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 #ifdef CONFIG_F2FS_FAULT_INJECTION
  38         if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO))
  39                 bio->bi_error = -EIO;
  40 #endif
  41
  42         if (f2fs_bio_encrypted(bio)) {
  43                 if (bio->bi_error) {
  44                         fscrypt_release_ctx(bio->bi_private);
  45                 } else {
  46                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  47                         return;
  48                 }
  49         }
  50
  51         bio_for_each_segment_all(bvec, bio, i) {
  52                 struct page *page = bvec->bv_page;
  53
  54                 if (!bio->bi_error) {
  55                         if (!PageUptodate(page))
  56                                 SetPageUptodate(page);
  57                 } else {
  58                         ClearPageUptodate(page);
  59                         SetPageError(page);
  60                 }
  61                 unlock_page(page);
  62         }
  63         bio_put(bio);
  64 }
  65
  66 static void f2fs_write_end_io(struct bio *bio)
  67 {
  68         struct f2fs_sb_info *sbi = bio->bi_private;
  69         struct bio_vec *bvec;
  70         int i;
  71
  72         bio_for_each_segment_all(bvec, bio, i) {
  73                 struct page *page = bvec->bv_page;
  74
  75                 fscrypt_pullback_bio_page(&page, true);
  76
  77                 if (unlikely(bio->bi_error)) {
  78                         mapping_set_error(page->mapping, -EIO);
  79                         f2fs_stop_checkpoint(sbi, true);
  80                 }
  81                 end_page_writeback(page);
  82         }
  83         if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
  84                                 wq_has_sleeper(&sbi->cp_wait))
  85                 wake_up(&sbi->cp_wait);
  86
  87         bio_put(bio);
  88 }
  89
  90 /*
  91  * Low-level block read/write IO operations.
  92  */
  93 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  94                                 int npages, bool is_read)
  95 {
  96         struct bio *bio;
  97
  98         bio = f2fs_bio_alloc(npages);
  99
 100         bio->bi_bdev = sbi->sb->s_bdev;
 101         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 102         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
 103         bio->bi_private = is_read ? NULL : sbi;
 104
 105         return bio;
 106 }
 107
 108 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 109                                 struct bio *bio, enum page_type type)
 110 {
 111         if (!is_read_io(bio_op(bio))) {
 112                 atomic_inc(&sbi->nr_wb_bios);
 113                 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
 114                         current->plug && (type == DATA || type == NODE))
 115                         blk_finish_plug(current->plug);
 116         }
 117         submit_bio(bio);
 118 }
 119
 120 static void __submit_merged_bio(struct f2fs_bio_info *io)
 121 {
 122         struct f2fs_io_info *fio = &io->fio;
 123
 124         if (!io->bio)
 125                 return;
 126
 127         if (is_read_io(fio->op))
 128                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 129         else
 130                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 131
 132         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 133
 134         __submit_bio(io->sbi, io->bio, fio->type);
 135         io->bio = NULL;
 136 }
 137
 138 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 139                                                 struct page *page, nid_t ino)
 140 {
 141         struct bio_vec *bvec;
 142         struct page *target;
 143         int i;
 144
 145         if (!io->bio)
 146                 return false;
 147
 148         if (!inode && !page && !ino)
 149                 return true;
 150
 151         bio_for_each_segment_all(bvec, io->bio, i) {
 152
 153                 if (bvec->bv_page->mapping)
 154                         target = bvec->bv_page;
 155                 else
 156                         target = fscrypt_control_page(bvec->bv_page);
 157
 158                 if (inode && inode == target->mapping->host)
 159                         return true;
 160                 if (page && page == target)
 161                         return true;
 162                 if (ino && ino == ino_of_node(target))
 163                         return true;
 164         }
 165
 166         return false;
 167 }
 168
 169 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 170                                                 struct page *page, nid_t ino,
 171                                                 enum page_type type)
 172 {
 173         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 174         struct f2fs_bio_info *io = &sbi->write_io[btype];
 175         bool ret;
 176
 177         down_read(&io->io_rwsem);
 178         ret = __has_merged_page(io, inode, page, ino);
 179         up_read(&io->io_rwsem);
 180         return ret;
 181 }
 182
 183 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 184                                 struct inode *inode, struct page *page,
 185                                 nid_t ino, enum page_type type, int rw)
 186 {
 187         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 188         struct f2fs_bio_info *io;
 189
 190         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 191
 192         down_write(&io->io_rwsem);
 193
 194         if (!__has_merged_page(io, inode, page, ino))
 195                 goto out;
 196
 197         /* change META to META_FLUSH in the checkpoint procedure */
 198         if (type >= META_FLUSH) {
 199                 io->fio.type = META_FLUSH;
 200                 io->fio.op = REQ_OP_WRITE;
 201                 if (test_opt(sbi, NOBARRIER))
 202                         io->fio.op_flags = WRITE_FLUSH | REQ_META | REQ_PRIO;
 203                 else
 204                         io->fio.op_flags = WRITE_FLUSH_FUA | REQ_META |
 205                                                                 REQ_PRIO;
 206         }
 207         __submit_merged_bio(io);
 208 out:
 209         up_write(&io->io_rwsem);
 210 }
 211
 212 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 213                                                                         int rw)
 214 {
 215         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 216 }
 217
 218 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 219                                 struct inode *inode, struct page *page,
 220                                 nid_t ino, enum page_type type, int rw)
 221 {
 222         if (has_merged_page(sbi, inode, page, ino, type))
 223                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 224 }
 225
 226 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 227 {
 228         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 229         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 230         f2fs_submit_merged_bio(sbi, META, WRITE);
 231 }
 232
 233 /*
 234  * Fill the locked page with data located in the block address.
 235  * Return unlocked page.
 236  */
 237 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 238 {
 239         struct bio *bio;
 240         struct page *page = fio->encrypted_page ?
