fs/f2fs/data.c

   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 void f2fs_read_end_io(struct bio *bio)
  33 {
  34         struct bio_vec *bvec;
  35         int i;
  36
  37         if (f2fs_bio_encrypted(bio)) {
  38                 if (bio->bi_error) {
  39                         fscrypt_release_ctx(bio->bi_private);
  40                 } else {
  41                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  42                         return;
  43                 }
  44         }
  45
  46         bio_for_each_segment_all(bvec, bio, i) {
  47                 struct page *page = bvec->bv_page;
  48
  49                 if (!bio->bi_error) {
  50                         if (!PageUptodate(page))
  51                                 SetPageUptodate(page);
  52                 } else {
  53                         ClearPageUptodate(page);
  54                         SetPageError(page);
  55                 }
  56                 unlock_page(page);
  57         }
  58         bio_put(bio);
  59 }
  60
  61 static void f2fs_write_end_io(struct bio *bio)
  62 {
  63         struct f2fs_sb_info *sbi = bio->bi_private;
  64         struct bio_vec *bvec;
  65         int i;
  66
  67         bio_for_each_segment_all(bvec, bio, i) {
  68                 struct page *page = bvec->bv_page;
  69
  70                 fscrypt_pullback_bio_page(&page, true);
  71
  72                 if (unlikely(bio->bi_error)) {
  73                         set_bit(AS_EIO, &page->mapping->flags);
  74                         f2fs_stop_checkpoint(sbi, true);
  75                 }
  76                 end_page_writeback(page);
  77         }
  78         if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
  79                                 wq_has_sleeper(&sbi->cp_wait))
  80                 wake_up(&sbi->cp_wait);
  81
  82         bio_put(bio);
  83 }
  84
  85 /*
  86  * Low-level block read/write IO operations.
  87  */
  88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  89                                 int npages, bool is_read)
  90 {
  91         struct bio *bio;
  92
  93         bio = f2fs_bio_alloc(npages);
  94
  95         bio->bi_bdev = sbi->sb->s_bdev;
  96         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
  97         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  98         bio->bi_private = is_read ? NULL : sbi;
  99
 100         return bio;
 101 }
 102
 103 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 104                                 struct bio *bio, enum page_type type)
 105 {
 106         if (!is_read_io(bio_op(bio))) {
 107                 atomic_inc(&sbi->nr_wb_bios);
 108                 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
 109                         current->plug && (type == DATA || type == NODE))
 110                         blk_finish_plug(current->plug);
 111         }
 112         submit_bio(bio);
 113 }
 114
 115 static void __submit_merged_bio(struct f2fs_bio_info *io)
 116 {
 117         struct f2fs_io_info *fio = &io->fio;
 118
 119         if (!io->bio)
 120                 return;
 121
 122         if (is_read_io(fio->op))
 123                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 124         else
 125                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 126
 127         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 128
 129         __submit_bio(io->sbi, io->bio, fio->type);
 130         io->bio = NULL;
 131 }
 132
 133 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 134                                                 struct page *page, nid_t ino)
 135 {
 136         struct bio_vec *bvec;
 137         struct page *target;
 138         int i;
 139
 140         if (!io->bio)
 141                 return false;
 142
 143         if (!inode && !page && !ino)
 144                 return true;
 145
 146         bio_for_each_segment_all(bvec, io->bio, i) {
 147
 148                 if (bvec->bv_page->mapping)
 149                         target = bvec->bv_page;
 150                 else
 151                         target = fscrypt_control_page(bvec->bv_page);
 152
 153                 if (inode && inode == target->mapping->host)
 154                         return true;
 155                 if (page && page == target)
 156                         return true;
 157                 if (ino && ino == ino_of_node(target))
 158                         return true;
 159         }
 160
 161         return false;
 162 }
 163
 164 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 165                                                 struct page *page, nid_t ino,
 166                                                 enum page_type type)
 167 {
 168         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 169         struct f2fs_bio_info *io = &sbi->write_io[btype];
 170         bool ret;
 171
 172         down_read(&io->io_rwsem);
 173         ret = __has_merged_page(io, inode, page, ino);
 174         up_read(&io->io_rwsem);
 175         return ret;
 176 }
 177
 178 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 179                                 struct inode *inode, struct page *page,
 180                                 nid_t ino, enum page_type type, int rw)
 181 {
 182         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 183         struct f2fs_bio_info *io;
 184
 185         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 186
 187         down_write(&io->io_rwsem);
 188
 189         if (!__has_merged_page(io, inode, page, ino))
 190                 goto out;
 191
 192         /* change META to META_FLUSH in the checkpoint procedure */
 193         if (type >= META_FLUSH) {
 194                 io->fio.type = META_FLUSH;
 195                 io->fio.op = REQ_OP_WRITE;
 196                 if (test_opt(sbi, NOBARRIER))
 197                         io->fio.op_flags = WRITE_FLUSH | REQ_META | REQ_PRIO;
 198                 else
 199                         io->fio.op_flags = WRITE_FLUSH_FUA | REQ_META |
 200                                                                 REQ_PRIO;
 201         }
 202         __submit_merged_bio(io);
 203 out:
 204         up_write(&io->io_rwsem);
 205 }
 206
 207 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 208                                                                         int rw)
 209 {
 210         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 211 }
 212
 213 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 214                                 struct inode *inode, struct page *page,
 215                                 nid_t ino, enum page_type type, int rw)
 216 {
 217         if (has_merged_page(sbi, inode, page, ino, type))
 218                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 219 }
 220
 221 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 222 {
 223         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 224         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 225         f2fs_submit_merged_bio(sbi, META, WRITE);
 226 }
 227
 228 /*
 229  * Fill the locked page with data located in the block address.
 230  * Return unlocked page.
 231  */
 232 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 233 {
 234         struct bio *bio;
 235         struct page *page = fio->encrypted_page ?
