1 /* 2 * fs/f2fs/data.c 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/fs.h> 12 #include <linux/f2fs_fs.h> 13 #include <linux/buffer_head.h> 14 #include <linux/mpage.h> 15 #include <linux/writeback.h> 16 #include <linux/backing-dev.h> 17 #include <linux/pagevec.h> 18 #include <linux/blkdev.h> 19 #include <linux/bio.h> 20 #include <linux/prefetch.h> 21 #include <linux/uio.h> 22 #include <linux/mm.h> 23 #include <linux/memcontrol.h> 24 #include <linux/cleancache.h> 25 #include <linux/sched/signal.h> 26 27 #include "f2fs.h" 28 #include "node.h" 29 #include "segment.h" 30 #include "trace.h" 31 #include <trace/events/f2fs.h> 32 33 static bool __is_cp_guaranteed(struct page *page) 34 { 35 struct address_space *mapping = page->mapping; 36 struct inode *inode; 37 struct f2fs_sb_info *sbi; 38 39 if (!mapping) 40 return false; 41 42 inode = mapping->host; 43 sbi = F2FS_I_SB(inode); 44 45 if (inode->i_ino == F2FS_META_INO(sbi) || 46 inode->i_ino == F2FS_NODE_INO(sbi) || 47 S_ISDIR(inode->i_mode) || 48 is_cold_data(page)) 49 return true; 50 return false; 51 } 52 53 static void f2fs_read_end_io(struct bio *bio) 54 { 55 struct bio_vec *bvec; 56 int i; 57 58 #ifdef CONFIG_F2FS_FAULT_INJECTION 59 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) { 60 f2fs_show_injection_info(FAULT_IO); 61 bio->bi_status = BLK_STS_IOERR; 62 } 63 #endif 64 65 if (f2fs_bio_encrypted(bio)) { 66 if (bio->bi_status) { 67 fscrypt_release_ctx(bio->bi_private); 68 } else { 69 fscrypt_decrypt_bio_pages(bio->bi_private, bio); 70 return; 71 } 72 } 73 74 bio_for_each_segment_all(bvec, bio, i) { 75 struct page *page = bvec->bv_page; 76 77 if (!bio->bi_status) { 78 if (!PageUptodate(page)) 79 SetPageUptodate(page); 80 } else { 81 ClearPageUptodate(page); 82 SetPageError(page); 83 } 84 unlock_page(page); 85 } 86 bio_put(bio); 87 } 88 89 static void f2fs_write_end_io(struct bio *bio) 90 { 91 struct f2fs_sb_info *sbi = bio->bi_private; 92 struct bio_vec *bvec; 93 int i; 94 95 bio_for_each_segment_all(bvec, bio, i) { 96 struct page *page = bvec->bv_page; 97 enum count_type type = WB_DATA_TYPE(page); 98 99 if (IS_DUMMY_WRITTEN_PAGE(page)) { 100 set_page_private(page, (unsigned long)NULL); 101 ClearPagePrivate(page); 102 unlock_page(page); 103 mempool_free(page, sbi->write_io_dummy); 104 105 if (unlikely(bio->bi_status)) 106 f2fs_stop_checkpoint(sbi, true); 107 continue; 108 } 109 110 fscrypt_pullback_bio_page(&page, true); 111 112 if (unlikely(bio->bi_status)) { 113 mapping_set_error(page->mapping, -EIO); 114 f2fs_stop_checkpoint(sbi, true); 115 } 116 dec_page_count(sbi, type); 117 clear_cold_data(page); 118 end_page_writeback(page); 119 } 120 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 121 wq_has_sleeper(&sbi->cp_wait)) 122 wake_up(&sbi->cp_wait); 123 124 bio_put(bio); 125 } 126 127 /* 128 * Return true, if pre_bio's bdev is same as its target device. 129 */ 130 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 131 block_t blk_addr, struct bio *bio) 132 { 133 struct block_device *bdev = sbi->sb->s_bdev; 134 int i; 135 136 for (i = 0; i < sbi->s_ndevs; i++) { 137 if (FDEV(i).start_blk <= blk_addr && 138 FDEV(i).end_blk >= blk_addr) { 139 blk_addr -= FDEV(i).start_blk; 140 bdev = FDEV(i).bdev; 141 break; 142 } 143 } 144 if (bio) { 145 bio_set_dev(bio, bdev); 146 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); 147 } 148 return bdev; 149 } 150 151 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 152 { 153 int i; 154 155 for (i = 0; i < sbi->s_ndevs; i++) 156 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 157 return i; 158 return 0; 159 } 160 161 static bool __same_bdev(struct f2fs_sb_info *sbi, 162 block_t blk_addr, struct bio *bio) 163 { 164 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL); 165 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno; 166 } 167 168 /* 169 * Low-level block read/write IO operations. 170 */ 171 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr, 172 int npages, bool is_read) 173 { 174 struct bio *bio; 175 176 bio = f2fs_bio_alloc(sbi, npages, true); 177 178 f2fs_target_device(sbi, blk_addr, bio); 179 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io; 180 bio->bi_private = is_read ? NULL : sbi; 181 182 return bio; 183 } 184 185 static inline void __submit_bio(struct f2fs_sb_info *sbi, 186 struct bio *bio, enum page_type type) 187 { 188 if (!is_read_io(bio_op(bio))) { 189 unsigned int start; 190 191 if (f2fs_sb_mounted_blkzoned(sbi->sb) && 192 current->plug && (type == DATA || type == NODE)) 193 blk_finish_plug(current->plug); 194 195 if (type != DATA && type != NODE) 196 goto submit_io; 197 198 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS; 199 start %= F2FS_IO_SIZE(sbi); 200 201 if (start == 0) 202 goto submit_io; 203 204 /* fill dummy pages */ 205 for (; start < F2FS_IO_SIZE(sbi); start++) { 206 struct page *page = 207 mempool_alloc(sbi->write_io_dummy, 208 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL); 209 f2fs_bug_on(sbi, !page); 210 211 SetPagePrivate(page); 212 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE); 213 lock_page(page); 214 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 215 f2fs_bug_on(sbi, 1); 216 } 217 /* 218 * In the NODE case, we lose next block address chain. So, we 219 * need to do checkpoint in f2fs_sync_file. 220 */ 221 if (type == NODE) 222 set_sbi_flag(sbi, SBI_NEED_CP); 223 } 224 submit_io: 225 if (is_read_io(bio_op(bio))) 226 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 227 else 228 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 229 submit_bio(bio); 230 } 231 232 static void __submit_merged_bio(struct f2fs_bio_info *io) 233 { 234 struct f2fs_io_info *fio = &io->fio; 235 236 if (!io->bio) 237 return; 238 239 bio_set_op_attrs(io->bio, fio->op, fio->op_flags); 240 241 if (is_read_io(fio->op)) 242 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 243 else 244 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 245 246 __submit_bio(io->sbi, io->bio, fio->type); 247 io->bio = NULL; 248 } 249 250 static bool __has_merged_page(struct f2fs_bio_info *io, 251 struct inode *inode, nid_t ino, pgoff_t idx) 252 { 253 struct bio_vec *bvec; 254 struct page *target; 255 int i; 256 257 if (!io->bio) 258 return false; 259 260 if (!inode && !ino) 261 return true; 262 263 bio_for_each_segment_all(bvec, io->bio, i) { 264 265 if (bvec->bv_page->mapping) 266 target = bvec->bv_page; 267 else 268 target = fscrypt_control_page(bvec->bv_page); 269 270 if (idx != target->index) 271 continue; 272 273 if (inode && inode == target->mapping->host) 274 return true; 275 if (ino && ino == ino_of_node(target)) 276 return true; 277 } 278 279 return false; 280 } 281 282 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode, 283 nid_t ino, pgoff_t idx, enum page_type type) 284 { 285 enum page_type btype = PAGE_TYPE_OF_BIO(type); 286 enum temp_type temp; 287 struct f2fs_bio_info *io; 288 bool ret = false; 289 290 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 291 io = sbi->write_io[btype] + temp; 292 293 down_read(&io->io_rwsem); 294 ret = __has_merged_page(io, inode, ino, idx); 295 up_read(&io->io_rwsem); 296 297 /* TODO: use HOT temp only for meta pages now. */ 298 if (ret || btype == META) 299 break; 300 } 301 return ret; 302 } 303 304 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 305 enum page_type type, enum temp_type temp) 306 { 307 enum page_type btype = PAGE_TYPE_OF_BIO(type); 308 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 309 310 down_write(&io->io_rwsem); 311 312 /* change META to META_FLUSH in the checkpoint procedure */ 313 if (type >= META_FLUSH) { 314 io->fio.type = META_FLUSH; 315 io->fio.op = REQ_OP_WRITE; 316 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC; 317 if (!test_opt(sbi, NOBARRIER)) 318 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 319 } 320 __submit_merged_bio(io); 321 up_write(&io->io_rwsem); 322 } 323 324 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 325 struct inode *inode, nid_t ino, pgoff_t idx, 326 enum page_type type, bool force) 327 { 328 enum temp_type temp; 329 330 if (!force && !has_merged_page(sbi, inode, ino, idx, type)) 331 return; 332 333 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 334 335 __f2fs_submit_merged_write(sbi, type, temp); 336 337 /* TODO: use HOT temp only for meta pages now. */ 338 if (type >= META) 339 break; 340 } 341 } 342 343 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 344 { 345 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true); 346 } 347 348 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 349 struct inode *inode, nid_t ino, pgoff_t idx, 350 enum page_type type) 351 { 352 __submit_merged_write_cond(sbi, inode, ino, idx, type, false); 353 } 354 355 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 356 { 357 f2fs_submit_merged_write(sbi, DATA); 358 f2fs_submit_merged_write(sbi, NODE); 359 f2fs_submit_merged_write(sbi, META); 360 } 361 362 /* 363 * Fill the locked page with data located in the block address. 364 * A caller needs to unlock the page on failure. 365 */ 366 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 367 { 368 struct bio *bio; 369 struct page *page = fio->encrypted_page ? 370 fio->encrypted_page : fio->page; 371 372 trace_f2fs_submit_page_bio(page, fio); 373 f2fs_trace_ios(fio, 0); 374 375 /* Allocate a new bio */ 376 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op)); 377 378 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 379 bio_put(bio); 380 return -EFAULT; 381 } 382 bio_set_op_attrs(bio, fio->op, fio->op_flags); 383 384 __submit_bio(fio->sbi, bio, fio->type); 385 386 if (!is_read_io(fio->op)) 387 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page)); 388 return 0; 389 } 390 391 int f2fs_submit_page_write(struct f2fs_io_info *fio) 392 { 393 struct f2fs_sb_info *sbi = fio->sbi; 394 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 395 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 396 struct page *bio_page; 397 int err = 0; 398 399 f2fs_bug_on(sbi, is_read_io(fio->op)); 400 401 down_write(&io->io_rwsem); 402 next: 403 if (fio->in_list) { 404 spin_lock(&io->io_lock); 405 if (list_empty(&io->io_list)) { 406 spin_unlock(&io->io_lock); 407 goto out_fail; 408 } 409 fio = list_first_entry(&io->io_list, 410 struct f2fs_io_info, list); 411 list_del(&fio->list); 412 spin_unlock(&io->io_lock); 413 } 414 415 if (fio->old_blkaddr != NEW_ADDR) 416 verify_block_addr(sbi, fio->old_blkaddr); 417 verify_block_addr(sbi, fio->new_blkaddr); 418 419 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page; 420 421 /* set submitted = true as a return value */ 422 fio->submitted = true; 423 424 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 425 426 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 || 427 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) || 428 !__same_bdev(sbi, fio->new_blkaddr, io->bio))) 429 __submit_merged_bio(io); 430 alloc_new: 431 if (io->bio == NULL) { 432 if ((fio->type == DATA || fio->type == NODE) && 433 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 434 err = -EAGAIN; 435 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 436 goto out_fail; 437 } 438 io->bio = __bio_alloc(sbi, fio->new_blkaddr, 439 BIO_MAX_PAGES, false); 440 io->fio = *fio; 441 } 442 443 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 444 __submit_merged_bio(io); 445 goto alloc_new; 446 } 447 448 io->last_block_in_bio = fio->new_blkaddr; 449 f2fs_trace_ios(fio, 0); 450 451 trace_f2fs_submit_page_write(fio->page, fio); 452 453 if (fio->in_list) 454 goto next; 455 out_fail: 456 up_write(&io->io_rwsem); 457 return err; 458 } 459 460 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 461 unsigned nr_pages) 462 { 463 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 464 struct fscrypt_ctx *ctx = NULL; 465 struct bio *bio; 466 467 if (f2fs_encrypted_file(inode)) { 468 ctx = fscrypt_get_ctx(inode, GFP_NOFS); 469 if (IS_ERR(ctx)) 470 return ERR_CAST(ctx); 471 472 /* wait the page to be moved by cleaning */ 473 f2fs_wait_on_block_writeback(sbi, blkaddr); 474 } 475 476 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false); 477 if (!bio) { 478 if (ctx) 479 fscrypt_release_ctx(ctx); 480 return ERR_PTR(-ENOMEM); 481 } 482 f2fs_target_device(sbi, blkaddr, bio); 483 bio->bi_end_io = f2fs_read_end_io; 484 bio->bi_private = ctx; 485 bio_set_op_attrs(bio, REQ_OP_READ, 0); 486 487 return bio; 488 } 489 490 /* This can handle encryption stuffs */ 491 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 492 block_t blkaddr) 493 { 494 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1); 495 496 if (IS_ERR(bio)) 497 return PTR_ERR(bio); 498 499 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 500 bio_put(bio); 501 return -EFAULT; 502 } 503 __submit_bio(F2FS_I_SB(inode), bio, DATA); 504 return 0; 505 } 506 507 static void __set_data_blkaddr(struct dnode_of_data *dn) 508 { 509 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 510 __le32 *addr_array; 511 int base = 0; 512 513 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 514 base = get_extra_isize(dn->inode); 515 516 /* Get physical address of data block */ 517 addr_array = blkaddr_in_node(rn); 518 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 519 } 520 521 /* 522 * Lock ordering for the change of data block address: 523 * ->data_page 524 * ->node_page 525 * update block addresses in the node page 526 */ 527 void set_data_blkaddr(struct dnode_of_data *dn) 528 { 529 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 530 __set_data_blkaddr(dn); 531 if (set_page_dirty(dn->node_page)) 532 dn->node_changed = true; 533 } 534 535 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 536 { 537 dn->data_blkaddr = blkaddr; 538 set_data_blkaddr(dn); 539 f2fs_update_extent_cache(dn); 540 } 541 542 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 543 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 544 { 545 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 546 int err; 547 548 if (!count) 549 return 0; 550 551 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 552 return -EPERM; 553 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 554 return err; 555 556 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 557 dn->ofs_in_node, count); 558 559 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 560 561 for (; count > 0; dn->ofs_in_node++) { 562 block_t blkaddr = datablock_addr(dn->inode, 563 dn->node_page, dn->ofs_in_node); 564 if (blkaddr == NULL_ADDR) { 565 dn->data_blkaddr = NEW_ADDR; 566 __set_data_blkaddr(dn); 567 count--; 568 } 569 } 570 571 if (set_page_dirty(dn->node_page)) 572 dn->node_changed = true; 573 return 0; 574 } 575 576 /* Should keep dn->ofs_in_node unchanged */ 577 int reserve_new_block(struct dnode_of_data *dn) 578 { 579 unsigned int ofs_in_node = dn->ofs_in_node; 580 int ret; 581 582 ret = reserve_new_blocks(dn, 1); 583 dn->ofs_in_node = ofs_in_node; 584 return ret; 585 } 586 587 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 588 { 589 bool need_put = dn->inode_page ? false : true; 590 int err; 591 592 err = get_dnode_of_data(dn, index, ALLOC_NODE); 593 if (err) 594 return err; 595 596 if (dn->data_blkaddr == NULL_ADDR) 597 err = reserve_new_block(dn); 598 if (err || need_put) 599 f2fs_put_dnode(dn); 600 return err; 601 } 602 603 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 604 { 605 struct extent_info ei = {0,0,0}; 606 struct inode *inode = dn->inode; 607 608 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 609 dn->data_blkaddr = ei.blk + index - ei.fofs; 610 return 0; 611 } 612 613 return f2fs_reserve_block(dn, index); 614 } 615 616 struct page *get_read_data_page(struct inode *inode, pgoff_t index, 617 int op_flags, bool for_write) 618 { 619 struct address_space *mapping = inode->i_mapping; 620 struct dnode_of_data dn; 621 struct page *page; 622 struct extent_info ei = {0,0,0}; 623 int err; 624 625 page = f2fs_grab_cache_page(mapping, index, for_write); 626 if (!page) 627 return ERR_PTR(-ENOMEM); 628 629 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 630 dn.data_blkaddr = ei.blk + index - ei.fofs; 631 goto got_it; 632 } 633 634 set_new_dnode(&dn, inode, NULL, NULL, 0); 635 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 636 if (err) 637 goto put_err; 638 f2fs_put_dnode(&dn); 639 640 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 641 err = -ENOENT; 642 goto put_err; 643 } 644 got_it: 645 if (PageUptodate(page)) { 646 unlock_page(page); 647 return page; 648 } 649 650 /* 651 * A new dentry page is allocated but not able to be written, since its 652 * new inode page couldn't be allocated due to -ENOSPC. 