Merge branch 'devel' of master.kernel.org:/home/rmk/linux-2.6-serial
[pandora-kernel.git] / fs / ntfs / file.c
index 7275338..ae2fe00 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * file.c - NTFS kernel file operations.  Part of the Linux-NTFS project.
  *
- * Copyright (c) 2001-2005 Anton Altaparmakov
+ * Copyright (c) 2001-2006 Anton Altaparmakov
  *
  * This program/include file is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License as published
@@ -106,7 +106,7 @@ static int ntfs_file_open(struct inode *vi, struct file *filp)
  * this is the case, the necessary zeroing will also have happened and that all
  * metadata is self-consistent.
  *
- * Locking: i_sem on the vfs inode corrseponsind to the ntfs inode @ni must be
+ * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
  *         held by the caller.
  */
 static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size,
@@ -231,8 +231,7 @@ do_non_resident_extend:
                 * Read the page.  If the page is not present, this will zero
                 * the uninitialized regions for us.
                 */
-               page = read_cache_page(mapping, index,
-                               (filler_t*)mapping->a_ops->readpage, NULL);
+               page = read_mapping_page(mapping, index, NULL);
                if (IS_ERR(page)) {
                        err = PTR_ERR(page);
                        goto init_err_out;
@@ -248,7 +247,7 @@ do_non_resident_extend:
                 * enough to make ntfs_writepage() work.
                 */
                write_lock_irqsave(&ni->size_lock, flags);
-               ni->initialized_size = (index + 1) << PAGE_CACHE_SHIFT;
+               ni->initialized_size = (s64)(index + 1) << PAGE_CACHE_SHIFT;
                if (ni->initialized_size > new_init_size)
                        ni->initialized_size = new_init_size;
                write_unlock_irqrestore(&ni->size_lock, flags);
@@ -473,7 +472,7 @@ static inline int ntfs_submit_bh_for_read(struct buffer_head *bh)
  * @bytes:     number of bytes to be written
  *
  * This is called for non-resident attributes from ntfs_file_buffered_write()
- * with i_sem held on the inode (@pages[0]->mapping->host).  There are
+ * with i_mutex held on the inode (@pages[0]->mapping->host).  There are
  * @nr_pages pages in @pages which are locked but not kmap()ped.  The source
  * data has not yet been copied into the @pages.
  * 
@@ -510,7 +509,7 @@ static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
        u32 attr_rec_len = 0;
        unsigned blocksize, u;
        int err, mp_size;
-       BOOL rl_write_locked, was_hole, is_retry;
+       bool rl_write_locked, was_hole, is_retry;
        unsigned char blocksize_bits;
        struct {
                u8 runlist_merged:1;
@@ -529,8 +528,8 @@ static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
                        "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
                        vi->i_ino, ni->type, pages[0]->index, nr_pages,
                        (long long)pos, bytes);
-       blocksize_bits = vi->i_blkbits;
-       blocksize = 1 << blocksize_bits;
+       blocksize = vol->sb->s_blocksize;
+       blocksize_bits = vol->sb->s_blocksize_bits;
        u = 0;
        do {
                struct page *page = pages[u];
@@ -544,13 +543,13 @@ static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
                                return -ENOMEM;
                }
        } while (++u < nr_pages);
-       rl_write_locked = FALSE;
+       rl_write_locked = false;
        rl = NULL;
        err = 0;
        vcn = lcn = -1;
        vcn_len = 0;
        lcn_block = -1;
-       was_hole = FALSE;
+       was_hole = false;
        cpos = pos >> vol->cluster_size_bits;
        end = pos + bytes;
        cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits;
@@ -761,7 +760,7 @@ map_buffer_cached:
                        }
                        continue;
                }
-               is_retry = FALSE;
+               is_retry = false;
                if (!rl) {
                        down_read(&ni->runlist.lock);
 retry_remap:
@@ -777,7 +776,7 @@ retry_remap:
                                 * Successful remap, setup the map cache and
                                 * use that to deal with the buffer.
                                 */
-                               was_hole = FALSE;
+                               was_hole = false;
                                vcn = bh_cpos;
                                vcn_len = rl[1].vcn - vcn;
                                lcn_block = lcn << (vol->cluster_size_bits -
@@ -793,7 +792,7 @@ retry_remap:
                                if (likely(vcn + vcn_len >= cend)) {
                                        if (rl_write_locked) {
                                                up_write(&ni->runlist.lock);
-                                               rl_write_locked = FALSE;
+                                               rl_write_locked = false;
