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/uio.h>
  26 #include <linux/writeback.h>
  27 #include "jfs_incore.h"
  28 #include "jfs_inode.h"
  29 #include "jfs_filsys.h"
  30 #include "jfs_imap.h"
  31 #include "jfs_extent.h"
  32 #include "jfs_unicode.h"
  33 #include "jfs_debug.h"
  34
  35
  36 struct inode *jfs_iget(struct super_block *sb, unsigned long ino)
  37 {
  38         struct inode *inode;
  39         int ret;
  40
  41         inode = iget_locked(sb, ino);
  42         if (!inode)
  43                 return ERR_PTR(-ENOMEM);
  44         if (!(inode->i_state & I_NEW))
  45                 return inode;
  46
  47         ret = diRead(inode);
  48         if (ret < 0) {
  49                 iget_failed(inode);
  50                 return ERR_PTR(ret);
  51         }
  52
  53         if (S_ISREG(inode->i_mode)) {
  54                 inode->i_op = &jfs_file_inode_operations;
  55                 inode->i_fop = &jfs_file_operations;
  56                 inode->i_mapping->a_ops = &jfs_aops;
  57         } else if (S_ISDIR(inode->i_mode)) {
  58                 inode->i_op = &jfs_dir_inode_operations;
  59                 inode->i_fop = &jfs_dir_operations;
  60         } else if (S_ISLNK(inode->i_mode)) {
  61                 if (inode->i_size >= IDATASIZE) {
  62                         inode->i_op = &page_symlink_inode_operations;
  63                         inode_nohighmem(inode);
  64                         inode->i_mapping->a_ops = &jfs_aops;
  65                 } else {
  66                         inode->i_op = &jfs_fast_symlink_inode_operations;
  67                         inode->i_link = JFS_IP(inode)->i_inline;
  68                         /*
  69                          * The inline data should be null-terminated, but
  70                          * don't let on-disk corruption crash the kernel
  71                          */
  72                         inode->i_link[inode->i_size] = '\0';
  73                 }
  74         } else {
  75                 inode->i_op = &jfs_file_inode_operations;
  76                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
  77         }
  78         unlock_new_inode(inode);
  79         return inode;
  80 }
  81
  82 /*
  83  * Workhorse of both fsync & write_inode
  84  */
  85 int jfs_commit_inode(struct inode *inode, int wait)
  86 {
  87         int rc = 0;
  88         tid_t tid;
  89         static int noisy = 5;
  90
  91         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
  92
  93         /*
  94          * Don't commit if inode has been committed since last being
  95          * marked dirty, or if it has been deleted.
  96          */
  97         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
  98                 return 0;
  99
 100         if (isReadOnly(inode)) {
 101                 /* kernel allows writes to devices on read-only
 102                  * partitions and may think inode is dirty
 103                  */
 104                 if (!special_file(inode->i_mode) && noisy) {
 105                         jfs_err("jfs_commit_inode(0x%p) called on read-only volume",
 106                                 inode);
 107                         jfs_err("Is remount racy?");
 108                         noisy--;
 109                 }
 110                 return 0;
 111         }
 112
 113         tid = txBegin(inode->i_sb, COMMIT_INODE);
 114         mutex_lock(&JFS_IP(inode)->commit_mutex);
 115
 116         /*
 117          * Retest inode state after taking commit_mutex
 118          */
 119         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
 120                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
 121
 122         txEnd(tid);
 123         mutex_unlock(&JFS_IP(inode)->commit_mutex);
 124         return rc;
 125 }
 126
 127 int jfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 128 {
 129         int wait = wbc->sync_mode == WB_SYNC_ALL;
 130
 131         if (inode->i_nlink == 0)
 132                 return 0;
 133         /*
 134          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
 135          * It has been committed since the last change, but was still
 136          * on the dirty inode list.
