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 26 #include "f2fs.h" 27 #include "node.h" 28 #include "segment.h" 29 #include "trace.h" 30 #include <trace/events/f2fs.h> 31 32 static bool __is_cp_guaranteed(struct page *page) 33 { 34 struct address_space *mapping = page->mapping; 35 struct inode *inode; 36 struct f2fs_sb_info *sbi; 37 38 if (!mapping) 39 return false; 40 41 inode = mapping->host; 42 sbi = F2FS_I_SB(inode); 43 44 if (inode->i_ino == F2FS_META_INO(sbi) || 45 inode->i_ino == F2FS_NODE_INO(sbi) || 46 S_ISDIR(inode->i_mode) || 47 is_cold_data(page)) 48 return true; 49 return false; 50 } 51 52 static void f2fs_read_end_io(struct bio *bio) 53 { 54 struct bio_vec *bvec; 55 int i; 56 57 #ifdef CONFIG_F2FS_FAULT_INJECTION 58 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) 59 bio->bi_error = -EIO; 60 #endif 61 62 if (f2fs_bio_encrypted(bio)) { 63 if (bio->bi_error) { 64 fscrypt_release_ctx(bio->bi_private); 65 } else { 66 fscrypt_decrypt_bio_pages(bio->bi_private, bio); 67 return; 68 } 69 } 70 71 bio_for_each_segment_all(bvec, bio, i) { 72 struct page *page = bvec->bv_page; 73 74 if (!bio->bi_error) { 75 if (!PageUptodate(page)) 76 SetPageUptodate(page); 77 } else { 78 ClearPageUptodate(page); 79 SetPageError(page); 80 } 81 unlock_page(page); 82 } 83 bio_put(bio); 84 } 85 86 static void f2fs_write_end_io(struct bio *bio) 87 { 88 struct f2fs_sb_info *sbi = bio->bi_private; 89 struct bio_vec *bvec; 90 int i; 91 92 bio_for_each_segment_all(bvec, bio, i) { 93 struct page *page = bvec->bv_page; 94 enum count_type type = WB_DATA_TYPE(page); 95 96 fscrypt_pullback_bio_page(&page, true); 97 98 if (unlikely(bio->bi_error)) { 99 mapping_set_error(page->mapping, -EIO); 100 f2fs_stop_checkpoint(sbi, true); 101 } 102 dec_page_count(sbi, type); 103 clear_cold_data(page); 104 end_page_writeback(page); 105 } 106 if (!get_pages(sbi, F2FS_WB_CP_DATA) && 107 wq_has_sleeper(&sbi->cp_wait)) 108 wake_up(&sbi->cp_wait); 109 110 bio_put(bio); 111 } 112 113 /* 114 * Return true, if pre_bio's bdev is same as its target device. 115 */ 116 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, 117 block_t blk_addr, struct bio *bio) 118 { 119 struct block_device *bdev = sbi->sb->s_bdev; 120 int i; 121 122 for (i = 0; i < sbi->s_ndevs; i++) { 123 if (FDEV(i).start_blk <= blk_addr && 124 FDEV(i).end_blk >= blk_addr) { 125 blk_addr -= FDEV(i).start_blk; 126 bdev = FDEV(i).bdev; 127 break; 128 } 129 } 130 if (bio) { 131 bio->bi_bdev = bdev; 132 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); 133 } 134 return bdev; 135 } 136 137 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) 138 { 139 int i; 140 141 for (i = 0; i < sbi->s_ndevs; i++) 142 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) 143 return i; 144 return 0; 145 } 146 147 static bool __same_bdev(struct f2fs_sb_info *sbi, 148 block_t blk_addr, struct bio *bio) 149 { 150 return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev; 151 } 152 153 /* 154 * Low-level block read/write IO operations. 155 */ 156 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr, 157 int npages, bool is_read) 158 { 159 struct bio *bio; 160 161 bio = f2fs_bio_alloc(npages); 162 163 f2fs_target_device(sbi, blk_addr, bio); 164 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io; 165 bio->bi_private = is_read ? NULL : sbi; 166 167 return bio; 168 } 169 170 static inline void __submit_bio(struct f2fs_sb_info *sbi, 171 struct bio *bio, enum page_type type) 172 { 173 if (!is_read_io(bio_op(bio))) { 174 if (f2fs_sb_mounted_blkzoned(sbi->sb) && 175 current->plug && (type == DATA || type == NODE)) 176 blk_finish_plug(current->plug); 177 } 178 submit_bio(bio); 179 } 180 181 static void __submit_merged_bio(struct f2fs_bio_info *io) 182 { 183 struct f2fs_io_info *fio = &io->fio; 184 185 if (!io->bio) 186 return; 187 188 if (is_read_io(fio->op)) 189 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio); 190 else 191 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio); 192 193 bio_set_op_attrs(io->bio, fio->op, fio->op_flags); 194 195 __submit_bio(io->sbi, io->bio, fio->type); 196 io->bio = NULL; 197 } 198 199 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode, 200 struct page *page, nid_t ino) 201 { 202 struct bio_vec *bvec; 203 struct page *target; 204 int i; 205 206 if (!io->bio) 207 return false; 208 209 if (!inode && !page && !ino) 210 return true; 211 212 bio_for_each_segment_all(bvec, io->bio, i) { 213 214 if (bvec->bv_page->mapping) 215 target = bvec->bv_page; 216 else 217 target = fscrypt_control_page(bvec->bv_page); 218 219 if (inode && inode == target->mapping->host) 220 return true; 221 if (page && page == target) 222 return true; 223 if (ino && ino == ino_of_node(target)) 224 return true; 225 } 226 227 return false; 228 } 229 230 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode, 231 struct page *page, nid_t ino, 232 enum page_type type) 233 { 234 enum page_type btype = PAGE_TYPE_OF_BIO(type); 235 struct f2fs_bio_info *io = &sbi->write_io[btype]; 236 bool ret; 237 238 down_read(&io->io_rwsem); 239 ret = __has_merged_page(io, inode, page, ino); 240 up_read(&io->io_rwsem); 241 return ret; 242 } 243 244 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, 245 struct inode *inode, struct page *page, 246 nid_t ino, enum page_type type, int rw) 247 { 248 enum page_type btype = PAGE_TYPE_OF_BIO(type); 249 struct f2fs_bio_info *io; 250 251 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype]; 252 253 down_write(&io->io_rwsem); 254 255 if (!__has_merged_page(io, inode, page, ino)) 256 goto out; 257 258 /* change META to META_FLUSH in the checkpoint procedure */ 259 if (type >= META_FLUSH) { 260 io->fio.type = META_FLUSH; 261 io->fio.op = REQ_OP_WRITE; 262 io->fio.op_flags = REQ_PREFLUSH | REQ_META | REQ_PRIO; 263 if (!test_opt(sbi, NOBARRIER)) 264 io->fio.op_flags |= REQ_FUA; 265 } 266 __submit_merged_bio(io); 267 out: 268 up_write(&io->io_rwsem); 269 } 270 271 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type, 272 int rw) 273 { 274 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw); 275 } 276 277 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi, 278 struct inode *inode, struct page *page, 279 nid_t ino, enum page_type type, int rw) 280 { 281 if (has_merged_page(sbi, inode, page, ino, type)) 282 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw); 283 } 284 285 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi) 286 { 287 f2fs_submit_merged_bio(sbi, DATA, WRITE); 288 f2fs_submit_merged_bio(sbi, NODE, WRITE); 289 f2fs_submit_merged_bio(sbi, META, WRITE); 290 } 291 292 /* 293 * Fill the locked page with data located in the block address. 