Skip to contentSkip to conversation

binfmt_misc: add two-steps registration (opt-in)

PATCH In review Domenico Andreoli 2 participants 4 messages Mar 1, 2022 – Mar 10, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 binfmt_misc: add two-steps registration (opt-in)

From: Domenico Andreoli <[email protected]>

Experimenting with new interpreter configurations can lead to annoying failures, when the system is left unable to load ELF binaries power cycling is the only way to get it back operational.

This patch tries to mitigate such conditions by adding an opt-in two-steps registration.

A new optional field is added to the configuration string, it's an expiration interval for the newly added interpreter. If the user is not able to confirm in time, possibly because the system is broken, the new interpreter is automatically disabled.

---
 Documentation/admin-guide/binfmt-misc.rst |   12 +++
 fs/binfmt_misc.c                          |  112 ++++++++++++++++++++++++++++--
 2 files changed, 113 insertions(+), 11 deletions(-)
Documentation/admin-guide/binfmt-misc.rst +10 -2
Index: b/Documentation/admin-guide/binfmt-misc.rst
===================================================================
@@ -16,8 +16,8 @@ First you must mount binfmt_misc::
 	mount binfmt_misc -t binfmt_misc /proc/sys/fs/binfmt_misc
 
 To actually register a new binary type, you have to set up a string looking like
-``:name:type:offset:magic:mask:interpreter:flags`` (where you can choose the
-``:`` upon your needs) and echo it to ``/proc/sys/fs/binfmt_misc/register``.
+``:name:type:offset:magic:mask:interpreter:flags:timeout`` (where you can choose
+the ``:`` upon your needs) and echo it to ``/proc/sys/fs/binfmt_misc/register``.
 
 Here is what the fields mean:
 
@@ -88,6 +88,14 @@ Here is what the fields mean:
 	    emulation is installed and uses the opened image to spawn the
 	    emulator, meaning it is always available once installed,
 	    regardless of how the environment changes.
+- ``timeout``
+  is an optional field; the newly added interpreter is automatically
+  disabled after the specified number of seconds. To cancel such
+  count down, cat or echo something to ``/proc/.../the_name``.  This
+  registration in two steps allows recovering a system left unusable
+  by some wrong configuration. A timeout of 0 seconds effectively adds
+  a disabled interpreter.  Values smaller than 0 or bigger than 120
+  are invalid.
 
 
 There are some restrictions:
fs/binfmt_misc.c +103 -9

209 lines not on this page · raw message

#1 lore raw

Hi Eric and Kees,

Quoted 6 lines · Domenico Andreoli

From: Domenico Andreoli <[email protected]>

Experimenting with new interpreter configurations can lead to annoying failures, when the system is left unable to load ELF binaries power cycling is the only way to get it back operational.

Quoted 7 lines

This patch tries to mitigate such conditions by adding an opt-in two-steps registration.

A new optional field is added to the configuration string, it's an expiration interval for the newly added interpreter. If the user is not able to confirm in time, possibly because the system is broken, the new interpreter is automatically disabled.

I was wondering whether, maybe, likely, you missed this patch of mine.

It would be great if you could just ack (or nack) it for later.

Thanks!
Domenico

Quoted 266 lines
---
 Documentation/admin-guide/binfmt-misc.rst |   12 +++
 fs/binfmt_misc.c                          |  112 ++++++++++++++++++++++++++++--
 2 files changed, 113 insertions(+), 11 deletions(-)

Index: b/Documentation/admin-guide/binfmt-misc.rst
===================================================================

Documentation/admin-guide/binfmt-misc.rst +10 -2
@@ -16,8 +16,8 @@ First you must mount binfmt_misc::
 	mount binfmt_misc -t binfmt_misc /proc/sys/fs/binfmt_misc
 
 To actually register a new binary type, you have to set up a string looking like
-``:name:type:offset:magic:mask:interpreter:flags`` (where you can choose the
-``:`` upon your needs) and echo it to ``/proc/sys/fs/binfmt_misc/register``.
+``:name:type:offset:magic:mask:interpreter:flags:timeout`` (where you can choose
+the ``:`` upon your needs) and echo it to ``/proc/sys/fs/binfmt_misc/register``.
 
