| Message ID | 20180501175124.192587-2-mjg59@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, 2018-05-01 at 10:51 -0700, Matthew Garrett wrote: > Sites may wish to provide additional metadata alongside files in order > to make more fine-grained security decisions[1]. The security of this is > enhanced if this metadata is protected, something that EVM makes > possible. However, the kernel cannot know about the set of extended > attributes that local admins may wish to protect, and hardcoding this > policy in the kernel makes it difficult to change over time and less > convenient for distributions to enable. > > This patch adds a new /sys/kernel/security/evm_xattrs node, which can be > read to obtain the current set of EVM-protected extended attributes or > written to in order to add new entries. Extending this list will not > change the validity of any existing signatures provided that the file > in question does not have any of the additional extended attributes - > missing xattrs are skipped when calculating the EVM hash. > > [1] For instance, a package manager could install information about the > package uploader in an additional extended attribute. Local LSM policy > could then be associated with that extended attribute in order to > restrict the privileges available to packages from less trusted > uploaders. > > Signed-off-by: Matthew Garrett <mjg59@google.com> > --- > Documentation/ABI/testing/evm | 13 ++++ > security/integrity/evm/evm_secfs.c | 102 +++++++++++++++++++++++++++-- > 2 files changed, 109 insertions(+), 6 deletions(-) > > diff --git a/Documentation/ABI/testing/evm b/Documentation/ABI/testing/evm > index d12cb2eae9ee..bfe529eb26b2 100644 > --- a/Documentation/ABI/testing/evm > +++ b/Documentation/ABI/testing/evm > @@ -57,3 +57,16 @@ Description: > dracut (via 97masterkey and 98integrity) and systemd (via > core/ima-setup) have support for loading keys at boot > time. > + > +What: security/evm_xattrs > +Date: April 2018 > +Contact: Matthew Garrett <mjg59@google.com> > +Description: > + Shows the set of extended attributes used to calculate or > + validate the EVM signature, and allows additional attributes > + to be added at runtime. Adding additional extended attributes > + will result in any existing signatures generated without the > + additional attributes becoming invalid, and any signatures > + generated after additional attributes are added will only be > + valid if the same additional attributes are configured on > + system boot. This needs to be updated ... > diff --git a/security/integrity/evm/evm_secfs.c b/security/integrity/evm/evm_secfs.c > index feba03bbedae..3b371125d439 100644 > --- a/security/integrity/evm/evm_secfs.c > +++ b/security/integrity/evm/evm_secfs.c > @@ -20,6 +20,7 @@ > #include "evm.h" > > static struct dentry *evm_init_tpm; > +static struct dentry *evm_xattrs; > > /** > * evm_read_key - read() for <securityfs>/evm > @@ -107,13 +108,102 @@ static const struct file_operations evm_key_ops = { > .write = evm_write_key, > }; > > -int __init evm_init_secfs(void) > +/** > + * evm_read_xattrs - read() for <securityfs>/evm_xattrs > + * > + * @filp: file pointer, not actually used > + * @buf: where to put the result > + * @count: maximum to send along > + * @ppos: where to start > + * > + * Returns number of bytes read or error code, as appropriate > + */ > +static ssize_t evm_read_xattrs(struct file *filp, char __user *buf, > + size_t count, loff_t *ppos) > { > - int error = 0; > + char *temp; > + int offset = 0; > + ssize_t rc, size = 0; > + struct xattr_list *xattr; > + > + if (*ppos != 0) > + return 0; > > - evm_init_tpm = securityfs_create_file("evm", S_IRUSR | S_IRGRP, > - NULL, NULL, &evm_key_ops); > + list_for_each_entry(xattr, &evm_config_xattrnames, list) > + size += strlen(xattr->name) + 1; > + > + temp = kmalloc(size + 1, GFP_KERNEL); > + if (!temp) > + return -ENOMEM; > + > + list_for_each_entry(xattr, &evm_config_xattrnames, list) { > + sprintf(temp + offset, "%s\n", xattr->name); > + offset += strlen(xattr->name) + 1; > + } > + > + rc = simple_read_from_buffer(buf, count, ppos, temp, strlen(temp)); > + > + return rc; > +} > + > +/** > + * evm_write_xattrs - write() for <securityfs>/evm_xattrs > + * @file: file pointer, not actually used > + * @buf: where to get the data from > + * @count: bytes sent > + * @ppos: where to start > + * > + * Returns number of bytes written or error code, as appropriate > + */ > +static ssize_t evm_write_xattrs(struct file *file, const char __user *buf, > + size_t count, loff_t *ppos) > +{ > + int len; > + struct xattr_list *xattr; > + > + if (!capable(CAP_SYS_ADMIN)) > + return -EPERM; > + > + if (*ppos != 0) > + return -EINVAL; Is there a maximum xattr name size? Should there be? > + > + xattr = kmalloc(sizeof(struct xattr_list), GFP_KERNEL); > + if (!xattr) > + return -ENOMEM; > + > + xattr->name = memdup_user_nul(buf, count); > + if (!xattr->name) { > + kfree(xattr); > + return -ENOMEM; > + } > + > + /* Remove any trailing newline */ > + len = strlen(xattr->name); > + if (xattr->name[len-1] == '\n') > + xattr->name[len-1] = '\0'; Shouldn't we somehow sanity check userspace provided strings, before adding them to the list of xattrs? Perhaps limit the string to the upper and lower case letters? > + > + list_add_tail(&xattr->list, &evm_config_xattrnames); > + return count; > +} > + > +static const struct file_operations evm_xattr_ops = { > + .read = evm_read_xattrs, > + .write = evm_write_xattrs, > +}; > + > +int __init evm_init_secfs(void) > +{ > + evm_init_tpm = securityfs_create_file("evm", 0440, NULL, NULL, > + &evm_key_ops); > if (!evm_init_tpm || IS_ERR(evm_init_tpm)) > - error = -EFAULT; > - return error; > + return -EFAULT; > + > + evm_xattrs = securityfs_create_file("evm_xattrs", 0440, NULL, NULL, > + &evm_xattr_ops); > + if (!evm_xattrs || IS_ERR(evm_xattrs)) { > + securityfs_remove(evm_init_tpm); > + return -EFAULT; > + } > + Do we really want this feature unconditionally enabled? How often do you expect to add xattrs? Is there ever a point where you would want to lock the list of xattrs from changing? Mimi > + return 0; > }
On Wed, May 2, 2018 at 8:17 PM Mimi Zohar <zohar@linux.vnet.ibm.com> wrote: > On Tue, 2018-05-01 at 10:51 -0700, Matthew Garrett wrote: > > + Shows the set of extended attributes used to calculate or > > + validate the EVM signature, and allows additional attributes > > + to be added at runtime. Adding additional extended attributes > > + will result in any existing signatures generated without the > > + additional attributes becoming invalid, and any signatures > > + generated after additional attributes are added will only be > > + valid if the same additional attributes are configured on > > + system boot. > This needs to be updated ... Yup. > > + if (*ppos != 0) > > + return -EINVAL; > Is there a maximum xattr name size? Should there be? There is - I'll add a check. > > + /* Remove any trailing newline */ > > + len = strlen(xattr->name); > > + if (xattr->name[len-1] == '\n') > > + xattr->name[len-1] = '\0'; > Shouldn't we somehow sanity check userspace provided strings, before > adding them to the list of xattrs? Perhaps limit the string to the > upper and lower case letters? As far as I can tell the VFS doesn't do that, and I wouldn't put it past someone to use UTF-8 at some point… > + evm_xattrs = securityfs_create_file("evm_xattrs", 0440, NULL, NULL, > > + &evm_xattr_ops); > > + if (!evm_xattrs || IS_ERR(evm_xattrs)) { > > + securityfs_remove(evm_init_tpm); > > + return -EFAULT; > > + } > > + > Do we really want this feature unconditionally enabled? How often do > you expect to add xattrs? Is there ever a point where you would want > to lock the list of xattrs from changing? I think a config option would make sense here. Locking doesn't seem unreasonable, but I'm not sure what the threat model would really be - adding new xattrs would only result in additional signatures validating if they were signed with a trusted key in the first place.
