linux/fs/f2fs/data.c
<<
>>
Prefs
   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/cleancache.h>
  23#include <linux/sched/signal.h>
  24
  25#include "f2fs.h"
  26#include "node.h"
  27#include "segment.h"
  28#include "trace.h"
  29#include <trace/events/f2fs.h>
  30
  31#define NUM_PREALLOC_POST_READ_CTXS     128
  32
  33static struct kmem_cache *bio_post_read_ctx_cache;
  34static mempool_t *bio_post_read_ctx_pool;
  35
  36static bool __is_cp_guaranteed(struct page *page)
  37{
  38        struct address_space *mapping = page->mapping;
  39        struct inode *inode;
  40        struct f2fs_sb_info *sbi;
  41
  42        if (!mapping)
  43                return false;
  44
  45        inode = mapping->host;
  46        sbi = F2FS_I_SB(inode);
  47
  48        if (inode->i_ino == F2FS_META_INO(sbi) ||
  49                        inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  50                        S_ISDIR(inode->i_mode) ||
  51                        (S_ISREG(inode->i_mode) &&
  52                        is_inode_flag_set(inode, FI_ATOMIC_FILE)) ||
  53                        is_cold_data(page))
  54                return true;
  55        return false;
  56}
  57
  58/* postprocessing steps for read bios */
  59enum bio_post_read_step {
  60        STEP_INITIAL = 0,
  61        STEP_DECRYPT,
  62};
  63
  64struct bio_post_read_ctx {
  65        struct bio *bio;
  66        struct work_struct work;
  67        unsigned int cur_step;
  68        unsigned int enabled_steps;
  69};
  70
  71static void __read_end_io(struct bio *bio)
  72{
  73        struct page *page;
  74        struct bio_vec *bv;
  75        int i;
  76
  77        bio_for_each_segment_all(bv, bio, i) {
  78                page = bv->bv_page;
  79
  80                /* PG_error was set if any post_read step failed */
  81                if (bio->bi_status || PageError(page)) {
  82                        ClearPageUptodate(page);
  83                        SetPageError(page);
  84                } else {
  85                        SetPageUptodate(page);
  86                }
  87                unlock_page(page);
  88        }
  89        if (bio->bi_private)
  90                mempool_free(bio->bi_private, bio_post_read_ctx_pool);
  91        bio_put(bio);
  92}
  93
  94static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
  95
  96static void decrypt_work(struct work_struct *work)
  97{
  98        struct bio_post_read_ctx *ctx =
  99                container_of(work, struct bio_post_read_ctx, work);
 100
 101        fscrypt_decrypt_bio(ctx->bio);
 102
 103        bio_post_read_processing(ctx);
 104}
 105
 106static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 107{
 108        switch (++ctx->cur_step) {
 109        case STEP_DECRYPT:
 110                if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 111                        INIT_WORK(&ctx->work, decrypt_work);
 112                        fscrypt_enqueue_decrypt_work(&ctx->work);
 113                        return;
 114                }
 115                ctx->cur_step++;
 116                /* fall-through */
 117        default:
 118                __read_end_io(ctx->bio);
 119        }
 120}
 121
 122static bool f2fs_bio_post_read_required(struct bio *bio)
 123{
 124        return bio->bi_private && !bio->bi_status;
 125}
 126
 127static void f2fs_read_end_io(struct bio *bio)
 128{
 129#ifdef CONFIG_F2FS_FAULT_INJECTION
 130        if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)), FAULT_IO)) {
 131                f2fs_show_injection_info(FAULT_IO);
 132                bio->bi_status = BLK_STS_IOERR;
 133        }
 134#endif
 135
 136        if (f2fs_bio_post_read_required(bio)) {
 137                struct bio_post_read_ctx *ctx = bio->bi_private;
 138
 139                ctx->cur_step = STEP_INITIAL;
 140                bio_post_read_processing(ctx);
 141                return;
 142        }
 143
 144        __read_end_io(bio);
 145}
 146
 147static void f2fs_write_end_io(struct bio *bio)
 148{
 149        struct f2fs_sb_info *sbi = bio->bi_private;
 150        struct bio_vec *bvec;
 151        int i;
 152
 153        bio_for_each_segment_all(bvec, bio, i) {
 154                struct page *page = bvec->bv_page;
 155                enum count_type type = WB_DATA_TYPE(page);
 156
 157                if (IS_DUMMY_WRITTEN_PAGE(page)) {
 158                        set_page_private(page, (unsigned long)NULL);
 159                        ClearPagePrivate(page);
 160                        unlock_page(page);
 161                        mempool_free(page, sbi->write_io_dummy);
 162
 163                        if (unlikely(bio->bi_status))
 164                                f2fs_stop_checkpoint(sbi, true);
 165                        continue;
 166                }
 167
 168                fscrypt_pullback_bio_page(&page, true);
 169
 170                if (unlikely(bio->bi_status)) {
 171                        mapping_set_error(page->mapping, -EIO);
 172                        if (type == F2FS_WB_CP_DATA)
 173                                f2fs_stop_checkpoint(sbi, true);
 174                }
 175
 176                f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 177                                        page->index != nid_of_node(page));
 178
 179                dec_page_count(sbi, type);
 180                clear_cold_data(page);
 181                end_page_writeback(page);
 182        }
 183        if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 184                                wq_has_sleeper(&sbi->cp_wait))
 185                wake_up(&sbi->cp_wait);
 186
 187        bio_put(bio);
 188}
 189
 190/*
 191 * Return true, if pre_bio's bdev is same as its target device.
 192 */
 193struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 194                                block_t blk_addr, struct bio *bio)
 195{
 196        struct block_device *bdev = sbi->sb->s_bdev;
 197        int i;
 198
 199        for (i = 0; i < sbi->s_ndevs; i++) {
 200                if (FDEV(i).start_blk <= blk_addr &&
 201                                        FDEV(i).end_blk >= blk_addr) {
 202                        blk_addr -= FDEV(i).start_blk;
 203                        bdev = FDEV(i).bdev;
 204                        break;
 205                }
 206        }
 207        if (bio) {
 208                bio_set_dev(bio, bdev);
 209                bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 210        }
 211        return bdev;
 212}
 213
 214int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 215{
 216        int i;
 217
 218        for (i = 0; i < sbi->s_ndevs; i++)
 219                if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 220                        return i;
 221        return 0;
 222}
 223
 224static bool __same_bdev(struct f2fs_sb_info *sbi,
 225                                block_t blk_addr, struct bio *bio)
 226{
 227        struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 228        return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 229}
 230
 231/*
 232 * Low-level block read/write IO operations.
 233 */
 234static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 235                                struct writeback_control *wbc,
 236                                int npages, bool is_read,
 237                                enum page_type type, enum temp_type temp)
 238{
 239        struct bio *bio;
 240
 241        bio = f2fs_bio_alloc(sbi, npages, true);
 242
 243        f2fs_target_device(sbi, blk_addr, bio);
 244        if (is_read) {
 245                bio->bi_end_io = f2fs_read_end_io;
 246                bio->bi_private = NULL;
 247        } else {
 248                bio->bi_end_io = f2fs_write_end_io;
 249                bio->bi_private = sbi;
 250                bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 251        }
 252        if (wbc)
 253                wbc_init_bio(wbc, bio);
 254
 255        return bio;
 256}
 257
 258static inline void __submit_bio(struct f2fs_sb_info *sbi,
 259                                struct bio *bio, enum page_type type)
 260{
 261        if (!is_read_io(bio_op(bio))) {
 262                unsigned int start;
 263
 264                if (type != DATA && type != NODE)
 265                        goto submit_io;
 266
 267                if (f2fs_sb_has_blkzoned(sbi->sb) && current->plug)
 268                        blk_finish_plug(current->plug);
 269
 270                start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 271                start %= F2FS_IO_SIZE(sbi);
 272
 273                if (start == 0)
 274                        goto submit_io;
 275
 276                /* fill dummy pages */
 277                for (; start < F2FS_IO_SIZE(sbi); start++) {
 278                        struct page *page =
 279                                mempool_alloc(sbi->write_io_dummy,
 280                                        GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 281                        f2fs_bug_on(sbi, !page);
 282
 283                        SetPagePrivate(page);
 284                        set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 285                        lock_page(page);
 286                        if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 287                                f2fs_bug_on(sbi, 1);
 288                }
 289                /*
 290                 * In the NODE case, we lose next block address chain. So, we
 291                 * need to do checkpoint in f2fs_sync_file.