653 * In such the case, its blkaddr can be remained as NEW_ADDR. 654 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. 655 */ 656 if (dn.data_blkaddr == NEW_ADDR) { 657 zero_user_segment(page, 0, PAGE_SIZE); 658 if (!PageUptodate(page)) 659 SetPageUptodate(page); 660 unlock_page(page); 661 return page; 662 } 663 664 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr); 665 if (err) 666 goto put_err; 667 return page; 668 669 put_err: 670 f2fs_put_page(page, 1); 671 return ERR_PTR(err); 672 } 673 674 struct page *find_data_page(struct inode *inode, pgoff_t index) 675 { 676 struct address_space *mapping = inode->i_mapping; 677 struct page *page; 678 679 page = find_get_page(mapping, index); 680 if (page && PageUptodate(page)) 681 return page; 682 f2fs_put_page(page, 0); 683 684 page = get_read_data_page(inode, index, 0, false); 685 if (IS_ERR(page)) 686 return page; 687 688 if (PageUptodate(page)) 689 return page; 690 691 wait_on_page_locked(page); 692 if (unlikely(!PageUptodate(page))) { 693 f2fs_put_page(page, 0); 694 return ERR_PTR(-EIO); 695 } 696 return page; 697 } 698 699 /* 700 * If it tries to access a hole, return an error. 701 * Because, the callers, functions in dir.c and GC, should be able to know 702 * whether this page exists or not. 703 */ 704 struct page *get_lock_data_page(struct inode *inode, pgoff_t index, 705 bool for_write) 706 { 707 struct address_space *mapping = inode->i_mapping; 708 struct page *page; 709 repeat: 710 page = get_read_data_page(inode, index, 0, for_write); 711 if (IS_ERR(page)) 712 return page; 713 714 /* wait for read completion */ 715 lock_page(page); 716 if (unlikely(page->mapping != mapping)) { 717 f2fs_put_page(page, 1); 718 goto repeat; 719 } 720 if (unlikely(!PageUptodate(page))) { 721 f2fs_put_page(page, 1); 722 return ERR_PTR(-EIO); 723 } 724 return page; 725 } 726 727 /* 728 * Caller ensures that this data page is never allocated. 729 * A new zero-filled data page is allocated in the page cache. 730 * 731 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 732 * f2fs_unlock_op(). 733 * Note that, ipage is set only by make_empty_dir, and if any error occur, 734 * ipage should be released by this function. 735 */ 736 struct page *get_new_data_page(struct inode *inode, 737 struct page *ipage, pgoff_t index, bool new_i_size) 738 { 739 struct address_space *mapping = inode->i_mapping; 740 struct page *page; 741 struct dnode_of_data dn; 742 int err; 743 744 page = f2fs_grab_cache_page(mapping, index, true); 745 if (!page) { 746 /* 747 * before exiting, we should make sure ipage will be released 748 * if any error occur. 749 */ 750 f2fs_put_page(ipage, 1); 751 return ERR_PTR(-ENOMEM); 752 } 753 754 set_new_dnode(&dn, inode, ipage, NULL, 0); 755 err = f2fs_reserve_block(&dn, index); 756 if (err) { 757 f2fs_put_page(page, 1); 758 return ERR_PTR(err); 759 } 760 if (!ipage) 761 f2fs_put_dnode(&dn); 762 763 if (PageUptodate(page)) 764 goto got_it; 765 766 if (dn.data_blkaddr == NEW_ADDR) { 767 zero_user_segment(page, 0, PAGE_SIZE); 768 if (!PageUptodate(page)) 769 SetPageUptodate(page); 770 } else { 771 f2fs_put_page(page, 1); 772 773 /* if ipage exists, blkaddr should be NEW_ADDR */ 774 f2fs_bug_on(F2FS_I_SB(inode), ipage); 775 page = get_lock_data_page(inode, index, true); 776 if (IS_ERR(page)) 777 return page; 778 } 779 got_it: 780 if (new_i_size && i_size_read(inode) < 781 ((loff_t)(index + 1) << PAGE_SHIFT)) 782 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 783 return page; 784 } 785 786 static int __allocate_data_block(struct dnode_of_data *dn) 787 { 788 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 789 struct f2fs_summary sum; 790 struct node_info ni; 791 pgoff_t fofs; 792 blkcnt_t count = 1; 793 int err; 794 795 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 796 return -EPERM; 797 798 dn->data_blkaddr = datablock_addr(dn->inode, 799 dn->node_page, dn->ofs_in_node); 800 if (dn->data_blkaddr == NEW_ADDR) 801 goto alloc; 802 803 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 804 return err; 805 806 alloc: 807 get_node_info(sbi, dn->nid, &ni); 808 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 809 810 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr, 811 &sum, CURSEG_WARM_DATA, NULL, false); 812 set_data_blkaddr(dn); 813 814 /* update i_size */ 815 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) + 816 dn->ofs_in_node; 817 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT)) 818 f2fs_i_size_write(dn->inode, 819 ((loff_t)(fofs + 1) << PAGE_SHIFT)); 820 return 0; 821 } 822 823 static inline bool __force_buffered_io(struct inode *inode, int rw) 824 { 825 return (f2fs_encrypted_file(inode) || 826 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) || 827 F2FS_I_SB(inode)->s_ndevs); 828 } 829 830 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from) 831 { 832 struct inode *inode = file_inode(iocb->ki_filp); 833 struct f2fs_map_blocks map; 834 int err = 0; 835 836 /* convert inline data for Direct I/O*/ 837 if (iocb->ki_flags & IOCB_DIRECT) { 838 err = f2fs_convert_inline_inode(inode); 839 if (err) 840 return err; 841 } 842 843 if (is_inode_flag_set(inode, FI_NO_PREALLOC)) 844 return 0; 845 846 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos); 847 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from)); 848 if (map.m_len > map.m_lblk) 849 map.m_len -= map.m_lblk; 850 else 851 map.m_len = 0; 852 853 map.m_next_pgofs = NULL; 854 855 if (iocb->ki_flags & IOCB_DIRECT) 856 return f2fs_map_blocks(inode, &map, 1, 857 __force_buffered_io(inode, WRITE) ? 858 F2FS_GET_BLOCK_PRE_AIO : 859 F2FS_GET_BLOCK_PRE_DIO); 860 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) { 861 err = f2fs_convert_inline_inode(inode); 862 if (err) 863 return err; 864 } 865 if (!f2fs_has_inline_data(inode)) { 866 err = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO); 867 if (map.m_len > 0 && err == -ENOSPC) { 868 set_inode_flag(inode, FI_NO_PREALLOC); 869 err = 0; 870 } 871 return err; 872 } 873 return err; 874 } 875 876 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 877 { 878 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 879 if (lock) 880 down_read(&sbi->node_change); 881 else 882 up_read(&sbi->node_change); 883 } else { 884 if (lock) 885 f2fs_lock_op(sbi); 886 else 887 f2fs_unlock_op(sbi); 888 } 889 } 890 891 /* 892 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with 893 * f2fs_map_blocks structure. 894 * If original data blocks are allocated, then give them to blockdev. 