summaryrefslogtreecommitdiff
path: root/source3/smbd/reply.c
diff options
context:
space:
mode:
Diffstat (limited to 'source3/smbd/reply.c')
-rw-r--r--source3/smbd/reply.c7184
1 files changed, 7184 insertions, 0 deletions
diff --git a/source3/smbd/reply.c b/source3/smbd/reply.c
new file mode 100644
index 0000000000..6933533672
--- /dev/null
+++ b/source3/smbd/reply.c
@@ -0,0 +1,7184 @@
+/*
+ Unix SMB/CIFS implementation.
+ Main SMB reply routines
+ Copyright (C) Andrew Tridgell 1992-1998
+ Copyright (C) Andrew Bartlett 2001
+ Copyright (C) Jeremy Allison 1992-2007.
+ Copyright (C) Volker Lendecke 2007
+
+ 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/>.
+*/
+/*
+ This file handles most of the reply_ calls that the server
+ makes to handle specific protocols
+*/
+
+#include "includes.h"
+
+/* look in server.c for some explanation of these variables */
+extern enum protocol_types Protocol;
+extern int max_recv;
+unsigned int smb_echo_count = 0;
+extern uint32 global_client_caps;
+
+extern bool global_encrypted_passwords_negotiated;
+
+/****************************************************************************
+ Ensure we check the path in *exactly* the same way as W2K for a findfirst/findnext
+ path or anything including wildcards.
+ We're assuming here that '/' is not the second byte in any multibyte char
+ set (a safe assumption). '\\' *may* be the second byte in a multibyte char
+ set.
+****************************************************************************/
+
+/* Custom version for processing POSIX paths. */
+#define IS_PATH_SEP(c,posix_only) ((c) == '/' || (!(posix_only) && (c) == '\\'))
+
+static NTSTATUS check_path_syntax_internal(char *path,
+ bool posix_path,
+ bool *p_last_component_contains_wcard)
+{
+ char *d = path;
+ const char *s = path;
+ NTSTATUS ret = NT_STATUS_OK;
+ bool start_of_name_component = True;
+
+ *p_last_component_contains_wcard = False;
+
+ while (*s) {
+ if (IS_PATH_SEP(*s,posix_path)) {
+ /*
+ * Safe to assume is not the second part of a mb char
+ * as this is handled below.
+ */
+ /* Eat multiple '/' or '\\' */
+ while (IS_PATH_SEP(*s,posix_path)) {
+ s++;
+ }
+ if ((d != path) && (*s != '\0')) {
+ /* We only care about non-leading or trailing '/' or '\\' */
+ *d++ = '/';
+ }
+
+ start_of_name_component = True;
+ /* New component. */
+ *p_last_component_contains_wcard = False;
+ continue;
+ }
+
+ if (start_of_name_component) {
+ if ((s[0] == '.') && (s[1] == '.') && (IS_PATH_SEP(s[2],posix_path) || s[2] == '\0')) {
+ /* Uh oh - "/../" or "\\..\\" or "/..\0" or "\\..\0" ! */
+
+ /*
+ * No mb char starts with '.' so we're safe checking the directory separator here.
+ */
+
+ /* If we just added a '/' - delete it */
+ if ((d > path) && (*(d-1) == '/')) {
+ *(d-1) = '\0';
+ d--;
+ }
+
+ /* Are we at the start ? Can't go back further if so. */
+ if (d <= path) {
+ ret = NT_STATUS_OBJECT_PATH_SYNTAX_BAD;
+ break;
+ }
+ /* Go back one level... */
+ /* We know this is safe as '/' cannot be part of a mb sequence. */
+ /* NOTE - if this assumption is invalid we are not in good shape... */
+ /* Decrement d first as d points to the *next* char to write into. */
+ for (d--; d > path; d--) {
+ if (*d == '/')
+ break;
+ }
+ s += 2; /* Else go past the .. */
+ /* We're still at the start of a name component, just the previous one. */
+ continue;
+
+ } else if ((s[0] == '.') && ((s[1] == '\0') || IS_PATH_SEP(s[1],posix_path))) {
+ if (posix_path) {
+ /* Eat the '.' */
+ s++;
+ continue;
+ }
+ }
+
+ }
+
+ if (!(*s & 0x80)) {
+ if (!posix_path) {
+ if (*s <= 0x1f) {
+ return NT_STATUS_OBJECT_NAME_INVALID;
+ }
+ switch (*s) {
+ case '*':
+ case '?':
+ case '<':
+ case '>':
+ case '"':
+ *p_last_component_contains_wcard = True;
+ break;
+ default:
+ break;
+ }
+ }
+ *d++ = *s++;
+ } else {
+ size_t siz;
+ /* Get the size of the next MB character. */
+ next_codepoint(s,&siz);
+ switch(siz) {
+ case 5:
+ *d++ = *s++;
+ /*fall through*/
+ case 4:
+ *d++ = *s++;
+ /*fall through*/
+ case 3:
+ *d++ = *s++;
+ /*fall through*/
+ case 2:
+ *d++ = *s++;
+ /*fall through*/
+ case 1:
+ *d++ = *s++;
+ break;
+ default:
+ DEBUG(0,("check_path_syntax_internal: character length assumptions invalid !\n"));
+ *d = '\0';
+ return NT_STATUS_INVALID_PARAMETER;
+ }
+ }
+ start_of_name_component = False;
+ }
+
+ *d = '\0';
+
+ return ret;
+}
+
+/****************************************************************************
+ Ensure we check the path in *exactly* the same way as W2K for regular pathnames.
+ No wildcards allowed.
+****************************************************************************/
+
+NTSTATUS check_path_syntax(char *path)
+{
+ bool ignore;
+ return check_path_syntax_internal(path, False, &ignore);
+}
+
+/****************************************************************************
+ Ensure we check the path in *exactly* the same way as W2K for regular pathnames.
+ Wildcards allowed - p_contains_wcard returns true if the last component contained
+ a wildcard.
+****************************************************************************/
+
+NTSTATUS check_path_syntax_wcard(char *path, bool *p_contains_wcard)
+{
+ return check_path_syntax_internal(path, False, p_contains_wcard);
+}
+
+/****************************************************************************
+ Check the path for a POSIX client.
+ We're assuming here that '/' is not the second byte in any multibyte char
+ set (a safe assumption).
+****************************************************************************/
+
+NTSTATUS check_path_syntax_posix(char *path)
+{
+ bool ignore;
+ return check_path_syntax_internal(path, True, &ignore);
+}
+
+/****************************************************************************
+ Pull a string and check the path allowing a wilcard - provide for error return.
+****************************************************************************/
+
+size_t srvstr_get_path_wcard(TALLOC_CTX *ctx,
+ const char *inbuf,
+ uint16 smb_flags2,
+ char **pp_dest,
+ const char *src,
+ size_t src_len,
+ int flags,
+ NTSTATUS *err,
+ bool *contains_wcard)
+{
+ size_t ret;
+
+ *pp_dest = NULL;
+
+ if (src_len == 0) {
+ ret = srvstr_pull_buf_talloc(ctx,
+ inbuf,
+ smb_flags2,
+ pp_dest,
+ src,
+ flags);
+ } else {
+ ret = srvstr_pull_talloc(ctx,
+ inbuf,
+ smb_flags2,
+ pp_dest,
+ src,
+ src_len,
+ flags);
+ }
+
+ if (!*pp_dest) {
+ *err = NT_STATUS_INVALID_PARAMETER;
+ return ret;
+ }
+
+ *contains_wcard = False;
+
+ if (smb_flags2 & FLAGS2_DFS_PATHNAMES) {
+ /*
+ * For a DFS path the function parse_dfs_path()
+ * will do the path processing, just make a copy.
+ */
+ *err = NT_STATUS_OK;
+ return ret;
+ }
+
+ if (lp_posix_pathnames()) {
+ *err = check_path_syntax_posix(*pp_dest);
+ } else {
+ *err = check_path_syntax_wcard(*pp_dest, contains_wcard);
+ }
+
+ return ret;
+}
+
+/****************************************************************************
+ Pull a string and check the path - provide for error return.
+****************************************************************************/
+
+size_t srvstr_get_path(TALLOC_CTX *ctx,
+ const char *inbuf,
+ uint16 smb_flags2,
+ char **pp_dest,
+ const char *src,
+ size_t src_len,
+ int flags,
+ NTSTATUS *err)
+{
+ size_t ret;
+
+ *pp_dest = NULL;
+
+ if (src_len == 0) {
+ ret = srvstr_pull_buf_talloc(ctx,
+ inbuf,
+ smb_flags2,
+ pp_dest,
+ src,
+ flags);
+ } else {
+ ret = srvstr_pull_talloc(ctx,
+ inbuf,
+ smb_flags2,
+ pp_dest,
+ src,
+ src_len,
+ flags);
+ }
+
+ if (!*pp_dest) {
+ *err = NT_STATUS_INVALID_PARAMETER;
+ return ret;
+ }
+
+ if (smb_flags2 & FLAGS2_DFS_PATHNAMES) {
+ /*
+ * For a DFS path the function parse_dfs_path()
+ * will do the path processing, just make a copy.
+ */
+ *err = NT_STATUS_OK;
+ return ret;
+ }
+
+ if (lp_posix_pathnames()) {
+ *err = check_path_syntax_posix(*pp_dest);
+ } else {
+ *err = check_path_syntax(*pp_dest);
+ }
+
+ return ret;
+}
+
+/****************************************************************************
+ Check if we have a correct fsp pointing to a file. Basic check for open fsp.
+****************************************************************************/
+
+bool check_fsp_open(connection_struct *conn, struct smb_request *req,
+ files_struct *fsp)
+{
+ if (!(fsp) || !(conn)) {
+ reply_nterror(req, NT_STATUS_INVALID_HANDLE);
+ return False;
+ }
+ if (((conn) != (fsp)->conn) || req->vuid != (fsp)->vuid) {
+ reply_nterror(req, NT_STATUS_INVALID_HANDLE);
+ return False;
+ }
+ return True;
+}
+
+/****************************************************************************
+ Check if we have a correct fsp pointing to a file. Replacement for the
+ CHECK_FSP macro.
+****************************************************************************/
+
+bool check_fsp(connection_struct *conn, struct smb_request *req,
+ files_struct *fsp)
+{
+ if (!check_fsp_open(conn, req, fsp)) {
+ return False;
+ }
+ if ((fsp)->is_directory) {
+ reply_nterror(req, NT_STATUS_INVALID_DEVICE_REQUEST);
+ return False;
+ }
+ if ((fsp)->fh->fd == -1) {
+ reply_nterror(req, NT_STATUS_ACCESS_DENIED);
+ return False;
+ }
+ (fsp)->num_smb_operations++;
+ return True;
+}
+
+/****************************************************************************
+ Check if we have a correct fsp pointing to a quota fake file. Replacement for
+ the CHECK_NTQUOTA_HANDLE_OK macro.
+****************************************************************************/
+
+bool check_fsp_ntquota_handle(connection_struct *conn, struct smb_request *req,
+ files_struct *fsp)
+{
+ if (!check_fsp_open(conn, req, fsp)) {
+ return false;
+ }
+
+ if (fsp->is_directory) {
+ return false;
+ }
+
+ if (fsp->fake_file_handle == NULL) {
+ return false;
+ }
+
+ if (fsp->fake_file_handle->type != FAKE_FILE_TYPE_QUOTA) {
+ return false;
+ }
+
+ if (fsp->fake_file_handle->private_data == NULL) {
+ return false;
+ }
+
+ return true;
+}
+
+/****************************************************************************
+ Check if we have a correct fsp. Replacement for the FSP_BELONGS_CONN macro
+****************************************************************************/
+
+bool fsp_belongs_conn(connection_struct *conn, struct smb_request *req,
+ files_struct *fsp)
+{
+ if ((fsp) && (conn) && ((conn)==(fsp)->conn)
+ && (req->vuid == (fsp)->vuid)) {
+ return True;
+ }
+
+ reply_nterror(req, NT_STATUS_INVALID_HANDLE);
+ return False;
+}
+
+/****************************************************************************
+ Reply to a (netbios-level) special message.
+****************************************************************************/
+
+void reply_special(char *inbuf)
+{
+ int msg_type = CVAL(inbuf,0);
+ int msg_flags = CVAL(inbuf,1);
+ fstring name1,name2;
+ char name_type = 0;
+
+ /*
+ * We only really use 4 bytes of the outbuf, but for the smb_setlen
+ * calculation & friends (srv_send_smb uses that) we need the full smb
+ * header.
+ */
+ char outbuf[smb_size];
+
+ static bool already_got_session = False;
+
+ *name1 = *name2 = 0;
+
+ memset(outbuf, '\0', sizeof(outbuf));
+
+ smb_setlen(outbuf,0);
+
+ switch (msg_type) {
+ case 0x81: /* session request */
+
+ if (already_got_session) {
+ exit_server_cleanly("multiple session request not permitted");
+ }
+
+ SCVAL(outbuf,0,0x82);
+ SCVAL(outbuf,3,0);
+ if (name_len(inbuf+4) > 50 ||
+ name_len(inbuf+4 + name_len(inbuf + 4)) > 50) {
+ DEBUG(0,("Invalid name length in session request\n"));
+ return;
+ }
+ name_extract(inbuf,4,name1);
+ name_type = name_extract(inbuf,4 + name_len(inbuf + 4),name2);
+ DEBUG(2,("netbios connect: name1=%s name2=%s\n",
+ name1,name2));
+
+ set_local_machine_name(name1, True);
+ set_remote_machine_name(name2, True);
+
+ DEBUG(2,("netbios connect: local=%s remote=%s, name type = %x\n",
+ get_local_machine_name(), get_remote_machine_name(),
+ name_type));
+
+ if (name_type == 'R') {
+ /* We are being asked for a pathworks session ---
+ no thanks! */
+ SCVAL(outbuf, 0,0x83);
+ break;
+ }
+
+ /* only add the client's machine name to the list
+ of possibly valid usernames if we are operating
+ in share mode security */
+ if (lp_security() == SEC_SHARE) {
+ add_session_user(get_remote_machine_name());
+ }
+
+ reload_services(True);
+ reopen_logs();
+
+ already_got_session = True;
+ break;
+
+ case 0x89: /* session keepalive request
+ (some old clients produce this?) */
+ SCVAL(outbuf,0,SMBkeepalive);
+ SCVAL(outbuf,3,0);
+ break;
+
+ case 0x82: /* positive session response */
+ case 0x83: /* negative session response */
+ case 0x84: /* retarget session response */
+ DEBUG(0,("Unexpected session response\n"));
+ break;
+
+ case SMBkeepalive: /* session keepalive */
+ default:
+ return;
+ }
+
+ DEBUG(5,("init msg_type=0x%x msg_flags=0x%x\n",
+ msg_type, msg_flags));
+
+ srv_send_smb(smbd_server_fd(), outbuf, false);
+ return;
+}
+
+/****************************************************************************
+ Reply to a tcon.
+ conn POINTER CAN BE NULL HERE !
+****************************************************************************/
+
+void reply_tcon(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ const char *service;
+ char *service_buf = NULL;
+ char *password = NULL;
+ char *dev = NULL;
+ int pwlen=0;
+ NTSTATUS nt_status;
+ char *p;
+ DATA_BLOB password_blob;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBtcon);
+
+ if (smb_buflen(req->inbuf) < 4) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBtcon);
+ return;
+ }
+
+ p = smb_buf(req->inbuf)+1;
+ p += srvstr_pull_buf_talloc(ctx, req->inbuf, req->flags2,
+ &service_buf, p, STR_TERMINATE) + 1;
+ pwlen = srvstr_pull_buf_talloc(ctx, req->inbuf, req->flags2,
+ &password, p, STR_TERMINATE) + 1;
+ p += pwlen;
+ p += srvstr_pull_buf_talloc(ctx, req->inbuf, req->flags2,
+ &dev, p, STR_TERMINATE) + 1;
+
+ if (service_buf == NULL || password == NULL || dev == NULL) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBtcon);
+ return;
+ }
+ p = strrchr_m(service_buf,'\\');
+ if (p) {
+ service = p+1;
+ } else {
+ service = service_buf;
+ }
+
+ password_blob = data_blob(password, pwlen+1);
+
+ conn = make_connection(service,password_blob,dev,req->vuid,&nt_status);
+ req->conn = conn;
+
+ data_blob_clear_free(&password_blob);
+
+ if (!conn) {
+ reply_nterror(req, nt_status);
+ END_PROFILE(SMBtcon);
+ return;
+ }
+
+ reply_outbuf(req, 2, 0);
+ SSVAL(req->outbuf,smb_vwv0,max_recv);
+ SSVAL(req->outbuf,smb_vwv1,conn->cnum);
+ SSVAL(req->outbuf,smb_tid,conn->cnum);
+
+ DEBUG(3,("tcon service=%s cnum=%d\n",
+ service, conn->cnum));
+
+ END_PROFILE(SMBtcon);
+ return;
+}
+
+/****************************************************************************
+ Reply to a tcon and X.
+ conn POINTER CAN BE NULL HERE !
+****************************************************************************/
+
+void reply_tcon_and_X(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *service = NULL;
+ DATA_BLOB password;
+ TALLOC_CTX *ctx = talloc_tos();
+ /* what the cleint thinks the device is */
+ char *client_devicetype = NULL;
+ /* what the server tells the client the share represents */
+ const char *server_devicetype;
+ NTSTATUS nt_status;
+ int passlen;
+ char *path = NULL;
+ char *p, *q;
+ uint16 tcon_flags;
+
+ START_PROFILE(SMBtconX);
+
+ if (req->wct < 4) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+
+ passlen = SVAL(req->inbuf,smb_vwv3);
+ tcon_flags = SVAL(req->inbuf,smb_vwv2);
+
+ /* we might have to close an old one */
+ if ((tcon_flags & 0x1) && conn) {
+ close_cnum(conn,req->vuid);
+ req->conn = NULL;
+ conn = NULL;
+ }
+
+ if ((passlen > MAX_PASS_LEN) || (passlen >= smb_buflen(req->inbuf))) {
+ reply_doserror(req, ERRDOS, ERRbuftoosmall);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+
+ if (global_encrypted_passwords_negotiated) {
+ password = data_blob_talloc(talloc_tos(), smb_buf(req->inbuf),
+ passlen);
+ if (lp_security() == SEC_SHARE) {
+ /*
+ * Security = share always has a pad byte
+ * after the password.
+ */
+ p = smb_buf(req->inbuf) + passlen + 1;
+ } else {
+ p = smb_buf(req->inbuf) + passlen;
+ }
+ } else {
+ password = data_blob_talloc(talloc_tos(), smb_buf(req->inbuf),
+ passlen+1);
+ /* Ensure correct termination */
+ password.data[passlen]=0;
+ p = smb_buf(req->inbuf) + passlen + 1;
+ }
+
+ p += srvstr_pull_buf_talloc(ctx, req->inbuf, req->flags2, &path, p,
+ STR_TERMINATE);
+
+ if (path == NULL) {
+ data_blob_clear_free(&password);
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+
+ /*
+ * the service name can be either: \\server\share
+ * or share directly like on the DELL PowerVault 705
+ */
+ if (*path=='\\') {
+ q = strchr_m(path+2,'\\');
+ if (!q) {
+ data_blob_clear_free(&password);
+ reply_doserror(req, ERRDOS, ERRnosuchshare);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+ service = q+1;
+ } else {
+ service = path;
+ }
+
+ p += srvstr_pull_talloc(ctx, req->inbuf, req->flags2,
+ &client_devicetype, p,
+ MIN(6,smb_bufrem(req->inbuf, p)), STR_ASCII);
+
+ if (client_devicetype == NULL) {
+ data_blob_clear_free(&password);
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+
+ DEBUG(4,("Client requested device type [%s] for share [%s]\n", client_devicetype, service));
+
+ conn = make_connection(service, password, client_devicetype,
+ req->vuid, &nt_status);
+ req->conn =conn;
+
+ data_blob_clear_free(&password);
+
+ if (!conn) {
+ reply_nterror(req, nt_status);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+
+ if ( IS_IPC(conn) )
+ server_devicetype = "IPC";
+ else if ( IS_PRINT(conn) )
+ server_devicetype = "LPT1:";
+ else
+ server_devicetype = "A:";
+
+ if (Protocol < PROTOCOL_NT1) {
+ reply_outbuf(req, 2, 0);
+ if (message_push_string(&req->outbuf, server_devicetype,
+ STR_TERMINATE|STR_ASCII) == -1) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+ } else {
+ /* NT sets the fstype of IPC$ to the null string */
+ const char *fstype = IS_IPC(conn) ? "" : lp_fstype(SNUM(conn));
+
+ if (tcon_flags & TCONX_FLAG_EXTENDED_RESPONSE) {
+ /* Return permissions. */
+ uint32 perm1 = 0;
+ uint32 perm2 = 0;
+
+ reply_outbuf(req, 7, 0);
+
+ if (IS_IPC(conn)) {
+ perm1 = FILE_ALL_ACCESS;
+ perm2 = FILE_ALL_ACCESS;
+ } else {
+ perm1 = CAN_WRITE(conn) ?
+ SHARE_ALL_ACCESS :
+ SHARE_READ_ONLY;
+ }
+
+ SIVAL(req->outbuf, smb_vwv3, perm1);
+ SIVAL(req->outbuf, smb_vwv5, perm2);
+ } else {
+ reply_outbuf(req, 3, 0);
+ }
+
+ if ((message_push_string(&req->outbuf, server_devicetype,
+ STR_TERMINATE|STR_ASCII) == -1)
+ || (message_push_string(&req->outbuf, fstype,
+ STR_TERMINATE) == -1)) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBtconX);
+ return;
+ }
+
+ /* what does setting this bit do? It is set by NT4 and
+ may affect the ability to autorun mounted cdroms */
+ SSVAL(req->outbuf, smb_vwv2, SMB_SUPPORT_SEARCH_BITS|
+ (lp_csc_policy(SNUM(conn)) << 2));
+
+ init_dfsroot(conn, req->inbuf, req->outbuf);
+ }
+
+
+ DEBUG(3,("tconX service=%s \n",
+ service));
+
+ /* set the incoming and outgoing tid to the just created one */
+ SSVAL(req->inbuf,smb_tid,conn->cnum);
+ SSVAL(req->outbuf,smb_tid,conn->cnum);
+
+ END_PROFILE(SMBtconX);
+
+ chain_reply(req);
+ return;
+}
+
+/****************************************************************************
+ Reply to an unknown type.
+****************************************************************************/
+
+void reply_unknown_new(struct smb_request *req, uint8 type)
+{
+ DEBUG(0, ("unknown command type (%s): type=%d (0x%X)\n",
+ smb_fn_name(type), type, type));
+ reply_doserror(req, ERRSRV, ERRunknownsmb);
+ return;
+}
+
+/****************************************************************************
+ Reply to an ioctl.
+ conn POINTER CAN BE NULL HERE !
+****************************************************************************/
+
+void reply_ioctl(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ uint16 device;
+ uint16 function;
+ uint32 ioctl_code;
+ int replysize;
+ char *p;
+
+ START_PROFILE(SMBioctl);
+
+ if (req->wct < 3) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBioctl);
+ return;
+ }
+
+ device = SVAL(req->inbuf,smb_vwv1);
+ function = SVAL(req->inbuf,smb_vwv2);
+ ioctl_code = (device << 16) + function;
+
+ DEBUG(4, ("Received IOCTL (code 0x%x)\n", ioctl_code));
+
+ switch (ioctl_code) {
+ case IOCTL_QUERY_JOB_INFO:
+ replysize = 32;
+ break;
+ default:
+ reply_doserror(req, ERRSRV, ERRnosupport);
+ END_PROFILE(SMBioctl);
+ return;
+ }
+
+ reply_outbuf(req, 8, replysize+1);
+ SSVAL(req->outbuf,smb_vwv1,replysize); /* Total data bytes returned */
+ SSVAL(req->outbuf,smb_vwv5,replysize); /* Data bytes this buffer */
+ SSVAL(req->outbuf,smb_vwv6,52); /* Offset to data */
+ p = smb_buf(req->outbuf);
+ memset(p, '\0', replysize+1); /* valgrind-safe. */
+ p += 1; /* Allow for alignment */
+
+ switch (ioctl_code) {
+ case IOCTL_QUERY_JOB_INFO:
+ {
+ files_struct *fsp = file_fsp(SVAL(req->inbuf,
+ smb_vwv0));
+ if (!fsp) {
+ reply_doserror(req, ERRDOS, ERRbadfid);
+ END_PROFILE(SMBioctl);
+ return;
+ }
+ SSVAL(p,0,fsp->rap_print_jobid); /* Job number */
+ srvstr_push((char *)req->outbuf, req->flags2, p+2,
+ global_myname(), 15,
+ STR_TERMINATE|STR_ASCII);
+ if (conn) {
+ srvstr_push((char *)req->outbuf, req->flags2,
+ p+18, lp_servicename(SNUM(conn)),
+ 13, STR_TERMINATE|STR_ASCII);
+ } else {
+ memset(p+18, 0, 13);
+ }
+ break;
+ }
+ }
+
+ END_PROFILE(SMBioctl);
+ return;
+}
+
+/****************************************************************************
+ Strange checkpath NTSTATUS mapping.
+****************************************************************************/
+
+static NTSTATUS map_checkpath_error(const char *inbuf, NTSTATUS status)
+{
+ /* Strange DOS error code semantics only for checkpath... */
+ if (!(SVAL(inbuf,smb_flg2) & FLAGS2_32_BIT_ERROR_CODES)) {
+ if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_INVALID,status)) {
+ /* We need to map to ERRbadpath */
+ return NT_STATUS_OBJECT_PATH_NOT_FOUND;
+ }
+ }
+ return status;
+}
+
+/****************************************************************************
+ Reply to a checkpath.
