SMACK netfilter smacklabel socket match
Subsystems (2)
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): fa.f7jv/+EnhNJziduAqQS3XHiU6/[email protected]
A message that is not in the archive (sent off-list or not yet received): fa.1A5YyyPb1uCn//[email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
PATCH 1/1 SMACK netfilter smacklabel socket match
hello,
i was playing with smack, trying to do funny things
Alas, when I use a 'labelled process' and try to access internet, packet are dropped sooner or later (because of ip options)
I tried to
echo 0.0.0.0/0 @ > /smack/netlabel
with no success...
looking at security/smack/smack_lsm.c:smack_host_label the following lines
bestmask.s_addr = 0;
...
if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr)
continue;
if the dest we try to reach match the 0.0.0.0/0, this condition will be true (either because we have a better match or because, well (miap->s_addr | bestmask.s_addr) == bestmask.s_addr == 0
So let the 0.0.0.0/0 a chance!
I realize this patch is a little ugly, a cleaner way would be to insert struct smk_netlbladdr sorted from longest to smallest mask and break the loop as soon as we have a match... regards,
Etienne
------
security/smack/smack_lsm.c +1 -1
@@ -1540,7 +1540,7 @@ static char *smack_host_label(struct sockaddr_in *sip)* If the list entry mask is less specific than the best * already found this entry is uninteresting. */ -if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) +if ( ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) && (miap->s_addr | bestmask.s_addr) !=0 )continue ;/* * This is better than any entry found so far.
PATCH 1/1 SMACK smacklabel : apply &MASK to IP inserted in /smack/netlabel
hello,
with current code it is possible to insert inconsistent IP/mask in /smack/netlabel
before patch :
==============
root@etienne-desktop:/home/etienne/linux-2.6# cat /smack/netlabel 12.67.3.2/15 @ 12.67.3.1/15 @ 12.67.2.1/15 @ 12.67.2.1/16 @ 12.67.1.1/16 @ 0.0.0.0/0 @
the solution is to apply the mask to the IP inserted in /smack/netlabel
after the patch:
================
root@etienne-desktop:/home/etienne/linux-2.6# echo 12.67.3.2/15 @ > /smack/netlabel root@etienne-desktop:/home/etienne/linux-2.6# cat /smack/netlabel 12.67.0.0/15 @ root@etienne-desktop:/home/etienne/linux-2.6# echo 12.67.3.1/15 @ > /smack/netlabel root@etienne-desktop:/home/etienne/linux-2.6# cat /smack/netlabel 12.67.0.0/15 @ root@etienne-desktop:/home/etienne/linux-2.6# echo 12.67.3.3/15 @ > /smack/netlabel root@etienne-desktop:/home/etienne/linux-2.6# cat /smack/netlabel 12.67.0.0/15 @
regards,
Etienne
----
security/smack/smackfs.c +1 -0
@@ -765,6 +765,7 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, mask.s_addr |= bebits; bebits <<=1 ; } + newname.sin_addr.s_addr &= mask.s_addr;/* * Only allow one writer at a time. Writes should be * quite rare and small in any case.
Re: [PATCH] SMACK smacklabel : apply &MASK to IP inserted in /smack/netlabel
security/smack/smackfs.c static ssize_t smk_write_netlbladdr(struct file *file, · line 771
Quoted 13 lines · etienne
----
security/smack/smackfs.c +1
-0@@ -765,6 +765,7 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, mask.s_addr |= bebits; bebits <<=1 ; } + newname.sin_addr.s_addr &= mask.s_addr;/* * Only allow one writer at a time. Writes should be * quite rare and small in any case.
If you do this you can simplify some of the code in smack_host_label() by removing the code which applies the mask to the stored addresses when comparing addresses. There may be other places as well.
Signature · 2 lines
paul moore linux @ hp
Casey Schaufler
maintainerRe: [PATCH] SMACK smacklabel : apply &MASK to IP inserted in /smack/netlabel
Quoted 12 lines · Paul Moore
Quoted 10 lines · etienne
----
security/smack/smackfs.c +1
-0@@ -765,6 +765,7 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, mask.s_addr |= bebits; bebits <<=1 ; } + newname.sin_addr.s_addr &= mask.s_addr;
Quoted 10 lines
security/smack/smackfs.c static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, · line 771
Quoted 4 lines
security/smack/smackfs.c +0
-0/* * Only allow one writer at a time. Writes should be * quite rare and small in any case. If you do this you can simplify some of the code in
smack_host_label()by removing the code which applies the mask to the stored addresses when comparing addresses. There may be other places as well.
1234567890123456789012345678901234567890123456789012345678901234567890
Thank all of you for your kind suggestions. I'm in the process of cleaning up after a meltdown in the Smack test lab, but I will look into this as soon as I can. Did I ever say how much I dislike netmasks?
Re: [PATCH] SMACK smacklabel : apply &MASK to IP inserted in /smack/netlabel
In reply to Paul MooreQuoted 10 lines · Paul Moore
Quoted 9 lines · etienne
----
security/smack/smackfs.c +0
-0@@ -765,6 +765,7 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, mask.s_addr |= bebits; bebits <<=1 ; }
Quoted 9 lines
security/smack/smackfs.c static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, · line 771
Quoted 4 lines
security/smack/smackfs.c +1
-0+ newname.sin_addr.s_addr &= mask.s_addr;/* * Only allow one writer at a time. Writes should be * quite rare and small in any case. If you do this you can simplify some of the code in
smack_host_label()by removing the code which applies the mask to the stored addresses when comparing addresses. There may be other places as well.
you're right, in mk_write_netlbladdr also, i'll have a look thanks,
Etienne
From: etienne <[email protected]>
Date: Tue, 17 Feb 2009 21:01:15 +0100
security/smack/smack_lsm.c static char *smack_host_label(struct sockaddr_in *sip) · line 1543
Quoted 2 lines
security/smack/smack_lsm.c +1
-1-if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) +if ( ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) && (miap->s_addr | bestmask.s_addr) !=0 )
Please don't add all of those new spaces and please add a newline after the "&&" so that the line of code does not exceed 80 columns.
