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 "jfs_incore.h"
  26 #include "jfs_filsys.h"
  27 #include "jfs_imap.h"
  28 #include "jfs_extent.h"
  29 #include "jfs_unicode.h"
  30 #include "jfs_debug.h"
  31
  32
  33 extern struct inode_operations jfs_dir_inode_operations;
  34 extern struct inode_operations jfs_file_inode_operations;
  35 extern struct inode_operations jfs_symlink_inode_operations;
  36 extern struct file_operations jfs_dir_operations;
  37 extern struct file_operations jfs_file_operations;
  38 struct address_space_operations jfs_aops;
  39 extern int freeZeroLink(struct inode *);
  40
  41 void jfs_read_inode(struct inode *inode)
  42 {
  43         if (diRead(inode)) {
  44                 make_bad_inode(inode);
  45                 return;
  46         }
  47
  48         if (S_ISREG(inode->i_mode)) {
  49                 inode->i_op = &jfs_file_inode_operations;
  50                 inode->i_fop = &jfs_file_operations;
  51                 inode->i_mapping->a_ops = &jfs_aops;
  52         } else if (S_ISDIR(inode->i_mode)) {
  53                 inode->i_op = &jfs_dir_inode_operations;
  54                 inode->i_fop = &jfs_dir_operations;
  55                 inode->i_mapping->a_ops = &jfs_aops;
  56                 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
  57         } else if (S_ISLNK(inode->i_mode)) {
  58                 if (inode->i_size >= IDATASIZE) {
  59                         inode->i_op = &page_symlink_inode_operations;
  60                         inode->i_mapping->a_ops = &jfs_aops;
  61                 } else
  62                         inode->i_op = &jfs_symlink_inode_operations;
  63         } else {
  64                 inode->i_op = &jfs_file_inode_operations;
  65                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
  66         }
  67 }
  68
  69 /*
  70  * Workhorse of both fsync & write_inode
  71  */
  72 int jfs_commit_inode(struct inode *inode, int wait)
  73 {
  74         int rc = 0;
  75         tid_t tid;
  76         static int noisy = 5;
  77
  78         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
  79
  80         /*
  81          * Don't commit if inode has been committed since last being
  82          * marked dirty, or if it has been deleted.
  83          */
  84         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
  85                 return 0;
  86
  87         if (isReadOnly(inode)) {
  88                 /* kernel allows writes to devices on read-only
  89                  * partitions and may think inode is dirty
  90                  */
  91                 if (!special_file(inode->i_mode) && noisy) {
  92                         jfs_err("jfs_commit_inode(0x%p) called on "
  93                                    "read-only volume", inode);
  94                         jfs_err("Is remount racy?");
  95                         noisy--;
  96                 }
  97                 return 0;
  98         }
  99
 100         tid = txBegin(inode->i_sb, COMMIT_INODE);
 101         down(&JFS_IP(inode)->commit_sem);
 102
 103         /*
 104          * Retest inode state after taking commit_sem
 105          */
 106         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
 107                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
 108
 109         txEnd(tid);
 110         up(&JFS_IP(inode)->commit_sem);
 111         return rc;
 112 }
 113
 114 int jfs_write_inode(struct inode *inode, int wait)
 115 {
 116         if (test_cflag(COMMIT_Nolink, inode))
 117                 return 0;
 118         /*
 119          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
 120          * It has been committed since the last change, but was still
 121          * on the dirty inode list.
 122          */
 123          if (!test_cflag(COMMIT_Dirty, inode)) {
 124                 /* Make sure committed changes hit the disk */
 125                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
 126                 return 0;
 127          }
 128
 129         if (jfs_commit_inode(inode, wait)) {
 130                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
 131                 return -EIO;
 132         } else
 133                 return 0;
 134 }
 135
 136 void jfs_delete_inode(struct inode *inode)
 137 {
 138         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
 139
 140         if (test_cflag(COMMIT_Freewmap, inode))
 141                 freeZeroLink(inode);
 142
 143         diFree(inode);
 144
 145         /*
 146          * Free the inode from the quota allocation.
