linux/fs/reiserfs/inode.c
<<
>>
Prefs
   1/*
   2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3 */
   4
   5#include <linux/time.h>
   6#include <linux/fs.h>
   7#include "reiserfs.h"
   8#include "acl.h"
   9#include "xattr.h"
  10#include <linux/exportfs.h>
  11#include <linux/pagemap.h>
  12#include <linux/highmem.h>
  13#include <linux/slab.h>
  14#include <linux/uaccess.h>
  15#include <asm/unaligned.h>
  16#include <linux/buffer_head.h>
  17#include <linux/mpage.h>
  18#include <linux/writeback.h>
  19#include <linux/quotaops.h>
  20#include <linux/swap.h>
  21#include <linux/aio.h>
  22
  23int reiserfs_commit_write(struct file *f, struct page *page,
  24                          unsigned from, unsigned to);
  25
  26void reiserfs_evict_inode(struct inode *inode)
  27{
  28        /*
  29         * We need blocks for transaction + (user+group) quota
  30         * update (possibly delete)
  31         */
  32        int jbegin_count =
  33            JOURNAL_PER_BALANCE_CNT * 2 +
  34            2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  35        struct reiserfs_transaction_handle th;
  36        int err;
  37
  38        if (!inode->i_nlink && !is_bad_inode(inode))
  39                dquot_initialize(inode);
  40
  41        truncate_inode_pages_final(&inode->i_data);
  42        if (inode->i_nlink)
  43                goto no_delete;
  44
  45        /*
  46         * The = 0 happens when we abort creating a new inode
  47         * for some reason like lack of space..
  48         * also handles bad_inode case
  49         */
  50        if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {
  51
  52                reiserfs_delete_xattrs(inode);
  53
  54                reiserfs_write_lock(inode->i_sb);
  55
  56                if (journal_begin(&th, inode->i_sb, jbegin_count))
  57                        goto out;
  58                reiserfs_update_inode_transaction(inode);
  59
  60                reiserfs_discard_prealloc(&th, inode);
  61
  62                err = reiserfs_delete_object(&th, inode);
  63
  64                /*
  65                 * Do quota update inside a transaction for journaled quotas.
  66                 * We must do that after delete_object so that quota updates
  67                 * go into the same transaction as stat data deletion
  68                 */
  69                if (!err) {
  70                        int depth = reiserfs_write_unlock_nested(inode->i_sb);
  71                        dquot_free_inode(inode);
  72                        reiserfs_write_lock_nested(inode->i_sb, depth);
  73                }
  74
  75                if (journal_end(&th))
  76                        goto out;
  77
  78                /*
  79                 * check return value from reiserfs_delete_object after
  80                 * ending the transaction
  81                 */
  82                if (err)
  83                    goto out;
  84
  85                /*
  86                 * all items of file are deleted, so we can remove
  87                 * "save" link
  88                 * we can't do anything about an error here
  89                 */
  90                remove_save_link(inode, 0 /* not truncate */);
  91out:
  92                reiserfs_write_unlock(inode->i_sb);
  93        } else {
  94                /* no object items are in the tree */
  95                ;
  96        }
  97
  98        /* note this must go after the journal_end to prevent deadlock */
  99        clear_inode(inode);
 100
 101        dquot_drop(inode);
 102        inode->i_blocks = 0;
 103        return;
 104
 105no_delete:
 106        clear_inode(inode);
 107        dquot_drop(inode);
 108}
 109
 110static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
 111                          __u32 objectid, loff_t offset, int type, int length)
 112{
 113        key->version = version;
 114
 115        key->on_disk_key.k_dir_id = dirid;
 116        key->on_disk_key.k_objectid = objectid;
 117        set_cpu_key_k_offset(key, offset);
 118        set_cpu_key_k_type(key, type);
 119        key->key_length = length;
 120}
 121
 122/*
 123 * take base of inode_key (it comes from inode always) (dirid, objectid)
 124 * and version from an inode, set offset and type of key
 125 */
 126void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
 127                  int type, int length)
 128{
 129        _make_cpu_key(key, get_inode_item_key_version(inode),
 130                      le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 131                      le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 132                      length);
 133}
 134
 135/* when key is 0, do not set version and short key */
 136inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 137                              int version,
 138                              loff_t offset, int type, int length,
 139                              int entry_count /*or ih_free_space */ )
 140{
 141        if (key) {
 142                ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 143                ih->ih_key.k_objectid =
 144                    cpu_to_le32(key->on_disk_key.k_objectid);
 145        }
 146        put_ih_version(ih, version);
 147        set_le_ih_k_offset(ih, offset);
 148        set_le_ih_k_type(ih, type);
 149        put_ih_item_len(ih, length);
 150        /*    set_ih_free_space (ih, 0); */
 151        /*
 152         * for directory items it is entry count, for directs and stat
 153         * datas - 0xffff, for indirects - 0
 154         */
 155        put_ih_entry_count(ih, entry_count);
 156}
 157
 158/*
 159 * FIXME: we might cache recently accessed indirect item
 160 * Ugh.  Not too eager for that....
 161 * I cut the code until such time as I see a convincing argument (benchmark).
 162 * I don't want a bloated inode struct..., and I don't like code complexity....
 163 */
 164
 165/*
 166 * cutting the code is fine, since it really isn't in use yet and is easy
 167 * to add back in.  But, Vladimir has a really good idea here.  Think
 168 * about what happens for reading a file.  For each page,
 169 * The VFS layer calls reiserfs_readpage, who searches the tree to find
 170 * an indirect item.  This indirect item has X number of pointers, where
 171 * X is a big number if we've done the block allocation right.  But,
 172 * we only use one or two of these pointers during each call to readpage,
 173 * needlessly researching again later on.
 174 *
 175 * The size of the cache could be dynamic based on the size of the file.
 176 *
 177 * I'd also like to see us cache the location the stat data item, since
 178 * we are needlessly researching for that frequently.
 179 *
 180 * --chris
 181 */
 182
 183/*
 184 * If this page has a file tail in it, and
 185 * it was read in by get_block_create_0, the page data is valid,
 186 * but tail is still sitting in a direct item, and we can't write to
 187 * it.  So, look through this page, and check all the mapped buffers
 188 * to make sure they have valid block numbers.  Any that don't need
 189 * to be unmapped, so that __block_write_begin will correctly call
 190 * reiserfs_get_block to convert the tail into an unformatted node
 191 */
 192static inline void fix_tail_page_for_writing(struct page *page)
 193{
 194        struct buffer_head *head, *next, *bh;
 195
 196        if (page && page_has_buffers(page)) {
 197                head = page_buffers(page);
 198                bh = head;
 199                do {
 200                        next = bh->b_this_page;
 201                        if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 202                                reiserfs_unmap_buffer(bh);
 203                        }
 204                        bh = next;
 205                } while (bh != head);
 206        }
 207}
 208
 209/*
 210 * reiserfs_get_block does not need to allocate a block only if it has been
 211 * done already or non-hole position has been found in the indirect item
 212 */
 213static inline int allocation_needed(int retval, b_blocknr_t allocated,
 214                                    struct item_head *ih,
 215                                    __le32 * item, int pos_in_item)
 216{
 217        if (allocated)
 218                return 0;
 219        if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 220            get_block_num(item, pos_in_item))
 221                return 0;
 222        return 1;
 223}
 224
 225static inline int indirect_item_found(int retval, struct item_head *ih)
 226{
 227        return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 228}
 229
 230static inline void set_block_dev_mapped(struct buffer_head *bh,
 231                                        b_blocknr_t block, struct inode *inode)
 232{
 233        map_bh(bh, inode->i_sb, block);
 234}
 235
 236/*
 237 * files which were created in the earlier version can not be longer,
 238 * than 2 gb
 239 */
 240static int file_capable(struct inode *inode, sector_t block)
 241{
 242        /* it is new file. */
 243        if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||
 244            /* old file, but 'block' is inside of 2gb */
 245            block < (1 << (31 - inode->i_sb->s_blocksize_bits)))
 246                return 1;
 247
 248        return 0;
 249}
 250
 251static int restart_transaction(struct reiserfs_transaction_handle *th,
 252                               struct inode *inode, struct treepath *path)
 253{
 254        struct super_block *s = th->t_super;
 255        int err;
 256
 257        BUG_ON(!th->t_trans_id);
 258        BUG_ON(!th->t_refcount);
 259
 260        pathrelse(path);
 261
 262        /* we cannot restart while nested */
 263        if (th->t_refcount > 1) {
 264                return 0;
 265        }
 266        reiserfs_update_sd(th, inode);
 267        err = journal_end(th);
 268        if (!err) {
 269                err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 270                if (!err)
 271                        reiserfs_update_inode_transaction(inode);
 272        }
 273        return err;
 274}
 275
 276/*
 277 * it is called by get_block when create == 0. Returns block number
 278 * for 'block'-th logical block of file. When it hits direct item it
 279 * returns 0 (being called from bmap) or read direct item into piece
 280 * of page (bh_result)
 281 * Please improve the english/clarity in the comment above, as it is
 282 * hard to understand.
 283 */
 284static int _get_block_create_0(struct inode *inode, sector_t block,
 285                               struct buffer_head *bh_result, int args)
 286{
 287        INITIALIZE_PATH(path);
 288        struct cpu_key key;
 289        struct buffer_head *bh;
 290        struct item_head *ih, tmp_ih;
 291        b_blocknr_t blocknr;
 292        char *p = NULL;
 293        int chars;
 294        int ret;
 295        int result;
 296        int done = 0;
 297        unsigned long offset;
 298
 299        /* prepare the key to look for the 'block'-th block of file */
 300        make_cpu_key(&key, inode,
 301                     (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 302                     3);
 303
 304        result = search_for_position_by_key(inode->i_sb, &key, &path);
 305        if (result != POSITION_FOUND) {
 306                pathrelse(&path);
 307                if (p)
 308                        kunmap(bh_result->b_page);
 309                if (result == IO_ERROR)
 310                        return -EIO;
 311                /*
 312                 * We do not return -ENOENT if there is a hole but page is
 313                 * uptodate, because it means that there is some MMAPED data
 314                 * associated with it that is yet to be written to disk.
 315                 */
 316                if ((args & GET_BLOCK_NO_HOLE)
 317                    && !PageUptodate(bh_result->b_page)) {
 318                        return -ENOENT;
 319                }
 320                return 0;
 321        }
 322
 323        bh = get_last_bh(&path);
 324        ih = tp_item_head(&path);
 325        if (is_indirect_le_ih(ih)) {
 326                __le32 *ind_item = (__le32 *) ih_item_body(bh, ih);
 327
 328                /*
 329                 * FIXME: here we could cache indirect item or part of it in
 330                 * the inode to avoid search_by_key in case of subsequent
 331                 * access to file
 332                 */
 333                blocknr = get_block_num(ind_item, path.pos_in_item);
 334                ret = 0;
 335                if (blocknr) {
 336                        map_bh(bh_result, inode->i_sb, blocknr);
 337                        if (path.pos_in_item ==
 338                            ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 339                                set_buffer_boundary(bh_result);
 340                        }
 341                } else
 342                        /*
 343                         * We do not return -ENOENT if there is a hole but
 344                         * page is uptodate, because it means that there is
 345                         * some MMAPED data associated with it that is
 346                         * yet to be written to disk.
 347                         */
 348                if ((args & GET_BLOCK_NO_HOLE)
 349                            && !PageUptodate(bh_result->b_page)) {
 350                        ret = -ENOENT;
 351                }
 352
 353                pathrelse(&path);
 354                if (p)
 355                        kunmap(bh_result->b_page);
 356                return ret;
 357        }
 358        /* requested data are in direct item(s) */
 359        if (!(args & GET_BLOCK_READ_DIRECT)) {
 360                /*
 361                 * we are called by bmap. FIXME: we can not map block of file
 362                 * when it is stored in direct item(s)
 363                 */
 364                pathrelse(&path);
 365                if (p)
 366                        kunmap(bh_result->b_page);
 367                return -ENOENT;
 368        }
 369
 370        /*
 371         * if we've got a direct item, and the buffer or page was uptodate,
 372         * we don't want to pull data off disk again.  skip to the
 373         * end, where we map the buffer and return
 374         */
 375        if (buffer_uptodate(bh_result)) {
 376                goto finished;
 377        } else
 378                /*
 379                 * grab_tail_page can trigger calls to reiserfs_get_block on
 380                 * up to date pages without any buffers.  If the page is up
 381                 * to date, we don't want read old data off disk.  Set the up
 382                 * to date bit on the buffer instead and jump to the end
 383                 */
 384        if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 385                set_buffer_uptodate(bh_result);
 386                goto finished;
 387        }
 388        /* read file tail into part of page */
 389        offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 390        copy_item_head(&tmp_ih, ih);
 391
 392        /*
 393         * we only want to kmap if we are reading the tail into the page.
 394         * this is not the common case, so we don't kmap until we are
 395         * sure we need to.  But, this means the item might move if
 396         * kmap schedules
 397         */
 398        if (!p)
 399                p = (char *)kmap(bh_result->b_page);
 400
 401        p += offset;
 402        memset(p, 0, inode->i_sb->s_blocksize);
 403        do {
 404                if (!is_direct_le_ih(ih)) {
 405                        BUG();
 406                }
 407                /*
 408                 * make sure we don't read more bytes than actually exist in
 409                 * the file.  This can happen in odd cases where i_size isn't
 410                 * correct, and when direct item padding results in a few
 411                 * extra bytes at the end of the direct item
 412                 */
 413                if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 414                        break;
 415                if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 416                        chars =
 417                            inode->i_size - (le_ih_k_offset(ih) - 1) -
 418                            path.pos_in_item;
 419                        done = 1;
 420                } else {
 421                        chars = ih_item_len(ih) - path.pos_in_item;
 422                }
 423                memcpy(p, ih_item_body(bh, ih) + path.pos_in_item, chars);
 424
 425                if (done)
 426                        break;
 427
 428                p += chars;
 429
 430                /*
 431                 * we done, if read direct item is not the last item of
 432                 * node FIXME: we could try to check right delimiting key
 433                 * to see whether direct item continues in the right
 434                 * neighbor or rely on i_size
 435                 */
 436                if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 437                        break;
 438
 439                /* update key to look for the next piece */
 440                set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 441                result = search_for_position_by_key(inode->i_sb, &key, &path);
 442                if (result != POSITION_FOUND)
 443                        /* i/o error most likely */
 444                        break;
 445                bh = get_last_bh(&path);
 446                ih = tp_item_head(&path);
 447        } while (1);
 448
 449        flush_dcache_page(bh_result->b_page);
 450        kunmap(bh_result->b_page);
 451
 452finished:
 453        pathrelse(&path);
 454
 455        if (result == IO_ERROR)
 456                return -EIO;
 457
 458        /*
 459         * this buffer has valid data, but isn't valid for io.  mapping it to
 460         * block #0 tells the rest of reiserfs it just has a tail in it
 461         */
 462        map_bh(bh_result, inode->i_sb, 0);
 463        set_buffer_uptodate(bh_result);
 464        return 0;
 465}
 466
 467/*
 468 * this is called to create file map. So, _get_block_create_0 will not
 469 * read direct item
 470 */
 471static int reiserfs_bmap(struct inode *inode, sector_t block,
 472                         struct buffer_head *bh_result, int create)
 473{
 474        if (!file_capable(inode, block))
 475                return -EFBIG;
 476
 477        reiserfs_write_lock(inode->i_sb);
 478        /* do not read the direct item */
 479        _get_block_create_0(inode, block, bh_result, 0);
 480        reiserfs_write_unlock(inode->i_sb);
 481        return 0;
 482}
 483
 484/*
 485 * special version of get_block that is only used by grab_tail_page right
 486 * now.  It is sent to __block_write_begin, and when you try to get a
 487 * block past the end of the file (or a block from a hole) it returns
 488 * -ENOENT instead of a valid buffer.  __block_write_begin expects to
 489 * be able to do i/o on the buffers returned, unless an error value
 490 * is also returned.
