The main purpose of this patch is to protect dynamic stack allocations from stack-clash attacks. I think the only notable change since V3 was to use a target hook instead of a target macro to control a bit of special behavior we're going to want on aarch64.
Bootstrapped and regression tested on x86. Installing on the trunk. Jeff
commit feb64697b00b4a52346eaa3b7f349a0636936f7c Author: Jeff Law <l...@torsion.usersys.redhat.com> Date: Wed Jun 28 14:02:16 2017 -0400 2017-09-18 Jeff Law <l...@redhat.com> * explow.c: Include "params.h". (anti_adjust_stack_and_probe_stack_clash): New function. (get_stack_check_protect): Likewise. (compute_stack_clash_protection_loop_data): Likewise. (emit_stack_clash_protection_loop_start): Likewise. (emit_stack_clash_protection_loop_end): Likewise. (allocate_dynamic_stack_space): Use get_stack_check_protect. Use anti_adjust_stack_and_probe_stack_clash. * explow.h (compute_stack_clash_protection_loop_data): Prototype. (emit_stack_clash_protection_loop_start): Likewise. (emit_stack_clash_protection_loop_end): Likewise. * rtl.h (get_stack_check_protect): Prototype. * target.def (stack_clash_protection_final_dynamic_probe): New hook. * targhooks.c (default_stack_clash_protection_final_dynamic_probe): New. * targhooks.h (default_stack_clash_protection_final_dynamic_probe): Prototype. * doc/tm.texi.in (TARGET_STACK_CLASH_PROTECTION_FINAL_DYNAMIC_PROBE): Add @hook. * doc/tm.texi: Rebuilt. * config/aarch64/aarch64.c (aarch64_expand_prologue): Use get_stack_check_protect. * config/alpha/alpha.c (alpha_expand_prologue): Likewise. * config/arm/arm.c (arm_expand_prologue): Likewise. (arm_frame_pointer_required): Likewise. * config/i386/i386.c (ix86_expand_prologue): Likewise. * config/ia64/ia64.c (ia64_expand_prologue): Likewise. * config/mips/mips.c (mips_expand_prologue): Likewise. * config/powerpcspe/powerpcspe.c (rs6000_emit_prologue): Likewise. * config/rs6000/rs6000.c (rs6000_emit_prologue): Likewise. * config/sparc/sparc.c (sparc_expand_prologue): Likewise. (sparc_flat_expand_prologue): Likewise. * gcc.dg/stack-check-3.c: New test. diff --git a/gcc/ChangeLog b/gcc/ChangeLog index a1aa3f6327d..b8012611405 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,5 +1,37 @@ 2017-09-19 Jeff Law <l...@redhat.com> + * explow.c: Include "params.h". + (anti_adjust_stack_and_probe_stack_clash): New function. + (get_stack_check_protect): Likewise. + (compute_stack_clash_protection_loop_data): Likewise. + (emit_stack_clash_protection_loop_start): Likewise. + (emit_stack_clash_protection_loop_end): Likewise. + (allocate_dynamic_stack_space): Use get_stack_check_protect. + Use anti_adjust_stack_and_probe_stack_clash. + * explow.h (compute_stack_clash_protection_loop_data): Prototype. + (emit_stack_clash_protection_loop_start): Likewise. + (emit_stack_clash_protection_loop_end): Likewise. + * rtl.h (get_stack_check_protect): Prototype. + * target.def (stack_clash_protection_final_dynamic_probe): New hook. + * targhooks.c (default_stack_clash_protection_final_dynamic_probe): New. + * targhooks.h (default_stack_clash_protection_final_dynamic_probe): + Prototype. + * doc/tm.texi.in (TARGET_STACK_CLASH_PROTECTION_FINAL_DYNAMIC_PROBE): + Add @hook. + * doc/tm.texi: Rebuilt. + * config/aarch64/aarch64.c (aarch64_expand_prologue): Use + get_stack_check_protect. + * config/alpha/alpha.c (alpha_expand_prologue): Likewise. + * config/arm/arm.c (arm_expand_prologue): Likewise. + (arm_frame_pointer_required): Likewise. + * config/i386/i386.c (ix86_expand_prologue): Likewise. + * config/ia64/ia64.c (ia64_expand_prologue): Likewise. + * config/mips/mips.c (mips_expand_prologue): Likewise. + * config/powerpcspe/powerpcspe.c (rs6000_emit_prologue): Likewise. + * config/rs6000/rs6000.c (rs6000_emit_prologue): Likewise. + * config/sparc/sparc.c (sparc_expand_prologue): Likewise. + (sparc_flat_expand_prologue): Likewise. + * common.opt (-fstack-clash-protection): New option. * flag-types.h (enum stack_check_type): Note difference between -fstack-check= and -fstack-clash-protection. diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c index 1c1400871d4..5e26cb70da0 100644 --- a/gcc/config/aarch64/aarch64.c +++ b/gcc/config/aarch64/aarch64.c @@ -3666,12 +3666,14 @@ aarch64_expand_prologue (void) { if (crtl->is_leaf && !cfun->calls_alloca) { - if (frame_size > PROBE_INTERVAL && frame_size > STACK_CHECK_PROTECT) - aarch64_emit_probe_stack_range (STACK_CHECK_PROTECT, - frame_size - STACK_CHECK_PROTECT); + if (frame_size > PROBE_INTERVAL + && frame_size > get_stack_check_protect ()) + aarch64_emit_probe_stack_range (get_stack_check_protect (), + (frame_size + - get_stack_check_protect ())); } else if (frame_size > 0) - aarch64_emit_probe_stack_range (STACK_CHECK_PROTECT, frame_size); + aarch64_emit_probe_stack_range (get_stack_check_protect (), frame_size); } aarch64_sub_sp (IP0_REGNUM, initial_adjust, true); diff --git a/gcc/config/alpha/alpha.c b/gcc/config/alpha/alpha.c index e0f458c9109..a4e8b2b6c30 100644 --- a/gcc/config/alpha/alpha.c +++ b/gcc/config/alpha/alpha.c @@ -7761,7 +7761,7 @@ alpha_expand_prologue (void) probed_size = frame_size; if (flag_stack_check) - probed_size += STACK_CHECK_PROTECT; + probed_size += get_stack_check_protect (); if (probed_size <= 32768) { diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index bc802adc3e3..679e838b0aa 100644 --- a/gcc/config/arm/arm.c +++ b/gcc/config/arm/arm.c @@ -21693,13 +21693,13 @@ arm_expand_prologue (void) if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - arm_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT, + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) + arm_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect (), regno, live_regs_mask); } else if (size > 0) - arm_emit_probe_stack_range (STACK_CHECK_PROTECT, size, + arm_emit_probe_stack_range (get_stack_check_protect (), size, regno, live_regs_mask); } @@ -27886,7 +27886,7 @@ arm_frame_pointer_required (void) { /* We don't have the final size of the frame so adjust. */ size += 32 * UNITS_PER_WORD; - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) return true; } else diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c index 1c765fbc74c..05b0520759c 100644 --- a/gcc/config/i386/i386.c +++ b/gcc/config/i386/i386.c @@ -14871,7 +14871,7 @@ ix86_expand_prologue (void) HOST_WIDE_INT size = allocate; if (TARGET_64BIT && size >= HOST_WIDE_INT_C (0x80000000)) - size = 0x80000000 - STACK_CHECK_PROTECT - 1; + size = 0x80000000 - get_stack_check_protect () - 1; if (TARGET_STACK_PROBE) { @@ -14881,18 +14881,20 @@ ix86_expand_prologue (void) ix86_emit_probe_stack_range (0, size); } else - ix86_emit_probe_stack_range (0, size + STACK_CHECK_PROTECT); + ix86_emit_probe_stack_range (0, + size + get_stack_check_protect ()); } else { if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - ix86_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT); + if (size > PROBE_INTERVAL + && size > get_stack_check_protect ()) + ix86_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect ()); } else - ix86_emit_probe_stack_range (STACK_CHECK_PROTECT, size); + ix86_emit_probe_stack_range (get_stack_check_protect (), size); } } } diff --git a/gcc/config/ia64/ia64.c b/gcc/config/ia64/ia64.c index b4d63592ea0..00ef2159323 100644 --- a/gcc/config/ia64/ia64.c +++ b/gcc/config/ia64/ia64.c @@ -3502,15 +3502,16 @@ ia64_expand_prologue (void) if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - ia64_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT, + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) + ia64_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect (), bs_size); - else if (size + bs_size > STACK_CHECK_PROTECT) - ia64_emit_probe_stack_range (STACK_CHECK_PROTECT, 0, bs_size); + else if (size + bs_size > get_stack_check_protect ()) + ia64_emit_probe_stack_range (get_stack_check_protect (), + 0, bs_size); } else if (size + bs_size > 0) - ia64_emit_probe_stack_range (STACK_CHECK_PROTECT, size, bs_size); + ia64_emit_probe_stack_range (get_stack_check_protect (), size, bs_size); } if (dump_file) diff --git a/gcc/config/mips/mips.c b/gcc/config/mips/mips.c index 7eaff14d472..67cee0bb051 100644 --- a/gcc/config/mips/mips.c +++ b/gcc/config/mips/mips.c @@ -12084,12 +12084,12 @@ mips_expand_prologue (void) { if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - mips_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT); + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) + mips_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect ()); } else if (size > 0) - mips_emit_probe_stack_range (STACK_CHECK_PROTECT, size); + mips_emit_probe_stack_range (get_stack_check_protect (), size); } /* Save the registers. Allocate up to MIPS_MAX_FIRST_STACK_STEP diff --git a/gcc/config/powerpcspe/powerpcspe.c b/gcc/config/powerpcspe/powerpcspe.c index 11664eed4e0..a956729f706 100644 --- a/gcc/config/powerpcspe/powerpcspe.c +++ b/gcc/config/powerpcspe/powerpcspe.c @@ -29693,12 +29693,12 @@ rs6000_emit_prologue (void) if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - rs6000_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT); + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) + rs6000_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect ()); } else if (size > 0) - rs6000_emit_probe_stack_range (STACK_CHECK_PROTECT, size); + rs6000_emit_probe_stack_range (get_stack_check_protect (), size); } if (TARGET_FIX_AND_CONTINUE) diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c index 1978634f02e..e5ef63889b7 100644 --- a/gcc/config/rs6000/rs6000.c +++ b/gcc/config/rs6000/rs6000.c @@ -26765,12 +26765,12 @@ rs6000_emit_prologue (void) if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - rs6000_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT); + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) + rs6000_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect ()); } else if (size > 0) - rs6000_emit_probe_stack_range (STACK_CHECK_PROTECT, size); + rs6000_emit_probe_stack_range (get_stack_check_protect (), size); } if (TARGET_FIX_AND_CONTINUE) diff --git a/gcc/config/sparc/sparc.c b/gcc/config/sparc/sparc.c index 749a7f866cb..906bd75b560 100644 --- a/gcc/config/sparc/sparc.c +++ b/gcc/config/sparc/sparc.c @@ -5738,12 +5738,12 @@ sparc_expand_prologue (void) { if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - sparc_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT); + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) + sparc_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect ()); } else if (size > 0) - sparc_emit_probe_stack_range (STACK_CHECK_PROTECT, size); + sparc_emit_probe_stack_range (get_stack_check_protect (), size); } if (size == 0) @@ -5849,12 +5849,12 @@ sparc_flat_expand_prologue (void) { if (crtl->is_leaf && !cfun->calls_alloca) { - if (size > PROBE_INTERVAL && size > STACK_CHECK_PROTECT) - sparc_emit_probe_stack_range (STACK_CHECK_PROTECT, - size - STACK_CHECK_PROTECT); + if (size > PROBE_INTERVAL && size > get_stack_check_protect ()) + sparc_emit_probe_stack_range (get_stack_check_protect (), + size - get_stack_check_protect ()); } else if (size > 0) - sparc_emit_probe_stack_range (STACK_CHECK_PROTECT, size); + sparc_emit_probe_stack_range (get_stack_check_protect (), size); } if (sparc_save_local_in_regs_p) diff --git a/gcc/doc/tm.texi b/gcc/doc/tm.texi index ae65e4f5edf..07ae66a2e61 100644 --- a/gcc/doc/tm.texi +++ b/gcc/doc/tm.texi @@ -3411,6 +3411,10 @@ GCC computed the default from the values of the above macros and you will normally not need to override that default. @end defmac +@deftypefn {Target Hook} bool TARGET_STACK_CLASH_PROTECTION_FINAL_DYNAMIC_PROBE (rtx @var{residual}) +Some targets make optimistic assumptions about the state of stack probing when they emit their prologues. On such targets a probe into the end of any dynamically allocated space is likely required for safety against stack clash style attacks. Define this variable to return nonzero if such a probe is required or zero otherwise. You need not define this macro if it would always have the value zero. +@end deftypefn + @need 2000 @node Frame Registers @subsection Registers That Address the Stack Frame diff --git a/gcc/doc/tm.texi.in b/gcc/doc/tm.texi.in index 733466dbb84..6a794371342 100644 --- a/gcc/doc/tm.texi.in +++ b/gcc/doc/tm.texi.in @@ -2847,6 +2847,8 @@ GCC computed the default from the values of the above macros and you will normally not need to override that default. @end defmac +@hook TARGET_STACK_CLASH_PROTECTION_FINAL_DYNAMIC_PROBE + @need 2000 @node Frame Registers @subsection Registers That Address the Stack Frame diff --git a/gcc/explow.c b/gcc/explow.c index 638dc5f8f0f..0f30507d1f5 100644 --- a/gcc/explow.c +++ b/gcc/explow.c @@ -40,8 +40,10 @@ along with GCC; see the file COPYING3. If not see #include "expr.h" #include "common/common-target.h" #include "output.h" +#include "params.h" static rtx break_out_memory_refs (rtx); +static void anti_adjust_stack_and_probe_stack_clash (rtx); /* Truncate and perhaps sign-extend C as appropriate for MODE. */ @@ -1283,6 +1285,29 @@ get_dynamic_stack_size (rtx *psize, unsigned size_align, *psize = size; } +/* Return the number of bytes to "protect" on the stack for -fstack-check. + + "protect" in the context of -fstack-check means how many bytes we + should always ensure are available on the stack. More importantly + this is how many bytes are skipped when probing the stack. + + On some targets we want to reuse the -fstack-check prologue support + to give a degree of protection against stack clashing style attacks. + + In that scenario we do not want to skip bytes before probing as that + would render the stack clash protections useless. + + So we never use STACK_CHECK_PROTECT directly. Instead we indirect though + this helper which allows us to provide different values for + -fstack-check and -fstack-clash-protection. */ +HOST_WIDE_INT +get_stack_check_protect (void) +{ + if (flag_stack_clash_protection) + return 0; + return STACK_CHECK_PROTECT; +} + /* Return an rtx representing the address of an area of memory dynamically pushed on the stack. @@ -1441,7 +1466,7 @@ allocate_dynamic_stack_space (rtx size, unsigned size_align, probe_stack_range (STACK_OLD_CHECK_PROTECT + STACK_CHECK_MAX_FRAME_SIZE, size); else if (flag_stack_check == STATIC_BUILTIN_STACK_CHECK) - probe_stack_range (STACK_CHECK_PROTECT, size); + probe_stack_range (get_stack_check_protect (), size); /* Don't let anti_adjust_stack emit notes. */ suppress_reg_args_size = true; @@ -1494,6 +1519,8 @@ allocate_dynamic_stack_space (rtx size, unsigned size_align, if (flag_stack_check && STACK_CHECK_MOVING_SP) anti_adjust_stack_and_probe (size, false); + else if (flag_stack_clash_protection) + anti_adjust_stack_and_probe_stack_clash (size); else anti_adjust_stack (size); @@ -1769,6 +1796,219 @@ probe_stack_range (HOST_WIDE_INT first, rtx size) emit_insn (gen_blockage ()); } +/* Compute parameters for stack clash probing a dynamic stack + allocation of SIZE bytes. + + We compute ROUNDED_SIZE, LAST_ADDR, RESIDUAL and PROBE_INTERVAL. + + Additionally we conditionally dump the type of probing that will + be needed given the values computed. */ + +void +compute_stack_clash_protection_loop_data (rtx *rounded_size, rtx *last_addr, + rtx *residual, + HOST_WIDE_INT *probe_interval, + rtx size) +{ + /* Round SIZE down to STACK_CLASH_PROTECTION_PROBE_INTERVAL */ + *probe_interval + = 1 << PARAM_VALUE (PARAM_STACK_CLASH_PROTECTION_PROBE_INTERVAL); + *rounded_size = simplify_gen_binary (AND, Pmode, size, + GEN_INT (-*probe_interval)); + + /* Compute the value of the stack pointer for the last iteration. + It's just SP + ROUNDED_SIZE. */ + rtx rounded_size_op = force_operand (*rounded_size, NULL_RTX); + *last_addr = force_operand (gen_rtx_fmt_ee (STACK_GROW_OP, Pmode, + stack_pointer_rtx, + rounded_size_op), + NULL_RTX); + + /* Compute any residuals not allocated by the loop above. Residuals + are just the ROUNDED_SIZE - SIZE. */ + *residual = simplify_gen_binary (MINUS, Pmode, size, *rounded_size); + + /* Dump key information to make writing tests easy. */ + if (dump_file) + { + if (*rounded_size == CONST0_RTX (Pmode)) + fprintf (dump_file, + "Stack clash skipped dynamic allocation and probing loop.