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/locks.h>
  22 #include "jfs_incore.h"
  23 #include "jfs_filsys.h"
  24 #include "jfs_dmap.h"
  25 #include "jfs_imap.h"
  26 #include "jfs_extent.h"
  27 #include "jfs_unicode.h"
  28 #include "jfs_debug.h"
  29
  30
  31 extern struct inode_operations jfs_dir_inode_operations;
  32 extern struct inode_operations jfs_file_inode_operations;
  33 extern struct inode_operations jfs_symlink_inode_operations;
  34 extern struct file_operations jfs_dir_operations;
  35 extern struct file_operations jfs_file_operations;
  36 struct address_space_operations jfs_aops;
  37 extern int freeZeroLink(struct inode *);
  38
  39 void jfs_clear_inode(struct inode *inode)
  40 {
  41         struct jfs_inode_info *ji = JFS_IP(inode);
  42
  43         if (is_bad_inode(inode))
  44                 /*
  45                  * We free the fs-dependent structure before making the
  46                  * inode bad
  47                  */
  48                 return;
  49
  50         jFYI(1, ("jfs_clear_inode called ip = 0x%p\n", inode));
  51
  52         if (ji->active_ag != -1) {
  53                 struct bmap *bmap = JFS_SBI(inode->i_sb)->bmap;
  54                 atomic_dec(&bmap->db_active[ji->active_ag]);
  55         }
  56
  57         ASSERT(list_empty(&ji->anon_inode_list));
  58
  59         if (ji->atlhead) {
  60                 jERROR(1, ("jfs_clear_inode: inode %p has anonymous tlocks\n",
  61                                         inode));
  62                 jERROR(1, ("i_state = 0x%lx, cflag = 0x%lx\n",
  63                                         inode->i_state, ji->cflag));
  64         }
  65
  66         free_jfs_inode(inode);
  67 }
  68
  69 void jfs_read_inode(struct inode *inode)
  70 {
  71         int rc;
  72
  73         rc = alloc_jfs_inode(inode);
  74         if (rc) {
  75                 jFYI(1, ("In jfs_read_inode, alloc_jfs_inode failed"));
  76                 goto bad_inode;
  77         }
  78         jFYI(1, ("In jfs_read_inode, inode = 0x%p\n", inode));
  79
  80         if (diRead(inode))
  81                 goto bad_inode_free;
  82
  83         if (S_ISREG(inode->i_mode)) {
  84                 inode->i_op = &jfs_file_inode_operations;
  85                 inode->i_fop = &jfs_file_operations;
  86                 inode->i_mapping->a_ops = &jfs_aops;
  87         } else if (S_ISDIR(inode->i_mode)) {
  88                 inode->i_op = &jfs_dir_inode_operations;
  89                 inode->i_fop = &jfs_dir_operations;
  90                 inode->i_mapping->a_ops = &jfs_aops;
  91                 inode->i_mapping->gfp_mask = GFP_NOFS;
  92         } else if (S_ISLNK(inode->i_mode)) {
  93                 if (inode->i_size >= IDATASIZE) {
  94                         inode->i_op = &page_symlink_inode_operations;
  95                         inode->i_mapping->a_ops = &jfs_aops;
  96                 } else
  97                         inode->i_op = &jfs_symlink_inode_operations;
  98         } else {
  99                 inode->i_op = &jfs_file_inode_operations;
 100                 init_special_inode(inode, inode->i_mode,
 101                                    kdev_t_to_nr(inode->i_rdev));
 102         }
 103
 104         return;
 105
 106       bad_inode_free:
 107         free_jfs_inode(inode);
 108       bad_inode:
 109         make_bad_inode(inode);
 110 }
 111
 112 /* This define is from fs/open.c */
 113 #define special_file(m) (S_ISCHR(m)||S_ISBLK(m)||S_ISFIFO(m)||S_ISSOCK(m))
 114
 115 /*
 116  * Workhorse of both fsync & write_inode
 117  */
 118 int jfs_commit_inode(struct inode *inode, int wait)
 119 {
 120         int rc = 0;
 121         tid_t tid;
 122         static int noisy = 5;
 123
 124         jFYI(1, ("In jfs_commit_inode, inode = 0x%p\n", inode));
 125
 126         /*
 127          * Don't commit if inode has been committed since last being
 128          * marked dirty, or if it has been deleted.
