build-sys: make readahead and vconsole optional

6 files changed, 1539 insertions, 0 deletions

diff --git a/src/readahead/readahead-collect.c b/src/readahead/readahead-collect.c
new file mode 100644
index 0000000000..eac11e7e5c
--- /dev/null
+++ b/src/readahead/readahead-collect.c
@@ -0,0 +1,692 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2010 Lennart Poettering
+
+  systemd 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 2 of the License, or
+  (at your option) any later version.
+
+  systemd 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 systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <errno.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <linux/limits.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/select.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <linux/fanotify.h>
+#include <sys/signalfd.h>
+#include <sys/poll.h>
+#include <sys/mman.h>
+#include <linux/fs.h>
+#include <linux/fiemap.h>
+#include <sys/ioctl.h>
+#include <sys/vfs.h>
+#include <getopt.h>
+#include <sys/inotify.h>
+
+#include "missing.h"
+#include "util.h"
+#include "set.h"
+#include "sd-daemon.h"
+#include "ioprio.h"
+#include "readahead-common.h"
+#include "virt.h"
+
+/* fixme:
+ *
+ * - detect ssd on btrfs/lvm...
+ * - read ahead directories
+ * - gzip?
+ * - remount rw?
+ * - handle files where nothing is in mincore
+ * - does ioprio_set work with fadvise()?
+ */
+
+static unsigned arg_files_max = 16*1024;
+static off_t arg_file_size_max = READAHEAD_FILE_SIZE_MAX;
+static usec_t arg_timeout = 2*USEC_PER_MINUTE;
+
+static ReadaheadShared *shared = NULL;
+
+/* Avoid collisions with the NULL pointer */
+#define SECTOR_TO_PTR(s) ULONG_TO_PTR((s)+1)
+#define PTR_TO_SECTOR(p) (PTR_TO_ULONG(p)-1)
+
+static int btrfs_defrag(int fd) {
+        struct btrfs_ioctl_vol_args data;
+
+        zero(data);
+        data.fd = fd;
+
+        return ioctl(fd, BTRFS_IOC_DEFRAG, &data);
+}
+
+static int pack_file(FILE *pack, const char *fn, bool on_btrfs) {
+        struct stat st;
+        void *start = MAP_FAILED;
+        uint8_t *vec;
+        uint32_t b, c;
+        size_t l, pages;
+        bool mapped;
+        int r = 0, fd = -1, k;
+
+        assert(pack);
+        assert(fn);
+
+        if ((fd = open(fn, O_RDONLY|O_CLOEXEC|O_NOATIME|O_NOCTTY|O_NOFOLLOW)) < 0) {
+
+                if (errno == ENOENT)
+                        return 0;
+
+                if (errno == EPERM || errno == EACCES)
+                        return 0;
+
+                log_warning("open(%s) failed: %m", fn);
+                r = -errno;
+                goto finish;
+        }
+
+        if ((k = file_verify(fd, fn, arg_file_size_max, &st)) <= 0) {
+                r = k;
+                goto finish;
+        }
+
+        if (on_btrfs)
+                btrfs_defrag(fd);
+
+        l = PAGE_ALIGN(st.st_size);
+        if ((start = mmap(NULL, l, PROT_READ, MAP_SHARED, fd, 0)) == MAP_FAILED) {
+                log_warning("mmap(%s) failed: %m", fn);
+                r = -errno;
+                goto finish;
+        }
+
+        pages = l / page_size();
+
+        vec = alloca(pages);
+        memset(vec, 0, pages);
+        if (mincore(start, l, vec) < 0) {
+                log_warning("mincore(%s) failed: %m", fn);
+                r = -errno;
+                goto finish;
+        }
+
+        fputs(fn, pack);
+        fputc('\n', pack);
+
+        mapped = false;
+        for (c = 0; c < pages; c++) {
+                bool new_mapped = !!(vec[c] & 1);
+
+                if (!mapped && new_mapped)
+                        b = c;
+                else if (mapped && !new_mapped) {
+                        fwrite(&b, sizeof(b), 1, pack);
+                        fwrite(&c, sizeof(c), 1, pack);
+
+                        log_debug("%s: page %u to %u", fn, b, c);
+                }
+
+                mapped = new_mapped;
+        }
+
+        /* We don't write any range data if we should read the entire file */
+        if (mapped && b > 0) {
+                fwrite(&b, sizeof(b), 1, pack);
+                fwrite(&c, sizeof(c), 1, pack);
+
+                log_debug("%s: page %u to %u", fn, b, c);
+        }
+
+        /* End marker */
+        b = 0;
+        fwrite(&b, sizeof(b), 1, pack);
+        fwrite(&b, sizeof(b), 1, pack);
+
+finish:
+        if (start != MAP_FAILED)
+                munmap(start, l);
+
+        if (fd >= 0)
+                close_nointr_nofail(fd);
+
+        return r;
+}
+
+static unsigned long fd_first_block(int fd) {
+        struct {
+                struct fiemap fiemap;
+                struct fiemap_extent extent;
+        } data;
+
+        zero(data);
+        data.fiemap.fm_length = ~0ULL;
+        data.fiemap.fm_extent_count = 1;
+
+        if (ioctl(fd, FS_IOC_FIEMAP, &data) < 0)
+                return 0;
+
+        if (data.fiemap.fm_mapped_extents <= 0)
+                return 0;
+
+        if (data.fiemap.fm_extents[0].fe_flags & FIEMAP_EXTENT_UNKNOWN)
+                return 0;
+
+        return (unsigned long) data.fiemap.fm_extents[0].fe_physical;
+}
+
+struct item {
+        const char *path;
+        unsigned long block;
+};
+
+static int qsort_compare(const void *a, const void *b) {
+        const struct item *i, *j;
+
+        i = a;
+        j = b;
+
+        if (i->block < j->block)
+                return -1;
+        if (i->block > j->block)
+                return 1;
+
+        return strcmp(i->path, j->path);
+}
+
+static int collect(const char *root) {
+        enum {
+                FD_FANOTIFY,  /* Get the actual fs events */
+                FD_SIGNAL,
+                FD_INOTIFY,   /* We get notifications to quit early via this fd */
+                _FD_MAX
+        };
+        struct pollfd pollfd[_FD_MAX];
+        int fanotify_fd = -1, signal_fd = -1, inotify_fd = -1, r = 0;
+        pid_t my_pid;
+        Hashmap *files = NULL;
+        Iterator i;
+        char *p, *q;
+        sigset_t mask;
+        FILE *pack = NULL;
+        char *pack_fn_new = NULL, *pack_fn = NULL;
