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

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/mpage.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/version.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "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
 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                 set_extent_delalloc(io_tree, start_pos, end_of_last_block,
 318                                  GFP_NOFS);
 319                 btrfs_add_ordered_inode(inode);
 320         } else {
 321                 u64 aligned_end;
 322                 /* step one, delete the existing extents in this range */
 323                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 324                         ~((u64)root->sectorsize - 1);
 325                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 326                                          aligned_end, aligned_end, &hint_byte);
 327                 if (err)
 328                         goto failed;
 329                 if (isize > inline_size)
 330                         inline_size = min_t(u64, isize, aligned_end);
 331                 inline_size -= start_pos;
 332                 err = insert_inline_extent(trans, root, inode, start_pos,
 333                                            inline_size, pages, 0, num_pages);
 334                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 335                 BUG_ON(err);
 336         }
 337         if (end_pos > isize) {
 338                 i_size_write(inode, end_pos);
 339                 btrfs_update_inode(trans, root, inode);
 340         }
 341 failed:
 342         err = btrfs_end_transaction(trans, root);
 343 out_unlock:
 344         mutex_unlock(&root->fs_info->fs_mutex);
 345         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 346         return err;
 347 }
 348
 349 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 350 {
 351         struct extent_map *em;
 352         struct extent_map *split = NULL;
 353         struct extent_map *split2 = NULL;
 354         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 355         u64 len = end - start + 1;
 356         int ret;
 357         int testend = 1;
 358
 359         if (end == (u64)-1) {
 360                 len = (u64)-1;
 361                 testend = 0;
 362         }
 363         while(1) {
 364                 if (!split)
 365                         split = alloc_extent_map(GFP_NOFS);
 366                 if (!split2)
 367                         split2 = alloc_extent_map(GFP_NOFS);
 368
 369                 spin_lock(&em_tree->lock);
 370                 em = lookup_extent_mapping(em_tree, start, len);
 371                 if (!em) {
 372                         spin_unlock(&em_tree->lock);
 373                         break;
 374                 }
 375                 remove_extent_mapping(em_tree, em);
 376
 377                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 378                     em->start < start) {
 379                         split->start = em->start;
 380                         split->len = start - em->start;
 381                         split->block_start = em->block_start;
 382                         split->bdev = em->bdev;
 383                         split->flags = em->flags;
 384                         ret = add_extent_mapping(em_tree, split);
 385                         BUG_ON(ret);
 386                         free_extent_map(split);
 387                         split = split2;
 388                         split2 = NULL;
 389                 }
 390                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 391                     testend && em->start + em->len > start + len) {
 392                         u64 diff = start + len - em->start;
 393
 394                         split->start = start + len;
 395                         split->len = em->start + em->len - (start + len);
 396                         split->bdev = em->bdev;
 397                         split->flags = em->flags;
 398
 399                         split->block_start = em->block_start + diff;
 400
 401                         ret = add_extent_mapping(em_tree, split);
 402                         BUG_ON(ret);
 403                         free_extent_map(split);
 404                         split = NULL;
 405                 }
 406                 spin_unlock(&em_tree->lock);
 407
 408                 /* once for us */
 409                 free_extent_map(em);
 410                 /* once for the tree*/
 411                 free_extent_map(em);
 412         }
 413         if (split)
 414                 free_extent_map(split);
 415         if (split2)
 416                 free_extent_map(split2);
 417         return 0;
 418 }
 419
 420 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 421 {
 422         return 0;
 423 #if 0
 424         struct btrfs_path *path;
 425         struct btrfs_key found_key;
 426         struct extent_buffer *leaf;
 427         struct btrfs_file_extent_item *extent;
 428         u64 last_offset = 0;
 429         int nritems;
 430         int slot;
 431         int found_type;
 432         int ret;
 433         int err = 0;
 434         u64 extent_end = 0;
 435
 436         path = btrfs_alloc_path();
 437         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 438                                        last_offset, 0);
 439         while(1) {
 440                 nritems = btrfs_header_nritems(path->nodes[0]);
 441                 if (path->slots[0] >= nritems) {
 442                         ret = btrfs_next_leaf(root, path);
 443                         if (ret)
 444                                 goto out;
