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

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/mpage.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/version.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "ioctl.h"
  38 #include "print-tree.h"
  39
  40
  41 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  42                                 struct page **prepared_pages,
  43                                 const char __user * buf)
  44 {
  45         long page_fault = 0;
  46         int i;
  47         int offset = pos & (PAGE_CACHE_SIZE - 1);
  48
  49         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  50                 size_t count = min_t(size_t,
  51                                      PAGE_CACHE_SIZE - offset, write_bytes);
  52                 struct page *page = prepared_pages[i];
  53                 fault_in_pages_readable(buf, count);
  54
  55                 /* Copy data from userspace to the current page */
  56                 kmap(page);
  57                 page_fault = __copy_from_user(page_address(page) + offset,
  58                                               buf, count);
  59                 /* Flush processor's dcache for this page */
  60                 flush_dcache_page(page);
  61                 kunmap(page);
  62                 buf += count;
  63                 write_bytes -= count;
  64
  65                 if (page_fault)
  66                         break;
  67         }
  68         return page_fault ? -EFAULT : 0;
  69 }
  70
  71 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  72 {
  73         size_t i;
  74         for (i = 0; i < num_pages; i++) {
  75                 if (!pages[i])
  76                         break;
  77                 unlock_page(pages[i]);
  78                 mark_page_accessed(pages[i]);
  79                 page_cache_release(pages[i]);
  80         }
  81 }
  82
  83 static int insert_inline_extent(struct btrfs_trans_handle *trans,
  84                                 struct btrfs_root *root, struct inode *inode,
  85                                 u64 offset, size_t size,
  86                                 struct page **pages, size_t page_offset,
  87                                 int num_pages)
  88 {
  89         struct btrfs_key key;
  90         struct btrfs_path *path;
  91         struct extent_buffer *leaf;
  92         char *kaddr;
  93         unsigned long ptr;
  94         struct btrfs_file_extent_item *ei;
  95         struct page *page;
  96         u32 datasize;
  97         int err = 0;
  98         int ret;
  99         int i;
 100         ssize_t cur_size;
 101
 102         path = btrfs_alloc_path();
 103         if (!path)
 104                 return -ENOMEM;
 105
 106         btrfs_set_trans_block_group(trans, inode);
 107
 108         key.objectid = inode->i_ino;
 109         key.offset = offset;
 110         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 111
 112         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 113         if (ret < 0) {
 114                 err = ret;
 115                 goto fail;
 116         }
 117         if (ret == 1) {
 118                 path->slots[0]--;
 119                 leaf = path->nodes[0];
 120                 ei = btrfs_item_ptr(leaf, path->slots[0],
 121                                     struct btrfs_file_extent_item);
 122
 123                 if (btrfs_file_extent_type(leaf, ei) !=
 124                     BTRFS_FILE_EXTENT_INLINE) {
 125                         goto insert;
 126                 }
 127                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 128                 ret = 0;
 129         }
 130         if (ret == 0) {
 131                 u32 found_size;
 132                 u64 found_end;
 133
 134                 leaf = path->nodes[0];
 135                 ei = btrfs_item_ptr(leaf, path->slots[0],
 136                                     struct btrfs_file_extent_item);
 137
 138                 if (btrfs_file_extent_type(leaf, ei) !=
 139                     BTRFS_FILE_EXTENT_INLINE) {
 140                         err = ret;
 141                         btrfs_print_leaf(root, leaf);
 142                         printk("found wasn't inline offset %Lu inode %lu\n",
 143                                offset, inode->i_ino);
 144                         goto fail;
 145                 }
 146                 found_size = btrfs_file_extent_inline_len(leaf,
 147                                           btrfs_item_nr(leaf, path->slots[0]));
 148                 found_end = key.offset + found_size;
 149
 150                 if (found_end < offset + size) {
 151                         btrfs_release_path(root, path);
 152                         ret = btrfs_search_slot(trans, root, &key, path,
 153                                                 offset + size - found_end, 1);
 154                         BUG_ON(ret != 0);
 155                         ret = btrfs_extend_item(trans, root, path,
 156                                                 offset + size - found_end);
 157                         if (ret) {
 158                                 err = ret;
 159                                 goto fail;
 160                         }
 161                         leaf = path->nodes[0];
 162                         ei = btrfs_item_ptr(leaf, path->slots[0],
 163                                             struct btrfs_file_extent_item);
 164                 }
 165                 if (found_end < offset) {
 166                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 167                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 168                 }
 169         } else {
 170 insert:
 171                 btrfs_release_path(root, path);
 172                 datasize = offset + size - key.offset;