 292                 */
 293                if (type == NODE)
 294                        set_sbi_flag(sbi, SBI_NEED_CP);
 295        }
 296submit_io:
 297        if (is_read_io(bio_op(bio)))
 298                trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 299        else
 300                trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 301        submit_bio(bio);
 302}
 303
 304static void __submit_merged_bio(struct f2fs_bio_info *io)
 305{
 306        struct f2fs_io_info *fio = &io->fio;
 307
 308        if (!io->bio)
 309                return;
 310
 311        bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 312
 313        if (is_read_io(fio->op))
 314                trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 315        else
 316                trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 317
 318        __submit_bio(io->sbi, io->bio, fio->type);
 319        io->bio = NULL;
 320}
 321
 322static bool __has_merged_page(struct f2fs_bio_info *io,
 323                                struct inode *inode, nid_t ino, pgoff_t idx)
 324{
 325        struct bio_vec *bvec;
 326        struct page *target;
 327        int i;
 328
 329        if (!io->bio)
 330                return false;
 331
 332        if (!inode && !ino)
 333                return true;
 334
 335        bio_for_each_segment_all(bvec, io->bio, i) {
 336
 337                if (bvec->bv_page->mapping)
 338                        target = bvec->bv_page;
 339                else
 340                        target = fscrypt_control_page(bvec->bv_page);
 341
 342                if (idx != target->index)
 343                        continue;
 344
 345                if (inode && inode == target->mapping->host)
 346                        return true;
 347                if (ino && ino == ino_of_node(target))
 348                        return true;
 349        }
 350
 351        return false;
 352}
 353
 354static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 355                                nid_t ino, pgoff_t idx, enum page_type type)
 356{
 357        enum page_type btype = PAGE_TYPE_OF_BIO(type);
 358        enum temp_type temp;
 359        struct f2fs_bio_info *io;
 360        bool ret = false;
 361
 362        for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 363                io = sbi->write_io[btype] + temp;
 364
 365                down_read(&io->io_rwsem);
 366                ret = __has_merged_page(io, inode, ino, idx);
 367                up_read(&io->io_rwsem);
 368
 369                /* TODO: use HOT temp only for meta pages now. */
 370                if (ret || btype == META)
 371                        break;
 372        }
 373        return ret;
 374}
 375
 376static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 377                                enum page_type type, enum temp_type temp)
 378{
 379        enum page_type btype = PAGE_TYPE_OF_BIO(type);
 380        struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 381
 382        down_write(&io->io_rwsem);
 383
 384        /* change META to META_FLUSH in the checkpoint procedure */
 385        if (type >= META_FLUSH) {
 386                io->fio.type = META_FLUSH;
 387                io->fio.op = REQ_OP_WRITE;
 388                io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 389                if (!test_opt(sbi, NOBARRIER))
 390                        io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 391        }
 392        __submit_merged_bio(io);
 393        up_write(&io->io_rwsem);
 394}
 395
 396static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 397                                struct inode *inode, nid_t ino, pgoff_t idx,
 398                                enum page_type type, bool force)
 399{
 400        enum temp_type temp;
 401
 402        if (!force && !has_merged_page(sbi, inode, ino, idx, type))
 403                return;
 404
 405        for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 406
 407                __f2fs_submit_merged_write(sbi, type, temp);
 408
 409                /* TODO: use HOT temp only for meta pages now. */
 410                if (type >= META)
 411                        break;
 412        }
 413}
 414
 415void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 416{
 417        __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 418}
 419
 420void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 421                                struct inode *inode, nid_t ino, pgoff_t idx,
 422                                enum page_type type)
 423{
 424        __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
 425}
 426
 427void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 428{
 429        f2fs_submit_merged_write(sbi, DATA);
 430        f2fs_submit_merged_write(sbi, NODE);
 431        f2fs_submit_merged_write(sbi, META);
 432}
 433
 434/*
 435 * Fill the locked page with data located in the block address.
 436 * A caller needs to unlock the page on failure.
 437 */
 438int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 439{
 440        struct bio *bio;
 441        struct page *page = fio->encrypted_page ?
 442                        fio->encrypted_page : fio->page;
 443
 444        verify_block_addr(fio, fio->new_blkaddr);
 445        trace_f2fs_submit_page_bio(page, fio);
 446        f2fs_trace_ios(fio, 0);
 447
 448        /* Allocate a new bio */
 449        bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 450                                1, is_read_io(fio->op), fio->type, fio->temp);
 451
 452        if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 453                bio_put(bio);
 454                return -EFAULT;
 455        }
 456        bio_set_op_attrs(bio, fio->op, fio->op_flags);
 457
 458        __submit_bio(fio->sbi, bio, fio->type);
 459
 460        if (!is_read_io(fio->op))
 461                inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 462        return 0;
 463}
 464
 465void f2fs_submit_page_write(struct f2fs_io_info *fio)
 466{
 467        struct f2fs_sb_info *sbi = fio->sbi;
 468        enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 469        struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 470        struct page *bio_page;
 471
 472        f2fs_bug_on(sbi, is_read_io(fio->op));
 473
 474        down_write(&io->io_rwsem);
 475next:
 476        if (fio->in_list) {
 477                spin_lock(&io->io_lock);
 478                if (list_empty(&io->io_list)) {
 479                        spin_unlock(&io->io_lock);
 480                        goto out;
 481                }
 482                fio = list_first_entry(&io->io_list,
 483                                                struct f2fs_io_info, list);
 484                list_del(&fio->list);
 485                spin_unlock(&io->io_lock);
 486        }
 487
 488        if (is_valid_blkaddr(fio->old_blkaddr))
 489                verify_block_addr(fio, fio->old_blkaddr);
 490        verify_block_addr(fio, fio->new_blkaddr);
 491
 492        bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 493
 494        /* set submitted = true as a return value */
 495        fio->submitted = true;
 496
 497        inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 498
 499        if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 500            (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 501                        !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 502                __submit_merged_bio(io);
 503alloc_new:
 504        if (io->bio == NULL) {
 505                if ((fio->type == DATA || fio->type == NODE) &&
 506                                fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 507                        dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 508                        fio->retry = true;
 509                        goto skip;
 510                }
 511                io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 512                                                BIO_MAX_PAGES, false,
 513                                                fio->type, fio->temp);
 514                io->fio = *fio;
 515        }
 516
 517        if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 518                __submit_merged_bio(io);
 519                goto alloc_new;
 520        }
 521
 522        if (fio->io_wbc)
 523                wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 524
 525        io->last_block_in_bio = fio->new_blkaddr;
 526        f2fs_trace_ios(fio, 0);
 527
 528        trace_f2fs_submit_page_write(fio->page, fio);
 529skip:
 530        if (fio->in_list)
 531                goto next;
 532out:
 533        up_write(&io->io_rwsem);
 534}
 535
 536static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 537                                                         unsigned nr_pages)
 538{
 539        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 540        struct bio *bio;
 541        struct bio_post_read_ctx *ctx;
 542        unsigned int post_read_steps = 0;
 543
 544        bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 545        if (!bio)
 546                return ERR_PTR(-ENOMEM);
 547        f2fs_target_device(sbi, blkaddr, bio);
 548        bio->bi_end_io = f2fs_read_end_io;
 549        bio_set_op_attrs(bio, REQ_OP_READ, 0);
 550
 551        if (f2fs_encrypted_file(inode))
 552                post_read_steps |= 1 << STEP_DECRYPT;
 553        if (post_read_steps) {
 554                ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 555                if (!ctx) {
 556                        bio_put(bio);
 557                        return ERR_PTR(-ENOMEM);
 558                }
 559                ctx->bio = bio;
 560                ctx->enabled_steps = post_read_steps;
 561                bio->bi_private = ctx;
 562
 563                /* wait the page to be moved by cleaning */
 564                f2fs_wait_on_block_writeback(sbi, blkaddr);
 565        }
 566
 567        return bio;
 568}
 569
 570/* This can handle encryption stuffs */
 571static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 572                                                        block_t blkaddr)
 573{
 574        struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1);
 575
 576        if (IS_ERR(bio))
 577                return PTR_ERR(bio);
 578
 579        if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 580                bio_put(bio);
 581                return -EFAULT;
 582        }
 583        __submit_bio(F2FS_I_SB(inode), bio, DATA);
 584        return 0;
 585}
 586
 587static void __set_data_blkaddr(struct dnode_of_data *dn)
 588{
 589        struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 590        __le32 *addr_array;
 591        int base = 0;
 592
 593        if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 594                base = get_extra_isize(dn->inode);
 595
 596        /* Get physical address of data block */
 597        addr_array = blkaddr_in_node(rn);
 598        addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 599}
 600
 601/*
 602 * Lock ordering for the change of data block address:
 603 * ->data_page
 604 *  ->node_page
 605 *    update block addresses in the node page
 606 */
 607void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 608{
 609        f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 610        __set_data_blkaddr(dn);
 611        if (set_page_dirty(dn->node_page))
 612                dn->node_changed = true;
 613}
 614
 615void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 616{
 617        dn->data_blkaddr = blkaddr;
 618        f2fs_set_data_blkaddr(dn);
 619        f2fs_update_extent_cache(dn);
 620}
 621
 622/* dn->ofs_in_node will be returned with up-to-date last block pointer */
 623int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 624{
 625        struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 626        int err;
 627
 628        if (!count)
 629                return 0;
 630
 631        if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 632                return -EPERM;
 633        if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 634                return err;
 635
 636        trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 637                                                dn->ofs_in_node, count);
 638
 639        f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 640
 641        for (; count > 0; dn->ofs_in_node++) {
 642                block_t blkaddr = datablock_addr(dn->inode,
 643                                        dn->node_page, dn->ofs_in_node);
 644                if (blkaddr == NULL_ADDR) {
 645                        dn->data_blkaddr = NEW_ADDR;
 646                        __set_data_blkaddr(dn);
 647                        count--;
 648                }
 649        }
 650
 651        if (set_page_dirty(dn->node_page))
 652                dn->node_changed = true;
 653        return 0;
 654}
 655
 656/* Should keep dn->ofs_in_node unchanged */
 657int f2fs_reserve_new_block(struct dnode_of_data *dn)
 658{
 659        unsigned int ofs_in_node = dn->ofs_in_node;
 660        int ret;
 661
 662        ret = f2fs_reserve_new_blocks(dn, 1);
 663        dn->ofs_in_node = ofs_in_node;
 664        return ret;
 665}
 666
 667int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 668{
 669        bool need_put = dn->inode_page ? false : true;
 670        int err;
 671
 672        err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 673        if (err)
 674                return err;
 675
 676        if (dn->data_blkaddr == NULL_ADDR)
 677                err = f2fs_reserve_new_block(dn);
 678        if (err || need_put)
 679                f2fs_put_dnode(dn);
 680        return err;
 681}
 682
 683int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 684{
 685        struct extent_info ei  = {0,0,0};
 686        struct inode *inode = dn->inode;
 687
 688        if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 689                dn->data_blkaddr = ei.blk + index - ei.fofs;
 690                return 0;
 691        }
 692
 693        return f2fs_reserve_block(dn, index);
 694}
 695
 696struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 697                                                int op_flags, bool for_write)
 698{
 699        struct address_space *mapping = inode->i_mapping;
 700        struct dnode_of_data dn;
 701        struct page *page;
 702        struct extent_info ei = {0,0,0};
 703        int err;
 704
 705        page = f2fs_grab_cache_page(mapping, index, for_write);
 706        if (!page)
 707                return ERR_PTR(-ENOMEM);
 708
 709        if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 710                dn.data_blkaddr = ei.blk + index - ei.fofs;
 711                goto got_it;
 712        }
 713
 714        set_new_dnode(&dn, inode, NULL, NULL, 0);
 715        err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 716        if (err)
 717                goto put_err;
 718        f2fs_put_dnode(&dn);
 719
 720        if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 721                err = -ENOENT;
 722                goto put_err;
 723        }
 724got_it:
 725        if (PageUptodate(page)) {
 726                unlock_page(page);
 727                return page;
 728        }
 729
 730        /*
 731         * A new dentry page is allocated but not able to be written, since its
 732         * new inode page couldn't be allocated due to -ENOSPC.
