1
0
mirror of https://github.com/samba-team/samba.git synced 2024-12-23 17:34:34 +03:00
samba-mirror/source3/smbd/open.c

3120 lines
82 KiB
C
Raw Normal View History

/*
Unix SMB/CIFS implementation.
file opening and share modes
Copyright (C) Andrew Tridgell 1992-1998
Copyright (C) Jeremy Allison 2001-2004
Copyright (C) Volker Lendecke 2005
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "includes.h"
extern const struct generic_mapping file_generic_mapping;
extern struct current_user current_user;
extern userdom_struct current_user_info;
extern bool global_client_failed_oplock_break;
struct deferred_open_record {
bool delayed_for_oplocks;
struct file_id id;
};
/****************************************************************************
fd support routines - attempt to do a dos_open.
****************************************************************************/
static NTSTATUS fd_open(struct connection_struct *conn,
const char *fname,
files_struct *fsp,
int flags,
mode_t mode)
{
NTSTATUS status = NT_STATUS_OK;
#ifdef O_NOFOLLOW
/*
* Never follow symlinks on a POSIX client. The
* client should be doing this.
*/
if (fsp->posix_open || !lp_symlinks(SNUM(conn))) {
flags |= O_NOFOLLOW;
}
#endif
fsp->fh->fd = SMB_VFS_OPEN(conn,fname,fsp,flags,mode);
if (fsp->fh->fd == -1) {
status = map_nt_error_from_unix(errno);
}
DEBUG(10,("fd_open: name %s, flags = 0%o mode = 0%o, fd = %d. %s\n",
fname, flags, (int)mode, fsp->fh->fd,
(fsp->fh->fd == -1) ? strerror(errno) : "" ));
return status;
}
/****************************************************************************
Close the file associated with a fsp.
****************************************************************************/
NTSTATUS fd_close(files_struct *fsp)
{
if (fsp->fh->fd == -1) {
return NT_STATUS_OK; /* What we used to call a stat open. */
}
if (fsp->fh->ref_count > 1) {
return NT_STATUS_OK; /* Shared handle. Only close last reference. */
}
return fd_close_posix(fsp);
}
/****************************************************************************
Change the ownership of a file to that of the parent directory.
Do this by fd if possible.
****************************************************************************/
static void change_file_owner_to_parent(connection_struct *conn,
const char *inherit_from_dir,
files_struct *fsp)
{
SMB_STRUCT_STAT parent_st;
int ret;
ret = SMB_VFS_STAT(conn, inherit_from_dir, &parent_st);
if (ret == -1) {
DEBUG(0,("change_file_owner_to_parent: failed to stat parent "
"directory %s. Error was %s\n",
inherit_from_dir, strerror(errno) ));
return;
}
become_root();
ret = SMB_VFS_FCHOWN(fsp, parent_st.st_uid, (gid_t)-1);
unbecome_root();
if (ret == -1) {
DEBUG(0,("change_file_owner_to_parent: failed to fchown "
"file %s to parent directory uid %u. Error "
"was %s\n", fsp->fsp_name,
(unsigned int)parent_st.st_uid,
strerror(errno) ));
}
DEBUG(10,("change_file_owner_to_parent: changed new file %s to "
"parent directory uid %u.\n", fsp->fsp_name,
(unsigned int)parent_st.st_uid ));
}
static NTSTATUS change_dir_owner_to_parent(connection_struct *conn,
const char *inherit_from_dir,
const char *fname,
SMB_STRUCT_STAT *psbuf)
{
char *saved_dir = NULL;
SMB_STRUCT_STAT sbuf;
SMB_STRUCT_STAT parent_st;
TALLOC_CTX *ctx = talloc_tos();
NTSTATUS status = NT_STATUS_OK;
int ret;
ret = SMB_VFS_STAT(conn, inherit_from_dir, &parent_st);
if (ret == -1) {
status = map_nt_error_from_unix(errno);
DEBUG(0,("change_dir_owner_to_parent: failed to stat parent "
"directory %s. Error was %s\n",
inherit_from_dir, strerror(errno) ));
return status;
}
/* We've already done an lstat into psbuf, and we know it's a
directory. If we can cd into the directory and the dev/ino
are the same then we can safely chown without races as
we're locking the directory in place by being in it. This
should work on any UNIX (thanks tridge :-). JRA.
*/
saved_dir = vfs_GetWd(ctx,conn);
if (!saved_dir) {
status = map_nt_error_from_unix(errno);
DEBUG(0,("change_dir_owner_to_parent: failed to get "
"current working directory. Error was %s\n",
strerror(errno)));
return status;
}
/* Chdir into the new path. */
if (vfs_ChDir(conn, fname) == -1) {
status = map_nt_error_from_unix(errno);
DEBUG(0,("change_dir_owner_to_parent: failed to change "
"current working directory to %s. Error "
"was %s\n", fname, strerror(errno) ));
goto out;
}
if (SMB_VFS_STAT(conn,".",&sbuf) == -1) {
status = map_nt_error_from_unix(errno);
DEBUG(0,("change_dir_owner_to_parent: failed to stat "
"directory '.' (%s) Error was %s\n",
fname, strerror(errno)));
goto out;
}
/* Ensure we're pointing at the same place. */
if (sbuf.st_dev != psbuf->st_dev ||
sbuf.st_ino != psbuf->st_ino ||
sbuf.st_mode != psbuf->st_mode ) {
DEBUG(0,("change_dir_owner_to_parent: "
"device/inode/mode on directory %s changed. "
"Refusing to chown !\n", fname ));
status = NT_STATUS_ACCESS_DENIED;
goto out;
}
become_root();
ret = SMB_VFS_CHOWN(conn, ".", parent_st.st_uid, (gid_t)-1);
unbecome_root();
if (ret == -1) {
status = map_nt_error_from_unix(errno);
DEBUG(10,("change_dir_owner_to_parent: failed to chown "
"directory %s to parent directory uid %u. "
"Error was %s\n", fname,
(unsigned int)parent_st.st_uid, strerror(errno) ));
goto out;
}
DEBUG(10,("change_dir_owner_to_parent: changed ownership of new "
"directory %s to parent directory uid %u.\n",
fname, (unsigned int)parent_st.st_uid ));
out:
vfs_ChDir(conn,saved_dir);
return status;
}
/****************************************************************************
Open a file.
****************************************************************************/
static NTSTATUS open_file(files_struct *fsp,
connection_struct *conn,
struct smb_request *req,
const char *parent_dir,
const char *name,
const char *path,
SMB_STRUCT_STAT *psbuf,
int flags,
mode_t unx_mode,
uint32 access_mask, /* client requested access mask. */
uint32 open_access_mask) /* what we're actually using in the open. */
{
NTSTATUS status = NT_STATUS_OK;
int accmode = (flags & O_ACCMODE);
int local_flags = flags;
bool file_existed = VALID_STAT(*psbuf);
fsp->fh->fd = -1;
errno = EPERM;
/* Check permissions */
/*
* This code was changed after seeing a client open request
* containing the open mode of (DENY_WRITE/read-only) with
* the 'create if not exist' bit set. The previous code
* would fail to open the file read only on a read-only share
* as it was checking the flags parameter directly against O_RDONLY,
* this was failing as the flags parameter was set to O_RDONLY|O_CREAT.
* JRA.
*/
if (!CAN_WRITE(conn)) {
/* It's a read-only share - fail if we wanted to write. */
if(accmode != O_RDONLY) {
DEBUG(3,("Permission denied opening %s\n", path));
return NT_STATUS_ACCESS_DENIED;
} else if(flags & O_CREAT) {
/* We don't want to write - but we must make sure that
O_CREAT doesn't create the file if we have write
access into the directory.
*/
flags &= ~O_CREAT;
local_flags &= ~O_CREAT;
}
}
/*
* This little piece of insanity is inspired by the
* fact that an NT client can open a file for O_RDONLY,
* but set the create disposition to FILE_EXISTS_TRUNCATE.
* If the client *can* write to the file, then it expects to
* truncate the file, even though it is opening for readonly.
* Quicken uses this stupid trick in backup file creation...
* Thanks *greatly* to "David W. Chapman Jr." <dwcjr@inethouston.net>
* for helping track this one down. It didn't bite us in 2.0.x
* as we always opened files read-write in that release. JRA.
*/
if ((accmode == O_RDONLY) && ((flags & O_TRUNC) == O_TRUNC)) {
DEBUG(10,("open_file: truncate requested on read-only open "
"for file %s\n", path));
local_flags = (flags & ~O_ACCMODE)|O_RDWR;
}
if ((open_access_mask & (FILE_READ_DATA|FILE_WRITE_DATA|FILE_APPEND_DATA|FILE_EXECUTE)) ||
(!file_existed && (local_flags & O_CREAT)) ||
((local_flags & O_TRUNC) == O_TRUNC) ) {
/*
* We can't actually truncate here as the file may be locked.
* open_file_ntcreate will take care of the truncate later. JRA.
*/
local_flags &= ~O_TRUNC;
#if defined(O_NONBLOCK) && defined(S_ISFIFO)
/*
* We would block on opening a FIFO with no one else on the
* other end. Do what we used to do and add O_NONBLOCK to the
* open flags. JRA.
*/
if (file_existed && S_ISFIFO(psbuf->st_mode)) {
local_flags |= O_NONBLOCK;
}
#endif
/* Don't create files with Microsoft wildcard characters. */
if ((local_flags & O_CREAT) && !file_existed &&
ms_has_wild(path)) {
return NT_STATUS_OBJECT_NAME_INVALID;
}
/* Actually do the open */
status = fd_open(conn, path, fsp, local_flags, unx_mode);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(3,("Error opening file %s (%s) (local_flags=%d) "
"(flags=%d)\n",
path,nt_errstr(status),local_flags,flags));
return status;
}
if ((local_flags & O_CREAT) && !file_existed) {
/* Inherit the ACL if required */
if (lp_inherit_perms(SNUM(conn))) {
inherit_access_acl(conn, parent_dir, path,
unx_mode);
}
/* Change the owner if required. */
if (lp_inherit_owner(SNUM(conn))) {
change_file_owner_to_parent(conn, parent_dir,
fsp);
}
notify_fname(conn, NOTIFY_ACTION_ADDED,
FILE_NOTIFY_CHANGE_FILE_NAME, path);
}
} else {
fsp->fh->fd = -1; /* What we used to call a stat open. */
}
if (!file_existed) {
int ret;
if (fsp->fh->fd == -1) {
ret = SMB_VFS_STAT(conn, path, psbuf);
} else {
ret = SMB_VFS_FSTAT(fsp, psbuf);
/* If we have an fd, this stat should succeed. */
if (ret == -1) {
DEBUG(0,("Error doing fstat on open file %s "
"(%s)\n", path,strerror(errno) ));
}
}
/* For a non-io open, this stat failing means file not found. JRA */
if (ret == -1) {
status = map_nt_error_from_unix(errno);
fd_close(fsp);
return status;
}
}
/*
* POSIX allows read-only opens of directories. We don't
* want to do this (we use a different code path for this)
* so catch a directory open and return an EISDIR. JRA.
*/
if(S_ISDIR(psbuf->st_mode)) {
fd_close(fsp);
errno = EISDIR;
return NT_STATUS_FILE_IS_A_DIRECTORY;
}
fsp->mode = psbuf->st_mode;
fsp->file_id = vfs_file_id_from_sbuf(conn, psbuf);
fsp->vuid = req ? req->vuid : UID_FIELD_INVALID;
fsp->file_pid = req ? req->smbpid : 0;
fsp->can_lock = True;
fsp->can_read = (access_mask & (FILE_READ_DATA)) ? True : False;
if (!CAN_WRITE(conn)) {
fsp->can_write = False;
} else {
fsp->can_write = (access_mask & (FILE_WRITE_DATA | FILE_APPEND_DATA)) ?