Quoted 3 lines · etienne
I realize this patch is a little ugly, a cleaner way would be to insert struct
smk_netlbladdrsorted from longest to smallest mask and break the loop as soon as we have a match... regards,
Agreed, the address matching code really should be improved; if you feel like you could contribute the changes I'm pretty sure Casey would welcome the patches :)
Regarding your fix below, I think a cleaner solution would be to do something like the following in place of the existing mask check ...
if ((miap->s_addr & bestmask.s_addr) || (bestmask.s_addr == 0)) {
bestmask.s_addr = miap->s_addr;
bestlabel = snp->smk_label;
} ... however there is one small problem with this approach (your proposal suffers from the same issue): normally the smack_host_label() code prefers the first matching entry in the list, the change above preserves that with the exception of a 0.0.0.0/0 entry. Granted, you shouldn't allow that in the first place but I believe it is possible so it is something that needs to be taken into consideration.
Quoted 20 lines
Signed-off-by Etienne <[email protected]>------
security/smack/smack_lsm.c +0
-0@@ -1540,7 +1540,7 @@ static char *smack_host_label(struct sockaddr_in *sip) * If the list entry mask is less specific than the best * alreadyfound this entry is uninteresting.
*/security/smack/smack_lsm.c +1
-1-if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) + if ( ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr)&& (
miap->s_addr| bestmask.s_addr) != 0 ) continue;
/*
* This is better than any entry found so far.
--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Signature · 2 lines
paul moore linux @ hp
Quoted 9 lines · Paul Moore
Quoted 3 lines · etienne
I realize this patch is a little ugly, a cleaner way would be to insert struct
smk_netlbladdrsorted from longest to smallest mask and break the loop as soon as we have a match... regards,Agreed, the address matching code really should be improved; if you feel like you could contribute the changes I'm pretty sure Casey would welcome the patches :)
yes I could try that, this week-end maybe
Quoted 8 lines
Regarding your fix below, I think a cleaner solution would be to do something like the following in place of the existing mask check ...
if ((miap->s_addr & bestmask.s_addr) || (bestmask.s_addr == 0)) { bestmask.s_addr = miap->s_addr; bestlabel = snp->smk_label; }
Quoted 7 lines
... however there is one small problem with this approach (your proposal suffers from the same issue): normally the
smack_host_label()code prefers the first matching entry in the list, the change above preserves that with the exception of a 0.0.0.0/0 entry. Granted, you shouldn't allow that in the first place but I believe it is possible so it is something that needs to be taken into consideration.
hummm... I didn't see it that way; I think this function is basically a reimplementation of IPv4 classless routing (longest match first)?
anyway, I think the cleanest way would be to, well, sort smk_netlbladdr by mask on insertion (perf doesn't matter here) and this way smack_host_label can stop the loop on first match. Plus, it would give a nicer /smack/netlabel ouptut :)
so, how should we handle it? apply the patches (with whitespaces damages corrected ;) ) now (as it corrects a bug) an elaborate the cleaner way later? I think this should go to stable too?
regards
Etienne
Quoted 15 lines
Quoted 15 lines
Signed-off-by Etienne <[email protected]>------
security/smack/smack_lsm.c +0
-0@@ -1540,7 +1540,7 @@ static char *smack_host_label(struct sockaddr_in *sip) * If the list entry mask is less specific than the best * alreadyfound this entry is uninteresting.
*/security/smack/smack_lsm.c +1
-1-if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) + if ( ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr)&& (
miap->s_addr| bestmask.s_addr) != 0 ) continue;
/*
* This is better than any entry found so far.
Quoted 4 lines · etienne
... anyway, I think the cleanest way would be to, well, sort
smk_netlbladdrby mask on insertion (perf doesn't matter here) and this waysmack_host_labelcan stop the loop on first match. Plus, it would give a nicer /smack/netlabel ouptut :)
Agreed.
Quoted 3 lines
so, how should we handle it? apply the patches (with whitespaces damages corrected ;) ) now (as it corrects a bug) an elaborate the cleaner way later?
Well, since you have some time and willingness to do things "the right way" I would recommend dropping these patches (which are really just band-aids) and working on the right solution to stored the addresses/masks in a sorted list with the mask already applied.
FWIW, the NetLabel code (net/netlabel) has to do very similar things with sorted address lists so I built an address list construct which builds on the list.h ideas and operates in a similar way. You may find it helpful.
Quoted 1 line
I think this should go to stable too?
I would worry about getting the patches developed, tested and in an acceptable form first, then we can worry about where they should be applied ;)
Signature · 2 lines
paul moore linux @ hp
Casey Schaufler
maintainerQuoted 10 lines · Paul Moore
Quoted 5 lines · etienne
... anyway, I think the cleanest way would be to, well, sort
smk_netlbladdrby mask on insertion (perf doesn't matter here) and this waysmack_host_labelcan stop the loop on first match. Plus, it would give a nicer /smack/netlabel ouptut :)Agreed.
Yes, it would make it nicer. You'll need to do a better job on the list management than I've been doing. It's probably well past time to introduce the Standard list management scheme to Smack, and you'll need to do so if you want to do insertions and/or deletions.
Quoted 15 lines
Quoted 4 lines
so, how should we handle it? apply the patches (with whitespaces damages corrected ;) ) now (as it corrects a bug) an elaborate the cleaner way later?
Well, since you have some time and willingness to do things "the right way" I would recommend dropping these patches (which are really just band-aids) and working on the right solution to stored the addresses/masks in a sorted list with the mask already applied.
FWIW, the NetLabel code (net/netlabel) has to do very similar things with sorted address lists so I built an address list construct which builds on the
list.hideas and operates in a similar way. You may find it helpful.
Quoted 7 lines
Quoted 2 lines
I think this should go to stable too?
I would worry about getting the patches developed, tested and in an acceptable form first, then we can worry about where they should be applied ;)
I would be delighted to see these changes. When you have preliminary versions I would be eager to see them and give them a try in the Smack test laboratory.
Etienne, thank you very much for the work you've done so far. Paul, thank you for your recommendations.
Quoted 2 lines · Casey Schaufler
Quoted 1 line · Paul Moore
On Wednesday 18 February 2009 02:23:24 am etienne wrote:
....
Quoted 6 lines
Yes, it would make it nicer. You'll need to do a better job on the list management than I've been doing. It's probably well past time to introduce the Standard list management scheme to Smack, and you'll need to do so if you want to do insertions and/or deletions.
well, we could maybe do that for smack_netlbladdrs.
for smk_rules, i don't know, depending to the use case, it could grow bigger and thus need a more efficient scheme than linked-list like hash-table.