                                        } else
                                                up_read(&ni->runlist.lock);
                                        rl = NULL;
@@ -819,13 +818,13 @@ retry_remap:
                                         */
                                        up_read(&ni->runlist.lock);
                                        down_write(&ni->runlist.lock);
-                                       rl_write_locked = TRUE;
+                                       rl_write_locked = true;
                                        goto retry_remap;
                                }
                                err = ntfs_map_runlist_nolock(ni, bh_cpos,
                                                NULL);
                                if (likely(!err)) {
-                                       is_retry = TRUE;
+                                       is_retry = true;
                                        goto retry_remap;
                                }
                                /*
@@ -904,7 +903,7 @@ rl_not_mapped_enoent:
                if (!rl_write_locked) {
                        up_read(&ni->runlist.lock);
                        down_write(&ni->runlist.lock);
-                       rl_write_locked = TRUE;
+                       rl_write_locked = true;
                        goto retry_remap;
                }
                /* Find the previous last allocated cluster. */
@@ -918,7 +917,7 @@ rl_not_mapped_enoent:
                        }
                }
                rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE,
-                               FALSE);
+                               false);
                if (IS_ERR(rl2)) {
                        err = PTR_ERR(rl2);
                        ntfs_debug("Failed to allocate cluster, error code %i.",
@@ -943,7 +942,8 @@ rl_not_mapped_enoent:
                }
                ni->runlist.rl = rl;
                status.runlist_merged = 1;
-               ntfs_debug("Allocated cluster, lcn 0x%llx.", lcn);
+               ntfs_debug("Allocated cluster, lcn 0x%llx.",
+                               (unsigned long long)lcn);
                /* Map and lock the mft record and get the attribute record. */
                if (!NInoAttr(ni))
                        base_ni = ni;
@@ -1093,7 +1093,7 @@ rl_not_mapped_enoent:
                status.mft_attr_mapped = 0;
                status.mp_rebuilt = 0;
                /* Setup the map cache and use that to deal with the buffer. */
-               was_hole = TRUE;
+               was_hole = true;
                vcn = bh_cpos;
                vcn_len = 1;
                lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits);
@@ -1105,7 +1105,7 @@ rl_not_mapped_enoent:
                 */
                if (likely(vcn + vcn_len >= cend)) {
                        up_write(&ni->runlist.lock);
-                       rl_write_locked = FALSE;
+                       rl_write_locked = false;
                        rl = NULL;
                }
                goto map_buffer_cached;
@@ -1117,7 +1117,7 @@ rl_not_mapped_enoent:
        if (likely(!err)) {
                if (unlikely(rl_write_locked)) {
                        up_write(&ni->runlist.lock);
-                       rl_write_locked = FALSE;
+                       rl_write_locked = false;
                } else if (unlikely(rl))
                        up_read(&ni->runlist.lock);
                rl = NULL;
@@ -1206,8 +1206,6 @@ rl_not_mapped_enoent:
                                        "attribute runlist in error code "
                                        "path.  Run chkdsk to recover the "
                                        "lost cluster.");
-                       make_bad_inode(vi);
-                       make_bad_inode(VFS_I(base_ni));
                        NVolSetErrors(vol);
                } else /* if (success) */ {
                        status.runlist_merged = 0;
@@ -1238,8 +1236,6 @@ rl_not_mapped_enoent:
                        ntfs_error(vol->sb, "Failed to restore attribute "
                                        "record in error code path.  Run "
                                        "chkdsk to recover.");
-                       make_bad_inode(vi);
-                       make_bad_inode(VFS_I(base_ni));
                        NVolSetErrors(vol);
                } else /* if (success) */ {
                        if (ntfs_mapping_pairs_build(vol, (u8*)a +
@@ -1252,8 +1248,6 @@ rl_not_mapped_enoent:
                                                "mapping pairs array in error "
                                                "code path.  Run chkdsk to "
                                                "recover.");
-                               make_bad_inode(vi);
-                               make_bad_inode(VFS_I(base_ni));
                                NVolSetErrors(vol);
                        }
                        flush_dcache_mft_record_page(ctx->ntfs_ino);
@@ -1364,7 +1358,7 @@ err_out:
        goto out;
 }
 