 733         * In such the case, its blkaddr can be remained as NEW_ADDR.
 734         * see, f2fs_add_link -> f2fs_get_new_data_page ->
 735         * f2fs_init_inode_metadata.
 736         */
 737        if (dn.data_blkaddr == NEW_ADDR) {
 738                zero_user_segment(page, 0, PAGE_SIZE);
 739                if (!PageUptodate(page))
 740                        SetPageUptodate(page);
 741                unlock_page(page);
 742                return page;
 743        }
 744
 745        err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 746        if (err)
 747                goto put_err;
 748        return page;
 749
 750put_err:
 751        f2fs_put_page(page, 1);
 752        return ERR_PTR(err);
 753}
 754
 755struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 756{
 757        struct address_space *mapping = inode->i_mapping;
 758        struct page *page;
 759
 760        page = find_get_page(mapping, index);
 761        if (page && PageUptodate(page))
 762                return page;
 763        f2fs_put_page(page, 0);
 764
 765        page = f2fs_get_read_data_page(inode, index, 0, false);
 766        if (IS_ERR(page))
 767                return page;
 768
 769        if (PageUptodate(page))
 770                return page;
 771
 772        wait_on_page_locked(page);
 773        if (unlikely(!PageUptodate(page))) {
 774                f2fs_put_page(page, 0);
 775                return ERR_PTR(-EIO);
 776        }
 777        return page;
 778}
 779
 780/*
 781 * If it tries to access a hole, return an error.
 782 * Because, the callers, functions in dir.c and GC, should be able to know
 783 * whether this page exists or not.
 784 */
 785struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 786                                                        bool for_write)
 787{
 788        struct address_space *mapping = inode->i_mapping;
 789        struct page *page;
 790repeat:
 791        page = f2fs_get_read_data_page(inode, index, 0, for_write);
 792        if (IS_ERR(page))
 793                return page;
 794
 795        /* wait for read completion */
 796        lock_page(page);
 797        if (unlikely(page->mapping != mapping)) {
 798                f2fs_put_page(page, 1);
 799                goto repeat;
 800        }
 801        if (unlikely(!PageUptodate(page))) {
 802                f2fs_put_page(page, 1);
 803                return ERR_PTR(-EIO);
 804        }
 805        return page;
 806}
 807
 808/*
 809 * Caller ensures that this data page is never allocated.
 810 * A new zero-filled data page is allocated in the page cache.
 811 *
 812 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 813 * f2fs_unlock_op().
 814 * Note that, ipage is set only by make_empty_dir, and if any error occur,
 815 * ipage should be released by this function.
 816 */
 817struct page *f2fs_get_new_data_page(struct inode *inode,
 818                struct page *ipage, pgoff_t index, bool new_i_size)
 819{
 820        struct address_space *mapping = inode->i_mapping;
 821        struct page *page;
 822        struct dnode_of_data dn;
 823        int err;
 824
 825        page = f2fs_grab_cache_page(mapping, index, true);
 826        if (!page) {
 827                /*
 828                 * before exiting, we should make sure ipage will be released
 829                 * if any error occur.
 830                 */
 831                f2fs_put_page(ipage, 1);
 832                return ERR_PTR(-ENOMEM);
 833        }
 834
 835        set_new_dnode(&dn, inode, ipage, NULL, 0);
 836        err = f2fs_reserve_block(&dn, index);
 837        if (err) {
 838                f2fs_put_page(page, 1);
 839                return ERR_PTR(err);
 840        }
 841        if (!ipage)
 842                f2fs_put_dnode(&dn);
 843
 844        if (PageUptodate(page))
 845                goto got_it;
 846
 847        if (dn.data_blkaddr == NEW_ADDR) {
 848                zero_user_segment(page, 0, PAGE_SIZE);
 849                if (!PageUptodate(page))
 850                        SetPageUptodate(page);
 851        } else {
 852                f2fs_put_page(page, 1);
 853
 854                /* if ipage exists, blkaddr should be NEW_ADDR */
 855                f2fs_bug_on(F2FS_I_SB(inode), ipage);
 856                page = f2fs_get_lock_data_page(inode, index, true);
 857                if (IS_ERR(page))
 858                        return page;
 859        }
 860got_it:
 861        if (new_i_size && i_size_read(inode) <
 862                                ((loff_t)(index + 1) << PAGE_SHIFT))
 863                f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 864        return page;
 865}
 866
 867static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 868{
 869        struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 870        struct f2fs_summary sum;
 871        struct node_info ni;
 872        pgoff_t fofs;
 873        blkcnt_t count = 1;
 874        int err;
 875
 876        if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 877                return -EPERM;
 878
 879        dn->data_blkaddr = datablock_addr(dn->inode,
 880                                dn->node_page, dn->ofs_in_node);
 881        if (dn->data_blkaddr == NEW_ADDR)
 882                goto alloc;
 883
 884        if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 885                return err;
 886
 887alloc:
 888        f2fs_get_node_info(sbi, dn->nid, &ni);
 889        set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 890
 891        f2fs_allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 892                                        &sum, seg_type, NULL, false);
 893        f2fs_set_data_blkaddr(dn);
 894
 895        /* update i_size */
 896        fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 897                                                        dn->ofs_in_node;
 898        if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 899                f2fs_i_size_write(dn->inode,
 900                                ((loff_t)(fofs + 1) << PAGE_SHIFT));
 901        return 0;
 902}
 903
 904int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 905{
 906        struct inode *inode = file_inode(iocb->ki_filp);
 907        struct f2fs_map_blocks map;
 908        int flag;
 909        int err = 0;
 910        bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 911
 912        /* convert inline data for Direct I/O*/
 913        if (direct_io) {
 914                err = f2fs_convert_inline_inode(inode);
 915                if (err)
 916                        return err;
 917        }
 918
 919        if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 920                return 0;
 921
 922        map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 923        map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 924        if (map.m_len > map.m_lblk)
 925                map.m_len -= map.m_lblk;
 926        else
 927                map.m_len = 0;
 928
 929        map.m_next_pgofs = NULL;
 930        map.m_next_extent = NULL;
 931        map.m_seg_type = NO_CHECK_TYPE;
 932
 933        if (direct_io) {
 934                map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 935                flag = f2fs_force_buffered_io(inode, WRITE) ?
 936                                        F2FS_GET_BLOCK_PRE_AIO :
 937                                        F2FS_GET_BLOCK_PRE_DIO;
 938                goto map_blocks;
 939        }
 940        if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 941                err = f2fs_convert_inline_inode(inode);
 942                if (err)
 943                        return err;
 944        }
 945        if (f2fs_has_inline_data(inode))
 946                return err;
 947
 948        flag = F2FS_GET_BLOCK_PRE_AIO;
 949
 950map_blocks:
 951        err = f2fs_map_blocks(inode, &map, 1, flag);
 952        if (map.m_len > 0 && err == -ENOSPC) {
 953                if (!direct_io)
 954                        set_inode_flag(inode, FI_NO_PREALLOC);
 955                err = 0;
 956        }
 957        return err;
 958}
 959
 960static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 961{
 962        if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 963                if (lock)
 964                        down_read(&sbi->node_change);
 965                else
 966                        up_read(&sbi->node_change);
 967        } else {
 968                if (lock)
 969                        f2fs_lock_op(sbi);
 970                else
 971                        f2fs_unlock_op(sbi);
 972        }
 973}
 974
 975/*
 976 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 977 * f2fs_map_blocks structure.
 978 * If original data blocks are allocated, then give them to blockdev.
