctx->enabled_steps & 310 (STEP_DECRYPT | STEP_DECOMPRESS); 311 312 /* 313 * If we have only decompression step between decompression and 314 * decrypt, we don't need post processing for this. 315 */ 316 if (enabled_steps == STEP_DECOMPRESS && 317 !f2fs_low_mem_mode(sbi)) { 318 f2fs_handle_step_decompress(ctx, intask); 319 } else if (enabled_steps) { 320 INIT_WORK(&ctx->work, f2fs_post_read_work); 321 queue_work(ctx->sbi->post_read_wq, &ctx->work); 322 return; 323 } 324 } 325 326 f2fs_verify_and_finish_bio(bio, intask); 327 } 328 329 static void f2fs_write_end_io(struct bio *bio) 330 { 331 struct f2fs_sb_info *sbi; 332 struct bio_vec *bvec; 333 struct bvec_iter_all iter_all; 334 335 iostat_update_and_unbind_ctx(bio, 1); 336 sbi = bio->bi_private; 337 338 if (time_to_inject(sbi, FAULT_WRITE_IO)) { 339 f2fs_show_injection_info(sbi, FAULT_WRITE_IO); 340 bio->bi_status = BLK_STS_IOERR; 341 } 342 343 bio_for_each_segment_all(bvec, bio, iter_all) { 344 struct page *page = bvec->bv_page; 345 enum count_type type = WB_DATA_TYPE(page); 346 347 if (page_private_dummy(page)) { 348 clear_page_private_dummy(page); 349 unlock_page(page); 350 mempool_free(page, sbi->write_io_dummy); 351 352 if (unlikely(bio->bi_status)) 353 f2fs_stop_checkpoint(sbi, true, 354 STOP_CP_REASON_WRITE_FAIL); 355 continue; 356 } 357 358 fscrypt_finalize_bounce_page(&page); 359 360 #ifdef CONFIG_F2FS_FS_COMPRESSION 361 if (f2fs_is_compressed_page(page)) { 362 f2fs_compress_write_end_io(bio, page); 363 continue; 364 } 365 #endif 366 367 if (unlikely(bio->bi_status)) { 368 mapping_set_error(page->mapping, -EIO); 369 if (type == F2FS_WB_CP_DATA) 370 f2fs_stop_checkpoint(sbi, true, 371 STOP_CP_REASON_WRITE_FAIL); 372 } 373 374 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 375 page->index != nid_of_node(page)); 376 377 dec_page_count(sbi, type); 378 if (f2fs_in_warm_node_list(sbi, page)) 379 f2fs_del_fsync_node_entry(sbi, page); 380 clear_page_private_gcing(page); 381 end_page_writeback(page); 382 } 383 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 384 wq_has_sleeper(&sbi->cp_wait)) 385 wake_up(&sbi->cp_wait); 386 387 bio_put(bio); 388 } 389 390 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 391 block_t blk_addr, sector_t *sector) 392 { 393 struct block_device *bdev = sbi->sb->s_bdev; 394 int i; 395 396 if (f2fs_is_multi_device(sbi)) { 397 for (i = 0; i < sbi->s_ndevs; i++) { 398 if (FDEV(i).start_blk <= blk_addr && 399 FDEV(i).end_blk >= blk_addr) { 400 blk_addr -= FDEV(i).start_blk; 401 bdev = FDEV(i).bdev; 402 break; 403 } 404 } 405 } 406 407 if (sector) 408 *sector = SECTOR_FROM_BLOCK(blk_addr); 409 return bdev; 410 } 411 412 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 413 { 414 int i; 415 416 if (!f2fs_is_multi_device(sbi)) 417 return 0; 418 419 for (i = 0; i < sbi->s_ndevs; i++) 420 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 421 return i; 422 return 0; 423 } 424 425 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio) 426 { 427 unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1; 428 unsigned int fua_flag, meta_flag, io_flag; 429 blk_opf_t op_flags = 0; 430 431 if (fio->op != REQ_OP_WRITE) 432 return 0; 433 if (fio->type == DATA) 434 io_flag = fio->sbi->data_io_flag; 435 else if (fio->type == NODE) 436 io_flag = fio->sbi->node_io_flag; 437 else 438 return 0; 439 440 fua_flag = io_flag & temp_mask; 441 meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask; 442 443 /* 444 * data/node io flag bits per temp: 445 * REQ_META | REQ_FUA | 446 * 5 | 4 | 3 | 2 | 1 | 0 | 447 * Cold | Warm | Hot | Cold | Warm | Hot | 448 */ 449 if ((1 << fio->temp) & meta_flag) 450 op_flags |= REQ_META; 451 if ((1 << fio->temp) & fua_flag) 452 op_flags |= REQ_FUA; 453 return op_flags; 454 } 455 456 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) 457 { 458 struct f2fs_sb_info *sbi = fio->sbi; 459 struct block_device *bdev; 460 sector_t sector; 461 struct bio *bio; 462 463 bdev = f2fs_target_device(sbi, fio->new_blkaddr, §or); 464 bio = bio_alloc_bioset(bdev, npages, 465 fio->op | fio->op_flags | f2fs_io_flags(fio), 466 GFP_NOIO, &f2fs_bioset); 467 bio->bi_iter.bi_sector = sector; 468 if (is_read_io(fio->op)) { 469 bio->bi_end_io = f2fs_read_end_io; 470 bio->bi_private = NULL; 471 } else { 472 bio->bi_end_io = f2fs_write_end_io; 473 bio->bi_private = sbi; 474 } 475 iostat_alloc_and_bind_ctx(sbi, bio, NULL); 476 477 if (fio->io_wbc) 478 wbc_init_bio(fio->io_wbc, bio); 479 480 return bio; 481 } 482 483 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode, 484 pgoff_t first_idx, 485 const struct f2fs_io_info *fio, 486 gfp_t gfp_mask) 487 { 488 /* 489 * The f2fs garbage collector sets ->encrypted_page when it wants to 490 * read/write raw data without encryption. 491 */ 492 if (!fio || !fio->encrypted_page) 493 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask); 494 } 495 496 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode, 497 pgoff_t next_idx, 498 const struct f2fs_io_info *fio) 499 { 500 /* 501 * The f2fs garbage collector sets ->encrypted_page when it wants to 502 * read/write raw data without encryption. 503 */ 504 if (fio && fio->encrypted_page) 505 return !bio_has_crypt_ctx(bio); 506 507 return fscrypt_mergeable_bio(bio, inode, next_idx); 508 } 509 510 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio, 511 enum page_type type) 512 { 513 WARN_ON_ONCE(!is_read_io(bio_op(bio))); 514 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 515 516 iostat_update_submit_ctx(bio, type); 517 submit_bio(bio); 518 } 519 520 static void f2fs_align_write_bio(struct f2fs_sb_info *sbi, struct bio *bio) 521 { 522 unsigned int start = 523 (bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS) % F2FS_IO_SIZE(sbi); 524 525 if (start == 0) 526 return; 527 528 /* fill dummy pages */ 529 for (; start < F2FS_IO_SIZE(sbi); start++) { 530 struct page *page = 531 mempool_alloc(sbi->write_io_dummy, 532 GFP_NOIO | __GFP_NOFAIL); 533 f2fs_bug_on(sbi, !page); 534 535 lock_page(page); 536 537 zero_user_segment(page, 0, PAGE_SIZE); 538 set_page_private_dummy(page); 539 540 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 541 f2fs_bug_on(sbi, 1); 542 } 543 } 544 545 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio, 546 enum page_type type) 547 { 548 WARN_ON_ONCE(is_read_io(bio_op(bio))); 549 550 if (type == DATA || type == NODE) { 551 if (f2fs_lfs_mode(sbi) && current->plug) 552 blk_finish_plug(current->plug); 553 554 if (F2FS_IO_ALIGNED(sbi)) { 555 f2fs_align_write_bio(sbi, bio); 556 /* 557 * In the NODE case, we lose next block address chain. 558 * So, we need to do checkpoint in f2fs_sync_file. 559 */ 560 if (type == NODE) 561 set_sbi_flag(sbi, SBI_NEED_CP); 562 } 563 } 564 565 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 566 iostat_update_submit_ctx(bio, type); 567 submit_bio(bio); 568 } 569 570 static void __submit_merged_bio(struct f2fs_bio_info *io) 571 { 572 struct f2fs_io_info *fio = &io->fio; 573 574 if (!io->bio) 575 return; 576 577 if (is_read_io(fio->op)) { 578 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 579 f2fs_submit_read_bio(io->sbi, io->bio, fio->type); 580 } else { 581 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 582 f2fs_submit_write_bio(io->sbi, io->bio, fio->type); 583 } 584 io->bio = NULL; 585 } 586 587 static bool __has_merged_page(struct bio *bio, struct inode *inode, 588 struct page *page, nid_t ino) 589 { 590 struct bio_vec *bvec; 591 struct bvec_iter_all iter_all; 592 593 if (!bio) 594 return false; 595 596 if (!inode && !page && !ino) 597 return true; 598 599 bio_for_each_segment_all(bvec, bio, iter_all) { 600 struct page *target = bvec->bv_page; 601 602 if (fscrypt_is_bounce_page(target)) { 603 target = fscrypt_pagecache_page(target); 604 if (IS_ERR(target)) 605 continue; 606 } 607 if (f2fs_is_compressed_page(target)) { 608 target = f2fs_compress_control_page(target); 609 if (IS_ERR(target)) 610 continue; 611 } 612 613 if (inode && inode == target->mapping->host) 614 return true; 615 if (page && page == target) 616 return true; 617 if (ino && ino == ino_of_node(target)) 618 return true; 619 } 620 621 return false; 622 } 623 624 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi) 625 { 626 int i; 627 628 for (i = 0; i < NR_PAGE_TYPE; i++) { 629 int n = (i == META) ? 1 : NR_TEMP_TYPE; 630 int j; 631 632 sbi->write_io[i] = f2fs_kmalloc(sbi, 633 array_size(n, sizeof(struct f2fs_bio_info)), 634 GFP_KERNEL); 635 if (!sbi->write_io[i]) 636 return -ENOMEM; 637 638 for (j = HOT; j < n; j++) { 639 init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem); 640 sbi->write_io[i][j].sbi = sbi; 641 sbi->write_io[i][j].bio = NULL; 642 spin_lock_init(&sbi->write_io[i][j].io_lock); 643 INIT_LIST_HEAD(&sbi->write_io[i][j].io_list); 644 INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list); 645 init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock); 646 } 647 } 648 649 return 0; 650 } 651 652 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 653 enum page_type type, enum temp_type temp) 654 { 655 enum page_type btype = PAGE_TYPE_OF_BIO(type); 656 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 657 658 f2fs_down_write(&io->io_rwsem); 659 660 /* change META to META_FLUSH in the checkpoint procedure */ 661 if (type >= META_FLUSH) { 662 io->fio.type = META_FLUSH; 663 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC; 664 if (!test_opt(sbi, NOBARRIER)) 665 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA; 666 } 667 __submit_merged_bio(io); 668 f2fs_up_write(&io->io_rwsem); 669 } 670 671 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 672 struct inode *inode, struct page *page, 673 nid_t ino, enum page_type type, bool force) 674 { 675 enum temp_type temp; 676 bool ret = true; 677 678 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 679 if (!force) { 680 enum page_type btype = PAGE_TYPE_OF_BIO(type); 681 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 682 683 f2fs_down_read(&io->io_rwsem); 684 ret = __has_merged_page(io->bio, inode, page, ino); 685 f2fs_up_read(&io->io_rwsem); 686 } 687 if (ret) 688 __f2fs_submit_merged_write(sbi, type, temp); 689 690 /* TODO: use HOT temp only for meta pages now. */ 691 if (type >= META) 692 break; 693 } 694 } 695 696 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 697 { 698 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true); 699 } 700 701 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 702 struct inode *inode, struct page *page, 703 nid_t ino, enum page_type type) 704 { 705 __submit_merged_write_cond(sbi, inode, page, ino, type, false); 706 } 707 708 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 709 { 710 f2fs_submit_merged_write(sbi, DATA); 711 f2fs_submit_merged_write(sbi, NODE); 712 f2fs_submit_merged_write(sbi, META); 713 } 714 715 /* 716 * Fill the locked page with data located in the block address. 717 * A caller needs to unlock the page on failure. 718 */ 719 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 720 { 721 struct bio *bio; 722 struct page *page = fio->encrypted_page ? 723 fio->encrypted_page : fio->page; 724 725 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 726 fio->is_por ? META_POR : (__is_meta_io(fio) ? 727 META_GENERIC : DATA_GENERIC_ENHANCE))) { 728 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR); 729 return -EFSCORRUPTED; 730 } 731 732 trace_f2fs_submit_page_bio(page, fio); 733 734 /* Allocate a new bio */ 735 bio = __bio_alloc(fio, 1); 736 737 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host, 738 fio->page->index, fio, GFP_NOIO); 739 740 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 741 bio_put(bio); 742 return -EFAULT; 743 } 744 745 if (fio->io_wbc && !is_read_io(fio->op)) 746 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 747 748 inc_page_count(fio->sbi, is_read_io(fio->op) ? 749 __read_io_type(page) : WB_DATA_TYPE(fio->page)); 750 751 if (is_read_io(bio_op(bio))) 752 f2fs_submit_read_bio(fio->sbi, bio, fio->type); 753 else 754 f2fs_submit_write_bio(fio->sbi, bio, fio->type); 755 return 0; 756 } 757 758 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 759 block_t last_blkaddr, block_t cur_blkaddr) 760 { 761 if (unlikely(sbi->max_io_bytes && 762 bio->bi_iter.bi_size >= sbi->max_io_bytes)) 763 return false; 764 if (last_blkaddr + 1 != cur_blkaddr) 765 return false; 766 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL); 767 } 768 769 static bool io_type_is_mergeable(struct f2fs_bio_info *io, 770 struct f2fs_io_info *fio) 771 { 772 if (io->fio.op != fio->op) 773 return false; 774 return io->fio.op_flags == fio->op_flags; 775 } 776 777 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, 778 struct f2fs_bio_info *io, 779 struct f2fs_io_info *fio, 780 block_t last_blkaddr, 781 block_t cur_blkaddr) 782 { 783 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) { 784 unsigned int filled_blocks = 785 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size); 786 unsigned int io_size = F2FS_IO_SIZE(sbi); 787 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt; 788 789 /* IOs in bio is aligned and left space of vectors is not enough */ 790 if (!