True : False;
}
fsp->print_file = False;
fsp->modified = False;
fsp->sent_oplock_break = NO_BREAK_SENT;
fsp->is_directory = False;
fsp->is_stat = False;
if (conn->aio_write_behind_list &&
is_in_path(path, conn->aio_write_behind_list, conn->case_sensitive)) {
fsp->aio_write_behind = True;
}
string_set(&fsp->fsp_name, path);
fsp->wcp = NULL; /* Write cache pointer. */
DEBUG(2,("%s opened file %s read=%s write=%s (numopen=%d)\n",
*current_user_info.smb_name ?
current_user_info.smb_name : conn->user,fsp->fsp_name,
BOOLSTR(fsp->can_read), BOOLSTR(fsp->can_write),
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
conn->num_files_open + 1));
errno = 0;
return NT_STATUS_OK;
Makefile.in: Fixed bug with continuation line causing proto to fail. Added $(PROGS) $(SPROGS) as targets for make clean. acconfig.h: Added HAVE_IRIX_SPECIFIC_CAPABILITIES. configure.in: Added sys/capability.h header check. Added function checks for srandom random srand rand. Added HAVE_IRIX_SPECIFIC_CAPABILITIES test. includes.h: Added #include <sys/capability.h>. ntdomain.h: Moved struct acct_info into here from smb.h smb.h: Added KERNEL_OPLOCK_CAPABILITY define. Moved enum action_type into rpcclient.h Moved struct cli_state into client.h Moved struct nt_client_info, struct tar_client_info, struct client_info into rpcclient.h lib/genrand.c: Changed to use sys_random() & friends. lib/smbrun.c: Lose capabilities after fork. lib/system.c: Added set_process_capability(), set_inherited_process_capability() sys_random(), sys_srandom(). lib/util.c: Added Ander's EFBIG lock check to fcntl_lock for 64 bit access to an 32 bit mounted NFS filesystem. nmbd/nmbd.c: Changed to use sys_random() & friends. nmbd/nmbd_browsesync.c: Changed to use sys_random() & friends. passdb/ldap.c: Missed one pdb_encode_acct_ctrl call. passdb/passdb.c: Changed to Ander's code for ' ' characters. passdb/smbpass.c: Added Ander's code to reset ACB_PWNOTREQ. script/mkproto.awk: Added 'long' to prototypes. smbd/chgpasswd.c: Lose capabilities after fork. smbd/open.c: Do the mmap *after* the kernel oplock. smbd/oplock.c: Removed stub code from kernel oplock path. Added set_process_capability(), set_inherited_process_capability() calls. smbd/reply.c: Initialize count = 0, offset = 0. smbd/server.c: Added set_process_capability(), set_inherited_process_capability() calls. tests/summary.c: Ensure we have RANDOM or RAND. utils/smbpasswd.c: Added Ander's code to reset ACB_PWNOTREQ. utils/torture.c: Changed to use sys_random() & friends. Jeremy. (This used to be commit e8be306f23963ac00b1a383ebe0cc1421529fb02)
1998-09-26 03:40:49 +04:00
}
/*******************************************************************
Return True if the filename is one of the special executable types.
********************************************************************/
static bool is_executable(const char *fname)
{
if ((fname = strrchr_m(fname,'.'))) {
if (strequal(fname,".com") ||
strequal(fname,".dll") ||
strequal(fname,".exe") ||
strequal(fname,".sym")) {
return True;
}
}
return False;
}
/****************************************************************************
Check if we can open a file with a share mode.
Returns True if conflict, False if not.
****************************************************************************/
static bool share_conflict(struct share_mode_entry *entry,
uint32 access_mask,
uint32 share_access)
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
{
DEBUG(10,("share_conflict: entry->access_mask = 0x%x, "
"entry->share_access = 0x%x, "
"entry->private_options = 0x%x\n",
(unsigned int)entry->access_mask,
(unsigned int)entry->share_access,
(unsigned int)entry->private_options));
DEBUG(10,("share_conflict: access_mask = 0x%x, share_access = 0x%x\n",
(unsigned int)access_mask, (unsigned int)share_access));
if ((entry->access_mask & (FILE_WRITE_DATA|
FILE_APPEND_DATA|
FILE_READ_DATA|
FILE_EXECUTE|
DELETE_ACCESS)) == 0) {
DEBUG(10,("share_conflict: No conflict due to "
"entry->access_mask = 0x%x\n",
(unsigned int)entry->access_mask ));
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
return False;
}
if ((access_mask & (FILE_WRITE_DATA|
FILE_APPEND_DATA|
FILE_READ_DATA|
FILE_EXECUTE|
DELETE_ACCESS)) == 0) {
DEBUG(10,("share_conflict: No conflict due to "
"access_mask = 0x%x\n",
(unsigned int)access_mask ));
This is a big, rather ugly patch. Whilst investigating the files not truncated when copying to a full disk problem, I discovered that we were not allowing the delete on close flag to be set properly, this led to other things, and after investigation of the proper delete on close semantics and their relationship to the file_share_delete flag I discovered there were some cases where we weren't doing the deny modes properly. And this after only 5 years working on them..... :-) :-). So here's the latest attempt. I realised the delete on close flag needs to be set across all smbds with a dev/ino pair open - in addition, the delete on close flag, allow share delete and delete access requested all need to be stored in the share mode tdb. The "delete_on_close" entry in the fsp struct is now redundant and should really be removed. This may also mean we can get rid of the "iterate_fsp" calls that I didn't like adding in the first place. Whilst doing this patch, I also discovered we needed to do the se_map_generic() call for file opens and POSIX ACL mapping, so I added that also. This code, although ugly, now passes the deny mode torture tests plus the delete on close tests I added. I do need to add one more multiple connection delete on close test to make sure I got the semantics exactly right, plus we should also (as Andrew suggested) move to random testing here. The good news is that NT should now correctly delete the file on disk full error when copying to a disk :-). Jeremy. (This used to be commit 51987684bd231c744da2e5f3705fd236d5616173)
2001-03-30 12:57:24 +04:00
return False;
}
#if 1 /* JRA TEST - Superdebug. */
#define CHECK_MASK(num, am, right, sa, share) \
DEBUG(10,("share_conflict: [%d] am (0x%x) & right (0x%x) = 0x%x\n", \
(unsigned int)(num), (unsigned int)(am), \
(unsigned int)(right), (unsigned int)(am)&(right) )); \
DEBUG(10,("share_conflict: [%d] sa (0x%x) & share (0x%x) = 0x%x\n", \
(unsigned int)(num), (unsigned int)(sa), \
(unsigned int)(share), (unsigned int)(sa)&(share) )); \
if (((am) & (right)) && !((sa) & (share))) { \
DEBUG(10,("share_conflict: check %d conflict am = 0x%x, right = 0x%x, \
sa = 0x%x, share = 0x%x\n", (num), (unsigned int)(am), (unsigned int)(right), (unsigned int)(sa), \
(unsigned int)(share) )); \
return True; \
}
#else
#define CHECK_MASK(num, am, right, sa, share) \
if (((am) & (right)) && !((sa) & (share))) { \
DEBUG(10,("share_conflict: check %d conflict am = 0x%x, right = 0x%x, \
sa = 0x%x, share = 0x%x\n", (num), (unsigned int)(am), (unsigned int)(right), (unsigned int)(sa), \
(unsigned int)(share) )); \
return True; \
This is a big, rather ugly patch. Whilst investigating the files not truncated when copying to a full disk problem, I discovered that we were not allowing the delete on close flag to be set properly, this led to other things, and after investigation of the proper delete on close semantics and their relationship to the file_share_delete flag I discovered there were some cases where we weren't doing the deny modes properly. And this after only 5 years working on them..... :-) :-). So here's the latest attempt. I realised the delete on close flag needs to be set across all smbds with a dev/ino pair open - in addition, the delete on close flag, allow share delete and delete access requested all need to be stored in the share mode tdb. The "delete_on_close" entry in the fsp struct is now redundant and should really be removed. This may also mean we can get rid of the "iterate_fsp" calls that I didn't like adding in the first place. Whilst doing this patch, I also discovered we needed to do the se_map_generic() call for file opens and POSIX ACL mapping, so I added that also. This code, although ugly, now passes the deny mode torture tests plus the delete on close tests I added. I do need to add one more multiple connection delete on close test to make sure I got the semantics exactly right, plus we should also (as Andrew suggested) move to random testing here. The good news is that NT should now correctly delete the file on disk full error when copying to a disk :-). Jeremy. (This used to be commit 51987684bd231c744da2e5f3705fd236d5616173)
2001-03-30 12:57:24 +04:00
}
#endif
This is a big, rather ugly patch. Whilst investigating the files not truncated when copying to a full disk problem, I discovered that we were not allowing the delete on close flag to be set properly, this led to other things, and after investigation of the proper delete on close semantics and their relationship to the file_share_delete flag I discovered there were some cases where we weren't doing the deny modes properly. And this after only 5 years working on them..... :-) :-). So here's the latest attempt. I realised the delete on close flag needs to be set across all smbds with a dev/ino pair open - in addition, the delete on close flag, allow share delete and delete access requested all need to be stored in the share mode tdb. The "delete_on_close" entry in the fsp struct is now redundant and should really be removed. This may also mean we can get rid of the "iterate_fsp" calls that I didn't like adding in the first place. Whilst doing this patch, I also discovered we needed to do the se_map_generic() call for file opens and POSIX ACL mapping, so I added that also. This code, although ugly, now passes the deny mode torture tests plus the delete on close tests I added. I do need to add one more multiple connection delete on close test to make sure I got the semantics exactly right, plus we should also (as Andrew suggested) move to random testing here. The good news is that NT should now correctly delete the file on disk full error when copying to a disk :-). Jeremy. (This used to be commit 51987684bd231c744da2e5f3705fd236d5616173)
2001-03-30 12:57:24 +04:00
CHECK_MASK(1, entry->access_mask, FILE_WRITE_DATA | FILE_APPEND_DATA,
share_access, FILE_SHARE_WRITE);
CHECK_MASK(2, access_mask, FILE_WRITE_DATA | FILE_APPEND_DATA,
entry->share_access, FILE_SHARE_WRITE);
CHECK_MASK(3, entry->access_mask, FILE_READ_DATA | FILE_EXECUTE,
share_access, FILE_SHARE_READ);
CHECK_MASK(4, access_mask, FILE_READ_DATA | FILE_EXECUTE,
entry->share_access, FILE_SHARE_READ);
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
CHECK_MASK(5, entry->access_mask, DELETE_ACCESS,
share_access, FILE_SHARE_DELETE);
CHECK_MASK(6, access_mask, DELETE_ACCESS,
entry->share_access, FILE_SHARE_DELETE);
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
DEBUG(10,("share_conflict: No conflict.\n"));
return False;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
#if defined(DEVELOPER)
static void validate_my_share_entries(int num,
struct share_mode_entry *share_entry)
{
files_struct *fsp;
if (!procid_is_me(&share_entry->pid)) {
return;
}
if (is_deferred_open_entry(share_entry) &&
!open_was_deferred(share_entry->op_mid)) {
char *str = talloc_asprintf(talloc_tos(),
"Got a deferred entry without a request: "
"PANIC: %s\n",
share_mode_str(talloc_tos(), num, share_entry));
smb_panic(str);
}
if (!is_valid_share_mode_entry(share_entry)) {
return;
}
fsp = file_find_dif(share_entry->id,
share_entry->share_file_id);
if (!fsp) {
DEBUG(0,("validate_my_share_entries: PANIC : %s\n",
share_mode_str(talloc_tos(), num, share_entry) ));
smb_panic("validate_my_share_entries: Cannot match a "
"share entry with an open file\n");
}
if (is_deferred_open_entry(share_entry) ||
is_unused_share_mode_entry(share_entry)) {
goto panic;
}
if ((share_entry->op_type == NO_OPLOCK) &&
(fsp->oplock_type == FAKE_LEVEL_II_OPLOCK)) {
/* Someone has already written to it, but I haven't yet
* noticed */
return;
}
if (((uint16)fsp->oplock_type) != share_entry->op_type) {
goto panic;
}
return;
panic:
{
char *str;
DEBUG(0,("validate_my_share_entries: PANIC : %s\n",
share_mode_str(talloc_tos(), num, share_entry) ));
str = talloc_asprintf(talloc_tos(),
"validate_my_share_entries: "
"file %s, oplock_type = 0x%x, op_type = 0x%x\n",
fsp->fsp_name, (unsigned int)fsp->oplock_type,
(unsigned int)share_entry->op_type );
smb_panic(str);
}
}
#endif
static bool is_stat_open(uint32 access_mask)
{
return (access_mask &&
((access_mask & ~(SYNCHRONIZE_ACCESS| FILE_READ_ATTRIBUTES|
FILE_WRITE_ATTRIBUTES))==0) &&
((access_mask & (SYNCHRONIZE_ACCESS|FILE_READ_ATTRIBUTES|
FILE_WRITE_ATTRIBUTES)) != 0));
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
/****************************************************************************
Deal with share modes
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
Invarient: Share mode must be locked on entry and exit.
Returns -1 on error, or number of share modes on success (may be zero).
