jail: Mount some things in the outer namespace and some in the inner one

[people/ms/pakfire.git] / src / libpakfire / jail.c

/*#############################################################################
#                                                                             #
# Pakfire - The IPFire package management system                              #
# Copyright (C) 2022 Pakfire development team                                 #
#                                                                             #
# This program is free software: you can redistribute it and/or modify        #
# it under the terms of the GNU General Public License as published by        #
# the Free Software Foundation, either version 3 of the License, or           #
# (at your option) any later version.                                         #
#                                                                             #
# This program is distributed in the hope that it will be useful,             #
# but WITHOUT ANY WARRANTY; without even the implied warranty of              #
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
# GNU General Public License for more details.                                #
#                                                                             #
# You should have received a copy of the GNU General Public License           #
# along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
#                                                                             #
#############################################################################*/

#include <errno.h>
#include <fcntl.h>
#include <linux/capability.h>
#include <linux/sched.h>
#include <sys/wait.h>
#include <linux/wait.h>
#include <sched.h>
#include <signal.h>
#include <stdlib.h>
#include <syscall.h>
#include <sys/capability.h>
#include <sys/epoll.h>
#include <sys/eventfd.h>
#include <sys/mount.h>
#include <sys/personality.h>
#include <sys/prctl.h>
#include <sys/resource.h>
#include <sys/timerfd.h>
#include <sys/types.h>
#include <sys/wait.h>

// libnl3
#include <net/if.h>
#include <netlink/route/link.h>

// libseccomp
#include <seccomp.h>

// libuuid
#include <uuid.h>

#include <pakfire/arch.h>
#include <pakfire/cgroup.h>
#include <pakfire/jail.h>
#include <pakfire/logging.h>
#include <pakfire/mount.h>
#include <pakfire/os.h>
#include <pakfire/pakfire.h>
#include <pakfire/path.h>
#include <pakfire/private.h>
#include <pakfire/pwd.h>
#include <pakfire/string.h>
#include <pakfire/util.h>

#define BUFFER_SIZE      1024 * 64
#define ENVIRON_SIZE     128
#define EPOLL_MAX_EVENTS 2
#define MAX_MOUNTPOINTS  8

// The default environment that will be set for every command
static const struct environ {
        const char* key;
        const char* val;
} ENV[] = {
        { "HOME", "/root" },
        { "LANG", "C.utf-8" },
        { "PATH", "/usr/local/sbin:/usr/sbin:/sbin:/usr/local/bin:/usr/bin:/bin", },
        { "TERM", "vt100" },

        // Tell everything that it is running inside a Pakfire container
        { "container", "pakfire" },
        { NULL, NULL },
};

struct pakfire_jail_mountpoint {
        char source[PATH_MAX];
        char target[PATH_MAX];
        int flags;
};

struct pakfire_jail {
        struct pakfire_ctx* ctx;
        struct pakfire* pakfire;
        int nrefs;

        // A unique ID for each jail
        uuid_t uuid;
        char __uuid[UUID_STR_LEN];

        // Resource Limits
        int nice;

        // Timeout
        struct itimerspec timeout;

        // CGroup
        struct pakfire_cgroup* cgroup;

        // Environment
        char* env[ENVIRON_SIZE];

        // Mountpoints
        struct pakfire_jail_mountpoint mountpoints[MAX_MOUNTPOINTS];
        unsigned int num_mountpoints;

        // Callbacks
        struct pakfire_jail_callbacks {
                // Log
                pakfire_jail_log_callback log;
                void* log_data;
        } callbacks;
};

struct pakfire_log_buffer {
        char data[BUFFER_SIZE];
        size_t used;
};

struct pakfire_jail_exec {
        int flags;

        // PIDs (of the children)
        int pidfd1;
        int pidfd2;

        // Socket to pass FDs
        int socket[2];

        // FD to notify the client that the parent has finished initialization
        int completed_fd;

        // Log pipes
        struct pakfire_jail_pipes {
                int stdin[2];
                int stdout[2];
                int stderr[2];

                // Logging
                int log_INFO[2];
                int log_ERROR[2];
#ifdef ENABLE_DEBUG
                int log_DEBUG[2];
#endif /* ENABLE_DEBUG */
        } pipes;

        // Communicate
        struct pakfire_jail_communicate {
                pakfire_jail_communicate_in  in;
                pakfire_jail_communicate_out out;
                void* data;
        } communicate;

        // Log buffers
        struct pakfire_jail_buffers {
                struct pakfire_log_buffer stdout;
                struct pakfire_log_buffer stderr;

                // Logging
                struct pakfire_log_buffer log_INFO;
                struct pakfire_log_buffer log_ERROR;
#ifdef ENABLE_DEBUG
                struct pakfire_log_buffer log_DEBUG;
#endif /* ENABLE_DEBUG */
        } buffers;

        struct pakfire_cgroup* cgroup;
        struct pakfire_cgroup_stats cgroup_stats;
};

static int clone3(struct clone_args* args, size_t size) {
        return syscall(__NR_clone3, args, size);
}

static int pidfd_send_signal(int pidfd, int sig, siginfo_t* info, unsigned int flags) {
        return syscall(SYS_pidfd_send_signal, pidfd, sig, info, flags);
}

static int pivot_root(const char* new_root, const char* old_root) {
        return syscall(SYS_pivot_root, new_root, old_root);
}

static int pakfire_jail_exec_has_flag(
                const struct pakfire_jail_exec* ctx, const enum pakfire_jail_exec_flags flag) {
        return ctx->flags & flag;
}

static void pakfire_jail_free(struct pakfire_jail* jail) {
        DEBUG(jail->pakfire, "Freeing jail at %p\n", jail);

        // Free environment
        for (unsigned int i = 0; jail->env[i]; i++)
                free(jail->env[i]);

        if (jail->cgroup)
                pakfire_cgroup_unref(jail->cgroup);
        if (jail->pakfire)
                pakfire_unref(jail->pakfire);
        if (jail->ctx)
                pakfire_ctx_unref(jail->ctx);
        free(jail);
}

/*
        Passes any log messages on to the default pakfire log callback
*/
static int pakfire_jail_default_log_callback(struct pakfire* pakfire, void* data,
                int priority, const char* line, size_t length) {
        switch (priority) {
                case LOG_INFO:
                        INFO(pakfire, "%s", line);
                        break;

                case LOG_ERR:
                        ERROR(pakfire, "%s", line);
                        break;

#ifdef ENABLE_DEBUG
                case LOG_DEBUG:
                        DEBUG(pakfire, "%s", line);
                        break;
#endif
        }

        return 0;
}

static const char* pakfire_jail_uuid(struct pakfire_jail* jail) {
        if (!*jail->__uuid)
                uuid_unparse_lower(jail->uuid, jail->__uuid);

        return jail->__uuid;
}

static int pakfire_jail_setup_interactive_env(struct pakfire_jail* jail) {
        // Set PS1
        int r = pakfire_jail_set_env(jail, "PS1", "pakfire-jail \\w> ");
        if (r)
                return r;

        // Copy TERM
        char* TERM = secure_getenv("TERM");
        if (TERM) {
                r = pakfire_jail_set_env(jail, "TERM", TERM);
                if (r)
                        return r;
        }

        // Copy LANG
        char* LANG = secure_getenv("LANG");
        if (LANG) {
                r = pakfire_jail_set_env(jail, "LANG", LANG);
                if (r)
                        return r;
        }

        return 0;
}

PAKFIRE_EXPORT int pakfire_jail_create(struct pakfire_jail** jail, struct pakfire* pakfire) {
        int r;

        const char* arch = pakfire_get_effective_arch(pakfire);

        // Allocate a new jail
        struct pakfire_jail* j = calloc(1, sizeof(*j));
        if (!j)
                return 1;

        // Reference context
        j->ctx = pakfire_ctx(pakfire);

        // Reference Pakfire
        j->pakfire = pakfire_ref(pakfire);

        // Initialize reference counter
        j->nrefs = 1;

        // Generate a random UUID
        uuid_generate_random(j->uuid);

        DEBUG(j->pakfire, "Allocated new jail at %p\n", j);

        // Set the default logging callback
        pakfire_jail_set_log_callback(j, pakfire_jail_default_log_callback, NULL);

        // Set default environment
        for (const struct environ* e = ENV; e->key; e++) {
                r = pakfire_jail_set_env(j, e->key, e->val);
                if (r)
                        goto ERROR;
        }

        // Enable all CPU features that CPU has to offer
        if (!pakfire_arch_is_supported_by_host(arch)) {
                r = pakfire_jail_set_env(j, "QEMU_CPU", "max");
                if (r)
                        goto ERROR;
        }

