fix: another attempt to avoid sigpipe on grep (#519)

Take this opportunity to factorize all the greps in /proc/cpuinfo
into a helper that avoids using a pipe to entirely avoid SIGPIPE
on a possibly gigantic /proc/cpuinfo

DEVELOPMENT.md

@@ -186,7 +186,7 @@ if [ $ret = $READ_CPUID_RET_OK ]; then
     cap_ssbd='Intel SSBD'
 elif [ $ret = $READ_CPUID_RET_ERR ] && [ "$g_mode" = live ]; then
     # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
-    if grep ^flags "$g_procfs/cpuinfo" | grep -qw ssbd; then
+    if cpuinfo_has_flag ssbd; then
         cap_ssbd='Intel SSBD (cpuinfo)'
         ret=$READ_CPUID_RET_OK
     fi

src/libs/210_cpu_detect.sh

@@ -195,6 +195,14 @@ is_arch_cap_mmio_immune() {
     [ "$cap_sbdr_ssdp_no" = 1 ] && [ "$cap_fbsdp_no" = 1 ] && [ "$cap_psdp_no" = 1 ]
 }
 
+# Whether the MMIO arch-cap immunity bits are undetermined because the
+# IA32_ARCH_CAPABILITIES MSR couldn't be read (msr module unavailable or kernel
+# lockdown).
+# Returns: 0 if undetermined, 1 otherwise
+is_arch_cap_mmio_undetermined() {
+    [ "$cap_sbdr_ssdp_no" = -1 ] || [ "$cap_fbsdp_no" = -1 ] || [ "$cap_psdp_no" = -1 ]
+}
+
 # Check whether the CPU is known to be unaffected by MMIO Stale Data (CVE-2022-21123/21125/21166)
 # Matches the kernel's NO_MMIO whitelist plus arch_cap_mmio_immune().
 # Model inventory and kernel-commit history are documented in check_mmio_linux().

src/libs/330_cpu_misc.sh

@@ -9,7 +9,7 @@ dmesg_grep() {
         # dmesg truncated
         return 2
     fi
-    ret_dmesg_grep_grepped=$(dmesg 2>/dev/null | grep -E "$1" | head -n1)
+    ret_dmesg_grep_grepped=$(dmesg 2>/dev/null | grep -m 1 -E "$1")
     # not found:
     [ -z "$ret_dmesg_grep_grepped" ] && return 1
     # found, output is in $ret_dmesg_grep_grepped
@@ -22,3 +22,9 @@ is_coreos() {
     command -v coreos-install >/dev/null 2>&1 && command -v toolbox >/dev/null 2>&1 && return 0
     return 1
 }
+
+# Check whether /proc/cpuinfo has $1 in the flags line
+# Returns: 0 if flag found, 1 otherwise
+cpuinfo_has_flag() {
+    grep -Eq '^flags\b.+\b'"$1"'\b' "$g_procfs/cpuinfo" 2>/dev/null
+}

src/libs/350_cpu_detect2.sh

@@ -20,8 +20,8 @@ parse_cpu_details() {
     cap_avx2=0
     cap_avx512=0
     if [ -e "$g_procfs/cpuinfo" ]; then
-        if grep -qw avx2 "$g_procfs/cpuinfo" 2>/dev/null; then cap_avx2=1; fi
-        if grep -qw avx512 "$g_procfs/cpuinfo" 2>/dev/null; then cap_avx512=1; fi
+        if cpuinfo_has_flag avx2; then cap_avx2=1; fi
+        if cpuinfo_has_flag avx512; then cap_avx512=1; fi
         cpu_vendor=$(grep '^vendor_id' "$g_procfs/cpuinfo" | awk '{print $3}' | head -n1)
         cpu_friendly_name=$(grep '^model name' "$g_procfs/cpuinfo" | cut -d: -f2- | head -n1 | sed -e 's/^ *//')
         # ARM-style cpuinfo: parse per-core implementer/part/arch/variant/revision lists