 236                         fio->encrypted_page : fio->page;
 237
 238         trace_f2fs_submit_page_bio(page, fio);
 239         f2fs_trace_ios(fio, 0);
 240
 241         /* Allocate a new bio */
 242         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
 243
 244         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 245                 bio_put(bio);
 246                 return -EFAULT;
 247         }
 248         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 249
 250         __submit_bio(fio->sbi, bio, fio->type);
 251         return 0;
 252 }
 253
 254 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 255 {
 256         struct f2fs_sb_info *sbi = fio->sbi;
 257         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 258         struct f2fs_bio_info *io;
 259         bool is_read = is_read_io(fio->op);
 260         struct page *bio_page;
 261
 262         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 263
 264         if (fio->old_blkaddr != NEW_ADDR)
 265                 verify_block_addr(sbi, fio->old_blkaddr);
 266         verify_block_addr(sbi, fio->new_blkaddr);
 267
 268         down_write(&io->io_rwsem);
 269
 270         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 271             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags)))
 272                 __submit_merged_bio(io);
 273 alloc_new:
 274         if (io->bio == NULL) {
 275                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 276
 277                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 278                                                 bio_blocks, is_read);
 279                 io->fio = *fio;
 280         }
 281
 282         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 283
 284         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 285                                                         PAGE_SIZE) {
 286                 __submit_merged_bio(io);
 287                 goto alloc_new;
 288         }
 289
 290         io->last_block_in_bio = fio->new_blkaddr;
 291         f2fs_trace_ios(fio, 0);
 292
 293         up_write(&io->io_rwsem);
 294         trace_f2fs_submit_page_mbio(fio->page, fio);
 295 }
 296
 297 static void __set_data_blkaddr(struct dnode_of_data *dn)
 298 {
 299         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 300         __le32 *addr_array;
 301
 302         /* Get physical address of data block */
 303         addr_array = blkaddr_in_node(rn);
 304         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 305 }
 306
 307 /*
 308  * Lock ordering for the change of data block address:
 309  * ->data_page
 310  *  ->node_page
 311  *    update block addresses in the node page
 312  */
 313 void set_data_blkaddr(struct dnode_of_data *dn)
 314 {
 315         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 316         __set_data_blkaddr(dn);
 317         if (set_page_dirty(dn->node_page))
 318                 dn->node_changed = true;
 319 }
 320
 321 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 322 {
 323         dn->data_blkaddr = blkaddr;
 324         set_data_blkaddr(dn);
 325         f2fs_update_extent_cache(dn);
 326 }
 327
 328 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 329 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 330 {
 331         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 332
 333         if (!count)
 334                 return 0;
 335
 336         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 337                 return -EPERM;
 338         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 339                 return -ENOSPC;
 340
 341         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 342                                                 dn->ofs_in_node, count);
 343
 344         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 345
 346         for (; count > 0; dn->ofs_in_node++) {
 347                 block_t blkaddr =
 348                         datablock_addr(dn->node_page, dn->ofs_in_node);
 349                 if (blkaddr == NULL_ADDR) {
 350                         dn->data_blkaddr = NEW_ADDR;
 351                         __set_data_blkaddr(dn);
 352                         count--;
 353                 }
 354         }
 355
 356         if (set_page_dirty(dn->node_page))
 357                 dn->node_changed = true;
 358         return 0;
 359 }
 360
 361 /* Should keep dn->ofs_in_node unchanged */
 362 int reserve_new_block(struct dnode_of_data *dn)
 363 {
 364         unsigned int ofs_in_node = dn->ofs_in_node;
 365         int ret;
 366
 367         ret = reserve_new_blocks(dn, 1);
 368         dn->ofs_in_node = ofs_in_node;
 369         return ret;
 370 }
 371
 372 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 373 {
 374         bool need_put = dn->inode_page ? false : true;
 375         int err;
 376
 377         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 378         if (err)
 379                 return err;
 380
 381         if (dn->data_blkaddr == NULL_ADDR)
 382                 err = reserve_new_block(dn);
 383         if (err || need_put)
 384                 f2fs_put_dnode(dn);
 385         return err;
 386 }
 387
 388 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 389 {
 390         struct extent_info ei;
 391         struct inode *inode = dn->inode;
 392
 393         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 394                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 395                 return 0;
 396         }
 397
 398         return f2fs_reserve_block(dn, index);
 399 }
 400
 401 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 402                                                 int op_flags, bool for_write)
 403 {
 404         struct address_space *mapping = inode->i_mapping;
 405         struct dnode_of_data dn;
 406         struct page *page;
 407         struct extent_info ei;
 408         int err;
 409         struct f2fs_io_info fio = {
 410                 .sbi = F2FS_I_SB(inode),
 411                 .type = DATA,
 412                 .op = REQ_OP_READ,
 413                 .op_flags = op_flags,
 414                 .encrypted_page = NULL,
 415         };
 416
 417         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 418                 return read_mapping_page(mapping, index, NULL);
 419
 420         page = f2fs_grab_cache_page(mapping, index, for_write);
 421         if (!page)
 422                 return ERR_PTR(-ENOMEM);
 423
 424         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 425                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 426                 goto got_it;
 427         }
 428
 429         set_new_dnode(&dn, inode, NULL, NULL, 0);
 430         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 431         if (err)
 432                 goto put_err;
 433         f2fs_put_dnode(&dn);
 434
 435         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 436                 err = -ENOENT;
 437                 goto put_err;
 438         }
 439 got_it:
 440         if (PageUptodate(page)) {
 441                 unlock_page(page);
 442                 return page;
 443         }
 444
 445         /*
 446          * A new dentry page is allocated but not able to be written, since its
 447          * new inode page couldn't be allocated due to -ENOSPC.
 448          * In such the case, its blkaddr can be remained as NEW_ADDR.
 449          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 450          */
 451         if (dn.data_blkaddr == NEW_ADDR) {
 452                 zero_user_segment(page, 0, PAGE_SIZE);
 453                 if (!PageUptodate(page))
 454                         SetPageUptodate(page);
 455                 unlock_page(page);
 456                 return page;
 457         }
 458
 459         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 460         fio.page = page;
 461         err = f2fs_submit_page_bio(&fio);
 462         if (err)
 463                 goto put_err;
 464         return page;
 465
 466 put_err:
 467         f2fs_put_page(page, 1);
 468         return ERR_PTR(err);
 469 }
 470
 471 struct page *find_data_page(struct inode *inode, pgoff_t index)
 472 {
 473         struct address_space *mapping = inode->i_mapping;
 474         struct page *page;
 475
 476         page = find_get_page(mapping, index);
 477         if (page && PageUptodate(page))
 478                 return page;
 479         f2fs_put_page(page, 0);
 480
 481         page = get_read_data_page(inode, index, READ_SYNC, false);
 482         if (IS_ERR(page))
 483                 return page;
 484
 485         if (PageUptodate(page))
 486                 return page;
 487
 488         wait_on_page_locked(page);
 489         if (unlikely(!PageUptodate(page))) {
 490                 f2fs_put_page(page, 0);
 491                 return ERR_PTR(-EIO);
 492         }
 493         return page;
 494 }
 495
 496 /*
 497  * If it tries to access a hole, return an error.
 498  * Because, the callers, functions in dir.c and GC, should be able to know
 499  * whether this page exists or not.
 500  */
 501 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 502                                                         bool for_write)
 503 {
 504         struct address_space *mapping = inode->i_mapping;
 505         struct page *page;
 506 repeat:
 507         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 508         if (IS_ERR(page))
 509                 return page;
 510
 511         /* wait for read completion */
 512         lock_page(page);
 513         if (unlikely(page->mapping != mapping)) {
 514                 f2fs_put_page(page, 1);
 515                 goto repeat;
 516         }
 517         if (unlikely(!PageUptodate(page))) {
 518                 f2fs_put_page(page, 1);
 519                 return ERR_PTR(-EIO);
 520         }
 521         return page;
 522 }
 523
 524 /*
 525  * Caller ensures that this data page is never allocated.
 526  * A new zero-filled data page is allocated in the page cache.
 527  *
 528  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 529  * f2fs_unlock_op().