+****************************************************************************/
+
+void reply_checkpath(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *name = NULL;
+ SMB_STRUCT_STAT sbuf;
+ NTSTATUS status;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBcheckpath);
+
+ srvstr_get_path(ctx,(char *)req->inbuf, req->flags2, &name,
+ smb_buf(req->inbuf) + 1, 0,
+ STR_TERMINATE, &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ status = map_checkpath_error((char *)req->inbuf, status);
+ reply_nterror(req, status);
+ END_PROFILE(SMBcheckpath);
+ return;
+ }
+
+ status = resolve_dfspath(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ name,
+ &name);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBcheckpath);
+ return;
+ }
+ goto path_err;
+ }
+
+ DEBUG(3,("reply_checkpath %s mode=%d\n", name, (int)SVAL(req->inbuf,smb_vwv0)));
+
+ status = unix_convert(ctx, conn, name, False, &name, NULL, &sbuf);
+ if (!NT_STATUS_IS_OK(status)) {
+ goto path_err;
+ }
+
+ status = check_name(conn, name);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3,("reply_checkpath: check_name of %s failed (%s)\n",name,nt_errstr(status)));
+ goto path_err;
+ }
+
+ if (!VALID_STAT(sbuf) && (SMB_VFS_STAT(conn,name,&sbuf) != 0)) {
+ DEBUG(3,("reply_checkpath: stat of %s failed (%s)\n",name,strerror(errno)));
+ status = map_nt_error_from_unix(errno);
+ goto path_err;
+ }
+
+ if (!S_ISDIR(sbuf.st_mode)) {
+ reply_botherror(req, NT_STATUS_NOT_A_DIRECTORY,
+ ERRDOS, ERRbadpath);
+ END_PROFILE(SMBcheckpath);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ END_PROFILE(SMBcheckpath);
+ return;
+
+ path_err:
+
+ END_PROFILE(SMBcheckpath);
+
+ /* We special case this - as when a Windows machine
+ is parsing a path is steps through the components
+ one at a time - if a component fails it expects
+ ERRbadpath, not ERRbadfile.
+ */
+ status = map_checkpath_error((char *)req->inbuf, status);
+ if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
+ /*
+ * Windows returns different error codes if
+ * the parent directory is valid but not the
+ * last component - it returns NT_STATUS_OBJECT_NAME_NOT_FOUND
+ * for that case and NT_STATUS_OBJECT_PATH_NOT_FOUND
+ * if the path is invalid.
+ */
+ reply_botherror(req, NT_STATUS_OBJECT_NAME_NOT_FOUND,
+ ERRDOS, ERRbadpath);
+ return;
+ }
+
+ reply_nterror(req, status);
+}
+
+/****************************************************************************
+ Reply to a getatr.
+****************************************************************************/
+
+void reply_getatr(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *fname = NULL;
+ SMB_STRUCT_STAT sbuf;
+ int mode=0;
+ SMB_OFF_T size=0;
+ time_t mtime=0;
+ char *p;
+ NTSTATUS status;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBgetatr);
+
+ p = smb_buf(req->inbuf) + 1;
+ p += srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &fname, p,
+ 0, STR_TERMINATE, &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBgetatr);
+ return;
+ }
+
+ status = resolve_dfspath(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ fname,
+ &fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBgetatr);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBgetatr);
+ return;
+ }
+
+ /* dos smetimes asks for a stat of "" - it returns a "hidden directory"
+ under WfWg - weird! */
+ if (*fname == '\0') {
+ mode = aHIDDEN | aDIR;
+ if (!CAN_WRITE(conn)) {
+ mode |= aRONLY;
+ }
+ size = 0;
+ mtime = 0;
+ } else {
+ status = unix_convert(ctx, conn, fname, False, &fname, NULL,&sbuf);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBgetatr);
+ return;
+ }
+ status = check_name(conn, fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3,("reply_getatr: check_name of %s failed (%s)\n",fname,nt_errstr(status)));
+ reply_nterror(req, status);
+ END_PROFILE(SMBgetatr);
+ return;
+ }
+ if (!VALID_STAT(sbuf) && (SMB_VFS_STAT(conn,fname,&sbuf) != 0)) {
+ DEBUG(3,("reply_getatr: stat of %s failed (%s)\n",fname,strerror(errno)));
+ reply_unixerror(req, ERRDOS,ERRbadfile);
+ END_PROFILE(SMBgetatr);
+ return;
+ }
+
+ mode = dos_mode(conn,fname,&sbuf);
+ size = sbuf.st_size;
+ mtime = sbuf.st_mtime;
+ if (mode & aDIR) {
+ size = 0;
+ }
+ }
+
+ reply_outbuf(req, 10, 0);
+
+ SSVAL(req->outbuf,smb_vwv0,mode);
+ if(lp_dos_filetime_resolution(SNUM(conn)) ) {
+ srv_put_dos_date3((char *)req->outbuf,smb_vwv1,mtime & ~1);
+ } else {
+ srv_put_dos_date3((char *)req->outbuf,smb_vwv1,mtime);
+ }
+ SIVAL(req->outbuf,smb_vwv3,(uint32)size);
+
+ if (Protocol >= PROTOCOL_NT1) {
+ SSVAL(req->outbuf, smb_flg2,
+ SVAL(req->outbuf, smb_flg2) | FLAGS2_IS_LONG_NAME);
+ }
+
+ DEBUG(3,("reply_getatr: name=%s mode=%d size=%u\n", fname, mode, (unsigned int)size ) );
+
+ END_PROFILE(SMBgetatr);
+ return;
+}
+
+/****************************************************************************
+ Reply to a setatr.
+****************************************************************************/
+
+void reply_setatr(struct smb_request *req)
+{
+ struct timespec ts[2];
+ connection_struct *conn = req->conn;
+ char *fname = NULL;
+ int mode;
+ time_t mtime;
+ SMB_STRUCT_STAT sbuf;
+ char *p;
+ NTSTATUS status;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBsetatr);
+
+ ZERO_STRUCT(ts);
+
+ if (req->wct < 2) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ return;
+ }
+
+ p = smb_buf(req->inbuf) + 1;
+ p += srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &fname, p,
+ 0, STR_TERMINATE, &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+
+ status = resolve_dfspath(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ fname,
+ &fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+
+ status = unix_convert(ctx, conn, fname, False, &fname, NULL, &sbuf);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+
+ status = check_name(conn, fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+
+ if (fname[0] == '.' && fname[1] == '\0') {
+ /*
+ * Not sure here is the right place to catch this
+ * condition. Might be moved to somewhere else later -- vl
+ */
+ reply_nterror(req, NT_STATUS_ACCESS_DENIED);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+
+ mode = SVAL(req->inbuf,smb_vwv0);
+ mtime = srv_make_unix_date3(req->inbuf+smb_vwv1);
+
+ ts[1] = convert_time_t_to_timespec(mtime);
+ status = smb_set_file_time(conn, NULL, fname,
+ &sbuf, ts, true);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+
+ if (mode != FILE_ATTRIBUTE_NORMAL) {
+ if (VALID_STAT_OF_DIR(sbuf))
+ mode |= aDIR;
+ else
+ mode &= ~aDIR;
+
+ if (file_set_dosmode(conn,fname,mode,&sbuf,NULL,false) != 0) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBsetatr);
+ return;
+ }
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ DEBUG( 3, ( "setatr name=%s mode=%d\n", fname, mode ) );
+
+ END_PROFILE(SMBsetatr);
+ return;
+}
+
+/****************************************************************************
+ Reply to a dskattr.
+****************************************************************************/
+
+void reply_dskattr(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ SMB_BIG_UINT dfree,dsize,bsize;
+ START_PROFILE(SMBdskattr);
+
+ if (get_dfree_info(conn,".",True,&bsize,&dfree,&dsize) == (SMB_BIG_UINT)-1) {
+ reply_unixerror(req, ERRHRD, ERRgeneral);
+ END_PROFILE(SMBdskattr);
+ return;
+ }
+
+ reply_outbuf(req, 5, 0);
+
+ if (Protocol <= PROTOCOL_LANMAN2) {
+ double total_space, free_space;
+ /* we need to scale this to a number that DOS6 can handle. We
+ use floating point so we can handle large drives on systems
+ that don't have 64 bit integers
+
+ we end up displaying a maximum of 2G to DOS systems
+ */
+ total_space = dsize * (double)bsize;
+ free_space = dfree * (double)bsize;
+
+ dsize = (SMB_BIG_UINT)((total_space+63*512) / (64*512));
+ dfree = (SMB_BIG_UINT)((free_space+63*512) / (64*512));
+
+ if (dsize > 0xFFFF) dsize = 0xFFFF;
+ if (dfree > 0xFFFF) dfree = 0xFFFF;
+
+ SSVAL(req->outbuf,smb_vwv0,dsize);
+ SSVAL(req->outbuf,smb_vwv1,64); /* this must be 64 for dos systems */
+ SSVAL(req->outbuf,smb_vwv2,512); /* and this must be 512 */
+ SSVAL(req->outbuf,smb_vwv3,dfree);
+ } else {
+ SSVAL(req->outbuf,smb_vwv0,dsize);
+ SSVAL(req->outbuf,smb_vwv1,bsize/512);
+ SSVAL(req->outbuf,smb_vwv2,512);
+ SSVAL(req->outbuf,smb_vwv3,dfree);
+ }
+
+ DEBUG(3,("dskattr dfree=%d\n", (unsigned int)dfree));
+
+ END_PROFILE(SMBdskattr);
+ return;
+}
+
+/****************************************************************************
+ Reply to a search.
+ Can be called from SMBsearch, SMBffirst or SMBfunique.
+****************************************************************************/
+
+void reply_search(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *mask = NULL;
+ char *directory = NULL;
+ char *fname = NULL;
+ SMB_OFF_T size;
+ uint32 mode;
+ time_t date;
+ uint32 dirtype;
+ unsigned int numentries = 0;
+ unsigned int maxentries = 0;
+ bool finished = False;
+ char *p;
+ int status_len;
+ char *path = NULL;
+ char status[21];
+ int dptr_num= -1;
+ bool check_descend = False;
+ bool expect_close = False;
+ NTSTATUS nt_status;
+ bool mask_contains_wcard = False;
+ bool allow_long_path_components = (req->flags2 & FLAGS2_LONG_PATH_COMPONENTS) ? True : False;
+ TALLOC_CTX *ctx = talloc_tos();
+ bool ask_sharemode = lp_parm_bool(SNUM(conn), "smbd", "search ask sharemode", true);
+
+ START_PROFILE(SMBsearch);
+
+ if (req->wct < 2) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+
+ if (lp_posix_pathnames()) {
+ reply_unknown_new(req, CVAL(req->inbuf, smb_com));
+ END_PROFILE(SMBsearch);
+ return;
+ }
+
+ /* If we were called as SMBffirst then we must expect close. */
+ if(CVAL(req->inbuf,smb_com) == SMBffirst) {
+ expect_close = True;
+ }
+
+ reply_outbuf(req, 1, 3);
+ maxentries = SVAL(req->inbuf,smb_vwv0);
+ dirtype = SVAL(req->inbuf,smb_vwv1);
+ p = smb_buf(req->inbuf) + 1;
+ p += srvstr_get_path_wcard(ctx,
+ (char *)req->inbuf,
+ req->flags2,
+ &path,
+ p,
+ 0,
+ STR_TERMINATE,
+ &nt_status,
+ &mask_contains_wcard);
+ if (!NT_STATUS_IS_OK(nt_status)) {
+ reply_nterror(req, nt_status);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+
+ nt_status = resolve_dfspath_wcard(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ path,
+ &path,
+ &mask_contains_wcard);
+ if (!NT_STATUS_IS_OK(nt_status)) {
+ if (NT_STATUS_EQUAL(nt_status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ reply_nterror(req, nt_status);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+
+ p++;
+ status_len = SVAL(p, 0);
+ p += 2;
+
+ /* dirtype &= ~aDIR; */
+
+ if (status_len == 0) {
+ SMB_STRUCT_STAT sbuf;
+
+ nt_status = unix_convert(ctx, conn, path, True,
+ &directory, NULL, &sbuf);
+ if (!NT_STATUS_IS_OK(nt_status)) {
+ reply_nterror(req, nt_status);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+
+ nt_status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(nt_status)) {
+ reply_nterror(req, nt_status);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+
+ p = strrchr_m(directory,'/');
+ if (!p) {
+ mask = directory;
+ directory = talloc_strdup(ctx,".");
+ if (!directory) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ } else {
+ *p = 0;
+ mask = p+1;
+ }
+
+ if (*directory == '\0') {
+ directory = talloc_strdup(ctx,".");
+ if (!directory) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ }
+ memset((char *)status,'\0',21);
+ SCVAL(status,0,(dirtype & 0x1F));
+
+ nt_status = dptr_create(conn,
+ directory,
+ True,
+ expect_close,
+ req->smbpid,
+ mask,
+ mask_contains_wcard,
+ dirtype,
+ &conn->dirptr);
+ if (!NT_STATUS_IS_OK(nt_status)) {
+ reply_nterror(req, nt_status);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ dptr_num = dptr_dnum(conn->dirptr);
+ } else {
+ int status_dirtype;
+
+ memcpy(status,p,21);
+ status_dirtype = CVAL(status,0) & 0x1F;
+ if (status_dirtype != (dirtype & 0x1F)) {
+ dirtype = status_dirtype;
+ }
+
+ conn->dirptr = dptr_fetch(status+12,&dptr_num);
+ if (!conn->dirptr) {
+ goto SearchEmpty;
+ }
+ string_set(&conn->dirpath,dptr_path(dptr_num));
+ mask = dptr_wcard(dptr_num);
+ if (!mask) {
+ goto SearchEmpty;
+ }
+ /*
+ * For a 'continue' search we have no string. So
+ * check from the initial saved string.
+ */
+ mask_contains_wcard = ms_has_wild(mask);
+ dirtype = dptr_attr(dptr_num);
+ }
+
+ DEBUG(4,("dptr_num is %d\n",dptr_num));
+
+ if ((dirtype&0x1F) == aVOLID) {
+ char buf[DIR_STRUCT_SIZE];
+ memcpy(buf,status,21);
+ if (!make_dir_struct(ctx,buf,"???????????",volume_label(SNUM(conn)),
+ 0,aVOLID,0,!allow_long_path_components)) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ dptr_fill(buf+12,dptr_num);
+ if (dptr_zero(buf+12) && (status_len==0)) {
+ numentries = 1;
+ } else {
+ numentries = 0;
+ }
+ if (message_push_blob(&req->outbuf,
+ data_blob_const(buf, sizeof(buf)))
+ == -1) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ } else {
+ unsigned int i;
+ maxentries = MIN(
+ maxentries,
+ ((BUFFER_SIZE -
+ ((uint8 *)smb_buf(req->outbuf) + 3 - req->outbuf))
+ /DIR_STRUCT_SIZE));
+
+ DEBUG(8,("dirpath=<%s> dontdescend=<%s>\n",
+ conn->dirpath,lp_dontdescend(SNUM(conn))));
+ if (in_list(conn->dirpath, lp_dontdescend(SNUM(conn)),True)) {
+ check_descend = True;
+ }
+
+ for (i=numentries;(i<maxentries) && !finished;i++) {
+ finished = !get_dir_entry(ctx,
+ conn,
+ mask,
+ dirtype,
+ &fname,
+ &size,
+ &mode,
+ &date,
+ check_descend,
+ ask_sharemode);
+ if (!finished) {
+ char buf[DIR_STRUCT_SIZE];
+ memcpy(buf,status,21);
+ if (!make_dir_struct(ctx,
+ buf,
+ mask,
+ fname,
+ size,
+ mode,
+ date,
+ !allow_long_path_components)) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ if (!dptr_fill(buf+12,dptr_num)) {
+ break;
+ }
+ if (message_push_blob(&req->outbuf,
+ data_blob_const(buf, sizeof(buf)))
+ == -1) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+ numentries++;
+ }
+ }
+ }
+
+ SearchEmpty:
+
+ /* If we were called as SMBffirst with smb_search_id == NULL
+ and no entries were found then return error and close dirptr
+ (X/Open spec) */
+
+ if (numentries == 0) {
+ dptr_close(&dptr_num);
+ } else if(expect_close && status_len == 0) {
+ /* Close the dptr - we know it's gone */
+ dptr_close(&dptr_num);
+ }
+
+ /* If we were called as SMBfunique, then we can close the dirptr now ! */
+ if(dptr_num >= 0 && CVAL(req->inbuf,smb_com) == SMBfunique) {
+ dptr_close(&dptr_num);
+ }
+
+ if ((numentries == 0) && !mask_contains_wcard) {
+ reply_botherror(req, STATUS_NO_MORE_FILES, ERRDOS, ERRnofiles);
+ END_PROFILE(SMBsearch);
+ return;
+ }
+
+ SSVAL(req->outbuf,smb_vwv0,numentries);
+ SSVAL(req->outbuf,smb_vwv1,3 + numentries * DIR_STRUCT_SIZE);
+ SCVAL(smb_buf(req->outbuf),0,5);
+ SSVAL(smb_buf(req->outbuf),1,numentries*DIR_STRUCT_SIZE);
+
+ /* The replies here are never long name. */
+ SSVAL(req->outbuf, smb_flg2,
+ SVAL(req->outbuf, smb_flg2) & (~FLAGS2_IS_LONG_NAME));
+ if (!allow_long_path_components) {
+ SSVAL(req->outbuf, smb_flg2,
+ SVAL(req->outbuf, smb_flg2)
+ & (~FLAGS2_LONG_PATH_COMPONENTS));
+ }
+
+ /* This SMB *always* returns ASCII names. Remove the unicode bit in flags2. */
+ SSVAL(req->outbuf, smb_flg2,
+ (SVAL(req->outbuf, smb_flg2) & (~FLAGS2_UNICODE_STRINGS)));
+
+ if (!directory) {
+ directory = dptr_path(dptr_num);
+ }
+
+ DEBUG(4,("%s mask=%s path=%s dtype=%d nument=%u of %u\n",
+ smb_fn_name(CVAL(req->inbuf,smb_com)),
+ mask,
+ directory ? directory : "./",
+ dirtype,
+ numentries,
+ maxentries ));
+
+ END_PROFILE(SMBsearch);
+ return;
+}
+
+/****************************************************************************
+ Reply to a fclose (stop directory search).
+****************************************************************************/
+
+void reply_fclose(struct smb_request *req)
+{
+ int status_len;
+ char status[21];
+ int dptr_num= -2;
+ char *p;
+ char *path = NULL;
+ NTSTATUS err;
+ bool path_contains_wcard = False;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBfclose);
+
+ if (lp_posix_pathnames()) {
+ reply_unknown_new(req, CVAL(req->inbuf, smb_com));
+ END_PROFILE(SMBfclose);
+ return;
+ }
+
+ p = smb_buf(req->inbuf) + 1;
+ p += srvstr_get_path_wcard(ctx,
+ (char *)req->inbuf,
+ req->flags2,
+ &path,
+ p,
+ 0,
+ STR_TERMINATE,
+ &err,
+ &path_contains_wcard);
+ if (!NT_STATUS_IS_OK(err)) {
+ reply_nterror(req, err);
+ END_PROFILE(SMBfclose);
+ return;
+ }
+ p++;
+ status_len = SVAL(p,0);
+ p += 2;
+
+ if (status_len == 0) {
+ reply_doserror(req, ERRSRV, ERRsrverror);
+ END_PROFILE(SMBfclose);
+ return;
+ }
+
+ memcpy(status,p,21);
+
+ if(dptr_fetch(status+12,&dptr_num)) {
+ /* Close the dptr - we know it's gone */
+ dptr_close(&dptr_num);
+ }
+
+ reply_outbuf(req, 1, 0);
+ SSVAL(req->outbuf,smb_vwv0,0);
+
+ DEBUG(3,("search close\n"));
+
+ END_PROFILE(SMBfclose);
+ return;
+}
+
+/****************************************************************************
+ Reply to an open.
+****************************************************************************/
+
+void reply_open(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *fname = NULL;
+ uint32 fattr=0;
+ SMB_OFF_T size = 0;
+ time_t mtime=0;
+ int info;
+ SMB_STRUCT_STAT sbuf;
+ files_struct *fsp;
+ int oplock_request;
+ int deny_mode;
+ uint32 dos_attr;
+ uint32 access_mask;
+ uint32 share_mode;
+ uint32 create_disposition;
+ uint32 create_options = 0;
+ NTSTATUS status;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBopen);
+
+ if (req->wct < 2) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBopen);
+ return;
+ }
+
+ oplock_request = CORE_OPLOCK_REQUEST(req->inbuf);
+ deny_mode = SVAL(req->inbuf,smb_vwv0);
+ dos_attr = SVAL(req->inbuf,smb_vwv1);
+
+ srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &fname,
+ smb_buf(req->inbuf)+1, 0,
+ STR_TERMINATE, &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBopen);
+ return;
+ }
+
+ if (!map_open_params_to_ntcreate(
+ fname, deny_mode, OPENX_FILE_EXISTS_OPEN, &access_mask,
+ &share_mode, &create_disposition, &create_options)) {
+ reply_nterror(req, NT_STATUS_DOS(ERRDOS, ERRbadaccess));
+ END_PROFILE(SMBopen);
+ return;
+ }
+
+ status = create_file(conn, /* conn */
+ req, /* req */
+ 0, /* root_dir_fid */
+ fname, /* fname */
+ access_mask, /* access_mask */
+ share_mode, /* share_access */
+ create_disposition, /* create_disposition*/
+ create_options, /* create_options */
+ dos_attr, /* file_attributes */
+ oplock_request, /* oplock_request */
+ 0, /* allocation_size */
+ NULL, /* sd */
+ NULL, /* ea_list */
+ &fsp, /* result */
+ &info, /* pinfo */
+ &sbuf); /* psbuf */
+
+ if (!NT_STATUS_IS_OK(status)) {
+ if (open_was_deferred(req->mid)) {
+ /* We have re-scheduled this call. */
+ END_PROFILE(SMBopen);
+ return;
+ }
+ reply_openerror(req, status);
+ END_PROFILE(SMBopen);
+ return;
+ }
+
+ size = sbuf.st_size;
+ fattr = dos_mode(conn,fsp->fsp_name,&sbuf);
+ mtime = sbuf.st_mtime;
+
+ if (fattr & aDIR) {
+ DEBUG(3,("attempt to open a directory %s\n",fsp->fsp_name));
+ close_file(fsp,ERROR_CLOSE);
+ reply_doserror(req, ERRDOS,ERRnoaccess);
+ END_PROFILE(SMBopen);
+ return;
+ }
+
+ reply_outbuf(req, 7, 0);
+ SSVAL(req->outbuf,smb_vwv0,fsp->fnum);
+ SSVAL(req->outbuf,smb_vwv1,fattr);
+ if(lp_dos_filetime_resolution(SNUM(conn)) ) {
+ srv_put_dos_date3((char *)req->outbuf,smb_vwv2,mtime & ~1);
+ } else {
+ srv_put_dos_date3((char *)req->outbuf,smb_vwv2,mtime);
+ }
+ SIVAL(req->outbuf,smb_vwv4,(uint32)size);
+ SSVAL(req->outbuf,smb_vwv6,deny_mode);
+
+ if (oplock_request && lp_fake_oplocks(SNUM(conn))) {
+ SCVAL(req->outbuf,smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+
+ if(EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+ SCVAL(req->outbuf,smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+ END_PROFILE(SMBopen);
+ return;
+}
+
+/****************************************************************************
+ Reply to an open and X.
