fs/jfs/inode.c

   1 /*
   2  *   Copyright (C) International Business Machines Corp., 2000-2004
   3  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
   4  *
   5  *   This program is free software;  you can redistribute it and/or modify
   6  *   it under the terms of the GNU General Public License as published by
   7  *   the Free Software Foundation; either version 2 of the License, or
   8  *   (at your option) any later version.
   9  *
  10  *   This program is distributed in the hope that it will be useful,
  11  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
  12  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  13  *   the GNU General Public License for more details.
  14  *
  15  *   You should have received a copy of the GNU General Public License
  16  *   along with this program;  if not, write to the Free Software
  17  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  18  */
  19
  20 #include <linux/fs.h>
  21 #include <linux/mpage.h>
  22 #include <linux/buffer_head.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/quotaops.h>
  25 #include <linux/vs_dlimit.h>
  26 #include "jfs_incore.h"
  27 #include "jfs_inode.h"
  28 #include "jfs_filsys.h"
  29 #include "jfs_imap.h"
  30 #include "jfs_extent.h"
  31 #include "jfs_unicode.h"
  32 #include "jfs_debug.h"
  33
  34
  35 void jfs_read_inode(struct inode *inode)
  36 {
  37         if (diRead(inode)) {
  38                 make_bad_inode(inode);
  39                 return;
  40         }
  41
  42         if (S_ISREG(inode->i_mode)) {
  43                 inode->i_op = &jfs_file_inode_operations;
  44                 inode->i_fop = &jfs_file_operations;
  45                 inode->i_mapping->a_ops = &jfs_aops;
  46         } else if (S_ISDIR(inode->i_mode)) {
  47                 inode->i_op = &jfs_dir_inode_operations;
  48                 inode->i_fop = &jfs_dir_operations;
  49         } else if (S_ISLNK(inode->i_mode)) {
  50                 if (inode->i_size >= IDATASIZE) {
  51                         inode->i_op = &page_symlink_inode_operations;
  52                         inode->i_mapping->a_ops = &jfs_aops;
  53                 } else
  54                         inode->i_op = &jfs_symlink_inode_operations;
  55         } else {
  56                 inode->i_op = &jfs_file_inode_operations;
  57                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
  58         }
  59         jfs_set_inode_flags(inode);
  60 }
  61
  62 /*
  63  * Workhorse of both fsync & write_inode
  64  */
  65 int jfs_commit_inode(struct inode *inode, int wait)
  66 {
  67         int rc = 0;
  68         tid_t tid;
  69         static int noisy = 5;
  70
  71         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
  72
  73         /*
  74          * Don't commit if inode has been committed since last being
  75          * marked dirty, or if it has been deleted.
  76          */
  77         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
  78                 return 0;
  79
  80         if (isReadOnly(inode)) {
  81                 /* kernel allows writes to devices on read-only
  82                  * partitions and may think inode is dirty
  83                  */
  84                 if (!special_file(inode->i_mode) && noisy) {
  85                         jfs_err("jfs_commit_inode(0x%p) called on "
  86                                    "read-only volume", inode);
  87                         jfs_err("Is remount racy?");
  88                         noisy--;
  89                 }
  90                 return 0;
  91         }
  92
  93         tid = txBegin(inode->i_sb, COMMIT_INODE);
  94         mutex_lock(&JFS_IP(inode)->commit_mutex);
  95
  96         /*
  97          * Retest inode state after taking commit_mutex
  98          */
  99         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
 100                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
 101
 102         txEnd(tid);
 103         mutex_unlock(&JFS_IP(inode)->commit_mutex);
 104         return rc;
 105 }
 106
 107 int jfs_write_inode(struct inode *inode, int wait)
 108 {
 109         if (test_cflag(COMMIT_Nolink, inode))
 110                 return 0;
 111         /*
 112          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
 113          * It has been committed since the last change, but was still
 114          * on the dirty inode list.
