#include <sys/vfs.h>
#include <getopt.h>
#include <sys/inotify.h>
+#include <math.h>
+
+#ifdef HAVE_LINUX_BTRFS_H
+#include <linux/btrfs.h>
+#endif
+
+#ifdef HAVE_FANOTIFY_INIT
+#include <sys/fanotify.h>
+#endif
#include <systemd/sd-daemon.h>
* - 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;
+static usec_t starttime;
/* 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;
+ struct btrfs_ioctl_vol_args data = { .fd = fd };
return ioctl(fd, BTRFS_IOC_DEFRAG, &data);
}
void *start = MAP_FAILED;
uint8_t *vec;
uint32_t b, c;
+ uint64_t inode;
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) {
+ fd = open(fn, O_RDONLY|O_CLOEXEC|O_NOATIME|O_NOCTTY|O_NOFOLLOW);
+ if (fd < 0) {
if (errno == ENOENT)
return 0;
goto finish;
}
- if ((k = file_verify(fd, fn, arg_file_size_max, &st)) <= 0) {
+ k = file_verify(fd, fn, arg_file_size_max, &st);
+ if (k <= 0) {
r = k;
goto finish;
}
btrfs_defrag(fd);
l = PAGE_ALIGN(st.st_size);
- if ((start = mmap(NULL, l, PROT_READ, MAP_SHARED, fd, 0)) == MAP_FAILED) {
+ start = mmap(NULL, l, PROT_READ, MAP_SHARED, fd, 0);
+ if (start == 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) {
fputs(fn, pack);
fputc('\n', pack);
+ /* Store the inode, so that we notice when the file is deleted */
+ inode = (uint64_t) st.st_ino;
+ fwrite(&inode, sizeof(inode), 1, pack);
+
mapped = false;
for (c = 0; c < pages; c++) {
bool new_mapped = !!(vec[c] & 1);
struct {
struct fiemap fiemap;
struct fiemap_extent extent;
- } data;
-
- zero(data);
- data.fiemap.fm_length = ~0ULL;
- data.fiemap.fm_extent_count = 1;
+ } data = {
+ .fiemap.fm_length = ~0ULL,
+ .fiemap.fm_extent_count = 1,
+ };
if (ioctl(fd, FS_IOC_FIEMAP, &data) < 0)
return 0;
struct item {
const char *path;
unsigned long block;
+ unsigned long bin;
};
static int qsort_compare(const void *a, const void *b) {
i = a;
j = b;
+ /* sort by bin first */
+ if (i->bin < j->bin)
+ return -1;
+ if (i->bin > j->bin)
+ return 1;
+
+ /* then sort by sector */
if (i->block < j->block)
return -1;
if (i->block > j->block)
FD_INOTIFY, /* We get notifications to quit early via this fd */
_FD_MAX
};
- struct pollfd pollfd[_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;
assert(root);
if (asprintf(&pack_fn, "%s/.readahead", root) < 0) {
- log_error("Out of memory");
- r = -ENOMEM;
+ r = log_oom();
goto finish;
}
+ starttime = now(CLOCK_MONOTONIC);
+
/* If there's no pack file yet we lower the kernel readahead
* so that mincore() is accurate. If there is a pack file
* already we assume it is accurate enough so that kernel
block_get_readahead(root, &previous_block_readahead) >= 0 &&
block_set_readahead(root, 8*1024) >= 0;
- 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");
goto finish;
}
- if (!(files = hashmap_new(string_hash_func, string_compare_func))) {
+ files = hashmap_new(string_hash_func, string_compare_func);
+ if (!files) {
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) {
+ fanotify_fd = fanotify_init(FAN_CLOEXEC|FAN_NONBLOCK, O_RDONLY|O_LARGEFILE|O_CLOEXEC|O_NOATIME);
+ if (fanotify_fd < 0) {
log_error("Failed to create fanotify object: %m");
r = -errno;
goto finish;
goto finish;
}
- if ((inotify_fd = open_inotify()) < 0) {
+ inotify_fd = open_inotify();
+ if (inotify_fd < 0) {
r = inotify_fd;
goto finish;
}
my_pid = getpid();
- zero(pollfd);
pollfd[FD_FANOTIFY].fd = fanotify_fd;
pollfd[FD_FANOTIFY].events = POLLIN;
pollfd[FD_SIGNAL].fd = signal_fd;
}
}
- if ((n = read(fanotify_fd, &data, sizeof(data))) < 0) {
+ n = read(fanotify_fd, &data, sizeof(data));
+ if (n < 0) {
if (errno == EINTR || errno == EAGAIN)
continue;
}
