net/core: BUG in copy_net_ns()
From: zzoru
Date: Fri Jan 11 2019 - 13:07:44 EST
net/core: BUG in copy_net_ns() (net_namespace.c)
Hello,
I've got the following error report while fuzzing the kernel with syzkaller.
On commit 1bdbe227492075d058e37cb3d400e6468d0095b5
Syzkaller hit 'WARNING in __alloc_pages_slowpath' bug.
syz-executor561 (17453) used greatest stack depth: 25056 bytes left
WARNING: CPU: 0 PID: 692 at mm/page_alloc.c:4415
__alloc_pages_slowpath+0x1cb1/0x2220 mm/page_alloc.c:4386
Kernel panic - not syncing: panic_on_warn set ...
CPU: 0 PID: 692 Comm: kswapd0 Not tainted 5.0.0-rc1+ #4
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
Ubuntu-1.8.2-1ubuntu1 04/01/2014
Call Trace:
Â__dump_stack lib/dump_stack.c:77 [inline]
Âdump_stack+0xca/0x13e lib/dump_stack.c:113
Âpanic+0x278/0x5bf kernel/panic.c:214
Â__warn.cold.10+0x20/0x45 kernel/panic.c:571
Âreport_bug+0x246/0x2d0 lib/bug.c:186
Âfixup_bug arch/x86/kernel/traps.c:178 [inline]
Âdo_error_trap+0x123/0x1e0 arch/x86/kernel/traps.c:271
Âdo_invalid_op+0x31/0x40 arch/x86/kernel/traps.c:290
Âinvalid_op+0x14/0x20 arch/x86/entry/entry_64.S:973
RIP: 0010:__alloc_pages_slowpath+0x1cb1/0x2220 mm/page_alloc.c:4415
Code: 8b 84 24 a8 00 00 00 e9 ea f1 ff ff 85 d2 0f 85 0b 01 00 00 48 c7
c7 c0 5e 55 84 e8 79 f8 23 02 e9 86 f9 ff ff 44 8b 74 24 0c <0f> 0b 48
b8 00 00 00 00 00 fc ff df 48 8b 54 24 18 48 c1 ea 03 80
RSP: 0018:ffff8880683fedb8 EFLAGS: 00010046
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 1ffff1100d07fda4
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88807ffdd528
RBP: dffffc0000000000 R08: 0000000000000000 R09: 000000000000067a
R10: 0000000000000000 R11: ffff88807ffdc487 R12: 0000000000000000
R13: ffff8880683ff010 R14: 0000000000415a00 R15: ffff8880683ff010
Â__alloc_pages_nodemask+0x521/0x5f0 mm/page_alloc.c:4555
Â__alloc_pages include/linux/gfp.h:473 [inline]
Â__alloc_pages_node include/linux/gfp.h:486 [inline]
Âkmem_getpages mm/slab.c:1398 [inline]
Âcache_grow_begin+0x95/0x300 mm/slab.c:2666
Âfallback_alloc+0x1ce/0x270 mm/slab.c:3208
Â__do_cache_alloc mm/slab.c:3345 [inline]
Âslab_alloc mm/slab.c:3373 [inline]
Âkmem_cache_alloc+0x286/0x2f0 mm/slab.c:3541
Âcreate_object+0x83/0x880 mm/kmemleak.c:578
Âkmemleak_alloc_recursive include/linux/kmemleak.h:55 [inline]
Âslab_post_alloc_hook mm/slab.h:442 [inline]
Âslab_alloc mm/slab.c:3381 [inline]
Âkmem_cache_alloc+0x18f/0x2f0 mm/slab.c:3541
Âmempool_alloc+0x13e/0x340 mm/mempool.c:385
Âbio_alloc_bioset+0x36f/0x5d0 block/bio.c:489
Âbio_alloc include/linux/bio.h:393 [inline]
Âsubmit_bh_wbc.isra.57+0x128/0x680 fs/buffer.c:3061
Â__block_write_full_page+0x6e8/0xcd0 fs/buffer.c:1765
Âblock_write_full_page+0x202/0x250 fs/buffer.c:2955
Âpageout mm/vmscan.c:865 [inline]
Âshrink_page_list+0x220f/0x3800 mm/vmscan.c:1383
Âshrink_inactive_list+0x3c2/0xaa0 mm/vmscan.c:1961
Âshrink_list mm/vmscan.c:2273 [inline]
Âshrink_node_memcg.constprop.83+0x4bf/0x10e0 mm/vmscan.c:2538
Âshrink_node+0x162/0xd10 mm/vmscan.c:2753
Âkswapd_shrink_node mm/vmscan.c:3516 [inline]
Âbalance_pgdat+0x47f/0xc00 mm/vmscan.c:3674
Âkswapd+0x57c/0xde0 mm/vmscan.c:3929
Âkthread+0x347/0x410 kernel/kthread.c:246
Âret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:352
Dumping ftrace buffer:
ÂÂ (ftrace buffer empty)
Kernel Offset: disabled
Rebooting in 86400 seconds..
