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 "nilfs.h"
  31 #include "mdt.h"
  32 #include "dat.h"
  33 #include "page.h"
  34 #include "btnode.h"
  35
  36
  37 static const struct address_space_operations def_btnode_aops = {
  38         .sync_page              = block_sync_page,
  39 };
  40
  41 void nilfs_btnode_cache_init(struct address_space *btnc,
  42                              struct backing_dev_info *bdi)
  43 {
  44         btnc->host = NULL;  /* can safely set to host inode ? */
  45         btnc->flags = 0;
  46         mapping_set_gfp_mask(btnc, GFP_NOFS);
  47         btnc->assoc_mapping = NULL;
  48         btnc->backing_dev_info = bdi;
  49         btnc->a_ops = &def_btnode_aops;
  50 }
  51
  52 void nilfs_btnode_cache_clear(struct address_space *btnc)
  53 {
  54         invalidate_mapping_pages(btnc, 0, -1);
  55         truncate_inode_pages(btnc, 0);
  56 }
  57
  58 int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
  59                               sector_t pblocknr, struct buffer_head **pbh,
  60                               int newblk)
  61 {
  62         struct buffer_head *bh;
  63         struct inode *inode = NILFS_BTNC_I(btnc);
  64         int err;
  65
  66         bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
  67         if (unlikely(!bh))
  68                 return -ENOMEM;
  69
  70         err = -EEXIST; /* internal code */
  71         if (newblk) {
  72                 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
  73                              buffer_dirty(bh))) {
  74                         brelse(bh);
  75                         BUG();
  76                 }
  77                 memset(bh->b_data, 0, 1 << inode->i_blkbits);
  78                 bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
  79                 bh->b_blocknr = blocknr;
  80                 set_buffer_mapped(bh);
  81                 set_buffer_uptodate(bh);
  82                 goto found;
  83         }
  84
  85         if (buffer_uptodate(bh) || buffer_dirty(bh))
  86                 goto found;
  87
  88         if (pblocknr == 0) {
  89                 pblocknr = blocknr;
  90                 if (inode->i_ino != NILFS_DAT_INO) {
  91                         struct inode *dat =
  92                                 nilfs_dat_inode(NILFS_I_NILFS(inode));
  93
  94                         /* blocknr is a virtual block number */
  95                         err = nilfs_dat_translate(dat, blocknr, &pblocknr);
  96                         if (unlikely(err)) {
  97                                 brelse(bh);
  98                                 goto out_locked;
  99                         }
 100                 }
 101         }
 102         lock_buffer(bh);
 103         if (buffer_uptodate(bh)) {
 104                 unlock_buffer(bh);
 105                 err = -EEXIST; /* internal code */
 106                 goto found;
 107         }
 108         set_buffer_mapped(bh);
 109         bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
 110         bh->b_blocknr = pblocknr; /* set block address for read */
 111         bh->b_end_io = end_buffer_read_sync;
 112         get_bh(bh);
 113         submit_bh(READ, bh);
 114         bh->b_blocknr = blocknr; /* set back to the given block address */
 115         err = 0;
 116 found:
 117         *pbh = bh;
 118
 119 out_locked:
 120         unlock_page(bh->b_page);
 121         page_cache_release(bh->b_page);
 122         return err;
 123 }
 124
 125 int nilfs_btnode_get(struct address_space *btnc, __u64 blocknr,
 126                      sector_t pblocknr, struct buffer_head **pbh, int newblk)
 127 {
 128         struct buffer_head *bh;
 129         int err;
 130
 131         err = nilfs_btnode_submit_block(btnc, blocknr, pblocknr, pbh, newblk);
 132         if (err == -EEXIST) /* internal code (cache hit) */
 133                 return 0;
 134         if (unlikely(err))
 135                 return err;
 136
 137         bh = *pbh;
 138         wait_on_buffer(bh);
 139         if (!buffer_uptodate(bh)) {
 140                 brelse(bh);
 141                 return -EIO;
 142         }
 143         return 0;
 144 }
 145
 146 /**
 147  * nilfs_btnode_delete - delete B-tree node buffer
 148  * @bh: buffer to be deleted
 149  *
 150  * nilfs_btnode_delete() invalidates the specified buffer and delete the page
 151  * including the buffer if the page gets unbusy.
 152  */
 153 void nilfs_btnode_delete(struct buffer_head *bh)
 154 {
 155         struct address_space *mapping;
 156         struct page *page = bh->b_page;
 157         pgoff_t index = page_index(page);
 158         int still_dirty;
 159
 160         page_cache_get(page);
 161         lock_page(page);
 162         wait_on_page_writeback(page);
 163
 164         nilfs_forget_buffer(bh);
 165         still_dirty = PageDirty(page);
 166         mapping = page->mapping;
 167         unlock_page(page);
 168         page_cache_release(page);
 169
 170         if (!still_dirty && mapping)
 171                 invalidate_inode_pages2_range(mapping, index, index);
 172 }
 173
 174 /**
 175  * nilfs_btnode_prepare_change_key
 176  *  prepare to move contents of the block for old key to one of new key.
