readahead: implement minimal readahead logic based on fanotify(), mincore() and readahead()

.gitignore

+systemd-readahead-collect
+systemd-readahead-replay
 systemd-reply-password
 systemd-ask-password-agent
 systemd-ask-password

Makefile.am

@@ -92,7 +92,9 @@ rootlibexec_PROGRAMS = \
 	systemd-remount-api-vfs \
 	systemd-kmsg-syslogd \
 	systemd-vconsole-setup \
-	systemd-reply-password
+	systemd-reply-password \
+	systemd-readahead-collect \
+	systemd-readahead-replay
 
 noinst_PROGRAMS = \
 	test-engine \
@@ -699,6 +701,30 @@ systemd_reply_password_SOURCES = \
 systemd_reply_password_LDADD = \
 	libsystemd-basic.la
 
+systemd_readahead_collect_SOURCES = \
+	src/readahead-collect.c \
+	src/sd-daemon.c \
+	src/readahead-common.c
+
+systemd_readahead_collect_CFLAGS = \
+	$(UDEV_CFLAGS)
+
+systemd_readahead_collect_LDADD = \
+	libsystemd-basic.la \
+	$(UDEV_LIBS)
+
+systemd_readahead_replay_SOURCES = \
+	src/readahead-replay.c \
+	src/sd-daemon.c \
+	src/readahead-common.c
+
+systemd_readahead_replay_CFLAGS = \
+	$(UDEV_CFLAGS)
+
+systemd_readahead_replay_LDADD = \
+	libsystemd-basic.la \
+	$(UDEV_LIBS)
+
 systemd_cgls_SOURCES = \
 	src/cgls.c \
 	src/cgroup-show.c \

configure.ac

@@ -27,6 +27,7 @@ AM_INIT_AUTOMAKE([foreign 1.11 -Wall -Wno-portability silent-rules tar-pax subdi
 AC_SUBST(PACKAGE_URL, [http://www.freedesktop.org/wiki/Software/systemd])
 
 AC_CANONICAL_HOST
+AC_DEFINE_UNQUOTED([CANONICAL_HOST], "$host", [Canonical host string.])
 
 AM_SILENT_RULES([yes])
 

src/hashmap.c

@@ -476,6 +476,21 @@ void* hashmap_steal_first(Hashmap *h) {
         return data;
 }
 
+void* hashmap_steal_first_key(Hashmap *h) {
+        void *key;
+
+        if (!h)
+                return NULL;
+
+        if (!h->iterate_list_head)
+                return NULL;
+
+        key = (void*) h->iterate_list_head->key;
+        remove_entry(h, h->iterate_list_head);
+
+        return key;
+}
+
 unsigned hashmap_size(Hashmap *h) {
 
         if (!h)

src/hashmap.h

@@ -72,6 +72,7 @@ void *hashmap_iterate_skip(Hashmap *h, const void *key, Iterator *i);
 
 void hashmap_clear(Hashmap *h);
 void *hashmap_steal_first(Hashmap *h);
+void *hashmap_steal_first_key(Hashmap *h);
 void* hashmap_first(Hashmap *h);
 void* hashmap_last(Hashmap *h);
 

