git.oblomov.eu Git - 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 "ordered-data.h"
  38 #include "ioctl.h"
  39 #include "print-tree.h"
  40 #include "compat.h"
  41
  42
  43 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  44                                 struct page **prepared_pages,
  45                                 const char __user * buf)
  46 {
  47         long page_fault = 0;
  48         int i;
  49         int offset = pos & (PAGE_CACHE_SIZE - 1);
  50
  51         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  52                 size_t count = min_t(size_t,
  53                                      PAGE_CACHE_SIZE - offset, write_bytes);
  54                 struct page *page = prepared_pages[i];
  55                 fault_in_pages_readable(buf, count);
  56
  57                 /* Copy data from userspace to the current page */
  58                 kmap(page);
  59                 page_fault = __copy_from_user(page_address(page) + offset,
  60                                               buf, count);
  61                 /* Flush processor's dcache for this page */
  62                 flush_dcache_page(page);
  63                 kunmap(page);
  64                 buf += count;
  65                 write_bytes -= count;
  66
  67                 if (page_fault)
  68                         break;
  69         }
  70         return page_fault ? -EFAULT : 0;
  71 }
  72
  73 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  74 {
  75         size_t i;
  76         for (i = 0; i < num_pages; i++) {
  77                 if (!pages[i])
  78                         break;
  79                 unlock_page(pages[i]);
  80                 mark_page_accessed(pages[i]);
  81                 page_cache_release(pages[i]);
  82         }
  83 }
  84
  85 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
  86                                 struct btrfs_root *root, struct inode *inode,
  87                                 u64 offset, size_t size,
  88                                 struct page **pages, size_t page_offset,
  89                                 int num_pages)
  90 {
  91         struct btrfs_key key;
  92         struct btrfs_path *path;
  93         struct extent_buffer *leaf;
  94         char *kaddr;
  95         unsigned long ptr;
  96         struct btrfs_file_extent_item *ei;
  97         struct page *page;
  98         u32 datasize;
  99         int err = 0;
 100         int ret;
 101         int i;
 102         ssize_t cur_size;
 103
 104         path = btrfs_alloc_path();
 105         if (!path)
 106                 return -ENOMEM;
 107
 108         btrfs_set_trans_block_group(trans, inode);
 109
 110         key.objectid = inode->i_ino;
 111         key.offset = offset;
 112         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 113
 114         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 115         if (ret < 0) {
 116                 err = ret;
 117                 goto fail;
 118         }
 119         if (ret == 1) {
 120                 struct btrfs_key found_key;
 121
 122                 if (path->slots[0] == 0)
 123                         goto insert;
 124
 125                 path->slots[0]--;
 126                 leaf = path->nodes[0];
 127                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 128
 129                 if (found_key.objectid != inode->i_ino)
 130                         goto insert;
 131
 132                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 133                         goto insert;
 134                 ei = btrfs_item_ptr(leaf, path->slots[0],
 135                                     struct btrfs_file_extent_item);
 136
 137                 if (btrfs_file_extent_type(leaf, ei) !=
 138                     BTRFS_FILE_EXTENT_INLINE) {
 139                         goto insert;
 140                 }
 141                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 142                 ret = 0;
 143         }
 144         if (ret == 0) {
 145                 u32 found_size;
 146                 u64 found_end;
 147
 148                 leaf = path->nodes[0];
 149                 ei = btrfs_item_ptr(leaf, path->slots[0],
 150                                     struct btrfs_file_extent_item);
 151
 152                 if (btrfs_file_extent_type(leaf, ei) !=
 153                     BTRFS_FILE_EXTENT_INLINE) {
 154                         err = ret;
 155                         btrfs_print_leaf(root, leaf);
 156                         printk("found wasn't inline offset %Lu inode %lu\n",
 157                                offset, inode->i_ino);
 158                         goto fail;
 159                 }
 160                 found_size = btrfs_file_extent_inline_len(leaf,
 161                                           btrfs_item_nr(leaf, path->slots[0]));
 162                 found_end = key.offset + found_size;
 163
 164                 if (found_end < offset + size) {
 165                         btrfs_release_path(root, path);
 166                         ret = btrfs_search_slot(trans, root, &key, path,
 167                                                 offset + size - found_end, 1);
 168                         BUG_ON(ret != 0);
 169
 170                         ret = btrfs_extend_item(trans, root, path,
 171                                                 offset + size - found_end);
 172                         if (ret) {
 173                                 err = ret;
 174                                 goto fail;
 175                         }
 176                         leaf = path->nodes[0];
 177                         ei = btrfs_item_ptr(leaf, path->slots[0],
 178                                             struct btrfs_file_extent_item);
