1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * probe-file.c : operate ftrace k/uprobe events files
5 * Written by Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
10 #include <sys/types.h>
13 #include <linux/zalloc.h>
14 #include "namespaces.h"
17 #include "strfilter.h"
24 #include <api/fs/tracing_path.h>
25 #include "probe-event.h"
26 #include "probe-file.h"
28 #include "perf_regs.h"
31 /* 4096 - 2 ('\n' + '\0') */
32 #define MAX_CMDLEN 4094
34 static void print_open_warning(int err, bool uprobe)
36 char sbuf[STRERR_BUFSIZE];
42 config = "CONFIG_UPROBE_EVENTS";
44 config = "CONFIG_KPROBE_EVENTS";
46 pr_warning("%cprobe_events file does not exist"
47 " - please rebuild kernel with %s.\n",
48 uprobe ? 'u' : 'k', config);
49 } else if (err == -ENOTSUP)
50 pr_warning("Tracefs or debugfs is not mounted.\n");
52 pr_warning("Failed to open %cprobe_events: %s\n",
54 str_error_r(-err, sbuf, sizeof(sbuf)));
57 static void print_both_open_warning(int kerr, int uerr)
59 /* Both kprobes and uprobes are disabled, warn it. */
60 if (kerr == -ENOTSUP && uerr == -ENOTSUP)
61 pr_warning("Tracefs or debugfs is not mounted.\n");
62 else if (kerr == -ENOENT && uerr == -ENOENT)
63 pr_warning("Please rebuild kernel with CONFIG_KPROBE_EVENTS "
64 "or/and CONFIG_UPROBE_EVENTS.\n");
66 char sbuf[STRERR_BUFSIZE];
67 pr_warning("Failed to open kprobe events: %s.\n",
68 str_error_r(-kerr, sbuf, sizeof(sbuf)));
69 pr_warning("Failed to open uprobe events: %s.\n",
70 str_error_r(-uerr, sbuf, sizeof(sbuf)));
74 int open_trace_file(const char *trace_file, bool readwrite)
79 ret = e_snprintf(buf, PATH_MAX, "%s/%s", tracing_path_mount(), trace_file);
81 pr_debug("Opening %s write=%d\n", buf, readwrite);
82 if (readwrite && !probe_event_dry_run)
83 ret = open(buf, O_RDWR | O_APPEND, 0);
85 ret = open(buf, O_RDONLY, 0);
93 static int open_kprobe_events(bool readwrite)
95 return open_trace_file("kprobe_events", readwrite);
98 static int open_uprobe_events(bool readwrite)
100 return open_trace_file("uprobe_events", readwrite);
103 int probe_file__open(int flag)
107 if (flag & PF_FL_UPROBE)
108 fd = open_uprobe_events(flag & PF_FL_RW);
110 fd = open_kprobe_events(flag & PF_FL_RW);
112 print_open_warning(fd, flag & PF_FL_UPROBE);
117 int probe_file__open_both(int *kfd, int *ufd, int flag)
122 *kfd = open_kprobe_events(flag & PF_FL_RW);
123 *ufd = open_uprobe_events(flag & PF_FL_RW);
124 if (*kfd < 0 && *ufd < 0) {
125 print_both_open_warning(*kfd, *ufd);
132 /* Get raw string list of current kprobe_events or uprobe_events */
133 struct strlist *probe_file__get_rawlist(int fd)
137 char buf[MAX_CMDLEN];
144 sl = strlist__new(NULL, NULL);
152 fp = fdopen(fddup, "r");
154 goto out_close_fddup;
157 p = fgets(buf, MAX_CMDLEN, fp);
164 ret = strlist__add(sl, buf);
166 pr_debug("strlist__add failed (%d)\n", ret);
184 static struct strlist *__probe_file__get_namelist(int fd, bool include_group)
187 struct strlist *sl, *rawlist;
188 struct str_node *ent;
189 struct probe_trace_event tev;
192 memset(&tev, 0, sizeof(tev));
193 rawlist = probe_file__get_rawlist(fd);
196 sl = strlist__new(NULL, NULL);
