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 38void nilfs_btnode_cache_init_once(struct address_space *btnc) 39{ 40 memset(btnc, 0, sizeof(*btnc)); 41 INIT_RADIX_TREE(&btnc->page_tree, GFP_ATOMIC); 42 spin_lock_init(&btnc->tree_lock); 43 INIT_LIST_HEAD(&btnc->private_list); 44 spin_lock_init(&btnc->private_lock); 45 46 spin_lock_init(&btnc->i_mmap_lock); 47 INIT_RAW_PRIO_TREE_ROOT(&btnc->i_mmap); 48 INIT_LIST_HEAD(&btnc->i_mmap_nonlinear); 49} 50 51static const struct address_space_operations def_btnode_aops = { 52 .sync_page = block_sync_page, 53}; 54 55void nilfs_btnode_cache_init(struct address_space *btnc, 56 struct backing_dev_info *bdi) 57{ 58 btnc->host = NULL; /* can safely set to host inode ? */ 59 btnc->flags = 0; 60 mapping_set_gfp_mask(btnc, GFP_NOFS); 61 btnc->assoc_mapping = NULL; 62 btnc->backing_dev_info = bdi; 63 btnc->a_ops = &def_btnode_aops; 64} 65 66void nilfs_btnode_cache_clear(struct address_space *btnc) 67{ 68 invalidate_mapping_pages(btnc, 0, -1); 69 truncate_inode_pages(btnc, 0); 70} 71 72struct buffer_head * 73nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr) 74{ 75 struct inode *inode = NILFS_BTNC_I(btnc); 76 struct buffer_head *bh; 77 78 bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node); 79 if (unlikely(!bh)) 80 return NULL; 81 82 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) || 83 buffer_dirty(bh))) { 84 brelse(bh); 85 BUG(); 86 } 87 memset(bh->b_data, 0, 1 << inode->i_blkbits); 88 bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev; 89 bh->b_blocknr = blocknr; 90 set_buffer_mapped(bh); 91 set_buffer_uptodate(bh); 92 93 unlock_page(bh->b_page); 94 page_cache_release(bh->b_page); 95 return bh; 96} 97 98int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr, 99 sector_t pblocknr, int mode, 100 struct buffer_head **pbh, sector_t *submit_ptr) 101{ 102 struct buffer_head *bh; 103 struct inode *inode = NILFS_BTNC_I(btnc); 104 struct page *page; 105 int err; 106 107 bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node); 108 if (unlikely(!bh)) 109 return -ENOMEM; 110 111 err = -EEXIST; /* internal code */ 112 page = bh->b_page; 113 114 if (buffer_uptodate(bh) || buffer_dirty(bh)) 115 goto found; 116 117 if (pblocknr == 0) { 118 pblocknr = blocknr; 119 if (inode->i_ino != NILFS_DAT_INO) { 120 struct inode *dat = 121 nilfs_dat_inode(NILFS_I_NILFS(inode)); 122 123 /* blocknr is a virtual block number */ 124 err = nilfs_dat_translate(dat, blocknr, &pblocknr); 125 if (unlikely(err)) { 126 brelse(bh); 127 goto out_locked; 128 } 129 } 130 } 131 132 if (mode == READA) { 133 if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) { 134 err = -EBUSY; /* internal code */ 135 brelse(bh); 136 goto out_locked; 137 } 138 } else { /* mode == READ */ 139 lock_buffer(bh); 140 } 141 if (buffer_uptodate(bh)) { 142 unlock_buffer(bh); 143 err = -EEXIST; /* internal code */ 144 goto found; 145 } 146 set_buffer_mapped(bh); 147 bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev; 148 bh->b_blocknr = pblocknr; /* set block address for read */ 149 bh->b_end_io = end_buffer_read_sync; 150 get_bh(bh); 151 submit_bh(mode, bh); 152 bh->b_blocknr = blocknr; /* set back to the given block address */ 153 *submit_ptr = pblocknr; 154 err = 0; 155found: 156 *pbh = bh; 157 158out_locked: 159 unlock_page(page); 160 page_cache_release(page); 161 return err; 162} 163 164/** 165 * nilfs_btnode_delete - delete B-tree node buffer 166 * @bh: buffer to be deleted 167 * 168 * nilfs_btnode_delete() invalidates the specified buffer and delete the page 169 * including the buffer if the page gets unbusy. 170 */ 171void nilfs_btnode_delete(struct buffer_head *bh) 172{ 173 struct address_space *mapping; 174 struct page *page = bh->b_page; 175 pgoff_t index = page_index(page); 176 int still_dirty; 177 178 page_cache_get(page); 179 lock_page(page); 180 wait_on_page_writeback(page); 181 182 nilfs_forget_buffer(bh); 183 still_dirty = PageDirty(page); 184 mapping = page->mapping; 185 unlock_page(page); 186 page_cache_release(page); 187 188 if (!still_dirty && mapping) 189 invalidate_inode_pages2_range(mapping, index, index); 190} 191 192/** 193 * nilfs_btnode_prepare_change_key 194 * prepare to move contents of the block for old key to one of new key. 195 * the old buffer will not be removed, but might be reused for new buffer. 