fs/btrfs/disk-io.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/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h> // for block_sync_page
  26 #include <linux/workqueue.h>
  27 #include "crc32c.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "transaction.h"
  31 #include "btrfs_inode.h"
  32 #include "volumes.h"
  33 #include "print-tree.h"
  34
  35 #if 0
  36 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  37 {
  38         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  39                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  40                        (unsigned long long)extent_buffer_blocknr(buf),
  41                        (unsigned long long)btrfs_header_blocknr(buf));
  42                 return 1;
  43         }
  44         return 0;
  45 }
  46 #endif
  47
  48 static struct extent_io_ops btree_extent_io_ops;
  49 static struct workqueue_struct *end_io_workqueue;
  50 static struct workqueue_struct *async_submit_workqueue;
  51
  52 struct end_io_wq {
  53         struct bio *bio;
  54         bio_end_io_t *end_io;
  55         void *private;
  56         struct btrfs_fs_info *info;
  57         int error;
  58         int metadata;
  59         struct list_head list;
  60 };
  61
  62 struct async_submit_bio {
  63         struct inode *inode;
  64         struct bio *bio;
  65         struct list_head list;
  66         extent_submit_bio_hook_t *submit_bio_hook;
  67         int rw;
  68         int mirror_num;
  69 };
  70
  71 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  72                                     size_t page_offset, u64 start, u64 len,
  73                                     int create)
  74 {
  75         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  76         struct extent_map *em;
  77         int ret;
  78
  79         spin_lock(&em_tree->lock);
  80         em = lookup_extent_mapping(em_tree, start, len);
  81         if (em) {
  82                 em->bdev =
  83                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  84                 spin_unlock(&em_tree->lock);
  85                 goto out;
  86         }
  87         spin_unlock(&em_tree->lock);
  88
  89         em = alloc_extent_map(GFP_NOFS);
  90         if (!em) {
  91                 em = ERR_PTR(-ENOMEM);
  92                 goto out;
  93         }
  94         em->start = 0;
  95         em->len = (u64)-1;
  96         em->block_start = 0;
  97         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  98
  99         spin_lock(&em_tree->lock);
 100         ret = add_extent_mapping(em_tree, em);
 101         if (ret == -EEXIST) {
 102                 u64 failed_start = em->start;
 103                 u64 failed_len = em->len;
 104
 105                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 106                        em->start, em->len, em->block_start);
 107                 free_extent_map(em);
 108                 em = lookup_extent_mapping(em_tree, start, len);
 109                 if (em) {
 110                         printk("after failing, found %Lu %Lu %Lu\n",
 111                                em->start, em->len, em->block_start);
 112                         ret = 0;
 113                 } else {
 114                         em = lookup_extent_mapping(em_tree, failed_start,
 115                                                    failed_len);
 116                         if (em) {
 117                                 printk("double failure lookup gives us "
 118                                        "%Lu %Lu -> %Lu\n", em->start,
 119                                        em->len, em->block_start);
 120                                 free_extent_map(em);
 121                         }
 122                         ret = -EIO;
 123                 }
 124         } else if (ret) {
 125                 free_extent_map(em);
 126                 em = NULL;
 127         }
 128         spin_unlock(&em_tree->lock);
 129
 130         if (ret)
 131                 em = ERR_PTR(ret);
 132 out:
 133         return em;
 134 }
 135
 136 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 137 {
 138         return btrfs_crc32c(seed, data, len);
 139 }
 140
 141 void btrfs_csum_final(u32 crc, char *result)
 142 {
 143         *(__le32 *)result = ~cpu_to_le32(crc);
 144 }
 145
 146 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 147                            int verify)
 148 {
 149         char result[BTRFS_CRC32_SIZE];
 150         unsigned long len;
 151         unsigned long cur_len;
 152         unsigned long offset = BTRFS_CSUM_SIZE;
 153         char *map_token = NULL;
 154         char *kaddr;
 155         unsigned long map_start;
 156         unsigned long map_len;
 157         int err;
 158         u32 crc = ~(u32)0;
 159
 160         len = buf->len - offset;
 161         while(len > 0) {
 162                 err = map_private_extent_buffer(buf, offset, 32,
 163                                         &map_token, &kaddr,
 164                                         &map_start, &map_len, KM_USER0);
 165                 if (err) {
 166                         printk("failed to map extent buffer! %lu\n",
 167                                offset);
 168                         return 1;
 169                 }
 170                 cur_len = min(len, map_len - (offset - map_start));
 171                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 172                                       crc, cur_len);
 173                 len -= cur_len;
 174                 offset += cur_len;
 175                 unmap_extent_buffer(buf, map_token, KM_USER0);
 176         }
 177         btrfs_csum_final(crc, result);
 178
 179         if (verify) {
 180                 int from_this_trans = 0;
 181
 182                 if (root->fs_info->running_transaction &&
 183                     btrfs_header_generation(buf) ==
 184                     root->fs_info->running_transaction->transid)
 185                         from_this_trans = 1;
 186
 187                 /* FIXME, this is not good */
 188                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 189                         u32 val;
 190                         u32 found = 0;
 191                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 192
 193                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 194                         printk("btrfs: %s checksum verify failed on %llu "
 195                                "wanted %X found %X from_this_trans %d "
 196                                "level %d\n",
 197                                root->fs_info->sb->s_id,
 198                                buf->start, val, found, from_this_trans,
 199                                btrfs_header_level(buf));
 200                         return 1;
 201                 }
 202         } else {
 203                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 204         }
 205         return 0;
 206 }
 207
 208 static int verify_parent_transid(struct extent_io_tree *io_tree,
 209                                  struct extent_buffer *eb, u64 parent_transid)
 210 {
 211         int ret;
 212
 213         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 214                 return 0;
 215
 216         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 217         if (extent_buffer_uptodate(io_tree, eb) &&
 218             btrfs_header_generation(eb) == parent_transid) {
 219                 ret = 0;
 220                 goto out;
 221         }
 222         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 223                (unsigned long long)eb->start,
 224                (unsigned long long)parent_transid,
 225                (unsigned long long)btrfs_header_generation(eb));
 226         ret = 1;
 227 out:
 228         clear_extent_buffer_uptodate(io_tree, eb);
 229         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 230                       GFP_NOFS);
 231         return ret;
 232
 233 }
 234
 235 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 236                                           struct extent_buffer *eb,
 237                                           u64 start, u64 parent_transid)
 238 {
 239         struct extent_io_tree *io_tree;
 240         int ret;
 241         int num_copies = 0;
 242         int mirror_num = 0;
 243
 244         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 245         while (1) {
 246                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 247                                                btree_get_extent, mirror_num);
