Message ID | e3e9c9da9115fd233c5a7895dbb4a698a365b1b0.1595884394.git.rgb@redhat.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [V3fix,ghak120] audit: initialize context values in case of mandatory events | expand |
On Mon, Jul 27, 2020 at 5:30 PM Richard Guy Briggs <rgb@redhat.com> wrote: > > Issue ghak120 enabled syscall records to accompany required records when > no rules are present to trigger the storage of syscall context. A > reported issue showed that the cwd was not always initialized. That > issue was already resolved ... Yes and no. Yes, it appears to be resolved in v5.8-rc1 and above, but the problematic commit is in v5.7 and I'm not sure backporting the fix in v5.8-rcX plus this patch is the right thing to do for a released kernel. The lowest risk fix for v5.7 at this point is to do a revert; regardless of what happens with this patch and v5.8-rcX please post a revert for the audit/stable-5.7 tree as soon as you can. > ... but a review of all other records that could > be triggered at the time of a syscall record revealed other potential > values that could be missing or misleading. Initialize them. > > The fds array is reset to -1 after the first syscall to indicate it > isn't valid any more, but was never set to -1 when the context was > allocated to indicate it wasn't yet valid. > > The audit_inode* functions can be called without going through > getname_flags() or getname_kernel() that sets audit_names and cwd, so > set the cwd if it has not already been done so due to audit_names being > valid. > > The LSM dump_common_audit_data() LSM_AUDIT_DATA_NET:AF_UNIX case was > missed with the ghak96 patch, so add that case here. > > Please see issue https://github.com/linux-audit/audit-kernel/issues/120 > Please see issue https://github.com/linux-audit/audit-kernel/issues/96 > Passes audit-testsuite. > > Signed-off-by: Richard Guy Briggs <rgb@redhat.com> > --- > kernel/auditsc.c | 3 +++ > security/lsm_audit.c | 1 + > 2 files changed, 4 insertions(+) > > diff --git a/kernel/auditsc.c b/kernel/auditsc.c > index 6884b50069d1..2f97618e6a34 100644 > --- a/kernel/auditsc.c > +++ b/kernel/auditsc.c > @@ -929,6 +929,7 @@ static inline struct audit_context *audit_alloc_context(enum audit_state state) > context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0; > INIT_LIST_HEAD(&context->killed_trees); > INIT_LIST_HEAD(&context->names_list); > + context->fds[0] = -1; > return context; > } > > @@ -2076,6 +2077,7 @@ void __audit_inode(struct filename *name, const struct dentry *dentry, > } > handle_path(dentry); > audit_copy_inode(n, dentry, inode, flags & AUDIT_INODE_NOEVAL); > + _audit_getcwd(context); > } > > void __audit_file(const struct file *file) > @@ -2194,6 +2196,7 @@ void __audit_inode_child(struct inode *parent, > audit_copy_inode(found_child, dentry, inode, 0); > else > found_child->ino = AUDIT_INO_UNSET; > + _audit_getcwd(context); > } > EXPORT_SYMBOL_GPL(__audit_inode_child); > > diff --git a/security/lsm_audit.c b/security/lsm_audit.c > index 53d0d183db8f..e93077612246 100644 > --- a/security/lsm_audit.c > +++ b/security/lsm_audit.c > @@ -369,6 +369,7 @@ static void dump_common_audit_data(struct audit_buffer *ab, > audit_log_untrustedstring(ab, p); > else > audit_log_n_hex(ab, p, len); > + audit_getcwd(); > break; > } > } I understand the "fds[0] = -1" fix in audit_alloc_context() (ironically, the kzalloc() which is supposed to help with cases like this, hurts us with this particular field), but I'm still not quite seeing why we need to sprinkle audit_getcwd() calls everywhere to fix this bug (this seems more like a feature add than a bigfix). Yes, they may fix the problem but it seems like simply adding a context->pwd test in audit_log_name() similar to what we do in audit_log_exit() is the correct fix. We are currently at -rc7 and this really needs to land before v5.8 is released, presumably this weekend; this means a small and limited bug fix patch is what is needed.
