JehanneOS/newlib
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

* fhandler_disk_file.cc (DIR_NUM_ENTRIES): Set to 100 to maximize

Browse Source 
performance on remote shares.
	(fhandler_disk_file::opendir): Move comment about Samba weirdness into
	fhandler_disk_file::readdir.  Don't disallow
	FileIdBothDirectoryInformation on Samba.
	(fhandler_disk_file::readdir): Workaround Samba problem with
	FileIdBothDirectoryInformation by rereading already read entries
	using FileBothDirectoryInformation.  Change comment about Samba
	weirdness explaining this change.
This commit is contained in:
Corinna Vinschen 2006-08-20 12:18:12 +00:00
parent ffa6849093
commit b0662a051b
2 changed files with 68 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
winsup/cygwin/ChangeLog
View File

@ -1,3 +1,15 @@
2006-08-20 Corinna Vinschen <corinna@vinschen.de>
* fhandler_disk_file.cc (DIR_NUM_ENTRIES): Set to 100 to maximize
performance on remote shares.
(fhandler_disk_file::opendir): Move comment about Samba weirdness into
fhandler_disk_file::readdir. Don't disallow
FileIdBothDirectoryInformation on Samba.
(fhandler_disk_file::readdir): Workaround Samba problem with
FileIdBothDirectoryInformation by rereading already read entries
using FileBothDirectoryInformation. Change comment about Samba
weirdness explaining this change.
2006-08-20 Hideki Iwamoto <h-iwamoto@kit.hi-ho.ne.jp> 2006-08-20 Hideki Iwamoto <h-iwamoto@kit.hi-ho.ne.jp>
* fhandler_disk_file.cc (fhandler_disk_file::pread): Properly check for * fhandler_disk_file.cc (fhandler_disk_file::pread): Properly check for