-static size_t __ntfs_copy_from_user_iovec(char *vaddr,
+static size_t __ntfs_copy_from_user_iovec_inatomic(char *vaddr,
                const struct iovec *iov, size_t iov_ofs, size_t bytes)
 {
        size_t total = 0;
@@ -1382,10 +1376,6 @@ static size_t __ntfs_copy_from_user_iovec(char *vaddr,
                bytes -= len;
                vaddr += len;
                if (unlikely(left)) {
-                       /*
-                        * Zero the rest of the target like __copy_from_user().
-                        */
-                       memset(vaddr, 0, bytes);
                        total -= left;
                        break;
                }
@@ -1426,11 +1416,13 @@ static inline void ntfs_set_next_iovec(const struct iovec **iovp,
  * pages (out to offset + bytes), to emulate ntfs_copy_from_user()'s
  * single-segment behaviour.
  *
- * We call the same helper (__ntfs_copy_from_user_iovec()) both when atomic and
- * when not atomic.  This is ok because __ntfs_copy_from_user_iovec() calls
- * __copy_from_user_inatomic() and it is ok to call this when non-atomic.  In
- * fact, the only difference between __copy_from_user_inatomic() and
- * __copy_from_user() is that the latter calls might_sleep().  And on many
+ * We call the same helper (__ntfs_copy_from_user_iovec_inatomic()) both
+ * when atomic and when not atomic.  This is ok because
+ * __ntfs_copy_from_user_iovec_inatomic() calls __copy_from_user_inatomic()
+ * and it is ok to call this when non-atomic.
+ * Infact, the only difference between __copy_from_user_inatomic() and
+ * __copy_from_user() is that the latter calls might_sleep() and the former
+ * should not zero the tail of the buffer on error.  And on many
  * architectures __copy_from_user_inatomic() is just defined to
  * __copy_from_user() so it makes no difference at all on those architectures.
  */
@@ -1447,14 +1439,18 @@ static inline size_t ntfs_copy_from_user_iovec(struct page **pages,
                if (len > bytes)
                        len = bytes;
                kaddr = kmap_atomic(*pages, KM_USER0);
-               copied = __ntfs_copy_from_user_iovec(kaddr + ofs,
+               copied = __ntfs_copy_from_user_iovec_inatomic(kaddr + ofs,
                                *iov, *iov_ofs, len);
                kunmap_atomic(kaddr, KM_USER0);
                if (unlikely(copied != len)) {
                        /* Do it the slow way. */
                        kaddr = kmap(*pages);
-                       copied = __ntfs_copy_from_user_iovec(kaddr + ofs,
+                       copied = __ntfs_copy_from_user_iovec_inatomic(kaddr + ofs,
                                        *iov, *iov_ofs, len);
+                       /*
+                        * Zero the rest of the target like __copy_from_user().
+                        */
+                       memset(kaddr + ofs + copied, 0, len - copied);
                        kunmap(*pages);
                        if (unlikely(copied != len))
                                goto err_out;
@@ -1489,14 +1485,15 @@ static inline void ntfs_flush_dcache_pages(struct page **pages,
                unsigned nr_pages)
 {
        BUG_ON(!nr_pages);
+       /*
+        * Warning: Do not do the decrement at the same time as the call to
+        * flush_dcache_page() because it is a NULL macro on i386 and hence the
+        * decrement never happens so the loop never terminates.
+        */
        do {
-               /*
-                * Warning: Do not do the decrement at the same time as the
-                * call because flush_dcache_page() is a NULL macro on i386
-                * and hence the decrement never happens.
-                */
+               --nr_pages;
                flush_dcache_page(pages[nr_pages]);
-       } while (--nr_pages > 0);
+       } while (nr_pages > 0);
 }
 