 445                         nritems = btrfs_header_nritems(path->nodes[0]);
 446                 }
 447                 slot = path->slots[0];
 448                 leaf = path->nodes[0];
 449                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 450                 if (found_key.objectid != inode->i_ino)
 451                         break;
 452                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 453                         goto out;
 454
 455                 if (found_key.offset < last_offset) {
 456                         WARN_ON(1);
 457                         btrfs_print_leaf(root, leaf);
 458                         printk("inode %lu found offset %Lu expected %Lu\n",
 459                                inode->i_ino, found_key.offset, last_offset);
 460                         err = 1;
 461                         goto out;
 462                 }
 463                 extent = btrfs_item_ptr(leaf, slot,
 464                                         struct btrfs_file_extent_item);
 465                 found_type = btrfs_file_extent_type(leaf, extent);
 466                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 467                         extent_end = found_key.offset +
 468                              btrfs_file_extent_num_bytes(leaf, extent);
 469                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 470                         struct btrfs_item *item;
 471                         item = btrfs_item_nr(leaf, slot);
 472                         extent_end = found_key.offset +
 473                              btrfs_file_extent_inline_len(leaf, item);
 474                         extent_end = (extent_end + root->sectorsize - 1) &
 475                                 ~((u64)root->sectorsize -1 );
 476                 }
 477                 last_offset = extent_end;
 478                 path->slots[0]++;
 479         }
 480         if (0 && last_offset < inode->i_size) {
 481                 WARN_ON(1);
 482                 btrfs_print_leaf(root, leaf);
 483                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 484                        last_offset, inode->i_size);
 485                 err = 1;
 486
 487         }
 488 out:
 489         btrfs_free_path(path);
 490         return err;
 491 #endif
 492 }
 493
 494 /*
 495  * this is very complex, but the basic idea is to drop all extents
 496  * in the range start - end.  hint_block is filled in with a block number
 497  * that would be a good hint to the block allocator for this file.
 498  *
 499  * If an extent intersects the range but is not entirely inside the range
 500  * it is either truncated or split.  Anything entirely inside the range
 501  * is deleted from the tree.
 502  */
 503 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 504                        struct btrfs_root *root, struct inode *inode,
 505                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 506 {
 507         u64 extent_end = 0;
 508         u64 search_start = start;
 509         struct extent_buffer *leaf;
 510         struct btrfs_file_extent_item *extent;
 511         struct btrfs_path *path;
 512         struct btrfs_key key;
 513         struct btrfs_file_extent_item old;
 514         int keep;
 515         int slot;
 516         int bookend;
 517         int found_type;
 518         int found_extent;
 519         int found_inline;
 520         int recow;
 521         int ret;
 522
 523         btrfs_drop_extent_cache(inode, start, end - 1);
 524
 525         path = btrfs_alloc_path();
 526         if (!path)
 527                 return -ENOMEM;
 528         while(1) {
 529                 recow = 0;
 530                 btrfs_release_path(root, path);
 531                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 532                                                search_start, -1);
 533                 if (ret < 0)
 534                         goto out;
 535                 if (ret > 0) {
 536                         if (path->slots[0] == 0) {
 537                                 ret = 0;
 538                                 goto out;
 539                         }
 540                         path->slots[0]--;
 541                 }
 542 next_slot:
 543                 keep = 0;
 544                 bookend = 0;
 545                 found_extent = 0;
 546                 found_inline = 0;
 547                 extent = NULL;
 548                 leaf = path->nodes[0];
 549                 slot = path->slots[0];
 550                 ret = 0;
 551                 btrfs_item_key_to_cpu(leaf, &key, slot);
 552                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 553                     key.offset >= end) {
 554                         goto out;
 555                 }
 556                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 557                     key.objectid != inode->i_ino) {
 558                         goto out;
 559                 }
 560                 if (recow) {
 561                         search_start = key.offset;
 562                         continue;
 563                 }
 564                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 565                         extent = btrfs_item_ptr(leaf, slot,
 566                                                 struct btrfs_file_extent_item);
 567                         found_type = btrfs_file_extent_type(leaf, extent);
 568                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 569                                 extent_end =
 570                                      btrfs_file_extent_disk_bytenr(leaf,
 571                                                                    extent);
