jail: Create a leaf cgroup

[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 <linux/capability.h>
#include <linux/fcntl.h>
#include <linux/sched.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/personality.h>
#include <sys/prctl.h>
#include <sys/resource.h>
#include <sys/types.h>
#include <sys/wait.h>

// libseccomp
#include <seccomp.h>

#include <pakfire/arch.h>
#include <pakfire/cgroup.h>
#include <pakfire/jail.h>
#include <pakfire/logging.h>
#include <pakfire/mount.h>
#include <pakfire/pakfire.h>
#include <pakfire/private.h>
#include <pakfire/util.h>

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

// The default environment that will be set for every command
static const struct environ {
        const char* key;
        const char* val;
} ENV[] = {
        { "LANG", "en_US.utf-8" },
        { "TERM", "vt100" },
        { NULL, NULL },
};

struct pakfire_jail {
        struct pakfire* pakfire;
        int nrefs;

        // Flags
        int flags;

        // Resource Limits
        int nice;

        // CGroup
        struct pakfire_cgroup* cgroup;

        // Environment
        char* env[ENVIRON_SIZE];

        // Logging
        pakfire_jail_log_callback log_callback;
        void* log_data;
};

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

struct pakfire_jail_exec {
        // PID (of the child)
        pid_t pid;
        int pidfd;

        // Process status (from waitid)
        siginfo_t status;

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

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

                // Logging
                int log_INFO[2];
                int log_ERROR[2];
                int log_DEBUG[2];
        } pipes;

        // 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;
                struct pakfire_log_buffer log_DEBUG;
        } buffers;

        struct pakfire_cgroup* cgroup;
};

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

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);

        pakfire_unref(jail->pakfire);
        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 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 flags) {
        int r;

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

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

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

        // Store flags
        j->flags = flags;

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

        // Set default log callback
        r = pakfire_jail_set_log_callback(j, pakfire_jail_default_log_callback, NULL);
        if (r)
                goto ERROR;

        // 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;
        }

        // Setup interactive stuff
        if (j->flags & PAKFIRE_JAIL_INTERACTIVE) {
                r = pakfire_jail_setup_interactive_env(j);
                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;
}

static int pakfire_jail_has_flag(struct pakfire_jail* jail, int flag) {
        return jail->flags & flag;
}

// 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;
        }

        char buffer[strlen(key) + 2];
        pakfire_string_format(buffer, "%s=", key);

        for (unsigned int i = 0; jail->env[i]; i++) {
                if (pakfire_string_startswith(jail->env[i], buffer))
                        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;
}

// Logging

PAKFIRE_EXPORT int pakfire_jail_set_log_callback(struct pakfire_jail* jail,
                pakfire_jail_log_callback callback, void* data) {
        jail->log_callback = callback;
        jail->log_data = data;

        return 0;
}

/*
        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(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)
                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_log_callback 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_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])
                        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(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) {
                close(*fd_write);
                *fd_write = 0;
        }

        // Return the read end
        return *fd_read;
}

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

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

        // Logging
        const int log_INFO  = pakfire_jail_get_pipe(jail, &ctx->pipes.log_INFO);
        const int log_ERROR = pakfire_jail_get_pipe(jail, &ctx->pipes.log_ERROR);
        const int log_DEBUG = pakfire_jail_get_pipe(jail, &ctx->pipes.log_DEBUG);

        // Make a list of all file descriptors we are interested in
        int fds[] = {
                stdout, stderr, pidfd, log_INFO, log_ERROR, log_DEBUG,
        };

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

        ev.events = EPOLLIN;

        // Turn file descriptors into non-blocking mode and add them to epoll()
        for (unsigned int i = 0; i < sizeof(fds) / sizeof(*fds); i++) {
                int fd = fds[i];

                // Skip fds which were not initialized
                if (fd <= 0)
                        continue;

                ev.data.fd = fd;

                if (epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &ev) < 0) {
                        ERROR(jail->pakfire, "Could not add file descriptor %d to epoll(): %m\n", fd);
                        r = 1;
                        goto ERROR;
                }
        }

        int ended = 0;

        // 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 fd = events[i].data.fd;

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

                        // Handle any changes to the PIDFD
                        if (fd == pidfd) {
                                // Call waidid() and store the result
                                r = waitid(P_PIDFD, ctx->pidfd, &ctx->status, WEXITED);
                                if (r) {
                                        ERROR(jail->pakfire, "waitid() failed: %m\n");
                                        goto ERROR;
                                }

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

                                callback = pakfire_jail_default_log_callback;

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

                                callback = pakfire_jail_default_log_callback;