\n"); + else if (GET_CODE (*rounded_size) == CONST_INT + && INTVAL (*rounded_size) <= 4 * *probe_interval) + fprintf (dump_file, + "Stack clash dynamic allocation and probing inline.\n"); + else if (GET_CODE (*rounded_size) == CONST_INT) + fprintf (dump_file, + "Stack clash dynamic allocation and probing in " + "rotated loop.\n"); + else + fprintf (dump_file, + "Stack clash dynamic allocation and probing in loop.\n"); + + if (*residual != CONST0_RTX (Pmode)) + fprintf (dump_file, + "Stack clash dynamic allocation and probing residuals.\n"); + else + fprintf (dump_file, + "Stack clash skipped dynamic allocation and " + "probing residuals.\n"); + } +} + +/* Emit the start of an allocate/probe loop for stack + clash protection. + + LOOP_LAB and END_LAB are returned for use when we emit the + end of the loop. + + LAST addr is the value for SP which stops the loop. */ +void +emit_stack_clash_protection_probe_loop_start (rtx *loop_lab, + rtx *end_lab, + rtx last_addr, + bool rotated) +{ + /* Essentially we want to emit any setup code, the top of loop + label and the comparison at the top of the loop. */ + *loop_lab = gen_label_rtx (); + *end_lab = gen_label_rtx (); + + emit_label (*loop_lab); + if (!rotated) + emit_cmp_and_jump_insns (stack_pointer_rtx, last_addr, EQ, NULL_RTX, + Pmode, 1, *end_lab); +} + +/* Emit the end of a stack clash probing loop. + + This consists of just the jump back to LOOP_LAB and + emitting END_LOOP after the loop. */ + +void +emit_stack_clash_protection_probe_loop_end (rtx loop_lab, rtx end_loop, + rtx last_addr, bool rotated) +{ + if (rotated) + emit_cmp_and_jump_insns (stack_pointer_rtx, last_addr, NE, NULL_RTX, + Pmode, 1, loop_lab); + else + emit_jump (loop_lab); + + emit_label (end_loop); + +} + +/* Adjust the stack pointer by minus SIZE (an rtx for a number of bytes) + while probing it. This pushes when SIZE is positive. SIZE need not + be constant. + + This is subtly different than anti_adjust_stack_and_probe to try and + prevent stack-clash attacks + + 1. It must assume no knowledge of the probing state, any allocation + must probe. + + Consider the case of a 1 byte alloca in a loop. If the sum of the + allocations is large, then this could be used to jump the guard if + probes were not emitted. + + 2. It never skips probes, whereas anti_adjust_stack_and_probe will + skip probes on the first couple PROBE_INTERVALs on the assumption + they're done elsewhere. + + 3. It only allocates and probes SIZE bytes, it does not need to + allocate/probe beyond that because this probing style does not + guarantee signal handling capability if the guard is hit. */ + +static void +anti_adjust_stack_and_probe_stack_clash (rtx size) +{ + /* First ensure SIZE is Pmode. */ + if (GET_MODE (size) != VOIDmode && GET_MODE (size) != Pmode) + size = convert_to_mode (Pmode, size, 1); + + /* We can get here with a constant size on some targets. */ + rtx rounded_size, last_addr, residual; + HOST_WIDE_INT probe_interval; + compute_stack_clash_protection_loop_data (&rounded_size, &last_addr, + &residual, &probe_interval, size); + + if (rounded_size != CONST0_RTX (Pmode)) + { + if (INTVAL (rounded_size) <= 4 * probe_interval) + { + for (HOST_WIDE_INT i = 0; + i < INTVAL (rounded_size); + i += probe_interval) + { + anti_adjust_stack (GEN_INT (probe_interval)); + + /* The prologue does not probe residuals. Thus the offset + here to probe just beyond what the prologue had already + allocated. */ + emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, + (probe_interval + - GET_MODE_SIZE (word_mode)))); + emit_insn (gen_blockage ()); + } + } + else + { + rtx loop_lab, end_loop; + bool rotate_loop = GET_CODE (rounded_size) == CONST_INT; + emit_stack_clash_protection_probe_loop_start (&loop_lab, &end_loop, + last_addr, rotate_loop); + + anti_adjust_stack (GEN_INT (probe_interval)); + + /* The prologue does not probe residuals. Thus the offset here + to probe just beyond what the prologue had already allocated. */ + emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, + (probe_interval + - GET_MODE_SIZE (word_mode)))); + + emit_stack_clash_protection_probe_loop_end (loop_lab, end_loop, + last_addr, rotate_loop); + emit_insn (gen_blockage ()); + } + } + + if (residual != CONST0_RTX (Pmode)) + { + rtx x = force_reg (Pmode, plus_constant (Pmode, residual, + -GET_MODE_SIZE (word_mode))); + anti_adjust_stack (residual); + emit_stack_probe (gen_rtx_PLUS (Pmode, stack_pointer_rtx, x)); + emit_insn (gen_blockage ()); + } + + /* Some targets make optimistic assumptions in their prologues about + how the