linux/fs/ext4/inode.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/ext4/inode.c
   3 *
   4 * Copyright (C) 1992, 1993, 1994, 1995
   5 * Remy Card (card@masi.ibp.fr)
   6 * Laboratoire MASI - Institut Blaise Pascal
   7 * Universite Pierre et Marie Curie (Paris VI)
   8 *
   9 *  from
  10 *
  11 *  linux/fs/minix/inode.c
  12 *
  13 *  Copyright (C) 1991, 1992  Linus Torvalds
  14 *
  15 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  16 *      (jj@sunsite.ms.mff.cuni.cz)
  17 *
  18 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  19 */
  20
  21#include <linux/fs.h>
  22#include <linux/time.h>
  23#include <linux/jbd2.h>
  24#include <linux/highuid.h>
  25#include <linux/pagemap.h>
  26#include <linux/quotaops.h>
  27#include <linux/string.h>
  28#include <linux/buffer_head.h>
  29#include <linux/writeback.h>
  30#include <linux/pagevec.h>
  31#include <linux/mpage.h>
  32#include <linux/namei.h>
  33#include <linux/uio.h>
  34#include <linux/bio.h>
  35#include <linux/workqueue.h>
  36#include <linux/kernel.h>
  37#include <linux/printk.h>
  38#include <linux/slab.h>
  39#include <linux/ratelimit.h>
  40
  41#include "ext4_jbd2.h"
  42#include "xattr.h"
  43#include "acl.h"
  44#include "truncate.h"
  45
  46#include <trace/events/ext4.h>
  47
  48#define MPAGE_DA_EXTENT_TAIL 0x01
  49
  50static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
  51                              struct ext4_inode_info *ei)
  52{
  53        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  54        __u16 csum_lo;
  55        __u16 csum_hi = 0;
  56        __u32 csum;
  57
  58        csum_lo = raw->i_checksum_lo;
  59        raw->i_checksum_lo = 0;
  60        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  61            EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
  62                csum_hi = raw->i_checksum_hi;
  63                raw->i_checksum_hi = 0;
  64        }
  65
  66        csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
  67                           EXT4_INODE_SIZE(inode->i_sb));
  68
  69        raw->i_checksum_lo = csum_lo;
  70        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  71            EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
  72                raw->i_checksum_hi = csum_hi;
  73
  74        return csum;
  75}
  76
  77static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
  78                                  struct ext4_inode_info *ei)
  79{
  80        __u32 provided, calculated;
  81
  82        if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  83            cpu_to_le32(EXT4_OS_LINUX) ||
  84            !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  85                EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  86                return 1;
  87
  88        provided = le16_to_cpu(raw->i_checksum_lo);
  89        calculated = ext4_inode_csum(inode, raw, ei);
  90        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  91            EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
  92                provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
  93        else
  94                calculated &= 0xFFFF;
  95
  96        return provided == calculated;
  97}
  98
  99static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
 100                                struct ext4_inode_info *ei)
 101{
 102        __u32 csum;
 103
 104        if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
 105            cpu_to_le32(EXT4_OS_LINUX) ||
 106            !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 107                EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 108                return;
 109
 110        csum = ext4_inode_csum(inode, raw, ei);
 111        raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
 112        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 113            EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 114                raw->i_checksum_hi = cpu_to_le16(csum >> 16);
 115}
 116
 117static inline int ext4_begin_ordered_truncate(struct inode *inode,
 118                                              loff_t new_size)
 119{
 120        trace_ext4_begin_ordered_truncate(inode, new_size);
 121        /*
 122         * If jinode is zero, then we never opened the file for
 123         * writing, so there's no need to call
 124         * jbd2_journal_begin_ordered_truncate() since there's no
 125         * outstanding writes we need to flush.
 126         */
 127        if (!EXT4_I(inode)->jinode)
 128                return 0;
 129        return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
 130                                                   EXT4_I(inode)->jinode,
 131                                                   new_size);
 132}
 133
 134static void ext4_invalidatepage(struct page *page, unsigned long offset);
 135static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
 136                                   struct buffer_head *bh_result, int create);
 137static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
 138static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
 139static int __ext4_journalled_writepage(struct page *page, unsigned int len);
 140static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
 141static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
 142                struct inode *inode, struct page *page, loff_t from,
 143                loff_t length, int flags);
 144
 145/*
 146 * Test whether an inode is a fast symlink.
 147 */
 148static int ext4_inode_is_fast_symlink(struct inode *inode)
 149{
 150        int ea_blocks = EXT4_I(inode)->i_file_acl ?
 151                (inode->i_sb->s_blocksize >> 9) : 0;
 152
 153        return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
 154}
 155
 156/*
 157 * Restart the transaction associated with *handle.  This does a commit,
 158 * so before we call here everything must be consistently dirtied against
 159 * this transaction.
 160 */
 161int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
 162                                 int nblocks)
 163{
 164        int ret;
 165
 166        /*
 167         * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
 168         * moment, get_block can be called only for blocks inside i_size since
 169         * page cache has been already dropped and writes are blocked by
 170         * i_mutex. So we can safely drop the i_data_sem here.
 171         */
 172        BUG_ON(EXT4_JOURNAL(inode) == NULL);
 173        jbd_debug(2, "restarting handle %p\n", handle);
 174        up_write(&EXT4_I(inode)->i_data_sem);
 175        ret = ext4_journal_restart(handle, nblocks);
 176        down_write(&EXT4_I(inode)->i_data_sem);
 177        ext4_discard_preallocations(inode);
 178
 179        return ret;
 180}
 181
 182/*
 183 * Called at the last iput() if i_nlink is zero.
 184 */
 185void ext4_evict_inode(struct inode *inode)
 186{
 187        handle_t *handle;
 188        int err;
 189
 190        trace_ext4_evict_inode(inode);
 191
 192        ext4_ioend_wait(inode);
 193
 194        if (inode->i_nlink) {
 195                /*
 196                 * When journalling data dirty buffers are tracked only in the
 197                 * journal. So although mm thinks everything is clean and
 198                 * ready for reaping the inode might still have some pages to
 199                 * write in the running transaction or waiting to be
 200                 * checkpointed. Thus calling jbd2_journal_invalidatepage()
 201                 * (via truncate_inode_pages()) to discard these buffers can
 202                 * cause data loss. Also even if we did not discard these
 203                 * buffers, we would have no way to find them after the inode
 204                 * is reaped and thus user could see stale data if he tries to
 205                 * read them before the transaction is checkpointed. So be
 206                 * careful and force everything to disk here... We use
 207                 * ei->i_datasync_tid to store the newest transaction
 208                 * containing inode's data.
 209                 *
 210                 * Note that directories do not have this problem because they
 211                 * don't use page cache.
 212                 */
 213                if (ext4_should_journal_data(inode) &&
 214                    (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
 215                        journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 216                        tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
 217
 218                        jbd2_log_start_commit(journal, commit_tid);
 219                        jbd2_log_wait_commit(journal, commit_tid);
 220                        filemap_write_and_wait(&inode->i_data);
 221                }
 222                truncate_inode_pages(&inode->i_data, 0);
 223                goto no_delete;
 224        }
 225
 226        if (!is_bad_inode(inode))
 227                dquot_initialize(inode);
 228
 229        if (ext4_should_order_data(inode))
 230                ext4_begin_ordered_truncate(inode, 0);
 231        truncate_inode_pages(&inode->i_data, 0);
 232
 233        if (is_bad_inode(inode))
 234                goto no_delete;
 235
 236        /*
 237         * Protect us against freezing - iput() caller didn't have to have any
 238         * protection against it
 239         */
 240        sb_start_intwrite(inode->i_sb);
 241        handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
 242        if (IS_ERR(handle)) {
 243                ext4_std_error(inode->i_sb, PTR_ERR(handle));
 244                /*
 245                 * If we're going to skip the normal cleanup, we still need to
 246                 * make sure that the in-core orphan linked list is properly
 247                 * cleaned up.
 248                 */
 249                ext4_orphan_del(NULL, inode);
 250                sb_end_intwrite(inode->i_sb);
 251                goto no_delete;
 252        }
 253
 254        if (IS_SYNC(inode))
 255                ext4_handle_sync(handle);
 256        inode->i_size = 0;
 257        err = ext4_mark_inode_dirty(handle, inode);
 258        if (err) {
 259                ext4_warning(inode->i_sb,
 260                             "couldn't mark inode dirty (err %d)", err);
 261                goto stop_handle;
 262        }
 263        if (inode->i_blocks)
 264                ext4_truncate(inode);
 265
 266        /*
 267         * ext4_ext_truncate() doesn't reserve any slop when it
 268         * restarts journal transactions; therefore there may not be
 269         * enough credits left in the handle to remove the inode from
 270         * the orphan list and set the dtime field.
 271         */
 272        if (!ext4_handle_has_enough_credits(handle, 3)) {
 273                err = ext4_journal_extend(handle, 3);
 274                if (err > 0)
 275                        err = ext4_journal_restart(handle, 3);
 276                if (err != 0) {
 277                        ext4_warning(inode->i_sb,
 278                                     "couldn't extend journal (err %d)", err);
 279                stop_handle:
 280                        ext4_journal_stop(handle);
 281                        ext4_orphan_del(NULL, inode);
 282                        sb_end_intwrite(inode->i_sb);
 283                        goto no_delete;
 284                }
 285        }
 286
 287        /*
 288         * Kill off the orphan record which ext4_truncate created.
 289         * AKPM: I think this can be inside the above `if'.
 290         * Note that ext4_orphan_del() has to be able to cope with the
 291         * deletion of a non-existent orphan - this is because we don't
 292         * know if ext4_truncate() actually created an orphan record.
 293         * (Well, we could do this if we need to, but heck - it works)
 294         */
 295        ext4_orphan_del(handle, inode);
 296        EXT4_I(inode)->i_dtime  = get_seconds();
 297
 298        /*
 299         * One subtle ordering requirement: if anything has gone wrong
 300         * (transaction abort, IO errors, whatever), then we can still
 301         * do these next steps (the fs will already have been marked as
 302         * having errors), but we can't free the inode if the mark_dirty
 303         * fails.
 304         */
 305        if (ext4_mark_inode_dirty(handle, inode))
 306                /* If that failed, just do the required in-core inode clear. */
 307                ext4_clear_inode(inode);
 308        else
 309                ext4_free_inode(handle, inode);
 310        ext4_journal_stop(handle);
 311        sb_end_intwrite(inode->i_sb);
 312        return;
 313no_delete:
 314        ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
 315}
 316
 317#ifdef CONFIG_QUOTA
 318qsize_t *ext4_get_reserved_space(struct inode *inode)
 319{
 320        return &EXT4_I(inode)->i_reserved_quota;
 321}
 322#endif
 323
 324/*
 325 * Calculate the number of metadata blocks need to reserve
 326 * to allocate a block located at @lblock
 327 */
 328static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 329{
 330        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 331                return ext4_ext_calc_metadata_amount(inode, lblock);
 332
 333        return ext4_ind_calc_metadata_amount(inode, lblock);
 334}
 335
 336/*
 337 * Called with i_data_sem down, which is important since we can call
 338 * ext4_discard_preallocations() from here.
 339 */
 340void ext4_da_update_reserve_space(struct inode *inode,
 341                                        int used, int quota_claim)
 342{
 343        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 344        struct ext4_inode_info *ei = EXT4_I(inode);
 345
 346        spin_lock(&ei->i_block_reservation_lock);
 347        trace_ext4_da_update_reserve_space(inode, used, quota_claim);
 348        if (unlikely(used > ei->i_reserved_data_blocks)) {
 349                ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
 350                         "with only %d reserved data blocks",
 351                         __func__, inode->i_ino, used,
 352                         ei->i_reserved_data_blocks);
 353                WARN_ON(1);
 354                used = ei->i_reserved_data_blocks;
 355        }
 356
 357        if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
 358                ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
 359                         "with only %d reserved metadata blocks\n", __func__,
 360                         inode->i_ino, ei->i_allocated_meta_blocks,
 361                         ei->i_reserved_meta_blocks);
 362                WARN_ON(1);
 363                ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
 364        }
 365
 366        /* Update per-inode reservations */
 367        ei->i_reserved_data_blocks -= used;
 368        ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
 369        percpu_counter_sub(&sbi->s_dirtyclusters_counter,
 370                           used + ei->i_allocated_meta_blocks);
 371        ei->i_allocated_meta_blocks = 0;
 372
 373        if (ei->i_reserved_data_blocks == 0) {
 374                /*
 375                 * We can release all of the reserved metadata blocks
 376                 * only when we have written all of the delayed
 377                 * allocation blocks.
 378                 */
 379                percpu_counter_sub(&sbi->s_dirtyclusters_counter,
 380                                   ei->i_reserved_meta_blocks);
 381                ei->i_reserved_meta_blocks = 0;
 382                ei->i_da_metadata_calc_len = 0;
 383        }
 384        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 385
 386        /* Update quota subsystem for data blocks */
 387        if (quota_claim)
 388                dquot_claim_block(inode, EXT4_C2B(sbi, used));
 389        else {
 390                /*
 391                 * We did fallocate with an offset that is already delayed
 392                 * allocated. So on delayed allocated writeback we should
 393                 * not re-claim the quota for fallocated blocks.
 394                 */
 395                dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
 396        }
 397
 398        /*
 399         * If we have done all the pending block allocations and if
 400         * there aren't any writers on the inode, we can discard the
 401         * inode's preallocations.
 402         */
 403        if ((ei->i_reserved_data_blocks == 0) &&
 404            (atomic_read(&inode->i_writecount) == 0))
 405                ext4_discard_preallocations(inode);
 406}
 407
 408static int __check_block_validity(struct inode *inode, const char *func,
 409                                unsigned int line,
 410                                struct ext4_map_blocks *map)
 411{
 412        if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
 413                                   map->m_len)) {
 414                ext4_error_inode(inode, func, line, map->m_pblk,
 415                                 "lblock %lu mapped to illegal pblock "
 416                                 "(length %d)", (unsigned long) map->m_lblk,
 417                                 map->m_len);
 418                return -EIO;
 419        }
 420        return 0;
 421}
 422
 423#define check_block_validity(inode, map)        \
 424        __check_block_validity((inode), __func__, __LINE__, (map))
 425
 426/*
 427 * Return the number of contiguous dirty pages in a given inode
 428 * starting at page frame idx.
 429 */
 430static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
 431                                    unsigned int max_pages)
 432{
 433        struct address_space *mapping = inode->i_mapping;
 434        pgoff_t index;
 435        struct pagevec pvec;
 436        pgoff_t num = 0;
 437        int i, nr_pages, done = 0;
 438
 439        if (max_pages == 0)
 440                return 0;
 441        pagevec_init(&pvec, 0);
 442        while (!done) {
 443                index = idx;
 444                nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 445                                              PAGECACHE_TAG_DIRTY,
 446                                              (pgoff_t)PAGEVEC_SIZE);
 447                if (nr_pages == 0)
 448                        break;
 449                for (i = 0; i < nr_pages; i++) {
 450                        struct page *page = pvec.pages[i];
 451                        struct buffer_head *bh, *head;
 452
 453                        lock_page(page);
 454                        if (unlikely(page->mapping != mapping) ||
 455                            !PageDirty(page) ||
 456                            PageWriteback(page) ||
 457                            page->index != idx) {
 458                                done = 1;
 459                                unlock_page(page);
 460                                break;
 461                        }
 462                        if (page_has_buffers(page)) {
 463                                bh = head = page_buffers(page);
 464                                do {
 465                                        if (!buffer_delay(bh) &&
 466                                            !buffer_unwritten(bh))
 467                                                done = 1;
 468                                        bh = bh->b_this_page;
 469                                } while (!done && (bh != head));
 470                        }
 471                        unlock_page(page);
 472                        if (done)
 473                                break;
 474                        idx++;
 475                        num++;
 476                        if (num >= max_pages) {
 477                                done = 1;
 478                                break;
 479                        }
 480                }
 481                pagevec_release(&pvec);
 482        }
 483        return num;
 484}
 485
 486/*
 487 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
 488 */
 489static void set_buffers_da_mapped(struct inode *inode,
 490                                   struct ext4_map_blocks *map)
 491{
 492        struct address_space *mapping = inode->i_mapping;
 493        struct pagevec pvec;
 494        int i, nr_pages;
 495        pgoff_t index, end;
 496
 497        index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
 498        end = (map->m_lblk + map->m_len - 1) >>
 499                (PAGE_CACHE_SHIFT - inode->i_blkbits);
 500
 501        pagevec_init(&pvec, 0);
 502        while (index <= end) {
 503                nr_pages = pagevec_lookup(&pvec, mapping, index,
 504                                          min(end - index + 1,
 505                                              (pgoff_t)PAGEVEC_SIZE));
 506                if (nr_pages == 0)
 507                        break;
 508                for (i = 0; i < nr_pages; i++) {
 509                        struct page *page = pvec.pages[i];
 510                        struct buffer_head *bh, *head;
 511
 512                        if (unlikely(page->mapping != mapping) ||
 513                            !PageDirty(page))
 514                                break;
 515
 516                        if (page_has_buffers(page)) {
 517                                bh = head = page_buffers(page);
 518                                do {
 519                                        set_buffer_da_mapped(bh);
 520                                        bh = bh->b_this_page;
 521                                } while (bh != head);
 522                        }
 523                        index++;
 524                }
 525                pagevec_release(&pvec);
 526        }
 527}
 528
 529/*
 530 * The ext4_map_blocks() function tries to look up the requested blocks,
 531 * and returns if the blocks are already mapped.
 532 *
 533 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 534 * and store the allocated blocks in the result buffer head and mark it
 535 * mapped.
 536 *
 537 * If file type is extents based, it will call ext4_ext_map_blocks(),
 538 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 539 * based files
 540 *
 541 * On success, it returns the number of blocks being mapped or allocate.
 542 * if create==0 and the blocks are pre-allocated and uninitialized block,
 543 * the result buffer head is unmapped. If the create ==1, it will make sure
 544 * the buffer head is mapped.
 545 *
 546 * It returns 0 if plain look up failed (blocks have not been allocated), in
 547 * that case, buffer head is unmapped
 548 *
 549 * It returns the error in case of allocation failure.
 550 */
 551int ext4_map_blocks(handle_t *handle, struct inode *inode,
 552                    struct ext4_map_blocks *map, int flags)
 553{
 554        int retval;
 555
 556        map->m_flags = 0;
 557        ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
 558                  "logical block %lu\n", inode->i_ino, flags, map->m_len,
 559                  (unsigned long) map->m_lblk);
 560        /*
 561         * Try to see if we can get the block without requesting a new
 562         * file system block.
 563         */
 564        if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
 565                down_read((&EXT4_I(inode)->i_data_sem));
 566        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 567                retval = ext4_ext_map_blocks(handle, inode, map, flags &
 568                                             EXT4_GET_BLOCKS_KEEP_SIZE);
 569        } else {
 570                retval = ext4_ind_map_blocks(handle, inode, map, flags &
 571                                             EXT4_GET_BLOCKS_KEEP_SIZE);
 572        }
 573        if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
 574                up_read((&EXT4_I(inode)->i_data_sem));
 575
 576        if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 577                int ret = check_block_validity(inode, map);
 578                if (ret != 0)
 579                        return ret;
 580        }
 581
 582        /* If it is only a block(s) look up */
 583        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 584                return retval;
 585
 586        /*
 587         * Returns if the blocks have already allocated
 588         *
 589         * Note that if blocks have been preallocated
 590         * ext4_ext_get_block() returns the create = 0
 591         * with buffer head unmapped.
 592         */
 593        if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 594                return retval;
 595
 596        /*
 597         * When we call get_blocks without the create flag, the
 598         * BH_Unwritten flag could have gotten set if the blocks
 599         * requested were part of a uninitialized extent.  We need to
 600         * clear this flag now that we are committed to convert all or
 601         * part of the uninitialized extent to be an initialized
 602         * extent.  This is because we need to avoid the combination
 603         * of BH_Unwritten and BH_Mapped flags being simultaneously
 604         * set on the buffer_head.
 605         */
 606        map->m_flags &= ~EXT4_MAP_UNWRITTEN;
 607
 608        /*
 609         * New blocks allocate and/or writing to uninitialized extent
 610         * will possibly result in updating i_data, so we take
 611         * the write lock of i_data_sem, and call get_blocks()
 612         * with create == 1 flag.