 241                         fio->encrypted_page : fio->page;
 242
 243         trace_f2fs_submit_page_bio(page, fio);
 244         f2fs_trace_ios(fio, 0);
 245
 246         /* Allocate a new bio */
 247         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
 248
 249         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 250                 bio_put(bio);
 251                 return -EFAULT;
 252         }
 253         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 254
 255         __submit_bio(fio->sbi, bio, fio->type);
 256         return 0;
 257 }
 258
 259 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 260 {
 261         struct f2fs_sb_info *sbi = fio->sbi;
 262         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 263         struct f2fs_bio_info *io;
 264         bool is_read = is_read_io(fio->op);
 265         struct page *bio_page;
 266
 267         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 268
 269         if (fio->old_blkaddr != NEW_ADDR)
 270                 verify_block_addr(sbi, fio->old_blkaddr);
 271         verify_block_addr(sbi, fio->new_blkaddr);
 272
 273         down_write(&io->io_rwsem);
 274
 275         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 276             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags)))
 277                 __submit_merged_bio(io);
 278 alloc_new:
 279         if (io->bio == NULL) {
 280                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 281                                                 BIO_MAX_PAGES, is_read);
 282                 io->fio = *fio;
 283         }
 284
 285         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 286
 287         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 288                                                         PAGE_SIZE) {
 289                 __submit_merged_bio(io);
 290                 goto alloc_new;
 291         }
 292
 293         io->last_block_in_bio = fio->new_blkaddr;
 294         f2fs_trace_ios(fio, 0);
 295
 296         up_write(&io->io_rwsem);
 297         trace_f2fs_submit_page_mbio(fio->page, fio);
 298 }
 299
 300 static void __set_data_blkaddr(struct dnode_of_data *dn)
 301 {
 302         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 303         __le32 *addr_array;
 304
 305         /* Get physical address of data block */
 306         addr_array = blkaddr_in_node(rn);
 307         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 308 }
 309
 310 /*
 311  * Lock ordering for the change of data block address:
 312  * ->data_page
 313  *  ->node_page
 314  *    update block addresses in the node page
 315  */
 316 void set_data_blkaddr(struct dnode_of_data *dn)
 317 {
 318         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 319         __set_data_blkaddr(dn);
 320         if (set_page_dirty(dn->node_page))
 321                 dn->node_changed = true;
 322 }
 323
 324 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 325 {
 326         dn->data_blkaddr = blkaddr;
 327         set_data_blkaddr(dn);
 328         f2fs_update_extent_cache(dn);
 329 }
 330
 331 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 332 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 333 {
 334         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 335
 336         if (!count)
 337                 return 0;
 338
 339         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 340                 return -EPERM;
 341         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 342                 return -ENOSPC;
 343
 344         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 345                                                 dn->ofs_in_node, count);
 346
 347         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 348
 349         for (; count > 0; dn->ofs_in_node++) {
 350                 block_t blkaddr =
 351                         datablock_addr(dn->node_page, dn->ofs_in_node);
 352                 if (blkaddr == NULL_ADDR) {
 353                         dn->data_blkaddr = NEW_ADDR;
 354                         __set_data_blkaddr(dn);
 355                         count--;
 356                 }
 357         }
 358
 359         if (set_page_dirty(dn->node_page))
 360                 dn->node_changed = true;
 361         return 0;
 362 }
 363
 364 /* Should keep dn->ofs_in_node unchanged */
 365 int reserve_new_block(struct dnode_of_data *dn)
 366 {
 367         unsigned int ofs_in_node = dn->ofs_in_node;
 368         int ret;
 369
 370         ret = reserve_new_blocks(dn, 1);
 371         dn->ofs_in_node = ofs_in_node;
 372         return ret;
 373 }
 374
 375 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 376 {
 377         bool need_put = dn->inode_page ? false : true;
 378         int err;
 379
 380         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 381         if (err)
 382                 return err;
 383
 384         if (dn->data_blkaddr == NULL_ADDR)
 385                 err = reserve_new_block(dn);
 386         if (err || need_put)
 387                 f2fs_put_dnode(dn);
 388         return err;
 389 }
 390
 391 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 392 {
 393         struct extent_info ei;
 394         struct inode *inode = dn->inode;
 395
 396         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 397                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 398                 return 0;
 399         }
 400
 401         return f2fs_reserve_block(dn, index);
 402 }
 403
 404 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 405                                                 int op_flags, bool for_write)
 406 {
 407         struct address_space *mapping = inode->i_mapping;
 408         struct dnode_of_data dn;
 409         struct page *page;
 410         struct extent_info ei;
 411         int err;
 412         struct f2fs_io_info fio = {
 413                 .sbi = F2FS_I_SB(inode),
 414                 .type = DATA,
 415                 .op = REQ_OP_READ,
 416                 .op_flags = op_flags,
 417                 .encrypted_page = NULL,
 418         };
 419
 420         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 421                 return read_mapping_page(mapping, index, NULL);
 422
 423         page = f2fs_grab_cache_page(mapping, index, for_write);
 424         if (!page)
 425                 return ERR_PTR(-ENOMEM);
 426
 427         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 428                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 429                 goto got_it;
 430         }
 431
 432         set_new_dnode(&dn, inode, NULL, NULL, 0);
 433         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 434         if (err)
 435                 goto put_err;
 436         f2fs_put_dnode(&dn);
 437
 438         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 439                 err = -ENOENT;
 440                 goto put_err;
 441         }
 442 got_it:
 443         if (PageUptodate(page)) {
 444                 unlock_page(page);
 445                 return page;
 446         }
 447
 448         /*
 449          * A new dentry page is allocated but not able to be written, since its
 450          * new inode page couldn't be allocated due to -ENOSPC.
 451          * In such the case, its blkaddr can be remained as NEW_ADDR.
 452          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 453          */
 454         if (dn.data_blkaddr == NEW_ADDR) {
 455                 zero_user_segment(page, 0, PAGE_SIZE);
 456                 if (!PageUptodate(page))
 457                         SetPageUptodate(page);
 458                 unlock_page(page);
 459                 return page;
 460         }
 461
 462         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 463         fio.page = page;
 464         err = f2fs_submit_page_bio(&fio);
 465         if (err)
 466                 goto put_err;
 467         return page;
 468
 469 put_err:
 470         f2fs_put_page(page, 1);
 471         return ERR_PTR(err);
 472 }
 473
 474 struct page *find_data_page(struct inode *inode, pgoff_t index)
 475 {
 476         struct address_space *mapping = inode->i_mapping;
 477         struct page *page;
 478
 479         page = find_get_page(mapping, index);
 480         if (page && PageUptodate(page))
 481                 return page;
 482         f2fs_put_page(page, 0);
 483
 484         page = get_read_data_page(inode, index, READ_SYNC, false);
 485         if (IS_ERR(page))
 486                 return page;
 487
 488         if (PageUptodate(page))
 489                 return page;
 490
 491         wait_on_page_locked(page);
 492         if (unlikely(!PageUptodate(page))) {
 493                 f2fs_put_page(page, 0);
 494                 return ERR_PTR(-EIO);
 495         }
 496         return page;
 497 }
 498
 499 /*
 500  * If it tries to access a hole, return an error.
 501  * Because, the callers, functions in dir.c and GC, should be able to know
 502  * whether this page exists or not.
