summaryrefslogtreecommitdiffstats
path: root/source4/client/cifsddio.c
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--source4/client/cifsddio.c524
1 files changed, 524 insertions, 0 deletions
diff --git a/source4/client/cifsddio.c b/source4/client/cifsddio.c
new file mode 100644
index 0000000..0163daa
--- /dev/null
+++ b/source4/client/cifsddio.c
@@ -0,0 +1,524 @@
+/*
+ CIFSDD - dd for SMB.
+ IO routines, generic and specific.
+
+ Copyright (C) James Peach 2005-2006
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+#include "libcli/libcli.h"
+#include "lib/cmdline/cmdline.h"
+
+#include "cifsdd.h"
+
+/* ------------------------------------------------------------------------- */
+/* UNIX file descriptor IO. */
+/* ------------------------------------------------------------------------- */
+
+struct fd_handle
+{
+ struct dd_iohandle h;
+ int fd;
+};
+
+#define IO_HANDLE_TO_FD(h) (((struct fd_handle *)(h))->fd)
+
+static bool fd_seek_func(void * handle, uint64_t offset)
+{
+ ssize_t ret;
+
+ ret = lseek(IO_HANDLE_TO_FD(handle), offset, SEEK_SET);
+ if (ret < 0) {
+ fprintf(stderr, "%s: seek failed: %s\n",
+ PROGNAME, strerror(errno));
+ return(false);
+ }
+
+ return(true);
+}
+
+static bool fd_read_func(void * handle,
+ uint8_t * buf,
+ uint64_t wanted,
+ uint64_t * actual)
+{
+ ssize_t ret;
+
+ ret = read(IO_HANDLE_TO_FD(handle), buf, wanted);
+ if (ret < 0) {
+ fprintf(stderr, "%s: %llu byte read failed: %s\n",
+ PROGNAME, (unsigned long long)wanted,
+ strerror(errno));
+ return(false);
+ }
+
+ *actual = (uint64_t)ret;
+ return(true);
+}
+
+static bool fd_write_func(void * handle,
+ uint8_t * buf,
+ uint64_t wanted,
+ uint64_t * actual)
+{
+ ssize_t ret;
+
+ ret = write(IO_HANDLE_TO_FD(handle), buf, wanted);
+ if (ret < 0) {
+ fprintf(stderr, "%s: %llu byte write failed: %s\n",
+ PROGNAME, (unsigned long long)wanted,
+ strerror(errno));
+ return(false);
+ }
+
+ *actual = (uint64_t)ret;
+ return(true);
+}
+
+static struct dd_iohandle * open_fd_handle(const char * path,
+ uint64_t io_size,
+ int options)
+{
+ struct fd_handle * fdh;
+ int oflags = 0;
+
+ DEBUG(4, ("opening fd stream for %s\n", path));
+ if ((fdh = talloc_zero(NULL, struct fd_handle)) == NULL) {
+ return(NULL);
+ }
+
+ fdh->h.io_read = fd_read_func;
+ fdh->h.io_write = fd_write_func;
+ fdh->h.io_seek = fd_seek_func;
+
+ if (options & DD_DIRECT_IO) {
+#ifdef HAVE_OPEN_O_DIRECT
+ oflags |= O_DIRECT;
+#else
+ DEBUG(1, ("no support for direct IO on this platform\n"));
+#endif
+ }
+
+ if (options & DD_SYNC_IO)
+ oflags |= O_SYNC;
+
+ oflags |= (options & DD_WRITE) ? (O_WRONLY | O_CREAT) : (O_RDONLY);
+
+ fdh->fd = open(path, oflags, 0644);
+ if (fdh->fd < 0) {
+ fprintf(stderr, "%s: %s: %s\n",
+ PROGNAME, path, strerror(errno));
+ talloc_free(fdh);
+ return(NULL);
+ }
+
+ if (options & DD_OPLOCK) {
+ DEBUG(2, ("FIXME: take local oplock on %s\n", path));
+ }
+
+ SMB_ASSERT((void *)fdh == (void *)&fdh->h);
+ return(&fdh->h);
+}
+
+/* ------------------------------------------------------------------------- */
+/* CIFS client IO. */
+/* ------------------------------------------------------------------------- */
+
+struct cifs_handle
+{
+ struct dd_iohandle h;
+ struct smbcli_state * cli;
+ int fnum;
+ uint64_t offset;
+};
+
+#define IO_HANDLE_TO_SMB(h) ((struct cifs_handle *)(h))
+
+static bool smb_seek_func(void * handle, uint64_t offset)
+{
+ IO_HANDLE_TO_SMB(handle)->offset = offset;
+ return(true);
+}
+
+static bool smb_read_func(void * handle, uint8_t * buf, uint64_t wanted,
+ uint64_t * actual)
+{
+ NTSTATUS ret;
+ union smb_read r;
+ struct cifs_handle * smbh;
+
+ ZERO_STRUCT(r);
+ smbh = IO_HANDLE_TO_SMB(handle);
+
+ r.generic.level = RAW_READ_READX;
+ r.readx.in.file.fnum = smbh->fnum;
+ r.readx.in.offset = smbh->offset;
+ r.readx.in.mincnt = wanted;
+ r.readx.in.maxcnt = wanted;
+ r.readx.out.data = buf;
+
+ /* FIXME: Should I really set readx.in.remaining? That just seems
+ * redundant.