 179                         inode->i_blocks += (offset + size - found_end) >> 9;
 180                 }
 181                 if (found_end < offset) {
 182                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 183                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 184                 }
 185         } else {
 186 insert:
 187                 btrfs_release_path(root, path);
 188                 datasize = offset + size - key.offset;
 189                 inode->i_blocks += datasize >> 9;
 190                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 191                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 192                                               datasize);
 193                 if (ret) {
 194                         err = ret;
 195                         printk("got bad ret %d\n", ret);
 196                         goto fail;
 197                 }
 198                 leaf = path->nodes[0];
 199                 ei = btrfs_item_ptr(leaf, path->slots[0],
 200                                     struct btrfs_file_extent_item);
 201                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 202                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 203         }
 204         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 205
 206         cur_size = size;
 207         i = 0;
 208         while (size > 0) {
 209                 page = pages[i];
 210                 kaddr = kmap_atomic(page, KM_USER0);
 211                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 212                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 213                 kunmap_atomic(kaddr, KM_USER0);
 214                 page_offset = 0;
 215                 ptr += cur_size;
 216                 size -= cur_size;
 217                 if (i >= num_pages) {
 218                         printk("i %d num_pages %d\n", i, num_pages);
 219                 }
 220                 i++;
 221         }
 222         btrfs_mark_buffer_dirty(leaf);
 223 fail:
 224         btrfs_free_path(path);
 225         return err;
 226 }
 227
 228 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 229                                    struct btrfs_root *root,
 230                                    struct file *file,
 231                                    struct page **pages,
 232                                    size_t num_pages,
 233                                    loff_t pos,
 234                                    size_t write_bytes)
 235 {
 236         int err = 0;
 237         int i;
 238         struct inode *inode = fdentry(file)->d_inode;
 239         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 240         u64 hint_byte;
 241         u64 num_bytes;
 242         u64 start_pos;
 243         u64 end_of_last_block;
 244         u64 end_pos = pos + write_bytes;
 245         u64 inline_size;
 246         loff_t isize = i_size_read(inode);
 247
 248         start_pos = pos & ~((u64)root->sectorsize - 1);
 249         num_bytes = (write_bytes + pos - start_pos +
 250                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 251
 252         end_of_last_block = start_pos + num_bytes - 1;
 253
 254         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 255         mutex_lock(&root->fs_info->fs_mutex);
 256         trans = btrfs_start_transaction(root, 1);
 257         if (!trans) {
 258                 err = -ENOMEM;
 259                 goto out_unlock;
 260         }
 261         btrfs_set_trans_block_group(trans, inode);
 262         hint_byte = 0;
 263
 264         if ((end_of_last_block & 4095) == 0) {
 265                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 266         }
 267         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 268
 269         /* FIXME...EIEIO, ENOSPC and more */
 270         /* insert any holes we need to create */
 271         if (isize < end_pos) {
 272                 u64 last_pos_in_file;
 273                 u64 hole_size;
 274                 u64 mask = root->sectorsize - 1;
 275                 last_pos_in_file = (isize + mask) & ~mask;
 276                 hole_size = (end_pos - last_pos_in_file + mask) & ~mask;
 277                 if (last_pos_in_file < end_pos) {
 278                         err = btrfs_drop_extents(trans, root, inode,
 279                                                  last_pos_in_file,
 280                                                  last_pos_in_file + hole_size,
 281                                                  last_pos_in_file,
 282                                                  &hint_byte);
 283                         if (err)
 284                                 goto failed;
 285
 286                         err = btrfs_insert_file_extent(trans, root,
 287                                                        inode->i_ino,
 288                                                        last_pos_in_file,
 289                                                        0, 0, hole_size, 0);
 290                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 291                                         last_pos_in_file + hole_size -1);
 292                         btrfs_check_file(root, inode);
 293                 }
 294                 if (err)
 295                         goto failed;
 296         }
 297
 298         /*
 299          * either allocate an extent for the new bytes or setup the key
 300          * to show we are doing inline data in the extent
 301          */
 302         inline_size = end_pos;
 303         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 304             inline_size > root->fs_info->max_inline ||
 305             (inline_size & (root->sectorsize -1)) == 0 ||
 306             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 307                 u64 last_end;