 503  */
 504 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 505                                                         bool for_write)
 506 {
 507         struct address_space *mapping = inode->i_mapping;
 508         struct page *page;
 509 repeat:
 510         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 511         if (IS_ERR(page))
 512                 return page;
 513
 514         /* wait for read completion */
 515         lock_page(page);
 516         if (unlikely(page->mapping != mapping)) {
 517                 f2fs_put_page(page, 1);
 518                 goto repeat;
 519         }
 520         if (unlikely(!PageUptodate(page))) {
 521                 f2fs_put_page(page, 1);
 522                 return ERR_PTR(-EIO);
 523         }
 524         return page;
 525 }
 526
 527 /*
 528  * Caller ensures that this data page is never allocated.
 529  * A new zero-filled data page is allocated in the page cache.
 530  *
 531  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 532  * f2fs_unlock_op().
 533  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 534  * ipage should be released by this function.
 535  */
 536 struct page *get_new_data_page(struct inode *inode,
 537                 struct page *ipage, pgoff_t index, bool new_i_size)
 538 {
 539         struct address_space *mapping = inode->i_mapping;
 540         struct page *page;
 541         struct dnode_of_data dn;
 542         int err;
 543
 544         page = f2fs_grab_cache_page(mapping, index, true);
 545         if (!page) {
 546                 /*
 547                  * before exiting, we should make sure ipage will be released
 548                  * if any error occur.
 549                  */
 550                 f2fs_put_page(ipage, 1);
 551                 return ERR_PTR(-ENOMEM);
 552         }
 553
 554         set_new_dnode(&dn, inode, ipage, NULL, 0);
 555         err = f2fs_reserve_block(&dn, index);
 556         if (err) {
 557                 f2fs_put_page(page, 1);
 558                 return ERR_PTR(err);
 559         }
 560         if (!ipage)
 561                 f2fs_put_dnode(&dn);
 562
 563         if (PageUptodate(page))
 564                 goto got_it;
 565
 566         if (dn.data_blkaddr == NEW_ADDR) {
 567                 zero_user_segment(page, 0, PAGE_SIZE);
 568                 if (!PageUptodate(page))
 569                         SetPageUptodate(page);
 570         } else {
 571                 f2fs_put_page(page, 1);
 572
 573                 /* if ipage exists, blkaddr should be NEW_ADDR */
 574                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 575                 page = get_lock_data_page(inode, index, true);
 576                 if (IS_ERR(page))
 577                         return page;
 578         }
 579 got_it:
 580         if (new_i_size && i_size_read(inode) <
 581                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 582                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 583         return page;
 584 }
 585
 586 static int __allocate_data_block(struct dnode_of_data *dn)
 587 {
 588         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 589         struct f2fs_summary sum;
 590         struct node_info ni;
 591         int seg = CURSEG_WARM_DATA;
 592         pgoff_t fofs;
 593         blkcnt_t count = 1;
 594
 595         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 596                 return -EPERM;
 597
 598         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 599         if (dn->data_blkaddr == NEW_ADDR)
 600                 goto alloc;
 601
 602         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 603                 return -ENOSPC;
 604
 605 alloc:
 606         get_node_info(sbi, dn->nid, &ni);
 607         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 608
 609         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 610                 seg = CURSEG_DIRECT_IO;
 611
 612         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 613                                                                 &sum, seg);
 614         set_data_blkaddr(dn);
 615
 616         /* update i_size */
 617         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 618                                                         dn->ofs_in_node;
 619         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 620                 f2fs_i_size_write(dn->inode,
 621                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 622         return 0;
 623 }
 624
 625 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 626 {
 627         struct inode *inode = file_inode(iocb->ki_filp);
 628         struct f2fs_map_blocks map;
 629         ssize_t ret = 0;
 630
 631         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 632         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 633         if (map.m_len > map.m_lblk)
 634                 map.m_len -= map.m_lblk;
 635         else
 636                 map.m_len = 0;
 637
 638         map.m_next_pgofs = NULL;
 639
 640         if (iocb->ki_flags & IOCB_DIRECT) {
 641                 ret = f2fs_convert_inline_inode(inode);
 642                 if (ret)
 643                         return ret;
 644                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 645         }
 646         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 647                 ret = f2fs_convert_inline_inode(inode);
 648                 if (ret)
 649                         return ret;
 650         }
 651         if (!f2fs_has_inline_data(inode))
 652                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 653         return ret;
 654 }
 655
 656 /*
 657  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 658  * f2fs_map_blocks structure.
 659  * If original data blocks are allocated, then give them to blockdev.
 660  * Otherwise,
 661  *     a. preallocate requested block addresses
 662  *     b. do not use extent cache for better performance
 663  *     c. give the block addresses to blockdev
 664  */
 665 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 666                                                 int create, int flag)
 667 {
 668         unsigned int maxblocks = map->m_len;
 669         struct dnode_of_data dn;
 670         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 671         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 672         pgoff_t pgofs, end_offset, end;
 673         int err = 0, ofs = 1;
 674         unsigned int ofs_in_node, last_ofs_in_node;
 675         blkcnt_t prealloc;
 676         struct extent_info ei;
 677         block_t blkaddr;
 678
 679         if (!maxblocks)
 680                 return 0;
 681
 682         map->m_len = 0;
 683         map->m_flags = 0;
 684
 685         /* it only supports block size == page size */
 686         pgofs = (pgoff_t)map->m_lblk;
 687         end = pgofs + maxblocks;
 688
 689         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 690                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 691                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 692                 map->m_flags = F2FS_MAP_MAPPED;
 693                 goto out;
 694         }
 695
 696 next_dnode:
 697         if (create)
 698                 f2fs_lock_op(sbi);
 699
 700         /* When reading holes, we need its node page */
 701         set_new_dnode(&dn, inode, NULL, NULL, 0);
 702         err = get_dnode_of_data(&dn, pgofs, mode);
 703         if (err) {
 704                 if (flag == F2FS_GET_BLOCK_BMAP)
 705                         map->m_pblk = 0;
 706                 if (err == -ENOENT) {
 707                         err = 0;
 708                         if (map->m_next_pgofs)
 709                                 *map->m_next_pgofs =
 710                                         get_next_page_offset(&dn, pgofs);
 711                 }
 712                 goto unlock_out;
 713         }
 714
 715         prealloc = 0;
 716         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
 717         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 718
 719 next_block:
 720         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 721
 722         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 723                 if (create) {
 724                         if (unlikely(f2fs_cp_error(sbi))) {
 725                                 err = -EIO;
 726                                 goto sync_out;
 727                         }
 728                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 729                                 if (blkaddr == NULL_ADDR) {
 730                                         prealloc++;
 731                                         last_ofs_in_node = dn.ofs_in_node;
 732                                 }
 733                         } else {
 734                                 err = __allocate_data_block(&dn);
 735                                 if (!err)
 736                                         set_inode_flag(inode, FI_APPEND_WRITE);
 737                         }
 738                         if (err)
 739                                 goto sync_out;
 740                         map->m_flags = F2FS_MAP_NEW;
