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