SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/file.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/mpage.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/version.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "ioctl.h"
  38 #include "print-tree.h"
  39 #include "tree-log.h"
  40 #include "locking.h"
  41 #include "compat.h"
  42
  43
  44 /* simple helper to fault in pages and copy.  This should go away
  45  * and be replaced with calls into generic code.
  46  */
  47 static int noinline btrfs_copy_from_user(loff_t pos, int num_pages,
  48                                          int write_bytes,
  49                                          struct page **prepared_pages,
  50                                          const char __user * buf)
  51 {
  52         long page_fault = 0;
  53         int i;
  54         int offset = pos & (PAGE_CACHE_SIZE - 1);
  55
  56         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  57                 size_t count = min_t(size_t,
  58                                      PAGE_CACHE_SIZE - offset, write_bytes);
  59                 struct page *page = prepared_pages[i];
  60                 fault_in_pages_readable(buf, count);
  61
  62                 /* Copy data from userspace to the current page */
  63                 kmap(page);
  64                 page_fault = __copy_from_user(page_address(page) + offset,
  65                                               buf, count);
  66                 /* Flush processor's dcache for this page */
  67                 flush_dcache_page(page);
  68                 kunmap(page);
  69                 buf += count;
  70                 write_bytes -= count;
  71
  72                 if (page_fault)
  73                         break;
  74         }
  75         return page_fault ? -EFAULT : 0;
  76 }
  77
  78 /*
  79  * unlocks pages after btrfs_file_write is done with them
  80  */
  81 static void noinline btrfs_drop_pages(struct page **pages, size_t num_pages)
  82 {
  83         size_t i;
  84         for (i = 0; i < num_pages; i++) {
  85                 if (!pages[i])
  86                         break;
  87                 /* page checked is some magic around finding pages that
  88                  * have been modified without going through btrfs_set_page_dirty
  89                  * clear it here
  90                  */
  91                 ClearPageChecked(pages[i]);
  92                 unlock_page(pages[i]);
  93                 mark_page_accessed(pages[i]);
  94                 page_cache_release(pages[i]);
  95         }
  96 }
  97
  98 /* this does all the hard work for inserting an inline extent into
  99  * the btree.  Any existing inline extent is extended as required to make room,
 100  * otherwise things are inserted as required into the btree
 101  */
 102 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
 103                                 struct btrfs_root *root, struct inode *inode,
 104                                 u64 offset, size_t size,
 105                                 struct page **pages, size_t page_offset,
 106                                 int num_pages)
 107 {
 108         struct btrfs_key key;
 109         struct btrfs_path *path;
 110         struct extent_buffer *leaf;
 111         char *kaddr;
 112         unsigned long ptr;
 113         struct btrfs_file_extent_item *ei;
 114         struct page *page;
 115         u32 datasize;
 116         int err = 0;
 117         int ret;
 118         int i;
 119         ssize_t cur_size;
 120
 121         path = btrfs_alloc_path();
 122         if (!path)
 123                 return -ENOMEM;
 124
 125         btrfs_set_trans_block_group(trans, inode);
 126
 127         key.objectid = inode->i_ino;
 128         key.offset = offset;
 129         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 130
 131         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 132         if (ret < 0) {
 133                 err = ret;
 134                 goto fail;
 135         }
 136         if (ret == 1) {
 137                 struct btrfs_key found_key;
 138
 139                 if (path->slots[0] == 0)
 140                         goto insert;
 141
 142                 path->slots[0]--;
 143                 leaf = path->nodes[0];
 144                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 145
 146                 if (found_key.objectid != inode->i_ino)
 147                         goto insert;
 148
 149                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 150                         goto insert;
 151                 ei = btrfs_item_ptr(leaf, path->slots[0],
 152                                     struct btrfs_file_extent_item);
 153
 154                 if (btrfs_file_extent_type(leaf, ei) !=
 155                     BTRFS_FILE_EXTENT_INLINE) {
 156                         goto insert;
 157                 }
 158                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 159                 ret = 0;
 160         }
 161         if (ret == 0) {
 162                 u32 found_size;
 163                 u64 found_end;
 164
 165                 leaf = path->nodes[0];
 166                 ei = btrfs_item_ptr(leaf, path->slots[0],
 167                                     struct btrfs_file_extent_item);
 168
 169                 if (btrfs_file_extent_type(leaf, ei) !=
 170                     BTRFS_FILE_EXTENT_INLINE) {
 171                         err = ret;
 172                         btrfs_print_leaf(root, leaf);
 173                         printk("found wasn't inline offset %Lu inode %lu\n",
 174                                offset, inode->i_ino);
 175                         goto fail;
 176                 }
 177                 found_size = btrfs_file_extent_inline_len(leaf,
 178                                           btrfs_item_nr(leaf, path->slots[0]));
 179                 found_end = key.offset + found_size;
 180
 181                 if (found_end < offset + size) {
 182                         btrfs_release_path(root, path);
 183                         ret = btrfs_search_slot(trans, root, &key, path,
 184                                                 offset + size - found_end, 1);
 185                         BUG_ON(ret != 0);
 186
 187                         ret = btrfs_extend_item(trans, root, path,
 188                                                 offset + size - found_end);
 189                         if (ret) {
 190                                 err = ret;
 191                                 goto fail;
 192                         }
 193                         leaf = path->nodes[0];
 194                         ei = btrfs_item_ptr(leaf, path->slots[0],