 572                                 if (extent_end)
 573                                         *hint_byte = extent_end;
 574
 575                                 extent_end = key.offset +
 576                                      btrfs_file_extent_num_bytes(leaf, extent);
 577                                 found_extent = 1;
 578                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 579                                 struct btrfs_item *item;
 580                                 item = btrfs_item_nr(leaf, slot);
 581                                 found_inline = 1;
 582                                 extent_end = key.offset +
 583                                      btrfs_file_extent_inline_len(leaf, item);
 584                         }
 585                 } else {
 586                         extent_end = search_start;
 587                 }
 588
 589                 /* we found nothing we can drop */
 590                 if ((!found_extent && !found_inline) ||
 591                     search_start >= extent_end) {
 592                         int nextret;
 593                         u32 nritems;
 594                         nritems = btrfs_header_nritems(leaf);
 595                         if (slot >= nritems - 1) {
 596                                 nextret = btrfs_next_leaf(root, path);
 597                                 if (nextret)
 598                                         goto out;
 599                                 recow = 1;
 600                         } else {
 601                                 path->slots[0]++;
 602                         }
 603                         goto next_slot;
 604                 }
 605
 606                 if (found_inline) {
 607                         u64 mask = root->sectorsize - 1;
 608                         search_start = (extent_end + mask) & ~mask;
 609                 } else
 610                         search_start = extent_end;
 611                 if (end <= extent_end && start >= key.offset && found_inline) {
 612                         *hint_byte = EXTENT_MAP_INLINE;
 613                         continue;
 614                 }
 615                 if (end < extent_end && end >= key.offset) {
 616                         if (found_extent) {
 617                                 u64 disk_bytenr =
 618                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 619                                 u64 disk_num_bytes =
 620                                     btrfs_file_extent_disk_num_bytes(leaf,
 621                                                                       extent);
 622                                 read_extent_buffer(leaf, &old,
 623                                                    (unsigned long)extent,
 624                                                    sizeof(old));
 625                                 if (disk_bytenr != 0) {
 626                                         ret = btrfs_inc_extent_ref(trans, root,
 627                                                  disk_bytenr, disk_num_bytes,
 628                                                  root->root_key.objectid,
 629                                                  trans->transid,
 630                                                  key.objectid, end);
 631                                         BUG_ON(ret);
 632                                 }
 633                         }
 634                         bookend = 1;
 635                         if (found_inline && start <= key.offset)
 636                                 keep = 1;
 637                 }
 638                 /* truncate existing extent */
 639                 if (start > key.offset) {
 640                         u64 new_num;
 641                         u64 old_num;
 642                         keep = 1;
 643                         WARN_ON(start & (root->sectorsize - 1));
 644                         if (found_extent) {
 645                                 new_num = start - key.offset;
 646                                 old_num = btrfs_file_extent_num_bytes(leaf,
 647                                                                       extent);
 648                                 *hint_byte =
 649                                         btrfs_file_extent_disk_bytenr(leaf,
 650                                                                       extent);
 651                                 if (btrfs_file_extent_disk_bytenr(leaf,
 652                                                                   extent)) {
 653                                         dec_i_blocks(inode, old_num - new_num);
 654                                 }
 655                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 656                                                                 new_num);
 657                                 btrfs_mark_buffer_dirty(leaf);
 658                         } else if (key.offset < inline_limit &&
 659                                    (end > extent_end) &&
 660                                    (inline_limit < extent_end)) {
 661                                 u32 new_size;
 662                                 new_size = btrfs_file_extent_calc_inline_size(
 663                                                    inline_limit - key.offset);
 664                                 dec_i_blocks(inode, (extent_end - key.offset) -
 665                                         (inline_limit - key.offset));
 666                                 btrfs_truncate_item(trans, root, path,
 667                                                     new_size, 1);
 668                         }
 669                 }
 670                 /* delete the entire extent */