****************************************************************************/
static NTSTATUS open_mode_check(connection_struct *conn,
const char *fname,
struct share_mode_lock *lck,
uint32 access_mask,
uint32 share_access,
uint32 create_options,
bool *file_existed)
{
int i;
if(lck->num_share_modes == 0) {
return NT_STATUS_OK;
}
*file_existed = True;
/* A delete on close prohibits everything */
if (lck->delete_on_close) {
return NT_STATUS_DELETE_PENDING;
}
if (is_stat_open(access_mask)) {
/* Stat open that doesn't trigger oplock breaks or share mode
* checks... ! JRA. */
return NT_STATUS_OK;
}
/*
* Check if the share modes will give us access.
*/
#if defined(DEVELOPER)
for(i = 0; i < lck->num_share_modes; i++) {
validate_my_share_entries(i, &lck->share_modes[i]);
}
#endif
if (!lp_share_modes(SNUM(conn))) {
return NT_STATUS_OK;
}
/* Now we check the share modes, after any oplock breaks. */
for(i = 0; i < lck->num_share_modes; i++) {
if (!is_valid_share_mode_entry(&lck->share_modes[i])) {
continue;
}
/* someone else has a share lock on it, check to see if we can
* too */
if (share_conflict(&lck->share_modes[i],
access_mask, share_access)) {
return NT_STATUS_SHARING_VIOLATION;
}
}
return NT_STATUS_OK;
}
static bool is_delete_request(files_struct *fsp) {
return ((fsp->access_mask == DELETE_ACCESS) &&
(fsp->oplock_type == NO_OPLOCK));
}
/*
* 1) No files open at all or internal open: Grant whatever the client wants.
*
* 2) Exclusive (or batch) oplock around: If the requested access is a delete
* request, break if the oplock around is a batch oplock. If it's another
* requested access type, break.
*
* 3) Only level2 around: Grant level2 and do nothing else.
*/
static bool delay_for_oplocks(struct share_mode_lock *lck,
files_struct *fsp,
uint16 mid,
int pass_number,
int oplock_request)
{
int i;
struct share_mode_entry *exclusive = NULL;
bool valid_entry = False;
bool delay_it = False;
bool have_level2 = False;
NTSTATUS status;
char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
if (oplock_request & INTERNAL_OPEN_ONLY) {
fsp->oplock_type = NO_OPLOCK;
}
if ((oplock_request & INTERNAL_OPEN_ONLY) || is_stat_open(fsp->access_mask)) {
return False;
}
for (i=0; i<lck->num_share_modes; i++) {
if (!is_valid_share_mode_entry(&lck->share_modes[i])) {
continue;
}
/* At least one entry is not an invalid or deferred entry. */
valid_entry = True;
if (pass_number == 1) {
if (BATCH_OPLOCK_TYPE(lck->share_modes[i].op_type)) {
SMB_ASSERT(exclusive == NULL);
exclusive = &lck->share_modes[i];
}
} else {
if (EXCLUSIVE_OPLOCK_TYPE(lck->share_modes[i].op_type)) {
SMB_ASSERT(exclusive == NULL);
exclusive = &lck->share_modes[i];
}
}
if (lck->share_modes[i].op_type == LEVEL_II_OPLOCK) {
SMB_ASSERT(exclusive == NULL);
have_level2 = True;
}
}
if (!valid_entry) {
/* All entries are placeholders or deferred.
* Directly grant whatever the client wants. */
if (fsp->oplock_type == NO_OPLOCK) {
/* Store a level2 oplock, but don't tell the client */
fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
}
return False;
}
if (exclusive != NULL) { /* Found an exclusive oplock */
SMB_ASSERT(!have_level2);
delay_it = is_delete_request(fsp) ?
BATCH_OPLOCK_TYPE(exclusive->op_type) : True;
}
if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
/* We can at most grant level2 as there are other
* level2 or NO_OPLOCK entries. */
fsp->oplock_type = LEVEL_II_OPLOCK;
}
if ((fsp->oplock_type == NO_OPLOCK) && have_level2) {
/* Store a level2 oplock, but don't tell the client */
fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
}
if (!delay_it) {
return False;
}
/*
* Send a break message to the oplock holder and delay the open for
* our client.
*/
DEBUG(10, ("Sending break request to PID %s\n",
procid_str_static(&exclusive->pid)));
exclusive->op_mid = mid;
/* Create the message. */
share_mode_entry_to_message(msg, exclusive);
/* Add in the FORCE_OPLOCK_BREAK_TO_NONE bit in the message if set. We
don't want this set in the share mode struct pointed to by lck. */
if (oplock_request & FORCE_OPLOCK_BREAK_TO_NONE) {
SSVAL(msg,6,exclusive->op_type | FORCE_OPLOCK_BREAK_TO_NONE);
}
status = messaging_send_buf(smbd_messaging_context(), exclusive->pid,
MSG_SMB_BREAK_REQUEST,
(uint8 *)msg,
MSG_SMB_SHARE_MODE_ENTRY_SIZE);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(3, ("Could not send oplock break message: %s\n",
nt_errstr(status)));
}
return True;
}
static bool request_timed_out(struct timeval request_time,
struct timeval timeout)
{
struct timeval now, end_time;
GetTimeOfDay(&now);
end_time = timeval_sum(&request_time, &timeout);
return (timeval_compare(&end_time, &now) < 0);
}
/****************************************************************************
Handle the 1 second delay in returning a SHARING_VIOLATION error.
****************************************************************************/
static void defer_open(struct share_mode_lock *lck,
struct timeval request_time,
struct timeval timeout,
struct smb_request *req,
struct deferred_open_record *state)
{
int i;
/* Paranoia check */
for (i=0; i<lck->num_share_modes; i++) {
struct share_mode_entry *e = &lck->share_modes[i];
if (!is_deferred_open_entry(e)) {
continue;
}
if (procid_is_me(&e->pid) && (e->op_mid == req->mid)) {
DEBUG(0, ("Trying to defer an already deferred "
"request: mid=%d, exiting\n", req->mid));
exit_server("attempt to defer a deferred request");
}
}
/* End paranoia check */
DEBUG(10,("defer_open_sharing_error: time [%u.%06u] adding deferred "
"open entry for mid %u\n",
(unsigned int)request_time.tv_sec,
(unsigned int)request_time.tv_usec,
(unsigned int)req->mid));
if (!push_deferred_smb_message(req, request_time, timeout,
(char *)state, sizeof(*state))) {
exit_server("push_deferred_smb_message failed");
}
add_deferred_open(lck, req->mid, request_time, state->id);
/*
* Push the MID of this packet on the signing queue.
* We only do this once, the first time we push the packet
* onto the deferred open queue, as this has a side effect
* of incrementing the response sequence number.
*/
srv_defer_sign_response(req->mid);
}
/****************************************************************************
On overwrite open ensure that the attributes match.
****************************************************************************/
static bool open_match_attributes(connection_struct *conn,
const char *path,
uint32 old_dos_attr,
uint32 new_dos_attr,
mode_t existing_unx_mode,
mode_t new_unx_mode,
mode_t *returned_unx_mode)
{
uint32 noarch_old_dos_attr, noarch_new_dos_attr;
noarch_old_dos_attr = (old_dos_attr & ~FILE_ATTRIBUTE_ARCHIVE);
noarch_new_dos_attr = (new_dos_attr & ~FILE_ATTRIBUTE_ARCHIVE);
if((noarch_old_dos_attr == 0 && noarch_new_dos_attr != 0) ||
(noarch_old_dos_attr != 0 && ((noarch_old_dos_attr & noarch_new_dos_attr) == noarch_old_dos_attr))) {
*returned_unx_mode = new_unx_mode;
} else {
*returned_unx_mode = (mode_t)0;
}
DEBUG(10,("open_match_attributes: file %s old_dos_attr = 0x%x, "
"existing_unx_mode = 0%o, new_dos_attr = 0x%x "
"returned_unx_mode = 0%o\n",
path,
(unsigned int)old_dos_attr,
(unsigned int)existing_unx_mode,
(unsigned int)new_dos_attr,
(unsigned int)*returned_unx_mode ));
/* If we're mapping SYSTEM and HIDDEN ensure they match. */
if (lp_map_system(SNUM(conn)) || lp_store_dos_attributes(SNUM(conn))) {
if ((old_dos_attr & FILE_ATTRIBUTE_SYSTEM) &&
!(new_dos_attr & FILE_ATTRIBUTE_SYSTEM)) {
return False;
}
}
if (lp_map_hidden(SNUM(conn)) || lp_store_dos_attributes(SNUM(conn))) {
if ((old_dos_attr & FILE_ATTRIBUTE_HIDDEN) &&
!(new_dos_attr & FILE_ATTRIBUTE_HIDDEN)) {
return False;
}
}
return True;
}
/****************************************************************************
Special FCB or DOS processing in the case of a sharing violation.
Try and find a duplicated file handle.
****************************************************************************/
static files_struct *fcb_or_dos_open(connection_struct *conn,
const char *fname,
struct file_id id,
uint16 file_pid,
uint16 vuid,
uint32 access_mask,
uint32 share_access,
uint32 create_options)
{
files_struct *fsp;
files_struct *dup_fsp;
DEBUG(5,("fcb_or_dos_open: attempting old open semantics for "
"file %s.\n", fname ));
for(fsp = file_find_di_first(id); fsp;
fsp = file_find_di_next(fsp)) {
DEBUG(10,("fcb_or_dos_open: checking file %s, fd = %d, "
"vuid = %u, file_pid = %u, private_options = 0x%x "
"access_mask = 0x%x\n", fsp->fsp_name,
fsp->fh->fd, (unsigned int)fsp->vuid,
(unsigned int)fsp->file_pid,
(unsigned int)fsp->fh->private_options,
(unsigned int)fsp->access_mask ));
if (fsp->fh->fd != -1 &&
fsp->vuid == vuid &&
fsp->file_pid == file_pid &&
(fsp->fh->private_options & (NTCREATEX_OPTIONS_PRIVATE_DENY_DOS |
NTCREATEX_OPTIONS_PRIVATE_DENY_FCB)) &&
(fsp->access_mask & FILE_WRITE_DATA) &&
strequal(fsp->fsp_name, fname)) {
DEBUG(10,("fcb_or_dos_open: file match\n"));
break;
}
}
if (!fsp) {
return NULL;
}
/* quite an insane set of semantics ... */
if (is_executable(fname) &&
(fsp->fh->private_options & NTCREATEX_OPTIONS_PRIVATE_DENY_DOS)) {
DEBUG(10,("fcb_or_dos_open: file fail due to is_executable.\n"));
return NULL;
}
/* We need to duplicate this fsp. */
if (!NT_STATUS_IS_OK(dup_file_fsp(fsp, access_mask, share_access,
create_options, &dup_fsp))) {
return NULL;
}
return dup_fsp;
}
/****************************************************************************
Open a file with a share mode - old openX method - map into NTCreate.