294 * Return unlocked page. 295 */ 296 int f2fs_submit_page_bio(struct f2fs_io_info *fio) 297 { 298 struct bio *bio; 299 struct page *page = fio->encrypted_page ? 300 fio->encrypted_page : fio->page; 301 302 trace_f2fs_submit_page_bio(page, fio); 303 f2fs_trace_ios(fio, 0); 304 305 /* Allocate a new bio */ 306 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op)); 307 308 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 309 bio_put(bio); 310 return -EFAULT; 311 } 312 bio_set_op_attrs(bio, fio->op, fio->op_flags); 313 314 __submit_bio(fio->sbi, bio, fio->type); 315 return 0; 316 } 317 318 void f2fs_submit_page_mbio(struct f2fs_io_info *fio) 319 { 320 struct f2fs_sb_info *sbi = fio->sbi; 321 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); 322 struct f2fs_bio_info *io; 323 bool is_read = is_read_io(fio->op); 324 struct page *bio_page; 325 326 io = is_read ? &sbi->read_io : &sbi->write_io[btype]; 327 328 if (fio->old_blkaddr != NEW_ADDR) 329 verify_block_addr(sbi, fio->old_blkaddr); 330 verify_block_addr(sbi, fio->new_blkaddr); 331 332 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page; 333 334 if (!is_read) 335 inc_page_count(sbi, WB_DATA_TYPE(bio_page)); 336 337 down_write(&io->io_rwsem); 338 339 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 || 340 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) || 341 !__same_bdev(sbi, fio->new_blkaddr, io->bio))) 342 __submit_merged_bio(io); 343 alloc_new: 344 if (io->bio == NULL) { 345 io->bio = __bio_alloc(sbi, fio->new_blkaddr, 346 BIO_MAX_PAGES, is_read); 347 io->fio = *fio; 348 } 349 350 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < 351 PAGE_SIZE) { 352 __submit_merged_bio(io); 353 goto alloc_new; 354 } 355 356 io->last_block_in_bio = fio->new_blkaddr; 357 f2fs_trace_ios(fio, 0); 358 359 up_write(&io->io_rwsem); 360 trace_f2fs_submit_page_mbio(fio->page, fio); 361 } 362 363 static void __set_data_blkaddr(struct dnode_of_data *dn) 364 { 365 struct f2fs_node *rn = F2FS_NODE(dn->node_page); 366 __le32 *addr_array; 367 368 /* Get physical address of data block */ 369 addr_array = blkaddr_in_node(rn); 370 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); 371 } 372 373 /* 374 * Lock ordering for the change of data block address: 375 * ->data_page 376 * ->node_page 377 * update block addresses in the node page 378 */ 379 void set_data_blkaddr(struct dnode_of_data *dn) 380 { 381 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 382 __set_data_blkaddr(dn); 383 if (set_page_dirty(dn->node_page)) 384 dn->node_changed = true; 385 } 386 387 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) 388 { 389 dn->data_blkaddr = blkaddr; 390 set_data_blkaddr(dn); 391 f2fs_update_extent_cache(dn); 392 } 393 394 /* dn->ofs_in_node will be returned with up-to-date last block pointer */ 395 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) 396 { 397 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 398 399 if (!count) 400 return 0; 401 402 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 403 return -EPERM; 404 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count))) 405 return -ENOSPC; 406 407 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid, 408 dn->ofs_in_node, count); 409 410 f2fs_wait_on_page_writeback(dn->node_page, NODE, true); 411 412 for (; count > 0; dn->ofs_in_node++) { 413 block_t blkaddr = 414 datablock_addr(dn->node_page, dn->ofs_in_node); 415 if (blkaddr == NULL_ADDR) { 416 dn->data_blkaddr = NEW_ADDR; 417 __set_data_blkaddr(dn); 418 count--; 419 } 420 } 421 422 if (set_page_dirty(dn->node_page)) 423 dn->node_changed = true; 424 return 0; 425 } 426 427 /* Should keep dn->ofs_in_node unchanged */ 428 int reserve_new_block(struct dnode_of_data *dn) 429 { 430 unsigned int ofs_in_node = dn->ofs_in_node; 431 int ret; 432 433 ret = reserve_new_blocks(dn, 1); 434 dn->ofs_in_node = ofs_in_node; 435 return ret; 436 } 437 438 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) 439 { 440 bool need_put = dn->inode_page ? false : true; 441 int err; 442 443 err = get_dnode_of_data(dn, index, ALLOC_NODE); 444 if (err) 445 return err; 446 447 if (dn->data_blkaddr == NULL_ADDR) 448 err = reserve_new_block(dn); 449 if (err || need_put) 450 f2fs_put_dnode(dn); 451 return err; 452 } 453 454 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index) 455 { 456 struct extent_info ei; 457 struct inode *inode = dn->inode; 458 459 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 460 dn->data_blkaddr = ei.blk + index - ei.fofs; 461 return 0; 462 } 463 464 return f2fs_reserve_block(dn, index); 465 } 466 467 struct page *get_read_data_page(struct inode *inode, pgoff_t index, 468 int op_flags, bool for_write) 469 { 470 struct address_space *mapping = inode->i_mapping; 471 struct dnode_of_data dn; 472 struct page *page; 473 struct extent_info ei; 474 int err; 475 struct f2fs_io_info fio = { 476 .sbi = F2FS_I_SB(inode), 477 .type = DATA, 478 .op = REQ_OP_READ, 479 .op_flags = op_flags, 480 .encrypted_page = NULL, 481 }; 482 483 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) 484 return read_mapping_page(mapping, index, NULL); 485 486 page = f2fs_grab_cache_page(mapping, index, for_write); 487 if (!page) 488 return ERR_PTR(-ENOMEM); 489 490 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 491 dn.data_blkaddr = ei.blk + index - ei.fofs; 492 goto got_it; 493 } 494 495 set_new_dnode(&dn, inode, NULL, NULL, 0); 496 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 497 if (err) 498 goto put_err; 499 f2fs_put_dnode(&dn); 500 501 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 502 err = -ENOENT; 503 goto put_err; 504 } 505 got_it: 506 if (PageUptodate(page)) { 507 unlock_page(page); 508 return page; 509 } 510 511 /* 512 * A new dentry page is allocated but not able to be written, since its 513 * new inode page couldn't be allocated due to -ENOSPC. 514 * In such the case, its blkaddr can be remained as NEW_ADDR. 515 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. 516 */ 517 if (dn.data_blkaddr == NEW_ADDR) { 518 zero_user_segment(page, 0, PAGE_SIZE); 519 if (!PageUptodate(page)) 520 SetPageUptodate(page); 521 unlock_page(page); 522 return page; 523 } 524 525 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 526 fio.page = page; 527 err = f2fs_submit_page_bio(&fio); 528 if (err) 529 goto put_err; 530 return page; 531 532 put_err: 533 f2fs_put_page(page, 1); 534 return ERR_PTR(err); 535 } 536 537 struct page *find_data_page(struct inode *inode, pgoff_t index) 538 { 539 struct address_space *mapping = inode->i_mapping; 540 struct page *page; 541 542 page = find_get_page(mapping, index); 543 if (page && PageUptodate(page)) 544 return page; 545 f2fs_put_page(page, 0); 546 547 page = get_read_data_page(inode, index, 0, false); 548 if (IS_ERR(page)) 549 return page; 550 551 if (PageUptodate(page)) 552 return page; 553 554 wait_on_page_locked(page); 555 if (unlikely(!PageUptodate(page))) { 556 f2fs_put_page(page, 0); 557 return ERR_PTR(-EIO); 558 } 559 return page; 560 } 561 562 /* 563 * If it tries to access a hole, return an error. 