 115          */
 116          if (!test_cflag(COMMIT_Dirty, inode)) {
 117                 /* Make sure committed changes hit the disk */
 118                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
 119                 return 0;
 120          }
 121
 122         if (jfs_commit_inode(inode, wait)) {
 123                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
 124                 return -EIO;
 125         } else
 126                 return 0;
 127 }
 128
 129 void jfs_delete_inode(struct inode *inode)
 130 {
 131         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
 132
 133         if (!is_bad_inode(inode) &&
 134             (JFS_IP(inode)->fileset == FILESYSTEM_I)) {
 135                 truncate_inode_pages(&inode->i_data, 0);
 136
 137                 if (test_cflag(COMMIT_Freewmap, inode))
 138                         jfs_free_zero_link(inode);
 139
 140                 diFree(inode);
 141
 142                 /*
 143                  * Free the inode from the quota allocation.
 144                  */
 145                 DQUOT_INIT(inode);
 146                 DQUOT_FREE_INODE(inode);
 147                 DQUOT_DROP(inode);
 148                 DLIMIT_FREE_INODE(inode);
 149         }
 150
 151         clear_inode(inode);
 152 }
 153
 154 void jfs_dirty_inode(struct inode *inode)
 155 {
 156         static int noisy = 5;
 157
 158         if (isReadOnly(inode)) {
 159                 if (!special_file(inode->i_mode) && noisy) {
 160                         /* kernel allows writes to devices on read-only
 161                          * partitions and may try to mark inode dirty
 162                          */
 163                         jfs_err("jfs_dirty_inode called on read-only volume");
 164                         jfs_err("Is remount racy?");
 165                         noisy--;
 166                 }
 167                 return;
 168         }
 169
 170         set_cflag(COMMIT_Dirty, inode);
 171 }
 172
 173 int jfs_get_block(struct inode *ip, sector_t lblock,
 174                   struct buffer_head *bh_result, int create)
 175 {
 176         s64 lblock64 = lblock;
 177         int rc = 0;
 178         xad_t xad;
 179         s64 xaddr;
 180         int xflag;
 181         s32 xlen = bh_result->b_size >> ip->i_blkbits;
 182
 183         /*
 184          * Take appropriate lock on inode
 185          */
 186         if (create)
 187                 IWRITE_LOCK(ip);
 188         else
 189                 IREAD_LOCK(ip);
 190
 191         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
 192             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
 193             xaddr) {
 194                 if (xflag & XAD_NOTRECORDED) {
 195                         if (!create)
 196                                 /*
 197                                  * Allocated but not recorded, read treats
 198                                  * this as a hole
 199                                  */
 200                                 goto unlock;
 201 #ifdef _JFS_4K
 202                         XADoffset(&xad, lblock64);
 203                         XADlength(&xad, xlen);
 204                         XADaddress(&xad, xaddr);
 205 #else                           /* _JFS_4K */
 206                         /*
 207                          * As long as block size = 4K, this isn't a problem.
 208                          * We should mark the whole page not ABNR, but how
 209                          * will we know to mark the other blocks BH_New?
