Skip to contentSkip to conversation

9p: unsigned/signed wrap in p9/unix modes.

PATCH In review Geyslan G. Bem 2 participants 10 messages Oct 7, 2013 – Oct 29, 2013 3 messages not in the archive 9P file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 9p: unsigned/signed wrap in p9/unix modes.

Changes the sign type to unsigned, avoiding the possibility of wrap when ORing the p9 or unix bit modes.

---
 fs/9p/vfs_inode.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/9p/vfs_inode.c +3 -3
index b352457..574095e 100644
@@ -63,7 +63,7 @@ static const struct inode_operations v9fs_symlink_inode_operations;
 
 static u32 unixmode2p9mode(struct v9fs_session_info *v9ses, umode_t mode)
 {
-	int res;
+	u32 res;
 	res = mode & 0777;
 	if (S_ISDIR(mode))
 		res |= P9_DMDIR;
@@ -125,7 +125,7 @@ static int p9mode2perm(struct v9fs_session_info *v9ses,
 static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
 			       struct p9_wstat *stat, dev_t *rdev)
 {
-	int res;
+	umode_t res;
 	u32 mode = stat->mode;
 
 	*rdev = 0;
@@ -144,7 +144,7 @@ static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
 	else if ((mode & P9_DMDEVICE) && (v9fs_proto_dotu(v9ses))
 		 && (v9ses->nodev == 0)) {
 		char type = 0, ext[32];
-		int major = -1, minor = -1;
+		u32 major = 0, minor = 0;
 
 		strlcpy(ext, stat->extension, sizeof(ext));
 		sscanf(ext, "%c %u %u", &type, &major, &minor);
Signature · 1 line
1.8.4
#1 lore raw
Quoted 2 lines · Geyslan G. Bem

Changes the sign type to unsigned, avoiding the possibility of wrap when ORing the p9 or unix bit modes.

[]

fs/9p/vfs_inode.c

Quoted 1 line
fs/9p/vfs_inode.c +0 -0

[]

fs/9p/vfs_inode.c static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses, · line 150

Quoted 9 lines
fs/9p/vfs_inode.c +1 -1
@@ -144,7 +144,7 @@ static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
 	else if ((mode & P9_DMDEVICE) && (v9fs_proto_dotu(v9ses))
 		 && (v9ses->nodev == 0)) {
 		char type = 0, ext[32];
-		int major = -1, minor = -1;
+		u32 major = 0, minor = 0;
 
 		strlcpy(ext, stat->extension, sizeof(ext));
 		sscanf(ext, "%c %u %u", &type, &major, &minor);

This bit changes the MKDEV below it.

I would think that the sscanf return should be
checked for 3 and maybe MKDEV should not be
constructed when it's not.

#2 lore raw

Joe,

Thank you for reply.

What do you think about:

                 strncpy(ext, stat->extension, sizeof(ext));

+ if (sscanf(ext, "%c %u %u", &type, &major, &minor) < 3) {
+ p9_debug(P9_DEBUG_ERROR,
+ "It's necessary define type, major
and minor values when using P9_DMDEVICE");
+ return res;
+ }

                 switch (type) {
                 case 'c':
                         res |= S_IFCHR;
                         break;
...
                 *rdev = MKDEV(major, minor);

Geyslan Gregório Bem
hackingbits.com

2013/10/7 Joe Perches <[email protected]>:

Quoted 23 lines · Geyslan G. Bem
Quoted 2 lines

Changes the sign type to unsigned, avoiding the possibility of wrap when ORing the p9 or unix bit modes.

[]

fs/9p/vfs_inode.c

Quoted 1 line
fs/9p/vfs_inode.c +0 -0

[]

fs/9p/vfs_inode.c static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses, · line 150

Quoted 9 lines
fs/9p/vfs_inode.c +1 -1
@@ -144,7 +144,7 @@ static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
      else if ((mode & P9_DMDEVICE) && (v9fs_proto_dotu(v9ses))
               && (v9ses->nodev == 0)) {
              char type = 0, ext[32];
-             int major = -1, minor = -1;
+             u32 major = 0, minor = 0;
 
              strlcpy(ext, stat->extension, sizeof(ext));
              sscanf(ext, "%c %u %u", &type, &major, &minor);

This bit changes the MKDEV below it.