 530  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 531  * ipage should be released by this function.
 532  */
 533 struct page *get_new_data_page(struct inode *inode,
 534                 struct page *ipage, pgoff_t index, bool new_i_size)
 535 {
 536         struct address_space *mapping = inode->i_mapping;
 537         struct page *page;
 538         struct dnode_of_data dn;
 539         int err;
 540
 541         page = f2fs_grab_cache_page(mapping, index, true);
 542         if (!page) {
 543                 /*
 544                  * before exiting, we should make sure ipage will be released
 545                  * if any error occur.
 546                  */
 547                 f2fs_put_page(ipage, 1);
 548                 return ERR_PTR(-ENOMEM);
 549         }
 550
 551         set_new_dnode(&dn, inode, ipage, NULL, 0);
 552         err = f2fs_reserve_block(&dn, index);
 553         if (err) {
 554                 f2fs_put_page(page, 1);
 555                 return ERR_PTR(err);
 556         }
 557         if (!ipage)
 558                 f2fs_put_dnode(&dn);
 559
 560         if (PageUptodate(page))
 561                 goto got_it;
 562
 563         if (dn.data_blkaddr == NEW_ADDR) {
 564                 zero_user_segment(page, 0, PAGE_SIZE);
 565                 if (!PageUptodate(page))
 566                         SetPageUptodate(page);
 567         } else {
 568                 f2fs_put_page(page, 1);
 569
 570                 /* if ipage exists, blkaddr should be NEW_ADDR */
 571                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 572                 page = get_lock_data_page(inode, index, true);
 573                 if (IS_ERR(page))
 574                         return page;
 575         }
 576 got_it:
 577         if (new_i_size && i_size_read(inode) <
 578                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 579                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 580         return page;
 581 }
 582
 583 static int __allocate_data_block(struct dnode_of_data *dn)
 584 {
 585         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 586         struct f2fs_summary sum;
 587         struct node_info ni;
 588         int seg = CURSEG_WARM_DATA;
 589         pgoff_t fofs;
 590         blkcnt_t count = 1;
 591
 592         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 593                 return -EPERM;
 594
 595         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 596         if (dn->data_blkaddr == NEW_ADDR)
 597                 goto alloc;
 598
 599         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 600                 return -ENOSPC;
 601
 602 alloc:
 603         get_node_info(sbi, dn->nid, &ni);
 604         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 605
 606         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 607                 seg = CURSEG_DIRECT_IO;
 608
 609         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 610                                                                 &sum, seg);
 611         set_data_blkaddr(dn);
 612
 613         /* update i_size */
 614         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 615                                                         dn->ofs_in_node;
 616         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 617                 f2fs_i_size_write(dn->inode,
 618                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 619         return 0;
 620 }
 621
 622 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 623 {
 624         struct inode *inode = file_inode(iocb->ki_filp);
 625         struct f2fs_map_blocks map;
 626         ssize_t ret = 0;
 627
 628         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 629         map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
 630         map.m_next_pgofs = NULL;
 631
 632         if (f2fs_encrypted_inode(inode))
 633                 return 0;
 634
 635         if (iocb->ki_flags & IOCB_DIRECT) {
 636                 ret = f2fs_convert_inline_inode(inode);
 637                 if (ret)
 638                         return ret;
 639                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 640         }
 641         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 642                 ret = f2fs_convert_inline_inode(inode);
 643                 if (ret)
 644                         return ret;
 645         }
 646         if (!f2fs_has_inline_data(inode))
 647                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 648         return ret;
 649 }
 650
 651 /*
 652  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 653  * f2fs_map_blocks structure.
 654  * If original data blocks are allocated, then give them to blockdev.
 655  * Otherwise,
 656  *     a. preallocate requested block addresses
 657  *     b. do not use extent cache for better performance
 658  *     c. give the block addresses to blockdev
 659  */
 660 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 661                                                 int create, int flag)
 662 {
 663         unsigned int maxblocks = map->m_len;
 664         struct dnode_of_data dn;
 665         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 666         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 667         pgoff_t pgofs, end_offset, end;
 668         int err = 0, ofs = 1;
 669         unsigned int ofs_in_node, last_ofs_in_node;
 670         blkcnt_t prealloc;
 671         struct extent_info ei;
 672         bool allocated = false;
 673         block_t blkaddr;
 674
 675         map->m_len = 0;
 676         map->m_flags = 0;
 677
 678         /* it only supports block size == page size */
 679         pgofs = (pgoff_t)map->m_lblk;
 680         end = pgofs + maxblocks;
 681
 682         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 683                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 684                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 685                 map->m_flags = F2FS_MAP_MAPPED;
 686                 goto out;
 687         }
 688
 689 next_dnode:
 690         if (create)
 691                 f2fs_lock_op(sbi);
 692
 693         /* When reading holes, we need its node page */
 694         set_new_dnode(&dn, inode, NULL, NULL, 0);
 695         err = get_dnode_of_data(&dn, pgofs, mode);
 696         if (err) {
 697                 if (flag == F2FS_GET_BLOCK_BMAP)
 698                         map->m_pblk = 0;
 699                 if (err == -ENOENT) {
 700                         err = 0;
 701                         if (map->m_next_pgofs)
 702                                 *map->m_next_pgofs =
 703                                         get_next_page_offset(&dn, pgofs);
 704                 }
 705                 goto unlock_out;
 706         }
 707
 708         prealloc = 0;
 709         ofs_in_node = dn.ofs_in_node;
 710         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 711
 712 next_block:
 713         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 714
 715         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 716                 if (create) {
 717                         if (unlikely(f2fs_cp_error(sbi))) {
 718                                 err = -EIO;
 719                                 goto sync_out;
 720                         }
 721                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 722                                 if (blkaddr == NULL_ADDR) {
 723                                         prealloc++;
 724                                         last_ofs_in_node = dn.ofs_in_node;
 725                                 }
 726                         } else {
 727                                 err = __allocate_data_block(&dn);
 728                                 if (!err) {
 729                                         set_inode_flag(inode, FI_APPEND_WRITE);
 730                                         allocated = true;
 731                                 }
 732                         }
 733                         if (err)
 734                                 goto sync_out;
 735                         map->m_flags = F2FS_MAP_NEW;
 736                         blkaddr = dn.data_blkaddr;
 737                 } else {
 738                         if (flag == F2FS_GET_BLOCK_BMAP) {
 739                                 map->m_pblk = 0;
 740                                 goto sync_out;
 741                         }
 742                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 743                                                 blkaddr == NULL_ADDR) {
 744                                 if (map->m_next_pgofs)
 745                                         *map->m_next_pgofs = pgofs + 1;
 746                         }
 747                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 748                                                 blkaddr != NEW_ADDR)
 749                                 goto sync_out;