 671                 if (!keep) {
 672                         u64 disk_bytenr = 0;
 673                         u64 disk_num_bytes = 0;
 674                         u64 extent_num_bytes = 0;
 675                         u64 root_gen;
 676                         u64 root_owner;
 677
 678                         root_gen = btrfs_header_generation(leaf);
 679                         root_owner = btrfs_header_owner(leaf);
 680                         if (found_extent) {
 681                                 disk_bytenr =
 682                                       btrfs_file_extent_disk_bytenr(leaf,
 683                                                                      extent);
 684                                 disk_num_bytes =
 685                                       btrfs_file_extent_disk_num_bytes(leaf,
 686                                                                        extent);
 687                                 extent_num_bytes =
 688                                       btrfs_file_extent_num_bytes(leaf, extent);
 689                                 *hint_byte =
 690                                         btrfs_file_extent_disk_bytenr(leaf,
 691                                                                       extent);
 692                         }
 693                         ret = btrfs_del_item(trans, root, path);
 694                         /* TODO update progress marker and return */
 695                         BUG_ON(ret);
 696                         btrfs_release_path(root, path);
 697                         extent = NULL;
 698                         if (found_extent && disk_bytenr != 0) {
 699                                 dec_i_blocks(inode, extent_num_bytes);
 700                                 ret = btrfs_free_extent(trans, root,
 701                                                 disk_bytenr,
 702                                                 disk_num_bytes,
 703                                                 root_owner,
 704                                                 root_gen, inode->i_ino,
 705                                                 key.offset, 0);
 706                         }
 707
 708                         BUG_ON(ret);
 709                         if (!bookend && search_start >= end) {
 710                                 ret = 0;
 711                                 goto out;
 712                         }
 713                         if (!bookend)
 714                                 continue;
 715                 }
 716                 if (bookend && found_inline && start <= key.offset) {
 717                         u32 new_size;
 718                         new_size = btrfs_file_extent_calc_inline_size(
 719                                                    extent_end - end);
 720                         dec_i_blocks(inode, (extent_end - key.offset) -
 721                                         (extent_end - end));
 722                         btrfs_truncate_item(trans, root, path, new_size, 0);
 723                 }
 724                 /* create bookend, splitting the extent in two */
 725                 if (bookend && found_extent) {
 726                         struct btrfs_key ins;
 727                         ins.objectid = inode->i_ino;
 728                         ins.offset = end;
 729                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 730                         btrfs_release_path(root, path);
 731                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 732                                                       sizeof(*extent));
 733
 734                         leaf = path->nodes[0];
 735                         if (ret) {
 736                                 btrfs_print_leaf(root, leaf);
 737                                 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);
 738                         }
 739                         BUG_ON(ret);
 740                         extent = btrfs_item_ptr(leaf, path->slots[0],
 741                                                 struct btrfs_file_extent_item);
 742                         write_extent_buffer(leaf, &old,
 743                                             (unsigned long)extent, sizeof(old));
 744
 745                         btrfs_set_file_extent_offset(leaf, extent,
 746                                     le64_to_cpu(old.offset) + end - key.offset);
 747                         WARN_ON(le64_to_cpu(old.num_bytes) <
 748                                 (extent_end - end));
 749                         btrfs_set_file_extent_num_bytes(leaf, extent,
 750                                                         extent_end - end);
 751                         btrfs_set_file_extent_type(leaf, extent,
 752                                                    BTRFS_FILE_EXTENT_REG);
 753
 754                         btrfs_mark_buffer_dirty(path->nodes[0]);
 755                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 756                                 inode->i_blocks +=
 757                                       btrfs_file_extent_num_bytes(leaf,
 758                                                                   extent) >> 9;
 759                         }
 760                         ret = 0;
 761                         goto out;
 762                 }
 763         }
 764 out:
 765         btrfs_free_path(path);
 766         btrfs_check_file(root, inode);
 767         return ret;
 768 }
 769
 770 /*
 771  * this gets pages into the page cache and locks them down
 772  */
 773 static int prepare_pages(struct btrfs_root *root, struct file *file,
 774                          struct page **pages, size_t num_pages,