[..]
Quoted 6 lines
Quoted 1 line
I would be delighted to see these changes. When you have preliminary versions I would be eager to see them and give them a try in the Smack test laboratory.
OK, will send tonight
Quoted 3 lines
Etienne, thank you very much for the work you've done so far. Paul, thank you for your recommendations.
well, I'll try to explain my use case for SMACK, could you please tell me if this makes sense and if it is doable and sane with SMACK :
I have single-user computer that, for simplicity sake, do only web browsing with firefox;
the attack vector i'm concerned with is malicious web pages, that could execute malicious code on my computer or worse erase some of my data;
so i express the following security policy in a tool-agnostic way :
1. firefox can access internet
2. firefox can read/write it's configuration directory in my $HOME
3. firefox can read/write to a download directory
4. firefox can execute kpdf, okular, vlc etc...
5. firefox can read system files
6. firefox can write to temporary folder
pretty simple. So I expect the 'tool' to express this policy in very few line; (i had a look at selinux/refpolicy, and I'm ashamed I was too lazy to test/understand further). And if possible a mainline tool would be a big bonus.
So I decided to give smack a try, and here are my notes/interrogations :
rule 1. if i understand correctly, I have to load the following smack rule
"firefox _ rwx"
well, as '_' is the default objectlabel for all system files, it means that firefox will have smack 'w' access on system.
So first issue : is it possible to express network access in another way? Or maybe I have to relabel /bin/, /sbin etc with a custom system label ?
rule 2-6 : easy to implement with smack, i label my $HOME with some label and download/cfg dir with other labels Firefox won't have rw access to my $HOME hehe
Second issue : what is the simplest way to start firefox with the firefox label?
I used the following hack : write a small program (i used cap_mac_admin, could have been suid) that :
a) set /proc/self/attr/current
b) drop capabilities
c) start firefox
Is there a cleanest way, can a process be started with its objectlabel?
Third issue : there seems to be no way to log/audit access violations, have you plans to implement that?
best regards,
Etienne
Quoted 11 lines · etienne
Quoted 5 lines · Casey Schaufler
Yes, it would make it nicer. You'll need to do a better job on the list management than I've been doing. It's probably well past time to introduce the Standard list management scheme to Smack, and you'll need to do so if you want to do insertions and/or deletions.
well, we could maybe do that for
smack_netlbladdrs.
forsmk_rules, i don't know, depending to the use case, it could grow bigger and thus need a more efficient scheme than linked-list like hash-table.
The code is easily changed because it is private to Smack and we don't have to worry about backwards compatibility issues. I would focus on improving the linked list approach (masked, sorted, etc.) and when traversing the list becomes a bottleneck we can look at alternative approaches. Others may have a better view, but from what I've seen the general approach taken in Linux Kernel optimization is to develop something that works then refine and optimize it once you have a better understanding of the common use cases.
Signature · 2 lines
paul moore linux @ hp
Casey Schaufler
maintainerQuoted 5 lines · etienne
...
Quoted 3 lines
Etienne, thank you very much for the work you've done so far. Paul, thank you for your recommendations.
Quoted 8 lines
well, I'll try to explain my use case for SMACK, could you please tell me if this makes sense and if it is doable and sane with SMACK :
I have single-user computer that, for simplicity sake, do only web browsing with firefox;
the attack vector i'm concerned with is malicious web pages, that could execute malicious code on my computer or worse erase some of my data;so i express the following security policy in a tool-agnostic way :
1. firefox can access internet
In Smack terms then you want the process label of your browser process to have access to hosts on the internet in general. The easy way to do this is for it to run with the ambient label (cat /smack/ambient to see it) which will be the floor label "_" unless you change it. Note that your browser will need to talk to the X11 server as well, so processes with the label of the browser will need write access to processes with the label of the X11 server, and visa versa.
Quoted 2 lines
2. firefox can read/write it's configuration directory in my $HOME
The label of the process will have to match the label of that directory for this to work.
Quoted 2 lines
3. firefox can read/write to a download directory
Again, the label will have to match.
Quoted 2 lines
4. firefox can execute kpdf, okular, vlc etc...
All these files will have the floor label "_" by design, so this is easy,
Quoted 2 lines
5. firefox can read system files
Again, system files will have the floor label, so this is easy.
Quoted 2 lines
6. firefox can write to temporary folder
Do you need to use /tmp, or does firefox respect $TMPDIR? You can set the label of /tmp to the star "*" label if worse comes to worst.
Quoted 1 line
pretty simple. So I expect the 'tool' to express this policy in very few line; (i had a look at selinux/refpolicy, and I'm ashamed I was too lazy to test/understand further).
Don't be ashamed. I wrote Smack because I was too lazy to figure out SELinux policy.
Quoted 11 lines
And if possible a mainline tool would be a big bonus.
So I decided to give smack a try, and here are my notes/interrogations :
rule 1. if i understand correctly, I have to load the following smack rule
"firefox _ rwx"
well, as '_' is the default objectlabel for all system files, it means that firefox will have smack 'w' access on system.So first issue : is it possible to express network access in another way? Or maybe I have to relabel
/bin/, /sbin etc with a custom system label ?
If you want firefox to talk on the internet, and have no other processes talk on the internet including the X11 server, you need to run firefox with a different label from everything else. This will make it difficult to talk to the X11 server.
Quoted 11 lines
rule 2-6 : easy to implement with smack, i label my $HOME with some label and download/cfg dir with other labels Firefox won't have rw access to my $HOME hehe
Second issue : what is the simplest way to start firefox with the firefox label?
I used the following hack : write a small program (i usedcap_mac_admin, could have been suid) that :
a) set/proc/self/attr/current
b) drop capabilities
c) start firefox
Is there a cleanest way, can a process be started with its objectlabel?
I have a newsmack program, but all that it does is what your "hack" does.
Quoted 2 lines
Third issue : there seems to be no way to log/audit access violations, have you plans to implement that?
Hmm. Audit should be working.
Quoted 10 lines
best regards,
Etienne
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Quoted 14 lines · Casey Schaufler
Quoted 4 lines · etienne
...
Quoted 3 lines
Etienne, thank you very much for the work you've done so far. Paul, thank you for your recommendations.