        // Set container UUID
        r = pakfire_jail_set_env(j, "container_uuid", pakfire_jail_uuid(j));
        if (r)
                goto ERROR;

        // Disable systemctl to talk to systemd
        if (!pakfire_on_root(j->pakfire)) {
                r = pakfire_jail_set_env(j, "SYSTEMD_OFFLINE", "1");
                if (r)
                        goto ERROR;
        }

        // Done
        *jail = j;
        return 0;

ERROR:
        pakfire_jail_free(j);

        return r;
}

PAKFIRE_EXPORT struct pakfire_jail* pakfire_jail_ref(struct pakfire_jail* jail) {
        ++jail->nrefs;

        return jail;
}

PAKFIRE_EXPORT struct pakfire_jail* pakfire_jail_unref(struct pakfire_jail* jail) {
        if (--jail->nrefs > 0)
                return jail;

        pakfire_jail_free(jail);
        return NULL;
}

// Logging Callback

PAKFIRE_EXPORT void pakfire_jail_set_log_callback(struct pakfire_jail* jail,
                pakfire_jail_log_callback callback, void* data) {
        jail->callbacks.log = callback;
        jail->callbacks.log_data = data;
}

// Resource Limits

PAKFIRE_EXPORT int pakfire_jail_nice(struct pakfire_jail* jail, int nice) {
        // Check if nice level is in range
        if (nice < -19 || nice > 20) {
                errno = EINVAL;
                return 1;
        }

        // Store nice level
        jail->nice = nice;

        return 0;
}

int pakfire_jail_set_cgroup(struct pakfire_jail* jail, struct pakfire_cgroup* cgroup) {
        // Free any previous cgroup
        if (jail->cgroup) {
                pakfire_cgroup_unref(jail->cgroup);
                jail->cgroup = NULL;
        }

        // Set any new cgroup
        if (cgroup) {
                DEBUG(jail->pakfire, "Setting cgroup %p\n", cgroup);

                jail->cgroup = pakfire_cgroup_ref(cgroup);
        }

        // Done
        return 0;
}

// Environment

// Returns the length of the environment
static unsigned int pakfire_jail_env_length(struct pakfire_jail* jail) {
        unsigned int i = 0;

        // Count everything in the environment
        for (char** e = jail->env; *e; e++)
                i++;

        return i;
}

// Finds an existing environment variable and returns its index or -1 if not found
static int pakfire_jail_find_env(struct pakfire_jail* jail, const char* key) {
        if (!key) {
                errno = EINVAL;
                return -1;
        }

        const size_t length = strlen(key);

        for (unsigned int i = 0; jail->env[i]; i++) {
                if ((pakfire_string_startswith(jail->env[i], key)
                                && *(jail->env[i] + length) == '=')) {
                        return i;
                }
        }

        // Nothing found
        return -1;
}

// Returns the value of an environment variable or NULL
PAKFIRE_EXPORT const char* pakfire_jail_get_env(struct pakfire_jail* jail,
                const char* key) {
        int i = pakfire_jail_find_env(jail, key);
        if (i < 0)
                return NULL;

        return jail->env[i] + strlen(key) + 1;
}

// Sets an environment variable
PAKFIRE_EXPORT int pakfire_jail_set_env(struct pakfire_jail* jail,
                const char* key, const char* value) {
        // Find the index where to write this value to
        int i = pakfire_jail_find_env(jail, key);
        if (i < 0)
                i = pakfire_jail_env_length(jail);

        // Return -ENOSPC when the environment is full
        if (i >= ENVIRON_SIZE) {
                errno = ENOSPC;
                return -1;
        }

        // Free any previous value
        if (jail->env[i])
                free(jail->env[i]);

        // Format and set environment variable
        asprintf(&jail->env[i], "%s=%s", key, value);

        DEBUG(jail->pakfire, "Set environment variable: %s\n", jail->env[i]);

        return 0;
}

// Imports an environment
PAKFIRE_EXPORT int pakfire_jail_import_env(struct pakfire_jail* jail, const char* env[]) {
        if (!env)
                return 0;

        char* key;
        char* val;
        int r;

        // Copy environment variables
        for (unsigned int i = 0; env[i]; i++) {
                r = pakfire_string_partition(env[i], "=", &key, &val);
                if (r)
                        continue;

                // Set value
                r = pakfire_jail_set_env(jail, key, val);

                if (key)
                        free(key);
                if (val)
                        free(val);

                // Break on error
                if (r)
                        return r;
        }

        return 0;
}

// Timeout

PAKFIRE_EXPORT int pakfire_jail_set_timeout(
                struct pakfire_jail* jail, unsigned int timeout) {
        // Store value
        jail->timeout.it_value.tv_sec = timeout;

        if (timeout > 0)
                DEBUG(jail->pakfire, "Timeout set to %u second(s)\n", timeout);
        else
                DEBUG(jail->pakfire, "Timeout disabled\n");

        return 0;
}

static int pakfire_jail_create_timer(struct pakfire_jail* jail) {
        int r;

        // Nothing to do if no timeout has been set
        if (!jail->timeout.it_value.tv_sec)
                return -1;

        // Create a new timer
        const int fd = timerfd_create(CLOCK_MONOTONIC, 0);
        if (fd < 0) {
                ERROR(jail->pakfire, "Could not create timer: %m\n");
                goto ERROR;
        }

        // Arm timer
        r = timerfd_settime(fd, 0, &jail->timeout, NULL);
        if (r) {
                ERROR(jail->pakfire, "Could not arm timer: %m\n");
                goto ERROR;
        }

        return fd;

ERROR:
        if (fd >= 0)
                close(fd);

        return -1;
}

/*
        This function replaces any logging in the child process.

        All log messages will be sent to the parent process through their respective pipes.
*/
static void pakfire_jail_log_redirect(void* data, int priority, const char* file,
                int line, const char* fn, const char* format, va_list args) {
        struct pakfire_jail_pipes* pipes = (struct pakfire_jail_pipes*)data;
        int fd;

        switch (priority) {
                case LOG_INFO:
                        fd = pipes->log_INFO[1];
                        break;

                case LOG_ERR:
                        fd = pipes->log_ERROR[1];
                        break;

#ifdef ENABLE_DEBUG
                case LOG_DEBUG:
                        fd = pipes->log_DEBUG[1];
                        break;
#endif /* ENABLE_DEBUG */

                // Ignore any messages of an unknown priority
                default:
                        return;
        }

        // Send the log message
        if (fd >= 0)
                vdprintf(fd, format, args);
}

static int pakfire_jail_log_buffer_is_full(const struct pakfire_log_buffer* buffer) {
        return (sizeof(buffer->data) == buffer->used);
}

/*
        This function reads as much data as it can from the file descriptor.
        If it finds a whole line in it, it will send it to the logger and repeat the process.
        If not newline character is found, it will try to read more data until it finds one.
*/
static int pakfire_jail_handle_log(struct pakfire_jail* jail,
                struct pakfire_jail_exec* ctx, int priority, int fd,
                struct pakfire_log_buffer* buffer, pakfire_jail_communicate_out callback, void* data) {
        char line[BUFFER_SIZE + 1];

        // Fill up buffer from fd
        if (buffer->used < sizeof(buffer->data)) {
                ssize_t bytes_read = read(fd, buffer->data + buffer->used,
                                sizeof(buffer->data) - buffer->used);

                // Handle errors
                if (bytes_read < 0) {
                        ERROR(jail->pakfire, "Could not read from fd %d: %m\n", fd);
                        return -1;
                }

                // Update buffer size
                buffer->used += bytes_read;
        }

        // See if we have any lines that we can write
        while (buffer->used) {
                // Search for the end of the first line
                char* eol = memchr(buffer->data, '\n', buffer->used);

                // No newline found
                if (!eol) {
                        // If the buffer is full, we send the content to the logger and try again
                        // This should not happen in practise
                        if (pakfire_jail_log_buffer_is_full(buffer)) {
                                DEBUG(jail->pakfire, "Logging buffer is full. Sending all content\n");

                                eol = buffer->data + sizeof(buffer->data) - 1;

                        // Otherwise we might have only read parts of the output
                        } else
                                break;
                }

                // Find the length of the string
                size_t length = eol - buffer->data + 1;

                // Copy the line into the buffer
                memcpy(line, buffer->data, length);

                // Terminate the string
                line[length] = '\0';

                // Log the line
                if (callback) {
                        int r = callback(jail->pakfire, data, priority, line, length);
                        if (r) {
                                ERROR(jail->pakfire, "The logging callback returned an error: %d\n", r);
                                return r;
                        }
                }