 308                 u64 existing_delalloc = 0;
 309
 310                 for (i = 0; i < num_pages; i++) {
 311                         struct page *p = pages[i];
 312                         SetPageUptodate(p);
 313                         set_page_dirty(p);
 314                 }
 315                 last_end = (u64)(pages[num_pages -1]->index) <<
 316                                 PAGE_CACHE_SHIFT;
 317                 last_end += PAGE_CACHE_SIZE - 1;
 318                 if (start_pos < isize) {
 319                         u64 delalloc_start = start_pos;
 320                         existing_delalloc = count_range_bits(io_tree,
 321                                              &delalloc_start,
 322                                              end_of_last_block, (u64)-1,
 323                                              EXTENT_DELALLOC);
 324                 }
 325                 set_extent_delalloc(io_tree, start_pos, end_of_last_block,
 326                                  GFP_NOFS);
 327                 btrfs_add_ordered_inode(inode);
 328         } else {
 329                 u64 aligned_end;
 330                 /* step one, delete the existing extents in this range */
 331                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 332                         ~((u64)root->sectorsize - 1);
 333                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 334                                          aligned_end, aligned_end, &hint_byte);
 335                 if (err)
 336                         goto failed;
 337                 if (isize > inline_size)
 338                         inline_size = min_t(u64, isize, aligned_end);
 339                 inline_size -= start_pos;
 340                 err = insert_inline_extent(trans, root, inode, start_pos,
 341                                            inline_size, pages, 0, num_pages);
 342                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 343                 BUG_ON(err);
 344         }
 345         if (end_pos > isize) {
 346                 i_size_write(inode, end_pos);
 347                 btrfs_update_inode(trans, root, inode);
 348         }
 349 failed:
 350         err = btrfs_end_transaction(trans, root);
 351 out_unlock:
 352         mutex_unlock(&root->fs_info->fs_mutex);
 353         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 354         return err;
 355 }
 356
 357 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 358 {
 359         struct extent_map *em;
 360         struct extent_map *split = NULL;
 361         struct extent_map *split2 = NULL;
 362         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 363         u64 len = end - start + 1;
 364         int ret;
 365         int testend = 1;
 366
 367         if (end == (u64)-1) {
 368                 len = (u64)-1;
 369                 testend = 0;
 370         }
 371         while(1) {
 372                 if (!split)
 373                         split = alloc_extent_map(GFP_NOFS);
 374                 if (!split2)
 375                         split2 = alloc_extent_map(GFP_NOFS);
 376
 377                 spin_lock(&em_tree->lock);
 378                 em = lookup_extent_mapping(em_tree, start, len);
 379                 if (!em) {
 380                         spin_unlock(&em_tree->lock);
 381                         break;
 382                 }
 383                 remove_extent_mapping(em_tree, em);
 384
 385                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 386                     em->start < start) {
 387                         split->start = em->start;
 388                         split->len = start - em->start;
 389                         split->block_start = em->block_start;
 390                         split->bdev = em->bdev;
 391                         split->flags = em->flags;
 392                         ret = add_extent_mapping(em_tree, split);
 393                         BUG_ON(ret);
 394                         free_extent_map(split);
 395                         split = split2;
 396                         split2 = NULL;
 397                 }
 398                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 399                     testend && em->start + em->len > start + len) {
 400                         u64 diff = start + len - em->start;
 401
 402                         split->start = start + len;
 403                         split->len = em->start + em->len - (start + len);
 404                         split->bdev = em->bdev;
 405                         split->flags = em->flags;
 406
 407                         split->block_start = em->block_start + diff;
 408
 409                         ret = add_extent_mapping(em_tree, split);
 410                         BUG_ON(ret);
 411                         free_extent_map(split);
 412                         split = NULL;
 413                 }
 414                 spin_unlock(&em_tree->lock);
 415
 416                 /* once for us */
 417                 free_extent_map(em);
 418                 /* once for the tree*/
 419                 free_extent_map(em);
 420         }
 421         if (split)
 422                 free_extent_map(split);
 423         if (split2)
 424                 free_extent_map(split2);
 425         return 0;
 426 }
 427
 428 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 429 {
 430         return 0;
 431 #if 0
 432         struct btrfs_path *path;
 433         struct btrfs_key found_key;
 434         struct extent_buffer *leaf;
 435         struct btrfs_file_extent_item *extent;
 436         u64 last_offset = 0;
 437         int nritems;
 438         int slot;
 439         int found_type;
 440         int ret;
 441         int err = 0;
 442         u64 extent_end = 0;
 443
