]> asedeno.scripts.mit.edu Git - linux.git/blob - kernel/trace/trace_uprobe.c
Merge tag 'armsoc-dt' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
[linux.git] / kernel / trace / trace_uprobe.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * uprobes-based tracing events
4  *
5  * Copyright (C) IBM Corporation, 2010-2012
6  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
7  */
8 #define pr_fmt(fmt)     "trace_uprobe: " fmt
9
10 #include <linux/ctype.h>
11 #include <linux/module.h>
12 #include <linux/uaccess.h>
13 #include <linux/uprobes.h>
14 #include <linux/namei.h>
15 #include <linux/string.h>
16 #include <linux/rculist.h>
17
18 #include "trace_dynevent.h"
19 #include "trace_probe.h"
20 #include "trace_probe_tmpl.h"
21
22 #define UPROBE_EVENT_SYSTEM     "uprobes"
23
24 struct uprobe_trace_entry_head {
25         struct trace_entry      ent;
26         unsigned long           vaddr[];
27 };
28
29 #define SIZEOF_TRACE_ENTRY(is_return)                   \
30         (sizeof(struct uprobe_trace_entry_head) +       \
31          sizeof(unsigned long) * (is_return ? 2 : 1))
32
33 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
34         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
35
36 struct trace_uprobe_filter {
37         rwlock_t                rwlock;
38         int                     nr_systemwide;
39         struct list_head        perf_events;
40 };
41
42 static int trace_uprobe_create(int argc, const char **argv);
43 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
44 static int trace_uprobe_release(struct dyn_event *ev);
45 static bool trace_uprobe_is_busy(struct dyn_event *ev);
46 static bool trace_uprobe_match(const char *system, const char *event,
47                                struct dyn_event *ev);
48
49 static struct dyn_event_operations trace_uprobe_ops = {
50         .create = trace_uprobe_create,
51         .show = trace_uprobe_show,
52         .is_busy = trace_uprobe_is_busy,
53         .free = trace_uprobe_release,
54         .match = trace_uprobe_match,
55 };
56
57 /*
58  * uprobe event core functions
59  */
60 struct trace_uprobe {
61         struct dyn_event                devent;
62         struct trace_uprobe_filter      filter;
63         struct uprobe_consumer          consumer;
64         struct path                     path;
65         struct inode                    *inode;
66         char                            *filename;
67         unsigned long                   offset;
68         unsigned long                   ref_ctr_offset;
69         unsigned long                   nhit;
70         struct trace_probe              tp;
71 };
72
73 static bool is_trace_uprobe(struct dyn_event *ev)
74 {
75         return ev->ops == &trace_uprobe_ops;
76 }
77
78 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
79 {
80         return container_of(ev, struct trace_uprobe, devent);
81 }
82
83 /**
84  * for_each_trace_uprobe - iterate over the trace_uprobe list
85  * @pos:        the struct trace_uprobe * for each entry
86  * @dpos:       the struct dyn_event * to use as a loop cursor
87  */
88 #define for_each_trace_uprobe(pos, dpos)        \
89         for_each_dyn_event(dpos)                \
90                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
91
92 #define SIZEOF_TRACE_UPROBE(n)                          \
93         (offsetof(struct trace_uprobe, tp.args) +       \
94         (sizeof(struct probe_arg) * (n)))
95
96 static int register_uprobe_event(struct trace_uprobe *tu);
97 static int unregister_uprobe_event(struct trace_uprobe *tu);
98
99 struct uprobe_dispatch_data {
100         struct trace_uprobe     *tu;
101         unsigned long           bp_addr;
102 };
103
104 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
105 static int uretprobe_dispatcher(struct uprobe_consumer *con,
106                                 unsigned long func, struct pt_regs *regs);
107
108 #ifdef CONFIG_STACK_GROWSUP
109 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
110 {
111         return addr - (n * sizeof(long));
112 }
113 #else
114 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
115 {
116         return addr + (n * sizeof(long));
117 }
118 #endif
119
120 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
121 {
122         unsigned long ret;
123         unsigned long addr = user_stack_pointer(regs);
124
125         addr = adjust_stack_addr(addr, n);
126
127         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
128                 return 0;
129
130         return ret;
131 }
132
133 /*
134  * Uprobes-specific fetch functions
135  */
136 static nokprobe_inline int
137 probe_mem_read(void *dest, void *src, size_t size)
138 {
139         void __user *vaddr = (void __force __user *)src;
140
141         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
142 }
143
144 static nokprobe_inline int
145 probe_mem_read_user(void *dest, void *src, size_t size)
146 {
147         return probe_mem_read(dest, src, size);
148 }
149
150 /*
151  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
152  * length and relative data location.