 195                                             struct btrfs_file_extent_item);
 196                         inode_add_bytes(inode, offset + size - found_end);
 197                 }
 198                 if (found_end < offset) {
 199                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 200                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 201                 }
 202         } else {
 203 insert:
 204                 btrfs_release_path(root, path);
 205                 datasize = offset + size - key.offset;
 206                 inode_add_bytes(inode, datasize);
 207                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 208                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 209                                               datasize);
 210                 if (ret) {
 211                         err = ret;
 212                         printk("got bad ret %d\n", ret);
 213                         goto fail;
 214                 }
 215                 leaf = path->nodes[0];
 216                 ei = btrfs_item_ptr(leaf, path->slots[0],
 217                                     struct btrfs_file_extent_item);
 218                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 219                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 220         }
 221         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 222
 223         cur_size = size;
 224         i = 0;
 225         while (size > 0) {
 226                 page = pages[i];
 227                 kaddr = kmap_atomic(page, KM_USER0);
 228                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 229                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 230                 kunmap_atomic(kaddr, KM_USER0);
 231                 page_offset = 0;
 232                 ptr += cur_size;
 233                 size -= cur_size;
 234                 if (i >= num_pages) {
 235                         printk("i %d num_pages %d\n", i, num_pages);
 236                 }
 237                 i++;
 238         }
 239         btrfs_mark_buffer_dirty(leaf);
 240 fail:
 241         btrfs_free_path(path);
 242         return err;
 243 }
 244
 245 /*
 246  * after copy_from_user, pages need to be dirtied and we need to make
 247  * sure holes are created between the current EOF and the start of
 248  * any next extents (if required).
 249  *
 250  * this also makes the decision about creating an inline extent vs
 251  * doing real data extents, marking pages dirty and delalloc as required.
 252  */
 253 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 254                                    struct btrfs_root *root,
 255                                    struct file *file,
 256                                    struct page **pages,
 257                                    size_t num_pages,
 258                                    loff_t pos,
 259                                    size_t write_bytes)
 260 {
 261         int err = 0;
 262         int i;
 263         struct inode *inode = fdentry(file)->d_inode;
 264         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 265         u64 hint_byte;
 266         u64 num_bytes;
 267         u64 start_pos;
 268         u64 end_of_last_block;
 269         u64 end_pos = pos + write_bytes;
 270         u64 inline_size;
 271         int did_inline = 0;
 272         loff_t isize = i_size_read(inode);
 273
 274         start_pos = pos & ~((u64)root->sectorsize - 1);
 275         num_bytes = (write_bytes + pos - start_pos +
 276                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 277
 278         end_of_last_block = start_pos + num_bytes - 1;
 279
 280         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 281         trans = btrfs_join_transaction(root, 1);
 282         if (!trans) {
 283                 err = -ENOMEM;
 284                 goto out_unlock;
 285         }
 286         btrfs_set_trans_block_group(trans, inode);
 287         hint_byte = 0;
 288
 289         if ((end_of_last_block & 4095) == 0) {
 290                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 291         }
 292         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 293
 294         /* FIXME...EIEIO, ENOSPC and more */
 295         /* insert any holes we need to create */
 296         if (isize < start_pos) {
 297                 u64 last_pos_in_file;
 298                 u64 hole_size;
 299                 u64 mask = root->sectorsize - 1;
 300                 last_pos_in_file = (isize + mask) & ~mask;
 301                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 302                 if (hole_size > 0) {
 303                         btrfs_wait_ordered_range(inode, last_pos_in_file,
 304                                                  last_pos_in_file + hole_size);
 305                         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 306                         err = btrfs_drop_extents(trans, root, inode,
 307                                                  last_pos_in_file,
 308                                                  last_pos_in_file + hole_size,
 309                                                  last_pos_in_file,
 310                                                  &hint_byte);
 311                         if (err)
 312                                 goto failed;
 313
 314                         err = btrfs_insert_file_extent(trans, root,
 315                                                        inode->i_ino,
 316                                                        last_pos_in_file,
 317                                                        0, 0, hole_size, 0);
 318                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 319                                         last_pos_in_file + hole_size - 1, 0);
 320                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 321                         btrfs_check_file(root, inode);
 322                 }
 323                 if (err)
 324                         goto failed;
 325         }
 326
 327         /*
 328          * either allocate an extent for the new bytes or setup the key
 329          * to show we are doing inline data in the extent
 330          */
 331         inline_size = end_pos;
 332         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 333             inline_size > root->fs_info->max_inline ||
 334             (inline_size & (root->sectorsize -1)) == 0 ||
 335             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 336                 /* check for reserved extents on each page, we don't want
 337                  * to reset the delalloc bit on things that already have
 338                  * extents reserved.