caller may have probed the stack. Make sure we honor + those assumptions when needed. */ + if (size != CONST0_RTX (Pmode) + && targetm.stack_clash_protection_final_dynamic_probe (residual)) + { + /* Ideally we would just probe at *sp. However, if SIZE is not + a compile-time constant, but is zero at runtime, then *sp + might hold live data. So probe at *sp if we know that + an allocation was made, otherwise probe into the red zone + which is obviously undesirable. */ + if (GET_CODE (size) == CONST_INT) + { + emit_stack_probe (stack_pointer_rtx); + emit_insn (gen_blockage ()); + } + else + { + emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, + -GET_MODE_SIZE (word_mode))); + emit_insn (gen_blockage ()); + } + } +} + + /* Adjust the stack pointer by minus SIZE (an rtx for a number of bytes) while probing it. This pushes when SIZE is positive. SIZE need not be constant. If ADJUST_BACK is true, adjust back the stack pointer diff --git a/gcc/explow.h b/gcc/explow.h index 217a3226adb..b85c051e8ce 100644 --- a/gcc/explow.h +++ b/gcc/explow.h @@ -69,6 +69,15 @@ extern void anti_adjust_stack (rtx); /* Add some bytes to the stack while probing it. An rtx says how many. */ extern void anti_adjust_stack_and_probe (rtx, bool); +/* Support for building allocation/probing loops for stack-clash + protection of dyamically allocated stack space. */ +extern void compute_stack_clash_protection_loop_data (rtx *, rtx *, rtx *, + HOST_WIDE_INT *, rtx); +extern void emit_stack_clash_protection_probe_loop_start (rtx *, rtx *, + rtx, bool); +extern void emit_stack_clash_protection_probe_loop_end (rtx, rtx, + rtx, bool); + /* This enum is used for the following two functions. */ enum save_level {SAVE_BLOCK, SAVE_FUNCTION, SAVE_NONLOCAL}; diff --git a/gcc/rtl.h b/gcc/rtl.h index a450a04c237..a63f33e747a 100644 --- a/gcc/rtl.h +++ b/gcc/rtl.h @@ -2722,6 +2722,7 @@ get_full_set_src_cost (rtx x, machine_mode mode, struct full_rtx_costs *c) /* In explow.c */ extern HOST_WIDE_INT trunc_int_for_mode (HOST_WIDE_INT, machine_mode); extern rtx plus_constant (machine_mode, rtx, HOST_WIDE_INT, bool = false); +extern HOST_WIDE_INT get_stack_check_protect (void); /* In rtl.c */ extern rtx rtx_alloc (RTX_CODE CXX_MEM_STAT_INFO); diff --git a/gcc/target.def b/gcc/target.def index f4c35763686..ae22d7a6183 100644 --- a/gcc/target.def +++ b/gcc/target.def @@ -5735,6 +5735,13 @@ these registers when the target switches are opposed to them.)", void, (void), hook_void_void) +DEFHOOK +(stack_clash_protection_final_dynamic_probe, + "Some targets make optimistic assumptions about the state of stack probing when they emit their prologues. On such targets a probe into the end of any dynamically allocated space is likely required for safety against stack clash style attacks. Define this variable to return nonzero if such a probe is required or zero otherwise. You need not define this macro if it would always have the value zero.", + bool, (rtx residual), + default_stack_clash_protection_final_dynamic_probe) + + /* Functions specific to the C family of frontends. */ #undef HOOK_PREFIX #define HOOK_PREFIX "TARGET_C_" diff --git a/gcc/targhooks.c b/gcc/targhooks.c index 4bd4833dfdd..d87d6c7f430 100644 --- a/gcc/targhooks.c +++ b/gcc/targhooks.c @@ -2207,4 +2207,10 @@ default_excess_precision (enum excess_precision_type ATTRIBUTE_UNUSED) return FLT_EVAL_METHOD_PROMOTE_TO_FLOAT; } +HOST_WIDE_INT +default_stack_clash_protection_final_dynamic_probe (rtx residual ATTRIBUTE_UNUSED) +{ + return 0; +} + #include "gt-targhooks.h" diff --git a/gcc/targhooks.h b/gcc/targhooks.h index 6d78508c8a2..a70992d034e 100644 --- a/gcc/targhooks.h +++ b/gcc/targhooks.h @@ -271,5 +271,6 @@ extern unsigned int default_min_arithmetic_precision (void); extern enum flt_eval_method default_excess_precision (enum excess_precision_type ATTRIBUTE_UNUSED); +extern bool default_stack_clash_protection_final_dynamic_probe (rtx); #endif /* GCC_TARGHOOKS_H */ diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index a77b54604e6..221956496ae 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,5 +1,7 @@ 2017-09-19 Jeff Law <l...@redhat.com> + * gcc.dg/stack-check-3.c: New test. + * gcc.dg/stack-check-2.c: New test. * lib/target-supports.exp (check_effective_target_supports_stack_clash_protection): New function. diff --git a/gcc/testsuite/gcc.dg/stack-check-3.c b/gcc/testsuite/gcc.dg/stack-check-3.c new file mode 100644 index 00000000000..58fb65649ee --- /dev/null +++ b/gcc/testsuite/gcc.dg/stack-check-3.c @@ -0,0 +1,86 @@ +/* The goal here is to ensure that dynamic allocations via vlas or + alloca calls receive probing. + + Scanning the RTL or assembly code seems like insanity here as does + checking for particular allocation sizes and probe offsets. For + now we just verify that there's an allocation + probe loop and + residual allocation + probe for f?. */ + +/* { dg-do compile } */ +/* { dg-options "-O2 -fstack-clash-protection -fdump-rtl-expand -fno-optimize-sibling-calls --param stack-clash-protection-probe-interval=4096 --param stack-clash-protection-guard-size=4096" } */ +/* { dg-require-effective-target supports_stack_clash_protection } */ + +__attribute__((noinline, noclone)) void +foo (char *p) +{ + asm volatile ("" : : "r" (p) : "memory"); +} + +/* Simple VLA, no other locals. */ +__attribute__((noinline, noclone)) void +f0 (int x) +{ + char vla[x]; + foo (vla); +} + +/* Simple VLA, small local frame. */ +__attribute__((noinline, noclone)) void +f1 (int x) +{ + char locals[128]; + char vla[x]; + foo (vla); +} + +/* Small constant alloca, no other locals. */ +__attribute__((noinline, noclone)) void +f2 (int x) +{ + char *vla = __builtin_alloca (128); + foo (vla); +} + +/* Big constant alloca, small local frame. */ +__attribute__((noinline, noclone)) void +f3 (int x) +{ + char locals[128]; + char *vla = __builtin_alloca (16384); + foo (vla); +} + +/* Big constant alloca, small local frame. */ +__attribute__((noinline, noclone)) void +f3a (int x) +{ + char locals[128]; + char *vla = __builtin_alloca (32768); + foo (vla); +} + +/* Nonconstant alloca, no other locals. */ +__attribute__((noinline, noclone)) void +f4 (int x) +{ + char *vla = __builtin_alloca (x); + foo (vla); +} + +/* Nonconstant alloca, small local frame. */ +__attribute__((noinline, noclone)) void +f5 (int x) +{ + char locals[128]; + char *vla = __builtin_alloca (x); + foo (vla); +} + +/* { dg-final { scan-rtl-dump-times "allocation and probing residuals" 7 "expand" } } */ + + +/* { dg-final { scan-rtl-dump-times "allocation and probing in loop" 7 "expand" { target callee_realigns_stack } } } */ +/* { dg-final { scan-rtl-dump-times "allocation and probing in loop" 4 "expand" { target { ! callee_realigns_stack } } } } */ +/* { dg-final { scan-rtl-dump-times "allocation and probing in rotated loop" 1 "expand" { target { ! callee_realigns_stack } } } } */ +/* { dg-final { scan-rtl-dump-times "allocation and probing inline" 1 "expand" { target { ! callee_realigns_stack } } } } */ +/* { dg-final { scan-rtl-dump-times "skipped dynamic allocation and probing loop" 1 "expand" { target { ! callee_realigns_stack } } } } */ commit 74af142db7e145d887fc63f9beb342027095e285 Author: law <law@138bc75d-0d04-0410-961f-82ee72b054a4> Date: Wed Sep 20 04:56:54 2017 +0000 * common.opt (-fstack-clash-protection): New option. * flag-types.h (enum stack_check_type): Note difference between -fstack-check= and -fstack-clash-protection. * params.def (PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE): New PARAM. (PARAM_STACK_CLASH_PROTECTION_PROBE_INTERVAL): Likewise. * toplev.c (process_options): Issue warnings/errors for cases not handled with -fstack-clash-protection. * doc/invoke.texi (-fstack-clash-protection): Document new option. (-fstack-check): Note additional problem with -fstack-check=generic. Note that -fstack-check is primarily for Ada and refer users to -fstack-clash-protection for stack-clash-protection. Document new params for stack clash protection. * gcc.dg/stack-check-2.c: New test. * lib/target-supports.exp (check_effective_target_supports_stack_clash_protection): New function. (check_effective_target_frame_pointer_for_non_leaf): Likewise. (check_effective_target_caller_implicit_probes): Likewise. git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@252994 138bc75d-0d04-0410-961f-82ee72b054a4 diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 9576c8c4f47..a1aa3f6327d 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,18 @@ +2017-09-19 Jeff Law <l...@redhat.com> + + * common.opt (-fstack-clash-protection): New option. + * flag-types.h (enum stack_check_type): Note difference between + -fstack-check= and -fstack-clash-protection. + * params.def (PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE): New PARAM. + (PARAM_STACK_CLASH_PROTECTION_PROBE_INTERVAL): Likewise. + * toplev.c (process_options): Issue warnings/errors for cases + not handled with -fstack-clash-protection. + * doc/invoke.texi (-fstack-clash-protection): Document new option. + (-fstack-check): Note additional problem with -fstack-check=generic. + Note that -fstack-check is primarily for Ada and refer users + to -fstack-clash-protection for stack-clash-protection. + Document new params for stack clash protection. + 2017-09-19 Uros Bizjak <ubiz...@gmail.com> * config/i386/i386.md (*scc_bt<mode>): New insn_and_split pattern. diff --git a/gcc/common.opt b/gcc/common.opt index fa6dd845d54..f22661cfbba 100644 --- a/gcc/common.opt +++ b/gcc/common.opt @@ -2320,13 +2320,18 @@ Common Report Var(flag_variable_expansion_in_unroller) Optimization Apply variable expansion when loops are unrolled. fstack-check= -Common Report RejectNegative Joined +Common Report RejectNegative Joined Optimization -fstack-check=[no|generic|specific] Insert stack checking code into the program. fstack-check Common Alias(fstack-check=, specific, no) Insert stack checking code into the program. Same as -fstack-check=specific. +fstack-clash-protection +Common Report Var(flag_stack_clash_protection) Optimization +Insert code to probe each page of stack space as it is allocated to protect +from stack-clash style attacks. + fstack-limit Common Var(common_deferred_options) Defer diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi index 204c9b77b61..f0f9559b024 100644 --- a/gcc/doc/invoke.texi +++ b/gcc/doc/invoke.texi @@ -10187,6 +10187,21 @@ compilation without. The value for compilation