fs/nilfs2/btnode.c

   1 /*
   2  * btnode.c - NILFS B-tree node cache
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * This file was originally written by Seiji Kihara <kihara@osrg.net>
  21  * and fully revised by Ryusuke Konishi <ryusuke@osrg.net> for
  22  * stabilization and simplification.
  23  *
  24  */
  25
  26 #include <linux/types.h>
  27 #include <linux/buffer_head.h>
  28 #include <linux/mm.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/gfp.h>
  31 #include "nilfs.h"
  32 #include "mdt.h"
  33 #include "dat.h"
  34 #include "page.h"
  35 #include "btnode.h"
  36
  37 void nilfs_btnode_cache_init(struct address_space *btnc,
  38                              struct backing_dev_info *bdi)
  39 {
  40         nilfs_mapping_init(btnc, bdi);
  41 }
  42
  43 void nilfs_btnode_cache_clear(struct address_space *btnc)
  44 {
  45         invalidate_mapping_pages(btnc, 0, -1);
  46         truncate_inode_pages(btnc, 0);
  47 }
  48
  49 struct buffer_head *
  50 nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
  51 {
  52         struct inode *inode = NILFS_BTNC_I(btnc);
  53         struct buffer_head *bh;
  54
  55         bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
  56         if (unlikely(!bh))
  57                 return NULL;
  58
  59         if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
  60                      buffer_dirty(bh))) {
  61                 brelse(bh);
  62                 BUG();
  63         }
  64         memset(bh->b_data, 0, 1 << inode->i_blkbits);
  65         bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
  66         bh->b_blocknr = blocknr;
  67         set_buffer_mapped(bh);
  68         set_buffer_uptodate(bh);
  69
  70         unlock_page(bh->b_page);
  71         page_cache_release(bh->b_page);
  72         return bh;
  73 }
  74
  75 int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
  76                               sector_t pblocknr, int mode,
  77                               struct buffer_head **pbh, sector_t *submit_ptr)
  78 {
  79         struct buffer_head *bh;
  80         struct inode *inode = NILFS_BTNC_I(btnc);
  81         struct page *page;
  82         int err;
  83
  84         bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
  85         if (unlikely(!bh))
  86                 return -ENOMEM;
  87
  88         err = -EEXIST; /* internal code */
  89         page = bh->b_page;
  90
  91         if (buffer_uptodate(bh) || buffer_dirty(bh))
  92                 goto found;
  93
  94         if (pblocknr == 0) {
  95                 pblocknr = blocknr;
  96                 if (inode->i_ino != NILFS_DAT_INO) {
  97                         struct inode *dat = NILFS_I_NILFS(inode)->ns_dat;
  98
  99                         /* blocknr is a virtual block number */
 100                         err = nilfs_dat_translate(dat, blocknr, &pblocknr);
 101                         if (unlikely(err)) {
 102                                 brelse(bh);
 103                                 goto out_locked;
 104                         }
 105                 }
 106         }
 107
 108         if (mode == READA) {
 109                 if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
 110                         err = -EBUSY; /* internal code */
 111                         brelse(bh);
 112                         goto out_locked;
 113                 }
 114         } else { /* mode == READ */
 115                 lock_buffer(bh);
 116         }
 117         if (buffer_uptodate(bh)) {
 118                 unlock_buffer(bh);
 119                 err = -EEXIST; /* internal code */
 120                 goto found;
 121         }
 122         set_buffer_mapped(bh);
 123         bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
 124         bh->b_blocknr = pblocknr; /* set block address for read */
 125         bh->b_end_io = end_buffer_read_sync;
 126         get_bh(bh);
 127         submit_bh(mode, bh);
 128         bh->b_blocknr = blocknr; /* set back to the given block address */
 129         *submit_ptr = pblocknr;
 130         err = 0;
 131 found:
 132         *pbh = bh;
 133
 134 out_locked:
 135         unlock_page(page);
 136         page_cache_release(page);
 137         return err;
 138 }
 139
 140 /**
 141  * nilfs_btnode_delete - delete B-tree node buffer
 142  * @bh: buffer to be deleted
 143  *
 144  * nilfs_btnode_delete() invalidates the specified buffer and delete the page
 145  * including the buffer if the page gets unbusy.
 146  */
 147 void nilfs_btnode_delete(struct buffer_head *bh)
 148 {
 149         struct address_space *mapping;
 150         struct page *page = bh->b_page;
 151         pgoff_t index = page_index(page);
 152         int still_dirty;
 153
 154         page_cache_get(page);
 155         lock_page(page);
 156         wait_on_page_writeback(page);
 157
 158         nilfs_forget_buffer(bh);
 159         still_dirty = PageDirty(page);
 160         mapping = page->mapping;
 161         unlock_page(page);
 162         page_cache_release(page);
 163
 164         if (!still_dirty && mapping)
 165                 invalidate_inode_pages2_range(mapping, index, index);
 166 }
 167
 168 /**
 169  * nilfs_btnode_prepare_change_key
 170  *  prepare to move contents of the block for old key to one of new key.