****************************************************************************/
bool map_open_params_to_ntcreate(const char *fname, int deny_mode, int open_func,
uint32 *paccess_mask,
uint32 *pshare_mode,
uint32 *pcreate_disposition,
uint32 *pcreate_options)
{
uint32 access_mask;
uint32 share_mode;
uint32 create_disposition;
uint32 create_options = 0;
DEBUG(10,("map_open_params_to_ntcreate: fname = %s, deny_mode = 0x%x, "
"open_func = 0x%x\n",
fname, (unsigned int)deny_mode, (unsigned int)open_func ));
/* Create the NT compatible access_mask. */
switch (GET_OPENX_MODE(deny_mode)) {
case DOS_OPEN_EXEC: /* Implies read-only - used to be FILE_READ_DATA */
case DOS_OPEN_RDONLY:
access_mask = FILE_GENERIC_READ;
break;
case DOS_OPEN_WRONLY:
access_mask = FILE_GENERIC_WRITE;
break;
case DOS_OPEN_RDWR:
case DOS_OPEN_FCB:
access_mask = FILE_GENERIC_READ|FILE_GENERIC_WRITE;
break;
default:
DEBUG(10,("map_open_params_to_ntcreate: bad open mode = 0x%x\n",
(unsigned int)GET_OPENX_MODE(deny_mode)));
return False;
}
/* Create the NT compatible create_disposition. */
switch (open_func) {
case OPENX_FILE_EXISTS_FAIL|OPENX_FILE_CREATE_IF_NOT_EXIST:
create_disposition = FILE_CREATE;
break;
case OPENX_FILE_EXISTS_OPEN:
create_disposition = FILE_OPEN;
break;
case OPENX_FILE_EXISTS_OPEN|OPENX_FILE_CREATE_IF_NOT_EXIST:
create_disposition = FILE_OPEN_IF;
break;
case OPENX_FILE_EXISTS_TRUNCATE:
create_disposition = FILE_OVERWRITE;
break;
case OPENX_FILE_EXISTS_TRUNCATE|OPENX_FILE_CREATE_IF_NOT_EXIST:
create_disposition = FILE_OVERWRITE_IF;
break;
default:
/* From samba4 - to be confirmed. */
if (GET_OPENX_MODE(deny_mode) == DOS_OPEN_EXEC) {
create_disposition = FILE_CREATE;
break;
}
DEBUG(10,("map_open_params_to_ntcreate: bad "
"open_func 0x%x\n", (unsigned int)open_func));
return False;
}
/* Create the NT compatible share modes. */
switch (GET_DENY_MODE(deny_mode)) {
case DENY_ALL:
share_mode = FILE_SHARE_NONE;
break;
case DENY_WRITE:
share_mode = FILE_SHARE_READ;
break;
case DENY_READ:
share_mode = FILE_SHARE_WRITE;
break;
case DENY_NONE:
share_mode = FILE_SHARE_READ|FILE_SHARE_WRITE;
break;
case DENY_DOS:
create_options |= NTCREATEX_OPTIONS_PRIVATE_DENY_DOS;
if (is_executable(fname)) {
share_mode = FILE_SHARE_READ|FILE_SHARE_WRITE;
} else {
if (GET_OPENX_MODE(deny_mode) == DOS_OPEN_RDONLY) {
share_mode = FILE_SHARE_READ;
} else {
share_mode = FILE_SHARE_NONE;
}
}
break;
case DENY_FCB:
create_options |= NTCREATEX_OPTIONS_PRIVATE_DENY_FCB;
share_mode = FILE_SHARE_NONE;
break;
default:
DEBUG(10,("map_open_params_to_ntcreate: bad deny_mode 0x%x\n",
(unsigned int)GET_DENY_MODE(deny_mode) ));
return False;
}
DEBUG(10,("map_open_params_to_ntcreate: file %s, access_mask = 0x%x, "
"share_mode = 0x%x, create_disposition = 0x%x, "
"create_options = 0x%x\n",
fname,
(unsigned int)access_mask,
(unsigned int)share_mode,
(unsigned int)create_disposition,
(unsigned int)create_options ));
if (paccess_mask) {
*paccess_mask = access_mask;
}
if (pshare_mode) {
*pshare_mode = share_mode;
}
if (pcreate_disposition) {
*pcreate_disposition = create_disposition;
}
if (pcreate_options) {
*pcreate_options = create_options;
}
return True;
}
static void schedule_defer_open(struct share_mode_lock *lck,
struct timeval request_time,
struct smb_request *req)
{
struct deferred_open_record state;
/* This is a relative time, added to the absolute
request_time value to get the absolute timeout time.
Note that if this is the second or greater time we enter
this codepath for this particular request mid then
request_time is left as the absolute time of the *first*
time this request mid was processed. This is what allows
the request to eventually time out. */
struct timeval timeout;
/* Normally the smbd we asked should respond within
* OPLOCK_BREAK_TIMEOUT seconds regardless of whether
* the client did, give twice the timeout as a safety
* measure here in case the other smbd is stuck
* somewhere else. */
timeout = timeval_set(OPLOCK_BREAK_TIMEOUT*2, 0);
/* Nothing actually uses state.delayed_for_oplocks
but it's handy to differentiate in debug messages
between a 30 second delay due to oplock break, and
a 1 second delay for share mode conflicts. */
state.delayed_for_oplocks = True;
state.id = lck->id;
if (!request_timed_out(request_time, timeout)) {
defer_open(lck, request_time, timeout, req, &state);
}
}
/****************************************************************************
Open a file with a share mode.
****************************************************************************/
NTSTATUS open_file_ntcreate(connection_struct *conn,
struct smb_request *req,
const char *fname,
SMB_STRUCT_STAT *psbuf,
uint32 access_mask, /* access bits (FILE_READ_DATA etc.) */
uint32 share_access, /* share constants (FILE_SHARE_READ etc) */
uint32 create_disposition, /* FILE_OPEN_IF etc. */
uint32 create_options, /* options such as delete on close. */
uint32 new_dos_attributes, /* attributes used for new file. */
int oplock_request, /* internal Samba oplock codes. */
/* Information (FILE_EXISTS etc.) */
int *pinfo,
files_struct **result)
{
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
int flags=0;
int flags2=0;
bool file_existed = VALID_STAT(*psbuf);
bool def_acl = False;
bool posix_open = False;
bool new_file_created = False;
struct file_id id;
NTSTATUS fsp_open = NT_STATUS_ACCESS_DENIED;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
files_struct *fsp = NULL;
mode_t new_unx_mode = (mode_t)0;
mode_t unx_mode = (mode_t)0;
int info;
uint32 existing_dos_attributes = 0;
struct pending_message_list *pml = NULL;
struct timeval request_time = timeval_zero();
struct share_mode_lock *lck = NULL;
uint32 open_access_mask = access_mask;
NTSTATUS status;
int ret_flock;
char *parent_dir;
const char *newname;
ZERO_STRUCT(id);
if (conn->printer) {
/*
* Printers are handled completely differently.
* Most of the passed parameters are ignored.
*/
if (pinfo) {
*pinfo = FILE_WAS_CREATED;
}
DEBUG(10, ("open_file_ntcreate: printer open fname=%s\n", fname));
return print_fsp_open(conn, fname, result);
}
if (!parent_dirname_talloc(talloc_tos(), fname, &parent_dir,
&newname)) {
return NT_STATUS_NO_MEMORY;
}
if (new_dos_attributes & FILE_FLAG_POSIX_SEMANTICS) {
posix_open = True;
unx_mode = (mode_t)(new_dos_attributes & ~FILE_FLAG_POSIX_SEMANTICS);
new_dos_attributes = 0;
} else {
/* We add aARCH to this as this mode is only used if the file is
* created new. */
unx_mode = unix_mode(conn, new_dos_attributes | aARCH, fname,
parent_dir);
}
DEBUG(10, ("open_file_ntcreate: fname=%s, dos_attrs=0x%x "
"access_mask=0x%x share_access=0x%x "
"create_disposition = 0x%x create_options=0x%x "
"unix mode=0%o oplock_request=%d\n",
fname, new_dos_attributes, access_mask, share_access,
create_disposition, create_options, unx_mode,
oplock_request));
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
if ((req == NULL) && ((oplock_request & INTERNAL_OPEN_ONLY) == 0)) {
DEBUG(0, ("No smb request but not an internal only open!\n"));
return NT_STATUS_INTERNAL_ERROR;
}
/*
* Only non-internal opens can be deferred at all
*/
if ((req != NULL)
&& ((pml = get_open_deferred_message(req->mid)) != NULL)) {
struct deferred_open_record *state =
(struct deferred_open_record *)pml->private_data.data;
/* Remember the absolute time of the original
request with this mid. We'll use it later to
see if this has timed out. */
request_time = pml->request_time;
/* Remove the deferred open entry under lock. */
lck = get_share_mode_lock(talloc_tos(), state->id, NULL, NULL);
if (lck == NULL) {
DEBUG(0, ("could not get share mode lock\n"));
} else {
del_deferred_open_entry(lck, req->mid);
TALLOC_FREE(lck);
}
/* Ensure we don't reprocess this message. */
remove_deferred_open_smb_message(req->mid);
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
status = check_name(conn, fname);
if (!NT_STATUS_IS_OK(status)) {
return status;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
if (!posix_open) {
new_dos_attributes &= SAMBA_ATTRIBUTES_MASK;
if (file_existed) {
existing_dos_attributes = dos_mode(conn, fname, psbuf);
}
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
/* ignore any oplock requests if oplocks are disabled */
if (!lp_oplocks(SNUM(conn)) || global_client_failed_oplock_break ||
IS_VETO_OPLOCK_PATH(conn, fname)) {
/* Mask off everything except the private Samba bits. */
oplock_request &= SAMBA_PRIVATE_OPLOCK_MASK;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
/* this is for OS/2 long file names - say we don't support them */
if (!lp_posix_pathnames() && strstr(fname,".+,;=[].")) {
/* OS/2 Workplace shell fix may be main code stream in a later
* release. */
DEBUG(5,("open_file_ntcreate: OS/2 long filenames are not "
"supported.\n"));
if (use_nt_status()) {
return NT_STATUS_OBJECT_NAME_NOT_FOUND;
}
return NT_STATUS_DOS(ERRDOS, ERRcannotopen);
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
switch( create_disposition ) {
/*
* Currently we're using FILE_SUPERSEDE as the same as
* FILE_OVERWRITE_IF but they really are
* different. FILE_SUPERSEDE deletes an existing file
* (requiring delete access) then recreates it.
*/
case FILE_SUPERSEDE:
/* If file exists replace/overwrite. If file doesn't
* exist create. */
flags2 |= (O_CREAT | O_TRUNC);
break;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
case FILE_OVERWRITE_IF:
/* If file exists replace/overwrite. If file doesn't
* exist create. */
flags2 |= (O_CREAT | O_TRUNC);
break;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
case FILE_OPEN:
/* If file exists open. If file doesn't exist error. */
if (!file_existed) {
DEBUG(5,("open_file_ntcreate: FILE_OPEN "
"requested for file %s and file "
"doesn't exist.\n", fname ));
errno = ENOENT;
return NT_STATUS_OBJECT_NAME_NOT_FOUND;
}
break;
case FILE_OVERWRITE:
/* If file exists overwrite. If file doesn't exist
* error. */
if (!file_existed) {
DEBUG(5,("open_file_ntcreate: FILE_OVERWRITE "
"requested for file %s and file "
"doesn't exist.\n", fname ));
errno = ENOENT;
return NT_STATUS_OBJECT_NAME_NOT_FOUND;
}
flags2 |= O_TRUNC;
break;
case FILE_CREATE:
/* If file exists error. If file doesn't exist
* create. */
if (file_existed) {
DEBUG(5,("open_file_ntcreate: FILE_CREATE "
"requested for file %s and file "
"already exists.\n", fname ));
if (S_ISDIR(psbuf->st_mode)) {
errno = EISDIR;
} else {
errno = EEXIST;
}
return map_nt_error_from_unix(errno);
}
flags2 |= (O_CREAT|O_EXCL);
break;
case FILE_OPEN_IF:
/* If file exists open. If file doesn't exist
* create. */
flags2 |= O_CREAT;
break;
default:
return NT_STATUS_INVALID_PARAMETER;
}
/* We only care about matching attributes on file exists and
* overwrite. */
if (!posix_open && file_existed && ((create_disposition == FILE_OVERWRITE) ||
(create_disposition == FILE_OVERWRITE_IF))) {
if (!open_match_attributes(conn, fname,
existing_dos_attributes,
new_dos_attributes, psbuf->st_mode,
unx_mode, &new_unx_mode)) {
DEBUG(5,("open_file_ntcreate: attributes missmatch "
"for file %s (%x %x) (0%o, 0%o)\n",
fname, existing_dos_attributes,
new_dos_attributes,
(unsigned int)psbuf->st_mode,
(unsigned int)unx_mode ));
errno = EACCES;
return NT_STATUS_ACCESS_DENIED;
}
}
/* This is a nasty hack - must fix... JRA. */
if (access_mask == MAXIMUM_ALLOWED_ACCESS) {
open_access_mask = access_mask = FILE_GENERIC_ALL;
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
/*
* Convert GENERIC bits to specific bits.
*/
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
se_map_generic(&access_mask, &file_generic_mapping);
open_access_mask = access_mask;
if (flags2 & O_TRUNC) {
open_access_mask |= FILE_WRITE_DATA; /* This will cause oplock breaks. */
}
DEBUG(10, ("open_file_ntcreate: fname=%s, after mapping "
"access_mask=0x%x\n", fname, access_mask ));
/*
* Note that we ignore the append flag as append does not
* mean the same thing under DOS and Unix.
*/
if (access_mask & (FILE_WRITE_DATA | FILE_APPEND_DATA)) {
/* DENY_DOS opens are always underlying read-write on the
file handle, no matter what the requested access mask
says. */
if ((create_options & NTCREATEX_OPTIONS_PRIVATE_DENY_DOS) ||
access_mask & (FILE_READ_ATTRIBUTES|FILE_READ_DATA|FILE_READ_EA|FILE_EXECUTE)) {
flags = O_RDWR;
} else {
flags = O_WRONLY;
}
} else {
flags = O_RDONLY;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
/*
* Currently we only look at FILE_WRITE_THROUGH for create options.