 339                  */
 340                 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 341                 for (i = 0; i < num_pages; i++) {
 342                         struct page *p = pages[i];
 343                         SetPageUptodate(p);
 344                         ClearPageChecked(p);
 345                         set_page_dirty(p);
 346                 }
 347         } else {
 348                 u64 aligned_end;
 349                 /* step one, delete the existing extents in this range */
 350                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 351                         ~((u64)root->sectorsize - 1);
 352                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 353                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 354                                          aligned_end, aligned_end, &hint_byte);
 355                 if (err)
 356                         goto failed;
 357                 if (isize > inline_size)
 358                         inline_size = min_t(u64, isize, aligned_end);
 359                 inline_size -= start_pos;
 360                 err = insert_inline_extent(trans, root, inode, start_pos,
 361                                            inline_size, pages, 0, num_pages);
 362                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1, 0);
 363                 BUG_ON(err);
 364                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 365
 366                 /*
 367                  * an ugly way to do all the prop accounting around
 368                  * the page bits and mapping tags
 369                  */
 370                 set_page_writeback(pages[0]);
 371                 end_page_writeback(pages[0]);
 372                 did_inline = 1;
 373         }
 374         if (end_pos > isize) {
 375                 i_size_write(inode, end_pos);
 376                 if (did_inline)
 377                         BTRFS_I(inode)->disk_i_size = end_pos;
 378                 btrfs_update_inode(trans, root, inode);
 379         }
 380 failed:
 381         err = btrfs_end_transaction(trans, root);
 382 out_unlock:
 383         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 384         return err;
 385 }
 386
 387 /*
 388  * this drops all the extents in the cache that intersect the range
 389  * [start, end].  Existing extents are split as required.
 390  */
 391 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 392                             int skip_pinned)
 393 {
 394         struct extent_map *em;
 395         struct extent_map *split = NULL;
 396         struct extent_map *split2 = NULL;
 397         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 398         u64 len = end - start + 1;
 399         int ret;
 400         int testend = 1;
 401         unsigned long flags;
 402
 403         WARN_ON(end < start);
 404         if (end == (u64)-1) {
 405                 len = (u64)-1;
 406                 testend = 0;
 407         }
 408         while(1) {
 409                 if (!split)
 410                         split = alloc_extent_map(GFP_NOFS);
 411                 if (!split2)
 412                         split2 = alloc_extent_map(GFP_NOFS);
 413
 414                 spin_lock(&em_tree->lock);
 415                 em = lookup_extent_mapping(em_tree, start, len);
 416                 if (!em) {
 417                         spin_unlock(&em_tree->lock);
 418                         break;
 419                 }
 420                 flags = em->flags;
 421                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 422                         spin_unlock(&em_tree->lock);
 423                         if (em->start <= start &&
 424                             (!testend || em->start + em->len >= start + len)) {
 425                                 free_extent_map(em);
 426                                 break;
 427                         }
 428                         if (start < em->start) {
 429                                 len = em->start - start;
 430                         } else {
 431                                 len = start + len - (em->start + em->len);
 432                                 start = em->start + em->len;
 433                         }
 434                         free_extent_map(em);
 435                         continue;
 436                 }
 437                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 438                 remove_extent_mapping(em_tree, em);
 439
 440                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 441                     em->start < start) {
 442                         split->start = em->start;
 443                         split->len = start - em->start;
 444                         split->block_start = em->block_start;
 445                         split->bdev = em->bdev;
 446                         split->flags = flags;
 447                         ret = add_extent_mapping(em_tree, split);
 448                         BUG_ON(ret);
 449                         free_extent_map(split);
 450                         split = split2;
 451                         split2 = NULL;
 452                 }
 453                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 454                     testend && em->start + em->len > start + len) {
 455                         u64 diff = start + len - em->start;
 456
 457                         split->start = start + len;
 458                         split->len = em->start + em->len - (start + len);
 459                         split->bdev = em->bdev;
 460                         split->flags = flags;
 461
 462                         split->block_start = em->block_start + diff;
 463
 464                         ret = add_extent_mapping(em_tree, split);
 465                         BUG_ON(ret);
 466                         free_extent_map(split);
 467                         split = NULL;
 468                 }
 469                 spin_unlock(&em_tree->lock);
 470
 471                 /* once for us */
 472                 free_extent_map(em);
 473                 /* once for the tree*/
 474                 free_extent_map(em);
 475         }
 476         if (split)
 477                 free_extent_map(split);
 478         if (split2)
 479                 free_extent_map(split2);
 480         return 0;
 481 }
 482
 483 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 484 {
 485         return 0;
 486 #if 0
 487         struct btrfs_path *path;
 488         struct btrfs_key found_key;
 489         struct extent_buffer *leaf;
 490         struct btrfs_file_extent_item *extent;
 491         u64 last_offset = 0;
 492         int nritems;
 493         int slot;
 494         int found_type;
 495         int ret;
 496         int err = 0;