 613         */
 614        down_write((&EXT4_I(inode)->i_data_sem));
 615
 616        /*
 617         * if the caller is from delayed allocation writeout path
 618         * we have already reserved fs blocks for allocation
 619         * let the underlying get_block() function know to
 620         * avoid double accounting
 621         */
 622        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 623                ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 624        /*
 625         * We need to check for EXT4 here because migrate
 626         * could have changed the inode type in between
 627         */
 628        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 629                retval = ext4_ext_map_blocks(handle, inode, map, flags);
 630        } else {
 631                retval = ext4_ind_map_blocks(handle, inode, map, flags);
 632
 633                if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 634                        /*
 635                         * We allocated new blocks which will result in
 636                         * i_data's format changing.  Force the migrate
 637                         * to fail by clearing migrate flags
 638                         */
 639                        ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 640                }
 641
 642                /*
 643                 * Update reserved blocks/metadata blocks after successful
 644                 * block allocation which had been deferred till now. We don't
 645                 * support fallocate for non extent files. So we can update
 646                 * reserve space here.
 647                 */
 648                if ((retval > 0) &&
 649                        (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
 650                        ext4_da_update_reserve_space(inode, retval, 1);
 651        }
 652        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
 653                ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 654
 655                /* If we have successfully mapped the delayed allocated blocks,
 656                 * set the BH_Da_Mapped bit on them. Its important to do this
 657                 * under the protection of i_data_sem.
 658                 */
 659                if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 660                        set_buffers_da_mapped(inode, map);
 661        }
 662
 663        up_write((&EXT4_I(inode)->i_data_sem));
 664        if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 665                int ret = check_block_validity(inode, map);
 666                if (ret != 0)
 667                        return ret;
 668        }
 669        return retval;
 670}
 671
 672/* Maximum number of blocks we map for direct IO at once. */
 673#define DIO_MAX_BLOCKS 4096
 674
 675static int _ext4_get_block(struct inode *inode, sector_t iblock,
 676                           struct buffer_head *bh, int flags)
 677{
 678        handle_t *handle = ext4_journal_current_handle();
 679        struct ext4_map_blocks map;
 680        int ret = 0, started = 0;
 681        int dio_credits;
 682
 683        map.m_lblk = iblock;
 684        map.m_len = bh->b_size >> inode->i_blkbits;
 685
 686        if (flags && !handle) {
 687                /* Direct IO write... */
 688                if (map.m_len > DIO_MAX_BLOCKS)
 689                        map.m_len = DIO_MAX_BLOCKS;
 690                dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
 691                handle = ext4_journal_start(inode, dio_credits);
 692                if (IS_ERR(handle)) {
 693                        ret = PTR_ERR(handle);
 694                        return ret;
 695                }
 696                started = 1;
 697        }
 698
 699        ret = ext4_map_blocks(handle, inode, &map, flags);
 700        if (ret > 0) {
 701                map_bh(bh, inode->i_sb, map.m_pblk);
 702                bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
 703                bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 704                ret = 0;
 705        }
 706        if (started)
 707                ext4_journal_stop(handle);
 708        return ret;
 709}
 710
 711int ext4_get_block(struct inode *inode, sector_t iblock,
 712                   struct buffer_head *bh, int create)
 713{
 714        return _ext4_get_block(inode, iblock, bh,
 715                               create ? EXT4_GET_BLOCKS_CREATE : 0);
 716}
 717
 718/*
 719 * `handle' can be NULL if create is zero
 720 */
 721struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 722                                ext4_lblk_t block, int create, int *errp)
 723{
 724        struct ext4_map_blocks map;
 725        struct buffer_head *bh;
 726        int fatal = 0, err;
 727
 728        J_ASSERT(handle != NULL || create == 0);
 729
 730        map.m_lblk = block;
 731        map.m_len = 1;
 732        err = ext4_map_blocks(handle, inode, &map,
 733                              create ? EXT4_GET_BLOCKS_CREATE : 0);
 734
 735        /* ensure we send some value back into *errp */
 736        *errp = 0;
 737
 738        if (err < 0)
 739                *errp = err;
 740        if (err <= 0)
 741                return NULL;
 742
 743        bh = sb_getblk(inode->i_sb, map.m_pblk);
 744        if (!bh) {
 745                *errp = -EIO;
 746                return NULL;
 747        }
 748        if (map.m_flags & EXT4_MAP_NEW) {
 749                J_ASSERT(create != 0);
 750                J_ASSERT(handle != NULL);
 751
 752                /*
 753                 * Now that we do not always journal data, we should
 754                 * keep in mind whether this should always journal the
 755                 * new buffer as metadata.  For now, regular file
 756                 * writes use ext4_get_block instead, so it's not a
 757                 * problem.
 758                 */
 759                lock_buffer(bh);
 760                BUFFER_TRACE(bh, "call get_create_access");
 761                fatal = ext4_journal_get_create_access(handle, bh);
 762                if (!fatal && !buffer_uptodate(bh)) {
 763                        memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 764                        set_buffer_uptodate(bh);
 765                }
 766                unlock_buffer(bh);
 767                BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 768                err = ext4_handle_dirty_metadata(handle, inode, bh);
 769                if (!fatal)
 770                        fatal = err;
 771        } else {
 772                BUFFER_TRACE(bh, "not a new buffer");
 773        }
 774        if (fatal) {
 775                *errp = fatal;
 776                brelse(bh);
 777                bh = NULL;
 778        }
 779        return bh;
 780}
 781
 782struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 783                               ext4_lblk_t block, int create, int *err)
 784{
 785        struct buffer_head *bh;
 786
 787        bh = ext4_getblk(handle, inode, block, create, err);
 788        if (!bh)
 789                return bh;
 790        if (buffer_uptodate(bh))
 791                return bh;
 792        ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
 793        wait_on_buffer(bh);
 794        if (buffer_uptodate(bh))
 795                return bh;
 796        put_bh(bh);
 797        *err = -EIO;
 798        return NULL;
 799}
 800
 801static int walk_page_buffers(handle_t *handle,
 802                             struct buffer_head *head,
 803                             unsigned from,
 804                             unsigned to,
 805                             int *partial,
 806                             int (*fn)(handle_t *handle,
 807                                       struct buffer_head *bh))
 808{
 809        struct buffer_head *bh;
 810        unsigned block_start, block_end;
 811        unsigned blocksize = head->b_size;
 812        int err, ret = 0;
 813        struct buffer_head *next;
 814
 815        for (bh = head, block_start = 0;
 816             ret == 0 && (bh != head || !block_start);
 817             block_start = block_end, bh = next) {
 818                next = bh->b_this_page;
 819                block_end = block_start + blocksize;
 820                if (block_end <= from || block_start >= to) {
 821                        if (partial && !buffer_uptodate(bh))
 822                                *partial = 1;
 823                        continue;
 824                }
 825                err = (*fn)(handle, bh);
 826                if (!ret)
 827                        ret = err;
 828        }
 829        return ret;
 830}
 831
 832/*
 833 * To preserve ordering, it is essential that the hole instantiation and
 834 * the data write be encapsulated in a single transaction.  We cannot
 835 * close off a transaction and start a new one between the ext4_get_block()
 836 * and the commit_write().  So doing the jbd2_journal_start at the start of
 837 * prepare_write() is the right place.
 838 *
 839 * Also, this function can nest inside ext4_writepage() ->
 840 * block_write_full_page(). In that case, we *know* that ext4_writepage()
 841 * has generated enough buffer credits to do the whole page.  So we won't
 842 * block on the journal in that case, which is good, because the caller may
 843 * be PF_MEMALLOC.
 844 *
 845 * By accident, ext4 can be reentered when a transaction is open via
 846 * quota file writes.  If we were to commit the transaction while thus
 847 * reentered, there can be a deadlock - we would be holding a quota
 848 * lock, and the commit would never complete if another thread had a
 849 * transaction open and was blocking on the quota lock - a ranking
 850 * violation.
 851 *
 852 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
 853 * will _not_ run commit under these circumstances because handle->h_ref
 854 * is elevated.  We'll still have enough credits for the tiny quotafile
 855 * write.
 856 */
 857static int do_journal_get_write_access(handle_t *handle,
 858                                       struct buffer_head *bh)
 859{
 860        int dirty = buffer_dirty(bh);
 861        int ret;
 862
 863        if (!buffer_mapped(bh) || buffer_freed(bh))
 864                return 0;
 865        /*
 866         * __block_write_begin() could have dirtied some buffers. Clean
 867         * the dirty bit as jbd2_journal_get_write_access() could complain
 868         * otherwise about fs integrity issues. Setting of the dirty bit
 869         * by __block_write_begin() isn't a real problem here as we clear
 870         * the bit before releasing a page lock and thus writeback cannot
 871         * ever write the buffer.
 872         */
 873        if (dirty)
 874                clear_buffer_dirty(bh);
 875        ret = ext4_journal_get_write_access(handle, bh);
 876        if (!ret && dirty)
 877                ret = ext4_handle_dirty_metadata(handle, NULL, bh);
 878        return ret;
 879}
 880
 881static int ext4_get_block_write(struct inode *inode, sector_t iblock,
 882                   struct buffer_head *bh_result, int create);
 883static int ext4_write_begin(struct file *file, struct address_space *mapping,
 884                            loff_t pos, unsigned len, unsigned flags,
 885                            struct page **pagep, void **fsdata)
 886{
 887        struct inode *inode = mapping->host;
 888        int ret, needed_blocks;
 889        handle_t *handle;
 890        int retries = 0;
 891        struct page *page;
 892        pgoff_t index;
 893        unsigned from, to;
 894
 895        trace_ext4_write_begin(inode, pos, len, flags);
 896        /*
 897         * Reserve one block more for addition to orphan list in case
 898         * we allocate blocks but write fails for some reason
 899         */
 900        needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
 901        index = pos >> PAGE_CACHE_SHIFT;
 902        from = pos & (PAGE_CACHE_SIZE - 1);
 903        to = from + len;
 904
 905retry:
 906        handle = ext4_journal_start(inode, needed_blocks);
 907        if (IS_ERR(handle)) {
 908                ret = PTR_ERR(handle);
 909                goto out;
 910        }
 911
 912        /* We cannot recurse into the filesystem as the transaction is already
 913         * started */
 914        flags |= AOP_FLAG_NOFS;
 915
 916        page = grab_cache_page_write_begin(mapping, index, flags);
 917        if (!page) {
 918                ext4_journal_stop(handle);
 919                ret = -ENOMEM;
 920                goto out;
 921        }
 922        *pagep = page;
 923
 924        if (ext4_should_dioread_nolock(inode))
 925                ret = __block_write_begin(page, pos, len, ext4_get_block_write);
 926        else
 927                ret = __block_write_begin(page, pos, len, ext4_get_block);
 928
 929        if (!ret && ext4_should_journal_data(inode)) {
 930                ret = walk_page_buffers(handle, page_buffers(page),
 931                                from, to, NULL, do_journal_get_write_access);
 932        }
 933
 934        if (ret) {
 935                unlock_page(page);
 936                page_cache_release(page);
 937                /*
 938                 * __block_write_begin may have instantiated a few blocks
 939                 * outside i_size.  Trim these off again. Don't need
 940                 * i_size_read because we hold i_mutex.
 941                 *
 942                 * Add inode to orphan list in case we crash before
 943                 * truncate finishes
 944                 */
 945                if (pos + len > inode->i_size && ext4_can_truncate(inode))
 946                        ext4_orphan_add(handle, inode);
 947
 948                ext4_journal_stop(handle);
 949                if (pos + len > inode->i_size) {
 950                        ext4_truncate_failed_write(inode);
 951                        /*
 952                         * If truncate failed early the inode might
 953                         * still be on the orphan list; we need to
 954                         * make sure the inode is removed from the
 955                         * orphan list in that case.
 956                         */
 957                        if (inode->i_nlink)
 958                                ext4_orphan_del(NULL, inode);
 959                }
 960        }
 961
 962        if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 963                goto retry;
 964out:
 965        return ret;
 966}
 967
 968/* For write_end() in data=journal mode */
 969static int write_end_fn(handle_t *handle, struct buffer_head *bh)
 970{
 971        if (!buffer_mapped(bh) || buffer_freed(bh))
 972                return 0;
 973        set_buffer_uptodate(bh);
 974        return ext4_handle_dirty_metadata(handle, NULL, bh);
 975}
 976
 977static int ext4_generic_write_end(struct file *file,
 978                                  struct address_space *mapping,
 979                                  loff_t pos, unsigned len, unsigned copied,
 980                                  struct page *page, void *fsdata)
 981{
 982        int i_size_changed = 0;
 983        struct inode *inode = mapping->host;
 984        handle_t *handle = ext4_journal_current_handle();
 985
 986        copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 987
 988        /*
 989         * No need to use i_size_read() here, the i_size
 990         * cannot change under us because we hold i_mutex.
 991         *
 992         * But it's important to update i_size while still holding page lock:
 993         * page writeout could otherwise come in and zero beyond i_size.
 994         */
 995        if (pos + copied > inode->i_size) {
 996                i_size_write(inode, pos + copied);
 997                i_size_changed = 1;
 998        }
 999
1000        if (pos + copied >  EXT4_I(inode)->i_disksize) {
1001                /* We need to mark inode dirty even if
1002                 * new_i_size is less that inode->i_size
1003                 * bu greater than i_disksize.(hint delalloc)
1004                 */
1005                ext4_update_i_disksize(inode, (pos + copied));
1006                i_size_changed = 1;
1007        }
1008        unlock_page(page);
1009        page_cache_release(page);
1010
1011        /*
1012         * Don't mark the inode dirty under page lock. First, it unnecessarily
1013         * makes the holding time of page lock longer. Second, it forces lock
1014         * ordering of page lock and transaction start for journaling
1015         * filesystems.
1016         */
1017        if (i_size_changed)
1018                ext4_mark_inode_dirty(handle, inode);
1019
1020        return copied;
1021}
1022
1023/*
1024 * We need to pick up the new inode size which generic_commit_write gave us
1025 * `file' can be NULL - eg, when called from page_symlink().
1026 *
1027 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1028 * buffers are managed internally.
1029 */
1030static int ext4_ordered_write_end(struct file *file,
1031                                  struct address_space *mapping,
1032                                  loff_t pos, unsigned len, unsigned copied,
1033                                  struct page *page, void *fsdata)
1034{
1035        handle_t *handle = ext4_journal_current_handle();
1036        struct inode *inode = mapping->host;
1037        int ret = 0, ret2;
1038
1039        trace_ext4_ordered_write_end(inode, pos, len, copied);
1040        ret = ext4_jbd2_file_inode(handle, inode);
1041
1042        if (ret == 0) {
1043                ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1044                                                        page, fsdata);
1045                copied = ret2;
1046                if (pos + len > inode->i_size && ext4_can_truncate(inode))
1047                        /* if we have allocated more blocks and copied
1048                         * less. We will have blocks allocated outside
1049                         * inode->i_size. So truncate them
1050                         */
1051                        ext4_orphan_add(handle, inode);
1052                if (ret2 < 0)
1053                        ret = ret2;
1054        } else {
1055                unlock_page(page);
1056                page_cache_release(page);
1057        }
1058
1059        ret2 = ext4_journal_stop(handle);
1060        if (!ret)
1061                ret = ret2;
1062
1063        if (pos + len > inode->i_size) {
1064                ext4_truncate_failed_write(inode);
1065                /*
1066                 * If truncate failed early the inode might still be
1067                 * on the orphan list; we need to make sure the inode
1068                 * is removed from the orphan list in that case.
1069                 */
1070                if (inode->i_nlink)
1071                        ext4_orphan_del(NULL, inode);
1072        }
1073
1074
1075        return ret ? ret : copied;
1076}
1077
1078static int ext4_writeback_write_end(struct file *file,
1079                                    struct address_space *mapping,
1080                                    loff_t pos, unsigned len, unsigned copied,
1081                                    struct page *page, void *fsdata)
1082{
1083        handle_t *handle = ext4_journal_current_handle();
1084        struct inode *inode = mapping->host;
1085        int ret = 0, ret2;
1086
1087        trace_ext4_writeback_write_end(inode, pos, len, copied);
1088        ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1089                                                        page, fsdata);
1090        copied = ret2;
1091        if (pos + len > inode->i_size && ext4_can_truncate(inode))
1092                /* if we have allocated more blocks and copied
1093                 * less. We will have blocks allocated outside
1094                 * inode->i_size. So truncate them
1095                 */
1096                ext4_orphan_add(handle, inode);
1097
1098        if (ret2 < 0)
1099                ret = ret2;
1100
1101        ret2 = ext4_journal_stop(handle);
1102        if (!ret)
1103                ret = ret2;
1104
1105        if (pos + len > inode->i_size) {
1106                ext4_truncate_failed_write(inode);
1107                /*
1108                 * If truncate failed early the inode might still be
1109                 * on the orphan list; we need to make sure the inode
1110                 * is removed from the orphan list in that case.
1111                 */
1112                if (inode->i_nlink)
1113                        ext4_orphan_del(NULL, inode);
1114        }
1115
1116        return ret ? ret : copied;
1117}
1118
1119static int ext4_journalled_write_end(struct file *file,
1120                                     struct address_space *mapping,
1121                                     loff_t pos, unsigned len, unsigned copied,
1122                                     struct page *page, void *fsdata)
1123{
1124        handle_t *handle = ext4_journal_current_handle();
1125        struct inode *inode = mapping->host;
1126        int ret = 0, ret2;
1127        int partial = 0;
1128        unsigned from, to;
1129        loff_t new_i_size;
1130
1131        trace_ext4_journalled_write_end(inode, pos, len, copied);
1132        from = pos & (PAGE_CACHE_SIZE - 1);
1133        to = from + len;
1134
1135        BUG_ON(!ext4_handle_valid(handle));
1136
1137        if (copied < len) {
1138                if (!PageUptodate(page))
1139                        copied = 0;
1140                page_zero_new_buffers(page, from+copied, to);
1141        }
1142
1143        ret = walk_page_buffers(handle, page_buffers(page), from,
1144                                to, &partial, write_end_fn);
1145        if (!partial)
1146                SetPageUptodate(page);
1147        new_i_size = pos + copied;
1148        if (new_i_size > inode->i_size)
1149                i_size_write(inode, pos+copied);
1150        ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1151        EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1152        if (new_i_size > EXT4_I(inode)->i_disksize) {
1153                ext4_update_i_disksize(inode, new_i_size);
1154                ret2 = ext4_mark_inode_dirty(handle, inode);
1155                if (!ret)
1156                        ret = ret2;
1157        }
1158
1159        unlock_page(page);
1160        page_cache_release(page);
1161        if (pos + len > inode->i_size && ext4_can_truncate(inode))
1162                /* if we have allocated more blocks and copied
1163                 * less. We will have blocks allocated outside
1164                 * inode->i_size. So truncate them
1165                 */
1166                ext4_orphan_add(handle, inode);
1167
1168        ret2 = ext4_journal_stop(handle);
1169        if (!ret)
1170                ret = ret2;
1171        if (pos + len > inode->i_size) {
1172                ext4_truncate_failed_write(inode);
1173                /*
1174                 * If truncate failed early the inode might still be
1175                 * on the orphan list; we need to make sure the inode
1176                 * is removed from the orphan list in that case.
1177                 */
1178                if (inode->i_nlink)
1179                        ext4_orphan_del(NULL, inode);
1180        }
1181
1182        return ret ? ret : copied;
1183}
1184
1185/*
1186 * Reserve a single cluster located at lblock
1187 */
1188static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1189{
1190        int retries = 0;
1191        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1192        struct ext4_inode_info *ei = EXT4_I(inode);
1193        unsigned int md_needed;
1194        int ret;
1195        ext4_lblk_t save_last_lblock;
1196        int save_len;
1197
1198        /*
1199         * We will charge metadata quota at writeout time; this saves
1200         * us from metadata over-estimation, though we may go over by
1201         * a small amount in the end.  Here we just reserve for data.
1202         */
1203        ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1204        if (ret)
1205                return ret;
1206
1207        /*
1208         * recalculate the amount of metadata blocks to reserve
1209         * in order to allocate nrblocks
1210         * worse case is one extent per block
1211         */
1212repeat:
1213        spin_lock(&ei->i_block_reservation_lock);
1214        /*
1215         * ext4_calc_metadata_amount() has side effects, which we have
1216         * to be prepared undo if we fail to claim space.
