fs/reiserfs/file.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/reiserfs_fs.h>
   7 #include <linux/reiserfs_acl.h>
   8 #include <linux/reiserfs_xattr.h>
   9 #include <asm/uaccess.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/swap.h>
  12 #include <linux/writeback.h>
  13 #include <linux/blkdev.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/quotaops.h>
  16
  17 /*
  18 ** We pack the tails of files on file close, not at the time they are written.
  19 ** This implies an unnecessary copy of the tail and an unnecessary indirect item
  20 ** insertion/balancing, for files that are written in one write.
  21 ** It avoids unnecessary tail packings (balances) for files that are written in
  22 ** multiple writes and are small enough to have tails.
  23 **
  24 ** file_release is called by the VFS layer when the file is closed.  If
  25 ** this is the last open file descriptor, and the file
  26 ** small enough to have a tail, and the tail is currently in an
  27 ** unformatted node, the tail is converted back into a direct item.
  28 **
  29 ** We use reiserfs_truncate_file to pack the tail, since it already has
  30 ** all the conditions coded.
  31 */
  32 static int reiserfs_file_release(struct inode *inode, struct file *filp)
  33 {
  34
  35         struct reiserfs_transaction_handle th;
  36         int err;
  37         int jbegin_failure = 0;
  38
  39         BUG_ON(!S_ISREG(inode->i_mode));
  40
  41         /* fast out for when nothing needs to be done */
  42         if ((atomic_read(&inode->i_count) > 1 ||
  43              !(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) ||
  44              !tail_has_to_be_packed(inode)) &&
  45             REISERFS_I(inode)->i_prealloc_count <= 0) {
  46                 return 0;
  47         }
  48
  49         mutex_lock(&inode->i_mutex);
  50
  51         mutex_lock(&(REISERFS_I(inode)->i_mmap));
  52         if (REISERFS_I(inode)->i_flags & i_ever_mapped)
  53                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
  54
  55         reiserfs_write_lock(inode->i_sb);
  56         /* freeing preallocation only involves relogging blocks that
  57          * are already in the current transaction.  preallocation gets
  58          * freed at the end of each transaction, so it is impossible for
  59          * us to log any additional blocks (including quota blocks)
  60          */
  61         err = journal_begin(&th, inode->i_sb, 1);
  62         if (err) {
  63                 /* uh oh, we can't allow the inode to go away while there
  64                  * is still preallocation blocks pending.  Try to join the
  65                  * aborted transaction
  66                  */
  67                 jbegin_failure = err;
  68                 err = journal_join_abort(&th, inode->i_sb, 1);
  69
  70                 if (err) {
  71                         /* hmpf, our choices here aren't good.  We can pin the inode
  72                          * which will disallow unmount from every happening, we can
  73                          * do nothing, which will corrupt random memory on unmount,
  74                          * or we can forcibly remove the file from the preallocation
  75                          * list, which will leak blocks on disk.  Lets pin the inode
  76                          * and let the admin know what is going on.
  77                          */
  78                         igrab(inode);
  79                         reiserfs_warning(inode->i_sb,
  80                                          "pinning inode %lu because the "
  81                                          "preallocation can't be freed",
  82                                          inode->i_ino);
  83                         goto out;
  84                 }
  85         }
  86         reiserfs_update_inode_transaction(inode);
  87
  88 #ifdef REISERFS_PREALLOCATE
  89         reiserfs_discard_prealloc(&th, inode);
  90 #endif
  91         err = journal_end(&th, inode->i_sb, 1);
  92
  93         /* copy back the error code from journal_begin */
  94         if (!err)
  95                 err = jbegin_failure;
  96
  97         if (!err && atomic_read(&inode->i_count) <= 1 &&
  98             (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) &&
  99             tail_has_to_be_packed(inode)) {
 100                 /* if regular file is released by last holder and it has been
 101                    appended (we append by unformatted node only) or its direct
 102                    item(s) had to be converted, then it may have to be
 103                    indirect2direct converted */
 104                 err = reiserfs_truncate_file(inode, 0);
 105         }
 106       out:
 107         mutex_unlock(&(REISERFS_I(inode)->i_mmap));
 108         mutex_unlock(&inode->i_mutex);
 109         reiserfs_write_unlock(inode->i_sb);
 110         return err;
 111 }
 112
 113 static int reiserfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 114 {
 115         struct inode *inode;
 116
 117         inode = file->f_path.dentry->d_inode;
 118         mutex_lock(&(REISERFS_I(inode)->i_mmap));
 119         REISERFS_I(inode)->i_flags |= i_ever_mapped;
 120         mutex_unlock(&(REISERFS_I(inode)->i_mmap));
 121
 122         return generic_file_mmap(file, vma);
 123 }
 124
 125 static void reiserfs_vfs_truncate_file(struct inode *inode)
 126 {
 127         reiserfs_truncate_file(inode, 1);
 128 }
 129
 130 /* Sync a reiserfs file. */
 131
 132 /*
 133  * FIXME: sync_mapping_buffers() never has anything to sync.  Can
 134  * be removed...
