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                 bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
  91                 bh->b_blocknr = blocknr;
  92                 set_buffer_mapped(bh);
  93                 set_buffer_uptodate(bh);
  94                 goto found;
  95         }
  96
  97         if (buffer_uptodate(bh) || buffer_dirty(bh))
  98                 goto found;
  99
 100         if (pblocknr == 0) {
 101                 pblocknr = blocknr;
 102                 if (inode->i_ino != NILFS_DAT_INO) {
 103                         struct inode *dat =
 104                                 nilfs_dat_inode(NILFS_I_NILFS(inode));
 105
 106                         /* blocknr is a virtual block number */
 107                         err = nilfs_dat_translate(dat, blocknr, &pblocknr);
 108                         if (unlikely(err)) {
 109                                 brelse(bh);
 110                                 goto out_locked;
 111                         }
 112                 }
 113         }
 114         lock_buffer(bh);
 115         if (buffer_uptodate(bh)) {
 116                 unlock_buffer(bh);
 117                 err = -EEXIST; /* internal code */
 118                 goto found;
 119         }
 120         set_buffer_mapped(bh);
 121         bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
 122         bh->b_blocknr = pblocknr; /* set block address for read */
 123         bh->b_end_io = end_buffer_read_sync;
 124         get_bh(bh);
 125         submit_bh(READ, bh);
 126         bh->b_blocknr = blocknr; /* set back to the given block address */
 127         err = 0;
 128 found:
 129         *pbh = bh;
 130
 131 out_locked:
 132         unlock_page(bh->b_page);
 133         page_cache_release(bh->b_page);
 134         return err;
 135 }
 136
 137 int nilfs_btnode_get(struct address_space *btnc, __u64 blocknr,
 138                      sector_t pblocknr, struct buffer_head **pbh, int newblk)
 139 {
 140         struct buffer_head *bh;
 141         int err;
 142
 143         err = nilfs_btnode_submit_block(btnc, blocknr, pblocknr, pbh, newblk);
 144         if (err == -EEXIST) /* internal code (cache hit) */
 145                 return 0;
 146         if (unlikely(err))
 147                 return err;
 148
 149         bh = *pbh;
 150         wait_on_buffer(bh);
 151         if (!buffer_uptodate(bh)) {
 152                 brelse(bh);
 153                 return -EIO;
 154         }
 155         return 0;
 156 }
 157
 158 /**
 159  * nilfs_btnode_delete - delete B-tree node buffer
 160  * @bh: buffer to be deleted
 161  *
 162  * nilfs_btnode_delete() invalidates the specified buffer and delete the page
 163  * including the buffer if the page gets unbusy.
 164  */
 165 void nilfs_btnode_delete(struct buffer_head *bh)
 166 {
 167         struct address_space *mapping;
 168         struct page *page = bh->b_page;
 169         pgoff_t index = page_index(page);
 170         int still_dirty;
 171
 172         page_cache_get(page);
 173         lock_page(page);
 174         wait_on_page_writeback(page);
 175
 176         nilfs_forget_buffer(bh);
 177         still_dirty = PageDirty(page);
 178         mapping = page->mapping;
 179         unlock_page(page);
 180         page_cache_release(page);
 181
 182         if (!still_dirty && mapping)
 183                 invalidate_inode_pages2_range(mapping, index, index);
 184 }
 185
 186 /**
 187  * nilfs_btnode_prepare_change_key
 188  *  prepare to move contents of the block for old key to one of new key.
 189  *  the old buffer will not be removed, but might be reused for new buffer.
 190  *  it might return -ENOMEM because of memory allocation errors,
 191  *  and might return -EIO because of disk read errors.
 192  */
 193 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
 194                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 195 {
 196         struct buffer_head *obh, *nbh;
 197         struct inode *inode = NILFS_BTNC_I(btnc);
 198         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 199         int err;
 200
 201         if (oldkey == newkey)
 202                 return 0;
 203
 204         obh = ctxt->bh;
 205         ctxt->newbh = NULL;
 206
 207         if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
 208                 lock_page(obh->b_page);
 209                 /*
 210                  * We cannot call radix_tree_preload for the kernels older
 211                  * than 2.6.23, because it is not exported for modules.
 212                  */
 213 retry:
 214                 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 215                 if (err)
 216                         goto failed_unlock;
 217                 /* BUG_ON(oldkey != obh->b_page->index); */
 218                 if (unlikely(oldkey != obh->b_page->index))
 219                         NILFS_PAGE_BUG(obh->b_page,
 220                                        "invalid oldkey %lld (newkey=%lld)",
 221                                        (unsigned long long)oldkey,
 222                                        (unsigned long long)newkey);
 223
 224                 spin_lock_irq(&btnc->tree_lock);
 225                 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
 226                 spin_unlock_irq(&btnc->tree_lock);
 227                 /*
 228                  * Note: page->index will not change to newkey until
 229                  * nilfs_btnode_commit_change_key() will be called.
 230                  * To protect the page in intermediate state, the page lock
 231                  * is held.
 232                  */
 233                 radix_tree_preload_end();
 234                 if (!err)
 235                         return 0;
 236                 else if (err != -EEXIST)
 237                         goto failed_unlock;
 238
 239                 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
 240                 if (!err)
 241                         goto retry;
 242                 /* fallback to copy mode */
 243                 unlock_page(obh->b_page);
 244         }
 245
 246         err = nilfs_btnode_get(btnc, newkey, 0, &nbh, 1);
 247         if (likely(!err)) {
 248                 BUG_ON(nbh == obh);
 249                 ctxt->newbh = nbh;
 250         }
 251         return err;
 252
 253  failed_unlock:
 254         unlock_page(obh->b_page);
 255         return err;
 256 }
 257
 258 /**
 259  * nilfs_btnode_commit_change_key
 260  *  commit the change_key operation prepared by prepare_change_key().
 261  */
 262 void nilfs_btnode_commit_change_key(struct address_space *btnc,
 263                                     struct nilfs_btnode_chkey_ctxt *ctxt)
 264 {
 265         struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
 266         __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
 267         struct page *opage;
 268
 269         if (oldkey == newkey)
 270                 return;
 271
 272         if (nbh == NULL) {      /* blocksize == pagesize */
 273                 opage = obh->b_page;
 274                 if (unlikely(oldkey != opage->index))
 275                         NILFS_PAGE_BUG(opage,
 276                                        "invalid oldkey %lld (newkey=%lld)",
 277                                        (unsigned long long)oldkey,
 278                                        (unsigned long long)newkey);
 279                 if (!test_set_buffer_dirty(obh) && TestSetPageDirty(opage))
 280                         BUG();
 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 }