                // Remove line from buffer
                memmove(buffer->data, buffer->data + length, buffer->used - length);
                buffer->used -= length;
        }

        return 0;
}

static int pakfire_jail_stream_stdin(struct pakfire_jail* jail,
                struct pakfire_jail_exec* ctx, const int fd) {
        int r;

        // Nothing to do if there is no stdin callback set
        if (!ctx->communicate.in) {
                DEBUG(jail->pakfire, "Callback for standard input is not set\n");
                return 0;
        }

        // Skip if the writing pipe has already been closed
        if (!ctx->pipes.stdin[1])
                return 0;

        DEBUG(jail->pakfire, "Streaming standard input...\n");

        // Calling the callback
        r = ctx->communicate.in(jail->pakfire, ctx->communicate.data, fd);

        DEBUG(jail->pakfire, "Standard input callback finished: %d\n", r);

        // The callback signaled that it has written everything
        if (r == EOF) {
                DEBUG(jail->pakfire, "Closing standard input pipe\n");

                // Close the file-descriptor
                close(fd);

                // Reset the file-descriptor so it won't be closed again later
                ctx->pipes.stdin[1] = -1;

                // Report success
                r = 0;
        }

        return r;
}

static int pakfire_jail_setup_pipe(struct pakfire_jail* jail, int (*fds)[2], const int flags) {
        int r = pipe2(*fds, flags);
        if (r < 0) {
                ERROR(jail->pakfire, "Could not setup pipe: %m\n");
                return 1;
        }

        return 0;
}

static void pakfire_jail_close_pipe(struct pakfire_jail* jail, int fds[2]) {
        for (unsigned int i = 0; i < 2; i++)
                if (fds[i] >= 0)
                        close(fds[i]);
}

/*
        This is a convenience function to fetch the reading end of a pipe and
        closes the write end.
*/
static int pakfire_jail_get_pipe_to_read(struct pakfire_jail* jail, int (*fds)[2]) {
        // Give the variables easier names to avoid confusion
        int* fd_read  = &(*fds)[0];
        int* fd_write = &(*fds)[1];

        // Close the write end of the pipe
        if (*fd_write >= 0) {
                close(*fd_write);
                *fd_write = -1;
        }

        // Return the read end
        if (*fd_read >= 0)
                return *fd_read;

        return -1;
}

static int pakfire_jail_get_pipe_to_write(struct pakfire_jail* jail, int (*fds)[2]) {
        // Give the variables easier names to avoid confusion
        int* fd_read  = &(*fds)[0];
        int* fd_write = &(*fds)[1];

        // Close the read end of the pipe
        if (*fd_read >= 0) {
                close(*fd_read);
                *fd_read = -1;
        }

        // Return the write end
        if (*fd_write >= 0)
                return *fd_write;

        return -1;
}

static int pakfire_jail_recv_fd(struct pakfire_jail* jail, int socket, int* fd) {
        const size_t payload_length = sizeof(fd);
        char buffer[CMSG_SPACE(payload_length)];
        int r;

        struct msghdr msg = {
                .msg_control    = buffer,
                .msg_controllen = sizeof(buffer),
        };

        // Receive the message
        r = recvmsg(socket, &msg, 0);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not receive file descriptor: %s\n", strerror(errno));
                return -errno;
        }

        // Fetch the payload
        struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg);
        if (!cmsg)
                return -EBADMSG;

        *fd = *((int*)CMSG_DATA(cmsg));

        CTX_DEBUG(jail->ctx, "Received fd %d from socket %d\n", *fd, socket);

        return 0;
}

static int pakfire_jail_send_fd(struct pakfire_jail* jail, int socket, int fd) {
        const size_t payload_length = sizeof(fd);
        char buffer[CMSG_SPACE(payload_length)];
        int r;

        CTX_DEBUG(jail->ctx, "Sending fd %d to socket %d\n", fd, socket);

        // Header
        struct msghdr msg = {
                .msg_control    = buffer,
                .msg_controllen = sizeof(buffer),
        };

        // Payload
        struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg);
        cmsg->cmsg_level = SOL_SOCKET;
        cmsg->cmsg_type  = SCM_RIGHTS;
        cmsg->cmsg_len   = CMSG_LEN(payload_length);

        // Set payload
        *((int*)CMSG_DATA(cmsg)) = fd;

        // Send the message
        r = sendmsg(socket, &msg, 0);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not send file descriptor: %s\n", strerror(errno));
                return -errno;
        }

        return 0;
}

static int pakfire_jail_log(struct pakfire* pakfire, void* data, int priority,
                const char* line, const size_t length) {
        // Pass everything to the parent logger
        pakfire_log_condition(pakfire, priority, 0, "%.*s", (int)length, line);

        return 0;
}

static int pakfire_jail_epoll_add_fd(struct pakfire_jail* jail, int epollfd, int fd, int events) {
        struct epoll_event event = {
                .events = events|EPOLLHUP,
                .data   = {
                        .fd = fd,
                },
        };
        int r;

        // Read flags
        int flags = fcntl(fd, F_GETFL, 0);

        // Set modified flags
        r  = fcntl(fd, F_SETFL, flags|O_NONBLOCK);
        if (r < 0) {
                CTX_ERROR(jail->ctx, "Could not set file descriptor %d into non-blocking mode: %s\n",
                        fd, strerror(errno));
                return -errno;
        }

        // Add the file descriptor to the loop
        r = epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &event);
        if (r < 0) {
                ERROR(jail->pakfire, "Could not add file descriptor %d to epoll(): %s\n",
                        fd, strerror(errno));
                return -errno;
        }

        return 0;
}

static int pakfire_jail_setup_child2(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx);

static int pakfire_jail_wait_on_child(struct pakfire_jail* jail, int pidfd) {
        siginfo_t status = {};
        int r;

        // Call waitid() and store the result
        r = waitid(P_PIDFD, pidfd, &status, WEXITED);
        if (r) {
                CTX_ERROR(jail->ctx, "waitid() failed: %s\n", strerror(errno));
                return -errno;
        }

        switch (status.si_code) {
                // If the process exited normally, we return the exit code
                case CLD_EXITED:
                        CTX_DEBUG(jail->ctx, "The child process exited with code %d\n", status.si_status);
                        return status.si_status;

                case CLD_KILLED:
                        CTX_ERROR(jail->ctx, "The child process was killed\n");
                        return 139;

                case CLD_DUMPED:
                        CTX_ERROR(jail->ctx, "The child process terminated abnormally\n");
                        return 139;

                // Log anything else
                default:
                        CTX_ERROR(jail->ctx, "Unknown child exit code: %d\n", status.si_code);
                        break;
        }

        return -EBADMSG;
}

static int pakfire_jail_wait(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) {
        int epollfd = -1;
        struct epoll_event events[EPOLL_MAX_EVENTS];
        char garbage[8];
        int r = 0;

        // Fetch the UNIX domain socket
        const int socket_recv = pakfire_jail_get_pipe_to_read(jail, &ctx->socket);

        // Fetch file descriptors from context
        const int stdin = pakfire_jail_get_pipe_to_write(jail, &ctx->pipes.stdin);
        const int stdout = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.stdout);
        const int stderr = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.stderr);

        // Timer
        const int timerfd = pakfire_jail_create_timer(jail);

        // Logging
        const int log_INFO  = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.log_INFO);
        const int log_ERROR = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.log_ERROR);
#ifdef ENABLE_DEBUG
        const int log_DEBUG = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.log_DEBUG);
#endif /* ENABLE_DEBUG */

        // Make a list of all file descriptors we are interested in
        const struct pakfire_wait_fds {
                const int fd;
                const int events;
        } fds[] = {
                { socket_recv, EPOLLIN },

                // Standard input/output
                { stdin,  EPOLLOUT },
                { stdout, EPOLLIN },
                { stderr, EPOLLIN },

                // Timer
                { timerfd, EPOLLIN },

                // Child Processes
                { ctx->pidfd1, EPOLLIN },

                // Log Pipes
                { log_INFO, EPOLLIN },
                { log_ERROR, EPOLLIN },
#ifdef ENABLE_DEBUG
                { log_DEBUG, EPOLLIN },
#endif /* ENABLE_DEBUG */

                // Sentinel
                { -1, 0 },
        };

        // Setup epoll
        epollfd = epoll_create1(0);
        if (epollfd < 0) {
                ERROR(jail->pakfire, "Could not initialize epoll(): %m\n");
                r = 1;
                goto ERROR;
        }