Syzkaller reproducer:
# {Threaded:false Collide:false Repeat:true RepeatTimes:0 Procs:8
Sandbox:none Fault:false FaultCall:-1 FaultNth:0 EnableTun:false
UseTmpDir:true EnableCgroups:false EnableNetdev:true ResetNet:false
HandleSegv:false Repro:false Trace:false}
unshare(0x40000000)
C reproducer:
// autogenerated by syzkaller (https://github.com/google/syzkaller)
#define _GNU_SOURCE
#include <arpa/inet.h>
#include <dirent.h>
#include <endian.h>
#include <errno.h>
#include <fcntl.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <netinet/in.h>
#include <sched.h>
#include <signal.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/mount.h>
#include <sys/prctl.h>
#include <sys/resource.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/syscall.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/uio.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>
#include <linux/if_addr.h>
#include <linux/if_ether.h>
#include <linux/if_link.h>
#include <linux/if_tun.h>
#include <linux/in6.h>
#include <linux/ip.h>
#include <linux/neighbour.h>
#include <linux/net.h>
#include <linux/netlink.h>
#include <linux/rtnetlink.h>
#include <linux/tcp.h>
#include <linux/veth.h>
unsigned long long procid;
static void sleep_ms(uint64_t ms)
{
 usleep(ms * 1000);
}
static uint64_t current_time_ms(void)
{
 struct timespec ts;
 if (clock_gettime(CLOCK_MONOTONIC, &ts))
ÂÂÂ exit(1);
 return (uint64_t)ts.tv_sec * 1000 + (uint64_t)ts.tv_nsec / 1000000;
}
static void use_temporary_dir(void)
{
 char tmpdir_template[] = "./syzkaller.XXXXXX";
 char* tmpdir = mkdtemp(tmpdir_template);
 if (!tmpdir)
ÂÂÂ exit(1);
 if (chmod(tmpdir, 0777))
ÂÂÂ exit(1);
 if (chdir(tmpdir))
ÂÂÂ exit(1);
}
static bool write_file(const char* file, const char* what, ...)
{
 char buf[1024];
 va_list args;
 va_start(args, what);
 vsnprintf(buf, sizeof(buf), what, args);
 va_end(args);
 buf[sizeof(buf) - 1] = 0;
 int len = strlen(buf);
 int fd = open(file, O_WRONLY | O_CLOEXEC);
 if (fd == -1)
ÂÂÂ return false;
 if (write(fd, buf, len) != len) {
ÂÂÂ int err = errno;
ÂÂÂ close(fd);
ÂÂÂ errno = err;
ÂÂÂ return false;
 }
 close(fd);
 return true;
}
static struct {
 char* pos;
 int nesting;
 struct nlattr* nested[8];
 char buf[1024];
} nlmsg;
static void netlink_init(int typ, int flags, const void* data, int size)
{
 memset(&nlmsg, 0, sizeof(nlmsg));
 struct nlmsghdr* hdr = (struct nlmsghdr*)nlmsg.buf;
 hdr->nlmsg_type = typ;
 hdr->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | flags;
 memcpy(hdr + 1, data, size);
 nlmsg.pos = (char*)(hdr + 1) + NLMSG_ALIGN(size);
}
static void netlink_attr(int typ, const void* data, int size)
{
 struct nlattr* attr = (struct nlattr*)nlmsg.pos;
