]>
Commit | Line | Data |
---|---|---|
22be093f LP |
1 | /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ |
2 | ||
3 | /*** | |
4 | This file is part of systemd. | |
5 | ||
6 | Copyright 2010 Lennart Poettering | |
7 | ||
8 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
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 | |
22be093f LP |
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 | |
5430f7f2 | 16 | Lesser General Public License for more details. |
22be093f | 17 | |
5430f7f2 | 18 | You should have received a copy of the GNU Lesser General Public License |
22be093f LP |
19 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
20 | ***/ | |
21 | ||
22 | #include <errno.h> | |
23 | #include <inttypes.h> | |
24 | #include <fcntl.h> | |
25 | #include <linux/limits.h> | |
26 | #include <stdbool.h> | |
27 | #include <stdio.h> | |
28 | #include <stdlib.h> | |
29 | #include <string.h> | |
30 | #include <sys/select.h> | |
31 | #include <sys/time.h> | |
32 | #include <sys/types.h> | |
33 | #include <sys/stat.h> | |
34 | #include <unistd.h> | |
35 | #include <linux/fanotify.h> | |
36 | #include <sys/signalfd.h> | |
37 | #include <sys/poll.h> | |
38 | #include <sys/mman.h> | |
39 | #include <linux/fs.h> | |
40 | #include <linux/fiemap.h> | |
41 | #include <sys/ioctl.h> | |
746f8906 | 42 | #include <sys/vfs.h> |
8260358d | 43 | #include <getopt.h> |
6624768c | 44 | #include <sys/inotify.h> |
94243ef2 | 45 | #include <math.h> |
22be093f | 46 | |
4b357e15 MM |
47 | #ifdef HAVE_LINUX_BTRFS_H |
48 | #include <linux/btrfs.h> | |
49 | #endif | |
50 | ||
a8348796 LP |
51 | #ifdef HAVE_FANOTIFY_INIT |
52 | #include <sys/fanotify.h> | |
53 | #endif | |
54 | ||
81527be1 LP |
55 | #include <systemd/sd-daemon.h> |
56 | ||
22be093f LP |
57 | #include "missing.h" |
58 | #include "util.h" | |
59 | #include "set.h" | |
22be093f LP |
60 | #include "ioprio.h" |
61 | #include "readahead-common.h" | |
b52aae1d | 62 | #include "virt.h" |
22be093f | 63 | |
41a598e2 LP |
64 | /* fixme: |
65 | * | |
408b85df | 66 | * - detect ssd on btrfs/lvm... |
41a598e2 | 67 | * - read ahead directories |
408b85df | 68 | * - gzip? |
8260358d | 69 | * - remount rw? |
6624768c | 70 | * - handle files where nothing is in mincore |
408b85df | 71 | * - does ioprio_set work with fadvise()? |
41a598e2 LP |
72 | */ |
73 | ||
d9c7a87b | 74 | static ReadaheadShared *shared = NULL; |
b0640287 | 75 | static usec_t starttime; |
d9c7a87b | 76 | |
2e7485f0 LP |
77 | /* Avoid collisions with the NULL pointer */ |
78 | #define SECTOR_TO_PTR(s) ULONG_TO_PTR((s)+1) | |
79 | #define PTR_TO_SECTOR(p) (PTR_TO_ULONG(p)-1) | |
80 | ||
746f8906 | 81 | static int btrfs_defrag(int fd) { |
b92bea5d | 82 | struct btrfs_ioctl_vol_args data = { .fd = fd }; |
22be093f | 83 | |
746f8906 LP |
84 | return ioctl(fd, BTRFS_IOC_DEFRAG, &data); |
85 | } | |
86 | ||
87 | static int pack_file(FILE *pack, const char *fn, bool on_btrfs) { | |
22be093f LP |
88 | struct stat st; |
89 | void *start = MAP_FAILED; | |
8260358d | 90 | uint8_t *vec; |
22be093f | 91 | uint32_t b, c; |
189455ab | 92 | uint64_t inode; |