 979 * Otherwise,
 980 *     a. preallocate requested block addresses
 981 *     b. do not use extent cache for better performance
 982 *     c. give the block addresses to blockdev
 983 */
 984int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 985                                                int create, int flag)
 986{
 987        unsigned int maxblocks = map->m_len;
 988        struct dnode_of_data dn;
 989        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 990        int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 991        pgoff_t pgofs, end_offset, end;
 992        int err = 0, ofs = 1;
 993        unsigned int ofs_in_node, last_ofs_in_node;
 994        blkcnt_t prealloc;
 995        struct extent_info ei = {0,0,0};
 996        block_t blkaddr;
 997        unsigned int start_pgofs;
 998
 999        if (!maxblocks)
1000                return 0;
1001
1002        map->m_len = 0;
1003        map->m_flags = 0;
1004
1005        /* it only supports block size == page size */
1006        pgofs = (pgoff_t)map->m_lblk;
1007        end = pgofs + maxblocks;
1008
1009        if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1010                map->m_pblk = ei.blk + pgofs - ei.fofs;
1011                map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1012                map->m_flags = F2FS_MAP_MAPPED;
1013                if (map->m_next_extent)
1014                        *map->m_next_extent = pgofs + map->m_len;
1015                goto out;
1016        }
1017
1018next_dnode:
1019        if (create)
1020                __do_map_lock(sbi, flag, true);
1021
1022        /* When reading holes, we need its node page */
1023        set_new_dnode(&dn, inode, NULL, NULL, 0);
1024        err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1025        if (err) {
1026                if (flag == F2FS_GET_BLOCK_BMAP)
1027                        map->m_pblk = 0;
1028                if (err == -ENOENT) {
1029                        err = 0;
1030                        if (map->m_next_pgofs)
1031                                *map->m_next_pgofs =
1032                                        f2fs_get_next_page_offset(&dn, pgofs);
1033                        if (map->m_next_extent)
1034                                *map->m_next_extent =
1035                                        f2fs_get_next_page_offset(&dn, pgofs);
1036                }
1037                goto unlock_out;
1038        }
1039
1040        start_pgofs = pgofs;
1041        prealloc = 0;
1042        last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1043        end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1044
1045next_block:
1046        blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1047
1048        if (!is_valid_blkaddr(blkaddr)) {
1049                if (create) {
1050                        if (unlikely(f2fs_cp_error(sbi))) {
1051                                err = -EIO;
1052                                goto sync_out;
1053                        }
1054                        if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1055                                if (blkaddr == NULL_ADDR) {
1056                                        prealloc++;
1057                                        last_ofs_in_node = dn.ofs_in_node;
1058                                }
1059                        } else {
1060                                err = __allocate_data_block(&dn,
1061                                                        map->m_seg_type);
1062                                if (!err)
1063                                        set_inode_flag(inode, FI_APPEND_WRITE);
1064                        }
1065                        if (err)
1066                                goto sync_out;
1067                        map->m_flags |= F2FS_MAP_NEW;
1068                        blkaddr = dn.data_blkaddr;
1069                } else {
1070                        if (flag == F2FS_GET_BLOCK_BMAP) {
1071                                map->m_pblk = 0;
1072                                goto sync_out;
1073                        }
1074                        if (flag == F2FS_GET_BLOCK_PRECACHE)
1075                                goto sync_out;
1076                        if (flag == F2FS_GET_BLOCK_FIEMAP &&
1077                                                blkaddr == NULL_ADDR) {
1078                                if (map->m_next_pgofs)
1079                                        *map->m_next_pgofs = pgofs + 1;
1080                                goto sync_out;
1081                        }
1082                        if (flag != F2FS_GET_BLOCK_FIEMAP) {
1083                                /* for defragment case */
1084                                if (map->m_next_pgofs)
1085                                        *map->m_next_pgofs = pgofs + 1;
1086                                goto sync_out;
1087                        }
1088                }
1089        }
1090
1091        if (flag == F2FS_GET_BLOCK_PRE_AIO)
1092                goto skip;
1093
1094        if (map->m_len == 0) {
1095                /* preallocated unwritten block should be mapped for fiemap. */
1096                if (blkaddr == NEW_ADDR)
1097                        map->m_flags |= F2FS_MAP_UNWRITTEN;
1098                map->m_flags |= F2FS_MAP_MAPPED;
1099
1100                map->m_pblk = blkaddr;
1101                map->m_len = 1;
1102        } else if ((map->m_pblk != NEW_ADDR &&
1103                        blkaddr == (map->m_pblk + ofs)) ||
1104                        (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1105                        flag == F2FS_GET_BLOCK_PRE_DIO) {
1106                ofs++;
1107                map->m_len++;
1108        } else {
1109                goto sync_out;
1110        }
1111
1112skip:
1113        dn.ofs_in_node++;
1114        pgofs++;
1115
1116        /* preallocate blocks in batch for one dnode page */
1117        if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1118                        (pgofs == end || dn.ofs_in_node == end_offset)) {
1119
1120                dn.ofs_in_node = ofs_in_node;
1121                err = f2fs_reserve_new_blocks(&dn, prealloc);
1122                if (err)
1123                        goto sync_out;
1124
1125                map->m_len += dn.ofs_in_node - ofs_in_node;
1126                if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1127                        err = -ENOSPC;
1128                        goto sync_out;
1129                }
1130                dn.ofs_in_node = end_offset;
1131        }
1132
1133        if (pgofs >= end)
1134                goto sync_out;
1135        else if (dn.ofs_in_node < end_offset)
1136                goto next_block;
1137
1138        if (flag == F2FS_GET_BLOCK_PRECACHE) {
1139                if (map->m_flags & F2FS_MAP_MAPPED) {
1140                        unsigned int ofs = start_pgofs - map->m_lblk;
1141
1142                        f2fs_update_extent_cache_range(&dn,
1143                                start_pgofs, map->m_pblk + ofs,
1144                                map->m_len - ofs);
1145                }
1146        }
1147
1148        f2fs_put_dnode(&dn);
1149
1150        if (create) {
1151                __do_map_lock(sbi, flag, false);
1152                f2fs_balance_fs(sbi, dn.node_changed);
1153        }
1154        goto next_dnode;
1155
1156sync_out:
1157        if (flag == F2FS_GET_BLOCK_PRECACHE) {
1158                if (map->m_flags & F2FS_MAP_MAPPED) {
1159                        unsigned int ofs = start_pgofs - map->m_lblk;
1160
1161                        f2fs_update_extent_cache_range(&dn,
1162                                start_pgofs, map->m_pblk + ofs,
1163                                map->m_len - ofs);
1164                }
1165                if (map->m_next_extent)
1166                        *map->m_next_extent = pgofs + 1;
1167        }
1168        f2fs_put_dnode(&dn);
1169unlock_out:
1170        if (create) {
1171                __do_map_lock(sbi, flag, false);
1172                f2fs_balance_fs(sbi, dn.node_changed);
1173        }
1174out:
1175        trace_f2fs_map_blocks(inode, map, err);
1176        return err;
1177}
1178
1179bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1180{
1181        struct f2fs_map_blocks map;
1182        block_t last_lblk;
1183        int err;
1184
1185        if (pos + len > i_size_read(inode))
1186                return false;
1187
1188        map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1189        map.m_next_pgofs = NULL;
1190        map.m_next_extent = NULL;
1191        map.m_seg_type = NO_CHECK_TYPE;
1192        last_lblk = F2FS_BLK_ALIGN(pos + len);
1193
1194        while (map.m_lblk < last_lblk) {
1195                map.m_len = last_lblk - map.m_lblk;
1196                err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1197                if (err || map.m_len == 0)
1198                        return false;
1199                map.m_lblk += map.m_len;
1200        }
1201        return true;
1202}
1203
1204static int __get_data_block(struct inode *inode, sector_t iblock,
1205                        struct buffer_head *bh, int create, int flag,
1206                        pgoff_t *next_pgofs, int seg_type)
1207{
1208        struct f2fs_map_blocks map;
1209        int err;
1210
1211        map.m_lblk = iblock;
1212        map.m_len = bh->b_size >> inode->i_blkbits;
1213        map.m_next_pgofs = next_pgofs;
1214        map.m_next_extent = NULL;
1215        map.m_seg_type = seg_type;
1216
1217        err = f2fs_map_blocks(inode, &map, create, flag);
1218        if (!err) {
1219                map_bh(bh, inode->i_sb, map.m_pblk);
1220                bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1221                bh->b_size = (u64)map.m_len << inode->i_blkbits;
1222        }
1223        return err;
1224}
1225
1226static int get_data_block(struct inode *inode, sector_t iblock,
1227                        struct buffer_head *bh_result, int create, int flag,
1228                        pgoff_t *next_pgofs)
1229{
1230        return __get_data_block(inode, iblock, bh_result, create,
1231                                                        flag, next_pgofs,
1232                                                        NO_CHECK_TYPE);
1233}
1234
1235static int get_data_block_dio(struct inode *inode, sector_t iblock,
1236                        struct buffer_head *bh_result, int create)
1237{
1238        return __get_data_block(inode, iblock, bh_result, create,
1239                                                F2FS_GET_BLOCK_DEFAULT, NULL,
1240                                                f2fs_rw_hint_to_seg_type(
1241                                                        inode->i_write_hint));
1242}
1243
1244static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1245                        struct buffer_head *bh_result, int create)
1246{
1247        /* Block number less than F2FS MAX BLOCKS */
1248        if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1249                return -EFBIG;
1250
1251        return __get_data_block(inode, iblock, bh_result, create,
1252                                                F2FS_GET_BLOCK_BMAP, NULL,
1253                                                NO_CHECK_TYPE);
1254}
1255
1256static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1257{
1258        return (offset >> inode->i_blkbits);
1259}
1260
1261static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1262{
1263        return (blk << inode->i_blkbits);
1264}
1265
1266static int f2fs_xattr_fiemap(struct inode *inode,
1267                                struct fiemap_extent_info *fieinfo)
1268{
1269        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1270        struct page *page;
1271        struct node_info ni;
1272        __u64 phys = 0, len;
1273        __u32 flags;
1274        nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1275        int err = 0;
1276
1277        if (f2fs_has_inline_xattr(inode)) {
1278                int offset;
1279
1280                page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1281                                                inode->i_ino, false);
1282                if (!page)
1283                        return -ENOMEM;
1284
1285                f2fs_get_node_info(sbi, inode->i_ino, &ni);
1286
1287                phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1288                offset = offsetof(struct f2fs_inode, i_addr) +
1289                                        sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1290                                        get_inline_xattr_addrs(inode));
1291
1292                phys += offset;
1293                len = inline_xattr_size(inode);
1294
1295                f2fs_put_page(page, 1);
1296
1297                flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1298
1299                if (!xnid)
1300                        flags |= FIEMAP_EXTENT_LAST;
1301
1302                err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1303                if (err || err == 1)
1304                        return err;
1305        }
1306
1307        if (xnid) {
1308                page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1309                if (!page)
1310                        return -ENOMEM;
1311
1312                f2fs_get_node_info(sbi, xnid, &ni);
1313
1314                phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1315                len = inode->i_sb->s_blocksize;
1316
1317                f2fs_put_page(page, 1);
1318
1319                flags = FIEMAP_EXTENT_LAST;
1320        }
1321
1322        if (phys)
1323                err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1324
1325        return (err < 0 ? err : 0);
1326}
1327
1328int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1329                u64 start, u64 len)
1330{
1331        struct buffer_head map_bh;
1332        sector_t start_blk, last_blk;
1333        pgoff_t next_pgofs;
1334        u64 logical = 0, phys = 0, size = 0;
1335        u32 flags = 0;
1336        int ret = 0;
1337
1338        if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1339                ret = f2fs_precache_extents(inode);
1340                if (ret)
1341                        return ret;
1342        }
1343
1344        ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1345        if (ret)
1346                return ret;
1347
1348        inode_lock(inode);
1349
1350        if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1351                ret = f2fs_xattr_fiemap(inode, fieinfo);
1352                goto out;
1353        }
1354
1355        if (f2fs_has_inline_data(inode)) {
1356                ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1357                if (ret != -EAGAIN)
1358                        goto out;
1359        }
1360
1361        if (logical_to_blk(inode, len) == 0)
1362                len = blk_to_logical(inode, 1);
1363
1364        start_blk = logical_to_blk(inode, start);
1365        last_blk = logical_to_blk(inode, start + len - 1);
1366
1367next:
1368        memset(&map_bh, 0, sizeof(struct buffer_head));
1369        map_bh.b_size = len;
1370
1371        ret = get_data_block(inode, start_blk, &map_bh, 0,
1372                                        F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1373        if (ret)
1374                goto out;
1375
1376        /* HOLE */
1377        if (!buffer_mapped(&map_bh)) {
1378                start_blk = next_pgofs;
1379
1380                if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1381                                        F2FS_I_SB(inode)->max_file_blocks))
1382                        goto prep_next;
1383
1384                flags |= FIEMAP_EXTENT_LAST;
1385        }
1386
1387        if (size) {
1388                if (f2fs_encrypted_inode(inode))
1389                        flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1390
1391                ret = fiemap_fill_next_extent(fieinfo, logical,
1392                                phys, size, flags);
1393        }
1394
1395        if (start_blk > last_blk || ret)
1396                goto out;
1397
1398        logical = blk_to_logical(inode, start_blk);
1399        phys = blk_to_logical(inode, map_bh.b_blocknr);
1400        size = map_bh.b_size;
1401        flags = 0;
1402        if (buffer_unwritten(&map_bh))
1403                flags = FIEMAP_EXTENT_UNWRITTEN;
1404
1405        start_blk += logical_to_blk(inode, size);
1406
1407prep_next:
1408        cond_resched();
1409        if (fatal_signal_pending(current))
1410                ret = -EINTR;
1411        else
1412                goto next;
1413out:
1414        if (ret == 1)
1415                ret = 0;
1416
1417        inode_unlock(inode);
1418        return ret;
1419}
1420
1421/*
1422 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1423 * Major change was from block_size == page_size in f2fs by default.
