@@ -935,6 +935,44 @@ void eventpoll_release_file(struct file *file)
mutex_unlock(&epmutex);
}
+static int ep_insert(struct eventpoll *ep, const struct epoll_event *event,
+ struct file *tfile, int fd, int full_check);
+
+/*
+ * This is called from eventpoll_replace() to replace a linked file in the epoll
+ * interface with a new file received from another process. This is useful in
+ * cases where a process is trying to install a new file for an existing one
+ * that is linked in the epoll interface
+ */
+void eventpoll_replace_file(struct file *toreplace, struct file *file)
+{
+ int fd;
+ struct eventpoll *ep;
+ struct epitem *epi;
+ struct hlist_node *next;
+ struct epoll_event event;
+
+ if (!file_can_poll(file))
+ return;
+
+ mutex_lock(&epmutex);
+ if (unlikely(!toreplace->f_ep)) {
+ mutex_unlock(&epmutex);
+ return;
+ }
+
+ hlist_for_each_entry_safe(epi, next, toreplace->f_ep, fllink) {
+ ep = epi->ep;
+ mutex_lock(&ep->mtx);
+ fd = epi->ffd.fd;
+ event = epi->event;
+ ep_remove(ep, epi);
+ ep_insert(ep, &event, file, fd, 1);
+ mutex_unlock(&ep->mtx);
+ }
+ mutex_unlock(&epmutex);
+}
+
static int ep_alloc(struct eventpoll **pep)
{
int error;
@@ -20,10 +20,18 @@
#include <linux/spinlock.h>
#include <linux/rcupdate.h>
#include <linux/close_range.h>
+#include <linux/eventpoll.h>
#include <net/sock.h>
#include "internal.h"
+struct replace_filefd {
+ struct files_struct *files;
+ unsigned fd;
+ struct file *fdfile;
+ struct file *file;
+};
+
unsigned int sysctl_nr_open __read_mostly = 1024*1024;
unsigned int sysctl_nr_open_min = BITS_PER_LONG;
/* our min() is unusable in constant expressions ;-/ */
@@ -1131,6 +1139,52 @@ __releases(&files->file_lock)
return -EBUSY;
}
+static int do_replace_fd_array(const void *v, struct file *tofree, unsigned int n)
+{
+ struct replace_filefd *ffd = (void *)v;
+ struct fdtable *fdt;
+
+ fdt = files_fdtable(ffd->files);
+
+ if ((n != ffd->fd) && (tofree == ffd->fdfile)) {
+ get_file(ffd->file);
+ rcu_assign_pointer(fdt->fd[n], ffd->file);
+ spin_unlock(&ffd->files->file_lock);
+ filp_close(ffd->fdfile, ffd->files);
+ cond_resched();
+ spin_lock(&ffd->files->file_lock);
+ }
+ return 0;
+}
+
+static void do_replace(struct files_struct *files,
+ struct file *file, unsigned fd, struct file *fdfile)
+{
+ unsigned n = 0;
+ struct replace_filefd ffd = {
+ .files = files,
+ .fd = fd,
+ .fdfile = fdfile,
+ .file = file
+ };
+
+ /*
+ * Check if the file referenced by the fd number is linked to the epoll
+ * interface. If yes, replace the reference with the received file in
+ * the epoll interface.
+ */
+ if (fdfile && fdfile->f_ep) {
+ eventpoll_replace_file(fdfile, file);
+ }
+ /*
+ * Install the received file in the file descriptor table for all fd
+ * numbers referencing the same file as the one we are trying to
+ * replace. Do not install it for the fd number received since that is
+ * handled in do_dup2()
+ */
+ iterate_fd(files, n, do_replace_fd_array, &ffd);
+}
+
int replace_fd(unsigned fd, struct file *file, unsigned flags)
{
int err;
@@ -1148,8 +1202,10 @@ int replace_fd(unsigned fd, struct file *file, unsigned flags)
if (unlikely(err < 0))
goto out_unlock;
err = do_dup2(files, file, fd, &fdfile, flags);
+ do_replace(files, file, fd, fdfile);
if (fdfile)
filp_close(fdfile, files);
+
return err;
out_unlock:
@@ -25,6 +25,14 @@ struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, unsigned long t
/* Used to release the epoll bits inside the "struct file" */
void eventpoll_release_file(struct file *file);
+/*
+ * This is called from fs/file.c:do_replace() to replace a linked file in the
+ * epoll interface with a new file received from another process. This is useful
+ * in cases where a process is trying to install a new file for an existing one
+ * that is linked in the epoll interface
+ */
+void eventpoll_replace_file(struct file *toreplace, struct file *file);
+
/*
* This is called from inside fs/file_table.c:__fput() to unlink files
* from the eventpoll interface. We need to have this facility to cleanup
@@ -47,6 +47,7 @@
#include <linux/kcmp.h>
#include <sys/resource.h>
#include <sys/capability.h>
+#include <sys/epoll.h>
#include <unistd.h>
#include <sys/syscall.h>
@@ -4179,6 +4180,102 @@ TEST(user_notification_addfd)
close(memfd);
}
+TEST(user_notification_addfd_with_epoll_replace)
+{
+ char c;
+ pid_t pid;
+ long ret;
+ int status, listener, fd;
+ int efd, sfd[4];
+ struct epoll_event e;
+ struct seccomp_notif_addfd addfd = {};
+ struct seccomp_notif req = {};
+ struct seccomp_notif_resp resp = {};
+
+ ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
+ ASSERT_EQ(0, ret) {
+ TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
+ }
+
+ listener = user_notif_syscall(__NR_getppid,
+ SECCOMP_FILTER_FLAG_NEW_LISTENER);
+
+ /* Create two socket pairs sfd[0] <-> sfd[1] and sfd[2] <-> sfd[3] */
+ ASSERT_EQ(socketpair(AF_UNIX, SOCK_STREAM, 0, &sfd[0]), 0);
+ ASSERT_EQ(socketpair(AF_UNIX, SOCK_STREAM, 0, &sfd[2]), 0);
+
+ pid = fork();
+ ASSERT_GE(pid, 0);
+
+ if (pid == 0) {
+ efd = epoll_create(1);
+ if (efd == -1)
+ exit(1);
+
+ e.events = EPOLLIN;
+ if (epoll_ctl(efd, EPOLL_CTL_ADD, sfd[0], &e) != 0)
+ exit(1);
+
+ /*
+ * fd will be added here to replace an existing one linked
+ * in the epoll interface.