 210                          */
 211                         BUG();
 212 #endif                          /* _JFS_4K */
 213                         rc = extRecord(ip, &xad);
 214                         if (rc)
 215                                 goto unlock;
 216                         set_buffer_new(bh_result);
 217                 }
 218
 219                 map_bh(bh_result, ip->i_sb, xaddr);
 220                 bh_result->b_size = xlen << ip->i_blkbits;
 221                 goto unlock;
 222         }
 223         if (!create)
 224                 goto unlock;
 225
 226         /*
 227          * Allocate a new block
 228          */
 229 #ifdef _JFS_4K
 230         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 231                 goto unlock;
 232         rc = extAlloc(ip, xlen, lblock64, &xad, FALSE);
 233         if (rc)
 234                 goto unlock;
 235
 236         set_buffer_new(bh_result);
 237         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
 238         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
 239
 240 #else                           /* _JFS_4K */
 241         /*
 242          * We need to do whatever it takes to keep all but the last buffers
 243          * in 4K pages - see jfs_write.c
 244          */
 245         BUG();
 246 #endif                          /* _JFS_4K */
 247
 248       unlock:
 249         /*
 250          * Release lock on inode
 251          */
 252         if (create)
 253                 IWRITE_UNLOCK(ip);
 254         else
 255                 IREAD_UNLOCK(ip);
 256         return rc;
 257 }
 258
 259 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
 260 {
 261         return nobh_writepage(page, jfs_get_block, wbc);
 262 }
 263
 264 static int jfs_writepages(struct address_space *mapping,
 265                         struct writeback_control *wbc)
 266 {
 267         return mpage_writepages(mapping, wbc, jfs_get_block);
 268 }
 269
 270 static int jfs_readpage(struct file *file, struct page *page)
 271 {
 272         return mpage_readpage(page, jfs_get_block);
 273 }
 274
 275 static int jfs_readpages(struct file *file, struct address_space *mapping,
 276                 struct list_head *pages, unsigned nr_pages)
 277 {
 278         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
 279 }
 280
 281 static int jfs_prepare_write(struct file *file,
 282                              struct page *page, unsigned from, unsigned to)
 283 {
 284         return nobh_prepare_write(page, from, to, jfs_get_block);
 285 }
 286
 287 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
 288 {
 289         return generic_block_bmap(mapping, block, jfs_get_block);
 290 }
 291
 292 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
 293         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
 294 {
 295         struct file *file = iocb->ki_filp;
 296         struct inode *inode = file->f_mapping->host;
 297
 298         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 299                                 offset, nr_segs, jfs_get_block, NULL);
 300 }
 301
 302 const struct address_space_operations jfs_aops = {
 303         .readpage       = jfs_readpage,
 304         .readpages      = jfs_readpages,
 305         .writepage      = jfs_writepage,
 306         .writepages     = jfs_writepages,
 307         .sync_page      = block_sync_page,
 308         .prepare_write  = jfs_prepare_write,
 309         .commit_write   = nobh_commit_write,
 310         .bmap           = jfs_bmap,
 311         .direct_IO      = jfs_direct_IO,
 312 };
 313
 314 /*
 315  * Guts of jfs_truncate.  Called with locks already held.  Can be called
 316  * with directory for truncating directory index table.
 317  */
 318 void jfs_truncate_nolock(struct inode *ip, loff_t length)
 319 {
 320         loff_t newsize;
 321         tid_t tid;
 322
 323         ASSERT(length >= 0);
 324
 325         if (test_cflag(COMMIT_Nolink, ip)) {
 326                 xtTruncate(0, ip, length, COMMIT_WMAP);
 327                 return;
 328         }
 329
 330         do {
 331                 tid = txBegin(ip->i_sb, 0);
 332
 333                 /*
 334                  * The commit_mutex cannot be taken before txBegin.
 335                  * txBegin may block and there is a chance the inode
 336                  * could be marked dirty and need to be committed
 337                  * before txBegin unblocks
 338                  */
 339                 mutex_lock(&JFS_IP(ip)->commit_mutex);
 340
 341                 newsize = xtTruncate(tid, ip, length,
 342                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
 343                 if (newsize < 0) {
 344                         txEnd(tid);
 345                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
 346                         break;
 347                 }
 348
 349                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
 350                 mark_inode_dirty(ip);
 351
 352                 txCommit(tid, 1, &ip, 0);
 353                 txEnd(tid);
 354                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
 355         } while (newsize > length);     /* Truncate isn't always atomic */
 356 }
 357
 358 void jfs_truncate(struct inode *ip)
 359 {
 360         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
 361
 362         nobh_truncate_page(ip->i_mapping, ip->i_size);
 363
 364         IWRITE_LOCK(ip);
 365         jfs_truncate_nolock(ip, ip->i_size);
 366         IWRITE_UNLOCK(ip);
 367 }