On Wed, Jun 01, 2022 at 02:09:47PM -0700, Axel Rasmussen wrote:

Historically, it has been shown that intercepting kernel faults with userfaultfd (thereby forcing the kernel to wait for an arbitrary amount of time) can be exploited, or at least can make some kinds of exploits easier. So, in 37cd0575b8 "userfaultfd: add UFFD_USER_MODE_ONLY" we changed things so, in order for kernel faults to be handled by userfaultfd, either the process needs CAP_SYS_PTRACE, or this sysctl must be configured so that any unprivileged user can do it.

In a typical implementation of a hypervisor with live migration (take QEMU/KVM as one such example), we do indeed need to be able to handle kernel faults. But, both options above are less than ideal:

• Toggling the sysctl increases attack surface by allowing any unprivileged user to do it.

• Granting the live migration process CAP_SYS_PTRACE gives it this ability, but *also* the ability to "observe and control the execution of another process [...], and examine and change [its] memory and registers" (from ptrace(2)). This isn't something we need or want to be able to do, so granting this permission violates the "principle of least privilege".

This is all a long winded way to say: we want a more fine-grained way to grant access to userfaultfd, without granting other additional permissions at the same time.

To achieve this, add a /dev/userfaultfd misc device. This device provides an alternative to the userfaultfd(2) syscall for the creation of new userfaultfds. The idea is, any userfaultfds created this way will be able to handle kernel faults, without the caller having any special capabilities. Access to this mechanism is instead restricted using e.g. standard filesystem permissions.

Signed-off-by: Axel Rasmussen axelrasmussen@google.com

fs/userfaultfd.c | 76 ++++++++++++++++++++++++++------ include/uapi/linux/userfaultfd.h | 4 ++ 2 files changed, 66 insertions(+), 14 deletions(-)

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c index e943370107d0..8b92c1398169 100644 --- a/fs/userfaultfd.c +++ b/fs/userfaultfd.c @@ -30,6 +30,7 @@ #include <linux/security.h> #include <linux/hugetlb.h> #include <linux/swapops.h> +#include <linux/miscdevice.h> int sysctl_unprivileged_userfaultfd __read_mostly; @@ -413,13 +414,8 @@ vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason) if (ctx->features & UFFD_FEATURE_SIGBUS) goto out;

• if ((vmf->flags & FAULT_FLAG_USER) == 0 &&

•    ctx->flags & UFFD_USER_MODE_ONLY) {
  

• printk_once(KERN_WARNING "uffd: Set unprivileged_userfaultfd "
  

• 	"sysctl knob to 1 if kernel faults must be handled "
  

• 	"without obtaining CAP_SYS_PTRACE capability\n");
  

• if (!(vmf->flags & FAULT_FLAG_USER) && (ctx->flags & UFFD_USER_MODE_ONLY)) goto out;

• }

/* * If it's already released don't get it. This avoids to loop @@ -2052,19 +2048,33 @@ static void init_once_userfaultfd_ctx(void *mem) seqcount_spinlock_init(&ctx->refile_seq, &ctx->fault_pending_wqh.lock); } -SYSCALL_DEFINE1(userfaultfd, int, flags) +static inline bool userfaultfd_allowed(bool is_syscall, int flags) +{

• bool kernel_faults = !(flags & UFFD_USER_MODE_ONLY);
• bool allow_unprivileged = sysctl_unprivileged_userfaultfd;
• /* userfaultfd(2) access is controlled by sysctl + capability. */
• if (is_syscall && kernel_faults) {

• if (!allow_unprivileged && !capable(CAP_SYS_PTRACE))
  

• 	return false;
  

• }
• /*

• * For /dev/userfaultfd, access is to be controlled using e.g.
  

• * permissions on the device node. We assume this is correctly
  

• * configured by userspace, so we simply allow access here.
  

• */
  

• return true;

+}

This helper reads a bit weird because potentially it constantly returns "true" for !syscall use case but it's very not obvious..

Would it be cleaner to not pass in the bool at all? Something like (I also un-nested some of the condition checks, hopefully it'll be easier to read):

bool userfaultfd_syscall_allowed(int flags) { /* Userspace-only page faults are always allowed */ if (flags & UFFD_USER_MODE_ONLY) return true;

/* * The user is requesting kernel fault capabilities. Privileged * users are always allowed even for kernel fault traps. */ if (capable(CAP_SYS_PTRACE)) return true;

/* Whether we allow unprivileged users for kernel faults? */ return sysctl_unprivileged_userfaultfd; }

Then below...