 129          */
 130         if (test_cflag(COMMIT_Nolink, inode) ||
 131             !test_cflag(COMMIT_Dirty, inode))
 132                 return 0;
 133
 134         if (isReadOnly(inode)) {
 135                 /* kernel allows writes to devices on read-only
 136                  * partitions and may think inode is dirty
 137                  */
 138                 if (!special_file(inode->i_mode) && noisy) {
 139                         jERROR(1, ("jfs_commit_inode(0x%p) called on "
 140                                    "read-only volume\n", inode));
 141                         jERROR(1, ("Is remount racy?\n"));
 142                         noisy--;
 143                 }
 144                 return 0;
 145         }
 146
 147         tid = txBegin(inode->i_sb, COMMIT_INODE);
 148         down(&JFS_IP(inode)->commit_sem);
 149         rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
 150         txEnd(tid);
 151         up(&JFS_IP(inode)->commit_sem);
 152         return -rc;
 153 }
 154
 155 void jfs_write_inode(struct inode *inode, int wait)
 156 {
 157         /*
 158          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
 159          * It has been committed since the last change, but was still
 160          * on the dirty inode list
 161          */
 162         if (test_cflag(COMMIT_Nolink, inode) ||
 163             !test_cflag(COMMIT_Dirty, inode))
 164                 return;
 165
 166         if (jfs_commit_inode(inode, wait)) {
 167                 jERROR(1, ("jfs_write_inode: jfs_commit_inode failed!\n"));
 168         }
 169 }
 170
 171 void jfs_delete_inode(struct inode *inode)
 172 {
 173         jFYI(1, ("In jfs_delete_inode, inode = 0x%p\n", inode));
 174
 175         if (test_cflag(COMMIT_Freewmap, inode))
 176                 freeZeroLink(inode);
 177
 178         diFree(inode);
 179
 180         clear_inode(inode);
 181 }
 182
 183 void jfs_dirty_inode(struct inode *inode)
 184 {
 185         static int noisy = 5;
 186
 187         if (isReadOnly(inode)) {
 188                 if (!special_file(inode->i_mode) && noisy) {
 189                         /* kernel allows writes to devices on read-only
 190                          * partitions and may try to mark inode dirty
 191                          */
 192                         jERROR(1, ("jfs_dirty_inode called on "
 193                                    "read-only volume\n"));
 194                         jERROR(1, ("Is remount racy?\n"));
 195                         noisy--;
 196                 }
 197                 return;
 198         }
 199
 200         set_cflag(COMMIT_Dirty, inode);
 201 }
 202
 203 static int jfs_get_block(struct inode *ip, long lblock,
 204                          struct buffer_head *bh_result, int create)
 205 {
 206         s64 lblock64 = lblock;
 207         int no_size_check = 0;
 208         int rc = 0;
 209         int take_locks;
 210         xad_t xad;
 211         s64 xaddr;
 212         int xflag;
 213         s32 xlen;
 214
 215         /*
 216          * If this is a special inode (imap, dmap) or directory,
 217          * the lock should already be taken
 218          */
 219         take_locks = ((JFS_IP(ip)->fileset != AGGREGATE_I) &&
 220                       !S_ISDIR(ip->i_mode));
 221         /*
 222          * Take appropriate lock on inode
 223          */
 224         if (take_locks) {
 225                 if (create)
 226                         IWRITE_LOCK(ip);
 227                 else
 228                         IREAD_LOCK(ip);
 229         }
 230
 231         /*
 232          * A directory's "data" is the inode index table, but i_size is the
 233          * size of the d-tree, so don't check the offset against i_size
 234          */
 235         if (S_ISDIR(ip->i_mode))
 236                 no_size_check = 1;
 237
 238         if ((no_size_check ||
 239              ((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size)) &&
 240             (xtLookup(ip, lblock64, 1, &xflag, &xaddr, &xlen, no_size_check)
 241              == 0) && xlen) {
 242                 if (xflag & XAD_NOTRECORDED) {
 243                         if (!create)
 244                                 /*
 245                                  * Allocated but not recorded, read treats
 246                                  * this as a hole
 247                                  */
 248                                 goto unlock;
 249 #ifdef _JFS_4K
 250                         XADoffset(&xad, lblock64);
 251                         XADlength(&xad, xlen);
 252                         XADaddress(&xad, xaddr);
 253 #else                           /* _JFS_4K */
 254                         /*
 255                          * As long as block size = 4K, this isn't a problem.
 256                          * We should mark the whole page not ABNR, but how
 257                          * will we know to mark the other blocks BH_New?