 491 *
 492 * So, this allows __block_write_begin to be used for reading a single block
 493 * in a page.  Where it does not produce a valid page for holes, or past the
 494 * end of the file.  This turns out to be exactly what we need for reading
 495 * tails for conversion.
 496 *
 497 * The point of the wrapper is forcing a certain value for create, even
 498 * though the VFS layer is calling this function with create==1.  If you
 499 * don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 500 * don't use this function.
 501*/
 502static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 503                                       struct buffer_head *bh_result,
 504                                       int create)
 505{
 506        return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 507}
 508
 509/*
 510 * This is special helper for reiserfs_get_block in case we are executing
 511 * direct_IO request.
 512 */
 513static int reiserfs_get_blocks_direct_io(struct inode *inode,
 514                                         sector_t iblock,
 515                                         struct buffer_head *bh_result,
 516                                         int create)
 517{
 518        int ret;
 519
 520        bh_result->b_page = NULL;
 521
 522        /*
 523         * We set the b_size before reiserfs_get_block call since it is
 524         * referenced in convert_tail_for_hole() that may be called from
 525         * reiserfs_get_block()
 526         */
 527        bh_result->b_size = (1 << inode->i_blkbits);
 528
 529        ret = reiserfs_get_block(inode, iblock, bh_result,
 530                                 create | GET_BLOCK_NO_DANGLE);
 531        if (ret)
 532                goto out;
 533
 534        /* don't allow direct io onto tail pages */
 535        if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 536                /*
 537                 * make sure future calls to the direct io funcs for this
 538                 * offset in the file fail by unmapping the buffer
 539                 */
 540                clear_buffer_mapped(bh_result);
 541                ret = -EINVAL;
 542        }
 543
 544        /*
 545         * Possible unpacked tail. Flush the data before pages have
 546         * disappeared
 547         */
 548        if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 549                int err;
 550
 551                reiserfs_write_lock(inode->i_sb);
 552
 553                err = reiserfs_commit_for_inode(inode);
 554                REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 555
 556                reiserfs_write_unlock(inode->i_sb);
 557
 558                if (err < 0)
 559                        ret = err;
 560        }
 561out:
 562        return ret;
 563}
 564
 565/*
 566 * helper function for when reiserfs_get_block is called for a hole
 567 * but the file tail is still in a direct item
 568 * bh_result is the buffer head for the hole
 569 * tail_offset is the offset of the start of the tail in the file
 570 *
 571 * This calls prepare_write, which will start a new transaction
 572 * you should not be in a transaction, or have any paths held when you
 573 * call this.
 574 */
 575static int convert_tail_for_hole(struct inode *inode,
 576                                 struct buffer_head *bh_result,
 577                                 loff_t tail_offset)
 578{
 579        unsigned long index;
 580        unsigned long tail_end;
 581        unsigned long tail_start;
 582        struct page *tail_page;
 583        struct page *hole_page = bh_result->b_page;
 584        int retval = 0;
 585
 586        if ((tail_offset & (bh_result->b_size - 1)) != 1)
 587                return -EIO;
 588
 589        /* always try to read until the end of the block */
 590        tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 591        tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 592
 593        index = tail_offset >> PAGE_CACHE_SHIFT;
 594        /*
 595         * hole_page can be zero in case of direct_io, we are sure
 596         * that we cannot get here if we write with O_DIRECT into tail page
 597         */
 598        if (!hole_page || index != hole_page->index) {
 599                tail_page = grab_cache_page(inode->i_mapping, index);
 600                retval = -ENOMEM;
 601                if (!tail_page) {
 602                        goto out;
 603                }
 604        } else {
 605                tail_page = hole_page;
 606        }
 607
 608        /*
 609         * we don't have to make sure the conversion did not happen while
 610         * we were locking the page because anyone that could convert
 611         * must first take i_mutex.
 612         *
 613         * We must fix the tail page for writing because it might have buffers
 614         * that are mapped, but have a block number of 0.  This indicates tail
 615         * data that has been read directly into the page, and
 616         * __block_write_begin won't trigger a get_block in this case.
 617         */
 618        fix_tail_page_for_writing(tail_page);
 619        retval = __reiserfs_write_begin(tail_page, tail_start,
 620                                      tail_end - tail_start);
 621        if (retval)
 622                goto unlock;
 623
 624        /* tail conversion might change the data in the page */
 625        flush_dcache_page(tail_page);
 626
 627        retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 628
 629unlock:
 630        if (tail_page != hole_page) {
 631                unlock_page(tail_page);
 632                page_cache_release(tail_page);
 633        }
 634out:
 635        return retval;
 636}
 637
 638static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 639                                  sector_t block,
 640                                  struct inode *inode,
 641                                  b_blocknr_t * allocated_block_nr,
 642                                  struct treepath *path, int flags)
 643{
 644        BUG_ON(!th->t_trans_id);
 645
 646#ifdef REISERFS_PREALLOCATE
 647        if (!(flags & GET_BLOCK_NO_IMUX)) {
 648                return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 649                                                  path, block);
 650        }
 651#endif
 652        return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 653                                         block);
 654}
 655
 656int reiserfs_get_block(struct inode *inode, sector_t block,
 657                       struct buffer_head *bh_result, int create)
 658{
 659        int repeat, retval = 0;
 660        /* b_blocknr_t is (unsigned) 32 bit int*/
 661        b_blocknr_t allocated_block_nr = 0;
 662        INITIALIZE_PATH(path);
 663        int pos_in_item;
 664        struct cpu_key key;
 665        struct buffer_head *bh, *unbh = NULL;
 666        struct item_head *ih, tmp_ih;
 667        __le32 *item;
 668        int done;
 669        int fs_gen;
 670        struct reiserfs_transaction_handle *th = NULL;
 671        /*
 672         * space reserved in transaction batch:
 673         * . 3 balancings in direct->indirect conversion
 674         * . 1 block involved into reiserfs_update_sd()
 675         * XXX in practically impossible worst case direct2indirect()
 676         * can incur (much) more than 3 balancings.
 677         * quota update for user, group
 678         */
 679        int jbegin_count =
 680            JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 681            2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 682        int version;
 683        int dangle = 1;
 684        loff_t new_offset =
 685            (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 686
 687        reiserfs_write_lock(inode->i_sb);
 688        version = get_inode_item_key_version(inode);
 689
 690        if (!file_capable(inode, block)) {
 691                reiserfs_write_unlock(inode->i_sb);
 692                return -EFBIG;
 693        }
 694
 695        /*
 696         * if !create, we aren't changing the FS, so we don't need to
 697         * log anything, so we don't need to start a transaction
 698         */
 699        if (!(create & GET_BLOCK_CREATE)) {
 700                int ret;
 701                /* find number of block-th logical block of the file */
 702                ret = _get_block_create_0(inode, block, bh_result,
 703                                          create | GET_BLOCK_READ_DIRECT);
 704                reiserfs_write_unlock(inode->i_sb);
 705                return ret;
 706        }
 707
 708        /*
 709         * if we're already in a transaction, make sure to close
 710         * any new transactions we start in this func
 711         */
 712        if ((create & GET_BLOCK_NO_DANGLE) ||
 713            reiserfs_transaction_running(inode->i_sb))
 714                dangle = 0;
 715
 716        /*
 717         * If file is of such a size, that it might have a tail and
 718         * tails are enabled  we should mark it as possibly needing
 719         * tail packing on close
 720         */
 721        if ((have_large_tails(inode->i_sb)
 722             && inode->i_size < i_block_size(inode) * 4)
 723            || (have_small_tails(inode->i_sb)
 724                && inode->i_size < i_block_size(inode)))
 725                REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 726
 727        /* set the key of the first byte in the 'block'-th block of file */
 728        make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 729        if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 730start_trans:
 731                th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 732                if (!th) {
 733                        retval = -ENOMEM;
 734                        goto failure;
 735                }
 736                reiserfs_update_inode_transaction(inode);
 737        }
 738research:
 739
 740        retval = search_for_position_by_key(inode->i_sb, &key, &path);
 741        if (retval == IO_ERROR) {
 742                retval = -EIO;
 743                goto failure;
 744        }
 745
 746        bh = get_last_bh(&path);
 747        ih = tp_item_head(&path);
 748        item = tp_item_body(&path);
 749        pos_in_item = path.pos_in_item;
 750
 751        fs_gen = get_generation(inode->i_sb);
 752        copy_item_head(&tmp_ih, ih);
 753
 754        if (allocation_needed
 755            (retval, allocated_block_nr, ih, item, pos_in_item)) {
 756                /* we have to allocate block for the unformatted node */
 757                if (!th) {
 758                        pathrelse(&path);
 759                        goto start_trans;
 760                }
 761
 762                repeat =
 763                    _allocate_block(th, block, inode, &allocated_block_nr,
 764                                    &path, create);
 765
 766                /*
 767                 * restart the transaction to give the journal a chance to free
 768                 * some blocks.  releases the path, so we have to go back to
 769                 * research if we succeed on the second try
 770                 */
 771                if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 772                        SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 773                        retval = restart_transaction(th, inode, &path);
 774                        if (retval)
 775                                goto failure;
 776                        repeat =
 777                            _allocate_block(th, block, inode,
 778                                            &allocated_block_nr, NULL, create);
 779
 780                        if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 781                                goto research;
 782                        }
 783                        if (repeat == QUOTA_EXCEEDED)
 784                                retval = -EDQUOT;
 785                        else
 786                                retval = -ENOSPC;
 787                        goto failure;
 788                }
 789
 790                if (fs_changed(fs_gen, inode->i_sb)
 791                    && item_moved(&tmp_ih, &path)) {
 792                        goto research;
 793                }
 794        }
 795
 796        if (indirect_item_found(retval, ih)) {
 797                b_blocknr_t unfm_ptr;
 798                /*
 799                 * 'block'-th block is in the file already (there is
 800                 * corresponding cell in some indirect item). But it may be
 801                 * zero unformatted node pointer (hole)
 802                 */
 803                unfm_ptr = get_block_num(item, pos_in_item);
 804                if (unfm_ptr == 0) {
 805                        /* use allocated block to plug the hole */
 806                        reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 807                        if (fs_changed(fs_gen, inode->i_sb)
 808                            && item_moved(&tmp_ih, &path)) {
 809                                reiserfs_restore_prepared_buffer(inode->i_sb,
 810                                                                 bh);
 811                                goto research;
 812                        }
 813                        set_buffer_new(bh_result);
 814                        if (buffer_dirty(bh_result)
 815                            && reiserfs_data_ordered(inode->i_sb))
 816                                reiserfs_add_ordered_list(inode, bh_result);
 817                        put_block_num(item, pos_in_item, allocated_block_nr);
 818                        unfm_ptr = allocated_block_nr;
 819                        journal_mark_dirty(th, bh);
 820                        reiserfs_update_sd(th, inode);
 821                }
 822                set_block_dev_mapped(bh_result, unfm_ptr, inode);
 823                pathrelse(&path);
 824                retval = 0;
 825                if (!dangle && th)
 826                        retval = reiserfs_end_persistent_transaction(th);
 827
 828                reiserfs_write_unlock(inode->i_sb);
 829
 830                /*
 831                 * the item was found, so new blocks were not added to the file
 832                 * there is no need to make sure the inode is updated with this
 833                 * transaction
 834                 */
 835                return retval;
 836        }
 837
 838        if (!th) {
 839                pathrelse(&path);
 840                goto start_trans;
 841        }
 842
 843        /*
 844         * desired position is not found or is in the direct item. We have
 845         * to append file with holes up to 'block'-th block converting
 846         * direct items to indirect one if necessary
 847         */
 848        done = 0;
 849        do {
 850                if (is_statdata_le_ih(ih)) {
 851                        __le32 unp = 0;
 852                        struct cpu_key tmp_key;
 853
 854                        /* indirect item has to be inserted */
 855                        make_le_item_head(&tmp_ih, &key, version, 1,
 856                                          TYPE_INDIRECT, UNFM_P_SIZE,
 857                                          0 /* free_space */ );
 858
 859                        /*
 860                         * we are going to add 'block'-th block to the file.