+****************************************************************************/
+
+void reply_open_and_X(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *fname = NULL;
+ uint16 open_flags;
+ int deny_mode;
+ uint32 smb_attr;
+ /* Breakout the oplock request bits so we can set the
+ reply bits separately. */
+ int ex_oplock_request;
+ int core_oplock_request;
+ int oplock_request;
+#if 0
+ int smb_sattr = SVAL(req->inbuf,smb_vwv4);
+ uint32 smb_time = make_unix_date3(req->inbuf+smb_vwv6);
+#endif
+ int smb_ofun;
+ uint32 fattr=0;
+ int mtime=0;
+ SMB_STRUCT_STAT sbuf;
+ int smb_action = 0;
+ files_struct *fsp;
+ NTSTATUS status;
+ SMB_BIG_UINT allocation_size;
+ ssize_t retval = -1;
+ uint32 access_mask;
+ uint32 share_mode;
+ uint32 create_disposition;
+ uint32 create_options = 0;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBopenX);
+
+ if (req->wct < 15) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBopenX);
+ return;
+ }
+
+ open_flags = SVAL(req->inbuf,smb_vwv2);
+ deny_mode = SVAL(req->inbuf,smb_vwv3);
+ smb_attr = SVAL(req->inbuf,smb_vwv5);
+ ex_oplock_request = EXTENDED_OPLOCK_REQUEST(req->inbuf);
+ core_oplock_request = CORE_OPLOCK_REQUEST(req->inbuf);
+ oplock_request = ex_oplock_request | core_oplock_request;
+ smb_ofun = SVAL(req->inbuf,smb_vwv8);
+ allocation_size = (SMB_BIG_UINT)IVAL(req->inbuf,smb_vwv9);
+
+ /* If it's an IPC, pass off the pipe handler. */
+ if (IS_IPC(conn)) {
+ if (lp_nt_pipe_support()) {
+ reply_open_pipe_and_X(conn, req);
+ } else {
+ reply_doserror(req, ERRSRV, ERRaccess);
+ }
+ END_PROFILE(SMBopenX);
+ return;
+ }
+
+ /* XXXX we need to handle passed times, sattr and flags */
+ srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &fname,
+ smb_buf(req->inbuf), 0, STR_TERMINATE,
+ &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBopenX);
+ return;
+ }
+
+ if (!map_open_params_to_ntcreate(
+ fname, deny_mode, smb_ofun, &access_mask,
+ &share_mode, &create_disposition, &create_options)) {
+ reply_nterror(req, NT_STATUS_DOS(ERRDOS, ERRbadaccess));
+ END_PROFILE(SMBopenX);
+ return;
+ }
+
+ status = create_file(conn, /* conn */
+ req, /* req */
+ 0, /* root_dir_fid */
+ fname, /* fname */
+ access_mask, /* access_mask */
+ share_mode, /* share_access */
+ create_disposition, /* create_disposition*/
+ create_options, /* create_options */
+ smb_attr, /* file_attributes */
+ oplock_request, /* oplock_request */
+ 0, /* allocation_size */
+ NULL, /* sd */
+ NULL, /* ea_list */
+ &fsp, /* result */
+ &smb_action, /* pinfo */
+ &sbuf); /* psbuf */
+
+ if (!NT_STATUS_IS_OK(status)) {
+ END_PROFILE(SMBopenX);
+ if (open_was_deferred(req->mid)) {
+ /* We have re-scheduled this call. */
+ return;
+ }
+ reply_openerror(req, status);
+ return;
+ }
+
+ /* Setting the "size" field in vwv9 and vwv10 causes the file to be set to this size,
+ if the file is truncated or created. */
+ if (((smb_action == FILE_WAS_CREATED) || (smb_action == FILE_WAS_OVERWRITTEN)) && allocation_size) {
+ fsp->initial_allocation_size = smb_roundup(fsp->conn, allocation_size);
+ if (vfs_allocate_file_space(fsp, fsp->initial_allocation_size) == -1) {
+ close_file(fsp,ERROR_CLOSE);
+ reply_nterror(req, NT_STATUS_DISK_FULL);
+ END_PROFILE(SMBopenX);
+ return;
+ }
+ retval = vfs_set_filelen(fsp, (SMB_OFF_T)allocation_size);
+ if (retval < 0) {
+ close_file(fsp,ERROR_CLOSE);
+ reply_nterror(req, NT_STATUS_DISK_FULL);
+ END_PROFILE(SMBopenX);
+ return;
+ }
+ sbuf.st_size = get_allocation_size(conn,fsp,&sbuf);
+ }
+
+ fattr = dos_mode(conn,fsp->fsp_name,&sbuf);
+ mtime = sbuf.st_mtime;
+ if (fattr & aDIR) {
+ close_file(fsp,ERROR_CLOSE);
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBopenX);
+ return;
+ }
+
+ /* If the caller set the extended oplock request bit
+ and we granted one (by whatever means) - set the
+ correct bit for extended oplock reply.
+ */
+
+ if (ex_oplock_request && lp_fake_oplocks(SNUM(conn))) {
+ smb_action |= EXTENDED_OPLOCK_GRANTED;
+ }
+
+ if(ex_oplock_request && EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+ smb_action |= EXTENDED_OPLOCK_GRANTED;
+ }
+
+ /* If the caller set the core oplock request bit
+ and we granted one (by whatever means) - set the
+ correct bit for core oplock reply.
+ */
+
+ if (open_flags & EXTENDED_RESPONSE_REQUIRED) {
+ reply_outbuf(req, 19, 0);
+ } else {
+ reply_outbuf(req, 15, 0);
+ }
+
+ if (core_oplock_request && lp_fake_oplocks(SNUM(conn))) {
+ SCVAL(req->outbuf, smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+
+ if(core_oplock_request && EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+ SCVAL(req->outbuf, smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+
+ SSVAL(req->outbuf,smb_vwv2,fsp->fnum);
+ SSVAL(req->outbuf,smb_vwv3,fattr);
+ if(lp_dos_filetime_resolution(SNUM(conn)) ) {
+ srv_put_dos_date3((char *)req->outbuf,smb_vwv4,mtime & ~1);
+ } else {
+ srv_put_dos_date3((char *)req->outbuf,smb_vwv4,mtime);
+ }
+ SIVAL(req->outbuf,smb_vwv6,(uint32)sbuf.st_size);
+ SSVAL(req->outbuf,smb_vwv8,GET_OPENX_MODE(deny_mode));
+ SSVAL(req->outbuf,smb_vwv11,smb_action);
+
+ if (open_flags & EXTENDED_RESPONSE_REQUIRED) {
+ SIVAL(req->outbuf, smb_vwv15, STD_RIGHT_ALL_ACCESS);
+ }
+
+ END_PROFILE(SMBopenX);
+ chain_reply(req);
+ return;
+}
+
+/****************************************************************************
+ Reply to a SMBulogoffX.
+****************************************************************************/
+
+void reply_ulogoffX(struct smb_request *req)
+{
+ user_struct *vuser;
+
+ START_PROFILE(SMBulogoffX);
+
+ vuser = get_valid_user_struct(req->vuid);
+
+ if(vuser == NULL) {
+ DEBUG(3,("ulogoff, vuser id %d does not map to user.\n",
+ req->vuid));
+ }
+
+ /* in user level security we are supposed to close any files
+ open by this user */
+ if ((vuser != NULL) && (lp_security() != SEC_SHARE)) {
+ file_close_user(req->vuid);
+ }
+
+ invalidate_vuid(req->vuid);
+
+ reply_outbuf(req, 2, 0);
+
+ DEBUG( 3, ( "ulogoffX vuid=%d\n", req->vuid ) );
+
+ END_PROFILE(SMBulogoffX);
+ chain_reply(req);
+}
+
+/****************************************************************************
+ Reply to a mknew or a create.
+****************************************************************************/
+
+void reply_mknew(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *fname = NULL;
+ int com;
+ uint32 fattr = 0;
+ struct timespec ts[2];
+ files_struct *fsp;
+ int oplock_request = 0;
+ SMB_STRUCT_STAT sbuf;
+ NTSTATUS status;
+ uint32 access_mask = FILE_GENERIC_READ | FILE_GENERIC_WRITE;
+ uint32 share_mode = FILE_SHARE_READ|FILE_SHARE_WRITE;
+ uint32 create_disposition;
+ uint32 create_options = 0;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBcreate);
+
+ if (req->wct < 3) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBcreate);
+ return;
+ }
+
+ fattr = SVAL(req->inbuf,smb_vwv0);
+ oplock_request = CORE_OPLOCK_REQUEST(req->inbuf);
+ com = SVAL(req->inbuf,smb_com);
+
+ ts[1] =convert_time_t_to_timespec(
+ srv_make_unix_date3(req->inbuf + smb_vwv1));
+ /* mtime. */
+
+ srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &fname,
+ smb_buf(req->inbuf) + 1, 0,
+ STR_TERMINATE, &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcreate);
+ return;
+ }
+
+ if (fattr & aVOLID) {
+ DEBUG(0,("Attempt to create file (%s) with volid set - "
+ "please report this\n", fname));
+ }
+
+ if(com == SMBmknew) {
+ /* We should fail if file exists. */
+ create_disposition = FILE_CREATE;
+ } else {
+ /* Create if file doesn't exist, truncate if it does. */
+ create_disposition = FILE_OVERWRITE_IF;
+ }
+
+ status = create_file(conn, /* conn */
+ req, /* req */
+ 0, /* root_dir_fid */
+ fname, /* fname */
+ access_mask, /* access_mask */
+ share_mode, /* share_access */
+ create_disposition, /* create_disposition*/
+ create_options, /* create_options */
+ fattr, /* file_attributes */
+ oplock_request, /* oplock_request */
+ 0, /* allocation_size */
+ NULL, /* sd */
+ NULL, /* ea_list */
+ &fsp, /* result */
+ NULL, /* pinfo */
+ &sbuf); /* psbuf */
+
+ if (!NT_STATUS_IS_OK(status)) {
+ END_PROFILE(SMBcreate);
+ if (open_was_deferred(req->mid)) {
+ /* We have re-scheduled this call. */
+ return;
+ }
+ reply_openerror(req, status);
+ return;
+ }
+
+ ts[0] = get_atimespec(&sbuf); /* atime. */
+ status = smb_set_file_time(conn, fsp, fsp->fsp_name, &sbuf, ts, true);
+ if (!NT_STATUS_IS_OK(status)) {
+ END_PROFILE(SMBcreate);
+ reply_openerror(req, status);
+ return;
+ }
+
+ reply_outbuf(req, 1, 0);
+ SSVAL(req->outbuf,smb_vwv0,fsp->fnum);
+
+ if (oplock_request && lp_fake_oplocks(SNUM(conn))) {
+ SCVAL(req->outbuf,smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+
+ if(EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+ SCVAL(req->outbuf,smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+
+ DEBUG( 2, ( "reply_mknew: file %s\n", fsp->fsp_name ) );
+ DEBUG( 3, ( "reply_mknew %s fd=%d dmode=0x%x\n",
+ fsp->fsp_name, fsp->fh->fd, (unsigned int)fattr ) );
+
+ END_PROFILE(SMBcreate);
+ return;
+}
+
+/****************************************************************************
+ Reply to a create temporary file.
+****************************************************************************/
+
+void reply_ctemp(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *fname = NULL;
+ uint32 fattr;
+ files_struct *fsp;
+ int oplock_request;
+ int tmpfd;
+ SMB_STRUCT_STAT sbuf;
+ char *s;
+ NTSTATUS status;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBctemp);
+
+ if (req->wct < 3) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ fattr = SVAL(req->inbuf,smb_vwv0);
+ oplock_request = CORE_OPLOCK_REQUEST(req->inbuf);
+
+ srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &fname,
+ smb_buf(req->inbuf)+1, 0, STR_TERMINATE,
+ &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+ if (*fname) {
+ fname = talloc_asprintf(ctx,
+ "%s/TMXXXXXX",
+ fname);
+ } else {
+ fname = talloc_strdup(ctx, "TMXXXXXX");
+ }
+
+ if (!fname) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ status = resolve_dfspath(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ fname,
+ &fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ status = unix_convert(ctx, conn, fname, False, &fname, NULL, &sbuf);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ status = check_name(conn, fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ tmpfd = smb_mkstemp(fname);
+ if (tmpfd == -1) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ SMB_VFS_STAT(conn,fname,&sbuf);
+
+ /* We should fail if file does not exist. */
+ status = open_file_ntcreate(conn, req, fname, &sbuf,
+ FILE_GENERIC_READ | FILE_GENERIC_WRITE,
+ FILE_SHARE_READ|FILE_SHARE_WRITE,
+ FILE_OPEN,
+ 0,
+ fattr,
+ oplock_request,
+ NULL, &fsp);
+
+ /* close fd from smb_mkstemp() */
+ close(tmpfd);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ if (open_was_deferred(req->mid)) {
+ /* We have re-scheduled this call. */
+ END_PROFILE(SMBctemp);
+ return;
+ }
+ reply_openerror(req, status);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ reply_outbuf(req, 1, 0);
+ SSVAL(req->outbuf,smb_vwv0,fsp->fnum);
+
+ /* the returned filename is relative to the directory */
+ s = strrchr_m(fsp->fsp_name, '/');
+ if (!s) {
+ s = fsp->fsp_name;
+ } else {
+ s++;
+ }
+
+#if 0
+ /* Tested vs W2K3 - this doesn't seem to be here - null terminated filename is the only
+ thing in the byte section. JRA */
+ SSVALS(p, 0, -1); /* what is this? not in spec */
+#endif
+ if (message_push_string(&req->outbuf, s, STR_ASCII|STR_TERMINATE)
+ == -1) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBctemp);
+ return;
+ }
+
+ if (oplock_request && lp_fake_oplocks(SNUM(conn))) {
+ SCVAL(req->outbuf, smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+
+ if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+ SCVAL(req->outbuf, smb_flg,
+ CVAL(req->outbuf,smb_flg)|CORE_OPLOCK_GRANTED);
+ }
+
+ DEBUG( 2, ( "reply_ctemp: created temp file %s\n", fsp->fsp_name ) );
+ DEBUG( 3, ( "reply_ctemp %s fd=%d umode=0%o\n", fsp->fsp_name,
+ fsp->fh->fd, (unsigned int)sbuf.st_mode ) );
+
+ END_PROFILE(SMBctemp);
+ return;
+}
+
+/*******************************************************************
+ Check if a user is allowed to rename a file.
+********************************************************************/
+
+static NTSTATUS can_rename(connection_struct *conn, files_struct *fsp,
+ uint16 dirtype, SMB_STRUCT_STAT *pst)
+{
+ uint32 fmode;
+
+ if (!CAN_WRITE(conn)) {
+ return NT_STATUS_MEDIA_WRITE_PROTECTED;
+ }
+
+ fmode = dos_mode(conn, fsp->fsp_name, pst);
+ if ((fmode & ~dirtype) & (aHIDDEN | aSYSTEM)) {
+ return NT_STATUS_NO_SUCH_FILE;
+ }
+
+ if (S_ISDIR(pst->st_mode)) {
+ return NT_STATUS_OK;
+ }
+
+ if (fsp->access_mask & (DELETE_ACCESS|FILE_WRITE_ATTRIBUTES)) {
+ return NT_STATUS_OK;
+ }
+
+ return NT_STATUS_ACCESS_DENIED;
+}
+
+/*******************************************************************
+ * unlink a file with all relevant access checks
+ *******************************************************************/
+
+static NTSTATUS do_unlink(connection_struct *conn,
+ struct smb_request *req,
+ const char *fname,
+ uint32 dirtype)
+{
+ SMB_STRUCT_STAT sbuf;
+ uint32 fattr;
+ files_struct *fsp;
+ uint32 dirtype_orig = dirtype;
+ NTSTATUS status;
+
+ DEBUG(10,("do_unlink: %s, dirtype = %d\n", fname, dirtype ));
+
+ if (!CAN_WRITE(conn)) {
+ return NT_STATUS_MEDIA_WRITE_PROTECTED;
+ }
+
+ if (SMB_VFS_LSTAT(conn,fname,&sbuf) != 0) {
+ return map_nt_error_from_unix(errno);
+ }
+
+ fattr = dos_mode(conn,fname,&sbuf);
+
+ if (dirtype & FILE_ATTRIBUTE_NORMAL) {
+ dirtype = aDIR|aARCH|aRONLY;
+ }
+
+ dirtype &= (aDIR|aARCH|aRONLY|aHIDDEN|aSYSTEM);
+ if (!dirtype) {
+ return NT_STATUS_NO_SUCH_FILE;
+ }
+
+ if (!dir_check_ftype(conn, fattr, dirtype)) {
+ if (fattr & aDIR) {
+ return NT_STATUS_FILE_IS_A_DIRECTORY;
+ }
+ return NT_STATUS_NO_SUCH_FILE;
+ }
+
+ if (dirtype_orig & 0x8000) {
+ /* These will never be set for POSIX. */
+ return NT_STATUS_NO_SUCH_FILE;
+ }
+
+#if 0
+ if ((fattr & dirtype) & FILE_ATTRIBUTE_DIRECTORY) {
+ return NT_STATUS_FILE_IS_A_DIRECTORY;
+ }
+
+ if ((fattr & ~dirtype) & (FILE_ATTRIBUTE_HIDDEN|FILE_ATTRIBUTE_SYSTEM)) {
+ return NT_STATUS_NO_SUCH_FILE;
+ }
+
+ if (dirtype & 0xFF00) {
+ /* These will never be set for POSIX. */
+ return NT_STATUS_NO_SUCH_FILE;
+ }
+
+ dirtype &= 0xFF;
+ if (!dirtype) {
+ return NT_STATUS_NO_SUCH_FILE;
+ }
+
+ /* Can't delete a directory. */
+ if (fattr & aDIR) {
+ return NT_STATUS_FILE_IS_A_DIRECTORY;
+ }
+#endif
+
+#if 0 /* JRATEST */
+ else if (dirtype & aDIR) /* Asked for a directory and it isn't. */
+ return NT_STATUS_OBJECT_NAME_INVALID;
+#endif /* JRATEST */
+
+ /* Fix for bug #3035 from SATOH Fumiyasu <fumiyas@miraclelinux.com>
+
+ On a Windows share, a file with read-only dosmode can be opened with
+ DELETE_ACCESS. But on a Samba share (delete readonly = no), it
+ fails with NT_STATUS_CANNOT_DELETE error.
+
+ This semantic causes a problem that a user can not
+ rename a file with read-only dosmode on a Samba share
+ from a Windows command prompt (i.e. cmd.exe, but can rename
+ from Windows Explorer).
+ */
+
+ if (!lp_delete_readonly(SNUM(conn))) {
+ if (fattr & aRONLY) {
+ return NT_STATUS_CANNOT_DELETE;
+ }
+ }
+
+ /* On open checks the open itself will check the share mode, so
+ don't do it here as we'll get it wrong. */
+
+ status = create_file_unixpath
+ (conn, /* conn */
+ req, /* req */
+ fname, /* fname */
+ DELETE_ACCESS, /* access_mask */
+ FILE_SHARE_NONE, /* share_access */
+ FILE_OPEN, /* create_disposition*/
+ FILE_NON_DIRECTORY_FILE, /* create_options */
+ FILE_ATTRIBUTE_NORMAL, /* file_attributes */
+ 0, /* oplock_request */
+ 0, /* allocation_size */
+ NULL, /* sd */
+ NULL, /* ea_list */
+ &fsp, /* result */
+ NULL, /* pinfo */
+ &sbuf); /* psbuf */
+
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(10, ("create_file_unixpath failed: %s\n",
+ nt_errstr(status)));
+ return status;
+ }
+
+ /* The set is across all open files on this dev/inode pair. */
+ if (!set_delete_on_close(fsp, True, &conn->server_info->utok)) {
+ close_file(fsp, NORMAL_CLOSE);
+ return NT_STATUS_ACCESS_DENIED;
+ }
+
+ return close_file(fsp,NORMAL_CLOSE);
+}
+
+/****************************************************************************
+ The guts of the unlink command, split out so it may be called by the NT SMB
+ code.
+****************************************************************************/
+
+NTSTATUS unlink_internals(connection_struct *conn, struct smb_request *req,
+ uint32 dirtype, const char *name_in, bool has_wild)
+{
+ const char *directory = NULL;
+ char *mask = NULL;
+ char *name = NULL;
+ char *p = NULL;
+ int count=0;
+ NTSTATUS status = NT_STATUS_OK;
+ SMB_STRUCT_STAT sbuf;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ status = unix_convert(ctx, conn, name_in, has_wild, &name, NULL, &sbuf);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ p = strrchr_m(name,'/');
+ if (!p) {
+ directory = talloc_strdup(ctx, ".");
+ if (!directory) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ mask = name;
+ } else {
+ *p = 0;
+ directory = name;
+ mask = p+1;
+ }
+
+ /*
+ * We should only check the mangled cache
+ * here if unix_convert failed. This means
+ * that the path in 'mask' doesn't exist
+ * on the file system and so we need to look
+ * for a possible mangle. This patch from
+ * Tine Smukavec <valentin.smukavec@hermes.si>.
+ */
+
+ if (!VALID_STAT(sbuf) && mangle_is_mangled(mask,conn->params)) {
+ char *new_mask = NULL;
+ mangle_lookup_name_from_8_3(ctx,
+ mask,
+ &new_mask,
+ conn->params );
+ if (new_mask) {
+ mask = new_mask;
+ }
+ }
+
+ if (!has_wild) {
+ directory = talloc_asprintf(ctx,
+ "%s/%s",
+ directory,
+ mask);
+ if (!directory) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ if (dirtype == 0) {
+ dirtype = FILE_ATTRIBUTE_NORMAL;
+ }
+
+ status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ status = do_unlink(conn, req, directory, dirtype);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ count++;
+ } else {
+ struct smb_Dir *dir_hnd = NULL;
+ long offset = 0;
+ const char *dname;
+
+ if ((dirtype & SAMBA_ATTRIBUTES_MASK) == aDIR) {
+ return NT_STATUS_OBJECT_NAME_INVALID;
+ }
+
+ if (strequal(mask,"????????.???")) {
+ mask[0] = '*';
+ mask[1] = '\0';
+ }
+
+ status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ dir_hnd = OpenDir(talloc_tos(), conn, directory, mask,
+ dirtype);
+ if (dir_hnd == NULL) {
+ return map_nt_error_from_unix(errno);
+ }
+
+ /* XXXX the CIFS spec says that if bit0 of the flags2 field is set then
+ the pattern matches against the long name, otherwise the short name
+ We don't implement this yet XXXX
+ */
+
+ status = NT_STATUS_NO_SUCH_FILE;
+
+ while ((dname = ReadDirName(dir_hnd, &offset))) {
+ SMB_STRUCT_STAT st;
+ char *fname = NULL;
+
+ if (!is_visible_file(conn, directory, dname, &st, True)) {
+ continue;
+ }
+
+ /* Quick check for "." and ".." */
+ if (ISDOT(dname) || ISDOTDOT(dname)) {
+ continue;
+ }
+
+ if(!mask_match(dname, mask, conn->case_sensitive)) {
+ continue;
+ }
+
+ fname = talloc_asprintf(ctx, "%s/%s",
+ directory,
+ dname);
+ if (!fname) {
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ status = check_name(conn, fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ TALLOC_FREE(dir_hnd);
+ return status;
+ }
+
+ status = do_unlink(conn, req, fname, dirtype);
+ if (!NT_STATUS_IS_OK(status)) {
+ TALLOC_FREE(fname);
+ continue;
+ }
+
+ count++;
+ DEBUG(3,("unlink_internals: successful unlink [%s]\n",
+ fname));
+
+ TALLOC_FREE(fname);
+ }
+ TALLOC_FREE(dir_hnd);
+ }
+
+ if (count == 0 && NT_STATUS_IS_OK(status) && errno != 0) {
+ status = map_nt_error_from_unix(errno);
+ }
+
+ return status;
+}
+
+/****************************************************************************
+ Reply to a unlink
+****************************************************************************/
+
+void reply_unlink(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *name = NULL;
+ uint32 dirtype;
+ NTSTATUS status;
+ bool path_contains_wcard = False;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBunlink);
+
+ if (req->wct < 1) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBunlink);
+ return;
+ }
+
+ dirtype = SVAL(req->inbuf,smb_vwv0);
+
+ srvstr_get_path_wcard(ctx, (char *)req->inbuf, req->flags2, &name,
+ smb_buf(req->inbuf) + 1, 0,
+ STR_TERMINATE, &status, &path_contains_wcard);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBunlink);
+ return;
+ }
+
+ status = resolve_dfspath_wcard(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ name,
+ &name,
+ &path_contains_wcard);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBunlink);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBunlink);
+ return;
+ }
+
+ DEBUG(3,("reply_unlink : %s\n",name));
+
+ status = unlink_internals(conn, req, dirtype, name,
+ path_contains_wcard);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (open_was_deferred(req->mid)) {
+ /* We have re-scheduled this call. */
+ END_PROFILE(SMBunlink);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBunlink);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+ END_PROFILE(SMBunlink);
+
+ return;
+}
+
+/****************************************************************************
+ Fail for readbraw.
+****************************************************************************/
+
+static void fail_readraw(void)
+{
+ const char *errstr = talloc_asprintf(talloc_tos(),
+ "FAIL ! reply_readbraw: socket write fail (%s)",
+ strerror(errno));
+ if (!errstr) {
+ errstr = "";
+ }
+ exit_server_cleanly(errstr);
+}
+
+/****************************************************************************
+ Fake (read/write) sendfile. Returns -1 on read or write fail.
+****************************************************************************/
+
+static ssize_t fake_sendfile(files_struct *fsp, SMB_OFF_T startpos,
+ size_t nread)
+{
+ size_t bufsize;
+ size_t tosend = nread;
+ char *buf;
+
+ if (nread == 0) {
+ return 0;
+ }
+
+ bufsize = MIN(nread, 65536);
+
+ if (!(buf = SMB_MALLOC_ARRAY(char, bufsize))) {
+ return -1;
+ }
+
+ while (tosend > 0) {
+ ssize_t ret;
+ size_t cur_read;
+
+ if (tosend > bufsize) {
+ cur_read = bufsize;
+ } else {
+ cur_read = tosend;
+ }
+ ret = read_file(fsp,buf,startpos,cur_read);
+ if (ret == -1) {
+ SAFE_FREE(buf);
+ return -1;
+ }
+
+ /* If we had a short read, fill with zeros. */
+ if (ret < cur_read) {
+ memset(buf, '\0', cur_read - ret);
+ }
+
+ if (write_data(smbd_server_fd(),buf,cur_read) != cur_read) {
+ SAFE_FREE(buf);
+ return -1;
+ }
+ tosend -= cur_read;
+ startpos += cur_read;
+ }
+
+ SAFE_FREE(buf);
+ return (ssize_t)nread;
+}
+
+/****************************************************************************
+ Return a readbraw error (4 bytes of zero).
+****************************************************************************/
+
+static void reply_readbraw_error(void)
+{
+ char header[4];
+ SIVAL(header,0,0);
+ if (write_data(smbd_server_fd(),header,4) != 4) {
+ fail_readraw();
+ }
+}
+
+/****************************************************************************
+ Use sendfile in readbraw.
+****************************************************************************/
+
+void send_file_readbraw(connection_struct *conn,
+ files_struct *fsp,
+ SMB_OFF_T startpos,
+ size_t nread,
+ ssize_t mincount)
+{
+ char *outbuf = NULL;
+ ssize_t ret=0;
+
+#if defined(WITH_SENDFILE)
+ /*
+ * We can only use sendfile on a non-chained packet
+ * but we can use on a non-oplocked file. tridge proved this
+ * on a train in Germany :-). JRA.