22be093f LP |
93 | size_t l, pages; |
94 | bool mapped; | |
95 | int r = 0, fd = -1, k; | |
96 | ||
97 | assert(pack); | |
98 | assert(fn); | |
99 | ||
189455ab LP |
100 | fd = open(fn, O_RDONLY|O_CLOEXEC|O_NOATIME|O_NOCTTY|O_NOFOLLOW); |
101 | if (fd < 0) { | |
a78899f5 LP |
102 | |
103 | if (errno == ENOENT) | |
104 | return 0; | |
105 | ||
a76fad09 LP |
106 | if (errno == EPERM || errno == EACCES) |
107 | return 0; | |
108 | ||
22be093f LP |
109 | log_warning("open(%s) failed: %m", fn); |
110 | r = -errno; | |
111 | goto finish; | |
112 | } | |
113 | ||
189455ab LP |
114 | k = file_verify(fd, fn, arg_file_size_max, &st); |
115 | if (k <= 0) { | |
22be093f LP |
116 | r = k; |
117 | goto finish; | |
118 | } | |
119 | ||
746f8906 LP |
120 | if (on_btrfs) |
121 | btrfs_defrag(fd); | |
122 | ||
22be093f | 123 | l = PAGE_ALIGN(st.st_size); |
189455ab LP |
124 | start = mmap(NULL, l, PROT_READ, MAP_SHARED, fd, 0); |
125 | if (start == MAP_FAILED) { | |
22be093f LP |
126 | log_warning("mmap(%s) failed: %m", fn); |
127 | r = -errno; | |
128 | goto finish; | |
129 | } | |
130 | ||
37f85e66 | 131 | pages = l / page_size(); |
8260358d | 132 | vec = alloca(pages); |
37f85e66 | 133 | memset(vec, 0, pages); |
22be093f LP |
134 | if (mincore(start, l, vec) < 0) { |
135 | log_warning("mincore(%s) failed: %m", fn); | |
136 | r = -errno; | |
137 | goto finish; | |
138 | } | |
139 | ||
140 | fputs(fn, pack); | |
141 | fputc('\n', pack); | |
142 | ||
189455ab LP |
143 | /* Store the inode, so that we notice when the file is deleted */ |
144 | inode = (uint64_t) st.st_ino; | |
145 | fwrite(&inode, sizeof(inode), 1, pack); | |
146 | ||
22be093f LP |
147 | mapped = false; |
148 | for (c = 0; c < pages; c++) { | |
408b85df | 149 | bool new_mapped = !!(vec[c] & 1); |
22be093f LP |
150 | |
151 | if (!mapped && new_mapped) | |
152 | b = c; | |
153 | else if (mapped && !new_mapped) { | |
154 | fwrite(&b, sizeof(b), 1, pack); | |
155 | fwrite(&c, sizeof(c), 1, pack); | |
156 | ||
157 | log_debug("%s: page %u to %u", fn, b, c); | |
158 | } | |
159 | ||
160 | mapped = new_mapped; | |
161 | } | |
162 | ||
163 | /* We don't write any range data if we should read the entire file */ | |
164 | if (mapped && b > 0) { | |
165 | fwrite(&b, sizeof(b), 1, pack); | |
166 | fwrite(&c, sizeof(c), 1, pack); | |
167 | ||
168 | log_debug("%s: page %u to %u", fn, b, c); | |
169 | } | |
170 | ||
171 | /* End marker */ | |
172 | b = 0; | |
173 | fwrite(&b, sizeof(b), 1, pack); | |
174 | fwrite(&b, sizeof(b), 1, pack); | |
175 | ||
176 | finish: | |
177 | if (start != MAP_FAILED) | |
178 | munmap(start, l); | |
179 | ||
180 | if (fd >= 0) | |
181 | close_nointr_nofail(fd); | |
182 | ||
183 | return r; | |
184 | } | |
185 | ||
186 | static unsigned long fd_first_block(int fd) { | |
187 | struct { | |
188 | struct fiemap fiemap; | |
189 | struct fiemap_extent extent; | |
b92bea5d ZJS |
190 | } data = { |
191 | .fiemap.fm_length = ~0ULL, | |
192 | .fiemap.fm_extent_count = 1, | |
193 | }; | |
22be093f LP |
194 | |
195 | if (ioctl(fd, FS_IOC_FIEMAP, &data) < 0) | |