1424 */
1425static int f2fs_mpage_readpages(struct address_space *mapping,
1426                        struct list_head *pages, struct page *page,
1427                        unsigned nr_pages)
1428{
1429        struct bio *bio = NULL;
1430        sector_t last_block_in_bio = 0;
1431        struct inode *inode = mapping->host;
1432        const unsigned blkbits = inode->i_blkbits;
1433        const unsigned blocksize = 1 << blkbits;
1434        sector_t block_in_file;
1435        sector_t last_block;
1436        sector_t last_block_in_file;
1437        sector_t block_nr;
1438        struct f2fs_map_blocks map;
1439
1440        map.m_pblk = 0;
1441        map.m_lblk = 0;
1442        map.m_len = 0;
1443        map.m_flags = 0;
1444        map.m_next_pgofs = NULL;
1445        map.m_next_extent = NULL;
1446        map.m_seg_type = NO_CHECK_TYPE;
1447
1448        for (; nr_pages; nr_pages--) {
1449                if (pages) {
1450                        page = list_last_entry(pages, struct page, lru);
1451
1452                        prefetchw(&page->flags);
1453                        list_del(&page->lru);
1454                        if (add_to_page_cache_lru(page, mapping,
1455                                                  page->index,
1456                                                  readahead_gfp_mask(mapping)))
1457                                goto next_page;
1458                }
1459
1460                block_in_file = (sector_t)page->index;
1461                last_block = block_in_file + nr_pages;
1462                last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1463                                                                blkbits;
1464                if (last_block > last_block_in_file)
1465                        last_block = last_block_in_file;
1466
1467                /*
1468                 * Map blocks using the previous result first.
1469                 */
1470                if ((map.m_flags & F2FS_MAP_MAPPED) &&
1471                                block_in_file > map.m_lblk &&
1472                                block_in_file < (map.m_lblk + map.m_len))
1473                        goto got_it;
1474
1475                /*
1476                 * Then do more f2fs_map_blocks() calls until we are
1477                 * done with this page.
1478                 */
1479                map.m_flags = 0;
1480
1481                if (block_in_file < last_block) {
1482                        map.m_lblk = block_in_file;
1483                        map.m_len = last_block - block_in_file;
1484
1485                        if (f2fs_map_blocks(inode, &map, 0,
1486                                                F2FS_GET_BLOCK_DEFAULT))
1487                                goto set_error_page;
1488                }
1489got_it:
1490                if ((map.m_flags & F2FS_MAP_MAPPED)) {
1491                        block_nr = map.m_pblk + block_in_file - map.m_lblk;
1492                        SetPageMappedToDisk(page);
1493
1494                        if (!PageUptodate(page) && !cleancache_get_page(page)) {
1495                                SetPageUptodate(page);
1496                                goto confused;
1497                        }
1498                } else {
1499                        zero_user_segment(page, 0, PAGE_SIZE);
1500                        if (!PageUptodate(page))
1501                                SetPageUptodate(page);
1502                        unlock_page(page);
1503                        goto next_page;
1504                }
1505
1506                /*
1507                 * This page will go to BIO.  Do we need to send this
1508                 * BIO off first?
1509                 */
1510                if (bio && (last_block_in_bio != block_nr - 1 ||
1511                        !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1512submit_and_realloc:
1513                        __submit_bio(F2FS_I_SB(inode), bio, DATA);
1514                        bio = NULL;
1515                }
1516                if (bio == NULL) {
1517                        bio = f2fs_grab_read_bio(inode, block_nr, nr_pages);
1518                        if (IS_ERR(bio)) {
1519                                bio = NULL;
1520                                goto set_error_page;
1521                        }
1522                }
1523
1524                if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1525                        goto submit_and_realloc;
1526
1527                last_block_in_bio = block_nr;
1528                goto next_page;
1529set_error_page:
1530                SetPageError(page);
1531                zero_user_segment(page, 0, PAGE_SIZE);
1532                unlock_page(page);
1533                goto next_page;
1534confused:
1535                if (bio) {
1536                        __submit_bio(F2FS_I_SB(inode), bio, DATA);
1537                        bio = NULL;
1538                }
1539                unlock_page(page);
1540next_page:
1541                if (pages)
1542                        put_page(page);
1543        }
1544        BUG_ON(pages && !list_empty(pages));
1545        if (bio)
1546                __submit_bio(F2FS_I_SB(inode), bio, DATA);
1547        return 0;
1548}
1549
1550static int f2fs_read_data_page(struct file *file, struct page *page)
1551{
1552        struct inode *inode = page->mapping->host;
1553        int ret = -EAGAIN;
1554
1555        trace_f2fs_readpage(page, DATA);
1556
1557        /* If the file has inline data, try to read it directly */
1558        if (f2fs_has_inline_data(inode))
1559                ret = f2fs_read_inline_data(inode, page);
1560        if (ret == -EAGAIN)
1561                ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1562        return ret;
1563}
1564
1565static int f2fs_read_data_pages(struct file *file,
1566                        struct address_space *mapping,
1567                        struct list_head *pages, unsigned nr_pages)
1568{
1569        struct inode *inode = mapping->host;
1570        struct page *page = list_last_entry(pages, struct page, lru);
1571
1572        trace_f2fs_readpages(inode, page, nr_pages);
1573
1574        /* If the file has inline data, skip readpages */
1575        if (f2fs_has_inline_data(inode))
1576                return 0;
1577
1578        return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1579}
1580
1581static int encrypt_one_page(struct f2fs_io_info *fio)
1582{
1583        struct inode *inode = fio->page->mapping->host;
1584        gfp_t gfp_flags = GFP_NOFS;
1585
1586        if (!f2fs_encrypted_file(inode))
1587                return 0;
1588
1589        /* wait for GCed page writeback via META_MAPPING */
1590        f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1591
1592retry_encrypt:
1593        fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1594                        PAGE_SIZE, 0, fio->page->index, gfp_flags);
1595        if (!IS_ERR(fio->encrypted_page))
1596                return 0;
1597
1598        /* flush pending IOs and wait for a while in the ENOMEM case */
1599        if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1600                f2fs_flush_merged_writes(fio->sbi);
1601                congestion_wait(BLK_RW_ASYNC, HZ/50);
1602                gfp_flags |= __GFP_NOFAIL;
1603                goto retry_encrypt;
1604        }
1605        return PTR_ERR(fio->encrypted_page);
1606}
1607
1608static inline bool check_inplace_update_policy(struct inode *inode,
1609                                struct f2fs_io_info *fio)
1610{
1611        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1612        unsigned int policy = SM_I(sbi)->ipu_policy;
1613
1614        if (policy & (0x1 << F2FS_IPU_FORCE))
1615                return true;
1616        if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1617                return true;
1618        if (policy & (0x1 << F2FS_IPU_UTIL) &&
1619                        utilization(sbi) > SM_I(sbi)->min_ipu_util)
1620                return true;
1621        if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1622                        utilization(sbi) > SM_I(sbi)->min_ipu_util)
1623                return true;
1624
1625        /*
1626         * IPU for rewrite async pages
1627         */
1628        if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1629                        fio && fio->op == REQ_OP_WRITE &&
1630                        !(fio->op_flags & REQ_SYNC) &&
1631                        !f2fs_encrypted_inode(inode))
1632                return true;
1633
1634        /* this is only set during fdatasync */
1635        if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1636                        is_inode_flag_set(inode, FI_NEED_IPU))
1637                return true;
1638
1639        return false;
1640}
1641
1642bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1643{
1644        if (f2fs_is_pinned_file(inode))
1645                return true;
1646
1647        /* if this is cold file, we should overwrite to avoid fragmentation */
1648        if (file_is_cold(inode))
1649                return true;
1650
1651        return check_inplace_update_policy(inode, fio);
1652}
1653
1654bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1655{
1656        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1657
1658        if (test_opt(sbi, LFS))
1659                return true;
1660        if (S_ISDIR(inode->i_mode))
1661                return true;
1662        if (f2fs_is_atomic_file(inode))
1663                return true;
1664        if (fio) {
1665                if (is_cold_data(fio->page))
1666                        return true;
1667                if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1668                        return true;
1669        }
1670        return false;
1671}
1672
1673static inline bool need_inplace_update(struct f2fs_io_info *fio)
1674{
1675        struct inode *inode = fio->page->mapping->host;
1676
1677        if (f2fs_should_update_outplace(inode, fio))
1678                return false;
1679
1680        return f2fs_should_update_inplace(inode, fio);
1681}
1682
1683int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1684{
1685        struct page *page = fio->page;
1686        struct inode *inode = page->mapping->host;
1687        struct dnode_of_data dn;
1688        struct extent_info ei = {0,0,0};
1689        bool ipu_force = false;
1690        int err = 0;
1691
1692        set_new_dnode(&dn, inode, NULL, NULL, 0);
1693        if (need_inplace_update(fio) &&
1694                        f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1695                fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1696
1697                if (is_valid_blkaddr(fio->old_blkaddr)) {
1698                        ipu_force = true;
1699                        fio->need_lock = LOCK_DONE;
1700                        goto got_it;
1701                }
1702        }
1703
1704        /* Deadlock due to between page->lock and f2fs_lock_op */
1705        if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1706                return -EAGAIN;
1707
1708        err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1709        if (err)
1710                goto out;
1711
1712        fio->old_blkaddr = dn.data_blkaddr;
1713
1714        /* This page is already truncated */
1715        if (fio->old_blkaddr == NULL_ADDR) {
1716                ClearPageUptodate(page);
1717                goto out_writepage;
1718        }
1719got_it:
1720        /*
1721         * If current allocation needs SSR,
1722         * it had better in-place writes for updated data.