895 * Otherwise, 896 * a. preallocate requested block addresses 897 * b. do not use extent cache for better performance 898 * c. give the block addresses to blockdev 899 */ 900 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 901 int create, int flag) 902 { 903 unsigned int maxblocks = map->m_len; 904 struct dnode_of_data dn; 905 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 906 int mode = create ? ALLOC_NODE : LOOKUP_NODE; 907 pgoff_t pgofs, end_offset, end; 908 int err = 0, ofs = 1; 909 unsigned int ofs_in_node, last_ofs_in_node; 910 blkcnt_t prealloc; 911 struct extent_info ei = {0,0,0}; 912 block_t blkaddr; 913 914 if (!maxblocks) 915 return 0; 916 917 map->m_len = 0; 918 map->m_flags = 0; 919 920 /* it only supports block size == page size */ 921 pgofs = (pgoff_t)map->m_lblk; 922 end = pgofs + maxblocks; 923 924 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 925 map->m_pblk = ei.blk + pgofs - ei.fofs; 926 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 927 map->m_flags = F2FS_MAP_MAPPED; 928 goto out; 929 } 930 931 next_dnode: 932 if (create) 933 __do_map_lock(sbi, flag, true); 934 935 /* When reading holes, we need its node page */ 936 set_new_dnode(&dn, inode, NULL, NULL, 0); 937 err = get_dnode_of_data(&dn, pgofs, mode); 938 if (err) { 939 if (flag == F2FS_GET_BLOCK_BMAP) 940 map->m_pblk = 0; 941 if (err == -ENOENT) { 942 err = 0; 943 if (map->m_next_pgofs) 944 *map->m_next_pgofs = 945 get_next_page_offset(&dn, pgofs); 946 } 947 goto unlock_out; 948 } 949 950 prealloc = 0; 951 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 952 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 953 954 next_block: 955 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node); 956 957 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) { 958 if (create) { 959 if (unlikely(f2fs_cp_error(sbi))) { 960 err = -EIO; 961 goto sync_out; 962 } 963 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 964 if (blkaddr == NULL_ADDR) { 965 prealloc++; 966 last_ofs_in_node = dn.ofs_in_node; 967 } 968 } else { 969 err = __allocate_data_block(&dn); 970 if (!err) 971 set_inode_flag(inode, FI_APPEND_WRITE); 972 } 973 if (err) 974 goto sync_out; 975 map->m_flags |= F2FS_MAP_NEW; 976 blkaddr = dn.data_blkaddr; 977 } else { 978 if (flag == F2FS_GET_BLOCK_BMAP) { 979 map->m_pblk = 0; 980 goto sync_out; 981 } 982 if (flag == F2FS_GET_BLOCK_FIEMAP && 983 blkaddr == NULL_ADDR) { 984 if (map->m_next_pgofs) 985 *map->m_next_pgofs = pgofs + 1; 986 } 987 if (flag != F2FS_GET_BLOCK_FIEMAP || 988 blkaddr != NEW_ADDR) 989 goto sync_out; 990 } 991 } 992 993 if (flag == F2FS_GET_BLOCK_PRE_AIO) 994 goto skip; 995 996 if (map->m_len == 0) { 997 /* preallocated unwritten block should be mapped for fiemap. */ 998 if (blkaddr == NEW_ADDR) 999 map->m_flags |= F2FS_MAP_UNWRITTEN; 1000 map->m_flags |= F2FS_MAP_MAPPED; 1001 1002 map->m_pblk = blkaddr; 1003 map->m_len = 1; 1004 } else if ((map->m_pblk != NEW_ADDR && 1005 blkaddr == (map->m_pblk + ofs)) || 1006 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1007 flag == F2FS_GET_BLOCK_PRE_DIO) { 1008 ofs++; 1009 map->m_len++; 1010 } else { 1011 goto sync_out; 1012 } 1013 1014 skip: 1015 dn.ofs_in_node++; 1016 pgofs++; 1017 1018 /* preallocate blocks in batch for one dnode page */ 1019 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1020 (pgofs == end || dn.ofs_in_node == end_offset)) { 1021 1022 dn.ofs_in_node = ofs_in_node; 1023 err = reserve_new_blocks(&dn, prealloc); 1024 if (err) 1025 goto sync_out; 1026 1027 map->m_len += dn.ofs_in_node - ofs_in_node; 1028 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1029 err = -ENOSPC; 1030 goto sync_out; 1031 } 1032 dn.ofs_in_node = end_offset; 1033 } 1034 1035 if (pgofs >= end) 1036 goto sync_out; 1037 else if (dn.ofs_in_node < end_offset) 1038 goto next_block; 1039 1040 f2fs_put_dnode(&dn); 1041 1042 if (create) { 1043 __do_map_lock(sbi, flag, false); 1044 f2fs_balance_fs(sbi, dn.node_changed); 1045 } 1046 goto next_dnode; 1047 1048 sync_out: 1049 f2fs_put_dnode(&dn); 1050 unlock_out: 1051 if (create) { 1052 __do_map_lock(sbi, flag, false); 1053 f2fs_balance_fs(sbi, dn.node_changed); 1054 } 1055 out: 1056 trace_f2fs_map_blocks(inode, map, err); 1057 return err; 1058 } 1059 1060 static int __get_data_block(struct inode *inode, sector_t iblock, 1061 struct buffer_head *bh, int create, int flag, 1062 pgoff_t *next_pgofs) 1063 { 1064 struct f2fs_map_blocks map; 1065 int err; 1066 1067 map.m_lblk = iblock; 1068 map.m_len = bh->b_size >> inode->i_blkbits; 1069 map.m_next_pgofs = next_pgofs; 1070 1071 err = f2fs_map_blocks(inode, &map, create, flag); 1072 if (!err) { 1073 map_bh(bh, inode->i_sb, map.m_pblk); 1074 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags; 1075 bh->b_size = (u64)map.m_len << inode->i_blkbits; 1076 } 1077 return err; 1078 } 1079 1080 static int get_data_block(struct inode *inode, sector_t iblock, 1081 struct buffer_head *bh_result, int create, int flag, 1082 pgoff_t *next_pgofs) 1083 { 1084 return __get_data_block(inode, iblock, bh_result, create, 1085 flag, next_pgofs); 1086 } 1087 1088 static int get_data_block_dio(struct inode *inode, sector_t iblock, 1089 struct buffer_head *bh_result, int create) 1090 { 1091 return __get_data_block(inode, iblock, bh_result, create, 1092 F2FS_GET_BLOCK_DEFAULT, NULL); 1093 } 1094 1095 static int get_data_block_bmap(struct inode *inode, sector_t iblock, 1096 struct buffer_head *bh_result, int create) 1097 { 1098 /* Block number less than F2FS MAX BLOCKS */ 1099 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks)) 1100 return -EFBIG; 1101 1102 return __get_data_block(inode, iblock, bh_result, create, 1103 F2FS_GET_BLOCK_BMAP, NULL); 1104 } 1105 1106 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset) 1107 { 1108 return (offset >> inode->i_blkbits); 1109 } 1110 1111 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk) 1112 { 1113 return (blk << inode->i_blkbits); 1114 } 1115 1116 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1117 u64 start, u64 len) 1118 { 1119 struct buffer_head map_bh; 1120 sector_t start_blk, last_blk; 1121 pgoff_t next_pgofs; 1122 u64 logical = 0, phys = 0, size = 0; 1123 u32 flags = 0; 1124 int ret = 0; 1125 1126 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC); 1127 if (ret) 1128 return ret; 1129 1130 if (f2fs_has_inline_data(inode)) { 1131 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1132 if (ret != -EAGAIN) 1133 return ret; 1134 } 1135 1136 inode_lock(inode); 1137 1138 if (logical_to_blk(inode, len) == 0) 1139 len = blk_to_logical(inode, 1); 1140 1141 start_blk = logical_to_blk(inode, start); 1142 last_blk = logical_to_blk(inode, start + len - 1); 1143 1144 next: 1145 memset(&map_bh, 0, sizeof(struct buffer_head)); 1146 map_bh.b_size = len; 1147 1148 ret = get_data_block(inode, start_blk, &map_bh, 0, 1149 F2FS_GET_BLOCK_FIEMAP, &next_pgofs); 1150 if (ret) 1151 goto out; 1152 1153 /* HOLE */ 1154 if (!buffer_mapped(&map_bh)) { 1155 start_blk = next_pgofs; 1156 1157 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode, 1158 F2FS_I_SB(inode)->max_file_blocks)) 1159 goto prep_next; 1160 1161 flags |= FIEMAP_EXTENT_LAST; 1162 } 1163 1164 if (size) { 1165 if (f2fs_encrypted_inode(inode)) 1166 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1167 1168 ret = fiemap_fill_next_extent(fieinfo, logical, 1169 phys, size, flags); 1170 } 1171 1172 if (start_blk > last_blk || ret) 1173 goto out; 1174 1175 logical = blk_to_logical(inode, start_blk); 1176 phys = blk_to_logical(inode, map_bh.b_blocknr); 1177 size = map_bh.b_size; 1178 flags = 0; 1179 if (buffer_unwritten(&map_bh)) 1180 flags = FIEMAP_EXTENT_UNWRITTEN; 1181 1182 start_blk += logical_to_blk(inode, size); 1183 1184 prep_next: 1185 cond_resched(); 1186 if (fatal_signal_pending(current)) 1187 ret = -EINTR; 1188 else 1189 goto next; 1190 out: 1191 if (ret == 1) 1192 ret = 0; 1193 1194 inode_unlock(inode); 1195 return ret; 1196 } 1197 1198 /* 1199 * This function was originally taken from fs/mpage.c, and customized for f2fs. 1200 * Major change was from block_size == page_size in f2fs by default. 