SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 void nilfs_btnode_cache_init_once(struct address_space *btnc)
  38 {
  39         memset(btnc, 0, sizeof(*btnc));
  40         INIT_RADIX_TREE(&btnc->page_tree, GFP_ATOMIC);
  41         spin_lock_init(&btnc->tree_lock);
  42         INIT_LIST_HEAD(&btnc->private_list);
  43         spin_lock_init(&btnc->private_lock);
  44
  45         spin_lock_init(&btnc->i_mmap_lock);
  46         INIT_RAW_PRIO_TREE_ROOT(&btnc->i_mmap);
  47         INIT_LIST_HEAD(&btnc->i_mmap_nonlinear);
  48 }
  49
  50 static const struct address_space_operations def_btnode_aops = {
  51         .sync_page              = block_sync_page,
  52 };
  53
  54 void nilfs_btnode_cache_init(struct address_space *btnc,
  55                              struct backing_dev_info *bdi)
  56 {
  57         btnc->host = NULL;  /* can safely set to host inode ? */
  58         btnc->flags = 0;
  59         mapping_set_gfp_mask(btnc, GFP_NOFS);
  60         btnc->assoc_mapping = NULL;
  61         btnc->backing_dev_info = bdi;
  62         btnc->a_ops = &def_btnode_aops;
  63 }
  64
  65 void nilfs_btnode_cache_clear(struct address_space *btnc)
  66 {
  67         invalidate_mapping_pages(btnc, 0, -1);
  68         truncate_inode_pages(btnc, 0);
  69 }
  70
  71 int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
  72                               sector_t pblocknr, struct buffer_head **pbh,
  73                               int newblk)
  74 {
  75         struct buffer_head *bh;
  76         struct inode *inode = NILFS_BTNC_I(btnc);
  77         int err;
  78
  79         bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
  80         if (unlikely(!bh))
  81                 return -ENOMEM;
  82
  83         err = -EEXIST; /* internal code */
  84         if (newblk) {
  85                 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
  86                              buffer_dirty(bh))) {
  87                         brelse(bh);
  88                         BUG();
  89                 }
  90                 memset(bh->b_data, 0, 1 << inode->i_blkbits);
  91                 bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
  92                 bh->b_blocknr = blocknr;
  93                 set_buffer_mapped(bh);
  94                 set_buffer_uptodate(bh);
  95                 goto found;
  96         }
  97
  98         if (buffer_uptodate(bh) || buffer_dirty(bh))
  99                 goto found;
 100
 101         if (pblocknr == 0) {
 102                 pblocknr = blocknr;
 103                 if (inode->i_ino != NILFS_DAT_INO) {
 104                         struct inode *dat =
 105                                 nilfs_dat_inode(NILFS_I_NILFS(inode));
 106
 107                         /* blocknr is a virtual block number */
 108                         err = nilfs_dat_translate(dat, blocknr, &pblocknr);
 109                         if (unlikely(err)) {
 110                                 brelse(bh);
 111                                 goto out_locked;
 112                         }
 113                 }
 114         }
 115         lock_buffer(bh);
 116         if (buffer_uptodate(bh)) {
 117                 unlock_buffer(bh);
 118                 err = -EEXIST; /* internal code */
 119                 goto found;
 120         }
 121         set_buffer_mapped(bh);
 122         bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
 123         bh->b_blocknr = pblocknr; /* set block address for read */
 124         bh->b_end_io = end_buffer_read_sync;
 125         get_bh(bh);
 126         submit_bh(READ, bh);
 127         bh->b_blocknr = blocknr; /* set back to the given block address */
 128         err = 0;
 129 found:
 130         *pbh = bh;
 131
 132 out_locked:
 133         unlock_page(bh->b_page);
 134         page_cache_release(bh->b_page);
 135         return err;
 136 }
 137
 138 int nilfs_btnode_get(struct address_space *btnc, __u64 blocknr,
 139                      sector_t pblocknr, struct buffer_head **pbh, int newblk)
 140 {
 141         struct buffer_head *bh;
 142         int err;
 143
 144         err = nilfs_btnode_submit_block(btnc, blocknr, pblocknr, pbh, newblk);
 145         if (err == -EEXIST) /* internal code (cache hit) */
 146                 return 0;
 147         if (unlikely(err))
 148                 return err;
 149
 150         bh = *pbh;
 151         wait_on_buffer(bh);
 152         if (!buffer_uptodate(bh)) {
 153                 brelse(bh);
 154                 return -EIO;
 155         }
 156         return 0;
 157 }
 158
 159 /**
 160  * nilfs_btnode_delete - delete B-tree node buffer
 161  * @bh: buffer to be deleted
 162  *
 163  * nilfs_btnode_delete() invalidates the specified buffer and delete the page
 164  * including the buffer if the page gets unbusy.