 248                 if (!ret &&
 249                     !verify_parent_transid(io_tree, eb, parent_transid))
 250                         return ret;
 251
 252                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 253                                               eb->start, eb->len);
 254                 if (num_copies == 1)
 255                         return ret;
 256
 257                 mirror_num++;
 258                 if (mirror_num > num_copies)
 259                         return ret;
 260         }
 261         return -EIO;
 262 }
 263
 264 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 265 {
 266         struct extent_io_tree *tree;
 267         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 268         u64 found_start;
 269         int found_level;
 270         unsigned long len;
 271         struct extent_buffer *eb;
 272         int ret;
 273
 274         tree = &BTRFS_I(page->mapping->host)->io_tree;
 275
 276         if (page->private == EXTENT_PAGE_PRIVATE)
 277                 goto out;
 278         if (!page->private)
 279                 goto out;
 280         len = page->private >> 2;
 281         if (len == 0) {
 282                 WARN_ON(1);
 283         }
 284         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 285         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 286                                              btrfs_header_generation(eb));
 287         BUG_ON(ret);
 288         btrfs_clear_buffer_defrag(eb);
 289         found_start = btrfs_header_bytenr(eb);
 290         if (found_start != start) {
 291                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 292                        start, found_start, len);
 293                 WARN_ON(1);
 294                 goto err;
 295         }
 296         if (eb->first_page != page) {
 297                 printk("bad first page %lu %lu\n", eb->first_page->index,
 298                        page->index);
 299                 WARN_ON(1);
 300                 goto err;
 301         }
 302         if (!PageUptodate(page)) {
 303                 printk("csum not up to date page %lu\n", page->index);
 304                 WARN_ON(1);
 305                 goto err;
 306         }
 307         found_level = btrfs_header_level(eb);
 308         spin_lock(&root->fs_info->hash_lock);
 309         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 310         spin_unlock(&root->fs_info->hash_lock);
 311         csum_tree_block(root, eb, 0);
 312 err:
 313         free_extent_buffer(eb);
 314 out:
 315         return 0;
 316 }
 317
 318 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 319 {
 320         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 321
 322         csum_dirty_buffer(root, page);
 323         return 0;
 324 }
 325
 326 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 327                                struct extent_state *state)
 328 {
 329         struct extent_io_tree *tree;
 330         u64 found_start;
 331         int found_level;
 332         unsigned long len;
 333         struct extent_buffer *eb;
 334         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 335         int ret = 0;
 336
 337         tree = &BTRFS_I(page->mapping->host)->io_tree;
 338         if (page->private == EXTENT_PAGE_PRIVATE)
 339                 goto out;
 340         if (!page->private)
 341                 goto out;
 342         len = page->private >> 2;
 343         if (len == 0) {
 344                 WARN_ON(1);
 345         }
 346         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 347
 348         btrfs_clear_buffer_defrag(eb);
 349         found_start = btrfs_header_bytenr(eb);
 350         if (found_start != start) {
 351                 ret = -EIO;
 352                 goto err;
 353         }
 354         if (eb->first_page != page) {
 355                 printk("bad first page %lu %lu\n", eb->first_page->index,
 356                        page->index);
 357                 WARN_ON(1);
 358                 ret = -EIO;
 359                 goto err;
 360         }
 361         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 362                                  (unsigned long)btrfs_header_fsid(eb),
 363                                  BTRFS_FSID_SIZE)) {
 364                 printk("bad fsid on block %Lu\n", eb->start);
 365                 ret = -EIO;
 366                 goto err;
 367         }
 368         found_level = btrfs_header_level(eb);
 369
 370         ret = csum_tree_block(root, eb, 1);
 371         if (ret)
 372                 ret = -EIO;
 373
 374         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 375         end = eb->start + end - 1;
 376         release_extent_buffer_tail_pages(eb);
 377 err:
 378         free_extent_buffer(eb);
 379 out:
 380         return ret;
 381 }
 382
 383 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 384 static void end_workqueue_bio(struct bio *bio, int err)
 385 #else
 386 static int end_workqueue_bio(struct bio *bio,
 387                                    unsigned int bytes_done, int err)
 388 #endif
 389 {
 390         struct end_io_wq *end_io_wq = bio->bi_private;
 391         struct btrfs_fs_info *fs_info;
 392         unsigned long flags;
 393
 394 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 395         if (bio->bi_size)
 396                 return 1;
 397 #endif
 398
 399         fs_info = end_io_wq->info;
 400         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 401         end_io_wq->error = err;
 402         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 403         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 404         queue_work(end_io_workqueue, &fs_info->end_io_work);
 405
 406 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 407         return 0;
 408 #endif
 409 }
 410
 411 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 412                         int metadata)
 413 {
 414         struct end_io_wq *end_io_wq;
 415         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 416         if (!end_io_wq)
 417                 return -ENOMEM;
 418
 419         end_io_wq->private = bio->bi_private;
 420         end_io_wq->end_io = bio->bi_end_io;
 421         end_io_wq->info = info;
 422         end_io_wq->error = 0;
 423         end_io_wq->bio = bio;
 424         end_io_wq->metadata = metadata;
 425
 426         bio->bi_private = end_io_wq;
 427         bio->bi_end_io = end_workqueue_bio;
 428         return 0;
 429 }
 430
 431 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 432                         int rw, struct bio *bio, int mirror_num,
 433                         extent_submit_bio_hook_t *submit_bio_hook)
 434 {
 435         struct async_submit_bio *async;
 436
 437         /*
 438          * inline writerback should stay inline, only hop to the async
 439          * queue if we're pdflush
 440          */
 441         if (!current_is_pdflush())
 442                 return submit_bio_hook(inode, rw, bio, mirror_num);
 443
 444         async = kmalloc(sizeof(*async), GFP_NOFS);
 445         if (!async)
 446                 return -ENOMEM;
 447
 448         async->inode = inode;
 449         async->rw = rw;
 450         async->bio = bio;
 451         async->mirror_num = mirror_num;
 452         async->submit_bio_hook = submit_bio_hook;
 453
 454         spin_lock(&fs_info->async_submit_work_lock);
 455         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 456         spin_unlock(&fs_info->async_submit_work_lock);
 457
 458         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 459         return 0;
 460 }
 461
 462 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 463                                  int mirror_num)
 464 {
 465         struct btrfs_root *root = BTRFS_I(inode)->root;
 466         u64 offset;
 467         int ret;
 468
 469         offset = bio->bi_sector << 9;
 470
 471         if (rw & (1 << BIO_RW)) {
 472                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 473         }
 474
 475         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 476         BUG_ON(ret);
 477
 478         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 479 }
 480
 481 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 482                                  int mirror_num)
 483 {
 484         if (!(rw & (1 << BIO_RW))) {
 485                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 486         }
 487         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 488                                    inode, rw, bio, mirror_num,