*/
#if defined(O_SYNC)
if ((create_options & FILE_WRITE_THROUGH) && lp_strict_sync(SNUM(conn))) {
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
flags2 |= O_SYNC;
}
#endif /* O_SYNC */
if (posix_open && (access_mask & FILE_APPEND_DATA)) {
flags2 |= O_APPEND;
}
if (!posix_open && !CAN_WRITE(conn)) {
/*
* We should really return a permission denied error if either
* O_CREAT or O_TRUNC are set, but for compatibility with
* older versions of Samba we just AND them out.
*/
flags2 &= ~(O_CREAT|O_TRUNC);
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
/*
* Ensure we can't write on a read-only share or file.
*/
if (flags != O_RDONLY && file_existed &&
(!CAN_WRITE(conn) || IS_DOS_READONLY(existing_dos_attributes))) {
DEBUG(5,("open_file_ntcreate: write access requested for "
"file %s on read only %s\n",
fname, !CAN_WRITE(conn) ? "share" : "file" ));
errno = EACCES;
return NT_STATUS_ACCESS_DENIED;
}
status = file_new(conn, &fsp);
if(!NT_STATUS_IS_OK(status)) {
return status;
}
fsp->file_id = vfs_file_id_from_sbuf(conn, psbuf);
fsp->share_access = share_access;
fsp->fh->private_options = create_options;
fsp->access_mask = open_access_mask; /* We change this to the
* requested access_mask after
* the open is done. */
fsp->posix_open = posix_open;
/* Ensure no SAMBA_PRIVATE bits can be set. */
fsp->oplock_type = (oplock_request & ~SAMBA_PRIVATE_OPLOCK_MASK);
if (timeval_is_zero(&request_time)) {
request_time = fsp->open_time;
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
if (file_existed) {
id = vfs_file_id_from_sbuf(conn, psbuf);
lck = get_share_mode_lock(talloc_tos(), id,
conn->connectpath,
fname);
if (lck == NULL) {
file_free(fsp);
DEBUG(0, ("Could not get share mode lock\n"));
return NT_STATUS_SHARING_VIOLATION;
}
/* First pass - send break only on batch oplocks. */
if ((req != NULL)
&& delay_for_oplocks(lck, fsp, req->mid, 1,
oplock_request)) {
schedule_defer_open(lck, request_time, req);
TALLOC_FREE(lck);
file_free(fsp);
return NT_STATUS_SHARING_VIOLATION;
}
/* Use the client requested access mask here, not the one we
* open with. */
status = open_mode_check(conn, fname, lck,
access_mask, share_access,
create_options, &file_existed);
if (NT_STATUS_IS_OK(status)) {
/* We might be going to allow this open. Check oplock
* status again. */
/* Second pass - send break for both batch or
* exclusive oplocks. */
if ((req != NULL)
&& delay_for_oplocks(lck, fsp, req->mid, 2,
oplock_request)) {
schedule_defer_open(lck, request_time, req);
TALLOC_FREE(lck);
file_free(fsp);
return NT_STATUS_SHARING_VIOLATION;
}
}
if (NT_STATUS_EQUAL(status, NT_STATUS_DELETE_PENDING)) {
/* DELETE_PENDING is not deferred for a second */
TALLOC_FREE(lck);
file_free(fsp);
return status;
}
if (!NT_STATUS_IS_OK(status)) {
uint32 can_access_mask;
bool can_access = True;
SMB_ASSERT(NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION));
/* Check if this can be done with the deny_dos and fcb
* calls. */
if (create_options &
(NTCREATEX_OPTIONS_PRIVATE_DENY_DOS|
NTCREATEX_OPTIONS_PRIVATE_DENY_FCB)) {
files_struct *fsp_dup;
if (req == NULL) {
DEBUG(0, ("DOS open without an SMB "
"request!\n"));
TALLOC_FREE(lck);
file_free(fsp);
return NT_STATUS_INTERNAL_ERROR;
}
/* Use the client requested access mask here,
* not the one we open with. */
fsp_dup = fcb_or_dos_open(conn, fname, id,
req->smbpid,
req->vuid,
access_mask,
share_access,
create_options);
if (fsp_dup) {
TALLOC_FREE(lck);
file_free(fsp);
if (pinfo) {
*pinfo = FILE_WAS_OPENED;
}
conn->num_files_open++;
*result = fsp_dup;
return NT_STATUS_OK;
}
}
/*
* This next line is a subtlety we need for
* MS-Access. If a file open will fail due to share
* permissions and also for security (access) reasons,
* we need to return the access failed error, not the
* share error. We can't open the file due to kernel
* oplock deadlock (it's possible we failed above on
* the open_mode_check()) so use a userspace check.
*/
if (flags & O_RDWR) {
can_access_mask = FILE_READ_DATA|FILE_WRITE_DATA;
} else if (flags & O_WRONLY) {
can_access_mask = FILE_WRITE_DATA;
} else {
can_access_mask = FILE_READ_DATA;
}
if (((can_access_mask & FILE_WRITE_DATA) && !CAN_WRITE(conn)) ||
!can_access_file(conn,fname,psbuf,can_access_mask)) {
can_access = False;
}
/*
* If we're returning a share violation, ensure we
* cope with the braindead 1 second delay.
*/
if (!(oplock_request & INTERNAL_OPEN_ONLY) &&
lp_defer_sharing_violations()) {
struct timeval timeout;
struct deferred_open_record state;
int timeout_usecs;
/* this is a hack to speed up torture tests
in 'make test' */
timeout_usecs = lp_parm_int(SNUM(conn),
"smbd","sharedelay",
SHARING_VIOLATION_USEC_WAIT);
/* This is a relative time, added to the absolute
request_time value to get the absolute timeout time.
Note that if this is the second or greater time we enter
this codepath for this particular request mid then
request_time is left as the absolute time of the *first*
time this request mid was processed. This is what allows
the request to eventually time out. */
timeout = timeval_set(0, timeout_usecs);
/* Nothing actually uses state.delayed_for_oplocks
but it's handy to differentiate in debug messages
between a 30 second delay due to oplock break, and
a 1 second delay for share mode conflicts. */
state.delayed_for_oplocks = False;
state.id = id;
if ((req != NULL)
&& !request_timed_out(request_time,
timeout)) {
defer_open(lck, request_time, timeout,
req, &state);
}
}
TALLOC_FREE(lck);
if (can_access) {
/*
* We have detected a sharing violation here
* so return the correct error code
*/
status = NT_STATUS_SHARING_VIOLATION;
} else {
status = NT_STATUS_ACCESS_DENIED;
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
file_free(fsp);
return status;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
/*
* We exit this block with the share entry *locked*.....
*/
}
SMB_ASSERT(!file_existed || (lck != NULL));
/*
* Ensure we pay attention to default ACLs on directories if required.
*/
if ((flags2 & O_CREAT) && lp_inherit_acls(SNUM(conn)) &&
(def_acl = directory_has_default_acl(conn, parent_dir))) {
unx_mode = 0777;
}
DEBUG(4,("calling open_file with flags=0x%X flags2=0x%X mode=0%o, "
"access_mask = 0x%x, open_access_mask = 0x%x\n",
(unsigned int)flags, (unsigned int)flags2,
(unsigned int)unx_mode, (unsigned int)access_mask,
(unsigned int)open_access_mask));
/*
* open_file strips any O_TRUNC flags itself.
*/
fsp_open = open_file(fsp, conn, req, parent_dir, newname, fname, psbuf,
flags|flags2, unx_mode, access_mask,
open_access_mask);
if (!NT_STATUS_IS_OK(fsp_open)) {
if (lck != NULL) {
TALLOC_FREE(lck);
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
file_free(fsp);
return fsp_open;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
if (!file_existed) {
/*
* Deal with the race condition where two smbd's detect the
* file doesn't exist and do the create at the same time. One
* of them will win and set a share mode, the other (ie. this
* one) should check if the requested share mode for this
* create is allowed.
*/
/*
* Now the file exists and fsp is successfully opened,
* fsp->dev and fsp->inode are valid and should replace the
* dev=0,inode=0 from a non existent file. Spotted by
* Nadav Danieli <nadavd@exanet.com>. JRA.
*/
id = fsp->file_id;
lck = get_share_mode_lock(talloc_tos(), id,
conn->connectpath,
fname);
if (lck == NULL) {
DEBUG(0, ("open_file_ntcreate: Could not get share "
"mode lock for %s\n", fname));
fd_close(fsp);
file_free(fsp);
return NT_STATUS_SHARING_VIOLATION;
}
/* First pass - send break only on batch oplocks. */
if ((req != NULL)
&& delay_for_oplocks(lck, fsp, req->mid, 1,
oplock_request)) {
schedule_defer_open(lck, request_time, req);
TALLOC_FREE(lck);
fd_close(fsp);
file_free(fsp);
return NT_STATUS_SHARING_VIOLATION;
}
status = open_mode_check(conn, fname, lck,
access_mask, share_access,
create_options, &file_existed);
if (NT_STATUS_IS_OK(status)) {
/* We might be going to allow this open. Check oplock
* status again. */
/* Second pass - send break for both batch or
* exclusive oplocks. */
if ((req != NULL)
&& delay_for_oplocks(lck, fsp, req->mid, 2,
oplock_request)) {
schedule_defer_open(lck, request_time, req);
TALLOC_FREE(lck);
fd_close(fsp);
file_free(fsp);
return NT_STATUS_SHARING_VIOLATION;
}
}
if (!NT_STATUS_IS_OK(status)) {
struct deferred_open_record state;
fd_close(fsp);
file_free(fsp);
state.delayed_for_oplocks = False;
state.id = id;
/* Do it all over again immediately. In the second
* round we will find that the file existed and handle
* the DELETE_PENDING and FCB cases correctly. No need
* to duplicate the code here. Essentially this is a
* "goto top of this function", but don't tell
* anybody... */
if (req != NULL) {
defer_open(lck, request_time, timeval_zero(),
req, &state);
}
TALLOC_FREE(lck);
return status;
}
/*
* We exit this block with the share entry *locked*.....
*/
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
SMB_ASSERT(lck != NULL);
/* note that we ignore failure for the following. It is
basically a hack for NFS, and NFS will never set one of
these only read them. Nobody but Samba can ever set a deny
mode and we have already checked our more authoritative
locking database for permission to set this deny mode. If
the kernel refuses the operations then the kernel is wrong.
note that GPFS supports it as well - jmcd */
if (fsp->fh->fd != -1) {
ret_flock = SMB_VFS_KERNEL_FLOCK(fsp, share_access);
if(ret_flock == -1 ){
TALLOC_FREE(lck);
fd_close(fsp);
file_free(fsp);
return NT_STATUS_SHARING_VIOLATION;
}
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
/*
* At this point onwards, we can guarentee that the share entry
* is locked, whether we created the file or not, and that the
* deny mode is compatible with all current opens.
*/
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
/*
* If requested, truncate the file.
*/
if (flags2&O_TRUNC) {
/*
* We are modifing the file after open - update the stat
* struct..
*/
if ((SMB_VFS_FTRUNCATE(fsp, 0) == -1) ||
(SMB_VFS_FSTAT(fsp, psbuf)==-1)) {
status = map_nt_error_from_unix(errno);
TALLOC_FREE(lck);
fd_close(fsp);
file_free(fsp);
return status;
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
/* Record the options we were opened with. */
fsp->share_access = share_access;
fsp->fh->private_options = create_options;
fsp->access_mask = access_mask;
if (file_existed) {
/* stat opens on existing files don't get oplocks. */
if (is_stat_open(open_access_mask)) {
fsp->oplock_type = NO_OPLOCK;
}
if (!(flags2 & O_TRUNC)) {
info = FILE_WAS_OPENED;
} else {
info = FILE_WAS_OVERWRITTEN;
}
} else {
info = FILE_WAS_CREATED;
}
if (pinfo) {
*pinfo = info;
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
/*
* Setup the oplock info in both the shared memory and
* file structs.