 258                          */
 259                         BUG();
 260 #endif                          /* _JFS_4K */
 261                         rc = extRecord(ip, &xad);
 262                         if (rc)
 263                                 goto unlock;
 264                         bh_result->b_state |= (1UL << BH_New);
 265                 }
 266
 267                 bh_result->b_dev = ip->i_dev;
 268                 bh_result->b_blocknr = xaddr;
 269                 bh_result->b_state |= (1UL << BH_Mapped);
 270                 goto unlock;
 271         }
 272         if (!create)
 273                 goto unlock;
 274
 275         /*
 276          * Allocate a new block
 277          */
 278 #ifdef _JFS_4K
 279         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 280                 goto unlock;
 281         rc = extAlloc(ip, 1, lblock64, &xad, FALSE);
 282         if (rc)
 283                 goto unlock;
 284
 285         bh_result->b_dev = ip->i_dev;
 286         bh_result->b_blocknr = addressXAD(&xad);
 287         bh_result->b_state |= ((1UL << BH_Mapped) | (1UL << BH_New));
 288
 289 #else                           /* _JFS_4K */
 290         /*
 291          * We need to do whatever it takes to keep all but the last buffers
 292          * in 4K pages - see jfs_write.c
 293          */
 294         BUG();
 295 #endif                          /* _JFS_4K */
 296
 297       unlock:
 298         /*
 299          * Release lock on inode
 300          */
 301         if (take_locks) {
 302                 if (create)
 303                         IWRITE_UNLOCK(ip);
 304                 else
 305                         IREAD_UNLOCK(ip);
 306         }
 307         return -rc;
 308 }
 309
 310 static int jfs_writepage(struct page *page)
 311 {
 312         return block_write_full_page(page, jfs_get_block);
 313 }
 314
 315 static int jfs_readpage(struct file *file, struct page *page)
 316 {
 317         return block_read_full_page(page, jfs_get_block);
 318 }
 319
 320 static int jfs_prepare_write(struct file *file,
 321                              struct page *page, unsigned from, unsigned to)
 322 {
 323         return block_prepare_write(page, from, to, jfs_get_block);
 324 }
 325
 326 static int jfs_bmap(struct address_space *mapping, long block)
 327 {
 328         return generic_block_bmap(mapping, block, jfs_get_block);
 329 }
 330
 331 static int jfs_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
 332                          unsigned long blocknr, int blocksize)
 333 {
 334         return generic_direct_IO(rw, inode, iobuf, blocknr,
 335                                  blocksize, jfs_get_block);
 336 }
 337
 338 struct address_space_operations jfs_aops = {
 339         .readpage       = jfs_readpage,
 340         .writepage      = jfs_writepage,
 341         .sync_page      = block_sync_page,
 342         .prepare_write  = jfs_prepare_write,
 343         .commit_write   = generic_commit_write,
 344         .bmap           = jfs_bmap,
 345         .direct_IO      = jfs_direct_IO,
 346 };
 347
 348 /*
 349  * Guts of jfs_truncate.  Called with locks already held.  Can be called
 350  * with directory for truncating directory index table.
 351  */
 352 void jfs_truncate_nolock(struct inode *ip, loff_t length)
 353 {
 354         loff_t newsize;
 355         tid_t tid;
 356
 357         ASSERT(length >= 0);
 358
 359         if (test_cflag(COMMIT_Nolink, ip)) {
 360                 xtTruncate(0, ip, length, COMMIT_WMAP);
 361                 return;
 362         }
 363
 364         do {
 365                 tid = txBegin(ip->i_sb, 0);
 366
 367                 /*
 368                  * The commit_sem cannot be taken before txBegin.
 369                  * txBegin may block and there is a chance the inode
 370                  * could be marked dirty and need to be committed
 371                  * before txBegin unblocks
 372                  */
 373                 down(&JFS_IP(ip)->commit_sem);
 374
 375                 newsize = xtTruncate(tid, ip, length,
 376                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
 377                 if (newsize < 0) {
 378                         txEnd(tid);
 379                         up(&JFS_IP(ip)->commit_sem);
 380                         break;
 381                 }
 382
 383                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
 384                 mark_inode_dirty(ip);
 385
 386                 txCommit(tid, 1, &ip, 0);
 387                 txEnd(tid);
 388                 up(&JFS_IP(ip)->commit_sem);
 389         } while (newsize > length);     /* Truncate isn't always atomic */
 390 }
 391
 392 void jfs_truncate(struct inode *ip)
 393 {
 394         jFYI(1, ("jfs_truncate: size = 0x%lx\n", (ulong) ip->i_size));
 395
 396         block_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
 397
 398         IWRITE_LOCK(ip);
 399         jfs_truncate_nolock(ip, ip->i_size);
 400         IWRITE_UNLOCK(ip);
 401 }