+        bool on_ssd, on_btrfs;
+        struct statfs sfs;
+        usec_t not_after;
+
+        assert(root);
+
+        write_one_line_file("/proc/self/oom_score_adj", "1000");
+
+        if (ioprio_set(IOPRIO_WHO_PROCESS, getpid(), IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0)) < 0)
+                log_warning("Failed to set IDLE IO priority class: %m");
+
+        assert_se(sigemptyset(&mask) == 0);
+        sigset_add_many(&mask, SIGINT, SIGTERM, -1);
+        assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
+
+        if ((signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
+                log_error("signalfd(): %m");
+                r = -errno;
+                goto finish;
+        }
+
+        if (!(files = hashmap_new(string_hash_func, string_compare_func))) {
+                log_error("Failed to allocate set.");
+                r = -ENOMEM;
+                goto finish;
+        }
+
+        if ((fanotify_fd = fanotify_init(FAN_CLOEXEC|FAN_NONBLOCK, O_RDONLY|O_LARGEFILE|O_CLOEXEC|O_NOATIME)) < 0)  {
+                log_error("Failed to create fanotify object: %m");
+                r = -errno;
+                goto finish;
+        }
+
+        if (fanotify_mark(fanotify_fd, FAN_MARK_ADD|FAN_MARK_MOUNT, FAN_OPEN, AT_FDCWD, root) < 0) {
+                log_error("Failed to mark %s: %m", root);
+                r = -errno;
+                goto finish;
+        }
+
+        if ((inotify_fd = open_inotify()) < 0) {
+                r = inotify_fd;
+                goto finish;
+        }
+
+        not_after = now(CLOCK_MONOTONIC) + arg_timeout;
+
+        my_pid = getpid();
+
+        zero(pollfd);
+        pollfd[FD_FANOTIFY].fd = fanotify_fd;
+        pollfd[FD_FANOTIFY].events = POLLIN;
+        pollfd[FD_SIGNAL].fd = signal_fd;
+        pollfd[FD_SIGNAL].events = POLLIN;
+        pollfd[FD_INOTIFY].fd = inotify_fd;
+        pollfd[FD_INOTIFY].events = POLLIN;
+
+        sd_notify(0,
+                  "READY=1\n"
+                  "STATUS=Collecting readahead data");
+
+        log_debug("Collecting...");
+
+        if (access("/run/systemd/readahead/cancel", F_OK) >= 0) {
+                log_debug("Collection canceled");
+                r = -ECANCELED;
+                goto finish;
+        }
+
+        if (access("/run/systemd/readahead/done", F_OK) >= 0) {
+                log_debug("Got termination request");
+                goto done;
+        }
+
+        for (;;) {
+                union {
+                        struct fanotify_event_metadata metadata;
+                        char buffer[4096];
+                } data;
+                ssize_t n;
+                struct fanotify_event_metadata *m;
+                usec_t t;
+                int h;
+
+                if (hashmap_size(files) > arg_files_max) {
+                        log_debug("Reached maximum number of read ahead files, ending collection.");
+                        break;
+                }
+
+                t = now(CLOCK_MONOTONIC);
+                if (t >= not_after) {
+                        log_debug("Reached maximum collection time, ending collection.");
+                        break;
+                }
+
+                if ((h = poll(pollfd, _FD_MAX, (int) ((not_after - t) / USEC_PER_MSEC))) < 0) {
+
+                        if (errno == EINTR)
+                                continue;
+
+                        log_error("poll(): %m");
+                        r = -errno;
+                        goto finish;
+                }
+
+                if (h == 0) {
+                        log_debug("Reached maximum collection time, ending collection.");
+                        break;
+                }
+
+                if (pollfd[FD_SIGNAL].revents) {
+                        log_debug("Got signal.");
+                        break;
+                }
+
+                if (pollfd[FD_INOTIFY].revents) {
+                        uint8_t inotify_buffer[sizeof(struct inotify_event) + FILENAME_MAX];
+                        struct inotify_event *e;
+
+                        if ((n = read(inotify_fd, &inotify_buffer, sizeof(inotify_buffer))) < 0) {
+                                if (errno == EINTR || errno == EAGAIN)
+                                        continue;
+
+                                log_error("Failed to read inotify event: %m");
+                                r = -errno;
+                                goto finish;
+                        }
+
+                        e = (struct inotify_event*) inotify_buffer;
+                        while (n > 0) {
+                                size_t step;
+
+                                if ((e->mask & IN_CREATE) && streq(e->name, "cancel")) {
+                                        log_debug("Collection canceled");
+                                        r = -ECANCELED;
+                                        goto finish;
+                                }
+
+                                if ((e->mask & IN_CREATE) && streq(e->name, "done")) {
+                                        log_debug("Got termination request");
+                                        goto done;
+                                }
+
+                                step = sizeof(struct inotify_event) + e->len;
+                                assert(step <= (size_t) n);
+
+                                e = (struct inotify_event*) ((uint8_t*) e + step);
+                                n -= step;
+                        }
+                }
+
+                if ((n = read(fanotify_fd, &data, sizeof(data))) < 0) {
+
+                        if (errno == EINTR || errno == EAGAIN)
+                                continue;
+
+                        /* fanotify sometimes returns EACCES on read()
+                         * where it shouldn't. For now let's just
+                         * ignore it here (which is safe), but
+                         * eventually this should be
+                         * dropped when the kernel is fixed.