*/
if ((fsp->oplock_type != NO_OPLOCK) &&
(fsp->oplock_type != FAKE_LEVEL_II_OPLOCK)) {
if (!set_file_oplock(fsp, fsp->oplock_type)) {
/* Could not get the kernel oplock */
fsp->oplock_type = NO_OPLOCK;
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
}
if (info == FILE_WAS_OVERWRITTEN || info == FILE_WAS_CREATED || info == FILE_WAS_SUPERSEDED) {
new_file_created = True;
}
set_share_mode(lck, fsp, current_user.ut.uid, 0, fsp->oplock_type, new_file_created);
/* Handle strange delete on close create semantics. */
if ((create_options & FILE_DELETE_ON_CLOSE)
&& (is_ntfs_stream_name(fname)
|| can_set_initial_delete_on_close(lck))) {
status = can_set_delete_on_close(fsp, True, new_dos_attributes);
if (!NT_STATUS_IS_OK(status)) {
/* Remember to delete the mode we just added. */
del_share_mode(lck, fsp);
TALLOC_FREE(lck);
fd_close(fsp);
file_free(fsp);
return status;
}
/* Note that here we set the *inital* delete on close flag,
not the regular one. The magic gets handled in close. */
fsp->initial_delete_on_close = True;
}
if (new_file_created) {
/* Files should be initially set as archive */
if (lp_map_archive(SNUM(conn)) ||
lp_store_dos_attributes(SNUM(conn))) {
if (!posix_open) {
SMB_STRUCT_STAT tmp_sbuf;
SET_STAT_INVALID(tmp_sbuf);
if (file_set_dosmode(
conn, fname,
new_dos_attributes | aARCH,
&tmp_sbuf, parent_dir,
true) == 0) {
unx_mode = tmp_sbuf.st_mode;
}
}
}
}
/*
* Take care of inherited ACLs on created files - if default ACL not
* selected.
*/
if (!posix_open && !file_existed && !def_acl) {
int saved_errno = errno; /* We might get ENOSYS in the next
* call.. */
if (SMB_VFS_FCHMOD_ACL(fsp, unx_mode) == -1 &&
errno == ENOSYS) {
errno = saved_errno; /* Ignore ENOSYS */
}
} else if (new_unx_mode) {
int ret = -1;
/* Attributes need changing. File already existed. */
{
int saved_errno = errno; /* We might get ENOSYS in the
* next call.. */
ret = SMB_VFS_FCHMOD_ACL(fsp, new_unx_mode);
if (ret == -1 && errno == ENOSYS) {
errno = saved_errno; /* Ignore ENOSYS */
} else {
DEBUG(5, ("open_file_ntcreate: reset "
"attributes of file %s to 0%o\n",
fname, (unsigned int)new_unx_mode));
ret = 0; /* Don't do the fchmod below. */
}
}
if ((ret == -1) &&
(SMB_VFS_FCHMOD(fsp, new_unx_mode) == -1))
DEBUG(5, ("open_file_ntcreate: failed to reset "
"attributes of file %s to 0%o\n",
fname, (unsigned int)new_unx_mode));
}
/* If this is a successful open, we must remove any deferred open
* records. */
if (req != NULL) {
del_deferred_open_entry(lck, req->mid);
}
TALLOC_FREE(lck);
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
conn->num_files_open++;
*result = fsp;
return NT_STATUS_OK;
}
/****************************************************************************
Open a file for for write to ensure that we can fchmod it.
****************************************************************************/
NTSTATUS open_file_fchmod(connection_struct *conn, const char *fname,
SMB_STRUCT_STAT *psbuf, files_struct **result)
{
files_struct *fsp = NULL;
NTSTATUS status;
if (!VALID_STAT(*psbuf)) {
return NT_STATUS_INVALID_PARAMETER;
}
status = file_new(conn, &fsp);
if(!NT_STATUS_IS_OK(status)) {
return status;
}
/* note! we must use a non-zero desired access or we don't get
a real file descriptor. Oh what a twisted web we weave. */
status = open_file(fsp, conn, NULL, NULL, NULL, fname, psbuf, O_WRONLY,
0, FILE_WRITE_DATA, FILE_WRITE_DATA);
/*
* This is not a user visible file open.
* Don't set a share mode and don't increment
* the conn->num_files_open.
*/
if (!NT_STATUS_IS_OK(status)) {
file_free(fsp);
return status;
}
*result = fsp;
return NT_STATUS_OK;
}
/****************************************************************************
Close the fchmod file fd - ensure no locks are lost.
****************************************************************************/
NTSTATUS close_file_fchmod(files_struct *fsp)
{
NTSTATUS status = fd_close(fsp);
file_free(fsp);
return status;
}
static NTSTATUS mkdir_internal(connection_struct *conn,
const char *name,
uint32 file_attributes,
SMB_STRUCT_STAT *psbuf)
{
mode_t mode;
char *parent_dir;
const char *dirname;
NTSTATUS status;
bool posix_open = false;
if(!CAN_WRITE(conn)) {
DEBUG(5,("mkdir_internal: failing create on read-only share "
"%s\n", lp_servicename(SNUM(conn))));
return NT_STATUS_ACCESS_DENIED;
}
status = check_name(conn, name);
if (!NT_STATUS_IS_OK(status)) {
return status;
}
if (!parent_dirname_talloc(talloc_tos(), name, &parent_dir,
&dirname)) {
return NT_STATUS_NO_MEMORY;
}
if (file_attributes & FILE_FLAG_POSIX_SEMANTICS) {
posix_open = true;
mode = (mode_t)(file_attributes & ~FILE_FLAG_POSIX_SEMANTICS);
} else {
mode = unix_mode(conn, aDIR, name, parent_dir);
}
if (SMB_VFS_MKDIR(conn, name, mode) != 0) {
return map_nt_error_from_unix(errno);
}
/* Ensure we're checking for a symlink here.... */
/* We don't want to get caught by a symlink racer. */
if (SMB_VFS_LSTAT(conn, name, psbuf) == -1) {
DEBUG(2, ("Could not stat directory '%s' just created: %s\n",
name, strerror(errno)));
return map_nt_error_from_unix(errno);
}
if (!S_ISDIR(psbuf->st_mode)) {
DEBUG(0, ("Directory just '%s' created is not a directory\n",
name));
return NT_STATUS_ACCESS_DENIED;
}
if (lp_store_dos_attributes(SNUM(conn))) {
if (!posix_open) {
file_set_dosmode(conn, name,
file_attributes | aDIR, NULL,
parent_dir,
true);
}
}
if (lp_inherit_perms(SNUM(conn))) {
inherit_access_acl(conn, parent_dir, name, mode);
}
if (!(file_attributes & FILE_FLAG_POSIX_SEMANTICS)) {
/*
* Check if high bits should have been set,
* then (if bits are missing): add them.
* Consider bits automagically set by UNIX, i.e. SGID bit from parent
* dir.
*/
if (mode & ~(S_IRWXU|S_IRWXG|S_IRWXO) && (mode & ~psbuf->st_mode)) {
SMB_VFS_CHMOD(conn, name,
psbuf->st_mode | (mode & ~psbuf->st_mode));
}
}
/* Change the owner if required. */
if (lp_inherit_owner(SNUM(conn))) {
change_dir_owner_to_parent(conn, parent_dir, name, psbuf);
}
notify_fname(conn, NOTIFY_ACTION_ADDED, FILE_NOTIFY_CHANGE_DIR_NAME,
name);
return NT_STATUS_OK;
}
/****************************************************************************
Open a directory from an NT SMB call.
****************************************************************************/
NTSTATUS open_directory(connection_struct *conn,
struct smb_request *req,
const char *fname,
SMB_STRUCT_STAT *psbuf,
uint32 access_mask,
uint32 share_access,
uint32 create_disposition,
uint32 create_options,
uint32 file_attributes,
int *pinfo,
files_struct **result)
{
files_struct *fsp = NULL;
bool dir_existed = VALID_STAT(*psbuf) ? True : False;
struct share_mode_lock *lck = NULL;
NTSTATUS status;
int info = 0;
DEBUG(5,("open_directory: opening directory %s, access_mask = 0x%x, "
"share_access = 0x%x create_options = 0x%x, "
"create_disposition = 0x%x, file_attributes = 0x%x\n",
fname,
(unsigned int)access_mask,
(unsigned int)share_access,
(unsigned int)create_options,
(unsigned int)create_disposition,
(unsigned int)file_attributes));
if (!(file_attributes & FILE_FLAG_POSIX_SEMANTICS) && is_ntfs_stream_name(fname)) {
DEBUG(2, ("open_directory: %s is a stream name!\n", fname));
return NT_STATUS_NOT_A_DIRECTORY;
}
switch( create_disposition ) {
case FILE_OPEN:
info = FILE_WAS_OPENED;
/*
* We want to follow symlinks here.
*/
if (SMB_VFS_STAT(conn, fname, psbuf) != 0) {
return map_nt_error_from_unix(errno);
}
break;
case FILE_CREATE:
/* If directory exists error. If directory doesn't
* exist create. */
status = mkdir_internal(conn,
fname,
file_attributes,
psbuf);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(2, ("open_directory: unable to create "
"%s. Error was %s\n", fname,
nt_errstr(status)));
return status;
}
info = FILE_WAS_CREATED;
break;
case FILE_OPEN_IF:
/*
* If directory exists open. If directory doesn't
* exist create.
*/
status = mkdir_internal(conn,
fname,
file_attributes,
psbuf);
if (NT_STATUS_IS_OK(status)) {
info = FILE_WAS_CREATED;
}
if (NT_STATUS_EQUAL(status,
NT_STATUS_OBJECT_NAME_COLLISION)) {
info = FILE_WAS_OPENED;
status = NT_STATUS_OK;
}
break;
case FILE_SUPERSEDE:
case FILE_OVERWRITE:
case FILE_OVERWRITE_IF:
default:
DEBUG(5,("open_directory: invalid create_disposition "
"0x%x for directory %s\n",
(unsigned int)create_disposition, fname));
return NT_STATUS_INVALID_PARAMETER;
}
if(!S_ISDIR(psbuf->st_mode)) {
DEBUG(5,("open_directory: %s is not a directory !\n",
fname ));
return NT_STATUS_NOT_A_DIRECTORY;
}
status = file_new(conn, &fsp);
if(!NT_STATUS_IS_OK(status)) {
return status;
}
/*
* Setup the files_struct for it.
*/
fsp->mode = psbuf->st_mode;
fsp->file_id = vfs_file_id_from_sbuf(conn, psbuf);
fsp->vuid = req ? req->vuid : UID_FIELD_INVALID;
fsp->file_pid = req ? req->smbpid : 0;
fsp->can_lock = False;
fsp->can_read = False;
fsp->can_write = False;
fsp->share_access = share_access;
fsp->fh->private_options = create_options;
fsp->access_mask = access_mask;
fsp->print_file = False;
fsp->modified = False;
fsp->oplock_type = NO_OPLOCK;
fsp->sent_oplock_break = NO_BREAK_SENT;
fsp->is_directory = True;
fsp->is_stat = False;
fsp->posix_open = (file_attributes & FILE_FLAG_POSIX_SEMANTICS) ? True : False;
string_set(&fsp->fsp_name,fname);
lck = get_share_mode_lock(talloc_tos(), fsp->file_id,
conn->connectpath,
fname);
if (lck == NULL) {
DEBUG(0, ("open_directory: Could not get share mode lock for %s\n", fname));
file_free(fsp);
return NT_STATUS_SHARING_VIOLATION;
}
status = open_mode_check(conn, fname, lck,
access_mask, share_access,
create_options, &dir_existed);
if (!NT_STATUS_IS_OK(status)) {
TALLOC_FREE(lck);
file_free(fsp);
return status;
}
set_share_mode(lck, fsp, current_user.ut.uid, 0, NO_OPLOCK, True);
/* For directories the delete on close bit at open time seems
always to be honored on close... See test 19 in Samba4 BASE-DELETE. */
if (create_options & FILE_DELETE_ON_CLOSE) {
status = can_set_delete_on_close(fsp, True, 0);
if (!NT_STATUS_IS_OK(status) && !NT_STATUS_EQUAL(status, NT_STATUS_DIRECTORY_NOT_EMPTY)) {
TALLOC_FREE(lck);
file_free(fsp);
return status;
}
if (NT_STATUS_IS_OK(status)) {
/* Note that here we set the *inital* delete on close flag,
not the regular one. The magic gets handled in close. */
fsp->initial_delete_on_close = True;
}
}
TALLOC_FREE(lck);
if (pinfo) {
*pinfo = info;
}
This is a *big* checkin that may break some things, but implements the new open mechanism Andrew & I discussed. config.sub: configure: Included the QNX patch. include/vfs.h: smbd/vfs-wrap.c: smbd/vfs.c: Added ftruncate vfs call (needed). Note that we will also need locking calls in the vfs (to be added). lib/util_unistr.c: nmbd/nmbd_processlogon.c: Fix for NT domain logons causing nmbd to core dump. Also fix for sidsize DOS bug. locking/locking.c: Check value of ret before using it for memdup. printing/printing.c: Convert print_fsp_open to return an allocated fsp. rpc_server/srv_lsa.c: Fix for NT domain logons. I have removed all use of lp_share_modes() from the code (although I left the parameter in the table for backwards compatibility). It no longer makes sense for this to exist. smbd/close.c: Removed lp_share_modes(). smbd/fileio.c: Fixed parameters to unlock_share_entry call in panic code. smbd/files.c: Correctly set the unix_ERR_code to ERRnofids on fsp allocation fail. smbd/nttrans.c: smbd/reply.c: smbd/trans2.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. smbd/open.c: Changed all occurrences of open_file_shared/open_directory/ open_file_stat to return an fsp from the call. In addition I have fixed a long standing race condition in the deny mode processing w.r.t. two smbd's creating a file. Andrew, please note that your original idea of using open with O_EXCL in this case would not work (I went over the races very carefully) and so we must re-check deny modes *after* the open() call returns. This is because there is a race between the open with O_EXCL and the lock of the share mode entry. Imagine the case where the first smbd does the open with O_EXCL and a deny mode of DENY_ALL, but is pre-empted before it locks the share modes and creates the deny mode entry for DENY_ALL. A second smbd could then come in with O_RDONLY and a deny mode of DENY_NONE and the two opens would be allowed. The *only* way to fix this race is to lock the share modes after the open and then do the deny mode checks *after* this lock in the case where the file did not originally exist. This code will need extensive testing but seems to initially work. Jeremy. (This used to be commit ab0ecc39d688f16b9692fe90b991f0b89287070a)
2000-04-22 04:33:16 +04:00
conn->num_files_open++;
*result = fsp;
return NT_STATUS_OK;
}
NTSTATUS create_directory(connection_struct *conn, struct smb_request *req, const char *directory)
{
NTSTATUS status;
SMB_STRUCT_STAT sbuf;
files_struct *fsp;
SET_STAT_INVALID(sbuf);
status = open_directory(conn, req, directory, &sbuf,
FILE_READ_ATTRIBUTES, /* Just a stat open */
FILE_SHARE_NONE, /* Ignored for stat opens */
FILE_CREATE,
0,
FILE_ATTRIBUTE_DIRECTORY,
NULL,
&fsp);
if (NT_STATUS_IS_OK(status)) {
close_file(fsp, NORMAL_CLOSE);
}
return status;
}
/****************************************************************************
Open a pseudo-file (no locking checks - a 'stat' open).