+ * reply_readbraw has already checked the length.
+ */
+
+ if ( (chain_size == 0) && (nread > 0) && (fsp->base_fsp == NULL) &&
+ (fsp->wcp == NULL) && lp_use_sendfile(SNUM(conn)) ) {
+ char header[4];
+ DATA_BLOB header_blob;
+
+ _smb_setlen(header,nread);
+ header_blob = data_blob_const(header, 4);
+
+ if (SMB_VFS_SENDFILE(smbd_server_fd(), fsp,
+ &header_blob, startpos, nread) == -1) {
+ /* Returning ENOSYS means no data at all was sent.
+ * Do this as a normal read. */
+ if (errno == ENOSYS) {
+ goto normal_readbraw;
+ }
+
+ /*
+ * Special hack for broken Linux with no working sendfile. If we
+ * return EINTR we sent the header but not the rest of the data.
+ * Fake this up by doing read/write calls.
+ */
+ if (errno == EINTR) {
+ /* Ensure we don't do this again. */
+ set_use_sendfile(SNUM(conn), False);
+ DEBUG(0,("send_file_readbraw: sendfile not available. Faking..\n"));
+
+ if (fake_sendfile(fsp, startpos, nread) == -1) {
+ DEBUG(0,("send_file_readbraw: fake_sendfile failed for file %s (%s).\n",
+ fsp->fsp_name, strerror(errno) ));
+ exit_server_cleanly("send_file_readbraw fake_sendfile failed");
+ }
+ return;
+ }
+
+ DEBUG(0,("send_file_readbraw: sendfile failed for file %s (%s). Terminating\n",
+ fsp->fsp_name, strerror(errno) ));
+ exit_server_cleanly("send_file_readbraw sendfile failed");
+ }
+
+ return;
+ }
+#endif
+
+normal_readbraw:
+
+ outbuf = TALLOC_ARRAY(NULL, char, nread+4);
+ if (!outbuf) {
+ DEBUG(0,("send_file_readbraw: TALLOC_ARRAY failed for size %u.\n",
+ (unsigned)(nread+4)));
+ reply_readbraw_error();
+ return;
+ }
+
+ if (nread > 0) {
+ ret = read_file(fsp,outbuf+4,startpos,nread);
+#if 0 /* mincount appears to be ignored in a W2K server. JRA. */
+ if (ret < mincount)
+ ret = 0;
+#else
+ if (ret < nread)
+ ret = 0;
+#endif
+ }
+
+ _smb_setlen(outbuf,ret);
+ if (write_data(smbd_server_fd(),outbuf,4+ret) != 4+ret)
+ fail_readraw();
+
+ TALLOC_FREE(outbuf);
+}
+
+/****************************************************************************
+ Reply to a readbraw (core+ protocol).
+****************************************************************************/
+
+void reply_readbraw(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ ssize_t maxcount,mincount;
+ size_t nread = 0;
+ SMB_OFF_T startpos;
+ files_struct *fsp;
+ SMB_STRUCT_STAT st;
+ SMB_OFF_T size = 0;
+
+ START_PROFILE(SMBreadbraw);
+
+ if (srv_is_signing_active() || is_encrypted_packet(req->inbuf)) {
+ exit_server_cleanly("reply_readbraw: SMB signing/sealing is active - "
+ "raw reads/writes are disallowed.");
+ }
+
+ if (req->wct < 8) {
+ reply_readbraw_error();
+ END_PROFILE(SMBreadbraw);
+ return;
+ }
+
+ /*
+ * Special check if an oplock break has been issued
+ * and the readraw request croses on the wire, we must
+ * return a zero length response here.
+ */
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ /*
+ * We have to do a check_fsp by hand here, as
+ * we must always return 4 zero bytes on error,
+ * not a NTSTATUS.
+ */
+
+ if (!fsp || !conn || conn != fsp->conn ||
+ req->vuid != fsp->vuid ||
+ fsp->is_directory || fsp->fh->fd == -1) {
+ /*
+ * fsp could be NULL here so use the value from the packet. JRA.
+ */
+ DEBUG(3,("reply_readbraw: fnum %d not valid "
+ "- cache prime?\n",
+ (int)SVAL(req->inbuf,smb_vwv0)));
+ reply_readbraw_error();
+ END_PROFILE(SMBreadbraw);
+ return;
+ }
+
+ /* Do a "by hand" version of CHECK_READ. */
+ if (!(fsp->can_read ||
+ ((req->flags2 & FLAGS2_READ_PERMIT_EXECUTE) &&
+ (fsp->access_mask & FILE_EXECUTE)))) {
+ DEBUG(3,("reply_readbraw: fnum %d not readable.\n",
+ (int)SVAL(req->inbuf,smb_vwv0)));
+ reply_readbraw_error();
+ END_PROFILE(SMBreadbraw);
+ return;
+ }
+
+ flush_write_cache(fsp, READRAW_FLUSH);
+
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv1);
+ if(req->wct == 10) {
+ /*
+ * This is a large offset (64 bit) read.
+ */
+#ifdef LARGE_SMB_OFF_T
+
+ startpos |= (((SMB_OFF_T)IVAL(req->inbuf,smb_vwv8)) << 32);
+
+#else /* !LARGE_SMB_OFF_T */
+
+ /*
+ * Ensure we haven't been sent a >32 bit offset.
+ */
+
+ if(IVAL(req->inbuf,smb_vwv8) != 0) {
+ DEBUG(0,("reply_readbraw: large offset "
+ "(%x << 32) used and we don't support "
+ "64 bit offsets.\n",
+ (unsigned int)IVAL(req->inbuf,smb_vwv8) ));
+ reply_readbraw_error();
+ END_PROFILE(SMBreadbraw);
+ return;
+ }
+
+#endif /* LARGE_SMB_OFF_T */
+
+ if(startpos < 0) {
+ DEBUG(0,("reply_readbraw: negative 64 bit "
+ "readraw offset (%.0f) !\n",
+ (double)startpos ));
+ reply_readbraw_error();
+ END_PROFILE(SMBreadbraw);
+ return;
+ }
+ }
+
+ maxcount = (SVAL(req->inbuf,smb_vwv3) & 0xFFFF);
+ mincount = (SVAL(req->inbuf,smb_vwv4) & 0xFFFF);
+
+ /* ensure we don't overrun the packet size */
+ maxcount = MIN(65535,maxcount);
+
+ if (is_locked(fsp,(uint32)req->smbpid,
+ (SMB_BIG_UINT)maxcount,
+ (SMB_BIG_UINT)startpos,
+ READ_LOCK)) {
+ reply_readbraw_error();
+ END_PROFILE(SMBreadbraw);
+ return;
+ }
+
+ if (SMB_VFS_FSTAT(fsp, &st) == 0) {
+ size = st.st_size;
+ }
+
+ if (startpos >= size) {
+ nread = 0;
+ } else {
+ nread = MIN(maxcount,(size - startpos));
+ }
+
+#if 0 /* mincount appears to be ignored in a W2K server. JRA. */
+ if (nread < mincount)
+ nread = 0;
+#endif
+
+ DEBUG( 3, ( "reply_readbraw: fnum=%d start=%.0f max=%lu "
+ "min=%lu nread=%lu\n",
+ fsp->fnum, (double)startpos,
+ (unsigned long)maxcount,
+ (unsigned long)mincount,
+ (unsigned long)nread ) );
+
+ send_file_readbraw(conn, fsp, startpos, nread, mincount);
+
+ DEBUG(5,("reply_readbraw finished\n"));
+ END_PROFILE(SMBreadbraw);
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_LOCKING
+
+/****************************************************************************
+ Reply to a lockread (core+ protocol).
+****************************************************************************/
+
+void reply_lockread(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ ssize_t nread = -1;
+ char *data;
+ SMB_OFF_T startpos;
+ size_t numtoread;
+ NTSTATUS status;
+ files_struct *fsp;
+ struct byte_range_lock *br_lck = NULL;
+ char *p = NULL;
+
+ START_PROFILE(SMBlockread);
+
+ if (req->wct < 5) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBlockread);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBlockread);
+ return;
+ }
+
+ if (!CHECK_READ(fsp,req->inbuf)) {
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ END_PROFILE(SMBlockread);
+ return;
+ }
+
+ release_level_2_oplocks_on_change(fsp);
+
+ numtoread = SVAL(req->inbuf,smb_vwv1);
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv2);
+
+ numtoread = MIN(BUFFER_SIZE - (smb_size + 3*2 + 3), numtoread);
+
+ reply_outbuf(req, 5, numtoread + 3);
+
+ data = smb_buf(req->outbuf) + 3;
+
+ /*
+ * NB. Discovered by Menny Hamburger at Mainsoft. This is a core+
+ * protocol request that predates the read/write lock concept.
+ * Thus instead of asking for a read lock here we need to ask
+ * for a write lock. JRA.
+ * Note that the requested lock size is unaffected by max_recv.
+ */
+
+ br_lck = do_lock(smbd_messaging_context(),
+ fsp,
+ req->smbpid,
+ (SMB_BIG_UINT)numtoread,
+ (SMB_BIG_UINT)startpos,
+ WRITE_LOCK,
+ WINDOWS_LOCK,
+ False, /* Non-blocking lock. */
+ &status,
+ NULL);
+ TALLOC_FREE(br_lck);
+
+ if (NT_STATUS_V(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBlockread);
+ return;
+ }
+
+ /*
+ * However the requested READ size IS affected by max_recv. Insanity.... JRA.
+ */
+
+ if (numtoread > max_recv) {
+ DEBUG(0,("reply_lockread: requested read size (%u) is greater than maximum allowed (%u). \
+Returning short read of maximum allowed for compatibility with Windows 2000.\n",
+ (unsigned int)numtoread, (unsigned int)max_recv ));
+ numtoread = MIN(numtoread,max_recv);
+ }
+ nread = read_file(fsp,data,startpos,numtoread);
+
+ if (nread < 0) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBlockread);
+ return;
+ }
+
+ srv_set_message((char *)req->outbuf, 5, nread+3, False);
+
+ SSVAL(req->outbuf,smb_vwv0,nread);
+ SSVAL(req->outbuf,smb_vwv5,nread+3);
+ p = smb_buf(req->outbuf);
+ SCVAL(p,0,0); /* pad byte. */
+ SSVAL(p,1,nread);
+
+ DEBUG(3,("lockread fnum=%d num=%d nread=%d\n",
+ fsp->fnum, (int)numtoread, (int)nread));
+
+ END_PROFILE(SMBlockread);
+ return;
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_ALL
+
+/****************************************************************************
+ Reply to a read.
+****************************************************************************/
+
+void reply_read(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ size_t numtoread;
+ ssize_t nread = 0;
+ char *data;
+ SMB_OFF_T startpos;
+ int outsize = 0;
+ files_struct *fsp;
+
+ START_PROFILE(SMBread);
+
+ if (req->wct < 3) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBread);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBread);
+ return;
+ }
+
+ if (!CHECK_READ(fsp,req->inbuf)) {
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ END_PROFILE(SMBread);
+ return;
+ }
+
+ numtoread = SVAL(req->inbuf,smb_vwv1);
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv2);
+
+ numtoread = MIN(BUFFER_SIZE-outsize,numtoread);
+
+ /*
+ * The requested read size cannot be greater than max_recv. JRA.
+ */
+ if (numtoread > max_recv) {
+ DEBUG(0,("reply_read: requested read size (%u) is greater than maximum allowed (%u). \
+Returning short read of maximum allowed for compatibility with Windows 2000.\n",
+ (unsigned int)numtoread, (unsigned int)max_recv ));
+ numtoread = MIN(numtoread,max_recv);
+ }
+
+ reply_outbuf(req, 5, numtoread+3);
+
+ data = smb_buf(req->outbuf) + 3;
+
+ if (is_locked(fsp, (uint32)req->smbpid, (SMB_BIG_UINT)numtoread,
+ (SMB_BIG_UINT)startpos, READ_LOCK)) {
+ reply_doserror(req, ERRDOS,ERRlock);
+ END_PROFILE(SMBread);
+ return;
+ }
+
+ if (numtoread > 0)
+ nread = read_file(fsp,data,startpos,numtoread);
+
+ if (nread < 0) {
+ reply_unixerror(req, ERRDOS,ERRnoaccess);
+ END_PROFILE(SMBread);
+ return;
+ }
+
+ srv_set_message((char *)req->outbuf, 5, nread+3, False);
+
+ SSVAL(req->outbuf,smb_vwv0,nread);
+ SSVAL(req->outbuf,smb_vwv5,nread+3);
+ SCVAL(smb_buf(req->outbuf),0,1);
+ SSVAL(smb_buf(req->outbuf),1,nread);
+
+ DEBUG( 3, ( "read fnum=%d num=%d nread=%d\n",
+ fsp->fnum, (int)numtoread, (int)nread ) );
+
+ END_PROFILE(SMBread);
+ return;
+}
+
+/****************************************************************************
+ Setup readX header.
+****************************************************************************/
+
+static int setup_readX_header(char *outbuf, size_t smb_maxcnt)
+{
+ int outsize;
+ char *data;
+
+ outsize = srv_set_message(outbuf,12,smb_maxcnt,False);
+ data = smb_buf(outbuf);
+
+ memset(outbuf+smb_vwv0,'\0',24); /* valgrind init. */
+
+ SCVAL(outbuf,smb_vwv0,0xFF);
+ SSVAL(outbuf,smb_vwv2,0xFFFF); /* Remaining - must be -1. */
+ SSVAL(outbuf,smb_vwv5,smb_maxcnt);
+ SSVAL(outbuf,smb_vwv6,smb_offset(data,outbuf));
+ SSVAL(outbuf,smb_vwv7,(smb_maxcnt >> 16));
+ SSVAL(smb_buf(outbuf),-2,smb_maxcnt);
+ /* Reset the outgoing length, set_message truncates at 0x1FFFF. */
+ _smb_setlen_large(outbuf,(smb_size + 12*2 + smb_maxcnt - 4));
+ return outsize;
+}
+
+/****************************************************************************
+ Reply to a read and X - possibly using sendfile.
+****************************************************************************/
+
+static void send_file_readX(connection_struct *conn, struct smb_request *req,
+ files_struct *fsp, SMB_OFF_T startpos,
+ size_t smb_maxcnt)
+{
+ SMB_STRUCT_STAT sbuf;
+ ssize_t nread = -1;
+
+ if(SMB_VFS_FSTAT(fsp, &sbuf) == -1) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ return;
+ }
+
+ if (startpos > sbuf.st_size) {
+ smb_maxcnt = 0;
+ } else if (smb_maxcnt > (sbuf.st_size - startpos)) {
+ smb_maxcnt = (sbuf.st_size - startpos);
+ }
+
+ if (smb_maxcnt == 0) {
+ goto normal_read;
+ }
+
+#if defined(WITH_SENDFILE)
+ /*
+ * We can only use sendfile on a non-chained packet
+ * but we can use on a non-oplocked file. tridge proved this
+ * on a train in Germany :-). JRA.
+ */
+
+ if ((chain_size == 0) && (CVAL(req->inbuf,smb_vwv0) == 0xFF) &&
+ !is_encrypted_packet(req->inbuf) && (fsp->base_fsp == NULL) &&
+ lp_use_sendfile(SNUM(conn)) && (fsp->wcp == NULL) ) {
+ uint8 headerbuf[smb_size + 12 * 2];
+ DATA_BLOB header;
+
+ /*
+ * Set up the packet header before send. We
+ * assume here the sendfile will work (get the
+ * correct amount of data).
+ */
+
+ header = data_blob_const(headerbuf, sizeof(headerbuf));
+
+ construct_reply_common((char *)req->inbuf, (char *)headerbuf);
+ setup_readX_header((char *)headerbuf, smb_maxcnt);
+
+ if ((nread = SMB_VFS_SENDFILE(smbd_server_fd(), fsp, &header, startpos, smb_maxcnt)) == -1) {
+ /* Returning ENOSYS or EINVAL means no data at all was sent.
+ Do this as a normal read. */
+ if (errno == ENOSYS || errno == EINVAL) {
+ goto normal_read;
+ }
+
+ /*
+ * Special hack for broken Linux with no working sendfile. If we
+ * return EINTR we sent the header but not the rest of the data.
+ * Fake this up by doing read/write calls.
+ */
+
+ if (errno == EINTR) {
+ /* Ensure we don't do this again. */
+ set_use_sendfile(SNUM(conn), False);
+ DEBUG(0,("send_file_readX: sendfile not available. Faking..\n"));
+ nread = fake_sendfile(fsp, startpos,
+ smb_maxcnt);
+ if (nread == -1) {
+ DEBUG(0,("send_file_readX: fake_sendfile failed for file %s (%s).\n",
+ fsp->fsp_name, strerror(errno) ));
+ exit_server_cleanly("send_file_readX: fake_sendfile failed");
+ }
+ DEBUG( 3, ( "send_file_readX: fake_sendfile fnum=%d max=%d nread=%d\n",
+ fsp->fnum, (int)smb_maxcnt, (int)nread ) );
+ /* No outbuf here means successful sendfile. */
+ TALLOC_FREE(req->outbuf);
+ return;
+ }
+
+ DEBUG(0,("send_file_readX: sendfile failed for file %s (%s). Terminating\n",
+ fsp->fsp_name, strerror(errno) ));
+ exit_server_cleanly("send_file_readX sendfile failed");
+ }
+
+ DEBUG( 3, ( "send_file_readX: sendfile fnum=%d max=%d nread=%d\n",
+ fsp->fnum, (int)smb_maxcnt, (int)nread ) );
+ /* No outbuf here means successful sendfile. */
+ TALLOC_FREE(req->outbuf);
+ return;
+ }
+#endif
+
+normal_read:
+
+ if ((smb_maxcnt & 0xFF0000) > 0x10000) {
+ uint8 headerbuf[smb_size + 2*12];
+
+ construct_reply_common((char *)req->inbuf, (char *)headerbuf);
+ setup_readX_header((char *)headerbuf, smb_maxcnt);
+
+ /* Send out the header. */
+ if (write_data(smbd_server_fd(), (char *)headerbuf,
+ sizeof(headerbuf)) != sizeof(headerbuf)) {
+ DEBUG(0,("send_file_readX: write_data failed for file %s (%s). Terminating\n",
+ fsp->fsp_name, strerror(errno) ));
+ exit_server_cleanly("send_file_readX sendfile failed");
+ }
+ nread = fake_sendfile(fsp, startpos, smb_maxcnt);
+ if (nread == -1) {
+ DEBUG(0,("send_file_readX: fake_sendfile failed for file %s (%s).\n",
+ fsp->fsp_name, strerror(errno) ));
+ exit_server_cleanly("send_file_readX: fake_sendfile failed");
+ }
+ TALLOC_FREE(req->outbuf);
+ return;
+ }
+
+ reply_outbuf(req, 12, smb_maxcnt);
+
+ nread = read_file(fsp, smb_buf(req->outbuf), startpos, smb_maxcnt);
+ if (nread < 0) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ return;
+ }
+
+ setup_readX_header((char *)req->outbuf, nread);
+
+ DEBUG( 3, ( "send_file_readX fnum=%d max=%d nread=%d\n",
+ fsp->fnum, (int)smb_maxcnt, (int)nread ) );
+
+ chain_reply(req);
+}
+
+/****************************************************************************
+ Reply to a read and X.
+****************************************************************************/
+
+void reply_read_and_X(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ files_struct *fsp;
+ SMB_OFF_T startpos;
+ size_t smb_maxcnt;
+ bool big_readX = False;
+#if 0
+ size_t smb_mincnt = SVAL(req->inbuf,smb_vwv6);
+#endif
+
+ START_PROFILE(SMBreadX);
+
+ if ((req->wct != 10) && (req->wct != 12)) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv2));
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv3);
+ smb_maxcnt = SVAL(req->inbuf,smb_vwv5);
+
+ /* If it's an IPC, pass off the pipe handler. */
+ if (IS_IPC(conn)) {
+ reply_pipe_read_and_X(req);
+ END_PROFILE(SMBreadX);
+ return;
+ }
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBreadX);
+ return;
+ }
+
+ if (!CHECK_READ(fsp,req->inbuf)) {
+ reply_doserror(req, ERRDOS,ERRbadaccess);
+ END_PROFILE(SMBreadX);
+ return;
+ }
+
+ if (global_client_caps & CAP_LARGE_READX) {
+ size_t upper_size = SVAL(req->inbuf,smb_vwv7);
+ smb_maxcnt |= (upper_size<<16);
+ if (upper_size > 1) {
+ /* Can't do this on a chained packet. */
+ if ((CVAL(req->inbuf,smb_vwv0) != 0xFF)) {
+ reply_nterror(req, NT_STATUS_NOT_SUPPORTED);
+ END_PROFILE(SMBreadX);
+ return;
+ }
+ /* We currently don't do this on signed or sealed data. */
+ if (srv_is_signing_active() || is_encrypted_packet(req->inbuf)) {
+ reply_nterror(req, NT_STATUS_NOT_SUPPORTED);
+ END_PROFILE(SMBreadX);
+ return;
+ }
+ /* Is there room in the reply for this data ? */
+ if (smb_maxcnt > (0xFFFFFF - (smb_size -4 + 12*2))) {
+ reply_nterror(req,
+ NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBreadX);
+ return;
+ }
+ big_readX = True;
+ }
+ }
+
+ if (req->wct == 12) {
+#ifdef LARGE_SMB_OFF_T
+ /*
+ * This is a large offset (64 bit) read.
+ */
+ startpos |= (((SMB_OFF_T)IVAL(req->inbuf,smb_vwv10)) << 32);
+
+#else /* !LARGE_SMB_OFF_T */
+
+ /*
+ * Ensure we haven't been sent a >32 bit offset.
+ */
+
+ if(IVAL(req->inbuf,smb_vwv10) != 0) {
+ DEBUG(0,("reply_read_and_X - large offset (%x << 32) "
+ "used and we don't support 64 bit offsets.\n",
+ (unsigned int)IVAL(req->inbuf,smb_vwv10) ));
+ END_PROFILE(SMBreadX);
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ return;
+ }
+
+#endif /* LARGE_SMB_OFF_T */
+
+ }
+
+ if (is_locked(fsp, (uint32)req->smbpid, (SMB_BIG_UINT)smb_maxcnt,
+ (SMB_BIG_UINT)startpos, READ_LOCK)) {
+ END_PROFILE(SMBreadX);
+ reply_doserror(req, ERRDOS, ERRlock);
+ return;
+ }
+
+ if (!big_readX &&
+ schedule_aio_read_and_X(conn, req, fsp, startpos, smb_maxcnt)) {
+ END_PROFILE(SMBreadX);
+ return;
+ }
+
+ send_file_readX(conn, req, fsp, startpos, smb_maxcnt);
+
+ END_PROFILE(SMBreadX);
+ return;
+}
+
+/****************************************************************************
+ Error replies to writebraw must have smb_wct == 1. Fix this up.
+****************************************************************************/
+
+void error_to_writebrawerr(struct smb_request *req)
+{
+ uint8 *old_outbuf = req->outbuf;
+
+ reply_outbuf(req, 1, 0);
+
+ memcpy(req->outbuf, old_outbuf, smb_size);
+ TALLOC_FREE(old_outbuf);
+}
+
+/****************************************************************************
+ Reply to a writebraw (core+ or LANMAN1.0 protocol).
+****************************************************************************/
+
+void reply_writebraw(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *buf = NULL;
+ ssize_t nwritten=0;
+ ssize_t total_written=0;
+ size_t numtowrite=0;
+ size_t tcount;
+ SMB_OFF_T startpos;
+ char *data=NULL;
+ bool write_through;
+ files_struct *fsp;
+ NTSTATUS status;
+
+ START_PROFILE(SMBwritebraw);
+
+ /*
+ * If we ever reply with an error, it must have the SMB command
+ * type of SMBwritec, not SMBwriteBraw, as this tells the client
+ * we're finished.
