summaryrefslogtreecommitdiff
path: root/src/readahead/readahead-collect.c
blob: 658c230d65e985cbdd43a753ac3f2a515562110d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/

/***
  This file is part of systemd.

  Copyright 2010 Lennart Poettering

  systemd is free software; you can redistribute it and/or modify it
  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
  Lesser General Public License for more details.

  You should have received a copy of the GNU Lesser General Public License
  along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/

#include <errno.h>
#include <inttypes.h>
#include <fcntl.h>
#include <linux/limits.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/select.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <linux/fanotify.h>
#include <sys/signalfd.h>
#include <sys/poll.h>
#include <sys/mman.h>
#include <linux/fs.h>
#include <linux/fiemap.h>
#include <linux/btrfs.h>
#include <sys/ioctl.h>
#include <sys/vfs.h>
#include <getopt.h>
#include <sys/inotify.h>
#include <math.h>

#ifdef HAVE_FANOTIFY_INIT
#include <sys/fanotify.h>
#endif

#include <systemd/sd-daemon.h>

#include "missing.h"
#include "util.h"
#include "set.h"
#include "ioprio.h"
#include "readahead-common.h"
#include "virt.h"

/* fixme:
 *
 * - detect ssd on btrfs/lvm...
 * - read ahead directories
 * - gzip?
 * - remount rw?
 * - handle files where nothing is in mincore
 * - does ioprio_set work with fadvise()?
 */

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 = { .fd = fd };

        return ioctl(fd, BTRFS_IOC_DEFRAG, &data);
}

static int pack_file(FILE *pack, const char *fn, bool on_btrfs) {
        struct stat st;
        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);

        fd = open(fn, O_RDONLY|O_CLOEXEC|O_NOATIME|O_NOCTTY|O_NOFOLLOW);
        if (fd < 0) {

                if (errno == ENOENT)
                        return 0;

                if (errno == EPERM || errno == EACCES)
                        return 0;

                log_warning("open(%s) failed: %m", fn);
                r = -errno;
                goto finish;
        }

        k = file_verify(fd, fn, arg_file_size_max, &st);
        if (k <= 0) {
                r = k;
                goto finish;
        }

        if (on_btrfs)
                btrfs_defrag(fd);

        l = PAGE_ALIGN(st.st_size);
        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) {
                log_warning("mincore(%s) failed: %m", fn);
                r = -errno;
                goto finish;
        }

        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);

                if (!mapped && new_mapped)
                        b = c;
                else if (mapped && !new_mapped) {
                        fwrite(&b, sizeof(b), 1, pack);
                        fwrite(&c, sizeof(c), 1, pack);

                        log_debug("%s: page %u to %u", fn, b, c);
                }

                mapped = new_mapped;
        }

        /* We don't write any range data if we should read the entire file */
        if (mapped && b > 0) {
                fwrite(&b, sizeof(b), 1, pack);
                fwrite(&c, sizeof(c), 1, pack);

                log_debug("%s: page %u to %u", fn, b, c);
        }

        /* End marker */
        b = 0;
        fwrite(&b, sizeof(b), 1, pack);
        fwrite(&b, sizeof(b), 1, pack);

finish:
        if (start != MAP_FAILED)
                munmap(start, l);

        if (fd >= 0)
                close_nointr_nofail(fd);

        return r;
}

static unsigned long fd_first_block(int fd) {
        struct {
                struct fiemap fiemap;
                struct fiemap_extent extent;
        } data = {
                .fiemap.fm_length = ~0ULL,
                .fiemap.fm_extent_count = 1,
        };

        if (ioctl(fd, FS_IOC_FIEMAP, &data) < 0)
                return 0;

        if (data.fiemap.fm_mapped_extents <= 0)
                return 0;

        if (data.fiemap.fm_extents[0].fe_flags & FIEMAP_EXTENT_UNKNOWN)
                return 0;

        return (unsigned long) data.fiemap.fm_extents[0].fe_physical;
}

struct item {
        const char *path;
        unsigned long block;
        unsigned long bin;
};

static int qsort_compare(const void *a, const void *b) {
        const struct item *i, *j;

