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 if (type == F2FS_WB_CP_DATA) 115 f2fs_stop_checkpoint(sbi, true); 116 } 117 118 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && 119 page->index != nid_of_node(page)); 120 121 dec_page_count(sbi, type); 122 clear_cold_data(page); 123 end_page_writeback(page); 124 } 125 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 126 wq_has_sleeper(&sbi->cp_wait)) 127 wake_up(&sbi->cp_wait); 128 129 bio_put(bio); 130 } 131 132 /* 133 * Return true, if pre_bio's bdev is same as its target device. 134 */ 135 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 136 block_t blk_addr, struct bio *bio) 137 { 138 struct block_device *bdev = sbi->sb->s_bdev; 139 int i; 140 141 for (i = 0; i < sbi->s_ndevs; i++) { 142 if (FDEV(i).start_blk <= blk_addr && 143 FDEV(i).end_blk >= blk_addr) { 144 blk_addr -= FDEV(i).start_blk; 145 bdev = FDEV(i).bdev; 146 break; 147 } 148 } 149 if (bio) { 150 bio_set_dev(bio, bdev); 151 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); 152 } 153 return bdev; 154 } 155 156 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 157 { 158 int i; 159 160 for (i = 0; i < sbi->s_ndevs; i++) 161 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 162 return i; 163 return 0; 164 } 165 166 static bool __same_bdev(struct f2fs_sb_info *sbi, 167 block_t blk_addr, struct bio *bio) 168 { 169 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL); 170 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno; 171 } 172 173 /* 174 * Low-level block read/write IO operations. 175 */ 176 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr, 177 struct writeback_control *wbc, 178 int npages, bool is_read) 179 { 180 struct bio *bio; 181 182 bio = f2fs_bio_alloc(sbi, npages, true); 183 184 f2fs_target_device(sbi, blk_addr, bio); 185 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io; 186 bio->bi_private = is_read ? NULL : sbi; 187 if (wbc) 188 wbc_init_bio(wbc, bio); 189 190 return bio; 191 } 192 193 static inline void __submit_bio(struct f2fs_sb_info *sbi, 194 struct bio *bio, enum page_type type) 195 { 196 if (!is_read_io(bio_op(bio))) { 197 unsigned int start; 198 199 if (f2fs_sb_mounted_blkzoned(sbi->sb) && 200 current->plug && (type == DATA || type == NODE)) 201 blk_finish_plug(current->plug); 202 203 if (type != DATA && type != NODE) 204 goto submit_io; 205 206 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS; 207 start %= F2FS_IO_SIZE(sbi); 208 209 if (start == 0) 210 goto submit_io; 211 212 /* fill dummy pages */ 213 for (; start < F2FS_IO_SIZE(sbi); start++) { 214 struct page *page = 215 mempool_alloc(sbi->write_io_dummy, 216 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL); 217 f2fs_bug_on(sbi, !page); 218 219 SetPagePrivate(page); 220 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE); 221 lock_page(page); 222 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 223 f2fs_bug_on(sbi, 1); 224 } 225 /* 226 * In the NODE case, we lose next block address chain. So, we 227 * need to do checkpoint in f2fs_sync_file. 228 */ 229 if (type == NODE) 230 set_sbi_flag(sbi, SBI_NEED_CP); 231 } 232 submit_io: 233 if (is_read_io(bio_op(bio))) 234 trace_f2fs_submit_read_bio(sbi->sb, type, bio); 235 else 236 trace_f2fs_submit_write_bio(sbi->sb, type, bio); 237 submit_bio(bio); 238 } 239 240 static void __submit_merged_bio(struct f2fs_bio_info *io) 241 { 242 struct f2fs_io_info *fio = &io->fio; 243 244 if (!io->bio) 245 return; 246 247 bio_set_op_attrs(io->bio, fio->op, fio->op_flags); 248 249 if (is_read_io(fio->op)) 250 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio); 251 else 252 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio); 253 254 __submit_bio(io->sbi, io->bio, fio->type); 255 io->bio = NULL; 256 } 257 258 static bool __has_merged_page(struct f2fs_bio_info *io, 259 struct inode *inode, nid_t ino, pgoff_t idx) 260 { 261 struct bio_vec *bvec; 262 struct page *target; 263 int i; 264 265 if (!io->bio) 266 return false; 267 268 if (!inode && !ino) 269 return true; 270 271 bio_for_each_segment_all(bvec, io->bio, i) { 272 273 if (bvec->bv_page->mapping) 274 target = bvec->bv_page; 275 else 276 target = fscrypt_control_page(bvec->bv_page); 277 278 if (idx != target->index) 279 continue; 280 281 if (inode && inode == target->mapping->host) 282 return true; 283 if (ino && ino == ino_of_node(target)) 284 return true; 285 } 286 287 return false; 288 } 289 290 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode, 291 nid_t ino, pgoff_t idx, enum page_type type) 292 { 293 enum page_type btype = PAGE_TYPE_OF_BIO(type); 294 enum temp_type temp; 295 struct f2fs_bio_info *io; 296 bool ret = false; 297 298 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 299 io = sbi->write_io[btype] + temp; 300 301 down_read(&io->io_rwsem); 302 ret = __has_merged_page(io, inode, ino, idx); 303 up_read(&io->io_rwsem); 304 305 /* TODO: use HOT temp only for meta pages now. */ 306 if (ret || btype == META) 307 break; 308 } 309 return ret; 310 } 311 312 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, 313 enum page_type type, enum temp_type temp) 314 { 315 enum page_type btype = PAGE_TYPE_OF_BIO(type); 316 struct f2fs_bio_info *io = sbi->write_io[btype] + temp; 317 318 down_write(&io->io_rwsem); 319 320 /* change META to META_FLUSH in the checkpoint procedure */ 321 if (type >= META_FLUSH) { 322 io->fio.type = META_FLUSH; 323 io->fio.op = REQ_OP_WRITE; 324 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC; 325 if (!test_opt(sbi, NOBARRIER)) 326 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA; 327 } 328 __submit_merged_bio(io); 329 up_write(&io->io_rwsem); 330 } 331 332 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, 333 struct inode *inode, nid_t ino, pgoff_t idx, 334 enum page_type type, bool force) 335 { 336 enum temp_type temp; 337 338 if (!force && !has_merged_page(sbi, inode, ino, idx, type)) 339 return; 340 341 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { 342 343 __f2fs_submit_merged_write(sbi, type, temp); 344 345 /* TODO: use HOT temp only for meta pages now. */ 346 if (type >= META) 347 break; 348 } 349 } 350 351 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) 352 { 353 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true); 354 } 355 356 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, 357 struct inode *inode, nid_t ino, pgoff_t idx, 358 enum page_type type) 359 { 360 __submit_merged_write_cond(sbi, inode, ino, idx, type, false); 361 } 362 363 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) 364 { 365 f2fs_submit_merged_write(sbi, DATA); 366 f2fs_submit_merged_write(sbi, NODE); 367 f2fs_submit_merged_write(sbi, META); 368 } 369 370 /* 371 * Fill the locked page with data located in the block address. 372 * A caller needs to unlock the page on failure. 373 */ 374 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 375 { 376 struct bio *bio; 377 struct page *page = fio->encrypted_page ? 