 /**
@@ -1525,25 +1522,25 @@ static inline int ntfs_commit_pages_after_non_resident_write(
 
        vi = pages[0]->mapping->host;
        ni = NTFS_I(vi);
-       blocksize = 1 << vi->i_blkbits;
+       blocksize = vi->i_sb->s_blocksize;
        end = pos + bytes;
        u = 0;
        do {
                s64 bh_pos;
                struct page *page;
-               BOOL partial;
+               bool partial;
 
                page = pages[u];
                bh_pos = (s64)page->index << PAGE_CACHE_SHIFT;
                bh = head = page_buffers(page);
-               partial = FALSE;
+               partial = false;
                do {
                        s64 bh_end;
 
                        bh_end = bh_pos + blocksize;
                        if (bh_end <= pos || bh_pos >= end) {
                                if (!buffer_uptodate(bh))
-                                       partial = TRUE;
+                                       partial = true;
                        } else {
                                set_buffer_uptodate(bh);
                                mark_buffer_dirty(bh);
@@ -1622,11 +1619,8 @@ err_out:
                unmap_mft_record(base_ni);
        ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error "
                        "code %i).", err);
-       if (err != -ENOMEM) {
+       if (err != -ENOMEM)
                NVolSetErrors(ni->vol);
-               make_bad_inode(VFS_I(base_ni));
-               make_bad_inode(vi);
-       }
        return err;
 }
 
@@ -1637,7 +1631,7 @@ err_out:
  * @pos:       byte position in file at which the write begins
  * @bytes:     number of bytes to be written
  *
- * This is called from ntfs_file_buffered_write() with i_sem held on the inode
+ * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
  * (@pages[0]->mapping->host).  There are @nr_pages pages in @pages which are
  * locked but not kmap()ped.  The source data has already been copied into the
  * @page.  ntfs_prepare_pages_for_non_resident_write() has been called before
@@ -1801,8 +1795,6 @@ err_out:
                ntfs_error(vi->i_sb, "Resident attribute commit write failed "
                                "with error %i.", err);
                NVolSetErrors(ni->vol);
-               make_bad_inode(VFS_I(base_ni));
-               make_bad_inode(vi);
        }
        if (ctx)
                ntfs_attr_put_search_ctx(ctx);
@@ -1814,7 +1806,7 @@ err_out:
 /**
  * ntfs_file_buffered_write -
  *
- * Locking: The vfs is holding ->i_sem on the inode.
+ * Locking: The vfs is holding ->i_mutex on the inode.
  */
 static ssize_t ntfs_file_buffered_write(struct kiocb *iocb,
                const struct iovec *iov, unsigned long nr_segs,
@@ -2005,7 +1997,7 @@ static ssize_t ntfs_file_buffered_write(struct kiocb *iocb,
                                 */
                                down_read(&ni->runlist.lock);
                                lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >>
-                                               vol->cluster_size_bits, FALSE);
+                                               vol->cluster_size_bits, false);
                                up_read(&ni->runlist.lock);
                                if (unlikely(lcn < LCN_HOLE)) {
                                        status = -EIO;
@@ -2173,7 +2165,7 @@ static ssize_t ntfs_file_aio_write_nolock(struct kiocb *iocb,
        err = remove_suid(file->f_dentry);
        if (err)
                goto out;
-       inode_update_time(inode, 1);
+       file_update_time(file);
        written = ntfs_file_buffered_write(iocb, iov, nr_segs, pos, ppos,
                        count);
 out:
@@ -2184,21 +2176,19 @@ out:
 /**
  * ntfs_file_aio_write -
  */
-static ssize_t ntfs_file_aio_write(struct kiocb *iocb, const char __user *buf,
-               size_t count, loff_t pos)
+static ssize_t ntfs_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
+               unsigned long nr_segs, loff_t pos)
 {
        struct file *file = iocb->ki_filp;
        struct address_space *mapping = file->f_mapping;
        struct inode *inode = mapping->host;
        ssize_t ret;
-       struct iovec local_iov = { .iov_base = (void __user *)buf,
-                                  .iov_len = count };
 