 741                         blkaddr = dn.data_blkaddr;
 742                 } else {
 743                         if (flag == F2FS_GET_BLOCK_BMAP) {
 744                                 map->m_pblk = 0;
 745                                 goto sync_out;
 746                         }
 747                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 748                                                 blkaddr == NULL_ADDR) {
 749                                 if (map->m_next_pgofs)
 750                                         *map->m_next_pgofs = pgofs + 1;
 751                         }
 752                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 753                                                 blkaddr != NEW_ADDR)
 754                                 goto sync_out;
 755                 }
 756         }
 757
 758         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 759                 goto skip;
 760
 761         if (map->m_len == 0) {
 762                 /* preallocated unwritten block should be mapped for fiemap. */
 763                 if (blkaddr == NEW_ADDR)
 764                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 765                 map->m_flags |= F2FS_MAP_MAPPED;
 766
 767                 map->m_pblk = blkaddr;
 768                 map->m_len = 1;
 769         } else if ((map->m_pblk != NEW_ADDR &&
 770                         blkaddr == (map->m_pblk + ofs)) ||
 771                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 772                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 773                 ofs++;
 774                 map->m_len++;
 775         } else {
 776                 goto sync_out;
 777         }
 778
 779 skip:
 780         dn.ofs_in_node++;
 781         pgofs++;
 782
 783         /* preallocate blocks in batch for one dnode page */
 784         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 785                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 786
 787                 dn.ofs_in_node = ofs_in_node;
 788                 err = reserve_new_blocks(&dn, prealloc);
 789                 if (err)
 790                         goto sync_out;
 791
 792                 map->m_len += dn.ofs_in_node - ofs_in_node;
 793                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 794                         err = -ENOSPC;
 795                         goto sync_out;
 796                 }
 797                 dn.ofs_in_node = end_offset;
 798         }
 799
 800         if (pgofs >= end)
 801                 goto sync_out;
 802         else if (dn.ofs_in_node < end_offset)
 803                 goto next_block;
 804
 805         f2fs_put_dnode(&dn);
 806
 807         if (create) {
 808                 f2fs_unlock_op(sbi);
 809                 f2fs_balance_fs(sbi, dn.node_changed);
 810         }
 811         goto next_dnode;
 812
 813 sync_out:
 814         f2fs_put_dnode(&dn);
 815 unlock_out:
 816         if (create) {
 817                 f2fs_unlock_op(sbi);
 818                 f2fs_balance_fs(sbi, dn.node_changed);
 819         }
 820 out:
 821         trace_f2fs_map_blocks(inode, map, err);
 822         return err;
 823 }
 824
 825 static int __get_data_block(struct inode *inode, sector_t iblock,
 826                         struct buffer_head *bh, int create, int flag,
 827                         pgoff_t *next_pgofs)
 828 {
 829         struct f2fs_map_blocks map;
 830         int ret;
 831
 832         map.m_lblk = iblock;
 833         map.m_len = bh->b_size >> inode->i_blkbits;
 834         map.m_next_pgofs = next_pgofs;
 835
 836         ret = f2fs_map_blocks(inode, &map, create, flag);
 837         if (!ret) {
 838                 map_bh(bh, inode->i_sb, map.m_pblk);
 839                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 840                 bh->b_size = map.m_len << inode->i_blkbits;
 841         }
 842         return ret;
 843 }
 844
 845 static int get_data_block(struct inode *inode, sector_t iblock,
 846                         struct buffer_head *bh_result, int create, int flag,
 847                         pgoff_t *next_pgofs)
 848 {
 849         return __get_data_block(inode, iblock, bh_result, create,
 850                                                         flag, next_pgofs);
 851 }
 852
 853 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 854                         struct buffer_head *bh_result, int create)
 855 {
 856         return __get_data_block(inode, iblock, bh_result, create,
 857                                                 F2FS_GET_BLOCK_DIO, NULL);
 858 }
 859
 860 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 861                         struct buffer_head *bh_result, int create)
 862 {
 863         /* Block number less than F2FS MAX BLOCKS */
 864         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 865                 return -EFBIG;
 866
 867         return __get_data_block(inode, iblock, bh_result, create,
 868                                                 F2FS_GET_BLOCK_BMAP, NULL);
 869 }
 870
 871 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 872 {
 873         return (offset >> inode->i_blkbits);
 874 }
 875
 876 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 877 {
 878         return (blk << inode->i_blkbits);
 879 }
 880
 881 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 882                 u64 start, u64 len)
 883 {
 884         struct buffer_head map_bh;
 885         sector_t start_blk, last_blk;
 886         pgoff_t next_pgofs;
 887         u64 logical = 0, phys = 0, size = 0;
 888         u32 flags = 0;
 889         int ret = 0;
 890
 891         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 892         if (ret)
 893                 return ret;
 894
 895         if (f2fs_has_inline_data(inode)) {
 896                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 897                 if (ret != -EAGAIN)
 898                         return ret;
 899         }
 900
 901         inode_lock(inode);
 902
 903         if (logical_to_blk(inode, len) == 0)
 904                 len = blk_to_logical(inode, 1);
 905
 906         start_blk = logical_to_blk(inode, start);
 907         last_blk = logical_to_blk(inode, start + len - 1);
 908
 909 next:
 910         memset(&map_bh, 0, sizeof(struct buffer_head));
 911         map_bh.b_size = len;
 912
 913         ret = get_data_block(inode, start_blk, &map_bh, 0,
 914                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 915         if (ret)
 916                 goto out;
 917
 918         /* HOLE */
 919         if (!buffer_mapped(&map_bh)) {
 920                 start_blk = next_pgofs;
 921
 922                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
 923                                         F2FS_I_SB(inode)->max_file_blocks))
 924                         goto prep_next;
 925
 926                 flags |= FIEMAP_EXTENT_LAST;
 927         }
 928
 929         if (size) {
 930                 if (f2fs_encrypted_inode(inode))
 931                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 932
 933                 ret = fiemap_fill_next_extent(fieinfo, logical,
 934                                 phys, size, flags);
 935         }
 936
 937         if (start_blk > last_blk || ret)
 938                 goto out;
 939
 940         logical = blk_to_logical(inode, start_blk);
 941         phys = blk_to_logical(inode, map_bh.b_blocknr);
 942         size = map_bh.b_size;
 943         flags = 0;
 944         if (buffer_unwritten(&map_bh))
 945                 flags = FIEMAP_EXTENT_UNWRITTEN;
 946
 947         start_blk += logical_to_blk(inode, size);
 948
 949 prep_next:
 950         cond_resched();
 951         if (fatal_signal_pending(current))
 952                 ret = -EINTR;
 953         else
 954                 goto next;
 955 out:
 956         if (ret == 1)
 957                 ret = 0;
 958
 959         inode_unlock(inode);
 960         return ret;
 961 }
 962
 963 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
 964                                  unsigned nr_pages)
 965 {
 966         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 967         struct fscrypt_ctx *ctx = NULL;
 968         struct block_device *bdev = sbi->sb->s_bdev;
 969         struct bio *bio;
 970
 971         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
 972                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
 973                 if (IS_ERR(ctx))
 974                         return ERR_CAST(ctx);
 975
 976                 /* wait the page to be moved by cleaning */
 977                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
 978         }
 979
 980         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
 981         if (!bio) {
 982                 if (ctx)
 983                         fscrypt_release_ctx(ctx);
 984                 return ERR_PTR(-ENOMEM);
 985         }
 986         bio->bi_bdev = bdev;
 987         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
 988         bio->bi_end_io = f2fs_read_end_io;
 989         bio->bi_private = ctx;
 990
 991         return bio;
 992 }
 993
 994 /*
 995  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 996  * Major change was from block_size == page_size in f2fs by default.