 171  *  the old buffer will not be removed, but might be reused for new buffer.
 172  *  it might return -ENOMEM because of memory allocation errors,
 173  *  and might return -EIO because of disk read errors.
 174  */
 175 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
 176                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 177 {
 178         struct buffer_head *obh, *nbh;
 179         struct inode *inode = NILFS_BTNC_I(btnc);
 180         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 181         int err;
 182
 183         if (oldkey == newkey)
 184                 return 0;
 185
 186         obh = ctxt->bh;
 187         ctxt->newbh = NULL;
 188
 189         if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
 190                 lock_page(obh->b_page);
 191                 /*
 192                  * We cannot call radix_tree_preload for the kernels older
 193                  * than 2.6.23, because it is not exported for modules.
 194                  */
 195 retry:
 196                 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 197                 if (err)
 198                         goto failed_unlock;
 199                 /* BUG_ON(oldkey != obh->b_page->index); */
 200                 if (unlikely(oldkey != obh->b_page->index))
 201                         NILFS_PAGE_BUG(obh->b_page,
 202                                        "invalid oldkey %lld (newkey=%lld)",
 203                                        (unsigned long long)oldkey,
 204                                        (unsigned long long)newkey);
 205
 206                 spin_lock_irq(&btnc->tree_lock);
 207                 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
 208                 spin_unlock_irq(&btnc->tree_lock);
 209                 /*
 210                  * Note: page->index will not change to newkey until
 211                  * nilfs_btnode_commit_change_key() will be called.
 212                  * To protect the page in intermediate state, the page lock
 213                  * is held.
 214                  */
 215                 radix_tree_preload_end();
 216                 if (!err)
 217                         return 0;
 218                 else if (err != -EEXIST)
 219                         goto failed_unlock;
 220
 221                 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
 222                 if (!err)
 223                         goto retry;
 224                 /* fallback to copy mode */
 225                 unlock_page(obh->b_page);
 226         }
 227
 228         nbh = nilfs_btnode_create_block(btnc, newkey);
 229         if (!nbh)
 230                 return -ENOMEM;
 231
 232         BUG_ON(nbh == obh);
 233         ctxt->newbh = nbh;
 234         return 0;
 235
 236  failed_unlock:
 237         unlock_page(obh->b_page);
 238         return err;
 239 }
 240
 241 /**
 242  * nilfs_btnode_commit_change_key
 243  *  commit the change_key operation prepared by prepare_change_key().
 244  */
 245 void nilfs_btnode_commit_change_key(struct address_space *btnc,
 246                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 247 {
 248         struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
 249         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 250         struct page *opage;
 251
 252         if (oldkey == newkey)
 253                 return;
 254
 255         if (nbh == NULL) {      /* blocksize == pagesize */
 256                 opage = obh->b_page;
 257                 if (unlikely(oldkey != opage->index))
 258                         NILFS_PAGE_BUG(opage,
 259                                        "invalid oldkey %lld (newkey=%lld)",
 260                                        (unsigned long long)oldkey,
 261                                        (unsigned long long)newkey);
 262                 nilfs_btnode_mark_dirty(obh);
 263
 264                 spin_lock_irq(&btnc->tree_lock);
 265                 radix_tree_delete(&btnc->page_tree, oldkey);
 266                 radix_tree_tag_set(&btnc->page_tree, newkey,
 267                                    PAGECACHE_TAG_DIRTY);
 268                 spin_unlock_irq(&btnc->tree_lock);
 269
 270                 opage->index = obh->b_blocknr = newkey;
 271                 unlock_page(opage);
 272         } else {
 273                 nilfs_copy_buffer(nbh, obh);
 274                 nilfs_btnode_mark_dirty(nbh);
 275
 276                 nbh->b_blocknr = newkey;
 277                 ctxt->bh = nbh;
 278                 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
 279         }
 280 }
 281
 282 /**
 283  * nilfs_btnode_abort_change_key
 284  *  abort the change_key operation prepared by prepare_change_key().
 285  */
 286 void nilfs_btnode_abort_change_key(struct address_space *btnc,
 287                                    struct nilfs_btnode_chkey_ctxt *ctxt)
 288 {
 289         struct buffer_head *nbh = ctxt->newbh;
 290         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 291
 292         if (oldkey == newkey)
 293                 return;
 294
 295         if (nbh == NULL) {      /* blocksize == pagesize */
 296                 spin_lock_irq(&btnc->tree_lock);
 297                 radix_tree_delete(&btnc->page_tree, newkey);
 298                 spin_unlock_irq(&btnc->tree_lock);
 299                 unlock_page(ctxt->bh->b_page);
 300         } else
 301                 brelse(nbh);
 302 }