 750                 }
 751         }
 752
 753         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 754                 goto skip;
 755
 756         if (map->m_len == 0) {
 757                 /* preallocated unwritten block should be mapped for fiemap. */
 758                 if (blkaddr == NEW_ADDR)
 759                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 760                 map->m_flags |= F2FS_MAP_MAPPED;
 761
 762                 map->m_pblk = blkaddr;
 763                 map->m_len = 1;
 764         } else if ((map->m_pblk != NEW_ADDR &&
 765                         blkaddr == (map->m_pblk + ofs)) ||
 766                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 767                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 768                 ofs++;
 769                 map->m_len++;
 770         } else {
 771                 goto sync_out;
 772         }
 773
 774 skip:
 775         dn.ofs_in_node++;
 776         pgofs++;
 777
 778         /* preallocate blocks in batch for one dnode page */
 779         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 780                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 781
 782                 dn.ofs_in_node = ofs_in_node;
 783                 err = reserve_new_blocks(&dn, prealloc);
 784                 if (err)
 785                         goto sync_out;
 786
 787                 map->m_len += dn.ofs_in_node - ofs_in_node;
 788                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 789                         err = -ENOSPC;
 790                         goto sync_out;
 791                 }
 792                 dn.ofs_in_node = end_offset;
 793         }
 794
 795         if (pgofs >= end)
 796                 goto sync_out;
 797         else if (dn.ofs_in_node < end_offset)
 798                 goto next_block;
 799
 800         f2fs_put_dnode(&dn);
 801
 802         if (create) {
 803                 f2fs_unlock_op(sbi);
 804                 f2fs_balance_fs(sbi, allocated);
 805         }
 806         allocated = false;
 807         goto next_dnode;
 808
 809 sync_out:
 810         f2fs_put_dnode(&dn);
 811 unlock_out:
 812         if (create) {
 813                 f2fs_unlock_op(sbi);
 814                 f2fs_balance_fs(sbi, allocated);
 815         }
 816 out:
 817         trace_f2fs_map_blocks(inode, map, err);
 818         return err;
 819 }
 820
 821 static int __get_data_block(struct inode *inode, sector_t iblock,
 822                         struct buffer_head *bh, int create, int flag,
 823                         pgoff_t *next_pgofs)
 824 {
 825         struct f2fs_map_blocks map;
 826         int ret;
 827
 828         map.m_lblk = iblock;
 829         map.m_len = bh->b_size >> inode->i_blkbits;
 830         map.m_next_pgofs = next_pgofs;
 831
 832         ret = f2fs_map_blocks(inode, &map, create, flag);
 833         if (!ret) {
 834                 map_bh(bh, inode->i_sb, map.m_pblk);
 835                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 836                 bh->b_size = map.m_len << inode->i_blkbits;
 837         }
 838         return ret;
 839 }
 840
 841 static int get_data_block(struct inode *inode, sector_t iblock,
 842                         struct buffer_head *bh_result, int create, int flag,
 843                         pgoff_t *next_pgofs)
 844 {
 845         return __get_data_block(inode, iblock, bh_result, create,
 846                                                         flag, next_pgofs);
 847 }
 848
 849 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 850                         struct buffer_head *bh_result, int create)
 851 {
 852         return __get_data_block(inode, iblock, bh_result, create,
 853                                                 F2FS_GET_BLOCK_DIO, NULL);
 854 }
 855
 856 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 857                         struct buffer_head *bh_result, int create)
 858 {
 859         /* Block number less than F2FS MAX BLOCKS */
 860         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 861                 return -EFBIG;
 862
 863         return __get_data_block(inode, iblock, bh_result, create,
 864                                                 F2FS_GET_BLOCK_BMAP, NULL);
 865 }
 866
 867 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 868 {
 869         return (offset >> inode->i_blkbits);
 870 }
 871
 872 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 873 {
 874         return (blk << inode->i_blkbits);
 875 }
 876
 877 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 878                 u64 start, u64 len)
 879 {
 880         struct buffer_head map_bh;
 881         sector_t start_blk, last_blk;
 882         pgoff_t next_pgofs;
 883         loff_t isize;
 884         u64 logical = 0, phys = 0, size = 0;
 885         u32 flags = 0;
 886         int ret = 0;
 887
 888         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 889         if (ret)
 890                 return ret;
 891
 892         if (f2fs_has_inline_data(inode)) {
 893                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 894                 if (ret != -EAGAIN)
 895                         return ret;
 896         }
 897
 898         inode_lock(inode);
 899
 900         isize = i_size_read(inode);
 901         if (start >= isize)
 902                 goto out;
 903
 904         if (start + len > isize)
 905                 len = isize - start;
 906
 907         if (logical_to_blk(inode, len) == 0)
 908                 len = blk_to_logical(inode, 1);
 909
 910         start_blk = logical_to_blk(inode, start);
 911         last_blk = logical_to_blk(inode, start + len - 1);
 912
 913 next:
 914         memset(&map_bh, 0, sizeof(struct buffer_head));
 915         map_bh.b_size = len;
 916
 917         ret = get_data_block(inode, start_blk, &map_bh, 0,
 918                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 919         if (ret)
 920                 goto out;
 921
 922         /* HOLE */
 923         if (!buffer_mapped(&map_bh)) {
 924                 start_blk = next_pgofs;
 925                 /* Go through holes util pass the EOF */
 926                 if (blk_to_logical(inode, start_blk) < isize)
 927                         goto prep_next;
 928                 /* Found a hole beyond isize means no more extents.
 929                  * Note that the premise is that filesystems don't
 930                  * punch holes beyond isize and keep size unchanged.
 931                  */
 932                 flags |= FIEMAP_EXTENT_LAST;
 933         }
 934
 935         if (size) {
 936                 if (f2fs_encrypted_inode(inode))
 937                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 938
 939                 ret = fiemap_fill_next_extent(fieinfo, logical,
 940                                 phys, size, flags);
 941         }
 942
 943         if (start_blk > last_blk || ret)
 944                 goto out;
 945
 946         logical = blk_to_logical(inode, start_blk);
 947         phys = blk_to_logical(inode, map_bh.b_blocknr);
 948         size = map_bh.b_size;
 949         flags = 0;
 950         if (buffer_unwritten(&map_bh))
 951                 flags = FIEMAP_EXTENT_UNWRITTEN;
 952
 953         start_blk += logical_to_blk(inode, size);
 954
 955 prep_next:
 956         cond_resched();
 957         if (fatal_signal_pending(current))
 958                 ret = -EINTR;
 959         else
 960                 goto next;
 961 out:
 962         if (ret == 1)
 963                 ret = 0;
 964
 965         inode_unlock(inode);
 966         return ret;
 967 }
 968
 969 struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
 970                                                         unsigned nr_pages)
 971 {
 972         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 973         struct fscrypt_ctx *ctx = NULL;
 974         struct block_device *bdev = sbi->sb->s_bdev;
 975         struct bio *bio;
 976
 977         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
 978                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
 979                 if (IS_ERR(ctx))
 980                         return ERR_CAST(ctx);
 981
 982                 /* wait the page to be moved by cleaning */
 983                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
 984         }
 985
 986         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
 987         if (!bio) {
 988                 if (ctx)
 989                         fscrypt_release_ctx(ctx);
 990                 return ERR_PTR(-ENOMEM);
 991         }
 992         bio->bi_bdev = bdev;
 993         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
 994         bio->bi_end_io = f2fs_read_end_io;
 995         bio->bi_private = ctx;
 996
 997         return bio;
 998 }
 999
1000 /*
1001  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1002  * Major change was from block_size == page_size in f2fs by default.