 Here is what the fields mean:
 
@@ -88,6 +88,14 @@ Here is what the fields mean:
 	    emulation is installed and uses the opened image to spawn the
 	    emulator, meaning it is always available once installed,
 	    regardless of how the environment changes.
+- ``timeout``
+  is an optional field; the newly added interpreter is automatically
+  disabled after the specified number of seconds. To cancel such
+  count down, cat or echo something to ``/proc/.../the_name``.  This
+  registration in two steps allows recovering a system left unusable
+  by some wrong configuration. A timeout of 0 seconds effectively adds
+  a disabled interpreter.  Values smaller than 0 or bigger than 120
+  are invalid.
 
 
 There are some restrictions:

Index: b/fs/binfmt_misc.c
===================================================================

fs/binfmt_misc.c +103 -9
@@ -27,6 +27,8 @@
 #include <linux/syscalls.h>
 #include <linux/fs.h>
 #include <linux/uaccess.h>
+#include <linux/spinlock.h>
+#include <linux/timer.h>
 
 #include "internal.h"
 
@@ -49,6 +51,8 @@ enum {Enabled, Magic};
 #define MISC_FMT_CREDENTIALS (1 << 29)
 #define MISC_FMT_OPEN_FILE (1 << 28)
 
+struct node_timer;
+
 typedef struct {
 	struct list_head list;
 	unsigned long flags;		/* type, status, etc. */
@@ -60,8 +64,15 @@ typedef struct {
 	char *name;
 	struct dentry *dentry;
 	struct file *interp_file;
+	struct node_timer *auto_disable;
+	spinlock_t auto_disable_lock;
 } Node;
 
+struct node_timer {
+	struct timer_list timer;
+	Node *node;
+};
+
 static DEFINE_RWLOCK(entries_lock);
 static struct file_system_type bm_fs_type;
 static struct vfsmount *bm_mnt;
@@ -69,19 +80,30 @@ static int entry_count;
 
 /*
  * Max length of the register string.  Determined by:
- *  - 7 delimiters
- *  - name:   ~50 bytes
- *  - type:   1 byte
- *  - offset: 3 bytes (has to be smaller than BINPRM_BUF_SIZE)
- *  - magic:  128 bytes (512 in escaped form)
- *  - mask:   128 bytes (512 in escaped form)
- *  - interp: ~50 bytes
- *  - flags:  5 bytes
+ *  - 8 delimiters
+ *  - name:    ~50 bytes
+ *  - type:    1 byte
+ *  - offset:  3 bytes (has to be smaller than BINPRM_BUF_SIZE)
+ *  - magic:   128 bytes (512 in escaped form)
+ *  - mask:    128 bytes (512 in escaped form)
+ *  - interp:  ~50 bytes
+ *  - flags:   5 bytes
+ *  - timeout: 3 bytes
  * Round that up a bit, and then back off to hold the internal data
  * (like struct Node).
  */
 #define MAX_REGISTER_LENGTH 1920
 
+#define MAX_AUTO_DISABLE_TIMEOUT 120
+
+static void auto_disable_timer_fn(struct timer_list *t)
+{
+	struct node_timer *timer = container_of(t, struct node_timer, timer);
+
+	clear_bit(Enabled, &timer->node->flags);
+	pr_info("%s: auto-disabled\n", timer->node->name);
+}
+
 /*
  * Check if we support the binfmt
  * if we do, return the node, else NULL
@@ -266,6 +288,41 @@ static char *check_special_flags(char *s
 	return p;
 }
 
