Re: [RFC PATCH tip/master V3 0/8] kprobes/x86: Make kprobes instruction buffers read-only
Ping? On Wed, 29 Mar 2017 13:55:44 +0900 Masami Hiramatsuwrote: > Hi, > > This is the 3rd version of the series. I just updated 4/8 according > to Ingo's comment. > > This series tries to make kprobes instruction buffers read-only > pages. Since those buffers are used for trampoline code, those > are a part of "text area" and it should be marked as ro for > avoiding unexpected modification. And this actually fix a warning > rodata sanity check reported by lkp-robot. > > https://lkml.org/lkml/2017/2/27/161 > > This change requires changing the kprobe-booster at first > because it can modify the instruction buffer to add a jump while > resuming from single-stepping. Of course after we make the buffer > readonly, we may not be able to modify it while probing. > > So, at first this series checks the current bootable instructions > and fixes a missed instruction (call far), modifies can_boost to > use x86 instruction decoder, and inserts "booster" jump while > preparing instruction buffer instead of resuming from single-stepping. > At last, it makes the buffers for kprobes and optprobe readonly. > > This series also has some cleanup patches related to above > changes. > > Changes from V2: > - [4/8]: Fix imbalanced curly braces and remove unneeded >(void *) cast. > > --- > > Masami Hiramatsu (8): > kprobes/x86: Fix not to boost call far instruction > kprobes/x86: Fix the description of __copy_instruction() > kprobes/x86: Use instruction decoder for booster > kprobes/x86: Do not modify singlestep buffer while resuming > kprobes/x86: Make boostable flag boolean > kprobes/x86: Set kprobes pages readonly > kprobes/x86: Use probe_kernel_read instead of memcpy > kprobes/x86: Consolidate insn decoder users for copying code > > > arch/x86/include/asm/kprobes.h |7 +- > arch/x86/kernel/kprobes/common.h |4 + > arch/x86/kernel/kprobes/core.c | 149 > +++--- > arch/x86/kernel/kprobes/ftrace.c |2 - > arch/x86/kernel/kprobes/opt.c| 13 +++ > 5 files changed, 90 insertions(+), 85 deletions(-) > > -- > Masami Hiramatsu (Linaro) -- Masami Hiramatsu
Re: [RFC PATCH tip/master V3 0/8] kprobes/x86: Make kprobes instruction buffers read-only
Ping? On Wed, 29 Mar 2017 13:55:44 +0900 Masami Hiramatsu wrote: > Hi, > > This is the 3rd version of the series. I just updated 4/8 according > to Ingo's comment. > > This series tries to make kprobes instruction buffers read-only > pages. Since those buffers are used for trampoline code, those > are a part of "text area" and it should be marked as ro for > avoiding unexpected modification. And this actually fix a warning > rodata sanity check reported by lkp-robot. > > https://lkml.org/lkml/2017/2/27/161 > > This change requires changing the kprobe-booster at first > because it can modify the instruction buffer to add a jump while > resuming from single-stepping. Of course after we make the buffer > readonly, we may not be able to modify it while probing. > > So, at first this series checks the current bootable instructions > and fixes a missed instruction (call far), modifies can_boost to > use x86 instruction decoder, and inserts "booster" jump while > preparing instruction buffer instead of resuming from single-stepping. > At last, it makes the buffers for kprobes and optprobe readonly. > > This series also has some cleanup patches related to above > changes. > > Changes from V2: > - [4/8]: Fix imbalanced curly braces and remove unneeded >(void *) cast. > > --- > > Masami Hiramatsu (8): > kprobes/x86: Fix not to boost call far instruction > kprobes/x86: Fix the description of __copy_instruction() > kprobes/x86: Use instruction decoder for booster > kprobes/x86: Do not modify singlestep buffer while resuming > kprobes/x86: Make boostable flag boolean > kprobes/x86: Set kprobes pages readonly > kprobes/x86: Use probe_kernel_read instead of memcpy > kprobes/x86: Consolidate insn decoder users for copying code > > > arch/x86/include/asm/kprobes.h |7 +- > arch/x86/kernel/kprobes/common.h |4 + > arch/x86/kernel/kprobes/core.c | 149 > +++--- > arch/x86/kernel/kprobes/ftrace.c |2 - > arch/x86/kernel/kprobes/opt.c| 13 +++ > 5 files changed, 90 insertions(+), 85 deletions(-) > > -- > Masami Hiramatsu (Linaro) -- Masami Hiramatsu