****************************************************************************/
NTSTATUS open_file_stat(connection_struct *conn, struct smb_request *req,
const char *fname, SMB_STRUCT_STAT *psbuf,
files_struct **result)
{
files_struct *fsp = NULL;
NTSTATUS status;
if (!VALID_STAT(*psbuf)) {
return NT_STATUS_INVALID_PARAMETER;
}
/* Can't 'stat' open directories. */
if(S_ISDIR(psbuf->st_mode)) {
return NT_STATUS_FILE_IS_A_DIRECTORY;
}
status = file_new(conn, &fsp);
if(!NT_STATUS_IS_OK(status)) {
return status;
}
DEBUG(5,("open_file_stat: 'opening' file %s\n", fname));
/*
* Setup the files_struct for it.
*/
fsp->mode = psbuf->st_mode;
fsp->file_id = vfs_file_id_from_sbuf(conn, psbuf);
fsp->vuid = req ? req->vuid : UID_FIELD_INVALID;
fsp->file_pid = req ? req->smbpid : 0;
fsp->can_lock = False;
fsp->can_read = False;
fsp->can_write = False;
fsp->print_file = False;
fsp->modified = False;
fsp->oplock_type = NO_OPLOCK;
fsp->sent_oplock_break = NO_BREAK_SENT;
fsp->is_directory = False;
fsp->is_stat = True;
string_set(&fsp->fsp_name,fname);
conn->num_files_open++;
*result = fsp;
return NT_STATUS_OK;
}
/****************************************************************************
Receive notification that one of our open files has been renamed by another
smbd process.
****************************************************************************/
void msg_file_was_renamed(struct messaging_context *msg,
void *private_data,
uint32_t msg_type,
struct server_id server_id,
DATA_BLOB *data)
{
files_struct *fsp;
char *frm = (char *)data->data;
struct file_id id;
const char *sharepath;
const char *newname;
size_t sp_len;
if (data->data == NULL
|| data->length < MSG_FILE_RENAMED_MIN_SIZE + 2) {
DEBUG(0, ("msg_file_was_renamed: Got invalid msg len %d\n",
(int)data->length));
return;
}
/* Unpack the message. */
pull_file_id_16(frm, &id);
sharepath = &frm[16];
newname = sharepath + strlen(sharepath) + 1;
sp_len = strlen(sharepath);
DEBUG(10,("msg_file_was_renamed: Got rename message for sharepath %s, new name %s, "
"file_id %s\n",
sharepath, newname, file_id_string_tos(&id)));
for(fsp = file_find_di_first(id); fsp; fsp = file_find_di_next(fsp)) {
if (memcmp(fsp->conn->connectpath, sharepath, sp_len) == 0) {
DEBUG(10,("msg_file_was_renamed: renaming file fnum %d from %s -> %s\n",
fsp->fnum, fsp->fsp_name, newname ));
string_set(&fsp->fsp_name, newname);
} else {
/* TODO. JRA. */
/* Now we have the complete path we can work out if this is
actually within this share and adjust newname accordingly. */
DEBUG(10,("msg_file_was_renamed: share mismatch (sharepath %s "
"not sharepath %s) "
"fnum %d from %s -> %s\n",
fsp->conn->connectpath,
sharepath,
fsp->fnum,
fsp->fsp_name,
newname ));
}
}
}
struct case_semantics_state {
connection_struct *conn;
bool case_sensitive;
bool case_preserve;
bool short_case_preserve;
};
/****************************************************************************
Restore case semantics.
****************************************************************************/
static int restore_case_semantics(struct case_semantics_state *state)
{
state->conn->case_sensitive = state->case_sensitive;
state->conn->case_preserve = state->case_preserve;
state->conn->short_case_preserve = state->short_case_preserve;
return 0;
}
/****************************************************************************
Save case semantics.
****************************************************************************/
static struct case_semantics_state *set_posix_case_semantics(TALLOC_CTX *mem_ctx,
connection_struct *conn)
{
struct case_semantics_state *result;
if (!(result = talloc(mem_ctx, struct case_semantics_state))) {
DEBUG(0, ("talloc failed\n"));
return NULL;
}
result->conn = conn;
result->case_sensitive = conn->case_sensitive;
result->case_preserve = conn->case_preserve;
result->short_case_preserve = conn->short_case_preserve;
/* Set to POSIX. */
conn->case_sensitive = True;
conn->case_preserve = True;
conn->short_case_preserve = True;
talloc_set_destructor(result, restore_case_semantics);
return result;
}
/*
* If a main file is opened for delete, all streams need to be checked for
* !FILE_SHARE_DELETE. Do this by opening with DELETE_ACCESS.
* If that works, delete them all by setting the delete on close and close.
*/
static NTSTATUS open_streams_for_delete(connection_struct *conn,
const char *fname)
{
struct stream_struct *stream_info;
files_struct **streams;
int i;
unsigned int num_streams;
TALLOC_CTX *frame = talloc_stackframe();
NTSTATUS status;
status = SMB_VFS_STREAMINFO(conn, NULL, fname, talloc_tos(),
&num_streams, &stream_info);
if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_IMPLEMENTED)) {
DEBUG(10, ("no streams around\n"));
TALLOC_FREE(frame);
return NT_STATUS_OK;
}
if (!NT_STATUS_IS_OK(status)) {
DEBUG(10, ("SMB_VFS_STREAMINFO failed: %s\n",
nt_errstr(status)));
goto fail;
}
DEBUG(10, ("open_streams_for_delete found %d streams\n",
num_streams));
if (num_streams == 0) {
TALLOC_FREE(frame);
return NT_STATUS_OK;
}
streams = TALLOC_ARRAY(talloc_tos(), files_struct *, num_streams);
if (streams == NULL) {
DEBUG(0, ("talloc failed\n"));
status = NT_STATUS_NO_MEMORY;
goto fail;
}
for (i=0; i<num_streams; i++) {
char *streamname;
if (strequal(stream_info[i].name, "::$DATA")) {
streams[i] = NULL;
continue;
}
streamname = talloc_asprintf(talloc_tos(), "%s%s", fname,
stream_info[i].name);
if (streamname == NULL) {
DEBUG(0, ("talloc_aprintf failed\n"));
status = NT_STATUS_NO_MEMORY;
goto fail;
}
status = create_file_unixpath
(conn, /* conn */
NULL, /* req */
streamname, /* fname */
DELETE_ACCESS, /* access_mask */
FILE_SHARE_READ | FILE_SHARE_WRITE
| FILE_SHARE_DELETE, /* share_access */
FILE_OPEN, /* create_disposition*/
NTCREATEX_OPTIONS_PRIVATE_STREAM_DELETE, /* create_options */
FILE_ATTRIBUTE_NORMAL, /* file_attributes */
0, /* oplock_request */
0, /* allocation_size */
NULL, /* sd */
NULL, /* ea_list */
&streams[i], /* result */
NULL, /* pinfo */
NULL); /* psbuf */
TALLOC_FREE(streamname);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(10, ("Could not open stream %s: %s\n",
streamname, nt_errstr(status)));
break;
}
}
/*
* don't touch the variable "status" beyond this point :-)
*/
for (i -= 1 ; i >= 0; i--) {
if (streams[i] == NULL) {
continue;
}
DEBUG(10, ("Closing stream # %d, %s\n", i,
streams[i]->fsp_name));
close_file(streams[i], NORMAL_CLOSE);
}
fail:
TALLOC_FREE(frame);
return status;
}
/*
* Wrapper around open_file_ntcreate and open_directory
*/
NTSTATUS create_file_unixpath(connection_struct *conn,
struct smb_request *req,
const char *fname,
uint32_t access_mask,
uint32_t share_access,
uint32_t create_disposition,
uint32_t create_options,
uint32_t file_attributes,
uint32_t oplock_request,
SMB_BIG_UINT allocation_size,
struct security_descriptor *sd,
struct ea_list *ea_list,
files_struct **result,
int *pinfo,
SMB_STRUCT_STAT *psbuf)
{
SMB_STRUCT_STAT sbuf;
int info = FILE_WAS_OPENED;
files_struct *base_fsp = NULL;
files_struct *fsp = NULL;
NTSTATUS status;
DEBUG(10,("create_file_unixpath: access_mask = 0x%x "
"file_attributes = 0x%x, share_access = 0x%x, "
"create_disposition = 0x%x create_options = 0x%x "
"oplock_request = 0x%x ea_list = 0x%p, sd = 0x%p, "
"fname = %s\n",
(unsigned int)access_mask,
(unsigned int)file_attributes,
(unsigned int)share_access,
(unsigned int)create_disposition,
(unsigned int)create_options,
(unsigned int)oplock_request,
ea_list, sd, fname));
if (create_options & FILE_OPEN_BY_FILE_ID) {
status = NT_STATUS_NOT_SUPPORTED;
goto fail;
}
if (req == NULL) {
oplock_request |= INTERNAL_OPEN_ONLY;
}
if (psbuf != NULL) {
sbuf = *psbuf;
}
else {
if (SMB_VFS_STAT(conn, fname, &sbuf) == -1) {
SET_STAT_INVALID(sbuf);
}
}
if ((conn->fs_capabilities & FILE_NAMED_STREAMS)
&& (access_mask & DELETE_ACCESS)
&& !is_ntfs_stream_name(fname)) {
/*
* We can't open a file with DELETE access if any of the
* streams is open without FILE_SHARE_DELETE
*/
status = open_streams_for_delete(conn, fname);
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
}
/* This is the correct thing to do (check every time) but can_delete
* is expensive (it may have to read the parent directory
* permissions). So for now we're not doing it unless we have a strong
* hint the client is really going to delete this file. If the client
* is forcing FILE_CREATE let the filesystem take care of the
* permissions. */
/* Setting FILE_SHARE_DELETE is the hint. */
if (lp_acl_check_permissions(SNUM(conn))
&& (create_disposition != FILE_CREATE)
&& (share_access & FILE_SHARE_DELETE)
&& (access_mask & DELETE_ACCESS)
&& (((dos_mode(conn, fname, &sbuf) & FILE_ATTRIBUTE_READONLY)
&& !lp_delete_readonly(SNUM(conn)))
|| !can_delete_file_in_directory(conn, fname))) {
status = NT_STATUS_ACCESS_DENIED;
goto fail;
}
#if 0
/* We need to support SeSecurityPrivilege for this. */
if ((access_mask & SEC_RIGHT_SYSTEM_SECURITY) &&
!user_has_privileges(current_user.nt_user_token,
&se_security)) {
status = NT_STATUS_PRIVILEGE_NOT_HELD;
goto fail;
}
#endif
if ((conn->fs_capabilities & FILE_NAMED_STREAMS)
&& is_ntfs_stream_name(fname)
&& (!(create_options & NTCREATEX_OPTIONS_PRIVATE_STREAM_DELETE))) {
char *base;
uint32 base_create_disposition;
if (create_options & FILE_DIRECTORY_FILE) {
status = NT_STATUS_NOT_A_DIRECTORY;
goto fail;
}
status = split_ntfs_stream_name(talloc_tos(), fname,
&base, NULL);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(10, ("split_ntfs_stream_name failed: %s\n",
nt_errstr(status)));
goto fail;
}
SMB_ASSERT(!is_ntfs_stream_name(base)); /* paranoia.. */
switch (create_disposition) {
case FILE_OPEN:
base_create_disposition = FILE_OPEN;
break;
default:
base_create_disposition = FILE_OPEN_IF;
break;
}
status = create_file_unixpath(conn, NULL, base, 0,
FILE_SHARE_READ
| FILE_SHARE_WRITE
| FILE_SHARE_DELETE,
base_create_disposition,
0, 0, 0, 0, NULL, NULL,
&base_fsp, NULL, NULL);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(10, ("create_file_unixpath for base %s failed: "
"%s\n", base, nt_errstr(status)));
goto fail;
}
}
/*
* If it's a request for a directory open, deal with it separately.