1201 */ 1202 static int f2fs_mpage_readpages(struct address_space *mapping, 1203 struct list_head *pages, struct page *page, 1204 unsigned nr_pages) 1205 { 1206 struct bio *bio = NULL; 1207 sector_t last_block_in_bio = 0; 1208 struct inode *inode = mapping->host; 1209 const unsigned blkbits = inode->i_blkbits; 1210 const unsigned blocksize = 1 << blkbits; 1211 sector_t block_in_file; 1212 sector_t last_block; 1213 sector_t last_block_in_file; 1214 sector_t block_nr; 1215 struct f2fs_map_blocks map; 1216 1217 map.m_pblk = 0; 1218 map.m_lblk = 0; 1219 map.m_len = 0; 1220 map.m_flags = 0; 1221 map.m_next_pgofs = NULL; 1222 1223 for (; nr_pages; nr_pages--) { 1224 if (pages) { 1225 page = list_last_entry(pages, struct page, lru); 1226 1227 prefetchw(&page->flags); 1228 list_del(&page->lru); 1229 if (add_to_page_cache_lru(page, mapping, 1230 page->index, 1231 readahead_gfp_mask(mapping))) 1232 goto next_page; 1233 } 1234 1235 block_in_file = (sector_t)page->index; 1236 last_block = block_in_file + nr_pages; 1237 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> 1238 blkbits; 1239 if (last_block > last_block_in_file) 1240 last_block = last_block_in_file; 1241 1242 /* 1243 * Map blocks using the previous result first. 1244 */ 1245 if ((map.m_flags & F2FS_MAP_MAPPED) && 1246 block_in_file > map.m_lblk && 1247 block_in_file < (map.m_lblk + map.m_len)) 1248 goto got_it; 1249 1250 /* 1251 * Then do more f2fs_map_blocks() calls until we are 1252 * done with this page. 1253 */ 1254 map.m_flags = 0; 1255 1256 if (block_in_file < last_block) { 1257 map.m_lblk = block_in_file; 1258 map.m_len = last_block - block_in_file; 1259 1260 if (f2fs_map_blocks(inode, &map, 0, 1261 F2FS_GET_BLOCK_DEFAULT)) 1262 goto set_error_page; 1263 } 1264 got_it: 1265 if ((map.m_flags & F2FS_MAP_MAPPED)) { 1266 block_nr = map.m_pblk + block_in_file - map.m_lblk; 1267 SetPageMappedToDisk(page); 1268 1269 if (!PageUptodate(page) && !cleancache_get_page(page)) { 1270 SetPageUptodate(page); 1271 goto confused; 1272 } 1273 } else { 1274 zero_user_segment(page, 0, PAGE_SIZE); 1275 if (!PageUptodate(page)) 1276 SetPageUptodate(page); 1277 unlock_page(page); 1278 goto next_page; 1279 } 1280 1281 /* 1282 * This page will go to BIO. Do we need to send this 1283 * BIO off first? 1284 */ 1285 if (bio && (last_block_in_bio != block_nr - 1 || 1286 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) { 1287 submit_and_realloc: 1288 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1289 bio = NULL; 1290 } 1291 if (bio == NULL) { 1292 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages); 1293 if (IS_ERR(bio)) { 1294 bio = NULL; 1295 goto set_error_page; 1296 } 1297 } 1298 1299 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 1300 goto submit_and_realloc; 1301 1302 last_block_in_bio = block_nr; 1303 goto next_page; 1304 set_error_page: 1305 SetPageError(page); 1306 zero_user_segment(page, 0, PAGE_SIZE); 1307 unlock_page(page); 1308 goto next_page; 1309 confused: 1310 if (bio) { 1311 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1312 bio = NULL; 1313 } 1314 unlock_page(page); 1315 next_page: 1316 if (pages) 1317 put_page(page); 1318 } 1319 BUG_ON(pages && !list_empty(pages)); 1320 if (bio) 1321 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1322 return 0; 1323 } 1324 1325 static int f2fs_read_data_page(struct file *file, struct page *page) 1326 { 1327 struct inode *inode = page->mapping->host; 1328 int ret = -EAGAIN; 1329 1330 trace_f2fs_readpage(page, DATA); 1331 1332 /* If the file has inline data, try to read it directly */ 1333 if (f2fs_has_inline_data(inode)) 1334 ret = f2fs_read_inline_data(inode, page); 1335 if (ret == -EAGAIN) 1336 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1); 1337 return ret; 1338 } 1339 1340 static int f2fs_read_data_pages(struct file *file, 1341 struct address_space *mapping, 1342 struct list_head *pages, unsigned nr_pages) 1343 { 1344 struct inode *inode = mapping->host; 1345 struct page *page = list_last_entry(pages, struct page, lru); 1346 1347 trace_f2fs_readpages(inode, page, nr_pages); 1348 1349 /* If the file has inline data, skip readpages */ 1350 if (f2fs_has_inline_data(inode)) 1351 return 0; 1352 1353 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages); 1354 } 1355 1356 static int encrypt_one_page(struct f2fs_io_info *fio) 1357 { 1358 struct inode *inode = fio->page->mapping->host; 1359 gfp_t gfp_flags = GFP_NOFS; 1360 1361 if (!f2fs_encrypted_file(inode)) 1362 return 0; 1363 1364 /* wait for GCed encrypted page writeback */ 1365 f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr); 1366 1367 retry_encrypt: 1368 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page, 1369 PAGE_SIZE, 0, fio->page->index, gfp_flags); 1370 if (!IS_ERR(fio->encrypted_page)) 1371 return 0; 1372 1373 /* flush pending IOs and wait for a while in the ENOMEM case */ 1374 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 1375 f2fs_flush_merged_writes(fio->sbi); 1376 congestion_wait(BLK_RW_ASYNC, HZ/50); 1377 gfp_flags |= __GFP_NOFAIL; 1378 goto retry_encrypt; 1379 } 1380 return PTR_ERR(fio->encrypted_page); 1381 } 1382 1383 static inline bool need_inplace_update(struct f2fs_io_info *fio) 1384 { 1385 struct inode *inode = fio->page->mapping->host; 1386 1387 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode)) 1388 return false; 1389 if (is_cold_data(fio->page)) 1390 return false; 1391 if (IS_ATOMIC_WRITTEN_PAGE(fio->page)) 1392 return false; 1393 1394 return need_inplace_update_policy(inode, fio); 1395 } 1396 1397 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio) 1398 { 1399 if (fio->old_blkaddr == NEW_ADDR) 1400 return false; 1401 if (fio->old_blkaddr == NULL_ADDR) 1402 return false; 1403 return true; 1404 } 1405 1406 int do_write_data_page(struct f2fs_io_info *fio) 1407 { 1408 struct page *page = fio->page; 1409 struct inode *inode = page->mapping->host; 1410 struct dnode_of_data dn; 1411 struct extent_info ei = {0,0,0}; 1412 bool ipu_force = false; 1413 int err = 0; 1414 1415 set_new_dnode(&dn, inode, NULL, NULL, 0); 1416 if (need_inplace_update(fio) && 1417 f2fs_lookup_extent_cache(inode, page->index, &ei)) { 1418 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 1419 1420 if (valid_ipu_blkaddr(fio)) { 1421 ipu_force = true; 1422 fio->need_lock = LOCK_DONE; 1423 goto got_it; 1424 } 1425 } 1426 1427 /* Deadlock due to between page->lock and f2fs_lock_op */ 1428 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 1429 return -EAGAIN; 1430 1431 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 1432 if (err) 1433 goto out; 1434 1435 fio->old_blkaddr = dn.data_blkaddr; 1436 1437 /* This page is already truncated */ 1438 if (fio->old_blkaddr == NULL_ADDR) { 1439 ClearPageUptodate(page); 1440 goto out_writepage; 1441 } 1442 got_it: 1443 /* 1444 * If current allocation needs SSR, 1445 * it had better in-place writes for updated data. 1446 */ 1447 if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) { 1448 err = encrypt_one_page(fio); 1449 if (err) 1450 goto out_writepage; 1451 1452 set_page_writeback(page); 1453 f2fs_put_dnode(&dn); 1454 if (fio->need_lock == LOCK_REQ) 1455 f2fs_unlock_op(fio->sbi); 1456 err = rewrite_data_page(fio); 1457 trace_f2fs_do_write_data_page(fio->page, IPU); 1458 set_inode_flag(inode, FI_UPDATE_WRITE); 1459 return err; 1460 } 1461 1462 if (fio->need_lock == LOCK_RETRY) { 1463 if (!f2fs_trylock_op(fio->sbi)) { 1464 err = -EAGAIN; 1465 goto out_writepage; 1466 } 1467 fio->need_lock = LOCK_REQ; 1468 } 1469 