196 | return 0; | |
197 | ||
198 | if (data.fiemap.fm_mapped_extents <= 0) | |
199 | return 0; | |
200 | ||
201 | if (data.fiemap.fm_extents[0].fe_flags & FIEMAP_EXTENT_UNKNOWN) | |
202 | return 0; | |
203 | ||
204 | return (unsigned long) data.fiemap.fm_extents[0].fe_physical; | |
205 | } | |
206 | ||
207 | struct item { | |
208 | const char *path; | |
209 | unsigned long block; | |
94243ef2 | 210 | unsigned long bin; |
22be093f LP |
211 | }; |
212 | ||
213 | static int qsort_compare(const void *a, const void *b) { | |
214 | const struct item *i, *j; | |
215 | ||
216 | i = a; | |
217 | j = b; | |
218 | ||
94243ef2 AK |
219 | /* sort by bin first */ |
220 | if (i->bin < j->bin) | |
221 | return -1; | |
222 | if (i->bin > j->bin) | |
223 | return 1; | |
224 | ||
225 | /* then sort by sector */ | |
22be093f LP |
226 | if (i->block < j->block) |
227 | return -1; | |
228 | if (i->block > j->block) | |
229 | return 1; | |
230 | ||
231 | return strcmp(i->path, j->path); | |
232 | } | |
233 | ||
234 | static int collect(const char *root) { | |
235 | enum { | |
858209c5 | 236 | FD_FANOTIFY, /* Get the actual fs events */ |
22be093f | 237 | FD_SIGNAL, |
6624768c | 238 | FD_INOTIFY, /* We get notifications to quit early via this fd */ |
22be093f LP |
239 | _FD_MAX |
240 | }; | |
b92bea5d | 241 | struct pollfd pollfd[_FD_MAX] = {}; |
6624768c | 242 | int fanotify_fd = -1, signal_fd = -1, inotify_fd = -1, r = 0; |
22be093f LP |
243 | pid_t my_pid; |
244 | Hashmap *files = NULL; | |
245 | Iterator i; | |
246 | char *p, *q; | |
247 | sigset_t mask; | |
248 | FILE *pack = NULL; | |
249 | char *pack_fn_new = NULL, *pack_fn = NULL; | |
746f8906 LP |
250 | bool on_ssd, on_btrfs; |
251 | struct statfs sfs; | |
408b85df | 252 | usec_t not_after; |
6de338a2 LP |
253 | uint64_t previous_block_readahead; |
254 | bool previous_block_readahead_set = false; | |
22be093f LP |
255 | |
256 | assert(root); | |
257 | ||
6de338a2 | 258 | if (asprintf(&pack_fn, "%s/.readahead", root) < 0) { |
0d0f0c50 | 259 | r = log_oom(); |
6de338a2 LP |
260 | goto finish; |
261 | } | |
262 | ||
b0640287 | 263 | starttime = now(CLOCK_MONOTONIC); |
94243ef2 | 264 | |
6de338a2 LP |
265 | /* If there's no pack file yet we lower the kernel readahead |
266 | * so that mincore() is accurate. If there is a pack file | |
267 | * already we assume it is accurate enough so that kernel | |
268 | * readahead is never triggered. */ | |
269 | previous_block_readahead_set = | |
270 | access(pack_fn, F_OK) < 0 && | |
271 | block_get_readahead(root, &previous_block_readahead) >= 0 && | |
272 | block_set_readahead(root, 8*1024) >= 0; | |
273 | ||
22be093f LP |
274 | if (ioprio_set(IOPRIO_WHO_PROCESS, getpid(), IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0)) < 0) |
275 | log_warning("Failed to set IDLE IO priority class: %m"); | |
276 | ||
277 | assert_se(sigemptyset(&mask) == 0); | |
278 | sigset_add_many(&mask, SIGINT, SIGTERM, -1); | |
279 | assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); | |
280 | ||
281 | if ((signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) { | |