src/linux/fanotify.h

+#ifndef _LINUX_FANOTIFY_H
+#define _LINUX_FANOTIFY_H
+
+#include <linux/types.h>
+
+/* the following events that user-space can register for */
+#define FAN_ACCESS		0x00000001	/* File was accessed */
+#define FAN_MODIFY		0x00000002	/* File was modified */
+#define FAN_CLOSE_WRITE		0x00000008	/* Unwrittable file closed */
+#define FAN_CLOSE_NOWRITE	0x00000010	/* Writtable file closed */
+#define FAN_OPEN		0x00000020	/* File was opened */
+
+#define FAN_EVENT_ON_CHILD	0x08000000	/* interested in child events */
+
+/* FIXME currently Q's have no limit.... */
+#define FAN_Q_OVERFLOW		0x00004000	/* Event queued overflowed */
+
+#define FAN_OPEN_PERM		0x00010000	/* File open in perm check */
+#define FAN_ACCESS_PERM		0x00020000	/* File accessed in perm check */
+
+/* helper events */
+#define FAN_CLOSE		(FAN_CLOSE_WRITE | FAN_CLOSE_NOWRITE) /* close */
+
+/* flags used for fanotify_init() */
+#define FAN_CLOEXEC		0x00000001
+#define FAN_NONBLOCK		0x00000002
+
+#define FAN_ALL_INIT_FLAGS	(FAN_CLOEXEC | FAN_NONBLOCK)
+
+/* flags used for fanotify_modify_mark() */
+#define FAN_MARK_ADD		0x00000001
+#define FAN_MARK_REMOVE		0x00000002
+#define FAN_MARK_DONT_FOLLOW	0x00000004
+#define FAN_MARK_ONLYDIR	0x00000008
+#define FAN_MARK_MOUNT		0x00000010
+#define FAN_MARK_IGNORED_MASK	0x00000020
+#define FAN_MARK_IGNORED_SURV_MODIFY	0x00000040
+#define FAN_MARK_FLUSH		0x00000080
+
+#define FAN_ALL_MARK_FLAGS	(FAN_MARK_ADD |\
+				 FAN_MARK_REMOVE |\
+				 FAN_MARK_DONT_FOLLOW |\
+				 FAN_MARK_ONLYDIR |\
+				 FAN_MARK_MOUNT |\
+				 FAN_MARK_IGNORED_MASK |\
+				 FAN_MARK_IGNORED_SURV_MODIFY)
+
+/*
+ * All of the events - we build the list by hand so that we can add flags in
+ * the future and not break backward compatibility.  Apps will get only the
+ * events that they originally wanted.  Be sure to add new events here!
+ */
+#define FAN_ALL_EVENTS (FAN_ACCESS |\
+			FAN_MODIFY |\
+			FAN_CLOSE |\
+			FAN_OPEN)
+
+/*
+ * All events which require a permission response from userspace
+ */
+#define FAN_ALL_PERM_EVENTS (FAN_OPEN_PERM |\
+			     FAN_ACCESS_PERM)
+
+#define FAN_ALL_OUTGOING_EVENTS	(FAN_ALL_EVENTS |\
+				 FAN_ALL_PERM_EVENTS |\
+				 FAN_Q_OVERFLOW)
+
+#define FANOTIFY_METADATA_VERSION	2
+
+struct fanotify_event_metadata {
+	__u32 event_len;
+	__u32 vers;
+	__u64 mask;
+	__s32 fd;
+	__s32 pid;
+} __attribute__ ((packed));
+
+struct fanotify_response {
+	__s32 fd;
+	__u32 response;
+} __attribute__ ((packed));
+
+/* Legit userspace responses to a _PERM event */
+#define FAN_ALLOW	0x01
+#define FAN_DENY	0x02
+
+/* Helper functions to deal with fanotify_event_metadata buffers */
+#define FAN_EVENT_METADATA_LEN (sizeof(struct fanotify_event_metadata))
+
+#define FAN_EVENT_NEXT(meta, len) ((len) -= (meta)->event_len, \
+				   (struct fanotify_event_metadata*)(((char *)(meta)) + \
+				   (meta)->event_len))
+
+#define FAN_EVENT_OK(meta, len)	((long)(len) >= (long)FAN_EVENT_METADATA_LEN && \
+				(long)(meta)->event_len >= (long)FAN_EVENT_METADATA_LEN && \
+				(long)(meta)->event_len <= (long)(len))
+
+#endif /* _LINUX_FANOTIFY_H */

src/macro.h

@@ -27,6 +27,8 @@
 #include <sys/uio.h>
 #include <inttypes.h>
 
+#define PAGE_SIZE 4096
+
 #define _printf_attr_(a,b) __attribute__ ((format (printf, a, b)))
 #define _sentinel_ __attribute__ ((sentinel))
 #define _noreturn_ __attribute__((noreturn))
@@ -49,6 +51,10 @@ static inline size_t ALIGN(size_t l) {
         return ((l + sizeof(void*) - 1) & ~(sizeof(void*) - 1));
 }
 
+static inline size_t PAGE_ALIGN(size_t l) {
+        return ((l + PAGE_SIZE - 1) & ~(PAGE_SIZE -1));
+}
+
 #define ELEMENTSOF(x) (sizeof(x)/sizeof((x)[0]))
 
 #define MAX(a,b)                                \

src/missing.h

@@ -76,4 +76,29 @@ static inline int pivot_root(const char *new_root, const char *put_old) {
         return syscall(SYS_pivot_root, new_root, put_old);
 }
 