+ */
+ ret = smb_raw_read(smbh->cli->tree, &r);
+ if (!NT_STATUS_IS_OK(ret)) {
+ fprintf(stderr, "%s: %llu byte read failed: %s\n",
+ PROGNAME, (unsigned long long)wanted,
+ nt_errstr(ret));
+ return(false);
+ }
+
+ /* Trap integer wrap. */
+ SMB_ASSERT((smbh->offset + r.readx.out.nread) >= smbh->offset);
+
+ *actual = r.readx.out.nread;
+ smbh->offset += r.readx.out.nread;
+ return(true);
+}
+
+static bool smb_write_func(void * handle, uint8_t * buf, uint64_t wanted,
+ uint64_t * actual)
+{
+ NTSTATUS ret;
+ union smb_write w;
+ struct cifs_handle * smbh;
+
+ ZERO_STRUCT(w);
+ smbh = IO_HANDLE_TO_SMB(handle);
+
+ w.generic.level = RAW_WRITE_WRITEX;
+ w.writex.in.file.fnum = smbh->fnum;
+ w.writex.in.offset = smbh->offset;
+ w.writex.in.count = wanted;
+ w.writex.in.data = buf;
+
+ ret = smb_raw_write(smbh->cli->tree, &w);
+ if (!NT_STATUS_IS_OK(ret)) {
+ fprintf(stderr, "%s: %llu byte write failed: %s\n",
+ PROGNAME, (unsigned long long)wanted,
+ nt_errstr(ret));
+ return(false);
+ }
+
+ *actual = w.writex.out.nwritten;
+ smbh->offset += w.writex.out.nwritten;
+ return(true);
+}
+
+static struct smbcli_state * init_smb_session(struct resolve_context *resolve_ctx,
+ struct tevent_context *ev,
+ const char * host,
+ const char **ports,
+ const char * share,
+ const char *socket_options,
+ struct smbcli_options *options,
+ struct smbcli_session_options *session_options,
+ struct gensec_settings *gensec_settings)
+{
+ NTSTATUS ret;
+ struct smbcli_state * cli = NULL;
+
+ /* When we support SMB URLs, we can get different user credentials for
+ * each connection, but for now, we just use the same one for both.
+ */
+ ret = smbcli_full_connection(NULL, &cli, host, ports, share,
+ NULL /* devtype */,
+ socket_options,
+ samba_cmdline_get_creds(),
+ resolve_ctx,
+ ev, options,
+ session_options,
+ gensec_settings);
+
+ if (!NT_STATUS_IS_OK(ret)) {
+ fprintf(stderr, "%s: connecting to //%s/%s: %s\n",
+ PROGNAME, host, share, nt_errstr(ret));
+ return(NULL);
+ }
+
+ return(cli);
+}
+
+static int open_smb_file(struct smbcli_state * cli,
+ const char * path,
+ int options)
+{
+ NTSTATUS ret;
+ union smb_open o;
+
+ ZERO_STRUCT(o);
+
+ o.ntcreatex.level = RAW_OPEN_NTCREATEX;
+ o.ntcreatex.in.fname = path;
+
+ /* TODO: It's not clear whether to use these flags or to use the
+ * similarly named NTCREATEX flags in the create_options field.
