fs/jfs/inode.c

   1 /*
   2  *   Copyright (c) International Business Machines Corp., 2000-2002
   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 "jfs_incore.h"
  25 #include "jfs_filsys.h"
  26 #include "jfs_imap.h"
  27 #include "jfs_extent.h"
  28 #include "jfs_unicode.h"
  29 #include "jfs_debug.h"
  30
  31
  32 extern struct inode_operations jfs_dir_inode_operations;
  33 extern struct inode_operations jfs_file_inode_operations;
  34 extern struct inode_operations jfs_symlink_inode_operations;
  35 extern struct file_operations jfs_dir_operations;
  36 extern struct file_operations jfs_file_operations;
  37 struct address_space_operations jfs_aops;
  38 extern int freeZeroLink(struct inode *);
  39
  40 void jfs_read_inode(struct inode *inode)
  41 {
  42         if (diRead(inode)) {
  43                 make_bad_inode(inode);
  44                 return;
  45         }
  46
  47         if (S_ISREG(inode->i_mode)) {
  48                 inode->i_op = &jfs_file_inode_operations;
  49                 inode->i_fop = &jfs_file_operations;
  50                 inode->i_mapping->a_ops = &jfs_aops;
  51         } else if (S_ISDIR(inode->i_mode)) {
  52                 inode->i_op = &jfs_dir_inode_operations;
  53                 inode->i_fop = &jfs_dir_operations;
  54                 inode->i_mapping->a_ops = &jfs_aops;
  55                 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
  56         } else if (S_ISLNK(inode->i_mode)) {
  57                 if (inode->i_size >= IDATASIZE) {
  58                         inode->i_op = &page_symlink_inode_operations;
  59                         inode->i_mapping->a_ops = &jfs_aops;
  60                 } else
  61                         inode->i_op = &jfs_symlink_inode_operations;
  62         } else {
  63                 inode->i_op = &jfs_file_inode_operations;
  64                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
  65         }
  66 }
  67
  68 /*
  69  * Workhorse of both fsync & write_inode
  70  */
  71 int jfs_commit_inode(struct inode *inode, int wait)
  72 {
  73         int rc = 0;
  74         tid_t tid;
  75         static int noisy = 5;
  76
  77         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
  78
  79         /*
  80          * Don't commit if inode has been committed since last being
  81          * marked dirty, or if it has been deleted.
  82          */
  83         if (test_cflag(COMMIT_Nolink, inode) ||
  84             !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         rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
 103         txEnd(tid);
 104         up(&JFS_IP(inode)->commit_sem);
 105         return rc;
 106 }
 107
 108 void jfs_write_inode(struct inode *inode, int wait)
 109 {
 110         if (test_cflag(COMMIT_Nolink, inode))
 111                 return;
 112         /*
 113          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
 114          * It has been committed since the last change, but was still
 115          * on the dirty inode list.
 116          */
 117          if (!test_cflag(COMMIT_Dirty, inode)) {
 118                 /* Make sure committed changes hit the disk */
 119                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
 120                 return;
 121          }
 122
 123         if (jfs_commit_inode(inode, wait)) {
 124                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
 125         }
 126 }
 127
 128 void jfs_delete_inode(struct inode *inode)
 129 {
 130         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
 131
 132         if (test_cflag(COMMIT_Freewmap, inode))
 133                 freeZeroLink(inode);
 134
 135         diFree(inode);
 136
 137         clear_inode(inode);
 138 }
 139
 140 void jfs_dirty_inode(struct inode *inode)
 141 {
 142         static int noisy = 5;
 143
 144         if (isReadOnly(inode)) {
 145                 if (!special_file(inode->i_mode) && noisy) {
 146                         /* kernel allows writes to devices on read-only
 147                          * partitions and may try to mark inode dirty
 148                          */
 149                         jfs_err("jfs_dirty_inode called on read-only volume");
 150                         jfs_err("Is remount racy?");
 151                         noisy--;
 152                 }
 153                 return;
 154         }
 155
 156         set_cflag(COMMIT_Dirty, inode);
 157 }
 158
 159 static int
 160 jfs_get_blocks(struct inode *ip, sector_t lblock, unsigned long max_blocks,
 161                         struct buffer_head *bh_result, int create)
 162 {
 163         s64 lblock64 = lblock;
 164         int no_size_check = 0;
 165         int rc = 0;
 166         int take_locks;
 167         xad_t xad;
 168         s64 xaddr;
 169         int xflag;
 170         s32 xlen;
 171
 172         /*
 173          * If this is a special inode (imap, dmap) or directory,
 174          * the lock should already be taken
 175          */
 176         take_locks = ((JFS_IP(ip)->fileset != AGGREGATE_I) &&
 177                       !S_ISDIR(ip->i_mode));
 178         /*
 179          * Take appropriate lock on inode
 180          */
 181         if (take_locks) {
 182                 if (create)
 183                         IWRITE_LOCK(ip);
 184                 else
 185                         IREAD_LOCK(ip);
 186         }
 187
 188         /*
 189          * A directory's "data" is the inode index table, but i_size is the
 190          * size of the d-tree, so don't check the offset against i_size
 191          */
 192         if (S_ISDIR(ip->i_mode))
 193                 no_size_check = 1;
 194
 195         if ((no_size_check ||
 196              ((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size)) &&
 197             (xtLookup(ip, lblock64, max_blocks, &xflag, &xaddr, &xlen, no_size_check)
 198              == 0) && xlen) {
 199                 if (xflag & XAD_NOTRECORDED) {
 200                         if (!create)
 201                                 /*
 202                                  * Allocated but not recorded, read treats
 203                                  * this as a hole
 204                                  */
 205                                 goto unlock;
 206 #ifdef _JFS_4K
 207                         XADoffset(&xad, lblock64);
 208                         XADlength(&xad, xlen);
 209                         XADaddress(&xad, xaddr);
 210 #else                           /* _JFS_4K */
 211                         /*
 212                          * As long as block size = 4K, this isn't a problem.