+static char *setup_auto_disable(char *p, char *endp, Node *e)
+{
+	unsigned int timeout;
+	char buf[4] = {0};
+
+	while (endp[-1] == '\n')
+		endp--;
+	if (p >= endp || *p != ':' || ++p == endp)
+		return p;
+
+	endp = min(endp, p + sizeof(buf) - 1);
+	memcpy(buf, p, (size_t) (endp - p));
+
+	if (kstrtouint(buf, 10, &timeout) || timeout > MAX_AUTO_DISABLE_TIMEOUT) {
+		pr_info("%s: invalid timeout: %s\n", e->name, buf);
+		return p;
+	}
+
+	if (timeout == 0) {
+		e->flags &= ~(1 << Enabled);
+		return endp;
+	}
+
+	e->auto_disable = kmalloc(sizeof(struct node_timer), GFP_KERNEL);
+	if (!e->auto_disable)
+		return ERR_PTR(-ENOMEM);
+
+	pr_info("%s: auto-disable in %u seconds\n", e->name, timeout);
+
+	timer_setup(&e->auto_disable->timer, auto_disable_timer_fn, 0);
+	e->auto_disable->timer.expires = jiffies + timeout * HZ;
+	e->auto_disable->node = e;
+	return endp;
+}
+
 /*
  * This registers a new binary format, it recognises the syntax
  * ':name:type:offset:magic:mask:interpreter:flags'
@@ -273,7 +330,7 @@ static char *check_special_flags(char *s
  */
 static Node *create_entry(const char __user *buffer, size_t count)
 {
-	Node *e;
+	Node *e = NULL;
 	int memsize, err;
 	char *buf, *p;
 	char del;
@@ -297,6 +354,8 @@ static Node *create_entry(const char __u
 	if (copy_from_user(buf, buffer, count))
 		goto efault;
 
+	spin_lock_init(&e->auto_disable_lock);
+
 	del = *p++;	/* delimeter */
 
 	pr_debug("register: delim: %#x {%c}\n", del, del);
@@ -454,6 +513,14 @@ static Node *create_entry(const char __u
 
 	/* Parse the 'flags' field. */
 	p = check_special_flags(p, e);
+
+	/* Parse the 'timeout' field and init the auto-disable timer. */
+	p = setup_auto_disable(p, buf + count, e);
+	if (IS_ERR(p)) {
+		err = PTR_ERR(p);
+		goto out;
+	}
+
 	if (*p == '\n')
 		p++;
 	if (p != buf + count)
@@ -462,12 +529,15 @@ static Node *create_entry(const char __u
 	return e;
 
 out:
+	kfree(e);
 	return ERR_PTR(err);
 
 efault:
 	kfree(e);
 	return ERR_PTR(-EFAULT);
 einval:
+	if (e)
+		kfree(e->auto_disable);
 	kfree(e);
 	return ERR_PTR(-EINVAL);
 }
@@ -499,6 +569,21 @@ static int parse_command(const char __us
 
 /* generic stuff */
 
+static void cancel_auto_disable(Node *e)
+{
+	struct node_timer *auto_disable = NULL;
+
+	spin_lock(&e->auto_disable_lock);
+	swap(e->auto_disable, auto_disable);
+	spin_unlock(&e->auto_disable_lock);
+
+	if (auto_disable) {
+		if (del_timer_sync(&auto_disable->timer))
+			pr_info("%s: cancelled auto-disable\n", e->name);
+		kfree(auto_disable);
+	}
+}
+
 static void entry_status(Node *e, char *page)
 {
 	char *dp = page;
@@ -559,6 +644,8 @@ static void bm_evict_inode(struct inode
 
 	if (e && e->flags & MISC_FMT_OPEN_FILE)
 		filp_close(e->interp_file, NULL);
+	if (e)
+		cancel_auto_disable(e);
 
 	clear_inode(inode);
 	kfree(e);
@@ -588,6 +675,8 @@ bm_entry_read(struct file *file, char __
 	ssize_t res;
 	char *page;
 
