[thirdparty/systemd.git] / src / journal / journal-vacuum.c

/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/

/***
  This file is part of systemd.

  Copyright 2011 Lennart Poettering

  systemd is free software; you can redistribute it and/or modify it
  under the terms of the GNU Lesser General Public License as published by
  the Free Software Foundation; either version 2.1 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
  Lesser General Public License for more details.

  You should have received a copy of the GNU Lesser General Public License
  along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/

#include <sys/types.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <sys/statvfs.h>
#include <unistd.h>

#ifdef HAVE_XATTR
#include <attr/xattr.h>
#endif

#include "journal-def.h"
#include "journal-file.h"
#include "journal-vacuum.h"
#include "sd-id128.h"
#include "util.h"

struct vacuum_info {
        uint64_t usage;
        char *filename;

        uint64_t realtime;
        sd_id128_t seqnum_id;
        uint64_t seqnum;

        bool have_seqnum;
};

static int vacuum_compare(const void *_a, const void *_b) {
        const struct vacuum_info *a, *b;

        a = _a;
        b = _b;

        if (a->have_seqnum && b->have_seqnum &&
            sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
                if (a->seqnum < b->seqnum)
                        return -1;
                else if (a->seqnum > b->seqnum)
                        return 1;
                else
                        return 0;
        }

        if (a->realtime < b->realtime)
                return -1;
        else if (a->realtime > b->realtime)
                return 1;
        else if (a->have_seqnum && b->have_seqnum)
                return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
        else
                return strcmp(a->filename, b->filename);
}

static void patch_realtime(
                const char *dir,
                const char *fn,
                const struct stat *st,
                unsigned long long *realtime) {

        usec_t x;

#ifdef HAVE_XATTR
        uint64_t crtime;
        _cleanup_free_ const char *path = NULL;
#endif

        /* The timestamp was determined by the file name, but let's
         * see if the file might actually be older than the file name
         * suggested... */

        assert(dir);
        assert(fn);
        assert(st);
        assert(realtime);

        x = timespec_load(&st->st_ctim);
        if (x > 0 && x != (usec_t) -1 && x < *realtime)
                *realtime = x;

        x = timespec_load(&st->st_atim);
        if (x > 0 && x != (usec_t) -1 && x < *realtime)
                *realtime = x;

        x = timespec_load(&st->st_mtim);
        if (x > 0 && x != (usec_t) -1 && x < *realtime)
                *realtime = x;

#ifdef HAVE_XATTR
        /* Let's read the original creation time, if possible. Ideally
         * we'd just query the creation time the FS might provide, but
         * unfortunately there's currently no sane API to query
         * it. Hence let's implement this manually... */

        /* Unfortunately there is is not fgetxattrat(), so we need to
         * go via path here. :-( */

        path = strjoin(dir, "/", fn, NULL);
        if (!path)
                return;

        if (getxattr(path, "user.crtime_usec", &crtime, sizeof(crtime)) == sizeof(crtime)) {
                crtime = le64toh(crtime);

                if (crtime > 0 && crtime != (uint64_t) -1 && crtime < *realtime)
                        *realtime = crtime;
        }
#endif
}

static int journal_file_empty(int dir_fd, const char *name) {
        int r;
        le64_t n_entries;
        _cleanup_close_ int fd;

        fd = openat(dir_fd, name, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
        if (fd < 0)
                return -errno;

        if (lseek(fd, offsetof(Header, n_entries), SEEK_SET) < 0)
                return -errno;

        r = read(fd, &n_entries, sizeof(n_entries));
        if (r != sizeof(n_entries))
                return r == 0 ? -EINVAL : -errno;

        return le64toh(n_entries) == 0;
}

int journal_directory_vacuum(
                const char *directory,
                uint64_t max_use,
                usec_t max_retention_usec,
                usec_t *oldest_usec) {

