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
59 * - handle files where nothing is in mincore
60 * - does ioprio_set work with fadvise()?
63 static unsigned arg_files_max = 16*1024;
64 static off_t arg_file_size_max = READAHEAD_FILE_SIZE_MAX;
65 static usec_t arg_timeout = 2*USEC_PER_MINUTE;
67 static ReadaheadShared *shared = NULL;
69 /* Avoid collisions with the NULL pointer */
70 #define SECTOR_TO_PTR(s) ULONG_TO_PTR((s)+1)
71 #define PTR_TO_SECTOR(p) (PTR_TO_ULONG(p)-1)
73 static int btrfs_defrag(int fd) {
74 struct btrfs_ioctl_vol_args data;
79 return ioctl(fd, BTRFS_IOC_DEFRAG, &data);
82 static int pack_file(FILE *pack, const char *fn, bool on_btrfs) {
84 void *start = MAP_FAILED;
89 int r = 0, fd = -1, k;
94 if ((fd = open(fn, O_RDONLY|O_CLOEXEC|O_NOATIME|O_NOCTTY|O_NOFOLLOW)) < 0) {
99 log_warning("open(%s) failed: %m", fn);
104 if ((k = file_verify(fd, fn, arg_file_size_max, &st)) <= 0) {
112 l = PAGE_ALIGN(st.st_size);
113 if ((start = mmap(NULL, l, PROT_READ, MAP_SHARED, fd, 0)) == MAP_FAILED) {
114 log_warning("mmap(%s) failed: %m", fn);
119 pages = l / PAGE_SIZE;
122 if (mincore(start, l, vec) < 0) {
123 log_warning("mincore(%s) failed: %m", fn);
132 for (c = 0; c < pages; c++) {
133 bool new_mapped = !!(vec[c] & 1);
135 if (!mapped && new_mapped)
137 else if (mapped && !new_mapped) {
138 fwrite(&b, sizeof(b), 1, pack);
139 fwrite(&c, sizeof(c), 1, pack);
141 log_debug("%s: page %u to %u", fn, b, c);
147 /* We don't write any range data if we should read the entire file */
148 if (mapped && b > 0) {
149 fwrite(&b, sizeof(b), 1, pack);
150 fwrite(&c, sizeof(c), 1, pack);
152 log_debug("%s: page %u to %u", fn, b, c);
157 fwrite(&b, sizeof(b), 1, pack);
158 fwrite(&b, sizeof(b), 1, pack);
161 if (start != MAP_FAILED)
165 close_nointr_nofail(fd);
170 static unsigned long fd_first_block(int fd) {
172 struct fiemap fiemap;
173 struct fiemap_extent extent;
177 data.fiemap.fm_length = ~0ULL;
178 data.fiemap.fm_extent_count = 1;
180 if (ioctl(fd, FS_IOC_FIEMAP, &data) < 0)
183 if (data.fiemap.fm_mapped_extents <= 0)
186 if (data.fiemap.fm_extents[0].fe_flags & FIEMAP_EXTENT_UNKNOWN)
189 return (unsigned long) data.fiemap.fm_extents[0].fe_physical;
197 static int qsort_compare(const void *a, const void *b) {
198 const struct item *i, *j;
203 if (i->block < j->block)
205 if (i->block > j->block)
208 return strcmp(i->path, j->path);
211 static int collect(const char *root) {
213 FD_FANOTIFY, /* Get the actual fs events */
215 FD_INOTIFY, /* We get notifications to quit early via this fd */
218 struct pollfd pollfd[_FD_MAX];
219 int fanotify_fd = -1, signal_fd = -1, inotify_fd = -1, r = 0;
221 Hashmap *files = NULL;
226 char *pack_fn_new = NULL, *pack_fn = NULL;
227 bool on_ssd, on_btrfs;
233 write_one_line_file("/proc/self/oom_score_adj", "1000");
235 if (ioprio_set(IOPRIO_WHO_PROCESS, getpid(), IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0)) < 0)
236 log_warning("Failed to set IDLE IO priority class: %m");
238 assert_se(sigemptyset(&mask) == 0);
239 sigset_add_many(&mask, SIGINT, SIGTERM, -1);
240 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
242 if ((signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
243 log_error("signalfd(): %m");