 444         path = btrfs_alloc_path();
 445         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 446                                        last_offset, 0);
 447         while(1) {
 448                 nritems = btrfs_header_nritems(path->nodes[0]);
 449                 if (path->slots[0] >= nritems) {
 450                         ret = btrfs_next_leaf(root, path);
 451                         if (ret)
 452                                 goto out;
 453                         nritems = btrfs_header_nritems(path->nodes[0]);
 454                 }
 455                 slot = path->slots[0];
 456                 leaf = path->nodes[0];
 457                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 458                 if (found_key.objectid != inode->i_ino)
 459                         break;
 460                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 461                         goto out;
 462
 463                 if (found_key.offset < last_offset) {
 464                         WARN_ON(1);
 465                         btrfs_print_leaf(root, leaf);
 466                         printk("inode %lu found offset %Lu expected %Lu\n",
 467                                inode->i_ino, found_key.offset, last_offset);
 468                         err = 1;
 469                         goto out;
 470                 }
 471                 extent = btrfs_item_ptr(leaf, slot,
 472                                         struct btrfs_file_extent_item);
 473                 found_type = btrfs_file_extent_type(leaf, extent);
 474                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 475                         extent_end = found_key.offset +
 476                              btrfs_file_extent_num_bytes(leaf, extent);
 477                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 478                         struct btrfs_item *item;
 479                         item = btrfs_item_nr(leaf, slot);
 480                         extent_end = found_key.offset +
 481                              btrfs_file_extent_inline_len(leaf, item);
 482                         extent_end = (extent_end + root->sectorsize - 1) &
 483                                 ~((u64)root->sectorsize -1 );
 484                 }
 485                 last_offset = extent_end;
 486                 path->slots[0]++;
 487         }
 488         if (0 && last_offset < inode->i_size) {
 489                 WARN_ON(1);
 490                 btrfs_print_leaf(root, leaf);
 491                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 492                        last_offset, inode->i_size);
 493                 err = 1;
 494
 495         }
 496 out:
 497         btrfs_free_path(path);
 498         return err;
 499 #endif
 500 }
 501
 502 /*
 503  * this is very complex, but the basic idea is to drop all extents
 504  * in the range start - end.  hint_block is filled in with a block number
 505  * that would be a good hint to the block allocator for this file.
 506  *
 507  * If an extent intersects the range but is not entirely inside the range
 508  * it is either truncated or split.  Anything entirely inside the range
 509  * is deleted from the tree.
 510  */
 511 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 512                        struct btrfs_root *root, struct inode *inode,
 513                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 514 {
 515         u64 extent_end = 0;
 516         u64 search_start = start;
 517         struct extent_buffer *leaf;
 518         struct btrfs_file_extent_item *extent;
 519         struct btrfs_path *path;
 520         struct btrfs_key key;
 521         struct btrfs_file_extent_item old;
 522         int keep;
 523         int slot;
 524         int bookend;
 525         int found_type;
 526         int found_extent;
 527         int found_inline;
 528         int recow;
 529         int ret;
 530
 531         btrfs_drop_extent_cache(inode, start, end - 1);
 532
 533         path = btrfs_alloc_path();
 534         if (!path)
 535                 return -ENOMEM;
 536         while(1) {
 537                 recow = 0;
 538                 btrfs_release_path(root, path);
 539                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 540                                                search_start, -1);
 541                 if (ret < 0)
 542                         goto out;
 543                 if (ret > 0) {
 544                         if (path->slots[0] == 0) {
 545                                 ret = 0;
 546                                 goto out;
 547                         }
 548                         path->slots[0]--;
 549                 }
 550 next_slot:
 551                 keep = 0;
 552                 bookend = 0;
 553                 found_extent = 0;
 554                 found_inline = 0;
 555                 extent = NULL;
 556                 leaf = path->nodes[0];
 557                 slot = path->slots[0];
 558                 ret = 0;
 559                 btrfs_item_key_to_cpu(leaf, &key, slot);
 560                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 561                     key.offset >= end) {
 562                         goto out;
 563                 }
 564                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 565                     key.objectid != inode->i_ino) {
 566                         goto out;
 567                 }
 568                 if (recow) {
 569                         search_start = key.offset;
 570                         continue;
 571                 }
 572                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 573                         extent = btrfs_item_ptr(leaf, slot,
 574                                                 struct btrfs_file_extent_item);