+ */
+ if (options & DD_DIRECT_IO)
+ o.ntcreatex.in.flags |= FILE_FLAG_NO_BUFFERING;
+
+ if (options & DD_SYNC_IO)
+ o.ntcreatex.in.flags |= FILE_FLAG_WRITE_THROUGH;
+
+ o.ntcreatex.in.access_mask |=
+ (options & DD_WRITE) ? SEC_FILE_WRITE_DATA
+ : SEC_FILE_READ_DATA;
+
+ /* Try to create the file only if we will be writing to it. */
+ o.ntcreatex.in.open_disposition =
+ (options & DD_WRITE) ? NTCREATEX_DISP_OPEN_IF
+ : NTCREATEX_DISP_OPEN;
+
+ o.ntcreatex.in.share_access =
+ NTCREATEX_SHARE_ACCESS_READ | NTCREATEX_SHARE_ACCESS_WRITE;
+
+ if (options & DD_OPLOCK) {
+ o.ntcreatex.in.flags |= NTCREATEX_FLAGS_REQUEST_OPLOCK;
+ }
+
+ ret = smb_raw_open(cli->tree, NULL, &o);
+ if (!NT_STATUS_IS_OK(ret)) {
+ fprintf(stderr, "%s: opening %s: %s\n",
+ PROGNAME, path, nt_errstr(ret));
+ return(-1);
+ }
+
+ return(o.ntcreatex.out.file.fnum);
+}
+
+static struct dd_iohandle * open_cifs_handle(struct resolve_context *resolve_ctx,
+ struct tevent_context *ev,
+ const char * host,
+ const char **ports,
+ const char * share,
+ const char * path,
+ uint64_t io_size,
+ int options,
+ const char *socket_options,
+ struct smbcli_options *smb_options,
+ struct smbcli_session_options *smb_session_options,
+ struct gensec_settings *gensec_settings)
+{
+ struct cifs_handle * smbh;
+
+ if (path == NULL || *path == '\0') {
+ fprintf(stderr, "%s: missing path name within share //%s/%s\n",
+ PROGNAME, host, share);
+ }
+
+ DEBUG(4, ("opening SMB stream to //%s/%s for %s\n",
+ host, share, path));
+
+ if ((smbh = talloc_zero(NULL, struct cifs_handle)) == NULL) {
+ return(NULL);
+ }
+
+ smbh->h.io_read = smb_read_func;
+ smbh->h.io_write = smb_write_func;
+ smbh->h.io_seek = smb_seek_func;
+
+ if ((smbh->cli = init_smb_session(resolve_ctx, ev, host, ports, share,
+ socket_options,
+ smb_options, smb_session_options,
+ gensec_settings)) == NULL) {
+ return(NULL);
+ }
+
+ DEBUG(4, ("connected to //%s/%s with xmit size of %u bytes\n",
+ host, share, smbh->cli->transport->negotiate.max_xmit));
+
+ smbh->fnum = open_smb_file(smbh->cli, path, options);
+ return(&smbh->h);
+}
+
+/* ------------------------------------------------------------------------- */
+/* Abstract IO interface. */
+/* ------------------------------------------------------------------------- */
+
+struct dd_iohandle * dd_open_path(struct resolve_context *resolve_ctx,
+ struct tevent_context *ev,
+ const char * path,
+ const char **ports,
+ uint64_t io_size,
+ int options,
+ const char *socket_options,
+ struct smbcli_options *smb_options,
+ struct smbcli_session_options *smb_session_options,
+ struct gensec_settings *gensec_settings)
+{
+ if (file_exist(path)) {
+ return(open_fd_handle(path, io_size, options));
+ } else {
+ char * host;
+ char * share;
+
+ if (smbcli_parse_unc(path, NULL, &host, &share)) {
+ const char * remain;
+ remain = strstr(path, share) + strlen(share);
+
+ /* Skip over leading directory separators. */
+ while (*remain == '/' || *remain == '\\') { remain++; }
+
+ return(open_cifs_handle(resolve_ctx, ev, host, ports,
+ share, remain,
+ io_size, options,
+ socket_options, smb_options,
+ smb_session_options,
+ gensec_settings));
+ }
+
+ return(open_fd_handle(path, io_size, options));
+ }
+}
+
+/* Fill the buffer till it has at least need_size bytes. Use read operations of
+ * block_size bytes. Return the number of bytes read and fill buf_size with
+ * the new buffer size.