 137          */
 138         if (!test_cflag(COMMIT_Dirty, inode)) {
 139                 /* Make sure committed changes hit the disk */
 140                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
 141                 return 0;
 142         }
 143
 144         if (jfs_commit_inode(inode, wait)) {
 145                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
 146                 return -EIO;
 147         } else
 148                 return 0;
 149 }
 150
 151 void jfs_evict_inode(struct inode *inode)
 152 {
 153         jfs_info("In jfs_evict_inode, inode = 0x%p", inode);
 154
 155         if (!inode->i_nlink && !is_bad_inode(inode)) {
 156                 dquot_initialize(inode);
 157
 158                 if (JFS_IP(inode)->fileset == FILESYSTEM_I) {
 159                         truncate_inode_pages_final(&inode->i_data);
 160
 161                         if (test_cflag(COMMIT_Freewmap, inode))
 162                                 jfs_free_zero_link(inode);
 163
 164                         diFree(inode);
 165
 166                         /*
 167                          * Free the inode from the quota allocation.
 168                          */
 169                         dquot_initialize(inode);
 170                         dquot_free_inode(inode);
 171                 }
 172         } else {
 173                 truncate_inode_pages_final(&inode->i_data);
 174         }
 175         clear_inode(inode);
 176         dquot_drop(inode);
 177 }
 178
 179 void jfs_dirty_inode(struct inode *inode, int flags)
 180 {
 181         static int noisy = 5;
 182
 183         if (isReadOnly(inode)) {
 184                 if (!special_file(inode->i_mode) && noisy) {
 185                         /* kernel allows writes to devices on read-only
 186                          * partitions and may try to mark inode dirty
 187                          */
 188                         jfs_err("jfs_dirty_inode called on read-only volume");
 189                         jfs_err("Is remount racy?");
 190                         noisy--;
 191                 }
 192                 return;
 193         }
 194
 195         set_cflag(COMMIT_Dirty, inode);
 196 }
 197
 198 int jfs_get_block(struct inode *ip, sector_t lblock,
 199                   struct buffer_head *bh_result, int create)
 200 {
 201         s64 lblock64 = lblock;
 202         int rc = 0;
 203         xad_t xad;
 204         s64 xaddr;
 205         int xflag;
 206         s32 xlen = bh_result->b_size >> ip->i_blkbits;
 207
 208         /*
 209          * Take appropriate lock on inode
 210          */
 211         if (create)
 212                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 213         else
 214                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
 215
 216         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
 217             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
 218             xaddr) {
 219                 if (xflag & XAD_NOTRECORDED) {
 220                         if (!create)
 221                                 /*
 222                                  * Allocated but not recorded, read treats
 223                                  * this as a hole
 224                                  */
 225                                 goto unlock;
 226 #ifdef _JFS_4K
 227                         XADoffset(&xad, lblock64);
 228                         XADlength(&xad, xlen);
 229                         XADaddress(&xad, xaddr);
 230 #else                           /* _JFS_4K */
 231                         /*
 232                          * As long as block size = 4K, this isn't a problem.
 233                          * We should mark the whole page not ABNR, but how
 234                          * will we know to mark the other blocks BH_New?
 235                          */
 236                         BUG();
 237 #endif                          /* _JFS_4K */
 238                         rc = extRecord(ip, &xad);
 239                         if (rc)
 240                                 goto unlock;
 241                         set_buffer_new(bh_result);
 242                 }
 243
 244                 map_bh(bh_result, ip->i_sb, xaddr);
 245                 bh_result->b_size = xlen << ip->i_blkbits;
 246                 goto unlock;
 247         }
 248         if (!create)
 249                 goto unlock;
 250
 251         /*
 252          * Allocate a new block
 253          */
 254 #ifdef _JFS_4K
 255         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 256                 goto unlock;
 257         rc = extAlloc(ip, xlen, lblock64, &xad, false);
 258         if (rc)
 259                 goto unlock;
 260
 261         set_buffer_new(bh_result);
 262         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
 263         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
 264
 265 #else                           /* _JFS_4K */
 266         /*
 267          * We need to do whatever it takes to keep all but the last buffers
 268          * in 4K pages - see jfs_write.c
 269          */
 270         BUG();
 271 #endif                          /* _JFS_4K */
 272
 273       unlock:
 274         /*