I would think that the sscanf return should be
checked for 3 and maybe MKDEV should not be
constructed when it's not.

#3 lore raw
Quoted 11 lines · Geyslan Gregório Bem

Joe,

Thank you for reply.

What do you think about:

                 strncpy(ext, stat->extension, sizeof(ext));
(quoted diff) +3 -0
+                 if (sscanf(ext, "%c %u %u", &type, &major, &minor) < 3) {
+                                  p9_debug(P9_DEBUG_ERROR,
+                                  "It's necessary define type, major

and minor values when using P9_DMDEVICE");

Review

Quoted 8 lines
(quoted diff) +2 -0
+                                  return res;
+                 }
                 switch (type) {
                 case 'c':
                         res |= S_IFCHR;
                         break;
...
                 *rdev = MKDEV(major, minor);

I think the plan 9 folk should figure out what's right.

#4 lore raw

Joe,

Nice, I'll wait their reply, there are other p9 patches that I have already sent and am awaiting Eric's.

Thank you again.

Geyslan Gregório Bem
hackingbits.com

2013/10/7 Joe Perches <[email protected]>:

Quoted 24 lines · Geyslan Gregório Bem
Quoted 11 lines

Joe,

Thank you for reply.

What do you think about:

                 strncpy(ext, stat->extension, sizeof(ext));
(quoted diff) +3 -0
+                 if (sscanf(ext, "%c %u %u", &type, &major, &minor) < 3) {
+                                  p9_debug(P9_DEBUG_ERROR,
+                                  "It's necessary define type, major

and minor values when using P9_DMDEVICE");

Review

Quoted 8 lines
(quoted diff) +2 -0
+                                  return res;
+                 }
                 switch (type) {
                 case 'c':
                         res |= S_IFCHR;
                         break;
...
                 *rdev = MKDEV(major, minor);

I think the plan 9 folk should figure out what's right.

A message that is not in the archive (sent off-list or not yet received): CAFkjPTmFOkw5n964zGuiON1oLSy1CsPQQ8TrGNho6L-+M5Z7SA@mail.gmail.com

#6 lore raw

2013/10/20 Eric Van Hensbergen <[email protected]>:

Quoted 18 lines

Please resubmit a clean patch which includes the check of sscanf for exactly the correct number of arguments and handles errors properly in other cases. That last bit may be a bit problematic since right now the only errors are prints and we seem to be otherwise silently failing. Of course, looks like nothing else is checking return values from that function for error. We could set rdev to an ERR_PTR(-EIO), and then go through and do a check in all the places which matter (looks like there are a few places where rdev just gets discarded -- might even be a good idea to not parse rdev unless we need to by passing NULL to p9mode2unixmode.

All in all, its a corner case which is only likely with a broken server, but the full clean up would seem to be:
a) switch to u32's
b) pass NULL when rdev just gets discarded and don't bother parsing when
it is
c) check the sscanf return validity
d) on error set ERR_PTR in rdev and check on return before proceeding

Quoted 3 lines

That's a lot of cleanup, I'll add it to my work queue if you don't have time to rework your patch.

Eric, I would like to try with your guidance.

Quoted 4 lines

For the other patches, anyone you didn't see a response from me on today is being pulled into my for-next queue. Thanks for the cleanups.

-eric

Thanks for accept them.

Quoted 46 lines
Quoted 38 lines · Geyslan Gregório Bem

Joe,

Nice, I'll wait their reply, there are other p9 patches that I have already sent and am awaiting Eric's.

Thank you again.

Geyslan Gregório Bem
hackingbits.com

2013/10/7 Joe Perches <[email protected]>:

Quoted 25 lines · Geyslan Gregório Bem
Quoted 12 lines

Joe,

Thank you for reply.

What do you think about:

                 strncpy(ext, stat->extension, sizeof(ext));

+ if (sscanf(ext, "%c %u %u", &type, &major, &minor) <
3) {

(quoted diff) +2 -0
+                                  p9_debug(P9_DEBUG_ERROR,
+                                  "It's necessary define type, major

and minor values when using P9_DMDEVICE");

Review

Quoted 8 lines
(quoted diff) +2 -0
+                                  return res;
+                 }
                 switch (type) {
                 case 'c':
                         res |= S_IFCHR;
                         break;
...
                 *rdev = MKDEV(major, minor);

I think the plan 9 folk should figure out what's right.

