/**
* The little endian Unicode string $I30 as a global constant.
*/
-uchar_t I30[5] = { const_cpu_to_le16('$'), const_cpu_to_le16('I'),
+ntfschar I30[5] = { const_cpu_to_le16('$'), const_cpu_to_le16('I'),
const_cpu_to_le16('3'), const_cpu_to_le16('0'),
const_cpu_to_le16(0) };
* This is to avoid polluting the dcache with short file names. We want them to
* work but we don't care for how quickly one can access them. This also fixes
* the dcache aliasing issues.
+ *
+ * Locking: - Caller must hold i_sem on the directory.
+ * - Each page cache page in the index allocation mapping must be
+ * locked whilst being accessed otherwise we may find a corrupt
+ * page due to it being under ->writepage at the moment which
+ * applies the mst protection fixups before writing out and then
+ * removes them again after the write is complete after which it
+ * unlocks the page.
*/
-MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname,
+MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const ntfschar *uname,
const int uname_len, ntfs_name **res)
{
ntfs_volume *vol = dir_ni->vol;
u8 *kaddr;
ntfs_name *name = NULL;
+ BUG_ON(!S_ISDIR(VFS_I(dir_ni)->i_mode));
+ BUG_ON(NInoAttr(dir_ni));
/* Get hold of the mft record for the directory. */
m = map_mft_record(dir_ni);
if (unlikely(IS_ERR(m))) {
* returning.
*/
if (ntfs_are_names_equal(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length,
CASE_SENSITIVE, vol->upcase, vol->upcase_len)) {
found_it:
if (!NVolCaseSensitive(vol) &&
ie->key.file_name.file_name_type &&
ntfs_are_names_equal(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length,
IGNORE_CASE, vol->upcase, vol->upcase_len)) {
int name_size = sizeof(ntfs_name);
}
if (type != FILE_NAME_DOS)
- name_size += len * sizeof(uchar_t);
+ name_size += len * sizeof(ntfschar);
name = kmalloc(name_size, GFP_NOFS);
if (!name) {
err = -ENOMEM;
if (type != FILE_NAME_DOS) {
name->len = len;
memcpy(name->name, ie->key.file_name.file_name,
- len * sizeof(uchar_t));
+ len * sizeof(ntfschar));
} else
name->len = 0;
*res = name;
* know which way in the B+tree we have to go.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
IGNORE_CASE, vol->upcase, vol->upcase_len);
/*
* collation.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
CASE_SENSITIVE, vol->upcase, vol->upcase_len);
if (rc == -1)
err = PTR_ERR(page);
goto err_out;
}
+ lock_page(page);
kaddr = (u8*)page_address(page);
fast_descend_into_child_node:
/* Get to the index allocation block. */
* returning.
*/
if (ntfs_are_names_equal(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length,
CASE_SENSITIVE, vol->upcase, vol->upcase_len)) {
found_it2:
*res = NULL;
}
mref = le64_to_cpu(ie->data.dir.indexed_file);
+ unlock_page(page);
ntfs_unmap_page(page);
return mref;
}
if (!NVolCaseSensitive(vol) &&
ie->key.file_name.file_name_type &&
ntfs_are_names_equal(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length,
IGNORE_CASE, vol->upcase, vol->upcase_len)) {
int name_size = sizeof(ntfs_name);
"this message to "
"linux-ntfs-dev@lists."
"sourceforge.net.");
+ unlock_page(page);
ntfs_unmap_page(page);
goto dir_err_out;
}
if (type != FILE_NAME_DOS)
- name_size += len * sizeof(uchar_t);
+ name_size += len * sizeof(ntfschar);
name = kmalloc(name_size, GFP_NOFS);
if (!name) {
err = -ENOMEM;
if (type != FILE_NAME_DOS) {
name->len = len;
memcpy(name->name, ie->key.file_name.file_name,
- len * sizeof(uchar_t));
+ len * sizeof(ntfschar));
} else
name->len = 0;
*res = name;
* know which way in the B+tree we have to go.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
IGNORE_CASE, vol->upcase, vol->upcase_len);
/*
* collation.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
CASE_SENSITIVE, vol->upcase, vol->upcase_len);
if (rc == -1)
vol->cluster_size_bits >>
PAGE_CACHE_SHIFT)
goto fast_descend_into_child_node;
+ unlock_page(page);
ntfs_unmap_page(page);
goto descend_into_child_node;
}
* associated with it.