282 | log_error("signalfd(): %m"); | |
283 | r = -errno; | |
284 | goto finish; | |
285 | } | |
286 | ||
e62d8c39 ZJS |
287 | files = hashmap_new(string_hash_func, string_compare_func); |
288 | if (!files) { | |
22be093f LP |
289 | log_error("Failed to allocate set."); |
290 | r = -ENOMEM; | |
291 | goto finish; | |
292 | } | |
293 | ||
7989e1f2 | 294 | fanotify_fd = fanotify_init(FAN_CLOEXEC|FAN_NONBLOCK, O_RDONLY|O_LARGEFILE|O_CLOEXEC|O_NOATIME); |
e62d8c39 | 295 | if (fanotify_fd < 0) { |
22be093f LP |
296 | log_error("Failed to create fanotify object: %m"); |
297 | r = -errno; | |
298 | goto finish; | |
299 | } | |
300 | ||
301 | if (fanotify_mark(fanotify_fd, FAN_MARK_ADD|FAN_MARK_MOUNT, FAN_OPEN, AT_FDCWD, root) < 0) { | |
302 | log_error("Failed to mark %s: %m", root); | |
303 | r = -errno; | |
304 | goto finish; | |
305 | } | |
306 | ||
e62d8c39 ZJS |
307 | inotify_fd = open_inotify(); |
308 | if (inotify_fd < 0) { | |
6624768c LP |
309 | r = inotify_fd; |
310 | goto finish; | |
311 | } | |
312 | ||
8260358d | 313 | not_after = now(CLOCK_MONOTONIC) + arg_timeout; |
408b85df | 314 | |
22be093f LP |
315 | my_pid = getpid(); |
316 | ||
22be093f LP |
317 | pollfd[FD_FANOTIFY].fd = fanotify_fd; |
318 | pollfd[FD_FANOTIFY].events = POLLIN; | |
319 | pollfd[FD_SIGNAL].fd = signal_fd; | |
320 | pollfd[FD_SIGNAL].events = POLLIN; | |
6624768c LP |
321 | pollfd[FD_INOTIFY].fd = inotify_fd; |
322 | pollfd[FD_INOTIFY].events = POLLIN; | |
22be093f LP |
323 | |
324 | sd_notify(0, | |
325 | "READY=1\n" | |
326 | "STATUS=Collecting readahead data"); | |
327 | ||
328 | log_debug("Collecting..."); | |
329 | ||
2b583ce6 | 330 | if (access("/run/systemd/readahead/cancel", F_OK) >= 0) { |
6624768c LP |
331 | log_debug("Collection canceled"); |
332 | r = -ECANCELED; | |
333 | goto finish; | |
334 | } | |
335 | ||
2b583ce6 | 336 | if (access("/run/systemd/readahead/done", F_OK) >= 0) { |
6624768c LP |
337 | log_debug("Got termination request"); |
338 | goto done; | |
339 | } | |
340 | ||
22be093f LP |
341 | for (;;) { |
342 | union { | |
343 | struct fanotify_event_metadata metadata; | |
344 | char buffer[4096]; | |
345 | } data; | |
346 | ssize_t n; | |
347 | struct fanotify_event_metadata *m; | |
408b85df LP |
348 | usec_t t; |
349 | int h; | |
22be093f | 350 | |
8260358d | 351 | if (hashmap_size(files) > arg_files_max) { |
408b85df | 352 | log_debug("Reached maximum number of read ahead files, ending collection."); |
6e3eb5ba | 353 | break; |
408b85df LP |
354 | } |
355 | ||
356 | t = now(CLOCK_MONOTONIC); | |
357 | if (t >= not_after) { | |
358 | log_debug("Reached maximum collection time, ending collection."); | |
359 | break; | |
360 | } | |
6e3eb5ba | 361 | |
408b85df | 362 | if ((h = poll(pollfd, _FD_MAX, (int) ((not_after - t) / USEC_PER_MSEC))) < 0) { |
22be093f LP |
363 | |
364 | if (errno == EINTR) | |
365 | continue; | |
366 | ||
367 | log_error("poll(): %m"); | |
368 | r = -errno; | |
369 | goto finish; | |
370 | } | |
371 | ||
408b85df LP |
372 | if (h == 0) { |
373 | log_debug("Reached maximum collection time, ending collection."); | |