 489                                    __btree_submit_bio_hook);
 490 }
 491
 492 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 493 {
 494         struct extent_io_tree *tree;
 495         tree = &BTRFS_I(page->mapping->host)->io_tree;
 496         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 497 }
 498
 499 static int btree_writepages(struct address_space *mapping,
 500                             struct writeback_control *wbc)
 501 {
 502         struct extent_io_tree *tree;
 503         tree = &BTRFS_I(mapping->host)->io_tree;
 504         if (wbc->sync_mode == WB_SYNC_NONE) {
 505                 u64 num_dirty;
 506                 u64 start = 0;
 507                 unsigned long thresh = 96 * 1024 * 1024;
 508
 509                 if (wbc->for_kupdate)
 510                         return 0;
 511
 512                 if (current_is_pdflush()) {
 513                         thresh = 96 * 1024 * 1024;
 514                 } else {
 515                         thresh = 8 * 1024 * 1024;
 516                 }
 517                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 518                                              thresh, EXTENT_DIRTY);
 519                 if (num_dirty < thresh) {
 520                         return 0;
 521                 }
 522         }
 523         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 524 }
 525
 526 int btree_readpage(struct file *file, struct page *page)
 527 {
 528         struct extent_io_tree *tree;
 529         tree = &BTRFS_I(page->mapping->host)->io_tree;
 530         return extent_read_full_page(tree, page, btree_get_extent);
 531 }
 532
 533 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 534 {
 535         struct extent_io_tree *tree;
 536         struct extent_map_tree *map;
 537         int ret;
 538
 539         if (page_count(page) > 3) {
 540                 /* once for page->private, once for the caller, once
 541                  * once for the page cache
 542                  */
 543                 return 0;
 544         }
 545         tree = &BTRFS_I(page->mapping->host)->io_tree;
 546         map = &BTRFS_I(page->mapping->host)->extent_tree;
 547         ret = try_release_extent_state(map, tree, page, gfp_flags);
 548         if (ret == 1) {
 549                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 550                 ClearPagePrivate(page);
 551                 set_page_private(page, 0);
 552                 page_cache_release(page);
 553         }
 554         return ret;
 555 }
 556
 557 static void btree_invalidatepage(struct page *page, unsigned long offset)
 558 {
 559         struct extent_io_tree *tree;
 560         tree = &BTRFS_I(page->mapping->host)->io_tree;
 561         extent_invalidatepage(tree, page, offset);
 562         btree_releasepage(page, GFP_NOFS);
 563         if (PagePrivate(page)) {
 564                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 565                 ClearPagePrivate(page);
 566                 set_page_private(page, 0);
 567                 page_cache_release(page);
 568         }
 569 }
 570
 571 #if 0
 572 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 573 {
 574         struct buffer_head *bh;
 575         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 576         struct buffer_head *head;
 577         if (!page_has_buffers(page)) {
 578                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 579                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 580         }
 581         head = page_buffers(page);
 582         bh = head;
 583         do {
 584                 if (buffer_dirty(bh))
 585                         csum_tree_block(root, bh, 0);
 586                 bh = bh->b_this_page;
 587         } while (bh != head);
 588         return block_write_full_page(page, btree_get_block, wbc);
 589 }
 590 #endif
 591
 592 static struct address_space_operations btree_aops = {
 593         .readpage       = btree_readpage,
 594         .writepage      = btree_writepage,
 595         .writepages     = btree_writepages,
 596         .releasepage    = btree_releasepage,
 597         .invalidatepage = btree_invalidatepage,
 598         .sync_page      = block_sync_page,
 599 };
 600
 601 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 602                          u64 parent_transid)
 603 {
 604         struct extent_buffer *buf = NULL;
 605         struct inode *btree_inode = root->fs_info->btree_inode;
 606         int ret = 0;
 607
 608         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 609         if (!buf)
 610                 return 0;
 611         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 612                                  buf, 0, 0, btree_get_extent, 0);
 613         free_extent_buffer(buf);
 614         return ret;
 615 }
 616
 617 static int close_all_devices(struct btrfs_fs_info *fs_info)
 618 {
 619         struct list_head *list;
 620         struct list_head *next;
 621         struct btrfs_device *device;
 622
 623         list = &fs_info->fs_devices->devices;
 624         list_for_each(next, list) {
 625                 device = list_entry(next, struct btrfs_device, dev_list);
 626                 close_bdev_excl(device->bdev);
 627                 device->bdev = NULL;
 628         }
 629         return 0;
 630 }
 631
 632 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 633                                             u64 bytenr, u32 blocksize)
 634 {
 635         struct inode *btree_inode = root->fs_info->btree_inode;
 636         struct extent_buffer *eb;
 637         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 638                                 bytenr, blocksize, GFP_NOFS);
 639         return eb;
 640 }
 641
 642 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 643                                                  u64 bytenr, u32 blocksize)
 644 {
 645         struct inode *btree_inode = root->fs_info->btree_inode;
 646         struct extent_buffer *eb;
 647
 648         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 649                                  bytenr, blocksize, NULL, GFP_NOFS);
 650         return eb;
 651 }
 652
 653
 654 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 655                                       u32 blocksize, u64 parent_transid)
 656 {
 657         struct extent_buffer *buf = NULL;
 658         struct inode *btree_inode = root->fs_info->btree_inode;
 659         struct extent_io_tree *io_tree;
 660         int ret;
 661
 662         io_tree = &BTRFS_I(btree_inode)->io_tree;
 663
 664         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 665         if (!buf)
 666                 return NULL;
 667
 668         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 669
 670         if (ret == 0) {
 671                 buf->flags |= EXTENT_UPTODATE;
 672         }
 673         return buf;
 674
 675 }
 676
 677 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 678                      struct extent_buffer *buf)
 679 {
 680         struct inode *btree_inode = root->fs_info->btree_inode;
 681         if (btrfs_header_generation(buf) ==
 682             root->fs_info->running_transaction->transid)
 683                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 684                                           buf);
 685         return 0;
 686 }
 687
 688 int wait_on_tree_block_writeback(struct btrfs_root *root,
 689                                  struct extent_buffer *buf)
 690 {
 691         struct inode *btree_inode = root->fs_info->btree_inode;
 692         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 693                                         buf);
 694         return 0;
 695 }
 696
 697 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 698                         u32 stripesize, struct btrfs_root *root,
 699                         struct btrfs_fs_info *fs_info,
 700                         u64 objectid)
 701 {
 702         root->node = NULL;
 703         root->inode = NULL;
 704         root->commit_root = NULL;
 705         root->sectorsize = sectorsize;
 706         root->nodesize = nodesize;
 707         root->leafsize = leafsize;
 708         root->stripesize = stripesize;
 709         root->ref_cows = 0;
 710         root->track_dirty = 0;
 711
 712         root->fs_info = fs_info;
 713         root->objectid = objectid;
 714         root->last_trans = 0;
 715         root->highest_inode = 0;
 716         root->last_inode_alloc = 0;
 717         root->name = NULL;
 718         root->in_sysfs = 0;