1217         */
1218        save_len = ei->i_da_metadata_calc_len;
1219        save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1220        md_needed = EXT4_NUM_B2C(sbi,
1221                                 ext4_calc_metadata_amount(inode, lblock));
1222        trace_ext4_da_reserve_space(inode, md_needed);
1223
1224        /*
1225         * We do still charge estimated metadata to the sb though;
1226         * we cannot afford to run out of free blocks.
1227         */
1228        if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1229                ei->i_da_metadata_calc_len = save_len;
1230                ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1231                spin_unlock(&ei->i_block_reservation_lock);
1232                if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1233                        yield();
1234                        goto repeat;
1235                }
1236                dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1237                return -ENOSPC;
1238        }
1239        ei->i_reserved_data_blocks++;
1240        ei->i_reserved_meta_blocks += md_needed;
1241        spin_unlock(&ei->i_block_reservation_lock);
1242
1243        return 0;       /* success */
1244}
1245
1246static void ext4_da_release_space(struct inode *inode, int to_free)
1247{
1248        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1249        struct ext4_inode_info *ei = EXT4_I(inode);
1250
1251        if (!to_free)
1252                return;         /* Nothing to release, exit */
1253
1254        spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1255
1256        trace_ext4_da_release_space(inode, to_free);
1257        if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1258                /*
1259                 * if there aren't enough reserved blocks, then the
1260                 * counter is messed up somewhere.  Since this
1261                 * function is called from invalidate page, it's
1262                 * harmless to return without any action.
1263                 */
1264                ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1265                         "ino %lu, to_free %d with only %d reserved "
1266                         "data blocks", inode->i_ino, to_free,
1267                         ei->i_reserved_data_blocks);
1268                WARN_ON(1);
1269                to_free = ei->i_reserved_data_blocks;
1270        }
1271        ei->i_reserved_data_blocks -= to_free;
1272
1273        if (ei->i_reserved_data_blocks == 0) {
1274                /*
1275                 * We can release all of the reserved metadata blocks
1276                 * only when we have written all of the delayed
1277                 * allocation blocks.
1278                 * Note that in case of bigalloc, i_reserved_meta_blocks,
1279                 * i_reserved_data_blocks, etc. refer to number of clusters.
1280                 */
1281                percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1282                                   ei->i_reserved_meta_blocks);
1283                ei->i_reserved_meta_blocks = 0;
1284                ei->i_da_metadata_calc_len = 0;
1285        }
1286
1287        /* update fs dirty data blocks counter */
1288        percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1289
1290        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1291
1292        dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1293}
1294
1295static void ext4_da_page_release_reservation(struct page *page,
1296                                             unsigned long offset)
1297{
1298        int to_release = 0;
1299        struct buffer_head *head, *bh;
1300        unsigned int curr_off = 0;
1301        struct inode *inode = page->mapping->host;
1302        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1303        int num_clusters;
1304
1305        head = page_buffers(page);
1306        bh = head;
1307        do {
1308                unsigned int next_off = curr_off + bh->b_size;
1309
1310                if ((offset <= curr_off) && (buffer_delay(bh))) {
1311                        to_release++;
1312                        clear_buffer_delay(bh);
1313                        clear_buffer_da_mapped(bh);
1314                }
1315                curr_off = next_off;
1316        } while ((bh = bh->b_this_page) != head);
1317
1318        /* If we have released all the blocks belonging to a cluster, then we
1319         * need to release the reserved space for that cluster. */
1320        num_clusters = EXT4_NUM_B2C(sbi, to_release);
1321        while (num_clusters > 0) {
1322                ext4_fsblk_t lblk;
1323                lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1324                        ((num_clusters - 1) << sbi->s_cluster_bits);
1325                if (sbi->s_cluster_ratio == 1 ||
1326                    !ext4_find_delalloc_cluster(inode, lblk, 1))
1327                        ext4_da_release_space(inode, 1);
1328
1329                num_clusters--;
1330        }
1331}
1332
1333/*
1334 * Delayed allocation stuff
1335 */
1336
1337/*
1338 * mpage_da_submit_io - walks through extent of pages and try to write
1339 * them with writepage() call back
1340 *
1341 * @mpd->inode: inode
1342 * @mpd->first_page: first page of the extent
1343 * @mpd->next_page: page after the last page of the extent
1344 *
1345 * By the time mpage_da_submit_io() is called we expect all blocks
1346 * to be allocated. this may be wrong if allocation failed.
1347 *
1348 * As pages are already locked by write_cache_pages(), we can't use it
1349 */
1350static int mpage_da_submit_io(struct mpage_da_data *mpd,
1351                              struct ext4_map_blocks *map)
1352{
1353        struct pagevec pvec;
1354        unsigned long index, end;
1355        int ret = 0, err, nr_pages, i;
1356        struct inode *inode = mpd->inode;
1357        struct address_space *mapping = inode->i_mapping;
1358        loff_t size = i_size_read(inode);
1359        unsigned int len, block_start;
1360        struct buffer_head *bh, *page_bufs = NULL;
1361        int journal_data = ext4_should_journal_data(inode);
1362        sector_t pblock = 0, cur_logical = 0;
1363        struct ext4_io_submit io_submit;
1364
1365        BUG_ON(mpd->next_page <= mpd->first_page);
1366        memset(&io_submit, 0, sizeof(io_submit));
1367        /*
1368         * We need to start from the first_page to the next_page - 1
1369         * to make sure we also write the mapped dirty buffer_heads.
1370         * If we look at mpd->b_blocknr we would only be looking
1371         * at the currently mapped buffer_heads.
1372         */
1373        index = mpd->first_page;
1374        end = mpd->next_page - 1;
1375
1376        pagevec_init(&pvec, 0);
1377        while (index <= end) {
1378                nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1379                if (nr_pages == 0)
1380                        break;
1381                for (i = 0; i < nr_pages; i++) {
1382                        int commit_write = 0, skip_page = 0;
1383                        struct page *page = pvec.pages[i];
1384
1385                        index = page->index;
1386                        if (index > end)
1387                                break;
1388
1389                        if (index == size >> PAGE_CACHE_SHIFT)
1390                                len = size & ~PAGE_CACHE_MASK;
1391                        else
1392                                len = PAGE_CACHE_SIZE;
1393                        if (map) {
1394                                cur_logical = index << (PAGE_CACHE_SHIFT -
1395                                                        inode->i_blkbits);
1396                                pblock = map->m_pblk + (cur_logical -
1397                                                        map->m_lblk);
1398                        }
1399                        index++;
1400
1401                        BUG_ON(!PageLocked(page));
1402                        BUG_ON(PageWriteback(page));
1403
1404                        /*
1405                         * If the page does not have buffers (for
1406                         * whatever reason), try to create them using
1407                         * __block_write_begin.  If this fails,
1408                         * skip the page and move on.
1409                         */
1410                        if (!page_has_buffers(page)) {
1411                                if (__block_write_begin(page, 0, len,
1412                                                noalloc_get_block_write)) {
1413                                skip_page:
1414                                        unlock_page(page);
1415                                        continue;
1416                                }
1417                                commit_write = 1;
1418                        }
1419
1420                        bh = page_bufs = page_buffers(page);
1421                        block_start = 0;
1422                        do {
1423                                if (!bh)
1424                                        goto skip_page;
1425                                if (map && (cur_logical >= map->m_lblk) &&
1426                                    (cur_logical <= (map->m_lblk +
1427                                                     (map->m_len - 1)))) {
1428                                        if (buffer_delay(bh)) {
1429                                                clear_buffer_delay(bh);
1430                                                bh->b_blocknr = pblock;
1431                                        }
1432                                        if (buffer_da_mapped(bh))
1433                                                clear_buffer_da_mapped(bh);
1434                                        if (buffer_unwritten(bh) ||
1435                                            buffer_mapped(bh))
1436                                                BUG_ON(bh->b_blocknr != pblock);
1437                                        if (map->m_flags & EXT4_MAP_UNINIT)
1438                                                set_buffer_uninit(bh);
1439                                        clear_buffer_unwritten(bh);
1440                                }
1441
1442                                /*
1443                                 * skip page if block allocation undone and
1444                                 * block is dirty
1445                                 */
1446                                if (ext4_bh_delay_or_unwritten(NULL, bh))
1447                                        skip_page = 1;
1448                                bh = bh->b_this_page;
1449                                block_start += bh->b_size;
1450                                cur_logical++;
1451                                pblock++;
1452                        } while (bh != page_bufs);
1453
1454                        if (skip_page)
1455                                goto skip_page;
1456
1457                        if (commit_write)
1458                                /* mark the buffer_heads as dirty & uptodate */
1459                                block_commit_write(page, 0, len);
1460
1461                        clear_page_dirty_for_io(page);
1462                        /*
1463                         * Delalloc doesn't support data journalling,
1464                         * but eventually maybe we'll lift this
1465                         * restriction.
1466                         */
1467                        if (unlikely(journal_data && PageChecked(page)))
1468                                err = __ext4_journalled_writepage(page, len);
1469                        else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1470                                err = ext4_bio_write_page(&io_submit, page,
1471                                                          len, mpd->wbc);
1472                        else if (buffer_uninit(page_bufs)) {
1473                                ext4_set_bh_endio(page_bufs, inode);
1474                                err = block_write_full_page_endio(page,
1475                                        noalloc_get_block_write,
1476                                        mpd->wbc, ext4_end_io_buffer_write);
1477                        } else
1478                                err = block_write_full_page(page,
1479                                        noalloc_get_block_write, mpd->wbc);
1480
1481                        if (!err)
1482                                mpd->pages_written++;
1483                        /*
1484                         * In error case, we have to continue because
1485                         * remaining pages are still locked
1486                         */
1487                        if (ret == 0)
1488                                ret = err;
1489                }
1490                pagevec_release(&pvec);
1491        }
1492        ext4_io_submit(&io_submit);
1493        return ret;
1494}
1495
1496static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1497{
1498        int nr_pages, i;
1499        pgoff_t index, end;
1500        struct pagevec pvec;
1501        struct inode *inode = mpd->inode;
1502        struct address_space *mapping = inode->i_mapping;
1503
1504        index = mpd->first_page;
1505        end   = mpd->next_page - 1;
1506        while (index <= end) {
1507                nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1508                if (nr_pages == 0)
1509                        break;
1510                for (i = 0; i < nr_pages; i++) {
1511                        struct page *page = pvec.pages[i];
1512                        if (page->index > end)
1513                                break;
1514                        BUG_ON(!PageLocked(page));
1515                        BUG_ON(PageWriteback(page));
1516                        block_invalidatepage(page, 0);
1517                        ClearPageUptodate(page);
1518                        unlock_page(page);
1519                }
1520                index = pvec.pages[nr_pages - 1]->index + 1;
1521                pagevec_release(&pvec);
1522        }
1523        return;
1524}
1525
1526static void ext4_print_free_blocks(struct inode *inode)
1527{
1528        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1529        struct super_block *sb = inode->i_sb;
1530
1531        ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1532               EXT4_C2B(EXT4_SB(inode->i_sb),
1533                        ext4_count_free_clusters(inode->i_sb)));
1534        ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1535        ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1536               (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1537                percpu_counter_sum(&sbi->s_freeclusters_counter)));
1538        ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1539               (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1540                percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1541        ext4_msg(sb, KERN_CRIT, "Block reservation details");
1542        ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1543                 EXT4_I(inode)->i_reserved_data_blocks);
1544        ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1545               EXT4_I(inode)->i_reserved_meta_blocks);
1546        return;
1547}
1548
1549/*
1550 * mpage_da_map_and_submit - go through given space, map them
1551 *       if necessary, and then submit them for I/O
1552 *
1553 * @mpd - bh describing space
1554 *
1555 * The function skips space we know is already mapped to disk blocks.
1556 *
1557 */
1558static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1559{
1560        int err, blks, get_blocks_flags;
1561        struct ext4_map_blocks map, *mapp = NULL;
1562        sector_t next = mpd->b_blocknr;
1563        unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1564        loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1565        handle_t *handle = NULL;
1566
1567        /*
1568         * If the blocks are mapped already, or we couldn't accumulate
1569         * any blocks, then proceed immediately to the submission stage.
1570         */
1571        if ((mpd->b_size == 0) ||
1572            ((mpd->b_state  & (1 << BH_Mapped)) &&
1573             !(mpd->b_state & (1 << BH_Delay)) &&
1574             !(mpd->b_state & (1 << BH_Unwritten))))
1575                goto submit_io;
1576
1577        handle = ext4_journal_current_handle();
1578        BUG_ON(!handle);
1579
1580        /*
1581         * Call ext4_map_blocks() to allocate any delayed allocation
1582         * blocks, or to convert an uninitialized extent to be
1583         * initialized (in the case where we have written into
1584         * one or more preallocated blocks).
1585         *
1586         * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1587         * indicate that we are on the delayed allocation path.  This
1588         * affects functions in many different parts of the allocation
1589         * call path.  This flag exists primarily because we don't
1590         * want to change *many* call functions, so ext4_map_blocks()
1591         * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1592         * inode's allocation semaphore is taken.
1593         *
1594         * If the blocks in questions were delalloc blocks, set
1595         * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1596         * variables are updated after the blocks have been allocated.
1597         */
1598        map.m_lblk = next;
1599        map.m_len = max_blocks;
1600        get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1601        if (ext4_should_dioread_nolock(mpd->inode))
1602                get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1603        if (mpd->b_state & (1 << BH_Delay))
1604                get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1605
1606        blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1607        if (blks < 0) {
1608                struct super_block *sb = mpd->inode->i_sb;
1609
1610                err = blks;
1611                /*
1612                 * If get block returns EAGAIN or ENOSPC and there
1613                 * appears to be free blocks we will just let
1614                 * mpage_da_submit_io() unlock all of the pages.
1615                 */
1616                if (err == -EAGAIN)
1617                        goto submit_io;
1618
1619                if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1620                        mpd->retval = err;
1621                        goto submit_io;
1622                }
1623
1624                /*
1625                 * get block failure will cause us to loop in
1626                 * writepages, because a_ops->writepage won't be able
1627                 * to make progress. The page will be redirtied by
1628                 * writepage and writepages will again try to write
1629                 * the same.
1630                 */
1631                if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1632                        ext4_msg(sb, KERN_CRIT,
1633                                 "delayed block allocation failed for inode %lu "
1634                                 "at logical offset %llu with max blocks %zd "
1635                                 "with error %d", mpd->inode->i_ino,
1636                                 (unsigned long long) next,
1637                                 mpd->b_size >> mpd->inode->i_blkbits, err);
1638                        ext4_msg(sb, KERN_CRIT,
1639                                "This should not happen!! Data will be lost\n");
1640                        if (err == -ENOSPC)
1641                                ext4_print_free_blocks(mpd->inode);
1642                }
1643                /* invalidate all the pages */
1644                ext4_da_block_invalidatepages(mpd);
1645
1646                /* Mark this page range as having been completed */
1647                mpd->io_done = 1;
1648                return;
1649        }
1650        BUG_ON(blks == 0);
1651
1652        mapp = &map;
1653        if (map.m_flags & EXT4_MAP_NEW) {
1654                struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1655                int i;
1656
1657                for (i = 0; i < map.m_len; i++)
1658                        unmap_underlying_metadata(bdev, map.m_pblk + i);
1659
1660                if (ext4_should_order_data(mpd->inode)) {
1661                        err = ext4_jbd2_file_inode(handle, mpd->inode);
1662                        if (err) {
1663                                /* Only if the journal is aborted */
1664                                mpd->retval = err;
1665                                goto submit_io;
1666                        }
1667                }
1668        }
1669
1670        /*
1671         * Update on-disk size along with block allocation.
1672         */
1673        disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1674        if (disksize > i_size_read(mpd->inode))
1675                disksize = i_size_read(mpd->inode);
1676        if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1677                ext4_update_i_disksize(mpd->inode, disksize);
1678                err = ext4_mark_inode_dirty(handle, mpd->inode);
1679                if (err)
1680                        ext4_error(mpd->inode->i_sb,
1681                                   "Failed to mark inode %lu dirty",
1682                                   mpd->inode->i_ino);
1683        }
1684
1685submit_io:
1686        mpage_da_submit_io(mpd, mapp);
1687        mpd->io_done = 1;
1688}
1689
1690#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1691                (1 << BH_Delay) | (1 << BH_Unwritten))
1692
1693/*
1694 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1695 *
1696 * @mpd->lbh - extent of blocks
1697 * @logical - logical number of the block in the file
1698 * @bh - bh of the block (used to access block's state)
1699 *
1700 * the function is used to collect contig. blocks in same state
1701 */
1702static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1703                                   sector_t logical, size_t b_size,
1704                                   unsigned long b_state)
1705{
1706        sector_t next;
1707        int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1708
1709        /*
1710         * XXX Don't go larger than mballoc is willing to allocate
1711         * This is a stopgap solution.  We eventually need to fold
1712         * mpage_da_submit_io() into this function and then call
1713         * ext4_map_blocks() multiple times in a loop
1714         */
1715        if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1716                goto flush_it;
1717
1718        /* check if thereserved journal credits might overflow */
1719        if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1720                if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1721                        /*
1722                         * With non-extent format we are limited by the journal
1723                         * credit available.  Total credit needed to insert
1724                         * nrblocks contiguous blocks is dependent on the
1725                         * nrblocks.  So limit nrblocks.
1726                         */
1727                        goto flush_it;
1728                } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1729                                EXT4_MAX_TRANS_DATA) {
1730                        /*
1731                         * Adding the new buffer_head would make it cross the
1732                         * allowed limit for which we have journal credit
1733                         * reserved. So limit the new bh->b_size
1734                         */
1735                        b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1736                                                mpd->inode->i_blkbits;
1737                        /* we will do mpage_da_submit_io in the next loop */
1738                }
1739        }
1740        /*
1741         * First block in the extent
1742         */
1743        if (mpd->b_size == 0) {
1744                mpd->b_blocknr = logical;
1745                mpd->b_size = b_size;
1746                mpd->b_state = b_state & BH_FLAGS;
1747                return;
1748        }
1749
1750        next = mpd->b_blocknr + nrblocks;
1751        /*
1752         * Can we merge the block to our big extent?
1753         */
1754        if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1755                mpd->b_size += b_size;
1756                return;
1757        }
1758
1759flush_it:
1760        /*
1761         * We couldn't merge the block to our extent, so we
1762         * need to flush current  extent and start new one
1763         */
1764        mpage_da_map_and_submit(mpd);
1765        return;
1766}
1767
1768static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1769{
1770        return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1771}
1772
1773/*
1774 * This function is grabs code from the very beginning of
1775 * ext4_map_blocks, but assumes that the caller is from delayed write
1776 * time. This function looks up the requested blocks and sets the
1777 * buffer delay bit under the protection of i_data_sem.
1778 */
1779static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1780                              struct ext4_map_blocks *map,
1781                              struct buffer_head *bh)
1782{
1783        int retval;
1784        sector_t invalid_block = ~((sector_t) 0xffff);
1785
1786        if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1787                invalid_block = ~0;
1788
1789        map->m_flags = 0;
1790        ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1791                  "logical block %lu\n", inode->i_ino, map->m_len,
1792                  (unsigned long) map->m_lblk);
1793        /*
1794         * Try to see if we can get the block without requesting a new
1795         * file system block.
1796         */
1797        down_read((&EXT4_I(inode)->i_data_sem));
1798        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1799                retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1800        else
1801                retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1802
1803        if (retval == 0) {
1804                /*
1805                 * XXX: __block_prepare_write() unmaps passed block,
1806                 * is it OK?
1807                 */
1808                /* If the block was allocated from previously allocated cluster,
1809                 * then we dont need to reserve it again. */
1810                if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1811                        retval = ext4_da_reserve_space(inode, iblock);
1812                        if (retval)
1813                                /* not enough space to reserve */
1814                                goto out_unlock;
1815                }
1816
1817                /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1818                 * and it should not appear on the bh->b_state.
1819                 */
1820                map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1821
1822                map_bh(bh, inode->i_sb, invalid_block);
1823                set_buffer_new(bh);
1824                set_buffer_delay(bh);
1825        }
1826
1827out_unlock:
1828        up_read((&EXT4_I(inode)->i_data_sem));
1829
1830        return retval;
1831}
1832
1833/*
1834 * This is a special get_blocks_t callback which is used by
1835 * ext4_da_write_begin().  It will either return mapped block or
1836 * reserve space for a single block.
