Skip to contentSkip to conversation

xenconsole: merge pty access check into when it is opened

PATCH Superseded Matthew Daley 3 participants 13 messages Nov 30, 2013 – Dec 16, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xenconsole: merge pty access check into when it is opened

This stops pty_path from being leaked, and removes the toctou race, FWIW.

Not sure why it's a separate check to begin with...

Coverity-ID: 1056047

---
 tools/console/client/main.c |   10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
tools/console/client/main.c +5 -5
index 38c856a..c32d3eb 100644
@@ -116,12 +116,12 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
 			 * disambiguate: just read the pty path */
 			pty_path = xs_read(xs, XBT_NULL, path, &len);
 			if (pty_path != NULL) {
-				if (access(pty_path, R_OK|W_OK) != 0)
-					continue;
 				pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-				if (pty_fd == -1) 
-					err(errno, "Could not open tty `%s'", 
-					    pty_path);
+				if (pty_fd == -1) {
+					if (errno != EACCES)
+						err(errno, "Could not open tty `%s'",
+							pty_path);
+				}
 				free(pty_path);
 			}
 		}
Signature · 1 line
1.7.10.4
#1 lore raw
Quoted 19 lines · Matthew Daley

This stops pty_path from being leaked, and removes the toctou race, FWIW.

Not sure why it's a separate check to begin with...

Coverity-ID: 1056047

---
 tools/console/client/main.c |   10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
tools/console/client/main.c +0 -0
index 38c856a..c32d3eb 100644
@@ -116,12 +116,12 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
 			 * disambiguate: just read the pty path */
 			pty_path = xs_read(xs, XBT_NULL, path, &len);
 			if (pty_path != NULL) {

tools/console/client/main.c static int get_pty_fd(struct xs_handle *xs, char *path, int seconds) · line 123

Quoted 10 lines
tools/console/client/main.c +4 -5
-				if (access(pty_path, R_OK|W_OK) != 0)
-					continue;
 				pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-				if (pty_fd == -1) 
-					err(errno, "Could not open tty `%s'", 
-					    pty_path);
+				if (pty_fd == -1) {
+					if (errno != EACCES)
+						err(errno, "Could not open tty `%s'",
+							pty_path);

access() can fail for many more reasons than just EACCES. I would skip the errno check entirely and always print the error.

~Andrew

Quoted 4 lines
tools/console/client/main.c +1 -0
+				}
 				free(pty_path);
 			}
 		}
#2 lore raw
Quoted 25 lines · Andrew Cooper
Quoted 24 lines · Matthew Daley

This stops pty_path from being leaked, and removes the toctou race, FWIW.

Not sure why it's a separate check to begin with...

Coverity-ID: 1056047

---
 tools/console/client/main.c |   10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
tools/console/client/main.c +0 -4
index 38c856a..c32d3eb 100644
@@ -116,12 +116,12 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
                       * disambiguate: just read the pty path */
                      pty_path = xs_read(xs, XBT_NULL, path, &len);
                      if (pty_path != NULL) {
-                             if (access(pty_path, R_OK|W_OK) != 0)
-                                     continue;
                              pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-                             if (pty_fd == -1)
-                                     err(errno, "Could not open tty `%s'",
Quoted 8 lines

tools/console/client/main.c static int get_pty_fd(struct xs_handle *xs, char *path, int seconds) · line 123

Quoted 5 lines
tools/console/client/main.c +4 -1
-                                         pty_path);
+                             if (pty_fd == -1) {
+                                     if (errno != EACCES)
+                                             err(errno, "Could not open tty `%s'",
+                                                     pty_path);

access() can fail for many more reasons than just EACCES. I would skip the errno check entirely and always print the error.

err() doesn't return though (calls exit()). Given that, is always calling it still acceptable?

- Matthew

#3 lore raw
Quoted 34 lines · Matthew Daley
Quoted 24 lines · Andrew Cooper
Quoted 23 lines · Matthew Daley

This stops pty_path from being leaked, and removes the toctou race, FWIW.

Not sure why it's a separate check to begin with...

Coverity-ID: 1056047

---
 tools/console/client/main.c |   10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
tools/console/client/main.c +0 -3
index 38c856a..c32d3eb 100644
@@ -116,12 +116,12 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
                       * disambiguate: just read the pty path */
                      pty_path = xs_read(xs, XBT_NULL, path, &len);
                      if (pty_path != NULL) {
-                             if (access(pty_path, R_OK|W_OK) != 0)
-                                     continue;
                              pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-                             if (pty_fd == -1)
Quoted 8 lines

tools/console/client/main.c static int get_pty_fd(struct xs_handle *xs, char *path, int seconds) · line 123

Quoted 6 lines
tools/console/client/main.c +4 -2
-                                     err(errno, "Could not open tty `%s'",
-                                         pty_path);
+                             if (pty_fd == -1) {
+                                     if (errno != EACCES)
+                                             err(errno, "Could not open tty `%s'",
+                                                     pty_path);

access() can fail for many more reasons than just EACCES. I would skip the errno check entirely and always print the error.