        // Turn file descriptors into non-blocking mode and add them to epoll()
        for (const struct pakfire_wait_fds* fd = fds; fd->events; fd++) {
                // Skip fds which were not initialized
                if (fd->fd < 0)
                        continue;

                // Add the FD to the event loop
                r = pakfire_jail_epoll_add_fd(jail, epollfd, fd->fd, fd->events);
                if (r)
                        goto ERROR;
        }

        int ended = 0;
        int exit = 0;

        CTX_DEBUG(jail->ctx, "Launching main loop...\n");

        // Loop for as long as the process is alive
        while (!ended) {
                int num = epoll_wait(epollfd, events, EPOLL_MAX_EVENTS, -1);
                if (num < 1) {
                        // Ignore if epoll_wait() has been interrupted
                        if (errno == EINTR)
                                continue;

                        ERROR(jail->pakfire, "epoll_wait() failed: %m\n");
                        r = 1;

                        goto ERROR;
                }

                for (int i = 0; i < num; i++) {
                        int e  = events[i].events;
                        int fd = events[i].data.fd;

                        struct pakfire_log_buffer* buffer = NULL;
                        pakfire_jail_communicate_out callback = NULL;
                        void* data = NULL;
                        int priority;

                        // Check if there is any data to be read
                        if (e & EPOLLIN) {
                                // Monitor the first child process
                                if (fd == ctx->pidfd1) {
                                        r = pakfire_jail_wait_on_child(jail, ctx->pidfd1);
                                        if (r) {
                                                CTX_ERROR(jail->ctx, "The first child exited with an error\n");
                                                goto ERROR;
                                        }

                                        close(ctx->pidfd1);
                                        ctx->pidfd1 = -1;

                                        continue;

                                // Monitor the second child process
                                } else if (fd == ctx->pidfd2) {
                                        exit = pakfire_jail_wait_on_child(jail, ctx->pidfd2);
                                        if (exit < 0) {
                                                CTX_ERROR(jail->ctx, "The second child exited with an error\n");
                                                goto ERROR;
                                        }

                                        close(ctx->pidfd2);
                                        ctx->pidfd2 = -1;

                                        // Mark that we have ended so that we will process the remaining
                                        // events from epoll() now, but won't restart the outer loop.
                                        ended = 1;

                                        continue;

                                // Handle timer events
                                } else if (fd == timerfd) {
                                        DEBUG(jail->pakfire, "Timer event received\n");

                                        // Disarm the timer
                                        r = read(timerfd, garbage, sizeof(garbage));
                                        if (r < 1) {
                                                ERROR(jail->pakfire, "Could not disarm timer: %m\n");
                                                r = 1;
                                                goto ERROR;
                                        }

                                        // Terminate the process if it hasn't already ended
                                        if (!ended) {
                                                DEBUG(jail->pakfire, "Terminating process...\n");

                                                // Send SIGTERM to the process
                                                r = pidfd_send_signal(ctx->pidfd2, SIGKILL, NULL, 0);
                                                if (r) {
                                                        ERROR(jail->pakfire, "Could not kill process: %m\n");
                                                        goto ERROR;
                                                }
                                        }

                                        // There is nothing else to do
                                        continue;

                                // Handle socket messages
                                } else if (fd == socket_recv) {
                                        // Receive the FD of the second child process
                                        r = pakfire_jail_recv_fd(jail, socket_recv, &ctx->pidfd2);
                                        if (r)
                                                goto ERROR;

                                        // Add it to the event loop
                                        r = pakfire_jail_epoll_add_fd(jail, epollfd, ctx->pidfd2, EPOLLIN);
                                        if (r)
                                                goto ERROR;

                                        // Setup the child process
                                        r = pakfire_jail_setup_child2(jail, ctx);
                                        if (r)
                                                goto ERROR;

                                        // Don't fall through to log processing
                                        continue;

                                // Handle logging messages
                                } else if (fd == log_INFO) {
                                        buffer = &ctx->buffers.log_INFO;
                                        priority = LOG_INFO;

                                        callback = pakfire_jail_log;

                                } else if (fd == log_ERROR) {
                                        buffer = &ctx->buffers.log_ERROR;
                                        priority = LOG_ERR;

                                        callback = pakfire_jail_log;

#ifdef ENABLE_DEBUG
                                } else if (fd == log_DEBUG) {
                                        buffer = &ctx->buffers.log_DEBUG;
                                        priority = LOG_DEBUG;

                                        callback = pakfire_jail_log;
#endif /* ENABLE_DEBUG */

                                // Handle anything from the log pipes
                                } else if (fd == stdout) {
                                        buffer = &ctx->buffers.stdout;
                                        priority = LOG_INFO;

                                        // Send any output to the default logger if no callback is set
                                        if (ctx->communicate.out) {
                                                callback = ctx->communicate.out;
                                                data     = ctx->communicate.data;
                                        } else {
                                                callback = jail->callbacks.log;
                                                data     = jail->callbacks.log_data;
                                        }

                                } else if (fd == stderr) {
                                        buffer = &ctx->buffers.stderr;
                                        priority = LOG_ERR;

                                        // Send any output to the default logger if no callback is set
                                        if (ctx->communicate.out) {
                                                callback = ctx->communicate.out;
                                                data     = ctx->communicate.data;
                                        } else {
                                                callback = jail->callbacks.log;
                                                data     = jail->callbacks.log_data;
                                        }

                                } else {
                                        DEBUG(jail->pakfire, "Received invalid file descriptor %d\n", fd);
                                        continue;
                                }

                                // Handle log event
                                r = pakfire_jail_handle_log(jail, ctx, priority, fd, buffer, callback, data);
                                if (r)
                                        goto ERROR;
                        }

                        if (e & EPOLLOUT) {
                                // Handle standard input
                                if (fd == stdin) {
                                        r = pakfire_jail_stream_stdin(jail, ctx, fd);
                                        if (r) {
                                                switch (errno) {
                                                        // Ignore if we filled up the buffer
                                                        case EAGAIN:
                                                                break;

                                                        default:
                                                                ERROR(jail->pakfire, "Could not write to stdin: %m\n");
                                                                goto ERROR;
                                                }
                                        }
                                }
                        }

                        // Check if any file descriptors have been closed
                        if (e & EPOLLHUP) {
                                // Remove the file descriptor
                                r = epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, NULL);
                                if (r) {
                                        ERROR(jail->pakfire, "Could not remove closed file-descriptor %d: %m\n", fd);
                                        goto ERROR;
                                }
                        }
                }
        }

        // Return the exit code
        r = exit;

ERROR:
        CTX_DEBUG(jail->ctx, "Main loop terminated\n");

        if (epollfd >= 0)
                close(epollfd);
        if (timerfd >= 0)
                close(timerfd);

        return r;
}

int pakfire_jail_capture_stdout(struct pakfire* pakfire, void* data,
                int priority, const char* line, size_t length) {
        char** output = (char**)data;
        int r;

        // Append everything from stdout to a buffer
        if (output && priority == LOG_INFO) {
                r = asprintf(output, "%s%s", (output && *output) ? *output : "", line);
                if (r < 0)
                        return 1;
                return 0;
        }

        // Send everything else to the default logger
        return pakfire_jail_default_log_callback(pakfire, NULL, priority, line, length);
}

// Capabilities

// Logs all capabilities of the current process
static int pakfire_jail_show_capabilities(struct pakfire_jail* jail) {
        cap_t caps = NULL;
        char* name = NULL;
        cap_flag_value_t value_e;
        cap_flag_value_t value_i;
        cap_flag_value_t value_p;
        int r;

        // Fetch PID
        pid_t pid = getpid();

        // Fetch all capabilities
        caps = cap_get_proc();
        if (!caps) {
                ERROR(jail->pakfire, "Could not fetch capabilities: %m\n");
                r = 1;
                goto ERROR;
        }

        DEBUG(jail->pakfire, "Capabilities of PID %d:\n", pid);

        // Iterate over all capabilities
        for (unsigned int cap = 0; cap_valid(cap); cap++) {
                name = cap_to_name(cap);

                // Fetch effective value
                r = cap_get_flag(caps, cap, CAP_EFFECTIVE, &value_e);
                if (r)
                        goto ERROR;

                // Fetch inheritable value
                r = cap_get_flag(caps, cap, CAP_INHERITABLE, &value_i);
                if (r)
                        goto ERROR;

                // Fetch permitted value
                r = cap_get_flag(caps, cap, CAP_PERMITTED, &value_p);
                if (r)
                        goto ERROR;