Quoted 8 lines
well, I'll try to explain my use case for SMACK, could you please tell me if this makes sense and if it is doable and sane with SMACK :
I have single-user computer that, for simplicity sake, do only web browsing with firefox;
the attack vector i'm concerned with is malicious web pages, that could execute malicious code on my computer or worse erase some of my data;so i express the following security policy in a tool-agnostic way :
1. firefox can access internet
Quoted 8 lines
In Smack terms then you want the process label of your browser process to have access to hosts on the internet in general. The easy way to do this is for it to run with the ambient label (cat /smack/ambient to see it) which will be the floor label "_" unless you change it. Note that your browser will need to talk to the X11 server as well, so processes with the label of the browser will need write access to processes with the label of the X11 server, and visa versa.
OK
Quoted 3 lines
Quoted 3 lines
2. firefox can read/write it's configuration directory in my $HOME
[snip]
Quoted 5 lines
Do you need to use /tmp, or does firefox respect $TMPDIR? You can set the label of /tmp to the star "*" label if worse comes to worst.
i don't really know now, i label /tmp/ /var/tmp with *
Quoted 5 lines
Quoted 1 line
pretty simple. So I expect the 'tool' to express this policy in very few line; (i had a look at selinux/refpolicy, and I'm ashamed I was too lazy to test/understand further).
Don't be ashamed. I wrote Smack because I was too lazy to figure out SELinux policy.
:-)
Quoted 4 lines
I have a newsmack program, but all that it does is what your "hack" does.
OK then. If it's the only way
Quoted 5 lines
Quoted 2 lines
Third issue : there seems to be no way to log/audit access violations, have you plans to implement that?
Hmm. Audit should be working.
I see some "audit" hook in the code, but i don't see a way to log _specific_ smack information ie
"smack_subject smack_object smack_access drop" (+of course process name, pid, path, and any relevant info)
like selinux would do in 'avc_audit'
regards
Etienne
hello,
Paul Moore wrote:
..
Quoted 5 lines
Well, since you have some time and willingness to do things "the right way" I would recommend dropping these patches (which are really just band-aids) and working on the right solution to stored the addresses/masks in a sorted list with the mask already applied.
OK, I'm about to send a new patch; but while testing my patches and reading code, I noticed another bug :
In smackfs.c:smk_write_netlbladdr
the netmask mask.s_addr is not handled correctly, the netmask should be :
1- computed in u32
2- converted to be32 !!
with current code, a "pseudo u32 mask" is applied to a be32 ipaddr; it occurs to works for "common netmasks" (multiple of 8), not for "intermediate" mask (/15, /25)
Quoted 4 lines
FWIW, the NetLabel code (net/netlabel) has to do very similar things with sorted address lists so I built an address list construct which builds on the
list.hideas and operates in a similar way. You may find it helpful.
OK, I tested some code in userspace and when i was confident enough coded it to kernel
Quoted 5 lines
Quoted 1 line
I think this should go to stable too?
I would worry about getting the patches developed, tested and in an acceptable form first, then we can worry about where they should be applied ;)
OK :)
Casey Schaufler
maintainerQuoted 11 lines · etienne
hello,
Paul Moore wrote:
..Quoted 6 lines
Well, since you have some time and willingness to do things "the right way" I would recommend dropping these patches (which are really just band-aids) and working on the right solution to stored the addresses/masks in a sorted list with the mask already applied.
Quoted 8 lines
OK, I'm about to send a new patch; but while testing my patches and reading code, I noticed another bug :
In smackfs.c:smk_write_netlbladdr
the netmask mask.s_addr is not handled correctly, the netmask should be :
1- computed in u32
2- converted to be32 !!
with current code, a "pseudo u32 mask" is applied to a be32 ipaddr; it occurs to works for "common netmasks" (multiple of 8), not for "intermediate" mask (/15, /25)
Well, that's embarrassing. I am really looking forward to your fixes!
PATCH 1/1 SMACK netlabel fixes
Hello,
the following patch (against 2.6.29rc5) fixes a few issues in the smack/netlabel area :
1) smack_host_label disregard a "0.0.0.0/0 @" rule (or other label), preventing 'tagged' tasks to access Internet (many systems drop packets with IP options)
2) netmasks were not handled correctly, they were stored in a way _not equivalent_ to conversion to be32 (it was equivalent for /0, /8, /16, /24, /32 masks but not other masks)
3) smack_netlbladdr prefixes (IP/mask) were not consistent (mask&IP was not done), so there could have been different list entries for the same IP prefix; if those entries had different labels, well ...
4) they were not sorted
1) 2) 3) are bugs, 4) is a more cosmetic issue.