1470 err = encrypt_one_page(fio); 1471 if (err) 1472 goto out_writepage; 1473 1474 set_page_writeback(page); 1475 1476 /* LFS mode write path */ 1477 write_data_page(&dn, fio); 1478 trace_f2fs_do_write_data_page(page, OPU); 1479 set_inode_flag(inode, FI_APPEND_WRITE); 1480 if (page->index == 0) 1481 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 1482 out_writepage: 1483 f2fs_put_dnode(&dn); 1484 out: 1485 if (fio->need_lock == LOCK_REQ) 1486 f2fs_unlock_op(fio->sbi); 1487 return err; 1488 } 1489 1490 static int __write_data_page(struct page *page, bool *submitted, 1491 struct writeback_control *wbc, 1492 enum iostat_type io_type) 1493 { 1494 struct inode *inode = page->mapping->host; 1495 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1496 loff_t i_size = i_size_read(inode); 1497 const pgoff_t end_index = ((unsigned long long) i_size) 1498 >> PAGE_SHIFT; 1499 loff_t psize = (page->index + 1) << PAGE_SHIFT; 1500 unsigned offset = 0; 1501 bool need_balance_fs = false; 1502 int err = 0; 1503 struct f2fs_io_info fio = { 1504 .sbi = sbi, 1505 .ino = inode->i_ino, 1506 .type = DATA, 1507 .op = REQ_OP_WRITE, 1508 .op_flags = wbc_to_write_flags(wbc), 1509 .old_blkaddr = NULL_ADDR, 1510 .page = page, 1511 .encrypted_page = NULL, 1512 .submitted = false, 1513 .need_lock = LOCK_RETRY, 1514 .io_type = io_type, 1515 }; 1516 1517 trace_f2fs_writepage(page, DATA); 1518 1519 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1520 goto redirty_out; 1521 1522 if (page->index < end_index) 1523 goto write; 1524 1525 /* 1526 * If the offset is out-of-range of file size, 1527 * this page does not have to be written to disk. 1528 */ 1529 offset = i_size & (PAGE_SIZE - 1); 1530 if ((page->index >= end_index + 1) || !offset) 1531 goto out; 1532 1533 zero_user_segment(page, offset, PAGE_SIZE); 1534 write: 1535 if (f2fs_is_drop_cache(inode)) 1536 goto out; 1537 /* we should not write 0'th page having journal header */ 1538 if (f2fs_is_volatile_file(inode) && (!page->index || 1539 (!wbc->for_reclaim && 1540 available_free_memory(sbi, BASE_CHECK)))) 1541 goto redirty_out; 1542 1543 /* we should bypass data pages to proceed the kworkder jobs */ 1544 if (unlikely(f2fs_cp_error(sbi))) { 1545 mapping_set_error(page->mapping, -EIO); 1546 goto out; 1547 } 1548 1549 /* Dentry blocks are controlled by checkpoint */ 1550 if (S_ISDIR(inode->i_mode)) { 1551 fio.need_lock = LOCK_DONE; 1552 err = do_write_data_page(&fio); 1553 goto done; 1554 } 1555 1556 if (!wbc->for_reclaim) 1557 need_balance_fs = true; 1558 else if (has_not_enough_free_secs(sbi, 0, 0)) 1559 goto redirty_out; 1560 else 1561 set_inode_flag(inode, FI_HOT_DATA); 1562 1563 err = -EAGAIN; 1564 if (f2fs_has_inline_data(inode)) { 1565 err = f2fs_write_inline_data(inode, page); 1566 if (!err) 1567 goto out; 1568 } 1569 1570 if (err == -EAGAIN) { 1571 err = do_write_data_page(&fio); 1572 if (err == -EAGAIN) { 1573 fio.need_lock = LOCK_REQ; 1574 err = do_write_data_page(&fio); 1575 } 1576 } 1577 1578 down_write(&F2FS_I(inode)->i_sem); 1579 if (F2FS_I(inode)->last_disk_size < psize) 1580 F2FS_I(inode)->last_disk_size = psize; 1581 up_write(&F2FS_I(inode)->i_sem); 1582 1583 done: 1584 if (err && err != -ENOENT) 1585 goto redirty_out; 1586 1587 out: 1588 inode_dec_dirty_pages(inode); 1589 if (err) 1590 ClearPageUptodate(page); 1591 1592 if (wbc->for_reclaim) { 1593 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA); 1594 clear_inode_flag(inode, FI_HOT_DATA); 1595 remove_dirty_inode(inode); 1596 submitted = NULL; 1597 } 1598 1599 unlock_page(page); 1600 if (!S_ISDIR(inode->i_mode)) 1601 f2fs_balance_fs(sbi, need_balance_fs); 1602 1603 if (unlikely(f2fs_cp_error(sbi))) { 1604 f2fs_submit_merged_write(sbi, DATA); 1605 submitted = NULL; 1606 } 1607 1608 if (submitted) 1609 *submitted = fio.submitted; 1610 1611 return 0; 1612 1613 redirty_out: 1614 redirty_page_for_writepage(wbc, page); 1615 if (!err) 1616 return AOP_WRITEPAGE_ACTIVATE; 1617 unlock_page(page); 1618 return err; 1619 } 1620 1621 static int f2fs_write_data_page(struct page *page, 1622 struct writeback_control *wbc) 1623 { 1624 return __write_data_page(page, NULL, wbc, FS_DATA_IO); 1625 } 1626 1627 /* 1628 * This function was copied from write_cche_pages from mm/page-writeback.c. 1629 * The major change is making write step of cold data page separately from 1630 * warm/hot data page. 1631 */ 1632 static int f2fs_write_cache_pages(struct address_space *mapping, 1633 struct writeback_control *wbc, 1634 enum iostat_type io_type) 1635 { 1636 int ret = 0; 1637 int done = 0; 1638 struct pagevec pvec; 1639 int nr_pages; 1640 pgoff_t uninitialized_var(writeback_index); 1641 pgoff_t index; 1642 pgoff_t end; /* Inclusive */ 1643 pgoff_t done_index; 1644 pgoff_t last_idx = ULONG_MAX; 1645 int cycled; 1646 int range_whole = 0; 1647 int tag; 1648 1649 pagevec_init(&pvec); 1650 1651 if (get_dirty_pages(mapping->host) <= 1652 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 1653 set_inode_flag(mapping->host, FI_HOT_DATA); 1654 else 1655 clear_inode_flag(mapping->host, FI_HOT_DATA); 1656 1657 if (wbc->range_cyclic) { 1658 writeback_index = mapping->writeback_index; /* prev offset */ 1659 index = writeback_index; 1660 if (index == 0) 1661 cycled = 1; 1662 else 1663 cycled = 0; 1664 end = -1; 1665 } else { 1666 index = wbc->range_start >> PAGE_SHIFT; 1667 end = wbc->range_end >> PAGE_SHIFT; 1668 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 1669 range_whole = 1; 1670 cycled = 1; /* ignore range_cyclic tests */ 1671 } 1672 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 1673 tag = PAGECACHE_TAG_TOWRITE; 1674 else 1675 tag = PAGECACHE_TAG_DIRTY; 1676 retry: 1677 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 1678 tag_pages_for_writeback(mapping, index, end); 1679 done_index = index; 1680 while (!done && (index <= end)) { 1681 int i; 1682 1683 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 1684 tag); 1685 if (nr_pages == 0) 1686 break; 1687 1688 for (i = 0; i < nr_pages; i++) { 1689 struct page *page = pvec.pages[i]; 1690 bool submitted = false; 1691 1692 done_index = page->index; 1693 retry_write: 1694 lock_page(page); 1695 1696 if (unlikely(page->mapping != mapping)) { 1697 continue_unlock: 1698 unlock_page(page); 1699 continue; 1700 } 1701 1702 if (!PageDirty(page)) { 1703 /* someone wrote it for us */ 1704 goto continue_unlock; 1705 } 1706 1707 if (PageWriteback(page)) { 1708 if (wbc->sync_mode != WB_SYNC_NONE) 1709 f2fs_wait_on_page_writeback(page, 1710 DATA, true); 1711 else 1712 goto continue_unlock; 1713 } 1714 1715 BUG_ON(PageWriteback(page)); 1716 if (!clear_page_dirty_for_io(page)) 1717 goto continue_unlock; 1718 1719 ret = __write_data_page(page, &submitted, wbc, io_type); 1720 if (unlikely(ret)) { 1721 /* 1722 * keep nr_to_write, since vfs uses this to 1723 * get # of written pages. 