| Message ID | 20180308171456.2597-1-tycho@tycho.ws (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Mar 8, 2018 at 7:14 PM, Tycho Andersen <tycho@tycho.ws> wrote: > In keeping with the directive to get rid of VLAs [1], let's drop the VLA > from ima_audit_measurement(). We need to adjust the return type of > ima_audit_measurement, because now this function can fail if an allocation > fails. > + algo_hash_len = hash_len + strlen(algo_name) + 2; > + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); > - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); > + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); kasprintf() ?
On Thu, Mar 08, 2018 at 07:47:37PM +0200, Andy Shevchenko wrote: > On Thu, Mar 8, 2018 at 7:14 PM, Tycho Andersen <tycho@tycho.ws> wrote: > > In keeping with the directive to get rid of VLAs [1], let's drop the VLA > > from ima_audit_measurement(). We need to adjust the return type of > > ima_audit_measurement, because now this function can fail if an allocation > > fails. > > > > > + algo_hash_len = hash_len + strlen(algo_name) + 2; > > + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); > > > - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); > > + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > kasprintf() ? Sure, in fact I think we could just do: - snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); - audit_log_untrustedstring(ab, algo_hash); + audit_log_untrustedstring(ab, algo_name); + audit_log_format(ab, ":"); + audit_log_untrustedstring(ab, hash); and get rid of the allocation entirely. I'll test and make sure it works and then re-send. Cheers, Tycho
On Thu, 2018-03-08 at 11:37 -0700, Tycho Andersen wrote: > On Thu, Mar 08, 2018 at 07:47:37PM +0200, Andy Shevchenko wrote: > > On Thu, Mar 8, 2018 at 7:14 PM, Tycho Andersen <tycho@tycho.ws> wrote: > > > In keeping with the directive to get rid of VLAs [1], let's drop the VLA > > > from ima_audit_measurement(). We need to adjust the return type of > > > ima_audit_measurement, because now this function can fail if an allocation > > > fails. > > > > > > > > > + algo_hash_len = hash_len + strlen(algo_name) + 2; > > > + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); > > > > > - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); > > > + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > > kasprintf() ? > > Sure, in fact I think we could just do: > > - snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > - audit_log_untrustedstring(ab, algo_hash); > + audit_log_untrustedstring(ab, algo_name); > + audit_log_format(ab, ":"); > + audit_log_untrustedstring(ab, hash); > > and get rid of the allocation entirely. I'll test and make sure it > works and then re-send. The hash algorithm name is an enumeration that comes from the kernel. It's defined in crypto/hash_info.c: hash_algo_name. Why do we need to use audit_log_untrustedstring()? Mimi
On Thu, Mar 08, 2018 at 01:50:30PM -0500, Mimi Zohar wrote: > On Thu, 2018-03-08 at 11:37 -0700, Tycho Andersen wrote: > > On Thu, Mar 08, 2018 at 07:47:37PM +0200, Andy Shevchenko wrote: > > > On Thu, Mar 8, 2018 at 7:14 PM, Tycho Andersen <tycho@tycho.ws> wrote: > > > > In keeping with the directive to get rid of VLAs [1], let's drop the VLA > > > > from ima_audit_measurement(). We need to adjust the return type of > > > > ima_audit_measurement, because now this function can fail if an allocation > > > > fails. > > > > > > > > > > > > > + algo_hash_len = hash_len + strlen(algo_name) + 2; > > > > + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); > > > > > > > - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); > > > > + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > > > > kasprintf() ? > > > > Sure, in fact I think we could just do: > > > > - snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > - audit_log_untrustedstring(ab, algo_hash); > > + audit_log_untrustedstring(ab, algo_name); > > + audit_log_format(ab, ":"); > > + audit_log_untrustedstring(ab, hash); > > > > and get rid of the allocation entirely. I'll test and make sure it > > works and then re-send. > > The hash algorithm name is an enumeration that comes from the kernel. > It's defined in crypto/hash_info.c: hash_algo_name. Why do we need > to use audit_log_untrustedstring()? Yes, I suppose we don't need it for the hash either, since we're generating that and we know it's just hex digits and not any audit control characters or "s or anything. It looks like we could get rid of the other allocation too by just using audit_log_n_hex, but that uses hex_byte_pack_upper, vs. the hex_byte_pack that's currently in use in this function. Is that too much of a breakage? Tycho