 997  */
 998 static int f2fs_mpage_readpages(struct address_space *mapping,
 999                         struct list_head *pages, struct page *page,
1000                         unsigned nr_pages)
1001 {
1002         struct bio *bio = NULL;
1003         unsigned page_idx;
1004         sector_t last_block_in_bio = 0;
1005         struct inode *inode = mapping->host;
1006         const unsigned blkbits = inode->i_blkbits;
1007         const unsigned blocksize = 1 << blkbits;
1008         sector_t block_in_file;
1009         sector_t last_block;
1010         sector_t last_block_in_file;
1011         sector_t block_nr;
1012         struct f2fs_map_blocks map;
1013
1014         map.m_pblk = 0;
1015         map.m_lblk = 0;
1016         map.m_len = 0;
1017         map.m_flags = 0;
1018         map.m_next_pgofs = NULL;
1019
1020         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1021
1022                 prefetchw(&page->flags);
1023                 if (pages) {
1024                         page = list_entry(pages->prev, struct page, lru);
1025                         list_del(&page->lru);
1026                         if (add_to_page_cache_lru(page, mapping,
1027                                                   page->index,
1028                                                   readahead_gfp_mask(mapping)))
1029                                 goto next_page;
1030                 }
1031
1032                 block_in_file = (sector_t)page->index;
1033                 last_block = block_in_file + nr_pages;
1034                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1035                                                                 blkbits;
1036                 if (last_block > last_block_in_file)
1037                         last_block = last_block_in_file;
1038
1039                 /*
1040                  * Map blocks using the previous result first.
1041                  */
1042                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1043                                 block_in_file > map.m_lblk &&
1044                                 block_in_file < (map.m_lblk + map.m_len))
1045                         goto got_it;
1046
1047                 /*
1048                  * Then do more f2fs_map_blocks() calls until we are
1049                  * done with this page.
1050                  */
1051                 map.m_flags = 0;
1052
1053                 if (block_in_file < last_block) {
1054                         map.m_lblk = block_in_file;
1055                         map.m_len = last_block - block_in_file;
1056
1057                         if (f2fs_map_blocks(inode, &map, 0,
1058                                                 F2FS_GET_BLOCK_READ))
1059                                 goto set_error_page;
1060                 }
1061 got_it:
1062                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1063                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1064                         SetPageMappedToDisk(page);
1065
1066                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1067                                 SetPageUptodate(page);
1068                                 goto confused;
1069                         }
1070                 } else {
1071                         zero_user_segment(page, 0, PAGE_SIZE);
1072                         if (!PageUptodate(page))
1073                                 SetPageUptodate(page);
1074                         unlock_page(page);
1075                         goto next_page;
1076                 }
1077
1078                 /*
1079                  * This page will go to BIO.  Do we need to send this
1080                  * BIO off first?
1081                  */
1082                 if (bio && (last_block_in_bio != block_nr - 1)) {
1083 submit_and_realloc:
1084                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1085                         bio = NULL;
1086                 }
1087                 if (bio == NULL) {
1088                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1089                         if (IS_ERR(bio)) {
1090                                 bio = NULL;
1091                                 goto set_error_page;
1092                         }
1093                         bio_set_op_attrs(bio, REQ_OP_READ, 0);
1094                 }
1095
1096                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1097                         goto submit_and_realloc;
1098
1099                 last_block_in_bio = block_nr;
1100                 goto next_page;
1101 set_error_page:
1102                 SetPageError(page);
1103                 zero_user_segment(page, 0, PAGE_SIZE);
1104                 unlock_page(page);
1105                 goto next_page;
1106 confused:
1107                 if (bio) {
1108                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1109                         bio = NULL;
1110                 }
1111                 unlock_page(page);
1112 next_page:
1113                 if (pages)
1114                         put_page(page);
1115         }
1116         BUG_ON(pages && !list_empty(pages));
1117         if (bio)
1118                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1119         return 0;
1120 }
1121
1122 static int f2fs_read_data_page(struct file *file, struct page *page)
1123 {
1124         struct inode *inode = page->mapping->host;
1125         int ret = -EAGAIN;
1126
1127         trace_f2fs_readpage(page, DATA);
1128
1129         /* If the file has inline data, try to read it directly */
1130         if (f2fs_has_inline_data(inode))
1131                 ret = f2fs_read_inline_data(inode, page);
1132         if (ret == -EAGAIN)
1133                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1134         return ret;
1135 }
1136
1137 static int f2fs_read_data_pages(struct file *file,
1138                         struct address_space *mapping,
1139                         struct list_head *pages, unsigned nr_pages)
1140 {
1141         struct inode *inode = file->f_mapping->host;
1142         struct page *page = list_entry(pages->prev, struct page, lru);
1143
1144         trace_f2fs_readpages(inode, page, nr_pages);
1145
1146         /* If the file has inline data, skip readpages */
1147         if (f2fs_has_inline_data(inode))
1148                 return 0;
1149
1150         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1151 }
1152
1153 int do_write_data_page(struct f2fs_io_info *fio)
1154 {
1155         struct page *page = fio->page;
1156         struct inode *inode = page->mapping->host;
1157         struct dnode_of_data dn;
1158         int err = 0;
1159
1160         set_new_dnode(&dn, inode, NULL, NULL, 0);
1161         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1162         if (err)
1163                 return err;
1164
1165         fio->old_blkaddr = dn.data_blkaddr;
1166
1167         /* This page is already truncated */
1168         if (fio->old_blkaddr == NULL_ADDR) {
1169                 ClearPageUptodate(page);
1170                 goto out_writepage;
1171         }
1172
1173         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1174                 gfp_t gfp_flags = GFP_NOFS;
1175
1176                 /* wait for GCed encrypted page writeback */
1177                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1178                                                         fio->old_blkaddr);
1179 retry_encrypt:
1180                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1181                                                                 gfp_flags);
1182                 if (IS_ERR(fio->encrypted_page)) {
1183                         err = PTR_ERR(fio->encrypted_page);
1184                         if (err == -ENOMEM) {
1185                                 /* flush pending ios and wait for a while */
1186                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1187                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1188                                 gfp_flags |= __GFP_NOFAIL;
1189                                 err = 0;
1190                                 goto retry_encrypt;
1191                         }
1192                         goto out_writepage;
1193                 }
1194         }
1195
1196         set_page_writeback(page);
1197
1198         /*
1199          * If current allocation needs SSR,
1200          * it had better in-place writes for updated data.