        BUG_ON(iocb->ki_pos != pos);
 
-       down(&inode->i_sem);
-       ret = ntfs_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
-       up(&inode->i_sem);
+       mutex_lock(&inode->i_mutex);
+       ret = ntfs_file_aio_write_nolock(iocb, iov, nr_segs, &iocb->ki_pos);
+       mutex_unlock(&inode->i_mutex);
        if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
                int err = sync_page_range(inode, mapping, pos, ret);
                if (err < 0)
@@ -2221,12 +2211,12 @@ static ssize_t ntfs_file_writev(struct file *file, const struct iovec *iov,
        struct kiocb kiocb;
        ssize_t ret;
 
-       down(&inode->i_sem);
+       mutex_lock(&inode->i_mutex);
        init_sync_kiocb(&kiocb, file);
        ret = ntfs_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
        if (ret == -EIOCBQUEUED)
                ret = wait_on_sync_kiocb(&kiocb);
-       up(&inode->i_sem);
+       mutex_unlock(&inode->i_mutex);
        if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
                int err = sync_page_range(inode, mapping, *ppos - ret, ret);
                if (err < 0)
@@ -2269,7 +2259,7 @@ static ssize_t ntfs_file_write(struct file *file, const char __user *buf,
  * Note: In the past @filp could be NULL so we ignore it as we don't need it
  * anyway.
  *
- * Locking: Caller must hold i_sem on the inode.
+ * Locking: Caller must hold i_mutex on the inode.
  *
  * TODO: We should probably also write all attribute/index inodes associated
  * with this inode but since we have no simple way of getting to them we ignore
@@ -2304,15 +2294,13 @@ static int ntfs_file_fsync(struct file *filp, struct dentry *dentry,
 
 #endif /* NTFS_RW */
 
-struct file_operations ntfs_file_ops = {
+const struct file_operations ntfs_file_ops = {
        .llseek         = generic_file_llseek,   /* Seek inside file. */
-       .read           = generic_file_read,     /* Read from file. */
+       .read           = do_sync_read,          /* Read from file. */
        .aio_read       = generic_file_aio_read, /* Async read from file. */
-       .readv          = generic_file_readv,    /* Read from file. */
 #ifdef NTFS_RW
        .write          = ntfs_file_write,       /* Write to file. */
        .aio_write      = ntfs_file_aio_write,   /* Async write to file. */
-       .writev         = ntfs_file_writev,      /* Write to file. */
        /*.release      = ,*/                    /* Last file is closed.  See
                                                    fs/ext2/file.c::
                                                    ext2_release_file() for
@@ -2347,6 +2335,6 @@ struct inode_operations ntfs_file_inode_ops = {
 #endif /* NTFS_RW */
 };
 
-struct file_operations ntfs_empty_file_ops = {};
+const struct file_operations ntfs_empty_file_ops = {};
 
 struct inode_operations ntfs_empty_inode_ops = {};