564 * Because, the callers, functions in dir.c and GC, should be able to know 565 * whether this page exists or not. 566 */ 567 struct page *get_lock_data_page(struct inode *inode, pgoff_t index, 568 bool for_write) 569 { 570 struct address_space *mapping = inode->i_mapping; 571 struct page *page; 572 repeat: 573 page = get_read_data_page(inode, index, 0, for_write); 574 if (IS_ERR(page)) 575 return page; 576 577 /* wait for read completion */ 578 lock_page(page); 579 if (unlikely(page->mapping != mapping)) { 580 f2fs_put_page(page, 1); 581 goto repeat; 582 } 583 if (unlikely(!PageUptodate(page))) { 584 f2fs_put_page(page, 1); 585 return ERR_PTR(-EIO); 586 } 587 return page; 588 } 589 590 /* 591 * Caller ensures that this data page is never allocated. 592 * A new zero-filled data page is allocated in the page cache. 593 * 594 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and 595 * f2fs_unlock_op(). 596 * Note that, ipage is set only by make_empty_dir, and if any error occur, 597 * ipage should be released by this function. 598 */ 599 struct page *get_new_data_page(struct inode *inode, 600 struct page *ipage, pgoff_t index, bool new_i_size) 601 { 602 struct address_space *mapping = inode->i_mapping; 603 struct page *page; 604 struct dnode_of_data dn; 605 int err; 606 607 page = f2fs_grab_cache_page(mapping, index, true); 608 if (!page) { 609 /* 610 * before exiting, we should make sure ipage will be released 611 * if any error occur. 612 */ 613 f2fs_put_page(ipage, 1); 614 return ERR_PTR(-ENOMEM); 615 } 616 617 set_new_dnode(&dn, inode, ipage, NULL, 0); 618 err = f2fs_reserve_block(&dn, index); 619 if (err) { 620 f2fs_put_page(page, 1); 621 return ERR_PTR(err); 622 } 623 if (!ipage) 624 f2fs_put_dnode(&dn); 625 626 if (PageUptodate(page)) 627 goto got_it; 628 629 if (dn.data_blkaddr == NEW_ADDR) { 630 zero_user_segment(page, 0, PAGE_SIZE); 631 if (!PageUptodate(page)) 632 SetPageUptodate(page); 633 } else { 634 f2fs_put_page(page, 1); 635 636 /* if ipage exists, blkaddr should be NEW_ADDR */ 637 f2fs_bug_on(F2FS_I_SB(inode), ipage); 638 page = get_lock_data_page(inode, index, true); 639 if (IS_ERR(page)) 640 return page; 641 } 642 got_it: 643 if (new_i_size && i_size_read(inode) < 644 ((loff_t)(index + 1) << PAGE_SHIFT)) 645 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT)); 646 return page; 647 } 648 649 static int __allocate_data_block(struct dnode_of_data *dn) 650 { 651 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 652 struct f2fs_summary sum; 653 struct node_info ni; 654 pgoff_t fofs; 655 blkcnt_t count = 1; 656 657 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) 658 return -EPERM; 659 660 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node); 661 if (dn->data_blkaddr == NEW_ADDR) 662 goto alloc; 663 664 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count))) 665 return -ENOSPC; 666 667 alloc: 668 get_node_info(sbi, dn->nid, &ni); 669 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 670 671 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr, 672 &sum, CURSEG_WARM_DATA); 673 set_data_blkaddr(dn); 674 675 /* update i_size */ 676 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) + 677 dn->ofs_in_node; 678 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT)) 679 f2fs_i_size_write(dn->inode, 680 ((loff_t)(fofs + 1) << PAGE_SHIFT)); 681 return 0; 682 } 683 684 static inline bool __force_buffered_io(struct inode *inode, int rw) 685 { 686 return ((f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) || 687 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) || 688 F2FS_I_SB(inode)->s_ndevs); 689 } 690 691 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from) 692 { 693 struct inode *inode = file_inode(iocb->ki_filp); 694 struct f2fs_map_blocks map; 695 int err = 0; 696 697 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos); 698 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from)); 699 if (map.m_len > map.m_lblk) 700 map.m_len -= map.m_lblk; 701 else 702 map.m_len = 0; 703 704 map.m_next_pgofs = NULL; 705 706 if (iocb->ki_flags & IOCB_DIRECT) { 707 err = f2fs_convert_inline_inode(inode); 708 if (err) 709 return err; 710 return f2fs_map_blocks(inode, &map, 1, 711 __force_buffered_io(inode, WRITE) ? 712 F2FS_GET_BLOCK_PRE_AIO : 713 F2FS_GET_BLOCK_PRE_DIO); 714 } 715 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) { 716 err = f2fs_convert_inline_inode(inode); 717 if (err) 718 return err; 719 } 720 if (!f2fs_has_inline_data(inode)) 721 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO); 722 return err; 723 } 724 725 /* 726 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with 727 * f2fs_map_blocks structure. 728 * If original data blocks are allocated, then give them to blockdev. 729 * Otherwise, 730 * a. preallocate requested block addresses 731 * b. do not use extent cache for better performance 732 * c. give the block addresses to blockdev 733 */ 734 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, 735 int create, int flag) 736 { 737 unsigned int maxblocks = map->m_len; 738 struct dnode_of_data dn; 739 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 740 int mode = create ? ALLOC_NODE : LOOKUP_NODE; 741 pgoff_t pgofs, end_offset, end; 742 int err = 0, ofs = 1; 743 unsigned int ofs_in_node, last_ofs_in_node; 744 blkcnt_t prealloc; 745 struct extent_info ei; 746 block_t blkaddr; 747 748 if (!maxblocks) 749 return 0; 750 751 map->m_len = 0; 752 map->m_flags = 0; 753 754 /* it only supports block size == page size */ 755 pgofs = (pgoff_t)map->m_lblk; 756 end = pgofs + maxblocks; 757 758 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) { 759 map->m_pblk = ei.blk + pgofs - ei.fofs; 760 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs); 761 map->m_flags = F2FS_MAP_MAPPED; 762 goto out; 763 } 764 765 next_dnode: 766 if (create) 767 f2fs_lock_op(sbi); 768 769 /* When reading holes, we need its node page */ 770 set_new_dnode(&dn, inode, NULL, NULL, 0); 771 err = get_dnode_of_data(&dn, pgofs, mode); 772 if (err) { 773 if (flag == F2FS_GET_BLOCK_BMAP) 774 map->m_pblk = 0; 775 if (err == -ENOENT) { 776 err = 0; 777 if (map->m_next_pgofs) 778 *map->m_next_pgofs = 779 get_next_page_offset(&dn, pgofs); 780 } 781 goto unlock_out; 782 } 783 784 prealloc = 0; 785 last_ofs_in_node = ofs_in_node = dn.ofs_in_node; 786 end_offset = ADDRS_PER_PAGE(dn.node_page, inode); 787 788 next_block: 789 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node); 790 791 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) { 792 if (create) { 793 if (unlikely(f2fs_cp_error(sbi))) { 794 err = -EIO; 795 goto sync_out; 796 } 797 if (flag == F2FS_GET_BLOCK_PRE_AIO) { 798 if (blkaddr == NULL_ADDR) { 799 prealloc++; 800 last_ofs_in_node = dn.ofs_in_node; 801 } 802 } else { 803 err = __allocate_data_block(&dn); 804 if (!err) 805 set_inode_flag(inode, FI_APPEND_WRITE); 806 } 807 if (err) 808 goto sync_out; 809 map->m_flags = F2FS_MAP_NEW; 810 blkaddr = dn.data_blkaddr; 811 } else { 812 if (flag == F2FS_GET_BLOCK_BMAP) { 813 map->m_pblk = 0; 814 goto sync_out; 815 } 816 if (flag == F2FS_GET_BLOCK_FIEMAP && 817 blkaddr == NULL_ADDR) { 818 if (map->m_next_pgofs) 819 *map->m_next_pgofs = pgofs + 1; 820 } 821 if (flag != F2FS_GET_BLOCK_FIEMAP || 822 blkaddr != NEW_ADDR) 823 goto sync_out; 824 } 825 } 826 827 if (flag == F2FS_GET_BLOCK_PRE_AIO) 828 goto skip; 829 830 if (map->m_len == 0) { 831 /* preallocated unwritten block should be mapped for fiemap. */ 832 if (blkaddr == NEW_ADDR) 833 map->m_flags |= F2FS_MAP_UNWRITTEN; 834 map->m_flags |= F2FS_MAP_MAPPED; 835 836 map->m_pblk = blkaddr; 837 map->m_len = 1; 838 } else if ((map->m_pblk != NEW_ADDR && 839 blkaddr == (map->m_pblk + ofs)) || 840 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || 841 flag == F2FS_GET_BLOCK_PRE_DIO) { 842 ofs++; 843 map->m_len++; 844 } else { 845 goto sync_out; 846 } 847 848 skip: 849 dn.ofs_in_node++; 850 pgofs++; 851 852 /* preallocate blocks in batch for one dnode page */ 853 if (flag == F2FS_GET_BLOCK_PRE_AIO && 854 (pgofs == end || dn.ofs_in_node == end_offset)) { 855 856 dn.ofs_in_node = ofs_in_node; 857 err = reserve_new_blocks(&dn, prealloc); 858 if (err) 859 goto sync_out; 860 861 map->m_len += dn.ofs_in_node - ofs_in_node; 862 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { 863 err = -ENOSPC; 864 goto sync_out; 865 } 866 dn.ofs_in_node = end_offset; 867 } 868 869 if (pgofs >= end) 870 goto sync_out; 871 else if (dn.ofs_in_node < end_offset) 872 goto next_block; 873 874 f2fs_put_dnode(&dn); 875 876 if (create) { 877 f2fs_unlock_op(sbi); 878 f2fs_balance_fs(sbi, dn.node_changed); 879 } 880 goto next_dnode; 881 882 sync_out: 883 f2fs_put_dnode(&dn); 884 unlock_out: 885 if (create) { 886 f2fs_unlock_op(sbi); 887 f2fs_balance_fs(sbi, dn.node_changed); 888 } 889 out: 890 trace_f2fs_map_blocks(inode, map, err); 891 return err; 892 } 893 894 static int __get_data_block(struct inode *inode, sector_t iblock, 895 struct buffer_head *bh, int create, int flag, 896 pgoff_t *next_pgofs) 897 { 898 struct f2fs_map_blocks map; 899 int err; 900 901 map.m_lblk = iblock; 902 map.m_len = bh->b_size >> inode->i_blkbits; 903 map.m_next_pgofs = next_pgofs; 904 905 err = f2fs_map_blocks(inode, &map, create, flag); 906 if (!err) { 907 map_bh(bh, inode->i_sb, map.m_pblk); 908 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags; 909 bh->b_size = map.m_len << inode->i_blkbits; 910 } 911 return err; 912 } 913 914 static int get_data_block(struct inode *inode, sector_t iblock, 915 struct buffer_head *bh_result, int create, int flag, 916 pgoff_t *next_pgofs) 917 { 918 return __get_data_block(inode, iblock, bh_result, create, 919 flag, next_pgofs); 920 } 921 922 static int get_data_block_dio(struct inode *inode, sector_t iblock, 923 struct buffer_head *bh_result, int create) 924 { 925 return __get_data_block(inode, iblock, bh_result, create, 926 F2FS_GET_BLOCK_DIO, NULL); 927 } 928 929 static int get_data_block_bmap(struct inode *inode, sector_t iblock, 930 struct buffer_head *bh_result, int create) 931 { 932 /* Block number less than F2FS MAX BLOCKS */ 933 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks)) 934 return -EFBIG; 935 936 return __get_data_block(inode, iblock, bh_result, create, 937 F2FS_GET_BLOCK_BMAP, NULL); 938 } 939 940 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset) 941 { 942 return (offset >> inode->i_blkbits); 943 } 944 945 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk) 946 { 947 return (blk << inode->i_blkbits); 948 } 949 950 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 951 u64 start, u64 len) 952 { 953 struct buffer_head map_bh; 954 sector_t start_blk, last_blk; 955 pgoff_t next_pgofs; 956 u64 logical = 0, phys = 0, size = 0; 957 u32 flags = 0; 958 int ret = 0; 959 960 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC); 961 if (ret) 962 return ret; 963 964 if (f2fs_has_inline_data(inode)) { 965 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); 966 if (ret != -EAGAIN) 967 return ret; 968 } 969 970 inode_lock(inode); 971 972 if (logical_to_blk(inode, len) == 0) 973 len = blk_to_logical(inode, 1); 974 975 start_blk = logical_to_blk(inode, start); 976 last_blk = logical_to_blk(inode, start + len - 1); 977 978 next: 979 memset(&map_bh, 0, sizeof(struct buffer_head)); 980 map_bh.b_size = len; 981 982 ret = get_data_block(inode, start_blk, &map_bh, 0, 983 F2FS_GET_BLOCK_FIEMAP, &next_pgofs); 984 if (ret) 985 goto out; 986 987 /* HOLE */ 988 if (!buffer_mapped(&map_bh)) { 989 start_blk = next_pgofs; 990 991 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode, 992 F2FS_I_SB(inode)->max_file_blocks)) 993 goto prep_next; 994 995 flags |= FIEMAP_EXTENT_LAST; 996 } 997 998 if (size) { 999 if (f2fs_encrypted_inode(inode)) 1000 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; 1001 1002 ret = fiemap_fill_next_extent(fieinfo, logical, 1003 phys, size, flags); 1004 } 1005 1006 if (start_blk > last_blk || ret) 1007 goto out; 1008 1009 logical = blk_to_logical(inode, start_blk); 1010 phys = blk_to_logical(inode, map_bh.b_blocknr); 1011 size = map_bh.b_size; 1012 flags = 0; 1013 if (buffer_unwritten(&map_bh)) 1014 flags = FIEMAP_EXTENT_UNWRITTEN; 1015 1016 start_blk += logical_to_blk(inode, size); 1017 1018 prep_next: 1019 cond_resched(); 1020 if (fatal_signal_pending(current)) 1021 ret = -EINTR; 1022 else 1023 goto next; 1024 out: 1025 if (ret == 1) 1026 ret = 0; 1027 1028 inode_unlock(inode); 1029 return ret; 1030 } 1031 1032 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr, 1033 unsigned nr_pages) 1034 { 1035 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1036 struct fscrypt_ctx *ctx = NULL; 1037 struct bio *bio; 1038 1039 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) { 1040 ctx = fscrypt_get_ctx(inode, GFP_NOFS); 1041 if (IS_ERR(ctx)) 1042 return ERR_CAST(ctx); 1043 1044 /* wait the page to be moved by cleaning */ 1045 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr); 1046 } 1047 1048 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES)); 1049 if (!bio) { 1050 if (ctx) 1051 fscrypt_release_ctx(ctx); 1052 return ERR_PTR(-ENOMEM); 1053 } 1054 f2fs_target_device(sbi, blkaddr, bio); 1055 bio->bi_end_io = f2fs_read_end_io; 1056 bio->bi_private = ctx; 1057 1058 return bio; 1059 } 1060 1061 /* 1062 * This function was originally taken from fs/mpage.c, and customized for f2fs. 1063 * Major change was from block_size == page_size in f2fs by default. 