 575                         found_type = btrfs_file_extent_type(leaf, extent);
 576                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 577                                 extent_end =
 578                                      btrfs_file_extent_disk_bytenr(leaf,
 579                                                                    extent);
 580                                 if (extent_end)
 581                                         *hint_byte = extent_end;
 582
 583                                 extent_end = key.offset +
 584                                      btrfs_file_extent_num_bytes(leaf, extent);
 585                                 found_extent = 1;
 586                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 587                                 struct btrfs_item *item;
 588                                 item = btrfs_item_nr(leaf, slot);
 589                                 found_inline = 1;
 590                                 extent_end = key.offset +
 591                                      btrfs_file_extent_inline_len(leaf, item);
 592                         }
 593                 } else {
 594                         extent_end = search_start;
 595                 }
 596
 597                 /* we found nothing we can drop */
 598                 if ((!found_extent && !found_inline) ||
 599                     search_start >= extent_end) {
 600                         int nextret;
 601                         u32 nritems;
 602                         nritems = btrfs_header_nritems(leaf);
 603                         if (slot >= nritems - 1) {
 604                                 nextret = btrfs_next_leaf(root, path);
 605                                 if (nextret)
 606                                         goto out;
 607                                 recow = 1;
 608                         } else {
 609                                 path->slots[0]++;
 610                         }
 611                         goto next_slot;
 612                 }
 613
 614                 if (found_inline) {
 615                         u64 mask = root->sectorsize - 1;
 616                         search_start = (extent_end + mask) & ~mask;
 617                 } else
 618                         search_start = extent_end;
 619                 if (end <= extent_end && start >= key.offset && found_inline) {
 620                         *hint_byte = EXTENT_MAP_INLINE;
 621                         continue;
 622                 }
 623                 if (end < extent_end && end >= key.offset) {
 624                         if (found_extent) {
 625                                 u64 disk_bytenr =
 626                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 627                                 u64 disk_num_bytes =
 628                                     btrfs_file_extent_disk_num_bytes(leaf,
 629                                                                       extent);
 630                                 read_extent_buffer(leaf, &old,
 631                                                    (unsigned long)extent,
 632                                                    sizeof(old));
 633                                 if (disk_bytenr != 0) {
 634                                         ret = btrfs_inc_extent_ref(trans, root,
 635                                                  disk_bytenr, disk_num_bytes,
 636                                                  root->root_key.objectid,
 637                                                  trans->transid,
 638                                                  key.objectid, end);
 639                                         BUG_ON(ret);
 640                                 }
 641                         }
 642                         bookend = 1;
 643                         if (found_inline && start <= key.offset)
 644                                 keep = 1;
 645                 }
 646                 /* truncate existing extent */
 647                 if (start > key.offset) {
 648                         u64 new_num;
 649                         u64 old_num;
 650                         keep = 1;
 651                         WARN_ON(start & (root->sectorsize - 1));
 652                         if (found_extent) {
 653                                 new_num = start - key.offset;
 654                                 old_num = btrfs_file_extent_num_bytes(leaf,
 655                                                                       extent);
 656                                 *hint_byte =
 657                                         btrfs_file_extent_disk_bytenr(leaf,
 658                                                                       extent);
 659                                 if (btrfs_file_extent_disk_bytenr(leaf,
 660                                                                   extent)) {
 661                                         dec_i_blocks(inode, old_num - new_num);
 662                                 }
 663                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 664                                                                 new_num);
 665                                 btrfs_mark_buffer_dirty(leaf);
 666                         } else if (key.offset < inline_limit &&
 667                                    (end > extent_end) &&
 668                                    (inline_limit < extent_end)) {
 669                                 u32 new_size;
 670                                 new_size = btrfs_file_extent_calc_inline_size(
 671                                                    inline_limit - key.offset);
 672                                 dec_i_blocks(inode, (extent_end - key.offset) -
 673                                         (inline_limit - key.offset));