 135  */
 136
 137 static int reiserfs_sync_file(struct file *p_s_filp,
 138                               struct dentry *p_s_dentry, int datasync)
 139 {
 140         struct inode *p_s_inode = p_s_dentry->d_inode;
 141         int n_err;
 142         int barrier_done;
 143
 144         BUG_ON(!S_ISREG(p_s_inode->i_mode));
 145         n_err = sync_mapping_buffers(p_s_inode->i_mapping);
 146         reiserfs_write_lock(p_s_inode->i_sb);
 147         barrier_done = reiserfs_commit_for_inode(p_s_inode);
 148         reiserfs_write_unlock(p_s_inode->i_sb);
 149         if (barrier_done != 1 && reiserfs_barrier_flush(p_s_inode->i_sb))
 150                 blkdev_issue_flush(p_s_inode->i_sb->s_bdev, NULL);
 151         if (barrier_done < 0)
 152                 return barrier_done;
 153         return (n_err < 0) ? -EIO : 0;
 154 }
 155
 156 /* taken fs/buffer.c:__block_commit_write */
 157 int reiserfs_commit_page(struct inode *inode, struct page *page,
 158                          unsigned from, unsigned to)
 159 {
 160         unsigned block_start, block_end;
 161         int partial = 0;
 162         unsigned blocksize;
 163         struct buffer_head *bh, *head;
 164         unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT;
 165         int new;
 166         int logit = reiserfs_file_data_log(inode);
 167         struct super_block *s = inode->i_sb;
 168         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
 169         struct reiserfs_transaction_handle th;
 170         int ret = 0;
 171
 172         th.t_trans_id = 0;
 173         blocksize = 1 << inode->i_blkbits;
 174
 175         if (logit) {
 176                 reiserfs_write_lock(s);
 177                 ret = journal_begin(&th, s, bh_per_page + 1);
 178                 if (ret)
 179                         goto drop_write_lock;
 180                 reiserfs_update_inode_transaction(inode);
 181         }
 182         for (bh = head = page_buffers(page), block_start = 0;
 183              bh != head || !block_start;
 184              block_start = block_end, bh = bh->b_this_page) {
 185
 186                 new = buffer_new(bh);
 187                 clear_buffer_new(bh);
 188                 block_end = block_start + blocksize;
 189                 if (block_end <= from || block_start >= to) {
 190                         if (!buffer_uptodate(bh))
 191                                 partial = 1;
 192                 } else {
 193                         set_buffer_uptodate(bh);
 194                         if (logit) {
 195                                 reiserfs_prepare_for_journal(s, bh, 1);
 196                                 journal_mark_dirty(&th, s, bh);
 197                         } else if (!buffer_dirty(bh)) {
 198                                 mark_buffer_dirty(bh);
 199                                 /* do data=ordered on any page past the end
 200                                  * of file and any buffer marked BH_New.
 201                                  */
 202                                 if (reiserfs_data_ordered(inode->i_sb) &&
 203                                     (new || page->index >= i_size_index)) {
 204                                         reiserfs_add_ordered_list(inode, bh);
 205                                 }
 206                         }
 207                 }
 208         }
 209         if (logit) {
 210                 ret = journal_end(&th, s, bh_per_page + 1);
 211               drop_write_lock:
 212                 reiserfs_write_unlock(s);
 213         }
 214         /*
 215          * If this is a partial write which happened to make all buffers
 216          * uptodate then we can optimize away a bogus readpage() for
 217          * the next read(). Here we 'discover' whether the page went
 218          * uptodate as a result of this (potentially partial) write.