 147          */
 148         DQUOT_INIT(inode);
 149         DQUOT_FREE_INODE(inode);
 150         DQUOT_DROP(inode);
 151
 152         clear_inode(inode);
 153 }
 154
 155 void jfs_dirty_inode(struct inode *inode)
 156 {
 157         static int noisy = 5;
 158
 159         if (isReadOnly(inode)) {
 160                 if (!special_file(inode->i_mode) && noisy) {
 161                         /* kernel allows writes to devices on read-only
 162                          * partitions and may try to mark inode dirty
 163                          */
 164                         jfs_err("jfs_dirty_inode called on read-only volume");
 165                         jfs_err("Is remount racy?");
 166                         noisy--;
 167                 }
 168                 return;
 169         }
 170
 171         set_cflag(COMMIT_Dirty, inode);
 172 }
 173
 174 static int
 175 jfs_get_blocks(struct inode *ip, sector_t lblock, unsigned long max_blocks,
 176                         struct buffer_head *bh_result, int create)
 177 {
 178         s64 lblock64 = lblock;
 179         int no_size_check = 0;
 180         int rc = 0;
 181         int take_locks;
 182         xad_t xad;
 183         s64 xaddr;
 184         int xflag;
 185         s32 xlen;
 186
 187         /*
 188          * If this is a special inode (imap, dmap) or directory,
 189          * the lock should already be taken
 190          */
 191         take_locks = ((JFS_IP(ip)->fileset != AGGREGATE_I) &&
 192                       !S_ISDIR(ip->i_mode));
 193         /*
 194          * Take appropriate lock on inode
 195          */
 196         if (take_locks) {
 197                 if (create)
 198                         IWRITE_LOCK(ip);
 199                 else
 200                         IREAD_LOCK(ip);
 201         }
 202
 203         /*
 204          * A directory's "data" is the inode index table, but i_size is the
 205          * size of the d-tree, so don't check the offset against i_size
 206          */
 207         if (S_ISDIR(ip->i_mode))
 208                 no_size_check = 1;
 209
 210         if ((no_size_check ||
 211              ((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size)) &&
 212             (xtLookup(ip, lblock64, max_blocks, &xflag, &xaddr, &xlen, no_size_check)
 213              == 0) && xlen) {
 214                 if (xflag & XAD_NOTRECORDED) {
 215                         if (!create)
 216                                 /*
 217                                  * Allocated but not recorded, read treats
 218                                  * this as a hole
 219                                  */
 220                                 goto unlock;
 221 #ifdef _JFS_4K
 222                         XADoffset(&xad, lblock64);
 223                         XADlength(&xad, xlen);
 224                         XADaddress(&xad, xaddr);
 225 #else                           /* _JFS_4K */
 226                         /*
 227                          * As long as block size = 4K, this isn't a problem.
 228                          * We should mark the whole page not ABNR, but how
 229                          * will we know to mark the other blocks BH_New?