153  */
154 static nokprobe_inline int
155 fetch_store_string(unsigned long addr, void *dest, void *base)
156 {
157         long ret;
158         u32 loc = *(u32 *)dest;
159         int maxlen  = get_loc_len(loc);
160         u8 *dst = get_loc_data(dest, base);
161         void __user *src = (void __force __user *) addr;
162
163         if (unlikely(!maxlen))
164                 return -ENOMEM;
165
166         if (addr == FETCH_TOKEN_COMM)
167                 ret = strlcpy(dst, current->comm, maxlen);
168         else
169                 ret = strncpy_from_user(dst, src, maxlen);
170         if (ret >= 0) {
171                 if (ret == maxlen)
172                         dst[ret - 1] = '\0';
173                 else
174                         /*
175                          * Include the terminating null byte. In this case it
176                          * was copied by strncpy_from_user but not accounted
177                          * for in ret.
178                          */
179                         ret++;
180                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
181         }
182
183         return ret;
184 }
185
186 static nokprobe_inline int
187 fetch_store_string_user(unsigned long addr, void *dest, void *base)
188 {
189         return fetch_store_string(addr, dest, base);
190 }
191
192 /* Return the length of string -- including null terminal byte */
193 static nokprobe_inline int
194 fetch_store_strlen(unsigned long addr)
195 {
196         int len;
197         void __user *vaddr = (void __force __user *) addr;
198
199         if (addr == FETCH_TOKEN_COMM)
200                 len = strlen(current->comm) + 1;
201         else
202                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
203
204         return (len > MAX_STRING_SIZE) ? 0 : len;
205 }
206
207 static nokprobe_inline int
208 fetch_store_strlen_user(unsigned long addr)
209 {
210         return fetch_store_strlen(addr);
211 }
212
213 static unsigned long translate_user_vaddr(unsigned long file_offset)
214 {
215         unsigned long base_addr;
216         struct uprobe_dispatch_data *udd;
217
218         udd = (void *) current->utask->vaddr;
219
220         base_addr = udd->bp_addr - udd->tu->offset;
221         return base_addr + file_offset;
222 }
223
224 /* Note that we don't verify it, since the code does not come from user space */
225 static int
226 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
227                    void *base)
228 {
229         unsigned long val;
230
231         /* 1st stage: get value from context */
232         switch (code->op) {
233         case FETCH_OP_REG:
234                 val = regs_get_register(regs, code->param);
235                 break;
236         case FETCH_OP_STACK:
237                 val = get_user_stack_nth(regs, code->param);
238                 break;
239         case FETCH_OP_STACKP:
240                 val = user_stack_pointer(regs);
241                 break;
242         case FETCH_OP_RETVAL:
243                 val = regs_return_value(regs);
244                 break;
245         case FETCH_OP_IMM:
246                 val = code->immediate;
247                 break;
248         case FETCH_OP_COMM:
249                 val = FETCH_TOKEN_COMM;
250                 break;
251         case FETCH_OP_FOFFS:
252                 val = translate_user_vaddr(code->immediate);
253                 break;
254         default:
255                 return -EILSEQ;
256         }
257         code++;
258
259         return process_fetch_insn_bottom(code, val, dest, base);
260 }
261 NOKPROBE_SYMBOL(process_fetch_insn)
262
263 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
264 {
265         rwlock_init(&filter->rwlock);
266         filter->nr_systemwide = 0;
267         INIT_LIST_HEAD(&filter->perf_events);
268 }
269
270 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
271 {
272         return !filter->nr_systemwide && list_empty(&filter->perf_events);
273 }
274
275 static inline bool is_ret_probe(struct trace_uprobe *tu)
276 {
277         return tu->consumer.ret_handler != NULL;
278 }
279
280 static bool trace_uprobe_is_busy(struct dyn_event *ev)
281 {
282         struct trace_uprobe *tu = to_trace_uprobe(ev);
283
284         return trace_probe_is_enabled(&tu->tp);
285 }
286
287 static bool trace_uprobe_match(const char *system, const char *event,
288                                struct dyn_event *ev)
289 {
290         struct trace_uprobe *tu = to_trace_uprobe(ev);
291
292         return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
293             (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0);
294 }
295
296 /*
297  * Allocate new trace_uprobe and initialize it (including uprobes).