 173                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 174                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 175                                               datasize);
 176                 if (ret) {
 177                         err = ret;
 178                         printk("got bad ret %d\n", ret);
 179                         goto fail;
 180                 }
 181                 leaf = path->nodes[0];
 182                 ei = btrfs_item_ptr(leaf, path->slots[0],
 183                                     struct btrfs_file_extent_item);
 184                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 185                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 186         }
 187         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 188
 189         cur_size = size;
 190         i = 0;
 191         while (size > 0) {
 192                 page = pages[i];
 193                 kaddr = kmap_atomic(page, KM_USER0);
 194                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 195                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 196                 kunmap_atomic(kaddr, KM_USER0);
 197                 page_offset = 0;
 198                 ptr += cur_size;
 199                 size -= cur_size;
 200                 if (i >= num_pages) {
 201                         printk("i %d num_pages %d\n", i, num_pages);
 202                 }
 203                 i++;
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206 fail:
 207         btrfs_free_path(path);
 208         return err;
 209 }
 210
 211 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 212                                    struct btrfs_root *root,
 213                                    struct file *file,
 214                                    struct page **pages,
 215                                    size_t num_pages,
 216                                    loff_t pos,
 217                                    size_t write_bytes)
 218 {
 219         int err = 0;
 220         int i;
 221         struct inode *inode = file->f_path.dentry->d_inode;
 222         struct extent_map *em;
 223         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 224         u64 hint_byte;
 225         u64 num_bytes;
 226         u64 start_pos;
 227         u64 end_of_last_block;
 228         u64 end_pos = pos + write_bytes;
 229         u32 inline_size;
 230         loff_t isize = i_size_read(inode);
 231
 232         em = alloc_extent_map(GFP_NOFS);
 233         if (!em)
 234                 return -ENOMEM;
 235
 236         em->bdev = inode->i_sb->s_bdev;
 237
 238         start_pos = pos & ~((u64)root->sectorsize - 1);
 239         num_bytes = (write_bytes + pos - start_pos +
 240                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 241
 242         down_read(&BTRFS_I(inode)->root->snap_sem);
 243         end_of_last_block = start_pos + num_bytes - 1;
 244
 245         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 246         mutex_lock(&root->fs_info->fs_mutex);
 247         trans = btrfs_start_transaction(root, 1);
 248         if (!trans) {
 249                 err = -ENOMEM;
 250                 goto out_unlock;
 251         }
 252         btrfs_set_trans_block_group(trans, inode);
 253         inode->i_blocks += num_bytes >> 9;
 254         hint_byte = 0;
 255
 256         if ((end_of_last_block & 4095) == 0) {
 257                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 258         }
 259         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 260
 261         /* FIXME...EIEIO, ENOSPC and more */
 262
 263         /* insert any holes we need to create */
 264         if (inode->i_size < start_pos) {
 265                 u64 last_pos_in_file;
 266                 u64 hole_size;
 267                 u64 mask = root->sectorsize - 1;
 268                 last_pos_in_file = (isize + mask) & ~mask;
 269                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 270
 271                 if (last_pos_in_file < start_pos) {
 272                         err = btrfs_drop_extents(trans, root, inode,
 273                                                  last_pos_in_file,
 274                                                  last_pos_in_file + hole_size,
 275                                                  last_pos_in_file,
 276                                                  &hint_byte);
 277                         if (err)
 278                                 goto failed;
 279
 280                         err = btrfs_insert_file_extent(trans, root,
 281                                                        inode->i_ino,
 282                                                        last_pos_in_file,
 283                                                        0, 0, hole_size);
 284                 }
 285                 if (err)
 286                         goto failed;
 287         }
 288
 289         /*
 290          * either allocate an extent for the new bytes or setup the key
 291          * to show we are doing inline data in the extent
 292          */
 293         inline_size = end_pos;
 294         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 295             inline_size > 8192 ||
 296             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 297                 u64 last_end;
 298
 299                 for (i = 0; i < num_pages; i++) {
 300                         struct page *p = pages[i];
 301                         SetPageUptodate(p);
 302                         set_page_dirty(p);
 303                 }
 304                 last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
 305                 last_end += PAGE_CACHE_SIZE - 1;
 306                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 307                                  GFP_NOFS);