374 | break; | |
375 | } | |
376 | ||
6624768c LP |
377 | if (pollfd[FD_SIGNAL].revents) { |
378 | log_debug("Got signal."); | |
379 | break; | |
380 | } | |
381 | ||
382 | if (pollfd[FD_INOTIFY].revents) { | |
383 | uint8_t inotify_buffer[sizeof(struct inotify_event) + FILENAME_MAX]; | |
384 | struct inotify_event *e; | |
385 | ||
386 | if ((n = read(inotify_fd, &inotify_buffer, sizeof(inotify_buffer))) < 0) { | |
387 | if (errno == EINTR || errno == EAGAIN) | |
388 | continue; | |
389 | ||
390 | log_error("Failed to read inotify event: %m"); | |
391 | r = -errno; | |
392 | goto finish; | |
393 | } | |
394 | ||
395 | e = (struct inotify_event*) inotify_buffer; | |
396 | while (n > 0) { | |
397 | size_t step; | |
398 | ||
399 | if ((e->mask & IN_CREATE) && streq(e->name, "cancel")) { | |
400 | log_debug("Collection canceled"); | |
401 | r = -ECANCELED; | |
402 | goto finish; | |
403 | } | |
404 | ||
405 | if ((e->mask & IN_CREATE) && streq(e->name, "done")) { | |
406 | log_debug("Got termination request"); | |
407 | goto done; | |
408 | } | |
409 | ||
410 | step = sizeof(struct inotify_event) + e->len; | |
411 | assert(step <= (size_t) n); | |
412 | ||
413 | e = (struct inotify_event*) ((uint8_t*) e + step); | |
414 | n -= step; | |
415 | } | |
416 | } | |
417 | ||
22be093f LP |
418 | if ((n = read(fanotify_fd, &data, sizeof(data))) < 0) { |
419 | ||
cf37e246 LP |
420 | if (errno == EINTR || errno == EAGAIN) |
421 | continue; | |
422 | ||
423 | /* fanotify sometimes returns EACCES on read() | |
424 | * where it shouldn't. For now let's just | |
425 | * ignore it here (which is safe), but | |
426 | * eventually this should be | |
427 | * dropped when the kernel is fixed. | |
428 | * | |
429 | * https://bugzilla.redhat.com/show_bug.cgi?id=707577 */ | |
430 | if (errno == EACCES) | |
22be093f LP |
431 | continue; |
432 | ||
433 | log_error("Failed to read event: %m"); | |
434 | r = -errno; | |
435 | goto finish; | |
436 | } | |
437 | ||
408b85df | 438 | for (m = &data.metadata; FAN_EVENT_OK(m, n); m = FAN_EVENT_NEXT(m, n)) { |
d9c7a87b LP |
439 | char fn[PATH_MAX]; |
440 | int k; | |
22be093f | 441 | |
d9c7a87b LP |
442 | if (m->fd < 0) |
443 | goto next_iteration; | |
22be093f | 444 | |
d9c7a87b LP |
445 | if (m->pid == my_pid) |
446 | goto next_iteration; | |
22be093f | 447 | |
d9c7a87b LP |
448 | __sync_synchronize(); |
449 | if (m->pid == shared->replay) | |
450 | goto next_iteration; | |
22be093f | 451 | |
d9c7a87b LP |
452 | snprintf(fn, sizeof(fn), "/proc/self/fd/%i", m->fd); |
453 | char_array_0(fn); | |
454 | ||
455 | if ((k = readlink_malloc(fn, &p)) >= 0) { | |
d9c7a87b | 456 | if (startswith(p, "/tmp") || |
0840ce2d | 457 | endswith(p, " (deleted)") || |
d9c7a87b LP |
458 | hashmap_get(files, p)) |
459 | /* Not interesting, or | |
460 | * already read */ | |
461 | free(p); | |
462 | else { | |
463 | unsigned long ul; | |
b0640287 | 464 | usec_t entrytime; |
94243ef2 AK |
465 | struct item *entry; |
466 | ||
467 | entry = new0(struct item, 1); | |
b0640287 AK |
468 | if (!entry) { |
469 | r = log_oom(); | |