378 fio->encrypted_page : fio->page; 379 380 trace_f2fs_submit_page_bio(page, fio); 381 f2fs_trace_ios(fio, 0); 382 383 /* Allocate a new bio */ 384 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc, 385 1, is_read_io(fio->op)); 386 387 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 388 bio_put(bio); 389 return -EFAULT; 390 } 391 bio_set_op_attrs(bio, fio->op, fio->op_flags); 392 393 __submit_bio(fio->sbi, bio, fio->type); 394 395 if (!is_read_io(fio->op)) 396 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page)); 397 return 0; 398 } 399 400 int f2fs_submit_page_write(struct f2fs_io_info *fio) 401 { 402 struct f2fs_sb_info *sbi = fio->sbi; 403 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 404 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; 405 struct page *bio_page; 406 int err = 0; 407 408 f2fs_bug_on(sbi, is_read_io(fio->op)); 409 410 down_write(&io->io_rwsem); 411 next: 412 if (fio->in_list) { 413 spin_lock(&io->io_lock); 414 if (list_empty(&io->io_list)) { 415 spin_unlock(&io->io_lock); 416 goto out_fail; 417 } 418 fio = list_first_entry(&io->io_list, 419 struct f2fs_io_info, list); 420 list_del(&fio->list); 421 spin_unlock(&io->io_lock); 422 } 423 424 if (fio->old_blkaddr != NEW_ADDR) 425 verify_block_addr(sbi, fio->old_blkaddr); 426 verify_block_addr(sbi, fio->new_blkaddr); 427 428 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page; 429 430 /* set submitted = true as a return value */ 431 fio->submitted = true; 432 433 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 434 435 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 || 436 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) || 437 !__same_bdev(sbi, fio->new_blkaddr, io->bio))) 438 __submit_merged_bio(io); 439 alloc_new: 440 if (io->bio == NULL) { 441 if ((fio->type == DATA || fio->type == NODE) && 442 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { 443 err = -EAGAIN; 444 dec_page_count(sbi, WB_DATA_TYPE(bio_page)); 445 goto out_fail; 446 } 447 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc, 448 BIO_MAX_PAGES, false); 449 io->fio = *fio; 450 } 451 452 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) { 453 __submit_merged_bio(io); 454 goto alloc_new; 455 } 456 457 if (fio->io_wbc) 458 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE); 459 460 io->last_block_in_bio = fio->new_blkaddr; 461 f2fs_trace_ios(fio, 0); 462 463 trace_f2fs_submit_page_write(fio->page, fio); 464 465 if (fio->in_list) 466 goto next; 467 out_fail: 468 up_write(&io->io_rwsem); 469 return err; 470 } 471 472 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, 473 unsigned nr_pages) 474 { 475 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 476 struct fscrypt_ctx *ctx = NULL; 477 struct bio *bio; 478 479 if (f2fs_encrypted_file(inode)) { 480 ctx = fscrypt_get_ctx(inode, GFP_NOFS); 481 if (IS_ERR(ctx)) 482 return ERR_CAST(ctx); 483 484 /* wait the page to be moved by cleaning */ 485 f2fs_wait_on_block_writeback(sbi, blkaddr); 486 } 487 488 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false); 489 if (!bio) { 490 if (ctx) 491 fscrypt_release_ctx(ctx); 492 return ERR_PTR(-ENOMEM); 493 } 494 f2fs_target_device(sbi, blkaddr, bio); 495 bio->bi_end_io = f2fs_read_end_io; 496 bio->bi_private = ctx; 497 bio_set_op_attrs(bio, REQ_OP_READ, 0); 498 499 return bio; 500 } 501 502 /* This can handle encryption stuffs */ 503 static int f2fs_submit_page_read(struct inode *inode, struct page *page, 504 block_t blkaddr) 505 { 506 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1); 507 508 if (IS_ERR(bio)) 509 return PTR_ERR(bio); 510 511 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 512 bio_put(bio); 513 return -EFAULT; 514 } 515 __submit_bio(F2FS_I_SB(inode), bio, DATA); 516 return 0; 517 } 518 519 static void __set_data_blkaddr(struct dnode_of_data *dn) 520 { 521 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 522 __le32 *addr_array; 523 int base = 0; 524 525 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode)) 526 base = get_extra_isize(dn->inode); 527 528 /* Get physical address of data block */ 529 addr_array = blkaddr_in_node(rn); 530 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 531 } 532 533 /* 534 * Lock ordering for the change of data block address: 535 * ->data_page 536 * ->node_page 537 * update block addresses in the node page 538 */ 539 void set_data_blkaddr(struct dnode_of_data *dn) 540 { 541 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 542 __set_data_blkaddr(dn); 543 if (set_page_dirty(dn->node_page)) 544 dn->node_changed = true; 545 } 546 547 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 548 { 549 dn->data_blkaddr = blkaddr; 550 set_data_blkaddr(dn); 551 f2fs_update_extent_cache(dn); 552 } 553 554 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 555 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 556 { 557 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 558 int err; 559 560 if (!count) 561 return 0; 562 563 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 564 return -EPERM; 565 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 566 return err; 567 568 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 569 dn->ofs_in_node, count); 570 571 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 572 573 for (; count > 0; dn->ofs_in_node++) { 574 block_t blkaddr = datablock_addr(dn->inode, 575 dn->node_page, dn->ofs_in_node); 576 if (blkaddr == NULL_ADDR) { 577 dn->data_blkaddr = NEW_ADDR; 578 __set_data_blkaddr(dn); 579 count--; 580 } 581 } 582 583 if (set_page_dirty(dn->node_page)) 584 dn->node_changed = true; 585 return 0; 586 } 587 588 /* Should keep dn->ofs_in_node unchanged */ 589 int reserve_new_block(struct dnode_of_data *dn) 590 { 591 unsigned int ofs_in_node = dn->ofs_in_node; 592 int ret; 593 594 ret = reserve_new_blocks(dn, 1); 595 dn->ofs_in_node = ofs_in_node; 596 return ret; 597 } 598 599 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 600 { 601 bool need_put = dn->inode_page ? false : true; 602 int err; 603 604 err = get_dnode_of_data(dn, index, ALLOC_NODE); 605 if (err) 606 return err; 607 608 if (dn->data_blkaddr == NULL_ADDR) 609 err = reserve_new_block(dn); 610 if (err || need_put) 611 f2fs_put_dnode(dn); 612 return err; 613 } 614 615 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 616 { 617 struct extent_info ei = {0,0,0}; 618 struct inode *inode = dn->inode; 619 620 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 621 dn->data_blkaddr = ei.blk + index - ei.fofs; 622 return 0; 623 } 624 625 return f2fs_reserve_block(dn, index); 626 } 627 628 struct page *get_read_data_page(struct inode *inode, pgoff_t index, 629 int op_flags, bool for_write) 630 { 631 struct address_space *mapping = inode->i_mapping; 632 struct dnode_of_data dn; 633 struct page *page; 634 struct extent_info ei = {0,0,0}; 635 int err; 636 637 page = f2fs_grab_cache_page(mapping, index, for_write); 638 if (!page) 639 return ERR_PTR(-ENOMEM); 640 641 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 642 dn.data_blkaddr = ei.blk + index - ei.fofs; 643 goto got_it; 644 } 645 646 set_new_dnode(&dn, inode, NULL, NULL, 0); 647 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 648 if (err) 649 goto put_err; 650 f2fs_put_dnode(&dn); 651 652 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 653 err = -ENOENT; 654 goto put_err; 655 } 656 got_it: 657 if (PageUptodate(page)) { 658 unlock_page(page); 659 return page; 660 } 661 662 /* 663 * A new dentry page is allocated but not able to be written, since its 664 * new inode page couldn't be allocated due to -ENOSPC. 665 * In such the case, its blkaddr can be remained as NEW_ADDR. 666 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. 667 */ 668 if (dn.data_blkaddr == NEW_ADDR) { 669 zero_user_segment(page, 0, PAGE_SIZE); 670 if (!PageUptodate(page)) 671 SetPageUptodate(page); 672 unlock_page(page); 673 return page; 674 } 675 676 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr); 677 if (err) 678 goto put_err; 679 return page; 680 681 put_err: 682 f2fs_put_page(page, 1); 683 return ERR_PTR(err); 684 } 685 686 struct page *find_data_page(struct inode *inode, pgoff_t index) 687 { 688 struct address_space *mapping = inode->i_mapping; 689 struct page *page; 690 691 page = find_get_page(mapping, index); 692 if (page && PageUptodate(page)) 693 return page; 694 f2fs_put_page(page, 0); 695 696 page = get_read_data_page(inode, index, 0, false); 697 if (IS_ERR(page)) 698 return page; 699 700 if (PageUptodate(page)) 701 return page; 702 703 wait_on_page_locked(page); 704 if (unlikely(!PageUptodate(page))) { 705 f2fs_put_page(page, 0); 706 return ERR_PTR(-EIO); 707 } 708 return page; 709 } 710 711 /* 712 * If it tries to access a hole, return an error. 