298  */
299 static struct trace_uprobe *
300 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
301 {
302         struct trace_uprobe *tu;
303         int ret;
304
305         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
306         if (!tu)
307                 return ERR_PTR(-ENOMEM);
308
309         ret = trace_probe_init(&tu->tp, event, group);
310         if (ret < 0)
311                 goto error;
312
313         dyn_event_init(&tu->devent, &trace_uprobe_ops);
314         tu->consumer.handler = uprobe_dispatcher;
315         if (is_ret)
316                 tu->consumer.ret_handler = uretprobe_dispatcher;
317         init_trace_uprobe_filter(&tu->filter);
318         return tu;
319
320 error:
321         kfree(tu);
322
323         return ERR_PTR(ret);
324 }
325
326 static void free_trace_uprobe(struct trace_uprobe *tu)
327 {
328         if (!tu)
329                 return;
330
331         path_put(&tu->path);
332         trace_probe_cleanup(&tu->tp);
333         kfree(tu->filename);
334         kfree(tu);
335 }
336
337 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
338 {
339         struct dyn_event *pos;
340         struct trace_uprobe *tu;
341
342         for_each_trace_uprobe(tu, pos)
343                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
344                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
345                         return tu;
346
347         return NULL;
348 }
349
350 /* Unregister a trace_uprobe and probe_event */
351 static int unregister_trace_uprobe(struct trace_uprobe *tu)
352 {
353         int ret;
354
355         ret = unregister_uprobe_event(tu);
356         if (ret)
357                 return ret;
358
359         dyn_event_remove(&tu->devent);
360         free_trace_uprobe(tu);
361         return 0;
362 }
363
364 /*
365  * Uprobe with multiple reference counter is not allowed. i.e.
366  * If inode and offset matches, reference counter offset *must*
367  * match as well. Though, there is one exception: If user is
368  * replacing old trace_uprobe with new one(same group/event),
369  * then we allow same uprobe with new reference counter as far
370  * as the new one does not conflict with any other existing
371  * ones.
372  */
373 static struct trace_uprobe *find_old_trace_uprobe(struct trace_uprobe *new)
374 {
375         struct dyn_event *pos;
376         struct trace_uprobe *tmp, *old = NULL;
377         struct inode *new_inode = d_real_inode(new->path.dentry);
378
379         old = find_probe_event(trace_probe_name(&new->tp),
380                                 trace_probe_group_name(&new->tp));
381
382         for_each_trace_uprobe(tmp, pos) {
383                 if ((old ? old != tmp : true) &&
384                     new_inode == d_real_inode(tmp->path.dentry) &&
385                     new->offset == tmp->offset &&
386                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
387                         pr_warn("Reference counter offset mismatch.");
388                         return ERR_PTR(-EINVAL);
389                 }
390         }
391         return old;
392 }
393
394 /* Register a trace_uprobe and probe_event */
395 static int register_trace_uprobe(struct trace_uprobe *tu)
396 {
397         struct trace_uprobe *old_tu;
398         int ret;
399
400         mutex_lock(&event_mutex);
401
402         /* register as an event */
403         old_tu = find_old_trace_uprobe(tu);
404         if (IS_ERR(old_tu)) {
405                 ret = PTR_ERR(old_tu);
406                 goto end;
407         }
408
409         if (old_tu) {
410                 /* delete old event */
411                 ret = unregister_trace_uprobe(old_tu);
412                 if (ret)
413                         goto end;
414         }
415
416         ret = register_uprobe_event(tu);
417         if (ret) {
418                 pr_warn("Failed to register probe event(%d)\n", ret);
419                 goto end;
420         }
421
422         dyn_event_add(&tu->devent);
423
424 end:
425         mutex_unlock(&event_mutex);
426
427         return ret;
428 }
429
430 /*
431  * Argument syntax:
432  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
433  */
434 static int trace_uprobe_create(int argc, const char **argv)
435 {
436         struct trace_uprobe *tu;
437         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
438         char *arg, *filename, *rctr, *rctr_end, *tmp;
439         char buf[MAX_EVENT_NAME_LEN];
440         struct path path;
441         unsigned long offset, ref_ctr_offset;
442         bool is_return = false;
443         int i, ret;
444
445         ret = 0;
446         ref_ctr_offset = 0;
447
448         switch (argv[0][0]) {
449         case 'r':
450                 is_return = true;
451                 break;
452         case 'p':
453                 break;
454         default:
455                 return -ECANCELED;
456         }
457
458         if (argc < 2)
459                 return -ECANCELED;
460
461         if (argv[0][1] == ':')
462                 event = &argv[0][2];
463
464         if (!strchr(argv[1], '/'))
465                 return -ECANCELED;
466
467         filename = kstrdup(argv[1], GFP_KERNEL);
468         if (!filename)
469                 return -ENOMEM;
470
471         /* Find the last occurrence, in case the path contains ':' too. */
472         arg = strrchr(filename, ':');
473         if (!arg || !isdigit(arg[1])) {
474                 kfree(filename);
475                 return -ECANCELED;