#7 lore raw

2013/10/21 Geyslan Gregório Bem <[email protected]>:

Quoted 23 lines

2013/10/20 Eric Van Hensbergen <[email protected]>:

Quoted 18 lines

Please resubmit a clean patch which includes the check of sscanf for exactly the correct number of arguments and handles errors properly in other cases. That last bit may be a bit problematic since right now the only errors are prints and we seem to be otherwise silently failing. Of course, looks like nothing else is checking return values from that function for error. We could set rdev to an ERR_PTR(-EIO), and then go through and do a check in all the places which matter (looks like there are a few places where rdev just gets discarded -- might even be a good idea to not parse rdev unless we need to by passing NULL to p9mode2unixmode.

All in all, its a corner case which is only likely with a broken server, but the full clean up would seem to be:
a) switch to u32's
b) pass NULL when rdev just gets discarded and don't bother parsing when
it is
c) check the sscanf return validity
d) on error set ERR_PTR in rdev and check on return before proceeding

Quoted 3 lines

That's a lot of cleanup, I'll add it to my work queue if you don't have time to rework your patch.

Quoted 26 lines

Eric, I would like to try with your guidance.

Quoted 4 lines

For the other patches, anyone you didn't see a response from me on today is being pulled into my for-next queue. Thanks for the cleanups.

-eric

Thanks for accept them.

Quoted 17 lines
Quoted 11 lines · Geyslan Gregório Bem

Joe,

Nice, I'll wait their reply, there are other p9 patches that I have already sent and am awaiting Eric's.

Thank you again.

Geyslan Gregório Bem
hackingbits.com

Let me know if I got your requests:

fs/9p/vfs_inode.c +0 -0
index 94de6d1..8a332d0 100644
@@ -63,7 +63,7 @@ static const struct inode_operations

v9fs_symlink_inode_operations;

static u32 unixmode2p9mode(struct v9fs_session_info *v9ses, umode_t mode) {
- int res;

+    u32 res;
     res = mode & 0777;
     if (S_ISDIR(mode))
         res |= P9_DMDIR;

@@ -125,7 +125,7 @@ static int p9mode2perm(struct v9fs_session_info *v9ses,
static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
struct p9_wstat *stat, dev_t *rdev)
{
- int res;

+    umode_t res;
     u32 mode = stat->mode;

     *rdev = 0;
@@ -144,10 +144,15 @@ static umode_t p9mode2unixmode(struct
v9fs_session_info *v9ses,
     else if ((mode & P9_DMDEVICE) && (v9fs_proto_dotu(v9ses))
          && (v9ses->nodev == 0)) {
         char type = 0, ext[32];

- int major = -1, minor = -1;
+ u32 major = 0, minor = 0;

         strlcpy(ext, stat->extension, sizeof(ext));

- sscanf(ext, "%c %u %u", &type, &major, &minor);
+ if (sscanf(ext, "%c %u %u", &type, &major, &minor) < 3) {
+ p9_debug(P9_DEBUG_ERROR,
+ "It's necessary define type [%u], major [u%] and minor [u%]" \
+ "values when using P9_DMDEVICE", type, major, minor);
+ goto err_dev;
+ }

         switch (type) {
         case 'c':
             res |= S_IFCHR;

@@ -158,11 +163,16 @@ static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
default:
p9_debug(P9_DEBUG_ERROR, "Unknown special type %c %s\n",
type, stat->extension);

+            goto err_dev;
         };
         *rdev = MKDEV(major, minor);
     } else
         res |= S_IFREG;

+ goto ret;

+err_dev:
+ rdev = ERR_PTR(-EIO);
+ret:

     return res;

}

@@ -460,13 +470,17 @@ void v9fs_evict_inode(struct inode *inode)

static int v9fs_test_inode(struct inode *inode, void *data) {
- int umode;

+    umode_t umode;
     dev_t rdev;
     struct v9fs_inode *v9inode = V9FS_I(inode);
     struct p9_wstat *st = (struct p9_wstat *)data;
     struct v9fs_session_info *v9ses = v9fs_inode2v9ses(inode);

     umode = p9mode2unixmode(v9ses, st, &rdev);