 attr->nla_len = sizeof(*attr) + size;
 attr->nla_type = typ;
 memcpy(attr + 1, data, size);
 nlmsg.pos += NLMSG_ALIGN(attr->nla_len);
}
static void netlink_nest(int typ)
{
 struct nlattr* attr = (struct nlattr*)nlmsg.pos;
 attr->nla_type = typ;
 nlmsg.pos += sizeof(*attr);
 nlmsg.nested[nlmsg.nesting++] = attr;
}
static void netlink_done(void)
{
 struct nlattr* attr = nlmsg.nested[--nlmsg.nesting];
 attr->nla_len = nlmsg.pos - (char*)attr;
}
static int netlink_send(int sock)
{
 if (nlmsg.pos > nlmsg.buf + sizeof(nlmsg.buf) || nlmsg.nesting)
ÂÂÂ exit(1);
 struct nlmsghdr* hdr = (struct nlmsghdr*)nlmsg.buf;
 hdr->nlmsg_len = nlmsg.pos - nlmsg.buf;
 struct sockaddr_nl addr;
 memset(&addr, 0, sizeof(addr));
 addr.nl_family = AF_NETLINK;
 unsigned n = sendto(sock, nlmsg.buf, hdr->nlmsg_len, 0,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ (struct sockaddr*)&addr, sizeof(addr));
 if (n != hdr->nlmsg_len)
ÂÂÂ exit(1);
 n = recv(sock, nlmsg.buf, sizeof(nlmsg.buf), 0);
 if (n < sizeof(struct nlmsghdr) + sizeof(struct nlmsgerr))
ÂÂÂ exit(1);
 if (hdr->nlmsg_type != NLMSG_ERROR)
ÂÂÂ exit(1);
 return -((struct nlmsgerr*)(hdr + 1))->error;
}
static void netlink_add_device_impl(const char* type, const char* name)
{
 struct ifinfomsg hdr;
 memset(&hdr, 0, sizeof(hdr));
 netlink_init(RTM_NEWLINK, NLM_F_EXCL | NLM_F_CREATE, &hdr, sizeof(hdr));
 if (name)
ÂÂÂ netlink_attr(IFLA_IFNAME, name, strlen(name));
 netlink_nest(IFLA_LINKINFO);
 netlink_attr(IFLA_INFO_KIND, type, strlen(type));
}
static void netlink_add_device(int sock, const char* type, const char* name)
{
 netlink_add_device_impl(type, name);
 netlink_done();
 int err = netlink_send(sock);
 (void)err;
}
static void netlink_add_veth(int sock, const char* name, const char* peer)
{
 netlink_add_device_impl("veth", name);
 netlink_nest(IFLA_INFO_DATA);
 netlink_nest(VETH_INFO_PEER);
 nlmsg.pos += sizeof(struct ifinfomsg);
 netlink_attr(IFLA_IFNAME, peer, strlen(peer));
 netlink_done();
 netlink_done();
 netlink_done();
 int err = netlink_send(sock);
 (void)err;
}
static void netlink_add_hsr(int sock, const char* name, const char* slave1,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ const char* slave2)
{
 netlink_add_device_impl("hsr", name);
 netlink_nest(IFLA_INFO_DATA);
 int ifindex1 = if_nametoindex(slave1);
 netlink_attr(IFLA_HSR_SLAVE1, &ifindex1, sizeof(ifindex1));
 int ifindex2 = if_nametoindex(slave2);
 netlink_attr(IFLA_HSR_SLAVE2, &ifindex2, sizeof(ifindex2));
 netlink_done();
 netlink_done();
 int err = netlink_send(sock);
 (void)err;
}