+ *
+ * NOTE: The IO buffer is guaranteed to be big enough to fit
+ * need_size + block_size bytes into it.
+ */
+bool dd_fill_block(struct dd_iohandle * h,
+ uint8_t * buf,
+ uint64_t * buf_size,
+ uint64_t need_size,
+ uint64_t block_size)
+{
+ uint64_t read_size;
+
+ SMB_ASSERT(block_size > 0);
+ SMB_ASSERT(need_size > 0);
+
+ while (*buf_size < need_size) {
+
+ if (!h->io_read(h, buf + (*buf_size), block_size, &read_size)) {
+ return(false);
+ }
+
+ if (read_size == 0) {
+ h->io_flags |= DD_END_OF_FILE;
+ break;
+ }
+
+ DEBUG(6, ("added %llu bytes to IO buffer (need %llu bytes)\n",
+ (unsigned long long)read_size,
+ (unsigned long long)need_size));
+
+ *buf_size += read_size;
+ dd_stats.in.bytes += read_size;
+
+ if (read_size == block_size) {
+ dd_stats.in.fblocks++;
+ } else {
+ DEBUG(3, ("partial read of %llu bytes (expected %llu)\n",
+ (unsigned long long)read_size,
+ (unsigned long long)block_size));
+ dd_stats.in.pblocks++;
+ }
+ }
+
+ return(true);
+}
+
+/* Flush a buffer that contains buf_size bytes. Use writes of block_size to do it,
+ * and shift any remaining bytes back to the head of the buffer when there are
+ * no more block_size sized IOs left.
+ */
+bool dd_flush_block(struct dd_iohandle * h,
+ uint8_t * buf,
+ uint64_t * buf_size,
+ uint64_t block_size)
+{
+ uint64_t write_size;
+ uint64_t total_size = 0;
+
+ SMB_ASSERT(block_size > 0);
+
+ /* We have explicitly been asked to write a partial block. */
+ if ((*buf_size) < block_size) {
+
+ if (!h->io_write(h, buf, *buf_size, &write_size)) {
+ return(false);
+ }
+
+ if (write_size == 0) {
+ fprintf(stderr, "%s: unexpectedly wrote 0 bytes\n",
+ PROGNAME);
+ return(false);
+ }
+
+ total_size += write_size;
+ dd_stats.out.bytes += write_size;
+ dd_stats.out.pblocks++;
+ }
+
+ /* Write as many full blocks as there are in the buffer. */
+ while (((*buf_size) - total_size) >= block_size) {
+
+ if (!h->io_write(h, buf + total_size, block_size, &write_size)) {
+ return(false);
+ }
+
+ if (write_size == 0) {
+ fprintf(stderr, "%s: unexpectedly wrote 0 bytes\n",
+ PROGNAME);
+ return(false);
+ }
+
+ if (write_size == block_size) {
+ dd_stats.out.fblocks++;
+ } else {
+ dd_stats.out.pblocks++;
+ }
+
+ total_size += write_size;
+ dd_stats.out.bytes += write_size;
+
+ DEBUG(6, ("flushed %llu bytes from IO buffer of %llu bytes (%llu remain)\n",
+ (unsigned long long)block_size,
+ (unsigned long long)block_size,
+ (unsigned long long)(block_size - total_size)));
+ }
+
+ SMB_ASSERT(total_size > 0);
+
+ /* We have flushed as much of the IO buffer as we can while
+ * still doing block_size'd operations. Shift any remaining data
+ * to the front of the IO buffer.
+ */
+ if ((*buf_size) > total_size) {
+ uint64_t remain = (*buf_size) - total_size;
+
+ DEBUG(3, ("shifting %llu remainder bytes to IO buffer head\n",
+ (unsigned long long)remain));
+
+ memmove(buf, buf + total_size, remain);
+ (*buf_size) = remain;
+ } else if ((*buf_size) == total_size) {
+ (*buf_size) = 0;
+ } else {
+ /* Else buffer contains buf_size bytes that we will append
+ * to next time round.
+ */
+ DEBUG(3, ("%llu unflushed bytes left in IO buffer\n",
+ (unsigned long long)(*buf_size)));
+ }
+
+ return(true);
+}
+
+/* vim: set sw=8 sts=8 ts=8 tw=79 : */