Quoted 4 lines

err() doesn't return though (calls exit()). Given that, is always calling it still acceptable?

- Matthew

Hmm - that is a point.

Even as the patch currently stands, the behaviour of the loop has changed. Before, it would eat any error from access(), and only die in the toctou case where open() subsequently failed.

The code is waiting for an individual pty path found from xenstore. I would think that any errors resulting from a bad path being present in xenstore should probably count as fatal. After all, this is a xenconsole client asking xenconsoled which pty to connect to (which now I come to think of it, given an implicit assumption that the client is in the same domain as xenconsoled).

~Andrew

#4 lore raw

PATCH v2 1/1 xenconsole: merge pty access check into when it is opened

This stops pty_path from being leaked, and removes the toctou race, FWIW.

Not sure why it's a separate check to begin with...

Coverity-ID: 1056047

---
v2: err() if open() fails with EACCES as well

 tools/console/client/main.c |    7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
tools/console/client/main.c +2 -5
index 38c856a..3242008 100644
@@ -116,12 +116,9 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
 			 * disambiguate: just read the pty path */
 			pty_path = xs_read(xs, XBT_NULL, path, &len);
 			if (pty_path != NULL) {
-				if (access(pty_path, R_OK|W_OK) != 0)
-					continue;
 				pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-				if (pty_fd == -1) 
-					err(errno, "Could not open tty `%s'", 
-					    pty_path);
+				if (pty_fd == -1)
+					err(errno, "Could not open tty `%s'", pty_path);
 				free(pty_path);
 			}
 		}
Signature · 1 line
1.7.10.4
Reviewed-by on v2 1/1
#5 lore raw
Quoted 7 lines · Matthew Daley

This stops pty_path from being leaked, and removes the toctou race, FWIW.

Not sure why it's a separate check to begin with...

Coverity-ID: 1056047

Quoted 25 lines

---
v2: err() if open() fails with EACCES as well

 tools/console/client/main.c |    7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
tools/console/client/main.c +2 -5
index 38c856a..3242008 100644
@@ -116,12 +116,9 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
 			 * disambiguate: just read the pty path */
 			pty_path = xs_read(xs, XBT_NULL, path, &len);
 			if (pty_path != NULL) {
-				if (access(pty_path, R_OK|W_OK) != 0)
-					continue;
 				pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-				if (pty_fd == -1) 
-					err(errno, "Could not open tty `%s'", 
-					    pty_path);
+				if (pty_fd == -1)
+					err(errno, "Could not open tty `%s'", pty_path);
 				free(pty_path);
 			}
 		}
#6 lore raw
In reply to Matthew Daley

Pingy ping?

Quoted 35 lines · Matthew Daley

This stops pty_path from being leaked, and removes the toctou race, FWIW.

Not sure why it's a separate check to begin with...

Coverity-ID: 1056047

---
v2: err() if open() fails with EACCES as well

 tools/console/client/main.c |    7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
tools/console/client/main.c +2 -6
index 38c856a..3242008 100644
@@ -116,12 +116,9 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
                         * disambiguate: just read the pty path */
                        pty_path = xs_read(xs, XBT_NULL, path, &len);
                        if (pty_path != NULL) {
-                               if (access(pty_path, R_OK|W_OK) != 0)
-                                       continue;
                                pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-                               if (pty_fd == -1)
-                                       err(errno, "Could not open tty `%s'",
-                                           pty_path);
+                               if (pty_fd == -1)
+                                       err(errno, "Could not open tty `%s'", pty_path);
                                free(pty_path);
                        }
                }
--

1.7.10.4

#7 lore raw
In reply to Andrew Cooper

Andrew Cooper writes ("Re: [Xen-devel] [PATCH] xenconsole: merge pty access check into when it is opened"):

Quoted 7 lines · Matthew Daley
Quoted 6 lines
Quoted 2 lines · Andrew Cooper

access() can fail for many more reasons than just EACCES. I would skip the errno check entirely and always print the error.

err() doesn't return though (calls exit()). Given that, is always calling it still acceptable?

...

Quoted 3 lines

Even as the patch currently stands, the behaviour of the loop has changed. Before, it would eat any error from access(), and only die in the toctou case where open() subsequently failed.

I had the impression there is sometimes a race during domain startup, between a toolstack invoking xenconsole and xenconsoled allocating the pty.

Ian.

#8 lore raw
Quoted 9 lines · Ian Jackson

Andrew Cooper writes ("Re: [Xen-devel] [PATCH] xenconsole: merge pty access check into when it is opened"):

Quoted 7 lines · Matthew Daley
Quoted 6 lines
Quoted 2 lines · Andrew Cooper

access() can fail for many more reasons than just EACCES. I would skip the errno check entirely and always print the error.

err() doesn't return though (calls exit()). Given that, is always calling it still acceptable?

