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