 674                                 btrfs_truncate_item(trans, root, path,
 675                                                     new_size, 1);
 676                         }
 677                 }
 678                 /* delete the entire extent */
 679                 if (!keep) {
 680                         u64 disk_bytenr = 0;
 681                         u64 disk_num_bytes = 0;
 682                         u64 extent_num_bytes = 0;
 683                         u64 root_gen;
 684                         u64 root_owner;
 685
 686                         root_gen = btrfs_header_generation(leaf);
 687                         root_owner = btrfs_header_owner(leaf);
 688                         if (found_extent) {
 689                                 disk_bytenr =
 690                                       btrfs_file_extent_disk_bytenr(leaf,
 691                                                                      extent);
 692                                 disk_num_bytes =
 693                                       btrfs_file_extent_disk_num_bytes(leaf,
 694                                                                        extent);
 695                                 extent_num_bytes =
 696                                       btrfs_file_extent_num_bytes(leaf, extent);
 697                                 *hint_byte =
 698                                         btrfs_file_extent_disk_bytenr(leaf,
 699                                                                       extent);
 700                         }
 701                         ret = btrfs_del_item(trans, root, path);
 702                         /* TODO update progress marker and return */
 703                         BUG_ON(ret);
 704                         btrfs_release_path(root, path);
 705                         extent = NULL;
 706                         if (found_extent && disk_bytenr != 0) {
 707                                 dec_i_blocks(inode, extent_num_bytes);
 708                                 ret = btrfs_free_extent(trans, root,
 709                                                 disk_bytenr,
 710                                                 disk_num_bytes,
 711                                                 root_owner,
 712                                                 root_gen, inode->i_ino,
 713                                                 key.offset, 0);
 714                         }
 715
 716                         BUG_ON(ret);
 717                         if (!bookend && search_start >= end) {
 718                                 ret = 0;
 719                                 goto out;
 720                         }
 721                         if (!bookend)
 722                                 continue;
 723                 }
 724                 if (bookend && found_inline && start <= key.offset) {
 725                         u32 new_size;
 726                         new_size = btrfs_file_extent_calc_inline_size(
 727                                                    extent_end - end);
 728                         dec_i_blocks(inode, (extent_end - key.offset) -
 729                                         (extent_end - end));
 730                         btrfs_truncate_item(trans, root, path, new_size, 0);
 731                 }
 732                 /* create bookend, splitting the extent in two */
 733                 if (bookend && found_extent) {
 734                         struct btrfs_key ins;
 735                         ins.objectid = inode->i_ino;
 736                         ins.offset = end;
 737                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 738                         btrfs_release_path(root, path);
 739                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 740                                                       sizeof(*extent));
 741
 742                         leaf = path->nodes[0];
 743                         if (ret) {
 744                                 btrfs_print_leaf(root, leaf);
 745                                 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret , ins.objectid, ins.type, ins.offset, start, end, key.offset, extent_end, keep);
 746                         }
 747                         BUG_ON(ret);
 748                         extent = btrfs_item_ptr(leaf, path->slots[0],
 749                                                 struct btrfs_file_extent_item);
 750                         write_extent_buffer(leaf, &old,
 751                                             (unsigned long)extent, sizeof(old));
 752
 753                         btrfs_set_file_extent_offset(leaf, extent,
 754                                     le64_to_cpu(old.offset) + end - key.offset);
 755                         WARN_ON(le64_to_cpu(old.num_bytes) <
 756                                 (extent_end - end));
 757                         btrfs_set_file_extent_num_bytes(leaf, extent,
 758                                                         extent_end - end);
 759                         btrfs_set_file_extent_type(leaf, extent,
 760                                                    BTRFS_FILE_EXTENT_REG);
 761
 762                         btrfs_mark_buffer_dirty(path->nodes[0]);
 763                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 764                                 inode->i_blocks +=
 765                                       btrfs_file_extent_num_bytes(leaf,
 766                                                                   extent) >> 9;
 767                         }
 768                         ret = 0;
 769                         goto out;
 770                 }
 771         }
 772 out:
 773         btrfs_free_path(path);
 774         btrfs_check_file(root, inode);
 775         return ret;
 776 }
 777
 778 /*