476         }
477
478         trace_probe_log_init("trace_uprobe", argc, argv);
479         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
480
481         *arg++ = '\0';
482         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
483         if (ret) {
484                 trace_probe_log_err(0, FILE_NOT_FOUND);
485                 kfree(filename);
486                 trace_probe_log_clear();
487                 return ret;
488         }
489         if (!d_is_reg(path.dentry)) {
490                 trace_probe_log_err(0, NO_REGULAR_FILE);
491                 ret = -EINVAL;
492                 goto fail_address_parse;
493         }
494
495         /* Parse reference counter offset if specified. */
496         rctr = strchr(arg, '(');
497         if (rctr) {
498                 rctr_end = strchr(rctr, ')');
499                 if (!rctr_end) {
500                         ret = -EINVAL;
501                         rctr_end = rctr + strlen(rctr);
502                         trace_probe_log_err(rctr_end - filename,
503                                             REFCNT_OPEN_BRACE);
504                         goto fail_address_parse;
505                 } else if (rctr_end[1] != '\0') {
506                         ret = -EINVAL;
507                         trace_probe_log_err(rctr_end + 1 - filename,
508                                             BAD_REFCNT_SUFFIX);
509                         goto fail_address_parse;
510                 }
511
512                 *rctr++ = '\0';
513                 *rctr_end = '\0';
514                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
515                 if (ret) {
516                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
517                         goto fail_address_parse;
518                 }
519         }
520
521         /* Parse uprobe offset. */
522         ret = kstrtoul(arg, 0, &offset);
523         if (ret) {
524                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
525                 goto fail_address_parse;
526         }
527
528         /* setup a probe */
529         trace_probe_log_set_index(0);
530         if (event) {
531                 ret = traceprobe_parse_event_name(&event, &group, buf,
532                                                   event - argv[0]);
533                 if (ret)
534                         goto fail_address_parse;
535         } else {
536                 char *tail;
537                 char *ptr;
538
539                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
540                 if (!tail) {
541                         ret = -ENOMEM;
542                         goto fail_address_parse;
543                 }
544
545                 ptr = strpbrk(tail, ".-_");
546                 if (ptr)
547                         *ptr = '\0';
548
549                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
550                 event = buf;
551                 kfree(tail);
552         }
553
554         argc -= 2;
555         argv += 2;
556
557         tu = alloc_trace_uprobe(group, event, argc, is_return);
558         if (IS_ERR(tu)) {
559                 ret = PTR_ERR(tu);
560                 /* This must return -ENOMEM otherwise there is a bug */
561                 WARN_ON_ONCE(ret != -ENOMEM);
562                 goto fail_address_parse;
563         }
564         tu->offset = offset;
565         tu->ref_ctr_offset = ref_ctr_offset;
566         tu->path = path;
567         tu->filename = filename;
568
569         /* parse arguments */
570         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
571                 tmp = kstrdup(argv[i], GFP_KERNEL);
572                 if (!tmp) {
573                         ret = -ENOMEM;
574                         goto error;
575                 }
576
577                 trace_probe_log_set_index(i + 2);
578                 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
579                                         is_return ? TPARG_FL_RETURN : 0);
580                 kfree(tmp);
581                 if (ret)
582                         goto error;
583         }
584
585         ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
586         if (ret < 0)
587                 goto error;
588
589         ret = register_trace_uprobe(tu);
590         if (!ret)
591                 goto out;
592
593 error:
594         free_trace_uprobe(tu);
595 out:
596         trace_probe_log_clear();
597         return ret;
598
599 fail_address_parse:
600         trace_probe_log_clear();
601         path_put(&path);
602         kfree(filename);
603
604         return ret;
605 }
606
607 static int create_or_delete_trace_uprobe(int argc, char **argv)
608 {
609         int ret;
610
611         if (argv[0][0] == '-')
612                 return dyn_event_release(argc, argv, &trace_uprobe_ops);
613
614         ret = trace_uprobe_create(argc, (const char **)argv);
615         return ret == -ECANCELED ? -EINVAL : ret;
616 }
617
618 static int trace_uprobe_release(struct dyn_event *ev)
619 {
620         struct trace_uprobe *tu = to_trace_uprobe(ev);
621
622         return unregister_trace_uprobe(tu);
623 }
624
625 /* Probes listing interfaces */
626 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
627 {
628         struct trace_uprobe *tu = to_trace_uprobe(ev);
629         char c = is_ret_probe(tu) ? 'r' : 'p';
630         int i;
631
632         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
633                         trace_probe_name(&tu->tp), tu->filename,