 308         } else {
 309                 u64 aligned_end;
 310                 /* step one, delete the existing extents in this range */
 311                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 312                         ~((u64)root->sectorsize - 1);
 313                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 314                                          aligned_end, end_pos, &hint_byte);
 315                 if (err)
 316                         goto failed;
 317                 err = insert_inline_extent(trans, root, inode, start_pos,
 318                                            end_pos - start_pos, pages, 0,
 319                                            num_pages);
 320                 BUG_ON(err);
 321         }
 322         if (end_pos > isize) {
 323                 i_size_write(inode, end_pos);
 324                 btrfs_update_inode(trans, root, inode);
 325         }
 326 failed:
 327         err = btrfs_end_transaction(trans, root);
 328 out_unlock:
 329         mutex_unlock(&root->fs_info->fs_mutex);
 330         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 331         free_extent_map(em);
 332         up_read(&BTRFS_I(inode)->root->snap_sem);
 333         return err;
 334 }
 335
 336 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 337 {
 338         struct extent_map *em;
 339         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 340
 341         while(1) {
 342                 em = lookup_extent_mapping(em_tree, start, end);
 343                 if (!em)
 344                         break;
 345                 remove_extent_mapping(em_tree, em);
 346                 /* once for us */
 347                 free_extent_map(em);
 348                 /* once for the tree*/
 349                 free_extent_map(em);
 350         }
 351         return 0;
 352 }
 353
 354 /*
 355  * this is very complex, but the basic idea is to drop all extents
 356  * in the range start - end.  hint_block is filled in with a block number
 357  * that would be a good hint to the block allocator for this file.
 358  *
 359  * If an extent intersects the range but is not entirely inside the range
 360  * it is either truncated or split.  Anything entirely inside the range
 361  * is deleted from the tree.
 362  */
 363 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 364                        struct btrfs_root *root, struct inode *inode,
 365                        u64 start, u64 end, u64 inline_end, u64 *hint_byte)
 366 {
 367         int ret;
 368         struct btrfs_key key;
 369         struct extent_buffer *leaf;
 370         int slot;
 371         struct btrfs_file_extent_item *extent;
 372         u64 extent_end = 0;
 373         int keep;
 374         struct btrfs_file_extent_item old;
 375         struct btrfs_path *path;
 376         u64 search_start = start;
 377         int bookend;
 378         int found_type;
 379         int found_extent;
 380         int found_inline;
 381         int recow;
 382
 383         btrfs_drop_extent_cache(inode, start, end - 1);
 384
 385         path = btrfs_alloc_path();
 386         if (!path)
 387                 return -ENOMEM;
 388         while(1) {
 389                 recow = 0;
 390                 btrfs_release_path(root, path);
 391                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 392                                                search_start, -1);
 393                 if (ret < 0)
 394                         goto out;
 395                 if (ret > 0) {
 396                         if (path->slots[0] == 0) {
 397                                 ret = 0;
 398                                 goto out;
 399                         }
 400                         path->slots[0]--;
 401                 }
 402 next_slot:
 403                 keep = 0;
 404                 bookend = 0;
 405                 found_extent = 0;
 406                 found_inline = 0;
 407                 extent = NULL;
 408                 leaf = path->nodes[0];
 409                 slot = path->slots[0];
 410                 ret = 0;
 411                 btrfs_item_key_to_cpu(leaf, &key, slot);
 412                 if (key.offset >= end || key.objectid != inode->i_ino) {
 413                         goto out;
 414                 }
 415                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 416                         goto out;
 417                 }
 418                 if (recow) {
 419                         search_start = key.offset;
 420                         continue;
 421                 }
 422                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 423                         extent = btrfs_item_ptr(leaf, slot,
 424                                                 struct btrfs_file_extent_item);
 425                         found_type = btrfs_file_extent_type(leaf, extent);
 426                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 427                                 extent_end = key.offset +
 428                                      btrfs_file_extent_num_bytes(leaf, extent);
 429                                 found_extent = 1;
 430                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 431                                 struct btrfs_item *item;
 432                                 item = btrfs_item_nr(leaf, slot);
 433                                 found_inline = 1;
 434                                 extent_end = key.offset +
 435                                      btrfs_file_extent_inline_len(leaf, item);
 436                         }
 437                 } else {
 438                         extent_end = search_start;
 439                 }
 440
 441                 /* we found nothing we can drop */