1003  */
1004 static int f2fs_mpage_readpages(struct address_space *mapping,
1005                         struct list_head *pages, struct page *page,
1006                         unsigned nr_pages)
1007 {
1008         struct bio *bio = NULL;
1009         unsigned page_idx;
1010         sector_t last_block_in_bio = 0;
1011         struct inode *inode = mapping->host;
1012         const unsigned blkbits = inode->i_blkbits;
1013         const unsigned blocksize = 1 << blkbits;
1014         sector_t block_in_file;
1015         sector_t last_block;
1016         sector_t last_block_in_file;
1017         sector_t block_nr;
1018         struct f2fs_map_blocks map;
1019
1020         map.m_pblk = 0;
1021         map.m_lblk = 0;
1022         map.m_len = 0;
1023         map.m_flags = 0;
1024         map.m_next_pgofs = NULL;
1025
1026         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1027
1028                 prefetchw(&page->flags);
1029                 if (pages) {
1030                         page = list_entry(pages->prev, struct page, lru);
1031                         list_del(&page->lru);
1032                         if (add_to_page_cache_lru(page, mapping,
1033                                                   page->index,
1034                                                   readahead_gfp_mask(mapping)))
1035                                 goto next_page;
1036                 }
1037
1038                 block_in_file = (sector_t)page->index;
1039                 last_block = block_in_file + nr_pages;
1040                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1041                                                                 blkbits;
1042                 if (last_block > last_block_in_file)
1043                         last_block = last_block_in_file;
1044
1045                 /*
1046                  * Map blocks using the previous result first.
1047                  */
1048                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1049                                 block_in_file > map.m_lblk &&
1050                                 block_in_file < (map.m_lblk + map.m_len))
1051                         goto got_it;
1052
1053                 /*
1054                  * Then do more f2fs_map_blocks() calls until we are
1055                  * done with this page.
1056                  */
1057                 map.m_flags = 0;
1058
1059                 if (block_in_file < last_block) {
1060                         map.m_lblk = block_in_file;
1061                         map.m_len = last_block - block_in_file;
1062
1063                         if (f2fs_map_blocks(inode, &map, 0,
1064                                                 F2FS_GET_BLOCK_READ))
1065                                 goto set_error_page;
1066                 }
1067 got_it:
1068                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1069                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1070                         SetPageMappedToDisk(page);
1071
1072                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1073                                 SetPageUptodate(page);
1074                                 goto confused;
1075                         }
1076                 } else {
1077                         zero_user_segment(page, 0, PAGE_SIZE);
1078                         if (!PageUptodate(page))
1079                                 SetPageUptodate(page);
1080                         unlock_page(page);
1081                         goto next_page;
1082                 }
1083
1084                 /*
1085                  * This page will go to BIO.  Do we need to send this
1086                  * BIO off first?
1087                  */
1088                 if (bio && (last_block_in_bio != block_nr - 1)) {
1089 submit_and_realloc:
1090                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1091                         bio = NULL;
1092                 }
1093                 if (bio == NULL) {
1094                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1095                         if (IS_ERR(bio)) {
1096                                 bio = NULL;
1097                                 goto set_error_page;
1098                         }
1099                         bio_set_op_attrs(bio, REQ_OP_READ, 0);
1100                 }
1101
1102                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1103                         goto submit_and_realloc;
1104
1105                 last_block_in_bio = block_nr;
1106                 goto next_page;
1107 set_error_page:
1108                 SetPageError(page);
1109                 zero_user_segment(page, 0, PAGE_SIZE);
1110                 unlock_page(page);
1111                 goto next_page;
1112 confused:
1113                 if (bio) {
1114                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1115                         bio = NULL;
1116                 }
1117                 unlock_page(page);
1118 next_page:
1119                 if (pages)
1120                         put_page(page);
1121         }
1122         BUG_ON(pages && !list_empty(pages));
1123         if (bio)
1124                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1125         return 0;
1126 }
1127
1128 static int f2fs_read_data_page(struct file *file, struct page *page)
1129 {
1130         struct inode *inode = page->mapping->host;
1131         int ret = -EAGAIN;
1132
1133         trace_f2fs_readpage(page, DATA);
1134
1135         /* If the file has inline data, try to read it directly */
1136         if (f2fs_has_inline_data(inode))
1137                 ret = f2fs_read_inline_data(inode, page);
1138         if (ret == -EAGAIN)
1139                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1140         return ret;
1141 }
1142
1143 static int f2fs_read_data_pages(struct file *file,
1144                         struct address_space *mapping,
1145                         struct list_head *pages, unsigned nr_pages)
1146 {
1147         struct inode *inode = file->f_mapping->host;
1148         struct page *page = list_entry(pages->prev, struct page, lru);
1149
1150         trace_f2fs_readpages(inode, page, nr_pages);
1151
1152         /* If the file has inline data, skip readpages */
1153         if (f2fs_has_inline_data(inode))
1154                 return 0;
1155
1156         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1157 }
1158
1159 int do_write_data_page(struct f2fs_io_info *fio)
1160 {
1161         struct page *page = fio->page;
1162         struct inode *inode = page->mapping->host;
1163         struct dnode_of_data dn;
1164         int err = 0;
1165
1166         set_new_dnode(&dn, inode, NULL, NULL, 0);
1167         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1168         if (err)
1169                 return err;
1170
1171         fio->old_blkaddr = dn.data_blkaddr;
1172
1173         /* This page is already truncated */
1174         if (fio->old_blkaddr == NULL_ADDR) {
1175                 ClearPageUptodate(page);
1176                 goto out_writepage;
1177         }
1178
1179         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1180                 gfp_t gfp_flags = GFP_NOFS;
1181
1182                 /* wait for GCed encrypted page writeback */
1183                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1184                                                         fio->old_blkaddr);
1185 retry_encrypt:
1186                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1187                                                                 gfp_flags);
1188                 if (IS_ERR(fio->encrypted_page)) {
1189                         err = PTR_ERR(fio->encrypted_page);
1190                         if (err == -ENOMEM) {
1191                                 /* flush pending ios and wait for a while */
1192                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1193                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1194                                 gfp_flags |= __GFP_NOFAIL;
1195                                 err = 0;
1196                                 goto retry_encrypt;
1197                         }
1198                         goto out_writepage;
1199                 }
1200         }
1201
1202         set_page_writeback(page);
1203
1204         /*
1205          * If current allocation needs SSR,
1206          * it had better in-place writes for updated data.