                DEBUG(jail->pakfire,
                        "  %-24s : %c%c%c\n",
                        name,
                        (value_e == CAP_SET) ? 'e' : '-',
                        (value_i == CAP_SET) ? 'i' : '-',
                        (value_p == CAP_SET) ? 'p' : '-'
                );

                // Free name
                cap_free(name);
                name = NULL;
        }

        // Success
        r = 0;

ERROR:
        if (name)
                cap_free(name);
        if (caps)
                cap_free(caps);

        return r;
}

static int pakfire_jail_set_capabilities(struct pakfire_jail* jail) {
        cap_t caps = NULL;
        char* name = NULL;
        int r;

        // Fetch capabilities
        caps = cap_get_proc();
        if (!caps) {
                ERROR(jail->pakfire, "Could not read capabilities: %m\n");
                r = 1;
                goto ERROR;
        }

        // Walk through all capabilities
        for (cap_value_t cap = 0; cap_valid(cap); cap++) {
                cap_value_t _caps[] = { cap };

                // Fetch the name of the capability
                name = cap_to_name(cap);

                r = cap_set_flag(caps, CAP_EFFECTIVE, 1, _caps, CAP_SET);
                if (r) {
                        ERROR(jail->pakfire, "Could not set %s: %m\n", name);
                        goto ERROR;
                }

                r = cap_set_flag(caps, CAP_INHERITABLE, 1, _caps, CAP_SET);
                if (r) {
                        ERROR(jail->pakfire, "Could not set %s: %m\n", name);
                        goto ERROR;
                }

                r = cap_set_flag(caps, CAP_PERMITTED, 1, _caps, CAP_SET);
                if (r) {
                        ERROR(jail->pakfire, "Could not set %s: %m\n", name);
                        goto ERROR;
                }

                // Free name
                cap_free(name);
                name = NULL;
        }

        // Restore all capabilities
        r = cap_set_proc(caps);
        if (r) {
                ERROR(jail->pakfire, "Restoring capabilities failed: %m\n");
                goto ERROR;
        }

        // Add all capabilities to the ambient set
        for (unsigned int cap = 0; cap_valid(cap); cap++) {
                name = cap_to_name(cap);

                // Raise the capability
                r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, cap, 0, 0);
                if (r) {
                        ERROR(jail->pakfire, "Could not set ambient capability %s: %m\n", name);
                        goto ERROR;
                }

                // Free name
                cap_free(name);
                name = NULL;
        }

        // Success
        r = 0;

ERROR:
        if (name)
                cap_free(name);
        if (caps)
                cap_free(caps);

        return r;
}

// Syscall Filter

static int pakfire_jail_limit_syscalls(struct pakfire_jail* jail) {
        const int syscalls[] = {
                // The kernel's keyring isn't namespaced
                SCMP_SYS(keyctl),
                SCMP_SYS(add_key),
                SCMP_SYS(request_key),

                // Disable userfaultfd
                SCMP_SYS(userfaultfd),

                // Disable perf which could leak a lot of information about the host
                SCMP_SYS(perf_event_open),

                0,
        };
        int r = 1;

        DEBUG(jail->pakfire, "Applying syscall filter...\n");

        // Setup a syscall filter which allows everything by default
        scmp_filter_ctx ctx = seccomp_init(SCMP_ACT_ALLOW);
        if (!ctx) {
                ERROR(jail->pakfire, "Could not setup seccomp filter: %m\n");
                goto ERROR;
        }

        // All all syscalls
        for (const int* syscall = syscalls; *syscall; syscall++) {
                r = seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), *syscall, 0);
                if (r) {
                        ERROR(jail->pakfire, "Could not configure syscall %d: %m\n", *syscall);
                        goto ERROR;
                }
        }

        // Load syscall filter into the kernel
        r = seccomp_load(ctx);
        if (r) {
                ERROR(jail->pakfire, "Could not load syscall filter into the kernel: %m\n");
                goto ERROR;
        }

ERROR:
        if (ctx)
                seccomp_release(ctx);

        return r;
}

// Mountpoints

PAKFIRE_EXPORT int pakfire_jail_bind(struct pakfire_jail* jail,
                const char* source, const char* target, int flags) {
        struct pakfire_jail_mountpoint* mp = NULL;
        int r;

        // Check if there is any space left
        if (jail->num_mountpoints >= MAX_MOUNTPOINTS) {
                errno = ENOSPC;
                return 1;
        }

        // Check for valid inputs
        if (!source || !target) {
                errno = EINVAL;
                return 1;
        }

        // Select the next free slot
        mp = &jail->mountpoints[jail->num_mountpoints];

        // Copy source
        r = pakfire_string_set(mp->source, source);
        if (r) {
                ERROR(jail->pakfire, "Could not copy source: %m\n");
                return r;
        }

        // Copy target
        r = pakfire_string_set(mp->target, target);
        if (r) {
                ERROR(jail->pakfire, "Could not copy target: %m\n");
                return r;
        }

        // Copy flags
        mp->flags = flags;

        // Increment counter
        jail->num_mountpoints++;

        return 0;
}

static int pakfire_jail_mount_networking(struct pakfire_jail* jail) {
        int r;

        const char* paths[] = {
                "/etc/hosts",
                "/etc/resolv.conf",
                NULL,
        };

        // Bind-mount all paths read-only
        for (const char** path = paths; *path; path++) {
                r = pakfire_bind(jail->pakfire, *path, NULL, MS_RDONLY);
                if (r) {
                        switch (errno) {
                                // Ignore if we don't have permission
                                case EPERM:
                                        continue;

                                default:
                                        break;
                        }
                        return r;
                }
        }

        return 0;
}

/*
        Mounts everything that we require in the new namespace
*/
static int pakfire_jail_mount(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) {
        struct pakfire_jail_mountpoint* mp = NULL;
        int flags = 0;
        int r;

        // Enable loop devices
        if (pakfire_jail_exec_has_flag(ctx, PAKFIRE_JAIL_HAS_LOOP_DEVICES))
                flags |= PAKFIRE_MOUNT_LOOP_DEVICES;

        // Mount all default stuff
        r = pakfire_mount_all(jail->pakfire, PAKFIRE_MNTNS_OUTER, flags);
        if (r)
                return r;

        // Populate /dev
        r = pakfire_populate_dev(jail->pakfire, flags);
        if (r)
                return r;

        // Mount the interpreter (if needed)
        r = pakfire_mount_interpreter(jail->pakfire);
        if (r)
                return r;

        // Mount networking stuff
        if (pakfire_jail_exec_has_flag(ctx, PAKFIRE_JAIL_HAS_NETWORKING)) {
                r = pakfire_jail_mount_networking(jail);
                if (r)
                        return r;
        }

        // Mount all custom stuff
        for (unsigned int i = 0; i < jail->num_mountpoints; i++) {
                // Fetch mountpoint
                mp = &jail->mountpoints[i];

                // Mount it
                r = pakfire_bind(jail->pakfire, mp->source, mp->target, mp->flags);
                if (r)
                        return r;
        }

        return 0;
}

// Networking

static int pakfire_jail_setup_loopback(struct pakfire_jail* jail) {
        struct nl_sock* nl = NULL;
        struct nl_cache* cache = NULL;
        struct rtnl_link* link = NULL;
        struct rtnl_link* change = NULL;
        int r;

        DEBUG(jail->pakfire, "Setting up loopback...\n");

        // Allocate a netlink socket
        nl = nl_socket_alloc();
        if (!nl) {
                ERROR(jail->pakfire, "Could not allocate a netlink socket: %m\n");
                r = 1;
                goto ERROR;
        }

        // Connect the socket
        r = nl_connect(nl, NETLINK_ROUTE);
        if (r) {
                ERROR(jail->pakfire, "Could not connect netlink socket: %s\n", nl_geterror(r));
                goto ERROR;
        }

        // Allocate the netlink cache
        r = rtnl_link_alloc_cache(nl, AF_UNSPEC, &cache);
        if (r < 0) {
                ERROR(jail->pakfire, "Unable to allocate netlink cache: %s\n", nl_geterror(r));
                goto ERROR;
        }

        // Fetch loopback interface
        link = rtnl_link_get_by_name(cache, "lo");
        if (!link) {
                ERROR(jail->pakfire, "Could not find lo interface. Ignoring.\n");
                r = 0;
                goto ERROR;
        }

        // Allocate a new link
        change = rtnl_link_alloc();
        if (!change) {
                ERROR(jail->pakfire, "Could not allocate change link\n");
                r = 1;
                goto ERROR;
        }

        // Set the link to UP
        rtnl_link_set_flags(change, IFF_UP);

        // Apply any changes
        r = rtnl_link_change(nl, link, change, 0);
        if (r) {
                ERROR(jail->pakfire, "Unable to activate loopback: %s\n", nl_geterror(r));
                goto ERROR;
        }

        // Success
        r = 0;

ERROR:
        if (nl)
                nl_socket_free(nl);

        return r;
}

// UID/GID Mapping

static int pakfire_jail_setup_uid_mapping(struct pakfire_jail* jail, pid_t pid) {
        char path[PATH_MAX];
        int r;

        // Skip mapping anything when running on /
        if (pakfire_on_root(jail->pakfire))
                return 0;

        // Make path
        r = pakfire_string_format(path, "/proc/%d/uid_map", pid);
        if (r)
                return r;

        // Fetch UID
        const uid_t uid = pakfire_uid(jail->pakfire);

        // Fetch SUBUID
        const struct pakfire_subid* subuid = pakfire_subuid(jail->pakfire);
        if (!subuid)
                return 1;

        /* When running as root, we will map the entire range.