for (m = &data.metadata; FAN_EVENT_OK(m, n); m = FAN_EVENT_NEXT(m, n)) {
- char fn[PATH_MAX];
+ char fn[sizeof("/proc/self/fd/") + DECIMAL_STR_MAX(int)];
int k;
if (m->fd < 0)
goto next_iteration;
snprintf(fn, sizeof(fn), "/proc/self/fd/%i", m->fd);
- char_array_0(fn);
-
- if ((k = readlink_malloc(fn, &p)) >= 0) {
+ k = readlink_malloc(fn, &p);
+ if (k >= 0) {
if (startswith(p, "/tmp") ||
endswith(p, " (deleted)") ||
hashmap_get(files, p))
free(p);
else {
unsigned long ul;
+ usec_t entrytime;
+ struct item *entry;
+
+ entry = new0(struct item, 1);
+ if (!entry) {
+ r = log_oom();
+ goto finish;
+ }
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));
+ entrytime = now(CLOCK_MONOTONIC);
+
+ entry->block = ul;
+ entry->path = strdup(p);
+ if (!entry->path) {
+ free(entry);
+ r = log_oom();
+ goto finish;
+ }
+ entry->bin = (entrytime - starttime) / 2000000;
+
+ k = hashmap_put(files, p, entry);
+ if (k < 0) {
+ log_warning("hashmap_put() failed: %s", strerror(-k));
free(p);
}
}
log_warning("readlink(%s) failed: %s", fn, strerror(-k));
next_iteration:
- if (m->fd)
+ if (m->fd >= 0)
close_nointr_nofail(m->fd);
}
}
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;
+ on_btrfs = statfs(root, &sfs) >= 0 && F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC);
log_debug("On btrfs: %s", yes_no(on_btrfs));
if (asprintf(&pack_fn_new, "%s/.readahead.new", root) < 0) {
- log_error("Out of memory");
- r = -ENOMEM;
+ r = log_oom();
goto finish;
}
- if (!(pack = fopen(pack_fn_new, "we"))) {
+ pack = fopen(pack_fn_new, "we");
+ if (!pack) {
log_error("Failed to open pack file: %m");
r = -errno;
goto finish;
}
- fputs(CANONICAL_HOST "\n", pack);
+ fputs(CANONICAL_HOST READAHEAD_PACK_FILE_VERSION, pack);
putc(on_ssd ? 'S' : 'R', pack);
if (on_ssd || on_btrfs) {
HASHMAP_FOREACH_KEY(q, p, files, i)
pack_file(pack, p, on_btrfs);
} else {
- struct item *ordered, *j;
- unsigned k, n;
+ unsigned 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++;
- }
+ if (n) {
+ _cleanup_free_ struct item *ordered;
+ struct item *j;
+ unsigned k;
+
+ ordered = new(struct item, n);
+ if (!ordered) {
+ r = log_oom();
+ goto finish;
+ }
- assert(ordered + n == j);
+ j = ordered;
+ HASHMAP_FOREACH_KEY(q, p, files, i) {
+ memcpy(j, q, sizeof(struct item));
+ j++;
+ }
- qsort(ordered, n, sizeof(struct item), qsort_compare);
+ assert(ordered + n == j);
- for (k = 0; k < n; k++)
- pack_file(pack, ordered[k].path, on_btrfs);
+ qsort(ordered, n, sizeof(struct item), qsort_compare);
- free(ordered);
+ for (k = 0; k < n; k++)
+ pack_file(pack, ordered[k].path, on_btrfs);
+ } else
+ log_warning("No pack files");
}
log_debug("Finalizing...");
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_AUTO);
- log_parse_environment();
- log_open();
-
- umask(0022);
-
- r = parse_argv(argc, argv);
- if (r <= 0)
- return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+int main_collect(const char *root) {
- root = optind < argc ? argv[optind] : "/";
+ if (!root)
+ root = "/";
/* Skip this step on read-only media. Note that we check the
* underlying block device here, not he read-only flag of the
* device is theoretically writable. */
if (fs_on_read_only(root) > 0) {
log_info("Disabling readahead collector due to read-only media.");
- return 0;
+ return EXIT_SUCCESS;
}
if (!enough_ram()) {
log_info("Disabling readahead collector due to low memory.");
- return 0;
+ return EXIT_SUCCESS;
}
shared = shared_get();
if (!shared)
- return 1;
+ return EXIT_FAILURE;
shared->collect = getpid();
__sync_synchronize();
if (collect(root) < 0)
- return 1;
+ return EXIT_FAILURE;
- return 0;
+ return EXIT_SUCCESS;
}