 275          * Release lock on inode
 276          */
 277         if (create)
 278                 IWRITE_UNLOCK(ip);
 279         else
 280                 IREAD_UNLOCK(ip);
 281         return rc;
 282 }
 283
 284 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
 285 {
 286         return block_write_full_page(page, jfs_get_block, wbc);
 287 }
 288
 289 static int jfs_writepages(struct address_space *mapping,
 290                         struct writeback_control *wbc)
 291 {
 292         return mpage_writepages(mapping, wbc, jfs_get_block);
 293 }
 294
 295 static int jfs_readpage(struct file *file, struct page *page)
 296 {
 297         return mpage_readpage(page, jfs_get_block);
 298 }
 299
 300 static int jfs_readpages(struct file *file, struct address_space *mapping,
 301                 struct list_head *pages, unsigned nr_pages)
 302 {
 303         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
 304 }
 305
 306 static void jfs_write_failed(struct address_space *mapping, loff_t to)
 307 {
 308         struct inode *inode = mapping->host;
 309
 310         if (to > inode->i_size) {
 311                 truncate_pagecache(inode, inode->i_size);
 312                 jfs_truncate(inode);
 313         }
 314 }
 315
 316 static int jfs_write_begin(struct file *file, struct address_space *mapping,
 317                                 loff_t pos, unsigned len, unsigned flags,
 318                                 struct page **pagep, void **fsdata)
 319 {
 320         int ret;
 321
 322         ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
 323                                 jfs_get_block);
 324         if (unlikely(ret))
 325                 jfs_write_failed(mapping, pos + len);
 326
 327         return ret;
 328 }
 329
 330 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
 331 {
 332         return generic_block_bmap(mapping, block, jfs_get_block);
 333 }
 334
 335 static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 336 {
 337         struct file *file = iocb->ki_filp;
 338         struct address_space *mapping = file->f_mapping;
 339         struct inode *inode = file->f_mapping->host;
 340         size_t count = iov_iter_count(iter);
 341         ssize_t ret;
 342
 343         ret = blockdev_direct_IO(iocb, inode, iter, jfs_get_block);
 344
 345         /*
 346          * In case of error extending write may have instantiated a few
 347          * blocks outside i_size. Trim these off again.
 348          */
 349         if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
 350                 loff_t isize = i_size_read(inode);
 351                 loff_t end = iocb->ki_pos + count;
 352
 353                 if (end > isize)
 354                         jfs_write_failed(mapping, end);
 355         }
 356
 357         return ret;
 358 }
 359
 360 const struct address_space_operations jfs_aops = {
 361         .readpage       = jfs_readpage,
 362         .readpages      = jfs_readpages,
 363         .writepage      = jfs_writepage,
 364         .writepages     = jfs_writepages,
 365         .write_begin    = jfs_write_begin,
 366         .write_end      = nobh_write_end,
 367         .bmap           = jfs_bmap,
 368         .direct_IO      = jfs_direct_IO,
 369 };
 370
 371 /*
 372  * Guts of jfs_truncate.  Called with locks already held.  Can be called
 373  * with directory for truncating directory index table.
 374  */
 375 void jfs_truncate_nolock(struct inode *ip, loff_t length)
 376 {
 377         loff_t newsize;
 378         tid_t tid;
 379
 380         ASSERT(length >= 0);
 381
 382         if (test_cflag(COMMIT_Nolink, ip)) {
 383                 xtTruncate(0, ip, length, COMMIT_WMAP);
 384                 return;
 385         }
 386
 387         do {
 388                 tid = txBegin(ip->i_sb, 0);
 389
 390                 /*
 391                  * The commit_mutex cannot be taken before txBegin.
 392                  * txBegin may block and there is a chance the inode
 393                  * could be marked dirty and need to be committed
 394                  * before txBegin unblocks
 395                  */
 396                 mutex_lock(&JFS_IP(ip)->commit_mutex);
 397
 398                 newsize = xtTruncate(tid, ip, length,
 399                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
 400                 if (newsize < 0) {
 401                         txEnd(tid);
 402                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
 403                         break;
 404                 }
 405
 406                 ip->i_mtime = ip->i_ctime = current_time(ip);
 407                 mark_inode_dirty(ip);
 408
 409                 txCommit(tid, 1, &ip, 0);
 410                 txEnd(tid);
 411                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
 412         } while (newsize > length);     /* Truncate isn't always atomic */
 413 }
 414
 415 void jfs_truncate(struct inode *ip)
 416 {
 417         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
 418
 419         nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
 420
 421         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 422         jfs_truncate_nolock(ip, ip->i_size);
 423         IWRITE_UNLOCK(ip);
 424 }