713 * Because, the callers, functions in dir.c and GC, should be able to know 714 * whether this page exists or not. 715 */ 716 struct page *get_lock_data_page(struct inode *inode, pgoff_t index, 717 bool for_write) 718 { 719 struct address_space *mapping = inode->i_mapping; 720 struct page *page; 721 repeat: 722 page = get_read_data_page(inode, index, 0, for_write); 723 if (IS_ERR(page)) 724 return page; 725 726 /* wait for read completion */ 727 lock_page(page); 728 if (unlikely(page->mapping != mapping)) { 729 f2fs_put_page(page, 1); 730 goto repeat; 731 } 732 if (unlikely(!PageUptodate(page))) { 733 f2fs_put_page(page, 1); 734 return ERR_PTR(-EIO); 735 } 736 return page; 737 } 738 739 /* 740 * Caller ensures that this data page is never allocated. 741 * A new zero-filled data page is allocated in the page cache. 742 * 743 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 744 * f2fs_unlock_op(). 745 * Note that, ipage is set only by make_empty_dir, and if any error occur, 746 * ipage should be released by this function. 747 */ 748 struct page *get_new_data_page(struct inode *inode, 749 struct page *ipage, pgoff_t index, bool new_i_size) 750 { 751 struct address_space *mapping = inode->i_mapping; 752 struct page *page; 753 struct dnode_of_data dn; 754 int err; 755 756 page = f2fs_grab_cache_page(mapping, index, true); 757 if (!page) { 758 /* 759 * before exiting, we should make sure ipage will be released 760 * if any error occur. 761 */ 762 f2fs_put_page(ipage, 1); 763 return ERR_PTR(-ENOMEM); 764 } 765 766 set_new_dnode(&dn, inode, ipage, NULL, 0); 767 err = f2fs_reserve_block(&dn, index); 768 if (err) { 769 f2fs_put_page(page, 1); 770 return ERR_PTR(err); 771 } 772 if (!ipage) 773 f2fs_put_dnode(&dn); 774 775 if (PageUptodate(page)) 776 goto got_it; 777 778 if (dn.data_blkaddr == NEW_ADDR) { 779 zero_user_segment(page, 0, PAGE_SIZE); 780 if (!PageUptodate(page)) 781 SetPageUptodate(page); 782 } else { 783 f2fs_put_page(page, 1); 784 785 /* if ipage exists, blkaddr should be NEW_ADDR */ 786 f2fs_bug_on(F2FS_I_SB(inode), ipage); 787 page = get_lock_data_page(inode, index, true); 788 if (IS_ERR(page)) 789 return page; 790 } 791 got_it: 792 if (new_i_size && i_size_read(inode) < 793 ((loff_t)(index + 1) << PAGE_SHIFT)) 794 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 795 return page; 796 } 797 798 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) 799 { 800 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 801 struct f2fs_summary sum; 802 struct node_info ni; 803 pgoff_t fofs; 804 blkcnt_t count = 1; 805 int err; 806 807 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 808 return -EPERM; 809 810 dn->data_blkaddr = datablock_addr(dn->inode, 811 dn->node_page, dn->ofs_in_node); 812 if (dn->data_blkaddr == NEW_ADDR) 813 goto alloc; 814 815 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) 816 return err; 817 818 alloc: 819 get_node_info(sbi, dn->nid, &ni); 820 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 821 822 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr, 823 &sum, seg_type, NULL, false); 824 set_data_blkaddr(dn); 825 826 /* update i_size */ 827 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) + 828 dn->ofs_in_node; 829 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT)) 830 f2fs_i_size_write(dn->inode, 831 ((loff_t)(fofs + 1) << PAGE_SHIFT)); 832 return 0; 833 } 834 835 static inline bool __force_buffered_io(struct inode *inode, int rw) 836 { 837 return (f2fs_encrypted_file(inode) || 838 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) || 839 F2FS_I_SB(inode)->s_ndevs); 840 } 841 842 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from) 843 { 844 struct inode *inode = file_inode(iocb->ki_filp); 845 struct f2fs_map_blocks map; 846 int flag; 847 int err = 0; 848 bool direct_io = iocb->ki_flags & IOCB_DIRECT; 849 850 /* convert inline data for Direct I/O*/ 851 if (direct_io) { 852 err = f2fs_convert_inline_inode(inode); 853 if (err) 854 return err; 855 } 856 857 if (is_inode_flag_set(inode, FI_NO_PREALLOC)) 858 return 0; 859 860 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos); 861 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from)); 862 if (map.m_len > map.m_lblk) 863 map.m_len -= map.m_lblk; 864 else 865 map.m_len = 0; 866 867 map.m_next_pgofs = NULL; 868 map.m_seg_type = NO_CHECK_TYPE; 869 870 if (direct_io) { 871 map.m_seg_type = rw_hint_to_seg_type(iocb->ki_hint); 872 flag = __force_buffered_io(inode, WRITE) ? 873 F2FS_GET_BLOCK_PRE_AIO : 874 F2FS_GET_BLOCK_PRE_DIO; 875 goto map_blocks; 876 } 877 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) { 878 err = f2fs_convert_inline_inode(inode); 879 if (err) 880 return err; 881 } 882 if (f2fs_has_inline_data(inode)) 883 return err; 884 885 flag = F2FS_GET_BLOCK_PRE_AIO; 886 887 map_blocks: 888 err = f2fs_map_blocks(inode, &map, 1, flag); 889 if (map.m_len > 0 && err == -ENOSPC) { 890 if (!direct_io) 891 set_inode_flag(inode, FI_NO_PREALLOC); 892 err = 0; 893 } 894 return err; 895 } 896 897 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock) 898 { 899 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 900 if (lock) 901 down_read(&sbi->node_change); 902 else 903 up_read(&sbi->node_change); 904 } else { 905 if (lock) 906 f2fs_lock_op(sbi); 907 else 908 f2fs_unlock_op(sbi); 909 } 910 } 911 912 /* 913 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with 914 * f2fs_map_blocks structure. 915 * If original data blocks are allocated, then give them to blockdev. 916 * Otherwise, 917 * a. preallocate requested block addresses 918 * b. do not use extent cache for better performance 919 * c. give the block addresses to blockdev 920 */ 921 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 922 int create, int flag) 923 { 924 unsigned int maxblocks = map->m_len; 925 struct dnode_of_data dn; 926 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 927 int mode = create ? ALLOC_NODE : LOOKUP_NODE; 928 pgoff_t pgofs, end_offset, end; 929 int err = 0, ofs = 1; 930 unsigned int ofs_in_node, last_ofs_in_node; 931 blkcnt_t prealloc; 932 struct extent_info ei = {0,0,0}; 933 block_t blkaddr; 934 935 if (!maxblocks) 936 return 0; 937 938 map->m_len = 0; 939 map->m_flags = 0; 940 941 /* it only supports block size == page size */ 942 pgofs = (pgoff_t)map->m_lblk; 943 end = pgofs + maxblocks; 944 945 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 946 map->m_pblk = ei.blk + pgofs - ei.fofs; 947 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 948 map->m_flags = F2FS_MAP_MAPPED; 949 goto out; 950 } 951 952 next_dnode: 953 if (create) 954 __do_map_lock(sbi, flag, true); 955 956 /* When reading holes, we need its node page */ 957 set_new_dnode(&dn, inode, NULL, NULL, 0); 958 err = get_dnode_of_data(&dn, pgofs, mode); 959 if (err) { 960 if (flag == F2FS_GET_BLOCK_BMAP) 961 map->m_pblk = 0; 962 if (err == -ENOENT) { 963 err = 0; 964 if (map->m_next_pgofs) 965 *map->m_next_pgofs = 966 get_next_page_offset(&dn, pgofs); 967 } 968 goto unlock_out; 969 } 970 971 prealloc = 0; 972 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 973 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 974 975 next_block: 976 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node); 977 978 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) { 979 if (create) { 980 if (unlikely(f2fs_cp_error(sbi))) { 981 err = -EIO; 982 goto sync_out; 983 } 984 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 985 if (blkaddr == NULL_ADDR) { 986 prealloc++; 987 last_ofs_in_node = dn.ofs_in_node; 988 } 989 } else { 990 err = __allocate_data_block(&dn, 991 map->m_seg_type); 992 if (!err) 993 set_inode_flag(inode, FI_APPEND_WRITE); 994 } 995 if (err) 996 goto sync_out; 997 map->m_flags |= F2FS_MAP_NEW; 998 blkaddr = dn.data_blkaddr; 999 } else { 1000 if (flag == F2FS_GET_BLOCK_BMAP) { 1001 map->m_pblk = 0; 1002 goto sync_out; 1003 } 1004 if (flag == F2FS_GET_BLOCK_FIEMAP && 1005 blkaddr == NULL_ADDR) { 1006 if (map->m_next_pgofs) 1007 *map->m_next_pgofs = pgofs + 1; 1008 goto sync_out; 1009 } 1010 if (flag != F2FS_GET_BLOCK_FIEMAP) 1011 goto sync_out; 1012 } 1013 } 1014 1015 if (flag == F2FS_GET_BLOCK_PRE_AIO) 1016 goto skip; 1017 1018 if (map->m_len == 0) { 1019 /* preallocated unwritten block should be mapped for fiemap. */ 1020 if (blkaddr == NEW_ADDR) 1021 map->m_flags |= F2FS_MAP_UNWRITTEN; 1022 map->m_flags |= F2FS_MAP_MAPPED; 1023 1024 map->m_pblk = blkaddr; 1025 map->m_len = 1; 1026 } else if ((map->m_pblk != NEW_ADDR && 1027 blkaddr == (map->m_pblk + ofs)) || 1028 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 1029 flag == F2FS_GET_BLOCK_PRE_DIO) { 1030 ofs++; 1031 map->m_len++; 1032 } else { 1033 goto sync_out; 1034 } 1035 1036 skip: 1037 dn.ofs_in_node++; 1038 pgofs++; 1039 1040 /* preallocate blocks in batch for one dnode page */ 1041 if (flag == F2FS_GET_BLOCK_PRE_AIO && 1042 (pgofs == end || dn.ofs_in_node == end_offset)) { 1043 1044 dn.ofs_in_node = ofs_in_node; 1045 err = reserve_new_blocks(&dn, prealloc); 1046 if (err) 1047 goto sync_out; 1048 1049 map->m_len += dn.ofs_in_node - ofs_in_node; 1050 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 1051 err = -ENOSPC; 1052 goto sync_out; 1053 } 1054 dn.ofs_in_node = end_offset; 1055 } 1056 1057 if (pgofs >= end) 1058 goto sync_out; 1059 else if (dn.ofs_in_node < end_offset) 1060 goto next_block; 1061 1062 f2fs_put_dnode(&dn); 1063 1064 if (create) { 1065 __do_map_lock(sbi, flag, false); 1066 f2fs_balance_fs(sbi, dn.node_changed); 1067 } 1068 goto next_dnode; 1069 1070 sync_out: 1071 f2fs_put_dnode(&dn); 1072 unlock_out: 1073 if (create) { 1074 __do_map_lock(sbi, flag, false); 1075 f2fs_balance_fs(sbi, dn.node_changed); 1076 } 1077 out: 1078 trace_f2fs_map_blocks(inode, map, err); 1079 return err; 1080 } 1081 1082 static int __get_data_block(struct inode *inode, sector_t iblock, 1083 struct buffer_head *bh, int create, int flag, 1084 pgoff_t *next_pgofs, int seg_type) 1085 { 1086 struct f2fs_map_blocks map; 1087 int err; 1088 1089 map.m_lblk = iblock; 1090 map.m_len = bh->b_size >> inode->i_blkbits; 1091 map.m_next_pgofs = next_pgofs; 1092 map.m_seg_type = seg_type; 1093 1094 err = f2fs_map_blocks(inode, &map, create, flag); 1095 if (!err) { 1096 map_bh(bh, inode->i_sb, map.m_pblk); 1097 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags; 1098 bh->b_size = (u64)map.m_len << inode->i_blkbits; 1099 } 1100 return err; 1101 } 1102 1103 static int get_data_block(struct inode *inode, sector_t iblock, 1104 struct buffer_head *bh_result, int create, int flag, 1105 pgoff_t *next_pgofs) 1106 { 1107 return __get_data_block(inode, iblock, bh_result, create, 1108 flag, next_pgofs, 1109 NO_CHECK_TYPE); 1110 } 1111 1112 static int get_data_block_dio(struct inode *inode, sector_t iblock, 1113 struct buffer_head *bh_result, int create) 1114 { 1115 return __get_data_block(inode, iblock, bh_result, create, 1116 F2FS_GET_BLOCK_DEFAULT, NULL, 1117 rw_hint_to_seg_type( 1118 inode->i_write_hint)); 1119 } 1120 1121 static int get_data_block_bmap(struct inode *inode, sector_t iblock, 1122 struct buffer_head *bh_result, int create) 1123 { 1124 /* Block number less than F2FS MAX BLOCKS */ 1125 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks)) 1126 return -EFBIG; 1127 1128 return __get_data_block(inode, iblock, bh_result, create, 1129 F2FS_GET_BLOCK_BMAP, NULL, 1130 NO_CHECK_TYPE); 1131 } 1132 1133 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset) 1134 { 1135 return (offset >> inode->i_blkbits); 1136 } 1137 1138 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk) 1139 { 1140 return (blk << inode->i_blkbits); 1141 } 1142 1143 static int f2fs_xattr_fiemap(struct inode *inode, 1144 struct fiemap_extent_info *fieinfo) 1145 { 1146 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1147 struct page *page; 1148 struct node_info ni; 1149 __u64 phys = 0, len; 1150 __u32 flags; 1151 nid_t xnid = F2FS_I(inode)->i_xattr_nid; 1152 int err = 0; 1153 1154 if (f2fs_has_inline_xattr(inode)) { 1155 int offset; 1156 1157 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), 1158 inode->i_ino, false); 1159 if (!page) 1160 return -ENOMEM; 1161 1162 get_node_info(sbi, inode->i_ino, &ni); 1163 1164 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1165 offset = offsetof(struct f2fs_inode, i_addr) + 1166 sizeof(__le32) * (DEF_ADDRS_PER_INODE - 1167 F2FS_INLINE_XATTR_ADDRS(inode)); 1168 1169 phys += offset; 1170 len = inline_xattr_size(inode); 1171 1172 f2fs_put_page(page, 1); 1173 1174 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; 1175 1176 if (!xnid) 1177 flags |= FIEMAP_EXTENT_LAST; 1178 1179 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1180 if (err || err == 1) 1181 return err; 1182 } 1183 1184 if (xnid) { 1185 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false); 1186 if (!page) 1187 return -ENOMEM; 1188 1189 get_node_info(sbi, xnid, &ni); 1190 1191 phys = (__u64)blk_to_logical(inode, ni.blk_addr); 1192 len = inode->i_sb->s_blocksize; 1193 1194 f2fs_put_page(page, 1); 1195 1196 flags = FIEMAP_EXTENT_LAST; 1197 } 1198 1199 if (phys) 1200 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags); 1201 1202 return (err < 0 ? err : 0); 1203 } 1204 1205 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 1206 u64 start, u64 len) 1207 { 1208 struct buffer_head map_bh; 1209 sector_t start_blk, last_blk; 1210 pgoff_t next_pgofs; 1211 u64 logical = 0, phys = 0, size = 0; 1212 u32 flags = 0; 1213 int ret = 0; 1214 1215 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR); 1216 if (ret) 1217 return ret; 1218 1219 inode_lock(inode); 1220 1221 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1222 ret = f2fs_xattr_fiemap(inode, fieinfo); 1223 goto out; 1224 } 1225 1226 if (f2fs_has_inline_data(inode)) { 1227 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 1228 if (ret != -EAGAIN) 1229 goto out; 1230 } 1231 1232 if (logical_to_blk(inode, len) == 0) 1233 len = blk_to_logical(inode, 1); 1234 1235 start_blk = logical_to_blk(inode, start); 1236 last_blk = logical_to_blk(inode, start + len - 1); 1237 1238 next: 1239 memset(&map_bh, 0, sizeof(struct buffer_head)); 1240 map_bh.b_size = len; 1241 1242 ret = get_data_block(inode, start_blk, &map_bh, 0, 1243 F2FS_GET_BLOCK_FIEMAP, &next_pgofs); 1244 if (ret) 1245 goto out; 1246 1247 /* HOLE */ 1248 if (!buffer_mapped(&map_bh)) { 1249 start_blk = next_pgofs; 1250 1251 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode, 1252 F2FS_I_SB(inode)->max_file_blocks)) 1253 goto prep_next; 1254 1255 flags |= FIEMAP_EXTENT_LAST; 1256 } 1257 1258 if (size) { 1259 if (f2fs_encrypted_inode(inode)) 1260 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1261 1262 ret = fiemap_fill_next_extent(fieinfo, logical, 1263 phys, size, flags); 1264 } 1265 1266 if (start_blk > last_blk || ret) 1267 goto out; 1268 1269 logical = blk_to_logical(inode, start_blk); 1270 phys = blk_to_logical(inode, map_bh.b_blocknr); 1271 size = map_bh.b_size; 1272 flags = 0; 1273 if (buffer_unwritten(&map_bh)) 1274 flags = FIEMAP_EXTENT_UNWRITTEN; 1275 1276 start_blk += logical_to_blk(inode, size); 1277 1278 prep_next: 1279 cond_resched(); 1280 if (fatal_signal_pending(current)) 1281 ret = -EINTR; 1282 else 1283 goto next; 1284 out: 1285 if (ret == 1) 1286 ret = 0; 1287 1288 inode_unlock(inode); 1289 return ret; 1290 } 1291 1292 /* 1293 * This function was originally taken from fs/mpage.c, and customized for f2fs. 1294 * Major change was from block_size == page_size in f2fs by default. 