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

                                callback = pakfire_jail_default_log_callback;

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

                                callback = jail->log_callback;
                                data = jail->log_data;

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

                                callback = jail->log_callback;
                                data = jail->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;
                }
        }

ERROR:
        if (epollfd > 0)
                close(epollfd);

        return r;
}

static 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 (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

static int pakfire_jail_drop_capabilities(struct pakfire_jail* jail) {
        const int capabilities[] = {
                // Deny access to the kernel's audit system
                CAP_AUDIT_CONTROL,
                CAP_AUDIT_READ,
                CAP_AUDIT_WRITE,

                // Deny suspending block devices
                CAP_BLOCK_SUSPEND,

                // Deny any stuff with BPF
                CAP_BPF,

                // Deny checkpoint restore
                CAP_CHECKPOINT_RESTORE,

                // Deny opening files by inode number (open_by_handle_at)
                CAP_DAC_READ_SEARCH,

                // Deny setting SUID bits
                CAP_FSETID,

                // Deny locking more memory
                CAP_IPC_LOCK,

                // Deny modifying any Apparmor/SELinux/SMACK configuration
                CAP_MAC_ADMIN,
                CAP_MAC_OVERRIDE,

                // Deny creating any special devices
                CAP_MKNOD,

                // Deny setting any capabilities
                CAP_SETFCAP,

                // Deny reading from syslog
                CAP_SYSLOG,

                // Deny any admin actions (mount, sethostname, ...)
                CAP_SYS_ADMIN,

                // Deny rebooting the system
                CAP_SYS_BOOT,

                // Deny loading kernel modules
                CAP_SYS_MODULE,

                // Deny setting nice level
                CAP_SYS_NICE,

                // Deny access to /proc/kcore, /dev/mem, /dev/kmem
                CAP_SYS_RAWIO,

                // Deny circumventing any resource limits
                CAP_SYS_RESOURCE,

                // Deny setting the system time
                CAP_SYS_TIME,

                // Deny playing with suspend
                CAP_WAKE_ALARM,

                0,
        };

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

        size_t num_caps = 0;
        int r;

        // Drop any capabilities
        for (const int* cap = capabilities; *cap; cap++) {
                r = prctl(PR_CAPBSET_DROP, *cap, 0, 0, 0);
                if (r) {
                        ERROR(jail->pakfire, "Could not drop capability %d: %m\n", *cap);
                        return r;
                }

                num_caps++;
        }

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

        /*
                Set inheritable capabilities

                This ensures that no processes will be able to gain any of the listed
                capabilities again.
        */
        r = cap_set_flag(caps, CAP_INHERITABLE, num_caps, capabilities, CAP_CLEAR);
        if (r) {
                ERROR(jail->pakfire, "cap_set_flag() failed: %m\n");
                goto ERROR;
        }

        // Restore capabilities
        r = cap_set_proc(caps);
        if (r) {
                ERROR(jail->pakfire, "Could not restore capabilities: %m\n");
                goto ERROR;
        }

ERROR:
        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;
}

// UID/GID Mapping

static int pakfire_jail_write_uidgid_mapping(struct pakfire_jail* jail,
                const char* path, uid_t mapped_id, size_t length) {
        int r = 1;

        // Open file for writing
        FILE* f = fopen(path, "w");
        if (!f) {
                ERROR(jail->pakfire, "Could not open %s for writing: %m\n", path);
                goto ERROR;
        }

        // Write configuration
        int bytes_written = fprintf(f, "%d %d %ld\n", 0, mapped_id, length);
        if (bytes_written <= 0) {
                ERROR(jail->pakfire, "Could not write UID/GID mapping: %m\n");
                goto ERROR;
        }

        // Close the file
        r = fclose(f);
        f = NULL;
        if (r) {
                ERROR(jail->pakfire, "Could not write UID/GID mapping: %m\n");

                goto ERROR;
        }

        // Success
        r = 0;

ERROR:
        if (f)
                fclose(f);

        return r;
}

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

        uid_t mapped_uid = 0;
        const size_t length = 1;

        // Fetch the UID of the calling process
        uid_t uid = getuid();

        // Have we been called by root?
        if (uid == 0) {
                mapped_uid = 0;

        // Have we been called by an unprivileged user?
        } else {
                // XXX fetch SUBUID
                mapped_uid = uid;
        }

        // Make path
        r = pakfire_string_format(path, "/proc/%d/uid_map", pid);
        if (r < 0)
                return 1;