The patch :
-creates a new helper smk_netlbladdr_insert to insert a smk_netlbladdr, sorted by netmask length -use the new sorted nature of smack_netlbladdrs list to simplify smack_host_label :
the first match _will_ be the more specific
-corrects endianness issues in smk_write_netlbladdr & netlbladdr_seq_show
The patch are "tested" so that they no crash the system; cat /smack/netlabel is now sorted and always consistent. Some basics ping tests to '@' and other label combination seems ok
See an extract of my tests bellow the patch
regards,
Etienne
--- security/smack/smack_lsm.c | 38 ++++++---------------------- security/smack/smackfs.c | 60 +++++++++++++++++++++++++++++++++---------- 2 files changed, 54 insertions(+), 44 deletions(-)
security/smack/smack_lsm.c +8 -30
@@ -1498,10 +1498,7 @@ static int smack_socket_post_create(struct socket *sock, int family,* looks for host based access restrictions * * This version will only be appropriate for really small -* sets of single label hosts. Because of the masking -* it cannot shortcut out on the first match. There are -* numerious ways to address the problem, but none of them -* have been applied here. +* sets of single label hosts. * * Returns the label of the far end or NULL if it's not special. */
@@ -1512,41 +1509,22 @@ static char *smack_host_label(struct sockaddr_in *sip)struct in_addr *siap = &sip->sin_addr;struct in_addr *liap;struct in_addr *miap; -struct in_addr bestmask;if (siap->s_addr ==0 )return NULL ; - bestmask.s_addr =0 ; -for (snp = smack_netlbladdrs; snp !=NULL ; snp = snp->smk_next) { liap = &snp->smk_host.sin_addr; miap = &snp->smk_mask;/* -* If the addresses match after applying the list entry mask -* the entry matches the address. If it doesn't move along to -* the next entry. -*/ -if ((liap->s_addr & miap->s_addr) != - (siap->s_addr & miap->s_addr)) -continue ; -/* -* If the list entry mask identifies a single address -* it can't get any more specific. -*/ -if (miap->s_addr ==0xffffffff ) -return snp->smk_label; -/* -* If the list entry mask is less specific than the best -* already found this entry is uninteresting. +* we break after finding the first match because +* the list is sorted from longest to shortest mask +* so we have found the most specific match */ -if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) -continue ; -/* -* This is better than any entry found so far. -*/ - bestmask.s_addr = miap->s_addr; - bestlabel = snp->smk_label; +if (liap->s_addr == (siap->s_addr & miap->s_addr)) { + bestlabel = snp->smk_label; +break ; + } }return bestlabel;
security/smack/smackfs.c +46 -14
107 lines not on this page · raw message
=============================================================
TESTS
=============================================================
root@etienne-desktop:/home/etienne/linux-2.6# cat /smack/netlabel 212.180.1.0/26 toto 212.180.1.0/25 @ 217.146.186.0/24 _ 212.180.0.0/23 tit 212.180.0.0/15 @
root@etienne-desktop:/home/etienne/linux-2.6# cat /proc/self/attr/current _root@etienne-desktop:/home/etienne/linux-2.6# root@etienne-desktop:/home/etienne/linux-2.6# ping 212.180.1.65 -c 1 PING 212.180.1.65 (212.180.1.65) 56(84) bytes of data. 64 bytes from 212.180.1.65: icmp_seq=1 ttl=56 time=7.04 ms
--- 212.180.1.65 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 7.040/7.040/7.040/0.000 ms
root@etienne-desktop:/home/etienne/linux-2.6# ping 212.180.1.1 -c 1 Do you want to ping broadcast? Then -b root@etienne-desktop:/home/etienne/linux-2.6# echo toto > /proc/self/attr/current root@etienne-desktop:/home/etienne/linux-2.6# ping 212.180.1.1 -c 1 PING 212.180.1.1 (212.180.1.1) 56(84) bytes of data. 64 bytes from 212.180.1.1: icmp_seq=1 ttl=248 time=13.6 ms
--- 212.180.1.1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 13.632/13.632/13.632/0.000 ms
root@etienne-desktop:/home/etienne/linux-2.6# ping 212.180.1.65 -c 1 PING 212.180.1.65 (212.180.1.65) 56(84) bytes of data. 64 bytes from 212.180.1.65: icmp_seq=1 ttl=56 time=9.87 ms
--- 212.180.1.65 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 9.876/9.876/9.876/0.000 ms
root@etienne-desktop:/home/etienne/linux-2.6# ping 212.180.0.1 -c 1 Do you want to ping broadcast? Then -b root@etienne-desktop:/home/etienne/linux-2.6# echo titi > /proc/self/attr/current root@etienne-desktop:/home/etienne/linux-2.6# ping 212.180.0.1 -c 1 Do you want to ping broadcast? Then -b root@etienne-desktop:/home/etienne/linux-2.6# echo tit > /proc/self/attr/current root@etienne-desktop:/home/etienne/linux-2.6# ping 212.180.0.1 -c 1 PING 212.180.0.1 (212.180.0.1) 56(84) bytes of data. ^C --- 212.180.0.1 ping statistics --- 1 packets transmitted, 0 received, 100% packet loss, time 0ms
Casey Schaufler
maintainerRe: [PATCH] SMACK netlabel fixes
Quoted 15 lines · etienne
Hello,
the following patch (against 2.6.29rc5) fixes a few issues in the smack/netlabel area :
1)smack_host_labeldisregard a "0.0.0.0/0 @" rule (or other label), preventing 'tagged' tasks to access Internet (many systems drop packets with IP options)
2) netmasks were not handled correctly, they were stored in a way_notequivalent_to conversion to be32 (it was equivalent for /0, /8, /16, /24, /32 masks but not other masks)
3)smack_netlbladdrprefixes (IP/mask) were not consistent (mask&IP was not done), so there could have been different list entries for the same IP prefix; if those entries had different labels, well ...
4) they were not sorted1) 2) 3) are bugs, 4) is a more cosmetic issue.
The patch :(quoted diff) +0
-3-creates a new helper smk_netlbladdr_insert to insert a smk_netlbladdr, sorted by netmask length -use the new sorted nature of smack_netlbladdrs list to simplify smack_host_label : the first match _will_ be the more specific -corrects endianness issues in smk_write_netlbladdr & netlbladdr_seq_show
Quoted 7 lines
The patch are "tested" so that they no crash the system; cat /smack/netlabel is now sorted and always consistent. Some basics ping tests to '@' and other label combination seems ok
See an extract of my tests bellow the patchregards,
Etienne
I am in the process of configuring the Smack test lab so that I can bang on this a little. It looks good so far. Thank you.
Re: [PATCH] SMACK netlabel fixes
In reply to etienneQuoted 20 lines · etienne
security/smack/smack_lsm.c +1
-4@@ -1498,10 +1498,7 @@ static int smack_socket_post_create(struct socket*sock, int family, * looks for host based access restrictions * * This version will only be appropriate for really small -* sets of single label hosts. Because of the masking -* it cannot shortcut out on the first match. There are -* numerious ways to address the problem, but none of them -* have been applied here. +* sets of single label hosts. * * Returns the label of the far end or NULL if it's not special. */ @@ -1512,41 +1509,22 @@ static char *smack_host_label(struct sockaddr_in *sip) struct in_addr *siap = &sip->sin_addr;struct in_addr *liap;struct in_addr *miap;
security/smack/smack_lsm.c static char *smack_host_label(struct sockaddr_in · line 1518
Quoted 10 lines
security/smack/smack_lsm.c +0
-3-struct in_addr bestmask;if (siap->s_addr ==0 )return NULL ; - bestmask.s_addr =0 ; -for (snp = smack_netlbladdrs; snp !=NULL ; snp = snp->smk_next) { liap = &snp->smk_host.sin_addr; miap = &snp->smk_mask;
Unless I'm mistaken the 'liap' and 'miap' variables are only used once inside the for loop, why not remove them and simply reference 'snp' directly?