On 2020-07-27 22:14, Paul Moore wrote: > On Mon, Jul 27, 2020 at 5:30 PM Richard Guy Briggs <rgb@redhat.com> wrote: > > Issue ghak120 enabled syscall records to accompany required records when > > no rules are present to trigger the storage of syscall context. A > > reported issue showed that the cwd was not always initialized. That > > issue was already resolved ... > > Yes and no. Yes, it appears to be resolved in v5.8-rc1 and above, but > the problematic commit is in v5.7 and I'm not sure backporting the fix > in v5.8-rcX plus this patch is the right thing to do for a released > kernel. The lowest risk fix for v5.7 at this point is to do a revert; Ok, fair enough. I don't understand why you didn't do the revert since it appears so trivial to you and this review and fix turned out to be marginally more work. I didn't understand what you wanted when you referred to stable. > regardless of what happens with this patch and v5.8-rcX please post a > revert for the audit/stable-5.7 tree as soon as you can. (more below...) > > ... but a review of all other records that could > > be triggered at the time of a syscall record revealed other potential > > values that could be missing or misleading. Initialize them. > > > > The fds array is reset to -1 after the first syscall to indicate it > > isn't valid any more, but was never set to -1 when the context was > > allocated to indicate it wasn't yet valid. > > > > The audit_inode* functions can be called without going through > > getname_flags() or getname_kernel() that sets audit_names and cwd, so > > set the cwd if it has not already been done so due to audit_names being > > valid. > > > > The LSM dump_common_audit_data() LSM_AUDIT_DATA_NET:AF_UNIX case was > > missed with the ghak96 patch, so add that case here. > > > > Please see issue https://github.com/linux-audit/audit-kernel/issues/120 > > Please see issue https://github.com/linux-audit/audit-kernel/issues/96 > > Passes audit-testsuite. > > > > Signed-off-by: Richard Guy Briggs <rgb@redhat.com> > > --- > > kernel/auditsc.c | 3 +++ > > security/lsm_audit.c | 1 + > > 2 files changed, 4 insertions(+) > > > > diff --git a/kernel/auditsc.c b/kernel/auditsc.c > > index 6884b50069d1..2f97618e6a34 100644 > > --- a/kernel/auditsc.c > > +++ b/kernel/auditsc.c > > @@ -929,6 +929,7 @@ static inline struct audit_context *audit_alloc_context(enum audit_state state) > > context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0; > > INIT_LIST_HEAD(&context->killed_trees); > > INIT_LIST_HEAD(&context->names_list); > > + context->fds[0] = -1; > > return context; > > } > > > > @@ -2076,6 +2077,7 @@ void __audit_inode(struct filename *name, const struct dentry *dentry, > > } > > handle_path(dentry); > > audit_copy_inode(n, dentry, inode, flags & AUDIT_INODE_NOEVAL); > > + _audit_getcwd(context); > > } > > > > void __audit_file(const struct file *file) > > @@ -2194,6 +2196,7 @@ void __audit_inode_child(struct inode *parent, > > audit_copy_inode(found_child, dentry, inode, 0); > > else > > found_child->ino = AUDIT_INO_UNSET; > > + _audit_getcwd(context); > > } > > EXPORT_SYMBOL_GPL(__audit_inode_child); > > > > diff --git a/security/lsm_audit.c b/security/lsm_audit.c > > index 53d0d183db8f..e93077612246 100644 > > --- a/security/lsm_audit.c > > +++ b/security/lsm_audit.c > > @@ -369,6 +369,7 @@ static void dump_common_audit_data(struct audit_buffer *ab, > > audit_log_untrustedstring(ab, p); > > else > > audit_log_n_hex(ab, p, len); > > + audit_getcwd(); > > break; > > } > > } > > I understand the "fds[0] = -1" fix in audit_alloc_context() > (ironically, the kzalloc() which is supposed to help with cases like > this, hurts us with this particular field), but I'm still not quite > seeing why we need to sprinkle audit_getcwd() calls everywhere to fix > this bug (this seems more like a feature add than a bigfix). Yes, > they may fix the problem but it seems like