+static int new_userfaultfd(bool is_syscall, int flags) { struct userfaultfd_ctx *ctx; int fd;

• if (!sysctl_unprivileged_userfaultfd &&

•    (flags & UFFD_USER_MODE_ONLY) == 0 &&
  

•    !capable(CAP_SYS_PTRACE)) {
  

• printk_once(KERN_WARNING "uffd: Set unprivileged_userfaultfd "
  

• 	"sysctl knob to 1 if kernel faults must be handled "
  

• 	"without obtaining CAP_SYS_PTRACE capability\n");
  

• if (!userfaultfd_allowed(is_syscall, flags)) return -EPERM;

• }

.. we could write it as:

if (is_syscall && !userfaultfd_syscall_allowed(flags)) return -EPERM;

What do you think?

BUG_ON(!current->mm); @@ -2083,6 +2093,10 @@ SYSCALL_DEFINE1(userfaultfd, int, flags) refcount_set(&ctx->refcount, 1); ctx->flags = flags; ctx->features = 0;

• /*

• * If UFFD_USER_MODE_ONLY is not set, then userfaultfd_allowed() above
  

• * decided that kernel faults were allowed and should be handled.
  

• */
  

Hmm.. why this needs to be added above "released=false"? Did you want to add this (perhaps) above "flags" instead?

IMHO when people reading the flags it'll be clear already on how it was handled, the thing is the comment probably hide deep anyway so I'd consider omitting it.

The rest looks good to me, thanks.

ctx->released = false; atomic_set(&ctx->mmap_changing, 0); ctx->mm = current->mm; @@ -2098,8 +2112,42 @@ SYSCALL_DEFINE1(userfaultfd, int, flags) return fd; } +SYSCALL_DEFINE1(userfaultfd, int, flags) +{

• return new_userfaultfd(true, flags);

+}

+static int userfaultfd_dev_open(struct inode *inode, struct file *file) +{

• return 0;

+}

+static long userfaultfd_dev_ioctl(struct file *file, unsigned int cmd, unsigned long flags) +{

• if (cmd != USERFAULTFD_IOC_NEW)

• return -EINVAL;
  

• return new_userfaultfd(false, flags);

+}

+static const struct file_operations userfaultfd_dev_fops = {

• .open = userfaultfd_dev_open,
• .unlocked_ioctl = userfaultfd_dev_ioctl,
• .compat_ioctl = userfaultfd_dev_ioctl,
• .owner = THIS_MODULE,
• .llseek = noop_llseek,

+};

+static struct miscdevice userfaultfd_misc = {

• .minor = MISC_DYNAMIC_MINOR,
• .name = "userfaultfd",
• .fops = &userfaultfd_dev_fops

+};

static int __init userfaultfd_init(void) {

• WARN_ON(misc_register(&userfaultfd_misc));
• userfaultfd_ctx_cachep = kmem_cache_create("userfaultfd_ctx_cache", sizeof(struct userfaultfd_ctx), 0,

diff --git a/include/uapi/linux/userfaultfd.h b/include/uapi/linux/userfaultfd.h index 7d32b1e797fb..005e5e306266 100644 --- a/include/uapi/linux/userfaultfd.h +++ b/include/uapi/linux/userfaultfd.h @@ -12,6 +12,10 @@ #include <linux/types.h> +/* ioctls for /dev/userfaultfd */ +#define USERFAULTFD_IOC 0xAA +#define USERFAULTFD_IOC_NEW _IO(USERFAULTFD_IOC, 0x00)

/*

• If the UFFDIO_API is upgraded someday, the UFFDIO_UNREGISTER and
• UFFDIO_WAKE ioctls should be defined as _IOW and not as _IOR. In

-- 2.36.1.255.ge46751e96f-goog

On Wed, Jun 01, 2022 at 02:09:49PM -0700, Axel Rasmussen wrote:

Explain the different ways to create a new userfaultfd, and how access control works for each way.

Signed-off-by: Axel Rasmussen axelrasmussen@google.com

Documentation/admin-guide/mm/userfaultfd.rst | 40 ++++++++++++++++++-- Documentation/admin-guide/sysctl/vm.rst | 3 ++ 2 files changed, 40 insertions(+), 3 deletions(-)

diff --git a/Documentation/admin-guide/mm/userfaultfd.rst b/Documentation/admin-guide/mm/userfaultfd.rst index 6528036093e1..9bae1acd431f 100644 --- a/Documentation/admin-guide/mm/userfaultfd.rst +++ b/Documentation/admin-guide/mm/userfaultfd.rst @@ -17,7 +17,10 @@ of the ``PROT_NONE+SIGSEGV`` trick. Design ====== -Userfaults are delivered and resolved through the ``userfaultfd`` syscall. +Userspace creates a new userfaultfd, initializes it, and registers one or more +regions of virtual memory with it. Then, any page faults which occur within the +region(s) result in a message being delivered to the userfaultfd, notifying +userspace of the fault. The ``userfaultfd`` (aside from registering and unregistering virtual memory ranges) provides two primary functionalities: @@ -34,12 +37,11 @@ The real advantage of userfaults if compared to regular virtual memory management of mremap/mprotect is that the userfaults in all their operations never involve heavyweight structures like vmas (in fact the ``userfaultfd`` runtime load never takes the mmap_lock for writing).