           When running as a non-privileged user, we will map the root user inside the jail
           to the user's UID outside of the jail, and we will map the rest starting from one.
        */

        // Running as root
        if (uid == 0) {
                r = pakfire_file_write(jail->pakfire, path, 0, 0, 0,
                        "0 %lu %lu\n", subuid->id, subuid->length);
        } else {
                r = pakfire_file_write(jail->pakfire, path, 0, 0, 0,
                        "0 %lu 1\n1 %lu %lu\n", uid, subuid->id, subuid->length);
        }

        if (r) {
                ERROR(jail->pakfire, "Could not map UIDs: %m\n");
                return r;
        }

        return r;
}

static int pakfire_jail_setup_gid_mapping(struct pakfire_jail* jail, pid_t pid) {
        char path[PATH_MAX];
        int r;

        // Skip mapping anything when running on /
        if (pakfire_on_root(jail->pakfire))
                return 0;

        // Fetch GID
        const gid_t gid = pakfire_gid(jail->pakfire);

        // Fetch SUBGID
        const struct pakfire_subid* subgid = pakfire_subgid(jail->pakfire);
        if (!subgid)
                return 1;

        // Make path
        r = pakfire_string_format(path, "/proc/%d/gid_map", pid);
        if (r)
                return r;

        // Running as root
        if (gid == 0) {
                r = pakfire_file_write(jail->pakfire, path, 0, 0, 0,
                        "0 %lu %lu\n", subgid->id, subgid->length);
        } else {
                r = pakfire_file_write(jail->pakfire, path, 0, 0, 0,
                        "0 %lu 1\n1 %lu %lu\n", gid, subgid->id, subgid->length);
        }

        if (r) {
                ERROR(jail->pakfire, "Could not map GIDs: %m\n");
                return r;
        }

        return r;
}

static int pakfire_jail_setgroups(struct pakfire_jail* jail, pid_t pid) {
        char path[PATH_MAX];
        int r;

        // Make path
        r = pakfire_string_format(path, "/proc/%d/setgroups", pid);
        if (r)
                return r;

        r = pakfire_file_write(jail->pakfire, path, 0, 0, 0, "deny\n");
        if (r) {
                CTX_ERROR(jail->ctx, "Could not set setgroups to deny: %s\n", strerror(errno));
                r = -errno;
        }

        return r;
}

static int pakfire_jail_send_signal(struct pakfire_jail* jail, int fd) {
        const uint64_t val = 1;
        int r = 0;

        DEBUG(jail->pakfire, "Sending signal...\n");

        // Write to the file descriptor
        r = eventfd_write(fd, val);
        if (r < 0) {
                ERROR(jail->pakfire, "Could not send signal: %s\n", strerror(errno));
                r = -errno;
        }

        // Close the file descriptor
        close(fd);

        return r;
}

static int pakfire_jail_wait_for_signal(struct pakfire_jail* jail, int fd) {
        uint64_t val = 0;
        int r = 0;

        DEBUG(jail->pakfire, "Waiting for signal...\n");

        r = eventfd_read(fd, &val);
        if (r < 0) {
                ERROR(jail->pakfire, "Error waiting for signal: %s\n", strerror(errno));
                r = -errno;
        }

        // Close the file descriptor
        close(fd);

        return r;
}

static int pakfire_jail_switch_root(struct pakfire_jail* jail, const char* root) {
        int r;

        // Change to the new root
        r = chdir(root);
        if (r) {
                ERROR(jail->pakfire, "chdir(%s) failed: %m\n", root);
                return r;
        }

        // Switch Root!
        r = pivot_root(".", ".");
        if (r) {
                ERROR(jail->pakfire, "Failed changing into the new root directory %s: %m\n", root);
                return r;
        }

        // Umount the old root
        r = umount2(".", MNT_DETACH);
        if (r) {
                ERROR(jail->pakfire, "Could not umount the old root filesystem: %m\n");
                return r;
        }

        return 0;
}

/*
        Called by the parent that sets up the second child process...
*/
static int pakfire_jail_setup_child2(
                struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) {
        pid_t pid = -1;
        int r;

        // Fetch the PID
        r = pidfd_get_pid(ctx->pidfd2, &pid);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not fetch PID: %s\n", strerror(-r));
                return r;
        }

        // Setup UID mapping
        r = pakfire_jail_setup_uid_mapping(jail, pid);
        if (r)
                return r;

        // Write "deny" to /proc/PID/setgroups
        r = pakfire_jail_setgroups(jail, pid);
        if (r)
                return r;

        // Setup GID mapping
        r = pakfire_jail_setup_gid_mapping(jail, pid);
        if (r)
                return r;

        // Parent has finished initialisation
        DEBUG(jail->pakfire, "Parent has finished initialization\n");

        // Send signal to client
        r = pakfire_jail_send_signal(jail, ctx->completed_fd);
        if (r)
                return r;

        return 0;
}

/*
        Child 2 is launched in their own user/mount/etc. namespace.
*/
static int pakfire_jail_child2(struct pakfire_jail* jail,
                struct pakfire_jail_exec* ctx, const char* argv[]) {
        int r;

        // Fetch my own PID
        pid_t pid = getpid();

        CTX_DEBUG(jail->ctx, "Launched child process in jail with PID %d\n", pid);

        // Make this process dumpable
        r = prctl (PR_SET_DUMPABLE, 1, 0, 0, 0);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not make the process dumpable: %m\n");
                return 126;
        }

        // Don't drop any capabilities on setuid()
        r = prctl(PR_SET_KEEPCAPS, 1);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not set PR_SET_KEEPCAPS: %m\n");
                return 126;
        }

        // Wait for the parent to finish initialization
        r = pakfire_jail_wait_for_signal(jail, ctx->completed_fd);
        if (r)
                return r;

        // Fetch UID/GID
        uid_t uid = getuid();
        gid_t gid = getgid();

        // Fetch EUID/EGID
        uid_t euid = geteuid();
        gid_t egid = getegid();

        DEBUG(jail->pakfire, "  UID: %u (effective %u)\n", uid, euid);
        DEBUG(jail->pakfire, "  GID: %u (effective %u)\n", gid, egid);

        // Log all mountpoints
        pakfire_mount_list(jail->ctx);

        // Fail if we are not PID 1
        if (pid != 1) {
                CTX_ERROR(jail->ctx, "Child process is not PID 1\n");
                return 126;
        }

        // Fail if we are not running as root
        if (uid || gid || euid || egid) {
                ERROR(jail->pakfire, "Child process is not running as root\n");
                return 126;
        }

        // Mount all default stuff
        r = pakfire_mount_all(jail->pakfire, PAKFIRE_MNTNS_INNER, 0);
        if (r)
                return 126;

        const char* arch = pakfire_get_effective_arch(jail->pakfire);

        // Set personality
        unsigned long persona = pakfire_arch_personality(arch);
        if (persona) {
                r = personality(persona);
                if (r < 0) {
                        ERROR(jail->pakfire, "Could not set personality (%x)\n", (unsigned int)persona);
                        return 126;
                }
        }

        // Setup networking
        if (!pakfire_jail_exec_has_flag(ctx, PAKFIRE_JAIL_HAS_NETWORKING)) {
                r = pakfire_jail_setup_loopback(jail);
                if (r)
                        return 1;
        }

        // Set nice level
        if (jail->nice) {
                DEBUG(jail->pakfire, "Setting nice level to %d\n", jail->nice);

                r = setpriority(PRIO_PROCESS, pid, jail->nice);
                if (r) {
                        ERROR(jail->pakfire, "Could not set nice level: %m\n");
                        return 1;
                }
        }