1837 *
1838 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1839 * We also have b_blocknr = -1 and b_bdev initialized properly
1840 *
1841 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1842 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1843 * initialized properly.
1844 */
1845static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1846                                  struct buffer_head *bh, int create)
1847{
1848        struct ext4_map_blocks map;
1849        int ret = 0;
1850
1851        BUG_ON(create == 0);
1852        BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1853
1854        map.m_lblk = iblock;
1855        map.m_len = 1;
1856
1857        /*
1858         * first, we need to know whether the block is allocated already
1859         * preallocated blocks are unmapped but should treated
1860         * the same as allocated blocks.
1861         */
1862        ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1863        if (ret <= 0)
1864                return ret;
1865
1866        map_bh(bh, inode->i_sb, map.m_pblk);
1867        bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1868
1869        if (buffer_unwritten(bh)) {
1870                /* A delayed write to unwritten bh should be marked
1871                 * new and mapped.  Mapped ensures that we don't do
1872                 * get_block multiple times when we write to the same
1873                 * offset and new ensures that we do proper zero out
1874                 * for partial write.
1875                 */
1876                set_buffer_new(bh);
1877                set_buffer_mapped(bh);
1878        }
1879        return 0;
1880}
1881
1882/*
1883 * This function is used as a standard get_block_t calback function
1884 * when there is no desire to allocate any blocks.  It is used as a
1885 * callback function for block_write_begin() and block_write_full_page().
1886 * These functions should only try to map a single block at a time.
1887 *
1888 * Since this function doesn't do block allocations even if the caller
1889 * requests it by passing in create=1, it is critically important that
1890 * any caller checks to make sure that any buffer heads are returned
1891 * by this function are either all already mapped or marked for
1892 * delayed allocation before calling  block_write_full_page().  Otherwise,
1893 * b_blocknr could be left unitialized, and the page write functions will
1894 * be taken by surprise.
1895 */
1896static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1897                                   struct buffer_head *bh_result, int create)
1898{
1899        BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1900        return _ext4_get_block(inode, iblock, bh_result, 0);
1901}
1902
1903static int bget_one(handle_t *handle, struct buffer_head *bh)
1904{
1905        get_bh(bh);
1906        return 0;
1907}
1908
1909static int bput_one(handle_t *handle, struct buffer_head *bh)
1910{
1911        put_bh(bh);
1912        return 0;
1913}
1914
1915static int __ext4_journalled_writepage(struct page *page,
1916                                       unsigned int len)
1917{
1918        struct address_space *mapping = page->mapping;
1919        struct inode *inode = mapping->host;
1920        struct buffer_head *page_bufs;
1921        handle_t *handle = NULL;
1922        int ret = 0;
1923        int err;
1924
1925        ClearPageChecked(page);
1926        page_bufs = page_buffers(page);
1927        BUG_ON(!page_bufs);
1928        walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1929        /* As soon as we unlock the page, it can go away, but we have
1930         * references to buffers so we are safe */
1931        unlock_page(page);
1932
1933        handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1934        if (IS_ERR(handle)) {
1935                ret = PTR_ERR(handle);
1936                goto out;
1937        }
1938
1939        BUG_ON(!ext4_handle_valid(handle));
1940
1941        ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1942                                do_journal_get_write_access);
1943
1944        err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1945                                write_end_fn);
1946        if (ret == 0)
1947                ret = err;
1948        EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1949        err = ext4_journal_stop(handle);
1950        if (!ret)
1951                ret = err;
1952
1953        walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1954        ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1955out:
1956        return ret;
1957}
1958
1959/*
1960 * Note that we don't need to start a transaction unless we're journaling data
1961 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1962 * need to file the inode to the transaction's list in ordered mode because if
1963 * we are writing back data added by write(), the inode is already there and if
1964 * we are writing back data modified via mmap(), no one guarantees in which
1965 * transaction the data will hit the disk. In case we are journaling data, we
1966 * cannot start transaction directly because transaction start ranks above page
1967 * lock so we have to do some magic.
1968 *
1969 * This function can get called via...
1970 *   - ext4_da_writepages after taking page lock (have journal handle)
1971 *   - journal_submit_inode_data_buffers (no journal handle)
1972 *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1973 *   - grab_page_cache when doing write_begin (have journal handle)
1974 *
1975 * We don't do any block allocation in this function. If we have page with
1976 * multiple blocks we need to write those buffer_heads that are mapped. This
1977 * is important for mmaped based write. So if we do with blocksize 1K
1978 * truncate(f, 1024);
1979 * a = mmap(f, 0, 4096);
1980 * a[0] = 'a';
1981 * truncate(f, 4096);
1982 * we have in the page first buffer_head mapped via page_mkwrite call back
1983 * but other buffer_heads would be unmapped but dirty (dirty done via the
1984 * do_wp_page). So writepage should write the first block. If we modify
1985 * the mmap area beyond 1024 we will again get a page_fault and the
1986 * page_mkwrite callback will do the block allocation and mark the
1987 * buffer_heads mapped.
1988 *
1989 * We redirty the page if we have any buffer_heads that is either delay or
1990 * unwritten in the page.
1991 *
1992 * We can get recursively called as show below.
1993 *
1994 *      ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1995 *              ext4_writepage()
1996 *
1997 * But since we don't do any block allocation we should not deadlock.
1998 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1999 */
2000static int ext4_writepage(struct page *page,
2001                          struct writeback_control *wbc)
2002{
2003        int ret = 0, commit_write = 0;
2004        loff_t size;
2005        unsigned int len;
2006        struct buffer_head *page_bufs = NULL;
2007        struct inode *inode = page->mapping->host;
2008
2009        trace_ext4_writepage(page);
2010        size = i_size_read(inode);
2011        if (page->index == size >> PAGE_CACHE_SHIFT)
2012                len = size & ~PAGE_CACHE_MASK;
2013        else
2014                len = PAGE_CACHE_SIZE;
2015
2016        /*
2017         * If the page does not have buffers (for whatever reason),
2018         * try to create them using __block_write_begin.  If this
2019         * fails, redirty the page and move on.
2020         */
2021        if (!page_has_buffers(page)) {
2022                if (__block_write_begin(page, 0, len,
2023                                        noalloc_get_block_write)) {
2024                redirty_page:
2025                        redirty_page_for_writepage(wbc, page);
2026                        unlock_page(page);
2027                        return 0;
2028                }
2029                commit_write = 1;
2030        }
2031        page_bufs = page_buffers(page);
2032        if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2033                              ext4_bh_delay_or_unwritten)) {
2034                /*
2035                 * We don't want to do block allocation, so redirty
2036                 * the page and return.  We may reach here when we do
2037                 * a journal commit via journal_submit_inode_data_buffers.
2038                 * We can also reach here via shrink_page_list but it
2039                 * should never be for direct reclaim so warn if that
2040                 * happens
2041                 */
2042                WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2043                                                                PF_MEMALLOC);
2044                goto redirty_page;
2045        }
2046        if (commit_write)
2047                /* now mark the buffer_heads as dirty and uptodate */
2048                block_commit_write(page, 0, len);
2049
2050        if (PageChecked(page) && ext4_should_journal_data(inode))
2051                /*
2052                 * It's mmapped pagecache.  Add buffers and journal it.  There
2053                 * doesn't seem much point in redirtying the page here.
2054                 */
2055                return __ext4_journalled_writepage(page, len);
2056
2057        if (buffer_uninit(page_bufs)) {
2058                ext4_set_bh_endio(page_bufs, inode);
2059                ret = block_write_full_page_endio(page, noalloc_get_block_write,
2060                                            wbc, ext4_end_io_buffer_write);
2061        } else
2062                ret = block_write_full_page(page, noalloc_get_block_write,
2063                                            wbc);
2064
2065        return ret;
2066}
2067
2068/*
2069 * This is called via ext4_da_writepages() to
2070 * calculate the total number of credits to reserve to fit
2071 * a single extent allocation into a single transaction,
2072 * ext4_da_writpeages() will loop calling this before
2073 * the block allocation.
2074 */
2075
2076static int ext4_da_writepages_trans_blocks(struct inode *inode)
2077{
2078        int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2079
2080        /*
2081         * With non-extent format the journal credit needed to
2082         * insert nrblocks contiguous block is dependent on
2083         * number of contiguous block. So we will limit
2084         * number of contiguous block to a sane value
2085         */
2086        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2087            (max_blocks > EXT4_MAX_TRANS_DATA))
2088                max_blocks = EXT4_MAX_TRANS_DATA;
2089
2090        return ext4_chunk_trans_blocks(inode, max_blocks);
2091}
2092
2093/*
2094 * write_cache_pages_da - walk the list of dirty pages of the given
2095 * address space and accumulate pages that need writing, and call
2096 * mpage_da_map_and_submit to map a single contiguous memory region
2097 * and then write them.
2098 */
2099static int write_cache_pages_da(struct address_space *mapping,
2100                                struct writeback_control *wbc,
2101                                struct mpage_da_data *mpd,
2102                                pgoff_t *done_index)
2103{
2104        struct buffer_head      *bh, *head;
2105        struct inode            *inode = mapping->host;
2106        struct pagevec          pvec;
2107        unsigned int            nr_pages;
2108        sector_t                logical;
2109        pgoff_t                 index, end;
2110        long                    nr_to_write = wbc->nr_to_write;
2111        int                     i, tag, ret = 0;
2112
2113        memset(mpd, 0, sizeof(struct mpage_da_data));
2114        mpd->wbc = wbc;
2115        mpd->inode = inode;
2116        pagevec_init(&pvec, 0);
2117        index = wbc->range_start >> PAGE_CACHE_SHIFT;
2118        end = wbc->range_end >> PAGE_CACHE_SHIFT;
2119
2120        if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2121                tag = PAGECACHE_TAG_TOWRITE;
2122        else
2123                tag = PAGECACHE_TAG_DIRTY;
2124
2125        *done_index = index;
2126        while (index <= end) {
2127                nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2128                              min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2129                if (nr_pages == 0)
2130                        return 0;
2131
2132                for (i = 0; i < nr_pages; i++) {
2133                        struct page *page = pvec.pages[i];
2134
2135                        /*
2136                         * At this point, the page may be truncated or
2137                         * invalidated (changing page->mapping to NULL), or
2138                         * even swizzled back from swapper_space to tmpfs file
2139                         * mapping. However, page->index will not change
2140                         * because we have a reference on the page.
2141                         */
2142                        if (page->index > end)
2143                                goto out;
2144
2145                        *done_index = page->index + 1;
2146
2147                        /*
2148                         * If we can't merge this page, and we have
2149                         * accumulated an contiguous region, write it
2150                         */
2151                        if ((mpd->next_page != page->index) &&
2152                            (mpd->next_page != mpd->first_page)) {
2153                                mpage_da_map_and_submit(mpd);
2154                                goto ret_extent_tail;
2155                        }
2156
2157                        lock_page(page);
2158
2159                        /*
2160                         * If the page is no longer dirty, or its
2161                         * mapping no longer corresponds to inode we
2162                         * are writing (which means it has been
2163                         * truncated or invalidated), or the page is
2164                         * already under writeback and we are not
2165                         * doing a data integrity writeback, skip the page
2166                         */
2167                        if (!PageDirty(page) ||
2168                            (PageWriteback(page) &&
2169                             (wbc->sync_mode == WB_SYNC_NONE)) ||
2170                            unlikely(page->mapping != mapping)) {
2171                                unlock_page(page);
2172                                continue;
2173                        }
2174
2175                        wait_on_page_writeback(page);
2176                        BUG_ON(PageWriteback(page));
2177
2178                        if (mpd->next_page != page->index)
2179                                mpd->first_page = page->index;
2180                        mpd->next_page = page->index + 1;
2181                        logical = (sector_t) page->index <<
2182                                (PAGE_CACHE_SHIFT - inode->i_blkbits);
2183
2184                        if (!page_has_buffers(page)) {
2185                                mpage_add_bh_to_extent(mpd, logical,
2186                                                       PAGE_CACHE_SIZE,
2187                                                       (1 << BH_Dirty) | (1 << BH_Uptodate));
2188                                if (mpd->io_done)
2189                                        goto ret_extent_tail;
2190                        } else {
2191                                /*
2192                                 * Page with regular buffer heads,
2193                                 * just add all dirty ones
2194                                 */
2195                                head = page_buffers(page);
2196                                bh = head;
2197                                do {
2198                                        BUG_ON(buffer_locked(bh));
2199                                        /*
2200                                         * We need to try to allocate
2201                                         * unmapped blocks in the same page.
2202                                         * Otherwise we won't make progress
2203                                         * with the page in ext4_writepage
2204                                         */
2205                                        if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2206                                                mpage_add_bh_to_extent(mpd, logical,
2207                                                                       bh->b_size,
2208                                                                       bh->b_state);
2209                                                if (mpd->io_done)
2210                                                        goto ret_extent_tail;
2211                                        } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2212                                                /*
2213                                                 * mapped dirty buffer. We need
2214                                                 * to update the b_state
2215                                                 * because we look at b_state
2216                                                 * in mpage_da_map_blocks.  We
2217                                                 * don't update b_size because
2218                                                 * if we find an unmapped
2219                                                 * buffer_head later we need to
2220                                                 * use the b_state flag of that
2221                                                 * buffer_head.
2222                                                 */
2223                                                if (mpd->b_size == 0)
2224                                                        mpd->b_state = bh->b_state & BH_FLAGS;
2225                                        }
2226                                        logical++;
2227                                } while ((bh = bh->b_this_page) != head);
2228                        }
2229
2230                        if (nr_to_write > 0) {
2231                                nr_to_write--;
2232                                if (nr_to_write == 0 &&
2233                                    wbc->sync_mode == WB_SYNC_NONE)
2234                                        /*
2235                                         * We stop writing back only if we are
2236                                         * not doing integrity sync. In case of
2237                                         * integrity sync we have to keep going
2238                                         * because someone may be concurrently
2239                                         * dirtying pages, and we might have
2240                                         * synced a lot of newly appeared dirty
2241                                         * pages, but have not synced all of the
2242                                         * old dirty pages.
2243                                         */
2244                                        goto out;
2245                        }
2246                }
2247                pagevec_release(&pvec);
2248                cond_resched();
2249        }
2250        return 0;
2251ret_extent_tail:
2252        ret = MPAGE_DA_EXTENT_TAIL;
2253out:
2254        pagevec_release(&pvec);
2255        cond_resched();
2256        return ret;
2257}
2258
2259
2260static int ext4_da_writepages(struct address_space *mapping,
2261                              struct writeback_control *wbc)
2262{
2263        pgoff_t index;
2264        int range_whole = 0;
2265        handle_t *handle = NULL;
2266        struct mpage_da_data mpd;
2267        struct inode *inode = mapping->host;
2268        int pages_written = 0;
2269        unsigned int max_pages;
2270        int range_cyclic, cycled = 1, io_done = 0;
2271        int needed_blocks, ret = 0;
2272        long desired_nr_to_write, nr_to_writebump = 0;
2273        loff_t range_start = wbc->range_start;
2274        struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2275        pgoff_t done_index = 0;
2276        pgoff_t end;
2277        struct blk_plug plug;
2278
2279        trace_ext4_da_writepages(inode, wbc);
2280
2281        /*
2282         * No pages to write? This is mainly a kludge to avoid starting
2283         * a transaction for special inodes like journal inode on last iput()
2284         * because that could violate lock ordering on umount
2285         */
2286        if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2287                return 0;
2288
2289        /*
2290         * If the filesystem has aborted, it is read-only, so return
2291         * right away instead of dumping stack traces later on that
2292         * will obscure the real source of the problem.  We test
2293         * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2294         * the latter could be true if the filesystem is mounted
2295         * read-only, and in that case, ext4_da_writepages should
2296         * *never* be called, so if that ever happens, we would want
2297         * the stack trace.
2298         */
2299        if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2300                return -EROFS;
2301
2302        if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2303                range_whole = 1;
2304
2305        range_cyclic = wbc->range_cyclic;
2306        if (wbc->range_cyclic) {
2307                index = mapping->writeback_index;
2308                if (index)
2309                        cycled = 0;
2310                wbc->range_start = index << PAGE_CACHE_SHIFT;
2311                wbc->range_end  = LLONG_MAX;
2312                wbc->range_cyclic = 0;
2313                end = -1;
2314        } else {
2315                index = wbc->range_start >> PAGE_CACHE_SHIFT;
2316                end = wbc->range_end >> PAGE_CACHE_SHIFT;
2317        }
2318
2319        /*
2320         * This works around two forms of stupidity.  The first is in
2321         * the writeback code, which caps the maximum number of pages
2322         * written to be 1024 pages.  This is wrong on multiple
2323         * levels; different architectues have a different page size,
2324         * which changes the maximum amount of data which gets
2325         * written.  Secondly, 4 megabytes is way too small.  XFS
2326         * forces this value to be 16 megabytes by multiplying
2327         * nr_to_write parameter by four, and then relies on its
2328         * allocator to allocate larger extents to make them
2329         * contiguous.  Unfortunately this brings us to the second
2330         * stupidity, which is that ext4's mballoc code only allocates
2331         * at most 2048 blocks.  So we force contiguous writes up to
2332         * the number of dirty blocks in the inode, or
2333         * sbi->max_writeback_mb_bump whichever is smaller.
2334         */
2335        max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2336        if (!range_cyclic && range_whole) {
2337                if (wbc->nr_to_write == LONG_MAX)
2338                        desired_nr_to_write = wbc->nr_to_write;
2339                else
2340                        desired_nr_to_write = wbc->nr_to_write * 8;
2341        } else
2342                desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2343                                                           max_pages);
2344        if (desired_nr_to_write > max_pages)
2345                desired_nr_to_write = max_pages;
2346
2347        if (wbc->nr_to_write < desired_nr_to_write) {
2348                nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2349                wbc->nr_to_write = desired_nr_to_write;
2350        }
2351
2352retry:
2353        if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2354                tag_pages_for_writeback(mapping, index, end);
2355
2356        blk_start_plug(&plug);
2357        while (!ret && wbc->nr_to_write > 0) {
2358
2359                /*
2360                 * we  insert one extent at a time. So we need
2361                 * credit needed for single extent allocation.
2362                 * journalled mode is currently not supported
2363                 * by delalloc
2364                 */
2365                BUG_ON(ext4_should_journal_data(inode));
2366                needed_blocks = ext4_da_writepages_trans_blocks(inode);
2367
2368                /* start a new transaction*/
2369                handle = ext4_journal_start(inode, needed_blocks);
2370                if (IS_ERR(handle)) {
2371                        ret = PTR_ERR(handle);
2372                        ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2373                               "%ld pages, ino %lu; err %d", __func__,
2374                                wbc->nr_to_write, inode->i_ino, ret);
2375                        blk_finish_plug(&plug);
2376                        goto out_writepages;
2377                }
2378
2379                /*
2380                 * Now call write_cache_pages_da() to find the next
2381                 * contiguous region of logical blocks that need
2382                 * blocks to be allocated by ext4 and submit them.
2383                 */
2384                ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2385                /*
2386                 * If we have a contiguous extent of pages and we
2387                 * haven't done the I/O yet, map the blocks and submit
2388                 * them for I/O.
2389                 */
2390                if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2391                        mpage_da_map_and_submit(&mpd);
2392                        ret = MPAGE_DA_EXTENT_TAIL;
2393                }
2394                trace_ext4_da_write_pages(inode, &mpd);
2395                wbc->nr_to_write -= mpd.pages_written;
2396
2397                ext4_journal_stop(handle);
2398
2399                if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2400                        /* commit the transaction which would
2401                         * free blocks released in the transaction
2402                         * and try again
2403                         */
2404                        jbd2_journal_force_commit_nested(sbi->s_journal);
2405                        ret = 0;
2406                } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2407                        /*
2408                         * Got one extent now try with rest of the pages.
2409                         * If mpd.retval is set -EIO, journal is aborted.
2410                         * So we don't need to write any more.