*/
if (name) {
+ unlock_page(page);
ntfs_unmap_page(page);
return name->mref;
}
ntfs_debug("Entry not found.");
err = -ENOENT;
unm_err_out:
+ unlock_page(page);
ntfs_unmap_page(page);
err_out:
if (ctx)
*
* Note, @uname_len does not include the (optional) terminating NULL character.
*/
-u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname,
+u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const ntfschar *uname,
const int uname_len)
{
ntfs_volume *vol = dir_ni->vol;
* convert it to cpu format before returning.
*/
if (ntfs_are_names_equal(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, ic,
vol->upcase, vol->upcase_len)) {
found_it:
* know which way in the B+tree we have to go.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
IGNORE_CASE, vol->upcase, vol->upcase_len);
/*
* collation.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
CASE_SENSITIVE, vol->upcase, vol->upcase_len);
if (rc == -1)
err = PTR_ERR(page);
goto err_out;
}
+ lock_page(page);
kaddr = (u8*)page_address(page);
fast_descend_into_child_node:
/* Get to the index allocation block. */
* convert it to cpu format before returning.
*/
if (ntfs_are_names_equal(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, ic,
vol->upcase, vol->upcase_len)) {
found_it2:
mref = le64_to_cpu(ie->data.dir.indexed_file);
+ unlock_page(page);
ntfs_unmap_page(page);
return mref;
}
* know which way in the B+tree we have to go.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
IGNORE_CASE, vol->upcase, vol->upcase_len);
/*
* collation.
*/
rc = ntfs_collate_names(uname, uname_len,
- (uchar_t*)&ie->key.file_name.file_name,
+ (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, 1,
CASE_SENSITIVE, vol->upcase, vol->upcase_len);
if (rc == -1)
vol->cluster_size_bits >>
PAGE_CACHE_SHIFT)
goto fast_descend_into_child_node;
+ unlock_page(page);
ntfs_unmap_page(page);
goto descend_into_child_node;
}
ntfs_debug("Entry not found.");
err = -ENOENT;
unm_err_out:
+ unlock_page(page);
ntfs_unmap_page(page);
err_out:
if (ctx)
* @ndir: ntfs inode of current directory
* @index_type: specifies whether @iu is an index root or an index allocation
* @iu: index root or index allocation attribute to which @ie belongs
+ * @ia_page: page in which the index allocation buffer @ie is in resides
* @ie: current index entry
* @name: buffer to use for the converted name
* @dirent: vfs filldir callback context
*
* Convert the Unicode @name to the loaded NLS and pass it to the @filldir
* callback.
+ *
+ * If @index_type is INDEX_TYPE_ALLOCATION, @ia_page is the locked page
+ * containing the index allocation block containing the index entry @ie.
+ * Otherwise, @ia_page is NULL.
+ *
+ * Note, we drop (and then reacquire) the page lock on @ia_page across the
+ * @filldir() call otherwise we would deadlock with NFSd when it calls ->lookup
+ * since ntfs_lookup() will lock the same page. As an optimization, we do not
+ * retake the lock if we are returning a non-zero value as ntfs_readdir()
+ * would need to drop the lock immediately anyway.
*/
static inline int ntfs_filldir(ntfs_volume *vol, loff_t *fpos,
ntfs_inode *ndir, const INDEX_TYPE index_type,
- index_union iu, INDEX_ENTRY *ie, u8 *name,
- void *dirent, filldir_t filldir)
+ index_union iu, struct page *ia_page, INDEX_ENTRY *ie,
+ u8 *name, void *dirent, filldir_t filldir)
{
- int name_len;
+ unsigned long mref;
+ int name_len, rc;
unsigned dt_type;
FILE_NAME_TYPE_FLAGS name_type;
ntfs_debug("Skipping system file.");
return 0;
}
- name_len = ntfs_ucstonls(vol, (uchar_t*)&ie->key.file_name.file_name,
+ name_len = ntfs_ucstonls(vol, (ntfschar*)&ie->key.file_name.file_name,
ie->key.file_name.file_name_length, &name,
NTFS_MAX_NAME_LEN * NLS_MAX_CHARSET_SIZE + 1);
if (name_len <= 0) {
dt_type = DT_DIR;
else
dt_type = DT_REG;
+ mref = MREF_LE(ie->data.dir.indexed_file);
+ /*
+ * Drop the page lock otherwise we deadlock with NFS when it calls
+ * ->lookup since ntfs_lookup() will lock the same page.
