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