        // Close other end of log pipes
        close(ctx->pipes.log_INFO[0]);
        close(ctx->pipes.log_ERROR[0]);
#ifdef ENABLE_DEBUG
        close(ctx->pipes.log_DEBUG[0]);
#endif /* ENABLE_DEBUG */

        // Connect standard input
        if (ctx->pipes.stdin[0] >= 0) {
                r = dup2(ctx->pipes.stdin[0], STDIN_FILENO);
                if (r < 0) {
                        ERROR(jail->pakfire, "Could not connect fd %d to stdin: %m\n",
                                ctx->pipes.stdin[0]);

                        return 1;
                }
        }

        // Connect standard output and error
        if (ctx->pipes.stdout[1] >= 0 && ctx->pipes.stderr[1] >= 0) {
                r = dup2(ctx->pipes.stdout[1], STDOUT_FILENO);
                if (r < 0) {
                        ERROR(jail->pakfire, "Could not connect fd %d to stdout: %m\n",
                                ctx->pipes.stdout[1]);

                        return 1;
                }

                r = dup2(ctx->pipes.stderr[1], STDERR_FILENO);
                if (r < 0) {
                        ERROR(jail->pakfire, "Could not connect fd %d to stderr: %m\n",
                                ctx->pipes.stderr[1]);

                        return 1;
                }

                // Close the pipe (as we have moved the original file descriptors)
                pakfire_jail_close_pipe(jail, ctx->pipes.stdin);
                pakfire_jail_close_pipe(jail, ctx->pipes.stdout);
                pakfire_jail_close_pipe(jail, ctx->pipes.stderr);
        }

        // Reset open file limit (http://0pointer.net/blog/file-descriptor-limits.html)
        r = pakfire_rlimit_reset_nofile(jail->pakfire);
        if (r)
                return r;

        // Set capabilities
        r = pakfire_jail_set_capabilities(jail);
        if (r)
                return r;

        // Show capabilities
        r = pakfire_jail_show_capabilities(jail);
        if (r)
                return r;

        // Filter syscalls
        r = pakfire_jail_limit_syscalls(jail);
        if (r)
                return r;

        CTX_DEBUG(jail->ctx, "Child process initialization done\n");
        CTX_DEBUG(jail->ctx, "Launching command:\n");

        // Log argv
        for (unsigned int i = 0; argv[i]; i++)
                CTX_DEBUG(jail->ctx, "  argv[%u] = %s\n", i, argv[i]);

        // exec() command
        r = execvpe(argv[0], (char**)argv, jail->env);
        if (r < 0) {
                // Translate errno into regular exit code
                switch (errno) {
                        case ENOENT:
                                // Ignore if the command doesn't exist
                                if (ctx->flags & PAKFIRE_JAIL_NOENT_OK)
                                        r = 0;
                                else
                                        r = 127;

                                break;

                        default:
                                r = 1;
                }

                CTX_ERROR(jail->ctx, "Could not execve(%s): %m\n", argv[0]);
        }

        // We should not get here
        return r;
}

/*
        Child 1 is launched in a new mount namespace...
*/
static int pakfire_jail_child1(struct pakfire_jail* jail,
                struct pakfire_jail_exec* ctx, const char* argv[]) {
        int r;

        // Redirect any logging to our log pipe
        pakfire_ctx_set_log_callback(jail->ctx, pakfire_jail_log_redirect, &ctx->pipes);

        CTX_DEBUG(jail->ctx, "First child process launched\n");

        const int socket_send = pakfire_jail_get_pipe_to_write(jail, &ctx->socket);

        const char* root = pakfire_get_path(jail->pakfire);

        // Die with parent
        r = prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not configure to die with parent: %s\n", strerror(errno));
                goto ERROR;
        }

        // Change mount propagation so that we will receive, but don't propagate back
        r = pakfire_mount_change_propagation(jail->ctx, "/", MS_SLAVE);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not change mount propagation to SLAVE: %s\n", strerror(r));
                goto ERROR;
        }

        // Make root a mountpoint in the new mount namespace
        r = pakfire_mount_make_mounpoint(jail->pakfire, root);
        if (r)
                goto ERROR;

        // Make everything private
        r = pakfire_mount_change_propagation(jail->ctx, root, MS_PRIVATE);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not change mount propagation to PRIVATE: %s\n", strerror(r));
                goto ERROR;
        }

        // Mount everything
        r = pakfire_jail_mount(jail, ctx);
        if (r)
                goto ERROR;

        // chroot()
        r = pakfire_jail_switch_root(jail, root);
        if (r)
                goto ERROR;

        // Change mount propagation so that we will propagate everything down
        r = pakfire_mount_change_propagation(jail->ctx, "/", MS_SHARED);
        if (r) {
                CTX_ERROR(jail->ctx, "Could not change mount propagation to SHARED: %s\n", strerror(r));
                goto ERROR;
        }

        // Configure child process
        struct clone_args args = {
                .flags =
                        CLONE_NEWCGROUP |
                        CLONE_NEWIPC |
                        CLONE_NEWNS |
                        CLONE_NEWPID |
                        CLONE_NEWTIME |
                        CLONE_NEWUSER |
                        CLONE_NEWUTS |
                        CLONE_PIDFD,
                .exit_signal = SIGCHLD,
                .pidfd = (long long unsigned int)&ctx->pidfd2,
        };

        // Launch the process into the configured cgroup
        if (ctx->cgroup) {
                args.flags |= CLONE_INTO_CGROUP;

                // Clone into this cgroup
                args.cgroup = pakfire_cgroup_fd(ctx->cgroup);
        }

        // Setup networking
        if (!pakfire_jail_exec_has_flag(ctx, PAKFIRE_JAIL_HAS_NETWORKING))
                args.flags |= CLONE_NEWNET;

        // Fork the second child process
        pid_t pid = clone3(&args, sizeof(args));
        if (pid < 0) {
                CTX_ERROR(jail->ctx, "Could not fork the first child process: %s\n", strerror(errno));
                r = -errno;
                goto ERROR;

        // Child process
        } else if (pid == 0) {
                r = pakfire_jail_child2(jail, ctx, argv);
                _exit(r);
        }

        // Send the pidfd of the child to the first parent
        r = pakfire_jail_send_fd(jail, socket_send, ctx->pidfd2);
        if (r)
                goto ERROR;

ERROR:
        return r;
}

// Run a command in the jail
static int __pakfire_jail_exec(struct pakfire_jail* jail, const char* argv[],
                const int interactive,
                pakfire_jail_communicate_in  communicate_in,
                pakfire_jail_communicate_out communicate_out,
                void* data, int flags) {
        int r;

        // Check if argv is valid
        if (!argv || !argv[0]) {
                errno = EINVAL;
                return -1;
        }

        // Initialize context for this call
        struct pakfire_jail_exec ctx = {
                .flags = flags,

                .socket = { -1, -1 },

                .pipes = {
                        .stdin     = { -1, -1 },
                        .stdout    = { -1, -1 },
                        .stderr    = { -1, -1 },
                        .log_INFO  = { -1, -1 },
                        .log_ERROR = { -1, -1 },
#ifdef ENABLE_DEBUG
                        .log_DEBUG = { -1, -1 },
#endif /* ENABLE_DEBUG */
                },

                .communicate = {
                        .in   = communicate_in,
                        .out  = communicate_out,
                        .data = data,
                },

                // PIDs
                .pidfd1 = -1,
                .pidfd2 = -1,
        };

        DEBUG(jail->pakfire, "Executing jail...\n");

        // Become the subreaper
        r = prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0);
        if (r < 0) {
                CTX_ERROR(jail->ctx, "Failed to become the sub-reaper: %s\n", strerror(errno));
                r = -errno;
                goto ERROR;
        }

        // Enable networking in interactive mode
        if (interactive)
                ctx.flags |= PAKFIRE_JAIL_HAS_NETWORKING;

        // Create a UNIX domain socket
        r = socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, ctx.socket);
        if (r < 0) {
                CTX_ERROR(jail->ctx, "Could not create UNIX socket: %s\n", strerror(errno));
                r = -errno;
                goto ERROR;
        }

        /*
                Setup a file descriptor which can be used to notify the client that the parent
                has completed configuration.
        */
        ctx.completed_fd = eventfd(0, EFD_CLOEXEC);
        if (ctx.completed_fd < 0) {
                ERROR(jail->pakfire, "eventfd() failed: %m\n");
                return -1;
        }

        // Create pipes to communicate with child process if we are not running interactively
        if (!interactive) {
                // stdin (only if callback is set)
                if (ctx.communicate.in) {
                        r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stdin, 0);
                        if (r)
                                goto ERROR;
                }

                // stdout
                r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stdout, 0);
                if (r)
                        goto ERROR;

                // stderr
                r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stderr, 0);
                if (r)
                        goto ERROR;
        }

        // Setup pipes for logging
        // INFO
        r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_INFO, O_CLOEXEC);
        if (r)
                goto ERROR;

        // ERROR
        r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_ERROR, O_CLOEXEC);
        if (r)
                goto ERROR;

#ifdef ENABLE_DEBUG
        // DEBUG
        r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_DEBUG, O_CLOEXEC);
        if (r)
                goto ERROR;
#endif /* ENABLE_DEBUG */

        // Launch the process in a cgroup that is a leaf of the configured cgroup
        if (jail->cgroup) {
                // Fetch our UUID
                const char* uuid = pakfire_jail_uuid(jail);

                // Create a temporary cgroup
                r = pakfire_cgroup_child(&ctx.cgroup, jail->cgroup, uuid, 0);
                if (r) {
                        ERROR(jail->pakfire, "Could not create cgroup for jail: %m\n");
                        goto ERROR;
                }
        }

        /*
                Initially, we will set up a new mount namespace and launch a child process in it.