On Thu, 2018-03-08 at 12:04 -0700, Tycho Andersen wrote: > On Thu, Mar 08, 2018 at 01:50:30PM -0500, Mimi Zohar wrote: > > On Thu, 2018-03-08 at 11:37 -0700, Tycho Andersen wrote: > > > On Thu, Mar 08, 2018 at 07:47:37PM +0200, Andy Shevchenko wrote: > > > > On Thu, Mar 8, 2018 at 7:14 PM, Tycho Andersen <tycho@tycho.ws> wrote: > > > > > In keeping with the directive to get rid of VLAs [1], let's drop the VLA > > > > > from ima_audit_measurement(). We need to adjust the return type of > > > > > ima_audit_measurement, because now this function can fail if an allocation > > > > > fails. > > > > > > > > > > > > > > > > > + algo_hash_len = hash_len + strlen(algo_name) + 2; > > > > > + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); > > > > > > > > > - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); > > > > > + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > > > > > > kasprintf() ? > > > > > > Sure, in fact I think we could just do: > > > > > > - snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > - audit_log_untrustedstring(ab, algo_hash); > > > + audit_log_untrustedstring(ab, algo_name); > > > + audit_log_format(ab, ":"); > > > + audit_log_untrustedstring(ab, hash); > > > > > > and get rid of the allocation entirely. I'll test and make sure it > > > works and then re-send. > > > > The hash algorithm name is an enumeration that comes from the kernel. > > It's defined in crypto/hash_info.c: hash_algo_name. Why do we need > > to use audit_log_untrustedstring()? > > Yes, I suppose we don't need it for the hash either, since we're > generating that and we know it's just hex digits and not any audit > control characters or "s or anything. > > It looks like we could get rid of the other allocation too by just > using audit_log_n_hex, but that uses hex_byte_pack_upper, vs. the > hex_byte_pack that's currently in use in this function. Is that too > much of a breakage? Based on the discussion with Richard Briggs, we need to differentiate between the ima_audit_measurement() and the ima_parse_rule() usage of AUDIT_INTEGRITY_RULE. The ima_parse_rule() will continue to use AUDIT_INTEGRITY_RULE. ima_audit_measurement() will need to define and use a new number. Auidt name suggestions would be appreciated. When we make that sort of change, any other changes are insignificant. How different are the two formats? Mimi
On Thu, Mar 08, 2018 at 02:20:17PM -0500, Mimi Zohar wrote: > On Thu, 2018-03-08 at 12:04 -0700, Tycho Andersen wrote: > > On Thu, Mar 08, 2018 at 01:50:30PM -0500, Mimi Zohar wrote: > > > On Thu, 2018-03-08 at 11:37 -0700, Tycho Andersen wrote: > > > > On Thu, Mar 08, 2018 at 07:47:37PM +0200, Andy Shevchenko wrote: > > > > > On Thu, Mar 8, 2018 at 7:14 PM, Tycho Andersen <tycho@tycho.ws> wrote: > > > > > > In keeping with the directive to get rid of VLAs [1], let's drop the VLA > > > > > > from ima_audit_measurement(). We need to adjust the return type of > > > > > > ima_audit_measurement, because now this function can fail if an allocation > > > > > > fails. > > > > > > > > > > > > > > > > > > > > > + algo_hash_len = hash_len + strlen(algo_name) + 2; > > > > > > + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); > > > > > > > > > > > - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); > > > > > > + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > > > > > > > > kasprintf() ? > > > > > > > > Sure, in fact I think we could just do: > > > > > > > > - snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > > - audit_log_untrustedstring(ab, algo_hash); > > > > + audit_log_untrustedstring(ab, algo_name); > > > > + audit_log_format(ab, ":"); > > > > + audit_log_untrustedstring(ab, hash); > > > > > > > > and get rid of the allocation entirely. I'll test and make sure it > > > > works and then re-send. > > > > > > The hash algorithm name is an enumeration that comes from the kernel. > > > It's defined in crypto/hash_info.c: hash_algo_name. Why do we need > > > to use audit_log_untrustedstring()? > > > > Yes, I suppose we don't need it for the hash either, since