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                         struct inode *ipimap = JFS_SBI(inode->i_sb)->ipimap;
 160                         truncate_inode_pages_final(&inode->i_data);
 161
 162                         if (test_cflag(COMMIT_Freewmap, inode))
 163                                 jfs_free_zero_link(inode);
 164
 165                         if (ipimap && JFS_IP(ipimap)->i_imap)
 166                                 diFree(inode);
 167
 168                         /*
 169                          * Free the inode from the quota allocation.
 170                          */
 171                         dquot_initialize(inode);
 172                         dquot_free_inode(inode);
 173                 }
 174         } else {
 175                 truncate_inode_pages_final(&inode->i_data);
 176         }
 177         clear_inode(inode);
 178         dquot_drop(inode);
 179 }
 180
 181 void jfs_dirty_inode(struct inode *inode, int flags)
 182 {
 183         static int noisy = 5;
 184
 185         if (isReadOnly(inode)) {
 186                 if (!special_file(inode->i_mode) && noisy) {
 187                         /* kernel allows writes to devices on read-only
 188                          * partitions and may try to mark inode dirty
 189                          */
 190                         jfs_err("jfs_dirty_inode called on read-only volume");
 191                         jfs_err("Is remount racy?");
 192                         noisy--;
 193                 }
 194                 return;
 195         }
 196
 197         set_cflag(COMMIT_Dirty, inode);
 198 }
 199
 200 int jfs_get_block(struct inode *ip, sector_t lblock,
 201                   struct buffer_head *bh_result, int create)
 202 {
 203         s64 lblock64 = lblock;
 204         int rc = 0;
 205         xad_t xad;
 206         s64 xaddr;
 207         int xflag;
 208         s32 xlen = bh_result->b_size >> ip->i_blkbits;
 209
 210         /*
 211          * Take appropriate lock on inode
 212          */
 213         if (create)
 214                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 215         else
 216                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
 217
 218         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
 219             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
 220             xaddr) {
 221                 if (xflag & XAD_NOTRECORDED) {
 222                         if (!create)
 223                                 /*
 224                                  * Allocated but not recorded, read treats
 225                                  * this as a hole
 226                                  */
 227                                 goto unlock;
 228 #ifdef _JFS_4K
 229                         XADoffset(&xad, lblock64);
 230                         XADlength(&xad, xlen);
 231                         XADaddress(&xad, xaddr);
 232 #else                           /* _JFS_4K */
 233                         /*
 234                          * As long as block size = 4K, this isn't a problem.
 235                          * We should mark the whole page not ABNR, but how
 236                          * will we know to mark the other blocks BH_New?
 237                          */
 238                         BUG();
 239 #endif                          /* _JFS_4K */
 240                         rc = extRecord(ip, &xad);
 241                         if (rc)
 242                                 goto unlock;
 243                         set_buffer_new(bh_result);
 244                 }
 245
 246                 map_bh(bh_result, ip->i_sb, xaddr);
 247                 bh_result->b_size = xlen << ip->i_blkbits;
 248                 goto unlock;
 249         }
 250         if (!create)
 251                 goto unlock;
 252
 253         /*
 254          * Allocate a new block
 255          */
 256 #ifdef _JFS_4K
 257         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 258                 goto unlock;
 259         rc = extAlloc(ip, xlen, lblock64, &xad, false);
 260         if (rc)
 261                 goto unlock;
 262
 263         set_buffer_new(bh_result);
 264         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
 265         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
 266
 267 #else                           /* _JFS_4K */
 268         /*
 269          * We need to do whatever it takes to keep all but the last buffers
 270          * in 4K pages - see jfs_write.c
 271          */
 272         BUG();
 273 #endif                          /* _JFS_4K */
 274
 275       unlock:
 276         /*
 277          * Release lock on inode
 278          */
 279         if (create)
 280                 IWRITE_UNLOCK(ip);
 281         else
 282                 IREAD_UNLOCK(ip);
 283         return rc;
 284 }
 285
 286 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
 287 {