                This process remains in the user/ipc/time/etc. namespace and will set up
                the mount namespace.
        */

        // Configure child process
        struct clone_args args = {
                .flags =
                        CLONE_NEWNS |
                        CLONE_PIDFD |
                        CLONE_CLEAR_SIGHAND,
                .exit_signal = SIGCHLD,
                .pidfd = (long long unsigned int)&ctx.pidfd1,
        };

        // Fork the first child process
        pid_t pid = clone3(&args, sizeof(args));
        if (pid < 0) {
                CTX_ERROR(jail->ctx, "Could not fork the first child process: %s\n", strerror(errno));
                r = -errno;
                goto ERROR;

        // Child process
        } else if (pid == 0) {
                r = pakfire_jail_child1(jail, &ctx, argv);
                _exit(r);
        }

        // Parent process
        r = pakfire_jail_wait(jail, &ctx);
        if (r)
                goto ERROR;

ERROR:
        // Destroy the temporary cgroup (if any)
        if (ctx.cgroup) {
                // Read cgroup stats
                pakfire_cgroup_stat(ctx.cgroup, &ctx.cgroup_stats);
                pakfire_cgroup_stat_dump(ctx.cgroup, &ctx.cgroup_stats);
                pakfire_cgroup_destroy(ctx.cgroup);
                pakfire_cgroup_unref(ctx.cgroup);
        }

        // Close any file descriptors
        pakfire_jail_close_pipe(jail, ctx.pipes.stdin);
        pakfire_jail_close_pipe(jail, ctx.pipes.stdout);
        pakfire_jail_close_pipe(jail, ctx.pipes.stderr);
        pakfire_jail_close_pipe(jail, ctx.pipes.log_INFO);
        pakfire_jail_close_pipe(jail, ctx.pipes.log_ERROR);
#ifdef ENABLE_DEBUG
        pakfire_jail_close_pipe(jail, ctx.pipes.log_DEBUG);
#endif /* ENABLE_DEBUG */
        if (ctx.pidfd1 >= 0)
                close(ctx.pidfd1);
        if (ctx.pidfd2 >= 0)
                close(ctx.pidfd2);

        // Close sockets
        pakfire_jail_close_pipe(jail, ctx.socket);

        return r;
}

PAKFIRE_EXPORT int pakfire_jail_exec(
                struct pakfire_jail* jail,
                const char* argv[],
                pakfire_jail_communicate_in  callback_in,
                pakfire_jail_communicate_out callback_out,
                void* data, int flags) {
        return __pakfire_jail_exec(jail, argv, 0, callback_in, callback_out, data, flags);
}

static int pakfire_jail_exec_interactive(
                struct pakfire_jail* jail, const char* argv[], int flags) {
        int r;

        // Setup interactive stuff
        r = pakfire_jail_setup_interactive_env(jail);
        if (r)
                return r;

        return __pakfire_jail_exec(jail, argv, 1, NULL, NULL, NULL, flags);
}

int pakfire_jail_exec_script(struct pakfire_jail* jail,
                const char* script,
                const size_t size,
                const char* args[],
                pakfire_jail_communicate_in  callback_in,
                pakfire_jail_communicate_out callback_out,
                void* data) {
        char path[PATH_MAX];
        const char** argv = NULL;
        FILE* f = NULL;
        int r;

        const char* root = pakfire_get_path(jail->pakfire);

        // Write the scriptlet to disk
        r = pakfire_path_append(path, root, PAKFIRE_TMP_DIR "/pakfire-script.XXXXXX");
        if (r)
                goto ERROR;

        // Create a temporary file
        f = pakfire_mktemp(path, 0700);
        if (!f) {
                ERROR(jail->pakfire, "Could not create temporary file: %m\n");
                goto ERROR;
        }

        DEBUG(jail->pakfire, "Writing script to %s:\n%.*s\n", path, (int)size, script);

        // Write data
        r = fprintf(f, "%s", script);
        if (r < 0) {
                ERROR(jail->pakfire, "Could not write script to file %s: %m\n", path);
                goto ERROR;
        }

        // Close file
        r = fclose(f);
        if (r) {
                ERROR(jail->pakfire, "Could not close script file %s: %m\n", path);
                goto ERROR;
        }

        f = NULL;

        // Count how many arguments were passed
        unsigned int argc = 1;
        if (args) {
                for (const char** arg = args; *arg; arg++)
                        argc++;
        }

        argv = calloc(argc + 1, sizeof(*argv));
        if (!argv) {
                ERROR(jail->pakfire, "Could not allocate argv: %m\n");
                goto ERROR;
        }

        // Set command
        argv[0] = (root) ? pakfire_path_relpath(root, path) : path;

        // Copy args
        for (unsigned int i = 1; i < argc; i++)
                argv[i] = args[i-1];

        // Run the script
        r = pakfire_jail_exec(jail, argv, callback_in, callback_out, data, 0);

ERROR:
        if (argv)
                free(argv);
        if (f)
                fclose(f);

        // Remove script from disk
        if (*path)
                unlink(path);

        return r;
}

/*
        A convenience function that creates a new jail, runs the given command and destroys
        the jail again.
*/
int pakfire_jail_run(struct pakfire* pakfire, const char* argv[], int flags, char** output) {
        struct pakfire_jail* jail = NULL;
        int r;

        // Create a new jail
        r = pakfire_jail_create(&jail, pakfire);
        if (r)
                goto ERROR;

        // Execute the command
        r = pakfire_jail_exec(jail, argv, NULL, pakfire_jail_capture_stdout, output, 0);

ERROR:
        if (jail)
                pakfire_jail_unref(jail);

        return r;
}

int pakfire_jail_run_script(struct pakfire* pakfire,
                const char* script, const size_t length, const char* argv[], int flags) {
        struct pakfire_jail* jail = NULL;
        int r;

        // Create a new jail
        r = pakfire_jail_create(&jail, pakfire);
        if (r)
                goto ERROR;

        // Execute the command
        r = pakfire_jail_exec_script(jail, script, length, argv, NULL, NULL, NULL);

ERROR:
        if (jail)
                pakfire_jail_unref(jail);

        return r;
}

int pakfire_jail_shell(struct pakfire_jail* jail) {
        int r;

        const char* argv[] = {
                "/bin/bash", "--login", NULL,
        };

        // Execute /bin/bash
        r = pakfire_jail_exec_interactive(jail, argv, 0);

        // Raise any errors
        if (r < 0)
                return r;

        // Ignore any return codes from the shell
        return 0;
}

static int pakfire_jail_run_if_possible(struct pakfire* pakfire, const char** argv) {
        char path[PATH_MAX];
        int r;

        r = pakfire_path(pakfire, path, "%s", *argv);
        if (r)
                return r;

        // Check if the file is executable
        r = access(path, X_OK);
        if (r) {
                DEBUG(pakfire, "%s is not executable. Skipping...\n", *argv);
                return 0;
        }

        return pakfire_jail_run(pakfire, argv, 0, NULL);
}

int pakfire_jail_ldconfig(struct pakfire* pakfire) {
        const char* argv[] = {
                "/sbin/ldconfig",
                NULL,
        };

        return pakfire_jail_run_if_possible(pakfire, argv);
}

int pakfire_jail_run_systemd_tmpfiles(struct pakfire* pakfire) {
        const char* argv[] = {
                "/usr/bin/systemd-tmpfiles",
                "--create",
                NULL,
        };

        return pakfire_jail_run_if_possible(pakfire, argv);
}