+                         *
+                         * https://bugzilla.redhat.com/show_bug.cgi?id=707577 */
+                        if (errno == EACCES)
+                                continue;
+
+                        log_error("Failed to read event: %m");
+                        r = -errno;
+                        goto finish;
+                }
+
+                for (m = &data.metadata; FAN_EVENT_OK(m, n); m = FAN_EVENT_NEXT(m, n)) {
+                        char fn[PATH_MAX];
+                        int k;
+
+                        if (m->fd < 0)
+                                goto next_iteration;
+
+                        if (m->pid == my_pid)
+                                goto next_iteration;
+
+                        __sync_synchronize();
+                        if (m->pid == shared->replay)
+                                goto next_iteration;
+
+                        snprintf(fn, sizeof(fn), "/proc/self/fd/%i", m->fd);
+                        char_array_0(fn);
+
+                        if ((k = readlink_malloc(fn, &p)) >= 0) {
+                                if (startswith(p, "/tmp") ||
+                                    endswith(p, " (deleted)") ||
+                                    hashmap_get(files, p))
+                                        /* Not interesting, or
+                                         * already read */
+                                        free(p);
+                                else {
+                                        unsigned long ul;
+
+                                        ul = fd_first_block(m->fd);
+
+                                        if ((k = hashmap_put(files, p, SECTOR_TO_PTR(ul))) < 0) {
+                                                log_warning("set_put() failed: %s", strerror(-k));
+                                                free(p);
+                                        }
+                                }
+
+                        } else
+                                log_warning("readlink(%s) failed: %s", fn, strerror(-k));
+
+                next_iteration:
+                        if (m->fd)
+                                close_nointr_nofail(m->fd);
+                }
+        }
+
+done:
+        if (fanotify_fd >= 0) {
+                close_nointr_nofail(fanotify_fd);
+                fanotify_fd = -1;
+        }
+
+        log_debug("Writing Pack File...");
+
+        on_ssd = fs_on_ssd(root) > 0;
+        log_debug("On SSD: %s", yes_no(on_ssd));
+
+        on_btrfs = statfs(root, &sfs) >= 0 && (long) sfs.f_type == (long) BTRFS_SUPER_MAGIC;
+        log_debug("On btrfs: %s", yes_no(on_btrfs));
+
+        asprintf(&pack_fn, "%s/.readahead", root);
+        asprintf(&pack_fn_new, "%s/.readahead.new", root);
+
+        if (!pack_fn || !pack_fn_new) {
+                log_error("Out of memory");
+                r = -ENOMEM;
+                goto finish;
+        }
+
+        if (!(pack = fopen(pack_fn_new, "we"))) {
+                log_error("Failed to open pack file: %m");
+                r = -errno;
+                goto finish;
+        }
+
+        fputs(CANONICAL_HOST "\n", pack);
+        putc(on_ssd ? 'S' : 'R', pack);
+
+        if (on_ssd || on_btrfs) {
+
+                /* On SSD or on btrfs, just write things out in the
+                 * order the files were accessed. */
+
+                HASHMAP_FOREACH_KEY(q, p, files, i)
+                        pack_file(pack, p, on_btrfs);
+        } else {
+                struct item *ordered, *j;
+                unsigned k, n;
+
+                /* On rotating media, order things by the block
+                 * numbers */
+
+                log_debug("Ordering...");
+
+                n = hashmap_size(files);
+                if (!(ordered = new(struct item, n))) {
+                        log_error("Out of memory");
+                        r = -ENOMEM;
+                        goto finish;
+                }
+
+                j = ordered;
+                HASHMAP_FOREACH_KEY(q, p, files, i) {
+                        j->path = p;
+                        j->block = PTR_TO_SECTOR(q);
+                        j++;
+                }
+
+                assert(ordered + n == j);
+
+                qsort(ordered, n, sizeof(struct item), qsort_compare);
+
+                for (k = 0; k < n; k++)
+                        pack_file(pack, ordered[k].path, on_btrfs);
+
+                free(ordered);
+        }
+
+        log_debug("Finalizing...");
+
+        fflush(pack);
+
+        if (ferror(pack)) {
+                log_error("Failed to write pack file.");
+                r = -EIO;
+                goto finish;
+        }
+
+        if (rename(pack_fn_new, pack_fn) < 0) {
+                log_error("Failed to rename readahead file: %m");
+                r = -errno;
+                goto finish;
+        }
+
+        fclose(pack);
+        pack = NULL;
+
+        log_debug("Done.");
+
+finish:
+        if (fanotify_fd >= 0)