+ */
+ SCVAL(req->inbuf,smb_com,SMBwritec);
+
+ if (srv_is_signing_active()) {
+ END_PROFILE(SMBwritebraw);
+ exit_server_cleanly("reply_writebraw: SMB signing is active - "
+ "raw reads/writes are disallowed.");
+ }
+
+ if (req->wct < 12) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+ if (!check_fsp(conn, req, fsp)) {
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ if (!CHECK_WRITE(fsp)) {
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ tcount = IVAL(req->inbuf,smb_vwv1);
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv3);
+ write_through = BITSETW(req->inbuf+smb_vwv7,0);
+
+ /* We have to deal with slightly different formats depending
+ on whether we are using the core+ or lanman1.0 protocol */
+
+ if(Protocol <= PROTOCOL_COREPLUS) {
+ numtowrite = SVAL(smb_buf(req->inbuf),-2);
+ data = smb_buf(req->inbuf);
+ } else {
+ numtowrite = SVAL(req->inbuf,smb_vwv10);
+ data = smb_base(req->inbuf) + SVAL(req->inbuf, smb_vwv11);
+ }
+
+ /* Ensure we don't write bytes past the end of this packet. */
+ if (data + numtowrite > smb_base(req->inbuf) + smb_len(req->inbuf)) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ if (is_locked(fsp,(uint32)req->smbpid,(SMB_BIG_UINT)tcount,
+ (SMB_BIG_UINT)startpos, WRITE_LOCK)) {
+ reply_doserror(req, ERRDOS, ERRlock);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ if (numtowrite>0) {
+ nwritten = write_file(req,fsp,data,startpos,numtowrite);
+ }
+
+ DEBUG(3,("reply_writebraw: initial write fnum=%d start=%.0f num=%d "
+ "wrote=%d sync=%d\n",
+ fsp->fnum, (double)startpos, (int)numtowrite,
+ (int)nwritten, (int)write_through));
+
+ if (nwritten < (ssize_t)numtowrite) {
+ reply_unixerror(req, ERRHRD, ERRdiskfull);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ total_written = nwritten;
+
+ /* Allocate a buffer of 64k + length. */
+ buf = TALLOC_ARRAY(NULL, char, 65540);
+ if (!buf) {
+ reply_doserror(req, ERRDOS, ERRnomem);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ /* Return a SMBwritebraw message to the redirector to tell
+ * it to send more bytes */
+
+ memcpy(buf, req->inbuf, smb_size);
+ srv_set_message(buf,Protocol>PROTOCOL_COREPLUS?1:0,0,True);
+ SCVAL(buf,smb_com,SMBwritebraw);
+ SSVALS(buf,smb_vwv0,0xFFFF);
+ show_msg(buf);
+ if (!srv_send_smb(smbd_server_fd(),
+ buf,
+ IS_CONN_ENCRYPTED(conn))) {
+ exit_server_cleanly("reply_writebraw: srv_send_smb "
+ "failed.");
+ }
+
+ /* Now read the raw data into the buffer and write it */
+ status = read_smb_length(smbd_server_fd(), buf, SMB_SECONDARY_WAIT,
+ &numtowrite);
+ if (!NT_STATUS_IS_OK(status)) {
+ exit_server_cleanly("secondary writebraw failed");
+ }
+
+ /* Set up outbuf to return the correct size */
+ reply_outbuf(req, 1, 0);
+
+ if (numtowrite != 0) {
+
+ if (numtowrite > 0xFFFF) {
+ DEBUG(0,("reply_writebraw: Oversize secondary write "
+ "raw requested (%u). Terminating\n",
+ (unsigned int)numtowrite ));
+ exit_server_cleanly("secondary writebraw failed");
+ }
+
+ if (tcount > nwritten+numtowrite) {
+ DEBUG(3,("reply_writebraw: Client overestimated the "
+ "write %d %d %d\n",
+ (int)tcount,(int)nwritten,(int)numtowrite));
+ }
+
+ status = read_data(smbd_server_fd(), buf+4, numtowrite);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(0,("reply_writebraw: Oversize secondary write "
+ "raw read failed (%s). Terminating\n",
+ nt_errstr(status)));
+ exit_server_cleanly("secondary writebraw failed");
+ }
+
+ nwritten = write_file(req,fsp,buf+4,startpos+nwritten,numtowrite);
+ if (nwritten == -1) {
+ TALLOC_FREE(buf);
+ reply_unixerror(req, ERRHRD, ERRdiskfull);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ if (nwritten < (ssize_t)numtowrite) {
+ SCVAL(req->outbuf,smb_rcls,ERRHRD);
+ SSVAL(req->outbuf,smb_err,ERRdiskfull);
+ }
+
+ if (nwritten > 0) {
+ total_written += nwritten;
+ }
+ }
+
+ TALLOC_FREE(buf);
+ SSVAL(req->outbuf,smb_vwv0,total_written);
+
+ status = sync_file(conn, fsp, write_through);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(5,("reply_writebraw: sync_file for %s returned %s\n",
+ fsp->fsp_name, nt_errstr(status) ));
+ reply_nterror(req, status);
+ error_to_writebrawerr(req);
+ END_PROFILE(SMBwritebraw);
+ return;
+ }
+
+ DEBUG(3,("reply_writebraw: secondart write fnum=%d start=%.0f num=%d "
+ "wrote=%d\n",
+ fsp->fnum, (double)startpos, (int)numtowrite,
+ (int)total_written));
+
+ /* We won't return a status if write through is not selected - this
+ * follows what WfWg does */
+ END_PROFILE(SMBwritebraw);
+
+ if (!write_through && total_written==tcount) {
+
+#if RABBIT_PELLET_FIX
+ /*
+ * Fix for "rabbit pellet" mode, trigger an early TCP ack by
+ * sending a SMBkeepalive. Thanks to DaveCB at Sun for this.
+ * JRA.
+ */
+ if (!send_keepalive(smbd_server_fd())) {
+ exit_server_cleanly("reply_writebraw: send of "
+ "keepalive failed");
+ }
+#endif
+ TALLOC_FREE(req->outbuf);
+ }
+ return;
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_LOCKING
+
+/****************************************************************************
+ Reply to a writeunlock (core+).
+****************************************************************************/
+
+void reply_writeunlock(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ ssize_t nwritten = -1;
+ size_t numtowrite;
+ SMB_OFF_T startpos;
+ char *data;
+ NTSTATUS status = NT_STATUS_OK;
+ files_struct *fsp;
+
+ START_PROFILE(SMBwriteunlock);
+
+ if (req->wct < 5) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBwriteunlock);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBwriteunlock);
+ return;
+ }
+
+ if (!CHECK_WRITE(fsp)) {
+ reply_doserror(req, ERRDOS,ERRbadaccess);
+ END_PROFILE(SMBwriteunlock);
+ return;
+ }
+
+ numtowrite = SVAL(req->inbuf,smb_vwv1);
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv2);
+ data = smb_buf(req->inbuf) + 3;
+
+ if (numtowrite
+ && is_locked(fsp, (uint32)req->smbpid, (SMB_BIG_UINT)numtowrite,
+ (SMB_BIG_UINT)startpos, WRITE_LOCK)) {
+ reply_doserror(req, ERRDOS, ERRlock);
+ END_PROFILE(SMBwriteunlock);
+ return;
+ }
+
+ /* The special X/Open SMB protocol handling of
+ zero length writes is *NOT* done for
+ this call */
+ if(numtowrite == 0) {
+ nwritten = 0;
+ } else {
+ nwritten = write_file(req,fsp,data,startpos,numtowrite);
+ }
+
+ status = sync_file(conn, fsp, False /* write through */);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(5,("reply_writeunlock: sync_file for %s returned %s\n",
+ fsp->fsp_name, nt_errstr(status) ));
+ reply_nterror(req, status);
+ END_PROFILE(SMBwriteunlock);
+ return;
+ }
+
+ if(((nwritten < numtowrite) && (numtowrite != 0))||(nwritten < 0)) {
+ reply_unixerror(req, ERRHRD, ERRdiskfull);
+ END_PROFILE(SMBwriteunlock);
+ return;
+ }
+
+ if (numtowrite) {
+ status = do_unlock(smbd_messaging_context(),
+ fsp,
+ req->smbpid,
+ (SMB_BIG_UINT)numtowrite,
+ (SMB_BIG_UINT)startpos,
+ WINDOWS_LOCK);
+
+ if (NT_STATUS_V(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBwriteunlock);
+ return;
+ }
+ }
+
+ reply_outbuf(req, 1, 0);
+
+ SSVAL(req->outbuf,smb_vwv0,nwritten);
+
+ DEBUG(3,("writeunlock fnum=%d num=%d wrote=%d\n",
+ fsp->fnum, (int)numtowrite, (int)nwritten));
+
+ END_PROFILE(SMBwriteunlock);
+ return;
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_ALL
+
+/****************************************************************************
+ Reply to a write.
+****************************************************************************/
+
+void reply_write(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ size_t numtowrite;
+ ssize_t nwritten = -1;
+ SMB_OFF_T startpos;
+ char *data;
+ files_struct *fsp;
+ NTSTATUS status;
+
+ START_PROFILE(SMBwrite);
+
+ if (req->wct < 5) {
+ END_PROFILE(SMBwrite);
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ return;
+ }
+
+ /* If it's an IPC, pass off the pipe handler. */
+ if (IS_IPC(conn)) {
+ reply_pipe_write(req);
+ END_PROFILE(SMBwrite);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBwrite);
+ return;
+ }
+
+ if (!CHECK_WRITE(fsp)) {
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ END_PROFILE(SMBwrite);
+ return;
+ }
+
+ numtowrite = SVAL(req->inbuf,smb_vwv1);
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv2);
+ data = smb_buf(req->inbuf) + 3;
+
+ if (is_locked(fsp, (uint32)req->smbpid, (SMB_BIG_UINT)numtowrite,
+ (SMB_BIG_UINT)startpos, WRITE_LOCK)) {
+ reply_doserror(req, ERRDOS, ERRlock);
+ END_PROFILE(SMBwrite);
+ return;
+ }
+
+ /*
+ * X/Open SMB protocol says that if smb_vwv1 is
+ * zero then the file size should be extended or
+ * truncated to the size given in smb_vwv[2-3].
+ */
+
+ if(numtowrite == 0) {
+ /*
+ * This is actually an allocate call, and set EOF. JRA.
+ */
+ nwritten = vfs_allocate_file_space(fsp, (SMB_OFF_T)startpos);
+ if (nwritten < 0) {
+ reply_nterror(req, NT_STATUS_DISK_FULL);
+ END_PROFILE(SMBwrite);
+ return;
+ }
+ nwritten = vfs_set_filelen(fsp, (SMB_OFF_T)startpos);
+ if (nwritten < 0) {
+ reply_nterror(req, NT_STATUS_DISK_FULL);
+ END_PROFILE(SMBwrite);
+ return;
+ }
+ trigger_write_time_update_immediate(fsp);
+ } else {
+ nwritten = write_file(req,fsp,data,startpos,numtowrite);
+ }
+
+ status = sync_file(conn, fsp, False);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(5,("reply_write: sync_file for %s returned %s\n",
+ fsp->fsp_name, nt_errstr(status) ));
+ reply_nterror(req, status);
+ END_PROFILE(SMBwrite);
+ return;
+ }
+
+ if(((nwritten == 0) && (numtowrite != 0))||(nwritten < 0)) {
+ reply_unixerror(req, ERRHRD, ERRdiskfull);
+ END_PROFILE(SMBwrite);
+ return;
+ }
+
+ reply_outbuf(req, 1, 0);
+
+ SSVAL(req->outbuf,smb_vwv0,nwritten);
+
+ if (nwritten < (ssize_t)numtowrite) {
+ SCVAL(req->outbuf,smb_rcls,ERRHRD);
+ SSVAL(req->outbuf,smb_err,ERRdiskfull);
+ }
+
+ DEBUG(3,("write fnum=%d num=%d wrote=%d\n", fsp->fnum, (int)numtowrite, (int)nwritten));
+
+ END_PROFILE(SMBwrite);
+ return;
+}
+
+/****************************************************************************
+ Ensure a buffer is a valid writeX for recvfile purposes.
+****************************************************************************/
+
+#define STANDARD_WRITE_AND_X_HEADER_SIZE (smb_size - 4 + /* basic header */ \
+ (2*14) + /* word count (including bcc) */ \
+ 1 /* pad byte */)
+
+bool is_valid_writeX_buffer(const uint8_t *inbuf)
+{
+ size_t numtowrite;
+ connection_struct *conn = NULL;
+ unsigned int doff = 0;
+ size_t len = smb_len_large(inbuf);
+
+ if (is_encrypted_packet(inbuf)) {
+ /* Can't do this on encrypted
+ * connections. */
+ return false;
+ }
+
+ if (CVAL(inbuf,smb_com) != SMBwriteX) {
+ return false;
+ }
+
+ if (CVAL(inbuf,smb_vwv0) != 0xFF ||
+ CVAL(inbuf,smb_wct) != 14) {
+ DEBUG(10,("is_valid_writeX_buffer: chained or "
+ "invalid word length.\n"));
+ return false;
+ }
+
+ conn = conn_find(SVAL(inbuf, smb_tid));
+ if (conn == NULL) {
+ DEBUG(10,("is_valid_writeX_buffer: bad tid\n"));
+ return false;
+ }
+ if (IS_IPC(conn)) {
+ DEBUG(10,("is_valid_writeX_buffer: IPC$ tid\n"));
+ return false;
+ }
+ doff = SVAL(inbuf,smb_vwv11);
+
+ numtowrite = SVAL(inbuf,smb_vwv10);
+
+ if (len > doff && len - doff > 0xFFFF) {
+ numtowrite |= (((size_t)SVAL(inbuf,smb_vwv9))<<16);
+ }
+
+ if (numtowrite == 0) {
+ DEBUG(10,("is_valid_writeX_buffer: zero write\n"));
+ return false;
+ }
+
+ /* Ensure the sizes match up. */
+ if (doff < STANDARD_WRITE_AND_X_HEADER_SIZE) {
+ /* no pad byte...old smbclient :-( */
+ DEBUG(10,("is_valid_writeX_buffer: small doff %u (min %u)\n",
+ (unsigned int)doff,
+ (unsigned int)STANDARD_WRITE_AND_X_HEADER_SIZE));
+ return false;
+ }
+
+ if (len - doff != numtowrite) {
+ DEBUG(10,("is_valid_writeX_buffer: doff mismatch "
+ "len = %u, doff = %u, numtowrite = %u\n",
+ (unsigned int)len,
+ (unsigned int)doff,
+ (unsigned int)numtowrite ));
+ return false;
+ }
+
+ DEBUG(10,("is_valid_writeX_buffer: true "
+ "len = %u, doff = %u, numtowrite = %u\n",
+ (unsigned int)len,
+ (unsigned int)doff,
+ (unsigned int)numtowrite ));
+
+ return true;
+}
+
+/****************************************************************************
+ Reply to a write and X.
+****************************************************************************/
+
+void reply_write_and_X(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ files_struct *fsp;
+ SMB_OFF_T startpos;
+ size_t numtowrite;
+ bool write_through;
+ ssize_t nwritten;
+ unsigned int smb_doff;
+ unsigned int smblen;
+ char *data;
+ NTSTATUS status;
+
+ START_PROFILE(SMBwriteX);
+
+ if ((req->wct != 12) && (req->wct != 14)) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ numtowrite = SVAL(req->inbuf,smb_vwv10);
+ smb_doff = SVAL(req->inbuf,smb_vwv11);
+ smblen = smb_len(req->inbuf);
+
+ if (req->unread_bytes > 0xFFFF ||
+ (smblen > smb_doff &&
+ smblen - smb_doff > 0xFFFF)) {
+ numtowrite |= (((size_t)SVAL(req->inbuf,smb_vwv9))<<16);
+ }
+
+ if (req->unread_bytes) {
+ /* Can't do a recvfile write on IPC$ */
+ if (IS_IPC(conn)) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+ if (numtowrite != req->unread_bytes) {
+ reply_doserror(req, ERRDOS, ERRbadmem);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+ } else {
+ if (smb_doff > smblen || smb_doff + numtowrite < numtowrite ||
+ smb_doff + numtowrite > smblen) {
+ reply_doserror(req, ERRDOS, ERRbadmem);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+ }
+
+ /* If it's an IPC, pass off the pipe handler. */
+ if (IS_IPC(conn)) {
+ if (req->unread_bytes) {
+ reply_doserror(req, ERRDOS, ERRbadmem);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+ reply_pipe_write_and_X(req);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv2));
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv3);
+ write_through = BITSETW(req->inbuf+smb_vwv7,0);
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ if (!CHECK_WRITE(fsp)) {
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ data = smb_base(req->inbuf) + smb_doff;
+
+ if(req->wct == 14) {
+#ifdef LARGE_SMB_OFF_T
+ /*
+ * This is a large offset (64 bit) write.
+ */
+ startpos |= (((SMB_OFF_T)IVAL(req->inbuf,smb_vwv12)) << 32);
+
+#else /* !LARGE_SMB_OFF_T */
+
+ /*
+ * Ensure we haven't been sent a >32 bit offset.
+ */
+
+ if(IVAL(req->inbuf,smb_vwv12) != 0) {
+ DEBUG(0,("reply_write_and_X - large offset (%x << 32) "
+ "used and we don't support 64 bit offsets.\n",
+ (unsigned int)IVAL(req->inbuf,smb_vwv12) ));
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+#endif /* LARGE_SMB_OFF_T */
+ }
+
+ if (is_locked(fsp,(uint32)req->smbpid,
+ (SMB_BIG_UINT)numtowrite,
+ (SMB_BIG_UINT)startpos, WRITE_LOCK)) {
+ reply_doserror(req, ERRDOS, ERRlock);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ /* X/Open SMB protocol says that, unlike SMBwrite
+ if the length is zero then NO truncation is
+ done, just a write of zero. To truncate a file,
+ use SMBwrite. */
+
+ if(numtowrite == 0) {
+ nwritten = 0;
+ } else {
+
+ if ((req->unread_bytes == 0) &&
+ schedule_aio_write_and_X(conn, req, fsp, data, startpos,
+ numtowrite)) {
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ nwritten = write_file(req,fsp,data,startpos,numtowrite);
+ }
+
+ if(((nwritten == 0) && (numtowrite != 0))||(nwritten < 0)) {
+ reply_unixerror(req, ERRHRD, ERRdiskfull);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ reply_outbuf(req, 6, 0);
+ SSVAL(req->outbuf,smb_vwv2,nwritten);
+ SSVAL(req->outbuf,smb_vwv4,nwritten>>16);
+
+ if (nwritten < (ssize_t)numtowrite) {
+ SCVAL(req->outbuf,smb_rcls,ERRHRD);
+ SSVAL(req->outbuf,smb_err,ERRdiskfull);
+ }
+
+ DEBUG(3,("writeX fnum=%d num=%d wrote=%d\n",
+ fsp->fnum, (int)numtowrite, (int)nwritten));
+
+ status = sync_file(conn, fsp, write_through);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(5,("reply_write_and_X: sync_file for %s returned %s\n",
+ fsp->fsp_name, nt_errstr(status) ));
+ reply_nterror(req, status);
+ END_PROFILE(SMBwriteX);
+ return;
+ }
+
+ END_PROFILE(SMBwriteX);
+ chain_reply(req);
+ return;
+}
+
+/****************************************************************************
+ Reply to a lseek.
+****************************************************************************/
+
+void reply_lseek(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ SMB_OFF_T startpos;
+ SMB_OFF_T res= -1;
+ int mode,umode;
+ files_struct *fsp;
+
+ START_PROFILE(SMBlseek);
+
+ if (req->wct < 4) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBlseek);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ return;
+ }
+
+ flush_write_cache(fsp, SEEK_FLUSH);
+
+ mode = SVAL(req->inbuf,smb_vwv1) & 3;
+ /* NB. This doesn't use IVAL_TO_SMB_OFF_T as startpos can be signed in this case. */
+ startpos = (SMB_OFF_T)IVALS(req->inbuf,smb_vwv2);
+
+ switch (mode) {
+ case 0:
+ umode = SEEK_SET;
+ res = startpos;
+ break;
+ case 1:
+ umode = SEEK_CUR;
+ res = fsp->fh->pos + startpos;
+ break;
+ case 2:
+ umode = SEEK_END;
+ break;
+ default:
+ umode = SEEK_SET;
+ res = startpos;
+ break;
+ }
+
+ if (umode == SEEK_END) {
+ if((res = SMB_VFS_LSEEK(fsp,startpos,umode)) == -1) {
+ if(errno == EINVAL) {
+ SMB_OFF_T current_pos = startpos;
+ SMB_STRUCT_STAT sbuf;
+
+ if(SMB_VFS_FSTAT(fsp, &sbuf) == -1) {
+ reply_unixerror(req, ERRDOS,
+ ERRnoaccess);
+ END_PROFILE(SMBlseek);
+ return;
+ }
+
+ current_pos += sbuf.st_size;
+ if(current_pos < 0)
+ res = SMB_VFS_LSEEK(fsp,0,SEEK_SET);
+ }
+ }
+
+ if(res == -1) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBlseek);
+ return;
+ }
+ }
+
+ fsp->fh->pos = res;
+
+ reply_outbuf(req, 2, 0);
+ SIVAL(req->outbuf,smb_vwv0,res);
+
+ DEBUG(3,("lseek fnum=%d ofs=%.0f newpos = %.0f mode=%d\n",
+ fsp->fnum, (double)startpos, (double)res, mode));
+
+ END_PROFILE(SMBlseek);
+ return;
+}
+
+/****************************************************************************
+ Reply to a flush.
+****************************************************************************/
+
+void reply_flush(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ uint16 fnum;
+ files_struct *fsp;
+
+ START_PROFILE(SMBflush);
+
+ if (req->wct < 1) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ return;
+ }
+
+ fnum = SVAL(req->inbuf,smb_vwv0);
+ fsp = file_fsp(fnum);
+
+ if ((fnum != 0xFFFF) && !check_fsp(conn, req, fsp)) {
+ return;
+ }
+
+ if (!fsp) {
+ file_sync_all(conn);
+ } else {
+ NTSTATUS status = sync_file(conn, fsp, True);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(5,("reply_flush: sync_file for %s returned %s\n",
+ fsp->fsp_name, nt_errstr(status) ));
+ reply_nterror(req, status);
+ END_PROFILE(SMBflush);
+ return;
+ }
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ DEBUG(3,("flush\n"));
+ END_PROFILE(SMBflush);
+ return;
+}
+
+/****************************************************************************
+ Reply to a exit.
+ conn POINTER CAN BE NULL HERE !
+****************************************************************************/
+
+void reply_exit(struct smb_request *req)
+{
+ START_PROFILE(SMBexit);
+
+ file_close_pid(req->smbpid, req->vuid);
+
+ reply_outbuf(req, 0, 0);
+
+ DEBUG(3,("exit\n"));
+
+ END_PROFILE(SMBexit);
+ return;
+}
+
+/****************************************************************************
+ Reply to a close - has to deal with closing a directory opened by NT SMB's.
+****************************************************************************/
+
+void reply_close(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ NTSTATUS status = NT_STATUS_OK;
+ files_struct *fsp = NULL;
+ START_PROFILE(SMBclose);
+
+ if (req->wct < 3) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBclose);
+ return;
+ }
+
+ /* If it's an IPC, pass off to the pipe handler. */
+ if (IS_IPC(conn)) {
+ reply_pipe_close(conn, req);
+ END_PROFILE(SMBclose);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ /*
+ * We can only use CHECK_FSP if we know it's not a directory.
+ */
+
+ if(!fsp || (fsp->conn != conn) || (fsp->vuid != req->vuid)) {
+ reply_doserror(req, ERRDOS, ERRbadfid);
+ END_PROFILE(SMBclose);
+ return;
+ }
+
+ if(fsp->is_directory) {
+ /*
+ * Special case - close NT SMB directory handle.
+ */
+ DEBUG(3,("close directory fnum=%d\n", fsp->fnum));
+ status = close_file(fsp,NORMAL_CLOSE);
+ } else {
+ time_t t;
+ /*
+ * Close ordinary file.
+ */
+
+ DEBUG(3,("close fd=%d fnum=%d (numopen=%d)\n",
+ fsp->fh->fd, fsp->fnum,
+ conn->num_files_open));
+
+ /*
+ * Take care of any time sent in the close.
+ */
+
+ t = srv_make_unix_date3(req->inbuf+smb_vwv1);
+ set_close_write_time(fsp, convert_time_t_to_timespec(t));
+
+ /*
+ * close_file() returns the unix errno if an error
+ * was detected on close - normally this is due to
+ * a disk full error. If not then it was probably an I/O error.
+ */
+
+ status = close_file(fsp,NORMAL_CLOSE);
+ }
+
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBclose);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+ END_PROFILE(SMBclose);
+ return;
+}
+
+/****************************************************************************
+ Reply to a writeclose (Core+ protocol).
+****************************************************************************/
+
+void reply_writeclose(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ size_t numtowrite;
+ ssize_t nwritten = -1;
+ NTSTATUS close_status = NT_STATUS_OK;
+ SMB_OFF_T startpos;
+ char *data;
+ struct timespec mtime;
+ files_struct *fsp;
+
+ START_PROFILE(SMBwriteclose);
+
+ if (req->wct < 6) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBwriteclose);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBwriteclose);
+ return;
+ }
+ if (!CHECK_WRITE(fsp)) {
+ reply_doserror(req, ERRDOS,ERRbadaccess);
+ END_PROFILE(SMBwriteclose);
+ return;
+ }
+
+ numtowrite = SVAL(req->inbuf,smb_vwv1);
+ startpos = IVAL_TO_SMB_OFF_T(req->inbuf,smb_vwv2);
+ mtime = convert_time_t_to_timespec(srv_make_unix_date3(
+ req->inbuf+smb_vwv4));
+ data = smb_buf(req->inbuf) + 1;
+
+ if (numtowrite
+ && is_locked(fsp, (uint32)req->smbpid, (SMB_BIG_UINT)numtowrite,
+ (SMB_BIG_UINT)startpos, WRITE_LOCK)) {
+ reply_doserror(req, ERRDOS,ERRlock);
+ END_PROFILE(SMBwriteclose);
+ return;
+ }
+
+ nwritten = write_file(req,fsp,data,startpos,numtowrite);
+
+ set_close_write_time(fsp, mtime);
+
+ /*
+ * More insanity. W2K only closes the file if writelen > 0.
+ * JRA.
+ */
+
+ if (numtowrite) {
+ DEBUG(3,("reply_writeclose: zero length write doesn't close file %s\n",
+ fsp->fsp_name ));
+ close_status = close_file(fsp,NORMAL_CLOSE);
+ }
+
+ DEBUG(3,("writeclose fnum=%d num=%d wrote=%d (numopen=%d)\n",
+ fsp->fnum, (int)numtowrite, (int)nwritten,
+ conn->num_files_open));
+
+ if(((nwritten == 0) && (numtowrite != 0))||(nwritten < 0)) {
+ reply_doserror(req, ERRHRD, ERRdiskfull);
+ END_PROFILE(SMBwriteclose);
+ return;
+ }
+
+ if(!NT_STATUS_IS_OK(close_status)) {
+ reply_nterror(req, close_status);
+ END_PROFILE(SMBwriteclose);
+ return;
+ }
+
+ reply_outbuf(req, 1, 0);
+
+ SSVAL(req->outbuf,smb_vwv0,nwritten);
+ END_PROFILE(SMBwriteclose);
+ return;
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_LOCKING
+
+/****************************************************************************
+ Reply to a lock.