 719
 720         INIT_LIST_HEAD(&root->dirty_list);
 721         memset(&root->root_key, 0, sizeof(root->root_key));
 722         memset(&root->root_item, 0, sizeof(root->root_item));
 723         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 724         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 725         init_completion(&root->kobj_unregister);
 726         root->defrag_running = 0;
 727         root->defrag_level = 0;
 728         root->root_key.objectid = objectid;
 729         return 0;
 730 }
 731
 732 static int find_and_setup_root(struct btrfs_root *tree_root,
 733                                struct btrfs_fs_info *fs_info,
 734                                u64 objectid,
 735                                struct btrfs_root *root)
 736 {
 737         int ret;
 738         u32 blocksize;
 739
 740         __setup_root(tree_root->nodesize, tree_root->leafsize,
 741                      tree_root->sectorsize, tree_root->stripesize,
 742                      root, fs_info, objectid);
 743         ret = btrfs_find_last_root(tree_root, objectid,
 744                                    &root->root_item, &root->root_key);
 745         BUG_ON(ret);
 746
 747         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 748         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 749                                      blocksize, 0);
 750         BUG_ON(!root->node);
 751         return 0;
 752 }
 753
 754 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 755                                                struct btrfs_key *location)
 756 {
 757         struct btrfs_root *root;
 758         struct btrfs_root *tree_root = fs_info->tree_root;
 759         struct btrfs_path *path;
 760         struct extent_buffer *l;
 761         u64 highest_inode;
 762         u32 blocksize;
 763         int ret = 0;
 764
 765         root = kzalloc(sizeof(*root), GFP_NOFS);
 766         if (!root)
 767                 return ERR_PTR(-ENOMEM);
 768         if (location->offset == (u64)-1) {
 769                 ret = find_and_setup_root(tree_root, fs_info,
 770                                           location->objectid, root);
 771                 if (ret) {
 772                         kfree(root);
 773                         return ERR_PTR(ret);
 774                 }
 775                 goto insert;
 776         }
 777
 778         __setup_root(tree_root->nodesize, tree_root->leafsize,
 779                      tree_root->sectorsize, tree_root->stripesize,
 780                      root, fs_info, location->objectid);
 781
 782         path = btrfs_alloc_path();
 783         BUG_ON(!path);
 784         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 785         if (ret != 0) {
 786                 if (ret > 0)
 787                         ret = -ENOENT;
 788                 goto out;
 789         }
 790         l = path->nodes[0];
 791         read_extent_buffer(l, &root->root_item,
 792                btrfs_item_ptr_offset(l, path->slots[0]),
 793                sizeof(root->root_item));
 794         memcpy(&root->root_key, location, sizeof(*location));
 795         ret = 0;
 796 out:
 797         btrfs_release_path(root, path);
 798         btrfs_free_path(path);
 799         if (ret) {
 800                 kfree(root);
 801                 return ERR_PTR(ret);
 802         }
 803         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 804         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 805                                      blocksize, 0);
 806         BUG_ON(!root->node);
 807 insert:
 808         root->ref_cows = 1;
 809         ret = btrfs_find_highest_inode(root, &highest_inode);
 810         if (ret == 0) {
 811                 root->highest_inode = highest_inode;
 812                 root->last_inode_alloc = highest_inode;
 813         }
 814         return root;
 815 }
 816
 817 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 818                                         u64 root_objectid)
 819 {
 820         struct btrfs_root *root;
 821
 822         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 823                 return fs_info->tree_root;
 824         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 825                 return fs_info->extent_root;
 826
 827         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 828                                  (unsigned long)root_objectid);
 829         return root;
 830 }
 831
 832 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 833                                               struct btrfs_key *location)
 834 {
 835         struct btrfs_root *root;
 836         int ret;
 837
 838         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 839                 return fs_info->tree_root;
 840         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 841                 return fs_info->extent_root;
 842         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 843                 return fs_info->chunk_root;
 844         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 845                 return fs_info->dev_root;
 846
 847         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 848                                  (unsigned long)location->objectid);
 849         if (root)
 850                 return root;
 851
 852         root = btrfs_read_fs_root_no_radix(fs_info, location);
 853         if (IS_ERR(root))
 854                 return root;
 855         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 856                                 (unsigned long)root->root_key.objectid,
 857                                 root);
 858         if (ret) {
 859                 free_extent_buffer(root->node);
 860                 kfree(root);
 861                 return ERR_PTR(ret);
 862         }
 863         ret = btrfs_find_dead_roots(fs_info->tree_root,
 864                                     root->root_key.objectid, root);
 865         BUG_ON(ret);
 866
 867         return root;
 868 }
 869
 870 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 871                                       struct btrfs_key *location,
 872                                       const char *name, int namelen)
 873 {
 874         struct btrfs_root *root;
 875         int ret;
 876
 877         root = btrfs_read_fs_root_no_name(fs_info, location);
 878         if (!root)
 879                 return NULL;
 880
 881         if (root->in_sysfs)
 882                 return root;
 883
 884         ret = btrfs_set_root_name(root, name, namelen);
 885         if (ret) {
 886                 free_extent_buffer(root->node);
 887                 kfree(root);
 888                 return ERR_PTR(ret);
 889         }
 890
 891         ret = btrfs_sysfs_add_root(root);
 892         if (ret) {
 893                 free_extent_buffer(root->node);
 894                 kfree(root->name);
 895                 kfree(root);
 896                 return ERR_PTR(ret);
 897         }
 898         root->in_sysfs = 1;
 899         return root;
 900 }
 901 #if 0
 902 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 903         struct btrfs_hasher *hasher;
 904
 905         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 906         if (!hasher)
 907                 return -ENOMEM;
 908         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 909         if (!hasher->hash_tfm) {
 910                 kfree(hasher);
 911                 return -EINVAL;
 912         }
 913         spin_lock(&info->hash_lock);
 914         list_add(&hasher->list, &info->hashers);
 915         spin_unlock(&info->hash_lock);
 916         return 0;
 917 }
 918 #endif
 919
 920 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 921 {
 922         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 923         int ret = 0;
 924         struct list_head *cur;
 925         struct btrfs_device *device;
 926         struct backing_dev_info *bdi;
 927
 928         list_for_each(cur, &info->fs_devices->devices) {
 929                 device = list_entry(cur, struct btrfs_device, dev_list);
 930                 bdi = blk_get_backing_dev_info(device->bdev);
 931                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 932                         ret = 1;
 933                         break;
 934                 }
 935         }
 936         return ret;
 937 }
 938
 939 /*
 940  * this unplugs every device on the box, and it is only used when page
 941  * is null
 942  */
 943 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 944 {
 945         struct list_head *cur;