1295 */ 1296 static int f2fs_mpage_readpages(struct address_space *mapping, 1297 struct list_head *pages, struct page *page, 1298 unsigned nr_pages) 1299 { 1300 struct bio *bio = NULL; 1301 sector_t last_block_in_bio = 0; 1302 struct inode *inode = mapping->host; 1303 const unsigned blkbits = inode->i_blkbits; 1304 const unsigned blocksize = 1 << blkbits; 1305 sector_t block_in_file; 1306 sector_t last_block; 1307 sector_t last_block_in_file; 1308 sector_t block_nr; 1309 struct f2fs_map_blocks map; 1310 1311 map.m_pblk = 0; 1312 map.m_lblk = 0; 1313 map.m_len = 0; 1314 map.m_flags = 0; 1315 map.m_next_pgofs = NULL; 1316 map.m_seg_type = NO_CHECK_TYPE; 1317 1318 for (; nr_pages; nr_pages--) { 1319 if (pages) { 1320 page = list_last_entry(pages, struct page, lru); 1321 1322 prefetchw(&page->flags); 1323 list_del(&page->lru); 1324 if (add_to_page_cache_lru(page, mapping, 1325 page->index, 1326 readahead_gfp_mask(mapping))) 1327 goto next_page; 1328 } 1329 1330 block_in_file = (sector_t)page->index; 1331 last_block = block_in_file + nr_pages; 1332 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> 1333 blkbits; 1334 if (last_block > last_block_in_file) 1335 last_block = last_block_in_file; 1336 1337 /* 1338 * Map blocks using the previous result first. 1339 */ 1340 if ((map.m_flags & F2FS_MAP_MAPPED) && 1341 block_in_file > map.m_lblk && 1342 block_in_file < (map.m_lblk + map.m_len)) 1343 goto got_it; 1344 1345 /* 1346 * Then do more f2fs_map_blocks() calls until we are 1347 * done with this page. 1348 */ 1349 map.m_flags = 0; 1350 1351 if (block_in_file < last_block) { 1352 map.m_lblk = block_in_file; 1353 map.m_len = last_block - block_in_file; 1354 1355 if (f2fs_map_blocks(inode, &map, 0, 1356 F2FS_GET_BLOCK_DEFAULT)) 1357 goto set_error_page; 1358 } 1359 got_it: 1360 if ((map.m_flags & F2FS_MAP_MAPPED)) { 1361 block_nr = map.m_pblk + block_in_file - map.m_lblk; 1362 SetPageMappedToDisk(page); 1363 1364 if (!PageUptodate(page) && !cleancache_get_page(page)) { 1365 SetPageUptodate(page); 1366 goto confused; 1367 } 1368 } else { 1369 zero_user_segment(page, 0, PAGE_SIZE); 1370 if (!PageUptodate(page)) 1371 SetPageUptodate(page); 1372 unlock_page(page); 1373 goto next_page; 1374 } 1375 1376 /* 1377 * This page will go to BIO. Do we need to send this 1378 * BIO off first? 1379 */ 1380 if (bio && (last_block_in_bio != block_nr - 1 || 1381 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) { 1382 submit_and_realloc: 1383 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1384 bio = NULL; 1385 } 1386 if (bio == NULL) { 1387 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages); 1388 if (IS_ERR(bio)) { 1389 bio = NULL; 1390 goto set_error_page; 1391 } 1392 } 1393 1394 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 1395 goto submit_and_realloc; 1396 1397 last_block_in_bio = block_nr; 1398 goto next_page; 1399 set_error_page: 1400 SetPageError(page); 1401 zero_user_segment(page, 0, PAGE_SIZE); 1402 unlock_page(page); 1403 goto next_page; 1404 confused: 1405 if (bio) { 1406 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1407 bio = NULL; 1408 } 1409 unlock_page(page); 1410 next_page: 1411 if (pages) 1412 put_page(page); 1413 } 1414 BUG_ON(pages && !list_empty(pages)); 1415 if (bio) 1416 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1417 return 0; 1418 } 1419 1420 static int f2fs_read_data_page(struct file *file, struct page *page) 1421 { 1422 struct inode *inode = page->mapping->host; 1423 int ret = -EAGAIN; 1424 1425 trace_f2fs_readpage(page, DATA); 1426 1427 /* If the file has inline data, try to read it directly */ 1428 if (f2fs_has_inline_data(inode)) 1429 ret = f2fs_read_inline_data(inode, page); 1430 if (ret == -EAGAIN) 1431 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1); 1432 return ret; 1433 } 1434 1435 static int f2fs_read_data_pages(struct file *file, 1436 struct address_space *mapping, 1437 struct list_head *pages, unsigned nr_pages) 1438 { 1439 struct inode *inode = mapping->host; 1440 struct page *page = list_last_entry(pages, struct page, lru); 1441 1442 trace_f2fs_readpages(inode, page, nr_pages); 1443 1444 /* If the file has inline data, skip readpages */ 1445 if (f2fs_has_inline_data(inode)) 1446 return 0; 1447 1448 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages); 1449 } 1450 1451 static int encrypt_one_page(struct f2fs_io_info *fio) 1452 { 1453 struct inode *inode = fio->page->mapping->host; 1454 gfp_t gfp_flags = GFP_NOFS; 1455 1456 if (!f2fs_encrypted_file(inode)) 1457 return 0; 1458 1459 /* wait for GCed encrypted page writeback */ 1460 f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr); 1461 1462 retry_encrypt: 1463 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page, 1464 PAGE_SIZE, 0, fio->page->index, gfp_flags); 1465 if (!IS_ERR(fio->encrypted_page)) 1466 return 0; 1467 1468 /* flush pending IOs and wait for a while in the ENOMEM case */ 1469 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) { 1470 f2fs_flush_merged_writes(fio->sbi); 1471 congestion_wait(BLK_RW_ASYNC, HZ/50); 1472 gfp_flags |= __GFP_NOFAIL; 1473 goto retry_encrypt; 1474 } 1475 return PTR_ERR(fio->encrypted_page); 1476 } 1477 1478 static inline bool need_inplace_update(struct f2fs_io_info *fio) 1479 { 1480 struct inode *inode = fio->page->mapping->host; 1481 1482 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode)) 1483 return false; 1484 if (is_cold_data(fio->page)) 1485 return false; 1486 if (IS_ATOMIC_WRITTEN_PAGE(fio->page)) 1487 return false; 1488 1489 return need_inplace_update_policy(inode, fio); 1490 } 1491 1492 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio) 1493 { 1494 if (fio->old_blkaddr == NEW_ADDR) 1495 return false; 1496 if (fio->old_blkaddr == NULL_ADDR) 1497 return false; 1498 return true; 1499 } 1500 1501 int do_write_data_page(struct f2fs_io_info *fio) 1502 { 1503 struct page *page = fio->page; 1504 struct inode *inode = page->mapping->host; 1505 struct dnode_of_data dn; 1506 struct extent_info ei = {0,0,0}; 1507 bool ipu_force = false; 1508 int err = 0; 1509 1510 set_new_dnode(&dn, inode, NULL, NULL, 0); 1511 if (need_inplace_update(fio) && 1512 f2fs_lookup_extent_cache(inode, page->index, &ei)) { 1513 fio->old_blkaddr = ei.blk + page->index - ei.fofs; 1514 1515 if (valid_ipu_blkaddr(fio)) { 1516 ipu_force = true; 1517 fio->need_lock = LOCK_DONE; 1518 goto got_it; 1519 } 1520 } 1521 1522 /* Deadlock due to between page->lock and f2fs_lock_op */ 1523 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi)) 1524 return -EAGAIN; 1525 1526 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 1527 if (err) 1528 goto out; 1529 1530 fio->old_blkaddr = dn.data_blkaddr; 1531 1532 /* This page is already truncated */ 1533 if (fio->old_blkaddr == NULL_ADDR) { 1534 ClearPageUptodate(page); 1535 goto out_writepage; 1536 } 1537 got_it: 1538 /* 1539 * If current allocation needs SSR, 1540 * it had better in-place writes for updated data. 1541 */ 1542 if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) { 1543 err = encrypt_one_page(fio); 1544 if (err) 1545 goto out_writepage; 1546 1547 set_page_writeback(page); 1548 f2fs_put_dnode(&dn); 1549 if (fio->need_lock == LOCK_REQ) 1550 f2fs_unlock_op(fio->sbi); 1551 err = rewrite_data_page(fio); 1552 trace_f2fs_do_write_data_page(fio->page, IPU); 1553 set_inode_flag(inode, FI_UPDATE_WRITE); 1554 return err; 1555 } 1556 1557 if (fio->need_lock == LOCK_RETRY) { 1558 if (!f2fs_trylock_op(fio->sbi)) { 1559 err = -EAGAIN; 1560 goto out_writepage; 1561 } 1562 fio->need_lock = LOCK_REQ; 1563 } 1564 1565 err = encrypt_one_page(fio); 