 861                         * Use allocated block for that
 862                         */
 863                        if (cpu_key_k_offset(&key) == 1) {
 864                                unp = cpu_to_le32(allocated_block_nr);
 865                                set_block_dev_mapped(bh_result,
 866                                                     allocated_block_nr, inode);
 867                                set_buffer_new(bh_result);
 868                                done = 1;
 869                        }
 870                        tmp_key = key;  /* ;) */
 871                        set_cpu_key_k_offset(&tmp_key, 1);
 872                        PATH_LAST_POSITION(&path)++;
 873
 874                        retval =
 875                            reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 876                                                 inode, (char *)&unp);
 877                        if (retval) {
 878                                reiserfs_free_block(th, inode,
 879                                                    allocated_block_nr, 1);
 880                                /*
 881                                 * retval == -ENOSPC, -EDQUOT or -EIO
 882                                 * or -EEXIST
 883                                 */
 884                                goto failure;
 885                        }
 886                } else if (is_direct_le_ih(ih)) {
 887                        /* direct item has to be converted */
 888                        loff_t tail_offset;
 889
 890                        tail_offset =
 891                            ((le_ih_k_offset(ih) -
 892                              1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 893
 894                        /*
 895                         * direct item we just found fits into block we have
 896                         * to map. Convert it into unformatted node: use
 897                         * bh_result for the conversion
 898                         */
 899                        if (tail_offset == cpu_key_k_offset(&key)) {
 900                                set_block_dev_mapped(bh_result,
 901                                                     allocated_block_nr, inode);
 902                                unbh = bh_result;
 903                                done = 1;
 904                        } else {
 905                                /*
 906                                 * we have to pad file tail stored in direct
 907                                 * item(s) up to block size and convert it
 908                                 * to unformatted node. FIXME: this should
 909                                 * also get into page cache
 910                                 */
 911
 912                                pathrelse(&path);
 913                                /*
 914                                 * ugly, but we can only end the transaction if
 915                                 * we aren't nested
 916                                 */
 917                                BUG_ON(!th->t_refcount);
 918                                if (th->t_refcount == 1) {
 919                                        retval =
 920                                            reiserfs_end_persistent_transaction
 921                                            (th);
 922                                        th = NULL;
 923                                        if (retval)
 924                                                goto failure;
 925                                }
 926
 927                                retval =
 928                                    convert_tail_for_hole(inode, bh_result,
 929                                                          tail_offset);
 930                                if (retval) {
 931                                        if (retval != -ENOSPC)
 932                                                reiserfs_error(inode->i_sb,
 933                                                        "clm-6004",
 934                                                        "convert tail failed "
 935                                                        "inode %lu, error %d",
 936                                                        inode->i_ino,
 937                                                        retval);
 938                                        if (allocated_block_nr) {
 939                                                /*
 940                                                 * the bitmap, the super,
 941                                                 * and the stat data == 3
 942                                                 */
 943                                                if (!th)
 944                                                        th = reiserfs_persistent_transaction(inode->i_sb, 3);
 945                                                if (th)
 946                                                        reiserfs_free_block(th,
 947                                                                            inode,
 948                                                                            allocated_block_nr,
 949                                                                            1);
 950                                        }
 951                                        goto failure;
 952                                }
 953                                goto research;
 954                        }
 955                        retval =
 956                            direct2indirect(th, inode, &path, unbh,
 957                                            tail_offset);
 958                        if (retval) {
 959                                reiserfs_unmap_buffer(unbh);
 960                                reiserfs_free_block(th, inode,
 961                                                    allocated_block_nr, 1);
 962                                goto failure;
 963                        }
 964                        /*
 965                         * it is important the set_buffer_uptodate is done
 966                         * after the direct2indirect.  The buffer might
 967                         * contain valid data newer than the data on disk
 968                         * (read by readpage, changed, and then sent here by
 969                         * writepage).  direct2indirect needs to know if unbh
 970                         * was already up to date, so it can decide if the
 971                         * data in unbh needs to be replaced with data from
 972                         * the disk
 973                         */
 974                        set_buffer_uptodate(unbh);
 975
 976                        /*
 977                         * unbh->b_page == NULL in case of DIRECT_IO request,
 978                         * this means buffer will disappear shortly, so it
 979                         * should not be added to
 980                         */
 981                        if (unbh->b_page) {
 982                                /*
 983                                 * we've converted the tail, so we must
 984                                 * flush unbh before the transaction commits
 985                                 */
 986                                reiserfs_add_tail_list(inode, unbh);
 987
 988                                /*
 989                                 * mark it dirty now to prevent commit_write
 990                                 * from adding this buffer to the inode's
 991                                 * dirty buffer list
 992                                 */
 993                                /*
 994                                 * AKPM: changed __mark_buffer_dirty to
 995                                 * mark_buffer_dirty().  It's still atomic,
 996                                 * but it sets the page dirty too, which makes
 997                                 * it eligible for writeback at any time by the
 998                                 * VM (which was also the case with
 999                                 * __mark_buffer_dirty())
1000                                 */
1001                                mark_buffer_dirty(unbh);
1002                        }
1003                } else {
1004                        /*
1005                         * append indirect item with holes if needed, when
1006                         * appending pointer to 'block'-th block use block,
1007                         * which is already allocated
1008                         */
1009                        struct cpu_key tmp_key;
1010                        /*
1011                         * We use this in case we need to allocate
1012                         * only one block which is a fastpath
1013                         */
1014                        unp_t unf_single = 0;
1015                        unp_t *un;
1016                        __u64 max_to_insert =
1017                            MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
1018                            UNFM_P_SIZE;
1019                        __u64 blocks_needed;
1020
1021                        RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
1022                               "vs-804: invalid position for append");
1023                        /*
1024                         * indirect item has to be appended,
1025                         * set up key of that position
1026                         * (key type is unimportant)
1027                         */
1028                        make_cpu_key(&tmp_key, inode,
1029                                     le_key_k_offset(version,
1030                                                     &ih->ih_key) +
1031                                     op_bytes_number(ih,
1032                                                     inode->i_sb->s_blocksize),
1033                                     TYPE_INDIRECT, 3);
1034
1035                        RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
1036                               "green-805: invalid offset");
1037                        blocks_needed =
1038                            1 +
1039                            ((cpu_key_k_offset(&key) -
1040                              cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
1041                             s_blocksize_bits);
1042
1043                        if (blocks_needed == 1) {
1044                                un = &unf_single;
1045                        } else {
1046                                un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
1047                                if (!un) {
1048                                        un = &unf_single;
1049                                        blocks_needed = 1;
1050                                        max_to_insert = 0;
1051                                }
1052                        }
1053                        if (blocks_needed <= max_to_insert) {
1054                                /*
1055                                 * we are going to add target block to
1056                                 * the file. Use allocated block for that
1057                                 */
1058                                un[blocks_needed - 1] =
1059                                    cpu_to_le32(allocated_block_nr);
1060                                set_block_dev_mapped(bh_result,
1061                                                     allocated_block_nr, inode);
1062                                set_buffer_new(bh_result);
1063                                done = 1;
1064                        } else {
1065                                /* paste hole to the indirect item */
1066                                /*
1067                                 * If kmalloc failed, max_to_insert becomes
1068                                 * zero and it means we only have space for
1069                                 * one block
1070                                 */
1071                                blocks_needed =
1072                                    max_to_insert ? max_to_insert : 1;
1073                        }
1074                        retval =
1075                            reiserfs_paste_into_item(th, &path, &tmp_key, inode,
1076                                                     (char *)un,
1077                                                     UNFM_P_SIZE *
1078                                                     blocks_needed);
1079
1080                        if (blocks_needed != 1)
1081                                kfree(un);
1082
1083                        if (retval) {
1084                                reiserfs_free_block(th, inode,
1085                                                    allocated_block_nr, 1);
1086                                goto failure;
1087                        }
1088                        if (!done) {
1089                                /*
1090                                 * We need to mark new file size in case
1091                                 * this function will be interrupted/aborted
1092                                 * later on. And we may do this only for
1093                                 * holes.
1094                                 */
1095                                inode->i_size +=
1096                                    inode->i_sb->s_blocksize * blocks_needed;
1097                        }
1098                }
1099
1100                if (done == 1)
1101                        break;
1102
1103                /*
1104                 * this loop could log more blocks than we had originally
1105                 * asked for.  So, we have to allow the transaction to end
1106                 * if it is too big or too full.  Update the inode so things
1107                 * are consistent if we crash before the function returns
1108                 * release the path so that anybody waiting on the path before
1109                 * ending their transaction will be able to continue.
1110                 */
1111                if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1112                        retval = restart_transaction(th, inode, &path);
1113                        if (retval)
1114                                goto failure;
1115                }
1116                /*
1117                 * inserting indirect pointers for a hole can take a
1118                 * long time.  reschedule if needed and also release the write
1119                 * lock for others.
1120                 */
1121                reiserfs_cond_resched(inode->i_sb);
1122
1123                retval = search_for_position_by_key(inode->i_sb, &key, &path);
1124                if (retval == IO_ERROR) {
1125                        retval = -EIO;
1126                        goto failure;
1127                }
1128                if (retval == POSITION_FOUND) {
1129                        reiserfs_warning(inode->i_sb, "vs-825",
1130                                         "%K should not be found", &key);
1131                        retval = -EEXIST;
1132                        if (allocated_block_nr)
1133                                reiserfs_free_block(th, inode,
1134                                                    allocated_block_nr, 1);
1135                        pathrelse(&path);
1136                        goto failure;
1137                }
1138                bh = get_last_bh(&path);
1139                ih = tp_item_head(&path);
1140                item = tp_item_body(&path);
1141                pos_in_item = path.pos_in_item;
1142        } while (1);
1143
1144        retval = 0;
1145
1146failure:
1147        if (th && (!dangle || (retval && !th->t_trans_id))) {
1148                int err;
1149                if (th->t_trans_id)
1150                        reiserfs_update_sd(th, inode);
1151                err = reiserfs_end_persistent_transaction(th);
1152                if (err)
1153                        retval = err;
1154        }
1155
1156        reiserfs_write_unlock(inode->i_sb);
1157        reiserfs_check_path(&path);
1158        return retval;
1159}
1160
1161static int
1162reiserfs_readpages(struct file *file, struct address_space *mapping,
1163                   struct list_head *pages, unsigned nr_pages)
1164{
1165        return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1166}
1167
1168/*
1169 * Compute real number of used bytes by file
1170 * Following three functions can go away when we'll have enough space in
1171 * stat item
1172 */
1173static int real_space_diff(struct inode *inode, int sd_size)
1174{
1175        int bytes;
1176        loff_t blocksize = inode->i_sb->s_blocksize;
1177
1178        if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1179                return sd_size;
1180
1181        /*
1182         * End of file is also in full block with indirect reference, so round
1183         * up to the next block.
1184         *
1185         * there is just no way to know if the tail is actually packed
1186         * on the file, so we have to assume it isn't.  When we pack the
1187         * tail, we add 4 bytes to pretend there really is an unformatted
1188         * node pointer
1189         */
1190        bytes =
1191            ((inode->i_size +
1192              (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1193            sd_size;
1194        return bytes;
1195}
1196
1197static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1198                                        int sd_size)
1199{
1200        if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1201                return inode->i_size +
1202                    (loff_t) (real_space_diff(inode, sd_size));
1203        }
1204        return ((loff_t) real_space_diff(inode, sd_size)) +
1205            (((loff_t) blocks) << 9);
1206}
1207
1208/* Compute number of blocks used by file in ReiserFS counting */
1209static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1210{
1211        loff_t bytes = inode_get_bytes(inode);
1212        loff_t real_space = real_space_diff(inode, sd_size);
1213
1214        /* keeps fsck and non-quota versions of reiserfs happy */
1215        if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1216                bytes += (loff_t) 511;
1217        }
1218
1219        /*
1220         * files from before the quota patch might i_blocks such that
1221         * bytes < real_space.  Deal with that here to prevent it from
1222         * going negative.
1223         */
1224        if (bytes < real_space)
1225                return 0;
1226        return (bytes - real_space) >> 9;
1227}
1228
1229/*
1230 * BAD: new directories have stat data of new type and all other items
1231 * of old type. Version stored in the inode says about body items, so
1232 * in update_stat_data we can not rely on inode, but have to check
1233 * item version directly
1234 */
1235
1236/* called by read_locked_inode */
1237static void init_inode(struct inode *inode, struct treepath *path)
1238{
1239        struct buffer_head *bh;
1240        struct item_head *ih;
1241        __u32 rdev;
1242
1243        bh = PATH_PLAST_BUFFER(path);
1244        ih = tp_item_head(path);
1245
1246        copy_key(INODE_PKEY(inode), &ih->ih_key);
1247
1248        INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
1249        REISERFS_I(inode)->i_flags = 0;
1250        REISERFS_I(inode)->i_prealloc_block = 0;
1251        REISERFS_I(inode)->i_prealloc_count = 0;
1252        REISERFS_I(inode)->i_trans_id = 0;
1253        REISERFS_I(inode)->i_jl = NULL;
1254        reiserfs_init_xattr_rwsem(inode);
1255
1256        if (stat_data_v1(ih)) {
1257                struct stat_data_v1 *sd =
1258                    (struct stat_data_v1 *)ih_item_body(bh, ih);
1259                unsigned long blocks;
1260
1261                set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1262                set_inode_sd_version(inode, STAT_DATA_V1);
1263                inode->i_mode = sd_v1_mode(sd);
1264                set_nlink(inode, sd_v1_nlink(sd));
1265                i_uid_write(inode, sd_v1_uid(sd));
1266                i_gid_write(inode, sd_v1_gid(sd));
1267                inode->i_size = sd_v1_size(sd);
1268                inode->i_atime.tv_sec = sd_v1_atime(sd);
1269                inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1270                inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1271                inode->i_atime.tv_nsec = 0;
1272                inode->i_ctime.tv_nsec = 0;
1273                inode->i_mtime.tv_nsec = 0;
1274
1275                inode->i_blocks = sd_v1_blocks(sd);
1276                inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1277                blocks = (inode->i_size + 511) >> 9;
1278                blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1279
1280                /*
1281                 * there was a bug in <=3.5.23 when i_blocks could take
1282                 * negative values. Starting from 3.5.17 this value could
1283                 * even be stored in stat data. For such files we set
1284                 * i_blocks based on file size. Just 2 notes: this can be
1285                 * wrong for sparse files. On-disk value will be only
1286                 * updated if file's inode will ever change
1287                 */
1288                if (inode->i_blocks > blocks) {
1289                        inode->i_blocks = blocks;
1290                }
1291
1292                rdev = sd_v1_rdev(sd);
1293                REISERFS_I(inode)->i_first_direct_byte =
1294                    sd_v1_first_direct_byte(sd);
1295
1296                /*
1297                 * an early bug in the quota code can give us an odd
1298                 * number for the block count.  This is incorrect, fix it here.