 442                 if ((!found_extent && !found_inline) ||
 443                     search_start >= extent_end) {
 444                         int nextret;
 445                         u32 nritems;
 446                         nritems = btrfs_header_nritems(leaf);
 447                         if (slot >= nritems - 1) {
 448                                 nextret = btrfs_next_leaf(root, path);
 449                                 if (nextret)
 450                                         goto out;
 451                                 recow = 1;
 452                         } else {
 453                                 path->slots[0]++;
 454                         }
 455                         goto next_slot;
 456                 }
 457
 458                 /* FIXME, there's only one inline extent allowed right now */
 459                 if (found_inline) {
 460                         u64 mask = root->sectorsize - 1;
 461                         search_start = (extent_end + mask) & ~mask;
 462                 } else
 463                         search_start = extent_end;
 464
 465                 if (end < extent_end && end >= key.offset) {
 466                         if (found_extent) {
 467                                 u64 disk_bytenr =
 468                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 469                                 u64 disk_num_bytes =
 470                                     btrfs_file_extent_disk_num_bytes(leaf,
 471                                                                       extent);
 472                                 read_extent_buffer(leaf, &old,
 473                                                    (unsigned long)extent,
 474                                                    sizeof(old));
 475                                 if (disk_bytenr != 0) {
 476                                         ret = btrfs_inc_extent_ref(trans, root,
 477                                                  disk_bytenr, disk_num_bytes);
 478                                         BUG_ON(ret);
 479                                 }
 480                         }
 481                         if (!found_inline)
 482                                 bookend = 1;
 483                 }
 484                 /* truncate existing extent */
 485                 if (start > key.offset) {
 486                         u64 new_num;
 487                         u64 old_num;
 488                         keep = 1;
 489                         WARN_ON(start & (root->sectorsize - 1));
 490                         if (found_extent) {
 491                                 new_num = start - key.offset;
 492                                 old_num = btrfs_file_extent_num_bytes(leaf,
 493                                                                       extent);
 494                                 *hint_byte =
 495                                         btrfs_file_extent_disk_bytenr(leaf,
 496                                                                       extent);
 497                                 if (btrfs_file_extent_disk_bytenr(leaf,
 498                                                                   extent)) {
 499                                         inode->i_blocks -=
 500                                                 (old_num - new_num) >> 9;
 501                                 }
 502                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 503                                                                 new_num);
 504                                 btrfs_mark_buffer_dirty(leaf);
 505                         } else if (end > extent_end &&
 506                                    key.offset < inline_end &&
 507                                    inline_end < extent_end) {
 508                                 u32 new_size;
 509                                 new_size = btrfs_file_extent_calc_inline_size(
 510                                                    inline_end - key.offset);
 511                                 btrfs_truncate_item(trans, root, path,
 512                                                     new_size);
 513                         }
 514                 }
 515                 /* delete the entire extent */
 516                 if (!keep) {
 517                         u64 disk_bytenr = 0;
 518                         u64 disk_num_bytes = 0;
 519                         u64 extent_num_bytes = 0;
 520                         if (found_extent) {
 521                                 disk_bytenr =
 522                                       btrfs_file_extent_disk_bytenr(leaf,
 523                                                                      extent);
 524                                 disk_num_bytes =
 525                                       btrfs_file_extent_disk_num_bytes(leaf,
 526                                                                        extent);
 527                                 extent_num_bytes =
 528                                       btrfs_file_extent_num_bytes(leaf, extent);
 529                                 *hint_byte =
 530                                         btrfs_file_extent_disk_bytenr(leaf,
 531                                                                       extent);
 532                         }
 533                         ret = btrfs_del_item(trans, root, path);
 534                         /* TODO update progress marker and return */
 535                         BUG_ON(ret);
 536                         btrfs_release_path(root, path);
 537                         extent = NULL;
 538                         if (found_extent && disk_bytenr != 0) {
 539                                 inode->i_blocks -= extent_num_bytes >> 9;
 540                                 ret = btrfs_free_extent(trans, root,