 946         struct btrfs_device *device;
 947         struct btrfs_fs_info *info;
 948
 949         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 950         list_for_each(cur, &info->fs_devices->devices) {
 951                 device = list_entry(cur, struct btrfs_device, dev_list);
 952                 bdi = blk_get_backing_dev_info(device->bdev);
 953                 if (bdi->unplug_io_fn) {
 954                         bdi->unplug_io_fn(bdi, page);
 955                 }
 956         }
 957 }
 958
 959 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 960 {
 961         struct inode *inode;
 962         struct extent_map_tree *em_tree;
 963         struct extent_map *em;
 964         struct address_space *mapping;
 965         u64 offset;
 966
 967         /* the generic O_DIRECT read code does this */
 968         if (!page) {
 969                 __unplug_io_fn(bdi, page);
 970                 return;
 971         }
 972
 973         /*
 974          * page->mapping may change at any time.  Get a consistent copy
 975          * and use that for everything below
 976          */
 977         smp_mb();
 978         mapping = page->mapping;
 979         if (!mapping)
 980                 return;
 981
 982         inode = mapping->host;
 983         offset = page_offset(page);
 984
 985         em_tree = &BTRFS_I(inode)->extent_tree;
 986         spin_lock(&em_tree->lock);
 987         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 988         spin_unlock(&em_tree->lock);
 989         if (!em)
 990                 return;
 991
 992         offset = offset - em->start;
 993         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
 994                           em->block_start + offset, page);
 995         free_extent_map(em);
 996 }
 997
 998 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 999 {
1000 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1001         bdi_init(bdi);
1002 #endif
1003         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1004         bdi->state              = 0;
1005         bdi->capabilities       = default_backing_dev_info.capabilities;
1006         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1007         bdi->unplug_io_data     = info;
1008         bdi->congested_fn       = btrfs_congested_fn;
1009         bdi->congested_data     = info;
1010         return 0;
1011 }
1012
1013 static int bio_ready_for_csum(struct bio *bio)
1014 {
1015         u64 length = 0;
1016         u64 buf_len = 0;
1017         u64 start = 0;
1018         struct page *page;
1019         struct extent_io_tree *io_tree = NULL;
1020         struct btrfs_fs_info *info = NULL;
1021         struct bio_vec *bvec;
1022         int i;
1023         int ret;
1024
1025         bio_for_each_segment(bvec, bio, i) {
1026                 page = bvec->bv_page;
1027                 if (page->private == EXTENT_PAGE_PRIVATE) {
1028                         length += bvec->bv_len;
1029                         continue;
1030                 }
1031                 if (!page->private) {
1032                         length += bvec->bv_len;
1033                         continue;
1034                 }
1035                 length = bvec->bv_len;
1036                 buf_len = page->private >> 2;
1037                 start = page_offset(page) + bvec->bv_offset;
1038                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1039                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1040         }
1041         /* are we fully contained in this bio? */
1042         if (buf_len <= length)
1043                 return 1;
1044
1045         ret = extent_range_uptodate(io_tree, start + length,
1046                                     start + buf_len - 1);
1047         if (ret == 1)
1048                 return ret;
1049         return ret;
1050 }
1051
1052 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1053 static void btrfs_end_io_csum(void *p)
1054 #else
1055 static void btrfs_end_io_csum(struct work_struct *work)
1056 #endif
1057 {
1058 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1059         struct btrfs_fs_info *fs_info = p;
1060 #else
1061         struct btrfs_fs_info *fs_info = container_of(work,
1062                                                      struct btrfs_fs_info,
1063                                                      end_io_work);
1064 #endif
1065         unsigned long flags;
1066         struct end_io_wq *end_io_wq;
1067         struct bio *bio;
1068         struct list_head *next;
1069         int error;
1070         int was_empty;
1071
1072         while(1) {
1073                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1074                 if (list_empty(&fs_info->end_io_work_list)) {
1075                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1076                                                flags);
1077                         return;
1078                 }
1079                 next = fs_info->end_io_work_list.next;
1080                 list_del(next);
1081                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
1082
1083                 end_io_wq = list_entry(next, struct end_io_wq, list);
1084
1085                 bio = end_io_wq->bio;
1086                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1087                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1088                         was_empty = list_empty(&fs_info->end_io_work_list);
1089                         list_add_tail(&end_io_wq->list,
1090                                       &fs_info->end_io_work_list);
1091                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1092                                                flags);
1093                         if (was_empty)
1094                                 return;
1095                         continue;
1096                 }
1097                 error = end_io_wq->error;
1098                 bio->bi_private = end_io_wq->private;
1099                 bio->bi_end_io = end_io_wq->end_io;
1100                 kfree(end_io_wq);
1101 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1102                 bio_endio(bio, bio->bi_size, error);
1103 #else
1104                 bio_endio(bio, error);
1105 #endif
1106         }
1107 }
1108
1109 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1110 static void btrfs_async_submit_work(void *p)
1111 #else
1112 static void btrfs_async_submit_work(struct work_struct *work)
1113 #endif
1114 {
1115 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1116         struct btrfs_fs_info *fs_info = p;
1117 #else
1118         struct btrfs_fs_info *fs_info = container_of(work,
1119                                                      struct btrfs_fs_info,
1120                                                      async_submit_work);
1121 #endif
1122         struct async_submit_bio *async;
1123         struct list_head *next;
1124
1125         while(1) {
1126                 spin_lock(&fs_info->async_submit_work_lock);
1127                 if (list_empty(&fs_info->async_submit_work_list)) {
1128                         spin_unlock(&fs_info->async_submit_work_lock);
1129                         return;
1130                 }
1131                 next = fs_info->async_submit_work_list.next;
1132                 list_del(next);
1133                 spin_unlock(&fs_info->async_submit_work_lock);
1134
1135                 async = list_entry(next, struct async_submit_bio, list);
1136                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1137                                        async->mirror_num);
1138                 kfree(async);
1139         }
1140 }
1141
1142 struct btrfs_root *open_ctree(struct super_block *sb,
1143                               struct btrfs_fs_devices *fs_devices)
1144 {
1145         u32 sectorsize;
1146         u32 nodesize;
1147         u32 leafsize;
1148         u32 blocksize;
1149         u32 stripesize;
1150         struct buffer_head *bh;
1151         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1152                                                  GFP_NOFS);
1153         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1154                                                GFP_NOFS);
1155         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1156                                                 GFP_NOFS);
1157         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1158                                                 GFP_NOFS);
1159         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1160                                               GFP_NOFS);