(filled_blocks % io_size) && left_vecs < io_size) 791 return false; 792 } 793 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr)) 794 return false; 795 return io_type_is_mergeable(io, fio); 796 } 797 798 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio, 799 struct page *page, enum temp_type temp) 800 { 801 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 802 struct bio_entry *be; 803 804 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL); 805 be->bio = bio; 806 bio_get(bio); 807 808 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE) 809 f2fs_bug_on(sbi, 1); 810 811 f2fs_down_write(&io->bio_list_lock); 812 list_add_tail(&be->list, &io->bio_list); 813 f2fs_up_write(&io->bio_list_lock); 814 } 815 816 static void del_bio_entry(struct bio_entry *be) 817 { 818 list_del(&be->list); 819 kmem_cache_free(bio_entry_slab, be); 820 } 821 822 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio, 823 struct page *page) 824 { 825 struct f2fs_sb_info *sbi = fio->sbi; 826 enum temp_type temp; 827 bool found = false; 828 int ret = -EAGAIN; 829 830 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 831 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 832 struct list_head *head = &io->bio_list; 833 struct bio_entry *be; 834 835 f2fs_down_write(&io->bio_list_lock); 836 list_for_each_entry(be, head, list) { 837 if (be->bio != *bio) 838 continue; 839 840 found = true; 841 842 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio, 843 *fio->last_block, 844 fio->new_blkaddr)); 845 if (f2fs_crypt_mergeable_bio(*bio, 846 fio->page->mapping->host, 847 fio->page->index, fio) && 848 bio_add_page(*bio, page, PAGE_SIZE, 0) == 849 PAGE_SIZE) { 850 ret = 0; 851 break; 852 } 853 854 /* page can't be merged into bio; submit the bio */ 855 del_bio_entry(be); 856 f2fs_submit_write_bio(sbi, *bio, DATA); 857 break; 858 } 859 f2fs_up_write(&io->bio_list_lock); 860 } 861 862 if (ret) { 863 bio_put(*bio); 864 *bio = NULL; 865 } 866 867 return ret; 868 } 869 870 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi, 871 struct bio **bio, struct page *page) 872 { 873 enum temp_type temp; 874 bool found = false; 875 struct bio *target = bio ? *bio : NULL; 876 877 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { 878 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; 879 struct list_head *head = &io->bio_list; 880 struct bio_entry *be; 881 882 if (list_empty(head)) 883 continue; 884 885 f2fs_down_read(&io->bio_list_lock); 886 list_for_each_entry(be, head, list) { 887 if (target) 888 found = (target == be->bio); 889 else 890 found = __has_merged_page(be->bio, NULL, 891 page, 0); 892 if (found) 893 break; 894 } 895 f2fs_up_read(&io->bio_list_lock); 896 897 if (!found) 898 continue; 899 900 found = false; 901 902 f2fs_down_write(&io->bio_list_lock); 903 list_for_each_entry(be, head, list) { 904 if (target) 905 found = (target == be->bio); 906 else 907 found = __has_merged_page(be->bio, NULL, 908 page, 0); 909 if (found) { 910 target = be->bio; 911 del_bio_entry(be); 912 break; 913 } 914 } 915 f2fs_up_write(&io->bio_list_lock); 916 } 917 918 if (found) 919 f2fs_submit_write_bio(sbi, target, DATA); 920 if (bio && *bio) { 921 bio_put(*bio); 922 *bio = NULL; 923 } 924 } 925 926 int f2fs_merge_page_bio(struct f2fs_io_info *fio) 927 { 928 struct bio *bio = *fio->bio; 929 struct page *page = fio->encrypted_page ? 930 fio->encrypted_page : fio->page; 931 932 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr, 933 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) { 934 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR); 935 return -EFSCORRUPTED; 936 } 937 938 trace_f2fs_submit_page_bio(page, fio); 939 940 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block, 941 fio->new_blkaddr)) 942 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL); 943 alloc_new: 944 if (!bio) { 945 bio = __bio_alloc(fio, BIO_MAX_VECS); 946 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host, 947 fio->page->index, fio, GFP_NOIO); 948 949 add_bio_entry(fio->sbi, bio, page, fio->temp); 950 } else { 951 if (add_ipu_page(fio, &bio, page)) 952 goto alloc_new; 953 } 954 955 if (fio->io_wbc) 956 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE); 957 958 inc_page_count(fio->sbi, WB_DATA_TYPE(page)); 959 960 *fio->last_block = fio->new_blkaddr; 961 *fio->bio = bio; 962 963 return 0; 964 } 965 966 void f2fs_submit_page_write(struct f2fs_io_info *fio) 967 { 968 struct f2fs_sb_info *sbi = fio->sbi; 969 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 970 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 971 struct page *bio_page; 972 973 f2fs_bug_on(sbi, is_read_io(fio->op)); 974 975 f2fs_down_write(&io->io_rwsem); 976 next: 977 if (fio->in_list) { 978 spin_lock(&io->io_lock); 979 if (list_empty(&io->io_list)) { 980 spin_unlock(&io->io_lock); 981 goto out; 982 } 983 fio = list_first_entry(&io->io_list, 984 struct f2fs_io_info, list); 985 list_del(&fio->list); 986 spin_unlock(&io->io_lock); 987 } 988 989 verify_fio_blkaddr(fio); 990 991 if (fio->encrypted_page) 992 bio_page = fio->encrypted_page; 993 else if (fio->compressed_page) 994 bio_page = fio->compressed_page; 995 else 996 bio_page = fio->page; 997 998 /* set submitted = true as a return value */ 999 fio->submitted = true; 1000 1001 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 1002 1003 if (io->bio && 1004 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio, 1005 fio->new_blkaddr) || 1006 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host, 1007 bio_page->index, fio))) 1008 __submit_merged_bio(io); 1009 alloc_new: 1010 if (io->bio == NULL) { 1011 if (F2FS_IO_ALIGNED(sbi) && 1012 (fio->type == DATA || fio->type == NODE) && 1013 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 1014 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 1015 fio->retry = true; 1016 goto skip; 1017 } 1018 io->bio = __bio_alloc(fio, BIO_MAX_VECS); 1019 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host, 1020 bio_page->index, fio, GFP_NOIO); 1021 io->fio = *fio; 1022 } 1023 1024 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 1025 __submit_merged_bio(io); 1026 goto alloc_new; 1027 } 1028 1029 if (fio->io_wbc) 1030 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE); 1031 1032 io->last_block_in_bio = fio->new_blkaddr; 1033 1034 trace_f2fs_submit_page_write(fio->page, fio); 1035 skip: 1036 if (fio->in_list) 1037 goto next; 1038 out: 1039 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 1040 !f2fs_is_checkpoint_ready(sbi)) 1041 __submit_merged_bio(io); 1042 f2fs_up_write(&io->io_rwsem); 1043 } 1044 1045 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 1046 unsigned nr_pages, blk_opf_t op_flag, 1047 pgoff_t first_idx, bool for_write) 1048 { 1049 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1050 struct bio *bio; 1051 struct bio_post_read_ctx *ctx = NULL; 1052 unsigned int post_read_steps = 0; 1053 sector_t sector; 1054 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or); 1055 1056 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages), 1057 REQ_OP_READ | op_flag, 1058 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset); 1059 if (!bio) 1060 return ERR_PTR(-ENOMEM); 1061 bio->bi_iter.bi_sector = sector; 1062 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS); 1063 bio->bi_end_io = f2fs_read_end_io; 1064 1065 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 1066 post_read_steps |= STEP_DECRYPT; 1067 1068 if (f2fs_need_verity(inode, first_idx)) 1069 post_read_steps |= STEP_VERITY; 1070 1071 /* 1072 * STEP_DECOMPRESS is handled specially, since a compressed file might 1073 * contain both compressed and uncompressed clusters. We'll allocate a 1074 * bio_post_read_ctx if the file is compressed, but the caller is 1075 * responsible for enabling STEP_DECOMPRESS if it's actually needed. 1076 */ 1077 1078 if (post_read_steps || f2fs_compressed_file(inode)) { 1079 /* Due to the mempool, this never fails. */ 1080 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 1081 ctx->bio = bio; 1082 ctx->sbi = sbi; 1083 ctx->enabled_steps = post_read_steps; 1084 ctx->fs_blkaddr = blkaddr; 1085 ctx->decompression_attempted = false; 1086 bio->bi_private = ctx; 1087 } 1088 iostat_alloc_and_bind_ctx(sbi, bio, ctx); 1089 1090 return bio; 1091 } 1092 1093 /* This can handle encryption stuffs */ 1094 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 1095 block_t blkaddr, blk_opf_t op_flags, 1096 bool for_write) 1097 { 1098 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1099 struct bio *bio; 1100 1101 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags, 1102 page->index, for_write); 1103 if (IS_ERR(bio)) 1104 return PTR_ERR(bio); 1105 1106 /* wait for GCed page writeback via META_MAPPING */ 1107 f2fs_wait_on_block_writeback(inode, blkaddr); 1108 1109 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 1110 bio_put(bio); 1111 return -EFAULT; 1112 } 1113 inc_page_count(sbi, F2FS_RD_DATA); 1114 f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE); 1115 f2fs_submit_read_bio(sbi, bio, DATA); 1116 return 0; 1117 } 1118 1119 static void __set_data_blkaddr(struct dnode_of_data *dn) 1120 { 1121 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 1122 __le32 *addr_array; 1123 int base = 0; 1124 1125 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 1126 base = get_extra_isize(dn->inode); 1127 1128 /* Get physical address of data block */ 1129 addr_array = blkaddr_in_node(rn); 1130 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 1131 } 1132 1133 /* 1134 * Lock ordering for the change of data block address: 1135 * ->data_page 1136 * ->node_page 1137 * update block addresses in the node page 1138 */ 1139 void f2fs_set_data_blkaddr(struct dnode_of_data *dn) 1140 { 1141 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1142 __set_data_blkaddr(dn); 1143 if (set_page_dirty(dn->node_page)) 1144 dn->node_changed = true; 1145 } 1146 1147 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 1148 { 1149 dn->data_blkaddr = blkaddr; 1150 f2fs_set_data_blkaddr(dn); 1151 f2fs_update_read_extent_cache(dn); 1152 } 1153 1154 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 1155 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 1156 { 1157 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1158 int err; 1159 1160 if (!count) 1161 return 0; 1162 1163 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1164 return -EPERM; 1165 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1166 return err; 1167 1168 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 1169 dn->ofs_in_node, count); 1170 1171 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true); 1172 1173 for (; count > 0; dn->ofs_in_node++) { 1174 block_t blkaddr = f2fs_data_blkaddr(dn); 1175 1176 if (blkaddr == NULL_ADDR) { 1177 dn->data_blkaddr = NEW_ADDR; 1178 __set_data_blkaddr(dn); 1179 count--; 1180 } 1181 } 1182 1183 if (set_page_dirty(dn->node_page)) 1184 dn->node_changed = true; 1185 return 0; 1186 } 1187 1188 /* Should keep dn->ofs_in_node unchanged */ 1189 int f2fs_reserve_new_block(struct dnode_of_data *dn) 1190 { 1191 unsigned int ofs_in_node = dn->ofs_in_node; 1192 int ret; 1193 1194 ret = f2fs_reserve_new_blocks(dn, 1); 1195 dn->ofs_in_node = ofs_in_node; 1196 return ret; 1197 } 1198 1199 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 1200 { 1201 bool need_put = dn->inode_page ? false : true; 1202 int err; 1203 1204 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE); 1205 if (err) 1206 return err; 1207 1208 if (dn->data_blkaddr == NULL_ADDR) 1209 err = f2fs_reserve_new_block(dn); 1210 if (err || need_put) 1211 f2fs_put_dnode(dn); 1212 return err; 1213 } 1214 1215 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 1216 { 1217 struct extent_info ei = {0, }; 1218 struct inode *inode = dn->inode; 1219 1220 if (f2fs_lookup_read_extent_cache(inode, index, &ei)) { 1221 dn->data_blkaddr = ei.blk + index - ei.fofs; 1222 return 0; 1223 } 1224 1225 return f2fs_reserve_block(dn, index); 1226 } 1227 1228 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, 1229 blk_opf_t op_flags, bool for_write, 1230 pgoff_t *next_pgofs) 1231 { 1232 struct address_space *mapping = inode->i_mapping; 1233 struct dnode_of_data dn; 1234 struct page *page; 1235 struct extent_info ei = {0, }; 1236 int err; 1237 1238 page = f2fs_grab_cache_page(mapping, index, for_write); 1239 if (!page) 1240 return ERR_PTR(-ENOMEM); 1241 1242 if (f2fs_lookup_read_extent_cache(inode, index, &ei)) { 1243 dn.data_blkaddr = ei.blk + index - ei.fofs; 1244 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr, 1245 DATA_GENERIC_ENHANCE_READ)) { 1246 err = -EFSCORRUPTED; 1247 f2fs_handle_error(F2FS_I_SB(inode), 1248 ERROR_INVALID_BLKADDR); 1249 goto put_err; 1250 } 1251 goto got_it; 1252 } 1253 1254 set_new_dnode(&dn, inode, NULL, NULL, 0); 1255 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1256 if (err) { 1257 if (err == -ENOENT && next_pgofs) 1258 *next_pgofs = f2fs_get_next_page_offset(&dn, index); 1259 goto put_err; 1260 } 1261 f2fs_put_dnode(&dn); 1262 1263 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1264 err = -ENOENT; 1265 if (next_pgofs) 1266 *next_pgofs = index + 1; 1267 goto put_err; 1268 } 1269 if (dn.data_blkaddr != NEW_ADDR && 1270 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode), 1271 dn.data_blkaddr, 1272 DATA_GENERIC_ENHANCE)) { 1273 err = -EFSCORRUPTED; 1274 f2fs_handle_error(F2FS_I_SB(inode), 1275 ERROR_INVALID_BLKADDR); 1276 goto put_err; 1277 } 1278 got_it: 1279 if (PageUptodate(page)) { 1280 unlock_page(page); 1281 return page; 1282 } 1283 1284 /* 1285 * A new dentry page is allocated but not able to be written, since its 1286 * new inode page couldn't be allocated due to -ENOSPC. 