1064 */ 1065 static int f2fs_mpage_readpages(struct address_space *mapping, 1066 struct list_head *pages, struct page *page, 1067 unsigned nr_pages) 1068 { 1069 struct bio *bio = NULL; 1070 unsigned page_idx; 1071 sector_t last_block_in_bio = 0; 1072 struct inode *inode = mapping->host; 1073 const unsigned blkbits = inode->i_blkbits; 1074 const unsigned blocksize = 1 << blkbits; 1075 sector_t block_in_file; 1076 sector_t last_block; 1077 sector_t last_block_in_file; 1078 sector_t block_nr; 1079 struct f2fs_map_blocks map; 1080 1081 map.m_pblk = 0; 1082 map.m_lblk = 0; 1083 map.m_len = 0; 1084 map.m_flags = 0; 1085 map.m_next_pgofs = NULL; 1086 1087 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) { 1088 1089 prefetchw(&page->flags); 1090 if (pages) { 1091 page = list_entry(pages->prev, struct page, lru); 1092 list_del(&page->lru); 1093 if (add_to_page_cache_lru(page, mapping, 1094 page->index, 1095 readahead_gfp_mask(mapping))) 1096 goto next_page; 1097 } 1098 1099 block_in_file = (sector_t)page->index; 1100 last_block = block_in_file + nr_pages; 1101 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> 1102 blkbits; 1103 if (last_block > last_block_in_file) 1104 last_block = last_block_in_file; 1105 1106 /* 1107 * Map blocks using the previous result first. 1108 */ 1109 if ((map.m_flags & F2FS_MAP_MAPPED) && 1110 block_in_file > map.m_lblk && 1111 block_in_file < (map.m_lblk + map.m_len)) 1112 goto got_it; 1113 1114 /* 1115 * Then do more f2fs_map_blocks() calls until we are 1116 * done with this page. 1117 */ 1118 map.m_flags = 0; 1119 1120 if (block_in_file < last_block) { 1121 map.m_lblk = block_in_file; 1122 map.m_len = last_block - block_in_file; 1123 1124 if (f2fs_map_blocks(inode, &map, 0, 1125 F2FS_GET_BLOCK_READ)) 1126 goto set_error_page; 1127 } 1128 got_it: 1129 if ((map.m_flags & F2FS_MAP_MAPPED)) { 1130 block_nr = map.m_pblk + block_in_file - map.m_lblk; 1131 SetPageMappedToDisk(page); 1132 1133 if (!PageUptodate(page) && !cleancache_get_page(page)) { 1134 SetPageUptodate(page); 1135 goto confused; 1136 } 1137 } else { 1138 zero_user_segment(page, 0, PAGE_SIZE); 1139 if (!PageUptodate(page)) 1140 SetPageUptodate(page); 1141 unlock_page(page); 1142 goto next_page; 1143 } 1144 1145 /* 1146 * This page will go to BIO. Do we need to send this 1147 * BIO off first? 1148 */ 1149 if (bio && (last_block_in_bio != block_nr - 1 || 1150 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) { 1151 submit_and_realloc: 1152 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1153 bio = NULL; 1154 } 1155 if (bio == NULL) { 1156 bio = f2fs_grab_bio(inode, block_nr, nr_pages); 1157 if (IS_ERR(bio)) { 1158 bio = NULL; 1159 goto set_error_page; 1160 } 1161 bio_set_op_attrs(bio, REQ_OP_READ, 0); 1162 } 1163 1164 if (bio_add_page(bio, page, blocksize, 0) < blocksize) 1165 goto submit_and_realloc; 1166 1167 last_block_in_bio = block_nr; 1168 goto next_page; 1169 set_error_page: 1170 SetPageError(page); 1171 zero_user_segment(page, 0, PAGE_SIZE); 1172 unlock_page(page); 1173 goto next_page; 1174 confused: 1175 if (bio) { 1176 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1177 bio = NULL; 1178 } 1179 unlock_page(page); 1180 next_page: 1181 if (pages) 1182 put_page(page); 1183 } 1184 BUG_ON(pages && !list_empty(pages)); 1185 if (bio) 1186 __submit_bio(F2FS_I_SB(inode), bio, DATA); 1187 return 0; 1188 } 1189 1190 static int f2fs_read_data_page(struct file *file, struct page *page) 1191 { 1192 struct inode *inode = page->mapping->host; 1193 int ret = -EAGAIN; 1194 1195 trace_f2fs_readpage(page, DATA); 1196 1197 /* If the file has inline data, try to read it directly */ 1198 if (f2fs_has_inline_data(inode)) 1199 ret = f2fs_read_inline_data(inode, page); 1200 if (ret == -EAGAIN) 1201 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1); 1202 return ret; 1203 } 1204 1205 static int f2fs_read_data_pages(struct file *file, 1206 struct address_space *mapping, 1207 struct list_head *pages, unsigned nr_pages) 1208 { 1209 struct inode *inode = file->f_mapping->host; 1210 struct page *page = list_entry(pages->prev, struct page, lru); 1211 1212 trace_f2fs_readpages(inode, page, nr_pages); 1213 1214 /* If the file has inline data, skip readpages */ 1215 if (f2fs_has_inline_data(inode)) 1216 return 0; 1217 1218 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages); 1219 } 1220 1221 int do_write_data_page(struct f2fs_io_info *fio) 1222 { 1223 struct page *page = fio->page; 1224 struct inode *inode = page->mapping->host; 1225 struct dnode_of_data dn; 1226 int err = 0; 1227 1228 set_new_dnode(&dn, inode, NULL, NULL, 0); 1229 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 1230 if (err) 1231 return err; 1232 1233 fio->old_blkaddr = dn.data_blkaddr; 1234 1235 /* This page is already truncated */ 1236 if (fio->old_blkaddr == NULL_ADDR) { 1237 ClearPageUptodate(page); 1238 goto out_writepage; 1239 } 1240 1241 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) { 1242 gfp_t gfp_flags = GFP_NOFS; 1243 1244 /* wait for GCed encrypted page writeback */ 1245 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode), 1246 fio->old_blkaddr); 1247 retry_encrypt: 1248 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page, 1249 PAGE_SIZE, 0, 1250 fio->page->index, 1251 gfp_flags); 1252 if (IS_ERR(fio->encrypted_page)) { 1253 err = PTR_ERR(fio->encrypted_page); 1254 if (err == -ENOMEM) { 1255 /* flush pending ios and wait for a while */ 1256 f2fs_flush_merged_bios(F2FS_I_SB(inode)); 1257 congestion_wait(BLK_RW_ASYNC, HZ/50); 1258 gfp_flags |= __GFP_NOFAIL; 1259 err = 0; 1260 goto retry_encrypt; 1261 } 1262 goto out_writepage; 1263 } 1264 } 1265 1266 set_page_writeback(page); 1267 1268 /* 1269 * If current allocation needs SSR, 1270 * it had better in-place writes for updated data. 1271 */ 1272 if (unlikely(fio->old_blkaddr != NEW_ADDR && 1273 !is_cold_data(page) && 1274 !IS_ATOMIC_WRITTEN_PAGE(page) && 1275 need_inplace_update(inode))) { 1276 rewrite_data_page(fio); 1277 set_inode_flag(inode, FI_UPDATE_WRITE); 1278 trace_f2fs_do_write_data_page(page, IPU); 1279 } else { 1280 write_data_page(&dn, fio); 1281 trace_f2fs_do_write_data_page(page, OPU); 1282 set_inode_flag(inode, FI_APPEND_WRITE); 1283 if (page->index == 0) 1284 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 1285 } 1286 out_writepage: 1287 f2fs_put_dnode(&dn); 1288 return err; 1289 } 1290 1291 static int f2fs_write_data_page(struct page *page, 1292 struct writeback_control *wbc) 1293 { 1294 struct inode *inode = page->mapping->host; 1295 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1296 loff_t i_size = i_size_read(inode); 1297 const pgoff_t end_index = ((unsigned long long) i_size) 1298 >> PAGE_SHIFT; 1299 loff_t psize = (page->index + 1) << PAGE_SHIFT; 1300 unsigned offset = 0; 1301 bool need_balance_fs = false; 1302 int err = 0; 1303 struct f2fs_io_info fio = { 1304 .sbi = sbi, 1305 .type = DATA, 1306 .op = REQ_OP_WRITE, 1307 .op_flags = wbc_to_write_flags(wbc), 1308 .page = page, 1309 .encrypted_page = NULL, 1310 }; 1311 1312 trace_f2fs_writepage(page, DATA); 1313 1314 if (page->index < end_index) 1315 goto write; 1316 1317 /* 1318 * If the offset is out-of-range of file size, 1319 * this page does not have to be written to disk. 1320 */ 1321 offset = i_size & (PAGE_SIZE - 1); 1322 if ((page->index >= end_index + 1) || !offset) 1323 goto out; 1324 1325 zero_user_segment(page, offset, PAGE_SIZE); 1326 write: 1327 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1328 goto redirty_out; 1329 if (f2fs_is_drop_cache(inode)) 1330 goto out; 1331 /* we should not write 0'th page having journal header */ 1332 if (f2fs_is_volatile_file(inode) && (!page->index || 1333 (!wbc->for_reclaim && 1334 available_free_memory(sbi, BASE_CHECK)))) 1335 goto redirty_out; 1336 1337 /* we should bypass data pages to proceed the kworkder jobs */ 1338 if (unlikely(f2fs_cp_error(sbi))) { 1339 mapping_set_error(page->mapping, -EIO); 1340 goto out; 1341 } 1342 1343 /* Dentry blocks are controlled by checkpoint */ 1344 if (S_ISDIR(inode->i_mode)) { 1345 err = do_write_data_page(&fio); 1346 goto done; 1347 } 1348 1349 if (!wbc->for_reclaim) 1350 need_balance_fs = true; 1351 else if (has_not_enough_free_secs(sbi, 0, 0)) 1352 goto redirty_out; 1353 1354 err = -EAGAIN; 1355 f2fs_lock_op(sbi); 1356 if (f2fs_has_inline_data(inode)) 1357 err = f2fs_write_inline_data(inode, page); 1358 if (err == -EAGAIN) 1359 err = do_write_data_page(&fio); 1360 if (F2FS_I(inode)->last_disk_size < psize) 1361 F2FS_I(inode)->last_disk_size = psize; 1362 f2fs_unlock_op(sbi); 1363 done: 1364 if (err && err != -ENOENT) 1365 goto redirty_out; 1366 1367 out: 1368 inode_dec_dirty_pages(inode); 1369 if (err) 1370 ClearPageUptodate(page); 1371 1372 if (wbc->for_reclaim) { 1373 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE); 1374 remove_dirty_inode(inode); 1375 } 1376 1377 unlock_page(page); 1378 f2fs_balance_fs(sbi, need_balance_fs); 1379 1380 if (unlikely(f2fs_cp_error(sbi))) 1381 f2fs_submit_merged_bio(sbi, DATA, WRITE); 1382 1383 return 0; 1384 1385 redirty_out: 1386 redirty_page_for_writepage(wbc, page); 1387 if (!err) 1388 return AOP_WRITEPAGE_ACTIVATE; 1389 unlock_page(page); 1390 return err; 1391 } 1392 1393 /* 1394 * This function was copied from write_cche_pages from mm/page-writeback.c. 1395 * The major change is making write step of cold data page separately from 1396 * warm/hot data page. 1397 */ 1398 static int f2fs_write_cache_pages(struct address_space *mapping, 1399 struct writeback_control *wbc) 1400 { 1401 int ret = 0; 1402 int done = 0; 1403 struct pagevec pvec; 1404 int nr_pages; 1405 pgoff_t uninitialized_var(writeback_index); 1406 pgoff_t index; 1407 pgoff_t end; /* Inclusive */ 1408 pgoff_t done_index; 1409 int cycled; 1410 int range_whole = 0; 1411 int tag; 1412 int nwritten = 0; 1413 1414 pagevec_init(&pvec, 0); 1415 1416 if (wbc->range_cyclic) { 1417 writeback_index = mapping->writeback_index; /* prev offset */ 1418 index = writeback_index; 1419 if (index == 0) 1420 cycled = 1; 1421 else 1422 cycled = 0; 1423 end = -1; 1424 } else { 1425 index = wbc->range_start >> PAGE_SHIFT; 1426 end = wbc->range_end >> PAGE_SHIFT; 1427 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 1428 range_whole = 1; 1429 cycled = 1; /* ignore range_cyclic tests */ 1430 } 1431 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 1432 tag = PAGECACHE_TAG_TOWRITE; 1433 else 1434 tag = PAGECACHE_TAG_DIRTY; 1435 retry: 1436 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 1437 tag_pages_for_writeback(mapping, index, end); 1438 done_index = index; 1439 while (!done && (index <= end)) { 1440 int i; 1441 1442 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, 1443 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1); 1444 if (nr_pages == 0) 1445 break; 1446 1447 for (i = 0; i < nr_pages; i++) { 1448 struct page *page = pvec.pages[i]; 1449 1450 if (page->index > end) { 1451 done = 1; 1452 break; 1453 } 1454 1455 done_index = page->index; 1456 1457 lock_page(page); 1458 1459 if (unlikely(page->mapping != mapping)) { 1460 continue_unlock: 1461 unlock_page(page); 1462 continue; 1463 } 1464 1465 if (!PageDirty(page)) { 1466 /* someone wrote it for us */ 1467 goto continue_unlock; 1468 } 1469 1470 if (PageWriteback(page)) { 1471 if (wbc->sync_mode != WB_SYNC_NONE) 1472 f2fs_wait_on_page_writeback(page, 1473 DATA, true); 1474 else 1475 goto continue_unlock; 1476 } 1477 1478 BUG_ON(PageWriteback(page)); 1479 if (!clear_page_dirty_for_io(page)) 1480 goto continue_unlock; 1481 1482 ret = mapping->a_ops->writepage(page, wbc); 1483 if (unlikely(ret)) { 1484 /* 1485 * keep nr_to_write, since vfs uses this to 1486 * get # of written pages. 1487 */ 1488 if (ret == AOP_WRITEPAGE_ACTIVATE) { 1489 unlock_page(page); 1490 ret = 0; 1491 continue; 1492 } 1493 done_index = page->index + 1; 1494 done = 1; 1495 break; 1496 } else { 1497 nwritten++; 1498 } 1499 1500 if (--wbc->nr_to_write <= 0 && 1501 wbc->sync_mode == WB_SYNC_NONE) { 1502 done = 1; 1503 break; 1504 } 1505 } 1506 pagevec_release(&pvec); 1507 cond_resched(); 1508 } 1509 1510 if (!cycled && !done) { 1511 cycled = 1; 1512 index = 0; 1513 end = writeback_index - 1; 1514 goto retry; 1515 } 1516 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 1517 mapping->writeback_index = done_index; 1518 1519 if (nwritten) 1520 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host, 1521 NULL, 0, DATA, WRITE); 1522 1523 return ret; 1524 } 1525 1526 static int f2fs_write_data_pages(struct address_space *mapping, 1527 struct writeback_control *wbc) 1528 { 1529 struct inode *inode = mapping->host; 1530 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1531 struct blk_plug plug; 1532 int ret; 1533 1534 /* deal with chardevs and other special file */ 1535 if (!mapping->a_ops->writepage) 1536 return 0; 1537 1538 /* skip writing if there is no dirty page in this inode */ 1539 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) 1540 return 0; 1541 1542 