+                close_nointr_nofail(fanotify_fd);
+
+        if (signal_fd >= 0)
+                close_nointr_nofail(signal_fd);
+
+        if (inotify_fd >= 0)
+                close_nointr_nofail(inotify_fd);
+
+        if (pack) {
+                fclose(pack);
+                unlink(pack_fn_new);
+        }
+
+        free(pack_fn_new);
+        free(pack_fn);
+
+        while ((p = hashmap_steal_first_key(files)))
+                free(p);
+
+        hashmap_free(files);
+
+        return r;
+}
+
+static int help(void) {
+
+        printf("%s [OPTIONS...] [DIRECTORY]\n\n"
+               "Collect read-ahead data on early boot.\n\n"
+               "  -h --help                 Show this help\n"
+               "     --max-files=INT        Maximum number of files to read ahead\n"
+               "     --max-file-size=BYTES  Maximum size of files to read ahead\n"
+               "     --timeout=USEC         Maximum time to spend collecting data\n",
+               program_invocation_short_name);
+
+        return 0;
+}
+
+static int parse_argv(int argc, char *argv[]) {
+
+        enum {
+                ARG_FILES_MAX = 0x100,
+                ARG_FILE_SIZE_MAX,
+                ARG_TIMEOUT
+        };
+
+        static const struct option options[] = {
+                { "help",          no_argument,       NULL, 'h'                },
+                { "files-max",     required_argument, NULL, ARG_FILES_MAX      },
+                { "file-size-max", required_argument, NULL, ARG_FILE_SIZE_MAX  },
+                { "timeout",       required_argument, NULL, ARG_TIMEOUT        },
+                { NULL,            0,                 NULL, 0                  }
+        };
+
+        int c;
+
+        assert(argc >= 0);
+        assert(argv);
+
+        while ((c = getopt_long(argc, argv, "h", options, NULL)) >= 0) {
+
+                switch (c) {
+
+                case 'h':
+                        help();
+                        return 0;
+
+                case ARG_FILES_MAX:
+                        if (safe_atou(optarg, &arg_files_max) < 0 || arg_files_max <= 0) {
+                                log_error("Failed to parse maximum number of files %s.", optarg);
+                                return -EINVAL;
+                        }
+                        break;
+
+                case ARG_FILE_SIZE_MAX: {
+                        unsigned long long ull;
+
+                        if (safe_atollu(optarg, &ull) < 0 || ull <= 0) {
+                                log_error("Failed to parse maximum file size %s.", optarg);
+                                return -EINVAL;
+                        }
+
+                        arg_file_size_max = (off_t) ull;
+                        break;
+                }
+
+                case ARG_TIMEOUT:
+                        if (parse_usec(optarg, &arg_timeout) < 0 || arg_timeout <= 0) {
+                                log_error("Failed to parse timeout %s.", optarg);
+                                return -EINVAL;
+                        }
+
+                        break;
+
+                case '?':
+                        return -EINVAL;
+
+                default:
+                        log_error("Unknown option code %c", c);
+                        return -EINVAL;
+                }
+        }
+
+        if (optind != argc &&
+            optind != argc-1) {
+                help();
+                return -EINVAL;
+        }
+
+        return 1;
+}
+
+int main(int argc, char *argv[]) {
+        int r;
+        const char *root;
+
+        log_set_target(LOG_TARGET_SYSLOG_OR_KMSG);
+        log_parse_environment();
+        log_open();
+
+        umask(0022);
+
+        if ((r = parse_argv(argc, argv)) <= 0)
+                return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+
+        root = optind < argc ? argv[optind] : "/";
+
+        if (fs_on_read_only(root) > 0) {
+                log_info("Disabling readahead collector due to read-only media.");
+                return 0;
+        }
+
+        if (!enough_ram()) {
+                log_info("Disabling readahead collector due to low memory.");
+                return 0;
+        }
+
+        if (detect_virtualization(NULL) > 0) {
+                log_info("Disabling readahead collector due to execution in virtualized environment.");
+                return 0;
+        }
+
+        if (!(shared = shared_get()))
+                return 1;
+
+        shared->collect = getpid();
+        __sync_synchronize();
+
+        if (collect(root) < 0)
+                return 1;
+
+        return 0;
+}
diff --git a/src/readahead/readahead-common.c b/src/readahead/readahead-common.c
new file mode 100644
index 0000000000..67214ec379
--- /dev/null
+++ b/src/readahead/readahead-common.c
@@ -0,0 +1,269 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2010 Lennart Poettering
+
+  systemd 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 2 of the License, or
+  (at your option) any later version.