1287 * In such the case, its blkaddr can be remained as NEW_ADDR. 1288 * see, f2fs_add_link -> f2fs_get_new_data_page -> 1289 * f2fs_init_inode_metadata. 1290 */ 1291 if (dn.data_blkaddr == NEW_ADDR) { 1292 zero_user_segment(page, 0, PAGE_SIZE); 1293 if (!PageUptodate(page)) 1294 SetPageUptodate(page); 1295 unlock_page(page); 1296 return page; 1297 } 1298 1299 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr, 1300 op_flags, for_write); 1301 if (err) 1302 goto put_err; 1303 return page; 1304 1305 put_err: 1306 f2fs_put_page(page, 1); 1307 return ERR_PTR(err); 1308 } 1309 1310 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index, 1311 pgoff_t *next_pgofs) 1312 { 1313 struct address_space *mapping = inode->i_mapping; 1314 struct page *page; 1315 1316 page = find_get_page(mapping, index); 1317 if (page && PageUptodate(page)) 1318 return page; 1319 f2fs_put_page(page, 0); 1320 1321 page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs); 1322 if (IS_ERR(page)) 1323 return page; 1324 1325 if (PageUptodate(page)) 1326 return page; 1327 1328 wait_on_page_locked(page); 1329 if (unlikely(!PageUptodate(page))) { 1330 f2fs_put_page(page, 0); 1331 return ERR_PTR(-EIO); 1332 } 1333 return page; 1334 } 1335 1336 /* 1337 * If it tries to access a hole, return an error. 1338 * Because, the callers, functions in dir.c and GC, should be able to know 1339 * whether this page exists or not. 1340 */ 1341 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, 1342 bool for_write) 1343 { 1344 struct address_space *mapping = inode->i_mapping; 1345 struct page *page; 1346 repeat: 1347 page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL); 1348 if (IS_ERR(page)) 1349 return page; 1350 1351 /* wait for read completion */ 1352 lock_page(page); 1353 if (unlikely(page->mapping != mapping)) { 1354 f2fs_put_page(page, 1); 1355 goto repeat; 1356 } 1357 if (unlikely(!PageUptodate(page))) { 1358 f2fs_put_page(page, 1); 1359 return ERR_PTR(-EIO); 1360 } 1361 return page; 1362 } 1363 1364 /* 1365 * Caller ensures that this data page is never allocated. 1366 * A new zero-filled data page is allocated in the page cache. 1367 * 1368 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 1369 * f2fs_unlock_op(). 1370 * Note that, ipage is set only by make_empty_dir, and if any error occur, 1371 * ipage should be released by this function. 1372 */ 1373 struct page *f2fs_get_new_data_page(struct inode *inode, 1374 struct page *ipage, pgoff_t index, bool new_i_size) 1375 { 1376 struct address_space *mapping = inode->i_mapping; 1377 struct page *page; 1378 struct dnode_of_data dn; 1379 int err; 1380 1381 page = f2fs_grab_cache_page(mapping, index, true); 1382 if (!page) { 1383 /* 1384 * before exiting, we should make sure ipage will be released 1385 * if any error occur. 1386 */ 1387 f2fs_put_page(ipage, 1); 1388 return ERR_PTR(-ENOMEM); 1389 } 1390 1391 set_new_dnode(&dn, inode, ipage, NULL, 0); 1392 err = f2fs_reserve_block(&dn, index); 1393 if (err) { 1394 f2fs_put_page(page, 1); 1395 return ERR_PTR(err); 1396 } 1397 if (!ipage) 1398 f2fs_put_dnode(&dn); 1399 1400 if (PageUptodate(page)) 1401 goto got_it; 1402 1403 if (dn.data_blkaddr == NEW_ADDR) { 1404 zero_user_segment(page, 0, PAGE_SIZE); 1405 if (!PageUptodate(page)) 1406 SetPageUptodate(page); 1407 } else { 1408 f2fs_put_page(page, 1); 1409 1410 /* if ipage exists, blkaddr should be NEW_ADDR */ 1411 f2fs_bug_on(F2FS_I_SB(inode), ipage); 1412 page = f2fs_get_lock_data_page(inode, index, true); 1413 if (IS_ERR(page)) 1414 return page; 1415 } 1416 got_it: 1417 if (new_i_size && i_size_read(inode) < 1418 ((loff_t)(index + 1) << PAGE_SHIFT)) 1419 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 1420 return page; 1421 } 1422 1423 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 1424 { 1425 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1426 struct f2fs_summary sum; 1427 struct node_info ni; 1428 block_t old_blkaddr; 1429 blkcnt_t count = 1; 1430 int err; 1431 1432 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 1433 return -EPERM; 1434 1435 err = f2fs_get_node_info(sbi, dn->nid, &ni, false); 1436 if (err) 1437 return err; 1438 1439 dn->data_blkaddr = f2fs_data_blkaddr(dn); 1440 if (dn->data_blkaddr != NULL_ADDR) 1441 goto alloc; 1442 1443 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 1444 return err; 1445 1446 alloc: 1447 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1448 old_blkaddr = dn->data_blkaddr; 1449 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr, 1450 &sum, seg_type, NULL); 1451 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 1452 invalidate_mapping_pages(META_MAPPING(sbi), 1453 old_blkaddr, old_blkaddr); 1454 f2fs_invalidate_compress_page(sbi, old_blkaddr); 1455 } 1456 f2fs_update_data_blkaddr(dn, dn->data_blkaddr); 1457 return 0; 1458 } 1459 1460 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 1461 { 1462 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1463 if (lock) 1464 f2fs_down_read(&sbi->node_change); 1465 else 1466 f2fs_up_read(&sbi->node_change); 1467 } else { 1468 if (lock) 1469 f2fs_lock_op(sbi); 1470 else 1471 f2fs_unlock_op(sbi); 1472 } 1473 } 1474 1475 /* 1476 * f2fs_map_blocks() tries to find or build mapping relationship which 1477 * maps continuous logical blocks to physical blocks, and return such 1478 * info via f2fs_map_blocks structure. 1479 */ 1480 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 1481 int create, int flag) 1482 { 1483 unsigned int maxblocks = map->m_len; 1484 struct dnode_of_data dn; 1485 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1486 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE; 1487 pgoff_t pgofs, end_offset, end; 1488 int err = 0, ofs = 1; 1489 unsigned int ofs_in_node, last_ofs_in_node; 1490 blkcnt_t prealloc; 1491 struct extent_info ei = {0, }; 1492 block_t blkaddr; 1493 unsigned int start_pgofs; 1494 int bidx = 0; 1495 1496 if (!maxblocks) 1497 return 0; 1498 1499 map->m_bdev = inode->i_sb->s_bdev; 1500 map->m_multidev_dio = 1501 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag); 1502 1503 map->m_len = 0; 1504 map->m_flags = 0; 1505 1506 /* it only supports block size == page size */ 1507 pgofs = (pgoff_t)map->m_lblk; 1508 end = pgofs + maxblocks; 1509 1510 if (!create && f2fs_lookup_read_extent_cache(inode, pgofs, &ei)) { 1511 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1512 map->m_may_create) 1513 goto next_dnode; 1514 1515 map->m_pblk = ei.blk + pgofs - ei.fofs; 1516 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 1517 map->m_flags = F2FS_MAP_MAPPED; 1518 if (map->m_next_extent) 1519 *map->m_next_extent = pgofs + map->m_len; 1520 1521 /* for hardware encryption, but to avoid potential issue in future */ 1522 if (flag == F2FS_GET_BLOCK_DIO) 1523 f2fs_wait_on_block_writeback_range(inode, 1524 map->m_pblk, map->m_len); 1525 1526 if (map->m_multidev_dio) { 1527 block_t blk_addr = map->m_pblk; 1528 1529 bidx = f2fs_target_device_index(sbi, map->m_pblk); 1530 1531 map->m_bdev = FDEV(bidx).bdev; 1532 map->m_pblk -= FDEV(bidx).start_blk; 1533 map->m_len = min(map->m_len, 1534 FDEV(bidx).end_blk + 1 - map->m_pblk); 1535 1536 if (map->m_may_create) 1537 f2fs_update_device_state(sbi, inode->i_ino, 1538 blk_addr, map->m_len); 1539 } 1540 goto out; 1541 } 1542 1543 next_dnode: 1544 if (map->m_may_create) 1545 f2fs_do_map_lock(sbi, flag, true); 1546 1547 /* When reading holes, we need its node page */ 1548 set_new_dnode(&dn, inode, NULL, NULL, 0); 1549 err = f2fs_get_dnode_of_data(&dn, pgofs, mode); 1550 if (err) { 1551 if (flag == F2FS_GET_BLOCK_BMAP) 1552 map->m_pblk = 0; 1553 1554 if (err == -ENOENT) { 1555 /* 1556 * There is one exceptional case that read_node_page() 1557 * may return -ENOENT due to filesystem has been 1558 * shutdown or cp_error, so force to convert error 1559 * number to EIO for such case. 1560 */ 1561 if (map->m_may_create && 1562 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || 1563 f2fs_cp_error(sbi))) { 1564 err = -EIO; 1565 goto unlock_out; 1566 } 1567 1568 err = 0; 1569 if (map->m_next_pgofs) 1570 *map->m_next_pgofs = 1571 f2fs_get_next_page_offset(&dn, pgofs); 1572 if (map->m_next_extent) 1573 *map->m_next_extent = 1574 f2fs_get_next_page_offset(&dn, pgofs); 1575 } 1576 goto unlock_out; 1577 } 1578 1579 start_pgofs = pgofs; 1580 prealloc = 0; 1581 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 1582 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 1583 1584 next_block: 1585 blkaddr = f2fs_data_blkaddr(&dn); 1586 1587 if (__is_valid_data_blkaddr(blkaddr) && 1588 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) { 1589 err = -EFSCORRUPTED; 1590 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1591 goto sync_out; 1592 } 1593 1594 if (__is_valid_data_blkaddr(blkaddr)) { 1595 /* use out-place-update for driect IO under LFS mode */ 1596 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO && 1597 map->m_may_create) { 1598 err = __allocate_data_block(&dn, map->m_seg_type); 1599 if (err) 1600 goto sync_out; 1601 blkaddr = dn.data_blkaddr; 1602 set_inode_flag(inode, FI_APPEND_WRITE); 1603 } 1604 } else { 1605 if (create) { 1606 if (unlikely(f2fs_cp_error(sbi))) { 1607 err = -EIO; 1608 goto sync_out; 1609 } 1610 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 1611 if (blkaddr == NULL_ADDR) { 1612 prealloc++; 1613 last_ofs_in_node = dn.ofs_in_node; 1614 } 1615 } else { 1616 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO && 1617 flag != F2FS_GET_BLOCK_DIO); 1618 err = __allocate_data_block(&dn, 1619 map->m_seg_type); 1620 if (!err) { 1621 if (flag == F2FS_GET_BLOCK_PRE_DIO) 1622 file_need_truncate(inode); 1623 set_inode_flag(inode, FI_APPEND_WRITE); 1624 } 1625 } 1626 if (err) 1627 goto sync_out; 1628 map->m_flags |= F2FS_MAP_NEW; 1629 blkaddr = dn.data_blkaddr; 1630 } else { 1631 if (f2fs_compressed_file(inode) && 1632 f2fs_sanity_check_cluster(&dn) && 1633 (flag != F2FS_GET_BLOCK_FIEMAP || 1634 IS_ENABLED(CONFIG_F2FS_CHECK_FS))) { 1635 err = -EFSCORRUPTED; 1636 f2fs_handle_error(sbi, 1637 ERROR_CORRUPTED_CLUSTER); 1638 goto sync_out; 1639 } 1640 if (flag == F2FS_GET_BLOCK_BMAP) { 1641 map->m_pblk = 0; 1642 goto sync_out; 1643 } 1644 if (flag == F2FS_GET_BLOCK_PRECACHE) 1645 goto sync_out; 1646 if (flag == F2FS_GET_BLOCK_FIEMAP && 1647 blkaddr == NULL_ADDR) { 1648 if (map->m_next_pgofs) 1649 *map->m_next_pgofs = pgofs + 1; 1650 goto sync_out; 1651 } 1652 if (flag != F2FS_GET_BLOCK_FIEMAP) { 1653 /* for defragment case */ 1654 if (map->m_next_pgofs) 1655 *map->m_next_pgofs = pgofs + 1; 1656 goto sync_out; 1657 } 1658 } 1659 } 1660 1661 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1662 goto skip; 1663 1664 if (map->m_multidev_dio) 1665 bidx = f2fs_target_device_index(sbi, blkaddr); 1666 1667 if (map->m_len == 0) { 1668 /* reserved delalloc block should be mapped for fiemap. */ 1669 if (blkaddr == NEW_ADDR) 1670 map->m_flags |= F2FS_MAP_DELALLOC; 1671 map->m_flags |= F2FS_MAP_MAPPED; 1672 1673 map->m_pblk = blkaddr; 1674 map->m_len = 1; 1675 1676 if (map->m_multidev_dio) 1677 map->m_bdev = FDEV(bidx).bdev; 1678 } else if ((map->m_pblk != NEW_ADDR && 1679 blkaddr == (map->m_pblk + ofs)) || 1680 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1681 flag == F2FS_GET_BLOCK_PRE_DIO) { 1682 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev) 1683 goto sync_out; 1684 ofs++; 1685 map->m_len++; 1686 } else { 1687 goto sync_out; 1688 } 1689 1690 skip: 1691 dn.ofs_in_node++; 1692 pgofs++; 1693 1694 /* preallocate blocks in batch for one dnode page */ 1695 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1696 (pgofs == end || dn.ofs_in_node == end_offset)) { 1697 1698 dn.ofs_in_node = ofs_in_node; 1699 err = f2fs_reserve_new_blocks(&dn, prealloc); 1700 if (err) 1701 goto sync_out; 1702 1703 map->m_len += dn.ofs_in_node - ofs_in_node; 1704 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1705 err = -ENOSPC; 1706 goto sync_out; 1707 } 1708 dn.ofs_in_node = end_offset; 1709 } 1710 1711 if (pgofs >= end) 1712 goto sync_out; 1713 else if (dn.ofs_in_node < end_offset) 1714 goto next_block; 1715 1716 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1717 if (map->m_flags & F2FS_MAP_MAPPED) { 1718 unsigned int ofs = start_pgofs - map->m_lblk; 1719 1720 f2fs_update_read_extent_cache_range(&dn, 1721 start_pgofs, map->m_pblk + ofs, 1722 map->m_len - ofs); 1723 } 1724 } 1725 1726 f2fs_put_dnode(&dn); 1727 1728 if (map->m_may_create) { 1729 f2fs_do_map_lock(sbi, flag, false); 1730 f2fs_balance_fs(sbi, dn.node_changed); 1731 } 1732 goto next_dnode; 1733 1734 sync_out: 1735 1736 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) { 1737 /* 1738 * for hardware encryption, but to avoid potential issue 1739 * in future 1740 */ 1741 f2fs_wait_on_block_writeback_range(inode, 1742 map->m_pblk, map->m_len); 1743 1744 if (map->m_multidev_dio) { 1745 block_t blk_addr = map->m_pblk; 1746 1747 bidx = f2fs_target_device_index(sbi, map->m_pblk); 1748 1749 map->m_bdev = FDEV(bidx).bdev; 1750 map->m_pblk -= FDEV(bidx).start_blk; 1751 1752 if (map->m_may_create) 1753 f2fs_update_device_state(sbi, inode->i_ino, 1754 blk_addr, map->m_len); 