+ */
+ if (syscall(__NR_getppid) != USER_NOTIF_MAGIC)
+ exit(1);
+
+ /*
+ * Write data to the sfd[3] connected to sfd[2], but due to
+ * the swap, we should see data on sfd[0]
+ */
+ if (write(sfd[3], "w", 1) != 1)
+ exit(1);
+
+ if (epoll_wait(efd, &e, 1, 0) != 1)
+ exit(1);
+
+ if (read(sfd[0], &c, 1) != 1)
+ exit(1);
+
+ if ('w' != c)
+ exit(1);
+
+ if (epoll_ctl(efd, EPOLL_CTL_DEL, sfd[0], &e) != 0)
+ exit(1);
+
+ close(efd);
+ close(sfd[0]);
+ close(sfd[1]);
+ close(sfd[2]);
+ close(sfd[3]);
+ exit(0);
+ }
+
+ ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
+
+ addfd.srcfd = sfd[2];
+ addfd.newfd = sfd[0];
+ addfd.id = req.id;
+ addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
+ addfd.newfd_flags = O_CLOEXEC;
+
+ /*
+ * Verfiy we can install and replace a file that is linked in the
+ * epoll interface. Replace the socket sfd[0] with sfd[2]
+ */
+ fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
+ EXPECT_EQ(fd, sfd[0]);
+
+ resp.id = req.id;
+ resp.error = 0;
+ resp.val = USER_NOTIF_MAGIC;
+ EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
+
+ /* Wait for child to finish. */
+ EXPECT_EQ(waitpid(pid, &status, 0), pid);
+ EXPECT_EQ(true, WIFEXITED(status));
+ EXPECT_EQ(0, WEXITSTATUS(status));
+}
+
TEST(user_notification_addfd_rlimit)
{
pid_t pid;
Introduce a mechanism to replace a file linked in the epoll interface or a file that has been dup'ed by a file received via the replace_fd() interface. eventpoll_replace() is called from do_replace() and finds all instances of the file to be replaced and replaces them with the new file. do_replace() also replaces the file in the file descriptor table for all fd numbers referencing it with the new file. We have a use case where multiple IPv6 only network namespaces can use a single IPv4 network namespace for IPv4 only egress connectivity by switching their sockets from IPv6 to IPv4 network namespace. This allows for migration of systems to IPv6 only while keeping their connectivity to IPv4 only destinations intact. Today, we achieve this by setting up seccomp filter to intercept network system calls like connect() from a container in a container manager which runs in an IPv4 only network namespace. The container manager creates a new IPv4 connection and injects the new file descriptor through SECCOMP_NOTIFY_IOCTL_ADDFD replacing the original file descriptor from the connect() call. This does not work for cases where the original file descriptor is handed off to a system like epoll before the connect() call. After a new file descriptor is injected the original file descriptor being referenced by the epoll fd is not longer valid leading to failures. As a workaround the container manager when intercepting connect() loops through all open socket file descriptors to check if they are referencing the socket attempting the connect() and replace the reference with the to be injected file descriptor. This workaround is cumbersome and makes the solution prone to similar yet to be discovered issues. The above change will enable us remove the workaround in the container manager and let the kernel handle the replacement correctly. Signed-off-by: aloktiagi <aloktiagi@gmail.com> --- Changes in v4: - address review comment to remove the redundant eventpoll_replace() function. - removed an extra empty line introduced in include/linux/file.h Changes in v3: - address review comment and iterate over the file table while holding the spin_lock(&files->file_lock). - address review comment and call filp_close() outside the spin_lock(&files->file_lock). --- fs/eventpoll.c | 38 ++++++++ fs/file.c | 56 +++++++++++ include/linux/eventpoll.h | 8 ++ tools/testing/selftests/seccomp/seccomp_bpf.c | 97 +++++++++++++++++++ 4 files changed, 199 insertions(+)