[RFC PATCH tip/master V3 0/8] kprobes/x86: Make kprobes instruction buffers read-only
Hi, This is the 3rd version of the series. I just updated 4/8 according to Ingo's comment. This series tries to make kprobes instruction buffers read-only pages. Since those buffers are used for trampoline code, those are a part of "text area" and it should be marked as ro for avoiding unexpected modification. And this actually fix a warning rodata sanity check reported by lkp-robot. https://lkml.org/lkml/2017/2/27/161 This change requires changing the kprobe-booster at first because it can modify the instruction buffer to add a jump while resuming from single-stepping. Of course after we make the buffer readonly, we may not be able to modify it while probing. So, at first this series checks the current bootable instructions and fixes a missed instruction (call far), modifies can_boost to use x86 instruction decoder, and inserts "booster" jump while preparing instruction buffer instead of resuming from single-stepping. At last, it makes the buffers for kprobes and optprobe readonly. This series also has some cleanup patches related to above changes. Changes from V2: - [4/8]: Fix imbalanced curly braces and remove unneeded (void *) cast. --- Masami Hiramatsu (8): kprobes/x86: Fix not to boost call far instruction kprobes/x86: Fix the description of __copy_instruction() kprobes/x86: Use instruction decoder for booster kprobes/x86: Do not modify singlestep buffer while resuming kprobes/x86: Make boostable flag boolean kprobes/x86: Set kprobes pages readonly kprobes/x86: Use probe_kernel_read instead of memcpy kprobes/x86: Consolidate insn decoder users for copying code arch/x86/include/asm/kprobes.h |7 +- arch/x86/kernel/kprobes/common.h |4 + arch/x86/kernel/kprobes/core.c | 149 +++--- arch/x86/kernel/kprobes/ftrace.c |2 - arch/x86/kernel/kprobes/opt.c| 13 +++ 5 files changed, 90 insertions(+), 85 deletions(-) -- Masami Hiramatsu (Linaro)
[RFC PATCH tip/master V3 0/8] kprobes/x86: Make kprobes instruction buffers read-only
Hi, This is the 3rd version of the series. I just updated 4/8 according to Ingo's comment. This series tries to make kprobes instruction buffers read-only pages. Since those buffers are used for trampoline code, those are a part of "text area" and it should be marked as ro for avoiding unexpected modification. And this actually fix a warning rodata sanity check reported by lkp-robot. https://lkml.org/lkml/2017/2/27/161 This change requires changing the kprobe-booster at first because it can modify the instruction buffer to add a jump while resuming from single-stepping. Of course after we make the buffer readonly, we may not be able to modify it while probing. So, at first this series checks the current bootable instructions and fixes a missed instruction (call far), modifies can_boost to use x86 instruction decoder, and inserts "booster" jump while preparing instruction buffer instead of resuming from single-stepping. At last, it makes the buffers for kprobes and optprobe readonly. This series also has some cleanup patches related to above changes. Changes from V2: - [4/8]: Fix imbalanced curly braces and remove unneeded (void *) cast. --- Masami Hiramatsu (8): kprobes/x86: Fix not to boost call far instruction kprobes/x86: Fix the description of __copy_instruction() kprobes/x86: Use instruction decoder for booster kprobes/x86: Do not modify singlestep buffer while resuming kprobes/x86: Make boostable flag boolean kprobes/x86: Set kprobes pages readonly kprobes/x86: Use probe_kernel_read instead of memcpy kprobes/x86: Consolidate insn decoder users for copying code arch/x86/include/asm/kprobes.h |7 +- arch/x86/kernel/kprobes/common.h |4 + arch/x86/kernel/kprobes/core.c | 149 +++--- arch/x86/kernel/kprobes/ftrace.c |2 - arch/x86/kernel/kprobes/opt.c| 13 +++ 5 files changed, 90 insertions(+), 85 deletions(-) -- Masami Hiramatsu (Linaro)