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 #include "jfs_dmap.h"
  35
  36
  37 struct inode *jfs_iget(struct super_block *sb, unsigned long ino)
  38 {
  39         struct inode *inode;
  40         int ret;
  41
  42         inode = iget_locked(sb, ino);
  43         if (!inode)
  44                 return ERR_PTR(-ENOMEM);
  45         if (!(inode->i_state & I_NEW))
  46                 return inode;
  47
  48         ret = diRead(inode);
  49         if (ret < 0) {
  50                 iget_failed(inode);
  51                 return ERR_PTR(ret);
  52         }
  53
  54         if (S_ISREG(inode->i_mode)) {
  55                 inode->i_op = &jfs_file_inode_operations;
  56                 inode->i_fop = &jfs_file_operations;
  57                 inode->i_mapping->a_ops = &jfs_aops;
  58         } else if (S_ISDIR(inode->i_mode)) {
  59                 inode->i_op = &jfs_dir_inode_operations;
  60                 inode->i_fop = &jfs_dir_operations;
  61         } else if (S_ISLNK(inode->i_mode)) {
  62                 if (inode->i_size >= IDATASIZE) {
  63                         inode->i_op = &page_symlink_inode_operations;
  64                         inode_nohighmem(inode);
  65                         inode->i_mapping->a_ops = &jfs_aops;
  66                 } else {
  67                         inode->i_op = &jfs_fast_symlink_inode_operations;
  68                         inode->i_link = JFS_IP(inode)->i_inline;
  69                         /*
  70                          * The inline data should be null-terminated, but
  71                          * don't let on-disk corruption crash the kernel
  72                          */
  73                         inode->i_link[inode->i_size] = '\0';
  74                 }
  75         } else {
  76                 inode->i_op = &jfs_file_inode_operations;
  77                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
  78         }
  79         unlock_new_inode(inode);
  80         return inode;
  81 }
  82
  83 /*
  84  * Workhorse of both fsync & write_inode
  85  */
  86 int jfs_commit_inode(struct inode *inode, int wait)
  87 {
  88         int rc = 0;
  89         tid_t tid;
  90         static int noisy = 5;
  91
  92         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
  93
  94         /*
  95          * Don't commit if inode has been committed since last being
  96          * marked dirty, or if it has been deleted.
  97          */
  98         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
  99                 return 0;
 100
 101         if (isReadOnly(inode)) {
 102                 /* kernel allows writes to devices on read-only
 103                  * partitions and may think inode is dirty
 104                  */
 105                 if (!special_file(inode->i_mode) && noisy) {
 106                         jfs_err("jfs_commit_inode(0x%p) called on read-only volume",
 107                                 inode);
 108                         jfs_err("Is remount racy?");
 109                         noisy--;
 110                 }
 111                 return 0;
 112         }
 113
 114         tid = txBegin(inode->i_sb, COMMIT_INODE);
 115         mutex_lock(&JFS_IP(inode)->commit_mutex);
 116
 117         /*
 118          * Retest inode state after taking commit_mutex
 119          */
 120         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
 121                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
 122
 123         txEnd(tid);
 124         mutex_unlock(&JFS_IP(inode)->commit_mutex);
 125         return rc;
 126 }
 127
 128 int jfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 129 {
 130         int wait = wbc->sync_mode == WB_SYNC_ALL;
 131
 132         if (inode->i_nlink == 0)
 133                 return 0;
 134         /*
 135          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
 136          * It has been committed since the last change, but was still
 137          * on the dirty inode list.
 138          */
 139         if (!test_cflag(COMMIT_Dirty, inode)) {
 140                 /* Make sure committed changes hit the disk */
 141                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
 142                 return 0;
 143         }
 144
 145         if (jfs_commit_inode(inode, wait)) {
 146                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
 147                 return -EIO;
 148         } else
 149                 return 0;
 150 }
 151
 152 void jfs_evict_inode(struct inode *inode)
 153 {
 154         struct jfs_inode_info *ji = JFS_IP(inode);
 155
 156         jfs_info("In jfs_evict_inode, inode = 0x%p", inode);
 157
 158         if (!inode->i_nlink && !is_bad_inode(inode)) {
 159                 dquot_initialize(inode);
 160
 161                 if (JFS_IP(inode)->fileset == FILESYSTEM_I) {
 162                         truncate_inode_pages_final(&inode->i_data);
 163
 164                         if (test_cflag(COMMIT_Freewmap, inode))
 165                                 jfs_free_zero_link(inode);
 166
 167                         diFree(inode);
 168
 169                         /*
 170                          * Free the inode from the quota allocation.