1207          */
1208         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1209                         !is_cold_data(page) &&
1210                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1211                         need_inplace_update(inode))) {
1212                 rewrite_data_page(fio);
1213                 set_inode_flag(inode, FI_UPDATE_WRITE);
1214                 trace_f2fs_do_write_data_page(page, IPU);
1215         } else {
1216                 write_data_page(&dn, fio);
1217                 trace_f2fs_do_write_data_page(page, OPU);
1218                 set_inode_flag(inode, FI_APPEND_WRITE);
1219                 if (page->index == 0)
1220                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1221         }
1222 out_writepage:
1223         f2fs_put_dnode(&dn);
1224         return err;
1225 }
1226
1227 static int f2fs_write_data_page(struct page *page,
1228                                         struct writeback_control *wbc)
1229 {
1230         struct inode *inode = page->mapping->host;
1231         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1232         loff_t i_size = i_size_read(inode);
1233         const pgoff_t end_index = ((unsigned long long) i_size)
1234                                                         >> PAGE_SHIFT;
1235         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1236         unsigned offset = 0;
1237         bool need_balance_fs = false;
1238         int err = 0;
1239         struct f2fs_io_info fio = {
1240                 .sbi = sbi,
1241                 .type = DATA,
1242                 .op = REQ_OP_WRITE,
1243                 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1244                 .page = page,
1245                 .encrypted_page = NULL,
1246         };
1247
1248         trace_f2fs_writepage(page, DATA);
1249
1250         if (page->index < end_index)
1251                 goto write;
1252
1253         /*
1254          * If the offset is out-of-range of file size,
1255          * this page does not have to be written to disk.
1256          */
1257         offset = i_size & (PAGE_SIZE - 1);
1258         if ((page->index >= end_index + 1) || !offset)
1259                 goto out;
1260
1261         zero_user_segment(page, offset, PAGE_SIZE);
1262 write:
1263         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1264                 goto redirty_out;
1265         if (f2fs_is_drop_cache(inode))
1266                 goto out;
1267         /* we should not write 0'th page having journal header */
1268         if (f2fs_is_volatile_file(inode) && (!page->index ||
1269                         (!wbc->for_reclaim &&
1270                         available_free_memory(sbi, BASE_CHECK))))
1271                 goto redirty_out;
1272
1273         /* we should bypass data pages to proceed the kworkder jobs */
1274         if (unlikely(f2fs_cp_error(sbi))) {
1275                 mapping_set_error(page->mapping, -EIO);
1276                 goto out;
1277         }
1278
1279         /* Dentry blocks are controlled by checkpoint */
1280         if (S_ISDIR(inode->i_mode)) {
1281                 err = do_write_data_page(&fio);
1282                 goto done;
1283         }
1284
1285         if (!wbc->for_reclaim)
1286                 need_balance_fs = true;
1287         else if (has_not_enough_free_secs(sbi, 0))
1288                 goto redirty_out;
1289
1290         err = -EAGAIN;
1291         f2fs_lock_op(sbi);
1292         if (f2fs_has_inline_data(inode))
1293                 err = f2fs_write_inline_data(inode, page);
1294         if (err == -EAGAIN)
1295                 err = do_write_data_page(&fio);
1296         if (F2FS_I(inode)->last_disk_size < psize)
1297                 F2FS_I(inode)->last_disk_size = psize;
1298         f2fs_unlock_op(sbi);
1299 done:
1300         if (err && err != -ENOENT)
1301                 goto redirty_out;
1302
1303         clear_cold_data(page);
1304 out:
1305         inode_dec_dirty_pages(inode);
1306         if (err)
1307                 ClearPageUptodate(page);
1308
1309         if (wbc->for_reclaim) {
1310                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1311                 remove_dirty_inode(inode);
1312         }
1313
1314         unlock_page(page);
1315         f2fs_balance_fs(sbi, need_balance_fs);
1316
1317         if (unlikely(f2fs_cp_error(sbi)))
1318                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1319
1320         return 0;
1321
1322 redirty_out:
1323         redirty_page_for_writepage(wbc, page);
1324         unlock_page(page);
1325         return err;
1326 }
1327
1328 /*
1329  * This function was copied from write_cche_pages from mm/page-writeback.c.
1330  * The major change is making write step of cold data page separately from
1331  * warm/hot data page.
1332  */
1333 static int f2fs_write_cache_pages(struct address_space *mapping,
1334                                         struct writeback_control *wbc)
1335 {
1336         int ret = 0;
1337         int done = 0;
1338         struct pagevec pvec;
1339         int nr_pages;
1340         pgoff_t uninitialized_var(writeback_index);
1341         pgoff_t index;
1342         pgoff_t end;            /* Inclusive */
1343         pgoff_t done_index;
1344         int cycled;
1345         int range_whole = 0;
1346         int tag;
1347
1348         pagevec_init(&pvec, 0);
1349
1350         if (wbc->range_cyclic) {
1351                 writeback_index = mapping->writeback_index; /* prev offset */
1352                 index = writeback_index;
1353                 if (index == 0)
1354                         cycled = 1;
1355                 else
1356                         cycled = 0;
1357                 end = -1;
1358         } else {
1359                 index = wbc->range_start >> PAGE_SHIFT;
1360                 end = wbc->range_end >> PAGE_SHIFT;
1361                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1362                         range_whole = 1;
1363                 cycled = 1; /* ignore range_cyclic tests */
1364         }
1365         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1366                 tag = PAGECACHE_TAG_TOWRITE;
1367         else
1368                 tag = PAGECACHE_TAG_DIRTY;
1369 retry:
1370         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1371                 tag_pages_for_writeback(mapping, index, end);
1372         done_index = index;
1373         while (!done && (index <= end)) {
1374                 int i;
1375
1376                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1377                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1378                 if (nr_pages == 0)
1379                         break;
1380
1381                 for (i = 0; i < nr_pages; i++) {
1382                         struct page *page = pvec.pages[i];
1383
1384                         if (page->index > end) {
1385                                 done = 1;
1386                                 break;
1387                         }
1388
1389                         done_index = page->index;
1390
1391                         lock_page(page);
1392
1393                         if (unlikely(page->mapping != mapping)) {
1394 continue_unlock:
1395                                 unlock_page(page);
1396                                 continue;
1397                         }
1398
1399                         if (!PageDirty(page)) {
1400                                 /* someone wrote it for us */
1401                                 goto continue_unlock;
1402                         }
1403
1404                         if (PageWriteback(page)) {
1405                                 if (wbc->sync_mode != WB_SYNC_NONE)
1406                                         f2fs_wait_on_page_writeback(page,
1407                                                                 DATA, true);
1408                                 else
1409                                         goto continue_unlock;
1410                         }
1411
1412                         BUG_ON(PageWriteback(page));
1413                         if (!clear_page_dirty_for_io(page))
1414                                 goto continue_unlock;
1415
1416                         ret = mapping->a_ops->writepage(page, wbc);
1417                         if (unlikely(ret)) {
1418                                 done_index = page->index + 1;
1419                                 done = 1;
1420                                 break;
1421                         }
1422
1423                         if (--wbc->nr_to_write <= 0 &&
1424                             wbc->sync_mode == WB_SYNC_NONE) {
1425                                 done = 1;
1426                                 break;
1427                         }
1428                 }
1429                 pagevec_release(&pvec);
1430                 cond_resched();
1431         }
1432
1433         if (!cycled && !done) {
1434                 cycled = 1;
1435                 index = 0;
1436                 end = writeback_index - 1;
1437                 goto retry;
1438         }
1439         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1440                 mapping->writeback_index = done_index;
1441
1442         return ret;
1443 }
1444
1445 static int f2fs_write_data_pages(struct address_space *mapping,
1446                             struct writeback_control *wbc)
1447 {
1448         struct inode *inode = mapping->host;
1449         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1450         struct blk_plug plug;
1451         int ret;
1452
1453         /* deal with chardevs and other special file */
1454         if (!mapping->a_ops->writepage)
1455                 return 0;
1456
1457         /* skip writing if there is no dirty page in this inode */
1458         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1459                 return 0;
1460
1461         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1462                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1463                         available_free_memory(sbi, DIRTY_DENTS))
1464                 goto skip_write;
1465
1466         /* skip writing during file defragment */
1467         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1468                 goto skip_write;
1469
1470         /* during POR, we don't need to trigger writepage at all. */
1471         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1472                 goto skip_write;
1473
1474         trace_f2fs_writepages(mapping->host, wbc, DATA);
1475
1476         blk_start_plug(&plug);
1477         ret = f2fs_write_cache_pages(mapping, wbc);
1478         blk_finish_plug(&plug);
1479         /*
1480          * if some pages were truncated, we cannot guarantee its mapping->host
1481          * to detect pending bios.
1482          */
1483         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1484
1485         remove_dirty_inode(inode);
1486         return ret;
1487
1488 skip_write:
1489         wbc->pages_skipped += get_dirty_pages(inode);
1490         trace_f2fs_writepages(mapping->host, wbc, DATA);
1491         return 0;
1492 }
1493
1494 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1495 {
1496         struct inode *inode = mapping->host;
1497         loff_t i_size = i_size_read(inode);
1498
1499         if (to > i_size) {
1500                 truncate_pagecache(inode, i_size);
1501                 truncate_blocks(inode, i_size, true);
1502         }
1503 }
1504
1505 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1506                         struct page *page, loff_t pos, unsigned len,
1507                         block_t *blk_addr, bool *node_changed)
1508 {
1509         struct inode *inode = page->mapping->host;
1510         pgoff_t index = page->index;
1511         struct dnode_of_data dn;
1512         struct page *ipage;
1513         bool locked = false;
1514         struct extent_info ei;
1515         int err = 0;
1516
1517         /*
1518          * we already allocated all the blocks, so we don't need to get
1519          * the block addresses when there is no need to fill the page.
1520          */
1521         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1522                                         len == PAGE_SIZE)
1523                 return 0;
1524
1525         if (f2fs_has_inline_data(inode) ||
1526                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1527                 f2fs_lock_op(sbi);
1528                 locked = true;
1529         }
1530 restart:
1531         /* check inline_data */
1532         ipage = get_node_page(sbi, inode->i_ino);
1533         if (IS_ERR(ipage)) {
1534                 err = PTR_ERR(ipage);
1535                 goto unlock_out;
1536         }
1537
1538         set_new_dnode(&dn, inode, ipage, ipage, 0);
1539
1540         if (f2fs_has_inline_data(inode)) {
1541                 if (pos + len <= MAX_INLINE_DATA) {
1542                         read_inline_data(page, ipage);
1543                         set_inode_flag(inode, FI_DATA_EXIST);
1544                         if (inode->i_nlink)
1545                                 set_inline_node(ipage);
1546                 } else {
1547                         err = f2fs_convert_inline_page(&dn, page);
1548                         if (err)
1549                                 goto out;
1550                         if (dn.data_blkaddr == NULL_ADDR)
1551                                 err = f2fs_get_block(&dn, index);
1552                 }
1553         } else if (locked) {
1554                 err = f2fs_get_block(&dn, index);
1555         } else {
1556                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1557                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1558                 } else {
1559                         /* hole case */
1560                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1561                         if (err || dn.data_blkaddr == NULL_ADDR) {
1562                                 f2fs_put_dnode(&dn);
1563                                 f2fs_lock_op(sbi);
1564                                 locked = true;
1565                                 goto restart;
1566                         }
1567                 }
1568         }
1569
1570         /* convert_inline_page can make node_changed */
1571         *blk_addr = dn.data_blkaddr;
1572         *node_changed = dn.node_changed;
1573 out:
1574         f2fs_put_dnode(&dn);
1575 unlock_out:
1576         if (locked)
1577                 f2fs_unlock_op(sbi);
1578         return err;
1579 }
1580
1581 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1582                 loff_t pos, unsigned len, unsigned flags,
1583                 struct page **pagep, void **fsdata)
1584 {
1585         struct inode *inode = mapping->host;
1586         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1587         struct page *page = NULL;
1588         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1589         bool need_balance = false;
1590         block_t blkaddr = NULL_ADDR;
1591         int err = 0;
1592
1593         trace_f2fs_write_begin(inode, pos, len, flags);
1594
1595         /*
1596          * We should check this at this moment to avoid deadlock on inode page
1597          * and #0 page. The locking rule for inline_data conversion should be:
1598          * lock_page(page #0) -> lock_page(inode_page)
1599          */
1600         if (index != 0) {
1601                 err = f2fs_convert_inline_inode(inode);
1602                 if (err)
1603                         goto fail;
1604         }
1605 repeat:
1606         page = grab_cache_page_write_begin(mapping, index, flags);
1607         if (!page) {
1608                 err = -ENOMEM;
1609                 goto fail;
1610         }
1611
1612         *pagep = page;
1613
1614         err = prepare_write_begin(sbi, page, pos, len,
1615                                         &blkaddr, &need_balance);
1616         if (err)
1617                 goto fail;
1618
1619         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1620                 unlock_page(page);
1621                 f2fs_balance_fs(sbi, true);
1622                 lock_page(page);
1623                 if (page->mapping != mapping) {
1624                         /* The page got truncated from under us */
1625                         f2fs_put_page(page, 1);
1626                         goto repeat;
1627                 }
1628         }
1629
1630         f2fs_wait_on_page_writeback(page, DATA, false);
1631
1632         /* wait for GCed encrypted page writeback */
1633         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1634                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1635
1636         if (len == PAGE_SIZE)
1637                 goto out_update;
1638         if (PageUptodate(page))
1639                 goto out_clear;
1640
1641         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1642                 unsigned start = pos & (PAGE_SIZE - 1);
1643                 unsigned end = start + len;
1644
1645                 /* Reading beyond i_size is simple: memset to zero */
1646                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1647                 goto out_update;
1648         }
1649
1650         if (blkaddr == NEW_ADDR) {
1651                 zero_user_segment(page, 0, PAGE_SIZE);
1652         } else {
1653                 struct bio *bio;
1654
1655                 bio = f2fs_grab_bio(inode, blkaddr, 1);
1656                 if (IS_ERR(bio)) {
1657                         err = PTR_ERR(bio);
1658                         goto fail;
1659                 }
1660                 bio_set_op_attrs(bio, REQ_OP_READ, READ_SYNC);
1661                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1662                         bio_put(bio);
1663                         err = -EFAULT;
1664                         goto fail;
1665                 }
1666
1667                 __submit_bio(sbi, bio, DATA);
1668
1669                 lock_page(page);
1670                 if (unlikely(page->mapping != mapping)) {
1671                         f2fs_put_page(page, 1);
1672                         goto repeat;
1673                 }
1674                 if (unlikely(!PageUptodate(page))) {
1675                         err = -EIO;
1676                         goto fail;
1677                 }
1678         }
1679 out_update:
1680         if (!PageUptodate(page))
1681                 SetPageUptodate(page);
1682 out_clear:
1683         clear_cold_data(page);
1684         return 0;
1685
1686 fail:
1687         f2fs_put_page(page, 1);
1688         f2fs_write_failed(mapping, pos + len);
1689         return err;
1690 }
1691
1692 static int f2fs_write_end(struct file *file,
1693                         struct address_space *mapping,
1694                         loff_t pos, unsigned len, unsigned copied,
1695                         struct page *page, void *fsdata)
1696 {
1697         struct inode *inode = page->mapping->host;
1698
1699         trace_f2fs_write_end(inode, pos, len, copied);
1700
1701         set_page_dirty(page);
1702
1703         if (pos + copied > i_size_read(inode))
1704                 f2fs_i_size_write(inode, pos + copied);
1705
1706         f2fs_put_page(page, 1);
1707         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1708         return copied;
1709 }
1710
1711 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1712                            loff_t offset)
1713 {
1714         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1715
1716         if (offset & blocksize_mask)
1717                 return -EINVAL;
1718
1719         if (iov_iter_alignment(iter) & blocksize_mask)
1720                 return -EINVAL;
1721
1722         return 0;
1723 }
1724
1725 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1726 {
1727         struct address_space *mapping = iocb->ki_filp->f_mapping;
1728         struct inode *inode = mapping->host;
1729         size_t count = iov_iter_count(iter);
1730         loff_t offset = iocb->ki_pos;
1731         int rw = iov_iter_rw(iter);
1732         int err;
1733
1734         err = check_direct_IO(inode, iter, offset);
1735         if (err)
1736                 return err;
1737
1738         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1739                 return 0;
1740         if (test_opt(F2FS_I_SB(inode), LFS))
1741                 return 0;
1742
1743         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1744
1745         down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1746         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1747         up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1748
1749         if (rw == WRITE) {
1750                 if (err > 0)
1751                         set_inode_flag(inode, FI_UPDATE_WRITE);
1752                 else if (err < 0)
1753                         f2fs_write_failed(mapping, offset + count);
1754         }
1755
1756         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1757
1758         return err;
1759 }
1760
1761 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1762                                                         unsigned int length)
1763 {
1764         struct inode *inode = page->mapping->host;
1765         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1766
1767         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1768                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1769                 return;
1770
1771         if (PageDirty(page)) {
1772                 if (inode->i_ino == F2FS_META_INO(sbi))
1773                         dec_page_count(sbi, F2FS_DIRTY_META);
1774                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1775                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1776                 else
1777                         inode_dec_dirty_pages(inode);
1778         }
1779
1780         /* This is atomic written page, keep Private */
1781         if (IS_ATOMIC_WRITTEN_PAGE(page))
1782                 return;
1783
1784         set_page_private(page, 0);
1785         ClearPagePrivate(page);
1786 }
1787
1788 int f2fs_release_page(struct page *page, gfp_t wait)
1789 {
1790         /* If this is dirty page, keep PagePrivate */
1791         if (PageDirty(page))
1792                 return 0;
1793
1794         /* This is atomic written page, keep Private */
1795         if (IS_ATOMIC_WRITTEN_PAGE(page))
1796                 return 0;
1797
1798         set_page_private(page, 0);
1799         ClearPagePrivate(page);
1800         return 1;
1801 }
1802
1803 /*
1804  * This was copied from __set_page_dirty_buffers which gives higher performance
1805  * in very high speed storages. (e.g., pmem)
1806  */
1807 void f2fs_set_page_dirty_nobuffers(struct page *page)
1808 {
1809         struct address_space *mapping = page->mapping;
1810         unsigned long flags;
1811
1812         if (unlikely(!mapping))
1813                 return;
1814
1815         spin_lock(&mapping->private_lock);
1816         lock_page_memcg(page);
1817         SetPageDirty(page);
1818         spin_unlock(&mapping->private_lock);
1819
1820         spin_lock_irqsave(&mapping->tree_lock, flags);
1821         WARN_ON_ONCE(!PageUptodate(page));
1822         account_page_dirtied(page, mapping);
1823         radix_tree_tag_set(&mapping->page_tree,
1824                         page_index(page), PAGECACHE_TAG_DIRTY);
1825         spin_unlock_irqrestore(&mapping->tree_lock, flags);
1826         unlock_page_memcg(page);
1827
1828         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1829         return;
1830 }
1831
1832 static int f2fs_set_data_page_dirty(struct page *page)
1833 {
1834         struct address_space *mapping = page->mapping;
1835         struct inode *inode = mapping->host;
1836
1837         trace_f2fs_set_page_dirty(page, DATA);
1838
1839         if (!PageUptodate(page))
1840                 SetPageUptodate(page);
1841
1842         if (f2fs_is_atomic_file(inode)) {
1843                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1844                         register_inmem_page(inode, page);
1845                         return 1;
1846                 }
1847                 /*
1848                  * Previously, this page has been registered, we just
1849                  * return here.
1850                  */
1851                 return 0;
1852         }
1853
1854         if (!PageDirty(page)) {
1855                 f2fs_set_page_dirty_nobuffers(page);
1856                 update_dirty_page(inode, page);
1857                 return 1;
1858         }
1859         return 0;
1860 }
1861
1862 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1863 {
1864         struct inode *inode = mapping->host;
1865
1866         if (f2fs_has_inline_data(inode))
1867                 return 0;
1868
1869         /* make sure allocating whole blocks */
1870         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1871                 filemap_write_and_wait(mapping);
1872
1873         return generic_block_bmap(mapping, block, get_data_block_bmap);
1874 }
1875
1876 const struct address_space_operations f2fs_dblock_aops = {
1877         .readpage       = f2fs_read_data_page,
1878         .readpages      = f2fs_read_data_pages,
1879         .writepage      = f2fs_write_data_page,
1880         .writepages     = f2fs_write_data_pages,
1881         .write_begin    = f2fs_write_begin,
1882         .write_end      = f2fs_write_end,
1883         .set_page_dirty = f2fs_set_data_page_dirty,
1884         .invalidatepage = f2fs_invalidate_page,
1885         .releasepage    = f2fs_release_page,
1886         .direct_IO      = f2fs_direct_IO,
1887         .bmap           = f2fs_bmap,
1888 };