hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I4PM10 CVE: NA
--------------------------------
During kernel copy_from_user processing, the kernel triggers a RAS exception when reading pages. In this solution, we identify this scenario in the kernel do_sea processing process, send SIGBUS signals to the process that triggers copy_from_user and isolate memory pages, preventing kernel panic.
At the same time, we use cmdline(uce_kernel_recovery) or proc (/proc/sys/kernel/uce_kernel_recovery) to control this feature on/off.
Usage: 1. Each bit controls whether this feature is turned on in a scene, 1 means turned on and 0 means turned off. 2. Bit2 represents copy_from_user scene, other bits are currently reserved.
eg: make copy_from_user scene open this feature: 1. echo 4 > /proc/sys/kernel/uce_kernel_recovery. or 2. uce_kernel_recovery=4 add to cmdline.
Signed-off-by: Tong Tiangen tongtiangen@huawei.com --- Documentation/admin-guide/sysctl/kernel.rst | 17 ++ arch/arm64/Kconfig | 9 + arch/arm64/include/asm/exception.h | 13 ++ arch/arm64/lib/copy_from_user.S | 11 ++ arch/arm64/mm/Makefile | 2 + arch/arm64/mm/fault.c | 4 + arch/arm64/mm/uce_kernel_recovery.c | 198 ++++++++++++++++++++ 7 files changed, 254 insertions(+) create mode 100644 arch/arm64/mm/uce_kernel_recovery.c
diff --git a/Documentation/admin-guide/sysctl/kernel.rst b/Documentation/admin-guide/sysctl/kernel.rst index 7d5e8a67c775..56bb3afe3794 100644 --- a/Documentation/admin-guide/sysctl/kernel.rst +++ b/Documentation/admin-guide/sysctl/kernel.rst @@ -1526,3 +1526,20 @@ is 10 seconds.
The softlockup threshold is (``2 * watchdog_thresh``). Setting this tunable to zero will disable lockup detection altogether. + +uce_kernel_recovery(ARM64 only) +=============================== + +This value can be used to control whether panic the kernel when UCE RAS +errors occur in a specific scenario. Each bit controls a scene, 1 means +avoid kernel panic when encountering UCE RAS error in this scenario, and +0 means kernel panic. + +Current usage of each bit: + +============ ============== +bit0 reserved +bit1 reserved +bit2 copy_from_user +bit3 ~ bit31 reserved +============ ============== diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index c0f6a275f798..226d3eb61524 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -1625,6 +1625,15 @@ config ARM64_CNP at runtime, and does not affect PEs that do not implement this feature.
+config ARM64_UCE_KERNEL_RECOVERY + bool "arm64 uce kernel recovery for special scenario" + depends on ACPI_APEI_SEA + help + With ARM v8.2 RAS Extension, SEA are usually triggered when memory + error are consumed. In some cases, if the error address is in a + user page there is a chance to recover. we can isolate this page + and killing process instead of die. + endmenu
menu "ARMv8.3 architectural features" diff --git a/arch/arm64/include/asm/exception.h b/arch/arm64/include/asm/exception.h index 0756191f44f6..731cf01d9296 100644 --- a/arch/arm64/include/asm/exception.h +++ b/arch/arm64/include/asm/exception.h @@ -19,6 +19,19 @@ #define __exception_irq_entry __kprobes #endif
