1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2 of the License, or
11 (at your option) any later version.
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 General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
25 #include <linux/limits.h>
30 #include <sys/select.h>
32 #include <sys/types.h>
35 #include <linux/fanotify.h>
36 #include <sys/signalfd.h>
40 #include <linux/fiemap.h>
41 #include <sys/ioctl.h>
44 #include <sys/inotify.h>
49 #include "sd-daemon.h"
51 #include "readahead-common.h"
55 * - detect ssd on btrfs/lvm...
56 * - read ahead directories
57 * - sd_readahead_cancel
60 * - handle files where nothing is in mincore
61 * - does ioprio_set work with fadvise()?
64 static unsigned arg_files_max = 16*1024;
65 static off_t arg_file_size_max = READAHEAD_FILE_SIZE_MAX;
66 static usec_t arg_timeout = 2*USEC_PER_MINUTE;
68 static int btrfs_defrag(int fd) {
69 struct btrfs_ioctl_vol_args data;
74 return ioctl(fd, BTRFS_IOC_DEFRAG, &data);
77 static int pack_file(FILE *pack, const char *fn, bool on_btrfs) {
79 void *start = MAP_FAILED;
84 int r = 0, fd = -1, k;
89 if ((fd = open(fn, O_RDONLY|O_CLOEXEC|O_NOATIME|O_NOCTTY|O_NOFOLLOW)) < 0) {
90 log_warning("open(%s) failed: %m", fn);
95 if ((k = file_verify(fd, fn, arg_file_size_max, &st)) <= 0) {
103 l = PAGE_ALIGN(st.st_size);
104 if ((start = mmap(NULL, l, PROT_READ, MAP_SHARED, fd, 0)) == MAP_FAILED) {
105 log_warning("mmap(%s) failed: %m", fn);
110 pages = l / PAGE_SIZE;
113 if (mincore(start, l, vec) < 0) {
114 log_warning("mincore(%s) failed: %m", fn);
123 for (c = 0; c < pages; c++) {
124 bool new_mapped = !!(vec[c] & 1);
126 if (!mapped && new_mapped)
128 else if (mapped && !new_mapped) {
129 fwrite(&b, sizeof(b), 1, pack);
130 fwrite(&c, sizeof(c), 1, pack);
132 log_debug("%s: page %u to %u", fn, b, c);
138 /* We don't write any range data if we should read the entire file */
139 if (mapped && b > 0) {
140 fwrite(&b, sizeof(b), 1, pack);
141 fwrite(&c, sizeof(c), 1, pack);
143 log_debug("%s: page %u to %u", fn, b, c);
148 fwrite(&b, sizeof(b), 1, pack);
149 fwrite(&b, sizeof(b), 1, pack);
152 if (start != MAP_FAILED)
156 close_nointr_nofail(fd);
161 static unsigned long fd_first_block(int fd) {
163 struct fiemap fiemap;
164 struct fiemap_extent extent;
168 data.fiemap.fm_length = ~0ULL;
169 data.fiemap.fm_extent_count = 1;
171 if (ioctl(fd, FS_IOC_FIEMAP, &data) < 0)
174 if (data.fiemap.fm_mapped_extents <= 0)
177 if (data.fiemap.fm_extents[0].fe_flags & FIEMAP_EXTENT_UNKNOWN)
180 return (unsigned long) data.fiemap.fm_extents[0].fe_physical;
188 static int qsort_compare(const void *a, const void *b) {
189 const struct item *i, *j;
194 if (i->block < j->block)
196 if (i->block > j->block)
199 return strcmp(i->path, j->path);
202 static int collect(const char *root) {
204 FD_FANOTIFY, /* Get the actualy fs events */
206 FD_INOTIFY, /* We get notifications to quit early via this fd */
209 struct pollfd pollfd[_FD_MAX];
210 int fanotify_fd = -1, signal_fd = -1, inotify_fd = -1, r = 0;
212 Hashmap *files = NULL;
217 char *pack_fn_new = NULL, *pack_fn = NULL;
218 bool on_ssd, on_btrfs;
224 write_one_line_file("/proc/self/oom_score_adj", "1000");
226 if (ioprio_set(IOPRIO_WHO_PROCESS, getpid(), IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0)) < 0)
227 log_warning("Failed to set IDLE IO priority class: %m");