 497         u64 extent_end = 0;
 498
 499         path = btrfs_alloc_path();
 500         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 501                                        last_offset, 0);
 502         while(1) {
 503                 nritems = btrfs_header_nritems(path->nodes[0]);
 504                 if (path->slots[0] >= nritems) {
 505                         ret = btrfs_next_leaf(root, path);
 506                         if (ret)
 507                                 goto out;
 508                         nritems = btrfs_header_nritems(path->nodes[0]);
 509                 }
 510                 slot = path->slots[0];
 511                 leaf = path->nodes[0];
 512                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 513                 if (found_key.objectid != inode->i_ino)
 514                         break;
 515                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 516                         goto out;
 517
 518                 if (found_key.offset < last_offset) {
 519                         WARN_ON(1);
 520                         btrfs_print_leaf(root, leaf);
 521                         printk("inode %lu found offset %Lu expected %Lu\n",
 522                                inode->i_ino, found_key.offset, last_offset);
 523                         err = 1;
 524                         goto out;
 525                 }
 526                 extent = btrfs_item_ptr(leaf, slot,
 527                                         struct btrfs_file_extent_item);
 528                 found_type = btrfs_file_extent_type(leaf, extent);
 529                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 530                         extent_end = found_key.offset +
 531                              btrfs_file_extent_num_bytes(leaf, extent);
 532                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 533                         struct btrfs_item *item;
 534                         item = btrfs_item_nr(leaf, slot);
 535                         extent_end = found_key.offset +
 536                              btrfs_file_extent_inline_len(leaf, item);
 537                         extent_end = (extent_end + root->sectorsize - 1) &
 538                                 ~((u64)root->sectorsize -1 );
 539                 }
 540                 last_offset = extent_end;
 541                 path->slots[0]++;
 542         }
 543         if (0 && last_offset < inode->i_size) {
 544                 WARN_ON(1);
 545                 btrfs_print_leaf(root, leaf);
 546                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 547                        last_offset, inode->i_size);
 548                 err = 1;
 549
 550         }
 551 out:
 552         btrfs_free_path(path);
 553         return err;
 554 #endif
 555 }
 556
 557 /*
 558  * this is very complex, but the basic idea is to drop all extents
 559  * in the range start - end.  hint_block is filled in with a block number
 560  * that would be a good hint to the block allocator for this file.
 561  *
 562  * If an extent intersects the range but is not entirely inside the range
 563  * it is either truncated or split.  Anything entirely inside the range
 564  * is deleted from the tree.
 565  *
 566  * inline_limit is used to tell this code which offsets in the file to keep
 567  * if they contain inline extents.
 568  */
 569 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 570                        struct btrfs_root *root, struct inode *inode,
 571                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 572 {
 573         u64 extent_end = 0;
 574         u64 search_start = start;
 575         u64 leaf_start;
 576         u64 root_gen;
 577         u64 root_owner;
 578         struct extent_buffer *leaf;
 579         struct btrfs_file_extent_item *extent;
 580         struct btrfs_path *path;
 581         struct btrfs_key key;
 582         struct btrfs_file_extent_item old;
 583         int keep;
 584         int slot;
 585         int bookend;
 586         int found_type;
 587         int found_extent;
 588         int found_inline;
 589         int recow;
 590         int ret;
 591
 592         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 593
 594         path = btrfs_alloc_path();
 595         if (!path)
 596                 return -ENOMEM;
 597         while(1) {
 598                 recow = 0;
 599                 btrfs_release_path(root, path);
 600                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 601                                                search_start, -1);
 602                 if (ret < 0)
 603                         goto out;
 604                 if (ret > 0) {
 605                         if (path->slots[0] == 0) {
 606                                 ret = 0;
 607                                 goto out;
 608                         }
 609                         path->slots[0]--;
 610                 }
 611 next_slot:
 612                 keep = 0;
 613                 bookend = 0;
 614                 found_extent = 0;
 615                 found_inline = 0;
 616                 leaf_start = 0;
 617                 root_gen = 0;
 618                 root_owner = 0;
 619                 extent = NULL;
 620                 leaf = path->nodes[0];
 621                 slot = path->slots[0];
 622                 ret = 0;
 623                 btrfs_item_key_to_cpu(leaf, &key, slot);
 624                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 625                     key.offset >= end) {
 626                         goto out;
 627                 }
 628                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 629                     key.objectid != inode->i_ino) {
 630                         goto out;
 631                 }
 632                 if (recow) {
 633                         search_start = key.offset;
 634                         continue;
 635                 }
 636                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 637                         extent = btrfs_item_ptr(leaf, slot,
 638                                                 struct btrfs_file_extent_item);
 639                         found_type = btrfs_file_extent_type(leaf, extent);
 640                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 641                                 extent_end =
 642                                      btrfs_file_extent_disk_bytenr(leaf,
 643                                                                    extent);
 644                                 if (extent_end)
 645                                         *hint_byte = extent_end;
 646
 647                                 extent_end = key.offset +
 648                                      btrfs_file_extent_num_bytes(leaf, extent);