470 | goto finish; | |
471 | } | |
22be093f | 472 | |
d9c7a87b LP |
473 | ul = fd_first_block(m->fd); |
474 | ||
b0640287 | 475 | entrytime = now(CLOCK_MONOTONIC); |
94243ef2 AK |
476 | |
477 | entry->block = ul; | |
478 | entry->path = strdup(p); | |
b0640287 AK |
479 | if (!entry->path) { |
480 | free(entry); | |
481 | r = log_oom(); | |
482 | goto finish; | |
483 | } | |
484 | entry->bin = (entrytime - starttime) / 2000000; | |
94243ef2 | 485 | |
ef42202a ZJS |
486 | k = hashmap_put(files, p, entry); |
487 | if (k < 0) { | |
488 | log_warning("hashmap_put() failed: %s", strerror(-k)); | |
d9c7a87b | 489 | free(p); |
22be093f | 490 | } |
d9c7a87b | 491 | } |
22be093f | 492 | |
d9c7a87b LP |
493 | } else |
494 | log_warning("readlink(%s) failed: %s", fn, strerror(-k)); | |
22be093f | 495 | |
d9c7a87b | 496 | next_iteration: |
c4b996bd | 497 | if (m->fd >= 0) |
22be093f | 498 | close_nointr_nofail(m->fd); |
22be093f | 499 | } |
22be093f LP |
500 | } |
501 | ||
6624768c | 502 | done: |
22be093f LP |
503 | if (fanotify_fd >= 0) { |
504 | close_nointr_nofail(fanotify_fd); | |
505 | fanotify_fd = -1; | |
506 | } | |
507 | ||
508 | log_debug("Writing Pack File..."); | |
509 | ||
55888fa4 | 510 | on_ssd = fs_on_ssd(root) > 0; |
22be093f LP |
511 | log_debug("On SSD: %s", yes_no(on_ssd)); |
512 | ||
c51cf056 | 513 | on_btrfs = statfs(root, &sfs) >= 0 && F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC); |
746f8906 LP |
514 | log_debug("On btrfs: %s", yes_no(on_btrfs)); |
515 | ||
6de338a2 | 516 | if (asprintf(&pack_fn_new, "%s/.readahead.new", root) < 0) { |
0d0f0c50 | 517 | r = log_oom(); |
22be093f LP |
518 | goto finish; |
519 | } | |
520 | ||
189455ab LP |
521 | pack = fopen(pack_fn_new, "we"); |
522 | if (!pack) { | |
22be093f LP |
523 | log_error("Failed to open pack file: %m"); |
524 | r = -errno; | |
525 | goto finish; | |
526 | } | |
527 | ||
cae544bc | 528 | fputs(CANONICAL_HOST READAHEAD_PACK_FILE_VERSION, pack); |
22be093f LP |
529 | putc(on_ssd ? 'S' : 'R', pack); |
530 | ||
746f8906 | 531 | if (on_ssd || on_btrfs) { |
22be093f | 532 | |
746f8906 | 533 | /* On SSD or on btrfs, just write things out in the |
41a598e2 | 534 | * order the files were accessed. */ |
22be093f LP |
535 | |
536 | HASHMAP_FOREACH_KEY(q, p, files, i) | |
746f8906 | 537 | pack_file(pack, p, on_btrfs); |
22be093f | 538 | } else { |
7ff7394d | 539 | unsigned n; |
22be093f LP |
540 | |
541 | /* On rotating media, order things by the block | |
542 | * numbers */ | |
543 | ||
544 | log_debug("Ordering..."); | |
545 | ||
546 | n = hashmap_size(files); | |
7ff7394d ZJS |
547 | if (n) { |
548 | _cleanup_free_ struct item *ordered; | |
549 | struct item *j; | |
550 | unsigned k; | |
551 | ||
552 | ordered = new(struct item, n); | |
553 | if (!ordered) { | |
554 | r = log_oom(); | |
555 | goto finish; | |
556 | } | |
22be093f | 557 | |
7ff7394d ZJS |
558 | j = ordered; |
559 | HASHMAP_FOREACH_KEY(q, p, files, i) { | |
560 | memcpy(j, q, sizeof(struct item)); | |
561 | j++; | |
562 | } | |
22be093f | 563 | |
7ff7394d | 564 | assert(ordered + n == j); |
22be093f | 565 | |