+
+  systemd 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 systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/sysinfo.h>
+#include <sys/inotify.h>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <unistd.h>
+#include <libudev.h>
+
+#include "log.h"
+#include "readahead-common.h"
+#include "util.h"
+
+int file_verify(int fd, const char *fn, off_t file_size_max, struct stat *st) {
+        assert(fd >= 0);
+        assert(fn);
+        assert(st);
+
+        if (fstat(fd, st) < 0) {
+                log_warning("fstat(%s) failed: %m", fn);
+                return -errno;
+        }
+
+        if (!S_ISREG(st->st_mode)) {
+                log_debug("Not preloading special file %s", fn);
+                return 0;
+        }
+
+        if (st->st_size <= 0 || st->st_size > file_size_max) {
+                log_debug("Not preloading file %s with size out of bounds %llu", fn, (unsigned long long) st->st_size);
+                return 0;
+        }
+
+        return 1;
+}
+
+int fs_on_ssd(const char *p) {
+        struct stat st;
+        struct udev *udev = NULL;
+        struct udev_device *udev_device = NULL, *look_at = NULL;
+        bool b = false;
+        const char *devtype, *rotational, *model, *id;
+
+        assert(p);
+
+        if (stat(p, &st) < 0)
+                return -errno;
+
+        if (major(st.st_dev) == 0)
+                return false;
+
+        if (!(udev = udev_new()))
+                return -ENOMEM;
+
+        if (!(udev_device = udev_device_new_from_devnum(udev, 'b', st.st_dev)))
+                goto finish;
+
+        if ((devtype = udev_device_get_property_value(udev_device, "DEVTYPE")) &&
+            streq(devtype, "partition"))
+                look_at = udev_device_get_parent(udev_device);
+        else
+                look_at = udev_device;
+
+        if (!look_at)
+                goto finish;
+
+        /* First, try high-level property */
+        if ((id = udev_device_get_property_value(look_at, "ID_SSD"))) {
+                b = streq(id, "1");
+                goto finish;
+        }
+
+        /* Second, try kernel attribute */
+        if ((rotational = udev_device_get_sysattr_value(look_at, "queue/rotational")))
+                if ((b = streq(rotational, "0")))
+                        goto finish;
+
+        /* Finally, fallback to heuristics */
+        if (!(look_at = udev_device_get_parent(look_at)))
+                goto finish;
+
+        if ((model = udev_device_get_sysattr_value(look_at, "model")))
+                b = !!strstr(model, "SSD");
+
+finish:
+        if (udev_device)
+                udev_device_unref(udev_device);
+
+        if (udev)
+                udev_unref(udev);
+
+        return b;
+}
+
+int fs_on_read_only(const char *p) {
+        struct stat st;
+        struct udev *udev = NULL;
+        struct udev_device *udev_device = NULL;
+        bool b = false;
+        const char *read_only;
+
+        assert(p);
+
+        if (stat(p, &st) < 0)
+                return -errno;
+
+        if (major(st.st_dev) == 0)
+                return false;
+
+        if (!(udev = udev_new()))
+                return -ENOMEM;
+
+        if (!(udev_device = udev_device_new_from_devnum(udev, 'b', st.st_dev)))
+                goto finish;
+
+        if ((read_only = udev_device_get_sysattr_value(udev_device, "ro")))
+                if ((b = streq(read_only, "1")))
+                        goto finish;
+
+finish:
+        if (udev_device)
+                udev_device_unref(udev_device);
+
+        if (udev)
+                udev_unref(udev);
+
+        return b;
+}
+
+bool enough_ram(void) {
+        struct sysinfo si;
+
+        assert_se(sysinfo(&si) >= 0);
+
+        /* Enable readahead only with at least 128MB memory */
+        return si.totalram > 127 * 1024*1024 / si.mem_unit;
+}
+
+int open_inotify(void) {
+        int fd;
+
+        if ((fd = inotify_init1(IN_CLOEXEC|IN_NONBLOCK)) < 0) {
+                log_error("Failed to create inotify handle: %m");
+                return -errno;
+        }
+
+        mkdir("/run/systemd", 0755);
+        mkdir("/run/systemd/readahead", 0755);
+
+        if (inotify_add_watch(fd, "/run/systemd/readahead", IN_CREATE) < 0) {
+                log_error("Failed to watch /run/systemd/readahead: %m");
+                close_nointr_nofail(fd);
+                return -errno;
+        }
+
+        return fd;
+}
+
+ReadaheadShared *shared_get(void) {
+        int fd;
+        ReadaheadShared *m = NULL;
+
+        mkdir("/run/systemd", 0755);
+        mkdir("/run/systemd/readahead", 0755);
+
+        if ((fd = open("/run/systemd/readahead/shared", O_CREAT|O_RDWR|O_CLOEXEC, 0644)) < 0) {
+                log_error("Failed to create shared memory segment: %m");
+                goto finish;
+        }
+
+        if (ftruncate(fd, sizeof(ReadaheadShared)) < 0) {
+                log_error("Failed to truncate shared memory segment: %m");
+                goto finish;
+        }
+
+        if ((m = mmap(NULL, sizeof(ReadaheadShared), PROT_WRITE|PROT_READ, MAP_SHARED, fd, 0)) == MAP_FAILED) {
+                log_error("Failed to mmap shared memory segment: %m");
+                m = NULL;
+                goto finish;
+        }
+
+finish:
+        if (fd >= 0)
+                close_nointr_nofail(fd);
+
+        return m;
+}
+
+#define BUMP_REQUEST_NR (16*1024)
+
+int bump_request_nr(const char *p) {
+        struct stat st;
+        uint64_t u;
+        char *ap = NULL, *line = NULL;
+        int r;
+        dev_t d;
+
+        assert(p);
+
+        if (stat(p, &st) < 0)
+                return -errno;
+
+        if (major(st.st_dev) == 0)
+                return 0;
+
+        d = st.st_dev;
+        block_get_whole_disk(d, &d);
+
+        if (asprintf(&ap, "/sys/dev/block/%u:%u/queue/nr_requests", major(d), minor(d)) < 0) {
+                r= -ENOMEM;
+                goto finish;
+        }
+
+        r = read_one_line_file(ap, &line);
+        if (r < 0) {
+                if (r == -ENOENT)
+                        r = 0;
+                goto finish;
+        }
+
+        r = safe_atou64(line, &u);
+        if (r >= 0 && u >= BUMP_REQUEST_NR) {
+                r = 0;
+                goto finish;
+        }
+
+        free(line);
+        line = NULL;
+
+        if (asprintf(&line, "%lu", (unsigned long) BUMP_REQUEST_NR) < 0) {
+                r = -ENOMEM;
+                goto finish;
+        }
+
+        r = write_one_line_file(ap, line);
+        if (r < 0)
+                goto finish;
+
+        log_info("Bumped block_nr parameter of %u:%u to %lu. This is a temporary hack and should be removed one day.", major(d), minor(d), (unsigned long) BUMP_REQUEST_NR);
+        r = 1;
+
+finish:
+        free(ap);
+        free(line);
+
+        return r;
+}
diff --git a/src/readahead/readahead-common.h b/src/readahead/readahead-common.h
new file mode 100644
index 0000000000..9547ad201c
--- /dev/null
+++ b/src/readahead/readahead-common.h
@@ -0,0 +1,50 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef fooreadaheadcommonhfoo
+#define fooreadaheadcommonhfoo
+
+/***
+  This file is part of systemd.