1299                 */
1300                if (inode->i_blocks & 1) {
1301                        inode->i_blocks++;
1302                }
1303                inode_set_bytes(inode,
1304                                to_real_used_space(inode, inode->i_blocks,
1305                                                   SD_V1_SIZE));
1306                /*
1307                 * nopack is initially zero for v1 objects. For v2 objects,
1308                 * nopack is initialised from sd_attrs
1309                 */
1310                REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1311        } else {
1312                /*
1313                 * new stat data found, but object may have old items
1314                 * (directories and symlinks)
1315                 */
1316                struct stat_data *sd = (struct stat_data *)ih_item_body(bh, ih);
1317
1318                inode->i_mode = sd_v2_mode(sd);
1319                set_nlink(inode, sd_v2_nlink(sd));
1320                i_uid_write(inode, sd_v2_uid(sd));
1321                inode->i_size = sd_v2_size(sd);
1322                i_gid_write(inode, sd_v2_gid(sd));
1323                inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1324                inode->i_atime.tv_sec = sd_v2_atime(sd);
1325                inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1326                inode->i_ctime.tv_nsec = 0;
1327                inode->i_mtime.tv_nsec = 0;
1328                inode->i_atime.tv_nsec = 0;
1329                inode->i_blocks = sd_v2_blocks(sd);
1330                rdev = sd_v2_rdev(sd);
1331                if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1332                        inode->i_generation =
1333                            le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1334                else
1335                        inode->i_generation = sd_v2_generation(sd);
1336
1337                if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1338                        set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1339                else
1340                        set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1341                REISERFS_I(inode)->i_first_direct_byte = 0;
1342                set_inode_sd_version(inode, STAT_DATA_V2);
1343                inode_set_bytes(inode,
1344                                to_real_used_space(inode, inode->i_blocks,
1345                                                   SD_V2_SIZE));
1346                /*
1347                 * read persistent inode attributes from sd and initialise
1348                 * generic inode flags from them
1349                 */
1350                REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1351                sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1352        }
1353
1354        pathrelse(path);
1355        if (S_ISREG(inode->i_mode)) {
1356                inode->i_op = &reiserfs_file_inode_operations;
1357                inode->i_fop = &reiserfs_file_operations;
1358                inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1359        } else if (S_ISDIR(inode->i_mode)) {
1360                inode->i_op = &reiserfs_dir_inode_operations;
1361                inode->i_fop = &reiserfs_dir_operations;
1362        } else if (S_ISLNK(inode->i_mode)) {
1363                inode->i_op = &reiserfs_symlink_inode_operations;
1364                inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1365        } else {
1366                inode->i_blocks = 0;
1367                inode->i_op = &reiserfs_special_inode_operations;
1368                init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1369        }
1370}
1371
1372/* update new stat data with inode fields */
1373static void inode2sd(void *sd, struct inode *inode, loff_t size)
1374{
1375        struct stat_data *sd_v2 = (struct stat_data *)sd;
1376        __u16 flags;
1377
1378        set_sd_v2_mode(sd_v2, inode->i_mode);
1379        set_sd_v2_nlink(sd_v2, inode->i_nlink);
1380        set_sd_v2_uid(sd_v2, i_uid_read(inode));
1381        set_sd_v2_size(sd_v2, size);
1382        set_sd_v2_gid(sd_v2, i_gid_read(inode));
1383        set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1384        set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1385        set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1386        set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1387        if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1388                set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1389        else
1390                set_sd_v2_generation(sd_v2, inode->i_generation);
1391        flags = REISERFS_I(inode)->i_attrs;
1392        i_attrs_to_sd_attrs(inode, &flags);
1393        set_sd_v2_attrs(sd_v2, flags);
1394}
1395
1396/* used to copy inode's fields to old stat data */
1397static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1398{
1399        struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1400
1401        set_sd_v1_mode(sd_v1, inode->i_mode);
1402        set_sd_v1_uid(sd_v1, i_uid_read(inode));
1403        set_sd_v1_gid(sd_v1, i_gid_read(inode));
1404        set_sd_v1_nlink(sd_v1, inode->i_nlink);
1405        set_sd_v1_size(sd_v1, size);
1406        set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1407        set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1408        set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1409
1410        if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1411                set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1412        else
1413                set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1414
1415        /* Sigh. i_first_direct_byte is back */
1416        set_sd_v1_first_direct_byte(sd_v1,
1417                                    REISERFS_I(inode)->i_first_direct_byte);
1418}
1419
1420/*
1421 * NOTE, you must prepare the buffer head before sending it here,
1422 * and then log it after the call
1423 */
1424static void update_stat_data(struct treepath *path, struct inode *inode,
1425                             loff_t size)
1426{
1427        struct buffer_head *bh;
1428        struct item_head *ih;
1429
1430        bh = PATH_PLAST_BUFFER(path);
1431        ih = tp_item_head(path);
1432
1433        if (!is_statdata_le_ih(ih))
1434                reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1435                               INODE_PKEY(inode), ih);
1436
1437        /* path points to old stat data */
1438        if (stat_data_v1(ih)) {
1439                inode2sd_v1(ih_item_body(bh, ih), inode, size);
1440        } else {
1441                inode2sd(ih_item_body(bh, ih), inode, size);
1442        }
1443
1444        return;
1445}
1446
1447void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1448                             struct inode *inode, loff_t size)
1449{
1450        struct cpu_key key;
1451        INITIALIZE_PATH(path);
1452        struct buffer_head *bh;
1453        int fs_gen;
1454        struct item_head *ih, tmp_ih;
1455        int retval;
1456
1457        BUG_ON(!th->t_trans_id);
1458
1459        /* key type is unimportant */
1460        make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);
1461
1462        for (;;) {
1463                int pos;
1464                /* look for the object's stat data */
1465                retval = search_item(inode->i_sb, &key, &path);
1466                if (retval == IO_ERROR) {
1467                        reiserfs_error(inode->i_sb, "vs-13050",
1468                                       "i/o failure occurred trying to "
1469                                       "update %K stat data", &key);
1470                        return;
1471                }
1472                if (retval == ITEM_NOT_FOUND) {
1473                        pos = PATH_LAST_POSITION(&path);
1474                        pathrelse(&path);
1475                        if (inode->i_nlink == 0) {
1476                                /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1477                                return;
1478                        }
1479                        reiserfs_warning(inode->i_sb, "vs-13060",
1480                                         "stat data of object %k (nlink == %d) "
1481                                         "not found (pos %d)",
1482                                         INODE_PKEY(inode), inode->i_nlink,
1483                                         pos);
1484                        reiserfs_check_path(&path);
1485                        return;
1486                }
1487
1488                /*
1489                 * sigh, prepare_for_journal might schedule.  When it
1490                 * schedules the FS might change.  We have to detect that,
1491                 * and loop back to the search if the stat data item has moved
1492                 */
1493                bh = get_last_bh(&path);
1494                ih = tp_item_head(&path);
1495                copy_item_head(&tmp_ih, ih);
1496                fs_gen = get_generation(inode->i_sb);
1497                reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1498
1499                /* Stat_data item has been moved after scheduling. */
1500                if (fs_changed(fs_gen, inode->i_sb)
1501                    && item_moved(&tmp_ih, &path)) {
1502                        reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1503                        continue;
1504                }
1505                break;
1506        }
1507        update_stat_data(&path, inode, size);
1508        journal_mark_dirty(th, bh);
1509        pathrelse(&path);
1510        return;
1511}
1512
1513/*
1514 * reiserfs_read_locked_inode is called to read the inode off disk, and it
1515 * does a make_bad_inode when things go wrong.  But, we need to make sure
1516 * and clear the key in the private portion of the inode, otherwise a
1517 * corresponding iput might try to delete whatever object the inode last
1518 * represented.
1519 */
1520static void reiserfs_make_bad_inode(struct inode *inode)
1521{
1522        memset(INODE_PKEY(inode), 0, KEY_SIZE);
1523        make_bad_inode(inode);
1524}
1525
1526/*
1527 * initially this function was derived from minix or ext2's analog and
1528 * evolved as the prototype did
1529 */
1530int reiserfs_init_locked_inode(struct inode *inode, void *p)
1531{
1532        struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1533        inode->i_ino = args->objectid;
1534        INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1535        return 0;
1536}
1537
1538/*
1539 * looks for stat data in the tree, and fills up the fields of in-core
1540 * inode stat data fields
1541 */
1542void reiserfs_read_locked_inode(struct inode *inode,
1543                                struct reiserfs_iget_args *args)
1544{
1545        INITIALIZE_PATH(path_to_sd);
1546        struct cpu_key key;
1547        unsigned long dirino;
1548        int retval;
1549
1550        dirino = args->dirid;
1551
1552        /*
1553         * set version 1, version 2 could be used too, because stat data
1554         * key is the same in both versions
1555         */
1556        key.version = KEY_FORMAT_3_5;
1557        key.on_disk_key.k_dir_id = dirino;
1558        key.on_disk_key.k_objectid = inode->i_ino;
1559        key.on_disk_key.k_offset = 0;
1560        key.on_disk_key.k_type = 0;
1561
1562        /* look for the object's stat data */
1563        retval = search_item(inode->i_sb, &key, &path_to_sd);
1564        if (retval == IO_ERROR) {
1565                reiserfs_error(inode->i_sb, "vs-13070",
1566                               "i/o failure occurred trying to find "
1567                               "stat data of %K", &key);
1568                reiserfs_make_bad_inode(inode);
1569                return;
1570        }
1571
1572        /* a stale NFS handle can trigger this without it being an error */
1573        if (retval != ITEM_FOUND) {
1574                pathrelse(&path_to_sd);
1575                reiserfs_make_bad_inode(inode);
1576                clear_nlink(inode);
1577                return;
1578        }
1579
1580        init_inode(inode, &path_to_sd);
1581
1582        /*
1583         * It is possible that knfsd is trying to access inode of a file
1584         * that is being removed from the disk by some other thread. As we
1585         * update sd on unlink all that is required is to check for nlink
1586         * here. This bug was first found by Sizif when debugging
1587         * SquidNG/Butterfly, forgotten, and found again after Philippe
1588         * Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1589
1590         * More logical fix would require changes in fs/inode.c:iput() to
1591         * remove inode from hash-table _after_ fs cleaned disk stuff up and
1592         * in iget() to return NULL if I_FREEING inode is found in
1593         * hash-table.
1594         */
1595
1596        /*
1597         * Currently there is one place where it's ok to meet inode with
1598         * nlink==0: processing of open-unlinked and half-truncated files
1599         * during mount (fs/reiserfs/super.c:finish_unfinished()).
1600         */
1601        if ((inode->i_nlink == 0) &&
1602            !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1603                reiserfs_warning(inode->i_sb, "vs-13075",
1604                                 "dead inode read from disk %K. "
1605                                 "This is likely to be race with knfsd. Ignore",
1606                                 &key);
1607                reiserfs_make_bad_inode(inode);
1608        }
1609
1610        /* init inode should be relsing */
1611        reiserfs_check_path(&path_to_sd);
1612
1613        /*
1614         * Stat data v1 doesn't support ACLs.
1615         */
1616        if (get_inode_sd_version(inode) == STAT_DATA_V1)
1617                cache_no_acl(inode);
1618}
1619
1620/*
1621 * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1622 *
1623 * @inode:    inode from hash table to check
1624 * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1625 *
1626 * This function is called by iget5_locked() to distinguish reiserfs inodes
1627 * having the same inode numbers. Such inodes can only exist due to some
1628 * error condition. One of them should be bad. Inodes with identical
1629 * inode numbers (objectids) are distinguished by parent directory ids.
1630 *
1631 */
1632int reiserfs_find_actor(struct inode *inode, void *opaque)
1633{
1634        struct reiserfs_iget_args *args;
1635
1636        args = opaque;
1637        /* args is already in CPU order */
1638        return (inode->i_ino == args->objectid) &&
1639            (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1640}
1641
1642struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1643{
1644        struct inode *inode;
1645        struct reiserfs_iget_args args;
1646        int depth;
1647
1648        args.objectid = key->on_disk_key.k_objectid;
1649        args.dirid = key->on_disk_key.k_dir_id;
1650        depth = reiserfs_write_unlock_nested(s);
1651        inode = iget5_locked(s, key->on_disk_key.k_objectid,
1652                             reiserfs_find_actor, reiserfs_init_locked_inode,
1653                             (void *)(&args));
1654        reiserfs_write_lock_nested(s, depth);
1655        if (!inode)
1656                return ERR_PTR(-ENOMEM);
1657
1658        if (inode->i_state & I_NEW) {
1659                reiserfs_read_locked_inode(inode, &args);
1660                unlock_new_inode(inode);
1661        }
1662
1663        if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1664                /* either due to i/o error or a stale NFS handle */
1665                iput(inode);
1666                inode = NULL;
1667        }
1668        return inode;
1669}
1670
1671static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1672        u32 objectid, u32 dir_id, u32 generation)
1673
1674{
1675        struct cpu_key key;
1676        struct inode *inode;
1677
1678        key.on_disk_key.k_objectid = objectid;
1679        key.on_disk_key.k_dir_id = dir_id;
1680        reiserfs_write_lock(sb);
1681        inode = reiserfs_iget(sb, &key);
1682        if (inode && !IS_ERR(inode) && generation != 0 &&
1683            generation != inode->i_generation) {
1684                iput(inode);
1685                inode = NULL;
1686        }
1687        reiserfs_write_unlock(sb);
1688
1689        return d_obtain_alias(inode);
1690}
1691
1692struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1693                int fh_len, int fh_type)
1694{
1695        /*
1696         * fhtype happens to reflect the number of u32s encoded.
1697         * due to a bug in earlier code, fhtype might indicate there
1698         * are more u32s then actually fitted.
1699         * so if fhtype seems to be more than len, reduce fhtype.