1755 1756 f2fs_bug_on(sbi, blk_addr + map->m_len > 1757 FDEV(bidx).end_blk + 1); 1758 } 1759 } 1760 1761 if (flag == F2FS_GET_BLOCK_PRECACHE) { 1762 if (map->m_flags & F2FS_MAP_MAPPED) { 1763 unsigned int ofs = start_pgofs - map->m_lblk; 1764 1765 f2fs_update_read_extent_cache_range(&dn, 1766 start_pgofs, map->m_pblk + ofs, 1767 map->m_len - ofs); 1768 } 1769 if (map->m_next_extent) 1770 *map->m_next_extent = pgofs + 1; 1771 } 1772 f2fs_put_dnode(&dn); 1773 unlock_out: 1774 if (map->m_may_create) { 1775 f2fs_do_map_lock(sbi, flag, false); 1776 f2fs_balance_fs(sbi, dn.node_changed); 1777 } 1778 out: 1779 trace_f2fs_map_blocks(inode, map, create, flag, err); 1780 return err; 1781 } 1782 1783 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) 1784 { 1785 struct f2fs_map_blocks map; 1786 block_t last_lblk; 1787 int err; 1788 1789 if (pos + len > i_size_read(inode)) 1790 return false; 1791 1792 map.m_lblk = F2FS_BYTES_TO_BLK(pos); 1793 map.m_next_pgofs = NULL; 1794 map.m_next_extent = NULL; 1795 map.m_seg_type = NO_CHECK_TYPE; 1796 map.m_may_create = false; 1797 last_lblk = F2FS_BLK_ALIGN(pos + len); 1798 1799 while (map.m_lblk < last_lblk) { 1800 map.m_len = last_lblk - map.m_lblk; 1801 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT); 1802 if (err || map.m_len == 0) 1803 return false; 1804 map.m_lblk += map.m_len; 1805 } 1806 return true; 1807 } 1808 1809 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes) 1810 { 1811 return (bytes >> inode->i_blkbits); 1812 } 1813 1814 static inline u64 blks_to_bytes(struct inode *inode, u64 blks) 1815 { 1816 return (blks << inode->i_blkbits); 1817 } 1818 1819 static int f2fs_xattr_fiemap(struct inode *inode, 1820 struct fiemap_extent_info *fieinfo) 1821 { 1822 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1823 struct page *page; 1824 struct node_info ni; 1825 __u64 phys = 0, len; 1826 __u32 flags; 1827 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1828 int err = 0; 1829 1830 if (f2fs_has_inline_xattr(inode)) { 1831 int offset; 1832 1833 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1834 inode->i_ino, false); 1835 if (!page) 1836 return -ENOMEM; 1837 1838 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false); 1839 if (err) { 1840 f2fs_put_page(page, 1); 1841 return err; 1842 } 1843 1844 phys = blks_to_bytes(inode, ni.blk_addr); 1845 offset = offsetof(struct f2fs_inode, i_addr) + 1846 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1847 get_inline_xattr_addrs(inode)); 1848 1849 phys += offset; 1850 len = inline_xattr_size(inode); 1851 1852 f2fs_put_page(page, 1); 1853 1854 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1855 1856 if (!xnid) 1857 flags |= FIEMAP_EXTENT_LAST; 1858 1859 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1860 trace_f2fs_fiemap(inode, 0, phys, len, flags, err); 1861 if (err) 1862 return err; 1863 } 1864 1865 if (xnid) { 1866 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1867 if (!page) 1868 return -ENOMEM; 1869 1870 err = f2fs_get_node_info(sbi, xnid, &ni, false); 1871 if (err) { 1872 f2fs_put_page(page, 1); 1873 return err; 1874 } 1875 1876 phys = blks_to_bytes(inode, ni.blk_addr); 1877 len = inode->i_sb->s_blocksize; 1878 1879 f2fs_put_page(page, 1); 1880 1881 flags = FIEMAP_EXTENT_LAST; 1882 } 1883 1884 if (phys) { 1885 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1886 trace_f2fs_fiemap(inode, 0, phys, len, flags, err); 1887 } 1888 1889 return (err < 0 ? err : 0); 1890 } 1891 1892 static loff_t max_inode_blocks(struct inode *inode) 1893 { 1894 loff_t result = ADDRS_PER_INODE(inode); 1895 loff_t leaf_count = ADDRS_PER_BLOCK(inode); 1896 1897 /* two direct node blocks */ 1898 result += (leaf_count * 2); 1899 1900 /* two indirect node blocks */ 1901 leaf_count *= NIDS_PER_BLOCK; 1902 result += (leaf_count * 2); 1903 1904 /* one double indirect node block */ 1905 leaf_count *= NIDS_PER_BLOCK; 1906 result += leaf_count; 1907 1908 return result; 1909 } 1910 1911 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1912 u64 start, u64 len) 1913 { 1914 struct f2fs_map_blocks map; 1915 sector_t start_blk, last_blk; 1916 pgoff_t next_pgofs; 1917 u64 logical = 0, phys = 0, size = 0; 1918 u32 flags = 0; 1919 int ret = 0; 1920 bool compr_cluster = false, compr_appended; 1921 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size; 1922 unsigned int count_in_cluster = 0; 1923 loff_t maxbytes; 1924 1925 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { 1926 ret = f2fs_precache_extents(inode); 1927 if (ret) 1928 return ret; 1929 } 1930 1931 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR); 1932 if (ret) 1933 return ret; 1934 1935 inode_lock(inode); 1936 1937 maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS; 1938 if (start > maxbytes) { 1939 ret = -EFBIG; 1940 goto out; 1941 } 1942 1943 if (len > maxbytes || (maxbytes - len) < start) 1944 len = maxbytes - start; 1945 1946 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1947 ret = f2fs_xattr_fiemap(inode, fieinfo); 1948 goto out; 1949 } 1950 1951 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) { 1952 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1953 if (ret != -EAGAIN) 1954 goto out; 1955 } 1956 1957 if (bytes_to_blks(inode, len) == 0) 1958 len = blks_to_bytes(inode, 1); 1959 1960 start_blk = bytes_to_blks(inode, start); 1961 last_blk = bytes_to_blks(inode, start + len - 1); 1962 1963 next: 1964 memset(&map, 0, sizeof(map)); 1965 map.m_lblk = start_blk; 1966 map.m_len = bytes_to_blks(inode, len); 1967 map.m_next_pgofs = &next_pgofs; 1968 map.m_seg_type = NO_CHECK_TYPE; 1969 1970 if (compr_cluster) { 1971 map.m_lblk += 1; 1972 map.m_len = cluster_size - count_in_cluster; 1973 } 1974 1975 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP); 1976 if (ret) 1977 goto out; 1978 1979 /* HOLE */ 1980 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) { 1981 start_blk = next_pgofs; 1982 1983 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode, 1984 max_inode_blocks(inode))) 1985 goto prep_next; 1986 1987 flags |= FIEMAP_EXTENT_LAST; 1988 } 1989 1990 compr_appended = false; 1991 /* In a case of compressed cluster, append this to the last extent */ 1992 if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) || 1993 !(map.m_flags & F2FS_MAP_FLAGS))) { 1994 compr_appended = true; 1995 goto skip_fill; 1996 } 1997 1998 if (size) { 1999 flags |= FIEMAP_EXTENT_MERGED; 2000 if (IS_ENCRYPTED(inode)) 2001 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 2002 2003 ret = fiemap_fill_next_extent(fieinfo, logical, 2004 phys, size, flags); 2005 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret); 2006 if (ret) 2007 goto out; 2008 size = 0; 2009 } 2010 2011 if (start_blk > last_blk) 2012 goto out; 2013 2014 skip_fill: 2015 if (map.m_pblk == COMPRESS_ADDR) { 2016 compr_cluster = true; 2017 count_in_cluster = 1; 2018 } else if (compr_appended) { 2019 unsigned int appended_blks = cluster_size - 2020 count_in_cluster + 1; 2021 size += blks_to_bytes(inode, appended_blks); 2022 start_blk += appended_blks; 2023 compr_cluster = false; 2024 } else { 2025 logical = blks_to_bytes(inode, start_blk); 2026 phys = __is_valid_data_blkaddr(map.m_pblk) ? 2027 blks_to_bytes(inode, map.m_pblk) : 0; 2028 size = blks_to_bytes(inode, map.m_len); 2029 flags = 0; 2030 2031 if (compr_cluster) { 2032 flags = FIEMAP_EXTENT_ENCODED; 2033 count_in_cluster += map.m_len; 2034 if (count_in_cluster == cluster_size) { 2035 compr_cluster = false; 2036 size += blks_to_bytes(inode, 1); 2037 } 2038 } else if (map.m_flags & F2FS_MAP_DELALLOC) { 2039 flags = FIEMAP_EXTENT_UNWRITTEN; 2040 } 2041 2042 start_blk += bytes_to_blks(inode, size); 2043 } 2044 2045 prep_next: 2046 cond_resched(); 2047 if (fatal_signal_pending(current)) 2048 ret = -EINTR; 2049 else 2050 goto next; 2051 out: 2052 if (ret == 1) 2053 ret = 0; 2054 2055 inode_unlock(inode); 2056 return ret; 2057 } 2058 2059 static inline loff_t f2fs_readpage_limit(struct inode *inode) 2060 { 2061 if (IS_ENABLED(CONFIG_FS_VERITY) && 2062 (IS_VERITY(inode) || f2fs_verity_in_progress(inode))) 2063 return inode->i_sb->s_maxbytes; 2064 2065 return i_size_read(inode); 2066 } 2067 2068 static int f2fs_read_single_page(struct inode *inode, struct page *page, 2069 unsigned nr_pages, 2070 struct f2fs_map_blocks *map, 2071 struct bio **bio_ret, 2072 sector_t *last_block_in_bio, 2073 bool is_readahead) 2074 { 2075 struct bio *bio = *bio_ret; 2076 const unsigned blocksize = blks_to_bytes(inode, 1); 2077 sector_t block_in_file; 2078 sector_t last_block; 2079 sector_t last_block_in_file; 2080 sector_t block_nr; 2081 int ret = 0; 2082 2083 block_in_file = (sector_t)page_index(page); 2084 last_block = block_in_file + nr_pages; 2085 last_block_in_file = bytes_to_blks(inode, 2086 f2fs_readpage_limit(inode) + blocksize - 1); 2087 if (last_block > last_block_in_file) 2088 last_block = last_block_in_file; 2089 2090 /* just zeroing out page which is beyond EOF */ 2091 if (block_in_file >= last_block) 2092 goto zero_out; 2093 /* 2094 * Map blocks using the previous result first. 2095 */ 2096 if ((map->m_flags & F2FS_MAP_MAPPED) && 2097 block_in_file > map->m_lblk && 2098 block_in_file < (map->m_lblk + map->m_len)) 2099 goto got_it; 2100 2101 /* 2102 * Then do more f2fs_map_blocks() calls until we are 2103 * done with this page. 2104 */ 2105 map->m_lblk = block_in_file; 2106 map->m_len = last_block - block_in_file; 2107 2108 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT); 2109 if (ret) 2110 goto out; 2111 got_it: 2112 if ((map->m_flags & F2FS_MAP_MAPPED)) { 2113 block_nr = map->m_pblk + block_in_file - map->m_lblk; 2114 SetPageMappedToDisk(page); 2115 2116 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr, 2117 DATA_GENERIC_ENHANCE_READ)) { 2118 ret = -EFSCORRUPTED; 2119 f2fs_handle_error(F2FS_I_SB(inode), 2120 ERROR_INVALID_BLKADDR); 2121 goto out; 2122 } 2123 } else { 2124 zero_out: 2125 zero_user_segment(page, 0, PAGE_SIZE); 2126 if (f2fs_need_verity(inode, page->index) && 2127 !fsverity_verify_page(page)) { 2128 ret = -EIO; 2129 goto out; 2130 } 2131 if (!PageUptodate(page)) 2132 SetPageUptodate(page); 2133 unlock_page(page); 2134 goto out; 2135 } 2136 2137 /* 2138 * This page will go to BIO. Do we need to send this 2139 * BIO off first? 2140 */ 2141 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio, 2142 *last_block_in_bio, block_nr) || 2143 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) { 2144 submit_and_realloc: 2145 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA); 2146 bio = NULL; 2147 } 2148 if (bio == NULL) { 2149 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages, 2150 is_readahead ? REQ_RAHEAD : 0, page->index, 2151 false); 2152 if (IS_ERR(bio)) { 2153 ret = PTR_ERR(bio); 2154 bio = NULL; 2155 goto out; 2156 } 2157 } 2158 2159 /* 2160 * If the page is under writeback, we need to wait for 2161 * its completion to see the correct decrypted data. 2162 */ 2163 f2fs_wait_on_block_writeback(inode, block_nr); 2164 2165 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2166 goto submit_and_realloc; 2167 2168 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA); 2169 f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO, 2170 F2FS_BLKSIZE); 2171 *last_block_in_bio = block_nr; 2172 goto out; 2173 out: 2174 *bio_ret = bio; 2175 return ret; 2176 } 2177 2178 #ifdef CONFIG_F2FS_FS_COMPRESSION 2179 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret, 2180 unsigned nr_pages, sector_t *last_block_in_bio, 2181 bool is_readahead, bool for_write) 2182 { 2183 struct dnode_of_data dn; 2184 struct inode *inode = cc->inode; 2185 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2186 struct bio *bio = *bio_ret; 2187 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size; 2188 sector_t last_block_in_file; 2189 const unsigned blocksize = blks_to_bytes(inode, 1); 2190 struct decompress_io_ctx *dic = NULL; 2191 struct extent_info ei = {}; 2192 bool from_dnode = true; 2193 int i; 2194 int ret = 0; 2195 2196 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc)); 2197 2198 last_block_in_file = bytes_to_blks(inode, 2199 f2fs_readpage_limit(inode) + blocksize - 1); 2200 2201 /* get rid of pages beyond EOF */ 2202 for (i = 0; i < cc->cluster_size; i++) { 2203 struct page *page = cc->rpages[i]; 2204 2205 if (!page) 2206 continue; 2207 if ((sector_t)page->index >= last_block_in_file) { 2208 zero_user_segment(page, 0, PAGE_SIZE); 2209 if (!PageUptodate(page)) 2210 SetPageUptodate(page); 2211 } else if (!PageUptodate(page)) { 2212 continue; 2213 } 2214 unlock_page(page); 2215 if (for_write) 2216 put_page(page); 2217 cc->rpages[i] = NULL; 2218 cc->nr_rpages--; 2219 } 2220 2221 /* we are done since all pages are beyond EOF */ 2222 if (f2fs_cluster_is_empty(cc)) 2223 goto out; 2224 2225 if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei)) 2226 from_dnode = false; 2227 2228 if (!from_dnode) 2229 goto skip_reading_dnode; 2230 2231 set_new_dnode(&dn, inode, NULL, NULL, 0); 2232 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 2233 if (ret) 2234 goto out; 2235 2236 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR); 2237 2238 skip_reading_dnode: 2239 for (i = 1; i < cc->cluster_size; i++) { 2240 block_t blkaddr; 2241 2242 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page, 2243 dn.ofs_in_node + i) : 2244 ei.blk + i - 1; 2245 2246 if (!