+ */
+ if (index_type == INDEX_TYPE_ALLOCATION)
+ unlock_page(ia_page);
ntfs_debug("Calling filldir for %s with len %i, fpos 0x%llx, inode "
- "0x%lx, DT_%s.", name, name_len, *fpos,
- MREF_LE(ie->data.dir.indexed_file),
+ "0x%lx, DT_%s.", name, name_len, *fpos, mref,
dt_type == DT_DIR ? "DIR" : "REG");
- return filldir(dirent, name, name_len, *fpos,
- MREF_LE(ie->data.dir.indexed_file), dt_type);
+ rc = filldir(dirent, name, name_len, *fpos, mref, dt_type);
+ /* Relock the page but not if we are aborting ->readdir. */
+ if (!rc && index_type == INDEX_TYPE_ALLOCATION)
+ lock_page(ia_page);
+ return rc;
}
/*
- * VFS calls readdir without BKL but with i_sem held. This protects the VFS
- * parts (e.g. ->f_pos and ->i_size, and it also protects against directory
- * modifications).
- *
* We use the same basic approach as the old NTFS driver, i.e. we parse the
* index root entries and then the index allocation entries that are marked
* as in use in the index bitmap.
+ *
* While this will return the names in random order this doesn't matter for
- * readdir but OTOH results in a faster readdir.
+ * ->readdir but OTOH results in a faster ->readdir.
+ *
+ * VFS calls ->readdir without BKL but with i_sem held. This protects the VFS
+ * parts (e.g. ->f_pos and ->i_size, and it also protects against directory
+ * modifications).
+ *
+ * Locking: - Caller must hold i_sem on the directory.
+ * - Each page cache page in the index allocation mapping must be
+ * locked whilst being accessed otherwise we may find a corrupt
+ * page due to it being under ->writepage at the moment which
+ * applies the mst protection fixups before writing out and then
+ * removes them again after the write is complete after which it
+ * unlocks the page.
*/
static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
{
ntfs_inode *ndir = NTFS_I(vdir);
ntfs_volume *vol = NTFS_SB(sb);
MFT_RECORD *m;
- INDEX_ROOT *ir;
+ INDEX_ROOT *ir = NULL;
INDEX_ENTRY *ie;
INDEX_ALLOCATION *ia;
u8 *name = NULL;
"inode 0x%lx.", vdir->i_ino);
goto err_out;
}
- /* Get to the index root value (it's been verified in read_inode). */
- ir = (INDEX_ROOT*)((u8*)ctx->attr +
- le16_to_cpu(ctx->attr->data.resident.value_offset));
+ /*
+ * Copy the index root attribute value to a buffer so that we can put
+ * the search context and unmap the mft record before calling the
+ * filldir() callback. We need to do this because of NFSd which calls
+ * ->lookup() from its filldir callback() and this causes NTFS to
+ * deadlock as ntfs_lookup() maps the mft record of the directory and
+ * we have got it mapped here already. The only solution is for us to
+ * unmap the mft record here so that a call to ntfs_lookup() is able to
+ * map the mft record without deadlocking.
+ */
+ rc = le32_to_cpu(ctx->attr->data.resident.value_length);
+ ir = (INDEX_ROOT*)kmalloc(rc, GFP_NOFS);
+ if (unlikely(!ir)) {
+ err = -ENOMEM;
+ goto err_out;
+ }
+ /* Copy the index root value (it has been verified in read_inode). */
+ memcpy(ir, (u8*)ctx->attr +
+ le16_to_cpu(ctx->attr->data.resident.value_offset), rc);
+ put_attr_search_ctx(ctx);
+ unmap_mft_record(ndir);
+ ctx = NULL;
+ m = NULL;
index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
/* The first index entry. */
ie = (INDEX_ENTRY*)((u8*)&ir->index +
* or signals an error (both covered by the rc test).