simply adding a > context->pwd test in audit_log_name() similar to what we do in > audit_log_exit() is the correct fix. Well, considering that ghak96 ended up being a bugfix (that wasn't its intent), I figured these audit_getcwd() were also bugfixes to prevent the same BUG under different calling conditions. > We are currently at -rc7 and this really needs to land before v5.8 is > released, presumably this weekend; this means a small and limited bug > fix patch is what is needed. Ok, so it sounds like rather than just fix it now, it would be better to revert it, then submit *one* patch for ghak120 plus this fix that will go tentatively upstream in 3 months, fully in 5. Arguably the last chunk above should be added to ghak96, so that should be reverted too, then resubmitted with this added chunk (or it could be a fixup chunk that would need to be sequenced with ghak120). As for the middle two chunks, they could either be resubmitted with a resubmitted ghak96, or with a resubmitted ghak120. As for the timing of all of these, ghak96 should be in place before the ghak120 patch, so even resubmitting one patch for the combined ghak120 and ghak96 might make more sense. I know you like only really minimal fixes this late, but this seemed pretty minimal to me... > paul moore - RGB -- Richard Guy Briggs <rgb@redhat.com> Sr. S/W Engineer, Kernel Security, Base Operating Systems Remote, Ottawa, Red Hat Canada IRC: rgb, SunRaycer Voice: +1.647.777.2635, Internal: (81) 32635
On Tue, Jul 28, 2020 at 12:27 PM Richard Guy Briggs <rgb@redhat.com> wrote: > On 2020-07-27 22:14, Paul Moore wrote: > > On Mon, Jul 27, 2020 at 5:30 PM Richard Guy Briggs <rgb@redhat.com> wrote: > > > Issue ghak120 enabled syscall records to accompany required records when > > > no rules are present to trigger the storage of syscall context. A > > > reported issue showed that the cwd was not always initialized. That > > > issue was already resolved ... > > > > Yes and no. Yes, it appears to be resolved in v5.8-rc1 and above, but > > the problematic commit is in v5.7 and I'm not sure backporting the fix > > in v5.8-rcX plus this patch is the right thing to do for a released > > kernel. The lowest risk fix for v5.7 at this point is to do a revert; > > Ok, fair enough. I don't understand why you didn't do the revert since > it appears so trivial to you and this review and fix turned out to be > marginally more work. I didn't understand what you wanted when you > referred to stable. I held off on the revert because I thought you might want the chance to submit the revert with your authorship. I made an assumption that it meant the same to you as it does to me; that's my mistake, I should have known better. I'll do the revert myself for stable-5.8 (which should trickle down to v5.7.z with the right metadata), don't bother with it. > > regardless of what happens with this patch and v5.8-rcX please post a > > revert for the audit/stable-5.7 tree as soon as you can. > > (more below...) > > > > ... but a review of all other records that could > > > be triggered at the time of a syscall record revealed other potential > > > values that could be missing or misleading. Initialize them. > > > > > > The fds array is reset to -1 after the first syscall to indicate it > > > isn't valid any more, but was never set to -1 when the context was > > > allocated to indicate it wasn't yet valid. > > > > > > The audit_inode* functions can be called without going through > > > getname_flags() or getname_kernel() that sets audit_names and cwd, so > > > set the cwd if it has not already been done so due to audit_names being > > > valid. > > > > > > The LSM dump_common_audit_data() LSM_AUDIT_DATA_NET:AF_UNIX case was > > > missed with the ghak96 patch, so add that case here. > > > > > > Please see issue https://github.com/linux-audit/audit-kernel/issues/120 > > > Please see issue https://github.com/linux-audit/audit-kernel/issues/96 > > > Passes audit-testsuite. > > > > > > Signed-off-by: Richard Guy Briggs <rgb@redhat.com> > > > --- > > > kernel/auditsc.c | 3 +++ > > > security/lsm_audit.c | 1 + > > > 2 files changed, 4 insertions(+) > > > > > > diff --git a/kernel/auditsc.c b/kernel/auditsc.c > > > index 6884b50069d1..2f97618e6a34 100644 > > > --- a/kernel/auditsc.c > > > +++ b/kernel/auditsc.c > > > @@ -929,6 +929,7 @@ static inline struct audit_context *audit_alloc_context(enum audit_state state) > > > context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0; > > > INIT_LIST_HEAD(&context->killed_trees); > > > INIT_LIST_HEAD(&context->names_list); > > > + context->fds[0] = -1; > > > return context; > > > } > > > > > > @@ -2076,6 +2077,7 @@ void __audit_inode(struct filename *name, const struct dentry *dentry, > > > } > > > handle_path(dentry); > > > audit_copy_inode(n, dentry, inode, flags & AUDIT_INODE_NOEVAL); > > > + _audit_getcwd(context); > > > } > > > > > > void __audit_file(const struct file *file) > > > @@ -2194,6 +2196,7 @@ void __audit_inode_child(struct inode *parent, > > > audit_copy_inode(found_child, dentry, inode, 0); > > > else > > > found_child->ino = AUDIT_INO_UNSET; > > > + _audit_getcwd(context); > > > } > > > EXPORT_SYMBOL_GPL(__audit_inode_child); > > > > > > diff --git a/security/lsm_audit.c b/security/lsm_audit.c > > > index 53d0d183db8f..e93077612246 100644 > > > --- a/security/lsm_audit.c > > > +++ b/security/lsm_audit.c > > > @@ -369,6 +369,7 @@ static void dump_common_audit_data(struct audit_buffer *ab, > > > audit_log_untrustedstring(ab, p); > > > else > > > audit_log_n_hex(ab, p, len); > > > + audit_getcwd(); > > > break; > > > } > > > } > > > > I understand the "fds[0] = -1" fix in audit_alloc_context() > > (ironically, the kzalloc() which is supposed to help with cases like > > this, hurts us with this particular field), but I'm still not quite > > seeing why we need to sprinkle audit_getcwd() calls everywhere to fix > > this bug (this seems more like a feature add than a bigfix). Yes, > > they may fix the problem but it seems like simply adding a > > context->pwd test in audit_log_name() similar to what we do in > > audit_log_exit() is the correct fix. > > Well, considering that ghak96 ended up being a bugfix (that wasn't its > intent), I figured these audit_getcwd() were also bugfixes to prevent > the same BUG under different calling conditions. > > > We are currently at -rc7 and this really needs to land before v5.8 is > > released, presumably this weekend; this means a small and limited bug > > fix patch is what is needed. > > Ok, so it sounds like rather than just fix it now, it would be better to > revert it, then submit *one* patch for ghak120 plus this fix that will > go tentatively upstream in 3 months, fully in 5. Arguably the last > chunk above should be added to ghak96, so that should be reverted too, > then resubmitted with this added chunk (or it could be a fixup chunk > that would need to be sequenced with ghak120). As for the middle two > chunks, they could either be resubmitted with a resubmitted ghak96, or > with a resubmitted ghak120. As for the timing of all of these, ghak96 > should be in place before the ghak120 patch, so even resubmitting one > patch for the combined ghak120 and ghak96 might make more sense. Sigh. I can't even reply to that paragraph above without going to GH and looking up all those different ghak references, which is annoying, and right now it seems like my time is better spent cleaning up this mess. I'm not exactly sure what you mean by "one patch", but right now we are at -rc7 and we've/I've got broken kernels to fix; submit whatever you want and we'll deal with it when it's posted. > I know you like only really minimal fixes this late, but this seemed > pretty minimal to me... Minimal is a one (two?) line NULL check in audit_log_name(), this patch is not that.