+****************************************************************************/
+
+void reply_lock(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ SMB_BIG_UINT count,offset;
+ NTSTATUS status;
+ files_struct *fsp;
+ struct byte_range_lock *br_lck = NULL;
+
+ START_PROFILE(SMBlock);
+
+ if (req->wct < 5) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBlock);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBlock);
+ return;
+ }
+
+ release_level_2_oplocks_on_change(fsp);
+
+ count = (SMB_BIG_UINT)IVAL(req->inbuf,smb_vwv1);
+ offset = (SMB_BIG_UINT)IVAL(req->inbuf,smb_vwv3);
+
+ DEBUG(3,("lock fd=%d fnum=%d offset=%.0f count=%.0f\n",
+ fsp->fh->fd, fsp->fnum, (double)offset, (double)count));
+
+ br_lck = do_lock(smbd_messaging_context(),
+ fsp,
+ req->smbpid,
+ count,
+ offset,
+ WRITE_LOCK,
+ WINDOWS_LOCK,
+ False, /* Non-blocking lock. */
+ &status,
+ NULL);
+
+ TALLOC_FREE(br_lck);
+
+ if (NT_STATUS_V(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBlock);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ END_PROFILE(SMBlock);
+ return;
+}
+
+/****************************************************************************
+ Reply to a unlock.
+****************************************************************************/
+
+void reply_unlock(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ SMB_BIG_UINT count,offset;
+ NTSTATUS status;
+ files_struct *fsp;
+
+ START_PROFILE(SMBunlock);
+
+ if (req->wct < 5) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBunlock);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBunlock);
+ return;
+ }
+
+ count = (SMB_BIG_UINT)IVAL(req->inbuf,smb_vwv1);
+ offset = (SMB_BIG_UINT)IVAL(req->inbuf,smb_vwv3);
+
+ status = do_unlock(smbd_messaging_context(),
+ fsp,
+ req->smbpid,
+ count,
+ offset,
+ WINDOWS_LOCK);
+
+ if (NT_STATUS_V(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBunlock);
+ return;
+ }
+
+ DEBUG( 3, ( "unlock fd=%d fnum=%d offset=%.0f count=%.0f\n",
+ fsp->fh->fd, fsp->fnum, (double)offset, (double)count ) );
+
+ reply_outbuf(req, 0, 0);
+
+ END_PROFILE(SMBunlock);
+ return;
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_ALL
+
+/****************************************************************************
+ Reply to a tdis.
+ conn POINTER CAN BE NULL HERE !
+****************************************************************************/
+
+void reply_tdis(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ START_PROFILE(SMBtdis);
+
+ if (!conn) {
+ DEBUG(4,("Invalid connection in tdis\n"));
+ reply_doserror(req, ERRSRV, ERRinvnid);
+ END_PROFILE(SMBtdis);
+ return;
+ }
+
+ conn->used = False;
+
+ close_cnum(conn,req->vuid);
+ req->conn = NULL;
+
+ reply_outbuf(req, 0, 0);
+ END_PROFILE(SMBtdis);
+ return;
+}
+
+/****************************************************************************
+ Reply to a echo.
+ conn POINTER CAN BE NULL HERE !
+****************************************************************************/
+
+void reply_echo(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ int smb_reverb;
+ int seq_num;
+ unsigned int data_len = smb_buflen(req->inbuf);
+
+ START_PROFILE(SMBecho);
+
+ if (req->wct < 1) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBecho);
+ return;
+ }
+
+ if (data_len > BUFFER_SIZE) {
+ DEBUG(0,("reply_echo: data_len too large.\n"));
+ reply_nterror(req, NT_STATUS_INSUFFICIENT_RESOURCES);
+ END_PROFILE(SMBecho);
+ return;
+ }
+
+ smb_reverb = SVAL(req->inbuf,smb_vwv0);
+
+ reply_outbuf(req, 1, data_len);
+
+ /* copy any incoming data back out */
+ if (data_len > 0) {
+ memcpy(smb_buf(req->outbuf),smb_buf(req->inbuf),data_len);
+ }
+
+ if (smb_reverb > 100) {
+ DEBUG(0,("large reverb (%d)?? Setting to 100\n",smb_reverb));
+ smb_reverb = 100;
+ }
+
+ for (seq_num =1 ; seq_num <= smb_reverb ; seq_num++) {
+ SSVAL(req->outbuf,smb_vwv0,seq_num);
+
+ show_msg((char *)req->outbuf);
+ if (!srv_send_smb(smbd_server_fd(),
+ (char *)req->outbuf,
+ IS_CONN_ENCRYPTED(conn)||req->encrypted))
+ exit_server_cleanly("reply_echo: srv_send_smb failed.");
+ }
+
+ DEBUG(3,("echo %d times\n", smb_reverb));
+
+ TALLOC_FREE(req->outbuf);
+
+ smb_echo_count++;
+
+ END_PROFILE(SMBecho);
+ return;
+}
+
+/****************************************************************************
+ Reply to a printopen.
+****************************************************************************/
+
+void reply_printopen(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ files_struct *fsp;
+ NTSTATUS status;
+
+ START_PROFILE(SMBsplopen);
+
+ if (req->wct < 2) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBsplopen);
+ return;
+ }
+
+ if (!CAN_PRINT(conn)) {
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBsplopen);
+ return;
+ }
+
+ /* Open for exclusive use, write only. */
+ status = print_fsp_open(conn, NULL, req->vuid, &fsp);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBsplopen);
+ return;
+ }
+
+ reply_outbuf(req, 1, 0);
+ SSVAL(req->outbuf,smb_vwv0,fsp->fnum);
+
+ DEBUG(3,("openprint fd=%d fnum=%d\n",
+ fsp->fh->fd, fsp->fnum));
+
+ END_PROFILE(SMBsplopen);
+ return;
+}
+
+/****************************************************************************
+ Reply to a printclose.
+****************************************************************************/
+
+void reply_printclose(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ files_struct *fsp;
+ NTSTATUS status;
+
+ START_PROFILE(SMBsplclose);
+
+ if (req->wct < 1) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBsplclose);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBsplclose);
+ return;
+ }
+
+ if (!CAN_PRINT(conn)) {
+ reply_nterror(req, NT_STATUS_DOS(ERRSRV, ERRerror));
+ END_PROFILE(SMBsplclose);
+ return;
+ }
+
+ DEBUG(3,("printclose fd=%d fnum=%d\n",
+ fsp->fh->fd,fsp->fnum));
+
+ status = close_file(fsp,NORMAL_CLOSE);
+
+ if(!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBsplclose);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ END_PROFILE(SMBsplclose);
+ return;
+}
+
+/****************************************************************************
+ Reply to a printqueue.
+****************************************************************************/
+
+void reply_printqueue(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ int max_count;
+ int start_index;
+
+ START_PROFILE(SMBsplretq);
+
+ if (req->wct < 2) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBsplretq);
+ return;
+ }
+
+ max_count = SVAL(req->inbuf,smb_vwv0);
+ start_index = SVAL(req->inbuf,smb_vwv1);
+
+ /* we used to allow the client to get the cnum wrong, but that
+ is really quite gross and only worked when there was only
+ one printer - I think we should now only accept it if they
+ get it right (tridge) */
+ if (!CAN_PRINT(conn)) {
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBsplretq);
+ return;
+ }
+
+ reply_outbuf(req, 2, 3);
+ SSVAL(req->outbuf,smb_vwv0,0);
+ SSVAL(req->outbuf,smb_vwv1,0);
+ SCVAL(smb_buf(req->outbuf),0,1);
+ SSVAL(smb_buf(req->outbuf),1,0);
+
+ DEBUG(3,("printqueue start_index=%d max_count=%d\n",
+ start_index, max_count));
+
+ {
+ print_queue_struct *queue = NULL;
+ print_status_struct status;
+ int count = print_queue_status(SNUM(conn), &queue, &status);
+ int num_to_get = ABS(max_count);
+ int first = (max_count>0?start_index:start_index+max_count+1);
+ int i;
+
+ if (first >= count)
+ num_to_get = 0;
+ else
+ num_to_get = MIN(num_to_get,count-first);
+
+
+ for (i=first;i<first+num_to_get;i++) {
+ char blob[28];
+ char *p = blob;
+
+ srv_put_dos_date2(p,0,queue[i].time);
+ SCVAL(p,4,(queue[i].status==LPQ_PRINTING?2:3));
+ SSVAL(p,5, queue[i].job);
+ SIVAL(p,7,queue[i].size);
+ SCVAL(p,11,0);
+ srvstr_push(blob, req->flags2, p+12,
+ queue[i].fs_user, 16, STR_ASCII);
+
+ if (message_push_blob(
+ &req->outbuf,
+ data_blob_const(
+ blob, sizeof(blob))) == -1) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBsplretq);
+ return;
+ }
+ }
+
+ if (count > 0) {
+ SSVAL(req->outbuf,smb_vwv0,count);
+ SSVAL(req->outbuf,smb_vwv1,
+ (max_count>0?first+count:first-1));
+ SCVAL(smb_buf(req->outbuf),0,1);
+ SSVAL(smb_buf(req->outbuf),1,28*count);
+ }
+
+ SAFE_FREE(queue);
+
+ DEBUG(3,("%d entries returned in queue\n",count));
+ }
+
+ END_PROFILE(SMBsplretq);
+ return;
+}
+
+/****************************************************************************
+ Reply to a printwrite.
+****************************************************************************/
+
+void reply_printwrite(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ int numtowrite;
+ char *data;
+ files_struct *fsp;
+
+ START_PROFILE(SMBsplwr);
+
+ if (req->wct < 1) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBsplwr);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBsplwr);
+ return;
+ }
+
+ if (!CAN_PRINT(conn)) {
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBsplwr);
+ return;
+ }
+
+ if (!CHECK_WRITE(fsp)) {
+ reply_doserror(req, ERRDOS, ERRbadaccess);
+ END_PROFILE(SMBsplwr);
+ return;
+ }
+
+ numtowrite = SVAL(smb_buf(req->inbuf),1);
+
+ if (smb_buflen(req->inbuf) < numtowrite + 3) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBsplwr);
+ return;
+ }
+
+ data = smb_buf(req->inbuf) + 3;
+
+ if (write_file(req,fsp,data,-1,numtowrite) != numtowrite) {
+ reply_unixerror(req, ERRHRD, ERRdiskfull);
+ END_PROFILE(SMBsplwr);
+ return;
+ }
+
+ DEBUG( 3, ( "printwrite fnum=%d num=%d\n", fsp->fnum, numtowrite ) );
+
+ END_PROFILE(SMBsplwr);
+ return;
+}
+
+/****************************************************************************
+ Reply to a mkdir.
+****************************************************************************/
+
+void reply_mkdir(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *directory = NULL;
+ NTSTATUS status;
+ SMB_STRUCT_STAT sbuf;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBmkdir);
+
+ srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &directory,
+ smb_buf(req->inbuf) + 1, 0,
+ STR_TERMINATE, &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBmkdir);
+ return;
+ }
+
+ status = resolve_dfspath(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ directory,
+ &directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBmkdir);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBmkdir);
+ return;
+ }
+
+ status = unix_convert(ctx, conn, directory, False, &directory, NULL, &sbuf);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBmkdir);
+ return;
+ }
+
+ status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBmkdir);
+ return;
+ }
+
+ status = create_directory(conn, req, directory);
+
+ DEBUG(5, ("create_directory returned %s\n", nt_errstr(status)));
+
+ if (!NT_STATUS_IS_OK(status)) {
+
+ if (!use_nt_status()
+ && NT_STATUS_EQUAL(status,
+ NT_STATUS_OBJECT_NAME_COLLISION)) {
+ /*
+ * Yes, in the DOS error code case we get a
+ * ERRDOS:ERRnoaccess here. See BASE-SAMBA3ERROR
+ * samba4 torture test.
+ */
+ status = NT_STATUS_DOS(ERRDOS, ERRnoaccess);
+ }
+
+ reply_nterror(req, status);
+ END_PROFILE(SMBmkdir);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ DEBUG( 3, ( "mkdir %s\n", directory ) );
+
+ END_PROFILE(SMBmkdir);
+ return;
+}
+
+/****************************************************************************
+ Static function used by reply_rmdir to delete an entire directory
+ tree recursively. Return True on ok, False on fail.
+****************************************************************************/
+
+static bool recursive_rmdir(TALLOC_CTX *ctx,
+ connection_struct *conn,
+ char *directory)
+{
+ const char *dname = NULL;
+ bool ret = True;
+ long offset = 0;
+ struct smb_Dir *dir_hnd = OpenDir(talloc_tos(), conn, directory,
+ NULL, 0);
+
+ if(dir_hnd == NULL)
+ return False;
+
+ while((dname = ReadDirName(dir_hnd, &offset))) {
+ char *fullname = NULL;
+ SMB_STRUCT_STAT st;
+
+ if (ISDOT(dname) || ISDOTDOT(dname)) {
+ continue;
+ }
+
+ if (!is_visible_file(conn, directory, dname, &st, False)) {
+ continue;
+ }
+
+ /* Construct the full name. */
+ fullname = talloc_asprintf(ctx,
+ "%s/%s",
+ directory,
+ dname);
+ if (!fullname) {
+ errno = ENOMEM;
+ ret = False;
+ break;
+ }
+
+ if(SMB_VFS_LSTAT(conn,fullname, &st) != 0) {
+ ret = False;
+ break;
+ }
+
+ if(st.st_mode & S_IFDIR) {
+ if(!recursive_rmdir(ctx, conn, fullname)) {
+ ret = False;
+ break;
+ }
+ if(SMB_VFS_RMDIR(conn,fullname) != 0) {
+ ret = False;
+ break;
+ }
+ } else if(SMB_VFS_UNLINK(conn,fullname) != 0) {
+ ret = False;
+ break;
+ }
+ TALLOC_FREE(fullname);
+ }
+ TALLOC_FREE(dir_hnd);
+ return ret;
+}
+
+/****************************************************************************
+ The internals of the rmdir code - called elsewhere.
+****************************************************************************/
+
+NTSTATUS rmdir_internals(TALLOC_CTX *ctx,
+ connection_struct *conn,
+ const char *directory)
+{
+ int ret;
+ SMB_STRUCT_STAT st;
+
+ /* Might be a symlink. */
+ if(SMB_VFS_LSTAT(conn, directory, &st) != 0) {
+ return map_nt_error_from_unix(errno);
+ }
+
+ if (S_ISLNK(st.st_mode)) {
+ /* Is what it points to a directory ? */
+ if(SMB_VFS_STAT(conn, directory, &st) != 0) {
+ return map_nt_error_from_unix(errno);
+ }
+ if (!(S_ISDIR(st.st_mode))) {
+ return NT_STATUS_NOT_A_DIRECTORY;
+ }
+ ret = SMB_VFS_UNLINK(conn,directory);
+ } else {
+ ret = SMB_VFS_RMDIR(conn,directory);
+ }
+ if (ret == 0) {
+ notify_fname(conn, NOTIFY_ACTION_REMOVED,
+ FILE_NOTIFY_CHANGE_DIR_NAME,
+ directory);
+ return NT_STATUS_OK;
+ }
+
+ if(((errno == ENOTEMPTY)||(errno == EEXIST)) && lp_veto_files(SNUM(conn))) {
+ /*
+ * Check to see if the only thing in this directory are
+ * vetoed files/directories. If so then delete them and
+ * retry. If we fail to delete any of them (and we *don't*
+ * do a recursive delete) then fail the rmdir.
+ */
+ const char *dname;
+ long dirpos = 0;
+ struct smb_Dir *dir_hnd = OpenDir(talloc_tos(), conn,
+ directory, NULL, 0);
+
+ if(dir_hnd == NULL) {
+ errno = ENOTEMPTY;
+ goto err;
+ }
+
+ while ((dname = ReadDirName(dir_hnd,&dirpos))) {
+ if((strcmp(dname, ".") == 0) || (strcmp(dname, "..")==0))
+ continue;
+ if (!is_visible_file(conn, directory, dname, &st, False))
+ continue;
+ if(!IS_VETO_PATH(conn, dname)) {
+ TALLOC_FREE(dir_hnd);
+ errno = ENOTEMPTY;
+ goto err;
+ }
+ }
+
+ /* We only have veto files/directories. Recursive delete. */
+
+ RewindDir(dir_hnd,&dirpos);
+ while ((dname = ReadDirName(dir_hnd,&dirpos))) {
+ char *fullname = NULL;
+
+ if (ISDOT(dname) || ISDOTDOT(dname)) {
+ continue;
+ }
+ if (!is_visible_file(conn, directory, dname, &st, False)) {
+ continue;
+ }
+
+ fullname = talloc_asprintf(ctx,
+ "%s/%s",
+ directory,
+ dname);
+
+ if(!fullname) {
+ errno = ENOMEM;
+ break;
+ }
+
+ if(SMB_VFS_LSTAT(conn,fullname, &st) != 0) {
+ break;
+ }
+ if(st.st_mode & S_IFDIR) {
+ if(lp_recursive_veto_delete(SNUM(conn))) {
+ if(!recursive_rmdir(ctx, conn, fullname))
+ break;
+ }
+ if(SMB_VFS_RMDIR(conn,fullname) != 0) {
+ break;
+ }
+ } else if(SMB_VFS_UNLINK(conn,fullname) != 0) {
+ break;
+ }
+ TALLOC_FREE(fullname);
+ }
+ TALLOC_FREE(dir_hnd);
+ /* Retry the rmdir */
+ ret = SMB_VFS_RMDIR(conn,directory);
+ }
+
+ err:
+
+ if (ret != 0) {
+ DEBUG(3,("rmdir_internals: couldn't remove directory %s : "
+ "%s\n", directory,strerror(errno)));
+ return map_nt_error_from_unix(errno);
+ }
+
+ notify_fname(conn, NOTIFY_ACTION_REMOVED,
+ FILE_NOTIFY_CHANGE_DIR_NAME,
+ directory);
+
+ return NT_STATUS_OK;
+}
+
+/****************************************************************************
+ Reply to a rmdir.
+****************************************************************************/
+
+void reply_rmdir(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *directory = NULL;
+ SMB_STRUCT_STAT sbuf;
+ NTSTATUS status;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBrmdir);
+
+ srvstr_get_path(ctx, (char *)req->inbuf, req->flags2, &directory,
+ smb_buf(req->inbuf) + 1, 0,
+ STR_TERMINATE, &status);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBrmdir);
+ return;
+ }
+
+ status = resolve_dfspath(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ directory,
+ &directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBrmdir);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBrmdir);
+ return;
+ }
+
+ status = unix_convert(ctx, conn, directory, False, &directory,
+ NULL, &sbuf);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBrmdir);
+ return;
+ }
+
+ status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBrmdir);
+ return;
+ }
+
+ dptr_closepath(directory, req->smbpid);
+ status = rmdir_internals(ctx, conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBrmdir);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ DEBUG( 3, ( "rmdir %s\n", directory ) );
+
+ END_PROFILE(SMBrmdir);
+ return;
+}
+
+/*******************************************************************
+ Resolve wildcards in a filename rename.
+********************************************************************/
+
+static bool resolve_wildcards(TALLOC_CTX *ctx,
+ const char *name1,
+ const char *name2,
+ char **pp_newname)
+{
+ char *name2_copy = NULL;
+ char *root1 = NULL;
+ char *root2 = NULL;
+ char *ext1 = NULL;
+ char *ext2 = NULL;
+ char *p,*p2, *pname1, *pname2;
+
+ name2_copy = talloc_strdup(ctx, name2);
+ if (!name2_copy) {
+ return False;
+ }
+
+ pname1 = strrchr_m(name1,'/');
+ pname2 = strrchr_m(name2_copy,'/');
+
+ if (!pname1 || !pname2) {
+ return False;
+ }
+
+ /* Truncate the copy of name2 at the last '/' */
+ *pname2 = '\0';
+
+ /* Now go past the '/' */
+ pname1++;
+ pname2++;
+
+ root1 = talloc_strdup(ctx, pname1);
+ root2 = talloc_strdup(ctx, pname2);
+
+ if (!root1 || !root2) {
+ return False;
+ }
+
+ p = strrchr_m(root1,'.');
+ if (p) {
+ *p = 0;
+ ext1 = talloc_strdup(ctx, p+1);
+ } else {
+ ext1 = talloc_strdup(ctx, "");
+ }
+ p = strrchr_m(root2,'.');
+ if (p) {
+ *p = 0;
+ ext2 = talloc_strdup(ctx, p+1);
+ } else {
+ ext2 = talloc_strdup(ctx, "");
+ }
+
+ if (!ext1 || !ext2) {
+ return False;
+ }
+
+ p = root1;
+ p2 = root2;
+ while (*p2) {
+ if (*p2 == '?') {
+ /* Hmmm. Should this be mb-aware ? */
+ *p2 = *p;
+ p2++;
+ } else if (*p2 == '*') {
+ *p2 = '\0';
+ root2 = talloc_asprintf(ctx, "%s%s",
+ root2,
+ p);
+ if (!root2) {
+ return False;
+ }
+ break;
+ } else {
+ p2++;
+ }
+ if (*p) {
+ p++;
+ }
+ }
+
+ p = ext1;
+ p2 = ext2;
+ while (*p2) {
+ if (*p2 == '?') {
+ /* Hmmm. Should this be mb-aware ? */
+ *p2 = *p;
+ p2++;
+ } else if (*p2 == '*') {
+ *p2 = '\0';
+ ext2 = talloc_asprintf(ctx, "%s%s",
+ ext2,
+ p);
+ if (!ext2) {
+ return False;
+ }
+ break;
+ } else {
+ p2++;
+ }
+ if (*p) {
+ p++;
+ }
+ }
+
+ if (*ext2) {
+ *pp_newname = talloc_asprintf(ctx, "%s/%s.%s",
+ name2_copy,
+ root2,
+ ext2);
+ } else {
+ *pp_newname = talloc_asprintf(ctx, "%s/%s",
+ name2_copy,
+ root2);
+ }
+
+ if (!*pp_newname) {
+ return False;
+ }
+
+ return True;
+}
+
+/****************************************************************************
+ Ensure open files have their names updated. Updated to notify other smbd's
+ asynchronously.
+****************************************************************************/
+
+static void rename_open_files(connection_struct *conn,
+ struct share_mode_lock *lck,
+ const char *newname)
+{
+ files_struct *fsp;
+ bool did_rename = False;
+
+ for(fsp = file_find_di_first(lck->id); fsp;
+ fsp = file_find_di_next(fsp)) {
+ /* fsp_name is a relative path under the fsp. To change this for other
+ sharepaths we need to manipulate relative paths. */
+ /* TODO - create the absolute path and manipulate the newname
+ relative to the sharepath. */
+ if (!strequal(fsp->conn->connectpath, conn->connectpath)) {
+ continue;
+ }
+ DEBUG(10,("rename_open_files: renaming file fnum %d (file_id %s) from %s -> %s\n",
+ fsp->fnum, file_id_string_tos(&fsp->file_id),
+ fsp->fsp_name, newname ));
+ string_set(&fsp->fsp_name, newname);
+ did_rename = True;
+ }
+
+ if (!did_rename) {
+ DEBUG(10,("rename_open_files: no open files on file_id %s for %s\n",
+ file_id_string_tos(&lck->id), newname ));
+ }
+
+ /* Send messages to all smbd's (not ourself) that the name has changed. */
+ rename_share_filename(smbd_messaging_context(), lck, conn->connectpath,
+ newname);
+}
+
+/****************************************************************************
+ We need to check if the source path is a parent directory of the destination
+ (ie. a rename of /foo/bar/baz -> /foo/bar/baz/bibble/bobble. If so we must
+ refuse the rename with a sharing violation. Under UNIX the above call can
+ *succeed* if /foo/bar/baz is a symlink to another area in the share. We
+ probably need to check that the client is a Windows one before disallowing
+ this as a UNIX client (one with UNIX extensions) can know the source is a
+ symlink and make this decision intelligently. Found by an excellent bug
+ report from <AndyLiebman@aol.com>.
+****************************************************************************/
+
+static bool rename_path_prefix_equal(const char *src, const char *dest)
+{
+ const char *psrc = src;
+ const char *pdst = dest;
+ size_t slen;
+
+ if (psrc[0] == '.' && psrc[1] == '/') {
+ psrc += 2;
+ }
+ if (pdst[0] == '.' && pdst[1] == '/') {
+ pdst += 2;
+ }
+ if ((slen = strlen(psrc)) > strlen(pdst)) {
+ return False;
+ }
+ return ((memcmp(psrc, pdst, slen) == 0) && pdst[slen] == '/');
+}
+
+/*
+ * Do the notify calls from a rename
+ */
+
+static void notify_rename(connection_struct *conn, bool is_dir,
+ const char *oldpath, const char *newpath)
+{
+ char *olddir, *newdir;
+ const char *oldname, *newname;
+ uint32 mask;
+
+ mask = is_dir ? FILE_NOTIFY_CHANGE_DIR_NAME
+ : FILE_NOTIFY_CHANGE_FILE_NAME;
+
+ if (!parent_dirname_talloc(NULL, oldpath, &olddir, &oldname)
+ || !parent_dirname_talloc(NULL, newpath, &newdir, &newname)) {
+ TALLOC_FREE(olddir);
+ return;
+ }
+
+ if (strcmp(olddir, newdir) == 0) {
+ notify_fname(conn, NOTIFY_ACTION_OLD_NAME, mask, oldpath);
+ notify_fname(conn, NOTIFY_ACTION_NEW_NAME, mask, newpath);
+ }
+ else {
+ notify_fname(conn, NOTIFY_ACTION_REMOVED, mask, oldpath);
+ notify_fname(conn, NOTIFY_ACTION_ADDED, mask, newpath);
+ }
+ TALLOC_FREE(olddir);
+ TALLOC_FREE(newdir);
+
+ /* this is a strange one. w2k3 gives an additional event for
+ CHANGE_ATTRIBUTES and CHANGE_CREATION on the new file when renaming
+ files, but not directories */
+ if (!is_dir) {
+ notify_fname(conn, NOTIFY_ACTION_MODIFIED,
+ FILE_NOTIFY_CHANGE_ATTRIBUTES
+ |FILE_NOTIFY_CHANGE_CREATION,
+ newpath);
+ }
+}
+
+/****************************************************************************
+ Rename an open file - given an fsp.