 779  * this gets pages into the page cache and locks them down
 780  */
 781 static int prepare_pages(struct btrfs_root *root, struct file *file,
 782                          struct page **pages, size_t num_pages,
 783                          loff_t pos, unsigned long first_index,
 784                          unsigned long last_index, size_t write_bytes)
 785 {
 786         int i;
 787         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 788         struct inode *inode = fdentry(file)->d_inode;
 789         int err = 0;
 790         u64 start_pos;
 791
 792         start_pos = pos & ~((u64)root->sectorsize - 1);
 793
 794         memset(pages, 0, num_pages * sizeof(struct page *));
 795
 796         for (i = 0; i < num_pages; i++) {
 797                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 798                 if (!pages[i]) {
 799                         err = -ENOMEM;
 800                         BUG_ON(1);
 801                 }
 802 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 803                 ClearPageDirty(pages[i]);
 804 #else
 805                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 806 #endif
 807                 wait_on_page_writeback(pages[i]);
 808                 set_page_extent_mapped(pages[i]);
 809                 WARN_ON(!PageLocked(pages[i]));
 810         }
 811         if (start_pos < inode->i_size) {
 812                 u64 last_pos;
 813                 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 814                 lock_extent(&BTRFS_I(inode)->io_tree,
 815                             start_pos, last_pos - 1, GFP_NOFS);
 816                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 817                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 818                                   GFP_NOFS);
 819                 unlock_extent(&BTRFS_I(inode)->io_tree,
 820                               start_pos, last_pos - 1, GFP_NOFS);
 821         }
 822         return 0;
 823 }
 824
 825 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 826                                 size_t count, loff_t *ppos)
 827 {
 828         loff_t pos;
 829         loff_t start_pos;
 830         ssize_t num_written = 0;
 831         ssize_t err = 0;
 832         int ret = 0;
 833         struct inode *inode = fdentry(file)->d_inode;
 834         struct btrfs_root *root = BTRFS_I(inode)->root;
 835         struct page **pages = NULL;
 836         int nrptrs;
 837         struct page *pinned[2];
 838         unsigned long first_index;
 839         unsigned long last_index;
 840
 841         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 842                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 843         pinned[0] = NULL;
 844         pinned[1] = NULL;
 845
 846         pos = *ppos;
 847         start_pos = pos;
 848
 849         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 850         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 851         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 852         if (err)
 853                 goto out_nolock;
 854         if (count == 0)
 855                 goto out_nolock;
 856 #ifdef REMOVE_SUID_PATH
 857         err = remove_suid(&file->f_path);
 858 #else
 859         err = remove_suid(fdentry(file));
 860 #endif
 861         if (err)
 862                 goto out_nolock;
 863         file_update_time(file);
 864
 865         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 866
 867         mutex_lock(&inode->i_mutex);
 868         first_index = pos >> PAGE_CACHE_SHIFT;
 869         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 870
 871         /*
 872          * if this is a nodatasum mount, force summing off for the inode
 873          * all the time.  That way a later mount with summing on won't
 874          * get confused
 875          */
 876         if (btrfs_test_opt(root, NODATASUM))
 877                 btrfs_set_flag(inode, NODATASUM);
 878
 879         /*
 880          * there are lots of better ways to do this, but this code
 881          * makes sure the first and last page in the file range are
 882          * up to date and ready for cow
 883          */
 884         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 885                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 886                 if (!PageUptodate(pinned[0])) {
 887                         ret = btrfs_readpage(NULL, pinned[0]);
 888                         BUG_ON(ret);
 889                         wait_on_page_locked(pinned[0]);
 890                 } else {
 891                         unlock_page(pinned[0]);
 892                 }
 893         }
 894         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 895                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 896                 if (!PageUptodate(pinned[1])) {
 897                         ret = btrfs_readpage(NULL, pinned[1]);
 898                         BUG_ON(ret);
 899                         wait_on_page_locked(pinned[1]);
 900                 } else {
 901                         unlock_page(pinned[1]);
 902                 }
 903         }
 904
 905         while(count > 0) {
 906                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 907                 size_t write_bytes = min(count, nrptrs *
 908                                         (size_t)PAGE_CACHE_SIZE -
 909                                          offset);
 910                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 911                                         PAGE_CACHE_SHIFT;
 912
 913                 WARN_ON(num_pages > nrptrs);
 914                 memset(pages, 0, sizeof(pages));
 915
 916                 mutex_lock(&root->fs_info->fs_mutex);
 917                 ret = btrfs_check_free_space(root, write_bytes, 0);