+
+  Copyright 2010 Lennart Poettering
+
+  systemd 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 2 of the License, or
+  (at your option) any later version.
+
+  systemd 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 systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include "macro.h"
+
+#define READAHEAD_FILE_SIZE_MAX (10*1024*1024)
+
+int file_verify(int fd, const char *fn, off_t file_size_max, struct stat *st);
+
+int fs_on_ssd(const char *p);
+int fs_on_read_only(const char *p);
+
+bool enough_ram(void);
+
+int open_inotify(void);
+
+typedef struct ReadaheadShared {
+        pid_t collect;
+        pid_t replay;
+} _packed_ ReadaheadShared;
+
+ReadaheadShared *shared_get(void);
+
+int bump_request_nr(const char *p);
+
+#endif
diff --git a/src/readahead/readahead-replay.c b/src/readahead/readahead-replay.c
new file mode 100644
index 0000000000..65011ac4be
--- /dev/null
+++ b/src/readahead/readahead-replay.c
@@ -0,0 +1,371 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2010 Lennart Poettering
+
+  systemd 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 2 of the License, or
+  (at your option) any later version.
+
+  systemd 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 systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <errno.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <linux/limits.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/select.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <getopt.h>
+#include <sys/inotify.h>
+
+#include "missing.h"
+#include "util.h"
+#include "set.h"
+#include "sd-daemon.h"
+#include "ioprio.h"
+#include "readahead-common.h"
+#include "virt.h"
+
+static off_t arg_file_size_max = READAHEAD_FILE_SIZE_MAX;
+
+static ReadaheadShared *shared = NULL;
+
+static int unpack_file(FILE *pack) {
+        char fn[PATH_MAX];
+        int r = 0, fd = -1;
+        bool any = false;
+        struct stat st;
+
+        assert(pack);
+
+        if (!fgets(fn, sizeof(fn), pack))
+                return 0;
+
+        char_array_0(fn);
+        truncate_nl(fn);
+
+        if ((fd = open(fn, O_RDONLY|O_CLOEXEC|O_NOATIME|O_NOCTTY|O_NOFOLLOW)) < 0) {
+
+                if (errno != ENOENT && errno != EPERM && errno != EACCES)
+                        log_warning("open(%s) failed: %m", fn);
+
+        } else if (file_verify(fd, fn, arg_file_size_max, &st) <= 0) {
+                close_nointr_nofail(fd);
+                fd = -1;
+        }
+
+        for (;;) {
+                uint32_t b, c;
+
+                if (fread(&b, sizeof(b), 1, pack) != 1 ||
+                    fread(&c, sizeof(c), 1, pack) != 1) {
+                        log_error("Premature end of pack file.");
+                        r = -EIO;
+                        goto finish;
+                }
+
+                if (b == 0 && c == 0)
+                        break;
+
+                if (c <= b) {
+                        log_error("Invalid pack file.");
+                        r = -EIO;
+                        goto finish;
+                }
+
+                log_debug("%s: page %u to %u", fn, b, c);
+
+                any = true;
+
+                if (fd >= 0)
+                        if (posix_fadvise(fd, b * page_size(), (c - b) * page_size(), POSIX_FADV_WILLNEED) < 0) {
+                                log_warning("posix_fadvise() failed: %m");
+                                goto finish;
+                        }
+        }
+
+        if (!any && fd >= 0) {
+                /* if no range is encoded in the pack file this is
+                 * intended to mean that the whole file shall be
+                 * read */
+
+                if (posix_fadvise(fd, 0, st.st_size, POSIX_FADV_WILLNEED) < 0) {
+                        log_warning("posix_fadvise() failed: %m");
+                        goto finish;
+                }
+        }
+
+finish:
+        if (fd >= 0)
+                close_nointr_nofail(fd);
+
+        return r;
+}
+
+static int replay(const char *root) {
+        FILE *pack = NULL;
+        char line[LINE_MAX];
+        int r = 0;
+        char *pack_fn = NULL;
+        int c;
+        bool on_ssd, ready = false;
+        int prio;
+        int inotify_fd = -1;
+
+        assert(root);
+
+        write_one_line_file("/proc/self/oom_score_adj", "1000");
+        bump_request_nr(root);
+
+        if (asprintf(&pack_fn, "%s/.readahead", root) < 0) {
+                log_error("Out of memory");
+                r = -ENOMEM;
+                goto finish;
+        }
+
+        if ((!(pack = fopen(pack_fn, "re")))) {
+                if (errno == ENOENT)
+                        log_debug("No pack file found.");
+                else {
+                        log_error("Failed to open pack file: %m");
+                        r = -errno;
+                }
+
+                goto finish;
+        }
+
+        posix_fadvise(fileno(pack), 0, 0, POSIX_FADV_WILLNEED);
+
+        if ((inotify_fd = open_inotify()) < 0) {
+                r = inotify_fd;
+                goto finish;
+        }
+
+        if (!(fgets(line, sizeof(line), pack))) {
+                log_error("Premature end of pack file.");
+                r = -EIO;
+                goto finish;
+        }
+
+        char_array_0(line);
+
+        if (!streq(line, CANONICAL_HOST "\n")) {
+                log_debug("Pack file host type mismatch.");
+                goto finish;
+        }
+
+        if ((c = getc(pack)) == EOF) {
+                log_debug("Premature end of pack file.");