 213                          * We should mark the whole page not ABNR, but how
 214                          * will we know to mark the other blocks BH_New?
 215                          */
 216                         BUG();
 217 #endif                          /* _JFS_4K */
 218                         rc = extRecord(ip, &xad);
 219                         if (rc)
 220                                 goto unlock;
 221                         set_buffer_new(bh_result);
 222                 }
 223
 224                 map_bh(bh_result, ip->i_sb, xaddr);
 225                 bh_result->b_size = xlen << ip->i_blkbits;
 226                 goto unlock;
 227         }
 228         if (!create)
 229                 goto unlock;
 230
 231         /*
 232          * Allocate a new block
 233          */
 234 #ifdef _JFS_4K
 235         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 236                 goto unlock;
 237         rc = extAlloc(ip, max_blocks, lblock64, &xad, FALSE);
 238         if (rc)
 239                 goto unlock;
 240
 241         set_buffer_new(bh_result);
 242         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
 243         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
 244
 245 #else                           /* _JFS_4K */
 246         /*
 247          * We need to do whatever it takes to keep all but the last buffers
 248          * in 4K pages - see jfs_write.c
 249          */
 250         BUG();
 251 #endif                          /* _JFS_4K */
 252
 253       unlock:
 254         /*
 255          * Release lock on inode
 256          */
 257         if (take_locks) {
 258                 if (create)
 259                         IWRITE_UNLOCK(ip);
 260                 else
 261                         IREAD_UNLOCK(ip);
 262         }
 263         return rc;
 264 }
 265
 266 static int jfs_get_block(struct inode *ip, sector_t lblock,
 267                          struct buffer_head *bh_result, int create)
 268 {
 269         return jfs_get_blocks(ip, lblock, 1, bh_result, create);
 270 }
 271
 272 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
 273 {
 274         return block_write_full_page(page, jfs_get_block, wbc);
 275 }
 276
 277 static int jfs_writepages(struct address_space *mapping,
 278                         struct writeback_control *wbc)
 279 {
 280         return mpage_writepages(mapping, wbc, jfs_get_block);
 281 }
 282
 283 static int jfs_readpage(struct file *file, struct page *page)
 284 {
 285         return mpage_readpage(page, jfs_get_block);
 286 }
 287
 288 static int jfs_readpages(struct file *file, struct address_space *mapping,
 289                 struct list_head *pages, unsigned nr_pages)
 290 {
 291         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
 292 }
 293
 294 static int jfs_prepare_write(struct file *file,
 295                              struct page *page, unsigned from, unsigned to)
 296 {
 297         return nobh_prepare_write(page, from, to, jfs_get_block);
 298 }
 299
 300 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
 301 {
 302         return generic_block_bmap(mapping, block, jfs_get_block);
 303 }
 304
 305 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
 306         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
 307 {
 308         struct file *file = iocb->ki_filp;
 309         struct inode *inode = file->f_mapping->host;
 310
 311         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 312                                 offset, nr_segs, jfs_get_blocks, NULL);
 313 }
 314
 315 struct address_space_operations jfs_aops = {
 316         .readpage       = jfs_readpage,
 317         .readpages      = jfs_readpages,
 318         .writepage      = jfs_writepage,
 319         .writepages     = jfs_writepages,
 320         .sync_page      = block_sync_page,
 321         .prepare_write  = jfs_prepare_write,
 322         .commit_write   = nobh_commit_write,
 323         .bmap           = jfs_bmap,
 324         .direct_IO      = jfs_direct_IO,
 325 };
 326
 327 /*
 328  * Guts of jfs_truncate.  Called with locks already held.  Can be called
 329  * with directory for truncating directory index table.
 330  */
 331 void jfs_truncate_nolock(struct inode *ip, loff_t length)
 332 {
 333         loff_t newsize;
 334         tid_t tid;
 335
 336         ASSERT(length >= 0);
 337
 338         if (test_cflag(COMMIT_Nolink, ip)) {
 339                 xtTruncate(0, ip, length, COMMIT_WMAP);
 340                 return;
 341         }
 342
 343         do {
 344                 tid = txBegin(ip->i_sb, 0);
 345
 346                 /*
 347                  * The commit_sem cannot be taken before txBegin.
 348                  * txBegin may block and there is a chance the inode
 349                  * could be marked dirty and need to be committed
 350                  * before txBegin unblocks
 351                  */
 352                 down(&JFS_IP(ip)->commit_sem);
 353
 354                 newsize = xtTruncate(tid, ip, length,
 355                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
 356                 if (newsize < 0) {
 357                         txEnd(tid);
 358                         up(&JFS_IP(ip)->commit_sem);
 359                         break;
 360                 }
 361
 362                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
 363                 mark_inode_dirty(ip);
 364
 365                 txCommit(tid, 1, &ip, 0);
 366                 txEnd(tid);
 367                 up(&JFS_IP(ip)->commit_sem);
 368         } while (newsize > length);     /* Truncate isn't always atomic */
 369 }
 370
 371 void jfs_truncate(struct inode *ip)
 372 {
 373         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
 374
 375         nobh_truncate_page(ip->i_mapping, ip->i_size);
 376
 377         IWRITE_LOCK(ip);
 378         jfs_truncate_nolock(ip, ip->i_size);
 379         IWRITE_UNLOCK(ip);
 380 }