1566 if (err) 1567 goto out_writepage; 1568 1569 set_page_writeback(page); 1570 1571 /* LFS mode write path */ 1572 write_data_page(&dn, fio); 1573 trace_f2fs_do_write_data_page(page, OPU); 1574 set_inode_flag(inode, FI_APPEND_WRITE); 1575 if (page->index == 0) 1576 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 1577 out_writepage: 1578 f2fs_put_dnode(&dn); 1579 out: 1580 if (fio->need_lock == LOCK_REQ) 1581 f2fs_unlock_op(fio->sbi); 1582 return err; 1583 } 1584 1585 static int __write_data_page(struct page *page, bool *submitted, 1586 struct writeback_control *wbc, 1587 enum iostat_type io_type) 1588 { 1589 struct inode *inode = page->mapping->host; 1590 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1591 loff_t i_size = i_size_read(inode); 1592 const pgoff_t end_index = ((unsigned long long) i_size) 1593 >> PAGE_SHIFT; 1594 loff_t psize = (page->index + 1) << PAGE_SHIFT; 1595 unsigned offset = 0; 1596 bool need_balance_fs = false; 1597 int err = 0; 1598 struct f2fs_io_info fio = { 1599 .sbi = sbi, 1600 .ino = inode->i_ino, 1601 .type = DATA, 1602 .op = REQ_OP_WRITE, 1603 .op_flags = wbc_to_write_flags(wbc), 1604 .old_blkaddr = NULL_ADDR, 1605 .page = page, 1606 .encrypted_page = NULL, 1607 .submitted = false, 1608 .need_lock = LOCK_RETRY, 1609 .io_type = io_type, 1610 .io_wbc = wbc, 1611 }; 1612 1613 trace_f2fs_writepage(page, DATA); 1614 1615 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1616 goto redirty_out; 1617 1618 if (page->index < end_index) 1619 goto write; 1620 1621 /* 1622 * If the offset is out-of-range of file size, 1623 * this page does not have to be written to disk. 1624 */ 1625 offset = i_size & (PAGE_SIZE - 1); 1626 if ((page->index >= end_index + 1) || !offset) 1627 goto out; 1628 1629 zero_user_segment(page, offset, PAGE_SIZE); 1630 write: 1631 if (f2fs_is_drop_cache(inode)) 1632 goto out; 1633 /* we should not write 0'th page having journal header */ 1634 if (f2fs_is_volatile_file(inode) && (!page->index || 1635 (!wbc->for_reclaim && 1636 available_free_memory(sbi, BASE_CHECK)))) 1637 goto redirty_out; 1638 1639 /* we should bypass data pages to proceed the kworkder jobs */ 1640 if (unlikely(f2fs_cp_error(sbi))) { 1641 mapping_set_error(page->mapping, -EIO); 1642 goto out; 1643 } 1644 1645 /* Dentry blocks are controlled by checkpoint */ 1646 if (S_ISDIR(inode->i_mode)) { 1647 fio.need_lock = LOCK_DONE; 1648 err = do_write_data_page(&fio); 1649 goto done; 1650 } 1651 1652 if (!wbc->for_reclaim) 1653 need_balance_fs = true; 1654 else if (has_not_enough_free_secs(sbi, 0, 0)) 1655 goto redirty_out; 1656 else 1657 set_inode_flag(inode, FI_HOT_DATA); 1658 1659 err = -EAGAIN; 1660 if (f2fs_has_inline_data(inode)) { 1661 err = f2fs_write_inline_data(inode, page); 1662 if (!err) 1663 goto out; 1664 } 1665 1666 if (err == -EAGAIN) { 1667 err = do_write_data_page(&fio); 1668 if (err == -EAGAIN) { 1669 fio.need_lock = LOCK_REQ; 1670 err = do_write_data_page(&fio); 1671 } 1672 } 1673 1674 down_write(&F2FS_I(inode)->i_sem); 1675 if (F2FS_I(inode)->last_disk_size < psize) 1676 F2FS_I(inode)->last_disk_size = psize; 1677 up_write(&F2FS_I(inode)->i_sem); 1678 1679 done: 1680 if (err && err != -ENOENT) 1681 goto redirty_out; 1682 1683 out: 1684 inode_dec_dirty_pages(inode); 1685 if (err) 1686 ClearPageUptodate(page); 1687 1688 if (wbc->for_reclaim) { 1689 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA); 1690 clear_inode_flag(inode, FI_HOT_DATA); 1691 remove_dirty_inode(inode); 1692 submitted = NULL; 1693 } 1694 1695 unlock_page(page); 1696 if (!S_ISDIR(inode->i_mode)) 1697 f2fs_balance_fs(sbi, need_balance_fs); 1698 1699 if (unlikely(f2fs_cp_error(sbi))) { 1700 f2fs_submit_merged_write(sbi, DATA); 1701 submitted = NULL; 1702 } 1703 1704 if (submitted) 1705 *submitted = fio.submitted; 1706 1707 return 0; 1708 1709 redirty_out: 1710 redirty_page_for_writepage(wbc, page); 1711 if (!err) 1712 return AOP_WRITEPAGE_ACTIVATE; 1713 unlock_page(page); 1714 return err; 1715 } 1716 1717 static int f2fs_write_data_page(struct page *page, 1718 struct writeback_control *wbc) 1719 { 1720 return __write_data_page(page, NULL, wbc, FS_DATA_IO); 1721 } 1722 1723 /* 1724 * This function was copied from write_cche_pages from mm/page-writeback.c. 1725 * The major change is making write step of cold data page separately from 1726 * warm/hot data page. 1727 */ 1728 static int f2fs_write_cache_pages(struct address_space *mapping, 1729 struct writeback_control *wbc, 1730 enum iostat_type io_type) 1731 { 1732 int ret = 0; 1733 int done = 0; 1734 struct pagevec pvec; 1735 int nr_pages; 1736 pgoff_t uninitialized_var(writeback_index); 1737 pgoff_t index; 1738 pgoff_t end; /* Inclusive */ 1739 pgoff_t done_index; 1740 pgoff_t last_idx = ULONG_MAX; 1741 int cycled; 1742 int range_whole = 0; 1743 int tag; 1744 1745 pagevec_init(&pvec); 1746 1747 if (get_dirty_pages(mapping->host) <= 1748 SM_I(F2FS_M_SB(mapping))->min_hot_blocks) 1749 set_inode_flag(mapping->host, FI_HOT_DATA); 1750 else 1751 clear_inode_flag(mapping->host, FI_HOT_DATA); 1752 1753 if (wbc->range_cyclic) { 1754 writeback_index = mapping->writeback_index; /* prev offset */ 1755 index = writeback_index; 1756 if (index == 0) 1757 cycled = 1; 1758 else 1759 cycled = 0; 1760 end = -1; 1761 } else { 1762 index = wbc->range_start >> PAGE_SHIFT; 1763 end = wbc->range_end >> PAGE_SHIFT; 1764 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 1765 range_whole = 1; 1766 cycled = 1; /* ignore range_cyclic tests */ 1767 } 1768 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 1769 tag = PAGECACHE_TAG_TOWRITE; 1770 else 1771 tag = PAGECACHE_TAG_DIRTY; 1772 retry: 1773 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 1774 tag_pages_for_writeback(mapping, index, end); 1775 done_index = index; 1776 while (!done && (index <= end)) { 1777 int i; 1778 1779 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 1780 tag); 1781 if (nr_pages == 0) 1782 break; 1783 1784 for (i = 0; i < nr_pages; i++) { 1785 struct page *page = pvec.pages[i]; 1786 bool submitted = false; 1787 1788 done_index = page->index; 1789 retry_write: 1790 lock_page(page); 1791 1792 if (unlikely(page->mapping != mapping)) { 1793 continue_unlock: 1794 unlock_page(page); 1795 continue; 1796 } 1797 1798 if (!PageDirty(page)) { 1799 /* someone wrote it for us */ 1800 goto continue_unlock; 1801 } 1802 1803 if (PageWriteback(page)) { 1804 if (wbc->sync_mode != WB_SYNC_NONE) 1805 f2fs_wait_on_page_writeback(page, 1806 DATA, true); 1807 else 1808 goto continue_unlock; 1809 } 1810 1811 BUG_ON(PageWriteback(page)); 1812 if (!clear_page_dirty_for_io(page)) 1813 goto continue_unlock; 1814 1815 ret = __write_data_page(page, &submitted, wbc, io_type); 1816 if (unlikely(ret)) { 1817 /* 1818 * keep nr_to_write, since vfs uses this to 1819 * get # of written pages. 