248 if (!(files = hashmap_new(string_hash_func, string_compare_func))) {
249 log_error("Failed to allocate set.");
254 if ((fanotify_fd = fanotify_init(FAN_CLOEXEC|FAN_NONBLOCK, O_RDONLY|O_LARGEFILE|O_CLOEXEC|O_NOATIME)) < 0) {
255 log_error("Failed to create fanotify object: %m");
260 if (fanotify_mark(fanotify_fd, FAN_MARK_ADD|FAN_MARK_MOUNT, FAN_OPEN, AT_FDCWD, root) < 0) {
261 log_error("Failed to mark %s: %m", root);
266 if ((inotify_fd = open_inotify()) < 0) {
271 not_after = now(CLOCK_MONOTONIC) + arg_timeout;
276 pollfd[FD_FANOTIFY].fd = fanotify_fd;
277 pollfd[FD_FANOTIFY].events = POLLIN;
278 pollfd[FD_SIGNAL].fd = signal_fd;
279 pollfd[FD_SIGNAL].events = POLLIN;
280 pollfd[FD_INOTIFY].fd = inotify_fd;
281 pollfd[FD_INOTIFY].events = POLLIN;
285 "STATUS=Collecting readahead data");
287 log_debug("Collecting...");
289 if (access("/dev/.systemd/readahead/cancel", F_OK) >= 0) {
290 log_debug("Collection canceled");
295 if (access("/dev/.systemd/readahead/done", F_OK) >= 0) {
296 log_debug("Got termination request");
302 struct fanotify_event_metadata metadata;
306 struct fanotify_event_metadata *m;
310 if (hashmap_size(files) > arg_files_max) {
311 log_debug("Reached maximum number of read ahead files, ending collection.");
315 t = now(CLOCK_MONOTONIC);
316 if (t >= not_after) {
317 log_debug("Reached maximum collection time, ending collection.");
321 if ((h = poll(pollfd, _FD_MAX, (int) ((not_after - t) / USEC_PER_MSEC))) < 0) {
326 log_error("poll(): %m");
332 log_debug("Reached maximum collection time, ending collection.");
336 if (pollfd[FD_SIGNAL].revents) {
337 log_debug("Got signal.");
341 if (pollfd[FD_INOTIFY].revents) {
342 uint8_t inotify_buffer[sizeof(struct inotify_event) + FILENAME_MAX];
343 struct inotify_event *e;
345 if ((n = read(inotify_fd, &inotify_buffer, sizeof(inotify_buffer))) < 0) {
346 if (errno == EINTR || errno == EAGAIN)
349 log_error("Failed to read inotify event: %m");
354 e = (struct inotify_event*) inotify_buffer;
358 if ((e->mask & IN_CREATE) && streq(e->name, "cancel")) {
359 log_debug("Collection canceled");
364 if ((e->mask & IN_CREATE) && streq(e->name, "done")) {
365 log_debug("Got termination request");
369 step = sizeof(struct inotify_event) + e->len;
370 assert(step <= (size_t) n);
372 e = (struct inotify_event*) ((uint8_t*) e + step);
377 if ((n = read(fanotify_fd, &data, sizeof(data))) < 0) {
379 if (errno == EINTR || errno == EAGAIN)
382 log_error("Failed to read event: %m");
387 for (m = &data.metadata; FAN_EVENT_OK(m, n); m = FAN_EVENT_NEXT(m, n)) {
394 if (m->pid == my_pid)
397 __sync_synchronize();
398 if (m->pid == shared->replay)
401 snprintf(fn, sizeof(fn), "/proc/self/fd/%i", m->fd);
404 if ((k = readlink_malloc(fn, &p)) >= 0) {
405 if (startswith(p, "/tmp") ||
406 endswith(p, " (deleted)") ||
407 hashmap_get(files, p))
408 /* Not interesting, or
414 ul = fd_first_block(m->fd);
416 if ((k = hashmap_put(files, p, SECTOR_TO_PTR(ul))) < 0) {
417 log_warning("set_put() failed: %s", strerror(-k));
423 log_warning("readlink(%s) failed: %s", fn, strerror(-k));
427 close_nointr_nofail(m->fd);
432 if (fanotify_fd >= 0) {
433 close_nointr_nofail(fanotify_fd);
437 log_debug("Writing Pack File...");
439 on_ssd = fs_on_ssd(root) > 0;
440 log_debug("On SSD: %s", yes_no(on_ssd));