__is_valid_data_blkaddr(blkaddr)) 2247 break; 2248 2249 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) { 2250 ret = -EFAULT; 2251 goto out_put_dnode; 2252 } 2253 cc->nr_cpages++; 2254 2255 if (!from_dnode && i >= ei.c_len) 2256 break; 2257 } 2258 2259 /* nothing to decompress */ 2260 if (cc->nr_cpages == 0) { 2261 ret = 0; 2262 goto out_put_dnode; 2263 } 2264 2265 dic = f2fs_alloc_dic(cc); 2266 if (IS_ERR(dic)) { 2267 ret = PTR_ERR(dic); 2268 goto out_put_dnode; 2269 } 2270 2271 for (i = 0; i < cc->nr_cpages; i++) { 2272 struct page *page = dic->cpages[i]; 2273 block_t blkaddr; 2274 struct bio_post_read_ctx *ctx; 2275 2276 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page, 2277 dn.ofs_in_node + i + 1) : 2278 ei.blk + i; 2279 2280 f2fs_wait_on_block_writeback(inode, blkaddr); 2281 2282 if (f2fs_load_compressed_page(sbi, page, blkaddr)) { 2283 if (atomic_dec_and_test(&dic->remaining_pages)) 2284 f2fs_decompress_cluster(dic, true); 2285 continue; 2286 } 2287 2288 if (bio && (!page_is_mergeable(sbi, bio, 2289 *last_block_in_bio, blkaddr) || 2290 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) { 2291 submit_and_realloc: 2292 f2fs_submit_read_bio(sbi, bio, DATA); 2293 bio = NULL; 2294 } 2295 2296 if (!bio) { 2297 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages, 2298 is_readahead ? REQ_RAHEAD : 0, 2299 page->index, for_write); 2300 if (IS_ERR(bio)) { 2301 ret = PTR_ERR(bio); 2302 f2fs_decompress_end_io(dic, ret, true); 2303 f2fs_put_dnode(&dn); 2304 *bio_ret = NULL; 2305 return ret; 2306 } 2307 } 2308 2309 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 2310 goto submit_and_realloc; 2311 2312 ctx = get_post_read_ctx(bio); 2313 ctx->enabled_steps |= STEP_DECOMPRESS; 2314 refcount_inc(&dic->refcnt); 2315 2316 inc_page_count(sbi, F2FS_RD_DATA); 2317 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE); 2318 *last_block_in_bio = blkaddr; 2319 } 2320 2321 if (from_dnode) 2322 f2fs_put_dnode(&dn); 2323 2324 *bio_ret = bio; 2325 return 0; 2326 2327 out_put_dnode: 2328 if (from_dnode) 2329 f2fs_put_dnode(&dn); 2330 out: 2331 for (i = 0; i < cc->cluster_size; i++) { 2332 if (cc->rpages[i]) { 2333 ClearPageUptodate(cc->rpages[i]); 2334 unlock_page(cc->rpages[i]); 2335 } 2336 } 2337 *bio_ret = bio; 2338 return ret; 2339 } 2340 #endif 2341 2342 /* 2343 * This function was originally taken from fs/mpage.c, and customized for f2fs. 2344 * Major change was from block_size == page_size in f2fs by default. 2345 */ 2346 static int f2fs_mpage_readpages(struct inode *inode, 2347 struct readahead_control *rac, struct page *page) 2348 { 2349 struct bio *bio = NULL; 2350 sector_t last_block_in_bio = 0; 2351 struct f2fs_map_blocks map; 2352 #ifdef CONFIG_F2FS_FS_COMPRESSION 2353 struct compress_ctx cc = { 2354 .inode = inode, 2355 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2356 .cluster_size = F2FS_I(inode)->i_cluster_size, 2357 .cluster_idx = NULL_CLUSTER, 2358 .rpages = NULL, 2359 .cpages = NULL, 2360 .nr_rpages = 0, 2361 .nr_cpages = 0, 2362 }; 2363 pgoff_t nc_cluster_idx = NULL_CLUSTER; 2364 #endif 2365 unsigned nr_pages = rac ? readahead_count(rac) : 1; 2366 unsigned max_nr_pages = nr_pages; 2367 int ret = 0; 2368 2369 map.m_pblk = 0; 2370 map.m_lblk = 0; 2371 map.m_len = 0; 2372 map.m_flags = 0; 2373 map.m_next_pgofs = NULL; 2374 map.m_next_extent = NULL; 2375 map.m_seg_type = NO_CHECK_TYPE; 2376 map.m_may_create = false; 2377 2378 for (; nr_pages; nr_pages--) { 2379 if (rac) { 2380 page = readahead_page(rac); 2381 prefetchw(&page->flags); 2382 } 2383 2384 #ifdef CONFIG_F2FS_FS_COMPRESSION 2385 if (f2fs_compressed_file(inode)) { 2386 /* there are remained comressed pages, submit them */ 2387 if (!f2fs_cluster_can_merge_page(&cc, page->index)) { 2388 ret = f2fs_read_multi_pages(&cc, &bio, 2389 max_nr_pages, 2390 &last_block_in_bio, 2391 rac != NULL, false); 2392 f2fs_destroy_compress_ctx(&cc, false); 2393 if (ret) 2394 goto set_error_page; 2395 } 2396 if (cc.cluster_idx == NULL_CLUSTER) { 2397 if (nc_cluster_idx == 2398 page->index >> cc.log_cluster_size) { 2399 goto read_single_page; 2400 } 2401 2402 ret = f2fs_is_compressed_cluster(inode, page->index); 2403 if (ret < 0) 2404 goto set_error_page; 2405 else if (!ret) { 2406 nc_cluster_idx = 2407 page->index >> cc.log_cluster_size; 2408 goto read_single_page; 2409 } 2410 2411 nc_cluster_idx = NULL_CLUSTER; 2412 } 2413 ret = f2fs_init_compress_ctx(&cc); 2414 if (ret) 2415 goto set_error_page; 2416 2417 f2fs_compress_ctx_add_page(&cc, page); 2418 2419 goto next_page; 2420 } 2421 read_single_page: 2422 #endif 2423 2424 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map, 2425 &bio, &last_block_in_bio, rac); 2426 if (ret) { 2427 #ifdef CONFIG_F2FS_FS_COMPRESSION 2428 set_error_page: 2429 #endif 2430 zero_user_segment(page, 0, PAGE_SIZE); 2431 unlock_page(page); 2432 } 2433 #ifdef CONFIG_F2FS_FS_COMPRESSION 2434 next_page: 2435 #endif 2436 if (rac) 2437 put_page(page); 2438 2439 #ifdef CONFIG_F2FS_FS_COMPRESSION 2440 if (f2fs_compressed_file(inode)) { 2441 /* last page */ 2442 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) { 2443 ret = f2fs_read_multi_pages(&cc, &bio, 2444 max_nr_pages, 2445 &last_block_in_bio, 2446 rac != NULL, false); 2447 f2fs_destroy_compress_ctx(&cc, false); 2448 } 2449 } 2450 #endif 2451 } 2452 if (bio) 2453 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA); 2454 return ret; 2455 } 2456 2457 static int f2fs_read_data_folio(struct file *file, struct folio *folio) 2458 { 2459 struct page *page = &folio->page; 2460 struct inode *inode = page_file_mapping(page)->host; 2461 int ret = -EAGAIN; 2462 2463 trace_f2fs_readpage(page, DATA); 2464 2465 if (!f2fs_is_compress_backend_ready(inode)) { 2466 unlock_page(page); 2467 return -EOPNOTSUPP; 2468 } 2469 2470 /* If the file has inline data, try to read it directly */ 2471 if (f2fs_has_inline_data(inode)) 2472 ret = f2fs_read_inline_data(inode, page); 2473 if (ret == -EAGAIN) 2474 ret = f2fs_mpage_readpages(inode, NULL, page); 2475 return ret; 2476 } 2477 2478 static void f2fs_readahead(struct readahead_control *rac) 2479 { 2480 struct inode *inode = rac->mapping->host; 2481 2482 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac)); 2483 2484 if (!f2fs_is_compress_backend_ready(inode)) 2485 return; 2486 2487 /* If the file has inline data, skip readahead */ 2488 if (f2fs_has_inline_data(inode)) 2489 return; 2490 2491 f2fs_mpage_readpages(inode, rac, NULL); 2492 } 2493 2494 int f2fs_encrypt_one_page(struct f2fs_io_info *fio) 2495 { 2496 struct inode *inode = fio->page->mapping->host; 2497 struct page *mpage, *page; 2498 gfp_t gfp_flags = GFP_NOFS; 2499 2500 if (!f2fs_encrypted_file(inode)) 2501 return 0; 2502 2503 page = fio->compressed_page ? fio->compressed_page : fio->page; 2504 2505 /* wait for GCed page writeback via META_MAPPING */ 2506 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr); 2507 2508 if (fscrypt_inode_uses_inline_crypto(inode)) 2509 return 0; 2510 2511 retry_encrypt: 2512 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page, 2513 PAGE_SIZE, 0, gfp_flags); 2514 if (IS_ERR(fio->encrypted_page)) { 2515 /* flush pending IOs and wait for a while in the ENOMEM case */ 2516 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 2517 f2fs_flush_merged_writes(fio->sbi); 2518 memalloc_retry_wait(GFP_NOFS); 2519 gfp_flags |= __GFP_NOFAIL; 2520 goto retry_encrypt; 2521 } 2522 return PTR_ERR(fio->encrypted_page); 2523 } 2524 2525 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr); 2526 if (mpage) { 2527 if (PageUptodate(mpage)) 2528 memcpy(page_address(mpage), 2529 page_address(fio->encrypted_page), PAGE_SIZE); 2530 f2fs_put_page(mpage, 1); 2531 } 2532 return 0; 2533 } 2534 2535 static inline bool check_inplace_update_policy(struct inode *inode, 2536 struct f2fs_io_info *fio) 2537 { 2538 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2539 unsigned int policy = SM_I(sbi)->ipu_policy; 2540 2541 if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) && 2542 is_inode_flag_set(inode, FI_OPU_WRITE)) 2543 return false; 2544 if (policy & (0x1 << F2FS_IPU_FORCE)) 2545 return true; 2546 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi)) 2547 return true; 2548 if (policy & (0x1 << F2FS_IPU_UTIL) && 2549 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2550 return true; 2551 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) && 2552 utilization(sbi) > SM_I(sbi)->min_ipu_util) 2553 return true; 2554 2555 /* 2556 * IPU for rewrite async pages 2557 */ 2558 if (policy & (0x1 << F2FS_IPU_ASYNC) && 2559 fio && fio->op == REQ_OP_WRITE && 2560 !(fio->op_flags & REQ_SYNC) && 2561 !IS_ENCRYPTED(inode)) 2562 return true; 2563 2564 /* this is only set during fdatasync */ 2565 if (policy & (0x1 << F2FS_IPU_FSYNC) && 2566 is_inode_flag_set(inode, FI_NEED_IPU)) 2567 return true; 2568 2569 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2570 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2571 return true; 2572 2573 return false; 2574 } 2575 2576 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) 2577 { 2578 /* swap file is migrating in aligned write mode */ 2579 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 2580 return false; 2581 2582 if (f2fs_is_pinned_file(inode)) 2583 return true; 2584 2585 /* if this is cold file, we should overwrite to avoid fragmentation */ 2586 if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE)) 2587 return true; 2588 2589 return check_inplace_update_policy(inode, fio); 2590 } 2591 2592 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) 2593 { 2594 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2595 2596 /* The below cases were checked when setting it. */ 2597 if (f2fs_is_pinned_file(inode)) 2598 return false; 2599 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 2600 return true; 2601 if (f2fs_lfs_mode(sbi)) 2602 return true; 2603 if (S_ISDIR(inode->i_mode)) 2604 return true; 2605 if (IS_NOQUOTA(inode)) 2606 return true; 2607 if (f2fs_is_atomic_file(inode)) 2608 return true; 2609 2610 /* swap file is migrating in aligned write mode */ 2611 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 2612 return true; 2613 2614 if (is_inode_flag_set(inode, FI_OPU_WRITE)) 2615 return true; 2616 2617 if (fio) { 2618 if (page_private_gcing(fio->page)) 2619 return true; 2620 if (page_private_dummy(fio->page)) 2621 return true; 2622 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 2623 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) 2624 return true; 2625 } 2626 return false; 2627 } 2628 2629 static inline bool need_inplace_update(struct f2fs_io_info *fio) 2630 { 2631 struct inode *inode = fio->page->mapping->host; 2632 2633 if (f2fs_should_update_outplace(inode, fio)) 2634 return false; 2635 2636 return f2fs_should_update_inplace(inode, fio); 2637 } 2638 2639 int f2fs_do_write_data_page(struct f2fs_io_info *fio) 2640 { 2641 struct page *page = fio->page; 2642 struct inode *inode = page->mapping->host; 2643 struct dnode_of_data dn; 2644 struct extent_info ei = {0, }; 2645 struct node_info ni; 2646 bool ipu_force = false; 2647 int err = 0; 2648 2649 /* Use COW inode to make dnode_of_data for atomic write */ 2650 if (f2fs_is_atomic_file(inode)) 2651 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0); 2652 else 2653 set_new_dnode(&dn, inode, NULL, NULL, 0); 2654 2655 if (need_inplace_update(fio) && 2656 f2fs_lookup_read_extent_cache(inode, page->index, &ei)) { 2657 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 2658 2659 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr, 2660 DATA_GENERIC_ENHANCE)) { 2661 f2fs_handle_error(fio->sbi, 2662 ERROR_INVALID_BLKADDR); 2663 return -EFSCORRUPTED; 2664 } 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, page->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 ClearPageUptodate(page); 2684 clear_page_private_gcing(page); 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 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR); 2693 goto out_writepage; 2694 } 2695 2696 /* 2697 * If current allocation needs SSR, 2698 * it had better in-place writes for updated data. 2699 */ 2700 if (ipu_force || 2701 (__is_valid_data_blkaddr(fio->old_blkaddr) && 2702 need_inplace_update(fio))) { 2703 err = f2fs_encrypt_one_page(fio); 2704 if (err) 2705 goto out_writepage; 2706 2707 set_page_writeback(page); 2708 ClearPageError(page); 2709 f2fs_put_dnode(&dn); 2710 if (fio->need_lock == LOCK_REQ) 2711 f2fs_unlock_op(fio->sbi); 2712 err = f2fs_inplace_write_data(fio); 2713 if (err) { 2714 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 2715 fscrypt_finalize_bounce_page(&fio->encrypted_page); 2716 if (PageWriteback(page)) 2717 end_page_writeback(page); 2718 } else { 2719 set_inode_flag(inode, FI_UPDATE_WRITE); 2720 } 2721 trace_f2fs_do_write_data_page(fio->page, IPU); 2722 return err; 2723 } 2724 2725 if (fio->need_lock == LOCK_RETRY) { 2726 if (!f2fs_trylock_op(fio->sbi)) { 2727 err = -EAGAIN; 2728 goto out_writepage; 2729 } 2730 fio->need_lock = LOCK_REQ; 2731 } 2732 2733 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false); 2734 if (err) 2735 goto out_writepage; 2736 2737 fio->version = ni.version; 2738 2739 err = f2fs_encrypt_one_page(fio); 2740 if (err) 2741 goto out_writepage; 2742 2743 set_page_writeback(page); 2744 ClearPageError(page); 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(page, OPU); 2752 set_inode_flag(inode, FI_APPEND_WRITE); 2753 if (page->index == 0) 2754 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 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 page *page, 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 = page->mapping->host; 2772 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2773 loff_t i_size = i_size_read(inode); 2774 const pgoff_t end_index = ((unsigned long long)i_size) 2775 >> PAGE_SHIFT; 2776 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT; 2777 unsigned offset = 0; 2778 bool need_balance_fs = false; 2779 int err = 0; 2780 struct f2fs_io_info fio = { 2781 .sbi = sbi, 2782 .ino = inode->i_ino, 2783 .type = DATA, 2784 .op = REQ_OP_WRITE, 2785 .op_flags = wbc_to_write_flags(wbc), 2786 .old_blkaddr = NULL_ADDR, 2787 .page = page, 2788 .encrypted_page = NULL, 2789 .submitted = false, 2790 .compr_blocks = compr_blocks, 2791 .need_lock = LOCK_RETRY, 2792 .post_read = f2fs_post_read_required(inode), 2793 .io_type = io_type, 2794 .io_wbc = wbc, 2795 .bio = bio, 2796 .last_block = last_block, 2797 }; 2798 2799 trace_f2fs_writepage(page, DATA); 2800 2801 /* we should bypass data pages to proceed the kworkder jobs */ 2802 if (unlikely(f2fs_cp_error(sbi))) { 2803 mapping_set_error(page->mapping, -EIO); 2804 /* 2805 * don't drop any dirty dentry pages for keeping lastest 2806 * directory structure. 