2411                         */
2412                        pages_written += mpd.pages_written;
2413                        ret = mpd.retval;
2414                        io_done = 1;
2415                } else if (wbc->nr_to_write)
2416                        /*
2417                         * There is no more writeout needed
2418                         * or we requested for a noblocking writeout
2419                         * and we found the device congested
2420                         */
2421                        break;
2422        }
2423        blk_finish_plug(&plug);
2424        if (!io_done && !cycled) {
2425                cycled = 1;
2426                index = 0;
2427                wbc->range_start = index << PAGE_CACHE_SHIFT;
2428                wbc->range_end  = mapping->writeback_index - 1;
2429                goto retry;
2430        }
2431
2432        /* Update index */
2433        wbc->range_cyclic = range_cyclic;
2434        if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2435                /*
2436                 * set the writeback_index so that range_cyclic
2437                 * mode will write it back later
2438                 */
2439                mapping->writeback_index = done_index;
2440
2441out_writepages:
2442        wbc->nr_to_write -= nr_to_writebump;
2443        wbc->range_start = range_start;
2444        trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2445        return ret;
2446}
2447
2448#define FALL_BACK_TO_NONDELALLOC 1
2449static int ext4_nonda_switch(struct super_block *sb)
2450{
2451        s64 free_blocks, dirty_blocks;
2452        struct ext4_sb_info *sbi = EXT4_SB(sb);
2453
2454        /*
2455         * switch to non delalloc mode if we are running low
2456         * on free block. The free block accounting via percpu
2457         * counters can get slightly wrong with percpu_counter_batch getting
2458         * accumulated on each CPU without updating global counters
2459         * Delalloc need an accurate free block accounting. So switch
2460         * to non delalloc when we are near to error range.
2461         */
2462        free_blocks  = EXT4_C2B(sbi,
2463                percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2464        dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2465        /*
2466         * Start pushing delalloc when 1/2 of free blocks are dirty.
2467         */
2468        if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2469            !writeback_in_progress(sb->s_bdi) &&
2470            down_read_trylock(&sb->s_umount)) {
2471                writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2472                up_read(&sb->s_umount);
2473        }
2474
2475        if (2 * free_blocks < 3 * dirty_blocks ||
2476                free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2477                /*
2478                 * free block count is less than 150% of dirty blocks
2479                 * or free blocks is less than watermark
2480                 */
2481                return 1;
2482        }
2483        return 0;
2484}
2485
2486static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2487                               loff_t pos, unsigned len, unsigned flags,
2488                               struct page **pagep, void **fsdata)
2489{
2490        int ret, retries = 0;
2491        struct page *page;
2492        pgoff_t index;
2493        struct inode *inode = mapping->host;
2494        handle_t *handle;
2495
2496        index = pos >> PAGE_CACHE_SHIFT;
2497
2498        if (ext4_nonda_switch(inode->i_sb)) {
2499                *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2500                return ext4_write_begin(file, mapping, pos,
2501                                        len, flags, pagep, fsdata);
2502        }
2503        *fsdata = (void *)0;
2504        trace_ext4_da_write_begin(inode, pos, len, flags);
2505retry:
2506        /*
2507         * With delayed allocation, we don't log the i_disksize update
2508         * if there is delayed block allocation. But we still need
2509         * to journalling the i_disksize update if writes to the end
2510         * of file which has an already mapped buffer.
2511         */
2512        handle = ext4_journal_start(inode, 1);
2513        if (IS_ERR(handle)) {
2514                ret = PTR_ERR(handle);
2515                goto out;
2516        }
2517        /* We cannot recurse into the filesystem as the transaction is already
2518         * started */
2519        flags |= AOP_FLAG_NOFS;
2520
2521        page = grab_cache_page_write_begin(mapping, index, flags);
2522        if (!page) {
2523                ext4_journal_stop(handle);
2524                ret = -ENOMEM;
2525                goto out;
2526        }
2527        *pagep = page;
2528
2529        ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2530        if (ret < 0) {
2531                unlock_page(page);
2532                ext4_journal_stop(handle);
2533                page_cache_release(page);
2534                /*
2535                 * block_write_begin may have instantiated a few blocks
2536                 * outside i_size.  Trim these off again. Don't need
2537                 * i_size_read because we hold i_mutex.
2538                 */
2539                if (pos + len > inode->i_size)
2540                        ext4_truncate_failed_write(inode);
2541        }
2542
2543        if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2544                goto retry;
2545out:
2546        return ret;
2547}
2548
2549/*
2550 * Check if we should update i_disksize
2551 * when write to the end of file but not require block allocation
2552 */
2553static int ext4_da_should_update_i_disksize(struct page *page,
2554                                            unsigned long offset)
2555{
2556        struct buffer_head *bh;
2557        struct inode *inode = page->mapping->host;
2558        unsigned int idx;
2559        int i;
2560
2561        bh = page_buffers(page);
2562        idx = offset >> inode->i_blkbits;
2563
2564        for (i = 0; i < idx; i++)
2565                bh = bh->b_this_page;
2566
2567        if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2568                return 0;
2569        return 1;
2570}
2571
2572static int ext4_da_write_end(struct file *file,
2573                             struct address_space *mapping,
2574                             loff_t pos, unsigned len, unsigned copied,
2575                             struct page *page, void *fsdata)
2576{
2577        struct inode *inode = mapping->host;
2578        int ret = 0, ret2;
2579        handle_t *handle = ext4_journal_current_handle();
2580        loff_t new_i_size;
2581        unsigned long start, end;
2582        int write_mode = (int)(unsigned long)fsdata;
2583
2584        if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2585                switch (ext4_inode_journal_mode(inode)) {
2586                case EXT4_INODE_ORDERED_DATA_MODE:
2587                        return ext4_ordered_write_end(file, mapping, pos,
2588                                        len, copied, page, fsdata);
2589                case EXT4_INODE_WRITEBACK_DATA_MODE:
2590                        return ext4_writeback_write_end(file, mapping, pos,
2591                                        len, copied, page, fsdata);
2592                default:
2593                        BUG();
2594                }
2595        }
2596
2597        trace_ext4_da_write_end(inode, pos, len, copied);
2598        start = pos & (PAGE_CACHE_SIZE - 1);
2599        end = start + copied - 1;
2600
2601        /*
2602         * generic_write_end() will run mark_inode_dirty() if i_size
2603         * changes.  So let's piggyback the i_disksize mark_inode_dirty
2604         * into that.
2605         */
2606
2607        new_i_size = pos + copied;
2608        if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2609                if (ext4_da_should_update_i_disksize(page, end)) {
2610                        down_write(&EXT4_I(inode)->i_data_sem);
2611                        if (new_i_size > EXT4_I(inode)->i_disksize) {
2612                                /*
2613                                 * Updating i_disksize when extending file
2614                                 * without needing block allocation
2615                                 */
2616                                if (ext4_should_order_data(inode))
2617                                        ret = ext4_jbd2_file_inode(handle,
2618                                                                   inode);
2619
2620                                EXT4_I(inode)->i_disksize = new_i_size;
2621                        }
2622                        up_write(&EXT4_I(inode)->i_data_sem);
2623                        /* We need to mark inode dirty even if
2624                         * new_i_size is less that inode->i_size
2625                         * bu greater than i_disksize.(hint delalloc)
2626                         */
2627                        ext4_mark_inode_dirty(handle, inode);
2628                }
2629        }
2630        ret2 = generic_write_end(file, mapping, pos, len, copied,
2631                                                        page, fsdata);
2632        copied = ret2;
2633        if (ret2 < 0)
2634                ret = ret2;
2635        ret2 = ext4_journal_stop(handle);
2636        if (!ret)
2637                ret = ret2;
2638
2639        return ret ? ret : copied;
2640}
2641
2642static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2643{
2644        /*
2645         * Drop reserved blocks
2646         */
2647        BUG_ON(!PageLocked(page));
2648        if (!page_has_buffers(page))
2649                goto out;
2650
2651        ext4_da_page_release_reservation(page, offset);
2652
2653out:
2654        ext4_invalidatepage(page, offset);
2655
2656        return;
2657}
2658
2659/*
2660 * Force all delayed allocation blocks to be allocated for a given inode.
2661 */
2662int ext4_alloc_da_blocks(struct inode *inode)
2663{
2664        trace_ext4_alloc_da_blocks(inode);
2665
2666        if (!EXT4_I(inode)->i_reserved_data_blocks &&
2667            !EXT4_I(inode)->i_reserved_meta_blocks)
2668                return 0;
2669
2670        /*
2671         * We do something simple for now.  The filemap_flush() will
2672         * also start triggering a write of the data blocks, which is
2673         * not strictly speaking necessary (and for users of
2674         * laptop_mode, not even desirable).  However, to do otherwise
2675         * would require replicating code paths in:
2676         *
2677         * ext4_da_writepages() ->
2678         *    write_cache_pages() ---> (via passed in callback function)
2679         *        __mpage_da_writepage() -->
2680         *           mpage_add_bh_to_extent()
2681         *           mpage_da_map_blocks()
2682         *
2683         * The problem is that write_cache_pages(), located in
2684         * mm/page-writeback.c, marks pages clean in preparation for
2685         * doing I/O, which is not desirable if we're not planning on
2686         * doing I/O at all.
2687         *
2688         * We could call write_cache_pages(), and then redirty all of
2689         * the pages by calling redirty_page_for_writepage() but that
2690         * would be ugly in the extreme.  So instead we would need to
2691         * replicate parts of the code in the above functions,
2692         * simplifying them because we wouldn't actually intend to
2693         * write out the pages, but rather only collect contiguous
2694         * logical block extents, call the multi-block allocator, and
2695         * then update the buffer heads with the block allocations.
2696         *
2697         * For now, though, we'll cheat by calling filemap_flush(),
2698         * which will map the blocks, and start the I/O, but not
2699         * actually wait for the I/O to complete.
2700         */
2701        return filemap_flush(inode->i_mapping);
2702}
2703
2704/*
2705 * bmap() is special.  It gets used by applications such as lilo and by
2706 * the swapper to find the on-disk block of a specific piece of data.
2707 *
2708 * Naturally, this is dangerous if the block concerned is still in the
2709 * journal.  If somebody makes a swapfile on an ext4 data-journaling
2710 * filesystem and enables swap, then they may get a nasty shock when the
2711 * data getting swapped to that swapfile suddenly gets overwritten by
2712 * the original zero's written out previously to the journal and
2713 * awaiting writeback in the kernel's buffer cache.
2714 *
2715 * So, if we see any bmap calls here on a modified, data-journaled file,
2716 * take extra steps to flush any blocks which might be in the cache.
2717 */
2718static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2719{
2720        struct inode *inode = mapping->host;
2721        journal_t *journal;
2722        int err;
2723
2724        if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2725                        test_opt(inode->i_sb, DELALLOC)) {
2726                /*
2727                 * With delalloc we want to sync the file
2728                 * so that we can make sure we allocate
2729                 * blocks for file
2730                 */
2731                filemap_write_and_wait(mapping);
2732        }
2733
2734        if (EXT4_JOURNAL(inode) &&
2735            ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2736                /*
2737                 * This is a REALLY heavyweight approach, but the use of
2738                 * bmap on dirty files is expected to be extremely rare:
2739                 * only if we run lilo or swapon on a freshly made file
2740                 * do we expect this to happen.
2741                 *
2742                 * (bmap requires CAP_SYS_RAWIO so this does not
2743                 * represent an unprivileged user DOS attack --- we'd be
2744                 * in trouble if mortal users could trigger this path at
2745                 * will.)
2746                 *
2747                 * NB. EXT4_STATE_JDATA is not set on files other than
2748                 * regular files.  If somebody wants to bmap a directory
2749                 * or symlink and gets confused because the buffer
2750                 * hasn't yet been flushed to disk, they deserve
2751                 * everything they get.
2752                 */
2753
2754                ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2755                journal = EXT4_JOURNAL(inode);
2756                jbd2_journal_lock_updates(journal);
2757                err = jbd2_journal_flush(journal);
2758                jbd2_journal_unlock_updates(journal);
2759
2760                if (err)
2761                        return 0;
2762        }
2763
2764        return generic_block_bmap(mapping, block, ext4_get_block);
2765}
2766
2767static int ext4_readpage(struct file *file, struct page *page)
2768{
2769        trace_ext4_readpage(page);
2770        return mpage_readpage(page, ext4_get_block);
2771}
2772
2773static int
2774ext4_readpages(struct file *file, struct address_space *mapping,
2775                struct list_head *pages, unsigned nr_pages)
2776{
2777        return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2778}
2779
2780static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2781{
2782        struct buffer_head *head, *bh;
2783        unsigned int curr_off = 0;
2784
2785        if (!page_has_buffers(page))
2786                return;
2787        head = bh = page_buffers(page);
2788        do {
2789                if (offset <= curr_off && test_clear_buffer_uninit(bh)
2790                                        && bh->b_private) {
2791                        ext4_free_io_end(bh->b_private);
2792                        bh->b_private = NULL;
2793                        bh->b_end_io = NULL;
2794                }
2795                curr_off = curr_off + bh->b_size;
2796                bh = bh->b_this_page;
2797        } while (bh != head);
2798}
2799
2800static void ext4_invalidatepage(struct page *page, unsigned long offset)
2801{
2802        journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2803
2804        trace_ext4_invalidatepage(page, offset);
2805
2806        /*
2807         * free any io_end structure allocated for buffers to be discarded
2808         */
2809        if (ext4_should_dioread_nolock(page->mapping->host))
2810                ext4_invalidatepage_free_endio(page, offset);
2811        /*
2812         * If it's a full truncate we just forget about the pending dirtying
2813         */
2814        if (offset == 0)
2815                ClearPageChecked(page);
2816
2817        if (journal)
2818                jbd2_journal_invalidatepage(journal, page, offset);
2819        else
2820                block_invalidatepage(page, offset);
2821}
2822
2823static int ext4_releasepage(struct page *page, gfp_t wait)
2824{
2825        journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2826
2827        trace_ext4_releasepage(page);
2828
2829        WARN_ON(PageChecked(page));
2830        if (!page_has_buffers(page))
2831                return 0;
2832        if (journal)
2833                return jbd2_journal_try_to_free_buffers(journal, page, wait);
2834        else
2835                return try_to_free_buffers(page);
2836}
2837
2838/*
2839 * ext4_get_block used when preparing for a DIO write or buffer write.
2840 * We allocate an uinitialized extent if blocks haven't been allocated.
2841 * The extent will be converted to initialized after the IO is complete.
2842 */
2843static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2844                   struct buffer_head *bh_result, int create)
2845{
2846        ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2847                   inode->i_ino, create);
2848        return _ext4_get_block(inode, iblock, bh_result,
2849                               EXT4_GET_BLOCKS_IO_CREATE_EXT);
2850}
2851
2852static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2853                   struct buffer_head *bh_result, int flags)
2854{
2855        handle_t *handle = ext4_journal_current_handle();
2856        struct ext4_map_blocks map;
2857        int ret = 0;
2858
2859        ext4_debug("ext4_get_block_write_nolock: inode %lu, flag %d\n",
2860                   inode->i_ino, flags);
2861
2862        flags = EXT4_GET_BLOCKS_NO_LOCK;
2863
2864        map.m_lblk = iblock;
2865        map.m_len = bh_result->b_size >> inode->i_blkbits;
2866
2867        ret = ext4_map_blocks(handle, inode, &map, flags);
2868        if (ret > 0) {
2869                map_bh(bh_result, inode->i_sb, map.m_pblk);
2870                bh_result->b_state = (bh_result->b_state & ~EXT4_MAP_FLAGS) |
2871                                        map.m_flags;
2872                bh_result->b_size = inode->i_sb->s_blocksize * map.m_len;
2873                ret = 0;
2874        }
2875        return ret;
2876}
2877
2878static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2879                            ssize_t size, void *private, int ret,
2880                            bool is_async)
2881{
2882        struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2883        ext4_io_end_t *io_end = iocb->private;
2884
2885        /* if not async direct IO or dio with 0 bytes write, just return */
2886        if (!io_end || !size)
2887                goto out;
2888
2889        ext_debug("ext4_end_io_dio(): io_end 0x%p "
2890                  "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2891                  iocb->private, io_end->inode->i_ino, iocb, offset,
2892                  size);
2893
2894        iocb->private = NULL;
2895
2896        /* if not aio dio with unwritten extents, just free io and return */
2897        if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2898                ext4_free_io_end(io_end);
2899out:
2900                if (is_async)
2901                        aio_complete(iocb, ret, 0);
2902                inode_dio_done(inode);
2903                return;
2904        }
2905
2906        io_end->offset = offset;
2907        io_end->size = size;
2908        if (is_async) {
2909                io_end->iocb = iocb;
2910                io_end->result = ret;
2911        }
2912
2913        ext4_add_complete_io(io_end);
2914}
2915
2916static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2917{
2918        ext4_io_end_t *io_end = bh->b_private;
2919        struct inode *inode;
2920
2921        if (!test_clear_buffer_uninit(bh) || !io_end)
2922                goto out;
2923
2924        if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2925                ext4_msg(io_end->inode->i_sb, KERN_INFO,
2926                         "sb umounted, discard end_io request for inode %lu",
2927                         io_end->inode->i_ino);
2928                ext4_free_io_end(io_end);
2929                goto out;
2930        }
2931
2932        /*
2933         * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2934         * but being more careful is always safe for the future change.
2935         */
2936        inode = io_end->inode;
2937        ext4_set_io_unwritten_flag(inode, io_end);
2938        ext4_add_complete_io(io_end);
2939out:
2940        bh->b_private = NULL;
2941        bh->b_end_io = NULL;
2942        clear_buffer_uninit(bh);
2943        end_buffer_async_write(bh, uptodate);
2944}
2945
2946static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2947{
2948        ext4_io_end_t *io_end;
2949        struct page *page = bh->b_page;
2950        loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2951        size_t size = bh->b_size;
2952
2953retry:
2954        io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2955        if (!io_end) {
2956                pr_warn_ratelimited("%s: allocation fail\n", __func__);
2957                schedule();
2958                goto retry;
2959        }
2960        io_end->offset = offset;
2961        io_end->size = size;
2962        /*
2963         * We need to hold a reference to the page to make sure it
2964         * doesn't get evicted before ext4_end_io_work() has a chance
2965         * to convert the extent from written to unwritten.
2966         */
2967        io_end->page = page;
2968        get_page(io_end->page);
2969
2970        bh->b_private = io_end;
2971        bh->b_end_io = ext4_end_io_buffer_write;
2972        return 0;
2973}
2974
2975/*
2976 * For ext4 extent files, ext4 will do direct-io write to holes,
2977 * preallocated extents, and those write extend the file, no need to
2978 * fall back to buffered IO.
2979 *
2980 * For holes, we fallocate those blocks, mark them as uninitialized
2981 * If those blocks were preallocated, we mark sure they are splited, but
2982 * still keep the range to write as uninitialized.
2983 *
2984 * The unwrritten extents will be converted to written when DIO is completed.
2985 * For async direct IO, since the IO may still pending when return, we
2986 * set up an end_io call back function, which will do the conversion
2987 * when async direct IO completed.
2988 *
2989 * If the O_DIRECT write will extend the file then add this inode to the
2990 * orphan list.  So recovery will truncate it back to the original size
2991 * if the machine crashes during the write.
2992 *
2993 */
2994static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2995                              const struct iovec *iov, loff_t offset,
2996                              unsigned long nr_segs)
2997{
2998        struct file *file = iocb->ki_filp;
2999        struct inode *inode = file->f_mapping->host;
3000        ssize_t ret;
3001        size_t count = iov_length(iov, nr_segs);
3002
3003        loff_t final_size = offset + count;
3004        if (rw == WRITE && final_size <= inode->i_size) {
3005                int overwrite = 0;
3006
3007                BUG_ON(iocb->private == NULL);
3008
3009                /* If we do a overwrite dio, i_mutex locking can be released */
3010                overwrite = *((int *)iocb->private);
3011
3012                if (overwrite) {
3013                        atomic_inc(&inode->i_dio_count);
3014                        down_read(&EXT4_I(inode)->i_data_sem);
3015                        mutex_unlock(&inode->i_mutex);
3016                }
3017
3018                /*
3019                 * We could direct write to holes and fallocate.
3020                 *
3021                 * Allocated blocks to fill the hole are marked as uninitialized
3022                 * to prevent parallel buffered read to expose the stale data
3023                 * before DIO complete the data IO.
3024                 *
3025                 * As to previously fallocated extents, ext4 get_block
3026                 * will just simply mark the buffer mapped but still
3027                 * keep the extents uninitialized.
3028                 *
3029                 * for non AIO case, we will convert those unwritten extents
3030                 * to written after return back from blockdev_direct_IO.
