releases/4.19.29/proc-fix-proc-net-after-setns-2.patch

   1 From b5eedf57e9d482f4426b037bc4c10379c10fb05d Mon Sep 17 00:00:00 2001
   2 From: Alexey Dobriyan <adobriyan@gmail.com>
   3 Date: Fri, 1 Feb 2019 14:20:01 -0800
   4 Subject: proc: fix /proc/net/* after setns(2)
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 [ Upstream commit 1fde6f21d90f8ba5da3cb9c54ca991ed72696c43 ]
  10
  11 /proc entries under /proc/net/* can't be cached into dcache because
  12 setns(2) can change current net namespace.
  13
  14 [akpm@linux-foundation.org: coding-style fixes]
  15 [akpm@linux-foundation.org: avoid vim miscolorization]
  16 [adobriyan@gmail.com: write test, add dummy ->d_revalidate hook: necessary if /proc/net/* is pinned at setns time]
  17   Link: http://lkml.kernel.org/r/20190108192350.GA12034@avx2
  18 Link: http://lkml.kernel.org/r/20190107162336.GA9239@avx2
  19 Fixes: 1da4d377f943fe4194ffb9fb9c26cc58fad4dd24 ("proc: revalidate misc dentries")
  20 Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
  21 Reported-by: Mateusz Stępień <mateusz.stepien@netrounds.com>
  22 Reported-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
  23 Cc: Al Viro <viro@zeniv.linux.org.uk>
  24 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  25 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  26 Signed-off-by: Sasha Levin <sashal@kernel.org>
  27 ---
  28  fs/proc/generic.c                           |   4 +-
  29  fs/proc/internal.h                          |   1 +
  30  fs/proc/proc_net.c                          |  20 +++
  31  tools/testing/selftests/proc/.gitignore     |   1 +
  32  tools/testing/selftests/proc/Makefile       |   1 +
  33  tools/testing/selftests/proc/setns-dcache.c | 129 ++++++++++++++++++++
  34  6 files changed, 155 insertions(+), 1 deletion(-)
  35  create mode 100644 tools/testing/selftests/proc/setns-dcache.c
  36
  37 diff --git a/fs/proc/generic.c b/fs/proc/generic.c
  38 index 8ae109429a88..e39bac94dead 100644
  39 --- a/fs/proc/generic.c
  40 +++ b/fs/proc/generic.c
  41 @@ -256,7 +256,7 @@ struct dentry *proc_lookup_de(struct inode *dir, struct dentry *dentry,
  42                 inode = proc_get_inode(dir->i_sb, de);
  43                 if (!inode)
  44                         return ERR_PTR(-ENOMEM);
  45 -               d_set_d_op(dentry, &proc_misc_dentry_ops);
  46 +               d_set_d_op(dentry, de->proc_dops);
  47                 return d_splice_alias(inode, dentry);
  48         }
  49         read_unlock(&proc_subdir_lock);
  50 @@ -429,6 +429,8 @@ static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
  51         INIT_LIST_HEAD(&ent->pde_openers);
  52         proc_set_user(ent, (*parent)->uid, (*parent)->gid);
  53
  54 +       ent->proc_dops = &proc_misc_dentry_ops;
  55 +
  56  out:
  57         return ent;
  58  }
  59 diff --git a/fs/proc/internal.h b/fs/proc/internal.h
  60 index 5185d7f6a51e..95b14196f284 100644
  61 --- a/fs/proc/internal.h
  62 +++ b/fs/proc/internal.h
  63 @@ -44,6 +44,7 @@ struct proc_dir_entry {
  64         struct completion *pde_unload_completion;
  65         const struct inode_operations *proc_iops;
  66         const struct file_operations *proc_fops;
  67 +       const struct dentry_operations *proc_dops;
  68         union {
  69                 const struct seq_operations *seq_ops;
  70                 int (*single_show)(struct seq_file *, void *);
  71 diff --git a/fs/proc/proc_net.c b/fs/proc/proc_net.c
  72 index d5e0fcb3439e..a7b12435519e 100644
  73 --- a/fs/proc/proc_net.c
  74 +++ b/fs/proc/proc_net.c
  75 @@ -38,6 +38,22 @@ static struct net *get_proc_net(const struct inode *inode)
