Skip to contentSkip to conversation

still incorrect cifs_reconnect fix?

CIFS Applied Petr Tesarik 3 participants 5 messages Nov 19, 2007 – Nov 20, 2007

lore
Share & tools
lore
#0 lore raw

Hi,

while merging commits f01d5e14e764b14b6bf5512678523d009254b209 and 638b250766272fcaaa0f7ed2776f58f4ac701914 into SLES10, I've noticed that there's apparently a bug. The code currently looks like this:

pdu_length = 4; /* enough to get RFC1001 header */
incomplete_rcv:

		length =
		    kernel_recvmsg(csocket, &smb_msg,
				&iov, 1, pdu_length, 0 /* BB other flags? */);

/* ... some irrelevant code left out ... */

		} else if (length < 4) {	/* <----- HERE IS THE PROBLEM
			cFYI(1, ("less than four bytes received (%d bytes)",
			      length));
			pdu_length -= length;
			msleep(1);
			goto incomplete_rcv;
		}

I think we should be checking for length < pdu_length, not for length < 4, because if we read 2 bytes in the first run and 2 bytes in the second un, CIFS will still treat the second run as incomplete (and possibly run in an infinite loop). Am I missing something obvious?

Kind regards,
Petr Tesarik

#1 lore raw
Quoted 23 lines · On Mon, 19 Nov 2007 10:30:45 +0100 Petr Tesarik <[email protected]> wrote:

Hi,

while merging commits f01d5e14e764b14b6bf5512678523d009254b209 and 638b250766272fcaaa0f7ed2776f58f4ac701914 into SLES10, I've noticed that there's apparently a bug. The code currently looks like this:

pdu_length = 4; /* enough to get RFC1001 header */
incomplete_rcv:

		length =
		    kernel_recvmsg(csocket, &smb_msg,
				&iov, 1, pdu_length, 0 /* BB other

flags? */);

/* ... some irrelevant code left out ... */

		} else if (length < 4) {	/* <----- HERE IS THE
PROBLEM cFYI(1, ("less than four bytes received (%d bytes)",
			      length));
			pdu_length -= length;
			msleep(1);
			goto incomplete_rcv;
		}
Quoted 5 lines

I think we should be checking for length < pdu_length, not for length < 4, because if we read 2 bytes in the first run and 2 bytes in the second un, CIFS will still treat the second run as incomplete (and possibly run in an infinite loop). Am I missing something obvious?

I think you're right that the logic there is wrong, but I don't see an infinite loop happening. It looks like in this situation, that the next call to kernel_recvmsg will be called a size of 0, which should eventually return 0. It'll then fall into the previous condition -- do a reconnect and then go around the big loop again.

A patch to clean that up would probably be a good thing. We likely don't need to do a reconnect here.

--
Jeff Layton <[email protected]>

#2 lore raw

Re: [linux-cifs-client] [PATCH] get rid of spurious session reconnects

Quoted 32 lines · Jeff Layton
Quoted 23 lines · On Mon, 19 Nov 2007 10:30:45 +0100 Petr Tesarik <[email protected]> wrote:

Hi,

while merging commits f01d5e14e764b14b6bf5512678523d009254b209 and 638b250766272fcaaa0f7ed2776f58f4ac701914 into SLES10, I've noticed that there's apparently a bug. The code currently looks like this:

pdu_length = 4; /* enough to get RFC1001 header */
incomplete_rcv:

		length =
		    kernel_recvmsg(csocket, &smb_msg,
				&iov, 1, pdu_length, 0 /* BB other

flags? */);

/* ... some irrelevant code left out ... */

		} else if (length < 4) {	/* <----- HERE IS THE
PROBLEM cFYI(1, ("less than four bytes received (%d bytes)",
			      length));
			pdu_length -= length;
			msleep(1);
			goto incomplete_rcv;
		}
Quoted 5 lines

I think we should be checking for length < pdu_length, not for length < 4, because if we read 2 bytes in the first run and 2 bytes in the second un, CIFS will still treat the second run as incomplete (and possibly run in an infinite loop). Am I missing something obvious?

Quoted 8 lines

I think you're right that the logic there is wrong, but I don't see an infinite loop happening. It looks like in this situation, that the next call to kernel_recvmsg will be called a size of 0, which should eventually return 0. It'll then fall into the previous condition -- do a reconnect and then go around the big loop again.

A patch to clean that up would probably be a good thing. We likely don't need to do a reconnect here.

OK, here we go:

When retrying kernel_recvmsg() because of a short read, check returned length against the remaining length, not against total length. This avoids unneeded session reconnects which would otherwise occur when kernel_recvmsg() finally returns zero when asked to read zero bytes.

 connect.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/cifs/connect.c +1 -1
index c52a76f..f972ec9 100644
@@ -438,7 +438,7 @@ incomplete_rcv:
 			csocket = server->ssocket;
 			wake_up(&server->response_q);
 			continue;
-		} else if (length < 4) {
+		} else if (length < pdu_length) {
 			cFYI(1, ("less than four bytes received (%d bytes)",
 			      length));
 			pdu_length -= length;
#3 lore raw

Re: [linux-cifs-client] [PATCH] get rid of spurious session reconnects

Quoted 43 lines · On Mon, 19 Nov 2007 14:23:13 +0100 Petr Tesarik <[email protected]> wrote:
Quoted 33 lines · Jeff Layton
Quoted 23 lines · On Mon, 19 Nov 2007 10:30:45 +0100 Petr Tesarik <[email protected]> wrote:

Hi,

while merging commits f01d5e14e764b14b6bf5512678523d009254b209 and 638b250766272fcaaa0f7ed2776f58f4ac701914 into SLES10, I've noticed that there's apparently a bug. The code currently looks like this:

pdu_length = 4; /* enough to get RFC1001 header */
incomplete_rcv:

		length =
		    kernel_recvmsg(csocket, &smb_msg,
				&iov, 1, pdu_length, 0 /* BB other

flags? */);

/* ... some irrelevant code left out ... */

		} else if (length < 4) {	/* <----- HERE IS
THE PROBLEM cFYI(1, ("less than four bytes received (%d bytes)",
			      length));
			pdu_length -= length;
			msleep(1);
			goto incomplete_rcv;
		}
Quoted 6 lines

I think we should be checking for length < pdu_length, not for length < 4, because if we read 2 bytes in the first run and 2 bytes in the second un, CIFS will still treat the second run as incomplete (and possibly run in an infinite loop). Am I missing something obvious?

Quoted 8 lines

I think you're right that the logic there is wrong, but I don't see an infinite loop happening. It looks like in this situation, that the next call to kernel_recvmsg will be called a size of 0, which should eventually return 0. It'll then fall into the previous condition -- do a reconnect and then go around the big loop again.

A patch to clean that up would probably be a good thing. We likely don't need to do a reconnect here.

Quoted 9 lines

OK, here we go:

When retrying kernel_recvmsg() because of a short read, check returned length against the remaining length, not against total length. This avoids unneeded session reconnects which would otherwise occur when kernel_recvmsg() finally returns zero when asked to read zero bytes.

Looks good (nice catch on this problem, btw). How about we fix up the cFYI at the same time? Note that this should probably be tested, even though it seems logical.

fs/cifs/connect.c +3 -3
index c4b32b7..b6ed933 100644
@@ -438,9 +438,9 @@ incomplete_rcv:
 			csocket = server->ssocket;
 			wake_up(&server->response_q);
 			continue;
-		} else if (length < 4) {
-			cFYI(1, ("less than four bytes received (%d bytes)",
-			      length));
+		} else if (length < pdu_length) {
+			cFYI(1, ("less than %d bytes received (%d bytes)",
+			      pdu_length, length));
 			pdu_length -= length;
 			msleep(1);
 			goto incomplete_rcv;
Applied
#4 lore raw

Re: [linux-cifs-client] [PATCH] get rid of spurious session reconnects

Merged into cifs-2.6.git

I still want to review/test more on the one other bug fix (quota or out of disk space, cifs sync issue which Jeff proposed a fix for earlier) and the ASCII vs. Unicode plain-text password check bug

Quoted 80 lines · Jeff Layton
Quoted 43 lines · On Mon, 19 Nov 2007 14:23:13 +0100 Petr Tesarik <[email protected]> wrote:
Quoted 33 lines · Jeff Layton
Quoted 23 lines · On Mon, 19 Nov 2007 10:30:45 +0100 Petr Tesarik <[email protected]> wrote:

Hi,

while merging commits f01d5e14e764b14b6bf5512678523d009254b209 and 638b250766272fcaaa0f7ed2776f58f4ac701914 into SLES10, I've noticed that there's apparently a bug. The code currently looks like this:

pdu_length = 4; /* enough to get RFC1001 header */
incomplete_rcv:

          length =
              kernel_recvmsg(csocket, &smb_msg,
                          &iov, 1, pdu_length, 0 /* BB other

flags? */);

/* ... some irrelevant code left out ... */

          } else if (length < 4) {        /* <----- HERE IS
THE PROBLEM cFYI(1, ("less than four bytes received (%d bytes)",
                        length));
                  pdu_length -= length;
                  msleep(1);
                  goto incomplete_rcv;
          }
Quoted 6 lines

I think we should be checking for length < pdu_length, not for length < 4, because if we read 2 bytes in the first run and 2 bytes in the second un, CIFS will still treat the second run as incomplete (and possibly run in an infinite loop). Am I missing something obvious?

Quoted 8 lines

I think you're right that the logic there is wrong, but I don't see an infinite loop happening. It looks like in this situation, that the next call to kernel_recvmsg will be called a size of 0, which should eventually return 0. It'll then fall into the previous condition -- do a reconnect and then go around the big loop again.

A patch to clean that up would probably be a good thing. We likely don't need to do a reconnect here.

Quoted 9 lines

OK, here we go:

When retrying kernel_recvmsg() because of a short read, check returned length against the remaining length, not against total length. This avoids unneeded session reconnects which would otherwise occur when kernel_recvmsg() finally returns zero when asked to read zero bytes.

Looks good (nice catch on this problem, btw). How about we fix up the cFYI at the same time? Note that this should probably be tested, even though it seems logical.

fs/cifs/connect.c +3 -3
index c4b32b7..b6ed933 100644
@@ -438,9 +438,9 @@ incomplete_rcv:
                        csocket = server->ssocket;
                        wake_up(&server->response_q);
                        continue;
-               } else if (length < 4) {
-                       cFYI(1, ("less than four bytes received (%d bytes)",
-                             length));
+               } else if (length < pdu_length) {
+                       cFYI(1, ("less than %d bytes received (%d bytes)",
+                             pdu_length, length));
                        pdu_length -= length;
                        msleep(1);
                        goto incomplete_rcv;
Signature · 3 lines
Thanks,

Steve