On Tue, 2018-05-08 at 21:30 +0000, Matthew Garrett wrote: > On Wed, May 2, 2018 at 8:17 PM Mimi Zohar <zohar@linux.vnet.ibm.com> wrote: > > On Tue, 2018-05-01 at 10:51 -0700, Matthew Garrett wrote: > > > + Shows the set of extended attributes used to calculate or > > > + validate the EVM signature, and allows additional > attributes > > > + to be added at runtime. Adding additional extended > attributes > > > + will result in any existing signatures generated without > the > > > + additional attributes becoming invalid, and any signatures > > > + generated after additional attributes are added will only > be > > > + valid if the same additional attributes are configured on > > > + system boot. > > > This needs to be updated ... > > Yup. > > > > + if (*ppos != 0) > > > + return -EINVAL; > > > Is there a maximum xattr name size? Should there be? > > There is - I'll add a check. > > > > + /* Remove any trailing newline */ > > > + len = strlen(xattr->name); > > > + if (xattr->name[len-1] == '\n') > > > + xattr->name[len-1] = '\0'; > > > Shouldn't we somehow sanity check userspace provided strings, before > > adding them to the list of xattrs? Perhaps limit the string to the > > upper and lower case letters? > > As far as I can tell the VFS doesn't do that, and I wouldn't put it past > someone to use UTF-8 at some point… The audit subsystem uses audit_log_untrustedstrings() to check for control characters. I'm not sure what should be included, but not checking userspace strings doesn't sound right. > > > + evm_xattrs = securityfs_create_file("evm_xattrs", 0440, NULL, NULL, > > > + &evm_xattr_ops); > > > + if (!evm_xattrs || IS_ERR(evm_xattrs)) { > > > + securityfs_remove(evm_init_tpm); > > > + return -EFAULT; > > > + } > > > + > > > Do we really want this feature unconditionally enabled? How often do > > you expect to add xattrs? Is there ever a point where you would want > > to lock the list of xattrs from changing? > > I think a config option would make sense here. Locking doesn't seem > unreasonable, but I'm not sure what the threat model would really be - > adding new xattrs would only result in additional signatures validating if > they were signed with a trusted key in the first place. Remember adding additional EVM xattrs isn't limited to just EVM signatures, but for the original EVM HMAC as well. Did you want to limit it to the EVM portable & immutable signatures? Mimi
On Tue, May 8, 2018 at 2:43 PM Mimi Zohar <zohar@linux.vnet.ibm.com> wrote: > On Tue, 2018-05-08 at 21:30 +0000, Matthew Garrett wrote: > > As far as I can tell the VFS doesn't do that, and I wouldn't put it past > > someone to use UTF-8 at some point… > The audit subsystem uses audit_log_untrustedstrings() to check for > control characters. I'm not sure what should be included, but not > checking userspace strings doesn't sound right. As far as I can tell it's legitimate for xattr names to include control characters. I can't think of /why/ someone would do that, but… > > I think a config option would make sense here. Locking doesn't seem > > unreasonable, but I'm not sure what the threat model would really be - > > adding new xattrs would only result in additional signatures validating if > > they were signed with a trusted key in the first place. > Remember adding additional EVM xattrs isn't limited to just EVM > signatures, but for the original EVM HMAC as well. Did you want to > limit it to the EVM portable & immutable signatures? If you add entries and then signatures are created you'll end up with signatures that won't validate on next boot until the same attributes are added. That feels at worst like root being able to trigger a DoS, which they'd be able to do by tampering with the signatures via the raw device node anyway.