1700         * Valid types are:
1701         *   2 - objectid + dir_id - legacy support
1702         *   3 - objectid + dir_id + generation
1703         *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1704         *   5 - objectid + dir_id + generation + objectid and dirid of parent
1705         *   6 - as above plus generation of directory
1706         * 6 does not fit in NFSv2 handles
1707         */
1708        if (fh_type > fh_len) {
1709                if (fh_type != 6 || fh_len != 5)
1710                        reiserfs_warning(sb, "reiserfs-13077",
1711                                "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1712                                fh_type, fh_len);
1713                fh_type = fh_len;
1714        }
1715        if (fh_len < 2)
1716                return NULL;
1717
1718        return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1719                (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1720}
1721
1722struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1723                int fh_len, int fh_type)
1724{
1725        if (fh_type > fh_len)
1726                fh_type = fh_len;
1727        if (fh_type < 4)
1728                return NULL;
1729
1730        return reiserfs_get_dentry(sb,
1731                (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1732                (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1733                (fh_type == 6) ? fid->raw[5] : 0);
1734}
1735
1736int reiserfs_encode_fh(struct inode *inode, __u32 * data, int *lenp,
1737                       struct inode *parent)
1738{
1739        int maxlen = *lenp;
1740
1741        if (parent && (maxlen < 5)) {
1742                *lenp = 5;
1743                return FILEID_INVALID;
1744        } else if (maxlen < 3) {
1745                *lenp = 3;
1746                return FILEID_INVALID;
1747        }
1748
1749        data[0] = inode->i_ino;
1750        data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1751        data[2] = inode->i_generation;
1752        *lenp = 3;
1753        if (parent) {
1754                data[3] = parent->i_ino;
1755                data[4] = le32_to_cpu(INODE_PKEY(parent)->k_dir_id);
1756                *lenp = 5;
1757                if (maxlen >= 6) {
1758                        data[5] = parent->i_generation;
1759                        *lenp = 6;
1760                }
1761        }
1762        return *lenp;
1763}
1764
1765/*
1766 * looks for stat data, then copies fields to it, marks the buffer
1767 * containing stat data as dirty
1768 */
1769/*
1770 * reiserfs inodes are never really dirty, since the dirty inode call
1771 * always logs them.  This call allows the VFS inode marking routines
1772 * to properly mark inodes for datasync and such, but only actually
1773 * does something when called for a synchronous update.
1774 */
1775int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1776{
1777        struct reiserfs_transaction_handle th;
1778        int jbegin_count = 1;
1779
1780        if (inode->i_sb->s_flags & MS_RDONLY)
1781                return -EROFS;
1782        /*
1783         * memory pressure can sometimes initiate write_inode calls with
1784         * sync == 1,
1785         * these cases are just when the system needs ram, not when the
1786         * inode needs to reach disk for safety, and they can safely be
1787         * ignored because the altered inode has already been logged.
1788         */
1789        if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1790                reiserfs_write_lock(inode->i_sb);
1791                if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1792                        reiserfs_update_sd(&th, inode);
1793                        journal_end_sync(&th);
1794                }
1795                reiserfs_write_unlock(inode->i_sb);
1796        }
1797        return 0;
1798}
1799
1800/*
1801 * stat data of new object is inserted already, this inserts the item
1802 * containing "." and ".." entries
1803 */
1804static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1805                                  struct inode *inode,
1806                                  struct item_head *ih, struct treepath *path,
1807                                  struct inode *dir)
1808{
1809        struct super_block *sb = th->t_super;
1810        char empty_dir[EMPTY_DIR_SIZE];
1811        char *body = empty_dir;
1812        struct cpu_key key;
1813        int retval;
1814
1815        BUG_ON(!th->t_trans_id);
1816
1817        _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1818                      le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1819                      TYPE_DIRENTRY, 3 /*key length */ );
1820
1821        /*
1822         * compose item head for new item. Directories consist of items of
1823         * old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1824         * is done by reiserfs_new_inode
1825         */
1826        if (old_format_only(sb)) {
1827                make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1828                                  TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1829
1830                make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1831                                       ih->ih_key.k_objectid,
1832                                       INODE_PKEY(dir)->k_dir_id,
1833                                       INODE_PKEY(dir)->k_objectid);
1834        } else {
1835                make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1836                                  TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1837
1838                make_empty_dir_item(body, ih->ih_key.k_dir_id,
1839                                    ih->ih_key.k_objectid,
1840                                    INODE_PKEY(dir)->k_dir_id,
1841                                    INODE_PKEY(dir)->k_objectid);
1842        }
1843
1844        /* look for place in the tree for new item */
1845        retval = search_item(sb, &key, path);
1846        if (retval == IO_ERROR) {
1847                reiserfs_error(sb, "vs-13080",
1848                               "i/o failure occurred creating new directory");
1849                return -EIO;
1850        }
1851        if (retval == ITEM_FOUND) {
1852                pathrelse(path);
1853                reiserfs_warning(sb, "vs-13070",
1854                                 "object with this key exists (%k)",
1855                                 &(ih->ih_key));
1856                return -EEXIST;
1857        }
1858
1859        /* insert item, that is empty directory item */
1860        return reiserfs_insert_item(th, path, &key, ih, inode, body);
1861}
1862
1863/*
1864 * stat data of object has been inserted, this inserts the item
1865 * containing the body of symlink
1866 */
1867static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th,
1868                                struct inode *inode,
1869                                struct item_head *ih,
1870                                struct treepath *path, const char *symname,
1871                                int item_len)
1872{
1873        struct super_block *sb = th->t_super;
1874        struct cpu_key key;
1875        int retval;
1876
1877        BUG_ON(!th->t_trans_id);
1878
1879        _make_cpu_key(&key, KEY_FORMAT_3_5,
1880                      le32_to_cpu(ih->ih_key.k_dir_id),
1881                      le32_to_cpu(ih->ih_key.k_objectid),
1882                      1, TYPE_DIRECT, 3 /*key length */ );
1883
1884        make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1885                          0 /*free_space */ );
1886
1887        /* look for place in the tree for new item */
1888        retval = search_item(sb, &key, path);
1889        if (retval == IO_ERROR) {
1890                reiserfs_error(sb, "vs-13080",
1891                               "i/o failure occurred creating new symlink");
1892                return -EIO;
1893        }
1894        if (retval == ITEM_FOUND) {
1895                pathrelse(path);
1896                reiserfs_warning(sb, "vs-13080",
1897                                 "object with this key exists (%k)",
1898                                 &(ih->ih_key));
1899                return -EEXIST;
1900        }
1901
1902        /* insert item, that is body of symlink */
1903        return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1904}
1905
1906/*
1907 * inserts the stat data into the tree, and then calls
1908 * reiserfs_new_directory (to insert ".", ".." item if new object is
1909 * directory) or reiserfs_new_symlink (to insert symlink body if new
1910 * object is symlink) or nothing (if new object is regular file)
1911
1912 * NOTE! uid and gid must already be set in the inode.  If we return
1913 * non-zero due to an error, we have to drop the quota previously allocated
1914 * for the fresh inode.  This can only be done outside a transaction, so
1915 * if we return non-zero, we also end the transaction.
1916 *
1917 * @th: active transaction handle
1918 * @dir: parent directory for new inode
1919 * @mode: mode of new inode
1920 * @symname: symlink contents if inode is symlink
1921 * @isize: 0 for regular file, EMPTY_DIR_SIZE for dirs, strlen(symname) for
1922 *         symlinks
1923 * @inode: inode to be filled
1924 * @security: optional security context to associate with this inode
1925 */
1926int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1927                       struct inode *dir, umode_t mode, const char *symname,
1928                       /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1929                          strlen (symname) for symlinks) */
1930                       loff_t i_size, struct dentry *dentry,
1931                       struct inode *inode,
1932                       struct reiserfs_security_handle *security)
1933{
1934        struct super_block *sb = dir->i_sb;
1935        struct reiserfs_iget_args args;
1936        INITIALIZE_PATH(path_to_key);
1937        struct cpu_key key;
1938        struct item_head ih;
1939        struct stat_data sd;
1940        int retval;
1941        int err;
1942        int depth;
1943
1944        BUG_ON(!th->t_trans_id);
1945
1946        depth = reiserfs_write_unlock_nested(sb);
1947        err = dquot_alloc_inode(inode);
1948        reiserfs_write_lock_nested(sb, depth);
1949        if (err)
1950                goto out_end_trans;
1951        if (!dir->i_nlink) {
1952                err = -EPERM;
1953                goto out_bad_inode;
1954        }
1955
1956        /* item head of new item */
1957        ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1958        ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1959        if (!ih.ih_key.k_objectid) {
1960                err = -ENOMEM;
1961                goto out_bad_inode;
1962        }
1963        args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1964        if (old_format_only(sb))
1965                make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1966                                  TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1967        else
1968                make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1969                                  TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1970        memcpy(INODE_PKEY(inode), &ih.ih_key, KEY_SIZE);
1971        args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1972
1973        depth = reiserfs_write_unlock_nested(inode->i_sb);
1974        err = insert_inode_locked4(inode, args.objectid,
1975                             reiserfs_find_actor, &args);
1976        reiserfs_write_lock_nested(inode->i_sb, depth);
1977        if (err) {
1978                err = -EINVAL;
1979                goto out_bad_inode;
1980        }
1981
1982        if (old_format_only(sb))
1983                /*
1984                 * not a perfect generation count, as object ids can be reused,
1985                 * but this is as good as reiserfs can do right now.
1986                 * note that the private part of inode isn't filled in yet,
1987                 * we have to use the directory.
1988                 */
1989                inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1990        else
1991#if defined( USE_INODE_GENERATION_COUNTER )
1992                inode->i_generation =
1993                    le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1994#else
1995                inode->i_generation = ++event;
1996#endif
1997
1998        /* fill stat data */
1999        set_nlink(inode, (S_ISDIR(mode) ? 2 : 1));
2000
2001        /* uid and gid must already be set by the caller for quota init */
2002
2003        /* symlink cannot be immutable or append only, right? */
2004        if (S_ISLNK(inode->i_mode))
2005                inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
2006
2007        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
2008        inode->i_size = i_size;
2009        inode->i_blocks = 0;
2010        inode->i_bytes = 0;
2011        REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
2012            U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
2013
2014        INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
2015        REISERFS_I(inode)->i_flags = 0;
2016        REISERFS_I(inode)->i_prealloc_block = 0;
2017        REISERFS_I(inode)->i_prealloc_count = 0;
2018        REISERFS_I(inode)->i_trans_id = 0;
2019        REISERFS_I(inode)->i_jl = NULL;
2020        REISERFS_I(inode)->i_attrs =
2021            REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
2022        sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
2023        reiserfs_init_xattr_rwsem(inode);
2024
2025        /* key to search for correct place for new stat data */
2026        _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
2027                      le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
2028                      TYPE_STAT_DATA, 3 /*key length */ );
2029
2030        /* find proper place for inserting of stat data */
2031        retval = search_item(sb, &key, &path_to_key);
2032        if (retval == IO_ERROR) {
2033                err = -EIO;
2034                goto out_bad_inode;
2035        }
2036        if (retval == ITEM_FOUND) {
2037                pathrelse(&path_to_key);
2038                err = -EEXIST;
2039                goto out_bad_inode;
2040        }
2041        if (old_format_only(sb)) {
2042                /* i_uid or i_gid is too big to be stored in stat data v3.5 */
2043                if (i_uid_read(inode) & ~0xffff || i_gid_read(inode) & ~0xffff) {
2044                        pathrelse(&path_to_key);
2045                        err = -EINVAL;
2046                        goto out_bad_inode;
2047                }
2048                inode2sd_v1(&sd, inode, inode->i_size);
2049        } else {
2050                inode2sd(&sd, inode, inode->i_size);
2051        }
2052        /*
2053         * store in in-core inode the key of stat data and version all
2054         * object items will have (directory items will have old offset
2055         * format, other new objects will consist of new items)
2056         */
2057        if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
2058                set_inode_item_key_version(inode, KEY_FORMAT_3_5);
2059        else
2060                set_inode_item_key_version(inode, KEY_FORMAT_3_6);
2061        if (old_format_only(sb))
2062                set_inode_sd_version(inode, STAT_DATA_V1);
2063        else
2064                set_inode_sd_version(inode, STAT_DATA_V2);
2065
2066        /* insert the stat data into the tree */
2067#ifdef DISPLACE_NEW_PACKING_LOCALITIES
2068        if (REISERFS_I(dir)->new_packing_locality)
2069                th->displace_new_blocks = 1;
2070#endif
2071        retval =
2072            reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
2073                                 (char *)(&sd));
2074        if (retval) {
2075                err = retval;
2076                reiserfs_check_path(&path_to_key);
2077                goto out_bad_inode;
2078        }
2079#ifdef DISPLACE_NEW_PACKING_LOCALITIES
2080        if (!th->displace_new_blocks)
2081                REISERFS_I(dir)->new_packing_locality = 0;
2082#endif
2083        if (S_ISDIR(mode)) {
2084                /* insert item with "." and ".." */
2085                retval =
2086                    reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
2087        }
2088
2089        if (S_ISLNK(mode)) {
2090                /* insert body of symlink */
2091                if (!old_format_only(sb))
2092                        i_size = ROUND_UP(i_size);
2093                retval =
2094                    reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
2095                                         i_size);
2096        }
2097        if (retval) {
2098                err = retval;
2099                reiserfs_check_path(&path_to_key);
2100                journal_end(th);
2101                goto out_inserted_sd;
2102        }
2103
2104        if (reiserfs_posixacl(inode->i_sb)) {
2105                reiserfs_write_unlock(inode->i_sb);
2106                retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
2107                reiserfs_write_lock(inode->i_sb);
2108                if (retval) {
2109                        err = retval;
2110                        reiserfs_check_path(&path_to_key);
2111                        journal_end(th);
2112                        goto out_inserted_sd;
2113                }
2114        } else if (inode->i_sb->s_flags & MS_POSIXACL) {
2115                reiserfs_warning(inode->i_sb, "jdm-13090",
2116                                 "ACLs aren't enabled in the fs, "
2117                                 "but vfs thinks they are!");
2118        } else if (IS_PRIVATE(dir))
2119                inode->i_flags |= S_PRIVATE;
2120
2121        if (security->name) {
2122                reiserfs_write_unlock(inode->i_sb);
2123                retval = reiserfs_security_write(th, inode, security);
2124                reiserfs_write_lock(inode->i_sb);
2125                if (retval) {
2126                        err = retval;
2127                        reiserfs_check_path(&path_to_key);
2128                        retval = journal_end(th);
2129                        if (retval)
2130                                err = retval;
2131                        goto out_inserted_sd;
2132                }
2133        }
2134
2135        reiserfs_update_sd(th, inode);
2136        reiserfs_check_path(&path_to_key);
2137
2138        return 0;
2139
2140out_bad_inode:
2141        /* Invalidate the object, nothing was inserted yet */
2142        INODE_PKEY(inode)->k_objectid = 0;
2143
2144        /* Quota change must be inside a transaction for journaling */
2145        depth = reiserfs_write_unlock_nested(inode->i_sb);
2146        dquot_free_inode(inode);
2147        reiserfs_write_lock_nested(inode->i_sb, depth);
2148
2149out_end_trans:
2150        journal_end(th);
2151        /*
2152         * Drop can be outside and it needs more credits so it's better
2153         * to have it outside
2154         */
2155        depth = reiserfs_write_unlock_nested(inode->i_sb);
2156        dquot_drop(inode);
2157        reiserfs_write_lock_nested(inode->i_sb, depth);
2158        inode->i_flags |= S_NOQUOTA;
2159        make_bad_inode(inode);
2160
2161out_inserted_sd:
2162        clear_nlink(inode);
2163        th->t_trans_id = 0;     /* so the caller can't use this handle later */
2164        unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
2165        iput(inode);
2166        return err;
2167}
2168
2169/*
2170 * finds the tail page in the page cache,
2171 * reads the last block in.