197 strlist__for_each_entry(ent, rawlist) {
198 ret = parse_probe_trace_command(ent->s, &tev);
202 ret = e_snprintf(buf, 128, "%s:%s", tev.group,
205 ret = strlist__add(sl, buf);
207 ret = strlist__add(sl, tev.event);
208 clear_probe_trace_event(&tev);
212 strlist__delete(rawlist);
221 /* Get current perf-probe event names */
222 struct strlist *probe_file__get_namelist(int fd)
224 return __probe_file__get_namelist(fd, false);
227 int probe_file__add_event(int fd, struct probe_trace_event *tev)
230 char *buf = synthesize_probe_trace_command(tev);
231 char sbuf[STRERR_BUFSIZE];
234 pr_debug("Failed to synthesize probe trace event.\n");
238 pr_debug("Writing event: %s\n", buf);
239 if (!probe_event_dry_run) {
240 if (write(fd, buf, strlen(buf)) < (int)strlen(buf)) {
242 pr_warning("Failed to write event: %s\n",
243 str_error_r(errno, sbuf, sizeof(sbuf)));
251 static int __del_trace_probe_event(int fd, struct str_node *ent)
257 /* Convert from perf-probe event to trace-probe event */
258 ret = e_snprintf(buf, 128, "-:%s", ent->s);
262 p = strchr(buf + 2, ':');
264 pr_debug("Internal error: %s should have ':' but not.\n",
271 pr_debug("Writing event: %s\n", buf);
272 ret = write(fd, buf, strlen(buf));
280 pr_warning("Failed to delete event: %s\n",
281 str_error_r(-ret, buf, sizeof(buf)));
285 int probe_file__get_events(int fd, struct strfilter *filter,
286 struct strlist *plist)
288 struct strlist *namelist;
289 struct str_node *ent;
296 namelist = __probe_file__get_namelist(fd, true);
300 strlist__for_each_entry(ent, namelist) {
301 p = strchr(ent->s, ':');
302 if ((p && strfilter__compare(filter, p + 1)) ||
303 strfilter__compare(filter, ent->s)) {
304 ret = strlist__add(plist, ent->s);
305 if (ret == -ENOMEM) {
306 pr_err("strlist__add failed with -ENOMEM\n");
313 strlist__delete(namelist);
318 int probe_file__del_strlist(int fd, struct strlist *namelist)
321 struct str_node *ent;
323 strlist__for_each_entry(ent, namelist) {
324 ret = __del_trace_probe_event(fd, ent);
331 int probe_file__del_events(int fd, struct strfilter *filter)
333 struct strlist *namelist;
336 namelist = strlist__new(NULL, NULL);
340 ret = probe_file__get_events(fd, filter, namelist);
344 ret = probe_file__del_strlist(fd, namelist);
345 strlist__delete(namelist);
350 /* Caller must ensure to remove this entry from list */
351 static void probe_cache_entry__delete(struct probe_cache_entry *entry)
354 BUG_ON(!list_empty(&entry->node));
356 strlist__delete(entry->tevlist);
357 clear_perf_probe_event(&entry->pev);
363 static struct probe_cache_entry *
364 probe_cache_entry__new(struct perf_probe_event *pev)
366 struct probe_cache_entry *entry = zalloc(sizeof(*entry));
369 INIT_LIST_HEAD(&entry->node);
370 entry->tevlist = strlist__new(NULL, NULL);
374 entry->spev = synthesize_perf_probe_command(pev);
376 perf_probe_event__copy(&entry->pev, pev) < 0) {
377 probe_cache_entry__delete(entry);
386 int probe_cache_entry__get_event(struct probe_cache_entry *entry,
387 struct probe_trace_event **tevs)
389 struct probe_trace_event *tev;
390 struct str_node *node;
393 ret = strlist__nr_entries(entry->tevlist);