static void netlink_device_change(int sock, const char* name, bool up,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ const char* master, const void* mac,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ int macsize)
{
 struct ifinfomsg hdr;
 memset(&hdr, 0, sizeof(hdr));
 if (up)
ÂÂÂ hdr.ifi_flags = hdr.ifi_change = IFF_UP;
 netlink_init(RTM_NEWLINK, 0, &hdr, sizeof(hdr));
 netlink_attr(IFLA_IFNAME, name, strlen(name));
 if (master) {
ÂÂÂ int ifindex = if_nametoindex(master);
ÂÂÂ netlink_attr(IFLA_MASTER, &ifindex, sizeof(ifindex));
 }
 if (macsize)
ÂÂÂ netlink_attr(IFLA_ADDRESS, mac, macsize);
 int err = netlink_send(sock);
 (void)err;
}
static int netlink_add_addr(int sock, const char* dev, const void* addr,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ int addrsize)
{
 struct ifaddrmsg hdr;
 memset(&hdr, 0, sizeof(hdr));
 hdr.ifa_family = addrsize == 4 ? AF_INET : AF_INET6;
 hdr.ifa_prefixlen = addrsize == 4 ? 24 : 120;
 hdr.ifa_scope = RT_SCOPE_UNIVERSE;
 hdr.ifa_index = if_nametoindex(dev);
 netlink_init(RTM_NEWADDR, NLM_F_CREATE | NLM_F_REPLACE, &hdr,
sizeof(hdr));
 netlink_attr(IFA_LOCAL, addr, addrsize);
 netlink_attr(IFA_ADDRESS, addr, addrsize);
 return netlink_send(sock);
}
static void netlink_add_addr4(int sock, const char* dev, const char* addr)
{
 struct in_addr in_addr;
 inet_pton(AF_INET, addr, &in_addr);
 int err = netlink_add_addr(sock, dev, &in_addr, sizeof(in_addr));
 (void)err;
}
static void netlink_add_addr6(int sock, const char* dev, const char* addr)
{
 struct in6_addr in6_addr;
 inet_pton(AF_INET6, addr, &in6_addr);
 int err = netlink_add_addr(sock, dev, &in6_addr, sizeof(in6_addr));
 (void)err;
}
#define DEV_IPV4 "172.20.20.%d"
#define DEV_IPV6 "fe80::%02hx"
#define DEV_MAC 0x00aaaaaaaaaa
static void initialize_netdevices(void)
{
 char netdevsim[16];
 sprintf(netdevsim, "netdevsim%d", (int)procid);
 struct {
ÂÂÂ const char* type;
ÂÂÂ const char* dev;
 } devtypes[] = {
ÂÂÂÂÂ {"ip6gretap", "ip6gretap0"}, {"bridge", "bridge0"},
ÂÂÂÂÂ {"vcan", "vcan0"},ÂÂÂÂÂÂÂÂÂÂ {"bond", "bond0"},
ÂÂÂÂÂ {"team", "team0"},ÂÂÂÂÂÂÂÂÂÂ {"dummy", "dummy0"},
ÂÂÂÂÂ {"nlmon", "nlmon0"},ÂÂÂÂÂÂÂÂ {"caif", "caif0"},
ÂÂÂÂÂ {"batadv", "batadv0"},ÂÂÂÂÂÂ {"vxcan", "vxcan1"},
ÂÂÂÂÂ {"netdevsim", netdevsim},ÂÂÂ {"veth", 0},
 };
 const char* devmasters[] = {"bridge", "bond", "team"};
 struct {
ÂÂÂ const char* name;
ÂÂÂ int macsize;
ÂÂÂ bool noipv6;
 } devices[] = {
ÂÂÂÂÂ {"lo", ETH_ALEN},
ÂÂÂÂÂ {"sit0", 0},
ÂÂÂÂÂ {"bridge0", ETH_ALEN},
ÂÂÂÂÂ {"vcan0", 0, true},
ÂÂÂÂÂ {"tunl0", 0},
ÂÂÂÂÂ {"gre0", 0},
ÂÂÂÂÂ {"gretap0", ETH_ALEN},
ÂÂÂÂÂ {"ip_vti0", 0},
ÂÂÂÂÂ {"ip6_vti0", 0},
ÂÂÂÂÂ {"ip6tnl0", 0},
ÂÂÂÂÂ {"ip6gre0", 0},
ÂÂÂÂÂ {"ip6gretap0", ETH_ALEN},
ÂÂÂÂÂ {"erspan0", ETH_ALEN},
ÂÂÂÂÂ {"bond0", ETH_ALEN},
ÂÂÂÂÂ {"veth0", ETH_ALEN},