*/
if (create_options & FILE_DIRECTORY_FILE) {
if (create_options & FILE_NON_DIRECTORY_FILE) {
status = NT_STATUS_INVALID_PARAMETER;
goto fail;
}
/* Can't open a temp directory. IFS kit test. */
if (file_attributes & FILE_ATTRIBUTE_TEMPORARY) {
status = NT_STATUS_INVALID_PARAMETER;
goto fail;
}
/*
* We will get a create directory here if the Win32
* app specified a security descriptor in the
* CreateDirectory() call.
*/
oplock_request = 0;
status = open_directory(
conn, req, fname, &sbuf, access_mask, share_access,
create_disposition, create_options, file_attributes,
&info, &fsp);
} else {
/*
* Ordinary file case.
*/
status = open_file_ntcreate(
conn, req, fname, &sbuf, access_mask, share_access,
create_disposition, create_options, file_attributes,
oplock_request, &info, &fsp);
if (NT_STATUS_EQUAL(status, NT_STATUS_FILE_IS_A_DIRECTORY)) {
/*
* Fail the open if it was explicitly a non-directory
* file.
*/
if (create_options & FILE_NON_DIRECTORY_FILE) {
status = NT_STATUS_FILE_IS_A_DIRECTORY;
goto fail;
}
oplock_request = 0;
status = open_directory(
conn, req, fname, &sbuf, access_mask,
share_access, create_disposition,
create_options, file_attributes,
&info, &fsp);
}
}
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
/*
* According to the MS documentation, the only time the security
* descriptor is applied to the opened file is iff we *created* the
* file; an existing file stays the same.
*
* Also, it seems (from observation) that you can open the file with
* any access mask but you can still write the sd. We need to override
* the granted access before we call set_sd
* Patch for bug #2242 from Tom Lackemann <cessnatomny@yahoo.com>.
*/
if ((sd != NULL) && (info == FILE_WAS_CREATED)
&& lp_nt_acl_support(SNUM(conn))) {
uint32_t sec_info_sent = ALL_SECURITY_INFORMATION;
uint32_t saved_access_mask = fsp->access_mask;
if (sd->owner_sid == NULL) {
sec_info_sent &= ~OWNER_SECURITY_INFORMATION;
}
if (sd->group_sid == NULL) {
sec_info_sent &= ~GROUP_SECURITY_INFORMATION;
}
if (sd->sacl == NULL) {
sec_info_sent &= ~SACL_SECURITY_INFORMATION;
}
if (sd->dacl == NULL) {
sec_info_sent &= ~DACL_SECURITY_INFORMATION;
}
fsp->access_mask = FILE_GENERIC_ALL;
status = SMB_VFS_FSET_NT_ACL(fsp, sec_info_sent, sd);
fsp->access_mask = saved_access_mask;
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
}
if ((ea_list != NULL) && (info == FILE_WAS_CREATED)) {
status = set_ea(conn, fsp, fname, ea_list);
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
}
if (!fsp->is_directory && S_ISDIR(sbuf.st_mode)) {
status = NT_STATUS_ACCESS_DENIED;
goto fail;
}
/* Save the requested allocation size. */
if ((info == FILE_WAS_CREATED) || (info == FILE_WAS_OVERWRITTEN)) {
if (allocation_size
&& (allocation_size > sbuf.st_size)) {
fsp->initial_allocation_size = smb_roundup(
fsp->conn, allocation_size);
if (fsp->is_directory) {
/* Can't set allocation size on a directory. */
status = NT_STATUS_ACCESS_DENIED;
goto fail;
}
if (vfs_allocate_file_space(
fsp, fsp->initial_allocation_size) == -1) {
status = NT_STATUS_DISK_FULL;
goto fail;
}
} else {
fsp->initial_allocation_size = smb_roundup(
fsp->conn, (SMB_BIG_UINT)sbuf.st_size);
}
}
DEBUG(10, ("create_file: info=%d\n", info));
/*
* Set fsp->base_fsp late enough that we can't "goto fail" anymore. In
* the fail: branch we call close_file(fsp, ERROR_CLOSE) which would
* also close fsp->base_fsp which we have to also do explicitly in
* this routine here, as not in all "goto fail:" we have the fsp set
* up already to be initialized with the base_fsp.
*/
fsp->base_fsp = base_fsp;
*result = fsp;
if (pinfo != NULL) {
*pinfo = info;
}
if (psbuf != NULL) {
if ((fsp->fh == NULL) || (fsp->fh->fd == -1)) {
*psbuf = sbuf;
}
else {
SMB_VFS_FSTAT(fsp, psbuf);
}
}
return NT_STATUS_OK;
fail:
DEBUG(10, ("create_file: %s\n", nt_errstr(status)));
if (fsp != NULL) {
close_file(fsp, ERROR_CLOSE);
fsp = NULL;
}
if (base_fsp != NULL) {
close_file(base_fsp, ERROR_CLOSE);
base_fsp = NULL;
}
return status;
}
NTSTATUS create_file(connection_struct *conn,
struct smb_request *req,
uint16_t root_dir_fid,
const char *fname,
uint32_t access_mask,
uint32_t share_access,
uint32_t create_disposition,
uint32_t create_options,
uint32_t file_attributes,
uint32_t oplock_request,
SMB_BIG_UINT allocation_size,
struct security_descriptor *sd,
struct ea_list *ea_list,
files_struct **result,
int *pinfo,
SMB_STRUCT_STAT *psbuf)
{
struct case_semantics_state *case_state = NULL;
SMB_STRUCT_STAT sbuf;
int info = FILE_WAS_OPENED;
files_struct *fsp = NULL;
NTSTATUS status;
DEBUG(10,("create_file: access_mask = 0x%x "
"file_attributes = 0x%x, share_access = 0x%x, "
"create_disposition = 0x%x create_options = 0x%x "
"oplock_request = 0x%x "
"root_dir_fid = 0x%x, ea_list = 0x%p, sd = 0x%p, "
"fname = %s\n",
(unsigned int)access_mask,
(unsigned int)file_attributes,
(unsigned int)share_access,
(unsigned int)create_disposition,
(unsigned int)create_options,
(unsigned int)oplock_request,
(unsigned int)root_dir_fid,
ea_list, sd, fname));
/*
* Get the file name.
*/
if (root_dir_fid != 0) {
/*
* This filename is relative to a directory fid.
*/
char *parent_fname = NULL;
files_struct *dir_fsp = file_fsp(root_dir_fid);
if (dir_fsp == NULL) {
status = NT_STATUS_INVALID_HANDLE;
goto fail;
}
if (!dir_fsp->is_directory) {
/*
* Check to see if this is a mac fork of some kind.
*/
if (is_ntfs_stream_name(fname)) {
status = NT_STATUS_OBJECT_PATH_NOT_FOUND;
goto fail;
}
/*
we need to handle the case when we get a
relative open relative to a file and the
pathname is blank - this is a reopen!
(hint from demyn plantenberg)
*/
status = NT_STATUS_INVALID_HANDLE;
goto fail;
}
if (ISDOT(dir_fsp->fsp_name)) {
/*
* We're at the toplevel dir, the final file name
* must not contain ./, as this is filtered out
* normally by srvstr_get_path and unix_convert
* explicitly rejects paths containing ./.
*/
parent_fname = talloc_strdup(talloc_tos(), "");
if (parent_fname == NULL) {
status = NT_STATUS_NO_MEMORY;
goto fail;
}
} else {
size_t dir_name_len = strlen(dir_fsp->fsp_name);
/*
* Copy in the base directory name.
*/
parent_fname = TALLOC_ARRAY(talloc_tos(), char,
dir_name_len+2);
if (parent_fname == NULL) {
status = NT_STATUS_NO_MEMORY;
goto fail;
}
memcpy(parent_fname, dir_fsp->fsp_name,
dir_name_len+1);
/*
* Ensure it ends in a '/'.
* We used TALLOC_SIZE +2 to add space for the '/'.
*/
if(dir_name_len
&& (parent_fname[dir_name_len-1] != '\\')
&& (parent_fname[dir_name_len-1] != '/')) {
parent_fname[dir_name_len] = '/';
parent_fname[dir_name_len+1] = '\0';
}
}
fname = talloc_asprintf(talloc_tos(), "%s%s", parent_fname,
fname);
if (fname == NULL) {
status = NT_STATUS_NO_MEMORY;
goto fail;
}
}
/*
* Check to see if this is a mac fork of some kind.
*/
if (is_ntfs_stream_name(fname)) {
enum FAKE_FILE_TYPE fake_file_type;
fake_file_type = is_fake_file(fname);
if (fake_file_type != FAKE_FILE_TYPE_NONE) {
/*
* Here we go! support for changing the disk quotas
* --metze
*
* We need to fake up to open this MAGIC QUOTA file
* and return a valid FID.
*
* w2k close this file directly after openening xp
* also tries a QUERY_FILE_INFO on the file and then
* close it
*/
status = open_fake_file(conn, fake_file_type, fname,
access_mask, &fsp);
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
SET_STAT_INVALID(sbuf);
goto done;
}
}
if ((req != NULL) && (req->flags2 & FLAGS2_DFS_PATHNAMES)) {
char *resolved_fname;
status = resolve_dfspath(talloc_tos(), conn, true, fname,
&resolved_fname);
if (!NT_STATUS_IS_OK(status)) {
/*
* For PATH_NOT_COVERED we had
* reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
* ERRSRV, ERRbadpath);
* Need to fix in callers
*/
goto fail;
}
fname = resolved_fname;
}
/*
* Check if POSIX semantics are wanted.
*/
if (file_attributes & FILE_FLAG_POSIX_SEMANTICS) {
case_state = set_posix_case_semantics(talloc_tos(), conn);
file_attributes &= ~FILE_FLAG_POSIX_SEMANTICS;
}
{
char *converted_fname;
SET_STAT_INVALID(sbuf);
status = unix_convert(talloc_tos(), conn, fname, False,
&converted_fname, NULL, &sbuf);
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
fname = converted_fname;
}
/* All file access must go through check_name() */
status = check_name(conn, fname);
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
status = create_file_unixpath(
conn, req, fname, access_mask, share_access,
create_disposition, create_options, file_attributes,
oplock_request, allocation_size, sd, ea_list,
&fsp, &info, &sbuf);
if (!NT_STATUS_IS_OK(status)) {
goto fail;
}
done:
DEBUG(10, ("create_file: info=%d\n", info));
*result = fsp;
if (pinfo != NULL) {
*pinfo = info;
}
if (psbuf != NULL) {
*psbuf = sbuf;
}
return NT_STATUS_OK;
fail:
DEBUG(10, ("create_file: %s\n", nt_errstr(status)));
if (fsp != NULL) {
close_file(fsp, ERROR_CLOSE);
fsp = NULL;
}
return status;
}