1723         */
1724        if (ipu_force || (is_valid_blkaddr(fio->old_blkaddr) &&
1725                                        need_inplace_update(fio))) {
1726                err = encrypt_one_page(fio);
1727                if (err)
1728                        goto out_writepage;
1729
1730                set_page_writeback(page);
1731                ClearPageError(page);
1732                f2fs_put_dnode(&dn);
1733                if (fio->need_lock == LOCK_REQ)
1734                        f2fs_unlock_op(fio->sbi);
1735                err = f2fs_inplace_write_data(fio);
1736                trace_f2fs_do_write_data_page(fio->page, IPU);
1737                set_inode_flag(inode, FI_UPDATE_WRITE);
1738                return err;
1739        }
1740
1741        if (fio->need_lock == LOCK_RETRY) {
1742                if (!f2fs_trylock_op(fio->sbi)) {
1743                        err = -EAGAIN;
1744                        goto out_writepage;
1745                }
1746                fio->need_lock = LOCK_REQ;
1747        }
1748
1749        err = encrypt_one_page(fio);
1750        if (err)
1751                goto out_writepage;
1752
1753        set_page_writeback(page);
1754        ClearPageError(page);
1755
1756        /* LFS mode write path */
1757        f2fs_outplace_write_data(&dn, fio);
1758        trace_f2fs_do_write_data_page(page, OPU);
1759        set_inode_flag(inode, FI_APPEND_WRITE);
1760        if (page->index == 0)
1761                set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1762out_writepage:
1763        f2fs_put_dnode(&dn);
1764out:
1765        if (fio->need_lock == LOCK_REQ)
1766                f2fs_unlock_op(fio->sbi);
1767        return err;
1768}
1769
1770static int __write_data_page(struct page *page, bool *submitted,
1771                                struct writeback_control *wbc,
1772                                enum iostat_type io_type)
1773{
1774        struct inode *inode = page->mapping->host;
1775        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1776        loff_t i_size = i_size_read(inode);
1777        const pgoff_t end_index = ((unsigned long long) i_size)
1778                                                        >> PAGE_SHIFT;
1779        loff_t psize = (page->index + 1) << PAGE_SHIFT;
1780        unsigned offset = 0;
1781        bool need_balance_fs = false;
1782        int err = 0;
1783        struct f2fs_io_info fio = {
1784                .sbi = sbi,
1785                .ino = inode->i_ino,
1786                .type = DATA,
1787                .op = REQ_OP_WRITE,
1788                .op_flags = wbc_to_write_flags(wbc),
1789                .old_blkaddr = NULL_ADDR,
1790                .page = page,
1791                .encrypted_page = NULL,
1792                .submitted = false,
1793                .need_lock = LOCK_RETRY,
1794                .io_type = io_type,
1795                .io_wbc = wbc,
1796        };
1797
1798        trace_f2fs_writepage(page, DATA);
1799
1800        /* we should bypass data pages to proceed the kworkder jobs */
1801        if (unlikely(f2fs_cp_error(sbi))) {
1802                mapping_set_error(page->mapping, -EIO);
1803                /*
1804                 * don't drop any dirty dentry pages for keeping lastest
1805                 * directory structure.
1806                 */
1807                if (S_ISDIR(inode->i_mode))
1808                        goto redirty_out;
1809                goto out;
1810        }
1811
1812        if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1813                goto redirty_out;
1814
1815        if (page->index < end_index)
1816                goto write;
1817
1818        /*
1819         * If the offset is out-of-range of file size,
1820         * this page does not have to be written to disk.
1821         */
1822        offset = i_size & (PAGE_SIZE - 1);
1823        if ((page->index >= end_index + 1) || !offset)
1824                goto out;
1825
1826        zero_user_segment(page, offset, PAGE_SIZE);
1827write:
1828        if (f2fs_is_drop_cache(inode))
1829                goto out;
1830        /* we should not write 0'th page having journal header */
1831        if (f2fs_is_volatile_file(inode) && (!page->index ||
1832                        (!wbc->for_reclaim &&
1833                        f2fs_available_free_memory(sbi, BASE_CHECK))))
1834                goto redirty_out;
1835
1836        /* Dentry blocks are controlled by checkpoint */
1837        if (S_ISDIR(inode->i_mode)) {
1838                fio.need_lock = LOCK_DONE;
1839                err = f2fs_do_write_data_page(&fio);
1840                goto done;
1841        }
1842
1843        if (!wbc->for_reclaim)
1844                need_balance_fs = true;
1845        else if (has_not_enough_free_secs(sbi, 0, 0))
1846                goto redirty_out;
1847        else
1848                set_inode_flag(inode, FI_HOT_DATA);
1849
1850        err = -EAGAIN;
1851        if (f2fs_has_inline_data(inode)) {
1852                err = f2fs_write_inline_data(inode, page);
1853                if (!err)
1854                        goto out;
1855        }
1856
1857        if (err == -EAGAIN) {
1858                err = f2fs_do_write_data_page(&fio);
1859                if (err == -EAGAIN) {
1860                        fio.need_lock = LOCK_REQ;
1861                        err = f2fs_do_write_data_page(&fio);
1862                }
1863        }
1864
1865        if (err) {
1866                file_set_keep_isize(inode);
1867        } else {
1868                down_write(&F2FS_I(inode)->i_sem);
1869                if (F2FS_I(inode)->last_disk_size < psize)
1870                        F2FS_I(inode)->last_disk_size = psize;
1871                up_write(&F2FS_I(inode)->i_sem);
1872        }
1873
1874done:
1875        if (err && err != -ENOENT)
1876                goto redirty_out;
1877
1878out:
1879        inode_dec_dirty_pages(inode);
1880        if (err)
1881                ClearPageUptodate(page);
1882
1883        if (wbc->for_reclaim) {
1884                f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1885                clear_inode_flag(inode, FI_HOT_DATA);
1886                f2fs_remove_dirty_inode(inode);
1887                submitted = NULL;
1888        }
1889
1890        unlock_page(page);
1891        if (!S_ISDIR(inode->i_mode))
1892                f2fs_balance_fs(sbi, need_balance_fs);
1893
1894        if (unlikely(f2fs_cp_error(sbi))) {
1895                f2fs_submit_merged_write(sbi, DATA);
1896                submitted = NULL;
1897        }
1898
1899        if (submitted)
1900                *submitted = fio.submitted;
1901
1902        return 0;
1903
1904redirty_out:
1905        redirty_page_for_writepage(wbc, page);
1906        /*
1907         * pageout() in MM traslates EAGAIN, so calls handle_write_error()
1908         * -> mapping_set_error() -> set_bit(AS_EIO, ...).
1909         * file_write_and_wait_range() will see EIO error, which is critical
1910         * to return value of fsync() followed by atomic_write failure to user.
1911         */
1912        if (!err || wbc->for_reclaim)
1913                return AOP_WRITEPAGE_ACTIVATE;
1914        unlock_page(page);
1915        return err;
1916}
1917
1918static int f2fs_write_data_page(struct page *page,
1919                                        struct writeback_control *wbc)
1920{
1921        return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1922}
1923
1924/*
1925 * This function was copied from write_cche_pages from mm/page-writeback.c.
1926 * The major change is making write step of cold data page separately from
1927 * warm/hot data page.
1928 */
1929static int f2fs_write_cache_pages(struct address_space *mapping,
1930                                        struct writeback_control *wbc,
1931                                        enum iostat_type io_type)
1932{
1933        int ret = 0;
1934        int done = 0;
1935        struct pagevec pvec;
1936        struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1937        int nr_pages;
1938        pgoff_t uninitialized_var(writeback_index);
1939        pgoff_t index;
1940        pgoff_t end;            /* Inclusive */
1941        pgoff_t done_index;
1942        pgoff_t last_idx = ULONG_MAX;
1943        int cycled;
1944        int range_whole = 0;
1945        int tag;
1946
1947        pagevec_init(&pvec);
1948
1949        if (get_dirty_pages(mapping->host) <=
1950                                SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1951                set_inode_flag(mapping->host, FI_HOT_DATA);
1952        else
1953                clear_inode_flag(mapping->host, FI_HOT_DATA);
1954
1955        if (wbc->range_cyclic) {
1956                writeback_index = mapping->writeback_index; /* prev offset */
1957                index = writeback_index;
1958                if (index == 0)
1959                        cycled = 1;
1960                else
1961                        cycled = 0;
1962                end = -1;
1963        } else {
1964                index = wbc->range_start >> PAGE_SHIFT;
1965                end = wbc->range_end >> PAGE_SHIFT;
1966                if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1967                        range_whole = 1;
1968                cycled = 1; /* ignore range_cyclic tests */
1969        }
1970        if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1971                tag = PAGECACHE_TAG_TOWRITE;
1972        else
1973                tag = PAGECACHE_TAG_DIRTY;
1974retry:
1975        if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1976                tag_pages_for_writeback(mapping, index, end);
1977        done_index = index;
1978        while (!done && (index <= end)) {
1979                int i;
1980
1981                nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
1982                                tag);
1983                if (nr_pages == 0)
1984                        break;
1985
1986                for (i = 0; i < nr_pages; i++) {
1987                        struct page *page = pvec.pages[i];
1988                        bool submitted = false;
1989
1990                        /* give a priority to WB_SYNC threads */
1991                        if (atomic_read(&sbi->wb_sync_req[DATA]) &&
1992                                        wbc->sync_mode == WB_SYNC_NONE) {
1993                                done = 1;
1994                                break;
1995                        }
1996
1997                        done_index = page->index;
1998retry_write:
1999                        lock_page(page);
2000
2001                        if (unlikely(page->mapping != mapping)) {
2002continue_unlock:
2003                                unlock_page(page);
2004                                continue;
2005                        }
2006
2007                        if (!PageDirty(page)) {
2008                                /* someone wrote it for us */
2009                                goto continue_unlock;
2010                        }
2011
2012                        if (PageWriteback(page)) {
2013                                if (wbc->sync_mode != WB_SYNC_NONE)
2014                                        f2fs_wait_on_page_writeback(page,
2015                                                                DATA, true);
2016                                else
2017                                        goto continue_unlock;
2018                        }
2019
2020                        BUG_ON(PageWriteback(page));
2021                        if (!clear_page_dirty_for_io(page))
2022                                goto continue_unlock;
2023
2024                        ret = __write_data_page(page, &submitted, wbc, io_type);
2025                        if (unlikely(ret)) {
2026                                /*
2027                                 * keep nr_to_write, since vfs uses this to
2028                                 * get # of written pages.