+
+ if (IS_ERR(rdev))
+ return 0;
+

     /* don't match inode of different type */
     if ((inode->i_mode & S_IFMT) != (umode & S_IFMT))
         return 0;
@@ -526,6 +540,11 @@ static struct inode *v9fs_qid_iget(struct super_block *sb,
      */
     inode->i_ino = i_ino;
     umode = p9mode2unixmode(v9ses, st, &rdev);

+ if (IS_ERR(rdev)) {
+ retval = rdev;
+ goto error;
+ }
+

     retval = v9fs_init_inode(v9ses, inode, umode, rdev);
     if (retval)
         goto error;

@@ -1461,9 +1480,10 @@ v9fs_vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rde

int v9fs_refresh_inode(struct p9_fid *fid, struct inode *inode) {
- int umode;

+    umode_t umode;
     dev_t rdev;
     loff_t i_size;
+    int ret = 0;
     struct p9_wstat *st;
     struct v9fs_session_info *v9ses;

@@ -1475,6 +1495,11 @@ int v9fs_refresh_inode(struct p9_fid *fid, struct inode *inode)

      * Don't update inode if the file type is different
      */
     umode = p9mode2unixmode(v9ses, st, &rdev);

+ if (IS_ERR(rdev)) {
+ ret = rdev;
+ goto out;
+ }
+

     if ((inode->i_mode & S_IFMT) != (umode & S_IFMT))
         goto out;

@@ -1491,7 +1516,7 @@ int v9fs_refresh_inode(struct p9_fid *fid, struct inode *inode)
out:

     p9stat_free(st);
     kfree(st);

- return 0;
+ return ret;
}

static const struct inode_operations v9fs_dir_inode_operations_dotu = {

Quoted 29 lines
Quoted 29 lines
Quoted 27 lines

2013/10/7 Joe Perches <[email protected]>:

Quoted 25 lines · Geyslan Gregório Bem
Quoted 12 lines

Joe,

Thank you for reply.

What do you think about:

                 strncpy(ext, stat->extension, sizeof(ext));

+ if (sscanf(ext, "%c %u %u", &type, &major, &minor) <
3) {

(quoted diff) +2 -0
+                                  p9_debug(P9_DEBUG_ERROR,
+                                  "It's necessary define type, major

and minor values when using P9_DMDEVICE");

Review

Quoted 8 lines
(quoted diff) +2 -0
+                                  return res;
+                 }
                 switch (type) {
                 case 'c':
                         res |= S_IFCHR;
                         break;
...
                 *rdev = MKDEV(major, minor);

I think the plan 9 folk should figure out what's right.

#8 lore raw

2013/10/21 Geyslan Gregório Bem <[email protected]>:

Quoted 179 lines

2013/10/21 Geyslan Gregório Bem <[email protected]>:

Quoted 23 lines

2013/10/20 Eric Van Hensbergen <[email protected]>:

Quoted 18 lines

Please resubmit a clean patch which includes the check of sscanf for exactly the correct number of arguments and handles errors properly in other cases. That last bit may be a bit problematic since right now the only errors are prints and we seem to be otherwise silently failing. Of course, looks like nothing else is checking return values from that function for error. We could set rdev to an ERR_PTR(-EIO), and then go through and do a check in all the places which matter (looks like there are a few places where rdev just gets discarded -- might even be a good idea to not parse rdev unless we need to by passing NULL to p9mode2unixmode.

All in all, its a corner case which is only likely with a broken server, but the full clean up would seem to be:
a) switch to u32's
b) pass NULL when rdev just gets discarded and don't bother parsing when
it is
c) check the sscanf return validity
d) on error set ERR_PTR in rdev and check on return before proceeding

Quoted 3 lines

That's a lot of cleanup, I'll add it to my work queue if you don't have time to rework your patch.

Quoted 26 lines

Eric, I would like to try with your guidance.

Quoted 4 lines

For the other patches, anyone you didn't see a response from me on today is being pulled into my for-next queue. Thanks for the cleanups.

-eric

Thanks for accept them.

Quoted 17 lines
Quoted 11 lines · Geyslan Gregório Bem

Joe,

Nice, I'll wait their reply, there are other p9 patches that I have already sent and am awaiting Eric's.