394 if (ret > probe_conf.max_probes)
397 *tevs = zalloc(ret * sizeof(*tev));
402 strlist__for_each_entry(node, entry->tevlist) {
404 ret = parse_probe_trace_command(node->s, tev);
411 /* For the kernel probe caches, pass target = NULL or DSO__NAME_KALLSYMS */
412 static int probe_cache__open(struct probe_cache *pcache, const char *target,
415 char cpath[PATH_MAX];
416 char sbuildid[SBUILD_ID_SIZE];
417 char *dir_name = NULL;
418 bool is_kallsyms = false;
422 if (target && build_id_cache__cached(target)) {
423 /* This is a cached buildid */
424 strlcpy(sbuildid, target, SBUILD_ID_SIZE);
425 dir_name = build_id_cache__linkname(sbuildid, NULL, 0);
429 if (!target || !strcmp(target, DSO__NAME_KALLSYMS)) {
430 target = DSO__NAME_KALLSYMS;
432 ret = sysfs__sprintf_build_id("/", sbuildid);
434 nsinfo__mountns_enter(nsi, &nsc);
435 ret = filename__sprintf_build_id(target, sbuildid);
436 nsinfo__mountns_exit(&nsc);
440 pr_debug("Failed to get build-id from %s.\n", target);
444 /* If we have no buildid cache, make it */
445 if (!build_id_cache__cached(sbuildid)) {
446 ret = build_id_cache__add_s(sbuildid, target, nsi,
449 pr_debug("Failed to add build-id cache: %s\n", target);
454 dir_name = build_id_cache__cachedir(sbuildid, target, nsi, is_kallsyms,
458 pr_debug("Failed to get cache from %s\n", target);
462 snprintf(cpath, PATH_MAX, "%s/probes", dir_name);
463 fd = open(cpath, O_CREAT | O_RDWR, 0644);
465 pr_debug("Failed to open cache(%d): %s\n", fd, cpath);
472 static int probe_cache__load(struct probe_cache *pcache)
474 struct probe_cache_entry *entry = NULL;
475 char buf[MAX_CMDLEN], *p;
479 fddup = dup(pcache->fd);
482 fp = fdopen(fddup, "r");
489 if (!fgets(buf, MAX_CMDLEN, fp))
491 p = strchr(buf, '\n');
494 /* #perf_probe_event or %sdt_event */
495 if (buf[0] == '#' || buf[0] == '%') {
496 entry = probe_cache_entry__new(NULL);
503 entry->spev = strdup(buf + 1);
505 ret = parse_perf_probe_command(buf + 1,
510 probe_cache_entry__delete(entry);
513 list_add_tail(&entry->node, &pcache->entries);
514 } else { /* trace_probe_event */
519 ret = strlist__add(entry->tevlist, buf);
520 if (ret == -ENOMEM) {
521 pr_err("strlist__add failed with -ENOMEM\n");
531 static struct probe_cache *probe_cache__alloc(void)
533 struct probe_cache *pcache = zalloc(sizeof(*pcache));
536 INIT_LIST_HEAD(&pcache->entries);
537 pcache->fd = -EINVAL;
542 void probe_cache__purge(struct probe_cache *pcache)
544 struct probe_cache_entry *entry, *n;
546 list_for_each_entry_safe(entry, n, &pcache->entries, node) {
547 list_del_init(&entry->node);
548 probe_cache_entry__delete(entry);
552 void probe_cache__delete(struct probe_cache *pcache)
557 probe_cache__purge(pcache);
563 struct probe_cache *probe_cache__new(const char *target, struct nsinfo *nsi)
565 struct probe_cache *pcache = probe_cache__alloc();
571 ret = probe_cache__open(pcache, target, nsi);
573 pr_debug("Cache open error: %d\n", ret);
577 ret = probe_cache__load(pcache);
579 pr_debug("Cache read error: %d\n", ret);
586 probe_cache__delete(pcache);
590 static bool streql(const char *a, const char *b)
598 return !strcmp(a, b);