        _cleanup_closedir_ DIR *d = NULL;
        int r = 0;
        struct vacuum_info *list = NULL;
        unsigned n_list = 0, i;
        size_t n_allocated = 0;
        uint64_t sum = 0, freed = 0;
        usec_t retention_limit = 0;

        assert(directory);

        if (max_use <= 0 && max_retention_usec <= 0)
                return 0;

        if (max_retention_usec > 0) {
                retention_limit = now(CLOCK_REALTIME);
                if (retention_limit > max_retention_usec)
                        retention_limit -= max_retention_usec;
                else
                        max_retention_usec = retention_limit = 0;
        }

        d = opendir(directory);
        if (!d)
                return -errno;

        for (;;) {
                struct dirent *de;
                size_t q;
                struct stat st;
                char *p;
                unsigned long long seqnum = 0, realtime;
                sd_id128_t seqnum_id;
                bool have_seqnum;

                errno = 0;
                de = readdir(d);
                if (!de && errno != 0) {
                        r = -errno;
                        goto finish;
                }

                if (!de)
                        break;

                if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
                        continue;

                if (!S_ISREG(st.st_mode))
                        continue;

                q = strlen(de->d_name);

                if (endswith(de->d_name, ".journal")) {

                        /* Vacuum archived files */

                        if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
                                continue;

                        if (de->d_name[q-8-16-1] != '-' ||
                            de->d_name[q-8-16-1-16-1] != '-' ||
                            de->d_name[q-8-16-1-16-1-32-1] != '@')
                                continue;

                        p = strdup(de->d_name);
                        if (!p) {
                                r = -ENOMEM;
                                goto finish;
                        }

                        de->d_name[q-8-16-1-16-1] = 0;
                        if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
                                free(p);
                                continue;
                        }

                        if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
                                free(p);
                                continue;
                        }

                        have_seqnum = true;

                } else if (endswith(de->d_name, ".journal~")) {
                        unsigned long long tmp;

                        /* Vacuum corrupted files */

                        if (q < 1 + 16 + 1 + 16 + 8 + 1)
                                continue;

                        if (de->d_name[q-1-8-16-1] != '-' ||
                            de->d_name[q-1-8-16-1-16-1] != '@')
                                continue;

                        p = strdup(de->d_name);
                        if (!p) {
                                r = -ENOMEM;
                                goto finish;
                        }

                        if (sscanf(de->d_name + q-1-8-16-1-16, "%16llx-%16llx.journal~", &realtime, &tmp) != 2) {
                                free(p);
                                continue;
                        }

                        have_seqnum = false;
                } else
                        /* We do not vacuum active files or unknown files! */
                        continue;

                if (journal_file_empty(dirfd(d), p)) {
                        /* Always vacuum empty non-online files. */

                        uint64_t size = 512UL * (uint64_t) st.st_blocks;

                        if (unlinkat(dirfd(d), p, 0) >= 0) {
                                log_info("Deleted empty journal %s/%s (%"PRIu64" bytes).",
                                         directory, p, size);
                                freed += size;
                        } else if (errno != ENOENT)
                                log_warning("Failed to delete %s/%s: %m", directory, p);

                        free(p);

                        continue;
                }

                patch_realtime(directory, p, &st, &realtime);

                GREEDY_REALLOC(list, n_allocated, n_list + 1);

                list[n_list].filename = p;
                list[n_list].usage = 512UL * (uint64_t) st.st_blocks;
                list[n_list].seqnum = seqnum;
                list[n_list].realtime = realtime;
                list[n_list].seqnum_id = seqnum_id;
                list[n_list].have_seqnum = have_seqnum;

                sum += list[n_list].usage;

                n_list ++;
        }

        qsort_safe(list, n_list, sizeof(struct vacuum_info), vacuum_compare);

        for (i = 0; i < n_list; i++) {
                struct statvfs ss;

                if (fstatvfs(dirfd(d), &ss) < 0) {
                        r = -errno;
                        goto finish;
                }

                if ((max_retention_usec <= 0 || list[i].realtime >= retention_limit) &&
                    (max_use <= 0 || sum <= max_use))
                        break;

                if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
                        log_debug("Deleted archived journal %s/%s (%"PRIu64" bytes).",
                                  directory, list[i].filename, list[i].usage);
                        freed += list[i].usage;

                        if (list[i].usage < sum)
                                sum -= list[i].usage;
                        else
                                sum = 0;