196 * it might return -ENOMEM because of memory allocation errors, 197 * and might return -EIO because of disk read errors. 198 */ 199int nilfs_btnode_prepare_change_key(struct address_space *btnc, 200 struct nilfs_btnode_chkey_ctxt *ctxt) 201{ 202 struct buffer_head *obh, *nbh; 203 struct inode *inode = NILFS_BTNC_I(btnc); 204 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey; 205 int err; 206 207 if (oldkey == newkey) 208 return 0; 209 210 obh = ctxt->bh; 211 ctxt->newbh = NULL; 212 213 if (inode->i_blkbits == PAGE_CACHE_SHIFT) { 214 lock_page(obh->b_page); 215 /* 216 * We cannot call radix_tree_preload for the kernels older 217 * than 2.6.23, because it is not exported for modules. 218 */ 219retry: 220 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 221 if (err) 222 goto failed_unlock; 223 /* BUG_ON(oldkey != obh->b_page->index); */ 224 if (unlikely(oldkey != obh->b_page->index)) 225 NILFS_PAGE_BUG(obh->b_page, 226 "invalid oldkey %lld (newkey=%lld)", 227 (unsigned long long)oldkey, 228 (unsigned long long)newkey); 229 230 spin_lock_irq(&btnc->tree_lock); 231 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page); 232 spin_unlock_irq(&btnc->tree_lock); 233 /* 234 * Note: page->index will not change to newkey until 235 * nilfs_btnode_commit_change_key() will be called. 236 * To protect the page in intermediate state, the page lock 237 * is held. 238 */ 239 radix_tree_preload_end(); 240 if (!err) 241 return 0; 242 else if (err != -EEXIST) 243 goto failed_unlock; 244 245 err = invalidate_inode_pages2_range(btnc, newkey, newkey); 246 if (!err) 247 goto retry; 248 /* fallback to copy mode */ 249 unlock_page(obh->b_page); 250 } 251 252 nbh = nilfs_btnode_create_block(btnc, newkey); 253 if (!nbh) 254 return -ENOMEM; 255 256 BUG_ON(nbh == obh); 257 ctxt->newbh = nbh; 258 return 0; 259 260 failed_unlock: 261 unlock_page(obh->b_page); 262 return err; 263} 264 265/** 266 * nilfs_btnode_commit_change_key 267 * commit the change_key operation prepared by prepare_change_key(). 268 */ 269void nilfs_btnode_commit_change_key(struct address_space *btnc, 270 struct nilfs_btnode_chkey_ctxt *ctxt) 271{ 272 struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh; 273 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey; 274 struct page *opage; 275 276 if (oldkey == newkey) 277 return; 278 279 if (nbh == NULL) { /* blocksize == pagesize */ 280 opage = obh->b_page; 281 if (unlikely(oldkey != opage->index)) 282 NILFS_PAGE_BUG(opage, 283 "invalid oldkey %lld (newkey=%lld)", 284 (unsigned long long)oldkey, 285 (unsigned long long)newkey); 286 nilfs_btnode_mark_dirty(obh); 287 288 spin_lock_irq(&btnc->tree_lock); 289 radix_tree_delete(&btnc->page_tree, oldkey); 290 radix_tree_tag_set(&btnc->page_tree, newkey, 291 PAGECACHE_TAG_DIRTY); 292 spin_unlock_irq(&btnc->tree_lock); 293 294 opage->index = obh->b_blocknr = newkey; 295 unlock_page(opage); 296 } else { 297 nilfs_copy_buffer(nbh, obh); 298 nilfs_btnode_mark_dirty(nbh); 299 300 nbh->b_blocknr = newkey; 301 ctxt->bh = nbh; 302 nilfs_btnode_delete(obh); /* will decrement bh->b_count */ 303 } 304} 305 306/** 307 * nilfs_btnode_abort_change_key 308 * abort the change_key operation prepared by prepare_change_key(). 309 */ 310void nilfs_btnode_abort_change_key(struct address_space *btnc, 311 struct nilfs_btnode_chkey_ctxt *ctxt) 312{ 313 struct buffer_head *nbh = ctxt->newbh; 314 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey; 315 316 if (oldkey == newkey) 317 return; 318 319 if (nbh == NULL) { /* blocksize == pagesize */ 320 spin_lock_irq(&btnc->tree_lock); 321 radix_tree_delete(&btnc->page_tree, newkey); 322 spin_unlock_irq(&btnc->tree_lock); 323 unlock_page(ctxt->bh->b_page); 324 } else 325 brelse(nbh); 326} 327