 918                 mutex_unlock(&root->fs_info->fs_mutex);
 919                 if (ret)
 920                         goto out;
 921
 922                 ret = prepare_pages(root, file, pages, num_pages,
 923                                     pos, first_index, last_index,
 924                                     write_bytes);
 925                 if (ret)
 926                         goto out;
 927
 928                 ret = btrfs_copy_from_user(pos, num_pages,
 929                                            write_bytes, pages, buf);
 930                 if (ret) {
 931                         btrfs_drop_pages(pages, num_pages);
 932                         goto out;
 933                 }
 934
 935                 ret = dirty_and_release_pages(NULL, root, file, pages,
 936                                               num_pages, pos, write_bytes);
 937                 btrfs_drop_pages(pages, num_pages);
 938                 if (ret)
 939                         goto out;
 940
 941                 buf += write_bytes;
 942                 count -= write_bytes;
 943                 pos += write_bytes;
 944                 num_written += write_bytes;
 945
 946                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 947                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 948                         btrfs_btree_balance_dirty(root, 1);
 949                 btrfs_throttle(root);
 950                 cond_resched();
 951         }
 952 out:
 953         mutex_unlock(&inode->i_mutex);
 954
 955 out_nolock:
 956         kfree(pages);
 957         if (pinned[0])
 958                 page_cache_release(pinned[0]);
 959         if (pinned[1])
 960                 page_cache_release(pinned[1]);
 961         *ppos = pos;
 962
 963         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 964                 err = sync_page_range(inode, inode->i_mapping,
 965                                       start_pos, num_written);
 966                 if (err < 0)
 967                         num_written = err;
 968         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
 969 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
 970                 do_sync_file_range(file, start_pos,
 971                                       start_pos + num_written - 1,
 972                                       SYNC_FILE_RANGE_WRITE |
 973                                       SYNC_FILE_RANGE_WAIT_AFTER);
 974 #else
 975                 do_sync_mapping_range(inode->i_mapping, start_pos,
 976                                       start_pos + num_written - 1,
 977                                       SYNC_FILE_RANGE_WRITE |
 978                                       SYNC_FILE_RANGE_WAIT_AFTER);
 979 #endif
 980                 invalidate_mapping_pages(inode->i_mapping,
 981                       start_pos >> PAGE_CACHE_SHIFT,
 982                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
 983         }
 984         current->backing_dev_info = NULL;
 985         btrfs_ordered_throttle(root, inode);
 986         return num_written ? num_written : err;
 987 }
 988
 989 static int btrfs_sync_file(struct file *file,
 990                            struct dentry *dentry, int datasync)
 991 {
 992         struct inode *inode = dentry->d_inode;
 993         struct btrfs_root *root = BTRFS_I(inode)->root;
 994         int ret = 0;
 995         struct btrfs_trans_handle *trans;
 996
 997         /*
 998          * check the transaction that last modified this inode
 999          * and see if its already been committed
1000          */
1001         mutex_lock(&root->fs_info->fs_mutex);
1002         if (!BTRFS_I(inode)->last_trans)
1003                 goto out;
1004         mutex_lock(&root->fs_info->trans_mutex);
1005         if (BTRFS_I(inode)->last_trans <=
1006             root->fs_info->last_trans_committed) {
1007                 BTRFS_I(inode)->last_trans = 0;
1008                 mutex_unlock(&root->fs_info->trans_mutex);
1009                 goto out;
1010         }
1011         mutex_unlock(&root->fs_info->trans_mutex);
1012
1013         /*
1014          * ok we haven't committed the transaction yet, lets do a commit
1015          */
1016         trans = btrfs_start_transaction(root, 1);
1017         if (!trans) {
1018                 ret = -ENOMEM;
1019                 goto out;
1020         }
1021         ret = btrfs_commit_transaction(trans, root);
1022 out:
1023         mutex_unlock(&root->fs_info->fs_mutex);
1024         return ret > 0 ? EIO : ret;
1025 }
1026
1027 static struct vm_operations_struct btrfs_file_vm_ops = {
1028 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1029         .nopage         = filemap_nopage,
1030         .populate       = filemap_populate,
1031 #else
1032         .fault          = filemap_fault,
1033 #endif
1034         .page_mkwrite   = btrfs_page_mkwrite,
1035 };
1036
1037 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1038 {
1039         vma->vm_ops = &btrfs_file_vm_ops;
1040         file_accessed(filp);
1041         return 0;
1042 }
1043
1044 struct file_operations btrfs_file_operations = {
1045         .llseek         = generic_file_llseek,
1046         .read           = do_sync_read,
1047         .aio_read       = generic_file_aio_read,
1048         .splice_read    = generic_file_splice_read,
1049 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1050         .sendfile       = generic_file_sendfile,
1051 #endif
1052         .write          = btrfs_file_write,
1053         .mmap           = btrfs_file_mmap,
1054         .open           = generic_file_open,
1055         .fsync          = btrfs_sync_file,
1056         .unlocked_ioctl = btrfs_ioctl,
1057 #ifdef CONFIG_COMPAT
1058         .compat_ioctl   = btrfs_ioctl,
1059 #endif
1060 };
1061