+                r = -EIO;
+                goto finish;
+        }
+
+        /* We do not retest SSD here, so that we can start replaying
+         * before udev is up.*/
+        on_ssd = c == 'S';
+        log_debug("On SSD: %s", yes_no(on_ssd));
+
+        if (on_ssd)
+                prio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0);
+        else
+                prio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_RT, 7);
+
+        if (ioprio_set(IOPRIO_WHO_PROCESS, getpid(), prio) < 0)
+                log_warning("Failed to set IDLE IO priority class: %m");
+
+        sd_notify(0, "STATUS=Replaying readahead data");
+
+        log_debug("Replaying...");
+
+        if (access("/run/systemd/readahead/noreplay", F_OK) >= 0) {
+                log_debug("Got termination request");
+                goto done;
+        }
+
+        while (!feof(pack) && !ferror(pack)) {
+                uint8_t inotify_buffer[sizeof(struct inotify_event) + FILENAME_MAX];
+                int k;
+                ssize_t n;
+
+                if ((n = read(inotify_fd, &inotify_buffer, sizeof(inotify_buffer))) < 0) {
+                        if (errno != EINTR && errno != EAGAIN) {
+                                log_error("Failed to read inotify event: %m");
+                                r = -errno;
+                                goto finish;
+                        }
+                } else {
+                        struct inotify_event *e = (struct inotify_event*) inotify_buffer;
+
+                        while (n > 0) {
+                                size_t step;
+
+                                if ((e->mask & IN_CREATE) && streq(e->name, "noreplay")) {
+                                        log_debug("Got termination request");
+                                        goto done;
+                                }
+
+                                step = sizeof(struct inotify_event) + e->len;
+                                assert(step <= (size_t) n);
+
+                                e = (struct inotify_event*) ((uint8_t*) e + step);
+                                n -= step;
+                        }
+                }
+
+                if ((k = unpack_file(pack)) < 0) {
+                        r = k;
+                        goto finish;
+                }
+
+                if (!ready) {
+                        /* We delay the ready notification until we
+                         * queued at least one read */
+                        sd_notify(0, "READY=1");
+                        ready = true;
+                }
+        }
+
+done:
+        if (!ready)
+                sd_notify(0, "READY=1");
+
+        if (ferror(pack)) {
+                log_error("Failed to read pack file.");
+                r = -EIO;
+                goto finish;
+        }
+
+        log_debug("Done.");
+
+finish:
+        if (pack)
+                fclose(pack);
+
+        if (inotify_fd >= 0)
+                close_nointr_nofail(inotify_fd);
+
+        free(pack_fn);
+
+        return r;
+}
+
+
+static int help(void) {
+
+        printf("%s [OPTIONS...] [DIRECTORY]\n\n"
+               "Replay collected read-ahead data on early boot.\n\n"
+               "  -h --help                 Show this help\n"
+               "     --max-file-size=BYTES  Maximum size of files to read ahead\n",
+               program_invocation_short_name);
+
+        return 0;
+}
+
+static int parse_argv(int argc, char *argv[]) {
+
+        enum {
+                ARG_FILE_SIZE_MAX
+        };
+
+        static const struct option options[] = {
+                { "help",          no_argument,       NULL, 'h'                },
+                { "file-size-max", required_argument, NULL, ARG_FILE_SIZE_MAX  },
+                { NULL,            0,                 NULL, 0                  }
+        };
+
+        int c;
+
+        assert(argc >= 0);
+        assert(argv);
+
+        while ((c = getopt_long(argc, argv, "h", options, NULL)) >= 0) {
+
+                switch (c) {
+
+                case 'h':
+                        help();
+                        return 0;
+
+                case ARG_FILE_SIZE_MAX: {
+                        unsigned long long ull;
+
+                        if (safe_atollu(optarg, &ull) < 0 || ull <= 0) {
+                                log_error("Failed to parse maximum file size %s.", optarg);
+                                return -EINVAL;
+                        }
+
+                        arg_file_size_max = (off_t) ull;
+                        break;
+                }
+
+                case '?':
+                        return -EINVAL;
+
+                default:
+                        log_error("Unknown option code %c", c);
+                        return -EINVAL;
+                }
+        }
+
+        if (optind != argc &&
+            optind != argc-1) {
+                help();
+                return -EINVAL;
+        }
+
+        return 1;
+}
+
+int main(int argc, char*argv[]) {
+        int r;
+        const char *root;
+
+        log_set_target(LOG_TARGET_SYSLOG_OR_KMSG);
+        log_parse_environment();
+        log_open();
+
+        umask(0022);
+
+        if ((r = parse_argv(argc, argv)) <= 0)
+                return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+
+        root = optind < argc ? argv[optind] : "/";
+
+        if (!enough_ram()) {
+                log_info("Disabling readahead replay due to low memory.");