 541                                                         disk_bytenr,
 542                                                         disk_num_bytes, 0);
 543                         }
 544
 545                         BUG_ON(ret);
 546                         if (!bookend && search_start >= end) {
 547                                 ret = 0;
 548                                 goto out;
 549                         }
 550                         if (!bookend)
 551                                 continue;
 552                 }
 553                 /* create bookend, splitting the extent in two */
 554                 if (bookend && found_extent) {
 555                         struct btrfs_key ins;
 556                         ins.objectid = inode->i_ino;
 557                         ins.offset = end;
 558                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 559                         btrfs_release_path(root, path);
 560                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 561                                                       sizeof(*extent));
 562
 563                         leaf = path->nodes[0];
 564                         if (ret) {
 565                                 btrfs_print_leaf(root, leaf);
 566                                 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);
 567                         }
 568                         BUG_ON(ret);
 569                         extent = btrfs_item_ptr(leaf, path->slots[0],
 570                                                 struct btrfs_file_extent_item);
 571                         write_extent_buffer(leaf, &old,
 572                                             (unsigned long)extent, sizeof(old));
 573
 574                         btrfs_set_file_extent_offset(leaf, extent,
 575                                     le64_to_cpu(old.offset) + end - key.offset);
 576                         WARN_ON(le64_to_cpu(old.num_bytes) <
 577                                 (extent_end - end));
 578                         btrfs_set_file_extent_num_bytes(leaf, extent,
 579                                                         extent_end - end);
 580                         btrfs_set_file_extent_type(leaf, extent,
 581                                                    BTRFS_FILE_EXTENT_REG);
 582
 583                         btrfs_mark_buffer_dirty(path->nodes[0]);
 584                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 585                                 inode->i_blocks +=
 586                                       btrfs_file_extent_num_bytes(leaf,
 587                                                                   extent) >> 9;
 588                         }
 589                         ret = 0;
 590                         goto out;
 591                 }
 592         }
 593 out:
 594         btrfs_free_path(path);
 595         return ret;
 596 }
 597
 598 /*
 599  * this gets pages into the page cache and locks them down
 600  */
 601 static int prepare_pages(struct btrfs_root *root,
 602                          struct file *file,
 603                          struct page **pages,
 604                          size_t num_pages,
 605                          loff_t pos,
 606                          unsigned long first_index,
 607                          unsigned long last_index,
 608                          size_t write_bytes)
 609 {
 610         int i;
 611         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 612         struct inode *inode = file->f_path.dentry->d_inode;
 613         int err = 0;
 614         u64 start_pos;
 615
 616         start_pos = pos & ~((u64)root->sectorsize - 1);
 617
 618         memset(pages, 0, num_pages * sizeof(struct page *));
 619
 620         for (i = 0; i < num_pages; i++) {
 621                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 622                 if (!pages[i]) {
 623                         err = -ENOMEM;
 624                         BUG_ON(1);
 625                 }
 626                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 627                 wait_on_page_writeback(pages[i]);
 628                 set_page_extent_mapped(pages[i]);
 629                 WARN_ON(!PageLocked(pages[i]));
 630         }
 631         return 0;
 632 }
 633
 634 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 635                                 size_t count, loff_t *ppos)
 636 {
 637         loff_t pos;
 638         size_t num_written = 0;
 639         int err = 0;
 640         int ret = 0;
 641         struct inode *inode = file->f_path.dentry->d_inode;
 642         struct btrfs_root *root = BTRFS_I(inode)->root;
 643         struct page **pages = NULL;
 644         int nrptrs;
 645         struct page *pinned[2];
 646         unsigned long first_index;
 647         unsigned long last_index;
 648
 649         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 650                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 651         pinned[0] = NULL;
 652         pinned[1] = NULL;
 653         if (file->f_flags & O_DIRECT)
 654                 return -EINVAL;
 655         pos = *ppos;
 656         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 657         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 658         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 659         if (err)
 660                 goto out;
 661         if (count == 0)
 662                 goto out;
 663         err = remove_suid(file->f_path.dentry);
 664         if (err)
 665                 goto out;
 666         file_update_time(file);
 667
 668         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 669
 670         mutex_lock(&inode->i_mutex);
 671         first_index = pos >> PAGE_CACHE_SHIFT;
 672         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 673