3031                 *
3032                 * for async DIO, the conversion needs to be defered when
3033                 * the IO is completed. The ext4 end_io callback function
3034                 * will be called to take care of the conversion work.
3035                 * Here for async case, we allocate an io_end structure to
3036                 * hook to the iocb.
3037                 */
3038                iocb->private = NULL;
3039                ext4_inode_aio_set(inode, NULL);
3040                if (!is_sync_kiocb(iocb)) {
3041                        ext4_io_end_t *io_end =
3042                                ext4_init_io_end(inode, GFP_NOFS);
3043                        if (!io_end) {
3044                                ret = -ENOMEM;
3045                                goto retake_lock;
3046                        }
3047                        io_end->flag |= EXT4_IO_END_DIRECT;
3048                        iocb->private = io_end;
3049                        /*
3050                         * we save the io structure for current async
3051                         * direct IO, so that later ext4_map_blocks()
3052                         * could flag the io structure whether there
3053                         * is a unwritten extents needs to be converted
3054                         * when IO is completed.
3055                         */
3056                        ext4_inode_aio_set(inode, io_end);
3057                }
3058
3059                if (overwrite)
3060                        ret = __blockdev_direct_IO(rw, iocb, inode,
3061                                                 inode->i_sb->s_bdev, iov,
3062                                                 offset, nr_segs,
3063                                                 ext4_get_block_write_nolock,
3064                                                 ext4_end_io_dio,
3065                                                 NULL,
3066                                                 0);
3067                else
3068                        ret = __blockdev_direct_IO(rw, iocb, inode,
3069                                                 inode->i_sb->s_bdev, iov,
3070                                                 offset, nr_segs,
3071                                                 ext4_get_block_write,
3072                                                 ext4_end_io_dio,
3073                                                 NULL,
3074                                                 DIO_LOCKING);
3075                if (iocb->private)
3076                        ext4_inode_aio_set(inode, NULL);
3077                /*
3078                 * The io_end structure takes a reference to the inode,
3079                 * that structure needs to be destroyed and the
3080                 * reference to the inode need to be dropped, when IO is
3081                 * complete, even with 0 byte write, or failed.
3082                 *
3083                 * In the successful AIO DIO case, the io_end structure will be
3084                 * desctroyed and the reference to the inode will be dropped
3085                 * after the end_io call back function is called.
3086                 *
3087                 * In the case there is 0 byte write, or error case, since
3088                 * VFS direct IO won't invoke the end_io call back function,
3089                 * we need to free the end_io structure here.
3090                 */
3091                if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3092                        ext4_free_io_end(iocb->private);
3093                        iocb->private = NULL;
3094                } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3095                                                EXT4_STATE_DIO_UNWRITTEN)) {
3096                        int err;
3097                        /*
3098                         * for non AIO case, since the IO is already
3099                         * completed, we could do the conversion right here
3100                         */
3101                        err = ext4_convert_unwritten_extents(inode,
3102                                                             offset, ret);
3103                        if (err < 0)
3104                                ret = err;
3105                        ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3106                }
3107
3108        retake_lock:
3109                /* take i_mutex locking again if we do a ovewrite dio */
3110                if (overwrite) {
3111                        inode_dio_done(inode);
3112                        up_read(&EXT4_I(inode)->i_data_sem);
3113                        mutex_lock(&inode->i_mutex);
3114                }
3115
3116                return ret;
3117        }
3118
3119        /* for write the the end of file case, we fall back to old way */
3120        return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3121}
3122
3123static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3124                              const struct iovec *iov, loff_t offset,
3125                              unsigned long nr_segs)
3126{
3127        struct file *file = iocb->ki_filp;
3128        struct inode *inode = file->f_mapping->host;
3129        ssize_t ret;
3130
3131        /*
3132         * If we are doing data journalling we don't support O_DIRECT
3133         */
3134        if (ext4_should_journal_data(inode))
3135                return 0;
3136
3137        trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3138        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3139                ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3140        else
3141                ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3142        trace_ext4_direct_IO_exit(inode, offset,
3143                                iov_length(iov, nr_segs), rw, ret);
3144        return ret;
3145}
3146
3147/*
3148 * Pages can be marked dirty completely asynchronously from ext4's journalling
3149 * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3150 * much here because ->set_page_dirty is called under VFS locks.  The page is
3151 * not necessarily locked.
3152 *
3153 * We cannot just dirty the page and leave attached buffers clean, because the
3154 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3155 * or jbddirty because all the journalling code will explode.
3156 *
3157 * So what we do is to mark the page "pending dirty" and next time writepage
3158 * is called, propagate that into the buffers appropriately.
3159 */
3160static int ext4_journalled_set_page_dirty(struct page *page)
3161{
3162        SetPageChecked(page);
3163        return __set_page_dirty_nobuffers(page);
3164}
3165
3166static const struct address_space_operations ext4_ordered_aops = {
3167        .readpage               = ext4_readpage,
3168        .readpages              = ext4_readpages,
3169        .writepage              = ext4_writepage,
3170        .write_begin            = ext4_write_begin,
3171        .write_end              = ext4_ordered_write_end,
3172        .bmap                   = ext4_bmap,
3173        .invalidatepage         = ext4_invalidatepage,
3174        .releasepage            = ext4_releasepage,
3175        .direct_IO              = ext4_direct_IO,
3176        .migratepage            = buffer_migrate_page,
3177        .is_partially_uptodate  = block_is_partially_uptodate,
3178        .error_remove_page      = generic_error_remove_page,
3179};
3180
3181static const struct address_space_operations ext4_writeback_aops = {
3182        .readpage               = ext4_readpage,
3183        .readpages              = ext4_readpages,
3184        .writepage              = ext4_writepage,
3185        .write_begin            = ext4_write_begin,
3186        .write_end              = ext4_writeback_write_end,
3187        .bmap                   = ext4_bmap,
3188        .invalidatepage         = ext4_invalidatepage,
3189        .releasepage            = ext4_releasepage,
3190        .direct_IO              = ext4_direct_IO,
3191        .migratepage            = buffer_migrate_page,
3192        .is_partially_uptodate  = block_is_partially_uptodate,
3193        .error_remove_page      = generic_error_remove_page,
3194};
3195
3196static const struct address_space_operations ext4_journalled_aops = {
3197        .readpage               = ext4_readpage,
3198        .readpages              = ext4_readpages,
3199        .writepage              = ext4_writepage,
3200        .write_begin            = ext4_write_begin,
3201        .write_end              = ext4_journalled_write_end,
3202        .set_page_dirty         = ext4_journalled_set_page_dirty,
3203        .bmap                   = ext4_bmap,
3204        .invalidatepage         = ext4_invalidatepage,
3205        .releasepage            = ext4_releasepage,
3206        .direct_IO              = ext4_direct_IO,
3207        .is_partially_uptodate  = block_is_partially_uptodate,
3208        .error_remove_page      = generic_error_remove_page,
3209};
3210
3211static const struct address_space_operations ext4_da_aops = {
3212        .readpage               = ext4_readpage,
3213        .readpages              = ext4_readpages,
3214        .writepage              = ext4_writepage,
3215        .writepages             = ext4_da_writepages,
3216        .write_begin            = ext4_da_write_begin,
3217        .write_end              = ext4_da_write_end,
3218        .bmap                   = ext4_bmap,
3219        .invalidatepage         = ext4_da_invalidatepage,
3220        .releasepage            = ext4_releasepage,
3221        .direct_IO              = ext4_direct_IO,
3222        .migratepage            = buffer_migrate_page,
3223        .is_partially_uptodate  = block_is_partially_uptodate,
3224        .error_remove_page      = generic_error_remove_page,
3225};
3226
3227void ext4_set_aops(struct inode *inode)
3228{
3229        switch (ext4_inode_journal_mode(inode)) {
3230        case EXT4_INODE_ORDERED_DATA_MODE:
3231                if (test_opt(inode->i_sb, DELALLOC))
3232                        inode->i_mapping->a_ops = &ext4_da_aops;
3233                else
3234                        inode->i_mapping->a_ops = &ext4_ordered_aops;
3235                break;
3236        case EXT4_INODE_WRITEBACK_DATA_MODE:
3237                if (test_opt(inode->i_sb, DELALLOC))
3238                        inode->i_mapping->a_ops = &ext4_da_aops;
3239                else
3240                        inode->i_mapping->a_ops = &ext4_writeback_aops;
3241                break;
3242        case EXT4_INODE_JOURNAL_DATA_MODE:
3243                inode->i_mapping->a_ops = &ext4_journalled_aops;
3244                break;
3245        default:
3246                BUG();
3247        }
3248}
3249
3250
3251/*
3252 * ext4_discard_partial_page_buffers()
3253 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3254 * This function finds and locks the page containing the offset
3255 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3256 * Calling functions that already have the page locked should call
3257 * ext4_discard_partial_page_buffers_no_lock directly.
3258 */
3259int ext4_discard_partial_page_buffers(handle_t *handle,
3260                struct address_space *mapping, loff_t from,
3261                loff_t length, int flags)
3262{
3263        struct inode *inode = mapping->host;
3264        struct page *page;
3265        int err = 0;
3266
3267        page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3268                                   mapping_gfp_mask(mapping) & ~__GFP_FS);
3269        if (!page)
3270                return -ENOMEM;
3271
3272        err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3273                from, length, flags);
3274
3275        unlock_page(page);
3276        page_cache_release(page);
3277        return err;
3278}
3279
3280/*
3281 * ext4_discard_partial_page_buffers_no_lock()
3282 * Zeros a page range of length 'length' starting from offset 'from'.
3283 * Buffer heads that correspond to the block aligned regions of the
3284 * zeroed range will be unmapped.  Unblock aligned regions
3285 * will have the corresponding buffer head mapped if needed so that
3286 * that region of the page can be updated with the partial zero out.
3287 *
3288 * This function assumes that the page has already been  locked.  The
3289 * The range to be discarded must be contained with in the given page.
3290 * If the specified range exceeds the end of the page it will be shortened
3291 * to the end of the page that corresponds to 'from'.  This function is
3292 * appropriate for updating a page and it buffer heads to be unmapped and
3293 * zeroed for blocks that have been either released, or are going to be
3294 * released.
3295 *
3296 * handle: The journal handle
3297 * inode:  The files inode
3298 * page:   A locked page that contains the offset "from"
3299 * from:   The starting byte offset (from the beginning of the file)
3300 *         to begin discarding
3301 * len:    The length of bytes to discard
3302 * flags:  Optional flags that may be used:
3303 *
3304 *         EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3305 *         Only zero the regions of the page whose buffer heads
3306 *         have already been unmapped.  This flag is appropriate
3307 *         for updating the contents of a page whose blocks may
3308 *         have already been released, and we only want to zero
3309 *         out the regions that correspond to those released blocks.
3310 *
3311 * Returns zero on success or negative on failure.
3312 */
3313static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3314                struct inode *inode, struct page *page, loff_t from,
3315                loff_t length, int flags)
3316{
3317        ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3318        unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3319        unsigned int blocksize, max, pos;
3320        ext4_lblk_t iblock;
3321        struct buffer_head *bh;
3322        int err = 0;
3323
3324        blocksize = inode->i_sb->s_blocksize;
3325        max = PAGE_CACHE_SIZE - offset;
3326
3327        if (index != page->index)
3328                return -EINVAL;
3329
3330        /*
3331         * correct length if it does not fall between
3332         * 'from' and the end of the page
3333         */
3334        if (length > max || length < 0)
3335                length = max;
3336
3337        iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3338
3339        if (!page_has_buffers(page))
3340                create_empty_buffers(page, blocksize, 0);
3341
3342        /* Find the buffer that contains "offset" */
3343        bh = page_buffers(page);
3344        pos = blocksize;
3345        while (offset >= pos) {
3346                bh = bh->b_this_page;
3347                iblock++;
3348                pos += blocksize;
3349        }
3350
3351        pos = offset;
3352        while (pos < offset + length) {
3353                unsigned int end_of_block, range_to_discard;
3354
3355                err = 0;
3356
3357                /* The length of space left to zero and unmap */
3358                range_to_discard = offset + length - pos;
3359
3360                /* The length of space until the end of the block */
3361                end_of_block = blocksize - (pos & (blocksize-1));
3362
3363                /*
3364                 * Do not unmap or zero past end of block
3365                 * for this buffer head
3366                 */
3367                if (range_to_discard > end_of_block)
3368                        range_to_discard = end_of_block;
3369
3370
3371                /*
3372                 * Skip this buffer head if we are only zeroing unampped
3373                 * regions of the page
3374                 */
3375                if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3376                        buffer_mapped(bh))
3377                                goto next;
3378
3379                /* If the range is block aligned, unmap */
3380                if (range_to_discard == blocksize) {
3381                        clear_buffer_dirty(bh);
3382                        bh->b_bdev = NULL;
3383                        clear_buffer_mapped(bh);
3384                        clear_buffer_req(bh);
3385                        clear_buffer_new(bh);
3386                        clear_buffer_delay(bh);
3387                        clear_buffer_unwritten(bh);
3388                        clear_buffer_uptodate(bh);
3389                        zero_user(page, pos, range_to_discard);
3390                        BUFFER_TRACE(bh, "Buffer discarded");
3391                        goto next;
3392                }
3393
3394                /*
3395                 * If this block is not completely contained in the range
3396                 * to be discarded, then it is not going to be released. Because
3397                 * we need to keep this block, we need to make sure this part
3398                 * of the page is uptodate before we modify it by writeing
3399                 * partial zeros on it.
3400                 */
3401                if (!buffer_mapped(bh)) {
3402                        /*
3403                         * Buffer head must be mapped before we can read
3404                         * from the block
3405                         */
3406                        BUFFER_TRACE(bh, "unmapped");
3407                        ext4_get_block(inode, iblock, bh, 0);
3408                        /* unmapped? It's a hole - nothing to do */
3409                        if (!buffer_mapped(bh)) {
3410                                BUFFER_TRACE(bh, "still unmapped");
3411                                goto next;
3412                        }
3413                }
3414
3415                /* Ok, it's mapped. Make sure it's up-to-date */
3416                if (PageUptodate(page))
3417                        set_buffer_uptodate(bh);
3418
3419                if (!buffer_uptodate(bh)) {
3420                        err = -EIO;
3421                        ll_rw_block(READ, 1, &bh);
3422                        wait_on_buffer(bh);
3423                        /* Uhhuh. Read error. Complain and punt.*/
3424                        if (!buffer_uptodate(bh))
3425                                goto next;
3426                }
3427
3428                if (ext4_should_journal_data(inode)) {
3429                        BUFFER_TRACE(bh, "get write access");
3430                        err = ext4_journal_get_write_access(handle, bh);
3431                        if (err)
3432                                goto next;
3433                }
3434
3435                zero_user(page, pos, range_to_discard);
3436
3437                err = 0;
3438                if (ext4_should_journal_data(inode)) {
3439                        err = ext4_handle_dirty_metadata(handle, inode, bh);
3440                } else
3441                        mark_buffer_dirty(bh);
3442
3443                BUFFER_TRACE(bh, "Partial buffer zeroed");
3444next:
3445                bh = bh->b_this_page;
3446                iblock++;
3447                pos += range_to_discard;
3448        }
3449
3450        return err;
3451}
3452
3453int ext4_can_truncate(struct inode *inode)
3454{
3455        if (S_ISREG(inode->i_mode))
3456                return 1;
3457        if (S_ISDIR(inode->i_mode))
3458                return 1;
3459        if (S_ISLNK(inode->i_mode))
3460                return !ext4_inode_is_fast_symlink(inode);
3461        return 0;
3462}
3463
3464/*
3465 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3466 * associated with the given offset and length
3467 *
3468 * @inode:  File inode
3469 * @offset: The offset where the hole will begin
3470 * @len:    The length of the hole
3471 *
3472 * Returns: 0 on success or negative on failure
3473 */
3474
3475int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3476{
3477        struct inode *inode = file->f_path.dentry->d_inode;
3478        if (!S_ISREG(inode->i_mode))
3479                return -EOPNOTSUPP;
3480
3481        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3482                /* TODO: Add support for non extent hole punching */
3483                return -EOPNOTSUPP;
3484        }
3485
3486        if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3487                /* TODO: Add support for bigalloc file systems */
3488                return -EOPNOTSUPP;
3489        }
3490
3491        return ext4_ext_punch_hole(file, offset, length);
3492}
3493
3494/*
3495 * ext4_truncate()
3496 *
3497 * We block out ext4_get_block() block instantiations across the entire
3498 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3499 * simultaneously on behalf of the same inode.
3500 *
3501 * As we work through the truncate and commit bits of it to the journal there
3502 * is one core, guiding principle: the file's tree must always be consistent on
3503 * disk.  We must be able to restart the truncate after a crash.
3504 *
3505 * The file's tree may be transiently inconsistent in memory (although it
3506 * probably isn't), but whenever we close off and commit a journal transaction,
3507 * the contents of (the filesystem + the journal) must be consistent and
3508 * restartable.  It's pretty simple, really: bottom up, right to left (although
3509 * left-to-right works OK too).
3510 *
3511 * Note that at recovery time, journal replay occurs *before* the restart of
3512 * truncate against the orphan inode list.
3513 *
3514 * The committed inode has the new, desired i_size (which is the same as
3515 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
3516 * that this inode's truncate did not complete and it will again call
3517 * ext4_truncate() to have another go.  So there will be instantiated blocks
3518 * to the right of the truncation point in a crashed ext4 filesystem.  But
3519 * that's fine - as long as they are linked from the inode, the post-crash
3520 * ext4_truncate() run will find them and release them.
3521 */
3522void ext4_truncate(struct inode *inode)
3523{
3524        trace_ext4_truncate_enter(inode);
3525
3526        if (!ext4_can_truncate(inode))
3527                return;
3528
3529        ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3530
3531        if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3532                ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3533
3534        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3535                ext4_ext_truncate(inode);
3536        else
3537                ext4_ind_truncate(inode);
3538
3539        trace_ext4_truncate_exit(inode);
3540}
3541
3542/*
3543 * ext4_get_inode_loc returns with an extra refcount against the inode's
3544 * underlying buffer_head on success. If 'in_mem' is true, we have all
3545 * data in memory that is needed to recreate the on-disk version of this
3546 * inode.
3547 */
3548static int __ext4_get_inode_loc(struct inode *inode,
3549                                struct ext4_iloc *iloc, int in_mem)
3550{
3551        struct ext4_group_desc  *gdp;
3552        struct buffer_head      *bh;
3553        struct super_block      *sb = inode->i_sb;
3554        ext4_fsblk_t            block;
3555        int                     inodes_per_block, inode_offset;
3556
3557        iloc->bh = NULL;
3558        if (!ext4_valid_inum(sb, inode->i_ino))
3559                return -EIO;
3560
3561        iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3562        gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3563        if (!gdp)
3564                return -EIO;
3565
3566        /*
3567         * Figure out the offset within the block group inode table
3568         */
3569        inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3570        inode_offset = ((inode->i_ino - 1) %
3571                        EXT4_INODES_PER_GROUP(sb));
3572        block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3573        iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3574
3575        bh = sb_getblk(sb, block);
3576        if (!bh) {
3577                EXT4_ERROR_INODE_BLOCK(inode, block,
3578                                       "unable to read itable block");
3579                return -EIO;
3580        }
3581        if (!buffer_uptodate(bh)) {
3582                lock_buffer(bh);
3583
3584                /*
3585                 * If the buffer has the write error flag, we have failed
3586                 * to write out another inode in the same block.  In this
3587                 * case, we don't have to read the block because we may
3588                 * read the old inode data successfully.
3589                 */
3590                if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3591                        set_buffer_uptodate(bh);
3592
3593                if (buffer_uptodate(bh)) {
3594                        /* someone brought it uptodate while we waited */
3595                        unlock_buffer(bh);
3596                        goto has_buffer;
3597                }
3598
3599                /*
3600                 * If we have all information of the inode in memory and this
3601                 * is the only valid inode in the block, we need not read the
3602                 * block.
3603                 */
3604                if (in_mem) {
3605                        struct buffer_head *bitmap_bh;
3606                        int i, start;
3607
3608                        start = inode_offset & ~(inodes_per_block - 1);
3609
3610                        /* Is the inode bitmap in cache? */
3611                        bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3612                        if (!bitmap_bh)
3613                                goto make_io;
3614
3615                        /*
3616                         * If the inode bitmap isn't in cache then the
3617                         * optimisation may end up performing two reads instead
3618                         * of one, so skip it.
