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