Vmas are not suitable for page- (or hugepage) granular fault tracking when dealing with virtual address spaces that could span Terabytes. Too many vmas would be needed for that. -The ``userfaultfd`` once opened by invoking the syscall, can also be +The ``userfaultfd``, once created, can also be passed using unix domain sockets to a manager process, so the same manager process could handle the userfaults of a multitude of different processes without them being aware about what is going on @@ -50,6 +52,38 @@ is a corner case that would currently return ``-EBUSY``). API === +Creating a userfaultfd +----------------------

+There are two ways to create a new userfaultfd, each of which provide ways to +restrict access to this functionality (since historically userfaultfds which +handle kernel page faults have been a useful tool for exploiting the kernel).

+The first way, supported by older kernels, is the userfaultfd(2) syscall.

How about "supported since userfaultfd was introduced"? Otherwise the reader can get a feeling that the syscall won't work on new kernels but it will.

+Access to this is controlled in several ways:

+- By default, the userfaultfd will be able to handle kernel page faults. This

s/kernel/both user and kernel/?

• can be disabled by passing in UFFD_USER_MODE_ONLY.

+- If vm.unprivileged_userfaultfd is 0, then the caller must *either* have

• CAP_SYS_PTRACE, or pass in UFFD_USER_MODE_ONLY.

+- If vm.unprivileged_userfaultfd is 1, then no particular privilege is needed to

• use this syscall, even if UFFD_USER_MODE_ONLY is *not* set.

The separation of above three paragraphs do not feel very clear to me to understand these flags.. Entry 1) was trying to define UFFD_USER_MODE_ONLY, but entry 2) was also referring to it in another context.

How about using two paragraphs to explain these two flags one by one? My try..

The user can always creates an userfaultfd that only traps userspace page faults only. To achieve it, one can create the userfaultfd object using the syscall userfaultfd() with flag UFFD_USER_MODE_ONLY passed in.

If the user would like to also trap kernel page faults for the address space, then either the process needs to have CAP_SYS_PTRACE capability, or the system must have vm.unprivileged_userfaultfd set to 1. By default, vm.unprivileged_userfaultfd is set to 0.

+The second way, added to the kernel more recently, is by opening and issuing a +USERFAULTFD_IOC_NEW ioctl to /dev/userfaultfd. This method yields equivalent +userfaultfds to the userfaultfd(2) syscall; its benefit is in how access to +creating userfaultfds is controlled.

Since the benefit is immediately mentioned next, how about dropping "its benefit is in how ... is controlled" and just connect these two paragraphs?

Again, please take it with a grain of salt on my English-related comments (it means all comment above :).

Thanks,

+Access to /dev/userfaultfd is controlled via normal filesystem permissions +(user/group/mode for example), which gives fine grained access to userfaultfd +specifically, without also granting other unrelated privileges at the same time +(as e.g. granting CAP_SYS_PTRACE would do).

+Initializing up a userfaultfd +-----------------------------

When first opened the ``userfaultfd`` must be enabled invoking the ``UFFDIO_API`` ioctl specifying a ``uffdio_api.api`` value set to ``UFFD_API`` (or a later API version) which will specify the ``read/POLLIN`` protocol diff --git a/Documentation/admin-guide/sysctl/vm.rst b/Documentation/admin-guide/sysctl/vm.rst index d7374a1e8ac9..e3a952d1fd35 100644 --- a/Documentation/admin-guide/sysctl/vm.rst +++ b/Documentation/admin-guide/sysctl/vm.rst @@ -927,6 +927,9 @@ calls without any restrictions. The default value is 0. +An alternative to this sysctl / the userfaultfd(2) syscall is to create +userfaultfds via /dev/userfaultfd. See +Documentation/admin-guide/mm/userfaultfd.rst. user_reserve_kbytes =================== -- 2.36.1.255.ge46751e96f-goog

On Wed, Jun 01, 2022 at 02:09:50PM -0700, Axel Rasmussen wrote:

Instead of always testing both userfaultfd(2) and /dev/userfaultfd, let the user choose which to test.