 230                          */
 231                         BUG();
 232 #endif                          /* _JFS_4K */
 233                         rc = extRecord(ip, &xad);
 234                         if (rc)
 235                                 goto unlock;
 236                         set_buffer_new(bh_result);
 237                 }
 238
 239                 map_bh(bh_result, ip->i_sb, xaddr);
 240                 bh_result->b_size = xlen << ip->i_blkbits;
 241                 goto unlock;
 242         }
 243         if (!create)
 244                 goto unlock;
 245
 246         /*
 247          * Allocate a new block
 248          */
 249 #ifdef _JFS_4K
 250         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 251                 goto unlock;
 252         rc = extAlloc(ip, max_blocks, lblock64, &xad, FALSE);
 253         if (rc)
 254                 goto unlock;
 255
 256         set_buffer_new(bh_result);
 257         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
 258         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
 259
 260 #else                           /* _JFS_4K */
 261         /*
 262          * We need to do whatever it takes to keep all but the last buffers
 263          * in 4K pages - see jfs_write.c
 264          */
 265         BUG();
 266 #endif                          /* _JFS_4K */
 267
 268       unlock:
 269         /*
 270          * Release lock on inode
 271          */
 272         if (take_locks) {
 273                 if (create)
 274                         IWRITE_UNLOCK(ip);
 275                 else
 276                         IREAD_UNLOCK(ip);
 277         }
 278         return rc;
 279 }
 280
 281 static int jfs_get_block(struct inode *ip, sector_t lblock,
 282                          struct buffer_head *bh_result, int create)
 283 {
 284         return jfs_get_blocks(ip, lblock, 1, bh_result, create);
 285 }
 286
 287 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
 288 {
 289         return block_write_full_page(page, jfs_get_block, wbc);
 290 }
 291
 292 static int jfs_writepages(struct address_space *mapping,
 293                         struct writeback_control *wbc)
 294 {
 295         return mpage_writepages(mapping, wbc, jfs_get_block);
 296 }
 297
 298 static int jfs_readpage(struct file *file, struct page *page)
 299 {
 300         return mpage_readpage(page, jfs_get_block);
 301 }
 302
 303 static int jfs_readpages(struct file *file, struct address_space *mapping,
 304                 struct list_head *pages, unsigned nr_pages)
 305 {
 306         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
 307 }
 308
 309 static int jfs_prepare_write(struct file *file,
 310                              struct page *page, unsigned from, unsigned to)
 311 {
 312         return nobh_prepare_write(page, from, to, jfs_get_block);
 313 }
 314
 315 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
 316 {
 317         return generic_block_bmap(mapping, block, jfs_get_block);
 318 }
 319
 320 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
 321         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
 322 {
 323         struct file *file = iocb->ki_filp;
 324         struct inode *inode = file->f_mapping->host;
 325
 326         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 327                                 offset, nr_segs, jfs_get_blocks, NULL);
 328 }
 329
 330 struct address_space_operations jfs_aops = {
 331         .readpage       = jfs_readpage,
 332         .readpages      = jfs_readpages,
 333         .writepage      = jfs_writepage,
 334         .writepages     = jfs_writepages,
 335         .sync_page      = block_sync_page,
 336         .prepare_write  = jfs_prepare_write,
 337         .commit_write   = nobh_commit_write,
 338         .bmap           = jfs_bmap,
 339         .direct_IO      = jfs_direct_IO,
 340 };
 341
 342 /*
 343  * Guts of jfs_truncate.  Called with locks already held.  Can be called
 344  * with directory for truncating directory index table.
 345  */
 346 void jfs_truncate_nolock(struct inode *ip, loff_t length)
 347 {
 348         loff_t newsize;
 349         tid_t tid;
 350
 351         ASSERT(length >= 0);
 352
 353         if (test_cflag(COMMIT_Nolink, ip)) {
 354                 xtTruncate(0, ip, length, COMMIT_WMAP);
 355                 return;
 356         }
 357
 358         do {
 359                 tid = txBegin(ip->i_sb, 0);
 360
 361                 /*
 362                  * The commit_sem cannot be taken before txBegin.
 363                  * txBegin may block and there is a chance the inode
 364                  * could be marked dirty and need to be committed
 365                  * before txBegin unblocks
 366                  */
 367                 down(&JFS_IP(ip)->commit_sem);
 368
 369                 newsize = xtTruncate(tid, ip, length,
 370                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
 371                 if (newsize < 0) {
 372                         txEnd(tid);
 373                         up(&JFS_IP(ip)->commit_sem);
 374                         break;
 375                 }
 376
 377                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
 378                 mark_inode_dirty(ip);
 379
 380                 txCommit(tid, 1, &ip, 0);
 381                 txEnd(tid);
 382                 up(&JFS_IP(ip)->commit_sem);
 383         } while (newsize > length);     /* Truncate isn't always atomic */
 384 }
 385
 386 void jfs_truncate(struct inode *ip)
 387 {
 388         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
 389
 390         nobh_truncate_page(ip->i_mapping, ip->i_size);
 391
 392         IWRITE_LOCK(ip);
 393         jfs_truncate_nolock(ip, ip->i_size);
 394         IWRITE_UNLOCK(ip);
 395 }