        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)
                return 1;

        return strcmp(i->path, j->path);
}

static int collect(const char *root) {
        enum {
                FD_FANOTIFY,  /* Get the actual fs events */
                FD_SIGNAL,
                FD_INOTIFY,   /* We get notifications to quit early via this fd */
                _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;
        Iterator i;
        char *p, *q;
        sigset_t mask;
        FILE *pack = NULL;
        char *pack_fn_new = NULL, *pack_fn = NULL;
        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);

        if (asprintf(&pack_fn, "%s/.readahead", root) < 0) {
                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
         * 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");

        assert_se(sigemptyset(&mask) == 0);
        sigset_add_many(&mask, SIGINT, SIGTERM, -1);
        assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);

        if ((signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
                log_error("signalfd(): %m");
                r = -errno;
                goto finish;
        }

        files = hashmap_new(string_hash_func, string_compare_func);
        if (!files) {
                log_error("Failed to allocate set.");
                r = -ENOMEM;
                goto finish;
        }

        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;
        }

        if (fanotify_mark(fanotify_fd, FAN_MARK_ADD|FAN_MARK_MOUNT, FAN_OPEN, AT_FDCWD, root) < 0) {
                log_error("Failed to mark %s: %m", root);
                r = -errno;
                goto finish;
        }

        inotify_fd = open_inotify();
        if (inotify_fd < 0) {
                r = inotify_fd;
                goto finish;
        }

        not_after = now(CLOCK_MONOTONIC) + arg_timeout;

        my_pid = getpid();

        pollfd[FD_FANOTIFY].fd = fanotify_fd;
        pollfd[FD_FANOTIFY].events = POLLIN;
        pollfd[FD_SIGNAL].fd = signal_fd;
        pollfd[FD_SIGNAL].events = POLLIN;
        pollfd[FD_INOTIFY].fd = inotify_fd;
        pollfd[FD_INOTIFY].events = POLLIN;

        sd_notify(0,
                  "READY=1\n"
                  "STATUS=Collecting readahead data");

        log_debug("Collecting...");

        if (access("/run/systemd/readahead/cancel", F_OK) >= 0) {
                log_debug("Collection canceled");
                r = -ECANCELED;
                goto finish;
        }

        if (access("/run/systemd/readahead/done", F_OK) >= 0) {
                log_debug("Got termination request");
                goto done;
        }

        for (;;) {
                union {
                        struct fanotify_event_metadata metadata;
                        char buffer[4096];
                } data;
                ssize_t n;
                struct fanotify_event_metadata *m;
                usec_t t;
                int h;

                if (hashmap_size(files) > arg_files_max) {
                        log_debug("Reached maximum number of read ahead files, ending collection.");
                        break;
                }

                t = now(CLOCK_MONOTONIC);
                if (t >= not_after) {
                        log_debug("Reached maximum collection time, ending collection.");
                        break;
                }

                if ((h = poll(pollfd, _FD_MAX, (int) ((not_after - t) / USEC_PER_MSEC))) < 0) {

                        if (errno == EINTR)
                                continue;

                        log_error("poll(): %m");
                        r = -errno;
                        goto finish;
                }

                if (h == 0) {
                        log_debug("Reached maximum collection time, ending collection.");
                        break;
                }

                if (pollfd[FD_SIGNAL].revents) {
                        log_debug("Got signal.");
                        break;
                }

                if (pollfd[FD_INOTIFY].revents) {
                        uint8_t inotify_buffer[sizeof(struct inotify_event) + FILENAME_MAX];
                        struct inotify_event *e;

                        if ((n = read(inotify_fd, &inotify_buffer, sizeof(inotify_buffer))) < 0) {
                                if (errno == EINTR || errno == EAGAIN)
                                        continue;

                                log_error("Failed to read inotify event: %m");
                                r = -errno;
                                goto finish;
                        }

                        e = (struct inotify_event*) inotify_buffer;
                        while (n > 0) {
                                size_t step;

                                if ((e->mask & IN_CREATE) && streq(e->name, "cancel")) {
                                        log_debug("Collection canceled");
                                        r = -ECANCELED;
                                        goto finish;
                                }

                                if ((e->mask & IN_CREATE) && streq(e->name, "done")) {
                                        log_debug("Got termination request");
                                        goto done;
                                }

                                step = sizeof(struct inotify_event) + e->len;
                                assert(step <= (size_t) n);

                                e = (struct inotify_event*) ((uint8_t*) e + step);
                                n -= step;
                        }
                }

                if ((n = read(fanotify_fd, &data, sizeof(data))) < 0) {

                        if (errno == EINTR || errno == EAGAIN)
                                continue;

                        /* fanotify sometimes returns EACCES on read()
                         * where it shouldn't. For now let's just
                         * ignore it here (which is safe), but
                         * eventually this should be
                         * dropped when the kernel is fixed.
                         *
                         * https://bugzilla.redhat.com/show_bug.cgi?id=707577 */
                        if (errno == EACCES)
                                continue;