601 struct probe_cache_entry *
602 probe_cache__find(struct probe_cache *pcache, struct perf_probe_event *pev)
604 struct probe_cache_entry *entry = NULL;
605 char *cmd = synthesize_perf_probe_command(pev);
610 for_each_probe_cache_entry(entry, pcache) {
612 if (entry->pev.event &&
613 streql(entry->pev.event, pev->event) &&
615 streql(entry->pev.group, pev->group)))
620 /* Hit if same event name or same command-string */
622 (streql(entry->pev.group, pev->group) &&
623 streql(entry->pev.event, pev->event))) ||
624 (!strcmp(entry->spev, cmd)))
634 struct probe_cache_entry *
635 probe_cache__find_by_name(struct probe_cache *pcache,
636 const char *group, const char *event)
638 struct probe_cache_entry *entry = NULL;
640 for_each_probe_cache_entry(entry, pcache) {
641 /* Hit if same event name or same command-string */
642 if (streql(entry->pev.group, group) &&
643 streql(entry->pev.event, event))
652 int probe_cache__add_entry(struct probe_cache *pcache,
653 struct perf_probe_event *pev,
654 struct probe_trace_event *tevs, int ntevs)
656 struct probe_cache_entry *entry = NULL;
660 if (!pcache || !pev || !tevs || ntevs <= 0) {
665 /* Remove old cache entry */
666 entry = probe_cache__find(pcache, pev);
668 list_del_init(&entry->node);
669 probe_cache_entry__delete(entry);
673 entry = probe_cache_entry__new(pev);
677 for (i = 0; i < ntevs; i++) {
678 if (!tevs[i].point.symbol)
681 command = synthesize_probe_trace_command(&tevs[i]);
684 ret = strlist__add(entry->tevlist, command);
685 if (ret == -ENOMEM) {
686 pr_err("strlist__add failed with -ENOMEM\n");
692 list_add_tail(&entry->node, &pcache->entries);
693 pr_debug("Added probe cache: %d\n", ntevs);
697 pr_debug("Failed to add probe caches\n");
698 probe_cache_entry__delete(entry);
702 #ifdef HAVE_GELF_GETNOTE_SUPPORT
703 static unsigned long long sdt_note__get_addr(struct sdt_note *note)
706 (unsigned long long)note->addr.a32[SDT_NOTE_IDX_LOC] :
707 (unsigned long long)note->addr.a64[SDT_NOTE_IDX_LOC];
710 static unsigned long long sdt_note__get_ref_ctr_offset(struct sdt_note *note)
713 (unsigned long long)note->addr.a32[SDT_NOTE_IDX_REFCTR] :
714 (unsigned long long)note->addr.a64[SDT_NOTE_IDX_REFCTR];
717 static const char * const type_to_suffix[] = {
718 ":s64", "", "", "", ":s32", "", ":s16", ":s8",
719 "", ":u8", ":u16", "", ":u32", "", "", "", ":u64"
723 * Isolate the string number and convert it into a decimal value;
724 * this will be an index to get suffix of the uprobe name (defining
727 static int sdt_arg_parse_size(char *n_ptr, const char **suffix)
731 type_idx = strtol(n_ptr, NULL, 10);
732 if (type_idx < -8 || type_idx > 8) {
733 pr_debug4("Failed to get a valid sdt type\n");
737 *suffix = type_to_suffix[type_idx + 8];
741 static int synthesize_sdt_probe_arg(struct strbuf *buf, int i, const char *arg)
743 char *op, *desc = strdup(arg), *new_op = NULL;
744 const char *suffix = "";
748 pr_debug4("Allocation error\n");
753 * Argument is in N@OP format. N is size of the argument and OP is
754 * the actual assembly operand. N can be omitted; in that case
755 * argument is just OP(without @).