634                         (int)(sizeof(void *) * 2), tu->offset);
635
636         if (tu->ref_ctr_offset)
637                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
638
639         for (i = 0; i < tu->tp.nr_args; i++)
640                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
641
642         seq_putc(m, '\n');
643         return 0;
644 }
645
646 static int probes_seq_show(struct seq_file *m, void *v)
647 {
648         struct dyn_event *ev = v;
649
650         if (!is_trace_uprobe(ev))
651                 return 0;
652
653         return trace_uprobe_show(m, ev);
654 }
655
656 static const struct seq_operations probes_seq_op = {
657         .start  = dyn_event_seq_start,
658         .next   = dyn_event_seq_next,
659         .stop   = dyn_event_seq_stop,
660         .show   = probes_seq_show
661 };
662
663 static int probes_open(struct inode *inode, struct file *file)
664 {
665         int ret;
666
667         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
668                 ret = dyn_events_release_all(&trace_uprobe_ops);
669                 if (ret)
670                         return ret;
671         }
672
673         return seq_open(file, &probes_seq_op);
674 }
675
676 static ssize_t probes_write(struct file *file, const char __user *buffer,
677                             size_t count, loff_t *ppos)
678 {
679         return trace_parse_run_command(file, buffer, count, ppos,
680                                         create_or_delete_trace_uprobe);
681 }
682
683 static const struct file_operations uprobe_events_ops = {
684         .owner          = THIS_MODULE,
685         .open           = probes_open,
686         .read           = seq_read,
687         .llseek         = seq_lseek,
688         .release        = seq_release,
689         .write          = probes_write,
690 };
691
692 /* Probes profiling interfaces */
693 static int probes_profile_seq_show(struct seq_file *m, void *v)
694 {
695         struct dyn_event *ev = v;
696         struct trace_uprobe *tu;
697
698         if (!is_trace_uprobe(ev))
699                 return 0;
700
701         tu = to_trace_uprobe(ev);
702         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
703                         trace_probe_name(&tu->tp), tu->nhit);
704         return 0;
705 }
706
707 static const struct seq_operations profile_seq_op = {
708         .start  = dyn_event_seq_start,
709         .next   = dyn_event_seq_next,
710         .stop   = dyn_event_seq_stop,
711         .show   = probes_profile_seq_show
712 };
713
714 static int profile_open(struct inode *inode, struct file *file)
715 {
716         return seq_open(file, &profile_seq_op);
717 }
718
719 static const struct file_operations uprobe_profile_ops = {
720         .owner          = THIS_MODULE,
721         .open           = profile_open,
722         .read           = seq_read,
723         .llseek         = seq_lseek,
724         .release        = seq_release,
725 };
726
727 struct uprobe_cpu_buffer {
728         struct mutex mutex;
729         void *buf;
730 };
731 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
732 static int uprobe_buffer_refcnt;
733
734 static int uprobe_buffer_init(void)
735 {
736         int cpu, err_cpu;
737
738         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
739         if (uprobe_cpu_buffer == NULL)
740                 return -ENOMEM;
741
742         for_each_possible_cpu(cpu) {
743                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
744                                                   GFP_KERNEL, 0);
745                 if (p == NULL) {
746                         err_cpu = cpu;
747                         goto err;
748                 }
749                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
750                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
751         }
752
753         return 0;
754
755 err:
756         for_each_possible_cpu(cpu) {
757                 if (cpu == err_cpu)
758                         break;
759                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
760         }
761
762         free_percpu(uprobe_cpu_buffer);
763         return -ENOMEM;
764 }
765
766 static int uprobe_buffer_enable(void)
767 {
768         int ret = 0;
769
770         BUG_ON(!mutex_is_locked(&event_mutex));
771
772         if (uprobe_buffer_refcnt++ == 0) {
773                 ret = uprobe_buffer_init();
774                 if (ret < 0)
775                         uprobe_buffer_refcnt--;
776         }
777
778         return ret;
779 }
780
781 static void uprobe_buffer_disable(void)
782 {
783         int cpu;
784
785         BUG_ON(!mutex_is_locked(&event_mutex));
786
787         if (--uprobe_buffer_refcnt == 0) {
788                 for_each_possible_cpu(cpu)
789                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
790                                                              cpu)->buf);
791
792                 free_percpu(uprobe_cpu_buffer);
793                 uprobe_cpu_buffer = NULL;
794         }
795 }
796
797 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
798 {
799         struct uprobe_cpu_buffer *ucb;
800         int cpu;
801
802         cpu = raw_smp_processor_id();
803         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
804
805         /*
806          * Use per-cpu buffers for fastest access, but we might migrate
807          * so the mutex makes sure we have sole access to it.