                        log_error("Failed to read event: %m");
                        r = -errno;
                        goto finish;
                }

                for (m = &data.metadata; FAN_EVENT_OK(m, n); m = FAN_EVENT_NEXT(m, n)) {
                        char fn[PATH_MAX];
                        int k;

                        if (m->fd < 0)
                                goto next_iteration;

                        if (m->pid == my_pid)
                                goto next_iteration;

                        __sync_synchronize();
                        if (m->pid == shared->replay)
                                goto next_iteration;

                        snprintf(fn, sizeof(fn), "/proc/self/fd/%i", m->fd);
                        char_array_0(fn);

                        if ((k = readlink_malloc(fn, &p)) >= 0) {
                                if (startswith(p, "/tmp") ||
                                    endswith(p, " (deleted)") ||
                                    hashmap_get(files, p))
                                        /* Not interesting, or
                                         * already read */
                                        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);

                                        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);
                                        }
                                }

                        } else
                                log_warning("readlink(%s) failed: %s", fn, strerror(-k));

                next_iteration:
                        if (m->fd >= 0)
                                close_nointr_nofail(m->fd);
                }
        }

done:
        if (fanotify_fd >= 0) {
                close_nointr_nofail(fanotify_fd);
                fanotify_fd = -1;
        }

        log_debug("Writing Pack File...");

        on_ssd = fs_on_ssd(root) > 0;
        log_debug("On SSD: %s", yes_no(on_ssd));

        on_btrfs = statfs(root, &sfs) >= 0 && F_TYPE_CMP(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) {
                r = log_oom();
                goto finish;
        }

        pack = fopen(pack_fn_new, "we");
        if (!pack) {
                log_error("Failed to open pack file: %m");
                r = -errno;
                goto finish;
        }

        fputs(CANONICAL_HOST READAHEAD_PACK_FILE_VERSION, pack);
        putc(on_ssd ? 'S' : 'R', pack);

        if (on_ssd || on_btrfs) {

                /* On SSD or on btrfs, just write things out in the
                 * order the files were accessed. */

                HASHMAP_FOREACH_KEY(q, p, files, i)
                        pack_file(pack, p, on_btrfs);
        } else {
                struct item *ordered, *j;
                unsigned k, n;

                /* On rotating media, order things by the block
                 * numbers */

                log_debug("Ordering...");

                n = hashmap_size(files);
                if (!(ordered = new(struct item, n))) {
                        r = log_oom();
                        goto finish;
                }

                j = ordered;
                HASHMAP_FOREACH_KEY(q, p, files, i) {
                        memcpy(j, q, sizeof(struct item));
                        j++;
                }

                assert(ordered + n == j);

                qsort(ordered, n, sizeof(struct item), qsort_compare);

                for (k = 0; k < n; k++)
                        pack_file(pack, ordered[k].path, on_btrfs);

                free(ordered);
        }

        log_debug("Finalizing...");

        fflush(pack);

        if (ferror(pack)) {
                log_error("Failed to write pack file.");
                r = -EIO;
                goto finish;
        }

        if (rename(pack_fn_new, pack_fn) < 0) {
                log_error("Failed to rename readahead file: %m");
                r = -errno;
                goto finish;
        }

        fclose(pack);
        pack = NULL;

        log_debug("Done.");

finish:
        if (fanotify_fd >= 0)
                close_nointr_nofail(fanotify_fd);

        if (signal_fd >= 0)
                close_nointr_nofail(signal_fd);

        if (inotify_fd >= 0)
                close_nointr_nofail(inotify_fd);

        if (pack) {
                fclose(pack);
                unlink(pack_fn_new);
        }
        free(pack_fn_new);
        free(pack_fn);

        while ((p = hashmap_steal_first_key(files)))
                free(p);

        hashmap_free(files);

        if (previous_block_readahead_set) {
                uint64_t bytes;

                /* 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);
        }

        return r;
}

int main_collect(const char *root) {

        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 EXIT_SUCCESS;
        }

        if (!enough_ram()) {
                log_info("Disabling readahead collector due to low memory.");
                return EXIT_SUCCESS;
        }

        shared = shared_get();
        if (!shared)
                return EXIT_FAILURE;

        shared->collect = getpid();
        __sync_synchronize();

        if (collect(root) < 0)
                return EXIT_FAILURE;

        return EXIT_SUCCESS;
}