        DEBUG(jail->pakfire, "Mapping UID range (%u - %lu)\n", mapped_uid, mapped_uid + length);

        return pakfire_jail_write_uidgid_mapping(jail, path, mapped_uid, length);
}

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

        gid_t mapped_gid = 0;
        const size_t length = 1;

        // Fetch the GID of the calling process
        gid_t gid = getgid();

        // Have we been called from the root group?
        if (gid == 0) {
                mapped_gid = 0;

        // Have we been called by an unprivileged group?
        } else {
                // XXX fetch SUBGID
                mapped_gid = gid;
        }

        // Make path
        r = pakfire_string_format(path, "/proc/%d/gid_map", pid);
        if (r < 0)
                return 1;

        DEBUG(jail->pakfire, "Mapping GID range (%u - %lu)\n", mapped_gid, mapped_gid + length);

        return pakfire_jail_write_uidgid_mapping(jail, path, mapped_gid, length);
}

static int pakfire_jail_setgroups(struct pakfire_jail* jail, pid_t pid) {
        char path[PATH_MAX];
        int r = 1;

        // Make path
        r = pakfire_string_format(path, "/proc/%d/setgroups", pid);
        if (r < 0)
                return 1;

        // Open file for writing
        FILE* f = fopen(path, "w");
        if (!f) {
                ERROR(jail->pakfire, "Could not open %s for writing: %m\n", path);
                goto ERROR;
        }

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

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

ERROR:
        if (f)
                fclose(f);

        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
        ssize_t bytes_written = write(fd, &val, sizeof(val));
        if (bytes_written < 0 || (size_t)bytes_written < sizeof(val)) {
                ERROR(jail->pakfire, "Could not send signal: %m\n");
                r = 1;
        }

        // 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");

        ssize_t bytes_read = read(fd, &val, sizeof(val));
        if (bytes_read < 0 || (size_t)bytes_read < sizeof(val)) {
                ERROR(jail->pakfire, "Error waiting for signal: %m\n");
                r = 1;
        }

        // Close the file descriptor
        close(fd);

        return r;
}

/*
        Performs the initialisation that needs to happen in the parent part
*/
static int pakfire_jail_parent(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) {
        int r;

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

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

        // Setup GID mapping
        r = pakfire_jail_setup_gid_mapping(jail, ctx->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;
}

static int pakfire_jail_child(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx,
                const char* argv[]) {
        int r;

        // Redirect any logging to our log pipe
        pakfire_set_log_callback(jail->pakfire, pakfire_jail_log, &ctx->pipes);

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

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

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

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

        // Perform further initialization

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

        // Check if we are (effectively running as root)
        if (uid != 0 || gid != 0) {
                ERROR(jail->pakfire, "Child process is not running as root\n");
                return 126;
        }

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

        // Change root (unless root is /)
        if (!pakfire_on_root(jail->pakfire)) {
                // Mount everything
                r = pakfire_mount_all(jail->pakfire);
                if (r)
                        return r;

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

                // Call chroot()
                r = chroot(root);
                if (r) {
                        ERROR(jail->pakfire, "chroot() to %s failed: %m\n", root);
                        return 1;
                }

                // Change directory to /
                r = chdir("/");
                if (r) {
                        ERROR(jail->pakfire, "chdir() after chroot() failed: %m\n");
                        return 1;
                }
        }

        // 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 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 output and error
        if (ctx->pipes.stdout[1] && ctx->pipes.stderr[1]) {
                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.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;

        // Drop capabilities
        r = pakfire_jail_drop_capabilities(jail);
        if (r)
                return r;

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

        // exec() command
        r = execvpe(argv[0], (char**)argv, jail->env);
        if (r < 0)
                ERROR(jail->pakfire, "Could not execve(): %m\n");

        // Translate errno into regular exit code
        switch (errno) {
                case ENOENT:
                        r = 127;
                        break;

                default:
                        r = 1;
        }

        // We should not get here
        return r;
}

// Run a command in the jail
static int __pakfire_jail_exec(struct pakfire_jail* jail, const char* argv[]) {
        int exit = -1;
        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 = {
                .pipes = {
                        .stdout   = { 0, 0 },
                        .stderr   = { 0, 0 },
                },
        };

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

        /*
                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 (!pakfire_jail_has_flag(jail, PAKFIRE_JAIL_INTERACTIVE)) {
                // 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 */

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

        // Launch the process in a cgroup that is a leaf of the configured cgroup
        if (jail->cgroup) {
                args.flags |= CLONE_INTO_CGROUP;

#warning TODO randomize the name

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

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

        // Fork this process
        ctx.pid = clone3(&args, sizeof(args));
        if (ctx.pid < 0) {
                ERROR(jail->pakfire, "Could not clone: %m\n");
                return -1;