808          */
809         mutex_lock(&ucb->mutex);
810
811         return ucb;
812 }
813
814 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
815 {
816         mutex_unlock(&ucb->mutex);
817 }
818
819 static void __uprobe_trace_func(struct trace_uprobe *tu,
820                                 unsigned long func, struct pt_regs *regs,
821                                 struct uprobe_cpu_buffer *ucb, int dsize,
822                                 struct trace_event_file *trace_file)
823 {
824         struct uprobe_trace_entry_head *entry;
825         struct ring_buffer_event *event;
826         struct ring_buffer *buffer;
827         void *data;
828         int size, esize;
829         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
830
831         WARN_ON(call != trace_file->event_call);
832
833         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
834                 return;
835
836         if (trace_trigger_soft_disabled(trace_file))
837                 return;
838
839         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
840         size = esize + tu->tp.size + dsize;
841         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
842                                                 call->event.type, size, 0, 0);
843         if (!event)
844                 return;
845
846         entry = ring_buffer_event_data(event);
847         if (is_ret_probe(tu)) {
848                 entry->vaddr[0] = func;
849                 entry->vaddr[1] = instruction_pointer(regs);
850                 data = DATAOF_TRACE_ENTRY(entry, true);
851         } else {
852                 entry->vaddr[0] = instruction_pointer(regs);
853                 data = DATAOF_TRACE_ENTRY(entry, false);
854         }
855
856         memcpy(data, ucb->buf, tu->tp.size + dsize);
857
858         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
859 }
860
861 /* uprobe handler */
862 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
863                              struct uprobe_cpu_buffer *ucb, int dsize)
864 {
865         struct event_file_link *link;
866
867         if (is_ret_probe(tu))
868                 return 0;
869
870         rcu_read_lock();
871         trace_probe_for_each_link_rcu(link, &tu->tp)
872                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
873         rcu_read_unlock();
874
875         return 0;
876 }
877
878 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
879                                  struct pt_regs *regs,
880                                  struct uprobe_cpu_buffer *ucb, int dsize)
881 {
882         struct event_file_link *link;
883
884         rcu_read_lock();
885         trace_probe_for_each_link_rcu(link, &tu->tp)
886                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
887         rcu_read_unlock();
888 }
889
890 /* Event entry printers */
891 static enum print_line_t
892 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
893 {
894         struct uprobe_trace_entry_head *entry;
895         struct trace_seq *s = &iter->seq;
896         struct trace_uprobe *tu;
897         u8 *data;
898
899         entry = (struct uprobe_trace_entry_head *)iter->ent;
900         tu = container_of(event, struct trace_uprobe, tp.call.event);
901
902         if (is_ret_probe(tu)) {
903                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
904                                  trace_probe_name(&tu->tp),
905                                  entry->vaddr[1], entry->vaddr[0]);
906                 data = DATAOF_TRACE_ENTRY(entry, true);
907         } else {
908                 trace_seq_printf(s, "%s: (0x%lx)",
909                                  trace_probe_name(&tu->tp),
910                                  entry->vaddr[0]);
911                 data = DATAOF_TRACE_ENTRY(entry, false);
912         }
913
914         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
915                 goto out;
916
917         trace_seq_putc(s, '\n');
918
919  out:
920         return trace_handle_return(s);
921 }
922
923 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
924                                 enum uprobe_filter_ctx ctx,
925                                 struct mm_struct *mm);
926
927 static int
928 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
929                    filter_func_t filter)
930 {
931         bool enabled = trace_probe_is_enabled(&tu->tp);
932         int ret;
933
934         if (file) {
935                 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
936                         return -EINTR;
937
938                 ret = trace_probe_add_file(&tu->tp, file);