+#ifdef CONFIG_ARM64_UCE_KERNEL_RECOVERY +bool arm64_process_kernel_sea(unsigned long addr, unsigned int esr, + struct pt_regs *regs, int sig, + int code, void __user *siaddr); +#else +static inline bool arm64_process_kernel_sea(unsigned long addr, unsigned int esr, + struct pt_regs *regs, int sig, + int code, void __user *siaddr) +{ + return false; +} +#endif + static inline u32 disr_to_esr(u64 disr) { unsigned int esr = ESR_ELx_EC_SERROR << ESR_ELx_EC_SHIFT; diff --git a/arch/arm64/lib/copy_from_user.S b/arch/arm64/lib/copy_from_user.S index 2cf999e41d30..100de4e2d9ee 100644 --- a/arch/arm64/lib/copy_from_user.S +++ b/arch/arm64/lib/copy_from_user.S @@ -60,6 +60,17 @@ SYM_FUNC_START(__arch_copy_from_user) #include "copy_template.S" mov x0, #0 // Nothing to copy ret + +/* + * In feature CONFIG_ARM64_UCE_KERNEL_RECOVERY, if RAS error is triggered + * in copy_from_user(), RAS error is processed in do_sea() and + * copy_from_user_sea_fallback will be assigned to regs->pc, finally return + * here to continue processing. + */ + .global copy_from_user_sea_fallback +copy_from_user_sea_fallback: + sub x0, end, dst // bytes not copied + ret SYM_FUNC_END(__arch_copy_from_user) EXPORT_SYMBOL(__arch_copy_from_user)
diff --git a/arch/arm64/mm/Makefile b/arch/arm64/mm/Makefile index 42e107d6da4f..3634ad81bdf1 100644 --- a/arch/arm64/mm/Makefile +++ b/arch/arm64/mm/Makefile @@ -11,6 +11,8 @@ obj-$(CONFIG_DEBUG_VIRTUAL) += physaddr.o obj-$(CONFIG_ARM64_MTE) += mteswap.o KASAN_SANITIZE_physaddr.o += n
+obj-$(CONFIG_ARM64_UCE_KERNEL_RECOVERY) += uce_kernel_recovery.o + obj-$(CONFIG_KASAN) += kasan_init.o KASAN_SANITIZE_kasan_init.o := n
diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c index 3fc5aceb72eb..7da2f8118b35 100644 --- a/arch/arm64/mm/fault.c +++ b/arch/arm64/mm/fault.c @@ -653,6 +653,10 @@ static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs) siaddr = NULL; else siaddr = (void __user *)addr; + + if (arm64_process_kernel_sea(addr, esr, regs, inf->sig, inf->code, siaddr)) + return 0; + arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
return 0; diff --git a/arch/arm64/mm/uce_kernel_recovery.c b/arch/arm64/mm/uce_kernel_recovery.c new file mode 100644 index 000000000000..c654dc6c4dfd --- /dev/null +++ b/arch/arm64/mm/uce_kernel_recovery.c @@ -0,0 +1,198 @@ +// SPDX-License-Identifier: GPL-2.0-only + +#define pr_fmt(fmt) "ARM64 UCE: " fmt + +#include <linux/acpi.h> +#include <linux/kallsyms.h> +#include <linux/mm.h> +#include <linux/kernel.h> +#include <linux/sysctl.h> + +#include <asm/acpi.h> +#include <asm/exception.h> +#include <asm/system_misc.h> +#include <asm/traps.h> +#include <asm/esr.h> + +struct uce_kernel_recovery_info { + int (*fn)(void); + const char *name; + unsigned long addr; + unsigned long size; +}; + +int copy_from_user_sea_fallback(void); + +static int kernel_access_sea_recovery; +static int kernel_uce_recovery_sysctl_max = 7; + +#define UCE_KER_REC_NUM ARRAY_SIZE(reco_info) +static struct uce_kernel_recovery_info reco_info[] = { + {NULL, NULL, 0, 0}, /* reserved */ + {NULL, NULL, 0, 0}, /* reserved */ + {copy_from_user_sea_fallback, "__arch_copy_from_user", (unsigned long)__arch_copy_from_user, 0}, +}; + +static struct ctl_table uce_kernel_recovery_ctl_table[] = { + { + .procname = "uce_kernel_recovery", + .data = &kernel_access_sea_recovery, + .maxlen = sizeof(kernel_access_sea_recovery), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &kernel_uce_recovery_sysctl_max, + }, + { } +}; + +static int __init kernel_access_sea_recovery_init(void) +{ + unsigned long addr, size, offset; + unsigned int i; + + for (i = 0; i < UCE_KER_REC_NUM; i++) { + addr = reco_info[i].addr; + + if (!addr) + continue; + + if (!kallsyms_lookup_size_offset(addr, &size, &offset)) { + pr_info("symbol %s lookup addr fail.