+	cancel_auto_disable(e);
+
 	page = (char *) __get_free_page(GFP_KERNEL);
 	if (!page)
 		return -ENOMEM;
@@ -607,6 +696,8 @@ static ssize_t bm_entry_write(struct fil
 	Node *e = file_inode(file)->i_private;
 	int res = parse_command(buffer, count);
 
+	cancel_auto_disable(e);
+
 	switch (res) {
 	case 1:
 		/* Disable this handler. */
@@ -699,6 +790,9 @@ static ssize_t bm_register_write(struct
 	list_add(&e->list, &entries);
 	write_unlock(&entries_lock);
 
+	if (e->auto_disable)
+		add_timer(&e->auto_disable->timer);
+
 	err = 0;
 out2:
 	dput(dentry);
Signature · 2 lines
rsa4096: 3B10 0CA1 8674 ACBA B4FE  FCD2 CE5B CF17 9960 DE13
ed25519: FFB4 0CC3 7F2E 091D F7DA  356E CC79 2832 ED38 CB05
#2 lore raw
In reply to Domenico Andreoli
Quoted 6 lines · Domenico Andreoli

From: Domenico Andreoli <[email protected]>

Experimenting with new interpreter configurations can lead to annoying failures, when the system is left unable to load ELF binaries power cycling is the only way to get it back operational.

Quoted 7 lines

This patch tries to mitigate such conditions by adding an opt-in two-steps registration.

A new optional field is added to the configuration string, it's an expiration interval for the newly added interpreter. If the user is not able to confirm in time, possibly because the system is broken, the new interpreter is automatically disabled.

Hi!

As this both changes the userspace API and adds timers, I'd like the change to be really well justified. Can you explain the conditions you get into that can't be escaped by just disabling the bad binfmt_misc entry?

-Kees

Signature · 1 line
Kees Cook
#3 lore raw
Quoted 10 lines · Kees Cook
Quoted 9 lines · Domenico Andreoli

From: Domenico Andreoli <[email protected]>

Experimenting with new interpreter configurations can lead to annoying failures, when the system is left unable to load ELF binaries power cycling is the only way to get it back operational.

This patch tries to mitigate such conditions by adding an opt-in two-steps registration.

Quoted 6 lines
Quoted 4 lines

A new optional field is added to the configuration string, it's an expiration interval for the newly added interpreter. If the user is not able to confirm in time, possibly because the system is broken, the new interpreter is automatically disabled.

Hi!

Hi!

Quoted 2 lines

As this both changes the userspace API and adds timers, I'd like the

Right but 1. it's backward compatible, 2. it fails on unsupporting kernels.

Curiosity, I understand why API changes require care but what's so special about the timers?

Quoted 3 lines

change to be really well justified. Can you explain the conditions you get into that can't be escaped by just disabling the bad binfmt_misc entry?

It happened when I somehow messed up with the ELF loader of my system, it was the very first time I tried to manually configure qemu-user-static for a foreign architecture.

Suddenly I could not do anything, no ls, no cat. Did not realize that my shell has built-in echo and that I could cut-and-paste the path for disabling the bad interpreter. I did not investigate what I did wrong or what I could do better, I simply didn't do it again.

I just got a deeper understanding of the note in Debian's update-binfmts manpage:

If you're not careful, you can break your system with update-binfmts. An easy way to do this is to register an ELF binary as a handler for ELF, which will almost certainly cause your system to hang immediately; even if it doesn't, you won't be able to run update-binfmts to fix it.

I shot on my foot and I thought the API could be made a bit more friendly.

Thanks,
Dom

Quoted 5 lines

-Kees

Signature · 1 line
Kees Cook
Signature · 2 lines
rsa4096: 3B10 0CA1 8674 ACBA B4FE  FCD2 CE5B CF17 9960 DE13
ed25519: FFB4 0CC3 7F2E 091D F7DA  356E CC79 2832 ED38 CB05