Quoted 21 lines
security/smack/smack_lsm.c +3
-16/* -* If the addresses match after applying the list entry mask -* the entry matches the address. If it doesn't move along to -* the next entry. -*/ -if ((liap->s_addr & miap->s_addr) != - (siap->s_addr & miap->s_addr)) -continue ; -/* -* If the list entry mask identifies a single address -* it can't get any more specific. -*/ -if (miap->s_addr ==0xffffffff ) -return snp->smk_label; -/* -* If the list entry mask is less specific than the best -* already found this entry is uninteresting. +* we break after finding the first match because +* the list is sorted from longest to shortest mask +* so we have found the most specific match */
security/smack/smack_lsm.c static char *smack_host_label(struct sockaddr_in *sip) · line 1526
Quoted 10 lines
security/smack/smack_lsm.c +3
-7-if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) -continue ; -/* -* This is better than any entry found so far. -*/ - bestmask.s_addr = miap->s_addr; - bestlabel = snp->smk_label; +if (liap->s_addr == (siap->s_addr & miap->s_addr)) { + bestlabel = snp->smk_label; +break ;
This is being a bit nit-picky, but why not get rid of 'bestlabel' completely and instead of breaking here simply reutn 'snp->smk_label'? If we ever reach the end of the function (no match) just return NULL.
Quoted 4 lines
security/smack/smack_lsm.c +1
-0+ } }return bestlabel;
...
security/smack/smack_lsm.c static int netlbladdr_seq_show(struct seq_file *s, · line 665
Quoted 12 lines
security/smack/smack_lsm.c +2
-4@@ -663,12 +659,10 @@ static int netlbladdr_seq_show(struct seq_file *s, void *v) {struct smk_netlbladdr *skp = (struct smk_netlbladdr *) v;unsigned char *hp = (char *) &skp->smk_host.sin_addr.s_addr; -__be32 bebits;int maskn =0 ; +u32 temp_mask =be32_to_cpu (skp->smk_mask.s_addr); -for (bebits =BEMASK ; bebits !=0 ; maskn++, bebits <<=1 ) -if ((skp->smk_mask.s_addr & bebits) ==0 ) -break ; +for ( ; temp_mask; temp_mask <<=1 , maskn++);
More nit-picky stuff :) Why not set 'maskn' to zero inside the for loop construct instead of at the top of the function? There is less chance for error this way if someone else touches the code.
Quoted 2 lines
security/smack/smack_lsm.c +0
-0seq_printf (s,"%u.%u.%u.%u/%d %s\n" , hp[0 ], hp[1 ], hp[2 ], hp[3 ], maskn, skp->smk_label);
...
security/smack/smack_lsm.c static int smk_open_netlbladdr(struct inode *inode, · line 707
Quoted 13 lines
security/smack/smack_lsm.c +9
-0@@ -702,6 +696,40 @@ static int smk_open_netlbladdr(struct inode *inode, struct file *file) }/** +* smk_netlbladdr_insert +* @new : netlabel to insert +* +* This helper insert netlabel in the smack_netlbladdrs list +* sorted by netmask length (longest to smallest) +*/ +static void smk_netlbladdr_insert (struct smk_netlbladdr *new) +{ +struct smk_netlbladdr *m;
An empty line here might be nice.
security/smack/smack_lsm.c static int smk_open_netlbladdr(struct inode *inode, · line 711
Quoted 4 lines
security/smack/smack_lsm.c +4
-0+if (smack_netlbladdrs ==NULL ) { + smack_netlbladdrs = new; +return ; + }
I would prefer another one here, if that is okay with you.
security/smack/smack_lsm.c static int smk_open_netlbladdr(struct inode *inode, · line 712
Quoted 1 line
security/smack/smack_lsm.c +1
-0+/** the comparison '>' is a bit hacky, but works */
Why start the comment with '/**', using a single '*' works just fine.
Quoted 6 lines
security/smack/smack_lsm.c +6
-0+if (new->smk_mask.s_addr > smack_netlbladdrs->smk_mask.s_addr) { + new->smk_next = smack_netlbladdrs; + smack_netlbladdrs = new; +return ; + } +for (m = smack_netlbladdrs; m !=NULL ; m = m->smk_next) {
security/smack/smack_lsm.c static int smk_open_netlbladdr(struct inode *inode, · line 728
Quoted 10 lines
security/smack/smack_lsm.c +10
-0+if (m->smk_next ==NULL ) { + m->smk_next = new; +return ; + } +if (new->smk_mask.s_addr > m->smk_next->smk_mask.s_addr) { + new->smk_next = m->smk_next; + m->smk_next = new; +return ; + } + }
As Casey mentioned earlier (and has been brough up in the past), you should heavily consider using the list.h constructs, it would make life much easier here and elsewhere.
Bonus points if you convert the other Smack lists to using the list.h bits.
Quoted 1 line
security/smack/smack_lsm.c +1
-0+}
...
security/smack/smack_lsm.c static ssize_t smk_write_netlbladdr(struct file *file, · line 755
Quoted 10 lines
security/smack/smack_lsm.c +2
-1+/** * smk_write_netlbladdr - write() for /smack/netlabel * @filp: file pointer, not actually used * @buf: where to get the data from @@ -724,8 +752,9 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, struct netlbl_audit audit_info;struct in_addr mask;unsigned int m; -__be32 bebits =BEMASK ; +u32 mask_bits = (1 <<31 );
Why not just enter the value directly here? It would be much clearer I think.
Signature · 2 lines
paul moore linux @ hp
PATCH 1/1 SMACK netlabel fixes v2
Hi Paul,
thanks for your comments, please find below an updated version of the patch, I hope it is fine with you and Casey. (i didnt remove the 'mask_bits = (1<<31)' of your last comment :
=> i think it's much obvious that it represent binary 1000..00 than 0x80000000, but that may be just me :)
I redid some basics tests, it seems i've not messed things too much :)
For the moving to standard list, I guess I could try to do that; (even other smack list, yes want my bonus points! ;) ) After a quick list.h grep i think i'll have to implement a 'sorted insert' anyway.