1161         int ret;
1162         int err = -EINVAL;
1163         struct btrfs_super_block *disk_super;
1164
1165         if (!extent_root || !tree_root || !fs_info) {
1166                 err = -ENOMEM;
1167                 goto fail;
1168         }
1169         end_io_workqueue = create_workqueue("btrfs-end-io");
1170         BUG_ON(!end_io_workqueue);
1171         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1172
1173         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1174         INIT_LIST_HEAD(&fs_info->trans_list);
1175         INIT_LIST_HEAD(&fs_info->dead_roots);
1176         INIT_LIST_HEAD(&fs_info->hashers);
1177         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1178         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1179         spin_lock_init(&fs_info->hash_lock);
1180         spin_lock_init(&fs_info->end_io_work_lock);
1181         spin_lock_init(&fs_info->async_submit_work_lock);
1182         spin_lock_init(&fs_info->delalloc_lock);
1183         spin_lock_init(&fs_info->new_trans_lock);
1184
1185         init_completion(&fs_info->kobj_unregister);
1186         fs_info->tree_root = tree_root;
1187         fs_info->extent_root = extent_root;
1188         fs_info->chunk_root = chunk_root;
1189         fs_info->dev_root = dev_root;
1190         fs_info->fs_devices = fs_devices;
1191         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1192         INIT_LIST_HEAD(&fs_info->space_info);
1193         btrfs_mapping_init(&fs_info->mapping_tree);
1194         fs_info->sb = sb;
1195         fs_info->max_extent = (u64)-1;
1196         fs_info->max_inline = 8192 * 1024;
1197         setup_bdi(fs_info, &fs_info->bdi);
1198         fs_info->btree_inode = new_inode(sb);
1199         fs_info->btree_inode->i_ino = 1;
1200         fs_info->btree_inode->i_nlink = 1;
1201
1202         sb->s_blocksize = 4096;
1203         sb->s_blocksize_bits = blksize_bits(4096);
1204
1205         /*
1206          * we set the i_size on the btree inode to the max possible int.
1207          * the real end of the address space is determined by all of
1208          * the devices in the system
1209          */
1210         fs_info->btree_inode->i_size = OFFSET_MAX;
1211         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1212         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1213
1214         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1215                              fs_info->btree_inode->i_mapping,
1216                              GFP_NOFS);
1217         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1218                              GFP_NOFS);
1219
1220         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1221
1222         extent_io_tree_init(&fs_info->free_space_cache,
1223                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1224         extent_io_tree_init(&fs_info->block_group_cache,
1225                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1226         extent_io_tree_init(&fs_info->pinned_extents,
1227                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1228         extent_io_tree_init(&fs_info->pending_del,
1229                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1230         extent_io_tree_init(&fs_info->extent_ins,
1231                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1232         fs_info->do_barriers = 1;
1233
1234 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1235         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1236         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1237                   fs_info);
1238         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1239 #else
1240         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1241         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1242         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1243 #endif
1244         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1245         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1246                sizeof(struct btrfs_key));
1247         insert_inode_hash(fs_info->btree_inode);
1248         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1249
1250         mutex_init(&fs_info->trans_mutex);
1251         mutex_init(&fs_info->fs_mutex);
1252
1253 #if 0
1254         ret = add_hasher(fs_info, "crc32c");
1255         if (ret) {
1256                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1257                 err = -ENOMEM;
1258                 goto fail_iput;
1259         }
1260 #endif
1261         __setup_root(4096, 4096, 4096, 4096, tree_root,
1262                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1263
1264
1265         bh = __bread(fs_devices->latest_bdev,
1266                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1267         if (!bh)
1268                 goto fail_iput;
1269
1270         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1271         brelse(bh);
1272
1273         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1274
1275         disk_super = &fs_info->super_copy;
1276         if (!btrfs_super_root(disk_super))
1277                 goto fail_sb_buffer;
1278
1279         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1280                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1281                        (unsigned long long)btrfs_super_num_devices(disk_super),
1282                        (unsigned long long)fs_devices->num_devices);
1283                 goto fail_sb_buffer;
1284         }
1285         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1286
1287         nodesize = btrfs_super_nodesize(disk_super);
1288         leafsize = btrfs_super_leafsize(disk_super);
1289         sectorsize = btrfs_super_sectorsize(disk_super);
1290         stripesize = btrfs_super_stripesize(disk_super);
1291         tree_root->nodesize = nodesize;
1292         tree_root->leafsize = leafsize;
1293         tree_root->sectorsize = sectorsize;
1294         tree_root->stripesize = stripesize;
1295
1296         sb->s_blocksize = sectorsize;
1297         sb->s_blocksize_bits = blksize_bits(sectorsize);
1298
1299         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1300                     sizeof(disk_super->magic))) {
1301                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1302                 goto fail_sb_buffer;
1303         }
1304
1305         mutex_lock(&fs_info->fs_mutex);
1306
1307         ret = btrfs_read_sys_array(tree_root);
1308         if (ret) {
1309                 printk("btrfs: failed to read the system array on %s\n",
1310                        sb->s_id);
1311                 goto fail_sys_array;
1312         }
1313
1314         blocksize = btrfs_level_size(tree_root,
1315                                      btrfs_super_chunk_root_level(disk_super));
1316
1317         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1318                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1319
1320         chunk_root->node = read_tree_block(chunk_root,
1321                                            btrfs_super_chunk_root(disk_super),
1322                                            blocksize, 0);
1323         BUG_ON(!chunk_root->node);
1324
1325         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1326                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1327                  BTRFS_UUID_SIZE);
1328
1329         ret = btrfs_read_chunk_tree(chunk_root);
1330         BUG_ON(ret);
1331
1332         blocksize = btrfs_level_size(tree_root,
1333                                      btrfs_super_root_level(disk_super));
1334
1335
1336         tree_root->node = read_tree_block(tree_root,
1337                                           btrfs_super_root(disk_super),
1338                                           blocksize, 0);
1339         if (!tree_root->node)
1340                 goto fail_sb_buffer;
1341
1342
1343         ret = find_and_setup_root(tree_root, fs_info,
1344                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1345         if (ret)
1346                 goto fail_tree_root;
1347         extent_root->track_dirty = 1;
1348
1349         ret = find_and_setup_root(tree_root, fs_info,
1350                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1351         dev_root->track_dirty = 1;
1352
1353         if (ret)
1354                 goto fail_extent_root;
1355
1356         btrfs_read_block_groups(extent_root);
1357