757 op = strchr(desc, '@');
762 if (sdt_arg_parse_size(desc, &suffix))
768 ret = arch_sdt_arg_parse_op(op, &new_op);
773 if (ret == SDT_ARG_VALID) {
774 ret = strbuf_addf(buf, " arg%d=%s%s", i + 1, new_op, suffix);
786 static char *synthesize_sdt_probe_command(struct sdt_note *note,
787 const char *pathname,
791 char *ret = NULL, **args;
792 int i, args_count, err;
793 unsigned long long ref_ctr_offset;
795 if (strbuf_init(&buf, 32) < 0)
798 err = strbuf_addf(&buf, "p:%s/%s %s:0x%llx",
799 sdtgrp, note->name, pathname,
800 sdt_note__get_addr(note));
802 ref_ctr_offset = sdt_note__get_ref_ctr_offset(note);
803 if (ref_ctr_offset && err >= 0)
804 err = strbuf_addf(&buf, "(0x%llx)", ref_ctr_offset);
813 args = argv_split(note->args, &args_count);
815 for (i = 0; i < args_count; ++i) {
816 if (synthesize_sdt_probe_arg(&buf, i, args[i]) < 0)
822 ret = strbuf_detach(&buf, NULL);
824 strbuf_release(&buf);
828 int probe_cache__scan_sdt(struct probe_cache *pcache, const char *pathname)
830 struct probe_cache_entry *entry = NULL;
831 struct list_head sdtlist;
832 struct sdt_note *note;
837 INIT_LIST_HEAD(&sdtlist);
838 ret = get_sdt_note_list(&sdtlist, pathname);
840 pr_debug4("Failed to get sdt note: %d\n", ret);
843 list_for_each_entry(note, &sdtlist, note_list) {
844 ret = snprintf(sdtgrp, 64, "sdt_%s", note->provider);
847 /* Try to find same-name entry */
848 entry = probe_cache__find_by_name(pcache, sdtgrp, note->name);
850 entry = probe_cache_entry__new(NULL);
856 ret = asprintf(&entry->spev, "%s:%s=%s", sdtgrp,
857 note->name, note->name);
860 entry->pev.event = strdup(note->name);
861 entry->pev.group = strdup(sdtgrp);
862 list_add_tail(&entry->node, &pcache->entries);
864 buf = synthesize_sdt_probe_command(note, pathname, sdtgrp);
870 ret = strlist__add(entry->tevlist, buf);
875 if (ret == -ENOMEM) {
876 pr_err("strlist__add failed with -ENOMEM\n");
881 list_del_init(&entry->node);
882 probe_cache_entry__delete(entry);
884 cleanup_sdt_note_list(&sdtlist);
889 static int probe_cache_entry__write(struct probe_cache_entry *entry, int fd)
891 struct str_node *snode;
894 const char *prefix = entry->sdt ? "%" : "#";
896 /* Save stat for rollback */
897 ret = fstat(fd, &st);
901 pr_debug("Writing cache: %s%s\n", prefix, entry->spev);
902 iov[0].iov_base = (void *)prefix; iov[0].iov_len = 1;
903 iov[1].iov_base = entry->spev; iov[1].iov_len = strlen(entry->spev);
904 iov[2].iov_base = (void *)"\n"; iov[2].iov_len = 1;
905 ret = writev(fd, iov, 3);
906 if (ret < (int)iov[1].iov_len + 2)
909 strlist__for_each_entry(snode, entry->tevlist) {
910 iov[0].iov_base = (void *)snode->s;
911 iov[0].iov_len = strlen(snode->s);
912 iov[1].iov_base = (void *)"\n"; iov[1].iov_len = 1;
913 ret = writev(fd, iov, 2);
914 if (ret < (int)iov[0].iov_len + 1)
920 /* Rollback to avoid cache file corruption */
923 if (ftruncate(fd, st.st_size) < 0)
929 int probe_cache__commit(struct probe_cache *pcache)
931 struct probe_cache_entry *entry;
934 /* TBD: if we do not update existing entries, skip it */
935 ret = lseek(pcache->fd, 0, SEEK_SET);
939 ret = ftruncate(pcache->fd, 0);
943 for_each_probe_cache_entry(entry, pcache) {
944 ret = probe_cache_entry__write(entry, pcache->fd);
945 pr_debug("Cache committed: %d\n", ret);
953 static bool probe_cache_entry__compare(struct probe_cache_entry *entry,
954 struct strfilter *filter)
956 char buf[128], *ptr = entry->spev;
958 if (entry->pev.event) {