                } else if (errno != ENOENT)
                        log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
        }

        if (oldest_usec && i < n_list && (*oldest_usec == 0 || list[i].realtime < *oldest_usec))
                *oldest_usec = list[i].realtime;

finish:
        for (i = 0; i < n_list; i++)
                free(list[i].filename);
        free(list);

        log_debug("Vacuuming done, freed %"PRIu64" bytes", freed);

        return r;
}
Commit	Line	Data
0284adc6 LP	1	/-- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil --/
	2
	3	/***
	4	This file is part of systemd.
	5
	6	Copyright 2011 Lennart Poettering
	7
	8	systemd is free software; you can redistribute it and/or modify it
	9	under the terms of the GNU Lesser General Public License as published by
	10	the Free Software Foundation; either version 2.1 of the License, or
	11	(at your option) any later version.
	12
	13	systemd is distributed in the hope that it will be useful, but
	14	WITHOUT ANY WARRANTY; without even the implied warranty of
	15	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	16	Lesser General Public License for more details.
	17
	18	You should have received a copy of the GNU Lesser General Public License
	19	along with systemd; If not, see <http://www.gnu.org/licenses/>.
	20	***/
	21
	22	#include <sys/types.h>
	23	#include <fcntl.h>
	24	#include <sys/stat.h>
	25	#include <sys/statvfs.h>
	26	#include <unistd.h>
	27
fb0951b0 LP	28	#ifdef HAVE_XATTR
	29	#include <attr/xattr.h>
	30	#endif
	31
0284adc6 LP	32	#include "journal-def.h"
	33	#include "journal-file.h"
	34	#include "journal-vacuum.h"
	35	#include "sd-id128.h"
	36	#include "util.h"
	37
	38	struct vacuum_info {
6c142648	39	uint64_t usage;
0284adc6 LP	40	char *filename;
	41
	42	uint64_t realtime;
	43	sd_id128_t seqnum_id;
	44	uint64_t seqnum;
	45
	46	bool have_seqnum;
	47	};
	48
	49	static int vacuum_compare(const void _a, const void _b) {
	50	const struct vacuum_info a, b;
	51
	52	a = _a;
	53	b = _b;
	54
	55	if (a->have_seqnum && b->have_seqnum &&
	56	sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
	57	if (a->seqnum < b->seqnum)
	58	return -1;
	59	else if (a->seqnum > b->seqnum)
	60	return 1;
	61	else
	62	return 0;
	63	}
	64
	65	if (a->realtime < b->realtime)
	66	return -1;
	67	else if (a->realtime > b->realtime)
	68	return 1;
	69	else if (a->have_seqnum && b->have_seqnum)
	70	return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
	71	else
	72	return strcmp(a->filename, b->filename);
	73	}
	74
fb0951b0 LP	75	static void patch_realtime(
	76	const char *dir,
	77	const char *fn,
	78	const struct stat *st,
	79	unsigned long long *realtime) {
	80
	81	usec_t x;
	82
	83	#ifdef HAVE_XATTR
	84	uint64_t crtime;
	85	_cleanup_free_ const char *path = NULL;
	86	#endif
	87
	88	/* The timestamp was determined by the file name, but let's
	89	* see if the file might actually be older than the file name
	90	* suggested... */
	91
	92	assert(dir);
	93	assert(fn);
	94	assert(st);
	95	assert(realtime);
	96
	97	x = timespec_load(&st->st_ctim);
	98	if (x > 0 && x != (usec_t) -1 && x < *realtime)
	99	*realtime = x;
	100
	101	x = timespec_load(&st->st_atim);
	102	if (x > 0 && x != (usec_t) -1 && x < *realtime)
	103	*realtime = x;
	104