2029                                 */
2030                                if (ret == AOP_WRITEPAGE_ACTIVATE) {
2031                                        unlock_page(page);
2032                                        ret = 0;
2033                                        continue;
2034                                } else if (ret == -EAGAIN) {
2035                                        ret = 0;
2036                                        if (wbc->sync_mode == WB_SYNC_ALL) {
2037                                                cond_resched();
2038                                                congestion_wait(BLK_RW_ASYNC,
2039                                                                        HZ/50);
2040                                                goto retry_write;
2041                                        }
2042                                        continue;
2043                                }
2044                                done_index = page->index + 1;
2045                                done = 1;
2046                                break;
2047                        } else if (submitted) {
2048                                last_idx = page->index;
2049                        }
2050
2051                        if (--wbc->nr_to_write <= 0 &&
2052                                        wbc->sync_mode == WB_SYNC_NONE) {
2053                                done = 1;
2054                                break;
2055                        }
2056                }
2057                pagevec_release(&pvec);
2058                cond_resched();
2059        }
2060
2061        if (!cycled && !done) {
2062                cycled = 1;
2063                index = 0;
2064                end = writeback_index - 1;
2065                goto retry;
2066        }
2067        if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2068                mapping->writeback_index = done_index;
2069
2070        if (last_idx != ULONG_MAX)
2071                f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2072                                                0, last_idx, DATA);
2073
2074        return ret;
2075}
2076
2077static int __f2fs_write_data_pages(struct address_space *mapping,
2078                                                struct writeback_control *wbc,
2079                                                enum iostat_type io_type)
2080{
2081        struct inode *inode = mapping->host;
2082        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2083        struct blk_plug plug;
2084        int ret;
2085
2086        /* deal with chardevs and other special file */
2087        if (!mapping->a_ops->writepage)
2088                return 0;
2089
2090        /* skip writing if there is no dirty page in this inode */
2091        if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2092                return 0;
2093
2094        /* during POR, we don't need to trigger writepage at all. */
2095        if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2096                goto skip_write;
2097
2098        if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
2099                        get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2100                        f2fs_available_free_memory(sbi, DIRTY_DENTS))
2101                goto skip_write;
2102
2103        /* skip writing during file defragment */
2104        if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2105                goto skip_write;
2106
2107        trace_f2fs_writepages(mapping->host, wbc, DATA);
2108
2109        /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2110        if (wbc->sync_mode == WB_SYNC_ALL)
2111                atomic_inc(&sbi->wb_sync_req[DATA]);
2112        else if (atomic_read(&sbi->wb_sync_req[DATA]))
2113                goto skip_write;
2114
2115        blk_start_plug(&plug);
2116        ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2117        blk_finish_plug(&plug);
2118
2119        if (wbc->sync_mode == WB_SYNC_ALL)
2120                atomic_dec(&sbi->wb_sync_req[DATA]);
2121        /*
2122         * if some pages were truncated, we cannot guarantee its mapping->host
2123         * to detect pending bios.
2124         */
2125
2126        f2fs_remove_dirty_inode(inode);
2127        return ret;
2128
2129skip_write:
2130        wbc->pages_skipped += get_dirty_pages(inode);
2131        trace_f2fs_writepages(mapping->host, wbc, DATA);
2132        return 0;
2133}
2134
2135static int f2fs_write_data_pages(struct address_space *mapping,
2136                            struct writeback_control *wbc)
2137{
2138        struct inode *inode = mapping->host;
2139
2140        return __f2fs_write_data_pages(mapping, wbc,
2141                        F2FS_I(inode)->cp_task == current ?
2142                        FS_CP_DATA_IO : FS_DATA_IO);
2143}
2144
2145static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2146{
2147        struct inode *inode = mapping->host;
2148        loff_t i_size = i_size_read(inode);
2149
2150        if (to > i_size) {
2151                down_write(&F2FS_I(inode)->i_mmap_sem);
2152                truncate_pagecache(inode, i_size);
2153                f2fs_truncate_blocks(inode, i_size, true);
2154                up_write(&F2FS_I(inode)->i_mmap_sem);
2155        }
2156}
2157
2158static int prepare_write_begin(struct f2fs_sb_info *sbi,
2159                        struct page *page, loff_t pos, unsigned len,
2160                        block_t *blk_addr, bool *node_changed)
2161{
2162        struct inode *inode = page->mapping->host;
2163        pgoff_t index = page->index;
2164        struct dnode_of_data dn;
2165        struct page *ipage;
2166        bool locked = false;
2167        struct extent_info ei = {0,0,0};
2168        int err = 0;
2169
2170        /*
2171         * we already allocated all the blocks, so we don't need to get
2172         * the block addresses when there is no need to fill the page.
2173         */
2174        if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2175                        !is_inode_flag_set(inode, FI_NO_PREALLOC))
2176                return 0;
2177
2178        if (f2fs_has_inline_data(inode) ||
2179                        (pos & PAGE_MASK) >= i_size_read(inode)) {
2180                __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
2181                locked = true;
2182        }
2183restart:
2184        /* check inline_data */
2185        ipage = f2fs_get_node_page(sbi, inode->i_ino);
2186        if (IS_ERR(ipage)) {
2187                err = PTR_ERR(ipage);
2188                goto unlock_out;
2189        }
2190
2191        set_new_dnode(&dn, inode, ipage, ipage, 0);
2192
2193        if (f2fs_has_inline_data(inode)) {
2194                if (pos + len <= MAX_INLINE_DATA(inode)) {
2195                        f2fs_do_read_inline_data(page, ipage);
2196                        set_inode_flag(inode, FI_DATA_EXIST);
2197                        if (inode->i_nlink)
2198                                set_inline_node(ipage);
2199                } else {
2200                        err = f2fs_convert_inline_page(&dn, page);
2201                        if (err)
2202                                goto out;
2203                        if (dn.data_blkaddr == NULL_ADDR)
2204                                err = f2fs_get_block(&dn, index);
2205                }
2206        } else if (locked) {
2207                err = f2fs_get_block(&dn, index);
2208        } else {
2209                if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2210                        dn.data_blkaddr = ei.blk + index - ei.fofs;
2211                } else {
2212                        /* hole case */
2213                        err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2214                        if (err || dn.data_blkaddr == NULL_ADDR) {
2215                                f2fs_put_dnode(&dn);
2216                                __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2217                                                                true);
2218                                locked = true;
2219                                goto restart;
2220                        }
2221                }
2222        }
2223
2224        /* convert_inline_page can make node_changed */
2225        *blk_addr = dn.data_blkaddr;
2226        *node_changed = dn.node_changed;
2227out:
2228        f2fs_put_dnode(&dn);
2229unlock_out:
2230        if (locked)
2231                __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
2232        return err;
2233}
2234
2235static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2236                loff_t pos, unsigned len, unsigned flags,
2237                struct page **pagep, void **fsdata)
2238{
2239        struct inode *inode = mapping->host;
2240        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2241        struct page *page = NULL;
2242        pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2243        bool need_balance = false, drop_atomic = false;
2244        block_t blkaddr = NULL_ADDR;
2245        int err = 0;
2246
2247        trace_f2fs_write_begin(inode, pos, len, flags);
2248
2249        if (f2fs_is_atomic_file(inode) &&
2250                        !f2fs_available_free_memory(sbi, INMEM_PAGES)) {
2251                err = -ENOMEM;
2252                drop_atomic = true;
2253                goto fail;
2254        }
2255
2256        /*
2257         * We should check this at this moment to avoid deadlock on inode page
2258         * and #0 page. The locking rule for inline_data conversion should be:
2259         * lock_page(page #0) -> lock_page(inode_page)
2260         */
2261        if (index != 0) {
2262                err = f2fs_convert_inline_inode(inode);
2263                if (err)
2264                        goto fail;
2265        }
2266repeat:
2267        /*
2268         * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2269         * wait_for_stable_page. Will wait that below with our IO control.
2270         */
2271        page = f2fs_pagecache_get_page(mapping, index,
2272                                FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2273        if (!page) {
2274                err = -ENOMEM;
2275                goto fail;
2276        }
2277
2278        *pagep = page;
2279
2280        err = prepare_write_begin(sbi, page, pos, len,
2281                                        &blkaddr, &need_balance);
2282        if (err)
2283                goto fail;
2284
2285        if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2286                unlock_page(page);
2287                f2fs_balance_fs(sbi, true);
2288                lock_page(page);
2289                if (page->mapping != mapping) {
2290                        /* The page got truncated from under us */
2291                        f2fs_put_page(page, 1);
2292                        goto repeat;
2293                }
2294        }
2295
2296        f2fs_wait_on_page_writeback(page, DATA, false);
2297
2298        /* wait for GCed page writeback via META_MAPPING */
2299        if (f2fs_post_read_required(inode))
2300                f2fs_wait_on_block_writeback(sbi, blkaddr);
2301
2302        if (len == PAGE_SIZE || PageUptodate(page))
2303                return 0;
2304
2305        if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2306                zero_user_segment(page, len, PAGE_SIZE);
2307                return 0;
2308        }
2309
2310        if (blkaddr == NEW_ADDR) {
2311                zero_user_segment(page, 0, PAGE_SIZE);
2312                SetPageUptodate(page);
2313        } else {
2314                err = f2fs_submit_page_read(inode, page, blkaddr);
2315                if (err)
2316                        goto fail;
2317
2318                lock_page(page);
2319                if (unlikely(page->mapping != mapping)) {
2320                        f2fs_put_page(page, 1);
2321                        goto repeat;
2322                }
2323                if (unlikely(!PageUptodate(page))) {
2324                        err = -EIO;
2325                        goto fail;
2326                }
2327        }
2328        return 0;
2329
2330fail:
2331        f2fs_put_page(page, 1);
2332        f2fs_write_failed(mapping, pos + len);
2333        if (drop_atomic)
2334                f2fs_drop_inmem_pages_all(sbi, false);
2335        return err;
2336}
2337
2338static int f2fs_write_end(struct file *file,
2339                        struct address_space *mapping,
2340                        loff_t pos, unsigned len, unsigned copied,
2341                        struct page *page, void *fsdata)
2342{
2343        struct inode *inode = page->mapping->host;
2344
2345        trace_f2fs_write_end(inode, pos, len, copied);
2346
2347        /*
2348         * This should be come from len == PAGE_SIZE, and we expect copied
2349         * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2350         * let generic_perform_write() try to copy data again through copied=0.