 649                                 found_extent = 1;
 650                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 651                                 struct btrfs_item *item;
 652                                 item = btrfs_item_nr(leaf, slot);
 653                                 found_inline = 1;
 654                                 extent_end = key.offset +
 655                                      btrfs_file_extent_inline_len(leaf, item);
 656                         }
 657                 } else {
 658                         extent_end = search_start;
 659                 }
 660
 661                 /* we found nothing we can drop */
 662                 if ((!found_extent && !found_inline) ||
 663                     search_start >= extent_end) {
 664                         int nextret;
 665                         u32 nritems;
 666                         nritems = btrfs_header_nritems(leaf);
 667                         if (slot >= nritems - 1) {
 668                                 nextret = btrfs_next_leaf(root, path);
 669                                 if (nextret)
 670                                         goto out;
 671                                 recow = 1;
 672                         } else {
 673                                 path->slots[0]++;
 674                         }
 675                         goto next_slot;
 676                 }
 677
 678                 if (found_inline) {
 679                         u64 mask = root->sectorsize - 1;
 680                         search_start = (extent_end + mask) & ~mask;
 681                 } else
 682                         search_start = extent_end;
 683                 if (end <= extent_end && start >= key.offset && found_inline) {
 684                         *hint_byte = EXTENT_MAP_INLINE;
 685                         goto out;
 686                 }
 687
 688                 if (found_extent) {
 689                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 690                                            sizeof(old));
 691                         root_gen = btrfs_header_generation(leaf);
 692                         root_owner = btrfs_header_owner(leaf);
 693                         leaf_start = leaf->start;
 694                 }
 695
 696                 if (end < extent_end && end >= key.offset) {
 697                         bookend = 1;
 698                         if (found_inline && start <= key.offset)
 699                                 keep = 1;
 700                 }
 701                 /* truncate existing extent */
 702                 if (start > key.offset) {
 703                         u64 new_num;
 704                         u64 old_num;
 705                         keep = 1;
 706                         WARN_ON(start & (root->sectorsize - 1));
 707                         if (found_extent) {
 708                                 new_num = start - key.offset;
 709                                 old_num = btrfs_file_extent_num_bytes(leaf,
 710                                                                       extent);
 711                                 *hint_byte =
 712                                         btrfs_file_extent_disk_bytenr(leaf,
 713                                                                       extent);
 714                                 if (btrfs_file_extent_disk_bytenr(leaf,
 715                                                                   extent)) {
 716                                         inode_sub_bytes(inode, old_num -
 717                                                         new_num);
 718                                 }
 719                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 720                                                                 new_num);
 721                                 btrfs_mark_buffer_dirty(leaf);
 722                         } else if (key.offset < inline_limit &&
 723                                    (end > extent_end) &&
 724                                    (inline_limit < extent_end)) {
 725                                 u32 new_size;
 726                                 new_size = btrfs_file_extent_calc_inline_size(
 727                                                    inline_limit - key.offset);
 728                                 inode_sub_bytes(inode, extent_end -
 729                                                 inline_limit);
 730                                 btrfs_truncate_item(trans, root, path,
 731                                                     new_size, 1);
 732                         }
 733                 }
 734                 /* delete the entire extent */
 735                 if (!keep) {
 736                         if (found_inline)
 737                                 inode_sub_bytes(inode, extent_end -
 738                                                 key.offset);
 739                         ret = btrfs_del_item(trans, root, path);
 740                         /* TODO update progress marker and return */
 741                         BUG_ON(ret);
 742                         extent = NULL;
 743                         btrfs_release_path(root, path);
 744                         /* the extent will be freed later */
 745                 }
 746                 if (bookend && found_inline && start <= key.offset) {
 747                         u32 new_size;
 748                         new_size = btrfs_file_extent_calc_inline_size(
 749                                                    extent_end - end);
 750                         inode_sub_bytes(inode, end - key.offset);
 751                         ret = btrfs_truncate_item(trans, root, path,
 752                                                   new_size, 0);
 753                         BUG_ON(ret);
 754                 }
 755                 /* create bookend, splitting the extent in two */
 756                 if (bookend && found_extent) {
 757                         u64 disk_bytenr;
 758                         struct btrfs_key ins;
 759                         ins.objectid = inode->i_ino;
 760                         ins.offset = end;
 761                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 762                         btrfs_release_path(root, path);
 763                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 764                                                       sizeof(*extent));
 765                         BUG_ON(ret);
 766
 767                         leaf = path->nodes[0];
 768                         extent = btrfs_item_ptr(leaf, path->slots[0],
 769                                                 struct btrfs_file_extent_item);
 770                         write_extent_buffer(leaf, &old,
 771                                             (unsigned long)extent, sizeof(old));