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE && 1543 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && 1544 available_free_memory(sbi, DIRTY_DENTS)) 1545 goto skip_write; 1546 1547 /* skip writing during file defragment */ 1548 if (is_inode_flag_set(inode, FI_DO_DEFRAG)) 1549 goto skip_write; 1550 1551 /* during POR, we don't need to trigger writepage at all. */ 1552 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 1553 goto skip_write; 1554 1555 trace_f2fs_writepages(mapping->host, wbc, DATA); 1556 1557 blk_start_plug(&plug); 1558 ret = f2fs_write_cache_pages(mapping, wbc); 1559 blk_finish_plug(&plug); 1560 /* 1561 * if some pages were truncated, we cannot guarantee its mapping->host 1562 * to detect pending bios. 1563 */ 1564 1565 remove_dirty_inode(inode); 1566 return ret; 1567 1568 skip_write: 1569 wbc->pages_skipped += get_dirty_pages(inode); 1570 trace_f2fs_writepages(mapping->host, wbc, DATA); 1571 return 0; 1572 } 1573 1574 static void f2fs_write_failed(struct address_space *mapping, loff_t to) 1575 { 1576 struct inode *inode = mapping->host; 1577 loff_t i_size = i_size_read(inode); 1578 1579 if (to > i_size) { 1580 truncate_pagecache(inode, i_size); 1581 truncate_blocks(inode, i_size, true); 1582 } 1583 } 1584 1585 static int prepare_write_begin(struct f2fs_sb_info *sbi, 1586 struct page *page, loff_t pos, unsigned len, 1587 block_t *blk_addr, bool *node_changed) 1588 { 1589 struct inode *inode = page->mapping->host; 1590 pgoff_t index = page->index; 1591 struct dnode_of_data dn; 1592 struct page *ipage; 1593 bool locked = false; 1594 struct extent_info ei; 1595 int err = 0; 1596 1597 /* 1598 * we already allocated all the blocks, so we don't need to get 1599 * the block addresses when there is no need to fill the page. 1600 */ 1601 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE) 1602 return 0; 1603 1604 if (f2fs_has_inline_data(inode) || 1605 (pos & PAGE_MASK) >= i_size_read(inode)) { 1606 f2fs_lock_op(sbi); 1607 locked = true; 1608 } 1609 restart: 1610 /* check inline_data */ 1611 ipage = get_node_page(sbi, inode->i_ino); 1612 if (IS_ERR(ipage)) { 1613 err = PTR_ERR(ipage); 1614 goto unlock_out; 1615 } 1616 1617 set_new_dnode(&dn, inode, ipage, ipage, 0); 1618 1619 if (f2fs_has_inline_data(inode)) { 1620 if (pos + len <= MAX_INLINE_DATA) { 1621 read_inline_data(page, ipage); 1622 set_inode_flag(inode, FI_DATA_EXIST); 1623 if (inode->i_nlink) 1624 set_inline_node(ipage); 1625 } else { 1626 err = f2fs_convert_inline_page(&dn, page); 1627 if (err) 1628 goto out; 1629 if (dn.data_blkaddr == NULL_ADDR) 1630 err = f2fs_get_block(&dn, index); 1631 } 1632 } else if (locked) { 1633 err = f2fs_get_block(&dn, index); 1634 } else { 1635 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1636 dn.data_blkaddr = ei.blk + index - ei.fofs; 1637 } else { 1638 /* hole case */ 1639 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 1640 if (err || dn.data_blkaddr == NULL_ADDR) { 1641 f2fs_put_dnode(&dn); 1642 f2fs_lock_op(sbi); 1643 locked = true; 1644 goto restart; 1645 } 1646 } 1647 } 1648 1649 /* convert_inline_page can make node_changed */ 1650 *blk_addr = dn.data_blkaddr; 1651 *node_changed = dn.node_changed; 1652 out: 1653 f2fs_put_dnode(&dn); 1654 unlock_out: 1655 if (locked) 1656 f2fs_unlock_op(sbi); 1657 return err; 1658 } 1659 1660 static int f2fs_write_begin(struct file *file, struct address_space *mapping, 1661 loff_t pos, unsigned len, unsigned flags, 1662 struct page **pagep, void **fsdata) 1663 { 1664 struct inode *inode = mapping->host; 1665 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1666 struct page *page = NULL; 1667 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; 1668 bool need_balance = false; 1669 block_t blkaddr = NULL_ADDR; 1670 int err = 0; 1671 1672 trace_f2fs_write_begin(inode, pos, len, flags); 1673 1674 /* 1675 * We should check this at this moment to avoid deadlock on inode page 1676 * and #0 page. The locking rule for inline_data conversion should be: 1677 * lock_page(page #0) -> lock_page(inode_page) 1678 */ 1679 if (index != 0) { 1680 err = f2fs_convert_inline_inode(inode); 1681 if (err) 1682 goto fail; 1683 } 1684 repeat: 1685 page = grab_cache_page_write_begin(mapping, index, flags); 1686 if (!page) { 1687 err = -ENOMEM; 1688 goto fail; 1689 } 1690 1691 *pagep = page; 1692 1693 err = prepare_write_begin(sbi, page, pos, len, 1694 &blkaddr, &need_balance); 1695 if (err) 1696 goto fail; 1697 1698 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) { 1699 unlock_page(page); 1700 f2fs_balance_fs(sbi, true); 1701 lock_page(page); 1702 if (page->mapping != mapping) { 1703 /* The page got truncated from under us */ 1704 f2fs_put_page(page, 1); 1705 goto repeat; 1706 } 1707 } 1708 1709 f2fs_wait_on_page_writeback(page, DATA, false); 1710 1711 /* wait for GCed encrypted page writeback */ 1712 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) 1713 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr); 1714 1715 if (len == PAGE_SIZE || PageUptodate(page)) 1716 return 0; 1717 1718 if (blkaddr == NEW_ADDR) { 1719 zero_user_segment(page, 0, PAGE_SIZE); 1720 SetPageUptodate(page); 1721 } else { 1722 struct bio *bio; 1723 1724 bio = f2fs_grab_bio(inode, blkaddr, 1); 1725 if (IS_ERR(bio)) { 1726 err = PTR_ERR(bio); 1727 goto fail; 1728 } 1729 bio->bi_opf = REQ_OP_READ; 1730 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) { 1731 bio_put(bio); 1732 err = -EFAULT; 1733 goto fail; 1734 } 1735 1736 __submit_bio(sbi, bio, DATA); 1737 1738 lock_page(page); 1739 if (unlikely(page->mapping != mapping)) { 1740 f2fs_put_page(page, 1); 1741 goto repeat; 1742 } 1743 if (unlikely(!PageUptodate(page))) { 1744 err = -EIO; 1745 goto fail; 1746 } 1747 } 1748 return 0; 1749 1750 fail: 1751 f2fs_put_page(page, 1); 1752 f2fs_write_failed(mapping, pos + len); 1753 return err; 1754 } 1755 1756 static int f2fs_write_end(struct file *file, 1757 struct address_space *mapping, 1758 loff_t pos, unsigned len, unsigned copied, 1759 struct page *page, void *fsdata) 1760 { 1761 struct inode *inode = page->mapping->host; 1762 1763 trace_f2fs_write_end(inode, pos, len, copied); 1764 1765 /* 1766 * This should be come from len == PAGE_SIZE, and we expect copied 1767 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and 1768 * let generic_perform_write() try to copy data again through copied=0. 