ÂÂÂÂÂ {"veth1", ETH_ALEN},
ÂÂÂÂÂ {"team0", ETH_ALEN},
ÂÂÂÂÂ {"veth0_to_bridge", ETH_ALEN},
ÂÂÂÂÂ {"veth1_to_bridge", ETH_ALEN},
ÂÂÂÂÂ {"veth0_to_bond", ETH_ALEN},
ÂÂÂÂÂ {"veth1_to_bond", ETH_ALEN},
ÂÂÂÂÂ {"veth0_to_team", ETH_ALEN},
ÂÂÂÂÂ {"veth1_to_team", ETH_ALEN},
ÂÂÂÂÂ {"veth0_to_hsr", ETH_ALEN},
ÂÂÂÂÂ {"veth1_to_hsr", ETH_ALEN},
ÂÂÂÂÂ {"hsr0", 0},
ÂÂÂÂÂ {"dummy0", ETH_ALEN},
ÂÂÂÂÂ {"nlmon0", 0},
ÂÂÂÂÂ {"vxcan1", 0, true},
ÂÂÂÂÂ {"caif0", ETH_ALEN},
ÂÂÂÂÂ {"batadv0", ETH_ALEN},
ÂÂÂÂÂ {netdevsim, ETH_ALEN},
 };
 int sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
 if (sock == -1)
ÂÂÂ exit(1);
 unsigned i;
 for (i = 0; i < sizeof(devtypes) / sizeof(devtypes[0]); i++)
ÂÂÂ netlink_add_device(sock, devtypes[i].type, devtypes[i].dev);
 for (i = 0; i < sizeof(devmasters) / (sizeof(devmasters[0])); i++) {
ÂÂÂ char master[32], slave0[32], veth0[32], slave1[32], veth1[32];
ÂÂÂ sprintf(slave0, "%s_slave_0", devmasters[i]);
ÂÂÂ sprintf(veth0, "veth0_to_%s", devmasters[i]);
ÂÂÂ netlink_add_veth(sock, slave0, veth0);
ÂÂÂ sprintf(slave1, "%s_slave_1", devmasters[i]);
ÂÂÂ sprintf(veth1, "veth1_to_%s", devmasters[i]);
ÂÂÂ netlink_add_veth(sock, slave1, veth1);
ÂÂÂ sprintf(master, "%s0", devmasters[i]);
ÂÂÂ netlink_device_change(sock, slave0, false, master, 0, 0);
ÂÂÂ netlink_device_change(sock, slave1, false, master, 0, 0);
 }
 netlink_device_change(sock, "bridge_slave_0", true, 0, 0, 0);
 netlink_device_change(sock, "bridge_slave_1", true, 0, 0, 0);
 netlink_add_veth(sock, "hsr_slave_0", "veth0_to_hsr");
 netlink_add_veth(sock, "hsr_slave_1", "veth1_to_hsr");
 netlink_add_hsr(sock, "hsr0", "hsr_slave_0", "hsr_slave_1");
 netlink_device_change(sock, "hsr_slave_0", true, 0, 0, 0);
 netlink_device_change(sock, "hsr_slave_1", true, 0, 0, 0);
 for (i = 0; i < sizeof(devices) / (sizeof(devices[0])); i++) {
ÂÂÂ char addr[32];
ÂÂÂ sprintf(addr, DEV_IPV4, i + 10);
ÂÂÂ netlink_add_addr4(sock, devices[i].name, addr);
ÂÂÂ if (!devices[i].noipv6) {
ÂÂÂÂÂ sprintf(addr, DEV_IPV6, i + 10);
ÂÂÂÂÂ netlink_add_addr6(sock, devices[i].name, addr);
ÂÂÂ }
ÂÂÂ uint64_t macaddr = DEV_MAC + ((i + 10ull) << 40);
ÂÂÂ netlink_device_change(sock, devices[i].name, true, 0, &macaddr,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ devices[i].macsize);
 }
 close(sock);
}
static void initialize_netdevices_init(void)
{
 int sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
 if (sock == -1)
ÂÂÂ exit(1);
 struct {
ÂÂÂ const char* type;
ÂÂÂ int macsize;
ÂÂÂ bool noipv6;
ÂÂÂ bool noup;
 } devtypes[] = {
ÂÂÂÂÂ {"nr", 7, true}, {"rose", 5, true, true},
 };
 unsigned i;
 for (i = 0; i < sizeof(devtypes) / sizeof(devtypes[0]); i++) {