 674         /*
 675          * there are lots of better ways to do this, but this code
 676          * makes sure the first and last page in the file range are
 677          * up to date and ready for cow
 678          */
 679         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 680                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 681                 if (!PageUptodate(pinned[0])) {
 682                         ret = btrfs_readpage(NULL, pinned[0]);
 683                         BUG_ON(ret);
 684                         wait_on_page_locked(pinned[0]);
 685                 } else {
 686                         unlock_page(pinned[0]);
 687                 }
 688         }
 689         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 690                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 691                 if (!PageUptodate(pinned[1])) {
 692                         ret = btrfs_readpage(NULL, pinned[1]);
 693                         BUG_ON(ret);
 694                         wait_on_page_locked(pinned[1]);
 695                 } else {
 696                         unlock_page(pinned[1]);
 697                 }
 698         }
 699
 700         while(count > 0) {
 701                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 702                 size_t write_bytes = min(count, nrptrs *
 703                                         (size_t)PAGE_CACHE_SIZE -
 704                                          offset);
 705                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 706                                         PAGE_CACHE_SHIFT;
 707
 708                 WARN_ON(num_pages > nrptrs);
 709                 memset(pages, 0, sizeof(pages));
 710                 ret = prepare_pages(root, file, pages, num_pages,
 711                                     pos, first_index, last_index,
 712                                     write_bytes);
 713                 if (ret)
 714                         goto out;
 715
 716                 ret = btrfs_copy_from_user(pos, num_pages,
 717                                            write_bytes, pages, buf);
 718                 if (ret) {
 719                         btrfs_drop_pages(pages, num_pages);
 720                         goto out;
 721                 }
 722
 723                 ret = dirty_and_release_pages(NULL, root, file, pages,
 724                                               num_pages, pos, write_bytes);
 725                 btrfs_drop_pages(pages, num_pages);
 726                 if (ret)
 727                         goto out;
 728
 729                 buf += write_bytes;
 730                 count -= write_bytes;
 731                 pos += write_bytes;
 732                 num_written += write_bytes;
 733
 734                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 735                 btrfs_btree_balance_dirty(root, 1);
 736                 cond_resched();
 737         }
 738         mutex_unlock(&inode->i_mutex);
 739 out:
 740         kfree(pages);
 741         if (pinned[0])
 742                 page_cache_release(pinned[0]);
 743         if (pinned[1])
 744                 page_cache_release(pinned[1]);
 745         *ppos = pos;
 746         current->backing_dev_info = NULL;
 747         return num_written ? num_written : err;
 748 }
 749
 750 static int btrfs_sync_file(struct file *file,
 751                            struct dentry *dentry, int datasync)
 752 {
 753         struct inode *inode = dentry->d_inode;
 754         struct btrfs_root *root = BTRFS_I(inode)->root;
 755         int ret = 0;
 756         struct btrfs_trans_handle *trans;
 757
 758         /*
 759          * check the transaction that last modified this inode
 760          * and see if its already been committed
 761          */
 762         mutex_lock(&root->fs_info->fs_mutex);
 763         if (!BTRFS_I(inode)->last_trans)
 764                 goto out;
 765         mutex_lock(&root->fs_info->trans_mutex);
 766         if (BTRFS_I(inode)->last_trans <=
 767             root->fs_info->last_trans_committed) {
 768                 BTRFS_I(inode)->last_trans = 0;
 769                 mutex_unlock(&root->fs_info->trans_mutex);
 770                 goto out;
 771         }
 772         mutex_unlock(&root->fs_info->trans_mutex);
 773
 774         /*
 775          * ok we haven't committed the transaction yet, lets do a commit
 776          */
 777         trans = btrfs_start_transaction(root, 1);
 778         if (!trans) {
 779                 ret = -ENOMEM;
 780                 goto out;
 781         }
 782         ret = btrfs_commit_transaction(trans, root);
 783 out:
 784         mutex_unlock(&root->fs_info->fs_mutex);
 785         return ret > 0 ? EIO : ret;
 786 }
 787
 788 static struct vm_operations_struct btrfs_file_vm_ops = {
 789 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 790         .nopage         = filemap_nopage,
 791         .populate       = filemap_populate,
 792 #else
 793         .fault          = filemap_fault,
 794 #endif
 795         .page_mkwrite   = btrfs_page_mkwrite,
 796 };
 797
 798 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 799 {
 800         vma->vm_ops = &btrfs_file_vm_ops;
 801         file_accessed(filp);
 802         return 0;
 803 }
 804
 805 struct file_operations btrfs_file_operations = {
 806         .llseek         = generic_file_llseek,
 807         .read           = do_sync_read,
 808         .aio_read       = generic_file_aio_read,
 809         .write          = btrfs_file_write,
 810         .mmap           = btrfs_file_mmap,
 811         .open           = generic_file_open,
 812         .fsync          = btrfs_sync_file,
 813         .unlocked_ioctl = btrfs_ioctl,
 814 #ifdef CONFIG_COMPAT
 815         .compat_ioctl   = btrfs_ioctl,
 816 #endif
 817 };
 818