1769 */ 1770 if (!PageUptodate(page)) { 1771 if (unlikely(copied != PAGE_SIZE)) 1772 copied = 0; 1773 else 1774 SetPageUptodate(page); 1775 } 1776 if (!copied) 1777 goto unlock_out; 1778 1779 set_page_dirty(page); 1780 1781 if (pos + copied > i_size_read(inode)) 1782 f2fs_i_size_write(inode, pos + copied); 1783 unlock_out: 1784 f2fs_put_page(page, 1); 1785 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 1786 return copied; 1787 } 1788 1789 static int check_direct_IO(struct inode *inode, struct iov_iter *iter, 1790 loff_t offset) 1791 { 1792 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1; 1793 1794 if (offset & blocksize_mask) 1795 return -EINVAL; 1796 1797 if (iov_iter_alignment(iter) & blocksize_mask) 1798 return -EINVAL; 1799 1800 return 0; 1801 } 1802 1803 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 1804 { 1805 struct address_space *mapping = iocb->ki_filp->f_mapping; 1806 struct inode *inode = mapping->host; 1807 size_t count = iov_iter_count(iter); 1808 loff_t offset = iocb->ki_pos; 1809 int rw = iov_iter_rw(iter); 1810 int err; 1811 1812 err = check_direct_IO(inode, iter, offset); 1813 if (err) 1814 return err; 1815 1816 if (__force_buffered_io(inode, rw)) 1817 return 0; 1818 1819 trace_f2fs_direct_IO_enter(inode, offset, count, rw); 1820 1821 down_read(&F2FS_I(inode)->dio_rwsem[rw]); 1822 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio); 1823 up_read(&F2FS_I(inode)->dio_rwsem[rw]); 1824 1825 if (rw == WRITE) { 1826 if (err > 0) 1827 set_inode_flag(inode, FI_UPDATE_WRITE); 1828 else if (err < 0) 1829 f2fs_write_failed(mapping, offset + count); 1830 } 1831 1832 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); 1833 1834 return err; 1835 } 1836 1837 void f2fs_invalidate_page(struct page *page, unsigned int offset, 1838 unsigned int length) 1839 { 1840 struct inode *inode = page->mapping->host; 1841 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1842 1843 if (inode->i_ino >= F2FS_ROOT_INO(sbi) && 1844 (offset % PAGE_SIZE || length != PAGE_SIZE)) 1845 return; 1846 1847 if (PageDirty(page)) { 1848 if (inode->i_ino == F2FS_META_INO(sbi)) { 1849 dec_page_count(sbi, F2FS_DIRTY_META); 1850 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { 1851 dec_page_count(sbi, F2FS_DIRTY_NODES); 1852 } else { 1853 inode_dec_dirty_pages(inode); 1854 remove_dirty_inode(inode); 1855 } 1856 } 1857 1858 /* This is atomic written page, keep Private */ 1859 if (IS_ATOMIC_WRITTEN_PAGE(page)) 1860 return; 1861 1862 set_page_private(page, 0); 1863 ClearPagePrivate(page); 1864 } 1865 1866 int f2fs_release_page(struct page *page, gfp_t wait) 1867 { 1868 /* If this is dirty page, keep PagePrivate */ 1869 if (PageDirty(page)) 1870 return 0; 1871 1872 /* This is atomic written page, keep Private */ 1873 if (IS_ATOMIC_WRITTEN_PAGE(page)) 1874 return 0; 1875 1876 set_page_private(page, 0); 1877 ClearPagePrivate(page); 1878 return 1; 1879 } 1880 1881 /* 1882 * This was copied from __set_page_dirty_buffers which gives higher performance 1883 * in very high speed storages. (e.g., pmem) 1884 */ 1885 void f2fs_set_page_dirty_nobuffers(struct page *page) 1886 { 1887 struct address_space *mapping = page->mapping; 1888 unsigned long flags; 1889 1890 if (unlikely(!mapping)) 1891 return; 1892 1893 spin_lock(&mapping->private_lock); 1894 lock_page_memcg(page); 1895 SetPageDirty(page); 1896 spin_unlock(&mapping->private_lock); 1897 1898 spin_lock_irqsave(&mapping->tree_lock, flags); 1899 WARN_ON_ONCE(!PageUptodate(page)); 1900 account_page_dirtied(page, mapping); 1901 radix_tree_tag_set(&mapping->page_tree, 1902 page_index(page), PAGECACHE_TAG_DIRTY); 1903 spin_unlock_irqrestore(&mapping->tree_lock, flags); 1904 unlock_page_memcg(page); 1905 1906 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1907 return; 1908 } 1909 1910 static int f2fs_set_data_page_dirty(struct page *page) 1911 { 1912 struct address_space *mapping = page->mapping; 1913 struct inode *inode = mapping->host; 1914 1915 trace_f2fs_set_page_dirty(page, DATA); 1916 1917 if (!PageUptodate(page)) 1918 SetPageUptodate(page); 1919 1920 if (f2fs_is_atomic_file(inode)) { 1921 if (!IS_ATOMIC_WRITTEN_PAGE(page)) { 1922 register_inmem_page(inode, page); 1923 return 1; 1924 } 1925 /* 1926 * Previously, this page has been registered, we just 1927 * return here. 1928 */ 1929 return 0; 1930 } 1931 1932 if (!PageDirty(page)) { 1933 f2fs_set_page_dirty_nobuffers(page); 1934 update_dirty_page(inode, page); 1935 return 1; 1936 } 1937 return 0; 1938 } 1939 1940 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) 1941 { 1942 struct inode *inode = mapping->host; 1943 1944 if (f2fs_has_inline_data(inode)) 1945 return 0; 1946 1947 /* make sure allocating whole blocks */ 1948 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 1949 filemap_write_and_wait(mapping); 1950 1951 return generic_block_bmap(mapping, block, get_data_block_bmap); 1952 } 1953 1954 #ifdef CONFIG_MIGRATION 1955 #include <linux/migrate.h> 1956 1957 int f2fs_migrate_page(struct address_space *mapping, 1958 struct page *newpage, struct page *page, enum migrate_mode mode) 1959 { 1960 int rc, extra_count; 1961 struct f2fs_inode_info *fi = F2FS_I(mapping->host); 1962 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page); 1963 1964 BUG_ON(PageWriteback(page)); 1965 1966 /* migrating an atomic written page is safe with the inmem_lock hold */ 1967 if (atomic_written && !mutex_trylock(&fi->inmem_lock)) 1968 return -EAGAIN; 1969 1970 /* 1971 * A reference is expected if PagePrivate set when move mapping, 1972 * however F2FS breaks this for maintaining dirty page counts when 1973 * truncating pages. So here adjusting the 'extra_count' make it work. 1974 */ 1975 extra_count = (atomic_written ? 1 : 0) - page_has_private(page); 1976 rc = migrate_page_move_mapping(mapping, newpage, 1977 page, NULL, mode, extra_count); 1978 if (rc != MIGRATEPAGE_SUCCESS) { 1979 if (atomic_written) 1980 mutex_unlock(&fi->inmem_lock); 1981 return rc; 1982 } 1983 1984 if (atomic_written) { 1985 struct inmem_pages *cur; 1986 list_for_each_entry(cur, &fi->inmem_pages, list) 1987 if (cur->page == page) { 1988 cur->page = newpage; 1989 break; 1990 } 1991 mutex_unlock(&fi->inmem_lock); 1992 put_page(page); 1993 get_page(newpage); 1994 } 1995 1996 if (PagePrivate(page)) 1997 SetPagePrivate(newpage); 1998 set_page_private(newpage, page_private(page)); 1999 2000 migrate_page_copy(newpage, page); 2001 2002 return MIGRATEPAGE_SUCCESS; 2003 } 2004 #endif 2005 2006 const struct address_space_operations f2fs_dblock_aops = { 2007 .readpage = f2fs_read_data_page, 2008 .readpages = f2fs_read_data_pages, 2009 .writepage = f2fs_write_data_page, 2010 .writepages = f2fs_write_data_pages, 2011 .write_begin = f2fs_write_begin, 2012 .write_end = f2fs_write_end, 2013 .set_page_dirty = f2fs_set_data_page_dirty, 2014 .invalidatepage = f2fs_invalidate_page, 2015 .releasepage = f2fs_release_page, 2016 .direct_IO = f2fs_direct_IO, 2017 .bmap = f2fs_bmap, 2018 #ifdef CONFIG_MIGRATION 2019 .migratepage = f2fs_migrate_page, 2020 #endif 2021 }; 2022