src/nspawn/nspawn-seccomp.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2016 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <errno.h>
  21 #include <linux/netlink.h>
  22 #include <sys/capability.h>
  23 #include <sys/types.h>
  24
  25 #ifdef HAVE_SECCOMP
  26 #include <seccomp.h>
  27 #endif
  28
  29 #include "log.h"
  30
  31 #ifdef HAVE_SECCOMP
  32 #include "seccomp-util.h"
  33 #endif
  34
  35 #include "nspawn-seccomp.h"
  36
  37 #ifdef HAVE_SECCOMP
  38
  39 static int seccomp_add_default_syscall_filter(scmp_filter_ctx ctx,
  40                                               uint64_t cap_list_retain) {
  41         unsigned i;
  42         int r;
  43         static const struct {
  44                 uint64_t capability;
  45                 int syscall_num;
  46         } blacklist[] = {
  47                 { 0,              SCMP_SYS(_sysctl)             }, /* obsolete syscall */
  48                 { 0,              SCMP_SYS(add_key)             }, /* keyring is not namespaced */
  49                 { 0,              SCMP_SYS(afs_syscall)         }, /* obsolete syscall */
  50                 { 0,              SCMP_SYS(bdflush)             },
  51 #ifdef __NR_bpf
  52                 { 0,              SCMP_SYS(bpf)                 },
  53 #endif
  54                 { 0,              SCMP_SYS(break)               }, /* obsolete syscall */
  55                 { 0,              SCMP_SYS(create_module)       }, /* obsolete syscall */
  56                 { 0,              SCMP_SYS(ftime)               }, /* obsolete syscall */
  57                 { 0,              SCMP_SYS(get_kernel_syms)     }, /* obsolete syscall */
  58                 { 0,              SCMP_SYS(getpmsg)             }, /* obsolete syscall */
  59                 { 0,              SCMP_SYS(gtty)                }, /* obsolete syscall */
  60 #ifdef __NR_kexec_file_load
  61                 { 0,              SCMP_SYS(kexec_file_load)     },
  62 #endif
  63                 { 0,              SCMP_SYS(kexec_load)          },
  64                 { 0,              SCMP_SYS(keyctl)              }, /* keyring is not namespaced */
  65                 { 0,              SCMP_SYS(lock)                }, /* obsolete syscall */
  66                 { 0,              SCMP_SYS(lookup_dcookie)      },
  67                 { 0,              SCMP_SYS(mpx)                 }, /* obsolete syscall */
  68                 { 0,              SCMP_SYS(nfsservctl)          }, /* obsolete syscall */
  69                 { 0,              SCMP_SYS(open_by_handle_at)   },
  70                 { 0,              SCMP_SYS(perf_event_open)     },
  71                 { 0,              SCMP_SYS(prof)                }, /* obsolete syscall */
  72                 { 0,              SCMP_SYS(profil)              }, /* obsolete syscall */
  73                 { 0,              SCMP_SYS(putpmsg)             }, /* obsolete syscall */
  74                 { 0,              SCMP_SYS(query_module)        }, /* obsolete syscall */
  75                 { 0,              SCMP_SYS(quotactl)            },
  76                 { 0,              SCMP_SYS(request_key)         }, /* keyring is not namespaced */
  77                 { 0,              SCMP_SYS(security)            }, /* obsolete syscall */
  78                 { 0,              SCMP_SYS(sgetmask)            }, /* obsolete syscall */
  79                 { 0,              SCMP_SYS(ssetmask)            }, /* obsolete syscall */
  80                 { 0,              SCMP_SYS(stty)                }, /* obsolete syscall */
  81                 { 0,              SCMP_SYS(swapoff)             },
  82                 { 0,              SCMP_SYS(swapon)              },
  83                 { 0,              SCMP_SYS(sysfs)               }, /* obsolete syscall */
  84                 { 0,              SCMP_SYS(tuxcall)             }, /* obsolete syscall */
  85                 { 0,              SCMP_SYS(ulimit)              }, /* obsolete syscall */
  86                 { 0,              SCMP_SYS(uselib)              }, /* obsolete syscall */
  87                 { 0,              SCMP_SYS(ustat)               }, /* obsolete syscall */
  88                 { 0,              SCMP_SYS(vserver)             }, /* obsolete syscall */
  89                 { CAP_SYSLOG,     SCMP_SYS(syslog)              },
  90                 { CAP_SYS_MODULE, SCMP_SYS(delete_module)       },