ÂÂÂ char dev[32], addr[32];
ÂÂÂ sprintf(dev, "%s%d", devtypes[i].type, (int)procid);
ÂÂÂ sprintf(addr, "172.30.%d.%d", i, (int)procid + 1);
ÂÂÂ netlink_add_addr4(sock, dev, addr);
ÂÂÂ if (!devtypes[i].noipv6) {
ÂÂÂÂÂ sprintf(addr, "fe88::%02hx:%02hx", i, (int)procid + 1);
ÂÂÂÂÂ netlink_add_addr6(sock, dev, addr);
ÂÂÂ }
ÂÂÂ int macsize = devtypes[i].macsize;
ÂÂÂ uint64_t macaddr = 0xbbbbbb +
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ ((unsigned long long)i << (8 * (macsize - 2))) +
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ (procid << (8 * (macsize - 1)));
ÂÂÂ netlink_device_change(sock, dev, !devtypes[i].noup, 0, &macaddr,
macsize);
 }
 close(sock);
}
static void setup_common()
{
 if (mount(0, "/sys/fs/fuse/connections", "fusectl", 0, 0)) {
 }
}
static void loop();
static void sandbox_common()
{
 prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0);
 setpgrp();
 setsid();
 struct rlimit rlim;
 rlim.rlim_cur = rlim.rlim_max = 200 << 20;
 setrlimit(RLIMIT_AS, &rlim);
 rlim.rlim_cur = rlim.rlim_max = 32 << 20;
 setrlimit(RLIMIT_MEMLOCK, &rlim);
 rlim.rlim_cur = rlim.rlim_max = 136 << 20;
 setrlimit(RLIMIT_FSIZE, &rlim);
 rlim.rlim_cur = rlim.rlim_max = 1 << 20;
 setrlimit(RLIMIT_STACK, &rlim);
 rlim.rlim_cur = rlim.rlim_max = 0;
 setrlimit(RLIMIT_CORE, &rlim);
 rlim.rlim_cur = rlim.rlim_max = 256;
 setrlimit(RLIMIT_NOFILE, &rlim);
 if (unshare(CLONE_NEWNS)) {
 }
 if (unshare(CLONE_NEWIPC)) {
 }
 if (unshare(0x02000000)) {
 }
 if (unshare(CLONE_NEWUTS)) {
 }
 if (unshare(CLONE_SYSVSEM)) {
 }
 typedef struct {
ÂÂÂ const char* name;
ÂÂÂ const char* value;
 } sysctl_t;
 static const sysctl_t sysctls[] = {
ÂÂÂÂÂ {"/proc/sys/kernel/shmmax", "16777216"},
ÂÂÂÂÂ {"/proc/sys/kernel/shmall", "536870912"},
ÂÂÂÂÂ {"/proc/sys/kernel/shmmni", "1024"},
ÂÂÂÂÂ {"/proc/sys/kernel/msgmax", "8192"},
ÂÂÂÂÂ {"/proc/sys/kernel/msgmni", "1024"},
ÂÂÂÂÂ {"/proc/sys/kernel/msgmnb", "1024"},
ÂÂÂÂÂ {"/proc/sys/kernel/sem", "1024 1048576 500 1024"},
 };
 unsigned i;
 for (i = 0; i < sizeof(sysctls) / sizeof(sysctls[0]); i++)
ÂÂÂ write_file(sysctls[i].name, sysctls[i].value);
}
int wait_for_loop(int pid)
{
 if (pid < 0)
ÂÂÂ exit(1);
 int status = 0;
 while (waitpid(-1, &status, __WALL) != pid) {
 }
 return WEXITSTATUS(status);
}
static int do_sandbox_none(void)
{
 if (unshare(CLONE_NEWPID)) {
 }
 int pid = fork();
 if (pid != 0)
ÂÂÂ return wait_for_loop(pid);
 setup_common();
 sandbox_common();
 initialize_netdevices_init();
 if (unshare(CLONE_NEWNET)) {
 }
 initialize_netdevices();
 loop();
 exit(1);
}
#define FS_IOC_SETFLAGS _IOW('f', 2, long)
static void remove_dir(const char* dir)
{
 DIR* dp;
 struct dirent* ep;
 int iter = 0;