1724 */ 1725 if (ret == AOP_WRITEPAGE_ACTIVATE) { 1726 unlock_page(page); 1727 ret = 0; 1728 continue; 1729 } else if (ret == -EAGAIN) { 1730 ret = 0; 1731 if (wbc->sync_mode == WB_SYNC_ALL) { 1732 cond_resched(); 1733 congestion_wait(BLK_RW_ASYNC, 1734 HZ/50); 1735 goto retry_write; 1736 } 1737 continue; 1738 } 1739 done_index = page->index + 1; 1740 done = 1; 1741 break; 1742 } else if (submitted) { 1743 last_idx = page->index; 1744 } 1745 1746 /* give a priority to WB_SYNC threads */ 1747 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) || 1748 --wbc->nr_to_write <= 0) && 1749 wbc->sync_mode == WB_SYNC_NONE) { 1750 done = 1; 1751 break; 1752 } 1753 } 1754 pagevec_release(&pvec); 1755 cond_resched(); 1756 } 1757 1758 if (!cycled && !done) { 1759 cycled = 1; 1760 index = 0; 1761 end = writeback_index - 1; 1762 goto retry; 1763 } 1764 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 1765 mapping->writeback_index = done_index; 1766 1767 if (last_idx != ULONG_MAX) 1768 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 1769 0, last_idx, DATA); 1770 1771 return ret; 1772 } 1773 1774 int __f2fs_write_data_pages(struct address_space *mapping, 1775 struct writeback_control *wbc, 1776 enum iostat_type io_type) 1777 { 1778 struct inode *inode = mapping->host; 1779 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1780 struct blk_plug plug; 1781 int ret; 1782 1783 /* deal with chardevs and other special file */ 1784 if (!mapping->a_ops->writepage) 1785 return 0; 1786 1787 /* skip writing if there is no dirty page in this inode */ 1788 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 1789 return 0; 1790 1791 /* during POR, we don't need to trigger writepage at all. */ 1792 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1793 goto skip_write; 1794 1795 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE && 1796 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 1797 available_free_memory(sbi, DIRTY_DENTS)) 1798 goto skip_write; 1799 1800 /* skip writing during file defragment */ 1801 if (is_inode_flag_set(inode, FI_DO_DEFRAG)) 1802 goto skip_write; 1803 1804 trace_f2fs_writepages(mapping->host, wbc, DATA); 1805 1806 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 1807 if (wbc->sync_mode == WB_SYNC_ALL) 1808 atomic_inc(&sbi->wb_sync_req); 1809 else if (atomic_read(&sbi->wb_sync_req)) 1810 goto skip_write; 1811 1812 blk_start_plug(&plug); 1813 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 1814 blk_finish_plug(&plug); 1815 1816 if (wbc->sync_mode == WB_SYNC_ALL) 1817 atomic_dec(&sbi->wb_sync_req); 1818 /* 1819 * if some pages were truncated, we cannot guarantee its mapping->host 1820 * to detect pending bios. 1821 */ 1822 1823 remove_dirty_inode(inode); 1824 return ret; 1825 1826 skip_write: 1827 wbc->pages_skipped += get_dirty_pages(inode); 1828 trace_f2fs_writepages(mapping->host, wbc, DATA); 1829 return 0; 1830 } 1831 1832 static int f2fs_write_data_pages(struct address_space *mapping, 1833 struct writeback_control *wbc) 1834 { 1835 struct inode *inode = mapping->host; 1836 1837 return __f2fs_write_data_pages(mapping, wbc, 1838 F2FS_I(inode)->cp_task == current ? 1839 FS_CP_DATA_IO : FS_DATA_IO); 1840 } 1841 1842 static void f2fs_write_failed(struct address_space *mapping, loff_t to) 1843 { 1844 struct inode *inode = mapping->host; 1845 loff_t i_size = i_size_read(inode); 1846 1847 if (to > i_size) { 1848 down_write(&F2FS_I(inode)->i_mmap_sem); 1849 truncate_pagecache(inode, i_size); 1850 truncate_blocks(inode, i_size, true); 1851 up_write(&F2FS_I(inode)->i_mmap_sem); 1852 } 1853 } 1854 1855 static int prepare_write_begin(struct f2fs_sb_info *sbi, 1856 struct page *page, loff_t pos, unsigned len, 1857 block_t *blk_addr, bool *node_changed) 1858 { 1859 struct inode *inode = page->mapping->host; 1860 pgoff_t index = page->index; 1861 struct dnode_of_data dn; 1862 struct page *ipage; 1863 bool locked = false; 1864 struct extent_info ei = {0,0,0}; 1865 int err = 0; 1866 1867 /* 1868 * we already allocated all the blocks, so we don't need to get 1869 * the block addresses when there is no need to fill the page. 1870 */ 1871 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE && 1872 !is_inode_flag_set(inode, FI_NO_PREALLOC)) 1873 return 0; 1874 1875 if (f2fs_has_inline_data(inode) || 1876 (pos & PAGE_MASK) >= i_size_read(inode)) { 1877 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true); 1878 locked = true; 1879 } 1880 restart: 1881 /* check inline_data */ 1882 ipage = get_node_page(sbi, inode->i_ino); 1883 if (IS_ERR(ipage)) { 1884 err = PTR_ERR(ipage); 1885 goto unlock_out; 1886 } 1887 1888 set_new_dnode(&dn, inode, ipage, ipage, 0); 1889 1890 if (f2fs_has_inline_data(inode)) { 1891 if (pos + len <= MAX_INLINE_DATA(inode)) { 1892 read_inline_data(page, ipage); 1893 set_inode_flag(inode, FI_DATA_EXIST); 1894 if (inode->i_nlink) 1895 set_inline_node(ipage); 1896 } else { 1897 err = f2fs_convert_inline_page(&dn, page); 1898 if (err) 1899 goto out; 1900 if (dn.data_blkaddr == NULL_ADDR) 1901 err = f2fs_get_block(&dn, index); 1902 } 1903 } else if (locked) { 1904 err = f2fs_get_block(&dn, index); 1905 } else { 1906 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1907 dn.data_blkaddr = ei.blk + index - ei.fofs; 1908 } else { 1909 /* hole case */ 1910 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 1911 if (err || dn.data_blkaddr == NULL_ADDR) { 1912 f2fs_put_dnode(&dn); 1913 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 1914 true); 1915 locked = true; 1916 goto restart; 1917 } 1918 } 1919 } 1920 1921 /* convert_inline_page can make node_changed */ 1922 *blk_addr = dn.data_blkaddr; 1923 *node_changed = dn.node_changed; 1924 out: 1925 f2fs_put_dnode(&dn); 1926 unlock_out: 1927 if (locked) 1928 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false); 1929 return err; 1930 } 1931 1932 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 1933 loff_t pos, unsigned len, unsigned flags, 1934 struct page **pagep, void **fsdata) 1935 { 1936 struct inode *inode = mapping->host; 1937 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1938 struct page *page = NULL; 1939 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 1940 bool need_balance = false; 1941 block_t blkaddr = NULL_ADDR; 1942 int err = 0; 1943 1944 trace_f2fs_write_begin(inode, pos, len, flags); 1945 1946 if (f2fs_is_atomic_file(inode) && 1947 !available_free_memory(sbi, INMEM_PAGES)) { 1948 err = -ENOMEM; 1949 goto fail; 1950 } 1951 1952 /* 1953 * We should check this at this moment to avoid deadlock on inode page 1954 * and #0 page. The locking rule for inline_data conversion should be: 1955 * lock_page(page #0) -> lock_page(inode_page) 1956 */ 1957 if (index != 0) { 1958 err = f2fs_convert_inline_inode(inode); 1959 if (err) 1960 goto fail; 1961 } 1962 repeat: 1963 /* 1964 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 1965 * wait_for_stable_page. Will wait that below with our IO control. 1966 */ 1967 page = f2fs_pagecache_get_page(mapping, index, 1968 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 1969 if (!page) { 1970 err = -ENOMEM; 1971 goto fail; 1972 } 1973 1974 *pagep = page; 1975 1976 err = prepare_write_begin(sbi, page, pos, len, 1977 &blkaddr, &need_balance); 1978 if (err) 1979 goto fail; 1980 1981 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) { 1982 unlock_page(page); 1983 f2fs_balance_fs(sbi, true); 1984 lock_page(page); 1985 if (page->mapping != mapping) { 1986 /* The page got truncated from under us */ 1987 f2fs_put_page(page, 1); 1988 goto repeat; 1989 } 1990 } 1991 1992 f2fs_wait_on_page_writeback(page, DATA, false); 1993 1994 /* wait for GCed encrypted page writeback */ 1995 if (f2fs_encrypted_file(inode)) 1996 f2fs_wait_on_block_writeback(sbi, blkaddr); 1997 1998 if (len == PAGE_SIZE || PageUptodate(page)) 1999 return 0; 2000 2001 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) { 2002 zero_user_segment(page, len, PAGE_SIZE); 2003 return 0; 2004 } 2005 2006 if (blkaddr == NEW_ADDR) { 2007 zero_user_segment(page, 0, PAGE_SIZE); 2008 SetPageUptodate(page); 2009 } else { 2010 err = f2fs_submit_page_read(inode, page, blkaddr); 2011 if (err) 2012 goto fail; 2013 2014 lock_page(page); 2015 if (unlikely(page->mapping != mapping)) { 2016 f2fs_put_page(page, 1); 2017 goto repeat; 2018 } 2019 if (unlikely(!PageUptodate(page))) { 2020 err = -EIO; 2021 goto fail; 2022 } 2023 } 2024 return 0; 2025 2026 fail: 2027 f2fs_put_page(page, 1); 2028 f2fs_write_failed(mapping, pos + len); 2029 if (f2fs_is_atomic_file(inode)) 2030 drop_inmem_pages_all(sbi); 2031 return err; 2032 } 2033 2034 static int f2fs_write_end(struct file *file, 2035 struct address_space *mapping, 2036 loff_t pos, unsigned len, unsigned copied, 2037 struct page *page, void *fsdata) 2038 { 2039 struct inode *inode = page->mapping->host; 2040 2041 trace_f2fs_write_end(inode, pos, len, copied); 2042 2043 /* 2044 * This should be come from len == PAGE_SIZE, and we expect copied 2045 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 2046 * let generic_perform_write() try to copy data again through copied=0. 