229 assert_se(sigemptyset(&mask) == 0);
230 sigset_add_many(&mask, SIGINT, SIGTERM, -1);
231 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
233 if ((signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
234 log_error("signalfd(): %m");
239 if (!(files = hashmap_new(string_hash_func, string_compare_func))) {
240 log_error("Failed to allocate set.");
245 if ((fanotify_fd = fanotify_init(FAN_CLOEXEC|FAN_NONBLOCK, O_RDONLY|O_LARGEFILE|O_CLOEXEC|O_NOATIME)) < 0) {
246 log_error("Failed to create fanotify object: %m");
251 if (fanotify_mark(fanotify_fd, FAN_MARK_ADD|FAN_MARK_MOUNT, FAN_OPEN, AT_FDCWD, root) < 0) {
252 log_error("Failed to mark %s: %m", root);
257 if ((inotify_fd = open_inotify()) < 0) {
262 not_after = now(CLOCK_MONOTONIC) + arg_timeout;
267 pollfd[FD_FANOTIFY].fd = fanotify_fd;
268 pollfd[FD_FANOTIFY].events = POLLIN;
269 pollfd[FD_SIGNAL].fd = signal_fd;
270 pollfd[FD_SIGNAL].events = POLLIN;
271 pollfd[FD_INOTIFY].fd = inotify_fd;
272 pollfd[FD_INOTIFY].events = POLLIN;
276 "STATUS=Collecting readahead data");
278 log_debug("Collecting...");
280 if (access("/dev/.systemd/readahead/cancel", F_OK) >= 0) {
281 log_debug("Collection canceled");
286 if (access("/dev/.systemd/readahead/done", F_OK) >= 0) {
287 log_debug("Got termination request");
293 struct fanotify_event_metadata metadata;
297 struct fanotify_event_metadata *m;
301 if (hashmap_size(files) > arg_files_max) {
302 log_debug("Reached maximum number of read ahead files, ending collection.");
306 t = now(CLOCK_MONOTONIC);
307 if (t >= not_after) {
308 log_debug("Reached maximum collection time, ending collection.");
312 if ((h = poll(pollfd, _FD_MAX, (int) ((not_after - t) / USEC_PER_MSEC))) < 0) {
317 log_error("poll(): %m");
323 log_debug("Reached maximum collection time, ending collection.");
327 if (pollfd[FD_SIGNAL].revents) {
328 log_debug("Got signal.");
332 if (pollfd[FD_INOTIFY].revents) {
333 uint8_t inotify_buffer[sizeof(struct inotify_event) + FILENAME_MAX];
334 struct inotify_event *e;
336 if ((n = read(inotify_fd, &inotify_buffer, sizeof(inotify_buffer))) < 0) {
337 if (errno == EINTR || errno == EAGAIN)
340 log_error("Failed to read inotify event: %m");
345 e = (struct inotify_event*) inotify_buffer;
349 if ((e->mask & IN_CREATE) && streq(e->name, "cancel")) {
350 log_debug("Collection canceled");
355 if ((e->mask & IN_CREATE) && streq(e->name, "done")) {
356 log_debug("Got termination request");
360 step = sizeof(struct inotify_event) + e->len;
361 assert(step <= (size_t) n);
363 e = (struct inotify_event*) ((uint8_t*) e + step);
368 if ((n = read(fanotify_fd, &data, sizeof(data))) < 0) {
370 if (errno == EINTR || errno == EAGAIN)
373 log_error("Failed to read event: %m");
378 for (m = &data.metadata; FAN_EVENT_OK(m, n); m = FAN_EVENT_NEXT(m, n)) {
380 if (m->pid != my_pid && m->fd >= 0) {
384 snprintf(fn, sizeof(fn), "/proc/self/fd/%i", m->fd);
387 if ((k = readlink_malloc(fn, &p)) >= 0) {
389 if (startswith(p, "/tmp") ||
390 hashmap_get(files, p))
391 /* Not interesting, or
397 ul = fd_first_block(m->fd);
399 if ((k = hashmap_put(files, p, ULONG_TO_PTR(ul))) < 0) {
400 log_warning("set_put() failed: %s", strerror(-k));
406 log_warning("readlink(%s) failed: %s", fn, strerror(-k));
410 close_nointr_nofail(m->fd);
415 if (fanotify_fd >= 0) {