with profile feedback needs to be more conservative (higher) in order to make tracer effective. +@item stack-clash-protection-guard-size +Specify the size of the operating system provided stack guard as +2 raised to @var{num} bytes. The default value is 12 (4096 bytes). +Acceptable values are between 12 and 30. Higher values may reduce the +number of explicit probes, but a value larger than the operating system +provided guard will leave code vulnerable to stack clash style attacks. + +@item stack-clash-protection-probe-interval +Stack clash protection involves probing stack space as it is allocated. This +param controls the maximum distance between probes into the stack as 2 raised +to @var{num} bytes. Acceptable values are between 10 and 16 and defaults to +12. Higher values may reduce the number of explicit probes, but a value +larger than the operating system provided guard will leave code vulnerable to +stack clash style attacks. + @item max-cse-path-length The maximum number of basic blocks on path that CSE considers. @@ -11412,7 +11427,8 @@ target support in the compiler but comes with the following drawbacks: @enumerate @item Modified allocation strategy for large objects: they are always -allocated dynamically if their size exceeds a fixed threshold. +allocated dynamically if their size exceeds a fixed threshold. Note this +may change the semantics of some code. @item Fixed limit on the size of the static frame of functions: when it is @@ -11427,6 +11443,25 @@ generic implementation, code performance is hampered. Note that old-style stack checking is also the fallback method for @samp{specific} if no target support has been added in the compiler. +@samp{-fstack-check=} is designed for Ada's needs to detect infinite recursion +and stack overflows. @samp{specific} is an excellent choice when compiling +Ada code. It is not generally sufficient to protect against stack-clash +attacks. To protect against those you want @samp{-fstack-clash-protection}. + +@item -fstack-clash-protection +@opindex fstack-clash-protection +Generate code to prevent stack clash style attacks. When this option is +enabled, the compiler will only allocate one page of stack space at a time +and each page is accessed immediately after allocation. Thus, it prevents +allocations from jumping over any stack guard page provided by the +operating system. + +Most targets do not fully support stack clash protection. However, on +those targets @option{-fstack-clash-protection} will protect dynamic stack +allocations. @option{-fstack-clash-protection} may also provide limited +protection for static stack allocations if the target supports +@option{-fstack-check=specific}. + @item -fstack-limit-register=@var{reg} @itemx -fstack-limit-symbol=@var{sym} @itemx -fno-stack-limit diff --git a/gcc/flag-types.h b/gcc/flag-types.h index 4938f69b1b6..1f439d35b07 100644 --- a/gcc/flag-types.h +++ b/gcc/flag-types.h @@ -166,7 +166,14 @@ enum permitted_flt_eval_methods PERMITTED_FLT_EVAL_METHODS_C11 }; -/* Type of stack check. */ +/* Type of stack check. + + Stack checking is designed to detect infinite recursion and stack + overflows for Ada programs. Furthermore stack checking tries to ensure + in that scenario that enough stack space is left to run a signal handler. + + -fstack-check= does not prevent stack-clash style attacks. For that + you want -fstack-clash-protection. */ enum stack_check_type { /* Do not check the stack. */ diff --git a/gcc/params.def b/gcc/params.def index 805302bb93e..860e79e3209 100644 --- a/gcc/params.def +++ b/gcc/params.def @@ -213,6 +213,16 @@ DEFPARAM(PARAM_STACK_FRAME_GROWTH, "Maximal stack frame growth due to inlining (in percent).", 1000, 0, 0) +DEFPARAM(PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE, + "stack-clash-protection-guard-size", + "Size of the stack guard expressed as a power of two.", + 12, 12, 30) + +DEFPARAM(PARAM_STACK_CLASH_PROTECTION_PROBE_INTERVAL, + "stack-clash-protection-probe-interval", + "Interval in which to probe the stack expressed as a power of two.", + 12, 10, 16) + /* The GCSE optimization will be disabled if it would require significantly more memory than this value. */ DEFPARAM(PARAM_MAX_GCSE_MEMORY, diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index 011420575b2..a77b54604e6 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,11 @@ +2017-09-19 Jeff Law <l...@redhat.com> + + * gcc.dg/stack-check-2.c: New test. + * lib/target-supports.exp + (check_effective_target_supports_stack_clash_protection): New function. + (check_effective_target_frame_pointer_for_non_leaf): Likewise. + (check_effective_target_caller_implicit_probes): Likewise. + 2017-09-19 Uros Bizjak <ubiz...