7ff7394d | 566 | qsort(ordered, n, sizeof(struct item), qsort_compare); |
22be093f | 567 | |
7ff7394d ZJS |
568 | for (k = 0; k < n; k++) |
569 | pack_file(pack, ordered[k].path, on_btrfs); | |
570 | } else | |
571 | log_warning("No pack files"); | |
22be093f LP |
572 | } |
573 | ||
574 | log_debug("Finalizing..."); | |
575 | ||
576 | fflush(pack); | |
577 | ||
578 | if (ferror(pack)) { | |
579 | log_error("Failed to write pack file."); | |
580 | r = -EIO; | |
581 | goto finish; | |
582 | } | |
583 | ||
584 | if (rename(pack_fn_new, pack_fn) < 0) { | |
585 | log_error("Failed to rename readahead file: %m"); | |
586 | r = -errno; | |
587 | goto finish; | |
588 | } | |
589 | ||
590 | fclose(pack); | |
591 | pack = NULL; | |
592 | ||
593 | log_debug("Done."); | |
594 | ||
595 | finish: | |
596 | if (fanotify_fd >= 0) | |
597 | close_nointr_nofail(fanotify_fd); | |
598 | ||
599 | if (signal_fd >= 0) | |
600 | close_nointr_nofail(signal_fd); | |
601 | ||
6624768c LP |
602 | if (inotify_fd >= 0) |
603 | close_nointr_nofail(inotify_fd); | |
604 | ||
22be093f LP |
605 | if (pack) { |
606 | fclose(pack); | |
607 | unlink(pack_fn_new); | |
608 | } | |
22be093f LP |
609 | free(pack_fn_new); |
610 | free(pack_fn); | |
611 | ||
612 | while ((p = hashmap_steal_first_key(files))) | |
f0cf061e | 613 | free(p); |
22be093f LP |
614 | |
615 | hashmap_free(files); | |
616 | ||
6de338a2 LP |
617 | if (previous_block_readahead_set) { |
618 | uint64_t bytes; | |
619 | ||
620 | /* Restore the original kernel readahead setting if we | |
621 | * changed it, and nobody has overwritten it since | |
622 | * yet. */ | |
623 | if (block_get_readahead(root, &bytes) >= 0 && bytes == 8*1024) | |
624 | block_set_readahead(root, previous_block_readahead); | |
625 | } | |
626 | ||
22be093f LP |
627 | return r; |
628 | } | |
629 | ||
87ce22cc | 630 | int main_collect(const char *root) { |
8260358d | 631 | |
87ce22cc LP |
632 | if (!root) |
633 | root = "/"; | |
2b590e13 | 634 | |
3b2d5b02 LP |
635 | /* Skip this step on read-only media. Note that we check the |
636 | * underlying block device here, not he read-only flag of the | |
637 | * file system on top, since that one is most likely mounted | |
638 | * read-only anyway at boot, even if the underlying block | |
639 | * device is theoretically writable. */ | |
2b590e13 LP |
640 | if (fs_on_read_only(root) > 0) { |
641 | log_info("Disabling readahead collector due to read-only media."); | |
87ce22cc | 642 | return EXIT_SUCCESS; |
2b590e13 LP |
643 | } |
644 | ||
41a598e2 LP |
645 | if (!enough_ram()) { |
646 | log_info("Disabling readahead collector due to low memory."); | |
87ce22cc | 647 | return EXIT_SUCCESS; |
41a598e2 LP |
648 | } |
649 | ||
3b2d5b02 LP |
650 | shared = shared_get(); |
651 | if (!shared) | |
87ce22cc | 652 | return EXIT_FAILURE; |
d9c7a87b LP |
653 | |
654 | shared->collect = getpid(); | |
655 | __sync_synchronize(); | |
656 | ||
2b590e13 | 657 | if (collect(root) < 0) |
87ce22cc | 658 | return EXIT_FAILURE; |
22be093f | 659 | |
87ce22cc | 660 | return EXIT_SUCCESS; |
22be093f | 661 | } |