939                 if (ret < 0)
940                         return ret;
941         } else {
942                 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
943                         return -EINTR;
944
945                 trace_probe_set_flag(&tu->tp, TP_FLAG_PROFILE);
946         }
947
948         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
949
950         if (enabled)
951                 return 0;
952
953         ret = uprobe_buffer_enable();
954         if (ret)
955                 goto err_flags;
956
957         tu->consumer.filter = filter;
958         tu->inode = d_real_inode(tu->path.dentry);
959         if (tu->ref_ctr_offset) {
960                 ret = uprobe_register_refctr(tu->inode, tu->offset,
961                                 tu->ref_ctr_offset, &tu->consumer);
962         } else {
963                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
964         }
965
966         if (ret)
967                 goto err_buffer;
968
969         return 0;
970
971  err_buffer:
972         uprobe_buffer_disable();
973
974  err_flags:
975         if (file)
976                 trace_probe_remove_file(&tu->tp, file);
977         else
978                 trace_probe_clear_flag(&tu->tp, TP_FLAG_PROFILE);
979
980         return ret;
981 }
982
983 static void
984 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
985 {
986         if (!trace_probe_is_enabled(&tu->tp))
987                 return;
988
989         if (file) {
990                 if (trace_probe_remove_file(&tu->tp, file) < 0)
991                         return;
992
993                 if (trace_probe_is_enabled(&tu->tp))
994                         return;
995         } else
996                 trace_probe_clear_flag(&tu->tp, TP_FLAG_PROFILE);
997
998         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
999
1000         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1001         tu->inode = NULL;
1002
1003         uprobe_buffer_disable();
1004 }
1005
1006 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1007 {
1008         int ret, size;
1009         struct uprobe_trace_entry_head field;
1010         struct trace_uprobe *tu = event_call->data;
1011
1012         if (is_ret_probe(tu)) {
1013                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1014                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1015                 size = SIZEOF_TRACE_ENTRY(true);
1016         } else {
1017                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1018                 size = SIZEOF_TRACE_ENTRY(false);
1019         }
1020
1021         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1022 }
1023
1024 #ifdef CONFIG_PERF_EVENTS
1025 static bool
1026 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1027 {
1028         struct perf_event *event;
1029
1030         if (filter->nr_systemwide)
1031                 return true;
1032
1033         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1034                 if (event->hw.target->mm == mm)
1035                         return true;
1036         }
1037
1038         return false;
1039 }
1040
1041 static inline bool
1042 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1043 {
1044         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1045 }
1046
1047 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1048 {
1049         bool done;
1050
1051         write_lock(&tu->filter.rwlock);
1052         if (event->hw.target) {
1053                 list_del(&event->hw.tp_list);
1054                 done = tu->filter.nr_systemwide ||
1055                         (event->hw.target->flags & PF_EXITING) ||
1056                         uprobe_filter_event(tu, event);
1057         } else {
1058                 tu->filter.nr_systemwide--;
1059                 done = tu->filter.nr_systemwide;
1060         }
1061         write_unlock(&tu->filter.rwlock);
1062
1063         if (!done)
1064                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1065
1066         return 0;
1067 }
1068
1069 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1070 {
1071         bool done;
1072         int err;
1073
1074         write_lock(&tu->filter.rwlock);
1075         if (event->hw.target) {
1076                 /*
1077                  * event->parent != NULL means copy_process(), we can avoid
1078                  * uprobe_apply(). current->mm must be probed and we can rely
1079                  * on dup_mmap() which preserves the already installed bp's.
1080                  *
1081                  * attr.enable_on_exec means that exec/mmap will install the
1082                  * breakpoints we need.