As with other test features, change the behavior based on a new command line flag. Introduce the idea of "test mods", which are generic (not specific to a test type) modifications to the behavior of the test. This is sort of borrowed from this RFC patch series [1], but simplified a bit.

The benefit is, in "typical" configurations this test is somewhat slow (say, 30sec or something). Testing both clearly doubles it, so it may not always be desirable, as users are likely to use one or the other, but never both, in the "real world".

Yes, when I was reading your other patch I was wishing no timeout of any of QE testbeds could trigger.

If you plan to remove it anyway, mind to squash these?

Signed-off-by: Axel Rasmussen axelrasmussen@google.com

tools/testing/selftests/vm/userfaultfd.c | 41 +++++++++++++++++------- 1 file changed, 30 insertions(+), 11 deletions(-)

diff --git a/tools/testing/selftests/vm/userfaultfd.c b/tools/testing/selftests/vm/userfaultfd.c index 1badb5d31bf9..aeee6f1ad8ef 100644 --- a/tools/testing/selftests/vm/userfaultfd.c +++ b/tools/testing/selftests/vm/userfaultfd.c @@ -128,6 +128,8 @@ struct uffd_stats { const char *examples = "# Run anonymous memory test on 100MiB region with 99999 bounces:\n" "./userfaultfd anon 100 99999\n\n"

• "# Run the same anonymous memory test, but using /dev/userfaultfd:\n"
• "./userfaultfd anon:dev 100 99999\n\n" "# Run share memory test on 1GiB region with 99 bounces:\n" "./userfaultfd shmem 1000 99\n\n" "# Run hugetlb memory test on 256MiB region with 50 bounces:\n"

@@ -144,6 +146,13 @@ static void usage(void) "[hugetlbfs_file]\n\n"); fprintf(stderr, "Supported <test type>: anon, hugetlb, " "hugetlb_shared, shmem\n\n");

• fprintf(stderr, "'Test mods' can be joined to the test type string with a ':'. "

• "Supported mods:\n");
  

• fprintf(stderr, "\tdev - Use /dev/userfaultfd instead of userfaultfd(2)\n");

That looks better than running everything by default twice. Though let's also provide "syscall" if we're at it, anyway?

Thanks,

• fprintf(stderr, "\nExample test mod usage:\n");
• fprintf(stderr, "# Run anonymous memory test with /dev/userfaultfd:\n");
• fprintf(stderr, "./userfaultfd anon:dev 100 99999\n\n");
• fprintf(stderr, "Examples:\n\n"); fprintf(stderr, "%s", examples); exit(1);

@@ -1607,8 +1616,6 @@ unsigned long default_huge_page_size(void) static void set_test_type(const char *type) {

• uint64_t features = UFFD_API_FEATURES;
• if (!strcmp(type, "anon")) { test_type = TEST_ANON; uffd_test_ops = &anon_uffd_test_ops;

@@ -1626,10 +1633,28 @@ static void set_test_type(const char *type) test_type = TEST_SHMEM; uffd_test_ops = &shmem_uffd_test_ops; test_uffdio_minor = true;

• } else {

• err("Unknown test type: %s", type);
  

• }

+}

+static void parse_test_type_arg(const char *raw_type) +{

• char *buf = strdup(raw_type);
• uint64_t features = UFFD_API_FEATURES;
• while (buf) {

• const char *token = strsep(&buf, ":");
  

• if (!test_type)
  

• 	set_test_type(token);
  

• else if (!strcmp(token, "dev"))
  

• 	test_dev_userfaultfd = true;
  

• else
  

• 	err("unrecognized test mod '%s'", token);
  

  }

• if (!test_type)

• err("failed to parse test type argument: '%s'", raw_type);
  

• if (test_type == TEST_HUGETLB) page_size = default_huge_page_size(); else

@@ -1676,7 +1701,7 @@ int main(int argc, char **argv) err("failed to arm SIGALRM"); alarm(ALARM_INTERVAL_SECS);

• set_test_type(argv[1]);

• parse_test_type_arg(argv[1]);

nr_cpus = sysconf(_SC_NPROCESSORS_ONLN); nr_pages_per_cpu = atol(argv[2]) * 1024*1024 / page_size / @@ -1714,12 +1739,6 @@ int main(int argc, char **argv) } printf("nr_pages: %lu, nr_pages_per_cpu: %lu\n", nr_pages, nr_pages_per_cpu);

• test_dev_userfaultfd = false;
• if (userfaultfd_stress())

• return 1;
  

• test_dev_userfaultfd = true; return userfaultfd_stress();

} -- 2.36.1.255.ge46751e96f-goog