 772
 773                         btrfs_set_file_extent_offset(leaf, extent,
 774                                     le64_to_cpu(old.offset) + end - key.offset);
 775                         WARN_ON(le64_to_cpu(old.num_bytes) <
 776                                 (extent_end - end));
 777                         btrfs_set_file_extent_num_bytes(leaf, extent,
 778                                                         extent_end - end);
 779                         btrfs_set_file_extent_type(leaf, extent,
 780                                                    BTRFS_FILE_EXTENT_REG);
 781
 782                         btrfs_mark_buffer_dirty(path->nodes[0]);
 783
 784                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 785                         if (disk_bytenr != 0) {
 786                                 ret = btrfs_inc_extent_ref(trans, root,
 787                                                 disk_bytenr,
 788                                                 le64_to_cpu(old.disk_num_bytes),
 789                                                 leaf->start,
 790                                                 root->root_key.objectid,
 791                                                 trans->transid,
 792                                                 ins.objectid, ins.offset);
 793                                 BUG_ON(ret);
 794                         }
 795                         btrfs_release_path(root, path);
 796                         if (disk_bytenr != 0) {
 797                                 inode_add_bytes(inode, extent_end - end);
 798                         }
 799                 }
 800
 801                 if (found_extent && !keep) {
 802                         u64 disk_bytenr = le64_to_cpu(old.disk_bytenr);
 803
 804                         if (disk_bytenr != 0) {
 805                                 inode_sub_bytes(inode,
 806                                                 le64_to_cpu(old.num_bytes));
 807                                 ret = btrfs_free_extent(trans, root,
 808                                                 disk_bytenr,
 809                                                 le64_to_cpu(old.disk_num_bytes),
 810                                                 leaf_start, root_owner,
 811                                                 root_gen, key.objectid,
 812                                                 key.offset, 0);
 813                                 BUG_ON(ret);
 814                                 *hint_byte = disk_bytenr;
 815                         }
 816                 }
 817
 818                 if (search_start >= end) {
 819                         ret = 0;
 820                         goto out;
 821                 }
 822         }
 823 out:
 824         btrfs_free_path(path);
 825         btrfs_check_file(root, inode);
 826         return ret;
 827 }
 828
 829 /*
 830  * this gets pages into the page cache and locks them down, it also properly
 831  * waits for data=ordered extents to finish before allowing the pages to be
 832  * modified.
 833  */
 834 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
 835                          struct page **pages, size_t num_pages,
 836                          loff_t pos, unsigned long first_index,
 837                          unsigned long last_index, size_t write_bytes)
 838 {
 839         int i;
 840         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 841         struct inode *inode = fdentry(file)->d_inode;
 842         int err = 0;
 843         u64 start_pos;
 844         u64 last_pos;
 845
 846         start_pos = pos & ~((u64)root->sectorsize - 1);
 847         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 848
 849         memset(pages, 0, num_pages * sizeof(struct page *));
 850 again:
 851         for (i = 0; i < num_pages; i++) {
 852                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 853                 if (!pages[i]) {
 854                         err = -ENOMEM;
 855                         BUG_ON(1);
 856                 }
 857                 wait_on_page_writeback(pages[i]);
 858         }
 859         if (start_pos < inode->i_size) {
 860                 struct btrfs_ordered_extent *ordered;
 861                 lock_extent(&BTRFS_I(inode)->io_tree,
 862                             start_pos, last_pos - 1, GFP_NOFS);
 863                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 864                 if (ordered &&
 865                     ordered->file_offset + ordered->len > start_pos &&
 866                     ordered->file_offset < last_pos) {
 867                         btrfs_put_ordered_extent(ordered);
 868                         unlock_extent(&BTRFS_I(inode)->io_tree,
 869                                       start_pos, last_pos - 1, GFP_NOFS);
 870                         for (i = 0; i < num_pages; i++) {
 871                                 unlock_page(pages[i]);
 872                                 page_cache_release(pages[i]);
 873                         }
 874                         btrfs_wait_ordered_range(inode, start_pos,
 875                                                  last_pos - start_pos);
 876                         goto again;
 877                 }
 878                 if (ordered)
 879                         btrfs_put_ordered_extent(ordered);
 880
 881                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 882                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 883                                   GFP_NOFS);
 884                 unlock_extent(&BTRFS_I(inode)->io_tree,
 885                               start_pos, last_pos - 1, GFP_NOFS);
 886         }
 887         for (i = 0; i < num_pages; i++) {
 888                 clear_page_dirty_for_io(pages[i]);
 889                 set_page_extent_mapped(pages[i]);
 890                 WARN_ON(!PageLocked(pages[i]));
 891         }
 892         return 0;
 893 }
 894
 895 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 896                                 size_t count, loff_t *ppos)
 897 {
 898         loff_t pos;
 899         loff_t start_pos;
 900         ssize_t num_written = 0;
 901         ssize_t err = 0;
 902         int ret = 0;
 903         struct inode *inode = fdentry(file)->d_inode;
 904         struct btrfs_root *root = BTRFS_I(inode)->root;
 905         struct page **pages = NULL;
 906         int nrptrs;
 907         struct page *pinned[2];
 908         unsigned long first_index;
 909         unsigned long last_index;
 910         int will_write;
 911
 912         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 913                       (file->f_flags & O_DIRECT));
 914
 915         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 916                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 917         pinned[0] = NULL;