src/libs/400_hw_check.sh

@@ -513,7 +513,7 @@ check_cpu() {
     fi
     if [ -z "$cap_ibrs" ] && [ $ret = $READ_CPUID_RET_ERR ] && has_runtime; then
         # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
-        if grep ^flags "$g_procfs/cpuinfo" | grep -qw ibrs; then
+        if cpuinfo_has_flag ibrs; then
             cap_ibrs='IBRS (cpuinfo)'
             cap_spec_ctrl=1
             pstatus green YES "ibrs flag in $g_procfs/cpuinfo"
@@ -588,7 +588,7 @@ check_cpu() {
         if [ $ret = $READ_CPUID_RET_OK ]; then
             cap_ibpb='IBPB_SUPPORT'
             pstatus green YES "IBPB_SUPPORT feature bit"
-        elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && grep ^flags "$g_procfs/cpuinfo" | grep -qw ibpb; then
+        elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && cpuinfo_has_flag ibpb; then
             # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
             cap_ibpb='IBPB (cpuinfo)'
             pstatus green YES "ibpb flag in $g_procfs/cpuinfo"
@@ -661,7 +661,7 @@ check_cpu() {
     fi
     if [ -z "$cap_stibp" ] && [ $ret = $READ_CPUID_RET_ERR ] && has_runtime; then
         # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
-        if grep ^flags "$g_procfs/cpuinfo" | grep -qw stibp; then
+        if cpuinfo_has_flag stibp; then
             cap_stibp='STIBP (cpuinfo)'
             pstatus green YES "stibp flag in $g_procfs/cpuinfo"
             ret=$READ_CPUID_RET_OK
@@ -733,9 +733,9 @@ check_cpu() {
 
     if [ -z "$cap_ssbd" ] && [ "$ret24" = $READ_CPUID_RET_ERR ] && [ "$ret25" = $READ_CPUID_RET_ERR ] && has_runtime; then
         # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
-        if grep ^flags "$g_procfs/cpuinfo" | grep -qw ssbd; then
+        if cpuinfo_has_flag ssbd; then
             cap_ssbd='SSBD (cpuinfo)'
-        elif grep ^flags "$g_procfs/cpuinfo" | grep -qw virt_ssbd; then
+        elif cpuinfo_has_flag virt_ssbd; then
             cap_ssbd='SSBD in VIRT_SPEC_CTRL (cpuinfo)'
         fi
     fi
@@ -795,7 +795,7 @@ check_cpu() {
     if [ $ret = $READ_CPUID_RET_OK ]; then
         pstatus green YES "L1D flush feature bit"
         cap_l1df=1
-    elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && grep ^flags "$g_procfs/cpuinfo" | grep -qw flush_l1d; then
+    elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && cpuinfo_has_flag flush_l1d; then
         # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
         pstatus green YES "flush_l1d flag in $g_procfs/cpuinfo"
         cap_l1df=1
@@ -815,7 +815,7 @@ check_cpu() {
         if [ $ret = $READ_CPUID_RET_OK ]; then
             cap_md_clear=1
             pstatus green YES "MD_CLEAR feature bit"
-        elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && grep ^flags "$g_procfs/cpuinfo" | grep -qw md_clear; then
+        elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && cpuinfo_has_flag md_clear; then
             # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
             cap_md_clear=1
             pstatus green YES "md_clear flag in $g_procfs/cpuinfo"
@@ -885,7 +885,7 @@ check_cpu() {
         if [ $ret = $READ_CPUID_RET_OK ]; then
             pstatus green YES
             cap_arch_capabilities=1
-        elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && grep ^flags "$g_procfs/cpuinfo" | grep -qw arch_capabilities; then
+        elif [ $ret = $READ_CPUID_RET_ERR ] && has_runtime && cpuinfo_has_flag arch_capabilities; then
             # CPUID device unavailable (e.g. in a VM): fall back to /proc/cpuinfo
             pstatus green YES "arch_capabilities flag in $g_procfs/cpuinfo"
             cap_arch_capabilities=1
@@ -987,8 +987,33 @@ check_cpu() {
                     pstatus yellow NO
                 fi
             elif [ $ret = $READ_MSR_RET_KO ]; then
+                # the MSR access faulted: the register is genuinely absent, so the
+                # pre-seeded 0 ("not advertised") values are correct.
                 pstatus yellow NO
             else
+                # RET_ERR (no msr module) or RET_LOCKDOWN (MSR reads restricted):
+                # CPUID told us the MSR exists but we couldn't read it, so its bits
+                # are undetermined, not 0. Leaving them at 0 would falsely claim the
+                # CPU "explicitly indicates not immune".
+                # Reset every arch-cap-derived value to -1 (UNKNOWN) instead.
+                cap_rdcl_no=-1
+                cap_taa_no=-1
+                cap_mds_no=-1
+                cap_ibrs_all=-1
+                cap_rsba=-1
+                cap_l1dflush_no=-1
+                cap_ssb_no=-1
+                cap_pschange_msc_no=-1
+                cap_tsx_ctrl_msr=-1
+                cap_gds_ctrl=-1
+                cap_gds_no=-1
+                cap_rfds_no=-1
+                cap_rfds_clear=-1
+                cap_its_no=-1
+                cap_sbdr_ssdp_no=-1
+                cap_fbsdp_no=-1
+                cap_psdp_no=-1
+                cap_fb_clear=-1
                 pstatus yellow UNKNOWN "$ret_read_msr_msg"
             fi
         fi