442 on_btrfs = statfs(root, &sfs) >= 0 && (long) sfs.f_type == (long) BTRFS_SUPER_MAGIC;
443 log_debug("On btrfs: %s", yes_no(on_btrfs));
445 asprintf(&pack_fn, "%s/.readahead", root);
446 asprintf(&pack_fn_new, "%s/.readahead.new", root);
448 if (!pack_fn || !pack_fn_new) {
449 log_error("Out of memory");
454 if (!(pack = fopen(pack_fn_new, "we"))) {
455 log_error("Failed to open pack file: %m");
460 fputs(CANONICAL_HOST "\n", pack);
461 putc(on_ssd ? 'S' : 'R', pack);
463 if (on_ssd || on_btrfs) {
465 /* On SSD or on btrfs, just write things out in the
466 * order the files were accessed. */
468 HASHMAP_FOREACH_KEY(q, p, files, i)
469 pack_file(pack, p, on_btrfs);
471 struct item *ordered, *j;
474 /* On rotating media, order things by the block
477 log_debug("Ordering...");
479 n = hashmap_size(files);
480 if (!(ordered = new(struct item, n))) {
481 log_error("Out of memory");
487 HASHMAP_FOREACH_KEY(q, p, files, i) {
489 j->block = PTR_TO_SECTOR(q);
493 assert(ordered + n == j);
495 qsort(ordered, n, sizeof(struct item), qsort_compare);
497 for (k = 0; k < n; k++)
498 pack_file(pack, ordered[k].path, on_btrfs);
503 log_debug("Finalizing...");
508 log_error("Failed to write pack file.");
513 if (rename(pack_fn_new, pack_fn) < 0) {
514 log_error("Failed to rename readahead file: %m");
525 if (fanotify_fd >= 0)
526 close_nointr_nofail(fanotify_fd);
529 close_nointr_nofail(signal_fd);
532 close_nointr_nofail(inotify_fd);
542 while ((p = hashmap_steal_first_key(files)))
550 static int help(void) {
552 printf("%s [OPTIONS...] [DIRECTORY]\n\n"
553 "Collect read-ahead data on early boot.\n\n"
554 " -h --help Show this help\n"
555 " --max-files=INT Maximum number of files to read ahead\n"
556 " --max-file-size=BYTES Maximum size of files to read ahead\n"
557 " --timeout=USEC Maximum time to spend collecting data\n",
558 program_invocation_short_name);
563 static int parse_argv(int argc, char *argv[]) {
566 ARG_FILES_MAX = 0x100,
571 static const struct option options[] = {
572 { "help", no_argument, NULL, 'h' },
573 { "files-max", required_argument, NULL, ARG_FILES_MAX },
574 { "file-size-max", required_argument, NULL, ARG_FILE_SIZE_MAX },
575 { "timeout", required_argument, NULL, ARG_TIMEOUT },
584 while ((c = getopt_long(argc, argv, "h", options, NULL)) >= 0) {
593 if (safe_atou(optarg, &arg_files_max) < 0 || arg_files_max <= 0) {
594 log_error("Failed to parse maximum number of files %s.", optarg);
599 case ARG_FILE_SIZE_MAX: {
600 unsigned long long ull;
602 if (safe_atollu(optarg, &ull) < 0 || ull <= 0) {
603 log_error("Failed to parse maximum file size %s.", optarg);
607 arg_file_size_max = (off_t) ull;
612 if (parse_usec(optarg, &arg_timeout) < 0 || arg_timeout <= 0) {
613 log_error("Failed to parse timeout %s.", optarg);
623 log_error("Unknown option code %c", c);
628 if (optind != argc &&
637 int main(int argc, char *argv[]) {
640 log_set_target(LOG_TARGET_SYSLOG_OR_KMSG);
641 log_parse_environment();
644 if ((r = parse_argv(argc, argv)) <= 0)
645 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
648 log_info("Disabling readahead collector due to low memory.");
652 if (!(shared = shared_get()))
655 shared->collect = getpid();
656 __sync_synchronize();
658 if (collect(optind < argc ? argv[optind] : "/") < 0)