@gmail.com> * gcc.target/i386/bt-5.c: New test. diff --git a/gcc/testsuite/gcc.dg/stack-check-2.c b/gcc/testsuite/gcc.dg/stack-check-2.c new file mode 100644 index 00000000000..196c4bbfbdd --- /dev/null +++ b/gcc/testsuite/gcc.dg/stack-check-2.c @@ -0,0 +1,66 @@ +/* The goal here is to ensure that we never consider a call to a noreturn + function as a potential tail call. + + Right now GCC discovers potential tail calls by looking at the + predecessors of the exit block. A call to a non-return function + has no successors and thus can never match that first filter. + + But that could change one day and we want to catch it. The problem + is the compiler could potentially optimize a tail call to a nonreturn + function, even if the caller has a frame. That breaks the assumption + that calls probe *sp when saving the return address that some targets + depend on to elide stack probes. */ + +/* { dg-do compile } */ +/* { dg-options "-O2 -fstack-clash-protection -fdump-tree-tailc -fdump-tree-optimized" } */ +/* { dg-require-effective-target supports_stack_clash_protection } */ + +extern void foo (void) __attribute__ ((__noreturn__)); + + +void +test_direct_1 (void) +{ + foo (); +} + +void +test_direct_2 (void) +{ + return foo (); +} + +void (*indirect)(void)__attribute__ ((noreturn)); + + +void +test_indirect_1 () +{ + (*indirect)(); +} + +void +test_indirect_2 (void) +{ + return (*indirect)();; +} + + +typedef void (*pvfn)() __attribute__ ((noreturn)); + +void (*indirect_casted)(void); + +void +test_indirect_casted_1 () +{ + (*(pvfn)indirect_casted)(); +} + +void +test_indirect_casted_2 (void) +{ + return (*(pvfn)indirect_casted)(); +} +/* { dg-final { scan-tree-dump-not "tail call" "tailc" } } */ +/* { dg-final { scan-tree-dump-not "tail call" "optimized" } } */ + diff --git a/gcc/testsuite/lib/target-supports.exp b/gcc/testsuite/lib/target-supports.exp index 6ea71222c6e..feed524680d 100644 --- a/gcc/testsuite/lib/target-supports.exp +++ b/gcc/testsuite/lib/target-supports.exp @@ -8621,3 +8621,80 @@ proc check_effective_target_autoincdec { } { } return 0 } + +# Return 1 if the target has support for stack probing designed +# to avoid stack-clash style attacks. +# +# This is used to restrict the stack-clash mitigation tests to +# just those targets that have been explicitly supported. +# +# In addition to the prologue work on those targets, each target's +# properties should be described in the functions below so that +# tests do not become a mess of unreadable target conditions. +# +proc check_effective_target_supports_stack_clash_protection { } { + + # Temporary until the target bits are fully ACK'd. +# if { [istarget aarch*-*-*] || [istarget x86_64-*-*] +# || [istarget i?86-*-*] || [istarget s390*-*-*] +# || [istarget powerpc*-*-*] || [istarget rs6000*-*-*] } { +# return 1 +# } + return 0 +} + +# Return 1 if the target creates a frame pointer for non-leaf functions +# Note we ignore cases where we apply tail call optimization here. +proc check_effective_target_frame_pointer_for_non_leaf { } { + if { [istarget aarch*-*-*] } { + return 1 + } + return 0 +} + +# Return 1 if the target's calling sequence or its ABI +# create implicit stack probes at or prior to function entry. +proc check_effective_target_caller_implicit_probes { } { + + # On x86/x86_64 the call instruction itself pushes the return + # address onto the stack. That is an implicit probe of *sp. + if { [istarget x86_64-*-*] || [istarget i?86-*-*] } { + return 1 + } + + # On PPC, the ABI mandates that the address of the outer + # frame be stored at *sp. Thus each allocation of stack + # space is itself an implicit probe of *sp. + if { [istarget powerpc*-*-*] || [istarget rs6000*-*-*] } { + return 1 + } + + # s390's ABI has a register save area allocated by the + # caller for use by the callee. The mere existence does + # not constitute a probe by the caller, but when the slots + # used by the callee those stores are implicit probes. + if { [istarget s390*-*-*] } { + return 1 + } + + # Not strictly true on aarch64, but we have agreed that we will + # consider any function that pushes SP more than 3kbytes into + # the guard page as broken. This essentially means that we can + # consider the aarch64 as having a caller implicit probe at + # *(sp + 1k). + if { [istarget aarch64*-*-*] } { + return 1; + } + + return 0 +} + +# Targets that potentially realign the stack pointer often cause residual +# stack allocations and make it difficult to elimination loops or residual +# allocations for dynamic stack allocations +proc check_effective_target_callee_realigns_stack { } { + if { [istarget x86_64-*-*] || [istarget i?86-*-*] } { + return 1 + } + return 0 +} diff --git a/gcc/toplev.c b/gcc/toplev.c index 7d2b8fffa0b..6f48e10850d 100644 --- a/gcc/toplev.c +++ b/gcc/toplev.c @@ -1605,6 +1605,26 @@ process_options (void) flag_associative_math = 0; } + /* -fstack-clash-protection is not currently supported on targets + where the stack grows up. */ + if (flag_stack_clash_protection && !STACK_GROWS_DOWNWARD) + { + warning_at (UNKNOWN_LOCATION, 0, + "%<-fstack-clash-protection%> is not supported on targets " + "where the stack grows from lower to higher addresses"); + flag_stack_clash_protection = 0; + } + + /* We can not support -fstack-check= and -fstack-clash-protection at + the same time. */ + if (flag_stack_check != NO_STACK_CHECK && flag_stack_clash_protection) + { + warning_at (UNKNOWN_LOCATION, 0, + "%<-fstack-check=%> and %<-fstack-clash_protection%> are " + "mutually exclusive. Disabling %<-fstack-check=%>"); + flag_stack_check = NO_STACK_CHECK; + } + /* With -fcx-limited-range, we do cheap and quick complex arithmetic. */ if (flag_cx_limited_range) flag_complex_method = 0;