 171                          */
 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         BUG_ON(!list_empty(&ji->anon_inode_list));
 181
 182         spin_lock_irq(&ji->ag_lock);
 183         if (ji->active_ag != -1) {
 184                 struct bmap *bmap = JFS_SBI(inode->i_sb)->bmap;
 185                 atomic_dec(&bmap->db_active[ji->active_ag]);
 186                 ji->active_ag = -1;
 187         }
 188         spin_unlock_irq(&ji->ag_lock);
 189 }
 190
 191 void jfs_dirty_inode(struct inode *inode, int flags)
 192 {
 193         static int noisy = 5;
 194
 195         if (isReadOnly(inode)) {
 196                 if (!special_file(inode->i_mode) && noisy) {
 197                         /* kernel allows writes to devices on read-only
 198                          * partitions and may try to mark inode dirty
 199                          */
 200                         jfs_err("jfs_dirty_inode called on read-only volume");
 201                         jfs_err("Is remount racy?");
 202                         noisy--;
 203                 }
 204                 return;
 205         }
 206
 207         set_cflag(COMMIT_Dirty, inode);
 208 }
 209
 210 int jfs_get_block(struct inode *ip, sector_t lblock,
 211                   struct buffer_head *bh_result, int create)
 212 {
 213         s64 lblock64 = lblock;
 214         int rc = 0;
 215         xad_t xad;
 216         s64 xaddr;
 217         int xflag;
 218         s32 xlen = bh_result->b_size >> ip->i_blkbits;
 219
 220         /*
 221          * Take appropriate lock on inode
 222          */
 223         if (create)
 224                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 225         else
 226                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
 227
 228         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
 229             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
 230             xaddr) {
 231                 if (xflag & XAD_NOTRECORDED) {
 232                         if (!create)
 233                                 /*
 234                                  * Allocated but not recorded, read treats
 235                                  * this as a hole
 236                                  */
 237                                 goto unlock;
 238 #ifdef _JFS_4K
 239                         XADoffset(&xad, lblock64);
 240                         XADlength(&xad, xlen);
 241                         XADaddress(&xad, xaddr);
 242 #else                           /* _JFS_4K */
 243                         /*
 244                          * As long as block size = 4K, this isn't a problem.
 245                          * We should mark the whole page not ABNR, but how
 246                          * will we know to mark the other blocks BH_New?
 247                          */
 248                         BUG();
 249 #endif                          /* _JFS_4K */
 250                         rc = extRecord(ip, &xad);
 251                         if (rc)
 252                                 goto unlock;
 253                         set_buffer_new(bh_result);
 254                 }
 255
 256                 map_bh(bh_result, ip->i_sb, xaddr);
 257                 bh_result->b_size = xlen << ip->i_blkbits;
 258                 goto unlock;
 259         }
 260         if (!create)
 261                 goto unlock;
 262
 263         /*
 264          * Allocate a new block
 265          */
 266 #ifdef _JFS_4K
 267         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 268                 goto unlock;
 269         rc = extAlloc(ip, xlen, lblock64, &xad, false);
 270         if (rc)
 271                 goto unlock;
 272
 273         set_buffer_new(bh_result);
 274         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
 275         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
 276
 277 #else                           /* _JFS_4K */
 278         /*
 279          * We need to do whatever it takes to keep all but the last buffers
 280          * in 4K pages - see jfs_write.c
 281          */
 282         BUG();
 283 #endif                          /* _JFS_4K */
 284
 285       unlock:
 286         /*
 287          * Release lock on inode
 288          */
 289         if (create)
 290                 IWRITE_UNLOCK(ip);
 291         else
 292                 IREAD_UNLOCK(ip);
 293         return rc;
 294 }
 295
 296 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
 297 {
 298         return block_write_full_page(page, jfs_get_block, wbc);
 299 }
 300
 301 static int jfs_writepages(struct address_space *mapping,
 302                         struct writeback_control *wbc)
 303 {
 304         return mpage_writepages(mapping, wbc, jfs_get_block);