 219          */
 220         if (!partial)
 221                 SetPageUptodate(page);
 222         return ret;
 223 }
 224
 225 /* Write @count bytes at position @ppos in a file indicated by @file
 226    from the buffer @buf.
 227
 228    generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want
 229    something simple that works.  It is not for serious use by general purpose filesystems, excepting the one that it was
 230    written for (ext2/3).  This is for several reasons:
 231
 232    * It has no understanding of any filesystem specific optimizations.
 233
 234    * It enters the filesystem repeatedly for each page that is written.
 235
 236    * It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key
 237    * operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time
 238    * to reiserfs which allows for fewer tree traversals.
 239
 240    * Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks.
 241
 242    * Asking the block allocation code for blocks one at a time is slightly less efficient.
 243
 244    All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to
 245    use it, but we were in a hurry to make code freeze, and so it couldn't be revised then.  This new code should make
 246    things right finally.
 247
 248    Future Features: providing search_by_key with hints.
 249
 250 */
 251 static ssize_t reiserfs_file_write(struct file *file,   /* the file we are going to write into */
 252                                    const char __user * buf,     /*  pointer to user supplied data
 253                                                                    (in userspace) */
 254                                    size_t count,        /* amount of bytes to write */
 255                                    loff_t * ppos        /* pointer to position in file that we start writing at. Should be updated to
 256                                                          * new current position before returning. */
 257                                    )
 258 {
 259         struct inode *inode = file->f_path.dentry->d_inode;     // Inode of the file that we are writing to.
 260         /* To simplify coding at this time, we store
 261            locked pages in array for now */
 262         struct reiserfs_transaction_handle th;
 263         th.t_trans_id = 0;
 264
 265         /* If a filesystem is converted from 3.5 to 3.6, we'll have v3.5 items
 266         * lying around (most of the disk, in fact). Despite the filesystem
 267         * now being a v3.6 format, the old items still can't support large
 268         * file sizes. Catch this case here, as the rest of the VFS layer is
 269         * oblivious to the different limitations between old and new items.
 270         * reiserfs_setattr catches this for truncates. This chunk is lifted
 271         * from generic_write_checks. */
 272         if (get_inode_item_key_version (inode) == KEY_FORMAT_3_5 &&
 273             *ppos + count > MAX_NON_LFS) {
 274                 if (*ppos >= MAX_NON_LFS) {
 275                         return -EFBIG;
 276                 }
 277                 if (count > MAX_NON_LFS - (unsigned long)*ppos)
 278                         count = MAX_NON_LFS - (unsigned long)*ppos;
 279         }
 280
 281         return do_sync_write(file, buf, count, ppos);
 282 }
 283
 284 const struct file_operations reiserfs_file_operations = {
 285         .read = do_sync_read,
 286         .write = reiserfs_file_write,
 287         .ioctl = reiserfs_ioctl,
 288 #ifdef CONFIG_COMPAT
 289         .compat_ioctl = reiserfs_compat_ioctl,
 290 #endif
 291         .mmap = reiserfs_file_mmap,
 292         .open = generic_file_open,
 293         .release = reiserfs_file_release,
 294         .fsync = reiserfs_sync_file,
 295         .aio_read = generic_file_aio_read,
 296         .aio_write = generic_file_aio_write,
 297         .splice_read = generic_file_splice_read,
 298         .splice_write = generic_file_splice_write,
 299 };
 300
 301 const struct inode_operations reiserfs_file_inode_operations = {
 302         .truncate = reiserfs_vfs_truncate_file,
 303         .setattr = reiserfs_setattr,
 304         .setxattr = reiserfs_setxattr,
 305         .getxattr = reiserfs_getxattr,
 306         .listxattr = reiserfs_listxattr,
 307         .removexattr = reiserfs_removexattr,
 308         .permission = reiserfs_permission,
 309 };