1358         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1359         fs_info->data_alloc_profile = (u64)-1;
1360         fs_info->metadata_alloc_profile = (u64)-1;
1361         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1362
1363         mutex_unlock(&fs_info->fs_mutex);
1364         return tree_root;
1365
1366 fail_extent_root:
1367         free_extent_buffer(extent_root->node);
1368 fail_tree_root:
1369         free_extent_buffer(tree_root->node);
1370 fail_sys_array:
1371         mutex_unlock(&fs_info->fs_mutex);
1372 fail_sb_buffer:
1373         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1374 fail_iput:
1375         iput(fs_info->btree_inode);
1376 fail:
1377         close_all_devices(fs_info);
1378         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1379
1380         kfree(extent_root);
1381         kfree(tree_root);
1382 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1383         bdi_destroy(&fs_info->bdi);
1384 #endif
1385         kfree(fs_info);
1386         return ERR_PTR(err);
1387 }
1388
1389 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1390 {
1391         char b[BDEVNAME_SIZE];
1392
1393         if (uptodate) {
1394                 set_buffer_uptodate(bh);
1395         } else {
1396                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1397                         printk(KERN_WARNING "lost page write due to "
1398                                         "I/O error on %s\n",
1399                                        bdevname(bh->b_bdev, b));
1400                 }
1401                 /* note, we dont' set_buffer_write_io_error because we have
1402                  * our own ways of dealing with the IO errors
1403                  */
1404                 clear_buffer_uptodate(bh);
1405         }
1406         unlock_buffer(bh);
1407         put_bh(bh);
1408 }
1409
1410 int write_all_supers(struct btrfs_root *root)
1411 {
1412         struct list_head *cur;
1413         struct list_head *head = &root->fs_info->fs_devices->devices;
1414         struct btrfs_device *dev;
1415         struct btrfs_super_block *sb;
1416         struct btrfs_dev_item *dev_item;
1417         struct buffer_head *bh;
1418         int ret;
1419         int do_barriers;
1420         int max_errors;
1421         int total_errors = 0;
1422         u32 crc;
1423         u64 flags;
1424
1425         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1426         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1427
1428         sb = &root->fs_info->super_for_commit;
1429         dev_item = &sb->dev_item;
1430         list_for_each(cur, head) {
1431                 dev = list_entry(cur, struct btrfs_device, dev_list);
1432                 btrfs_set_stack_device_type(dev_item, dev->type);
1433                 btrfs_set_stack_device_id(dev_item, dev->devid);
1434                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1435                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1436                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1437                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1438                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1439                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1440                 flags = btrfs_super_flags(sb);
1441                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1442
1443
1444                 crc = ~(u32)0;
1445                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1446                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1447                 btrfs_csum_final(crc, sb->csum);
1448
1449                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1450                               BTRFS_SUPER_INFO_SIZE);
1451
1452                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1453                 dev->pending_io = bh;
1454
1455                 get_bh(bh);
1456                 set_buffer_uptodate(bh);
1457                 lock_buffer(bh);
1458                 bh->b_end_io = btrfs_end_buffer_write_sync;
1459
1460                 if (do_barriers && dev->barriers) {
1461                         ret = submit_bh(WRITE_BARRIER, bh);
1462                         if (ret == -EOPNOTSUPP) {
1463                                 printk("btrfs: disabling barriers on dev %s\n",
1464                                        dev->name);
1465                                 set_buffer_uptodate(bh);
1466                                 dev->barriers = 0;
1467                                 get_bh(bh);
1468                                 lock_buffer(bh);
1469                                 ret = submit_bh(WRITE, bh);
1470                         }
1471                 } else {
1472                         ret = submit_bh(WRITE, bh);
1473                 }
1474                 if (ret)
1475                         total_errors++;
1476         }
1477         if (total_errors > max_errors) {
1478                 printk("btrfs: %d errors while writing supers\n", total_errors);
1479                 BUG();
1480         }
1481         total_errors = 0;
1482
1483         list_for_each(cur, head) {
1484                 dev = list_entry(cur, struct btrfs_device, dev_list);
1485                 BUG_ON(!dev->pending_io);
1486                 bh = dev->pending_io;
1487                 wait_on_buffer(bh);
1488                 if (!buffer_uptodate(dev->pending_io)) {
1489                         if (do_barriers && dev->barriers) {
1490                                 printk("btrfs: disabling barriers on dev %s\n",
1491                                        dev->name);
1492                                 set_buffer_uptodate(bh);
1493                                 get_bh(bh);
1494                                 lock_buffer(bh);
1495                                 dev->barriers = 0;
1496                                 ret = submit_bh(WRITE, bh);
1497                                 BUG_ON(ret);
1498                                 wait_on_buffer(bh);
1499                                 if (!buffer_uptodate(bh))
1500                                         total_errors++;
1501                         } else {
1502                                 total_errors++;
1503                         }
1504
1505                 }
1506                 dev->pending_io = NULL;
1507                 brelse(bh);
1508         }
1509         if (total_errors > max_errors) {
1510                 printk("btrfs: %d errors while writing supers\n", total_errors);
1511                 BUG();
1512         }
1513         return 0;
1514 }
1515
1516 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1517                       *root)
1518 {
1519         int ret;
1520
1521         ret = write_all_supers(root);
1522         return ret;
1523 }
1524
1525 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1526 {
1527         radix_tree_delete(&fs_info->fs_roots_radix,
1528                           (unsigned long)root->root_key.objectid);
1529         if (root->in_sysfs)
1530                 btrfs_sysfs_del_root(root);
1531         if (root->inode)
1532                 iput(root->inode);
1533         if (root->node)
1534                 free_extent_buffer(root->node);
1535         if (root->commit_root)
1536                 free_extent_buffer(root->commit_root);
1537         if (root->name)
1538                 kfree(root->name);
1539         kfree(root);
1540         return 0;
1541 }
1542
1543 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1544 {
1545         int ret;
1546         struct btrfs_root *gang[8];
1547         int i;
1548
1549         while(1) {
1550                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1551                                              (void **)gang, 0,
1552                                              ARRAY_SIZE(gang));
1553                 if (!ret)
1554                         break;
1555                 for (i = 0; i < ret; i++)
1556                         btrfs_free_fs_root(fs_info, gang[i]);
1557         }
1558         return 0;
1559 }
1560
1561 int close_ctree(struct btrfs_root *root)
1562 {
1563         int ret;
1564         struct btrfs_trans_handle *trans;
1565         struct btrfs_fs_info *fs_info = root->fs_info;
1566
1567         fs_info->closing = 1;
1568         btrfs_transaction_flush_work(root);
1569         mutex_lock(&fs_info->fs_mutex);
1570         btrfs_defrag_dirty_roots(root->fs_info);
1571         trans = btrfs_start_transaction(root, 1);
1572         ret = btrfs_commit_transaction(trans, root);
1573         /* run commit again to  drop the original snapshot */