src/vulns-helpers/check_mds.sh

@@ -138,7 +138,7 @@ check_mds_linux() {
         if is_x86_kernel; then
             pr_info_nol "* Kernel supports using MD_CLEAR mitigation: "
             kernel_md_clear_can_tell=1
-            if [ "$g_mode" = live ] && grep ^flags "$g_procfs/cpuinfo" | grep -qw md_clear; then
+            if [ "$g_mode" = live ] && cpuinfo_has_flag md_clear; then
                 kernel_md_clear="md_clear found in $g_procfs/cpuinfo"
                 pstatus green YES "$kernel_md_clear"
             fi

src/vulns-helpers/check_mmio.sh

@@ -6,16 +6,23 @@ check_mmio_bsd() {
     # the only partial defense available, and without OS-level VERW invocation it
     # cannot close the vulnerability.
     local unk
-    unk="your CPU's MMIO Stale Data status is unknown (Intel never officially assessed this CPU, its servicing period has ended)"
     if ! is_cpu_affected "$cve"; then
         pvulnstatus "$cve" OK "your CPU vendor reported your CPU model as not affected"
     elif is_cpu_mmio_unknown; then
+        if is_arch_cap_mmio_undetermined; then
+            # We only landed in the "unknown" bucket because the IA32_ARCH_CAPABILITIES
+            # MSR couldn't be read: the CPU might actually advertise MMIO immunity.
+            unk="your CPU's MMIO Stale Data status could not be determined: the IA32_ARCH_CAPABILITIES MSR (0x10a) couldn't be read"
+            pvulnstatus "$cve" UNK "$unk; load the cpuctl module and/or re-run as root to get a definitive answer"
+        else
+            unk="your CPU's MMIO Stale Data status is unknown (Intel never officially assessed this CPU, its servicing period has ended)"
             if [ "$opt_paranoid" = 1 ]; then
                 pvulnstatus "$cve" VULN "$unk, and no BSD mitigation exists"
                 explain "There is no known mitigation for this CPU model. Even with up-to-date microcode, BSD kernels do not invoke VERW for MMIO Stale Data clearing. Only a hardware replacement can fully address this."
             else
                 pvulnstatus "$cve" UNK "$unk; no BSD mitigation exists in any case"
             fi
+        fi
     else
         pvulnstatus "$cve" VULN "your CPU is affected and no BSD has implemented an MMIO Stale Data mitigation"
         explain "No BSD kernel currently implements an MMIO Stale Data mitigation (which would require invoking VERW at context switches and VM-entries). Updating CPU microcode alone does not mitigate this vulnerability without OS cooperation."
@@ -24,7 +31,7 @@ check_mmio_bsd() {
 