  91                 { CAP_SYS_MODULE, SCMP_SYS(finit_module)        },
  92                 { CAP_SYS_MODULE, SCMP_SYS(init_module)         },
  93                 { CAP_SYS_PACCT,  SCMP_SYS(acct)                },
  94                 { CAP_SYS_PTRACE, SCMP_SYS(process_vm_readv)    },
  95                 { CAP_SYS_PTRACE, SCMP_SYS(process_vm_writev)   },
  96                 { CAP_SYS_PTRACE, SCMP_SYS(ptrace)              },
  97                 { CAP_SYS_RAWIO,  SCMP_SYS(ioperm)              },
  98                 { CAP_SYS_RAWIO,  SCMP_SYS(iopl)                },
  99                 { CAP_SYS_RAWIO,  SCMP_SYS(pciconfig_iobase)    },
 100                 { CAP_SYS_RAWIO,  SCMP_SYS(pciconfig_read)      },
 101                 { CAP_SYS_RAWIO,  SCMP_SYS(pciconfig_write)     },
 102 #ifdef __NR_s390_pci_mmio_read
 103                 { CAP_SYS_RAWIO,  SCMP_SYS(s390_pci_mmio_read)  },
 104 #endif
 105 #ifdef __NR_s390_pci_mmio_write
 106                 { CAP_SYS_RAWIO,  SCMP_SYS(s390_pci_mmio_write) },
 107 #endif
 108                 { CAP_SYS_TIME,   SCMP_SYS(adjtimex)            },
 109                 { CAP_SYS_TIME,   SCMP_SYS(clock_adjtime)       },
 110                 { CAP_SYS_TIME,   SCMP_SYS(clock_settime)       },
 111                 { CAP_SYS_TIME,   SCMP_SYS(settimeofday)        },
 112                 { CAP_SYS_TIME,   SCMP_SYS(stime)               },
 113         };
 114
 115         for (i = 0; i < ELEMENTSOF(blacklist); i++) {
 116                 if (blacklist[i].capability != 0 && (cap_list_retain & (1ULL << blacklist[i].capability)))
 117                         continue;
 118
 119                 r = seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), blacklist[i].syscall_num, 0);
 120                 if (r == -EFAULT)
 121                         continue; /* unknown syscall */
 122                 if (r < 0)
 123                         return log_error_errno(r, "Failed to block syscall: %m");
 124         }
 125
 126         return 0;
 127 }
 128
 129 int setup_seccomp(uint64_t cap_list_retain) {
 130         scmp_filter_ctx seccomp;
 131         int r;
 132
 133         seccomp = seccomp_init(SCMP_ACT_ALLOW);
 134         if (!seccomp)
 135                 return log_oom();
 136
 137         r = seccomp_add_secondary_archs(seccomp);
 138         if (r < 0) {
 139                 log_error_errno(r, "Failed to add secondary archs to seccomp filter: %m");
 140                 goto finish;
 141         }
 142
 143         r = seccomp_add_default_syscall_filter(seccomp, cap_list_retain);
 144         if (r < 0)
 145                 goto finish;
 146
 147         /*
 148            Audit is broken in containers, much of the userspace audit
 149            hookup will fail if running inside a container. We don't
 150            care and just turn off creation of audit sockets.
 151
 152            This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail
 153            with EAFNOSUPPORT which audit userspace uses as indication
 154            that audit is disabled in the kernel.
 155          */
 156
 157         r = seccomp_rule_add(
 158                         seccomp,
 159                         SCMP_ACT_ERRNO(EAFNOSUPPORT),
 160                         SCMP_SYS(socket),
 161                         2,
 162                         SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
 163                         SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
 164         if (r < 0) {
 165                 log_error_errno(r, "Failed to add audit seccomp rule: %m");
 166                 goto finish;
 167         }
 168
 169         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
 170         if (r < 0) {
 171                 log_error_errno(r, "Failed to unset NO_NEW_PRIVS: %m");
 172                 goto finish;
 173         }
 174
 175         r = seccomp_load(seccomp);
 176         if (r == -EINVAL) {
 177                 log_debug_errno(r, "Kernel is probably not configured with CONFIG_SECCOMP. Disabling seccomp audit filter: %m");
 178                 r = 0;
 179                 goto finish;
 180         }
 181         if (r < 0) {
 182                 log_error_errno(r, "Failed to install seccomp audit filter: %m");
 183                 goto finish;
 184         }
 185
 186 finish:
 187         seccomp_release(seccomp);
 188         return r;
 189 }
 190
 191 #else
 192
 193 int setup_seccomp(uint64_t cap_list_retain) {
 194         return 0;
 195 }
 196
 197 #endif