But please allow me to do that on top of my patch
regards
Etienne
--
the following patch (against 2.6.29rc5) fixes a few issues in the smack/netlabel area :
1) smack_host_label disregard a "0.0.0.0/0 @" rule (or other label), preventing 'tagged' tasks to access Internet (many systems drop packets with IP options)
2) netmasks were not handled correctly, they were stored in a way _not equivalent_ to conversion to be32 (it was equivalent for /0, /8, /16, /24, /32 masks but not other masks)
3) smack_netlbladdr prefixes (IP/mask) were not consistent (mask&IP was not done), so there could have been different list entries for the same IP prefix; if those entries had different labels, well ...
4) they were not sorted
1) 2) 3) are bugs, 4) is a more cosmetic issue.
The patch :
-creates a new helper smk_netlbladdr_insert to insert a smk_netlbladdr, sorted by netmask length -use the new sorted nature of smack_netlbladdrs list to simplify smack_host_label :
the first match _will_ be the more specific
-corrects endianness issues in smk_write_netlbladdr & netlbladdr_seq_show
The patch are "tested" so that they no crash the system; cat /smack/netlabel is now sorted and always consistent. Some basics ping tests to '@' and other label combination seems ok
See an extract of my tests bellow the patch
regards,
Etienne
---
security/smack/smack_lsm.c | 43 +++++------------------------
security/smack/smackfs.c | 64 +++++++++++++++++++++++++++++++++---------- 2 files changed, 57 insertions(+), 50 deletions(-)
security/smack/smack_lsm.c +8 -35
@@ -1498,58 +1498,31 @@ static int smack_socket_post_create(struct socket *sock, int family,* looks for host based access restrictions * * This version will only be appropriate for really small -* sets of single label hosts. Because of the masking -* it cannot shortcut out on the first match. There are -* numerious ways to address the problem, but none of them -* have been applied here. +* sets of single label hosts. * * Returns the label of the far end or NULL if it's not special. */ static char *smack_host_label (struct sockaddr_in *sip) {struct smk_netlbladdr *snp; -char *bestlabel =NULL ;struct in_addr *siap = &sip->sin_addr; -struct in_addr *liap; -struct in_addr *miap; -struct in_addr bestmask;if (siap->s_addr ==0 )return NULL ; - bestmask.s_addr =0 ; -for (snp = smack_netlbladdrs; snp !=NULL ; snp = snp->smk_next) { - liap = &snp->smk_host.sin_addr; - miap = &snp->smk_mask; -/* -* If the addresses match after applying the list entry mask -* the entry matches the address. If it doesn't move along to -* the next entry. -*/ -if ((liap->s_addr & miap->s_addr) != - (siap->s_addr & miap->s_addr)) -continue ;/* -* If the list entry mask identifies a single address -* it can't get any more specific. +* we break after finding the first match because +* the list is sorted from longest to shortest mask +* so we have found the most specific match */ -if (miap->s_addr ==0xffffffff ) +if ((&snp->smk_host.sin_addr)->s_addr == + (siap->s_addr & (&snp->smk_mask)->s_addr)) {return snp->smk_label; -/* -* If the list entry mask is less specific than the best -* already found this entry is uninteresting. -*/ -if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) -continue ; -/* -* This is better than any entry found so far. -*/ - bestmask.s_addr = miap->s_addr; - bestlabel = snp->smk_label; + } } -return bestlabel; +return NULL ; }/**
security/smack/smackfs.c +49 -15
110 lines not on this page · raw message
Re: [PATCH] SMACK netlabel fixes v2
Quoted 7 lines · etienne
Hi Paul,
thanks for your comments, please find below an updated version of the patch, I hope it is fine with you and Casey. (i didnt remove the '
mask_bits= (1<<31)' of your last comment :
=> i think it's much obvious that it represent binary 1000..00 than0x80000000, but that may be just me :)
Fair enough, it was a pretty nit-picky comment :)
Quoted 6 lines
I redid some basics tests, it seems i've not messed things too much :)
For the moving to standard list, I guess I could try to do that; (even other smack list, yes want my bonus points! ;) ) After a quick
list.hgrep i think i'll have to implement a 'sorted insert' anyway. But please allow me to do that on top of my patch
That is fine, the patch below is still a good improvement, no need to hold it up for the list conversion.
Quoted 231 lines
--
the following patch (against 2.6.29rc5) fixes a few issues in the smack/netlabel area : 1)
smack_host_labeldisregard a "0.0.0.0/0 @" rule (or other label), preventing 'tagged' tasks to access Internet (many systems drop packets with IP options) 2) netmasks were not handled correctly, they were stored in a way_notequivalent_to conversion to be32 (it was equivalent for /0, /8, /16, /24, /32 masks but not other masks) 3)smack_netlbladdrprefixes (IP/mask) were not consistent (mask&IP was not done), so there could have been different list entries for the same IP prefix; if those entries had different labels, well ... 4) they were not sorted1) 2) 3) are bugs, 4) is a more cosmetic issue.