3619                         */
3620                        if (!buffer_uptodate(bitmap_bh)) {
3621                                brelse(bitmap_bh);
3622                                goto make_io;
3623                        }
3624                        for (i = start; i < start + inodes_per_block; i++) {
3625                                if (i == inode_offset)
3626                                        continue;
3627                                if (ext4_test_bit(i, bitmap_bh->b_data))
3628                                        break;
3629                        }
3630                        brelse(bitmap_bh);
3631                        if (i == start + inodes_per_block) {
3632                                /* all other inodes are free, so skip I/O */
3633                                memset(bh->b_data, 0, bh->b_size);
3634                                set_buffer_uptodate(bh);
3635                                unlock_buffer(bh);
3636                                goto has_buffer;
3637                        }
3638                }
3639
3640make_io:
3641                /*
3642                 * If we need to do any I/O, try to pre-readahead extra
3643                 * blocks from the inode table.
3644                 */
3645                if (EXT4_SB(sb)->s_inode_readahead_blks) {
3646                        ext4_fsblk_t b, end, table;
3647                        unsigned num;
3648
3649                        table = ext4_inode_table(sb, gdp);
3650                        /* s_inode_readahead_blks is always a power of 2 */
3651                        b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3652                        if (table > b)
3653                                b = table;
3654                        end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3655                        num = EXT4_INODES_PER_GROUP(sb);
3656                        if (ext4_has_group_desc_csum(sb))
3657                                num -= ext4_itable_unused_count(sb, gdp);
3658                        table += num / inodes_per_block;
3659                        if (end > table)
3660                                end = table;
3661                        while (b <= end)
3662                                sb_breadahead(sb, b++);
3663                }
3664
3665                /*
3666                 * There are other valid inodes in the buffer, this inode
3667                 * has in-inode xattrs, or we don't have this inode in memory.
3668                 * Read the block from disk.
3669                 */
3670                trace_ext4_load_inode(inode);
3671                get_bh(bh);
3672                bh->b_end_io = end_buffer_read_sync;
3673                submit_bh(READ | REQ_META | REQ_PRIO, bh);
3674                wait_on_buffer(bh);
3675                if (!buffer_uptodate(bh)) {
3676                        EXT4_ERROR_INODE_BLOCK(inode, block,
3677                                               "unable to read itable block");
3678                        brelse(bh);
3679                        return -EIO;
3680                }
3681        }
3682has_buffer:
3683        iloc->bh = bh;
3684        return 0;
3685}
3686
3687int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3688{
3689        /* We have all inode data except xattrs in memory here. */
3690        return __ext4_get_inode_loc(inode, iloc,
3691                !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3692}
3693
3694void ext4_set_inode_flags(struct inode *inode)
3695{
3696        unsigned int flags = EXT4_I(inode)->i_flags;
3697
3698        inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3699        if (flags & EXT4_SYNC_FL)
3700                inode->i_flags |= S_SYNC;
3701        if (flags & EXT4_APPEND_FL)
3702                inode->i_flags |= S_APPEND;
3703        if (flags & EXT4_IMMUTABLE_FL)
3704                inode->i_flags |= S_IMMUTABLE;
3705        if (flags & EXT4_NOATIME_FL)
3706                inode->i_flags |= S_NOATIME;
3707        if (flags & EXT4_DIRSYNC_FL)
3708                inode->i_flags |= S_DIRSYNC;
3709}
3710
3711/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3712void ext4_get_inode_flags(struct ext4_inode_info *ei)
3713{
3714        unsigned int vfs_fl;
3715        unsigned long old_fl, new_fl;
3716
3717        do {
3718                vfs_fl = ei->vfs_inode.i_flags;
3719                old_fl = ei->i_flags;
3720                new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3721                                EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3722                                EXT4_DIRSYNC_FL);
3723                if (vfs_fl & S_SYNC)
3724                        new_fl |= EXT4_SYNC_FL;
3725                if (vfs_fl & S_APPEND)
3726                        new_fl |= EXT4_APPEND_FL;
3727                if (vfs_fl & S_IMMUTABLE)
3728                        new_fl |= EXT4_IMMUTABLE_FL;
3729                if (vfs_fl & S_NOATIME)
3730                        new_fl |= EXT4_NOATIME_FL;
3731                if (vfs_fl & S_DIRSYNC)
3732                        new_fl |= EXT4_DIRSYNC_FL;
3733        } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3734}
3735
3736static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3737                                  struct ext4_inode_info *ei)
3738{
3739        blkcnt_t i_blocks ;
3740        struct inode *inode = &(ei->vfs_inode);
3741        struct super_block *sb = inode->i_sb;
3742
3743        if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3744                                EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3745                /* we are using combined 48 bit field */
3746                i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3747                                        le32_to_cpu(raw_inode->i_blocks_lo);
3748                if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3749                        /* i_blocks represent file system block size */
3750                        return i_blocks  << (inode->i_blkbits - 9);
3751                } else {
3752                        return i_blocks;
3753                }
3754        } else {
3755                return le32_to_cpu(raw_inode->i_blocks_lo);
3756        }
3757}
3758
3759struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3760{
3761        struct ext4_iloc iloc;
3762        struct ext4_inode *raw_inode;
3763        struct ext4_inode_info *ei;
3764        struct inode *inode;
3765        journal_t *journal = EXT4_SB(sb)->s_journal;
3766        long ret;
3767        int block;
3768        uid_t i_uid;
3769        gid_t i_gid;
3770
3771        inode = iget_locked(sb, ino);
3772        if (!inode)
3773                return ERR_PTR(-ENOMEM);
3774        if (!(inode->i_state & I_NEW))
3775                return inode;
3776
3777        ei = EXT4_I(inode);
3778        iloc.bh = NULL;
3779
3780        ret = __ext4_get_inode_loc(inode, &iloc, 0);
3781        if (ret < 0)
3782                goto bad_inode;
3783        raw_inode = ext4_raw_inode(&iloc);
3784
3785        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3786                ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3787                if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3788                    EXT4_INODE_SIZE(inode->i_sb)) {
3789                        EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3790                                EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3791                                EXT4_INODE_SIZE(inode->i_sb));
3792                        ret = -EIO;
3793                        goto bad_inode;
3794                }
3795        } else
3796                ei->i_extra_isize = 0;
3797
3798        /* Precompute checksum seed for inode metadata */
3799        if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3800                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3801                struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3802                __u32 csum;
3803                __le32 inum = cpu_to_le32(inode->i_ino);
3804                __le32 gen = raw_inode->i_generation;
3805                csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3806                                   sizeof(inum));
3807                ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3808                                              sizeof(gen));
3809        }
3810
3811        if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3812                EXT4_ERROR_INODE(inode, "checksum invalid");
3813                ret = -EIO;
3814                goto bad_inode;
3815        }
3816
3817        inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3818        i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3819        i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3820        if (!(test_opt(inode->i_sb, NO_UID32))) {
3821                i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3822                i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3823        }
3824        i_uid_write(inode, i_uid);
3825        i_gid_write(inode, i_gid);
3826        set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3827
3828        ext4_clear_state_flags(ei);     /* Only relevant on 32-bit archs */
3829        ei->i_dir_start_lookup = 0;
3830        ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3831        /* We now have enough fields to check if the inode was active or not.
3832         * This is needed because nfsd might try to access dead inodes
3833         * the test is that same one that e2fsck uses
3834         * NeilBrown 1999oct15
3835         */
3836        if (inode->i_nlink == 0) {
3837                if (inode->i_mode == 0 ||
3838                    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3839                        /* this inode is deleted */
3840                        ret = -ESTALE;
3841                        goto bad_inode;
3842                }
3843                /* The only unlinked inodes we let through here have
3844                 * valid i_mode and are being read by the orphan
3845                 * recovery code: that's fine, we're about to complete
3846                 * the process of deleting those. */
3847        }
3848        ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3849        inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3850        ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3851        if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3852                ei->i_file_acl |=
3853                        ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3854        inode->i_size = ext4_isize(raw_inode);
3855        ei->i_disksize = inode->i_size;
3856#ifdef CONFIG_QUOTA
3857        ei->i_reserved_quota = 0;
3858#endif
3859        inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3860        ei->i_block_group = iloc.block_group;
3861        ei->i_last_alloc_group = ~0;
3862        /*
3863         * NOTE! The in-memory inode i_data array is in little-endian order
3864         * even on big-endian machines: we do NOT byteswap the block numbers!
3865         */
3866        for (block = 0; block < EXT4_N_BLOCKS; block++)
3867                ei->i_data[block] = raw_inode->i_block[block];
3868        INIT_LIST_HEAD(&ei->i_orphan);
3869
3870        /*
3871         * Set transaction id's of transactions that have to be committed
3872         * to finish f[data]sync. We set them to currently running transaction
3873         * as we cannot be sure that the inode or some of its metadata isn't
3874         * part of the transaction - the inode could have been reclaimed and
3875         * now it is reread from disk.
3876         */
3877        if (journal) {
3878                transaction_t *transaction;
3879                tid_t tid;
3880
3881                read_lock(&journal->j_state_lock);
3882                if (journal->j_running_transaction)
3883                        transaction = journal->j_running_transaction;
3884                else
3885                        transaction = journal->j_committing_transaction;
3886                if (transaction)
3887                        tid = transaction->t_tid;
3888                else
3889                        tid = journal->j_commit_sequence;
3890                read_unlock(&journal->j_state_lock);
3891                ei->i_sync_tid = tid;
3892                ei->i_datasync_tid = tid;
3893        }
3894
3895        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3896                if (ei->i_extra_isize == 0) {
3897                        /* The extra space is currently unused. Use it. */
3898                        ei->i_extra_isize = sizeof(struct ext4_inode) -
3899                                            EXT4_GOOD_OLD_INODE_SIZE;
3900                } else {
3901                        __le32 *magic = (void *)raw_inode +
3902                                        EXT4_GOOD_OLD_INODE_SIZE +
3903                                        ei->i_extra_isize;
3904                        if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3905                                ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3906                }
3907        }
3908
3909        EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3910        EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3911        EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3912        EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3913
3914        inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3915        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3916                if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3917                        inode->i_version |=
3918                        (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3919        }
3920
3921        ret = 0;
3922        if (ei->i_file_acl &&
3923            !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3924                EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3925                                 ei->i_file_acl);
3926                ret = -EIO;
3927                goto bad_inode;
3928        } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3929                if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3930                    (S_ISLNK(inode->i_mode) &&
3931                     !ext4_inode_is_fast_symlink(inode)))
3932                        /* Validate extent which is part of inode */
3933                        ret = ext4_ext_check_inode(inode);
3934        } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3935                   (S_ISLNK(inode->i_mode) &&
3936                    !ext4_inode_is_fast_symlink(inode))) {
3937                /* Validate block references which are part of inode */
3938                ret = ext4_ind_check_inode(inode);
3939        }
3940        if (ret)
3941                goto bad_inode;
3942
3943        if (S_ISREG(inode->i_mode)) {
3944                inode->i_op = &ext4_file_inode_operations;
3945                inode->i_fop = &ext4_file_operations;
3946                ext4_set_aops(inode);
3947        } else if (S_ISDIR(inode->i_mode)) {
3948                inode->i_op = &ext4_dir_inode_operations;
3949                inode->i_fop = &ext4_dir_operations;
3950        } else if (S_ISLNK(inode->i_mode)) {
3951                if (ext4_inode_is_fast_symlink(inode)) {
3952                        inode->i_op = &ext4_fast_symlink_inode_operations;
3953                        nd_terminate_link(ei->i_data, inode->i_size,
3954                                sizeof(ei->i_data) - 1);
3955                } else {
3956                        inode->i_op = &ext4_symlink_inode_operations;
3957                        ext4_set_aops(inode);
3958                }
3959        } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3960              S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3961                inode->i_op = &ext4_special_inode_operations;
3962                if (raw_inode->i_block[0])
3963                        init_special_inode(inode, inode->i_mode,
3964                           old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3965                else
3966                        init_special_inode(inode, inode->i_mode,
3967                           new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3968        } else {
3969                ret = -EIO;
3970                EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3971                goto bad_inode;
3972        }
3973        brelse(iloc.bh);
3974        ext4_set_inode_flags(inode);
3975        unlock_new_inode(inode);
3976        return inode;
3977
3978bad_inode:
3979        brelse(iloc.bh);
3980        iget_failed(inode);
3981        return ERR_PTR(ret);
3982}
3983
3984static int ext4_inode_blocks_set(handle_t *handle,
3985                                struct ext4_inode *raw_inode,
3986                                struct ext4_inode_info *ei)
3987{
3988        struct inode *inode = &(ei->vfs_inode);
3989        u64 i_blocks = inode->i_blocks;
3990        struct super_block *sb = inode->i_sb;
3991
3992        if (i_blocks <= ~0U) {
3993                /*
3994                 * i_blocks can be represented in a 32 bit variable
3995                 * as multiple of 512 bytes
3996                 */
3997                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3998                raw_inode->i_blocks_high = 0;
3999                ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4000                return 0;
4001        }
4002        if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4003                return -EFBIG;
4004
4005        if (i_blocks <= 0xffffffffffffULL) {
4006                /*
4007                 * i_blocks can be represented in a 48 bit variable
4008                 * as multiple of 512 bytes
4009                 */
4010                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4011                raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4012                ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4013        } else {
4014                ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4015                /* i_block is stored in file system block size */
4016                i_blocks = i_blocks >> (inode->i_blkbits - 9);
4017                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4018                raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4019        }
4020        return 0;
4021}
4022
4023/*
4024 * Post the struct inode info into an on-disk inode location in the
4025 * buffer-cache.  This gobbles the caller's reference to the
4026 * buffer_head in the inode location struct.
4027 *
4028 * The caller must have write access to iloc->bh.
4029 */
4030static int ext4_do_update_inode(handle_t *handle,
4031                                struct inode *inode,
4032                                struct ext4_iloc *iloc)
4033{
4034        struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4035        struct ext4_inode_info *ei = EXT4_I(inode);
4036        struct buffer_head *bh = iloc->bh;
4037        int err = 0, rc, block;
4038        int need_datasync = 0;
4039        uid_t i_uid;
4040        gid_t i_gid;
4041
4042        /* For fields not not tracking in the in-memory inode,
4043         * initialise them to zero for new inodes. */
4044        if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4045                memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4046
4047        ext4_get_inode_flags(ei);
4048        raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4049        i_uid = i_uid_read(inode);
4050        i_gid = i_gid_read(inode);
4051        if (!(test_opt(inode->i_sb, NO_UID32))) {
4052                raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4053                raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4054/*
4055 * Fix up interoperability with old kernels. Otherwise, old inodes get
4056 * re-used with the upper 16 bits of the uid/gid intact
4057 */
4058                if (!ei->i_dtime) {
4059                        raw_inode->i_uid_high =
4060                                cpu_to_le16(high_16_bits(i_uid));
4061                        raw_inode->i_gid_high =
4062                                cpu_to_le16(high_16_bits(i_gid));
4063                } else {
4064                        raw_inode->i_uid_high = 0;
4065                        raw_inode->i_gid_high = 0;
4066                }
4067        } else {
4068                raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4069                raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4070                raw_inode->i_uid_high = 0;
4071                raw_inode->i_gid_high = 0;
4072        }
4073        raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4074
4075        EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4076        EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4077        EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4078        EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4079
4080        if (ext4_inode_blocks_set(handle, raw_inode, ei))
4081                goto out_brelse;
4082        raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4083        raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4084        if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4085            cpu_to_le32(EXT4_OS_HURD))
4086                raw_inode->i_file_acl_high =
4087                        cpu_to_le16(ei->i_file_acl >> 32);
4088        raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4089        if (ei->i_disksize != ext4_isize(raw_inode)) {
4090                ext4_isize_set(raw_inode, ei->i_disksize);
4091                need_datasync = 1;
4092        }
4093        if (ei->i_disksize > 0x7fffffffULL) {
4094                struct super_block *sb = inode->i_sb;
4095                if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4096                                EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4097                                EXT4_SB(sb)->s_es->s_rev_level ==
4098                                cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4099                        /* If this is the first large file
4100                         * created, add a flag to the superblock.
4101                         */
4102                        err = ext4_journal_get_write_access(handle,
4103                                        EXT4_SB(sb)->s_sbh);
4104                        if (err)
4105                                goto out_brelse;
4106                        ext4_update_dynamic_rev(sb);
4107                        EXT4_SET_RO_COMPAT_FEATURE(sb,
4108                                        EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4109                        ext4_handle_sync(handle);
4110                        err = ext4_handle_dirty_super(handle, sb);
4111                }
4112        }
4113        raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4114        if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4115                if (old_valid_dev(inode->i_rdev)) {
4116                        raw_inode->i_block[0] =
4117                                cpu_to_le32(old_encode_dev(inode->i_rdev));
4118                        raw_inode->i_block[1] = 0;
4119                } else {
4120                        raw_inode->i_block[0] = 0;
4121                        raw_inode->i_block[1] =
4122                                cpu_to_le32(new_encode_dev(inode->i_rdev));
4123                        raw_inode->i_block[2] = 0;
4124                }
4125        } else
4126                for (block = 0; block < EXT4_N_BLOCKS; block++)
4127                        raw_inode->i_block[block] = ei->i_data[block];
4128
4129        raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4130        if (ei->i_extra_isize) {
4131                if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4132                        raw_inode->i_version_hi =
4133                        cpu_to_le32(inode->i_version >> 32);
4134                raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4135        }
4136
4137        ext4_inode_csum_set(inode, raw_inode, ei);
4138
4139        BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4140        rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4141        if (!err)
4142                err = rc;
4143        ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4144
4145        ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4146out_brelse:
4147        brelse(bh);
4148        ext4_std_error(inode->i_sb, err);
4149        return err;
4150}
4151
4152/*
4153 * ext4_write_inode()
4154 *
4155 * We are called from a few places:
4156 *
4157 * - Within generic_file_write() for O_SYNC files.
4158 *   Here, there will be no transaction running. We wait for any running
4159 *   transaction to commit.
4160 *
4161 * - Within sys_sync(), kupdate and such.
4162 *   We wait on commit, if tol to.
4163 *
4164 * - Within prune_icache() (PF_MEMALLOC == true)
4165 *   Here we simply return.  We can't afford to block kswapd on the
4166 *   journal commit.
4167 *
4168 * In all cases it is actually safe for us to return without doing anything,
4169 * because the inode has been copied into a raw inode buffer in
4170 * ext4_mark_inode_dirty().  This is a correctness thing for O_SYNC and for
4171 * knfsd.
4172 *
4173 * Note that we are absolutely dependent upon all inode dirtiers doing the
4174 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4175 * which we are interested.
4176 *
4177 * It would be a bug for them to not do this.  The code:
4178 *
4179 *      mark_inode_dirty(inode)
4180 *      stuff();
4181 *      inode->i_size = expr;
4182 *
4183 * is in error because a kswapd-driven write_inode() could occur while
4184 * `stuff()' is running, and the new i_size will be lost.  Plus the inode
4185 * will no longer be on the superblock's dirty inode list.
4186 */
4187int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4188{
4189        int err;
4190
4191        if (current->flags & PF_MEMALLOC)
4192                return 0;
4193
4194        if (EXT4_SB(inode->i_sb)->s_journal) {
4195                if (ext4_journal_current_handle()) {
4196                        jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4197                        dump_stack();
4198                        return -EIO;
4199                }
4200
4201                if (wbc->sync_mode != WB_SYNC_ALL)
4202                        return 0;
4203
4204                err = ext4_force_commit(inode->i_sb);
4205        } else {
4206                struct ext4_iloc iloc;
4207
4208                err = __ext4_get_inode_loc(inode, &iloc, 0);
4209                if (err)
4210                        return err;
4211                if (wbc->sync_mode == WB_SYNC_ALL)
4212                        sync_dirty_buffer(iloc.bh);
4213                if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4214                        EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4215                                         "IO error syncing inode");
4216                        err = -EIO;
4217                }
4218                brelse(iloc.bh);
4219        }
4220        return err;
4221}
4222
4223/*
4224 * ext4_setattr()
4225 *
4226 * Called from notify_change.