1201          */
1202         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1203                         !is_cold_data(page) &&
1204                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1205                         need_inplace_update(inode))) {
1206                 rewrite_data_page(fio);
1207                 set_inode_flag(inode, FI_UPDATE_WRITE);
1208                 trace_f2fs_do_write_data_page(page, IPU);
1209         } else {
1210                 write_data_page(&dn, fio);
1211                 trace_f2fs_do_write_data_page(page, OPU);
1212                 set_inode_flag(inode, FI_APPEND_WRITE);
1213                 if (page->index == 0)
1214                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1215         }
1216 out_writepage:
1217         f2fs_put_dnode(&dn);
1218         return err;
1219 }
1220
1221 static int f2fs_write_data_page(struct page *page,
1222                                         struct writeback_control *wbc)
1223 {
1224         struct inode *inode = page->mapping->host;
1225         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1226         loff_t i_size = i_size_read(inode);
1227         const pgoff_t end_index = ((unsigned long long) i_size)
1228                                                         >> PAGE_SHIFT;
1229         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1230         unsigned offset = 0;
1231         bool need_balance_fs = false;
1232         int err = 0;
1233         struct f2fs_io_info fio = {
1234                 .sbi = sbi,
1235                 .type = DATA,
1236                 .op = REQ_OP_WRITE,
1237                 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1238                 .page = page,
1239                 .encrypted_page = NULL,
1240         };
1241
1242         trace_f2fs_writepage(page, DATA);
1243
1244         if (page->index < end_index)
1245                 goto write;
1246
1247         /*
1248          * If the offset is out-of-range of file size,
1249          * this page does not have to be written to disk.
1250          */
1251         offset = i_size & (PAGE_SIZE - 1);
1252         if ((page->index >= end_index + 1) || !offset)
1253                 goto out;
1254
1255         zero_user_segment(page, offset, PAGE_SIZE);
1256 write:
1257         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1258                 goto redirty_out;
1259         if (f2fs_is_drop_cache(inode))
1260                 goto out;
1261         /* we should not write 0'th page having journal header */
1262         if (f2fs_is_volatile_file(inode) && (!page->index ||
1263                         (!wbc->for_reclaim &&
1264                         available_free_memory(sbi, BASE_CHECK))))
1265                 goto redirty_out;
1266
1267         /* we should bypass data pages to proceed the kworkder jobs */
1268         if (unlikely(f2fs_cp_error(sbi))) {
1269                 mapping_set_error(page->mapping, -EIO);
1270                 goto out;
1271         }
1272
1273         /* Dentry blocks are controlled by checkpoint */
1274         if (S_ISDIR(inode->i_mode)) {
1275                 err = do_write_data_page(&fio);
1276                 goto done;
1277         }
1278
1279         if (!wbc->for_reclaim)
1280                 need_balance_fs = true;
1281         else if (has_not_enough_free_secs(sbi, 0, 0))
1282                 goto redirty_out;
1283
1284         err = -EAGAIN;
1285         f2fs_lock_op(sbi);
1286         if (f2fs_has_inline_data(inode))
1287                 err = f2fs_write_inline_data(inode, page);
1288         if (err == -EAGAIN)
1289                 err = do_write_data_page(&fio);
1290         if (F2FS_I(inode)->last_disk_size < psize)
1291                 F2FS_I(inode)->last_disk_size = psize;
1292         f2fs_unlock_op(sbi);
1293 done:
1294         if (err && err != -ENOENT)
1295                 goto redirty_out;
1296
1297         clear_cold_data(page);
1298 out:
1299         inode_dec_dirty_pages(inode);
1300         if (err)
1301                 ClearPageUptodate(page);
1302
1303         if (wbc->for_reclaim) {
1304                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1305                 remove_dirty_inode(inode);
1306         }
1307
1308         unlock_page(page);
1309         f2fs_balance_fs(sbi, need_balance_fs);
1310
1311         if (unlikely(f2fs_cp_error(sbi)))
1312                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1313
1314         return 0;
1315
1316 redirty_out:
1317         redirty_page_for_writepage(wbc, page);
1318         unlock_page(page);
1319         return err;
1320 }
1321
1322 /*
1323  * This function was copied from write_cche_pages from mm/page-writeback.c.
1324  * The major change is making write step of cold data page separately from
1325  * warm/hot data page.
1326  */
1327 static int f2fs_write_cache_pages(struct address_space *mapping,
1328                                         struct writeback_control *wbc)
1329 {
1330         int ret = 0;
1331         int done = 0;
1332         struct pagevec pvec;
1333         int nr_pages;
1334         pgoff_t uninitialized_var(writeback_index);
1335         pgoff_t index;
1336         pgoff_t end;            /* Inclusive */
1337         pgoff_t done_index;
1338         int cycled;
1339         int range_whole = 0;
1340         int tag;
1341         int nwritten = 0;
1342
1343         pagevec_init(&pvec, 0);
1344
1345         if (wbc->range_cyclic) {
1346                 writeback_index = mapping->writeback_index; /* prev offset */
1347                 index = writeback_index;
1348                 if (index == 0)
1349                         cycled = 1;
1350                 else
1351                         cycled = 0;
1352                 end = -1;
1353         } else {
1354                 index = wbc->range_start >> PAGE_SHIFT;
1355                 end = wbc->range_end >> PAGE_SHIFT;
1356                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1357                         range_whole = 1;
1358                 cycled = 1; /* ignore range_cyclic tests */
1359         }
1360         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1361                 tag = PAGECACHE_TAG_TOWRITE;
1362         else
1363                 tag = PAGECACHE_TAG_DIRTY;
1364 retry:
1365         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1366                 tag_pages_for_writeback(mapping, index, end);
1367         done_index = index;
1368         while (!done && (index <= end)) {
1369                 int i;
1370
1371                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1372                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1373                 if (nr_pages == 0)
1374                         break;
1375
1376                 for (i = 0; i < nr_pages; i++) {
1377                         struct page *page = pvec.pages[i];
1378
1379                         if (page->index > end) {
1380                                 done = 1;
1381                                 break;
1382                         }
1383
1384                         done_index = page->index;
1385
1386                         lock_page(page);
1387
1388                         if (unlikely(page->mapping != mapping)) {
1389 continue_unlock:
1390                                 unlock_page(page);
1391                                 continue;
1392                         }
1393
1394                         if (!PageDirty(page)) {
1395                                 /* someone wrote it for us */
1396                                 goto continue_unlock;
1397                         }
1398
1399                         if (PageWriteback(page)) {
1400                                 if (wbc->sync_mode != WB_SYNC_NONE)
1401                                         f2fs_wait_on_page_writeback(page,
1402                                                                 DATA, true);
1403                                 else
1404                                         goto continue_unlock;
1405                         }
1406
1407                         BUG_ON(PageWriteback(page));
1408                         if (!clear_page_dirty_for_io(page))
1409                                 goto continue_unlock;
1410
1411                         ret = mapping->a_ops->writepage(page, wbc);
1412                         if (unlikely(ret)) {
1413                                 done_index = page->index + 1;
1414                                 done = 1;
1415                                 break;
1416                         } else {
1417                                 nwritten++;
1418                         }
1419
1420                         if (--wbc->nr_to_write <= 0 &&
1421                             wbc->sync_mode == WB_SYNC_NONE) {
1422                                 done = 1;
1423                                 break;
1424                         }
1425                 }
1426                 pagevec_release(&pvec);
1427                 cond_resched();
1428         }
1429
1430         if (!cycled && !done) {
1431                 cycled = 1;
1432                 index = 0;
1433                 end = writeback_index - 1;
1434                 goto retry;
1435         }
1436         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1437                 mapping->writeback_index = done_index;
1438
1439         if (nwritten)
1440                 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1441                                                         NULL, 0, DATA, WRITE);
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
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) && len == PAGE_SIZE)
1522                 return 0;
1523
1524         if (f2fs_has_inline_data(inode) ||
1525                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1526                 f2fs_lock_op(sbi);
1527                 locked = true;
1528         }
1529 restart:
1530         /* check inline_data */
1531         ipage = get_node_page(sbi, inode->i_ino);
1532         if (IS_ERR(ipage)) {
1533                 err = PTR_ERR(ipage);
1534                 goto unlock_out;
1535         }
1536
1537         set_new_dnode(&dn, inode, ipage, ipage, 0);
1538
1539         if (f2fs_has_inline_data(inode)) {
1540                 if (pos + len <= MAX_INLINE_DATA) {
1541                         read_inline_data(page, ipage);
1542                         set_inode_flag(inode, FI_DATA_EXIST);
1543                         if (inode->i_nlink)
1544                                 set_inline_node(ipage);
1545                 } else {
1546                         err = f2fs_convert_inline_page(&dn, page);
1547                         if (err)
1548                                 goto out;
1549                         if (dn.data_blkaddr == NULL_ADDR)
1550                                 err = f2fs_get_block(&dn, index);
1551                 }
1552         } else if (locked) {
1553                 err = f2fs_get_block(&dn, index);
1554         } else {
1555                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1556                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1557                 } else {
1558                         /* hole case */
1559                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1560                         if (err || dn.data_blkaddr == NULL_ADDR) {
1561                                 f2fs_put_dnode(&dn);
1562                                 f2fs_lock_op(sbi);
1563                                 locked = true;
1564                                 goto restart;
1565                         }
1566                 }
1567         }
1568
1569         /* convert_inline_page can make node_changed */
1570         *blk_addr = dn.data_blkaddr;
1571         *node_changed = dn.node_changed;
1572 out:
1573         f2fs_put_dnode(&dn);
1574 unlock_out:
1575         if (locked)
1576                 f2fs_unlock_op(sbi);
1577         return err;
1578 }
1579
1580 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1581                 loff_t pos, unsigned len, unsigned flags,
1582                 struct page **pagep, void **fsdata)
1583 {
1584         struct inode *inode = mapping->host;
1585         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1586         struct page *page = NULL;
1587         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1588         bool need_balance = false;
1589         block_t blkaddr = NULL_ADDR;
1590         int err = 0;
1591
1592         trace_f2fs_write_begin(inode, pos, len, flags);
1593
1594         /*
1595          * We should check this at this moment to avoid deadlock on inode page
1596          * and #0 page. The locking rule for inline_data conversion should be:
1597          * lock_page(page #0) -> lock_page(inode_page)
1598          */
1599         if (index != 0) {
1600                 err = f2fs_convert_inline_inode(inode);
1601                 if (err)
1602                         goto fail;
1603         }
1604 repeat:
1605         page = grab_cache_page_write_begin(mapping, index, flags);
1606         if (!page) {
1607                 err = -ENOMEM;
1608                 goto fail;
1609         }
1610
1611         *pagep = page;
1612
1613         err = prepare_write_begin(sbi, page, pos, len,
1614                                         &blkaddr, &need_balance);
1615         if (err)
1616                 goto fail;
1617
1618         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1619                 unlock_page(page);
1620                 f2fs_balance_fs(sbi, true);
1621                 lock_page(page);
1622                 if (page->mapping != mapping) {
1623                         /* The page got truncated from under us */
1624                         f2fs_put_page(page, 1);
1625                         goto repeat;
1626                 }
1627         }
1628
1629         f2fs_wait_on_page_writeback(page, DATA, false);
1630
1631         /* wait for GCed encrypted page writeback */
1632         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1633                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1634
1635         if (len == PAGE_SIZE || PageUptodate(page))
1636                 return 0;
1637
1638         if (blkaddr == NEW_ADDR) {
1639                 zero_user_segment(page, 0, PAGE_SIZE);
1640                 SetPageUptodate(page);
1641         } else {
1642                 struct bio *bio;
1643
1644                 bio = f2fs_grab_bio(inode, blkaddr, 1);
1645                 if (IS_ERR(bio)) {
1646                         err = PTR_ERR(bio);
1647                         goto fail;
1648                 }
1649                 bio_set_op_attrs(bio, REQ_OP_READ, READ_SYNC);
1650                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1651                         bio_put(bio);
1652                         err = -EFAULT;
1653                         goto fail;
1654                 }
1655
1656                 __submit_bio(sbi, bio, DATA);
1657
1658                 lock_page(page);
1659                 if (unlikely(page->mapping != mapping)) {
1660                         f2fs_put_page(page, 1);
1661                         goto repeat;
1662                 }
1663                 if (unlikely(!PageUptodate(page))) {
1664                         err = -EIO;
1665                         goto fail;
1666                 }
1667         }
1668         return 0;
1669
1670 fail:
1671         f2fs_put_page(page, 1);
1672         f2fs_write_failed(mapping, pos + len);
1673         return err;
1674 }
1675
1676 static int f2fs_write_end(struct file *file,
1677                         struct address_space *mapping,
1678                         loff_t pos, unsigned len, unsigned copied,
1679                         struct page *page, void *fsdata)
1680 {
1681         struct inode *inode = page->mapping->host;
1682
1683         trace_f2fs_write_end(inode, pos, len, copied);
1684
1685         /*
1686          * This should be come from len == PAGE_SIZE, and we expect copied
1687          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1688          * let generic_perform_write() try to copy data again through copied=0.