*/
for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
- ntfs_debug("In index root, offset 0x%x.", (u8*)ie - (u8*)ir);
+ ntfs_debug("In index root, offset 0x%zx.", (u8*)ie - (u8*)ir);
/* Bounds checks. */
- if (unlikely((u8*)ie < (u8*)ctx->mrec || (u8*)ie +
+ if (unlikely((u8*)ie < (u8*)ir || (u8*)ie +
sizeof(INDEX_ENTRY_HEADER) > index_end ||
(u8*)ie + le16_to_cpu(ie->key_length) >
index_end))
if (ir_pos > (u8*)ie - (u8*)ir)
continue;
/* Submit the name to the filldir callback. */
- rc = ntfs_filldir(vol, &fpos, ndir, INDEX_TYPE_ROOT, ir, ie,
- name, dirent, filldir);
+ rc = ntfs_filldir(vol, &fpos, ndir, INDEX_TYPE_ROOT, ir, NULL,
+ ie, name, dirent, filldir);
if (rc) {
- put_attr_search_ctx(ctx);
- unmap_mft_record(ndir);
+ kfree(ir);
goto abort;
}
}
- /*
- * We are done with the index root and the mft record for that matter.
- * We need to release it, otherwise we deadlock on ntfs_attr_iget()
- * and/or ntfs_read_page().
- */
- put_attr_search_ctx(ctx);
- unmap_mft_record(ndir);
- m = NULL;
- ctx = NULL;
+ /* We are done with the index root and can free the buffer. */
+ kfree(ir);
+ ir = NULL;
/* If there is no index allocation attribute we are finished. */
if (!NInoIndexAllocPresent(ndir))
goto EOD;
ia_mapping = vdir->i_mapping;
bmp_vi = ndir->itype.index.bmp_ino;
if (unlikely(!bmp_vi)) {
- ntfs_debug("Inode %lu, regetting index bitmap.", vdir->i_ino);
+ ntfs_debug("Inode 0x%lx, regetting index bitmap.", vdir->i_ino);
bmp_vi = ntfs_attr_iget(vdir, AT_BITMAP, I30, 4);
if (unlikely(IS_ERR(bmp_vi))) {
ntfs_error(sb, "Failed to get bitmap attribute.");
/* If the current index buffer is in the same page we reuse the page. */
if ((prev_ia_pos & PAGE_CACHE_MASK) != (ia_pos & PAGE_CACHE_MASK)) {
prev_ia_pos = ia_pos;
- if (likely(ia_page != NULL))
+ if (likely(ia_page != NULL)) {
+ unlock_page(ia_page);
ntfs_unmap_page(ia_page);
+ }
/*
* Map the page cache page containing the current ia_pos,
* reading it from disk if necessary.
ia_page = NULL;
goto err_out;
}
+ lock_page(ia_page);
kaddr = (u8*)page_address(ia_page);
}
/* Get the current index buffer. */
/* Skip index block entry if continuing previous readdir. */
if (ia_pos - ia_start > (u8*)ie - (u8*)ia)
continue;
- /* Submit the name to the filldir callback. */
+ /*
+ * Submit the name to the @filldir callback. Note,
+ * ntfs_filldir() drops the lock on @ia_page but it retakes it
+ * before returning, unless a non-zero value is returned in
+ * which case the page is left unlocked.
+ */
rc = ntfs_filldir(vol, &fpos, ndir, INDEX_TYPE_ALLOCATION, ia,
- ie, name, dirent, filldir);
+ ia_page, ie, name, dirent, filldir);
if (rc) {
+ /* @ia_page is already unlocked in this case. */
ntfs_unmap_page(ia_page);
ntfs_unmap_page(bmp_page);
goto abort;
}
goto find_next_index_buffer;
unm_EOD:
- if (ia_page)
+ if (ia_page) {
+ unlock_page(ia_page);
ntfs_unmap_page(ia_page);
+ }
ntfs_unmap_page(bmp_page);
EOD:
/* We are finished, set fpos to EOD. */
err_out:
if (bmp_page)
ntfs_unmap_page(bmp_page);
- if (ia_page)
+ if (ia_page) {
+ unlock_page(ia_page);
ntfs_unmap_page(ia_page);
+ }
+ if (ir)
+ kfree(ir);
if (name)
kfree(name);
if (ctx)