 775                          loff_t pos, unsigned long first_index,
 776                          unsigned long last_index, size_t write_bytes)
 777 {
 778         int i;
 779         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 780         struct inode *inode = fdentry(file)->d_inode;
 781         int err = 0;
 782         u64 start_pos;
 783
 784         start_pos = pos & ~((u64)root->sectorsize - 1);
 785
 786         memset(pages, 0, num_pages * sizeof(struct page *));
 787
 788         for (i = 0; i < num_pages; i++) {
 789                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 790                 if (!pages[i]) {
 791                         err = -ENOMEM;
 792                         BUG_ON(1);
 793                 }
 794 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 795                 ClearPageDirty(pages[i]);
 796 #else
 797                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 798 #endif
 799                 wait_on_page_writeback(pages[i]);
 800                 set_page_extent_mapped(pages[i]);
 801                 WARN_ON(!PageLocked(pages[i]));
 802         }
 803         if (start_pos < inode->i_size) {
 804                 u64 last_pos;
 805                 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 806                 lock_extent(&BTRFS_I(inode)->io_tree,
 807                             start_pos, last_pos - 1, GFP_NOFS);
 808                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 809                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 810                                   GFP_NOFS);
 811                 unlock_extent(&BTRFS_I(inode)->io_tree,
 812                               start_pos, last_pos - 1, GFP_NOFS);
 813         }
 814         return 0;
 815 }
 816
 817 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 818                                 size_t count, loff_t *ppos)
 819 {
 820         loff_t pos;
 821         loff_t start_pos;
 822         ssize_t num_written = 0;
 823         ssize_t err = 0;
 824         int ret = 0;
 825         struct inode *inode = fdentry(file)->d_inode;
 826         struct btrfs_root *root = BTRFS_I(inode)->root;
 827         struct page **pages = NULL;
 828         int nrptrs;
 829         struct page *pinned[2];
 830         unsigned long first_index;
 831         unsigned long last_index;
 832
 833         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 834                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 835         pinned[0] = NULL;
 836         pinned[1] = NULL;
 837
 838         pos = *ppos;
 839         start_pos = pos;
 840
 841         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 842         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 843         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 844         if (err)
 845                 goto out_nolock;
 846         if (count == 0)
 847                 goto out_nolock;
 848 #ifdef REMOVE_SUID_PATH
 849         err = remove_suid(&file->f_path);
 850 #else
 851         err = remove_suid(fdentry(file));
 852 #endif
 853         if (err)
 854                 goto out_nolock;
 855         file_update_time(file);
 856
 857         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 858
 859         mutex_lock(&inode->i_mutex);
 860         first_index = pos >> PAGE_CACHE_SHIFT;
 861         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 862
 863         /*
 864          * if this is a nodatasum mount, force summing off for the inode
 865          * all the time.  That way a later mount with summing on won't
 866          * get confused
 867          */
 868         if (btrfs_test_opt(root, NODATASUM))
 869                 btrfs_set_flag(inode, NODATASUM);
 870
 871         /*
 872          * there are lots of better ways to do this, but this code
 873          * makes sure the first and last page in the file range are
 874          * up to date and ready for cow
 875          */
 876         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 877                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 878                 if (!PageUptodate(pinned[0])) {
 879                         ret = btrfs_readpage(NULL, pinned[0]);
 880                         BUG_ON(ret);
 881                         wait_on_page_locked(pinned[0]);
 882                 } else {
 883                         unlock_page(pinned[0]);
 884                 }
 885         }
 886         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 887                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 888                 if (!PageUptodate(pinned[1])) {
 889                         ret = btrfs_readpage(NULL, pinned[1]);
 890                         BUG_ON(ret);
 891                         wait_on_page_locked(pinned[1]);
 892                 } else {
 893                         unlock_page(pinned[1]);
 894                 }
 895         }
 896
 897         while(count > 0) {
 898                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 899                 size_t write_bytes = min(count, nrptrs *
 900                                         (size_t)PAGE_CACHE_SIZE -
 901                                          offset);
 902                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 903                                         PAGE_CACHE_SHIFT;
 904
 905                 WARN_ON(num_pages > nrptrs);
 906                 memset(pages, 0, sizeof(pages));
 907
 908                 mutex_lock(&root->fs_info->fs_mutex);
 909                 ret = btrfs_check_free_space(root, write_bytes, 0);
 910                 mutex_unlock(&root->fs_info->fs_mutex);
 911                 if (ret)