2172 *
2173 * On success, page_result is set to a locked, pinned page, and bh_result
2174 * is set to an up to date buffer for the last block in the file.  returns 0.
2175 *
2176 * tail conversion is not done, so bh_result might not be valid for writing
2177 * check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2178 * trying to write the block.
2179 *
2180 * on failure, nonzero is returned, page_result and bh_result are untouched.
2181 */
2182static int grab_tail_page(struct inode *inode,
2183                          struct page **page_result,
2184                          struct buffer_head **bh_result)
2185{
2186
2187        /*
2188         * we want the page with the last byte in the file,
2189         * not the page that will hold the next byte for appending
2190         */
2191        unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2192        unsigned long pos = 0;
2193        unsigned long start = 0;
2194        unsigned long blocksize = inode->i_sb->s_blocksize;
2195        unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2196        struct buffer_head *bh;
2197        struct buffer_head *head;
2198        struct page *page;
2199        int error;
2200
2201        /*
2202         * we know that we are only called with inode->i_size > 0.
2203         * we also know that a file tail can never be as big as a block
2204         * If i_size % blocksize == 0, our file is currently block aligned
2205         * and it won't need converting or zeroing after a truncate.
2206         */
2207        if ((offset & (blocksize - 1)) == 0) {
2208                return -ENOENT;
2209        }
2210        page = grab_cache_page(inode->i_mapping, index);
2211        error = -ENOMEM;
2212        if (!page) {
2213                goto out;
2214        }
2215        /* start within the page of the last block in the file */
2216        start = (offset / blocksize) * blocksize;
2217
2218        error = __block_write_begin(page, start, offset - start,
2219                                    reiserfs_get_block_create_0);
2220        if (error)
2221                goto unlock;
2222
2223        head = page_buffers(page);
2224        bh = head;
2225        do {
2226                if (pos >= start) {
2227                        break;
2228                }
2229                bh = bh->b_this_page;
2230                pos += blocksize;
2231        } while (bh != head);
2232
2233        if (!buffer_uptodate(bh)) {
2234                /*
2235                 * note, this should never happen, prepare_write should be
2236                 * taking care of this for us.  If the buffer isn't up to
2237                 * date, I've screwed up the code to find the buffer, or the
2238                 * code to call prepare_write
2239                 */
2240                reiserfs_error(inode->i_sb, "clm-6000",
2241                               "error reading block %lu", bh->b_blocknr);
2242                error = -EIO;
2243                goto unlock;
2244        }
2245        *bh_result = bh;
2246        *page_result = page;
2247
2248out:
2249        return error;
2250
2251unlock:
2252        unlock_page(page);
2253        page_cache_release(page);
2254        return error;
2255}
2256
2257/*
2258 * vfs version of truncate file.  Must NOT be called with
2259 * a transaction already started.
2260 *
2261 * some code taken from block_truncate_page
2262 */
2263int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2264{
2265        struct reiserfs_transaction_handle th;
2266        /* we want the offset for the first byte after the end of the file */
2267        unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2268        unsigned blocksize = inode->i_sb->s_blocksize;
2269        unsigned length;
2270        struct page *page = NULL;
2271        int error;
2272        struct buffer_head *bh = NULL;
2273        int err2;
2274
2275        reiserfs_write_lock(inode->i_sb);
2276
2277        if (inode->i_size > 0) {
2278                error = grab_tail_page(inode, &page, &bh);
2279                if (error) {
2280                        /*
2281                         * -ENOENT means we truncated past the end of the
2282                         * file, and get_block_create_0 could not find a
2283                         * block to read in, which is ok.
2284                         */
2285                        if (error != -ENOENT)
2286                                reiserfs_error(inode->i_sb, "clm-6001",
2287                                               "grab_tail_page failed %d",
2288                                               error);
2289                        page = NULL;
2290                        bh = NULL;
2291                }
2292        }
2293
2294        /*
2295         * so, if page != NULL, we have a buffer head for the offset at
2296         * the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2297         * then we have an unformatted node.  Otherwise, we have a direct item,
2298         * and no zeroing is required on disk.  We zero after the truncate,
2299         * because the truncate might pack the item anyway
2300         * (it will unmap bh if it packs).
2301         *
2302         * it is enough to reserve space in transaction for 2 balancings:
2303         * one for "save" link adding and another for the first
2304         * cut_from_item. 1 is for update_sd
2305         */
2306        error = journal_begin(&th, inode->i_sb,
2307                              JOURNAL_PER_BALANCE_CNT * 2 + 1);
2308        if (error)
2309                goto out;
2310        reiserfs_update_inode_transaction(inode);
2311        if (update_timestamps)
2312                /*
2313                 * we are doing real truncate: if the system crashes
2314                 * before the last transaction of truncating gets committed
2315                 * - on reboot the file either appears truncated properly
2316                 * or not truncated at all
2317                 */
2318                add_save_link(&th, inode, 1);
2319        err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2320        error = journal_end(&th);
2321        if (error)
2322                goto out;
2323
2324        /* check reiserfs_do_truncate after ending the transaction */
2325        if (err2) {
2326                error = err2;
2327                goto out;
2328        }
2329        
2330        if (update_timestamps) {
2331                error = remove_save_link(inode, 1 /* truncate */);
2332                if (error)
2333                        goto out;
2334        }
2335
2336        if (page) {
2337                length = offset & (blocksize - 1);
2338                /* if we are not on a block boundary */
2339                if (length) {
2340                        length = blocksize - length;
2341                        zero_user(page, offset, length);
2342                        if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2343                                mark_buffer_dirty(bh);
2344                        }
2345                }
2346                unlock_page(page);
2347                page_cache_release(page);
2348        }
2349
2350        reiserfs_write_unlock(inode->i_sb);
2351
2352        return 0;
2353out:
2354        if (page) {
2355                unlock_page(page);
2356                page_cache_release(page);
2357        }
2358
2359        reiserfs_write_unlock(inode->i_sb);
2360
2361        return error;
2362}
2363
2364static int map_block_for_writepage(struct inode *inode,
2365                                   struct buffer_head *bh_result,
2366                                   unsigned long block)
2367{
2368        struct reiserfs_transaction_handle th;
2369        int fs_gen;
2370        struct item_head tmp_ih;
2371        struct item_head *ih;
2372        struct buffer_head *bh;
2373        __le32 *item;
2374        struct cpu_key key;
2375        INITIALIZE_PATH(path);
2376        int pos_in_item;
2377        int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2378        loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2379        int retval;
2380        int use_get_block = 0;
2381        int bytes_copied = 0;
2382        int copy_size;
2383        int trans_running = 0;
2384
2385        /*
2386         * catch places below that try to log something without
2387         * starting a trans
2388         */
2389        th.t_trans_id = 0;
2390
2391        if (!buffer_uptodate(bh_result)) {
2392                return -EIO;
2393        }
2394
2395        kmap(bh_result->b_page);
2396start_over:
2397        reiserfs_write_lock(inode->i_sb);
2398        make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2399
2400research:
2401        retval = search_for_position_by_key(inode->i_sb, &key, &path);
2402        if (retval != POSITION_FOUND) {
2403                use_get_block = 1;
2404                goto out;
2405        }
2406
2407        bh = get_last_bh(&path);
2408        ih = tp_item_head(&path);
2409        item = tp_item_body(&path);
2410        pos_in_item = path.pos_in_item;
2411
2412        /* we've found an unformatted node */
2413        if (indirect_item_found(retval, ih)) {
2414                if (bytes_copied > 0) {
2415                        reiserfs_warning(inode->i_sb, "clm-6002",
2416                                         "bytes_copied %d", bytes_copied);
2417                }
2418                if (!get_block_num(item, pos_in_item)) {
2419                        /* crap, we are writing to a hole */
2420                        use_get_block = 1;
2421                        goto out;
2422                }
2423                set_block_dev_mapped(bh_result,
2424                                     get_block_num(item, pos_in_item), inode);
2425        } else if (is_direct_le_ih(ih)) {
2426                char *p;
2427                p = page_address(bh_result->b_page);
2428                p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2429                copy_size = ih_item_len(ih) - pos_in_item;
2430
2431                fs_gen = get_generation(inode->i_sb);
2432                copy_item_head(&tmp_ih, ih);
2433
2434                if (!trans_running) {
2435                        /* vs-3050 is gone, no need to drop the path */
2436                        retval = journal_begin(&th, inode->i_sb, jbegin_count);
2437                        if (retval)
2438                                goto out;
2439                        reiserfs_update_inode_transaction(inode);
2440                        trans_running = 1;
2441                        if (fs_changed(fs_gen, inode->i_sb)
2442                            && item_moved(&tmp_ih, &path)) {
2443                                reiserfs_restore_prepared_buffer(inode->i_sb,
2444                                                                 bh);
2445                                goto research;
2446                        }
2447                }
2448
2449                reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2450
2451                if (fs_changed(fs_gen, inode->i_sb)
2452                    && item_moved(&tmp_ih, &path)) {
2453                        reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2454                        goto research;
2455                }
2456
2457                memcpy(ih_item_body(bh, ih) + pos_in_item, p + bytes_copied,
2458                       copy_size);
2459
2460                journal_mark_dirty(&th, bh);
2461                bytes_copied += copy_size;
2462                set_block_dev_mapped(bh_result, 0, inode);
2463
2464                /* are there still bytes left? */
2465                if (bytes_copied < bh_result->b_size &&
2466                    (byte_offset + bytes_copied) < inode->i_size) {
2467                        set_cpu_key_k_offset(&key,
2468                                             cpu_key_k_offset(&key) +
2469                                             copy_size);
2470                        goto research;
2471                }
2472        } else {
2473                reiserfs_warning(inode->i_sb, "clm-6003",
2474                                 "bad item inode %lu", inode->i_ino);
2475                retval = -EIO;
2476                goto out;
2477        }
2478        retval = 0;
2479
2480out:
2481        pathrelse(&path);
2482        if (trans_running) {
2483                int err = journal_end(&th);
2484                if (err)
2485                        retval = err;
2486                trans_running = 0;
2487        }
2488        reiserfs_write_unlock(inode->i_sb);
2489
2490        /* this is where we fill in holes in the file. */
2491        if (use_get_block) {
2492                retval = reiserfs_get_block(inode, block, bh_result,
2493                                            GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2494                                            | GET_BLOCK_NO_DANGLE);
2495                if (!retval) {
2496                        if (!buffer_mapped(bh_result)
2497                            || bh_result->b_blocknr == 0) {
2498                                /* get_block failed to find a mapped unformatted node. */
2499                                use_get_block = 0;
2500                                goto start_over;
2501                        }
2502                }
2503        }
2504        kunmap(bh_result->b_page);
2505
2506        if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2507                /*
2508                 * we've copied data from the page into the direct item, so the
2509                 * buffer in the page is now clean, mark it to reflect that.
2510                 */
2511                lock_buffer(bh_result);
2512                clear_buffer_dirty(bh_result);
2513                unlock_buffer(bh_result);
2514        }
2515        return retval;
2516}
2517
2518/*
2519 * mason@suse.com: updated in 2.5.54 to follow the same general io
2520 * start/recovery path as __block_write_full_page, along with special
2521 * code to handle reiserfs tails.
2522 */
2523static int reiserfs_write_full_page(struct page *page,
2524                                    struct writeback_control *wbc)
2525{
2526        struct inode *inode = page->mapping->host;
2527        unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2528        int error = 0;
2529        unsigned long block;
2530        sector_t last_block;
2531        struct buffer_head *head, *bh;
2532        int partial = 0;
2533        int nr = 0;
2534        int checked = PageChecked(page);
2535        struct reiserfs_transaction_handle th;
2536        struct super_block *s = inode->i_sb;
2537        int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2538        th.t_trans_id = 0;
2539
2540        /* no logging allowed when nonblocking or from PF_MEMALLOC */
2541        if (checked && (current->flags & PF_MEMALLOC)) {
2542                redirty_page_for_writepage(wbc, page);
2543                unlock_page(page);
2544                return 0;
2545        }
2546
2547        /*
2548         * The page dirty bit is cleared before writepage is called, which
2549         * means we have to tell create_empty_buffers to make dirty buffers
2550         * The page really should be up to date at this point, so tossing
2551         * in the BH_Uptodate is just a sanity check.
2552         */
2553        if (!page_has_buffers(page)) {
2554                create_empty_buffers(page, s->s_blocksize,
2555                                     (1 << BH_Dirty) | (1 << BH_Uptodate));
2556        }
2557        head = page_buffers(page);
2558
2559        /*
2560         * last page in the file, zero out any contents past the
2561         * last byte in the file
2562         */
2563        if (page->index >= end_index) {
2564                unsigned last_offset;
2565
2566                last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2567                /* no file contents in this page */
2568                if (page->index >= end_index + 1 || !last_offset) {
2569                        unlock_page(page);
2570                        return 0;
2571                }
2572                zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2573        }
2574        bh = head;
2575        block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2576        last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2577        /* first map all the buffers, logging any direct items we find */
2578        do {
2579                if (block > last_block) {
2580                        /*
2581                         * This can happen when the block size is less than
2582                         * the page size.  The corresponding bytes in the page
2583                         * were zero filled above
2584                         */
2585                        clear_buffer_dirty(bh);
2586                        set_buffer_uptodate(bh);
2587                } else if ((checked || buffer_dirty(bh)) &&
2588                           (!buffer_mapped(bh) || (buffer_mapped(bh)
2589                                                       && bh->b_blocknr ==
2590                                                       0))) {
2591                        /*
2592                         * not mapped yet, or it points to a direct item, search
2593                         * the btree for the mapping info, and log any direct
2594                         * items found
2595                         */
2596                        if ((error = map_block_for_writepage(inode, bh, block))) {
2597                                goto fail;
2598                        }
2599                }
2600                bh = bh->b_this_page;
2601                block++;
2602        } while (bh != head);
2603
2604        /*
2605         * we start the transaction after map_block_for_writepage,
2606         * because it can create holes in the file (an unbounded operation).
