fs/reiserfs/inode.c

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