+****************************************************************************/
+
+NTSTATUS rename_internals_fsp(connection_struct *conn,
+ files_struct *fsp,
+ char *newname,
+ const char *newname_last_component,
+ uint32 attrs,
+ bool replace_if_exists)
+{
+ TALLOC_CTX *ctx = talloc_tos();
+ SMB_STRUCT_STAT sbuf, sbuf1;
+ NTSTATUS status = NT_STATUS_OK;
+ struct share_mode_lock *lck = NULL;
+ bool dst_exists;
+
+ ZERO_STRUCT(sbuf);
+
+ status = check_name(conn, newname);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ /* Ensure newname contains a '/' */
+ if(strrchr_m(newname,'/') == 0) {
+ newname = talloc_asprintf(ctx,
+ "./%s",
+ newname);
+ if (!newname) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ }
+
+ /*
+ * Check for special case with case preserving and not
+ * case sensitive. If the old last component differs from the original
+ * last component only by case, then we should allow
+ * the rename (user is trying to change the case of the
+ * filename).
+ */
+
+ if((conn->case_sensitive == False) && (conn->case_preserve == True) &&
+ strequal(newname, fsp->fsp_name)) {
+ char *p;
+ char *newname_modified_last_component = NULL;
+
+ /*
+ * Get the last component of the modified name.
+ * Note that we guarantee that newname contains a '/'
+ * character above.
+ */
+ p = strrchr_m(newname,'/');
+ newname_modified_last_component = talloc_strdup(ctx,
+ p+1);
+ if (!newname_modified_last_component) {
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ if(strcsequal(newname_modified_last_component,
+ newname_last_component) == False) {
+ /*
+ * Replace the modified last component with
+ * the original.
+ */
+ *p = '\0'; /* Truncate at the '/' */
+ newname = talloc_asprintf(ctx,
+ "%s/%s",
+ newname,
+ newname_last_component);
+ }
+ }
+
+ /*
+ * If the src and dest names are identical - including case,
+ * don't do the rename, just return success.
+ */
+
+ if (strcsequal(fsp->fsp_name, newname)) {
+ DEBUG(3,("rename_internals_fsp: identical names in rename %s - returning success\n",
+ newname));
+ return NT_STATUS_OK;
+ }
+
+ /*
+ * Have vfs_object_exist also fill sbuf1
+ */
+ dst_exists = vfs_object_exist(conn, newname, &sbuf1);
+
+ if(!replace_if_exists && dst_exists) {
+ DEBUG(3,("rename_internals_fsp: dest exists doing rename %s -> %s\n",
+ fsp->fsp_name,newname));
+ return NT_STATUS_OBJECT_NAME_COLLISION;
+ }
+
+ if (dst_exists) {
+ struct file_id fileid = vfs_file_id_from_sbuf(conn, &sbuf1);
+ files_struct *dst_fsp = file_find_di_first(fileid);
+ if (dst_fsp) {
+ DEBUG(3, ("rename_internals_fsp: Target file open\n"));
+ return NT_STATUS_ACCESS_DENIED;
+ }
+ }
+
+ /* Ensure we have a valid stat struct for the source. */
+ if (fsp->fh->fd != -1) {
+ if (SMB_VFS_FSTAT(fsp, &sbuf) == -1) {
+ return map_nt_error_from_unix(errno);
+ }
+ } else {
+ if (SMB_VFS_STAT(conn,fsp->fsp_name,&sbuf) == -1) {
+ return map_nt_error_from_unix(errno);
+ }
+ }
+
+ status = can_rename(conn, fsp, attrs, &sbuf);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3,("rename_internals_fsp: Error %s rename %s -> %s\n",
+ nt_errstr(status), fsp->fsp_name,newname));
+ if (NT_STATUS_EQUAL(status,NT_STATUS_SHARING_VIOLATION))
+ status = NT_STATUS_ACCESS_DENIED;
+ return status;
+ }
+
+ if (rename_path_prefix_equal(fsp->fsp_name, newname)) {
+ return NT_STATUS_ACCESS_DENIED;
+ }
+
+ lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
+ NULL);
+
+ /*
+ * We have the file open ourselves, so not being able to get the
+ * corresponding share mode lock is a fatal error.
+ */
+
+ SMB_ASSERT(lck != NULL);
+
+ if(SMB_VFS_RENAME(conn,fsp->fsp_name, newname) == 0) {
+ uint32 create_options = fsp->fh->private_options;
+
+ DEBUG(3,("rename_internals_fsp: succeeded doing rename on %s -> %s\n",
+ fsp->fsp_name,newname));
+
+ rename_open_files(conn, lck, newname);
+
+ notify_rename(conn, fsp->is_directory, fsp->fsp_name, newname);
+
+ /*
+ * A rename acts as a new file create w.r.t. allowing an initial delete
+ * on close, probably because in Windows there is a new handle to the
+ * new file. If initial delete on close was requested but not
+ * originally set, we need to set it here. This is probably not 100% correct,
+ * but will work for the CIFSFS client which in non-posix mode
+ * depends on these semantics. JRA.
+ */
+
+ set_allow_initial_delete_on_close(lck, fsp, True);
+
+ if (create_options & FILE_DELETE_ON_CLOSE) {
+ status = can_set_delete_on_close(fsp, True, 0);
+
+ 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);
+ return NT_STATUS_OK;
+ }
+
+ TALLOC_FREE(lck);
+
+ if (errno == ENOTDIR || errno == EISDIR) {
+ status = NT_STATUS_OBJECT_NAME_COLLISION;
+ } else {
+ status = map_nt_error_from_unix(errno);
+ }
+
+ DEBUG(3,("rename_internals_fsp: Error %s rename %s -> %s\n",
+ nt_errstr(status), fsp->fsp_name,newname));
+
+ return status;
+}
+
+/****************************************************************************
+ The guts of the rename command, split out so it may be called by the NT SMB
+ code.
+****************************************************************************/
+
+NTSTATUS rename_internals(TALLOC_CTX *ctx,
+ connection_struct *conn,
+ struct smb_request *req,
+ const char *name_in,
+ const char *newname_in,
+ uint32 attrs,
+ bool replace_if_exists,
+ bool src_has_wild,
+ bool dest_has_wild,
+ uint32_t access_mask)
+{
+ char *directory = NULL;
+ char *mask = NULL;
+ char *last_component_src = NULL;
+ char *last_component_dest = NULL;
+ char *name = NULL;
+ char *newname = NULL;
+ char *p;
+ int count=0;
+ NTSTATUS status = NT_STATUS_OK;
+ SMB_STRUCT_STAT sbuf1, sbuf2;
+ struct smb_Dir *dir_hnd = NULL;
+ const char *dname;
+ long offset = 0;
+
+ ZERO_STRUCT(sbuf1);
+ ZERO_STRUCT(sbuf2);
+
+ status = unix_convert(ctx, conn, name_in, src_has_wild, &name,
+ &last_component_src, &sbuf1);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ status = unix_convert(ctx, conn, newname_in, dest_has_wild, &newname,
+ &last_component_dest, &sbuf2);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ /*
+ * Split the old name into directory and last component
+ * strings. Note that unix_convert may have stripped off a
+ * leading ./ from both name and newname if the rename is
+ * at the root of the share. We need to make sure either both
+ * name and newname contain a / character or neither of them do
+ * as this is checked in resolve_wildcards().
+ */
+
+ p = strrchr_m(name,'/');
+ if (!p) {
+ directory = talloc_strdup(ctx, ".");
+ if (!directory) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ mask = name;
+ } else {
+ *p = 0;
+ directory = talloc_strdup(ctx, name);
+ if (!directory) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ mask = p+1;
+ *p = '/'; /* Replace needed for exceptional test below. */
+ }
+
+ /*
+ * We should only check the mangled cache
+ * here if unix_convert failed. This means
+ * that the path in 'mask' doesn't exist
+ * on the file system and so we need to look
+ * for a possible mangle. This patch from
+ * Tine Smukavec <valentin.smukavec@hermes.si>.
+ */
+
+ if (!VALID_STAT(sbuf1) && mangle_is_mangled(mask, conn->params)) {
+ char *new_mask = NULL;
+ mangle_lookup_name_from_8_3(ctx,
+ mask,
+ &new_mask,
+ conn->params );
+ if (new_mask) {
+ mask = new_mask;
+ }
+ }
+
+ if (!src_has_wild) {
+ files_struct *fsp;
+
+ /*
+ * No wildcards - just process the one file.
+ */
+ bool is_short_name = mangle_is_8_3(name, True, conn->params);
+
+ /* Add a terminating '/' to the directory name. */
+ directory = talloc_asprintf_append(directory,
+ "/%s",
+ mask);
+ if (!directory) {
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ /* Ensure newname contains a '/' also */
+ if(strrchr_m(newname,'/') == 0) {
+ newname = talloc_asprintf(ctx,
+ "./%s",
+ newname);
+ if (!newname) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ }
+
+ DEBUG(3, ("rename_internals: case_sensitive = %d, "
+ "case_preserve = %d, short case preserve = %d, "
+ "directory = %s, newname = %s, "
+ "last_component_dest = %s, is_8_3 = %d\n",
+ conn->case_sensitive, conn->case_preserve,
+ conn->short_case_preserve, directory,
+ newname, last_component_dest, is_short_name));
+
+ /* The dest name still may have wildcards. */
+ if (dest_has_wild) {
+ char *mod_newname = NULL;
+ if (!resolve_wildcards(ctx,
+ directory,newname,&mod_newname)) {
+ DEBUG(6, ("rename_internals: resolve_wildcards "
+ "%s %s failed\n",
+ directory,
+ newname));
+ return NT_STATUS_NO_MEMORY;
+ }
+ newname = mod_newname;
+ }
+
+ ZERO_STRUCT(sbuf1);
+ SMB_VFS_STAT(conn, directory, &sbuf1);
+
+ status = S_ISDIR(sbuf1.st_mode) ?
+ open_directory(conn, req, directory, &sbuf1,
+ access_mask,
+ FILE_SHARE_READ|FILE_SHARE_WRITE,
+ FILE_OPEN, 0, 0, NULL,
+ &fsp)
+ : open_file_ntcreate(conn, req, directory, &sbuf1,
+ access_mask,
+ FILE_SHARE_READ|FILE_SHARE_WRITE,
+ FILE_OPEN, 0, 0, 0, NULL,
+ &fsp);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3, ("Could not open rename source %s: %s\n",
+ directory, nt_errstr(status)));
+ return status;
+ }
+
+ status = rename_internals_fsp(conn, fsp, newname,
+ last_component_dest,
+ attrs, replace_if_exists);
+
+ close_file(fsp, NORMAL_CLOSE);
+
+ DEBUG(3, ("rename_internals: Error %s rename %s -> %s\n",
+ nt_errstr(status), directory,newname));
+
+ return status;
+ }
+
+ /*
+ * Wildcards - process each file that matches.
+ */
+ if (strequal(mask,"????????.???")) {
+ mask[0] = '*';
+ mask[1] = '\0';
+ }
+
+ status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ dir_hnd = OpenDir(talloc_tos(), conn, directory, mask, attrs);
+ if (dir_hnd == NULL) {
+ return map_nt_error_from_unix(errno);
+ }
+
+ status = NT_STATUS_NO_SUCH_FILE;
+ /*
+ * Was status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
+ * - gentest fix. JRA
+ */
+
+ while ((dname = ReadDirName(dir_hnd, &offset))) {
+ files_struct *fsp = NULL;
+ char *fname = NULL;
+ char *destname = NULL;
+ bool sysdir_entry = False;
+
+ /* Quick check for "." and ".." */
+ if (ISDOT(dname) || ISDOTDOT(dname)) {
+ if (attrs & aDIR) {
+ sysdir_entry = True;
+ } else {
+ continue;
+ }
+ }
+
+ if (!is_visible_file(conn, directory, dname, &sbuf1, False)) {
+ continue;
+ }
+
+ if(!mask_match(dname, mask, conn->case_sensitive)) {
+ continue;
+ }
+
+ if (sysdir_entry) {
+ status = NT_STATUS_OBJECT_NAME_INVALID;
+ break;
+ }
+
+ fname = talloc_asprintf(ctx,
+ "%s/%s",
+ directory,
+ dname);
+ if (!fname) {
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ if (!resolve_wildcards(ctx,
+ fname,newname,&destname)) {
+ DEBUG(6, ("resolve_wildcards %s %s failed\n",
+ fname, destname));
+ TALLOC_FREE(fname);
+ continue;
+ }
+ if (!destname) {
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ ZERO_STRUCT(sbuf1);
+ SMB_VFS_STAT(conn, fname, &sbuf1);
+
+ status = S_ISDIR(sbuf1.st_mode) ?
+ open_directory(conn, req, fname, &sbuf1,
+ access_mask,
+ FILE_SHARE_READ|FILE_SHARE_WRITE,
+ FILE_OPEN, 0, 0, NULL,
+ &fsp)
+ : open_file_ntcreate(conn, req, fname, &sbuf1,
+ access_mask,
+ FILE_SHARE_READ|FILE_SHARE_WRITE,
+ FILE_OPEN, 0, 0, 0, NULL,
+ &fsp);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3,("rename_internals: open_file_ntcreate "
+ "returned %s rename %s -> %s\n",
+ nt_errstr(status), directory, newname));
+ break;
+ }
+
+ status = rename_internals_fsp(conn, fsp, destname, dname,
+ attrs, replace_if_exists);
+
+ close_file(fsp, NORMAL_CLOSE);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3, ("rename_internals_fsp returned %s for "
+ "rename %s -> %s\n", nt_errstr(status),
+ directory, newname));
+ break;
+ }
+
+ count++;
+
+ DEBUG(3,("rename_internals: doing rename on %s -> "
+ "%s\n",fname,destname));
+
+ TALLOC_FREE(fname);
+ TALLOC_FREE(destname);
+ }
+ TALLOC_FREE(dir_hnd);
+
+ if (count == 0 && NT_STATUS_IS_OK(status) && errno != 0) {
+ status = map_nt_error_from_unix(errno);
+ }
+
+ return status;
+}
+
+/****************************************************************************
+ Reply to a mv.
+****************************************************************************/
+
+void reply_mv(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *name = NULL;
+ char *newname = NULL;
+ char *p;
+ uint32 attrs;
+ NTSTATUS status;
+ bool src_has_wcard = False;
+ bool dest_has_wcard = False;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBmv);
+
+ if (req->wct < 1) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBmv);
+ return;
+ }
+
+ attrs = SVAL(req->inbuf,smb_vwv0);
+
+ p = smb_buf(req->inbuf) + 1;
+ p += srvstr_get_path_wcard(ctx, (char *)req->inbuf, req->flags2, &name, p,
+ 0, STR_TERMINATE, &status,
+ &src_has_wcard);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBmv);
+ return;
+ }
+ p++;
+ p += srvstr_get_path_wcard(ctx, (char *)req->inbuf, req->flags2, &newname, p,
+ 0, STR_TERMINATE, &status,
+ &dest_has_wcard);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBmv);
+ return;
+ }
+
+ status = resolve_dfspath_wcard(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ name,
+ &name,
+ &src_has_wcard);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBmv);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBmv);
+ return;
+ }
+
+ status = resolve_dfspath_wcard(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ newname,
+ &newname,
+ &dest_has_wcard);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBmv);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBmv);
+ return;
+ }
+
+ DEBUG(3,("reply_mv : %s -> %s\n",name,newname));
+
+ status = rename_internals(ctx, conn, req, name, newname, attrs, False,
+ src_has_wcard, dest_has_wcard, DELETE_ACCESS);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (open_was_deferred(req->mid)) {
+ /* We have re-scheduled this call. */
+ END_PROFILE(SMBmv);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBmv);
+ return;
+ }
+
+ reply_outbuf(req, 0, 0);
+
+ END_PROFILE(SMBmv);
+ return;
+}
+
+/*******************************************************************
+ Copy a file as part of a reply_copy.
+******************************************************************/
+
+/*
+ * TODO: check error codes on all callers
+ */
+
+NTSTATUS copy_file(TALLOC_CTX *ctx,
+ connection_struct *conn,
+ const char *src,
+ const char *dest1,
+ int ofun,
+ int count,
+ bool target_is_directory)
+{
+ SMB_STRUCT_STAT src_sbuf, sbuf2;
+ SMB_OFF_T ret=-1;
+ files_struct *fsp1,*fsp2;
+ char *dest = NULL;
+ uint32 dosattrs;
+ uint32 new_create_disposition;
+ NTSTATUS status;
+
+ dest = talloc_strdup(ctx, dest1);
+ if (!dest) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ if (target_is_directory) {
+ const char *p = strrchr_m(src,'/');
+ if (p) {
+ p++;
+ } else {
+ p = src;
+ }
+ dest = talloc_asprintf_append(dest,
+ "/%s",
+ p);
+ if (!dest) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ }
+
+ if (!vfs_file_exist(conn,src,&src_sbuf)) {
+ TALLOC_FREE(dest);
+ return NT_STATUS_OBJECT_NAME_NOT_FOUND;
+ }
+
+ if (!target_is_directory && count) {
+ new_create_disposition = FILE_OPEN;
+ } else {
+ if (!map_open_params_to_ntcreate(dest1,0,ofun,
+ NULL, NULL, &new_create_disposition, NULL)) {
+ TALLOC_FREE(dest);
+ return NT_STATUS_INVALID_PARAMETER;
+ }
+ }
+
+ status = open_file_ntcreate(conn, NULL, src, &src_sbuf,
+ FILE_GENERIC_READ,
+ FILE_SHARE_READ|FILE_SHARE_WRITE,
+ FILE_OPEN,
+ 0,
+ FILE_ATTRIBUTE_NORMAL,
+ INTERNAL_OPEN_ONLY,
+ NULL, &fsp1);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ TALLOC_FREE(dest);
+ return status;
+ }
+
+ dosattrs = dos_mode(conn, src, &src_sbuf);
+ if (SMB_VFS_STAT(conn,dest,&sbuf2) == -1) {
+ ZERO_STRUCTP(&sbuf2);
+ }
+
+ status = open_file_ntcreate(conn, NULL, dest, &sbuf2,
+ FILE_GENERIC_WRITE,
+ FILE_SHARE_READ|FILE_SHARE_WRITE,
+ new_create_disposition,
+ 0,
+ dosattrs,
+ INTERNAL_OPEN_ONLY,
+ NULL, &fsp2);
+
+ TALLOC_FREE(dest);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ close_file(fsp1,ERROR_CLOSE);
+ return status;
+ }
+
+ if ((ofun&3) == 1) {
+ if(SMB_VFS_LSEEK(fsp2,0,SEEK_END) == -1) {
+ DEBUG(0,("copy_file: error - vfs lseek returned error %s\n", strerror(errno) ));
+ /*
+ * Stop the copy from occurring.
+ */
+ ret = -1;
+ src_sbuf.st_size = 0;
+ }
+ }
+
+ if (src_sbuf.st_size) {
+ ret = vfs_transfer_file(fsp1, fsp2, src_sbuf.st_size);
+ }
+
+ close_file(fsp1,NORMAL_CLOSE);
+
+ /* Ensure the modtime is set correctly on the destination file. */
+ set_close_write_time(fsp2, get_mtimespec(&src_sbuf));
+
+ /*
+ * As we are opening fsp1 read-only we only expect
+ * an error on close on fsp2 if we are out of space.
+ * Thus we don't look at the error return from the
+ * close of fsp1.
+ */
+ status = close_file(fsp2,NORMAL_CLOSE);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ if (ret != (SMB_OFF_T)src_sbuf.st_size) {
+ return NT_STATUS_DISK_FULL;
+ }
+
+ return NT_STATUS_OK;
+}
+
+/****************************************************************************
+ Reply to a file copy.
+****************************************************************************/
+
+void reply_copy(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ char *name = NULL;
+ char *newname = NULL;
+ char *directory = NULL;
+ char *mask = NULL;
+ char *p;
+ int count=0;
+ int error = ERRnoaccess;
+ int err = 0;
+ int tid2;
+ int ofun;
+ int flags;
+ bool target_is_directory=False;
+ bool source_has_wild = False;
+ bool dest_has_wild = False;
+ SMB_STRUCT_STAT sbuf1, sbuf2;
+ NTSTATUS status;
+ TALLOC_CTX *ctx = talloc_tos();
+
+ START_PROFILE(SMBcopy);
+
+ if (req->wct < 3) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ tid2 = SVAL(req->inbuf,smb_vwv0);
+ ofun = SVAL(req->inbuf,smb_vwv1);
+ flags = SVAL(req->inbuf,smb_vwv2);
+
+ p = smb_buf(req->inbuf);
+ p += srvstr_get_path_wcard(ctx, (char *)req->inbuf, req->flags2, &name, p,
+ 0, STR_TERMINATE, &status,
+ &source_has_wild);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+ p += srvstr_get_path_wcard(ctx, (char *)req->inbuf, req->flags2, &newname, p,
+ 0, STR_TERMINATE, &status,
+ &dest_has_wild);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ DEBUG(3,("reply_copy : %s -> %s\n",name,newname));
+
+ if (tid2 != conn->cnum) {
+ /* can't currently handle inter share copies XXXX */
+ DEBUG(3,("Rejecting inter-share copy\n"));
+ reply_doserror(req, ERRSRV, ERRinvdevice);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = resolve_dfspath_wcard(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ name,
+ &name,
+ &source_has_wild);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = resolve_dfspath_wcard(ctx, conn,
+ req->flags2 & FLAGS2_DFS_PATHNAMES,
+ newname,
+ &newname,
+ &dest_has_wild);
+ if (!NT_STATUS_IS_OK(status)) {
+ if (NT_STATUS_EQUAL(status,NT_STATUS_PATH_NOT_COVERED)) {
+ reply_botherror(req, NT_STATUS_PATH_NOT_COVERED,
+ ERRSRV, ERRbadpath);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = unix_convert(ctx, conn, name, source_has_wild,
+ &name, NULL, &sbuf1);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = unix_convert(ctx, conn, newname, dest_has_wild,
+ &newname, NULL, &sbuf2);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ target_is_directory = VALID_STAT_OF_DIR(sbuf2);
+
+ if ((flags&1) && target_is_directory) {
+ reply_doserror(req, ERRDOS, ERRbadfile);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ if ((flags&2) && !target_is_directory) {
+ reply_doserror(req, ERRDOS, ERRbadpath);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ if ((flags&(1<<5)) && VALID_STAT_OF_DIR(sbuf1)) {
+ /* wants a tree copy! XXXX */
+ DEBUG(3,("Rejecting tree copy\n"));
+ reply_doserror(req, ERRSRV, ERRerror);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ p = strrchr_m(name,'/');
+ if (!p) {
+ directory = talloc_strdup(ctx, "./");
+ if (!directory) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+ mask = name;
+ } else {
+ *p = 0;
+ directory = talloc_strdup(ctx, name);
+ if (!directory) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+ mask = p+1;
+ }
+
+ /*
+ * We should only check the mangled cache
+ * here if unix_convert failed. This means
+ * that the path in 'mask' doesn't exist
+ * on the file system and so we need to look
+ * for a possible mangle. This patch from
+ * Tine Smukavec <valentin.smukavec@hermes.si>.