80
winsup/cygwin/fhandler_disk_file.cc
View File

@ -1468,15 +1468,15 @@ fhandler_disk_file::rmdir ()
/* This kludge detects if we are attempting to remove the current working /* This kludge detects if we are attempting to remove the current working
directory. If so, we will move elsewhere to potentially allow the directory. If so, we will move elsewhere to potentially allow the
rmdir to succeed. This means that cygwin's concept of the current working rmdir to succeed. This means that cygwin's concept of the current
directory != Windows concept but, hey, whaddaregonnado? working directory != Windows concept but, hey, whaddaregonnado?
Note that this will not cause something like the following to work: Note that this will not cause something like the following to work:
$ cd foo $ cd foo
$ rmdir . $ rmdir .
since the shell will have foo "open" in the above case and so Windows will since the shell will have foo "open" in the above case and so Windows
not allow the deletion. (Actually it does on 9X.) will not allow the deletion. (Actually it does on 9X.)
FIXME: A potential workaround for this is for cygwin apps to *never* call FIXME: A potential workaround for this is for cygwin apps to *never*
SetCurrentDirectory. */ call SetCurrentDirectory. */
extern char windows_system_directory[]; extern char windows_system_directory[];
if (strcasematch (get_win32_name (), cygheap->cwd.win32) if (strcasematch (get_win32_name (), cygheap->cwd.win32)
@ -1514,7 +1514,7 @@ fhandler_disk_file::rmdir ()
To tune caching, just tweak this number. To get a feeling for the size, To tune caching, just tweak this number. To get a feeling for the size,
the size of the readdir cache is DIR_NUM_ENTRIES * 632 + 264 bytes. */ the size of the readdir cache is DIR_NUM_ENTRIES * 632 + 264 bytes. */
#define DIR_NUM_ENTRIES 25 /* Cache size 16064 bytes */ #define DIR_NUM_ENTRIES 100 /* Cache size 63464 bytes */
#define DIR_BUF_SIZE (DIR_NUM_ENTRIES \ #define DIR_BUF_SIZE (DIR_NUM_ENTRIES \
* (sizeof (FILE_ID_BOTH_DIR_INFORMATION) \ * (sizeof (FILE_ID_BOTH_DIR_INFORMATION) \
@ -1621,23 +1621,7 @@ fhandler_disk_file::opendir ()
if (pc.hasgood_inode ()) if (pc.hasgood_inode ())
{ {
dir->__flags |= dirent_set_d_ino; dir->__flags |= dirent_set_d_ino;
/* Something weird happens on Samba up to version 3.0.21c, if (wincap.has_fileid_dirinfo ())
which is fixed in 3.0.22. FileIdBothDirectoryInformation
seems to work nicely, but only up to the 128th entry in
the directory. After reaching this entry, the next call
to NtQueryDirectoryFile(FileIdBothDirectoryInformation)
returns STATUS_INVALID_LEVEL. Why should we care, we can
just switch to FileBothDirectoryInformation, isn't it?
Nope! The next call to
NtQueryDirectoryFile(FileBothDirectoryInformation)
actually returns STATUS_NO_MORE_FILES, regardless how
many files are left unread in the directory. This does
not happen when using FileBothDirectoryInformation right
from the start. In that case we can read the whole
directory unmolested. So we have to excempt Samba from
the usage of FileIdBothDirectoryInformation entirely,
even though Samba returns valid File IDs. */
if (wincap.has_fileid_dirinfo () && !pc.fs_is_samba ())
dir->__flags |= dirent_get_d_ino; dir->__flags |= dirent_get_d_ino;
} }
} }
@ -1790,6 +1774,52 @@ fhandler_disk_file::readdir (DIR *dir, dirent *de)
|| status == STATUS_INVALID_PARAMETER || status == STATUS_INVALID_PARAMETER
|| status == STATUS_INVALID_INFO_CLASS) || status == STATUS_INVALID_INFO_CLASS)
dir->__flags &= ~dirent_get_d_ino; dir->__flags &= ~dirent_get_d_ino;
/* Something weird happens on Samba up to version 3.0.21c, which is
fixed in 3.0.22. FileIdBothDirectoryInformation seems to work
nicely, but only up to the 128th entry in the directory. After
reaching this entry, the next call to NtQueryDirectoryFile
(FileIdBothDirectoryInformation) returns STATUS_INVALID_LEVEL.
Why should we care, we can just switch to
FileBothDirectoryInformation, isn't it? Nope! The next call to
NtQueryDirectoryFile(FileBothDirectoryInformation) actually
returns STATUS_NO_MORE_FILES, regardless how many files are left
unread in the directory. This does not happen when using
FileBothDirectoryInformation right from the start, but since
we can't decide whether the server we're talking with has this
bug or not, we end up serving Samba shares always in the slow
mode using FileBothDirectoryInformation. So, what we do here is
to implement the solution suggested by Andrew Tridgell, we just
reread all entries up to dir->d_position using
FileBothDirectoryInformation.
However, We do *not* mark this server as broken and fall back to
using FileBothDirectoryInformation further on. This would slow
down every access to such a server, even for directories under
128 entries. Also, bigger dirs only suffer from one additional
call per full directory scan, which shouldn't be too big a hit.
This can easily be changed if necessary. */
if (status == STATUS_INVALID_LEVEL && dir->__d_position)
{
d_cachepos (dir) = 0;
for (int cnt = 0; cnt < dir->__d_position; ++cnt)
{
if (d_cachepos (dir) == 0)
{
status = NtQueryDirectoryFile (dir->__handle, NULL, NULL,
0, &io, d_cache (dir), DIR_BUF_SIZE,
FileBothDirectoryInformation,
FALSE, NULL, cnt == 0);
if (!NT_SUCCESS (status))
goto go_ahead;
}
buf = (PFILE_ID_BOTH_DIR_INFORMATION) (d_cache (dir)
+ d_cachepos (dir));
if (buf->NextEntryOffset == 0)
d_cachepos (dir) = 0;
else
d_cachepos (dir) += buf->NextEntryOffset;
}
goto go_ahead;
}
} }
if (!(dir->__flags & dirent_get_d_ino)) if (!(dir->__flags & dirent_get_d_ino))
status = NtQueryDirectoryFile (dir->__handle, NULL, NULL, 0, &io, status = NtQueryDirectoryFile (dir->__handle, NULL, NULL, 0, &io,
@ -1798,6 +1828,8 @@ fhandler_disk_file::readdir (DIR *dir, dirent *de)
FALSE, NULL, dir->__d_position == 0); FALSE, NULL, dir->__d_position == 0);
} }
go_ahead:
if (NT_SUCCESS (status)) if (NT_SUCCESS (status))
{ {
buf = (PFILE_ID_BOTH_DIR_INFORMATION) (d_cache (dir) + d_cachepos (dir)); buf = (PFILE_ID_BOTH_DIR_INFORMATION) (d_cache (dir) + d_cachepos (dir));