2607         * starting it here, we can make a reliable estimate for how many
2608         * blocks we're going to log
2609         */
2610        if (checked) {
2611                ClearPageChecked(page);
2612                reiserfs_write_lock(s);
2613                error = journal_begin(&th, s, bh_per_page + 1);
2614                if (error) {
2615                        reiserfs_write_unlock(s);
2616                        goto fail;
2617                }
2618                reiserfs_update_inode_transaction(inode);
2619        }
2620        /* now go through and lock any dirty buffers on the page */
2621        do {
2622                get_bh(bh);
2623                if (!buffer_mapped(bh))
2624                        continue;
2625                if (buffer_mapped(bh) && bh->b_blocknr == 0)
2626                        continue;
2627
2628                if (checked) {
2629                        reiserfs_prepare_for_journal(s, bh, 1);
2630                        journal_mark_dirty(&th, bh);
2631                        continue;
2632                }
2633                /*
2634                 * from this point on, we know the buffer is mapped to a
2635                 * real block and not a direct item
2636                 */
2637                if (wbc->sync_mode != WB_SYNC_NONE) {
2638                        lock_buffer(bh);
2639                } else {
2640                        if (!trylock_buffer(bh)) {
2641                                redirty_page_for_writepage(wbc, page);
2642                                continue;
2643                        }
2644                }
2645                if (test_clear_buffer_dirty(bh)) {
2646                        mark_buffer_async_write(bh);
2647                } else {
2648                        unlock_buffer(bh);
2649                }
2650        } while ((bh = bh->b_this_page) != head);
2651
2652        if (checked) {
2653                error = journal_end(&th);
2654                reiserfs_write_unlock(s);
2655                if (error)
2656                        goto fail;
2657        }
2658        BUG_ON(PageWriteback(page));
2659        set_page_writeback(page);
2660        unlock_page(page);
2661
2662        /*
2663         * since any buffer might be the only dirty buffer on the page,
2664         * the first submit_bh can bring the page out of writeback.
2665         * be careful with the buffers.
2666         */
2667        do {
2668                struct buffer_head *next = bh->b_this_page;
2669                if (buffer_async_write(bh)) {
2670                        submit_bh(WRITE, bh);
2671                        nr++;
2672                }
2673                put_bh(bh);
2674                bh = next;
2675        } while (bh != head);
2676
2677        error = 0;
2678done:
2679        if (nr == 0) {
2680                /*
2681                 * if this page only had a direct item, it is very possible for
2682                 * no io to be required without there being an error.  Or,
2683                 * someone else could have locked them and sent them down the
2684                 * pipe without locking the page
2685                 */
2686                bh = head;
2687                do {
2688                        if (!buffer_uptodate(bh)) {
2689                                partial = 1;
2690                                break;
2691                        }
2692                        bh = bh->b_this_page;
2693                } while (bh != head);
2694                if (!partial)
2695                        SetPageUptodate(page);
2696                end_page_writeback(page);
2697        }
2698        return error;
2699
2700fail:
2701        /*
2702         * catches various errors, we need to make sure any valid dirty blocks
2703         * get to the media.  The page is currently locked and not marked for
2704         * writeback
2705         */
2706        ClearPageUptodate(page);
2707        bh = head;
2708        do {
2709                get_bh(bh);
2710                if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2711                        lock_buffer(bh);
2712                        mark_buffer_async_write(bh);
2713                } else {
2714                        /*
2715                         * clear any dirty bits that might have come from
2716                         * getting attached to a dirty page
2717                         */
2718                        clear_buffer_dirty(bh);
2719                }
2720                bh = bh->b_this_page;
2721        } while (bh != head);
2722        SetPageError(page);
2723        BUG_ON(PageWriteback(page));
2724        set_page_writeback(page);
2725        unlock_page(page);
2726        do {
2727                struct buffer_head *next = bh->b_this_page;
2728                if (buffer_async_write(bh)) {
2729                        clear_buffer_dirty(bh);
2730                        submit_bh(WRITE, bh);
2731                        nr++;
2732                }
2733                put_bh(bh);
2734                bh = next;
2735        } while (bh != head);
2736        goto done;
2737}
2738
2739static int reiserfs_readpage(struct file *f, struct page *page)
2740{
2741        return block_read_full_page(page, reiserfs_get_block);
2742}
2743
2744static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2745{
2746        struct inode *inode = page->mapping->host;
2747        reiserfs_wait_on_write_block(inode->i_sb);
2748        return reiserfs_write_full_page(page, wbc);
2749}
2750
2751static void reiserfs_truncate_failed_write(struct inode *inode)
2752{
2753        truncate_inode_pages(inode->i_mapping, inode->i_size);
2754        reiserfs_truncate_file(inode, 0);
2755}
2756
2757static int reiserfs_write_begin(struct file *file,
2758                                struct address_space *mapping,
2759                                loff_t pos, unsigned len, unsigned flags,
2760                                struct page **pagep, void **fsdata)
2761{
2762        struct inode *inode;
2763        struct page *page;
2764        pgoff_t index;
2765        int ret;
2766        int old_ref = 0;
2767
2768        inode = mapping->host;
2769        *fsdata = 0;
2770        if (flags & AOP_FLAG_CONT_EXPAND &&
2771            (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2772                pos ++;
2773                *fsdata = (void *)(unsigned long)flags;
2774        }
2775
2776        index = pos >> PAGE_CACHE_SHIFT;
2777        page = grab_cache_page_write_begin(mapping, index, flags);
2778        if (!page)
2779                return -ENOMEM;
2780        *pagep = page;
2781
2782        reiserfs_wait_on_write_block(inode->i_sb);
2783        fix_tail_page_for_writing(page);
2784        if (reiserfs_transaction_running(inode->i_sb)) {
2785                struct reiserfs_transaction_handle *th;
2786                th = (struct reiserfs_transaction_handle *)current->
2787                    journal_info;
2788                BUG_ON(!th->t_refcount);
2789                BUG_ON(!th->t_trans_id);
2790                old_ref = th->t_refcount;
2791                th->t_refcount++;
2792        }
2793        ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2794        if (ret && reiserfs_transaction_running(inode->i_sb)) {
2795                struct reiserfs_transaction_handle *th = current->journal_info;
2796                /*
2797                 * this gets a little ugly.  If reiserfs_get_block returned an
2798                 * error and left a transacstion running, we've got to close
2799                 * it, and we've got to free handle if it was a persistent
2800                 * transaction.
2801                 *
2802                 * But, if we had nested into an existing transaction, we need
2803                 * to just drop the ref count on the handle.
2804                 *
2805                 * If old_ref == 0, the transaction is from reiserfs_get_block,
2806                 * and it was a persistent trans.  Otherwise, it was nested
2807                 * above.
2808                 */
2809                if (th->t_refcount > old_ref) {
2810                        if (old_ref)
2811                                th->t_refcount--;
2812                        else {
2813                                int err;
2814                                reiserfs_write_lock(inode->i_sb);
2815                                err = reiserfs_end_persistent_transaction(th);
2816                                reiserfs_write_unlock(inode->i_sb);
2817                                if (err)
2818                                        ret = err;
2819                        }
2820                }
2821        }
2822        if (ret) {
2823                unlock_page(page);
2824                page_cache_release(page);
2825                /* Truncate allocated blocks */
2826                reiserfs_truncate_failed_write(inode);
2827        }
2828        return ret;
2829}
2830
2831int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2832{
2833        struct inode *inode = page->mapping->host;
2834        int ret;
2835        int old_ref = 0;
2836        int depth;
2837
2838        depth = reiserfs_write_unlock_nested(inode->i_sb);
2839        reiserfs_wait_on_write_block(inode->i_sb);
2840        reiserfs_write_lock_nested(inode->i_sb, depth);
2841
2842        fix_tail_page_for_writing(page);
2843        if (reiserfs_transaction_running(inode->i_sb)) {
2844                struct reiserfs_transaction_handle *th;
2845                th = (struct reiserfs_transaction_handle *)current->
2846                    journal_info;
2847                BUG_ON(!th->t_refcount);
2848                BUG_ON(!th->t_trans_id);
2849                old_ref = th->t_refcount;
2850                th->t_refcount++;
2851        }
2852
2853        ret = __block_write_begin(page, from, len, reiserfs_get_block);
2854        if (ret && reiserfs_transaction_running(inode->i_sb)) {
2855                struct reiserfs_transaction_handle *th = current->journal_info;
2856                /*
2857                 * this gets a little ugly.  If reiserfs_get_block returned an
2858                 * error and left a transacstion running, we've got to close
2859                 * it, and we've got to free handle if it was a persistent
2860                 * transaction.
2861                 *
2862                 * But, if we had nested into an existing transaction, we need
2863                 * to just drop the ref count on the handle.
2864                 *
2865                 * If old_ref == 0, the transaction is from reiserfs_get_block,
2866                 * and it was a persistent trans.  Otherwise, it was nested
2867                 * above.
2868                 */
2869                if (th->t_refcount > old_ref) {
2870                        if (old_ref)
2871                                th->t_refcount--;
2872                        else {
2873                                int err;
2874                                reiserfs_write_lock(inode->i_sb);
2875                                err = reiserfs_end_persistent_transaction(th);
2876                                reiserfs_write_unlock(inode->i_sb);
2877                                if (err)
2878                                        ret = err;
2879                        }
2880                }
2881        }
2882        return ret;
2883
2884}
2885
2886static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2887{
2888        return generic_block_bmap(as, block, reiserfs_bmap);
2889}
2890
2891static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2892                              loff_t pos, unsigned len, unsigned copied,
2893                              struct page *page, void *fsdata)
2894{
2895        struct inode *inode = page->mapping->host;
2896        int ret = 0;
2897        int update_sd = 0;
2898        struct reiserfs_transaction_handle *th;
2899        unsigned start;
2900        bool locked = false;
2901
2902        if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2903                pos ++;
2904
2905        reiserfs_wait_on_write_block(inode->i_sb);
2906        if (reiserfs_transaction_running(inode->i_sb))
2907                th = current->journal_info;
2908        else
2909                th = NULL;
2910
2911        start = pos & (PAGE_CACHE_SIZE - 1);
2912        if (unlikely(copied < len)) {
2913                if (!PageUptodate(page))
2914                        copied = 0;
2915
2916                page_zero_new_buffers(page, start + copied, start + len);
2917        }
2918        flush_dcache_page(page);
2919
2920        reiserfs_commit_page(inode, page, start, start + copied);
2921
2922        /*
2923         * generic_commit_write does this for us, but does not update the
2924         * transaction tracking stuff when the size changes.  So, we have
2925         * to do the i_size updates here.
2926         */
2927        if (pos + copied > inode->i_size) {
2928                struct reiserfs_transaction_handle myth;
2929                reiserfs_write_lock(inode->i_sb);
2930                locked = true;
2931                /*
2932                 * If the file have grown beyond the border where it
2933                 * can have a tail, unmark it as needing a tail
2934                 * packing
2935                 */
2936                if ((have_large_tails(inode->i_sb)
2937                     && inode->i_size > i_block_size(inode) * 4)
2938                    || (have_small_tails(inode->i_sb)
2939                        && inode->i_size > i_block_size(inode)))
2940                        REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2941
2942                ret = journal_begin(&myth, inode->i_sb, 1);
2943                if (ret)
2944                        goto journal_error;
2945
2946                reiserfs_update_inode_transaction(inode);
2947                inode->i_size = pos + copied;
2948                /*
2949                 * this will just nest into our transaction.  It's important
2950                 * to use mark_inode_dirty so the inode gets pushed around on
2951                 * the dirty lists, and so that O_SYNC works as expected
2952                 */
2953                mark_inode_dirty(inode);
2954                reiserfs_update_sd(&myth, inode);
2955                update_sd = 1;
2956                ret = journal_end(&myth);
2957                if (ret)
2958                        goto journal_error;
2959        }
2960        if (th) {
2961                if (!locked) {
2962                        reiserfs_write_lock(inode->i_sb);
2963                        locked = true;
2964                }
2965                if (!update_sd)
2966                        mark_inode_dirty(inode);
2967                ret = reiserfs_end_persistent_transaction(th);
2968                if (ret)
2969                        goto out;
2970        }
2971
2972out:
2973        if (locked)
2974                reiserfs_write_unlock(inode->i_sb);
2975        unlock_page(page);
2976        page_cache_release(page);
2977
2978        if (pos + len > inode->i_size)
2979                reiserfs_truncate_failed_write(inode);
2980
2981        return ret == 0 ? copied : ret;
2982
2983journal_error:
2984        reiserfs_write_unlock(inode->i_sb);
2985        locked = false;
2986        if (th) {
2987                if (!update_sd)
2988                        reiserfs_update_sd(th, inode);
2989                ret = reiserfs_end_persistent_transaction(th);
2990        }
2991        goto out;
2992}
2993
2994int reiserfs_commit_write(struct file *f, struct page *page,
2995                          unsigned from, unsigned to)
2996{
2997        struct inode *inode = page->mapping->host;
2998        loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2999        int ret = 0;
3000        int update_sd = 0;
3001        struct reiserfs_transaction_handle *th = NULL;
3002        int depth;
3003
3004        depth = reiserfs_write_unlock_nested(inode->i_sb);
3005        reiserfs_wait_on_write_block(inode->i_sb);
3006        reiserfs_write_lock_nested(inode->i_sb, depth);
3007
3008        if (reiserfs_transaction_running(inode->i_sb)) {
3009                th = current->journal_info;
3010        }
3011        reiserfs_commit_page(inode, page, from, to);
3012
3013        /*
3014         * generic_commit_write does this for us, but does not update the
3015         * transaction tracking stuff when the size changes.  So, we have
3016         * to do the i_size updates here.
