Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1===============================================
2Guarded Control Stack support for AArch64 Linux
3===============================================
4
5This document outlines briefly the interface provided to userspace by Linux in
6order to support use of the ARM Guarded Control Stack (GCS) feature.
7
8This is an outline of the most important features and issues only and not
9intended to be exhaustive.
10
11
12
131. General
14-----------
15
16* GCS is an architecture feature intended to provide greater protection
17 against return oriented programming (ROP) attacks and to simplify the
18 implementation of features that need to collect stack traces such as
19 profiling.
20
21* When GCS is enabled a separate guarded control stack is maintained by the
22 PE which is writeable only through specific GCS operations. This
23 stores the call stack only, when a procedure call instruction is
24 performed the current PC is pushed onto the GCS and on RET the
25 address in the LR is verified against that on the top of the GCS.
26
27* When active the current GCS pointer is stored in the system register
28 GCSPR_EL0. This is readable by userspace but can only be updated
29 via specific GCS instructions.
30
31* The architecture provides instructions for switching between guarded
32 control stacks with checks to ensure that the new stack is a valid
33 target for switching.
34
35* The functionality of GCS is similar to that provided by the x86 Shadow
36 Stack feature, due to sharing of userspace interfaces the ABI refers to
37 shadow stacks rather than GCS.
38
39* Support for GCS is reported to userspace via HWCAP_GCS in the aux vector
40 AT_HWCAP entry.
41
42* GCS is enabled per thread. While there is support for disabling GCS
43 at runtime this should be done with great care.
44
45* GCS memory access faults are reported as normal memory access faults.
46
47* GCS specific errors (those reported with EC 0x2d) will be reported as
48 SIGSEGV with a si_code of SEGV_CPERR (control protection error).
49
50* GCS is supported only for AArch64.
51
52* On systems where GCS is supported GCSPR_EL0 is always readable by EL0
53 regardless of the GCS configuration for the thread.
54
55* The architecture supports enabling GCS without verifying that return values
56 in LR match those in the GCS, the LR will be ignored. This is not supported
57 by Linux.
58
59
60
612. Enabling and disabling Guarded Control Stacks
62-------------------------------------------------
63
64* GCS is enabled and disabled for a thread via the PR_SET_SHADOW_STACK_STATUS
65 prctl(), this takes a single flags argument specifying which GCS features
66 should be used.
67
68* When set PR_SHADOW_STACK_ENABLE flag allocates a Guarded Control Stack
69 and enables GCS for the thread, enabling the functionality controlled by
70 GCSCRE0_EL1.{nTR, RVCHKEN, PCRSEL}.
71
72* When set the PR_SHADOW_STACK_PUSH flag enables the functionality controlled
73 by GCSCRE0_EL1.PUSHMEn, allowing explicit GCS pushes.
74
75* When set the PR_SHADOW_STACK_WRITE flag enables the functionality controlled
76 by GCSCRE0_EL1.STREn, allowing explicit stores to the Guarded Control Stack.
77
78* Any unknown flags will cause PR_SET_SHADOW_STACK_STATUS to return -EINVAL.
79
80* PR_LOCK_SHADOW_STACK_STATUS is passed a bitmask of features with the same
81 values as used for PR_SET_SHADOW_STACK_STATUS. Any future changes to the
82 status of the specified GCS mode bits will be rejected.
83
84* PR_LOCK_SHADOW_STACK_STATUS allows any bit to be locked, this allows
85 userspace to prevent changes to any future features.
86
87* There is no support for a process to remove a lock that has been set for
88 it.
89
90* PR_SET_SHADOW_STACK_STATUS and PR_LOCK_SHADOW_STACK_STATUS affect only the
91 thread that called them, any other running threads will be unaffected.
92
93* New threads inherit the GCS configuration of the thread that created them.
94
95* GCS is disabled on exec().
96
97* The current GCS configuration for a thread may be read with the
98 PR_GET_SHADOW_STACK_STATUS prctl(), this returns the same flags that
99 are passed to PR_SET_SHADOW_STACK_STATUS.
100
101* If GCS is disabled for a thread after having previously been enabled then
102 the stack will remain allocated for the lifetime of the thread. At present
103 any attempt to reenable GCS for the thread will be rejected, this may be
104 revisited in future.
105
106* It should be noted that since enabling GCS will result in GCS becoming
107 active immediately it is not normally possible to return from the function
108 that invoked the prctl() that enabled GCS. It is expected that the normal
109 usage will be that GCS is enabled very early in execution of a program.