+#ifdef __x86_64__
+#ifndef __NR_fanotify_init
+#define __NR_fanotify_init 300
+#endif
+#ifndef __NR_fanotify_mark
+#define __NR_fanotify_mark 301
+#endif
+#else
+#ifndef __NR_fanotify_init
+#define __NR_fanotify_init 338
+#endif
+#ifndef __NR_fanotify_mark
+#define __NR_fanotify_mark 339
+#endif
+#endif
+
+static inline int fanotify_init(unsigned int flags, unsigned int event_f_flags) {
+	return syscall(__NR_fanotify_init, flags, event_f_flags);
+}
+
+static inline int fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
+                                int dfd, const char *pathname) {
+	return syscall(__NR_fanotify_mark, fanotify_fd, flags, mask, dfd, pathname);
+}
+
 #endif

src/readahead-collect.c

+/*-*- 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 "missing.h"
+#include "util.h"
+#include "set.h"
+#include "sd-daemon.h"
+#include "ioprio.h"
+#include "readahead-common.h"
+
+/*
+   fixme:
+
+     - BTRFS_IOC_DEFRAG
+*/
+
+#define MINCORE_VEC_SIZE (READAHEAD_FILE_SIZE_MAX/PAGE_SIZE)
+
+static int pack_file(FILE *pack, const char *fn) {
+        struct stat st;
+        void *start = MAP_FAILED;
+        uint8_t vec[MINCORE_VEC_SIZE];
+        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) {
+                log_warning("open(%s) failed: %m", fn);
+                r = -errno;
+                goto finish;
+        }
+
+        if ((k = file_verify(fd, fn, &st)) <= 0) {
+                r = k;
+                goto finish;
+        }
+
+        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;
+        }
+
+        if (mincore(start, l, vec) < 0) {
+                log_warning("mincore(%s) failed: %m", fn);
+                r = -errno;
+                goto finish;
+        }
+
+        fputs(fn, pack);
+        fputc('\n', pack);
+
+        pages = l / PAGE_SIZE;
+        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,
+                FD_SIGNAL,
+                _FD_MAX
+        };
+        struct pollfd pollfd[_FD_MAX];
+        int fanotify_fd = -1, signal_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;
+
+        assert(root);
+
+        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, 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;
+        }
+
+        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;
+
+        sd_notify(0,
+                  "READY=1\n"
+                  "STATUS=Collecting readahead data");
+
+        log_debug("Collecting...");
+
+        for (;;) {
+                union {
+                        struct fanotify_event_metadata metadata;
+                        char buffer[4096];
+                } data;
+                ssize_t n;
+                struct fanotify_event_metadata *m;
+
+                if (poll(pollfd, _FD_MAX, -1) < 0) {
+
+                        if (errno == EINTR)
+                                continue;
+
+                        log_error("poll(): %m");
+                        r = -errno;
+                        goto finish;
+                }
+
+                if (pollfd[FD_SIGNAL].revents != 0)
+                        break;
+
+                if ((n = read(fanotify_fd, &data, sizeof(data))) < 0) {
+
+                        if (errno == EINTR || errno == EAGAIN)
+                                continue;
+
+                        log_error("Failed to read event: %m");
+                        r = -errno;
+                        goto finish;
+                }
+
+                m = &data.metadata;
+                while (FAN_EVENT_OK(m, n)) {
+
+                        if (m->pid != my_pid && m->fd >= 0) {
+                                char fn[PATH_MAX];
+                                int k;
+
+                                snprintf(fn, sizeof(fn), "/proc/self/fd/%i", m->fd);
+                                char_array_0(fn);
+
+                                if ((k = readlink_malloc(fn, &p)) >= 0) {
+
+                                        if (hashmap_get(files, p))
+                                                /* Already read */
+                                                free(p);
+                                        else {
+                                                unsigned long ul;
+
+                                                ul = fd_first_block(m->fd);
+
+                                                if ((k = hashmap_put(files, p, ULONG_TO_PTR(ul))) < 0) {
+
+                                                        if (k != -EEXIST)
+                                                                log_warning("set_put() failed: %s", strerror(-k));
+
+                                                        free(p);
+                                                }
+                                        }
+
+                                } else
+                                        log_warning("readlink(%s) failed: %s", fn, strerror(-k));
+                        }
+
+                        if (m->fd)
+                                close_nointr_nofail(m->fd);
+
+                        m = FAN_EVENT_NEXT(m, n);
+                }
+
+        }
+
+        if (fanotify_fd >= 0) {
+                close_nointr_nofail(fanotify_fd);
+                fanotify_fd = -1;
+        }
+
+        log_debug("Writing Pack File...");
+
+        on_ssd = fs_on_ssd(root);
+        log_debug("On SSD: %s", yes_no(on_ssd));
+
+        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 SSD, just write things out in the order the
+                 * files where accessed */
+
+                HASHMAP_FOREACH_KEY(q, p, files, i)
+                        pack_file(pack, p);
+        } 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_ULONG(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);
+
+                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;