3017         */
3018        if (pos > inode->i_size) {
3019                struct reiserfs_transaction_handle myth;
3020                /*
3021                 * If the file have grown beyond the border where it
3022                 * can have a tail, unmark it as needing a tail
3023                 * packing
3024                 */
3025                if ((have_large_tails(inode->i_sb)
3026                     && inode->i_size > i_block_size(inode) * 4)
3027                    || (have_small_tails(inode->i_sb)
3028                        && inode->i_size > i_block_size(inode)))
3029                        REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
3030
3031                ret = journal_begin(&myth, inode->i_sb, 1);
3032                if (ret)
3033                        goto journal_error;
3034
3035                reiserfs_update_inode_transaction(inode);
3036                inode->i_size = pos;
3037                /*
3038                 * this will just nest into our transaction.  It's important
3039                 * to use mark_inode_dirty so the inode gets pushed around
3040                 * on the dirty lists, and so that O_SYNC works as expected
3041                 */
3042                mark_inode_dirty(inode);
3043                reiserfs_update_sd(&myth, inode);
3044                update_sd = 1;
3045                ret = journal_end(&myth);
3046                if (ret)
3047                        goto journal_error;
3048        }
3049        if (th) {
3050                if (!update_sd)
3051                        mark_inode_dirty(inode);
3052                ret = reiserfs_end_persistent_transaction(th);
3053                if (ret)
3054                        goto out;
3055        }
3056
3057out:
3058        return ret;
3059
3060journal_error:
3061        if (th) {
3062                if (!update_sd)
3063                        reiserfs_update_sd(th, inode);
3064                ret = reiserfs_end_persistent_transaction(th);
3065        }
3066
3067        return ret;
3068}
3069
3070void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
3071{
3072        if (reiserfs_attrs(inode->i_sb)) {
3073                if (sd_attrs & REISERFS_SYNC_FL)
3074                        inode->i_flags |= S_SYNC;
3075                else
3076                        inode->i_flags &= ~S_SYNC;
3077                if (sd_attrs & REISERFS_IMMUTABLE_FL)
3078                        inode->i_flags |= S_IMMUTABLE;
3079                else
3080                        inode->i_flags &= ~S_IMMUTABLE;
3081                if (sd_attrs & REISERFS_APPEND_FL)
3082                        inode->i_flags |= S_APPEND;
3083                else
3084                        inode->i_flags &= ~S_APPEND;
3085                if (sd_attrs & REISERFS_NOATIME_FL)
3086                        inode->i_flags |= S_NOATIME;
3087                else
3088                        inode->i_flags &= ~S_NOATIME;
3089                if (sd_attrs & REISERFS_NOTAIL_FL)
3090                        REISERFS_I(inode)->i_flags |= i_nopack_mask;
3091                else
3092                        REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
3093        }
3094}
3095
3096void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
3097{
3098        if (reiserfs_attrs(inode->i_sb)) {
3099                if (inode->i_flags & S_IMMUTABLE)
3100                        *sd_attrs |= REISERFS_IMMUTABLE_FL;
3101                else
3102                        *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
3103                if (inode->i_flags & S_SYNC)
3104                        *sd_attrs |= REISERFS_SYNC_FL;
3105                else
3106                        *sd_attrs &= ~REISERFS_SYNC_FL;
3107                if (inode->i_flags & S_NOATIME)
3108                        *sd_attrs |= REISERFS_NOATIME_FL;
3109                else
3110                        *sd_attrs &= ~REISERFS_NOATIME_FL;
3111                if (REISERFS_I(inode)->i_flags & i_nopack_mask)
3112                        *sd_attrs |= REISERFS_NOTAIL_FL;
3113                else
3114                        *sd_attrs &= ~REISERFS_NOTAIL_FL;
3115        }
3116}
3117
3118/*
3119 * decide if this buffer needs to stay around for data logging or ordered
3120 * write purposes
3121 */
3122static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
3123{
3124        int ret = 1;
3125        struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3126
3127        lock_buffer(bh);
3128        spin_lock(&j->j_dirty_buffers_lock);
3129        if (!buffer_mapped(bh)) {
3130                goto free_jh;
3131        }
3132        /*
3133         * the page is locked, and the only places that log a data buffer
3134         * also lock the page.
3135         */
3136        if (reiserfs_file_data_log(inode)) {
3137                /*
3138                 * very conservative, leave the buffer pinned if
3139                 * anyone might need it.
3140                 */
3141                if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
3142                        ret = 0;
3143                }
3144        } else  if (buffer_dirty(bh)) {
3145                struct reiserfs_journal_list *jl;
3146                struct reiserfs_jh *jh = bh->b_private;
3147
3148                /*
3149                 * why is this safe?
3150                 * reiserfs_setattr updates i_size in the on disk
3151                 * stat data before allowing vmtruncate to be called.
3152                 *
3153                 * If buffer was put onto the ordered list for this
3154                 * transaction, we know for sure either this transaction
3155                 * or an older one already has updated i_size on disk,
3156                 * and this ordered data won't be referenced in the file
3157                 * if we crash.
3158                 *
3159                 * if the buffer was put onto the ordered list for an older
3160                 * transaction, we need to leave it around
3161                 */
3162                if (jh && (jl = jh->jl)
3163                    && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
3164                        ret = 0;
3165        }
3166free_jh:
3167        if (ret && bh->b_private) {
3168                reiserfs_free_jh(bh);
3169        }
3170        spin_unlock(&j->j_dirty_buffers_lock);
3171        unlock_buffer(bh);
3172        return ret;
3173}
3174
3175/* clm -- taken from fs/buffer.c:block_invalidate_page */
3176static void reiserfs_invalidatepage(struct page *page, unsigned int offset,
3177                                    unsigned int length)
3178{
3179        struct buffer_head *head, *bh, *next;
3180        struct inode *inode = page->mapping->host;
3181        unsigned int curr_off = 0;
3182        unsigned int stop = offset + length;
3183        int partial_page = (offset || length < PAGE_CACHE_SIZE);
3184        int ret = 1;
3185
3186        BUG_ON(!PageLocked(page));
3187
3188        if (!partial_page)
3189                ClearPageChecked(page);
3190
3191        if (!page_has_buffers(page))
3192                goto out;
3193
3194        head = page_buffers(page);
3195        bh = head;
3196        do {
3197                unsigned int next_off = curr_off + bh->b_size;
3198                next = bh->b_this_page;
3199
3200                if (next_off > stop)
3201                        goto out;
3202
3203                /*
3204                 * is this block fully invalidated?
3205                 */
3206                if (offset <= curr_off) {
3207                        if (invalidatepage_can_drop(inode, bh))
3208                                reiserfs_unmap_buffer(bh);
3209                        else
3210                                ret = 0;
3211                }
3212                curr_off = next_off;
3213                bh = next;
3214        } while (bh != head);
3215
3216        /*
3217         * We release buffers only if the entire page is being invalidated.
3218         * The get_block cached value has been unconditionally invalidated,
3219         * so real IO is not possible anymore.
3220         */
3221        if (!partial_page && ret) {
3222                ret = try_to_release_page(page, 0);
3223                /* maybe should BUG_ON(!ret); - neilb */
3224        }
3225out:
3226        return;
3227}
3228
3229static int reiserfs_set_page_dirty(struct page *page)
3230{
3231        struct inode *inode = page->mapping->host;
3232        if (reiserfs_file_data_log(inode)) {
3233                SetPageChecked(page);
3234                return __set_page_dirty_nobuffers(page);
3235        }
3236        return __set_page_dirty_buffers(page);
3237}
3238
3239/*
3240 * Returns 1 if the page's buffers were dropped.  The page is locked.
3241 *
3242 * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3243 * in the buffers at page_buffers(page).
3244 *
3245 * even in -o notail mode, we can't be sure an old mount without -o notail
3246 * didn't create files with tails.
3247 */
3248static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3249{
3250        struct inode *inode = page->mapping->host;
3251        struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3252        struct buffer_head *head;
3253        struct buffer_head *bh;
3254        int ret = 1;
3255
3256        WARN_ON(PageChecked(page));
3257        spin_lock(&j->j_dirty_buffers_lock);
3258        head = page_buffers(page);
3259        bh = head;
3260        do {
3261                if (bh->b_private) {
3262                        if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3263                                reiserfs_free_jh(bh);
3264                        } else {
3265                                ret = 0;
3266                                break;
3267                        }
3268                }
3269                bh = bh->b_this_page;
3270        } while (bh != head);
3271        if (ret)
3272                ret = try_to_free_buffers(page);
3273        spin_unlock(&j->j_dirty_buffers_lock);
3274        return ret;
3275}
3276
3277/*
3278 * We thank Mingming Cao for helping us understand in great detail what
3279 * to do in this section of the code.
3280 */
3281static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3282                                  struct iov_iter *iter, loff_t offset)
3283{
3284        struct file *file = iocb->ki_filp;
3285        struct inode *inode = file->f_mapping->host;
3286        size_t count = iov_iter_count(iter);
3287        ssize_t ret;
3288
3289        ret = blockdev_direct_IO(rw, iocb, inode, iter, offset,
3290                                 reiserfs_get_blocks_direct_io);
3291
3292        /*
3293         * In case of error extending write may have instantiated a few
3294         * blocks outside i_size. Trim these off again.
3295         */
3296        if (unlikely((rw & WRITE) && ret < 0)) {
3297                loff_t isize = i_size_read(inode);
3298                loff_t end = offset + count;
3299
3300                if ((end > isize) && inode_newsize_ok(inode, isize) == 0) {
3301                        truncate_setsize(inode, isize);
3302                        reiserfs_vfs_truncate_file(inode);
3303                }
3304        }
3305
3306        return ret;
3307}
3308
3309int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3310{
3311        struct inode *inode = dentry->d_inode;
3312        unsigned int ia_valid;
3313        int error;
3314
3315        error = inode_change_ok(inode, attr);
3316        if (error)
3317                return error;
3318
3319        /* must be turned off for recursive notify_change calls */
3320        ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3321
3322        if (is_quota_modification(inode, attr))
3323                dquot_initialize(inode);
3324        reiserfs_write_lock(inode->i_sb);
3325        if (attr->ia_valid & ATTR_SIZE) {
3326                /*
3327                 * version 2 items will be caught by the s_maxbytes check
3328                 * done for us in vmtruncate
3329                 */
3330                if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3331                    attr->ia_size > MAX_NON_LFS) {
3332                        reiserfs_write_unlock(inode->i_sb);
3333                        error = -EFBIG;
3334                        goto out;
3335                }
3336
3337                inode_dio_wait(inode);
3338
3339                /* fill in hole pointers in the expanding truncate case. */
3340                if (attr->ia_size > inode->i_size) {
3341                        error = generic_cont_expand_simple(inode, attr->ia_size);
3342                        if (REISERFS_I(inode)->i_prealloc_count > 0) {
3343                                int err;
3344                                struct reiserfs_transaction_handle th;
3345                                /* we're changing at most 2 bitmaps, inode + super */
3346                                err = journal_begin(&th, inode->i_sb, 4);
3347                                if (!err) {
3348                                        reiserfs_discard_prealloc(&th, inode);
3349                                        err = journal_end(&th);
3350                                }
3351                                if (err)
3352                                        error = err;
3353                        }
3354                        if (error) {
3355                                reiserfs_write_unlock(inode->i_sb);
3356                                goto out;
3357                        }
3358                        /*
3359                         * file size is changed, ctime and mtime are
3360                         * to be updated
3361                         */
3362                        attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3363                }
3364        }
3365        reiserfs_write_unlock(inode->i_sb);
3366
3367        if ((((attr->ia_valid & ATTR_UID) && (from_kuid(&init_user_ns, attr->ia_uid) & ~0xffff)) ||
3368             ((attr->ia_valid & ATTR_GID) && (from_kgid(&init_user_ns, attr->ia_gid) & ~0xffff))) &&
3369            (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3370                /* stat data of format v3.5 has 16 bit uid and gid */
3371                error = -EINVAL;
3372                goto out;
3373        }
3374
3375        if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3376            (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3377                struct reiserfs_transaction_handle th;
3378                int jbegin_count =
3379                    2 *
3380                    (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3381                     REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3382                    2;
3383
3384                error = reiserfs_chown_xattrs(inode, attr);
3385
3386                if (error)
3387                        return error;
3388
3389                /*
3390                 * (user+group)*(old+new) structure - we count quota
3391                 * info and , inode write (sb, inode)
3392                 */
3393                reiserfs_write_lock(inode->i_sb);
3394                error = journal_begin(&th, inode->i_sb, jbegin_count);
3395                reiserfs_write_unlock(inode->i_sb);
3396                if (error)
3397                        goto out;
3398                error = dquot_transfer(inode, attr);
3399                reiserfs_write_lock(inode->i_sb);
3400                if (error) {
3401                        journal_end(&th);
3402                        reiserfs_write_unlock(inode->i_sb);
3403                        goto out;
3404                }
3405
3406                /*
3407                 * Update corresponding info in inode so that everything
3408                 * is in one transaction
3409                 */
3410                if (attr->ia_valid & ATTR_UID)
3411                        inode->i_uid = attr->ia_uid;
3412                if (attr->ia_valid & ATTR_GID)
3413                        inode->i_gid = attr->ia_gid;
3414                mark_inode_dirty(inode);
3415                error = journal_end(&th);
3416                reiserfs_write_unlock(inode->i_sb);
3417                if (error)
3418                        goto out;
3419        }
3420
3421        if ((attr->ia_valid & ATTR_SIZE) &&
3422            attr->ia_size != i_size_read(inode)) {
3423                error = inode_newsize_ok(inode, attr->ia_size);
3424                if (!error) {
3425                        /*
3426                         * Could race against reiserfs_file_release
3427                         * if called from NFS, so take tailpack mutex.
3428                         */
3429                        mutex_lock(&REISERFS_I(inode)->tailpack);
3430                        truncate_setsize(inode, attr->ia_size);
3431                        reiserfs_truncate_file(inode, 1);
3432                        mutex_unlock(&REISERFS_I(inode)->tailpack);
3433                }
3434        }
3435
3436        if (!error) {
3437                setattr_copy(inode, attr);
3438                mark_inode_dirty(inode);
3439        }
3440
3441        if (!error && reiserfs_posixacl(inode->i_sb)) {
3442                if (attr->ia_valid & ATTR_MODE)
3443                        error = reiserfs_acl_chmod(inode);
3444        }
3445
3446out:
3447        return error;
3448}
3449
3450const struct address_space_operations reiserfs_address_space_operations = {
3451        .writepage = reiserfs_writepage,
3452        .readpage = reiserfs_readpage,
3453        .readpages = reiserfs_readpages,
3454        .releasepage = reiserfs_releasepage,
3455        .invalidatepage = reiserfs_invalidatepage,
3456        .write_begin = reiserfs_write_begin,
3457        .write_end = reiserfs_write_end,
3458        .bmap = reiserfs_aop_bmap,
3459        .direct_IO = reiserfs_direct_IO,
3460        .set_page_dirty = reiserfs_set_page_dirty,
3461};
3462