Thank you again.

Geyslan Gregório Bem
hackingbits.com

Let me know if I got your requests:

fs/9p/vfs_inode.c +1 -1
index 94de6d1..8a332d0 100644
@@ -63,7 +63,7 @@ static const struct inode_operations
 v9fs_symlink_inode_operations;
 
 static u32 unixmode2p9mode(struct v9fs_session_info *v9ses, umode_t mode)
 {
-    int res;
+    u32 res;
     res = mode & 0777;
     if (S_ISDIR(mode))
         res |= P9_DMDIR;
fs/9p/vfs_inode.c +8 -3
@@ -125,7 +125,7 @@ static int p9mode2perm(struct v9fs_session_info *v9ses,
 static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
                    struct p9_wstat *stat, dev_t *rdev)
 {
-    int res;
+    umode_t res;
     u32 mode = stat->mode;
 
     *rdev = 0;
@@ -144,10 +144,15 @@ static umode_t p9mode2unixmode(struct
 v9fs_session_info *v9ses,
     else if ((mode & P9_DMDEVICE) && (v9fs_proto_dotu(v9ses))
          && (v9ses->nodev == 0)) {
         char type = 0, ext[32];
-        int major = -1, minor = -1;
+        u32 major = 0, minor = 0;
 
         strlcpy(ext, stat->extension, sizeof(ext));
-        sscanf(ext, "%c %u %u", &type, &major, &minor);
+        if (sscanf(ext, "%c %u %u", &type, &major, &minor) < 3) {
+            p9_debug(P9_DEBUG_ERROR,
+                 "It's necessary define type [%u], major [u%] and minor [u%]" \
+                 "values when using P9_DMDEVICE", type, major, minor);
+            goto err_dev;
+        }
         switch (type) {
         case 'c':
             res |= S_IFCHR;
fs/9p/vfs_inode.c +23 -3
@@ -158,11 +163,16 @@ static umode_t p9mode2unixmode(struct
 v9fs_session_info *v9ses,
         default:
             p9_debug(P9_DEBUG_ERROR, "Unknown special type %c %s\n",
                  type, stat->extension);
+            goto err_dev;
         };
         *rdev = MKDEV(major, minor);
     } else
         res |= S_IFREG;
+    goto ret;
 
+err_dev:
+    rdev = ERR_PTR(-EIO);
+ret:
     return res;
 }
@@ -460,13 +470,17 @@ void v9fs_evict_inode(struct inode *inode)
 
 static int v9fs_test_inode(struct inode *inode, void *data)
 {
-    int umode;
+    umode_t umode;
     dev_t rdev;
     struct v9fs_inode *v9inode = V9FS_I(inode);
     struct p9_wstat *st = (struct p9_wstat *)data;
     struct v9fs_session_info *v9ses = v9fs_inode2v9ses(inode);
 
     umode = p9mode2unixmode(v9ses, st, &rdev);
+
+    if (IS_ERR(rdev))
+        return 0;
+
     /* don't match inode of different type */
     if ((inode->i_mode & S_IFMT) != (umode & S_IFMT))
         return 0;
@@ -526,6 +540,11 @@ static struct inode *v9fs_qid_iget(struct super_block *sb,
      */
     inode->i_ino = i_ino;
     umode = p9mode2unixmode(v9ses, st, &rdev);
+    if (IS_ERR(rdev)) {
+        retval = rdev;
+        goto error;
+    }
+
     retval = v9fs_init_inode(v9ses, inode, umode, rdev);
     if (retval)
         goto error;
@@ -1461,9 +1480,10 @@ v9fs_vfs_mknod(struct inode *dir, struct dentry
 *dentry, umode_t mode, dev_t rde
 
 int v9fs_refresh_inode(struct p9_fid *fid, struct inode *inode)
 {
-    int umode;
+    umode_t umode;
     dev_t rdev;
     loff_t i_size;
+    int ret = 0;
     struct p9_wstat *st;
     struct v9fs_session_info *v9ses;
@@ -1475,6 +1495,11 @@ int v9fs_refresh_inode(struct p9_fid *fid,
 struct inode *inode)
      * Don't update inode if the file type is different
      */
     umode = p9mode2unixmode(v9ses, st, &rdev);
+    if (IS_ERR(rdev)) {
+        ret = rdev;
+        goto out;
+    }
+
     if ((inode->i_mode & S_IFMT) != (umode & S_IFMT))
         goto out;
@@ -1491,7 +1516,7 @@ int v9fs_refresh_inode(struct p9_fid *fid,
 struct inode *inode)
 out:
     p9stat_free(st);
     kfree(st);
-    return 0;
+    return ret;
 }
 
Quoted 2 lines

static const struct inode_operations v9fs_dir_inode_operations_dotu = {

I didn't compile it. It's just a sketch.

The rdev tests (in the callers) must be:
if (rdev < 0) {
...

Waiting your reply.

Cheers.

Quoted 30 lines
Quoted 30 lines
Quoted 30 lines
Quoted 28 lines

2013/10/7 Joe Perches <[email protected]>:

Quoted 25 lines · Geyslan Gregório Bem
Quoted 12 lines

Joe,

Thank you for reply.

What do you think about:

                 strncpy(ext, stat->extension, sizeof(ext));

+ if (sscanf(ext, "%c %u %u", &type, &major, &minor) <
3) {

(quoted diff) +2 -0
+                                  p9_debug(P9_DEBUG_ERROR,
+                                  "It's necessary define type, major

and minor values when using P9_DMDEVICE");

Review

Quoted 8 lines
(quoted diff) +2 -0
+                                  return res;
+                 }
                 switch (type) {
                 case 'c':
                         res |= S_IFCHR;
                         break;
...
                 *rdev = MKDEV(major, minor);

I think the plan 9 folk should figure out what's right.

A message that is not in the archive (sent off-list or not yet received): CAFkjPTm_JhrWVzk0njqaUwpn5RxqYgRWYV2ZvPyrYoFKZHkgog@mail.gmail.com

A message that is not in the archive (sent off-list or not yet received): CAGG-pURzs512=6zu2_vBz5vtJu3Dm8hsdV=qZuUa5GJng6Eceg@mail.gmail.com

#11 lore raw

2013/10/28 Geyslan Gregório Bem <[email protected]>

Quoted 10 lines

2013/10/27 Eric Van Hensbergen <[email protected]>

Quoted 5 lines

Looks like the right approach. The one other optional thing I mentioned was support for passing NULL for rdev and not trying to parse the device info when rdev == NULL. Its a very slight optimization in the grand scheme of things, but would seem to be cleaner for the folks calling the function who don't touch rdev after the fact...

-eric

Great. Let me do the changes this afternoon.

Hi Eric and all.

You requested to avoid the parsing of device when rdev is NULL, all right? But I'm afraid that that manner the res (return value) can be returned wrong when the bit mode is a device. Well, I did some changes. In this new approach, when rdev is NULL, the function only doesn't make the device, but returns the res (umode_t) nicely.

Tell me if this approach is correct. Do I have to modify something else?

fs/9p/vfs_inode.c +0 -0
index 94de6d1..e3d56f1 100644
@@ -63,7 +63,7 @@ static const struct inode_operations

v9fs_symlink_inode_operations;

static u32 unixmode2p9mode(struct v9fs_session_info *v9ses, umode_t mode) {
- int res;

+    u32 res;
     res = mode & 0777;
     if (S_ISDIR(mode))
         res |= P9_DMDIR;

@@ -125,10 +125,9 @@ static int p9mode2perm(struct v9fs_session_info *v9ses,
static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
struct p9_wstat *stat, dev_t *rdev)
{
- int res;

+    umode_t res;
     u32 mode = stat->mode;
-    *rdev = 0;
     res = p9mode2perm(v9ses, stat);

     if ((mode & P9_DMDIR) == P9_DMDIR)
@@ -144,10 +143,15 @@ static umode_t p9mode2unixmode(struct
v9fs_session_info *v9ses,
     else if ((mode & P9_DMDEVICE) && (v9fs_proto_dotu(v9ses))
          && (v9ses->nodev == 0)) {
         char type = 0, ext[32];

- int major = -1, minor = -1;
+ u32 major = 0, minor = 0;

         strlcpy(ext, stat->extension, sizeof(ext));

- sscanf(ext, "%c %u %u", &type, &major, &minor);
+ if (sscanf(ext, "%c %u %u", &type, &major, &minor) < 3) {
+ p9_debug(P9_DEBUG_ERROR,
+ "It's necessary define type [%c], major [%u] and minor [%u]" \
+ "values when mode is P9_DMDEVICE\n", type, major, minor);
+ goto err_dev;
+ }

         switch (type) {
         case 'c':
             res |= S_IFCHR;

@@ -158,11 +162,18 @@ static umode_t p9mode2unixmode(struct v9fs_session_info *v9ses,
default:
p9_debug(P9_DEBUG_ERROR, "Unknown special type %c %s\n",
type, stat->extension);
- };
- *rdev = MKDEV(major, minor);
+ goto err_dev;

+        }
+        /* Only make device if rdev pointer isn't NULL */
+        if (rdev)
+            *rdev = MKDEV(major, minor);
     } else
         res |= S_IFREG;

-
+ goto ret;
+err_dev:
+ if (rdev)
+ rdev = ERR_PTR(-EIO);
+ret:

     return res;

}

@@ -460,13 +471,12 @@ void v9fs_evict_inode(struct inode *inode)

static int v9fs_test_inode(struct inode *inode, void *data) {

-    int umode;
-    dev_t rdev;
+    umode_t umode;
     struct v9fs_inode *v9inode = V9FS_I(inode);
     struct p9_wstat *st = (struct p9_wstat *)data;
     struct v9fs_session_info *v9ses = v9fs_inode2v9ses(inode);

- umode = p9mode2unixmode(v9ses, st, &rdev);

+    umode = p9mode2unixmode(v9ses, st, NULL);
     /* don't match inode of different type */
     if ((inode->i_mode & S_IFMT) != (umode & S_IFMT))
         return 0;
@@ -526,6 +536,10 @@ static struct inode *v9fs_qid_iget(struct super_block *sb,
      */
     inode->i_ino = i_ino;
     umode = p9mode2unixmode(v9ses, st, &rdev);

+ if (IS_ERR_VALUE(rdev)) {
+ retval = rdev;
+ goto error;
+ }

     retval = v9fs_init_inode(v9ses, inode, umode, rdev);
     if (retval)
         goto error;

@@ -1461,8 +1475,7 @@ v9fs_vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rde

int v9fs_refresh_inode(struct p9_fid *fid, struct inode *inode) {

-    int umode;
-    dev_t rdev;
+    umode_t umode;
     loff_t i_size;
     struct p9_wstat *st;
     struct v9fs_session_info *v9ses;
@@ -1474,7 +1487,7 @@ int v9fs_refresh_inode(struct p9_fid *fid,
struct inode *inode)
     /*
      * Don't update inode if the file type is different
      */
-    umode = p9mode2unixmode(v9ses, st, &rdev);
+    umode = p9mode2unixmode(v9ses, st, NULL);
     if ((inode->i_mode & S_IFMT) != (umode & S_IFMT))
         goto out;
Signature · 4 lines
Regards,

Geyslan G. Bem
hackingbits.com
#12 lore raw

2013/10/28 Geyslan Gregório Bem <[email protected]>:

Quoted 19 lines

2013/10/28 Geyslan Gregório Bem <[email protected]>

Quoted 10 lines

2013/10/27 Eric Van Hensbergen <[email protected]>

Quoted 5 lines

Looks like the right approach. The one other optional thing I mentioned was support for passing NULL for rdev and not trying to parse the device info when rdev == NULL. Its a very slight optimization in the grand scheme of things, but would seem to be cleaner for the folks calling the function who don't touch rdev after the fact...

-eric

Great. Let me do the changes this afternoon.

Hi Eric and all.

You requested to avoid the parsing of device when rdev is NULL, all right? But I'm afraid that that manner the res (return value) can be returned wrong when the bit mode is a device. Well, I did some changes. In this new approach, when rdev is NULL, the function only doesn't make the device, but returns the res (umode_t) nicely.

Quoted 7 lines

Tell me if this approach is correct. Do I have to modify something else?

--
Regards,

Geyslan G. Bem
hackingbits.com

Eric, I sent the new patch:
[PATCH] 9p: code refactor in vfs_inode.c

Signature · 4 lines
Regards,

Geyslan G. Bem
hackingbits.com