	105	x = timespec_load(&st->st_mtim);
	106	if (x > 0 && x != (usec_t) -1 && x < *realtime)
	107	*realtime = x;
	108
	109	#ifdef HAVE_XATTR
	110	/* Let's read the original creation time, if possible. Ideally
	111	* we'd just query the creation time the FS might provide, but
	112	* unfortunately there's currently no sane API to query
	113	* it. Hence let's implement this manually... */
	114
	115	/* Unfortunately there is is not fgetxattrat(), so we need to
	116	* go via path here. :-( */
	117
	118	path = strjoin(dir, "/", fn, NULL);
	119	if (!path)
	120	return;
	121
	122	if (getxattr(path, "user.crtime_usec", &crtime, sizeof(crtime)) == sizeof(crtime)) {
	123	crtime = le64toh(crtime);
	124
	125	if (crtime > 0 && crtime != (uint64_t) -1 && crtime < *realtime)
	126	*realtime = crtime;
	127	}
	128	#endif
	129	}
	130
9d647740	131	static int journal_file_empty(int dir_fd, const char *name) {
48979861	132	int r;
9d647740	133	le64_t n_entries;
48979861	134	_cleanup_close_ int fd;
9d647740 ZJS	135
	136	fd = openat(dir_fd, name, O_RDONLY\|O_CLOEXEC\|O_NOFOLLOW\|O_NONBLOCK);
	137	if (fd < 0)
	138	return -errno;
	139
	140	if (lseek(fd, offsetof(Header, n_entries), SEEK_SET) < 0)
	141	return -errno;
	142
	143	r = read(fd, &n_entries, sizeof(n_entries));
	144	if (r != sizeof(n_entries))
	145	return r == 0 ? -EINVAL : -errno;
	146
	147	return le64toh(n_entries) == 0;
	148	}
	149
fb0951b0 LP	150	int journal_directory_vacuum(
	151	const char *directory,
	152	uint64_t max_use,
fb0951b0 LP	153	usec_t max_retention_usec,
	154	usec_t *oldest_usec) {
	155
30cb029b	156	_cleanup_closedir_ DIR *d = NULL;
0284adc6 LP	157	int r = 0;
0284adc6 LP	158	struct vacuum_info *list = NULL;
30cb029b ZJS	159	unsigned n_list = 0, i;
30cb029b ZJS	160	size_t n_allocated = 0;
289f910e	161	uint64_t sum = 0, freed = 0;
fb0951b0	162	usec_t retention_limit = 0;
0284adc6 LP	163
	164	assert(directory);
	165
348ced90	166	if (max_use <= 0 && max_retention_usec <= 0)
0284adc6 LP	167	return 0;
0284adc6 LP	168
fb0951b0 LP	169	if (max_retention_usec > 0) {
	170	retention_limit = now(CLOCK_REALTIME);
	171	if (retention_limit > max_retention_usec)
	172	retention_limit -= max_retention_usec;
	173	else
	174	max_retention_usec = retention_limit = 0;
	175	}
	176
0284adc6 LP	177	d = opendir(directory);
	178	if (!d)
	179	return -errno;
	180
	181	for (;;) {
7d5e9c0f	182	struct dirent *de;
0284adc6 LP	183	size_t q;
	184	struct stat st;
	185	char *p;
	186	unsigned long long seqnum = 0, realtime;
	187	sd_id128_t seqnum_id;
	188	bool have_seqnum;
	189
78a5d04d FW	190	errno = 0;
	191	de = readdir(d);
	192	if (!de && errno != 0) {
	193	r = -errno;
0284adc6 LP	194	goto finish;
	195	}
	196
	197	if (!de)
	198	break;
	199
	200	if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
	201	continue;
	202
	203	if (!S_ISREG(st.st_mode))
	204	continue;
	205
	206	q = strlen(de->d_name);
	207
	208	if (endswith(de->d_name, ".journal")) {
	209
	210	/* Vacuum archived files */
	211
	212	if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
	213	continue;
	214
	215	if (de->d_name[q-8-16-1] != '-' \|\|
	216	de->d_name[q-8-16-1-16-1] != '-' \|\|
	217	de->d_name[q-8-16-1-16-1-32-1] != '@')
	218	continue;
	219
	220	p = strdup(de->d_name);
	221	if (!p) {
	222	r = -ENOMEM;
	223	goto finish;