4227 *
4228 * We want to trap VFS attempts to truncate the file as soon as
4229 * possible.  In particular, we want to make sure that when the VFS
4230 * shrinks i_size, we put the inode on the orphan list and modify
4231 * i_disksize immediately, so that during the subsequent flushing of
4232 * dirty pages and freeing of disk blocks, we can guarantee that any
4233 * commit will leave the blocks being flushed in an unused state on
4234 * disk.  (On recovery, the inode will get truncated and the blocks will
4235 * be freed, so we have a strong guarantee that no future commit will
4236 * leave these blocks visible to the user.)
4237 *
4238 * Another thing we have to assure is that if we are in ordered mode
4239 * and inode is still attached to the committing transaction, we must
4240 * we start writeout of all the dirty pages which are being truncated.
4241 * This way we are sure that all the data written in the previous
4242 * transaction are already on disk (truncate waits for pages under
4243 * writeback).
4244 *
4245 * Called with inode->i_mutex down.
4246 */
4247int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4248{
4249        struct inode *inode = dentry->d_inode;
4250        int error, rc = 0;
4251        int orphan = 0;
4252        const unsigned int ia_valid = attr->ia_valid;
4253
4254        error = inode_change_ok(inode, attr);
4255        if (error)
4256                return error;
4257
4258        if (is_quota_modification(inode, attr))
4259                dquot_initialize(inode);
4260        if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4261            (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4262                handle_t *handle;
4263
4264                /* (user+group)*(old+new) structure, inode write (sb,
4265                 * inode block, ? - but truncate inode update has it) */
4266                handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4267                                        EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4268                if (IS_ERR(handle)) {
4269                        error = PTR_ERR(handle);
4270                        goto err_out;
4271                }
4272                error = dquot_transfer(inode, attr);
4273                if (error) {
4274                        ext4_journal_stop(handle);
4275                        return error;
4276                }
4277                /* Update corresponding info in inode so that everything is in
4278                 * one transaction */
4279                if (attr->ia_valid & ATTR_UID)
4280                        inode->i_uid = attr->ia_uid;
4281                if (attr->ia_valid & ATTR_GID)
4282                        inode->i_gid = attr->ia_gid;
4283                error = ext4_mark_inode_dirty(handle, inode);
4284                ext4_journal_stop(handle);
4285        }
4286
4287        if (attr->ia_valid & ATTR_SIZE) {
4288
4289                if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4290                        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4291
4292                        if (attr->ia_size > sbi->s_bitmap_maxbytes)
4293                                return -EFBIG;
4294                }
4295        }
4296
4297        if (S_ISREG(inode->i_mode) &&
4298            attr->ia_valid & ATTR_SIZE &&
4299            (attr->ia_size < inode->i_size)) {
4300                handle_t *handle;
4301
4302                handle = ext4_journal_start(inode, 3);
4303                if (IS_ERR(handle)) {
4304                        error = PTR_ERR(handle);
4305                        goto err_out;
4306                }
4307                if (ext4_handle_valid(handle)) {
4308                        error = ext4_orphan_add(handle, inode);
4309                        orphan = 1;
4310                }
4311                EXT4_I(inode)->i_disksize = attr->ia_size;
4312                rc = ext4_mark_inode_dirty(handle, inode);
4313                if (!error)
4314                        error = rc;
4315                ext4_journal_stop(handle);
4316
4317                if (ext4_should_order_data(inode)) {
4318                        error = ext4_begin_ordered_truncate(inode,
4319                                                            attr->ia_size);
4320                        if (error) {
4321                                /* Do as much error cleanup as possible */
4322                                handle = ext4_journal_start(inode, 3);
4323                                if (IS_ERR(handle)) {
4324                                        ext4_orphan_del(NULL, inode);
4325                                        goto err_out;
4326                                }
4327                                ext4_orphan_del(handle, inode);
4328                                orphan = 0;
4329                                ext4_journal_stop(handle);
4330                                goto err_out;
4331                        }
4332                }
4333        }
4334
4335        if (attr->ia_valid & ATTR_SIZE) {
4336                if (attr->ia_size != i_size_read(inode)) {
4337                        truncate_setsize(inode, attr->ia_size);
4338                        /* Inode size will be reduced, wait for dio in flight.
4339                         * Temporarily disable dioread_nolock to prevent
4340                         * livelock. */
4341                        if (orphan) {
4342                                ext4_inode_block_unlocked_dio(inode);
4343                                inode_dio_wait(inode);
4344                                ext4_inode_resume_unlocked_dio(inode);
4345                        }
4346                }
4347                ext4_truncate(inode);
4348        }
4349
4350        if (!rc) {
4351                setattr_copy(inode, attr);
4352                mark_inode_dirty(inode);
4353        }
4354
4355        /*
4356         * If the call to ext4_truncate failed to get a transaction handle at
4357         * all, we need to clean up the in-core orphan list manually.
4358         */
4359        if (orphan && inode->i_nlink)
4360                ext4_orphan_del(NULL, inode);
4361
4362        if (!rc && (ia_valid & ATTR_MODE))
4363                rc = ext4_acl_chmod(inode);
4364
4365err_out:
4366        ext4_std_error(inode->i_sb, error);
4367        if (!error)
4368                error = rc;
4369        return error;
4370}
4371
4372int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4373                 struct kstat *stat)
4374{
4375        struct inode *inode;
4376        unsigned long delalloc_blocks;
4377
4378        inode = dentry->d_inode;
4379        generic_fillattr(inode, stat);
4380
4381        /*
4382         * We can't update i_blocks if the block allocation is delayed
4383         * otherwise in the case of system crash before the real block
4384         * allocation is done, we will have i_blocks inconsistent with
4385         * on-disk file blocks.
4386         * We always keep i_blocks updated together with real
4387         * allocation. But to not confuse with user, stat
4388         * will return the blocks that include the delayed allocation
4389         * blocks for this file.
4390         */
4391        delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4392                                EXT4_I(inode)->i_reserved_data_blocks);
4393
4394        stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4395        return 0;
4396}
4397
4398static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4399{
4400        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4401                return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4402        return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4403}
4404
4405/*
4406 * Account for index blocks, block groups bitmaps and block group
4407 * descriptor blocks if modify datablocks and index blocks
4408 * worse case, the indexs blocks spread over different block groups
4409 *
4410 * If datablocks are discontiguous, they are possible to spread over
4411 * different block groups too. If they are contiguous, with flexbg,
4412 * they could still across block group boundary.
4413 *
4414 * Also account for superblock, inode, quota and xattr blocks
4415 */
4416static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4417{
4418        ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4419        int gdpblocks;
4420        int idxblocks;
4421        int ret = 0;
4422
4423        /*
4424         * How many index blocks need to touch to modify nrblocks?
4425         * The "Chunk" flag indicating whether the nrblocks is
4426         * physically contiguous on disk
4427         *
4428         * For Direct IO and fallocate, they calls get_block to allocate
4429         * one single extent at a time, so they could set the "Chunk" flag
4430         */
4431        idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4432
4433        ret = idxblocks;
4434
4435        /*
4436         * Now let's see how many group bitmaps and group descriptors need
4437         * to account
4438         */
4439        groups = idxblocks;
4440        if (chunk)
4441                groups += 1;
4442        else
4443                groups += nrblocks;
4444
4445        gdpblocks = groups;
4446        if (groups > ngroups)
4447                groups = ngroups;
4448        if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4449                gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4450
4451        /* bitmaps and block group descriptor blocks */
4452        ret += groups + gdpblocks;
4453
4454        /* Blocks for super block, inode, quota and xattr blocks */
4455        ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4456
4457        return ret;
4458}
4459
4460/*
4461 * Calculate the total number of credits to reserve to fit
4462 * the modification of a single pages into a single transaction,
4463 * which may include multiple chunks of block allocations.
4464 *
4465 * This could be called via ext4_write_begin()
4466 *
4467 * We need to consider the worse case, when
4468 * one new block per extent.
4469 */
4470int ext4_writepage_trans_blocks(struct inode *inode)
4471{
4472        int bpp = ext4_journal_blocks_per_page(inode);
4473        int ret;
4474
4475        ret = ext4_meta_trans_blocks(inode, bpp, 0);
4476
4477        /* Account for data blocks for journalled mode */
4478        if (ext4_should_journal_data(inode))
4479                ret += bpp;
4480        return ret;
4481}
4482
4483/*
4484 * Calculate the journal credits for a chunk of data modification.
4485 *
4486 * This is called from DIO, fallocate or whoever calling
4487 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4488 *
4489 * journal buffers for data blocks are not included here, as DIO
4490 * and fallocate do no need to journal data buffers.
4491 */
4492int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4493{
4494        return ext4_meta_trans_blocks(inode, nrblocks, 1);
4495}
4496
4497/*
4498 * The caller must have previously called ext4_reserve_inode_write().
4499 * Give this, we know that the caller already has write access to iloc->bh.
4500 */
4501int ext4_mark_iloc_dirty(handle_t *handle,
4502                         struct inode *inode, struct ext4_iloc *iloc)
4503{
4504        int err = 0;
4505
4506        if (IS_I_VERSION(inode))
4507                inode_inc_iversion(inode);
4508
4509        /* the do_update_inode consumes one bh->b_count */
4510        get_bh(iloc->bh);
4511
4512        /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4513        err = ext4_do_update_inode(handle, inode, iloc);
4514        put_bh(iloc->bh);
4515        return err;
4516}
4517
4518/*
4519 * On success, We end up with an outstanding reference count against
4520 * iloc->bh.  This _must_ be cleaned up later.
4521 */
4522
4523int
4524ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4525                         struct ext4_iloc *iloc)
4526{
4527        int err;
4528
4529        err = ext4_get_inode_loc(inode, iloc);
4530        if (!err) {
4531                BUFFER_TRACE(iloc->bh, "get_write_access");
4532                err = ext4_journal_get_write_access(handle, iloc->bh);
4533                if (err) {
4534                        brelse(iloc->bh);
4535                        iloc->bh = NULL;
4536                }
4537        }
4538        ext4_std_error(inode->i_sb, err);
4539        return err;
4540}
4541
4542/*
4543 * Expand an inode by new_extra_isize bytes.
4544 * Returns 0 on success or negative error number on failure.
4545 */
4546static int ext4_expand_extra_isize(struct inode *inode,
4547                                   unsigned int new_extra_isize,
4548                                   struct ext4_iloc iloc,
4549                                   handle_t *handle)
4550{
4551        struct ext4_inode *raw_inode;
4552        struct ext4_xattr_ibody_header *header;
4553
4554        if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4555                return 0;
4556
4557        raw_inode = ext4_raw_inode(&iloc);
4558
4559        header = IHDR(inode, raw_inode);
4560
4561        /* No extended attributes present */
4562        if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4563            header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4564                memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4565                        new_extra_isize);
4566                EXT4_I(inode)->i_extra_isize = new_extra_isize;
4567                return 0;
4568        }
4569
4570        /* try to expand with EAs present */
4571        return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4572                                          raw_inode, handle);
4573}
4574
4575/*
4576 * What we do here is to mark the in-core inode as clean with respect to inode
4577 * dirtiness (it may still be data-dirty).
4578 * This means that the in-core inode may be reaped by prune_icache
4579 * without having to perform any I/O.  This is a very good thing,
4580 * because *any* task may call prune_icache - even ones which
4581 * have a transaction open against a different journal.
4582 *
4583 * Is this cheating?  Not really.  Sure, we haven't written the
4584 * inode out, but prune_icache isn't a user-visible syncing function.
4585 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4586 * we start and wait on commits.
4587 */
4588int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4589{
4590        struct ext4_iloc iloc;
4591        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4592        static unsigned int mnt_count;
4593        int err, ret;
4594
4595        might_sleep();
4596        trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4597        err = ext4_reserve_inode_write(handle, inode, &iloc);
4598        if (ext4_handle_valid(handle) &&
4599            EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4600            !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4601                /*
4602                 * We need extra buffer credits since we may write into EA block
4603                 * with this same handle. If journal_extend fails, then it will
4604                 * only result in a minor loss of functionality for that inode.
4605                 * If this is felt to be critical, then e2fsck should be run to
4606                 * force a large enough s_min_extra_isize.
4607                 */
4608                if ((jbd2_journal_extend(handle,
4609                             EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4610                        ret = ext4_expand_extra_isize(inode,
4611                                                      sbi->s_want_extra_isize,
4612                                                      iloc, handle);
4613                        if (ret) {
4614                                ext4_set_inode_state(inode,
4615                                                     EXT4_STATE_NO_EXPAND);
4616                                if (mnt_count !=
4617                                        le16_to_cpu(sbi->s_es->s_mnt_count)) {
4618                                        ext4_warning(inode->i_sb,
4619                                        "Unable to expand inode %lu. Delete"
4620                                        " some EAs or run e2fsck.",
4621                                        inode->i_ino);
4622                                        mnt_count =
4623                                          le16_to_cpu(sbi->s_es->s_mnt_count);
4624                                }
4625                        }
4626                }
4627        }
4628        if (!err)
4629                err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4630        return err;
4631}
4632
4633/*
4634 * ext4_dirty_inode() is called from __mark_inode_dirty()
4635 *
4636 * We're really interested in the case where a file is being extended.
4637 * i_size has been changed by generic_commit_write() and we thus need
4638 * to include the updated inode in the current transaction.
4639 *
4640 * Also, dquot_alloc_block() will always dirty the inode when blocks
4641 * are allocated to the file.
4642 *
4643 * If the inode is marked synchronous, we don't honour that here - doing
4644 * so would cause a commit on atime updates, which we don't bother doing.
4645 * We handle synchronous inodes at the highest possible level.
4646 */
4647void ext4_dirty_inode(struct inode *inode, int flags)
4648{
4649        handle_t *handle;
4650
4651        handle = ext4_journal_start(inode, 2);
4652        if (IS_ERR(handle))
4653                goto out;
4654
4655        ext4_mark_inode_dirty(handle, inode);
4656
4657        ext4_journal_stop(handle);
4658out:
4659        return;
4660}
4661
4662#if 0
4663/*
4664 * Bind an inode's backing buffer_head into this transaction, to prevent
4665 * it from being flushed to disk early.  Unlike
4666 * ext4_reserve_inode_write, this leaves behind no bh reference and
4667 * returns no iloc structure, so the caller needs to repeat the iloc
4668 * lookup to mark the inode dirty later.
4669 */
4670static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4671{
4672        struct ext4_iloc iloc;
4673
4674        int err = 0;
4675        if (handle) {
4676                err = ext4_get_inode_loc(inode, &iloc);
4677                if (!err) {
4678                        BUFFER_TRACE(iloc.bh, "get_write_access");
4679                        err = jbd2_journal_get_write_access(handle, iloc.bh);
4680                        if (!err)
4681                                err = ext4_handle_dirty_metadata(handle,
4682                                                                 NULL,
4683                                                                 iloc.bh);
4684                        brelse(iloc.bh);
4685                }
4686        }
4687        ext4_std_error(inode->i_sb, err);
4688        return err;
4689}
4690#endif
4691
4692int ext4_change_inode_journal_flag(struct inode *inode, int val)
4693{
4694        journal_t *journal;
4695        handle_t *handle;
4696        int err;
4697
4698        /*
4699         * We have to be very careful here: changing a data block's
4700         * journaling status dynamically is dangerous.  If we write a
4701         * data block to the journal, change the status and then delete
4702         * that block, we risk forgetting to revoke the old log record
4703         * from the journal and so a subsequent replay can corrupt data.
4704         * So, first we make sure that the journal is empty and that
4705         * nobody is changing anything.
4706         */
4707
4708        journal = EXT4_JOURNAL(inode);
4709        if (!journal)
4710                return 0;
4711        if (is_journal_aborted(journal))
4712                return -EROFS;
4713        /* We have to allocate physical blocks for delalloc blocks
4714         * before flushing journal. otherwise delalloc blocks can not
4715         * be allocated any more. even more truncate on delalloc blocks
4716         * could trigger BUG by flushing delalloc blocks in journal.
4717         * There is no delalloc block in non-journal data mode.
4718         */
4719        if (val && test_opt(inode->i_sb, DELALLOC)) {
4720                err = ext4_alloc_da_blocks(inode);
4721                if (err < 0)
4722                        return err;
4723        }
4724
4725        /* Wait for all existing dio workers */
4726        ext4_inode_block_unlocked_dio(inode);
4727        inode_dio_wait(inode);
4728
4729        jbd2_journal_lock_updates(journal);
4730
4731        /*
4732         * OK, there are no updates running now, and all cached data is
4733         * synced to disk.  We are now in a completely consistent state
4734         * which doesn't have anything in the journal, and we know that
4735         * no filesystem updates are running, so it is safe to modify
4736         * the inode's in-core data-journaling state flag now.
4737         */
4738
4739        if (val)
4740                ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4741        else {
4742                jbd2_journal_flush(journal);
4743                ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4744        }
4745        ext4_set_aops(inode);
4746
4747        jbd2_journal_unlock_updates(journal);
4748        ext4_inode_resume_unlocked_dio(inode);
4749
4750        /* Finally we can mark the inode as dirty. */
4751
4752        handle = ext4_journal_start(inode, 1);
4753        if (IS_ERR(handle))
4754                return PTR_ERR(handle);
4755
4756        err = ext4_mark_inode_dirty(handle, inode);
4757        ext4_handle_sync(handle);
4758        ext4_journal_stop(handle);
4759        ext4_std_error(inode->i_sb, err);
4760
4761        return err;
4762}
4763
4764static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4765{
4766        return !buffer_mapped(bh);
4767}
4768
4769int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4770{
4771        struct page *page = vmf->page;
4772        loff_t size;
4773        unsigned long len;
4774        int ret;
4775        struct file *file = vma->vm_file;
4776        struct inode *inode = file->f_path.dentry->d_inode;
4777        struct address_space *mapping = inode->i_mapping;
4778        handle_t *handle;
4779        get_block_t *get_block;
4780        int retries = 0;
4781
4782        sb_start_pagefault(inode->i_sb);
4783        file_update_time(vma->vm_file);
4784        /* Delalloc case is easy... */
4785        if (test_opt(inode->i_sb, DELALLOC) &&
4786            !ext4_should_journal_data(inode) &&
4787            !ext4_nonda_switch(inode->i_sb)) {
4788                do {
4789                        ret = __block_page_mkwrite(vma, vmf,
4790                                                   ext4_da_get_block_prep);
4791                } while (ret == -ENOSPC &&
4792                       ext4_should_retry_alloc(inode->i_sb, &retries));
4793                goto out_ret;
4794        }
4795
4796        lock_page(page);
4797        size = i_size_read(inode);
4798        /* Page got truncated from under us? */
4799        if (page->mapping != mapping || page_offset(page) > size) {
4800                unlock_page(page);
4801                ret = VM_FAULT_NOPAGE;
4802                goto out;
4803        }
4804
4805        if (page->index == size >> PAGE_CACHE_SHIFT)
4806                len = size & ~PAGE_CACHE_MASK;
4807        else
4808                len = PAGE_CACHE_SIZE;
4809        /*
4810         * Return if we have all the buffers mapped. This avoids the need to do
4811         * journal_start/journal_stop which can block and take a long time
4812         */
4813        if (page_has_buffers(page)) {
4814                if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4815                                        ext4_bh_unmapped)) {
4816                        /* Wait so that we don't change page under IO */
4817                        wait_on_page_writeback(page);
4818                        ret = VM_FAULT_LOCKED;
4819                        goto out;
4820                }
4821        }
4822        unlock_page(page);
4823        /* OK, we need to fill the hole... */
4824        if (ext4_should_dioread_nolock(inode))
4825                get_block = ext4_get_block_write;
4826        else
4827                get_block = ext4_get_block;
4828retry_alloc:
4829        handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4830        if (IS_ERR(handle)) {
4831                ret = VM_FAULT_SIGBUS;
4832                goto out;
4833        }
4834        ret = __block_page_mkwrite(vma, vmf, get_block);
4835        if (!ret && ext4_should_journal_data(inode)) {
4836                if (walk_page_buffers(handle, page_buffers(page), 0,
4837                          PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4838                        unlock_page(page);
4839                        ret = VM_FAULT_SIGBUS;
4840                        ext4_journal_stop(handle);
4841                        goto out;
4842                }
4843                ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4844        }
4845        ext4_journal_stop(handle);
4846        if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4847                goto retry_alloc;
4848out_ret:
4849        ret = block_page_mkwrite_return(ret);
4850out:
4851        sb_end_pagefault(inode->i_sb);
4852        return ret;
4853}
4854
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.