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
+ 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
- General Public License for more details.
+ Lesser General Public License for more details.
- You should have received a copy of the GNU General Public License
+ You should have received a copy of the GNU Lesser General Public License
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
* - 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;
/* Avoid collisions with the NULL pointer */
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);
bool on_ssd, on_btrfs;
struct statfs sfs;
usec_t not_after;
+ uint64_t previous_block_readahead;
+ bool previous_block_readahead_set = false;
assert(root);
- write_one_line_file("/proc/self/oom_score_adj", "1000");
+ if (asprintf(&pack_fn, "%s/.readahead", root) < 0) {
+ r = log_oom();
+ goto finish;
+ }
+
+ /* 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
+ * readahead is never triggered. */
+ previous_block_readahead_set =
+ access(pack_fn, F_OK) < 0 &&
+ block_get_readahead(root, &previous_block_readahead) >= 0 &&
+ block_set_readahead(root, 8*1024) >= 0;
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");
on_btrfs = statfs(root, &sfs) >= 0 && (long) sfs.f_type == (long) BTRFS_SUPER_MAGIC;
log_debug("On btrfs: %s", yes_no(on_btrfs));
- 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;
+ if (asprintf(&pack_fn_new, "%s/.readahead.new", root) < 0) {
+ 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) {
n = hashmap_size(files);
if (!(ordered = new(struct item, n))) {
- log_error("Out of memory");
- r = -ENOMEM;
+ r = log_oom();
goto finish;
}
fclose(pack);
unlink(pack_fn_new);
}
-
free(pack_fn_new);
free(pack_fn);
hashmap_free(files);
- 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 }
- };
+ if (previous_block_readahead_set) {
+ uint64_t bytes;
- 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;
- }
+ /* Restore the original kernel readahead setting if we
+ * changed it, and nobody has overwritten it since
+ * yet. */
+ if (block_get_readahead(root, &bytes) >= 0 && bytes == 8*1024)
+ block_set_readahead(root, previous_block_readahead);
}
- if (optind != argc &&
- optind != argc-1) {
- help();
- return -EINVAL;
- }
-
- return 1;
+ return r;
}
-int main(int argc, char *argv[]) {
- int r;
- const char *root;
-
- log_set_target(LOG_TARGET_AUTO);
- log_parse_environment();
- log_open();
+int main_collect(const char *root) {
- umask(0022);
-
- if ((r = parse_argv(argc, argv)) <= 0)
- return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
-
- 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
+ * file system on top, since that one is most likely mounted
+ * read-only anyway at boot, even if the underlying block
+ * 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;
- }
-
- if (detect_virtualization(NULL) > 0) {
- log_info("Disabling readahead collector due to execution in virtualized environment.");
- return 0;
+ return EXIT_SUCCESS;
}
- if (!(shared = shared_get()))
- return 1;
+ shared = shared_get();
+ if (!shared)
+ return EXIT_FAILURE;
shared->collect = getpid();
__sync_synchronize();
if (collect(root) < 0)
- return 1;
+ return EXIT_FAILURE;
- return 0;
+ return EXIT_SUCCESS;
}