 165  */
 166 void nilfs_btnode_delete(struct buffer_head *bh)
 167 {
 168         struct address_space *mapping;
 169         struct page *page = bh->b_page;
 170         pgoff_t index = page_index(page);
 171         int still_dirty;
 172
 173         page_cache_get(page);
 174         lock_page(page);
 175         wait_on_page_writeback(page);
 176
 177         nilfs_forget_buffer(bh);
 178         still_dirty = PageDirty(page);
 179         mapping = page->mapping;
 180         unlock_page(page);
 181         page_cache_release(page);
 182
 183         if (!still_dirty && mapping)
 184                 invalidate_inode_pages2_range(mapping, index, index);
 185 }
 186
 187 /**
 188  * nilfs_btnode_prepare_change_key
 189  *  prepare to move contents of the block for old key to one of new key.
 190  *  the old buffer will not be removed, but might be reused for new buffer.
 191  *  it might return -ENOMEM because of memory allocation errors,
 192  *  and might return -EIO because of disk read errors.
 193  */
 194 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
 195                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 196 {
 197         struct buffer_head *obh, *nbh;
 198         struct inode *inode = NILFS_BTNC_I(btnc);
 199         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 200         int err;
 201
 202         if (oldkey == newkey)
 203                 return 0;
 204
 205         obh = ctxt->bh;
 206         ctxt->newbh = NULL;
 207
 208         if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
 209                 lock_page(obh->b_page);
 210                 /*
 211                  * We cannot call radix_tree_preload for the kernels older
 212                  * than 2.6.23, because it is not exported for modules.
 213                  */
 214 retry:
 215                 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 216                 if (err)
 217                         goto failed_unlock;
 218                 /* BUG_ON(oldkey != obh->b_page->index); */
 219                 if (unlikely(oldkey != obh->b_page->index))
 220                         NILFS_PAGE_BUG(obh->b_page,
 221                                        "invalid oldkey %lld (newkey=%lld)",
 222                                        (unsigned long long)oldkey,
 223                                        (unsigned long long)newkey);
 224
 225                 spin_lock_irq(&btnc->tree_lock);
 226                 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
 227                 spin_unlock_irq(&btnc->tree_lock);
 228                 /*
 229                  * Note: page->index will not change to newkey until
 230                  * nilfs_btnode_commit_change_key() will be called.
 231                  * To protect the page in intermediate state, the page lock
 232                  * is held.
 233                  */
 234                 radix_tree_preload_end();
 235                 if (!err)
 236                         return 0;
 237                 else if (err != -EEXIST)
 238                         goto failed_unlock;
 239
 240                 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
 241                 if (!err)
 242                         goto retry;
 243                 /* fallback to copy mode */
 244                 unlock_page(obh->b_page);
 245         }
 246
 247         err = nilfs_btnode_get(btnc, newkey, 0, &nbh, 1);
 248         if (likely(!err)) {
 249                 BUG_ON(nbh == obh);
 250                 ctxt->newbh = nbh;
 251         }
 252         return err;
 253
 254  failed_unlock:
 255         unlock_page(obh->b_page);
 256         return err;
 257 }
 258
 259 /**
 260  * nilfs_btnode_commit_change_key
 261  *  commit the change_key operation prepared by prepare_change_key().
 262  */
 263 void nilfs_btnode_commit_change_key(struct address_space *btnc,
 264                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 265 {
 266         struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
 267         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 268         struct page *opage;
 269
 270         if (oldkey == newkey)
 271                 return;
 272
 273         if (nbh == NULL) {      /* blocksize == pagesize */
 274                 opage = obh->b_page;
 275                 if (unlikely(oldkey != opage->index))
 276                         NILFS_PAGE_BUG(opage,
 277                                        "invalid oldkey %lld (newkey=%lld)",
 278                                        (unsigned long long)oldkey,
 279                                        (unsigned long long)newkey);
 280                 nilfs_btnode_mark_dirty(obh);
 281
 282                 spin_lock_irq(&btnc->tree_lock);
 283                 radix_tree_delete(&btnc->page_tree, oldkey);
 284                 radix_tree_tag_set(&btnc->page_tree, newkey,
 285                                    PAGECACHE_TAG_DIRTY);
 286                 spin_unlock_irq(&btnc->tree_lock);
 287
 288                 opage->index = obh->b_blocknr = newkey;
 289                 unlock_page(opage);
 290         } else {
 291                 nilfs_copy_buffer(nbh, obh);
 292                 nilfs_btnode_mark_dirty(nbh);
 293
 294                 nbh->b_blocknr = newkey;
 295                 ctxt->bh = nbh;
 296                 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
 297         }
 298 }
 299
 300 /**
 301  * nilfs_btnode_abort_change_key
 302  *  abort the change_key operation prepared by prepare_change_key().
 303  */
 304 void nilfs_btnode_abort_change_key(struct address_space *btnc,
 305                                    struct nilfs_btnode_chkey_ctxt *ctxt)
 306 {
 307         struct buffer_head *nbh = ctxt->newbh;
 308         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 309
 310         if (oldkey == newkey)
 311                 return;
 312
 313         if (nbh == NULL) {      /* blocksize == pagesize */
 314                 spin_lock_irq(&btnc->tree_lock);
 315                 radix_tree_delete(&btnc->page_tree, newkey);
 316                 spin_unlock_irq(&btnc->tree_lock);
 317                 unlock_page(ctxt->bh->b_page);
 318         } else
 319                 brelse(nbh);
 320 }