On 2020-07-28 14:47, Paul Moore wrote: > On Tue, Jul 28, 2020 at 12:27 PM Richard Guy Briggs <rgb@redhat.com> wrote: > > I know you like only really minimal fixes this late, but this seemed > > pretty minimal to me... > > Minimal is a one (two?) line NULL check in audit_log_name(), this > patch is not that. I didn't try and test that since I'm not sure that would have worked because there appeared to be a low non-NULL value in it. brauer1's trace had 0x60 and mine had 0xd0. Or am I missing something obvious? The patch provided the information rather than ignoring the problem (which maybe should have been caught by WARN_ONCE?). > paul moore - RGB -- Richard Guy Briggs <rgb@redhat.com> Sr. S/W Engineer, Kernel Security, Base Operating Systems Remote, Ottawa, Red Hat Canada IRC: rgb, SunRaycer Voice: +1.647.777.2635, Internal: (81) 32635
On Tue, Jul 28, 2020 at 10:01 PM Richard Guy Briggs <rgb@redhat.com> wrote: > > On 2020-07-28 14:47, Paul Moore wrote: > > On Tue, Jul 28, 2020 at 12:27 PM Richard Guy Briggs <rgb@redhat.com> wrote: > > > I know you like only really minimal fixes this late, but this seemed > > > pretty minimal to me... > > > > Minimal is a one (two?) line NULL check in audit_log_name(), this > > patch is not that. > > I didn't try and test that since I'm not sure that would have worked > because there appeared to be a low non-NULL value in it. brauer1's trace had > 0x60 and mine had 0xd0. Or am I missing something obvious? Well, you mentioned the obvious already: both 0x60 and 0xd0 are not NULL. We already have a NULL check for context->pwd elsewhere so there is precedence for this solving a similar problem, although without going through the git log I'm not sure what problem that solved, or if it was precautionary. I agree the low value looks suspicious, it almost looks like an offset to me, ideally it would be good to understand how/why that value is "off'. It could be that the audit_context is not being properly initialized, reset, or something unrelated is clobbering the value; all things that would be nice to know. > The patch provided the information rather than ignoring the problem ... I disagree. -- paul moore www.paul-moore.com
diff --git a/kernel/auditsc.c b/kernel/auditsc.c index 6884b50069d1..2f97618e6a34 100644 --- a/kernel/auditsc.c +++ b/kernel/auditsc.c @@ -929,6 +929,7 @@ static inline struct audit_context *audit_alloc_context(enum audit_state state) context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0; INIT_LIST_HEAD(&context->killed_trees); INIT_LIST_HEAD(&context->names_list); + context->fds[0] = -1; return context; } @@ -2076,6 +2077,7 @@ void __audit_inode(struct filename *name, const struct dentry *dentry, } handle_path(dentry); audit_copy_inode(n, dentry, inode, flags & AUDIT_INODE_NOEVAL); + _audit_getcwd(context); } void __audit_file(const struct file *file) @@ -2194,6 +2196,7 @@ void __audit_inode_child(struct inode *parent, audit_copy_inode(found_child, dentry, inode, 0); else found_child->ino = AUDIT_INO_UNSET; + _audit_getcwd(context); } EXPORT_SYMBOL_GPL(__audit_inode_child); diff --git a/security/lsm_audit.c b/security/lsm_audit.c index 53d0d183db8f..e93077612246 100644 --- a/security/lsm_audit.c +++ b/security/lsm_audit.c @@ -369,6 +369,7 @@ static void dump_common_audit_data(struct audit_buffer *ab, audit_log_untrustedstring(ab, p); else audit_log_n_hex(ab, p, len); + audit_getcwd(); break; } }
Issue ghak120 enabled syscall records to accompany required records when no rules are present to trigger the storage of syscall context. A reported issue showed that the cwd was not always initialized. That issue was already resolved, but a review of all other records that could be triggered at the time of a syscall record revealed other potential values that could be missing or misleading. Initialize them. The fds array is reset to -1 after the first syscall to indicate it isn't valid any more, but was never set to -1 when the context was allocated to indicate it wasn't yet valid. The audit_inode* functions can be called without going through getname_flags() or getname_kernel() that sets audit_names and cwd, so set the cwd if it has not already been done so due to audit_names being valid. The LSM dump_common_audit_data() LSM_AUDIT_DATA_NET:AF_UNIX case was missed with the ghak96 patch, so add that case here. Please see issue https://github.com/linux-audit/audit-kernel/issues/120 Please see issue https://github.com/linux-audit/audit-kernel/issues/96 Passes audit-testsuite. Signed-off-by: Richard Guy Briggs <rgb@redhat.com> --- kernel/auditsc.c | 3 +++ security/lsm_audit.c | 1 + 2 files changed, 4 insertions(+)