 # MMIO Stale Data (Processor MMIO Stale Data Vulnerabilities) - Linux mitigation check
 check_mmio_linux() {
-    local status sys_interface_available msg kernel_mmio kernel_mmio_can_tell mmio_mitigated mmio_smt_mitigated mystatus mymsg unk
+    local status sys_interface_available msg kernel_mmio kernel_mmio_can_tell kernel_mmio_unknown_aware mmio_mitigated mmio_smt_mitigated mystatus mymsg unk
     status=UNK
     sys_interface_available=0
     msg=''
@@ -166,6 +173,11 @@ check_mmio_linux() {
         # MMIO Stale Data is Intel-only; skip x86-specific kernel/MSR checks on non-x86 kernels
         kernel_mmio=''
         kernel_mmio_can_tell=0
+        # Whether this kernel implements the X86_BUG_MMIO_UNKNOWN distinction, i.e. can
+        # report "Unknown: No mitigations" for CPUs Intel never assessed. Only such kernels
+        # emit a *trustworthy* "Not affected": they would have said "Unknown" instead if the
+        # CPU were in the unknown bucket. Detected by the presence of the literal sysfs string in the kernel image.
+        kernel_mmio_unknown_aware=0
         if is_x86_kernel; then
             pr_info_nol "* Kernel supports MMIO Stale Data mitigation: "
             kernel_mmio_can_tell=1
@@ -176,6 +188,10 @@ check_mmio_linux() {
                 kernel_mmio='found MMIO Stale Data mitigation evidence in kernel image'
                 pstatus green YES "$kernel_mmio"
             fi
+            if [ -z "$g_kernel_err" ] && grep -qF 'Unknown: No mitigations' "$g_kernel" 2>/dev/null; then
+                pr_debug "mmio: kernel image knows the 'Unknown: No mitigations' state (X86_BUG_MMIO_UNKNOWN-aware)"
+                kernel_mmio_unknown_aware=1
+            fi
             if [ -z "$kernel_mmio" ] && [ -n "$opt_config" ] && grep -q '^CONFIG_MITIGATION_MMIO_STALE_DATA=y' "$opt_config"; then
                 kernel_mmio='found MMIO Stale Data mitigation config option enabled'
                 pstatus green YES "$kernel_mmio"
@@ -241,13 +257,33 @@ check_mmio_linux() {
         # Bypass the normal sysfs reconciliation: sysfs reports "Unknown: No mitigations"
         # only on v6.0-v6.15. On earlier and on v6.16+ kernels it wrongly says "Not affected"
         # for these CPUs (which predate FB_CLEAR microcode and Intel's affected-processor list).
+        if is_arch_cap_mmio_undetermined; then
+            # We landed in the "unknown" bucket only because the IA32_ARCH_CAPABILITIES
+            # MSR couldn't be read from userspace (no msr module, or kernel lockdown under
+            # Secure Boot): the CPU might actually advertise MMIO immunity
+            # through FBSDP_NO/PSDP_NO/SBDR_SSDP_NO, but we can't read it, however the kernel can.
+            #
+            # We can trust a sysfs "Not affected" only if this kernel is X86_BUG_MMIO_UNKNOWN-aware:
+            # such a kernel would have reported "Unknown: No mitigations" instead if the CPU were in
+            # the unknown bucket, so "Not affected" genuinely means arch-cap immune.
+            # On kernels that lack that distinction, a "Not affected" is not trustworthy for these CPUs,
+            # so we keep UNK.
+            if [ "$g_mode" = live ] && [ "$sys_interface_available" = 1 ] &&
+                [ "$kernel_mmio_unknown_aware" = 1 ] && [ "$status" = OK ]; then
+                pvulnstatus "$cve" OK "your kernel reports your CPU as not affected, and this kernel distinguishes the MMIO 'unknown' state, so its verdict is trustworthy (we couldn't read the IA32_ARCH_CAPABILITIES MSR ourselves)"
+            else
+                unk="your CPU's MMIO Stale Data status could not be determined: the IA32_ARCH_CAPABILITIES MSR (0x10a) couldn't be read"
+                pvulnstatus "$cve" UNK "$unk; load the msr module and/or disable kernel lockdown, then re-run as root to get a definitive answer"
+            fi
+        else
             unk="your CPU's MMIO Stale Data status is unknown (Intel never officially assessed this CPU, its servicing period has ended)"
             if [ "$opt_paranoid" = 1 ]; then
                 pvulnstatus "$cve" VULN "$unk, and no mitigation is available"
-            explain "There is no known mitigation for this CPU model. Intel ended its servicing period without evaluating whether it is affected by MMIO Stale Data vulnerabilities, so no FB_CLEAR-capable microcode was released. Consider replacing affected hardware."
+                explain "There is no known mitigation for this CPU model. Intel ended its servicing period without evaluating whether it is affected by MMIO Stale Data vulnerabilities, so no FB_CLEAR-capable microcode was released."
             else
                 pvulnstatus "$cve" UNK "$unk; no mitigation is available in any case"
             fi
+        fi
     else
         if [ "$opt_sysfs_only" != 1 ]; then
             # compute mystatus and mymsg from our own logic