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

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

        DEBUG(jail->pakfire, "Waiting for PID %d to finish its work\n", ctx.pid);

        // Read output of the child process
        r = pakfire_jail_wait(jail, &ctx);
        if (r)
                goto ERROR;

        // Handle exit status
        switch (ctx.status.si_code) {
                case CLD_EXITED:
                        DEBUG(jail->pakfire, "The child process exited with code %d\n",
                                ctx.status.si_status);

                        // Pass exit code
                        exit = ctx.status.si_status;
                        break;

                case CLD_KILLED:
                case CLD_DUMPED:
                        ERROR(jail->pakfire, "The child process was killed\n");
                        break;

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

ERROR:
        // Destroy the temporary cgroup (if any)
        if (ctx.cgroup) {
                pakfire_cgroup_destroy(ctx.cgroup);
                pakfire_cgroup_unref(ctx.cgroup);
        }

        // Close any file descriptors
        pakfire_jail_close_pipe(jail, ctx.pipes.stdout);
        pakfire_jail_close_pipe(jail, ctx.pipes.stderr);
        if (ctx.pidfd)
                close(ctx.pidfd);
        pakfire_jail_close_pipe(jail, ctx.pipes.log_INFO);
        pakfire_jail_close_pipe(jail, ctx.pipes.log_ERROR);
        pakfire_jail_close_pipe(jail, ctx.pipes.log_DEBUG);

        // Umount everything
        if (!pakfire_on_root(jail->pakfire))
                pakfire_umount_all(jail->pakfire);

        return exit;
}

PAKFIRE_EXPORT int pakfire_jail_exec(struct pakfire_jail* jail,
                const char* argv[], char** output) {
        int r;

        // Store logging callback
        pakfire_jail_log_callback log_callback = jail->log_callback;
        void* log_data = jail->log_data;

        // Capture output if requested by user
        if (output)
                pakfire_jail_set_log_callback(jail, pakfire_jail_capture_stdout, output);

        // Run exec()
        r = __pakfire_jail_exec(jail, argv);

        // Restore log callback
        pakfire_jail_set_log_callback(jail, log_callback, log_data);

        return r;
}

PAKFIRE_EXPORT int pakfire_jail_exec_script(struct pakfire_jail* jail,
                const char* script, const size_t size, const char* args[], char** output) {
        char path[PATH_MAX];
        const char** argv = NULL;
        int r;

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

        // Write the scriptlet to disk
        r = pakfire_path_join(path, root, "pakfire-script.XXXXXX");
        if (r < 0)
                goto ERROR;

        // Open a temporary file
        int fd = mkstemp(path);
        if (fd < 0) {
                ERROR(jail->pakfire, "Could not open a temporary file: %m\n");
                r = 1;
                goto ERROR;
        }

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

        // Write data
        ssize_t bytes_written = write(fd, script, size);
        if (bytes_written < (ssize_t)size) {
                ERROR(jail->pakfire, "Could not write script to file %s: %m\n", path);
                r = 1;
                goto ERROR;
        }

        // Make the script executable
        r = fchmod(fd, S_IRUSR|S_IWUSR|S_IXUSR);
        if (r) {
                ERROR(jail->pakfire, "Could not set executable permissions on %s: %m\n", path);
                goto ERROR;
        }

        // Close file
        r = close(fd);
        if (r) {
                ERROR(jail->pakfire, "Could not close script file %s: %m\n", path);
                r = 1;
                goto ERROR;
        }

        // 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, output);

ERROR:
        if (argv)
                free(argv);

        // 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, flags);
        if (r)
                goto ERROR;

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

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, char** output) {
        struct pakfire_jail* jail = NULL;
        int r;

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

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

ERROR:
        if (jail)
                pakfire_jail_unref(jail);

        return r;
}


int pakfire_jail_shell(struct pakfire* pakfire) {
        const char* argv[] = {
                "/bin/bash", "--login", NULL,
        };

        // Execute /bin/bash
        return pakfire_jail_run(pakfire, argv, PAKFIRE_JAIL_INTERACTIVE, NULL);
}

int pakfire_jail_ldconfig(struct pakfire* pakfire) {
        char path[PATH_MAX];

        const char* ldconfig = "/sbin/ldconfig";

        // Check if ldconfig exists before calling it to avoid overhead
        int r = pakfire_make_path(pakfire, path, ldconfig);
        if (r < 0)
                return 1;

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

        const char* argv[] = {
                ldconfig, NULL,
        };

        // Run ldconfig
        return pakfire_jail_run(pakfire, argv, 0, NULL);
}