  76         return maybe_get_net(PDE_NET(PDE(inode)));
  77  }
  78
  79 +static int proc_net_d_revalidate(struct dentry *dentry, unsigned int flags)
  80 +{
  81 +       return 0;
  82 +}
  83 +
  84 +static const struct dentry_operations proc_net_dentry_ops = {
  85 +       .d_revalidate   = proc_net_d_revalidate,
  86 +       .d_delete       = always_delete_dentry,
  87 +};
  88 +
  89 +static void pde_force_lookup(struct proc_dir_entry *pde)
  90 +{
  91 +       /* /proc/net/ entries can be changed under us by setns(CLONE_NEWNET) */
  92 +       pde->proc_dops = &proc_net_dentry_ops;
  93 +}
  94 +
  95  static int seq_open_net(struct inode *inode, struct file *file)
  96  {
  97         unsigned int state_size = PDE(inode)->state_size;
  98 @@ -90,6 +106,7 @@ struct proc_dir_entry *proc_create_net_data(const char *name, umode_t mode,
  99         p = proc_create_reg(name, mode, &parent, data);
 100         if (!p)
 101                 return NULL;
 102 +       pde_force_lookup(p);
 103         p->proc_fops = &proc_net_seq_fops;
 104         p->seq_ops = ops;
 105         p->state_size = state_size;
 106 @@ -133,6 +150,7 @@ struct proc_dir_entry *proc_create_net_data_write(const char *name, umode_t mode
 107         p = proc_create_reg(name, mode, &parent, data);
 108         if (!p)
 109                 return NULL;
 110 +       pde_force_lookup(p);
 111         p->proc_fops = &proc_net_seq_fops;
 112         p->seq_ops = ops;
 113         p->state_size = state_size;
 114 @@ -181,6 +199,7 @@ struct proc_dir_entry *proc_create_net_single(const char *name, umode_t mode,
 115         p = proc_create_reg(name, mode, &parent, data);
 116         if (!p)
 117                 return NULL;
 118 +       pde_force_lookup(p);
 119         p->proc_fops = &proc_net_single_fops;
 120         p->single_show = show;
 121         return proc_register(parent, p);
 122 @@ -223,6 +242,7 @@ struct proc_dir_entry *proc_create_net_single_write(const char *name, umode_t mo
 123         p = proc_create_reg(name, mode, &parent, data);
 124         if (!p)
 125                 return NULL;
 126 +       pde_force_lookup(p);
 127         p->proc_fops = &proc_net_single_fops;
 128         p->single_show = show;
 129         p->write = write;
 130 diff --git a/tools/testing/selftests/proc/.gitignore b/tools/testing/selftests/proc/.gitignore
 131 index 82121a81681f..29bac5ef9a93 100644
 132 --- a/tools/testing/selftests/proc/.gitignore
 133 +++ b/tools/testing/selftests/proc/.gitignore
 134 @@ -10,4 +10,5 @@
 135  /proc-uptime-002
 136  /read
 137  /self
 138 +/setns-dcache
 139  /thread-self
 140 diff --git a/tools/testing/selftests/proc/Makefile b/tools/testing/selftests/proc/Makefile
 141 index 1c12c34cf85d..434d033ee067 100644
 142 --- a/tools/testing/selftests/proc/Makefile
 143 +++ b/tools/testing/selftests/proc/Makefile
 144 @@ -14,6 +14,7 @@ TEST_GEN_PROGS += proc-uptime-001
 145  TEST_GEN_PROGS += proc-uptime-002
 146  TEST_GEN_PROGS += read
 147  TEST_GEN_PROGS += self
 148 +TEST_GEN_PROGS += setns-dcache
 149  TEST_GEN_PROGS += thread-self
 150
 151  include ../lib.mk
 152 diff --git a/tools/testing/selftests/proc/setns-dcache.c b/tools/testing/selftests/proc/setns-dcache.c
 153 new file mode 100644
 154 index 000000000000..60ab197a73fc
 155 --- /dev/null
 156 +++ b/tools/testing/selftests/proc/setns-dcache.c
 157 @@ -0,0 +1,129 @@
 158 +/*
 159 + * Copyright © 2019 Alexey Dobriyan <adobriyan@gmail.com>
 160 + *
 161 + * Permission to use, copy, modify, and distribute this software for any
 162 + * purpose with or without fee is hereby granted, provided that the above
 163 + * copyright notice and this permission notice appear in all copies.