we're > > generating that and we know it's just hex digits and not any audit > > control characters or "s or anything. > > > > It looks like we could get rid of the other allocation too by just > > using audit_log_n_hex, but that uses hex_byte_pack_upper, vs. the > > hex_byte_pack that's currently in use in this function. Is that too > > much of a breakage? > > Based on the discussion with Richard Briggs, we need to differentiate > between the ima_audit_measurement() and the ima_parse_rule() usage of > AUDIT_INTEGRITY_RULE. The ima_parse_rule() will continue to use > AUDIT_INTEGRITY_RULE. ima_audit_measurement() will need to define and > use a new number. Auidt name suggestions would be appreciated. > > When we make that sort of change, any other changes are insignificant. > How different are the two formats? It's just uppercase and lowercase in the hash value, so: Mar 8 16:56:46 ima kernel: [ 104.922927] audit: type=1805 audit(1520528206.082:53): file="/bin/cat" hash="sha1:79e52322102f073684e2dd0ab7653c7c6fcc49b4" ppid=2049 pid=2123 auid=1000 uid=0 gid=0 euid=0 suid=0 fsuid=0 egid=0 sgid=0 fsgid=0 tty=pts0 ses=1 comm="bash" exe="/bin/bash" vs. Mar 8 19:45:12 ima kernel: [ 207.124383] audit: type=1805 audit(1520538312.740:239): file="/root/.viminfo" hash="sha1:3322BE0C00190AB0D20C47574575842EC3020BF5" ppid=2045 pid=2195 auid=1000 uid=0 gid=0 euid=0 suid=0 fsuid=0 egid=0 sgid=0 fsgid=0 tty=pts0 ses=1 comm="vi" exe="/usr/bin/vim.basic" I'm happy to do either way. Tycho
On Thu, 2018-03-08 at 12:47 -0700, Tycho Andersen wrote: > On Thu, Mar 08, 2018 at 02:20:17PM -0500, Mimi Zohar wrote: > > On Thu, 2018-03-08 at 12:04 -0700, Tycho Andersen wrote: > > > On Thu, Mar 08, 2018 at 01:50:30PM -0500, Mimi Zohar wrote: > > > > On Thu, 2018-03-08 at 11:37 -0700, Tycho Andersen wrote: > > > > > On Thu, Mar 08, 2018 at 07:47:37PM +0200, Andy Shevchenko wrote: > > > > > > On Thu, Mar 8, 2018 at 7:14 PM, Tycho Andersen <tycho@tycho.ws> wrote: > > > > > > > In keeping with the directive to get rid of VLAs [1], let's drop the VLA > > > > > > > from ima_audit_measurement(). We need to adjust the return type of > > > > > > > ima_audit_measurement, because now this function can fail if an allocation > > > > > > > fails. > > > > > > > > > > > > > > > > > > > > > > > > > + algo_hash_len = hash_len + strlen(algo_name) + 2; > > > > > > > + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); > > > > > > > > > > > > > - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); > > > > > > > + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > > > > > > > > > > kasprintf() ? > > > > > > > > > > Sure, in fact I think we could just do: > > > > > > > > > > - snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); > > > > > - audit_log_untrustedstring(ab, algo_hash); > > > > > + audit_log_untrustedstring(ab, algo_name); > > > > > + audit_log_format(ab, ":"); > > > > > + audit_log_untrustedstring(ab, hash); > > > > > > > > > > and get rid of the allocation entirely. I'll test and make sure it > > > > > works and then re-send. > > > > > > > > The hash algorithm name is an enumeration that comes from the kernel. > > > > It's defined in crypto/hash_info.c: hash_algo_name. Why do we need > > > > to use audit_log_untrustedstring()? > > > > > > Yes, I suppose we don't need it for the hash either, since we're > > > generating that and we know it's just hex digits and not any audit > > > control characters or "s or anything. > > > > > > It looks like we could get rid of the other allocation too by just > > > using audit_log_n_hex, but that uses hex_byte_pack_upper, vs. the > > > hex_byte_pack that's currently in use in this function. Is that too > > > much of a breakage? > > > > Based on the discussion with Richard Briggs, we need to differentiate > > between the ima_audit_measurement() and the ima_parse_rule() usage of > > AUDIT_INTEGRITY_RULE. The ima_parse_rule() will continue to use > > AUDIT_INTEGRITY_RULE. ima_audit_measurement() will need to define and > > use a new number. Auidt name suggestions would be appreciated. > > > > When we make that sort of change, any other changes are insignificant. > > How different are the two formats? > > It's just uppercase and lowercase in the hash value, so: > > Mar 8 16:56:46 ima kernel: [ 104.922927] audit: type=1805 audit(1520528206.082:53): file="/bin/cat" hash="sha1:79e52322102f073684e2dd0ab7653c7c6fcc49b4" ppid=2049 