 918         pinned[1] = NULL;
 919
 920         pos = *ppos;
 921         start_pos = pos;
 922
 923         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 924         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 925         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 926         if (err)
 927                 goto out_nolock;
 928         if (count == 0)
 929                 goto out_nolock;
 930
 931         err = file_remove_suid(file);
 932         if (err)
 933                 goto out_nolock;
 934         file_update_time(file);
 935
 936         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 937
 938         mutex_lock(&inode->i_mutex);
 939         first_index = pos >> PAGE_CACHE_SHIFT;
 940         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 941
 942         /*
 943          * if this is a nodatasum mount, force summing off for the inode
 944          * all the time.  That way a later mount with summing on won't
 945          * get confused
 946          */
 947         if (btrfs_test_opt(root, NODATASUM))
 948                 btrfs_set_flag(inode, NODATASUM);
 949
 950         /*
 951          * there are lots of better ways to do this, but this code
 952          * makes sure the first and last page in the file range are
 953          * up to date and ready for cow
 954          */
 955         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 956                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 957                 if (!PageUptodate(pinned[0])) {
 958                         ret = btrfs_readpage(NULL, pinned[0]);
 959                         BUG_ON(ret);
 960                         wait_on_page_locked(pinned[0]);
 961                 } else {
 962                         unlock_page(pinned[0]);
 963                 }
 964         }
 965         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 966                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 967                 if (!PageUptodate(pinned[1])) {
 968                         ret = btrfs_readpage(NULL, pinned[1]);
 969                         BUG_ON(ret);
 970                         wait_on_page_locked(pinned[1]);
 971                 } else {
 972                         unlock_page(pinned[1]);
 973                 }
 974         }
 975
 976         while(count > 0) {
 977                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 978                 size_t write_bytes = min(count, nrptrs *
 979                                         (size_t)PAGE_CACHE_SIZE -
 980                                          offset);
 981                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 982                                         PAGE_CACHE_SHIFT;
 983
 984                 WARN_ON(num_pages > nrptrs);
 985                 memset(pages, 0, sizeof(pages));
 986
 987                 ret = btrfs_check_free_space(root, write_bytes, 0);
 988                 if (ret)
 989                         goto out;
 990
 991                 ret = prepare_pages(root, file, pages, num_pages,
 992                                     pos, first_index, last_index,
 993                                     write_bytes);
 994                 if (ret)
 995                         goto out;
 996
 997                 ret = btrfs_copy_from_user(pos, num_pages,
 998                                            write_bytes, pages, buf);
 999                 if (ret) {
1000                         btrfs_drop_pages(pages, num_pages);
1001                         goto out;
1002                 }
1003
1004                 ret = dirty_and_release_pages(NULL, root, file, pages,
1005                                               num_pages, pos, write_bytes);
1006                 btrfs_drop_pages(pages, num_pages);
1007                 if (ret)
1008                         goto out;
1009
1010                 if (will_write) {
1011                         btrfs_fdatawrite_range(inode->i_mapping, pos,
1012                                                pos + write_bytes - 1,
1013                                                WB_SYNC_NONE);
1014                 } else {
1015                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1016                                                            num_pages);
1017                         if (num_pages <
1018                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1019                                 btrfs_btree_balance_dirty(root, 1);
1020                         btrfs_throttle(root);
1021                 }
1022
1023                 buf += write_bytes;
1024                 count -= write_bytes;
1025                 pos += write_bytes;
1026                 num_written += write_bytes;
1027
1028                 cond_resched();
1029         }
1030 out:
1031         mutex_unlock(&inode->i_mutex);
1032
1033 out_nolock:
1034         kfree(pages);
1035         if (pinned[0])
1036                 page_cache_release(pinned[0]);
1037         if (pinned[1])
1038                 page_cache_release(pinned[1]);
1039         *ppos = pos;
1040
1041         if (num_written > 0 && will_write) {
1042                 struct btrfs_trans_handle *trans;
1043
1044                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1045                 if (err)
1046                         num_written = err;
1047
1048                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1049                         trans = btrfs_start_transaction(root, 1);
1050                         ret = btrfs_log_dentry_safe(trans, root,
1051                                                     file->f_dentry);
1052                         if (ret == 0) {
1053                                 btrfs_sync_log(trans, root);
1054                                 btrfs_end_transaction(trans, root);
1055                         } else {
1056                                 btrfs_commit_transaction(trans, root);
1057                         }
1058                 }
1059                 if (file->f_flags & O_DIRECT) {
1060                         invalidate_mapping_pages(inode->i_mapping,
1061                               start_pos >> PAGE_CACHE_SHIFT,
1062                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1063                 }
1064         }
1065         current->backing_dev_info = NULL;
1066         return num_written ? num_written : err;
1067 }
1068
1069 int btrfs_release_file(struct inode * inode, struct file * filp)
1070 {
1071         if (filp->private_data)
1072                 btrfs_ioctl_trans_end(filp);
1073         return 0;
1074 }
1075
1076 /*
1077  * fsync call for both files and directories.  This logs the inode into
1078  * the tree log instead of forcing full commits whenever possible.