2807 */ 2808 if (S_ISDIR(inode->i_mode)) 2809 goto redirty_out; 2810 goto out; 2811 } 2812 2813 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 2814 goto redirty_out; 2815 2816 if (page->index < end_index || 2817 f2fs_verity_in_progress(inode) || 2818 compr_blocks) 2819 goto write; 2820 2821 /* 2822 * If the offset is out-of-range of file size, 2823 * this page does not have to be written to disk. 2824 */ 2825 offset = i_size & (PAGE_SIZE - 1); 2826 if ((page->index >= end_index + 1) || !offset) 2827 goto out; 2828 2829 zero_user_segment(page, offset, PAGE_SIZE); 2830 write: 2831 if (f2fs_is_drop_cache(inode)) 2832 goto out; 2833 2834 /* Dentry/quota blocks are controlled by checkpoint */ 2835 if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) { 2836 /* 2837 * We need to wait for node_write to avoid block allocation during 2838 * checkpoint. This can only happen to quota writes which can cause 2839 * the below discard race condition. 2840 */ 2841 if (IS_NOQUOTA(inode)) 2842 f2fs_down_read(&sbi->node_write); 2843 2844 fio.need_lock = LOCK_DONE; 2845 err = f2fs_do_write_data_page(&fio); 2846 2847 if (IS_NOQUOTA(inode)) 2848 f2fs_up_read(&sbi->node_write); 2849 2850 goto done; 2851 } 2852 2853 if (!wbc->for_reclaim) 2854 need_balance_fs = true; 2855 else if (has_not_enough_free_secs(sbi, 0, 0)) 2856 goto redirty_out; 2857 else 2858 set_inode_flag(inode, FI_HOT_DATA); 2859 2860 err = -EAGAIN; 2861 if (f2fs_has_inline_data(inode)) { 2862 err = f2fs_write_inline_data(inode, page); 2863 if (!err) 2864 goto out; 2865 } 2866 2867 if (err == -EAGAIN) { 2868 err = f2fs_do_write_data_page(&fio); 2869 if (err == -EAGAIN) { 2870 fio.need_lock = LOCK_REQ; 2871 err = f2fs_do_write_data_page(&fio); 2872 } 2873 } 2874 2875 if (err) { 2876 file_set_keep_isize(inode); 2877 } else { 2878 spin_lock(&F2FS_I(inode)->i_size_lock); 2879 if (F2FS_I(inode)->last_disk_size < psize) 2880 F2FS_I(inode)->last_disk_size = psize; 2881 spin_unlock(&F2FS_I(inode)->i_size_lock); 2882 } 2883 2884 done: 2885 if (err && err != -ENOENT) 2886 goto redirty_out; 2887 2888 out: 2889 inode_dec_dirty_pages(inode); 2890 if (err) { 2891 ClearPageUptodate(page); 2892 clear_page_private_gcing(page); 2893 } 2894 2895 if (wbc->for_reclaim) { 2896 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA); 2897 clear_inode_flag(inode, FI_HOT_DATA); 2898 f2fs_remove_dirty_inode(inode); 2899 submitted = NULL; 2900 } 2901 unlock_page(page); 2902 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) && 2903 !F2FS_I(inode)->wb_task && allow_balance) 2904 f2fs_balance_fs(sbi, need_balance_fs); 2905 2906 if (unlikely(f2fs_cp_error(sbi))) { 2907 f2fs_submit_merged_write(sbi, DATA); 2908 f2fs_submit_merged_ipu_write(sbi, bio, NULL); 2909 submitted = NULL; 2910 } 2911 2912 if (submitted) 2913 *submitted = fio.submitted ? 1 : 0; 2914 2915 return 0; 2916 2917 redirty_out: 2918 redirty_page_for_writepage(wbc, page); 2919 /* 2920 * pageout() in MM traslates EAGAIN, so calls handle_write_error() 2921 * -> mapping_set_error() -> set_bit(AS_EIO, ...). 2922 * file_write_and_wait_range() will see EIO error, which is critical 2923 * to return value of fsync() followed by atomic_write failure to user. 2924 */ 2925 if (!err || wbc->for_reclaim) 2926 return AOP_WRITEPAGE_ACTIVATE; 2927 unlock_page(page); 2928 return err; 2929 } 2930 2931 static int f2fs_write_data_page(struct page *page, 2932 struct writeback_control *wbc) 2933 { 2934 #ifdef CONFIG_F2FS_FS_COMPRESSION 2935 struct inode *inode = page->mapping->host; 2936 2937 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 2938 goto out; 2939 2940 if (f2fs_compressed_file(inode)) { 2941 if (f2fs_is_compressed_cluster(inode, page->index)) { 2942 redirty_page_for_writepage(wbc, page); 2943 return AOP_WRITEPAGE_ACTIVATE; 2944 } 2945 } 2946 out: 2947 #endif 2948 2949 return f2fs_write_single_data_page(page, NULL, NULL, NULL, 2950 wbc, FS_DATA_IO, 0, true); 2951 } 2952 2953 /* 2954 * This function was copied from write_cche_pages from mm/page-writeback.c. 2955 * The major change is making write step of cold data page separately from 2956 * warm/hot data page. 2957 */ 2958 static int f2fs_write_cache_pages(struct address_space *mapping, 2959 struct writeback_control *wbc, 2960 enum iostat_type io_type) 2961 { 2962 int ret = 0; 2963 int done = 0, retry = 0; 2964 struct page *pages[F2FS_ONSTACK_PAGES]; 2965 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 2966 struct bio *bio = NULL; 2967 sector_t last_block; 2968 #ifdef CONFIG_F2FS_FS_COMPRESSION 2969 struct inode *inode = mapping->host; 2970 struct compress_ctx cc = { 2971 .inode = inode, 2972 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 2973 .cluster_size = F2FS_I(inode)->i_cluster_size, 2974 .cluster_idx = NULL_CLUSTER, 2975 .rpages = NULL, 2976 .nr_rpages = 0, 2977 .cpages = NULL, 2978 .valid_nr_cpages = 0, 2979 .rbuf = NULL, 2980 .cbuf = NULL, 2981 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size, 2982 .private = NULL, 2983 }; 2984 #endif 2985 int nr_pages; 2986 pgoff_t index; 2987 pgoff_t end; /* Inclusive */ 2988 pgoff_t done_index; 2989 int range_whole = 0; 2990 xa_mark_t tag; 2991 int nwritten = 0; 2992 int submitted = 0; 2993 int i; 2994 2995 if (get_dirty_pages(mapping->host) <= 2996 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 2997 set_inode_flag(mapping->host, FI_HOT_DATA); 2998 else 2999 clear_inode_flag(mapping->host, FI_HOT_DATA); 3000 3001 if (wbc->range_cyclic) { 3002 index = mapping->writeback_index; /* prev offset */ 3003 end = -1; 3004 } else { 3005 index = wbc->range_start >> PAGE_SHIFT; 3006 end = wbc->range_end >> PAGE_SHIFT; 3007 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 3008 range_whole = 1; 3009 } 3010 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 3011 tag = PAGECACHE_TAG_TOWRITE; 3012 else 3013 tag = PAGECACHE_TAG_DIRTY; 3014 retry: 3015 retry = 0; 3016 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 3017 tag_pages_for_writeback(mapping, index, end); 3018 done_index = index; 3019 while (!done && !retry && (index <= end)) { 3020 nr_pages = find_get_pages_range_tag(mapping, &index, end, 3021 tag, F2FS_ONSTACK_PAGES, pages); 3022 if (nr_pages == 0) 3023 break; 3024 3025 for (i = 0; i < nr_pages; i++) { 3026 struct page *page = pages[i]; 3027 bool need_readd; 3028 readd: 3029 need_readd = false; 3030 #ifdef CONFIG_F2FS_FS_COMPRESSION 3031 if (f2fs_compressed_file(inode)) { 3032 void *fsdata = NULL; 3033 struct page *pagep; 3034 int ret2; 3035 3036 ret = f2fs_init_compress_ctx(&cc); 3037 if (ret) { 3038 done = 1; 3039 break; 3040 } 3041 3042 if (!f2fs_cluster_can_merge_page(&cc, 3043 page->index)) { 3044 ret = f2fs_write_multi_pages(&cc, 3045 &submitted, wbc, io_type); 3046 if (!ret) 3047 need_readd = true; 3048 goto result; 3049 } 3050 3051 if (unlikely(f2fs_cp_error(sbi))) 3052 goto lock_page; 3053 3054 if (!f2fs_cluster_is_empty(&cc)) 3055 goto lock_page; 3056 3057 if (f2fs_all_cluster_page_ready(&cc, 3058 pages, i, nr_pages, true)) 3059 goto lock_page; 3060 3061 ret2 = f2fs_prepare_compress_overwrite( 3062 inode, &pagep, 3063 page->index, &fsdata); 3064 if (ret2 < 0) { 3065 ret = ret2; 3066 done = 1; 3067 break; 3068 } else if (ret2 && 3069 (!f2fs_compress_write_end(inode, 3070 fsdata, page->index, 1) || 3071 !f2fs_all_cluster_page_ready(&cc, 3072 pages, i, nr_pages, false))) { 3073 retry = 1; 3074 break; 3075 } 3076 } 3077 #endif 3078 /* give a priority to WB_SYNC threads */ 3079 if (atomic_read(&sbi->wb_sync_req[DATA]) && 3080 wbc->sync_mode == WB_SYNC_NONE) { 3081 done = 1; 3082 break; 3083 } 3084 #ifdef CONFIG_F2FS_FS_COMPRESSION 3085 lock_page: 3086 #endif 3087 done_index = page->index; 3088 retry_write: 3089 lock_page(page); 3090 3091 if (unlikely(page->mapping != mapping)) { 3092 continue_unlock: 3093 unlock_page(page); 3094 continue; 3095 } 3096 3097 if (!PageDirty(page)) { 3098 /* someone wrote it for us */ 3099 goto continue_unlock; 3100 } 3101 3102 if (PageWriteback(page)) { 3103 if (wbc->sync_mode != WB_SYNC_NONE) 3104 f2fs_wait_on_page_writeback(page, 3105 DATA, true, true); 3106 else 3107 goto continue_unlock; 3108 } 3109 3110 if (!clear_page_dirty_for_io(page)) 3111 goto continue_unlock; 3112 3113 #ifdef CONFIG_F2FS_FS_COMPRESSION 3114 if (f2fs_compressed_file(inode)) { 3115 get_page(page); 3116 f2fs_compress_ctx_add_page(&cc, page); 3117 continue; 3118 } 3119 #endif 3120 ret = f2fs_write_single_data_page(page, &submitted, 3121 &bio, &last_block, wbc, io_type, 3122 0, true); 3123 if (ret == AOP_WRITEPAGE_ACTIVATE) 3124 unlock_page(page); 3125 #ifdef CONFIG_F2FS_FS_COMPRESSION 3126 result: 3127 #endif 3128 nwritten += submitted; 3129 wbc->nr_to_write -= submitted; 3130 3131 if (unlikely(ret)) { 3132 /* 3133 * keep nr_to_write, since vfs uses this to 3134 * get # of written pages. 3135 */ 3136 if (ret == AOP_WRITEPAGE_ACTIVATE) { 3137 ret = 0; 3138 goto next; 3139 } else if (ret == -EAGAIN) { 3140 ret = 0; 3141 if (wbc->sync_mode == WB_SYNC_ALL) { 3142 f2fs_io_schedule_timeout( 3143 DEFAULT_IO_TIMEOUT); 3144 goto retry_write; 3145 } 3146 goto next; 3147 } 3148 done_index = page->index + 1; 3149 done = 1; 3150 break; 3151 } 3152 3153 if (wbc->nr_to_write <= 0 && 3154 wbc->sync_mode == WB_SYNC_NONE) { 3155 done = 1; 3156 break; 3157 } 3158 next: 3159 if (need_readd) 3160 goto readd; 3161 } 3162 release_pages(pages, nr_pages); 3163 cond_resched(); 3164 } 3165 #ifdef CONFIG_F2FS_FS_COMPRESSION 3166 /* flush remained pages in compress cluster */ 3167 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) { 3168 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type); 3169 nwritten += submitted; 3170 wbc->nr_to_write -= submitted; 3171 if (ret) { 3172 done = 1; 3173 retry = 0; 3174 } 3175 } 3176 if (f2fs_compressed_file(inode)) 3177 f2fs_destroy_compress_ctx(&cc, false); 3178 #endif 3179 if (retry) { 3180 index = 0; 3181 end = -1; 3182 goto retry; 3183 } 3184 if (wbc->range_cyclic && !done) 3185 done_index = 0; 3186 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 3187 mapping->writeback_index = done_index; 3188 3189 if (nwritten) 3190 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 3191 NULL, 0, DATA); 3192 /* submit cached bio of IPU write */ 3193 if (bio) 3194 f2fs_submit_merged_ipu_write(sbi, &bio, NULL); 3195 3196 return ret; 3197 } 3198 3199 static inline bool __should_serialize_io(struct inode *inode, 3200 struct writeback_control *wbc) 3201 { 3202 /* to avoid deadlock in path of data flush */ 3203 if (F2FS_I(inode)->wb_task) 3204 return false; 3205 3206 if (!S_ISREG(inode->i_mode)) 3207 return false; 3208 if (IS_NOQUOTA(inode)) 3209 return false; 3210 3211 if (f2fs_need_compress_data(inode)) 3212 return true; 3213 if (wbc->sync_mode != WB_SYNC_ALL) 3214 return true; 3215 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks) 3216 return true; 3217 return false; 3218 } 3219 3220 static int __f2fs_write_data_pages(struct address_space *mapping, 3221 struct writeback_control *wbc, 3222 enum iostat_type io_type) 3223 { 3224 struct inode *inode = mapping->host; 3225 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3226 struct blk_plug plug; 3227 int ret; 3228 bool locked = false; 3229 3230 /* deal with chardevs and other special file */ 3231 if (!mapping->a_ops->writepage) 3232 return 0; 3233 3234 /* skip writing if there is no dirty page in this inode */ 3235 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 3236 return 0; 3237 3238 /* during POR, we don't need to trigger writepage at all. */ 3239 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 3240 goto skip_write; 3241 3242 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) && 3243 wbc->sync_mode == WB_SYNC_NONE && 3244 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 3245 f2fs_available_free_memory(sbi, DIRTY_DENTS)) 3246 goto skip_write; 3247 3248 /* skip writing in file defragment preparing stage */ 3249 if (is_inode_flag_set(inode, FI_SKIP_WRITES)) 3250 goto skip_write; 3251 3252 trace_f2fs_writepages(mapping->host, wbc, DATA); 3253 3254 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 3255 if (wbc->sync_mode == WB_SYNC_ALL) 3256 atomic_inc(&sbi->wb_sync_req[DATA]); 3257 else if (atomic_read(&sbi->wb_sync_req[DATA])) { 3258 /* to avoid potential deadlock */ 3259 if (current->plug) 3260 blk_finish_plug(current->plug); 3261 goto skip_write; 3262 } 3263 3264 if (__should_serialize_io(inode, wbc)) { 3265 mutex_lock(&sbi->writepages); 3266 locked = true; 3267 } 3268 3269 blk_start_plug(&plug); 3270 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 3271 blk_finish_plug(&plug); 3272 3273 if (locked) 3274 mutex_unlock(&sbi->writepages); 3275 3276 if (wbc->sync_mode == WB_SYNC_ALL) 3277 atomic_dec(&sbi->wb_sync_req[DATA]); 3278 /* 3279 * if some pages were truncated, we cannot guarantee its mapping->host 3280 * to detect pending bios. 3281 */ 3282 3283 f2fs_remove_dirty_inode(inode); 3284 return ret; 3285 3286 skip_write: 3287 wbc->pages_skipped += get_dirty_pages(inode); 3288 trace_f2fs_writepages(mapping->host, wbc, DATA); 3289 return 0; 3290 } 3291 3292 static int f2fs_write_data_pages(struct address_space *mapping, 3293 struct writeback_control *wbc) 3294 { 3295 struct inode *inode = mapping->host; 3296 3297 return __f2fs_write_data_pages(mapping, wbc, 3298 F2FS_I(inode)->cp_task == current ? 