pid=2123 auid=1000 uid=0 gid=0 euid=0 suid=0 fsuid=0 egid=0 sgid=0 fsgid=0 tty=pts0 ses=1 comm="bash" exe="/bin/bash" > > vs. > > Mar 8 19:45:12 ima kernel: [ 207.124383] audit: type=1805 audit(1520538312.740:239): file="/root/.viminfo" hash="sha1:3322BE0C00190AB0D20C47574575842EC3020BF5" ppid=2045 pid=2195 auid=1000 uid=0 gid=0 euid=0 suid=0 fsuid=0 egid=0 sgid=0 fsgid=0 tty=pts0 ses=1 comm="vi" exe="/usr/bin/vim.basic" > > I'm happy to do either way. If you're willing to wait until the container-id issue is resolved/upstreamed, then either way is fine. If you want the change to go in sooner, then keep it as it currently is. Mimi
diff --git a/security/integrity/ima/ima.h b/security/integrity/ima/ima.h index d52b487ad259..8e2470f72f7f 100644 --- a/security/integrity/ima/ima.h +++ b/security/integrity/ima/ima.h @@ -201,8 +201,8 @@ void ima_store_measurement(struct integrity_iint_cache *iint, struct file *file, const unsigned char *filename, struct evm_ima_xattr_data *xattr_value, int xattr_len, int pcr); -void ima_audit_measurement(struct integrity_iint_cache *iint, - const unsigned char *filename); +int ima_audit_measurement(struct integrity_iint_cache *iint, + const unsigned char *filename); int ima_alloc_init_template(struct ima_event_data *event_data, struct ima_template_entry **entry); int ima_store_template(struct ima_template_entry *entry, int violation, diff --git a/security/integrity/ima/ima_api.c b/security/integrity/ima/ima_api.c index 08fe405338e1..008d3887ae00 100644 --- a/security/integrity/ima/ima_api.c +++ b/security/integrity/ima/ima_api.c @@ -304,17 +304,28 @@ void ima_store_measurement(struct integrity_iint_cache *iint, ima_free_template_entry(entry); } -void ima_audit_measurement(struct integrity_iint_cache *iint, - const unsigned char *filename) +int ima_audit_measurement(struct integrity_iint_cache *iint, + const unsigned char *filename) { struct audit_buffer *ab; - char hash[(iint->ima_hash->length * 2) + 1]; + char *hash, *algo_hash; const char *algo_name = hash_algo_name[iint->ima_hash->algo]; - char algo_hash[sizeof(hash) + strlen(algo_name) + 2]; - int i; + int i, hash_len, algo_hash_len; if (iint->flags & IMA_AUDITED) - return; + return 0; + + hash_len = (iint->ima_hash->length * 2) + 1; + hash = kzalloc(hash_len, GFP_KERNEL); + if (!hash) + return -ENOMEM; + + algo_hash_len = hash_len + strlen(algo_name) + 2; + algo_hash = kzalloc(algo_hash_len, GFP_KERNEL); + if (!algo_hash) { + kfree(hash); + return -ENOMEM; + } for (i = 0; i < iint->ima_hash->length; i++) hex_byte_pack(hash + (i * 2), iint->ima_hash->digest[i]); @@ -323,18 +334,22 @@ void ima_audit_measurement(struct integrity_iint_cache *iint, ab = audit_log_start(current->audit_context, GFP_KERNEL, AUDIT_INTEGRITY_RULE); if (!ab) - return; + goto out; audit_log_format(ab, "file="); audit_log_untrustedstring(ab, filename); audit_log_format(ab, " hash="); - snprintf(algo_hash, sizeof(algo_hash), "%s:%s", algo_name, hash); + snprintf(algo_hash, algo_hash_len, "%s:%s", algo_name, hash); audit_log_untrustedstring(ab, algo_hash); audit_log_task_info(ab, current); audit_log_end(ab); iint->flags |= IMA_AUDITED; +out: + kfree(hash); + kfree(algo_hash); + return 0; } /* diff --git a/security/integrity/ima/ima_main.c b/security/integrity/ima/ima_main.c index 2cfb0c714967..356faae6f09c 100644 --- a/security/integrity/ima/ima_main.c +++ b/security/integrity/ima/ima_main.c @@ -288,8 +288,11 @@ static int process_measurement(struct file *file, char *buf, loff_t size, xattr_value, xattr_len, opened); inode_unlock(inode); } - if (action & IMA_AUDIT) - ima_audit_measurement(iint, pathname); + if (action & IMA_AUDIT) { + rc = ima_audit_measurement(iint, pathname); + if (rc < 0) + goto out_locked; + } if ((file->f_flags & O_DIRECT) && (iint->flags & IMA_PERMIT_DIRECTIO)) rc = 0;
In keeping with the directive to get rid of VLAs [1], let's drop the VLA from ima_audit_measurement(). We need to adjust the return type of ima_audit_measurement, because now this function can fail if an allocation fails. [1]: https://lkml.org/lkml/2018/3/7/621 Signed-off-by: Tycho Andersen <tycho@tycho.ws> --- security/integrity/ima/ima.h | 4 ++-- security/integrity/ima/ima_api.c | 31 +++++++++++++++++++++++-------- security/integrity/ima/ima_main.c | 7 +++++-- 3 files changed, 30 insertions(+), 12 deletions(-)