1820 */ 1821 if (ret == AOP_WRITEPAGE_ACTIVATE) { 1822 unlock_page(page); 1823 ret = 0; 1824 continue; 1825 } else if (ret == -EAGAIN) { 1826 ret = 0; 1827 if (wbc->sync_mode == WB_SYNC_ALL) { 1828 cond_resched(); 1829 congestion_wait(BLK_RW_ASYNC, 1830 HZ/50); 1831 goto retry_write; 1832 } 1833 continue; 1834 } 1835 done_index = page->index + 1; 1836 done = 1; 1837 break; 1838 } else if (submitted) { 1839 last_idx = page->index; 1840 } 1841 1842 /* give a priority to WB_SYNC threads */ 1843 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) || 1844 --wbc->nr_to_write <= 0) && 1845 wbc->sync_mode == WB_SYNC_NONE) { 1846 done = 1; 1847 break; 1848 } 1849 } 1850 pagevec_release(&pvec); 1851 cond_resched(); 1852 } 1853 1854 if (!cycled && !done) { 1855 cycled = 1; 1856 index = 0; 1857 end = writeback_index - 1; 1858 goto retry; 1859 } 1860 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 1861 mapping->writeback_index = done_index; 1862 1863 if (last_idx != ULONG_MAX) 1864 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host, 1865 0, last_idx, DATA); 1866 1867 return ret; 1868 } 1869 1870 int __f2fs_write_data_pages(struct address_space *mapping, 1871 struct writeback_control *wbc, 1872 enum iostat_type io_type) 1873 { 1874 struct inode *inode = mapping->host; 1875 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1876 struct blk_plug plug; 1877 int ret; 1878 1879 /* deal with chardevs and other special file */ 1880 if (!mapping->a_ops->writepage) 1881 return 0; 1882 1883 /* skip writing if there is no dirty page in this inode */ 1884 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 1885 return 0; 1886 1887 /* during POR, we don't need to trigger writepage at all. */ 1888 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1889 goto skip_write; 1890 1891 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE && 1892 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 1893 available_free_memory(sbi, DIRTY_DENTS)) 1894 goto skip_write; 1895 1896 /* skip writing during file defragment */ 1897 if (is_inode_flag_set(inode, FI_DO_DEFRAG)) 1898 goto skip_write; 1899 1900 trace_f2fs_writepages(mapping->host, wbc, DATA); 1901 1902 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ 1903 if (wbc->sync_mode == WB_SYNC_ALL) 1904 atomic_inc(&sbi->wb_sync_req); 1905 else if (atomic_read(&sbi->wb_sync_req)) 1906 goto skip_write; 1907 1908 blk_start_plug(&plug); 1909 ret = f2fs_write_cache_pages(mapping, wbc, io_type); 1910 blk_finish_plug(&plug); 1911 1912 if (wbc->sync_mode == WB_SYNC_ALL) 1913 atomic_dec(&sbi->wb_sync_req); 1914 /* 1915 * if some pages were truncated, we cannot guarantee its mapping->host 1916 * to detect pending bios. 1917 */ 1918 1919 remove_dirty_inode(inode); 1920 return ret; 1921 1922 skip_write: 1923 wbc->pages_skipped += get_dirty_pages(inode); 1924 trace_f2fs_writepages(mapping->host, wbc, DATA); 1925 return 0; 1926 } 1927 1928 static int f2fs_write_data_pages(struct address_space *mapping, 1929 struct writeback_control *wbc) 1930 { 1931 struct inode *inode = mapping->host; 1932 1933 return __f2fs_write_data_pages(mapping, wbc, 1934 F2FS_I(inode)->cp_task == current ? 1935 FS_CP_DATA_IO : FS_DATA_IO); 1936 } 1937 1938 static void f2fs_write_failed(struct address_space *mapping, loff_t to) 1939 { 1940 struct inode *inode = mapping->host; 1941 loff_t i_size = i_size_read(inode); 1942 1943 if (to > i_size) { 1944 down_write(&F2FS_I(inode)->i_mmap_sem); 1945 truncate_pagecache(inode, i_size); 1946 truncate_blocks(inode, i_size, true); 1947 up_write(&F2FS_I(inode)->i_mmap_sem); 1948 } 1949 } 1950 1951 static int prepare_write_begin(struct f2fs_sb_info *sbi, 1952 struct page *page, loff_t pos, unsigned len, 1953 block_t *blk_addr, bool *node_changed) 1954 { 1955 struct inode *inode = page->mapping->host; 1956 pgoff_t index = page->index; 1957 struct dnode_of_data dn; 1958 struct page *ipage; 1959 bool locked = false; 1960 struct extent_info ei = {0,0,0}; 1961 int err = 0; 1962 1963 /* 1964 * we already allocated all the blocks, so we don't need to get 1965 * the block addresses when there is no need to fill the page. 1966 */ 1967 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE && 1968 !is_inode_flag_set(inode, FI_NO_PREALLOC)) 1969 return 0; 1970 1971 if (f2fs_has_inline_data(inode) || 1972 (pos & PAGE_MASK) >= i_size_read(inode)) { 1973 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true); 1974 locked = true; 1975 } 1976 restart: 1977 /* check inline_data */ 1978 ipage = get_node_page(sbi, inode->i_ino); 1979 if (IS_ERR(ipage)) { 1980 err = PTR_ERR(ipage); 1981 goto unlock_out; 1982 } 1983 1984 set_new_dnode(&dn, inode, ipage, ipage, 0); 1985 1986 if (f2fs_has_inline_data(inode)) { 1987 if (pos + len <= MAX_INLINE_DATA(inode)) { 1988 read_inline_data(page, ipage); 1989 set_inode_flag(inode, FI_DATA_EXIST); 1990 if (inode->i_nlink) 1991 set_inline_node(ipage); 1992 } else { 1993 err = f2fs_convert_inline_page(&dn, page); 1994 if (err) 1995 goto out; 1996 if (dn.data_blkaddr == NULL_ADDR) 1997 err = f2fs_get_block(&dn, index); 1998 } 1999 } else if (locked) { 2000 err = f2fs_get_block(&dn, index); 2001 } else { 2002 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 2003 dn.data_blkaddr = ei.blk + index - ei.fofs; 2004 } else { 2005 /* hole case */ 2006 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 2007 if (err || dn.data_blkaddr == NULL_ADDR) { 2008 f2fs_put_dnode(&dn); 2009 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, 2010 true); 2011 locked = true; 2012 goto restart; 2013 } 2014 } 2015 } 2016 2017 /* convert_inline_page can make node_changed */ 2018 *blk_addr = dn.data_blkaddr; 2019 *node_changed = dn.node_changed; 2020 out: 2021 f2fs_put_dnode(&dn); 2022 unlock_out: 2023 if (locked) 2024 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false); 2025 return err; 2026 } 2027 2028 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 2029 loff_t pos, unsigned len, unsigned flags, 2030 struct page **pagep, void **fsdata) 2031 { 2032 struct inode *inode = mapping->host; 2033 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2034 struct page *page = NULL; 2035 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 2036 bool need_balance = false; 2037 block_t blkaddr = NULL_ADDR; 2038 int err = 0; 2039 2040 trace_f2fs_write_begin(inode, pos, len, flags); 2041 2042 if (f2fs_is_atomic_file(inode) && 2043 !available_free_memory(sbi, INMEM_PAGES)) { 2044 err = -ENOMEM; 2045 goto fail; 2046 } 2047 2048 /* 2049 * We should check this at this moment to avoid deadlock on inode page 2050 * and #0 page. The locking rule for inline_data conversion should be: 2051 * lock_page(page #0) -> lock_page(inode_page) 2052 */ 2053 if (index != 0) { 2054 err = f2fs_convert_inline_inode(inode); 2055 if (err) 2056 goto fail; 2057 } 2058 repeat: 2059 /* 2060 * Do not use grab_cache_page_write_begin() to avoid deadlock due to 2061 * wait_for_stable_page. Will wait that below with our IO control. 2062 */ 2063 page = f2fs_pagecache_get_page(mapping, index, 2064 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); 2065 if (!page) { 2066 err = -ENOMEM; 2067 goto fail; 2068 } 2069 2070 *pagep = page; 2071 2072 err = prepare_write_begin(sbi, page, pos, len, 2073 &blkaddr, &need_balance); 2074 if (err) 2075 goto fail; 2076 2077 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) { 2078 unlock_page(page); 2079 f2fs_balance_fs(sbi, true); 2080 lock_page(page); 2081 if (page->mapping != mapping) { 2082 /* The page got truncated from under us */ 2083 f2fs_put_page(page, 1); 2084 goto repeat; 2085 } 2086 } 2087 2088 f2fs_wait_on_page_writeback(page, DATA, false); 2089 2090 /* wait for GCed encrypted page writeback */ 2091 if (f2fs_encrypted_file(inode)) 2092 f2fs_wait_on_block_writeback(sbi, blkaddr); 2093 2094 if (len == PAGE_SIZE || PageUptodate(page)) 2095 return 0; 2096 2097 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) { 2098 zero_user_segment(page, len, PAGE_SIZE); 2099 return 0; 2100 } 2101 2102 if (blkaddr == NEW_ADDR) { 2103 zero_user_segment(page, 0, PAGE_SIZE); 2104 SetPageUptodate(page); 2105 } else { 2106 err = f2fs_submit_page_read(inode, page, blkaddr); 2107 if (err) 2108 goto fail; 2109 2110 lock_page(page); 2111 if (unlikely(page->mapping != mapping)) { 2112 f2fs_put_page(page, 1); 2113 goto repeat; 2114 } 2115 if (unlikely(!PageUptodate(page))) { 2116 err = -EIO; 2117 goto fail; 2118 } 2119 } 2120 return 0; 2121 2122 fail: 2123 f2fs_put_page(page, 1); 2124 f2fs_write_failed(mapping, pos + len); 2125 if (f2fs_is_atomic_file(inode)) 2126 drop_inmem_pages_all(sbi); 2127 return err; 2128 } 2129 2130 static int f2fs_write_end(struct file *file, 2131 struct address_space *mapping, 2132 loff_t pos, unsigned len, unsigned copied, 2133 struct page *page, void *fsdata) 2134 { 2135 struct inode *inode = page->mapping->host; 2136 2137 trace_f2fs_write_end(inode, pos, len, copied); 2138 2139 /* 2140 * This should be come from len == PAGE_SIZE, and we expect copied 2141 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 2142 * let generic_perform_write() try to copy data again through copied=0. 