	224	}
	225
	226	de->d_name[q-8-16-1-16-1] = 0;
	227	if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
	228	free(p);
	229	continue;
	230	}
	231
	232	if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
	233	free(p);
	234	continue;
	235	}
	236
	237	have_seqnum = true;
	238
	239	} else if (endswith(de->d_name, ".journal~")) {
	240	unsigned long long tmp;
	241
	242	/* Vacuum corrupted files */
	243
	244	if (q < 1 + 16 + 1 + 16 + 8 + 1)
	245	continue;
	246
	247	if (de->d_name[q-1-8-16-1] != '-' \|\|
	248	de->d_name[q-1-8-16-1-16-1] != '@')
	249	continue;
	250
	251	p = strdup(de->d_name);
	252	if (!p) {
	253	r = -ENOMEM;
	254	goto finish;
	255	}
	256
	257	if (sscanf(de->d_name + q-1-8-16-1-16, "%16llx-%16llx.journal~", &realtime, &tmp) != 2) {
258	free(p);
259	continue;
260	}
261
262	have_seqnum = false;
263	} else
4fa25d62	264	/* We do not vacuum active files or unknown files! */
0284adc6 LP	265	continue;
0284adc6 LP	266
629bfc5a	267	if (journal_file_empty(dirfd(d), p)) {
9d647740 ZJS	268	/* Always vacuum empty non-online files. */
9d647740 ZJS	269
289f910e ZJS	270	uint64_t size = 512UL * (uint64_t) st.st_blocks;
	271
	272	if (unlinkat(dirfd(d), p, 0) >= 0) {
	273	log_info("Deleted empty journal %s/%s (%"PRIu64" bytes).",
	274	directory, p, size);
	275	freed += size;
	276	} else if (errno != ENOENT)
629bfc5a	277	log_warning("Failed to delete %s/%s: %m", directory, p);
289f910e	278
2ee0591d LP	279	free(p);
2ee0591d LP	280
9d647740 ZJS	281	continue;
	282	}
	283
629bfc5a	284	patch_realtime(directory, p, &st, &realtime);
fb0951b0	285
30cb029b	286	GREEDY_REALLOC(list, n_allocated, n_list + 1);
0284adc6 LP	287
	288	list[n_list].filename = p;
	289	list[n_list].usage = 512UL * (uint64_t) st.st_blocks;
	290	list[n_list].seqnum = seqnum;
	291	list[n_list].realtime = realtime;
	292	list[n_list].seqnum_id = seqnum_id;
	293	list[n_list].have_seqnum = have_seqnum;
	294
	295	sum += list[n_list].usage;
	296
	297	n_list ++;
	298	}
	299
7ff7394d	300	qsort_safe(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
0284adc6	301
fb0951b0	302	for (i = 0; i < n_list; i++) {
0284adc6 LP	303	struct statvfs ss;
	304
	305	if (fstatvfs(dirfd(d), &ss) < 0) {
	306	r = -errno;
	307	goto finish;
	308	}
	309
fb0951b0	310	if ((max_retention_usec <= 0 \|\| list[i].realtime >= retention_limit) &&
348ced90	311	(max_use <= 0 \|\| sum <= max_use))
0284adc6 LP	312	break;
	313
	314	if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
289f910e ZJS	315	log_debug("Deleted archived journal %s/%s (%"PRIu64" bytes).",
	316	directory, list[i].filename, list[i].usage);
	317	freed += list[i].usage;
4fa25d62	318
6c142648	319	if (list[i].usage < sum)
4fa25d62 LP	320	sum -= list[i].usage;
	321	else
	322	sum = 0;
	323
0284adc6 LP	324	} else if (errno != ENOENT)
	325	log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
	326	}
	327
fb0951b0 LP	328	if (oldest_usec && i < n_list && (oldest_usec == 0 \|\| list[i].realtime < oldest_usec))
	329	*oldest_usec = list[i].realtime;
	330
0284adc6 LP	331	finish:
	332	for (i = 0; i < n_list; i++)
	333	free(list[i].filename);
0284adc6 LP	334	free(list);
0284adc6 LP	335
3bb621e1	336	log_debug("Vacuuming done, freed %"PRIu64" bytes", freed);
289f910e	337
0284adc6 LP	338	return r;
0284adc6 LP	339	}