1079  *
1080  * It needs to call filemap_fdatawait so that all ordered extent updates are
1081  * in the metadata btree are up to date for copying to the log.
1082  *
1083  * It drops the inode mutex before doing the tree log commit.  This is an
1084  * important optimization for directories because holding the mutex prevents
1085  * new operations on the dir while we write to disk.
1086  */
1087 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1088 {
1089         struct inode *inode = dentry->d_inode;
1090         struct btrfs_root *root = BTRFS_I(inode)->root;
1091         int ret = 0;
1092         struct btrfs_trans_handle *trans;
1093
1094         /*
1095          * check the transaction that last modified this inode
1096          * and see if its already been committed
1097          */
1098         if (!BTRFS_I(inode)->last_trans)
1099                 goto out;
1100
1101         mutex_lock(&root->fs_info->trans_mutex);
1102         if (BTRFS_I(inode)->last_trans <=
1103             root->fs_info->last_trans_committed) {
1104                 BTRFS_I(inode)->last_trans = 0;
1105                 mutex_unlock(&root->fs_info->trans_mutex);
1106                 goto out;
1107         }
1108         mutex_unlock(&root->fs_info->trans_mutex);
1109
1110         root->fs_info->tree_log_batch++;
1111         filemap_fdatawait(inode->i_mapping);
1112         root->fs_info->tree_log_batch++;
1113
1114         /*
1115          * ok we haven't committed the transaction yet, lets do a commit
1116          */
1117         if (file->private_data)
1118                 btrfs_ioctl_trans_end(file);
1119
1120         trans = btrfs_start_transaction(root, 1);
1121         if (!trans) {
1122                 ret = -ENOMEM;
1123                 goto out;
1124         }
1125
1126         ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1127         if (ret < 0) {
1128                 goto out;
1129         }
1130
1131         /* we've logged all the items and now have a consistent
1132          * version of the file in the log.  It is possible that
1133          * someone will come in and modify the file, but that's
1134          * fine because the log is consistent on disk, and we
1135          * have references to all of the file's extents
1136          *
1137          * It is possible that someone will come in and log the
1138          * file again, but that will end up using the synchronization
1139          * inside btrfs_sync_log to keep things safe.
1140          */
1141         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
1142
1143         if (ret > 0) {
1144                 ret = btrfs_commit_transaction(trans, root);
1145         } else {
1146                 btrfs_sync_log(trans, root);
1147                 ret = btrfs_end_transaction(trans, root);
1148         }
1149         mutex_lock(&file->f_dentry->d_inode->i_mutex);
1150 out:
1151         return ret > 0 ? EIO : ret;
1152 }
1153
1154 static struct vm_operations_struct btrfs_file_vm_ops = {
1155         .fault          = filemap_fault,
1156         .page_mkwrite   = btrfs_page_mkwrite,
1157 };
1158
1159 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1160 {
1161         vma->vm_ops = &btrfs_file_vm_ops;
1162         file_accessed(filp);
1163         return 0;
1164 }
1165
1166 struct file_operations btrfs_file_operations = {
1167         .llseek         = generic_file_llseek,
1168         .read           = do_sync_read,
1169         .aio_read       = generic_file_aio_read,
1170         .splice_read    = generic_file_splice_read,
1171         .write          = btrfs_file_write,
1172         .mmap           = btrfs_file_mmap,
1173         .open           = generic_file_open,
1174         .release        = btrfs_release_file,
1175         .fsync          = btrfs_sync_file,
1176         .unlocked_ioctl = btrfs_ioctl,
1177 #ifdef CONFIG_COMPAT
1178         .compat_ioctl   = btrfs_ioctl,
1179 #endif
1180 };