110
111
112
1133. Allocation of Guarded Control Stacks
114----------------------------------------
115
116* When GCS is enabled for a thread a new Guarded Control Stack will be
117 allocated for it of half the standard stack size or 2 gigabytes,
118 whichever is smaller.
119
120* When a new thread is created by a thread which has GCS enabled then a
121 new Guarded Control Stack will be allocated for the new thread with
122 half the size of the standard stack.
123
124* When a stack is allocated by enabling GCS or during thread creation then
125 the top 8 bytes of the stack will be initialised to 0 and GCSPR_EL0 will
126 be set to point to the address of this 0 value, this can be used to
127 detect the top of the stack.
128
129* Additional Guarded Control Stacks can be allocated using the
130 map_shadow_stack() system call.
131
132* Stacks allocated using map_shadow_stack() can optionally have an end of
133 stack marker and cap placed at the top of the stack. If the flag
134 SHADOW_STACK_SET_TOKEN is specified a cap will be placed on the stack,
135 if SHADOW_STACK_SET_MARKER is not specified the cap will be the top 8
136 bytes of the stack and if it is specified then the cap will be the next
137 8 bytes. While specifying just SHADOW_STACK_SET_MARKER by itself is
138 valid since the marker is all bits 0 it has no observable effect.
139
140* Stacks allocated using map_shadow_stack() must have a size which is a
141 multiple of 8 bytes larger than 8 bytes and must be 8 bytes aligned.
142
143* An address can be specified to map_shadow_stack(), if one is provided then
144 it must be aligned to a page boundary.
145
146* When a thread is freed the Guarded Control Stack initially allocated for
147 that thread will be freed. Note carefully that if the stack has been
148 switched this may not be the stack currently in use by the thread.
149
150
1514. Signal handling
152--------------------
153
154* A new signal frame record gcs_context encodes the current GCS mode and
155 pointer for the interrupted context on signal delivery. This will always
156 be present on systems that support GCS.
157
158* The record contains a flag field which reports the current GCS configuration
159 for the interrupted context as PR_GET_SHADOW_STACK_STATUS would.
160
161* The signal handler is run with the same GCS configuration as the interrupted
162 context.
163
164* When GCS is enabled for the interrupted thread a signal handling specific
165 GCS cap token will be written to the GCS, this is an architectural GCS cap
166 with the token type (bits 0..11) all clear. The GCSPR_EL0 reported in the
167 signal frame will point to this cap token.
168
169* The signal handler will use the same GCS as the interrupted context.
170
171* When GCS is enabled on signal entry a frame with the address of the signal
172 return handler will be pushed onto the GCS, allowing return from the signal
173 handler via RET as normal. This will not be reported in the gcs_context in
174 the signal frame.
175
176
1775. Signal return
178-----------------
179
180When returning from a signal handler:
181
182* If there is a gcs_context record in the signal frame then the GCS flags
183 and GCSPR_EL0 will be restored from that context prior to further
184 validation.
185
186* If there is no gcs_context record in the signal frame then the GCS
187 configuration will be unchanged.
188
189* If GCS is enabled on return from a signal handler then GCSPR_EL0 must
190 point to a valid GCS signal cap record, this will be popped from the
191 GCS prior to signal return.
192
193* If the GCS configuration is locked when returning from a signal then any
194 attempt to change the GCS configuration will be treated as an error. This
195 is true even if GCS was not enabled prior to signal entry.
196
197* GCS may be disabled via signal return but any attempt to enable GCS via
198 signal return will be rejected.
199
200
2016. ptrace extensions
202---------------------
203
204* A new regset NT_ARM_GCS is defined for use with PTRACE_GETREGSET and
205 PTRACE_SETREGSET.
206
207* The GCS mode, including enable and disable, may be configured via ptrace.
208 If GCS is enabled via ptrace no new GCS will be allocated for the thread.
209
210* Configuration via ptrace ignores locking of GCS mode bits.
211
212
2137. ELF coredump extensions
214---------------------------
215
216* NT_ARM_GCS notes will be added to each coredump for each thread of the
217 dumped process. The contents will be equivalent to the data that would
218 have been read if a PTRACE_GETREGSET of the corresponding type were
219 executed for each thread when the coredump was generated.
220
221
222
2238. /proc extensions
224--------------------
225
226* Guarded Control Stack pages will include "ss" in their VmFlags in
227 /proc/<pid>/smaps.