+ */
+
+ if (!VALID_STAT(sbuf1) && mangle_is_mangled(mask, conn->params)) {
+ char *new_mask = NULL;
+ mangle_lookup_name_from_8_3(ctx,
+ mask,
+ &new_mask,
+ conn->params );
+ if (new_mask) {
+ mask = new_mask;
+ }
+ }
+
+ if (!source_has_wild) {
+ directory = talloc_asprintf_append(directory,
+ "/%s",
+ mask);
+ if (dest_has_wild) {
+ char *mod_newname = NULL;
+ if (!resolve_wildcards(ctx,
+ directory,newname,&mod_newname)) {
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+ newname = mod_newname;
+ }
+
+ status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = check_name(conn, newname);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = copy_file(ctx,conn,directory,newname,ofun,
+ count,target_is_directory);
+
+ if(!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ } else {
+ count++;
+ }
+ } else {
+ struct smb_Dir *dir_hnd = NULL;
+ const char *dname = NULL;
+ long offset = 0;
+
+ if (strequal(mask,"????????.???")) {
+ mask[0] = '*';
+ mask[1] = '\0';
+ }
+
+ status = check_name(conn, directory);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ dir_hnd = OpenDir(talloc_tos(), conn, directory, mask, 0);
+ if (dir_hnd == NULL) {
+ status = map_nt_error_from_unix(errno);
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ error = ERRbadfile;
+
+ while ((dname = ReadDirName(dir_hnd, &offset))) {
+ char *destname = NULL;
+ char *fname = NULL;
+
+ if (ISDOT(dname) || ISDOTDOT(dname)) {
+ continue;
+ }
+
+ if (!is_visible_file(conn, directory, dname, &sbuf1, False)) {
+ continue;
+ }
+
+ if(!mask_match(dname, mask, conn->case_sensitive)) {
+ continue;
+ }
+
+ error = ERRnoaccess;
+ fname = talloc_asprintf(ctx,
+ "%s/%s",
+ directory,
+ dname);
+ if (!fname) {
+ TALLOC_FREE(dir_hnd);
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ if (!resolve_wildcards(ctx,
+ fname,newname,&destname)) {
+ continue;
+ }
+ if (!destname) {
+ TALLOC_FREE(dir_hnd);
+ reply_nterror(req, NT_STATUS_NO_MEMORY);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = check_name(conn, fname);
+ if (!NT_STATUS_IS_OK(status)) {
+ TALLOC_FREE(dir_hnd);
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ status = check_name(conn, destname);
+ if (!NT_STATUS_IS_OK(status)) {
+ TALLOC_FREE(dir_hnd);
+ reply_nterror(req, status);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ DEBUG(3,("reply_copy : doing copy on %s -> %s\n",fname, destname));
+
+ status = copy_file(ctx,conn,fname,destname,ofun,
+ count,target_is_directory);
+ if (NT_STATUS_IS_OK(status)) {
+ count++;
+ }
+ TALLOC_FREE(fname);
+ TALLOC_FREE(destname);
+ }
+ TALLOC_FREE(dir_hnd);
+ }
+
+ if (count == 0) {
+ if(err) {
+ /* Error on close... */
+ errno = err;
+ reply_unixerror(req, ERRHRD, ERRgeneral);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ reply_doserror(req, ERRDOS, error);
+ END_PROFILE(SMBcopy);
+ return;
+ }
+
+ reply_outbuf(req, 1, 0);
+ SSVAL(req->outbuf,smb_vwv0,count);
+
+ END_PROFILE(SMBcopy);
+ return;
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_LOCKING
+
+/****************************************************************************
+ Get a lock pid, dealing with large count requests.
+****************************************************************************/
+
+uint32 get_lock_pid( char *data, int data_offset, bool large_file_format)
+{
+ if(!large_file_format)
+ return (uint32)SVAL(data,SMB_LPID_OFFSET(data_offset));
+ else
+ return (uint32)SVAL(data,SMB_LARGE_LPID_OFFSET(data_offset));
+}
+
+/****************************************************************************
+ Get a lock count, dealing with large count requests.
+****************************************************************************/
+
+SMB_BIG_UINT get_lock_count( char *data, int data_offset, bool large_file_format)
+{
+ SMB_BIG_UINT count = 0;
+
+ if(!large_file_format) {
+ count = (SMB_BIG_UINT)IVAL(data,SMB_LKLEN_OFFSET(data_offset));
+ } else {
+
+#if defined(HAVE_LONGLONG)
+ count = (((SMB_BIG_UINT) IVAL(data,SMB_LARGE_LKLEN_OFFSET_HIGH(data_offset))) << 32) |
+ ((SMB_BIG_UINT) IVAL(data,SMB_LARGE_LKLEN_OFFSET_LOW(data_offset)));
+#else /* HAVE_LONGLONG */
+
+ /*
+ * NT4.x seems to be broken in that it sends large file (64 bit)
+ * lockingX calls even if the CAP_LARGE_FILES was *not*
+ * negotiated. For boxes without large unsigned ints truncate the
+ * lock count by dropping the top 32 bits.
+ */
+
+ if(IVAL(data,SMB_LARGE_LKLEN_OFFSET_HIGH(data_offset)) != 0) {
+ DEBUG(3,("get_lock_count: truncating lock count (high)0x%x (low)0x%x to just low count.\n",
+ (unsigned int)IVAL(data,SMB_LARGE_LKLEN_OFFSET_HIGH(data_offset)),
+ (unsigned int)IVAL(data,SMB_LARGE_LKLEN_OFFSET_LOW(data_offset)) ));
+ SIVAL(data,SMB_LARGE_LKLEN_OFFSET_HIGH(data_offset),0);
+ }
+
+ count = (SMB_BIG_UINT)IVAL(data,SMB_LARGE_LKLEN_OFFSET_LOW(data_offset));
+#endif /* HAVE_LONGLONG */
+ }
+
+ return count;
+}
+
+#if !defined(HAVE_LONGLONG)
+/****************************************************************************
+ Pathetically try and map a 64 bit lock offset into 31 bits. I hate Windows :-).
+****************************************************************************/
+
+static uint32 map_lock_offset(uint32 high, uint32 low)
+{
+ unsigned int i;
+ uint32 mask = 0;
+ uint32 highcopy = high;
+
+ /*
+ * Try and find out how many significant bits there are in high.
+ */
+
+ for(i = 0; highcopy; i++)
+ highcopy >>= 1;
+
+ /*
+ * We use 31 bits not 32 here as POSIX
+ * lock offsets may not be negative.
+ */
+
+ mask = (~0) << (31 - i);
+
+ if(low & mask)
+ return 0; /* Fail. */
+
+ high <<= (31 - i);
+
+ return (high|low);
+}
+#endif /* !defined(HAVE_LONGLONG) */
+
+/****************************************************************************
+ Get a lock offset, dealing with large offset requests.
+****************************************************************************/
+
+SMB_BIG_UINT get_lock_offset( char *data, int data_offset, bool large_file_format, bool *err)
+{
+ SMB_BIG_UINT offset = 0;
+
+ *err = False;
+
+ if(!large_file_format) {
+ offset = (SMB_BIG_UINT)IVAL(data,SMB_LKOFF_OFFSET(data_offset));
+ } else {
+
+#if defined(HAVE_LONGLONG)
+ offset = (((SMB_BIG_UINT) IVAL(data,SMB_LARGE_LKOFF_OFFSET_HIGH(data_offset))) << 32) |
+ ((SMB_BIG_UINT) IVAL(data,SMB_LARGE_LKOFF_OFFSET_LOW(data_offset)));
+#else /* HAVE_LONGLONG */
+
+ /*
+ * NT4.x seems to be broken in that it sends large file (64 bit)
+ * lockingX calls even if the CAP_LARGE_FILES was *not*
+ * negotiated. For boxes without large unsigned ints mangle the
+ * lock offset by mapping the top 32 bits onto the lower 32.
+ */
+
+ if(IVAL(data,SMB_LARGE_LKOFF_OFFSET_HIGH(data_offset)) != 0) {
+ uint32 low = IVAL(data,SMB_LARGE_LKOFF_OFFSET_LOW(data_offset));
+ uint32 high = IVAL(data,SMB_LARGE_LKOFF_OFFSET_HIGH(data_offset));
+ uint32 new_low = 0;
+
+ if((new_low = map_lock_offset(high, low)) == 0) {
+ *err = True;
+ return (SMB_BIG_UINT)-1;
+ }
+
+ DEBUG(3,("get_lock_offset: truncating lock offset (high)0x%x (low)0x%x to offset 0x%x.\n",
+ (unsigned int)high, (unsigned int)low, (unsigned int)new_low ));
+ SIVAL(data,SMB_LARGE_LKOFF_OFFSET_HIGH(data_offset),0);
+ SIVAL(data,SMB_LARGE_LKOFF_OFFSET_LOW(data_offset),new_low);
+ }
+
+ offset = (SMB_BIG_UINT)IVAL(data,SMB_LARGE_LKOFF_OFFSET_LOW(data_offset));
+#endif /* HAVE_LONGLONG */
+ }
+
+ return offset;
+}
+
+/****************************************************************************
+ Reply to a lockingX request.
+****************************************************************************/
+
+void reply_lockingX(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ files_struct *fsp;
+ unsigned char locktype;
+ unsigned char oplocklevel;
+ uint16 num_ulocks;
+ uint16 num_locks;
+ SMB_BIG_UINT count = 0, offset = 0;
+ uint32 lock_pid;
+ int32 lock_timeout;
+ int i;
+ char *data;
+ bool large_file_format;
+ bool err;
+ NTSTATUS status = NT_STATUS_UNSUCCESSFUL;
+
+ START_PROFILE(SMBlockingX);
+
+ if (req->wct < 8) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBlockingX);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv2));
+ locktype = CVAL(req->inbuf,smb_vwv3);
+ oplocklevel = CVAL(req->inbuf,smb_vwv3+1);
+ num_ulocks = SVAL(req->inbuf,smb_vwv6);
+ num_locks = SVAL(req->inbuf,smb_vwv7);
+ lock_timeout = IVAL(req->inbuf,smb_vwv4);
+ large_file_format = (locktype & LOCKING_ANDX_LARGE_FILES)?True:False;
+
+ if (!check_fsp(conn, req, fsp)) {
+ END_PROFILE(SMBlockingX);
+ return;
+ }
+
+ data = smb_buf(req->inbuf);
+
+ if (locktype & LOCKING_ANDX_CHANGE_LOCKTYPE) {
+ /* we don't support these - and CANCEL_LOCK makes w2k
+ and XP reboot so I don't really want to be
+ compatible! (tridge) */
+ reply_nterror(req, NT_STATUS_DOS(ERRDOS, ERRnoatomiclocks));
+ END_PROFILE(SMBlockingX);
+ return;
+ }
+
+ /* Check if this is an oplock break on a file
+ we have granted an oplock on.
+ */
+ if ((locktype & LOCKING_ANDX_OPLOCK_RELEASE)) {
+ /* Client can insist on breaking to none. */
+ bool break_to_none = (oplocklevel == 0);
+ bool result;
+
+ DEBUG(5,("reply_lockingX: oplock break reply (%u) from client "
+ "for fnum = %d\n", (unsigned int)oplocklevel,
+ fsp->fnum ));
+
+ /*
+ * Make sure we have granted an exclusive or batch oplock on
+ * this file.
+ */
+
+ if (fsp->oplock_type == 0) {
+
+ /* The Samba4 nbench simulator doesn't understand
+ the difference between break to level2 and break
+ to none from level2 - it sends oplock break
+ replies in both cases. Don't keep logging an error
+ message here - just ignore it. JRA. */
+
+ DEBUG(5,("reply_lockingX: Error : oplock break from "
+ "client for fnum = %d (oplock=%d) and no "
+ "oplock granted on this file (%s).\n",
+ fsp->fnum, fsp->oplock_type, fsp->fsp_name));
+
+ /* if this is a pure oplock break request then don't
+ * send a reply */
+ if (num_locks == 0 && num_ulocks == 0) {
+ END_PROFILE(SMBlockingX);
+ return;
+ } else {
+ END_PROFILE(SMBlockingX);
+ reply_doserror(req, ERRDOS, ERRlock);
+ return;
+ }
+ }
+
+ if ((fsp->sent_oplock_break == BREAK_TO_NONE_SENT) ||
+ (break_to_none)) {
+ result = remove_oplock(fsp);
+ } else {
+ result = downgrade_oplock(fsp);
+ }
+
+ if (!result) {
+ DEBUG(0, ("reply_lockingX: error in removing "
+ "oplock on file %s\n", fsp->fsp_name));
+ /* Hmmm. Is this panic justified? */
+ smb_panic("internal tdb error");
+ }
+
+ reply_to_oplock_break_requests(fsp);
+
+ /* if this is a pure oplock break request then don't send a
+ * reply */
+ if (num_locks == 0 && num_ulocks == 0) {
+ /* Sanity check - ensure a pure oplock break is not a
+ chained request. */
+ if(CVAL(req->inbuf,smb_vwv0) != 0xff)
+ DEBUG(0,("reply_lockingX: Error : pure oplock "
+ "break is a chained %d request !\n",
+ (unsigned int)CVAL(req->inbuf,
+ smb_vwv0) ));
+ END_PROFILE(SMBlockingX);
+ return;
+ }
+ }
+
+ /*
+ * We do this check *after* we have checked this is not a oplock break
+ * response message. JRA.
+ */
+
+ release_level_2_oplocks_on_change(fsp);
+
+ if (smb_buflen(req->inbuf) <
+ (num_ulocks + num_locks) * (large_file_format ? 20 : 10)) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBlockingX);
+ return;
+ }
+
+ /* Data now points at the beginning of the list
+ of smb_unlkrng structs */
+ for(i = 0; i < (int)num_ulocks; i++) {
+ lock_pid = get_lock_pid( data, i, large_file_format);
+ count = get_lock_count( data, i, large_file_format);
+ offset = get_lock_offset( data, i, large_file_format, &err);
+
+ /*
+ * There is no error code marked "stupid client bug".... :-).
+ */
+ if(err) {
+ END_PROFILE(SMBlockingX);
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ return;
+ }
+
+ DEBUG(10,("reply_lockingX: unlock start=%.0f, len=%.0f for "
+ "pid %u, file %s\n", (double)offset, (double)count,
+ (unsigned int)lock_pid, fsp->fsp_name ));
+
+ status = do_unlock(smbd_messaging_context(),
+ fsp,
+ lock_pid,
+ count,
+ offset,
+ WINDOWS_LOCK);
+
+ if (NT_STATUS_V(status)) {
+ END_PROFILE(SMBlockingX);
+ reply_nterror(req, status);
+ return;
+ }
+ }
+
+ /* Setup the timeout in seconds. */
+
+ if (!lp_blocking_locks(SNUM(conn))) {
+ lock_timeout = 0;
+ }
+
+ /* Now do any requested locks */
+ data += ((large_file_format ? 20 : 10)*num_ulocks);
+
+ /* Data now points at the beginning of the list
+ of smb_lkrng structs */
+
+ for(i = 0; i < (int)num_locks; i++) {
+ enum brl_type lock_type = ((locktype & LOCKING_ANDX_SHARED_LOCK) ?
+ READ_LOCK:WRITE_LOCK);
+ lock_pid = get_lock_pid( data, i, large_file_format);
+ count = get_lock_count( data, i, large_file_format);
+ offset = get_lock_offset( data, i, large_file_format, &err);
+
+ /*
+ * There is no error code marked "stupid client bug".... :-).
+ */
+ if(err) {
+ END_PROFILE(SMBlockingX);
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ return;
+ }
+
+ DEBUG(10,("reply_lockingX: lock start=%.0f, len=%.0f for pid "
+ "%u, file %s timeout = %d\n", (double)offset,
+ (double)count, (unsigned int)lock_pid,
+ fsp->fsp_name, (int)lock_timeout ));
+
+ if (locktype & LOCKING_ANDX_CANCEL_LOCK) {
+ if (lp_blocking_locks(SNUM(conn))) {
+
+ /* Schedule a message to ourselves to
+ remove the blocking lock record and
+ return the right error. */
+
+ if (!blocking_lock_cancel(fsp,
+ lock_pid,
+ offset,
+ count,
+ WINDOWS_LOCK,
+ locktype,
+ NT_STATUS_FILE_LOCK_CONFLICT)) {
+ END_PROFILE(SMBlockingX);
+ reply_nterror(
+ req,
+ NT_STATUS_DOS(
+ ERRDOS,
+ ERRcancelviolation));
+ return;
+ }
+ }
+ /* Remove a matching pending lock. */
+ status = do_lock_cancel(fsp,
+ lock_pid,
+ count,
+ offset,
+ WINDOWS_LOCK);
+ } else {
+ bool blocking_lock = lock_timeout ? True : False;
+ bool defer_lock = False;
+ struct byte_range_lock *br_lck;
+ uint32 block_smbpid;
+
+ br_lck = do_lock(smbd_messaging_context(),
+ fsp,
+ lock_pid,
+ count,
+ offset,
+ lock_type,
+ WINDOWS_LOCK,
+ blocking_lock,
+ &status,
+ &block_smbpid);
+
+ if (br_lck && blocking_lock && ERROR_WAS_LOCK_DENIED(status)) {
+ /* Windows internal resolution for blocking locks seems
+ to be about 200ms... Don't wait for less than that. JRA. */
+ if (lock_timeout != -1 && lock_timeout < lp_lock_spin_time()) {
+ lock_timeout = lp_lock_spin_time();
+ }
+ defer_lock = True;
+ }
+
+ /* This heuristic seems to match W2K3 very well. If a
+ lock sent with timeout of zero would fail with NT_STATUS_FILE_LOCK_CONFLICT
+ it pretends we asked for a timeout of between 150 - 300 milliseconds as
+ far as I can tell. Replacement for do_lock_spin(). JRA. */
+
+ if (br_lck && lp_blocking_locks(SNUM(conn)) && !blocking_lock &&
+ NT_STATUS_EQUAL((status), NT_STATUS_FILE_LOCK_CONFLICT)) {
+ defer_lock = True;
+ lock_timeout = lp_lock_spin_time();
+ }
+
+ if (br_lck && defer_lock) {
+ /*
+ * A blocking lock was requested. Package up
+ * this smb into a queued request and push it
+ * onto the blocking lock queue.
+ */
+ if(push_blocking_lock_request(br_lck,
+ req,
+ fsp,
+ lock_timeout,
+ i,
+ lock_pid,
+ lock_type,
+ WINDOWS_LOCK,
+ offset,
+ count,
+ block_smbpid)) {
+ TALLOC_FREE(br_lck);
+ END_PROFILE(SMBlockingX);
+ return;
+ }
+ }
+
+ TALLOC_FREE(br_lck);
+ }
+
+ if (NT_STATUS_V(status)) {
+ END_PROFILE(SMBlockingX);
+ reply_nterror(req, status);
+ return;
+ }
+ }
+
+ /* If any of the above locks failed, then we must unlock
+ all of the previous locks (X/Open spec). */
+
+ if (!(locktype & LOCKING_ANDX_CANCEL_LOCK) &&
+ (i != num_locks) &&
+ (num_locks != 0)) {
+ /*
+ * Ensure we don't do a remove on the lock that just failed,
+ * as under POSIX rules, if we have a lock already there, we
+ * will delete it (and we shouldn't) .....
+ */
+ for(i--; i >= 0; i--) {
+ lock_pid = get_lock_pid( data, i, large_file_format);
+ count = get_lock_count( data, i, large_file_format);
+ offset = get_lock_offset( data, i, large_file_format,
+ &err);
+
+ /*
+ * There is no error code marked "stupid client
+ * bug".... :-).
+ */
+ if(err) {
+ END_PROFILE(SMBlockingX);
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ return;
+ }
+
+ do_unlock(smbd_messaging_context(),
+ fsp,
+ lock_pid,
+ count,
+ offset,
+ WINDOWS_LOCK);
+ }
+ END_PROFILE(SMBlockingX);
+ reply_nterror(req, status);
+ return;
+ }
+
+ reply_outbuf(req, 2, 0);
+
+ DEBUG(3, ("lockingX fnum=%d type=%d num_locks=%d num_ulocks=%d\n",
+ fsp->fnum, (unsigned int)locktype, num_locks, num_ulocks));
+
+ END_PROFILE(SMBlockingX);
+ chain_reply(req);
+}
+
+#undef DBGC_CLASS
+#define DBGC_CLASS DBGC_ALL
+
+/****************************************************************************
+ Reply to a SMBreadbmpx (read block multiplex) request.
+ Always reply with an error, if someone has a platform really needs this,
+ please contact vl@samba.org
+****************************************************************************/
+
+void reply_readbmpx(struct smb_request *req)
+{
+ START_PROFILE(SMBreadBmpx);
+ reply_doserror(req, ERRSRV, ERRuseSTD);
+ END_PROFILE(SMBreadBmpx);
+ return;
+}
+
+/****************************************************************************
+ Reply to a SMBreadbs (read block multiplex secondary) request.
+ Always reply with an error, if someone has a platform really needs this,
+ please contact vl@samba.org
+****************************************************************************/
+
+void reply_readbs(struct smb_request *req)
+{
+ START_PROFILE(SMBreadBs);
+ reply_doserror(req, ERRSRV, ERRuseSTD);
+ END_PROFILE(SMBreadBs);
+ return;
+}
+
+/****************************************************************************
+ Reply to a SMBsetattrE.
+****************************************************************************/
+
+void reply_setattrE(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ struct timespec ts[2];
+ files_struct *fsp;
+ SMB_STRUCT_STAT sbuf;
+ NTSTATUS status;
+
+ START_PROFILE(SMBsetattrE);
+
+ if (req->wct < 7) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBsetattrE);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if(!fsp || (fsp->conn != conn)) {
+ reply_doserror(req, ERRDOS, ERRbadfid);
+ END_PROFILE(SMBsetattrE);
+ return;
+ }
+
+
+ /*
+ * Convert the DOS times into unix times. Ignore create
+ * time as UNIX can't set this.
+ */
+
+ ts[0] = convert_time_t_to_timespec(
+ srv_make_unix_date2(req->inbuf+smb_vwv3)); /* atime. */
+ ts[1] = convert_time_t_to_timespec(
+ srv_make_unix_date2(req->inbuf+smb_vwv5)); /* mtime. */
+
+ reply_outbuf(req, 0, 0);
+
+ /*
+ * Patch from Ray Frush <frush@engr.colostate.edu>
+ * Sometimes times are sent as zero - ignore them.
+ */
+
+ /* Ensure we have a valid stat struct for the source. */
+ if (fsp->fh->fd != -1) {
+ if (SMB_VFS_FSTAT(fsp, &sbuf) == -1) {
+ status = map_nt_error_from_unix(errno);
+ reply_nterror(req, status);
+ END_PROFILE(SMBsetattrE);
+ return;
+ }
+ } else {
+ if (SMB_VFS_STAT(conn, fsp->fsp_name, &sbuf) == -1) {
+ status = map_nt_error_from_unix(errno);
+ reply_nterror(req, status);
+ END_PROFILE(SMBsetattrE);
+ return;
+ }
+ }
+
+ status = smb_set_file_time(conn, fsp, fsp->fsp_name,
+ &sbuf, ts, true);
+ if (!NT_STATUS_IS_OK(status)) {
+ reply_doserror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBsetattrE);
+ return;
+ }
+
+ DEBUG( 3, ( "reply_setattrE fnum=%d actime=%u modtime=%u\n",
+ fsp->fnum,
+ (unsigned int)ts[0].tv_sec,
+ (unsigned int)ts[1].tv_sec));
+
+ END_PROFILE(SMBsetattrE);
+ return;
+}
+
+
+/* Back from the dead for OS/2..... JRA. */
+
+/****************************************************************************
+ Reply to a SMBwritebmpx (write block multiplex primary) request.
+ Always reply with an error, if someone has a platform really needs this,
+ please contact vl@samba.org
+****************************************************************************/
+
+void reply_writebmpx(struct smb_request *req)
+{
+ START_PROFILE(SMBwriteBmpx);
+ reply_doserror(req, ERRSRV, ERRuseSTD);
+ END_PROFILE(SMBwriteBmpx);
+ return;
+}
+
+/****************************************************************************
+ Reply to a SMBwritebs (write block multiplex secondary) request.
+ Always reply with an error, if someone has a platform really needs this,
+ please contact vl@samba.org
+****************************************************************************/
+
+void reply_writebs(struct smb_request *req)
+{
+ START_PROFILE(SMBwriteBs);
+ reply_doserror(req, ERRSRV, ERRuseSTD);
+ END_PROFILE(SMBwriteBs);
+ return;
+}
+
+/****************************************************************************
+ Reply to a SMBgetattrE.
+****************************************************************************/
+
+void reply_getattrE(struct smb_request *req)
+{
+ connection_struct *conn = req->conn;
+ SMB_STRUCT_STAT sbuf;
+ int mode;
+ files_struct *fsp;
+ struct timespec create_ts;
+
+ START_PROFILE(SMBgetattrE);
+
+ if (req->wct < 1) {
+ reply_nterror(req, NT_STATUS_INVALID_PARAMETER);
+ END_PROFILE(SMBgetattrE);
+ return;
+ }
+
+ fsp = file_fsp(SVAL(req->inbuf,smb_vwv0));
+
+ if(!fsp || (fsp->conn != conn)) {
+ reply_doserror(req, ERRDOS, ERRbadfid);
+ END_PROFILE(SMBgetattrE);
+ return;
+ }
+
+ /* Do an fstat on this file */
+ if(fsp_stat(fsp, &sbuf)) {
+ reply_unixerror(req, ERRDOS, ERRnoaccess);
+ END_PROFILE(SMBgetattrE);
+ return;
+ }
+
+ mode = dos_mode(conn,fsp->fsp_name,&sbuf);
+
+ /*
+ * Convert the times into dos times. Set create
+ * date to be last modify date as UNIX doesn't save
+ * this.
+ */
+
+ reply_outbuf(req, 11, 0);
+
+ create_ts = get_create_timespec(&sbuf,
+ lp_fake_dir_create_times(SNUM(conn)));
+ srv_put_dos_date2((char *)req->outbuf, smb_vwv0, create_ts.tv_sec);
+ srv_put_dos_date2((char *)req->outbuf, smb_vwv2, sbuf.st_atime);
+ /* Should we check pending modtime here ? JRA */
+ srv_put_dos_date2((char *)req->outbuf, smb_vwv4, sbuf.st_mtime);
+
+ if (mode & aDIR) {
+ SIVAL(req->outbuf, smb_vwv6, 0);
+ SIVAL(req->outbuf, smb_vwv8, 0);
+ } else {
+ uint32 allocation_size = get_allocation_size(conn,fsp, &sbuf);
+ SIVAL(req->outbuf, smb_vwv6, (uint32)sbuf.st_size);
+ SIVAL(req->outbuf, smb_vwv8, allocation_size);
+ }
+ SSVAL(req->outbuf,smb_vwv10, mode);
+
+ DEBUG( 3, ( "reply_getattrE fnum=%d\n", fsp->fnum));
+
+ END_PROFILE(SMBgetattrE);
+ return;
+}