2143 */ 2144 if (!PageUptodate(page)) { 2145 if (unlikely(copied != len)) 2146 copied = 0; 2147 else 2148 SetPageUptodate(page); 2149 } 2150 if (!copied) 2151 goto unlock_out; 2152 2153 set_page_dirty(page); 2154 2155 if (pos + copied > i_size_read(inode)) 2156 f2fs_i_size_write(inode, pos + copied); 2157 unlock_out: 2158 f2fs_put_page(page, 1); 2159 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 2160 return copied; 2161 } 2162 2163 static int check_direct_IO(struct inode *inode, struct iov_iter *iter, 2164 loff_t offset) 2165 { 2166 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1; 2167 2168 if (offset & blocksize_mask) 2169 return -EINVAL; 2170 2171 if (iov_iter_alignment(iter) & blocksize_mask) 2172 return -EINVAL; 2173 2174 return 0; 2175 } 2176 2177 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 2178 { 2179 struct address_space *mapping = iocb->ki_filp->f_mapping; 2180 struct inode *inode = mapping->host; 2181 size_t count = iov_iter_count(iter); 2182 loff_t offset = iocb->ki_pos; 2183 int rw = iov_iter_rw(iter); 2184 int err; 2185 2186 err = check_direct_IO(inode, iter, offset); 2187 if (err) 2188 return err; 2189 2190 if (__force_buffered_io(inode, rw)) 2191 return 0; 2192 2193 trace_f2fs_direct_IO_enter(inode, offset, count, rw); 2194 2195 down_read(&F2FS_I(inode)->dio_rwsem[rw]); 2196 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio); 2197 up_read(&F2FS_I(inode)->dio_rwsem[rw]); 2198 2199 if (rw == WRITE) { 2200 if (err > 0) { 2201 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO, 2202 err); 2203 set_inode_flag(inode, FI_UPDATE_WRITE); 2204 } else if (err < 0) { 2205 f2fs_write_failed(mapping, offset + count); 2206 } 2207 } 2208 2209 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); 2210 2211 return err; 2212 } 2213 2214 void f2fs_invalidate_page(struct page *page, unsigned int offset, 2215 unsigned int length) 2216 { 2217 struct inode *inode = page->mapping->host; 2218 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2219 2220 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 2221 (offset % PAGE_SIZE || length != PAGE_SIZE)) 2222 return; 2223 2224 if (PageDirty(page)) { 2225 if (inode->i_ino == F2FS_META_INO(sbi)) { 2226 dec_page_count(sbi, F2FS_DIRTY_META); 2227 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 2228 dec_page_count(sbi, F2FS_DIRTY_NODES); 2229 } else { 2230 inode_dec_dirty_pages(inode); 2231 remove_dirty_inode(inode); 2232 } 2233 } 2234 2235 /* This is atomic written page, keep Private */ 2236 if (IS_ATOMIC_WRITTEN_PAGE(page)) 2237 return drop_inmem_page(inode, page); 2238 2239 set_page_private(page, 0); 2240 ClearPagePrivate(page); 2241 } 2242 2243 int f2fs_release_page(struct page *page, gfp_t wait) 2244 { 2245 /* If this is dirty page, keep PagePrivate */ 2246 if (PageDirty(page)) 2247 return 0; 2248 2249 /* This is atomic written page, keep Private */ 2250 if (IS_ATOMIC_WRITTEN_PAGE(page)) 2251 return 0; 2252 2253 set_page_private(page, 0); 2254 ClearPagePrivate(page); 2255 return 1; 2256 } 2257 2258 /* 2259 * This was copied from __set_page_dirty_buffers which gives higher performance 2260 * in very high speed storages. (e.g., pmem) 2261 */ 2262 void f2fs_set_page_dirty_nobuffers(struct page *page) 2263 { 2264 struct address_space *mapping = page->mapping; 2265 unsigned long flags; 2266 2267 if (unlikely(!mapping)) 2268 return; 2269 2270 spin_lock(&mapping->private_lock); 2271 lock_page_memcg(page); 2272 SetPageDirty(page); 2273 spin_unlock(&mapping->private_lock); 2274 2275 spin_lock_irqsave(&mapping->tree_lock, flags); 2276 WARN_ON_ONCE(!PageUptodate(page)); 2277 account_page_dirtied(page, mapping); 2278 radix_tree_tag_set(&mapping->page_tree, 2279 page_index(page), PAGECACHE_TAG_DIRTY); 2280 spin_unlock_irqrestore(&mapping->tree_lock, flags); 2281 unlock_page_memcg(page); 2282 2283 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 2284 return; 2285 } 2286 2287 static int f2fs_set_data_page_dirty(struct page *page) 2288 { 2289 struct address_space *mapping = page->mapping; 2290 struct inode *inode = mapping->host; 2291 2292 trace_f2fs_set_page_dirty(page, DATA); 2293 2294 if (!PageUptodate(page)) 2295 SetPageUptodate(page); 2296 2297 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) { 2298 if (!IS_ATOMIC_WRITTEN_PAGE(page)) { 2299 register_inmem_page(inode, page); 2300 return 1; 2301 } 2302 /* 2303 * Previously, this page has been registered, we just 2304 * return here. 2305 */ 2306 return 0; 2307 } 2308 2309 if (!PageDirty(page)) { 2310 f2fs_set_page_dirty_nobuffers(page); 2311 update_dirty_page(inode, page); 2312 return 1; 2313 } 2314 return 0; 2315 } 2316 2317 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 2318 { 2319 struct inode *inode = mapping->host; 2320 2321 if (f2fs_has_inline_data(inode)) 2322 return 0; 2323 2324 /* make sure allocating whole blocks */ 2325 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 2326 filemap_write_and_wait(mapping); 2327 2328 return generic_block_bmap(mapping, block, get_data_block_bmap); 2329 } 2330 2331 #ifdef CONFIG_MIGRATION 2332 #include <linux/migrate.h> 2333 2334 int f2fs_migrate_page(struct address_space *mapping, 2335 struct page *newpage, struct page *page, enum migrate_mode mode) 2336 { 2337 int rc, extra_count; 2338 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 2339 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page); 2340 2341 BUG_ON(PageWriteback(page)); 2342 2343 /* migrating an atomic written page is safe with the inmem_lock hold */ 2344 if (atomic_written) { 2345 if (mode != MIGRATE_SYNC) 2346 return -EBUSY; 2347 if (!mutex_trylock(&fi->inmem_lock)) 2348 return -EAGAIN; 2349 } 2350 2351 /* 2352 * A reference is expected if PagePrivate set when move mapping, 2353 * however F2FS breaks this for maintaining dirty page counts when 2354 * truncating pages. So here adjusting the 'extra_count' make it work. 2355 */ 2356 extra_count = (atomic_written ? 1 : 0) - page_has_private(page); 2357 rc = migrate_page_move_mapping(mapping, newpage, 2358 page, NULL, mode, extra_count); 2359 if (rc != MIGRATEPAGE_SUCCESS) { 2360 if (atomic_written) 2361 mutex_unlock(&fi->inmem_lock); 2362 return rc; 2363 } 2364 2365 if (atomic_written) { 2366 struct inmem_pages *cur; 2367 list_for_each_entry(cur, &fi->inmem_pages, list) 2368 if (cur->page == page) { 2369 cur->page = newpage; 2370 break; 2371 } 2372 mutex_unlock(&fi->inmem_lock); 2373 put_page(page); 2374 get_page(newpage); 2375 } 2376 2377 if (PagePrivate(page)) 2378 SetPagePrivate(newpage); 2379 set_page_private(newpage, page_private(page)); 2380 2381 if (mode != MIGRATE_SYNC_NO_COPY) 2382 migrate_page_copy(newpage, page); 2383 else 2384 migrate_page_states(newpage, page); 2385 2386 return MIGRATEPAGE_SUCCESS; 2387 } 2388 #endif 2389 2390 const struct address_space_operations f2fs_dblock_aops = { 2391 .readpage = f2fs_read_data_page, 2392 .readpages = f2fs_read_data_pages, 2393 .writepage = f2fs_write_data_page, 2394 .writepages = f2fs_write_data_pages, 2395 .write_begin = f2fs_write_begin, 2396 .write_end = f2fs_write_end, 2397 .set_page_dirty = f2fs_set_data_page_dirty, 2398 .invalidatepage = f2fs_invalidate_page, 2399 .releasepage = f2fs_release_page, 2400 .direct_IO = f2fs_direct_IO, 2401 .bmap = f2fs_bmap, 2402 #ifdef CONFIG_MIGRATION 2403 .migratepage = f2fs_migrate_page, 2404 #endif 2405 }; 2406