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
  38 static const struct address_space_operations def_btnode_aops = {
  39         .sync_page              = block_sync_page,
  40 };
  41
  42 void nilfs_btnode_cache_init(struct address_space *btnc,
  43                              struct backing_dev_info *bdi)
  44 {
  45         nilfs_mapping_init(btnc, bdi, &def_btnode_aops);
  46 }
  47
  48 void nilfs_btnode_cache_clear(struct address_space *btnc)
  49 {
  50         invalidate_mapping_pages(btnc, 0, -1);
  51         truncate_inode_pages(btnc, 0);
  52 }
  53
  54 struct buffer_head *
  55 nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
  56 {
  57         struct inode *inode = NILFS_BTNC_I(btnc);
  58         struct buffer_head *bh;
  59
  60         bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
  61         if (unlikely(!bh))
  62                 return NULL;
  63
  64         if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
  65                      buffer_dirty(bh))) {
  66                 brelse(bh);
  67                 BUG();
  68         }
  69         memset(bh->b_data, 0, 1 << inode->i_blkbits);
  70         bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
  71         bh->b_blocknr = blocknr;
  72         set_buffer_mapped(bh);
  73         set_buffer_uptodate(bh);
  74
  75         unlock_page(bh->b_page);
  76         page_cache_release(bh->b_page);
  77         return bh;
  78 }
  79
  80 int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
  81                               sector_t pblocknr, int mode,
  82                               struct buffer_head **pbh, sector_t *submit_ptr)
  83 {
  84         struct buffer_head *bh;
  85         struct inode *inode = NILFS_BTNC_I(btnc);
  86         struct page *page;
  87         int err;
  88
  89         bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
  90         if (unlikely(!bh))
  91                 return -ENOMEM;
  92
  93         err = -EEXIST; /* internal code */
  94         page = bh->b_page;
  95
  96         if (buffer_uptodate(bh) || buffer_dirty(bh))
  97                 goto found;
  98
  99         if (pblocknr == 0) {
 100                 pblocknr = blocknr;
 101                 if (inode->i_ino != NILFS_DAT_INO) {
 102                         struct inode *dat = NILFS_I_NILFS(inode)->ns_dat;
 103
 104                         /* blocknr is a virtual block number */
 105                         err = nilfs_dat_translate(dat, blocknr, &pblocknr);
 106                         if (unlikely(err)) {
 107                                 brelse(bh);
 108                                 goto out_locked;
 109                         }
 110                 }
 111         }
 112
 113         if (mode == READA) {
 114                 if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
 115                         err = -EBUSY; /* internal code */
 116                         brelse(bh);
 117                         goto out_locked;
 118                 }
 119         } else { /* mode == READ */
 120                 lock_buffer(bh);
 121         }
 122         if (buffer_uptodate(bh)) {
 123                 unlock_buffer(bh);
 124                 err = -EEXIST; /* internal code */
 125                 goto found;
 126         }
 127         set_buffer_mapped(bh);
 128         bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
 129         bh->b_blocknr = pblocknr; /* set block address for read */
 130         bh->b_end_io = end_buffer_read_sync;
 131         get_bh(bh);
 132         submit_bh(mode, bh);
 133         bh->b_blocknr = blocknr; /* set back to the given block address */
 134         *submit_ptr = pblocknr;
 135         err = 0;
 136 found:
 137         *pbh = bh;
 138
 139 out_locked:
 140         unlock_page(page);
 141         page_cache_release(page);
 142         return err;
 143 }
 144
 145 /**
 146  * nilfs_btnode_delete - delete B-tree node buffer
 147  * @bh: buffer to be deleted
 148  *
 149  * nilfs_btnode_delete() invalidates the specified buffer and delete the page
 150  * including the buffer if the page gets unbusy.
 151  */
 152 void nilfs_btnode_delete(struct buffer_head *bh)
 153 {
 154         struct address_space *mapping;
 155         struct page *page = bh->b_page;
 156         pgoff_t index = page_index(page);
 157         int still_dirty;
 158
 159         page_cache_get(page);
 160         lock_page(page);
 161         wait_on_page_writeback(page);
 162
 163         nilfs_forget_buffer(bh);
 164         still_dirty = PageDirty(page);
 165         mapping = page->mapping;
 166         unlock_page(page);
 167         page_cache_release(page);
 168
 169         if (!still_dirty && mapping)
 170                 invalidate_inode_pages2_range(mapping, index, index);
 171 }
 172
 173 /**
 174  * nilfs_btnode_prepare_change_key
 175  *  prepare to move contents of the block for old key to one of new key.
 176  *  the old buffer will not be removed, but might be reused for new buffer.
 177  *  it might return -ENOMEM because of memory allocation errors,
 178  *  and might return -EIO because of disk read errors.
 179  */
 180 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
 181                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 182 {
 183         struct buffer_head *obh, *nbh;
 184         struct inode *inode = NILFS_BTNC_I(btnc);
 185         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 186         int err;
 187
 188         if (oldkey == newkey)
 189                 return 0;
 190
 191         obh = ctxt->bh;
 192         ctxt->newbh = NULL;
 193
 194         if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
 195                 lock_page(obh->b_page);
 196                 /*
 197                  * We cannot call radix_tree_preload for the kernels older
 198                  * than 2.6.23, because it is not exported for modules.
 199                  */
 200 retry:
 201                 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 202                 if (err)
 203                         goto failed_unlock;
 204                 /* BUG_ON(oldkey != obh->b_page->index); */
 205                 if (unlikely(oldkey != obh->b_page->index))
 206                         NILFS_PAGE_BUG(obh->b_page,
 207                                        "invalid oldkey %lld (newkey=%lld)",
 208                                        (unsigned long long)oldkey,
 209                                        (unsigned long long)newkey);
 210
 211                 spin_lock_irq(&btnc->tree_lock);
 212                 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
 213                 spin_unlock_irq(&btnc->tree_lock);
 214                 /*
 215                  * Note: page->index will not change to newkey until
 216                  * nilfs_btnode_commit_change_key() will be called.
 217                  * To protect the page in intermediate state, the page lock
 218                  * is held.
 219                  */
 220                 radix_tree_preload_end();
 221                 if (!err)
 222                         return 0;
 223                 else if (err != -EEXIST)
 224                         goto failed_unlock;
 225
 226                 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
 227                 if (!err)
 228                         goto retry;
 229                 /* fallback to copy mode */
 230                 unlock_page(obh->b_page);
 231         }
 232
 233         nbh = nilfs_btnode_create_block(btnc, newkey);
 234         if (!nbh)
 235                 return -ENOMEM;
 236
 237         BUG_ON(nbh == obh);
 238         ctxt->newbh = nbh;
 239         return 0;
 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 }