 164 + *
 165 + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
 166 + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
 167 + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
 168 + * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
 169 + * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
 170 + * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 171 + * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 172 + */
 173 +/*
 174 + * Test that setns(CLONE_NEWNET) points to new /proc/net content even
 175 + * if old one is in dcache.
 176 + *
 177 + * FIXME /proc/net/unix is under CONFIG_UNIX which can be disabled.
 178 + */
 179 +#undef NDEBUG
 180 +#include <assert.h>
 181 +#include <errno.h>
 182 +#include <sched.h>
 183 +#include <signal.h>
 184 +#include <stdio.h>
 185 +#include <stdlib.h>
 186 +#include <string.h>
 187 +#include <unistd.h>
 188 +#include <sys/types.h>
 189 +#include <sys/stat.h>
 190 +#include <fcntl.h>
 191 +#include <sys/socket.h>
 192 +
 193 +static pid_t pid = -1;
 194 +
 195 +static void f(void)
 196 +{
 197 +       if (pid > 0) {
 198 +               kill(pid, SIGTERM);
 199 +       }
 200 +}
 201 +
 202 +int main(void)
 203 +{
 204 +       int fd[2];
 205 +       char _ = 0;
 206 +       int nsfd;
 207 +
 208 +       atexit(f);
 209 +
 210 +       /* Check for priviledges and syscall availability straight away. */
 211 +       if (unshare(CLONE_NEWNET) == -1) {
 212 +               if (errno == ENOSYS || errno == EPERM) {
 213 +                       return 4;
 214 +               }
 215 +               return 1;
 216 +       }
 217 +       /* Distinguisher between two otherwise empty net namespaces. */
 218 +       if (socket(AF_UNIX, SOCK_STREAM, 0) == -1) {
 219 +               return 1;
 220 +       }
 221 +
 222 +       if (pipe(fd) == -1) {
 223 +               return 1;
 224 +       }
 225 +
 226 +       pid = fork();
 227 +       if (pid == -1) {
 228 +               return 1;
 229 +       }
 230 +
 231 +       if (pid == 0) {
 232 +               if (unshare(CLONE_NEWNET) == -1) {
 233 +                       return 1;
 234 +               }
 235 +
 236 +               if (write(fd[1], &_, 1) != 1) {
 237 +                       return 1;
 238 +               }
 239 +
 240 +               pause();
 241 +
 242 +               return 0;
 243 +       }
 244 +
 245 +       if (read(fd[0], &_, 1) != 1) {
 246 +               return 1;
 247 +       }
 248 +
 249 +       {
 250 +               char buf[64];
 251 +               snprintf(buf, sizeof(buf), "/proc/%u/ns/net", pid);
 252 +               nsfd = open(buf, O_RDONLY);
 253 +               if (nsfd == -1) {
 254 +                       return 1;
 255 +               }
 256 +       }
 257 +
 258 +       /* Reliably pin dentry into dcache. */
 259 +       (void)open("/proc/net/unix", O_RDONLY);
 260 +
 261 +       if (setns(nsfd, CLONE_NEWNET) == -1) {
 262 +               return 1;
 263 +       }
 264 +
 265 +       kill(pid, SIGTERM);
 266 +       pid = 0;
 267 +
 268 +       {
 269 +               char buf[4096];
 270 +               ssize_t rv;
 271 +               int fd;
 272 +
 273 +               fd = open("/proc/net/unix", O_RDONLY);
 274 +               if (fd == -1) {
 275 +                       return 1;
 276 +               }
 277 +
 278 +#define S "Num       RefCount Protocol Flags    Type St Inode Path\n"
 279 +               rv = read(fd, buf, sizeof(buf));
 280 +
 281 +               assert(rv == strlen(S));
 282 +               assert(memcmp(buf, S, strlen(S)) == 0);
 283 +       }
 284 +
 285 +       return 0;
 286 +}
 287 --
 288 2.19.1
 289