 912                         goto out;
 913
 914                 ret = prepare_pages(root, file, pages, num_pages,
 915                                     pos, first_index, last_index,
 916                                     write_bytes);
 917                 if (ret)
 918                         goto out;
 919
 920                 ret = btrfs_copy_from_user(pos, num_pages,
 921                                            write_bytes, pages, buf);
 922                 if (ret) {
 923                         btrfs_drop_pages(pages, num_pages);
 924                         goto out;
 925                 }
 926
 927                 ret = dirty_and_release_pages(NULL, root, file, pages,
 928                                               num_pages, pos, write_bytes);
 929                 btrfs_drop_pages(pages, num_pages);
 930                 if (ret)
 931                         goto out;
 932
 933                 buf += write_bytes;
 934                 count -= write_bytes;
 935                 pos += write_bytes;
 936                 num_written += write_bytes;
 937
 938                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 939                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 940                         btrfs_btree_balance_dirty(root, 1);
 941                 btrfs_throttle(root);
 942                 cond_resched();
 943         }
 944 out:
 945         mutex_unlock(&inode->i_mutex);
 946
 947 out_nolock:
 948         kfree(pages);
 949         if (pinned[0])
 950                 page_cache_release(pinned[0]);
 951         if (pinned[1])
 952                 page_cache_release(pinned[1]);
 953         *ppos = pos;
 954
 955         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 956                 err = sync_page_range(inode, inode->i_mapping,
 957                                       start_pos, num_written);
 958                 if (err < 0)
 959                         num_written = err;
 960         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
 961 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
 962                 do_sync_file_range(file, start_pos,
 963                                       start_pos + num_written - 1,
 964                                       SYNC_FILE_RANGE_WRITE |
 965                                       SYNC_FILE_RANGE_WAIT_AFTER);
 966 #else
 967                 do_sync_mapping_range(inode->i_mapping, start_pos,
 968                                       start_pos + num_written - 1,
 969                                       SYNC_FILE_RANGE_WRITE |
 970                                       SYNC_FILE_RANGE_WAIT_AFTER);
 971 #endif
 972                 invalidate_mapping_pages(inode->i_mapping,
 973                       start_pos >> PAGE_CACHE_SHIFT,
 974                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
 975         }
 976         current->backing_dev_info = NULL;
 977         btrfs_ordered_throttle(root, inode);
 978         return num_written ? num_written : err;
 979 }
 980
 981 static int btrfs_sync_file(struct file *file,
 982                            struct dentry *dentry, int datasync)
 983 {
 984         struct inode *inode = dentry->d_inode;
 985         struct btrfs_root *root = BTRFS_I(inode)->root;
 986         int ret = 0;
 987         struct btrfs_trans_handle *trans;
 988
 989         /*
 990          * check the transaction that last modified this inode
 991          * and see if its already been committed
 992          */
 993         mutex_lock(&root->fs_info->fs_mutex);
 994         if (!BTRFS_I(inode)->last_trans)
 995                 goto out;
 996         mutex_lock(&root->fs_info->trans_mutex);
 997         if (BTRFS_I(inode)->last_trans <=
 998             root->fs_info->last_trans_committed) {
 999                 BTRFS_I(inode)->last_trans = 0;
1000                 mutex_unlock(&root->fs_info->trans_mutex);
1001                 goto out;
1002         }
1003         mutex_unlock(&root->fs_info->trans_mutex);
1004
1005         /*
1006          * ok we haven't committed the transaction yet, lets do a commit
1007          */
1008         trans = btrfs_start_transaction(root, 1);
1009         if (!trans) {
1010                 ret = -ENOMEM;
1011                 goto out;
1012         }
1013         ret = btrfs_commit_transaction(trans, root);
1014 out:
1015         mutex_unlock(&root->fs_info->fs_mutex);
1016         return ret > 0 ? EIO : ret;
1017 }
1018
1019 static struct vm_operations_struct btrfs_file_vm_ops = {
1020 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1021         .nopage         = filemap_nopage,
1022         .populate       = filemap_populate,
1023 #else
1024         .fault          = filemap_fault,
1025 #endif
1026         .page_mkwrite   = btrfs_page_mkwrite,
1027 };
1028
1029 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1030 {
1031         vma->vm_ops = &btrfs_file_vm_ops;
1032         file_accessed(filp);
1033         return 0;
1034 }
1035
1036 struct file_operations btrfs_file_operations = {
1037         .llseek         = generic_file_llseek,
1038         .read           = do_sync_read,
1039         .aio_read       = generic_file_aio_read,
1040         .splice_read    = generic_file_splice_read,
1041 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1042         .sendfile       = generic_file_sendfile,
1043 #endif
1044         .write          = btrfs_file_write,
1045         .mmap           = btrfs_file_mmap,
1046         .open           = generic_file_open,
1047         .fsync          = btrfs_sync_file,
1048         .unlocked_ioctl = btrfs_ioctl,
1049 #ifdef CONFIG_COMPAT
1050         .compat_ioctl   = btrfs_ioctl,
1051 #endif
1052 };
1053