The patch :
-creates a new helpersmk_netlbladdr_insertto insert asmk_netlbladdr, sorted by netmask length -use the new sorted nature ofsmack_netlbladdrslist to simplifysmack_host_label: the first match _will_ be the more specific
-corrects endianness issues insmk_write_netlbladdr&netlbladdr_seq_showThe patch are "tested" so that they no crash the system; cat /smack/netlabel is now sorted and always consistent. Some basics ping tests
to '@' and other label combination seems ok See an extract of my tests
bellow the patchregards,
EtienneSigned-off-by <[email protected]>---
security/smack/smack_lsm.c| 43 +++++------------------------security/smack/smackfs.c | 64+++++++++++++++++++++++++++++++++---------- 2 files changed, 57 insertions(+), 50 deletions(-)
security/smack/smack_lsm.c +8
-35@@ -1498,58 +1498,31 @@ static int smack_socket_post_create(struct socket*sock, int family, * looks for host based access restrictions * * This version will only be appropriate for really small -* sets of single label hosts. Because of the masking -* it cannot shortcut out on the first match. There are -* numerious ways to address the problem, but none of them -* have been applied here. +* sets of single label hosts. * * Returns the label of the far end or NULL if it's not special. */ static char *smack_host_label (struct sockaddr_in *sip) {struct smk_netlbladdr *snp; -char *bestlabel =NULL ;struct in_addr *siap = &sip->sin_addr; -struct in_addr *liap; -struct in_addr *miap; -struct in_addr bestmask;if (siap->s_addr ==0 )return NULL ; - bestmask.s_addr =0 ; -for (snp = smack_netlbladdrs; snp !=NULL ; snp = snp->smk_next) { - liap = &snp->smk_host.sin_addr; - miap = &snp->smk_mask; -/* -* If the addresses match after applying the list entry mask -* the entry matches the address. If it doesn't move along to -* the next entry. -*/ -if ((liap->s_addr & miap->s_addr) != - (siap->s_addr & miap->s_addr)) -continue ;/* -* If the list entry mask identifies a single address -* it can't get any more specific. +* we break after finding the first match because +* the list is sorted from longest to shortest mask +* so we have found the most specific match */ -if (miap->s_addr ==0xffffffff ) +if ((&snp->smk_host.sin_addr)->s_addr == + (siap->s_addr & (&snp->smk_mask)->s_addr)) {return snp->smk_label; -/* -* If the list entry mask is less specific than the best -* already found this entry is uninteresting. -*/ -if ((miap->s_addr | bestmask.s_addr) == bestmask.s_addr) -continue ; -/* -* This is better than any entry found so far. -*/ - bestmask.s_addr = miap->s_addr; - bestlabel = snp->smk_label; + } } -return bestlabel; +return NULL ; }/** security/smack/smackfs.c +0
-4@@ -650,10 +650,6 @@ static void *netlbladdr_seq_next(struct seq_file *s, void *v, loff_t *pos) return skp; } -/* -#define BEMASK 0x80000000 -*/ -#define BEMASK 0x00000001#define BEBITS (sizeof(__be32) * 8)/*
security/smack/smackfs.c +49
-11@@ -663,12 +659,10 @@ static int netlbladdr_seq_show(struct seq_file *s, void *v) {struct smk_netlbladdr *skp = (struct smk_netlbladdr *) v;unsigned char *hp = (char *) &skp->smk_host.sin_addr.s_addr; -__be32 bebits; -int maskn =0 ; +int maskn; +u32 temp_mask =be32_to_cpu (skp->smk_mask.s_addr); -for (bebits =BEMASK ; bebits !=0 ; maskn++, bebits <<=1 ) -if ((skp->smk_mask.s_addr & bebits) ==0 ) -break ; +for (maskn =0 ; temp_mask; temp_mask <<=1 , maskn++);seq_printf (s,"%u.%u.%u.%u/%d %s\n" , hp[0 ], hp[1 ], hp[2 ], hp[3 ], maskn, skp->smk_label);@@ -702,6 +696,42 @@ static int smk_open_netlbladdr(struct inode *inode, struct file *file) }/** +* smk_netlbladdr_insert +* @new : netlabel to insert +* +* This helper insert netlabel in the smack_netlbladdrs list +* sorted by netmask length (longest to smallest) +*/ +static void smk_netlbladdr_insert (struct smk_netlbladdr *new) +{ +struct smk_netlbladdr *m; + +if (smack_netlbladdrs ==NULL ) { + smack_netlbladdrs = new; +return ; + } + +/* the comparison '>' is a bit hacky, but works */ +if (new->smk_mask.s_addr > smack_netlbladdrs->smk_mask.s_addr) { + new->smk_next = smack_netlbladdrs; + smack_netlbladdrs = new; +return ; + } +for (m = smack_netlbladdrs; m !=NULL ; m = m->smk_next) { +if (m->smk_next ==NULL ) { + m->smk_next = new; +return ; + } +if (new->smk_mask.s_addr > m->smk_next->smk_mask.s_addr) { + new->smk_next = m->smk_next; + m->smk_next = new; +return ; + } + } +} + + +/** * smk_write_netlbladdr - write() for /smack/netlabel * @filp: file pointer, not actually used * @buf: where to get the data from @@ -724,8 +754,9 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, struct netlbl_audit audit_info;struct in_addr mask;unsigned int m; -__be32 bebits =BEMASK ; +u32 mask_bits = (1 <<31 );__be32 nsa; +u32 temp_mask;/* * Must have privilege. @@ -761,10 +792,13 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, if (sp == NULL) return -EINVAL; -for (mask.s_addr =0 ; m >0 ; m--) { - mask.s_addr |= bebits; - bebits <<=1 ; +for (temp_mask =0 ; m >0 ; m--) { + temp_mask |= mask_bits; + mask_bits >>=1 ; } + mask.s_addr =cpu_to_be32 (temp_mask); + + newname.sin_addr.s_addr &= mask.s_addr;/* * Only allow one writer at a time. Writes should be * quite rare and small in any case. @@ -772,6 +806,7 @@ static ssize_t smk_write_netlbladdr(struct file *file, const char __user *buf, mutex_lock(&smk_netlbladdr_lock); nsa = newname.sin_addr.s_addr; +/* try to find if the prefix is already in the list */ for (skp = smack_netlbladdrs; skp !=NULL ; skp = skp->smk_next)if (skp->smk_host.sin_addr.s_addr == nsa && skp->smk_mask.s_addr == mask.s_addr)@@ -787,9 +822,8 @@ static ssize_t smk_write_netlbladdr(struct file *file,const char __user *buf, rc =0 ; skp->smk_host.sin_addr.s_addr = newname.sin_addr.s_addr; skp->smk_mask.s_addr = mask.s_addr; - skp->smk_next = smack_netlbladdrs; skp->smk_label = sp; - smack_netlbladdrs = skp; +smk_netlbladdr_insert (skp); } }else { rc = netlbl_cfg_unlbl_static_del(&init_net, NULL,
Signature · 2 lines
paul moore linux @ hp
Quoted 10 lines · etienne
OK, I'm about to send a new patch; but while testing my patches and reading code, I noticed another bug :
In smackfs.c:smk_write_netlbladdr
the netmask mask.s_addr is not handled correctly, the netmask should be :
1- computed in u32
2- converted to be32 !!
with current code, a "pseudo u32 mask" is applied to a be32 ipaddr; it occurs to works for "common netmasks" (multiple of 8), not for "intermediate" mask (/15, /25)
Heh, back when Casey was first drafting this code I mentioned the same issue regarding byte ordering but Casey assured me that everything was correct. I didn't have a Smack test system at the time so I couldn't verify the behavior. I'm glad you had a chance to test it, needless to say you should fix that when you submit your patch.
Signature · 2 lines
paul moore linux @ hp