3299 FS_CP_DATA_IO : FS_DATA_IO); 3300 } 3301 3302 void f2fs_write_failed(struct inode *inode, loff_t to) 3303 { 3304 loff_t i_size = i_size_read(inode); 3305 3306 if (IS_NOQUOTA(inode)) 3307 return; 3308 3309 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */ 3310 if (to > i_size && !f2fs_verity_in_progress(inode)) { 3311 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3312 filemap_invalidate_lock(inode->i_mapping); 3313 3314 truncate_pagecache(inode, i_size); 3315 f2fs_truncate_blocks(inode, i_size, true); 3316 3317 filemap_invalidate_unlock(inode->i_mapping); 3318 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3319 } 3320 } 3321 3322 static int prepare_write_begin(struct f2fs_sb_info *sbi, 3323 struct page *page, loff_t pos, unsigned len, 3324 block_t *blk_addr, bool *node_changed) 3325 { 3326 struct inode *inode = page->mapping->host; 3327 pgoff_t index = page->index; 3328 struct dnode_of_data dn; 3329 struct page *ipage; 3330 bool locked = false; 3331 struct extent_info ei = {0, }; 3332 int err = 0; 3333 int flag; 3334 3335 /* 3336 * If a whole page is being written and we already preallocated all the 3337 * blocks, then there is no need to get a block address now. 3338 */ 3339 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL)) 3340 return 0; 3341 3342 /* f2fs_lock_op avoids race between write CP and convert_inline_page */ 3343 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode)) 3344 flag = F2FS_GET_BLOCK_DEFAULT; 3345 else 3346 flag = F2FS_GET_BLOCK_PRE_AIO; 3347 3348 if (f2fs_has_inline_data(inode) || 3349 (pos & PAGE_MASK) >= i_size_read(inode)) { 3350 f2fs_do_map_lock(sbi, flag, true); 3351 locked = true; 3352 } 3353 3354 restart: 3355 /* check inline_data */ 3356 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3357 if (IS_ERR(ipage)) { 3358 err = PTR_ERR(ipage); 3359 goto unlock_out; 3360 } 3361 3362 set_new_dnode(&dn, inode, ipage, ipage, 0); 3363 3364 if (f2fs_has_inline_data(inode)) { 3365 if (pos + len <= MAX_INLINE_DATA(inode)) { 3366 f2fs_do_read_inline_data(page, ipage); 3367 set_inode_flag(inode, FI_DATA_EXIST); 3368 if (inode->i_nlink) 3369 set_page_private_inline(ipage); 3370 } else { 3371 err = f2fs_convert_inline_page(&dn, page); 3372 if (err) 3373 goto out; 3374 if (dn.data_blkaddr == NULL_ADDR) 3375 err = f2fs_get_block(&dn, index); 3376 } 3377 } else if (locked) { 3378 err = f2fs_get_block(&dn, index); 3379 } else { 3380 if (f2fs_lookup_read_extent_cache(inode, index, &ei)) { 3381 dn.data_blkaddr = ei.blk + index - ei.fofs; 3382 } else { 3383 /* hole case */ 3384 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3385 if (err || dn.data_blkaddr == NULL_ADDR) { 3386 f2fs_put_dnode(&dn); 3387 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 3388 true); 3389 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO); 3390 locked = true; 3391 goto restart; 3392 } 3393 } 3394 } 3395 3396 /* convert_inline_page can make node_changed */ 3397 *blk_addr = dn.data_blkaddr; 3398 *node_changed = dn.node_changed; 3399 out: 3400 f2fs_put_dnode(&dn); 3401 unlock_out: 3402 if (locked) 3403 f2fs_do_map_lock(sbi, flag, false); 3404 return err; 3405 } 3406 3407 static int __find_data_block(struct inode *inode, pgoff_t index, 3408 block_t *blk_addr) 3409 { 3410 struct dnode_of_data dn; 3411 struct page *ipage; 3412 struct extent_info ei = {0, }; 3413 int err = 0; 3414 3415 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino); 3416 if (IS_ERR(ipage)) 3417 return PTR_ERR(ipage); 3418 3419 set_new_dnode(&dn, inode, ipage, ipage, 0); 3420 3421 if (f2fs_lookup_read_extent_cache(inode, index, &ei)) { 3422 dn.data_blkaddr = ei.blk + index - ei.fofs; 3423 } else { 3424 /* hole case */ 3425 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 3426 if (err) { 3427 dn.data_blkaddr = NULL_ADDR; 3428 err = 0; 3429 } 3430 } 3431 *blk_addr = dn.data_blkaddr; 3432 f2fs_put_dnode(&dn); 3433 return err; 3434 } 3435 3436 static int __reserve_data_block(struct inode *inode, pgoff_t index, 3437 block_t *blk_addr, bool *node_changed) 3438 { 3439 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3440 struct dnode_of_data dn; 3441 struct page *ipage; 3442 int err = 0; 3443 3444 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true); 3445 3446 ipage = f2fs_get_node_page(sbi, inode->i_ino); 3447 if (IS_ERR(ipage)) { 3448 err = PTR_ERR(ipage); 3449 goto unlock_out; 3450 } 3451 set_new_dnode(&dn, inode, ipage, ipage, 0); 3452 3453 err = f2fs_get_block(&dn, index); 3454 3455 *blk_addr = dn.data_blkaddr; 3456 *node_changed = dn.node_changed; 3457 f2fs_put_dnode(&dn); 3458 3459 unlock_out: 3460 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false); 3461 return err; 3462 } 3463 3464 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi, 3465 struct page *page, loff_t pos, unsigned int len, 3466 block_t *blk_addr, bool *node_changed) 3467 { 3468 struct inode *inode = page->mapping->host; 3469 struct inode *cow_inode = F2FS_I(inode)->cow_inode; 3470 pgoff_t index = page->index; 3471 int err = 0; 3472 block_t ori_blk_addr = NULL_ADDR; 3473 3474 /* If pos is beyond the end of file, reserve a new block in COW inode */ 3475 if ((pos & PAGE_MASK) >= i_size_read(inode)) 3476 goto reserve_block; 3477 3478 /* Look for the block in COW inode first */ 3479 err = __find_data_block(cow_inode, index, blk_addr); 3480 if (err) 3481 return err; 3482 else if (*blk_addr != NULL_ADDR) 3483 return 0; 3484 3485 if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE)) 3486 goto reserve_block; 3487 3488 /* Look for the block in the original inode */ 3489 err = __find_data_block(inode, index, &ori_blk_addr); 3490 if (err) 3491 return err; 3492 3493 reserve_block: 3494 /* Finally, we should reserve a new block in COW inode for the update */ 3495 err = __reserve_data_block(cow_inode, index, blk_addr, node_changed); 3496 if (err) 3497 return err; 3498 inc_atomic_write_cnt(inode); 3499 3500 if (ori_blk_addr != NULL_ADDR) 3501 *blk_addr = ori_blk_addr; 3502 return 0; 3503 } 3504 3505 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 3506 loff_t pos, unsigned len, struct page **pagep, void **fsdata) 3507 { 3508 struct inode *inode = mapping->host; 3509 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3510 struct page *page = NULL; 3511 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 3512 bool need_balance = false; 3513 block_t blkaddr = NULL_ADDR; 3514 int err = 0; 3515 3516 trace_f2fs_write_begin(inode, pos, len); 3517 3518 if (!f2fs_is_checkpoint_ready(sbi)) { 3519 err = -ENOSPC; 3520 goto fail; 3521 } 3522 3523 /* 3524 * We should check this at this moment to avoid deadlock on inode page 3525 * and #0 page. The locking rule for inline_data conversion should be: 3526 * lock_page(page #0) -> lock_page(inode_page) 3527 */ 3528 if (index != 0) { 3529 err = f2fs_convert_inline_inode(inode); 3530 if (err) 3531 goto fail; 3532 } 3533 3534 #ifdef CONFIG_F2FS_FS_COMPRESSION 3535 if (f2fs_compressed_file(inode)) { 3536 int ret; 3537 3538 *fsdata = NULL; 3539 3540 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode))) 3541 goto repeat; 3542 3543 ret = f2fs_prepare_compress_overwrite(inode, pagep, 3544 index, fsdata); 3545 if (ret < 0) { 3546 err = ret; 3547 goto fail; 3548 } else if (ret) { 3549 return 0; 3550 } 3551 } 3552 #endif 3553 3554 repeat: 3555 /* 3556 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 3557 * wait_for_stable_page. Will wait that below with our IO control. 3558 */ 3559 page = f2fs_pagecache_get_page(mapping, index, 3560 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 3561 if (!page) { 3562 err = -ENOMEM; 3563 goto fail; 3564 } 3565 3566 /* TODO: cluster can be compressed due to race with .writepage */ 3567 3568 *pagep = page; 3569 3570 if (f2fs_is_atomic_file(inode)) 3571 err = prepare_atomic_write_begin(sbi, page, pos, len, 3572 &blkaddr, &need_balance); 3573 else 3574 err = prepare_write_begin(sbi, page, pos, len, 3575 &blkaddr, &need_balance); 3576 if (err) 3577 goto fail; 3578 3579 if (need_balance && !IS_NOQUOTA(inode) && 3580 has_not_enough_free_secs(sbi, 0, 0)) { 3581 unlock_page(page); 3582 f2fs_balance_fs(sbi, true); 3583 lock_page(page); 3584 if (page->mapping != mapping) { 3585 /* The page got truncated from under us */ 3586 f2fs_put_page(page, 1); 3587 goto repeat; 3588 } 3589 } 3590 3591 f2fs_wait_on_page_writeback(page, DATA, false, true); 3592 3593 if (len == PAGE_SIZE || PageUptodate(page)) 3594 return 0; 3595 3596 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) && 3597 !f2fs_verity_in_progress(inode)) { 3598 zero_user_segment(page, len, PAGE_SIZE); 3599 return 0; 3600 } 3601 3602 if (blkaddr == NEW_ADDR) { 3603 zero_user_segment(page, 0, PAGE_SIZE); 3604 SetPageUptodate(page); 3605 } else { 3606 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 3607 DATA_GENERIC_ENHANCE_READ)) { 3608 err = -EFSCORRUPTED; 3609 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 3610 goto fail; 3611 } 3612 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true); 3613 if (err) 3614 goto fail; 3615 3616 lock_page(page); 3617 if (unlikely(page->mapping != mapping)) { 3618 f2fs_put_page(page, 1); 3619 goto repeat; 3620 } 3621 if (unlikely(!PageUptodate(page))) { 3622 err = -EIO; 3623 goto fail; 3624 } 3625 } 3626 return 0; 3627 3628 fail: 3629 f2fs_put_page(page, 1); 3630 f2fs_write_failed(inode, pos + len); 3631 return err; 3632 } 3633 3634 static int f2fs_write_end(struct file *file, 3635 struct address_space *mapping, 3636 loff_t pos, unsigned len, unsigned copied, 3637 struct page *page, void *fsdata) 3638 { 3639 struct inode *inode = page->mapping->host; 3640 3641 trace_f2fs_write_end(inode, pos, len, copied); 3642 3643 /* 3644 * This should be come from len == PAGE_SIZE, and we expect copied 3645 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 3646 * let generic_perform_write() try to copy data again through copied=0. 3647 */ 3648 if (!PageUptodate(page)) { 3649 if (unlikely(copied != len)) 3650 copied = 0; 3651 else 3652 SetPageUptodate(page); 3653 } 3654 3655 #ifdef CONFIG_F2FS_FS_COMPRESSION 3656 /* overwrite compressed file */ 3657 if (f2fs_compressed_file(inode) && fsdata) { 3658 f2fs_compress_write_end(inode, fsdata, page->index, copied); 3659 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3660 3661 if (pos + copied > i_size_read(inode) && 3662 !f2fs_verity_in_progress(inode)) 3663 f2fs_i_size_write(inode, pos + copied); 3664 return copied; 3665 } 3666 #endif 3667 3668 if (!copied) 3669 goto unlock_out; 3670 3671 set_page_dirty(page); 3672 3673 if (pos + copied > i_size_read(inode) && 3674 !f2fs_verity_in_progress(inode)) { 3675 f2fs_i_size_write(inode, pos + copied); 3676 if (f2fs_is_atomic_file(inode)) 3677 f2fs_i_size_write(F2FS_I(inode)->cow_inode, 3678 pos + copied); 3679 } 3680 unlock_out: 3681 f2fs_put_page(page, 1); 3682 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3683 return copied; 3684 } 3685 3686 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length) 3687 { 3688 struct inode *inode = folio->mapping->host; 3689 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3690 3691 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 3692 (offset || length != folio_size(folio))) 3693 return; 3694 3695 if (folio_test_dirty(folio)) { 3696 if (inode->i_ino == F2FS_META_INO(sbi)) { 3697 dec_page_count(sbi, F2FS_DIRTY_META); 3698 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 3699 dec_page_count(sbi, F2FS_DIRTY_NODES); 3700 } else { 3701 inode_dec_dirty_pages(inode); 3702 f2fs_remove_dirty_inode(inode); 3703 } 3704 } 3705 3706 clear_page_private_reference(&folio->page); 3707 clear_page_private_gcing(&folio->page); 3708 3709 if (test_opt(sbi, COMPRESS_CACHE) && 3710 inode->i_ino == F2FS_COMPRESS_INO(sbi)) 3711 clear_page_private_data(&folio->page); 3712 3713 folio_detach_private(folio); 3714 } 3715 3716 bool f2fs_release_folio(struct folio *folio, gfp_t wait) 3717 { 3718 struct f2fs_sb_info *sbi; 3719 3720 /* If this is dirty folio, keep private data */ 3721 if (folio_test_dirty(folio)) 3722 return false; 3723 3724 sbi = F2FS_M_SB(folio->mapping); 3725 if (test_opt(sbi, COMPRESS_CACHE)) { 3726 struct inode *inode = folio->mapping->host; 3727 3728 if (inode->i_ino == F2FS_COMPRESS_INO(sbi)) 3729 clear_page_private_data(&folio->page); 3730 } 3731 3732 clear_page_private_reference(&folio->page); 3733 clear_page_private_gcing(&folio->page); 3734 3735 folio_detach_private(folio); 3736 return true; 3737 } 3738 3739 static bool f2fs_dirty_data_folio(struct address_space *mapping, 3740 struct folio *folio) 3741 { 3742 struct inode *inode = mapping->host; 3743 3744 trace_f2fs_set_page_dirty(&folio->page, DATA); 3745 3746 if (!folio_test_uptodate(folio)) 3747 folio_mark_uptodate(folio); 3748 BUG_ON(folio_test_swapcache(folio)); 3749 3750 if (filemap_dirty_folio(mapping, folio)) { 3751 f2fs_update_dirty_folio(inode, folio); 3752 return true; 3753 } 3754 return false; 3755 } 3756 3757 3758 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block) 3759 { 3760 #ifdef CONFIG_F2FS_FS_COMPRESSION 3761 struct dnode_of_data dn; 3762 sector_t start_idx, blknr = 0; 3763 int ret; 3764 3765 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size); 3766 3767 set_new_dnode(&dn, inode, NULL, NULL, 0); 3768 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 3769 if (ret) 3770 return 0; 3771 3772 if (dn.data_blkaddr != COMPRESS_ADDR) { 3773 dn.ofs_in_node += block - start_idx; 3774 blknr = f2fs_data_blkaddr(&dn); 3775 if (!