1574         trans = btrfs_start_transaction(root, 1);
1575         btrfs_commit_transaction(trans, root);
1576         ret = btrfs_write_and_wait_transaction(NULL, root);
1577         BUG_ON(ret);
1578
1579         write_ctree_super(NULL, root);
1580         mutex_unlock(&fs_info->fs_mutex);
1581
1582         btrfs_transaction_flush_work(root);
1583
1584         if (fs_info->delalloc_bytes) {
1585                 printk("btrfs: at unmount delalloc count %Lu\n",
1586                        fs_info->delalloc_bytes);
1587         }
1588         if (fs_info->extent_root->node)
1589                 free_extent_buffer(fs_info->extent_root->node);
1590
1591         if (fs_info->tree_root->node)
1592                 free_extent_buffer(fs_info->tree_root->node);
1593
1594         if (root->fs_info->chunk_root->node);
1595                 free_extent_buffer(root->fs_info->chunk_root->node);
1596
1597         if (root->fs_info->dev_root->node);
1598                 free_extent_buffer(root->fs_info->dev_root->node);
1599
1600         btrfs_free_block_groups(root->fs_info);
1601         del_fs_roots(fs_info);
1602
1603         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1604
1605         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1606         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1607         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1608         extent_io_tree_empty_lru(&fs_info->pending_del);
1609         extent_io_tree_empty_lru(&fs_info->extent_ins);
1610         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1611
1612         flush_workqueue(async_submit_workqueue);
1613         flush_workqueue(end_io_workqueue);
1614
1615         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1616
1617         flush_workqueue(async_submit_workqueue);
1618         destroy_workqueue(async_submit_workqueue);
1619
1620         flush_workqueue(end_io_workqueue);
1621         destroy_workqueue(end_io_workqueue);
1622
1623         iput(fs_info->btree_inode);
1624 #if 0
1625         while(!list_empty(&fs_info->hashers)) {
1626                 struct btrfs_hasher *hasher;
1627                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1628                                     hashers);
1629                 list_del(&hasher->hashers);
1630                 crypto_free_hash(&fs_info->hash_tfm);
1631                 kfree(hasher);
1632         }
1633 #endif
1634         close_all_devices(fs_info);
1635         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1636
1637 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1638         bdi_destroy(&fs_info->bdi);
1639 #endif
1640
1641         kfree(fs_info->extent_root);
1642         kfree(fs_info->tree_root);
1643         kfree(fs_info->chunk_root);
1644         kfree(fs_info->dev_root);
1645         return 0;
1646 }
1647
1648 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1649 {
1650         int ret;
1651         struct inode *btree_inode = buf->first_page->mapping->host;
1652
1653         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1654         if (!ret)
1655                 return ret;
1656
1657         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1658                                     parent_transid);
1659         return !ret;
1660 }
1661
1662 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1663 {
1664         struct inode *btree_inode = buf->first_page->mapping->host;
1665         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1666                                           buf);
1667 }
1668
1669 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1670 {
1671         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1672         u64 transid = btrfs_header_generation(buf);
1673         struct inode *btree_inode = root->fs_info->btree_inode;
1674
1675         if (transid != root->fs_info->generation) {
1676                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1677                         (unsigned long long)buf->start,
1678                         transid, root->fs_info->generation);
1679                 WARN_ON(1);
1680         }
1681         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1682 }
1683
1684 void btrfs_throttle(struct btrfs_root *root)
1685 {
1686         struct backing_dev_info *bdi;
1687
1688         bdi = &root->fs_info->bdi;
1689         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1690 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1691                 congestion_wait(WRITE, HZ/20);
1692 #else
1693                 blk_congestion_wait(WRITE, HZ/20);
1694 #endif
1695         }
1696 }
1697
1698 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1699 {
1700         /*
1701          * looks as though older kernels can get into trouble with
1702          * this code, they end up stuck in balance_dirty_pages forever
1703          */
1704 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1705         struct extent_io_tree *tree;
1706         u64 num_dirty;
1707         u64 start = 0;
1708         unsigned long thresh = 16 * 1024 * 1024;
1709         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1710
1711         if (current_is_pdflush())
1712                 return;
1713
1714         num_dirty = count_range_bits(tree, &start, (u64)-1,
1715                                      thresh, EXTENT_DIRTY);
1716         if (num_dirty > thresh) {
1717                 balance_dirty_pages_ratelimited_nr(
1718                                    root->fs_info->btree_inode->i_mapping, 1);
1719         }
1720 #else
1721         return;
1722 #endif
1723 }
1724
1725 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1726 {
1727         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1728         struct inode *btree_inode = root->fs_info->btree_inode;
1729         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1730                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1731 }
1732
1733 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1734 {
1735         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1736         struct inode *btree_inode = root->fs_info->btree_inode;
1737         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1738                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1739                         GFP_NOFS);
1740 }
1741
1742 int btrfs_buffer_defrag(struct extent_buffer *buf)
1743 {
1744         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1745         struct inode *btree_inode = root->fs_info->btree_inode;
1746         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1747                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1748 }
1749
1750 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1751 {
1752         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1753         struct inode *btree_inode = root->fs_info->btree_inode;
1754         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1755                      buf->start, buf->start + buf->len - 1,
1756                      EXTENT_DEFRAG_DONE, 0);
1757 }
1758
1759 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1760 {
1761         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1762         struct inode *btree_inode = root->fs_info->btree_inode;
1763         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1764                      buf->start, buf->start + buf->len - 1,
1765                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1766 }
1767
1768 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1769 {
1770         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1771         struct inode *btree_inode = root->fs_info->btree_inode;
1772         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1773                      buf->start, buf->start + buf->len - 1,
1774                      EXTENT_DEFRAG, GFP_NOFS);
1775 }
1776
1777 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1778 {
1779         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1780         int ret;
1781         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1782         if (ret == 0) {
1783                 buf->flags |= EXTENT_UPTODATE;
1784         }
1785         return ret;
1786 }
1787
1788 static struct extent_io_ops btree_extent_io_ops = {
1789         .writepage_io_hook = btree_writepage_io_hook,
1790         .readpage_end_io_hook = btree_readpage_end_io_hook,
1791         .submit_bio_hook = btree_submit_bio_hook,
1792         /* note we're sharing with inode.c for the merge bio hook */
1793         .merge_bio_hook = btrfs_merge_bio_hook,
1794 };