416 close_nointr_nofail(fanotify_fd);
420 log_debug("Writing Pack File...");
422 on_ssd = fs_on_ssd(root) == 0;
423 log_debug("On SSD: %s", yes_no(on_ssd));
425 on_btrfs = statfs(root, &sfs) >= 0 && sfs.f_type == BTRFS_SUPER_MAGIC;
426 log_debug("On btrfs: %s", yes_no(on_btrfs));
428 asprintf(&pack_fn, "%s/.readahead", root);
429 asprintf(&pack_fn_new, "%s/.readahead.new", root);
431 if (!pack_fn || !pack_fn_new) {
432 log_error("Out of memory");
437 if (!(pack = fopen(pack_fn_new, "we"))) {
438 log_error("Failed to open pack file: %m");
443 fputs(CANONICAL_HOST "\n", pack);
444 putc(on_ssd ? 'S' : 'R', pack);
446 if (on_ssd || on_btrfs) {
448 /* On SSD or on btrfs, just write things out in the
449 * order the files were accessed. */
451 HASHMAP_FOREACH_KEY(q, p, files, i)
452 pack_file(pack, p, on_btrfs);
454 struct item *ordered, *j;
457 /* On rotating media, order things by the block
460 log_debug("Ordering...");
462 n = hashmap_size(files);
463 if (!(ordered = new(struct item, n))) {
464 log_error("Out of memory");
470 HASHMAP_FOREACH_KEY(q, p, files, i) {
472 j->block = PTR_TO_ULONG(q);
476 assert(ordered + n == j);
478 qsort(ordered, n, sizeof(struct item), qsort_compare);
480 for (k = 0; k < n; k++)
481 pack_file(pack, ordered[k].path, on_btrfs);
486 log_debug("Finalizing...");
491 log_error("Failed to write pack file.");
496 if (rename(pack_fn_new, pack_fn) < 0) {
497 log_error("Failed to rename readahead file: %m");
508 if (fanotify_fd >= 0)
509 close_nointr_nofail(fanotify_fd);
512 close_nointr_nofail(signal_fd);
515 close_nointr_nofail(inotify_fd);
525 while ((p = hashmap_steal_first_key(files)))
533 static int help(void) {
535 printf("%s [OPTIONS...] [DIRECTORY]\n\n"
536 "Collect read-ahead data on early boot.\n\n"
537 " -h --help Show this help\n"
538 " --max-files=INT Maximum number of files to read ahead\n"
539 " --max-file-size=BYTES Maximum size of files to read ahead\n"
540 " --timeout=USEC Maximum time to spend collecting data\n",
541 program_invocation_short_name);
546 static int parse_argv(int argc, char *argv[]) {
549 ARG_FILES_MAX = 0x100,
554 static const struct option options[] = {
555 { "help", no_argument, NULL, 'h' },
556 { "files-max", required_argument, NULL, ARG_FILES_MAX },
557 { "file-size-max", required_argument, NULL, ARG_FILE_SIZE_MAX },
558 { "timeout", required_argument, NULL, ARG_TIMEOUT },
567 while ((c = getopt_long(argc, argv, "h", options, NULL)) >= 0) {
576 if (safe_atou(optarg, &arg_files_max) < 0 || arg_files_max <= 0) {
577 log_error("Failed to parse maximum number of files %s.", optarg);
582 case ARG_FILE_SIZE_MAX: {
583 unsigned long long ull;
585 if (safe_atollu(optarg, &ull) < 0 || ull <= 0) {
586 log_error("Failed to parse maximum file size %s.", optarg);
590 arg_file_size_max = (off_t) ull;
595 if (parse_usec(optarg, &arg_timeout) < 0 || arg_timeout <= 0) {
596 log_error("Failed to parse timeout %s.", optarg);
606 log_error("Unknown option code %c", c);
611 if (optind != argc &&
620 int main(int argc, char *argv[]) {
623 log_set_target(LOG_TARGET_SYSLOG_OR_KMSG);
624 log_parse_environment();
627 if ((r = parse_argv(argc, argv)) <= 0)
628 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
631 log_info("Disabling readahead collector due to low memory.");
635 if (collect(optind < argc ? argv[optind] : "/") < 0)