2351         */
2352        if (!PageUptodate(page)) {
2353                if (unlikely(copied != len))
2354                        copied = 0;
2355                else
2356                        SetPageUptodate(page);
2357        }
2358        if (!copied)
2359                goto unlock_out;
2360
2361        set_page_dirty(page);
2362
2363        if (pos + copied > i_size_read(inode))
2364                f2fs_i_size_write(inode, pos + copied);
2365unlock_out:
2366        f2fs_put_page(page, 1);
2367        f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2368        return copied;
2369}
2370
2371static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2372                           loff_t offset)
2373{
2374        unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2375
2376        if (offset & blocksize_mask)
2377                return -EINVAL;
2378
2379        if (iov_iter_alignment(iter) & blocksize_mask)
2380                return -EINVAL;
2381
2382        return 0;
2383}
2384
2385static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2386{
2387        struct address_space *mapping = iocb->ki_filp->f_mapping;
2388        struct inode *inode = mapping->host;
2389        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2390        size_t count = iov_iter_count(iter);
2391        loff_t offset = iocb->ki_pos;
2392        int rw = iov_iter_rw(iter);
2393        int err;
2394        enum rw_hint hint = iocb->ki_hint;
2395        int whint_mode = F2FS_OPTION(sbi).whint_mode;
2396
2397        err = check_direct_IO(inode, iter, offset);
2398        if (err)
2399                return err;
2400
2401        if (f2fs_force_buffered_io(inode, rw))
2402                return 0;
2403
2404        trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2405
2406        if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2407                iocb->ki_hint = WRITE_LIFE_NOT_SET;
2408
2409        if (!down_read_trylock(&F2FS_I(inode)->i_gc_rwsem[rw])) {
2410                if (iocb->ki_flags & IOCB_NOWAIT) {
2411                        iocb->ki_hint = hint;
2412                        err = -EAGAIN;
2413                        goto out;
2414                }
2415                down_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2416        }
2417
2418        err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2419        up_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2420
2421        if (rw == WRITE) {
2422                if (whint_mode == WHINT_MODE_OFF)
2423                        iocb->ki_hint = hint;
2424                if (err > 0) {
2425                        f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2426                                                                        err);
2427                        set_inode_flag(inode, FI_UPDATE_WRITE);
2428                } else if (err < 0) {
2429                        f2fs_write_failed(mapping, offset + count);
2430                }
2431        }
2432
2433out:
2434        trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2435
2436        return err;
2437}
2438
2439void f2fs_invalidate_page(struct page *page, unsigned int offset,
2440                                                        unsigned int length)
2441{
2442        struct inode *inode = page->mapping->host;
2443        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2444
2445        if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2446                (offset % PAGE_SIZE || length != PAGE_SIZE))
2447                return;
2448
2449        if (PageDirty(page)) {
2450                if (inode->i_ino == F2FS_META_INO(sbi)) {
2451                        dec_page_count(sbi, F2FS_DIRTY_META);
2452                } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2453                        dec_page_count(sbi, F2FS_DIRTY_NODES);
2454                } else {
2455                        inode_dec_dirty_pages(inode);
2456                        f2fs_remove_dirty_inode(inode);
2457                }
2458        }
2459
2460        /* This is atomic written page, keep Private */
2461        if (IS_ATOMIC_WRITTEN_PAGE(page))
2462                return f2fs_drop_inmem_page(inode, page);
2463
2464        set_page_private(page, 0);
2465        ClearPagePrivate(page);
2466}
2467
2468int f2fs_release_page(struct page *page, gfp_t wait)
2469{
2470        /* If this is dirty page, keep PagePrivate */
2471        if (PageDirty(page))
2472                return 0;
2473
2474        /* This is atomic written page, keep Private */
2475        if (IS_ATOMIC_WRITTEN_PAGE(page))
2476                return 0;
2477
2478        set_page_private(page, 0);
2479        ClearPagePrivate(page);
2480        return 1;
2481}
2482
2483static int f2fs_set_data_page_dirty(struct page *page)
2484{
2485        struct address_space *mapping = page->mapping;
2486        struct inode *inode = mapping->host;
2487
2488        trace_f2fs_set_page_dirty(page, DATA);
2489
2490        if (!PageUptodate(page))
2491                SetPageUptodate(page);
2492
2493        if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2494                if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2495                        f2fs_register_inmem_page(inode, page);
2496                        return 1;
2497                }
2498                /*
2499                 * Previously, this page has been registered, we just
2500                 * return here.
2501                 */
2502                return 0;
2503        }
2504
2505        if (!PageDirty(page)) {
2506                __set_page_dirty_nobuffers(page);
2507                f2fs_update_dirty_page(inode, page);
2508                return 1;
2509        }
2510        return 0;
2511}
2512
2513static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2514{
2515        struct inode *inode = mapping->host;
2516
2517        if (f2fs_has_inline_data(inode))
2518                return 0;
2519
2520        /* make sure allocating whole blocks */
2521        if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2522                filemap_write_and_wait(mapping);
2523
2524        return generic_block_bmap(mapping, block, get_data_block_bmap);
2525}
2526
2527#ifdef CONFIG_MIGRATION
2528#include <linux/migrate.h>
2529
2530int f2fs_migrate_page(struct address_space *mapping,
2531                struct page *newpage, struct page *page, enum migrate_mode mode)
2532{
2533        int rc, extra_count;
2534        struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2535        bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2536
2537        BUG_ON(PageWriteback(page));
2538
2539        /* migrating an atomic written page is safe with the inmem_lock hold */
2540        if (atomic_written) {
2541                if (mode != MIGRATE_SYNC)
2542                        return -EBUSY;
2543                if (!mutex_trylock(&fi->inmem_lock))
2544                        return -EAGAIN;
2545        }
2546
2547        /*
2548         * A reference is expected if PagePrivate set when move mapping,
2549         * however F2FS breaks this for maintaining dirty page counts when
2550         * truncating pages. So here adjusting the 'extra_count' make it work.
2551         */
2552        extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2553        rc = migrate_page_move_mapping(mapping, newpage,
2554                                page, NULL, mode, extra_count);
2555        if (rc != MIGRATEPAGE_SUCCESS) {
2556                if (atomic_written)
2557                        mutex_unlock(&fi->inmem_lock);
2558                return rc;
2559        }
2560
2561        if (atomic_written) {
2562                struct inmem_pages *cur;
2563                list_for_each_entry(cur, &fi->inmem_pages, list)
2564                        if (cur->page == page) {
2565                                cur->page = newpage;
2566                                break;
2567                        }
2568                mutex_unlock(&fi->inmem_lock);
2569                put_page(page);
2570                get_page(newpage);
2571        }
2572
2573        if (PagePrivate(page))
2574                SetPagePrivate(newpage);
2575        set_page_private(newpage, page_private(page));
2576
2577        if (mode != MIGRATE_SYNC_NO_COPY)
2578                migrate_page_copy(newpage, page);
2579        else
2580                migrate_page_states(newpage, page);
2581
2582        return MIGRATEPAGE_SUCCESS;
2583}
2584#endif
2585
2586const struct address_space_operations f2fs_dblock_aops = {
2587        .readpage       = f2fs_read_data_page,
2588        .readpages      = f2fs_read_data_pages,
2589        .writepage      = f2fs_write_data_page,
2590        .writepages     = f2fs_write_data_pages,
2591        .write_begin    = f2fs_write_begin,
2592        .write_end      = f2fs_write_end,
2593        .set_page_dirty = f2fs_set_data_page_dirty,
2594        .invalidatepage = f2fs_invalidate_page,
2595        .releasepage    = f2fs_release_page,
2596        .direct_IO      = f2fs_direct_IO,
2597        .bmap           = f2fs_bmap,
2598#ifdef CONFIG_MIGRATION
2599        .migratepage    = f2fs_migrate_page,
2600#endif
2601};
2602
2603void f2fs_clear_radix_tree_dirty_tag(struct page *page)
2604{
2605        struct address_space *mapping = page_mapping(page);
2606        unsigned long flags;
2607
2608        xa_lock_irqsave(&mapping->i_pages, flags);
2609        radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2610                                                PAGECACHE_TAG_DIRTY);
2611        xa_unlock_irqrestore(&mapping->i_pages, flags);
2612}
2613
2614int __init f2fs_init_post_read_processing(void)
2615{
2616        bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2617        if (!bio_post_read_ctx_cache)
2618                goto fail;
2619        bio_post_read_ctx_pool =
2620                mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2621                                         bio_post_read_ctx_cache);
2622        if (!bio_post_read_ctx_pool)
2623                goto fail_free_cache;
2624        return 0;
2625
2626fail_free_cache:
2627        kmem_cache_destroy(bio_post_read_ctx_cache);
2628fail:
2629        return -ENOMEM;
2630}
2631
2632void __exit f2fs_destroy_post_read_processing(void)
2633{
2634        mempool_destroy(bio_post_read_ctx_pool);
2635        kmem_cache_destroy(bio_post_read_ctx_cache);
2636}
2637