 288         return block_write_full_page(page, jfs_get_block, wbc);
 289 }
 290
 291 static int jfs_writepages(struct address_space *mapping,
 292                         struct writeback_control *wbc)
 293 {
 294         return mpage_writepages(mapping, wbc, jfs_get_block);
 295 }
 296
 297 static int jfs_readpage(struct file *file, struct page *page)
 298 {
 299         return mpage_readpage(page, jfs_get_block);
 300 }
 301
 302 static int jfs_readpages(struct file *file, struct address_space *mapping,
 303                 struct list_head *pages, unsigned nr_pages)
 304 {
 305         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
 306 }
 307
 308 static void jfs_write_failed(struct address_space *mapping, loff_t to)
 309 {
 310         struct inode *inode = mapping->host;
 311
 312         if (to > inode->i_size) {
 313                 truncate_pagecache(inode, inode->i_size);
 314                 jfs_truncate(inode);
 315         }
 316 }
 317
 318 static int jfs_write_begin(struct file *file, struct address_space *mapping,
 319                                 loff_t pos, unsigned len, unsigned flags,
 320                                 struct page **pagep, void **fsdata)
 321 {
 322         int ret;
 323
 324         ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
 325                                 jfs_get_block);
 326         if (unlikely(ret))
 327                 jfs_write_failed(mapping, pos + len);
 328
 329         return ret;
 330 }
 331
 332 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
 333 {
 334         return generic_block_bmap(mapping, block, jfs_get_block);
 335 }
 336
 337 static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 338 {
 339         struct file *file = iocb->ki_filp;
 340         struct address_space *mapping = file->f_mapping;
 341         struct inode *inode = file->f_mapping->host;
 342         size_t count = iov_iter_count(iter);
 343         ssize_t ret;
 344
 345         ret = blockdev_direct_IO(iocb, inode, iter, jfs_get_block);
 346
 347         /*
 348          * In case of error extending write may have instantiated a few
 349          * blocks outside i_size. Trim these off again.
 350          */
 351         if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
 352                 loff_t isize = i_size_read(inode);
 353                 loff_t end = iocb->ki_pos + count;
 354
 355                 if (end > isize)
 356                         jfs_write_failed(mapping, end);
 357         }
 358
 359         return ret;
 360 }
 361
 362 const struct address_space_operations jfs_aops = {
 363         .readpage       = jfs_readpage,
 364         .readpages      = jfs_readpages,
 365         .writepage      = jfs_writepage,
 366         .writepages     = jfs_writepages,
 367         .write_begin    = jfs_write_begin,
 368         .write_end      = nobh_write_end,
 369         .bmap           = jfs_bmap,
 370         .direct_IO      = jfs_direct_IO,
 371 };
 372
 373 /*
 374  * Guts of jfs_truncate.  Called with locks already held.  Can be called
 375  * with directory for truncating directory index table.
 376  */
 377 void jfs_truncate_nolock(struct inode *ip, loff_t length)
 378 {
 379         loff_t newsize;
 380         tid_t tid;
 381
 382         ASSERT(length >= 0);
 383
 384         if (test_cflag(COMMIT_Nolink, ip)) {
 385                 xtTruncate(0, ip, length, COMMIT_WMAP);
 386                 return;
 387         }
 388
 389         do {
 390                 tid = txBegin(ip->i_sb, 0);
 391
 392                 /*
 393                  * The commit_mutex cannot be taken before txBegin.
 394                  * txBegin may block and there is a chance the inode
 395                  * could be marked dirty and need to be committed
 396                  * before txBegin unblocks
 397                  */
 398                 mutex_lock(&JFS_IP(ip)->commit_mutex);
 399
 400                 newsize = xtTruncate(tid, ip, length,
 401                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
 402                 if (newsize < 0) {
 403                         txEnd(tid);
 404                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
 405                         break;
 406                 }
 407
 408                 ip->i_mtime = ip->i_ctime = current_time(ip);
 409                 mark_inode_dirty(ip);
 410
 411                 txCommit(tid, 1, &ip, 0);
 412                 txEnd(tid);
 413                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
 414         } while (newsize > length);     /* Truncate isn't always atomic */
 415 }
 416
 417 void jfs_truncate(struct inode *ip)
 418 {
 419         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
 420
 421         nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
 422
 423         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 424         jfs_truncate_nolock(ip, ip->i_size);
 425         IWRITE_UNLOCK(ip);
 426 }