On Tue, 2018-05-08 at 22:51 +0000, Matthew Garrett wrote: > On Tue, May 8, 2018 at 2:43 PM Mimi Zohar <zohar@linux.vnet.ibm.com> wrote: > > > On Tue, 2018-05-08 at 21:30 +0000, Matthew Garrett wrote: > > > As far as I can tell the VFS doesn't do that, and I wouldn't put it past > > > someone to use UTF-8 at some point… > > > The audit subsystem uses audit_log_untrustedstrings() to check for > > control characters. I'm not sure what should be included, but not > > checking userspace strings doesn't sound right. > > As far as I can tell it's legitimate for xattr names to include control > characters. I can't think of /why/ someone would do that, but… > > > > I think a config option would make sense here. Locking doesn't seem > > > unreasonable, but I'm not sure what the threat model would really be - > > > adding new xattrs would only result in additional signatures validating > if > > > they were signed with a trusted key in the first place. > > > Remember adding additional EVM xattrs isn't limited to just EVM > > signatures, but for the original EVM HMAC as well. Did you want to > > limit it to the EVM portable & immutable signatures? > > If you add entries and then signatures are created you'll end up with > signatures that won't validate on next boot until the same attributes are > added. That feels at worst like root being able to trigger a DoS, which > they'd be able to do by tampering with the signatures via the raw device > node anyway. With a system with EVM HMAC enabled and an "ima_policy=appraise_tcb" policy, which appraises all executables and files owned by root, any new xattrs would cause the EVM HMAC to be recalculated immediately. On boot, if any of these files are accessed prior to the additional xattrs are added, it could cause the system not to boot properly. True in both cases it is a DoS, but there is a difference between accessing the raw device and normal usage. Adding support for additional xattr names is fine. Making it a Kconfig option is required. Being able to lock adding additional xattr names should also be required. Mimi
On Wed, May 9, 2018 at 8:30 AM Mimi Zohar <zohar@linux.vnet.ibm.com> wrote: > Adding support for additional xattr names is fine. Making it a > Kconfig option is required. Being able to lock adding additional > xattr names should also be required. Ok! I'll add that.
diff --git a/Documentation/ABI/testing/evm b/Documentation/ABI/testing/evm index d12cb2eae9ee..bfe529eb26b2 100644 --- a/Documentation/ABI/testing/evm +++ b/Documentation/ABI/testing/evm @@ -57,3 +57,16 @@ Description: dracut (via 97masterkey and 98integrity) and systemd (via core/ima-setup) have support for loading keys at boot time. + +What: security/evm_xattrs +Date: April 2018 +Contact: Matthew Garrett <mjg59@google.com> +Description: + Shows the set of extended attributes used to calculate or + validate the EVM signature, and allows additional attributes + to be added at runtime. Adding additional extended attributes + will result in any existing signatures generated without the + additional attributes becoming invalid, and any signatures + generated after additional attributes are added will only be + valid if the same additional attributes are configured on + system boot. diff --git a/security/integrity/evm/evm_secfs.c b/security/integrity/evm/evm_secfs.c index feba03bbedae..3b371125d439 100644 --- a/security/integrity/evm/evm_secfs.c +++ b/security/integrity/evm/evm_secfs.c @@ -20,6 +20,7 @@ #include "evm.h" static struct dentry *evm_init_tpm; +static struct dentry *evm_xattrs; /** * evm_read_key - read() for <securityfs>/evm @@ -107,13 +108,102 @@ static const struct file_operations evm_key_ops = { .write = evm_write_key, }; -int __init evm_init_secfs(void) +/** + * evm_read_xattrs - read() for <securityfs>/evm_xattrs + * + * @filp: file pointer, not actually used + * @buf: where to put the result + * @count: maximum to send along + * @ppos: where to start + * + * Returns number of bytes read or error code, as appropriate + */ +static ssize_t evm_read_xattrs(struct file *filp, char __user *buf, + size_t count, loff_t *ppos) { - int error = 0; + char *temp; + int offset = 0; + ssize_t rc, size = 0; + struct xattr_list *xattr; + + if (*ppos != 0) + return 0; - evm_init_tpm = securityfs_create_file("evm", S_IRUSR | S_IRGRP, - NULL, NULL, &evm_key_ops); + list_for_each_entry(xattr, &evm_config_xattrnames, list) + size += strlen(xattr->name) + 1; + + temp = kmalloc(size + 1, GFP_KERNEL); + if (!temp) + return -ENOMEM; + + list_for_each_entry(xattr, &evm_config_xattrnames, list) { + sprintf(temp + offset, "%s\n", xattr->name); + offset += strlen(xattr->name) + 1; + } + + rc = simple_read_from_buffer(buf, count, ppos, temp, strlen(temp)); + + return rc; +} + +/** + * evm_write_xattrs - write() for <securityfs>/evm_xattrs + * @file: file pointer, not actually used + * @buf: where to get the data from + * @count: bytes sent + * @ppos: where to start + * + * Returns number of bytes written or error code, as appropriate + */ +static ssize_t evm_write_xattrs(struct file *file, const char __user *buf, + size_t count, loff_t *ppos) +{ + int len; + struct xattr_list *xattr; + + if (!capable(CAP_SYS_ADMIN)) + return -EPERM; + + if (*ppos != 0) + return -EINVAL; + + xattr = kmalloc(sizeof(struct xattr_list), GFP_KERNEL); + if (!xattr) + return -ENOMEM; + + xattr->name = memdup_user_nul(buf, count); + if (!xattr->name) { + kfree(xattr); + return -ENOMEM; + } + + /* Remove any trailing newline */ + len = strlen(xattr->name); + if (xattr->name[len-1] == '\n') + xattr->name[len-1] = '\0'; + + list_add_tail(&xattr->list, &evm_config_xattrnames); + return count; +} + +static const struct file_operations evm_xattr_ops = { + .read = evm_read_xattrs, + .write = evm_write_xattrs, +}; + +int __init evm_init_secfs(void) +{ + evm_init_tpm = securityfs_create_file("evm", 0440, NULL, NULL, + &evm_key_ops); if (!evm_init_tpm || IS_ERR(evm_init_tpm)) - error = -EFAULT; - return error; + return -EFAULT; + + evm_xattrs = securityfs_create_file("evm_xattrs", 0440, NULL, NULL, + &evm_xattr_ops); + if (!evm_xattrs || IS_ERR(evm_xattrs)) { + securityfs_remove(evm_init_tpm); + return -EFAULT; + } + + return 0; }
Sites may wish to provide additional metadata alongside files in order to make more fine-grained security decisions[1]. The security of this is enhanced if this metadata is protected, something that EVM makes possible. However, the kernel cannot know about the set of extended attributes that local admins may wish to protect, and hardcoding this policy in the kernel makes it difficult to change over time and less convenient for distributions to enable. This patch adds a new /sys/kernel/security/evm_xattrs node, which can be read to obtain the current set of EVM-protected extended attributes or written to in order to add new entries. Extending this list will not change the validity of any existing signatures provided that the file in question does not have any of the additional extended attributes - missing xattrs are skipped when calculating the EVM hash. [1] For instance, a package manager could install information about the package uploader in an additional extended attribute. Local LSM policy could then be associated with that extended attribute in order to restrict the privileges available to packages from less trusted uploaders. Signed-off-by: Matthew Garrett <mjg59@google.com> --- Documentation/ABI/testing/evm | 13 ++++ security/integrity/evm/evm_secfs.c | 102 +++++++++++++++++++++++++++-- 2 files changed, 109 insertions(+), 6 deletions(-)