src/vulns/CVE-2017-5715.sh

@@ -306,7 +306,7 @@ check_CVE_2017_5715_linux() {
                 # which in that case means ibrs is supported *and* enabled for kernel & user
                 # as per the ibrs patch series v3
                 if [ -z "$g_ibrs_supported" ]; then
-                    if grep ^flags "$g_procfs/cpuinfo" | grep -qw spec_ctrl_ibrs; then
+                    if cpuinfo_has_flag spec_ctrl_ibrs; then
                         pr_debug "ibrs: found spec_ctrl_ibrs flag in $g_procfs/cpuinfo"
                         g_ibrs_supported="spec_ctrl_ibrs flag in $g_procfs/cpuinfo"
                         # enabled=2 -> kernel & user

src/vulns/CVE-2017-5754.sh

@@ -11,7 +11,7 @@
 pti_performance_check() {
     local ret pcid invpcid
     pr_info_nol "  * Reduced performance impact of PTI: "
-    if [ -e "$g_procfs/cpuinfo" ] && grep ^flags "$g_procfs/cpuinfo" | grep -qw pcid; then
+    if cpuinfo_has_flag pcid; then
         pcid=1
     else
         read_cpuid 0x1 0x0 "$ECX" 17 1 1
@@ -21,7 +21,7 @@ pti_performance_check() {
         fi
     fi
 
-    if [ -e "$g_procfs/cpuinfo" ] && grep ^flags "$g_procfs/cpuinfo" | grep -qw invpcid; then
+    if cpuinfo_has_flag invpcid; then
         invpcid=1
     else
         read_cpuid 0x7 0x0 "$EBX" 10 1 1
@@ -110,11 +110,11 @@ check_CVE_2017_5754_linux() {
             dmesg_grep="$dmesg_grep|x86/pti: Unmapping kernel while in userspace"
             # aarch64
             dmesg_grep="$dmesg_grep|CPU features: detected( feature)?: Kernel page table isolation \(KPTI\)"
-            if grep ^flags "$g_procfs/cpuinfo" | grep -qw pti; then
+            if cpuinfo_has_flag pti; then
                 # vanilla PTI patch sets the 'pti' flag in cpuinfo
                 pr_debug "kpti_enabled: found 'pti' flag in $g_procfs/cpuinfo"
                 kpti_enabled=1
-            elif grep ^flags "$g_procfs/cpuinfo" | grep -qw kaiser; then
+            elif cpuinfo_has_flag kaiser; then
                 # kernel line 4.9 sets the 'kaiser' flag in cpuinfo
                 pr_debug "kpti_enabled: found 'kaiser' flag in $g_procfs/cpuinfo"
                 kpti_enabled=1

src/vulns/CVE-2018-3646.sh

@@ -89,7 +89,7 @@ check_CVE_2018_3646_linux() {
 
         pr_info "* Mitigation 2"
         pr_info_nol "  * L1D flush is supported by kernel: "
-        if [ "$g_mode" = live ] && grep -qw flush_l1d "$g_procfs/cpuinfo"; then
+        if [ "$g_mode" = live ] && cpuinfo_has_flag flush_l1d; then
             l1d_kernel="found flush_l1d in $g_procfs/cpuinfo"
         fi
         if [ -z "$l1d_kernel" ]; then
@@ -162,7 +162,7 @@ check_CVE_2018_3646_linux() {
 
         pr_info_nol "  * Hardware-backed L1D flush supported: "
         if [ "$g_mode" = live ]; then
-            if grep -qw flush_l1d "$g_procfs/cpuinfo" || [ -n "$l1d_xen_hardware" ]; then
+            if cpuinfo_has_flag flush_l1d || [ -n "$l1d_xen_hardware" ]; then
                 pstatus green YES "performance impact of the mitigation will be greatly reduced"
             else
                 pstatus blue NO "flush will be done in software, this is slower"