retry:
 while (umount2(dir, MNT_DETACH) == 0) {
 }
 dp = opendir(dir);
 if (dp == NULL) {
ÂÂÂ if (errno == EMFILE) {
ÂÂÂÂÂ exit(1);
ÂÂÂ }
ÂÂÂ exit(1);
 }
 while ((ep = readdir(dp))) {
ÂÂÂ if (strcmp(ep->d_name, ".") == 0 || strcmp(ep->d_name, "..") == 0)
ÂÂÂÂÂ continue;
ÂÂÂ char filename[FILENAME_MAX];
ÂÂÂ snprintf(filename, sizeof(filename), "%s/%s", dir, ep->d_name);
ÂÂÂ while (umount2(filename, MNT_DETACH) == 0) {
ÂÂÂ }
ÂÂÂ struct stat st;
ÂÂÂ if (lstat(filename, &st))
ÂÂÂÂÂ exit(1);
ÂÂÂ if (S_ISDIR(st.st_mode)) {
ÂÂÂÂÂ remove_dir(filename);
ÂÂÂÂÂ continue;
ÂÂÂ }
ÂÂÂ int i;
ÂÂÂ for (i = 0;; i++) {
ÂÂÂÂÂ if (unlink(filename) == 0)
ÂÂÂÂÂÂÂ break;
ÂÂÂÂÂ if (errno == EPERM) {
ÂÂÂÂÂÂÂ int fd = open(filename, O_RDONLY);
ÂÂÂÂÂÂÂ if (fd != -1) {
ÂÂÂÂÂÂÂÂÂ long flags = 0;
ÂÂÂÂÂÂÂÂÂ if (ioctl(fd, FS_IOC_SETFLAGS, &flags) == 0)
ÂÂÂÂÂÂÂÂÂÂÂ close(fd);
ÂÂÂÂÂÂÂÂÂ continue;
ÂÂÂÂÂÂÂ }
ÂÂÂÂÂ }
ÂÂÂÂÂ if (errno == EROFS) {
ÂÂÂÂÂÂÂ break;
ÂÂÂÂÂ }
ÂÂÂÂÂ if (errno != EBUSY || i > 100)
ÂÂÂÂÂÂÂ exit(1);
ÂÂÂÂÂ if (umount2(filename, MNT_DETACH))
ÂÂÂÂÂÂÂ exit(1);
ÂÂÂ }
 }
 closedir(dp);
 int i;
 for (i = 0;; i++) {
ÂÂÂ if (rmdir(dir) == 0)
ÂÂÂÂÂ break;
ÂÂÂ if (i < 100) {
ÂÂÂÂÂ if (errno == EPERM) {
ÂÂÂÂÂÂÂ int fd = open(dir, O_RDONLY);
ÂÂÂÂÂÂÂ if (fd != -1) {
ÂÂÂÂÂÂÂÂÂ long flags = 0;
ÂÂÂÂÂÂÂÂÂ if (ioctl(fd, FS_IOC_SETFLAGS, &flags) == 0)
ÂÂÂÂÂÂÂÂÂÂÂ close(fd);
ÂÂÂÂÂÂÂÂÂ continue;
ÂÂÂÂÂÂÂ }
ÂÂÂÂÂ }
ÂÂÂÂÂ if (errno == EROFS) {
ÂÂÂÂÂÂÂ break;
ÂÂÂÂÂ }
ÂÂÂÂÂ if (errno == EBUSY) {
ÂÂÂÂÂÂÂ if (umount2(dir, MNT_DETACH))
ÂÂÂÂÂÂÂÂÂ exit(1);
ÂÂÂÂÂÂÂ continue;
ÂÂÂÂÂ }
ÂÂÂÂÂ if (errno == ENOTEMPTY) {
ÂÂÂÂÂÂÂ if (iter < 100) {
ÂÂÂÂÂÂÂÂÂ iter++;
ÂÂÂÂÂÂÂÂÂ goto retry;
ÂÂÂÂÂÂÂ }
ÂÂÂÂÂ }
ÂÂÂ }
ÂÂÂ exit(1);
 }
}
static void kill_and_wait(int pid, int* status)
{
 kill(-pid, SIGKILL);
 kill(pid, SIGKILL);
 int i;
 for (i = 0; i < 100; i++) {
ÂÂÂ if (waitpid(-1, status, WNOHANG | __WALL) == pid)
ÂÂÂÂÂ return;
ÂÂÂ usleep(1000);
 }
 DIR* dir = opendir("/sys/fs/fuse/connections");
 if (dir) {
ÂÂÂ for (;;) {
ÂÂÂÂÂ struct dirent* ent = readdir(dir);
ÂÂÂÂÂ if (!ent)
ÂÂÂÂÂÂÂ break;
ÂÂÂÂÂ if (strcmp(ent->d_name, ".") == 0 || strcmp(ent->d_name, "..") == 0)
ÂÂÂÂÂÂÂ continue;
ÂÂÂÂÂ char abort[300];
ÂÂÂÂÂ snprintf(abort, sizeof(abort), "/sys/fs/fuse/connections/%s/abort",
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂ ent->d_name);
ÂÂÂÂÂ int fd = open(abort, O_WRONLY);
ÂÂÂÂÂ if (fd == -1) {
ÂÂÂÂÂÂÂ continue;
ÂÂÂÂÂ }
ÂÂÂÂÂ if (write(fd, abort, 1) < 0) {
ÂÂÂÂÂ }
ÂÂÂÂÂ close(fd);
ÂÂÂ }
ÂÂÂ closedir(dir);
 } else {
 }
 while (waitpid(-1, status, __WALL) != pid) {
 }
}
#define SYZ_HAVE_SETUP_TEST 1
static void setup_test()
{
 prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0);