__is_valid_data_blkaddr(blknr)) 3776 blknr = 0; 3777 } 3778 3779 f2fs_put_dnode(&dn); 3780 return blknr; 3781 #else 3782 return 0; 3783 #endif 3784 } 3785 3786 3787 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 3788 { 3789 struct inode *inode = mapping->host; 3790 sector_t blknr = 0; 3791 3792 if (f2fs_has_inline_data(inode)) 3793 goto out; 3794 3795 /* make sure allocating whole blocks */ 3796 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 3797 filemap_write_and_wait(mapping); 3798 3799 /* Block number less than F2FS MAX BLOCKS */ 3800 if (unlikely(block >= max_file_blocks(inode))) 3801 goto out; 3802 3803 if (f2fs_compressed_file(inode)) { 3804 blknr = f2fs_bmap_compress(inode, block); 3805 } else { 3806 struct f2fs_map_blocks map; 3807 3808 memset(&map, 0, sizeof(map)); 3809 map.m_lblk = block; 3810 map.m_len = 1; 3811 map.m_next_pgofs = NULL; 3812 map.m_seg_type = NO_CHECK_TYPE; 3813 3814 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP)) 3815 blknr = map.m_pblk; 3816 } 3817 out: 3818 trace_f2fs_bmap(inode, block, blknr); 3819 return blknr; 3820 } 3821 3822 #ifdef CONFIG_SWAP 3823 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk, 3824 unsigned int blkcnt) 3825 { 3826 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3827 unsigned int blkofs; 3828 unsigned int blk_per_sec = BLKS_PER_SEC(sbi); 3829 unsigned int secidx = start_blk / blk_per_sec; 3830 unsigned int end_sec = secidx + blkcnt / blk_per_sec; 3831 int ret = 0; 3832 3833 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3834 filemap_invalidate_lock(inode->i_mapping); 3835 3836 set_inode_flag(inode, FI_ALIGNED_WRITE); 3837 set_inode_flag(inode, FI_OPU_WRITE); 3838 3839 for (; secidx < end_sec; secidx++) { 3840 f2fs_down_write(&sbi->pin_sem); 3841 3842 f2fs_lock_op(sbi); 3843 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false); 3844 f2fs_unlock_op(sbi); 3845 3846 set_inode_flag(inode, FI_SKIP_WRITES); 3847 3848 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) { 3849 struct page *page; 3850 unsigned int blkidx = secidx * blk_per_sec + blkofs; 3851 3852 page = f2fs_get_lock_data_page(inode, blkidx, true); 3853 if (IS_ERR(page)) { 3854 f2fs_up_write(&sbi->pin_sem); 3855 ret = PTR_ERR(page); 3856 goto done; 3857 } 3858 3859 set_page_dirty(page); 3860 f2fs_put_page(page, 1); 3861 } 3862 3863 clear_inode_flag(inode, FI_SKIP_WRITES); 3864 3865 ret = filemap_fdatawrite(inode->i_mapping); 3866 3867 f2fs_up_write(&sbi->pin_sem); 3868 3869 if (ret) 3870 break; 3871 } 3872 3873 done: 3874 clear_inode_flag(inode, FI_SKIP_WRITES); 3875 clear_inode_flag(inode, FI_OPU_WRITE); 3876 clear_inode_flag(inode, FI_ALIGNED_WRITE); 3877 3878 filemap_invalidate_unlock(inode->i_mapping); 3879 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 3880 3881 return ret; 3882 } 3883 3884 static int check_swap_activate(struct swap_info_struct *sis, 3885 struct file *swap_file, sector_t *span) 3886 { 3887 struct address_space *mapping = swap_file->f_mapping; 3888 struct inode *inode = mapping->host; 3889 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3890 sector_t cur_lblock; 3891 sector_t last_lblock; 3892 sector_t pblock; 3893 sector_t lowest_pblock = -1; 3894 sector_t highest_pblock = 0; 3895 int nr_extents = 0; 3896 unsigned long nr_pblocks; 3897 unsigned int blks_per_sec = BLKS_PER_SEC(sbi); 3898 unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1; 3899 unsigned int not_aligned = 0; 3900 int ret = 0; 3901 3902 /* 3903 * Map all the blocks into the extent list. This code doesn't try 3904 * to be very smart. 3905 */ 3906 cur_lblock = 0; 3907 last_lblock = bytes_to_blks(inode, i_size_read(inode)); 3908 3909 while (cur_lblock < last_lblock && cur_lblock < sis->max) { 3910 struct f2fs_map_blocks map; 3911 retry: 3912 cond_resched(); 3913 3914 memset(&map, 0, sizeof(map)); 3915 map.m_lblk = cur_lblock; 3916 map.m_len = last_lblock - cur_lblock; 3917 map.m_next_pgofs = NULL; 3918 map.m_next_extent = NULL; 3919 map.m_seg_type = NO_CHECK_TYPE; 3920 map.m_may_create = false; 3921 3922 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP); 3923 if (ret) 3924 goto out; 3925 3926 /* hole */ 3927 if (!(map.m_flags & F2FS_MAP_FLAGS)) { 3928 f2fs_err(sbi, "Swapfile has holes"); 3929 ret = -EINVAL; 3930 goto out; 3931 } 3932 3933 pblock = map.m_pblk; 3934 nr_pblocks = map.m_len; 3935 3936 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask || 3937 nr_pblocks & sec_blks_mask) { 3938 not_aligned++; 3939 3940 nr_pblocks = roundup(nr_pblocks, blks_per_sec); 3941 if (cur_lblock + nr_pblocks > sis->max) 3942 nr_pblocks -= blks_per_sec; 3943 3944 if (!nr_pblocks) { 3945 /* this extent is last one */ 3946 nr_pblocks = map.m_len; 3947 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section"); 3948 goto next; 3949 } 3950 3951 ret = f2fs_migrate_blocks(inode, cur_lblock, 3952 nr_pblocks); 3953 if (ret) 3954 goto out; 3955 goto retry; 3956 } 3957 next: 3958 if (cur_lblock + nr_pblocks >= sis->max) 3959 nr_pblocks = sis->max - cur_lblock; 3960 3961 if (cur_lblock) { /* exclude the header page */ 3962 if (pblock < lowest_pblock) 3963 lowest_pblock = pblock; 3964 if (pblock + nr_pblocks - 1 > highest_pblock) 3965 highest_pblock = pblock + nr_pblocks - 1; 3966 } 3967 3968 /* 3969 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks 3970 */ 3971 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock); 3972 if (ret < 0) 3973 goto out; 3974 nr_extents += ret; 3975 cur_lblock += nr_pblocks; 3976 } 3977 ret = nr_extents; 3978 *span = 1 + highest_pblock - lowest_pblock; 3979 if (cur_lblock == 0) 3980 cur_lblock = 1; /* force Empty message */ 3981 sis->max = cur_lblock; 3982 sis->pages = cur_lblock - 1; 3983 sis->highest_bit = cur_lblock - 1; 3984 out: 3985 if (not_aligned) 3986 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)", 3987 not_aligned, blks_per_sec * F2FS_BLKSIZE); 3988 return ret; 3989 } 3990 3991 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 3992 sector_t *span) 3993 { 3994 struct inode *inode = file_inode(file); 3995 int ret; 3996 3997 if (!S_ISREG(inode->i_mode)) 3998 return -EINVAL; 3999 4000 if (f2fs_readonly(F2FS_I_SB(inode)->sb)) 4001 return -EROFS; 4002 4003 if (f2fs_lfs_mode(F2FS_I_SB(inode))) { 4004 f2fs_err(F2FS_I_SB(inode), 4005 "Swapfile not supported in LFS mode"); 4006 return -EINVAL; 4007 } 4008 4009 ret = f2fs_convert_inline_inode(inode); 4010 if (ret) 4011 return ret; 4012 4013 if (!f2fs_disable_compressed_file(inode)) 4014 return -EINVAL; 4015 4016 f2fs_precache_extents(inode); 4017 4018 ret = check_swap_activate(sis, file, span); 4019 if (ret < 0) 4020 return ret; 4021 4022 stat_inc_swapfile_inode(inode); 4023 set_inode_flag(inode, FI_PIN_FILE); 4024 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 4025 return ret; 4026 } 4027 4028 static void f2fs_swap_deactivate(struct file *file) 4029 { 4030 struct inode *inode = file_inode(file); 4031 4032 stat_dec_swapfile_inode(inode); 4033 clear_inode_flag(inode, FI_PIN_FILE); 4034 } 4035 #else 4036 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, 4037 sector_t *span) 4038 { 4039 return -EOPNOTSUPP; 4040 } 4041 4042 static void f2fs_swap_deactivate(struct file *file) 4043 { 4044 } 4045 #endif 4046 4047 const struct address_space_operations f2fs_dblock_aops = { 4048 .read_folio = f2fs_read_data_folio, 4049 .readahead = f2fs_readahead, 4050 .writepage = f2fs_write_data_page, 4051 .writepages = f2fs_write_data_pages, 4052 .write_begin = f2fs_write_begin, 4053 .write_end = f2fs_write_end, 4054 .dirty_folio = f2fs_dirty_data_folio, 4055 .migrate_folio = filemap_migrate_folio, 4056 .invalidate_folio = f2fs_invalidate_folio, 4057 .release_folio = f2fs_release_folio, 4058 .direct_IO = noop_direct_IO, 4059 .bmap = f2fs_bmap, 4060 .swap_activate = f2fs_swap_activate, 4061 .swap_deactivate = f2fs_swap_deactivate, 4062 }; 4063 4064 void f2fs_clear_page_cache_dirty_tag(struct page *page) 4065 { 4066 struct address_space *mapping = page_mapping(page); 4067 unsigned long flags; 4068 4069 xa_lock_irqsave(&mapping->i_pages, flags); 4070 __xa_clear_mark(&mapping->i_pages, page_index(page), 4071 PAGECACHE_TAG_DIRTY); 4072 xa_unlock_irqrestore(&mapping->i_pages, flags); 4073 } 4074 4075 int __init f2fs_init_post_read_processing(void) 4076 { 4077 bio_post_read_ctx_cache = 4078 kmem_cache_create("f2fs_bio_post_read_ctx", 4079 sizeof(struct bio_post_read_ctx), 0, 0, NULL); 4080 if (!bio_post_read_ctx_cache) 4081 goto fail; 4082 bio_post_read_ctx_pool = 4083 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 4084 bio_post_read_ctx_cache); 4085 if (!bio_post_read_ctx_pool) 4086 goto fail_free_cache; 4087 return 0; 4088 4089 fail_free_cache: 4090 kmem_cache_destroy(bio_post_read_ctx_cache); 4091 fail: 4092 return -ENOMEM; 4093 } 4094 4095 void f2fs_destroy_post_read_processing(void) 4096 { 4097 mempool_destroy(bio_post_read_ctx_pool); 4098 kmem_cache_destroy(bio_post_read_ctx_cache); 4099 } 4100 4101 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) 4102 { 4103 if (!f2fs_sb_has_encrypt(sbi) && 4104 !f2fs_sb_has_verity(sbi) && 4105 !f2fs_sb_has_compression(sbi)) 4106 return 0; 4107 4108 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq", 4109 WQ_UNBOUND | WQ_HIGHPRI, 4110 num_online_cpus()); 4111 return sbi->post_read_wq ? 0 : -ENOMEM; 4112 } 4113 4114 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) 4115 { 4116 if (sbi->post_read_wq) 4117 destroy_workqueue(sbi->post_read_wq); 4118 } 4119 4120 int __init f2fs_init_bio_entry_cache(void) 4121 { 4122 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab", 4123 sizeof(struct bio_entry)); 4124 return bio_entry_slab ? 0 : -ENOMEM; 4125 } 4126 4127 void f2fs_destroy_bio_entry_cache(void) 4128 { 4129 kmem_cache_destroy(bio_entry_slab); 4130 } 4131 4132 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 4133 unsigned int flags, struct iomap *iomap, 4134 struct iomap *srcmap) 4135 { 4136 struct f2fs_map_blocks map = {}; 4137 pgoff_t next_pgofs = 0; 4138 int err; 4139 4140 map.m_lblk = bytes_to_blks(inode, offset); 4141 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1; 4142 map.m_next_pgofs = &next_pgofs; 4143 map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint); 4144 if (flags & IOMAP_WRITE) 4145 map.m_may_create = true; 4146 4147 err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE, 4148 F2FS_GET_BLOCK_DIO); 4149 if (err) 4150 return err; 4151 4152 iomap->offset = blks_to_bytes(inode, map.m_lblk); 4153 4154 /* 4155 * When inline encryption is enabled, sometimes I/O to an encrypted file 4156 * has to be broken up to guarantee DUN contiguity. Handle this by 4157 * limiting the length of the mapping returned. 4158 */ 4159 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len); 4160 4161 /* 4162 * We should never see delalloc or compressed extents here based on 4163 * prior flushing and checks. 4164 */ 4165 if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR)) 4166 return -EINVAL; 4167 if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR)) 4168 return -EINVAL; 4169 4170 if (map.m_pblk != NULL_ADDR) { 4171 iomap->length = blks_to_bytes(inode, map.m_len); 4172 iomap->type = IOMAP_MAPPED; 4173 iomap->flags |= IOMAP_F_MERGED; 4174 iomap->bdev = map.m_bdev; 4175 iomap->addr = blks_to_bytes(inode, map.m_pblk); 4176 } else { 4177 if (flags & IOMAP_WRITE) 4178 return -ENOTBLK; 4179 iomap->length = blks_to_bytes(inode, next_pgofs) - 4180 iomap->offset; 4181 iomap->type = IOMAP_HOLE; 4182 iomap->addr = IOMAP_NULL_ADDR; 4183 } 4184 4185 if (map.m_flags & F2FS_MAP_NEW) 4186 iomap->flags |= IOMAP_F_NEW; 4187 if ((inode->i_state & I_DIRTY_DATASYNC) || 4188 offset + length > i_size_read(inode)) 4189 iomap->flags |= IOMAP_F_DIRTY; 4190 4191 return 0; 4192 } 4193 4194 const struct iomap_ops f2fs_iomap_ops = { 4195 .iomap_begin = f2fs_iomap_begin, 4196 }; 4197