 177  *  the old buffer will not be removed, but might be reused for new buffer.
 178  *  it might return -ENOMEM because of memory allocation errors,
 179  *  and might return -EIO because of disk read errors.
 180  */
 181 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
 182                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 183 {
 184         struct buffer_head *obh, *nbh;
 185         struct inode *inode = NILFS_BTNC_I(btnc);
 186         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 187         int err;
 188
 189         if (oldkey == newkey)
 190                 return 0;
 191
 192         obh = ctxt->bh;
 193         ctxt->newbh = NULL;
 194
 195         if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
 196                 lock_page(obh->b_page);
 197                 /*
 198                  * We cannot call radix_tree_preload for the kernels older
 199                  * than 2.6.23, because it is not exported for modules.
 200                  */
 201 retry:
 202                 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 203                 if (err)
 204                         goto failed_unlock;
 205                 /* BUG_ON(oldkey != obh->b_page->index); */
 206                 if (unlikely(oldkey != obh->b_page->index))
 207                         NILFS_PAGE_BUG(obh->b_page,
 208                                        "invalid oldkey %lld (newkey=%lld)",
 209                                        (unsigned long long)oldkey,
 210                                        (unsigned long long)newkey);
 211
 212                 spin_lock_irq(&btnc->tree_lock);
 213                 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
 214                 spin_unlock_irq(&btnc->tree_lock);
 215                 /*
 216                  * Note: page->index will not change to newkey until
 217                  * nilfs_btnode_commit_change_key() will be called.
 218                  * To protect the page in intermediate state, the page lock
 219                  * is held.
 220                  */
 221                 radix_tree_preload_end();
 222                 if (!err)
 223                         return 0;
 224                 else if (err != -EEXIST)
 225                         goto failed_unlock;
 226
 227                 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
 228                 if (!err)
 229                         goto retry;
 230                 /* fallback to copy mode */
 231                 unlock_page(obh->b_page);
 232         }
 233
 234         err = nilfs_btnode_get(btnc, newkey, 0, &nbh, 1);
 235         if (likely(!err)) {
 236                 BUG_ON(nbh == obh);
 237                 ctxt->newbh = nbh;
 238         }
 239         return err;
 240
 241  failed_unlock:
 242         unlock_page(obh->b_page);
 243         return err;
 244 }
 245
 246 /**
 247  * nilfs_btnode_commit_change_key
 248  *  commit the change_key operation prepared by prepare_change_key().
 249  */
 250 void nilfs_btnode_commit_change_key(struct address_space *btnc,
 251                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 252 {
 253         struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
 254         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 255         struct page *opage;
 256
 257         if (oldkey == newkey)
 258                 return;
 259
 260         if (nbh == NULL) {      /* blocksize == pagesize */
 261                 opage = obh->b_page;
 262                 if (unlikely(oldkey != opage->index))
 263                         NILFS_PAGE_BUG(opage,
 264                                        "invalid oldkey %lld (newkey=%lld)",
 265                                        (unsigned long long)oldkey,
 266                                        (unsigned long long)newkey);
 267                 nilfs_btnode_mark_dirty(obh);
 268
 269                 spin_lock_irq(&btnc->tree_lock);
 270                 radix_tree_delete(&btnc->page_tree, oldkey);
 271                 radix_tree_tag_set(&btnc->page_tree, newkey,
 272                                    PAGECACHE_TAG_DIRTY);
 273                 spin_unlock_irq(&btnc->tree_lock);
 274
 275                 opage->index = obh->b_blocknr = newkey;
 276                 unlock_page(opage);
 277         } else {
 278                 nilfs_copy_buffer(nbh, obh);
 279                 nilfs_btnode_mark_dirty(nbh);
 280
 281                 nbh->b_blocknr = newkey;
 282                 ctxt->bh = nbh;
 283                 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
 284         }
 285 }
 286
 287 /**
 288  * nilfs_btnode_abort_change_key
 289  *  abort the change_key operation prepared by prepare_change_key().
 290  */
 291 void nilfs_btnode_abort_change_key(struct address_space *btnc,
 292                                    struct nilfs_btnode_chkey_ctxt *ctxt)
 293 {
 294         struct buffer_head *nbh = ctxt->newbh;
 295         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 296
 297         if (oldkey == newkey)
 298                 return;
 299
 300         if (nbh == NULL) {      /* blocksize == pagesize */
 301                 spin_lock_irq(&btnc->tree_lock);
 302                 radix_tree_delete(&btnc->page_tree, newkey);
 303                 spin_unlock_irq(&btnc->tree_lock);
 304                 unlock_page(ctxt->bh->b_page);
 305         } else
 306                 brelse(nbh);
 307 }