 setpgrp();
}
#define SYZ_HAVE_RESET_TEST 1
static void reset_test()
{
 int fd;
 for (fd = 3; fd < 30; fd++)
ÂÂÂ close(fd);
}
static void execute_one(void);
#define WAIT_FLAGS __WALL
static void loop(void)
{
 int iter;
 for (iter = 0;; iter++) {
ÂÂÂ char cwdbuf[32];
ÂÂÂ sprintf(cwdbuf, "./%d", iter);
ÂÂÂ if (mkdir(cwdbuf, 0777))
ÂÂÂÂÂ exit(1);
ÂÂÂ int pid = fork();
ÂÂÂ if (pid < 0)
ÂÂÂÂÂ exit(1);
ÂÂÂ if (pid == 0) {
ÂÂÂÂÂ if (chdir(cwdbuf))
ÂÂÂÂÂÂÂ exit(1);
ÂÂÂÂÂ setup_test();
ÂÂÂÂÂ execute_one();
ÂÂÂÂÂ reset_test();
ÂÂÂÂÂ exit(0);
ÂÂÂ }
ÂÂÂ int status = 0;
ÂÂÂ uint64_t start = current_time_ms();
ÂÂÂ for (;;) {
ÂÂÂÂÂ if (waitpid(-1, &status, WNOHANG | WAIT_FLAGS) == pid)
ÂÂÂÂÂÂÂ break;
ÂÂÂÂÂ sleep_ms(1);
ÂÂÂÂÂ if (current_time_ms() - start < 5 * 1000)
ÂÂÂÂÂÂÂ continue;
ÂÂÂÂÂ kill_and_wait(pid, &status);
ÂÂÂÂÂ break;
ÂÂÂ }
ÂÂÂ remove_dir(cwdbuf);
 }
}
void execute_one(void)
{
 syscall(__NR_unshare, 0x40000000);
}
int main(void)
{
 syscall(__NR_mmap, 0x20000000, 0x1000000, 3, 0x32, -1, 0);
 for (procid = 0; procid < 8; procid++) {
ÂÂÂ if (fork() == 0) {
ÂÂÂÂÂ use_temporary_dir();
ÂÂÂÂÂ do_sandbox_none();
ÂÂÂ }
 }
 sleep(1000000);
 return 0;
}
I reviewed kernel code and found a bug that
net_drop_ns func doesn't call net_free func when refcount_dec_and_test's
return value is zero.
or
when rv = down_read_killable(&pernet_ops_rwsem) < 0, it doesn't need to
call refcount_dec_and_test.
https://github.com/torvalds/linux/commit/5ba049a5cc8e24a1643df75bbf65b4efa070fa74#diff-9312644e2968a45510bacdd2b2872ad2
(I can't reproduce this bug on v4.15 , and
1bdbe227492075d058e37cb3d400e6468d0095b5 with my patch. Because of the
previous version of kernel doesn't have this bug.)
This bug can lead to memory leak or DOS.
I made a patch for this bug. (just revert to a before commit)
diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index b02fb19df2cc..9de0ade14956 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -431,15 +431,18 @@ struct net *copy_net_ns(unsigned long flags,
ÂÂÂÂÂÂÂ get_user_ns(user_ns);
ÂÂÂÂÂÂÂ rv = down_read_killable(&pernet_ops_rwsem);
-ÂÂÂÂÂÂ if (rv < 0)
-ÂÂÂÂÂÂÂÂÂÂÂÂÂÂ goto put_userns;
+ÂÂÂÂÂÂ if (rv < 0){
+ÂÂÂÂÂÂÂ net_free(net);
+ÂÂÂÂÂÂÂ dec_net_namespaces(ucounts);
+ÂÂÂÂÂÂÂ put_user_ns(user_ns);
+ÂÂÂÂÂÂÂ return ERR_PTR(rv);
+ÂÂÂ }
ÂÂÂÂÂÂÂ rv = setup_net(net, user_ns);
ÂÂÂÂÂÂÂ up_read(&pernet_ops_rwsem);
ÂÂÂÂÂÂÂ if (rv < 0) {
-put_userns:
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ put_user_ns(user_ns);
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ net_drop_ns(net);
Âdec_ucounts:
and, sorry for my encrypted mails.
Attachment:
signature.asc
Description: OpenPGP digital signature