1689          */
1690         if (!PageUptodate(page)) {
1691                 if (unlikely(copied != PAGE_SIZE))
1692                         copied = 0;
1693                 else
1694                         SetPageUptodate(page);
1695         }
1696         if (!copied)
1697                 goto unlock_out;
1698
1699         set_page_dirty(page);
1700         clear_cold_data(page);
1701
1702         if (pos + copied > i_size_read(inode))
1703                 f2fs_i_size_write(inode, pos + copied);
1704 unlock_out:
1705         f2fs_put_page(page, 1);
1706         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1707         return copied;
1708 }
1709
1710 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1711                            loff_t offset)
1712 {
1713         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1714
1715         if (offset & blocksize_mask)
1716                 return -EINVAL;
1717
1718         if (iov_iter_alignment(iter) & blocksize_mask)
1719                 return -EINVAL;
1720
1721         return 0;
1722 }
1723
1724 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1725 {
1726         struct address_space *mapping = iocb->ki_filp->f_mapping;
1727         struct inode *inode = mapping->host;
1728         size_t count = iov_iter_count(iter);
1729         loff_t offset = iocb->ki_pos;
1730         int rw = iov_iter_rw(iter);
1731         int err;
1732
1733         err = check_direct_IO(inode, iter, offset);
1734         if (err)
1735                 return err;
1736
1737         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1738                 return 0;
1739         if (test_opt(F2FS_I_SB(inode), LFS))
1740                 return 0;
1741
1742         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1743
1744         down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1745         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1746         up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1747
1748         if (rw == WRITE) {
1749                 if (err > 0)
1750                         set_inode_flag(inode, FI_UPDATE_WRITE);
1751                 else if (err < 0)
1752                         f2fs_write_failed(mapping, offset + count);
1753         }
1754
1755         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1756
1757         return err;
1758 }
1759
1760 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1761                                                         unsigned int length)
1762 {
1763         struct inode *inode = page->mapping->host;
1764         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1765
1766         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1767                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1768                 return;
1769
1770         if (PageDirty(page)) {
1771                 if (inode->i_ino == F2FS_META_INO(sbi)) {
1772                         dec_page_count(sbi, F2FS_DIRTY_META);
1773                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
1774                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1775                 } else {
1776                         inode_dec_dirty_pages(inode);
1777                         remove_dirty_inode(inode);
1778                 }
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 #ifdef CONFIG_MIGRATION
1878 #include <linux/migrate.h>
1879
1880 int f2fs_migrate_page(struct address_space *mapping,
1881                 struct page *newpage, struct page *page, enum migrate_mode mode)
1882 {
1883         int rc, extra_count;
1884         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
1885         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
1886
1887         BUG_ON(PageWriteback(page));
1888
1889         /* migrating an atomic written page is safe with the inmem_lock hold */
1890         if (atomic_written && !mutex_trylock(&fi->inmem_lock))
1891                 return -EAGAIN;
1892
1893         /*
1894          * A reference is expected if PagePrivate set when move mapping,
1895          * however F2FS breaks this for maintaining dirty page counts when
1896          * truncating pages. So here adjusting the 'extra_count' make it work.
1897          */
1898         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
1899         rc = migrate_page_move_mapping(mapping, newpage,
1900                                 page, NULL, mode, extra_count);
1901         if (rc != MIGRATEPAGE_SUCCESS) {
1902                 if (atomic_written)
1903                         mutex_unlock(&fi->inmem_lock);
1904                 return rc;
1905         }
1906
1907         if (atomic_written) {
1908                 struct inmem_pages *cur;
1909                 list_for_each_entry(cur, &fi->inmem_pages, list)
1910                         if (cur->page == page) {
1911                                 cur->page = newpage;
1912                                 break;
1913                         }
1914                 mutex_unlock(&fi->inmem_lock);
1915                 put_page(page);
1916                 get_page(newpage);
1917         }
1918
1919         if (PagePrivate(page))
1920                 SetPagePrivate(newpage);
1921         set_page_private(newpage, page_private(page));
1922
1923         migrate_page_copy(newpage, page);
1924
1925         return MIGRATEPAGE_SUCCESS;
1926 }
1927 #endif
1928
1929 const struct address_space_operations f2fs_dblock_aops = {
1930         .readpage       = f2fs_read_data_page,
1931         .readpages      = f2fs_read_data_pages,
1932         .writepage      = f2fs_write_data_page,
1933         .writepages     = f2fs_write_data_pages,
1934         .write_begin    = f2fs_write_begin,
1935         .write_end      = f2fs_write_end,
1936         .set_page_dirty = f2fs_set_data_page_dirty,
1937         .invalidatepage = f2fs_invalidate_page,
1938         .releasepage    = f2fs_release_page,
1939         .direct_IO      = f2fs_direct_IO,
1940         .bmap           = f2fs_bmap,
1941 #ifdef CONFIG_MIGRATION
1942         .migratepage    = f2fs_migrate_page,
1943 #endif
1944 };