2047 */ 2048 if (!PageUptodate(page)) { 2049 if (unlikely(copied != len)) 2050 copied = 0; 2051 else 2052 SetPageUptodate(page); 2053 } 2054 if (!copied) 2055 goto unlock_out; 2056 2057 set_page_dirty(page); 2058 2059 if (pos + copied > i_size_read(inode)) 2060 f2fs_i_size_write(inode, pos + copied); 2061 unlock_out: 2062 f2fs_put_page(page, 1); 2063 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 2064 return copied; 2065 } 2066 2067 static int check_direct_IO(struct inode *inode, struct iov_iter *iter, 2068 loff_t offset) 2069 { 2070 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1; 2071 2072 if (offset & blocksize_mask) 2073 return -EINVAL; 2074 2075 if (iov_iter_alignment(iter) & blocksize_mask) 2076 return -EINVAL; 2077 2078 return 0; 2079 } 2080 2081 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 2082 { 2083 struct address_space *mapping = iocb->ki_filp->f_mapping; 2084 struct inode *inode = mapping->host; 2085 size_t count = iov_iter_count(iter); 2086 loff_t offset = iocb->ki_pos; 2087 int rw = iov_iter_rw(iter); 2088 int err; 2089 2090 err = check_direct_IO(inode, iter, offset); 2091 if (err) 2092 return err; 2093 2094 if (__force_buffered_io(inode, rw)) 2095 return 0; 2096 2097 trace_f2fs_direct_IO_enter(inode, offset, count, rw); 2098 2099 down_read(&F2FS_I(inode)->dio_rwsem[rw]); 2100 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio); 2101 up_read(&F2FS_I(inode)->dio_rwsem[rw]); 2102 2103 if (rw == WRITE) { 2104 if (err > 0) { 2105 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO, 2106 err); 2107 set_inode_flag(inode, FI_UPDATE_WRITE); 2108 } else if (err < 0) { 2109 f2fs_write_failed(mapping, offset + count); 2110 } 2111 } 2112 2113 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); 2114 2115 return err; 2116 } 2117 2118 void f2fs_invalidate_page(struct page *page, unsigned int offset, 2119 unsigned int length) 2120 { 2121 struct inode *inode = page->mapping->host; 2122 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2123 2124 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 2125 (offset % PAGE_SIZE || length != PAGE_SIZE)) 2126 return; 2127 2128 if (PageDirty(page)) { 2129 if (inode->i_ino == F2FS_META_INO(sbi)) { 2130 dec_page_count(sbi, F2FS_DIRTY_META); 2131 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 2132 dec_page_count(sbi, F2FS_DIRTY_NODES); 2133 } else { 2134 inode_dec_dirty_pages(inode); 2135 remove_dirty_inode(inode); 2136 } 2137 } 2138 2139 /* This is atomic written page, keep Private */ 2140 if (IS_ATOMIC_WRITTEN_PAGE(page)) 2141 return drop_inmem_page(inode, page); 2142 2143 set_page_private(page, 0); 2144 ClearPagePrivate(page); 2145 } 2146 2147 int f2fs_release_page(struct page *page, gfp_t wait) 2148 { 2149 /* If this is dirty page, keep PagePrivate */ 2150 if (PageDirty(page)) 2151 return 0; 2152 2153 /* This is atomic written page, keep Private */ 2154 if (IS_ATOMIC_WRITTEN_PAGE(page)) 2155 return 0; 2156 2157 set_page_private(page, 0); 2158 ClearPagePrivate(page); 2159 return 1; 2160 } 2161 2162 /* 2163 * This was copied from __set_page_dirty_buffers which gives higher performance 2164 * in very high speed storages. (e.g., pmem) 2165 */ 2166 void f2fs_set_page_dirty_nobuffers(struct page *page) 2167 { 2168 struct address_space *mapping = page->mapping; 2169 unsigned long flags; 2170 2171 if (unlikely(!mapping)) 2172 return; 2173 2174 spin_lock(&mapping->private_lock); 2175 lock_page_memcg(page); 2176 SetPageDirty(page); 2177 spin_unlock(&mapping->private_lock); 2178 2179 spin_lock_irqsave(&mapping->tree_lock, flags); 2180 WARN_ON_ONCE(!PageUptodate(page)); 2181 account_page_dirtied(page, mapping); 2182 radix_tree_tag_set(&mapping->page_tree, 2183 page_index(page), PAGECACHE_TAG_DIRTY); 2184 spin_unlock_irqrestore(&mapping->tree_lock, flags); 2185 unlock_page_memcg(page); 2186 2187 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 2188 return; 2189 } 2190 2191 static int f2fs_set_data_page_dirty(struct page *page) 2192 { 2193 struct address_space *mapping = page->mapping; 2194 struct inode *inode = mapping->host; 2195 2196 trace_f2fs_set_page_dirty(page, DATA); 2197 2198 if (!PageUptodate(page)) 2199 SetPageUptodate(page); 2200 2201 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) { 2202 if (!IS_ATOMIC_WRITTEN_PAGE(page)) { 2203 register_inmem_page(inode, page); 2204 return 1; 2205 } 2206 /* 2207 * Previously, this page has been registered, we just 2208 * return here. 2209 */ 2210 return 0; 2211 } 2212 2213 if (!PageDirty(page)) { 2214 f2fs_set_page_dirty_nobuffers(page); 2215 update_dirty_page(inode, page); 2216 return 1; 2217 } 2218 return 0; 2219 } 2220 2221 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 2222 { 2223 struct inode *inode = mapping->host; 2224 2225 if (f2fs_has_inline_data(inode)) 2226 return 0; 2227 2228 /* make sure allocating whole blocks */ 2229 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 2230 filemap_write_and_wait(mapping); 2231 2232 return generic_block_bmap(mapping, block, get_data_block_bmap); 2233 } 2234 2235 #ifdef CONFIG_MIGRATION 2236 #include <linux/migrate.h> 2237 2238 int f2fs_migrate_page(struct address_space *mapping, 2239 struct page *newpage, struct page *page, enum migrate_mode mode) 2240 { 2241 int rc, extra_count; 2242 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 2243 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page); 2244 2245 BUG_ON(PageWriteback(page)); 2246 2247 /* migrating an atomic written page is safe with the inmem_lock hold */ 2248 if (atomic_written) { 2249 if (mode != MIGRATE_SYNC) 2250 return -EBUSY; 2251 if (!mutex_trylock(&fi->inmem_lock)) 2252 return -EAGAIN; 2253 } 2254 2255 /* 2256 * A reference is expected if PagePrivate set when move mapping, 2257 * however F2FS breaks this for maintaining dirty page counts when 2258 * truncating pages. So here adjusting the 'extra_count' make it work. 2259 */ 2260 extra_count = (atomic_written ? 1 : 0) - page_has_private(page); 2261 rc = migrate_page_move_mapping(mapping, newpage, 2262 page, NULL, mode, extra_count); 2263 if (rc != MIGRATEPAGE_SUCCESS) { 2264 if (atomic_written) 2265 mutex_unlock(&fi->inmem_lock); 2266 return rc; 2267 } 2268 2269 if (atomic_written) { 2270 struct inmem_pages *cur; 2271 list_for_each_entry(cur, &fi->inmem_pages, list) 2272 if (cur->page == page) { 2273 cur->page = newpage; 2274 break; 2275 } 2276 mutex_unlock(&fi->inmem_lock); 2277 put_page(page); 2278 get_page(newpage); 2279 } 2280 2281 if (PagePrivate(page)) 2282 SetPagePrivate(newpage); 2283 set_page_private(newpage, page_private(page)); 2284 2285 if (mode != MIGRATE_SYNC_NO_COPY) 2286 migrate_page_copy(newpage, page); 2287 else 2288 migrate_page_states(newpage, page); 2289 2290 return MIGRATEPAGE_SUCCESS; 2291 } 2292 #endif 2293 2294 const struct address_space_operations f2fs_dblock_aops = { 2295 .readpage = f2fs_read_data_page, 2296 .readpages = f2fs_read_data_pages, 2297 .writepage = f2fs_write_data_page, 2298 .writepages = f2fs_write_data_pages, 2299 .write_begin = f2fs_write_begin, 2300 .write_end = f2fs_write_end, 2301 .set_page_dirty = f2fs_set_data_page_dirty, 2302 .invalidatepage = f2fs_invalidate_page, 2303 .releasepage = f2fs_release_page, 2304 .direct_IO = f2fs_direct_IO, 2305 .bmap = f2fs_bmap, 2306 #ifdef CONFIG_MIGRATION 2307 .migratepage = f2fs_migrate_page, 2308 #endif 2309 }; 2310