+                return 0;
+        }
+
+        if (detect_virtualization(NULL) > 0) {
+                log_info("Disabling readahead replay due to execution in virtualized environment.");
+                return 0;
+        }
+
+        if (!(shared = shared_get()))
+                return 1;
+
+        shared->replay = getpid();
+        __sync_synchronize();
+
+        if (replay(root) < 0)
+                return 1;
+
+        return 0;
+}
diff --git a/src/readahead/sd-readahead.c b/src/readahead/sd-readahead.c
new file mode 100644
index 0000000000..c5cfe67107
--- /dev/null
+++ b/src/readahead/sd-readahead.c
@@ -0,0 +1,76 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  Copyright 2010 Lennart Poettering
+
+  Permission is hereby granted, free of charge, to any person
+  obtaining a copy of this software and associated documentation files
+  (the "Software"), to deal in the Software without restriction,
+  including without limitation the rights to use, copy, modify, merge,
+  publish, distribute, sublicense, and/or sell copies of the Software,
+  and to permit persons to whom the Software is furnished to do so,
+  subject to the following conditions:
+
+  The above copyright notice and this permission notice shall be
+  included in all copies or substantial portions of the Software.
+
+  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+  BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+  ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+  CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+  SOFTWARE.
+***/
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+
+#include <unistd.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <fcntl.h>
+#include <string.h>
+
+#include "sd-readahead.h"
+
+static int touch(const char *path) {
+
+#if !defined(DISABLE_SYSTEMD) && defined(__linux__)
+        int fd;
+
+        mkdir("/run/systemd", 0755);
+        mkdir("/run/systemd/readahead", 0755);
+
+        if ((fd = open(path, O_WRONLY|O_CREAT|O_CLOEXEC|O_NOCTTY, 0666)) < 0)
+                return -errno;
+
+        for (;;) {
+                if (close(fd) >= 0)
+                        break;
+
+                if (errno != -EINTR)
+                        return -errno;
+        }
+
+#endif
+        return 0;
+}
+
+int sd_readahead(const char *action) {
+
+        if (!action)
+                return -EINVAL;
+
+        if (strcmp(action, "cancel") == 0)
+                return touch("/run/systemd/readahead/cancel");
+        else if (strcmp(action, "done") == 0)
+                return touch("/run/systemd/readahead/done");
+        else if (strcmp(action, "noreplay") == 0)
+                return touch("/run/systemd/readahead/noreplay");
+
+        return -EINVAL;
+}
diff --git a/src/readahead/sd-readahead.h b/src/readahead/sd-readahead.h
new file mode 100644
index 0000000000..5bf975a741
--- /dev/null
+++ b/src/readahead/sd-readahead.h
@@ -0,0 +1,81 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef foosdreadaheadhfoo
+#define foosdreadaheadhfoo
+
+/***
+  Copyright 2010 Lennart Poettering
+
+  Permission is hereby granted, free of charge, to any person
+  obtaining a copy of this software and associated documentation files
+  (the "Software"), to deal in the Software without restriction,
+  including without limitation the rights to use, copy, modify, merge,
+  publish, distribute, sublicense, and/or sell copies of the Software,
+  and to permit persons to whom the Software is furnished to do so,
+  subject to the following conditions:
+
+  The above copyright notice and this permission notice shall be
+  included in all copies or substantial portions of the Software.
+
+  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+  BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+  ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+  CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+  SOFTWARE.
+***/
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+  Reference implementation of a few boot readahead related
+  interfaces. These interfaces are trivial to implement. To simplify
+  porting we provide this reference implementation.  Applications are
+  welcome to reimplement the algorithms described here if they do not
+  want to include these two source files.
+
+  You may compile this with -DDISABLE_SYSTEMD to disable systemd
+  support. This makes all calls NOPs.
+
+  Since this is drop-in code we don't want any of our symbols to be
+  exported in any case. Hence we declare hidden visibility for all of
+  them.
+
+  You may find an up-to-date version of these source files online:
+
+  http://cgit.freedesktop.org/systemd/plain/src/sd-readahead.h
+  http://cgit.freedesktop.org/systemd/plain/src/sd-readahead.c
+
+  This should compile on non-Linux systems, too, but all functions
+  will become NOPs.
+
+  See sd-readahead(7) for more information.
+*/
+
+#ifndef _sd_hidden_
+#if (__GNUC__ >= 4) && !defined(SD_EXPORT_SYMBOLS)
+#define _sd_hidden_ __attribute__ ((visibility("hidden")))
+#else
+#define _sd_hidden_
+#endif
+#endif
+
+/*
+  Controls ongoing disk read-ahead operations during boot-up. The argument
+  must be a string, and either "cancel", "done" or "noreplay".
+
+  cancel = terminate read-ahead data collection, drop collected information
+  done = terminate read-ahead data collection, keep collected information
+  noreplay = terminate read-ahead replay
+*/
+int sd_readahead(const char *action) _sd_hidden_;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif