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