 305 }
 306
 307 static int jfs_readpage(struct file *file, struct page *page)
 308 {
 309         return mpage_readpage(page, jfs_get_block);
 310 }
 311
 312 static int jfs_readpages(struct file *file, struct address_space *mapping,
 313                 struct list_head *pages, unsigned nr_pages)
 314 {
 315         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
 316 }
 317
 318 static void jfs_write_failed(struct address_space *mapping, loff_t to)
 319 {
 320         struct inode *inode = mapping->host;
 321
 322         if (to > inode->i_size) {
 323                 truncate_pagecache(inode, inode->i_size);
 324                 jfs_truncate(inode);
 325         }
 326 }
 327
 328 static int jfs_write_begin(struct file *file, struct address_space *mapping,
 329                                 loff_t pos, unsigned len, unsigned flags,
 330                                 struct page **pagep, void **fsdata)
 331 {
 332         int ret;
 333
 334         ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
 335                                 jfs_get_block);
 336         if (unlikely(ret))
 337                 jfs_write_failed(mapping, pos + len);
 338
 339         return ret;
 340 }
 341
 342 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
 343 {
 344         return generic_block_bmap(mapping, block, jfs_get_block);
 345 }
 346
 347 static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 348 {
 349         struct file *file = iocb->ki_filp;
 350         struct address_space *mapping = file->f_mapping;
 351         struct inode *inode = file->f_mapping->host;
 352         size_t count = iov_iter_count(iter);
 353         ssize_t ret;
 354
 355         ret = blockdev_direct_IO(iocb, inode, iter, jfs_get_block);
 356
 357         /*
 358          * In case of error extending write may have instantiated a few
 359          * blocks outside i_size. Trim these off again.
 360          */
 361         if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
 362                 loff_t isize = i_size_read(inode);
 363                 loff_t end = iocb->ki_pos + count;
 364
 365                 if (end > isize)
 366                         jfs_write_failed(mapping, end);
 367         }
 368
 369         return ret;
 370 }
 371
 372 const struct address_space_operations jfs_aops = {
 373         .readpage       = jfs_readpage,
 374         .readpages      = jfs_readpages,
 375         .writepage      = jfs_writepage,
 376         .writepages     = jfs_writepages,
 377         .write_begin    = jfs_write_begin,
 378         .write_end      = nobh_write_end,
 379         .bmap           = jfs_bmap,
 380         .direct_IO      = jfs_direct_IO,
 381 };
 382
 383 /*
 384  * Guts of jfs_truncate.  Called with locks already held.  Can be called
 385  * with directory for truncating directory index table.
 386  */
 387 void jfs_truncate_nolock(struct inode *ip, loff_t length)
 388 {
 389         loff_t newsize;
 390         tid_t tid;
 391
 392         ASSERT(length >= 0);
 393
 394         if (test_cflag(COMMIT_Nolink, ip)) {
 395                 xtTruncate(0, ip, length, COMMIT_WMAP);
 396                 return;
 397         }
 398
 399         do {
 400                 tid = txBegin(ip->i_sb, 0);
 401
 402                 /*
 403                  * The commit_mutex cannot be taken before txBegin.
 404                  * txBegin may block and there is a chance the inode
 405                  * could be marked dirty and need to be committed
 406                  * before txBegin unblocks
 407                  */
 408                 mutex_lock(&JFS_IP(ip)->commit_mutex);
 409
 410                 newsize = xtTruncate(tid, ip, length,
 411                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
 412                 if (newsize < 0) {
 413                         txEnd(tid);
 414                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
 415                         break;
 416                 }
 417
 418                 ip->i_mtime = ip->i_ctime = current_time(ip);
 419                 mark_inode_dirty(ip);
 420
 421                 txCommit(tid, 1, &ip, 0);
 422                 txEnd(tid);
 423                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
 424         } while (newsize > length);     /* Truncate isn't always atomic */
 425 }
 426
 427 void jfs_truncate(struct inode *ip)
 428 {
 429         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
 430
 431         nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
 432
 433         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 434         jfs_truncate_nolock(ip, ip->i_size);
 435         IWRITE_UNLOCK(ip);
 436 }