...

Quoted 8 lines
Quoted 3 lines

Even as the patch currently stands, the behaviour of the loop has changed. Before, it would eat any error from access(), and only die in the toctou case where open() subsequently failed.

I had the impression there is sometimes a race during domain startup, between a toolstack invoking xenconsole and xenconsoled allocating the pty.

Ian.

In which case the xenstore path should be NULL and we wander around loop and wait on the watch again.

~Andrew

#9 lore raw

Re: [PATCH] xenconsole: merge pty access check into when it is opened [and 1 more messages]

In reply to Andrew Cooper

Andrew Cooper writes ("Re: [Xen-devel] [PATCH] xenconsole: merge pty access check into when it is opened"):

Quoted 3 lines

Even as the patch currently stands, the behaviour of the loop has changed. Before, it would eat any error from access(), and only die in the toctou case where open() subsequently failed.

I concur, and v2 of the patch does the same. This is probably correct as Andrew and I discussed.

But the deliberate (or, at least, probably-desirable) behaviour change is not reflected in the commit message which only discusses removing the race and saving a leak.

Matthew Daley writes ("Re: [PATCH v2] xenconsole: merge pty access check into when it is opened"):

Quoted 1 line

Pingy ping?

Sorry for not making this clear sooner, but:

I think this patch is a bugfix and should go into 4.4 so I am happy to apply it but the commit message should be improved. Would you like to do that ?

Thanks,
Ian.

#10 lore raw

Re: [PATCH] xenconsole: merge pty access check into when it is opened [and 1 more messages]

Quoted 12 lines · Ian Jackson

Andrew Cooper writes ("Re: [Xen-devel] [PATCH] xenconsole: merge pty access check into when it is opened"):

Quoted 3 lines

Even as the patch currently stands, the behaviour of the loop has changed. Before, it would eat any error from access(), and only die in the toctou case where open() subsequently failed.

I concur, and v2 of the patch does the same. This is probably correct as Andrew and I discussed.

But the deliberate (or, at least, probably-desirable) behaviour change is not reflected in the commit message which only discusses removing the race and saving a leak.

Quoted 8 lines

Matthew Daley writes ("Re: [PATCH v2] xenconsole: merge pty access check into when it is opened"):

Quoted 1 line

Pingy ping?

Sorry for not making this clear sooner, but:

I think this patch is a bugfix and should go into 4.4 so I am happy to apply it but the commit message should be improved. Would you like to do that ?

Sure thing. v3 on its way...

- Matthew

Quoted 3 lines

Thanks,
Ian.

#11 lore raw

PATCH v3 1/1 xenconsole: adjust pty opening error checking and handling

In reply to Ian Jackson

Currently we check the pty path received from xenstore with access(); if it indicates that the pty is not accessible, we loop around and wait for a new path to appear in xenstore.

This has several issues:
* If a path has been written to xenstore, it can be assumed that that pty should already be accessible to xenconsole, and hence any error that occurs while trying to open it should be fatal and not ignored
* If access() indicates no access to the pty, the memory allocated for the path is leaked when going around the loop again
* The accessibility of the pty could change between the access() and open() calls, leading to a TOCTOU race (this is what Coverity is complaining about).

By removing the explicit access() check and just erroring out whenever open() fails, we fix all these issues.

Coverity-ID: 1056047

---
v3: Improve commit message to explain what behaviours are fixed (and hence changed)

 tools/console/client/main.c |    7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
tools/console/client/main.c +2 -5
index 38c856a..3242008 100644
@@ -116,12 +116,9 @@ static int get_pty_fd(struct xs_handle *xs, char *path, int seconds)
 			 * disambiguate: just read the pty path */
 			pty_path = xs_read(xs, XBT_NULL, path, &len);
 			if (pty_path != NULL) {
-				if (access(pty_path, R_OK|W_OK) != 0)
-					continue;
 				pty_fd = open(pty_path, O_RDWR | O_NOCTTY);
-				if (pty_fd == -1) 
-					err(errno, "Could not open tty `%s'", 
-					    pty_path);
+				if (pty_fd == -1)
+					err(errno, "Could not open tty `%s'", pty_path);
 				free(pty_path);
 			}
 		}
Signature · 1 line
1.7.10.4
Acked-by on v3 1/1
#12 lore raw

Re: [PATCH v3] xenconsole: adjust pty opening error checking and handling

Matthew Daley writes ("[PATCH v3] xenconsole: adjust pty opening error checking and handling"):

Quoted 3 lines

Currently we check the pty path received from xenstore with access(); if it indicates that the pty is not accessible, we loop around and wait for a new path to appear in xenstore.

Thanks, and sorry to have made such a meal of this.