\n", + reco_info[i].name); + size = 0; + } + + reco_info[i].size = size; + } + + if (!register_sysctl("kernel", uce_kernel_recovery_ctl_table)) + pr_err("register sysctl table fail.\n"); + + return 1; +} +fs_initcall(kernel_access_sea_recovery_init); + +static int __init enable_kernel_access_sea_recovery(char *str) +{ + int max = (1 << UCE_KER_REC_NUM) - 1; + int val; + + if (kstrtoint(str, 0, &val)) + return -EINVAL; + + if (val < 0 || val > max) { + pr_info("invalid uce_kernel_recovery value %d", val); + return -EINVAL; + } + + kernel_access_sea_recovery = val; + + return 1; +} +__setup("uce_kernel_recovery=", enable_kernel_access_sea_recovery); + +/* + * what is kernel recovery? + * If the process's private data is accessed in the kernel mode to trigger + * special sea fault, it can controlled by killing the process and isolating + * the failure pages instead of die. + */ +static int is_in_kernel_recovery(unsigned int esr, struct pt_regs *regs) +{ + /* + * target insn: ldp-pre, ldp-post, ldp-offset, + * ldr-64bit-pre/pose, ldr-32bit-pre/post, ldrb-pre/post, ldrh-pre/post + */ + u32 target_insn[] = {0xa8c, 0xa9c, 0xa94, 0xf84, 0x784, 0x384, 0xb84}; + void *pc = (void *)instruction_pointer(regs); + struct uce_kernel_recovery_info *info; + bool insn_match = false; + u32 insn; + int i; + + pr_emerg("%s-%d, kernel recovery: 0x%x, esr: 0x%08x -- %s, %pS\n", + current->comm, current->pid, kernel_access_sea_recovery, esr, + esr_get_class_string(esr), pc); + + if (aarch64_insn_read((void *)pc, &insn)) { + pr_emerg("insn read fail.\n"); + return -EFAULT; + } + + /* + * We process special ESR: + * EC : 0b100101 Data Abort taken without a change in Exception level. + * DFSC : 0b010000 Synchronous External abort, not on translation table + * walk or hardware update of translation table. + * eg: 0x96000610 + */ + if (ESR_ELx_EC(esr) != ESR_ELx_EC_DABT_CUR || + (esr & ESR_ELx_FSC) != ESR_ELx_FSC_EXTABT) { + pr_emerg("esr not match.\n"); + return -EINVAL; + } + + insn = (insn >> 20) & 0xffc; + for (i = 0; i < ARRAY_SIZE(target_insn); i++) { + if (insn == target_insn[i]) { + insn_match = true; + break; + } + } + + if (!insn_match) { + pr_emerg("insn 0x%x is not match.\n", insn); + return -EINVAL; + } + + for (i = 0; i < UCE_KER_REC_NUM; i++) { + if (!((kernel_access_sea_recovery >> i) & 0x1)) + continue; + + info = &reco_info[i]; + if (info->fn && regs->pc >= info->addr && + regs->pc < (info->addr + info->size)) { + pr_emerg("total match %s success.\n", info->name); + return i; + } + } + + pr_emerg("scene is not match, kernel recovery %d.\n", + kernel_access_sea_recovery); + return -EINVAL; +} + +bool arm64_process_kernel_sea(unsigned long addr, unsigned int esr, + struct pt_regs *regs, int sig, + int code, void __user *siaddr) +{ + int idx; + + if (user_mode(regs) || apei_claim_sea(regs) < 0) + return false; + + if (!current->mm || !kernel_access_sea_recovery) { + pr_emerg("kernel recovery %d, %s-%d is %s-thread.\n", + kernel_access_sea_recovery, + current->comm, current->pid, + (current->mm) ? "user" : "kernel"); + + return false; + } + + idx = is_in_kernel_recovery(esr, regs); + if (idx < 0 || idx >= UCE_KER_REC_NUM) { + pr_emerg("Uncorrected hardware memory error (sence not match or sence switch is off) in kernel-access\n"); + return false; + } + + current->thread.fault_address = 0; + current->thread.fault_code = esr; + regs->pc = (unsigned long)reco_info[idx].fn; + + arm64_force_sig_fault(sig, code, siaddr, + "Uncorrected hardware memory use with kernel recovery in kernel-access\n"); + + return true; +}