959 snprintf(buf, 128, "%s:%s", entry->pev.group, entry->pev.event);
962 return strfilter__compare(filter, ptr);
965 int probe_cache__filter_purge(struct probe_cache *pcache,
966 struct strfilter *filter)
968 struct probe_cache_entry *entry, *tmp;
970 list_for_each_entry_safe(entry, tmp, &pcache->entries, node) {
971 if (probe_cache_entry__compare(entry, filter)) {
972 pr_info("Removed cached event: %s\n", entry->spev);
973 list_del_init(&entry->node);
974 probe_cache_entry__delete(entry);
980 static int probe_cache__show_entries(struct probe_cache *pcache,
981 struct strfilter *filter)
983 struct probe_cache_entry *entry;
985 for_each_probe_cache_entry(entry, pcache) {
986 if (probe_cache_entry__compare(entry, filter))
987 printf("%s\n", entry->spev);
992 /* Show all cached probes */
993 int probe_cache__show_all_caches(struct strfilter *filter)
995 struct probe_cache *pcache;
996 struct strlist *bidlist;
998 char *buf = strfilter__string(filter);
1000 pr_debug("list cache with filter: %s\n", buf);
1003 bidlist = build_id_cache__list_all(true);
1005 pr_debug("Failed to get buildids: %d\n", errno);
1008 strlist__for_each_entry(nd, bidlist) {
1009 pcache = probe_cache__new(nd->s, NULL);
1012 if (!list_empty(&pcache->entries)) {
1013 buf = build_id_cache__origname(nd->s);
1014 printf("%s (%s):\n", buf, nd->s);
1016 probe_cache__show_entries(pcache, filter);
1018 probe_cache__delete(pcache);
1020 strlist__delete(bidlist);
1025 enum ftrace_readme {
1026 FTRACE_README_PROBE_TYPE_X = 0,
1027 FTRACE_README_KRETPROBE_OFFSET,
1028 FTRACE_README_UPROBE_REF_CTR,
1029 FTRACE_README_USER_ACCESS,
1030 FTRACE_README_MULTIPROBE_EVENT,
1031 FTRACE_README_IMMEDIATE_VALUE,
1036 const char *pattern;
1038 } ftrace_readme_table[] = {
1039 #define DEFINE_TYPE(idx, pat) \
1040 [idx] = {.pattern = pat, .avail = false}
1041 DEFINE_TYPE(FTRACE_README_PROBE_TYPE_X, "*type: * x8/16/32/64,*"),
1042 DEFINE_TYPE(FTRACE_README_KRETPROBE_OFFSET, "*place (kretprobe): *"),
1043 DEFINE_TYPE(FTRACE_README_UPROBE_REF_CTR, "*ref_ctr_offset*"),
1044 DEFINE_TYPE(FTRACE_README_USER_ACCESS, "*[u]<offset>*"),
1045 DEFINE_TYPE(FTRACE_README_MULTIPROBE_EVENT, "*Create/append/*"),
1046 DEFINE_TYPE(FTRACE_README_IMMEDIATE_VALUE, "*\\imm-value,*"),
1049 static bool scan_ftrace_readme(enum ftrace_readme type)
1056 static bool scanned = false;
1061 fd = open_trace_file("README", false);
1065 fp = fdopen(fd, "r");
1071 while (getline(&buf, &len, fp) > 0)
1072 for (enum ftrace_readme i = 0; i < FTRACE_README_END; i++)
1073 if (!ftrace_readme_table[i].avail)
1074 ftrace_readme_table[i].avail =
1075 strglobmatch(buf, ftrace_readme_table[i].pattern);
1082 if (type >= FTRACE_README_END)
1085 return ftrace_readme_table[type].avail;
1088 bool probe_type_is_available(enum probe_type type)
1090 if (type >= PROBE_TYPE_END)
1092 else if (type == PROBE_TYPE_X)
1093 return scan_ftrace_readme(FTRACE_README_PROBE_TYPE_X);
1098 bool kretprobe_offset_is_supported(void)
1100 return scan_ftrace_readme(FTRACE_README_KRETPROBE_OFFSET);
1103 bool uprobe_ref_ctr_is_supported(void)
1105 return scan_ftrace_readme(FTRACE_README_UPROBE_REF_CTR);
1108 bool user_access_is_supported(void)
1110 return scan_ftrace_readme(FTRACE_README_USER_ACCESS);
1113 bool multiprobe_event_is_supported(void)
1115 return scan_ftrace_readme(FTRACE_README_MULTIPROBE_EVENT);
1118 bool immediate_value_is_supported(void)
1120 return scan_ftrace_readme(FTRACE_README_IMMEDIATE_VALUE);