1083                  */
1084                 done = tu->filter.nr_systemwide ||
1085                         event->parent || event->attr.enable_on_exec ||
1086                         uprobe_filter_event(tu, event);
1087                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1088         } else {
1089                 done = tu->filter.nr_systemwide;
1090                 tu->filter.nr_systemwide++;
1091         }
1092         write_unlock(&tu->filter.rwlock);
1093
1094         err = 0;
1095         if (!done) {
1096                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1097                 if (err)
1098                         uprobe_perf_close(tu, event);
1099         }
1100         return err;
1101 }
1102
1103 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1104                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1105 {
1106         struct trace_uprobe *tu;
1107         int ret;
1108
1109         tu = container_of(uc, struct trace_uprobe, consumer);
1110         read_lock(&tu->filter.rwlock);
1111         ret = __uprobe_perf_filter(&tu->filter, mm);
1112         read_unlock(&tu->filter.rwlock);
1113
1114         return ret;
1115 }
1116
1117 static void __uprobe_perf_func(struct trace_uprobe *tu,
1118                                unsigned long func, struct pt_regs *regs,
1119                                struct uprobe_cpu_buffer *ucb, int dsize)
1120 {
1121         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1122         struct uprobe_trace_entry_head *entry;
1123         struct hlist_head *head;
1124         void *data;
1125         int size, esize;
1126         int rctx;
1127
1128         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1129                 return;
1130
1131         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1132
1133         size = esize + tu->tp.size + dsize;
1134         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1135         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1136                 return;
1137
1138         preempt_disable();
1139         head = this_cpu_ptr(call->perf_events);
1140         if (hlist_empty(head))
1141                 goto out;
1142
1143         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1144         if (!entry)
1145                 goto out;
1146
1147         if (is_ret_probe(tu)) {
1148                 entry->vaddr[0] = func;
1149                 entry->vaddr[1] = instruction_pointer(regs);
1150                 data = DATAOF_TRACE_ENTRY(entry, true);
1151         } else {
1152                 entry->vaddr[0] = instruction_pointer(regs);
1153                 data = DATAOF_TRACE_ENTRY(entry, false);
1154         }
1155
1156         memcpy(data, ucb->buf, tu->tp.size + dsize);
1157
1158         if (size - esize > tu->tp.size + dsize) {
1159                 int len = tu->tp.size + dsize;
1160
1161                 memset(data + len, 0, size - esize - len);
1162         }
1163
1164         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1165                               head, NULL);
1166  out:
1167         preempt_enable();
1168 }
1169
1170 /* uprobe profile handler */
1171 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1172                             struct uprobe_cpu_buffer *ucb, int dsize)
1173 {
1174         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1175                 return UPROBE_HANDLER_REMOVE;
1176
1177         if (!is_ret_probe(tu))
1178                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1179         return 0;
1180 }
1181
1182 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1183                                 struct pt_regs *regs,
1184                                 struct uprobe_cpu_buffer *ucb, int dsize)
1185 {
1186         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1187 }
1188
1189 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1190                         const char **filename, u64 *probe_offset,
1191                         bool perf_type_tracepoint)
1192 {
1193         const char *pevent = trace_event_name(event->tp_event);
1194         const char *group = event->tp_event->class->system;
1195         struct trace_uprobe *tu;
1196
1197         if (perf_type_tracepoint)
1198                 tu = find_probe_event(pevent, group);
1199         else
1200                 tu = event->tp_event->data;
1201         if (!tu)
1202                 return -EINVAL;
1203
1204         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1205                                     : BPF_FD_TYPE_UPROBE;
1206         *filename = tu->filename;
1207         *probe_offset = tu->offset;
1208         return 0;
1209 }
1210 #endif  /* CONFIG_PERF_EVENTS */
1211
1212 static int
1213 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1214                       void *data)
1215 {
1216         struct trace_uprobe *tu = event->data;
1217         struct trace_event_file *file = data;
1218
1219         switch (type) {
1220         case TRACE_REG_REGISTER:
1221                 return probe_event_enable(tu, file, NULL);
1222
1223         case TRACE_REG_UNREGISTER:
1224                 probe_event_disable(tu, file);
1225                 return 0;
1226
1227 #ifdef CONFIG_PERF_EVENTS
1228         case TRACE_REG_PERF_REGISTER:
1229                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1230
1231         case TRACE_REG_PERF_UNREGISTER:
1232                 probe_event_disable(tu, NULL);
1233                 return 0;
1234
1235         case TRACE_REG_PERF_OPEN:
1236                 return uprobe_perf_open(tu, data);
1237
1238         case TRACE_REG_PERF_CLOSE:
1239                 return uprobe_perf_close(tu, data);
1240
1241 #endif
1242         default:
1243                 return 0;
1244         }
1245         return 0;
1246 }
1247
1248 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1249 {
1250         struct trace_uprobe *tu;
1251         struct uprobe_dispatch_data udd;
1252         struct uprobe_cpu_buffer *ucb;
1253         int dsize, esize;
1254         int ret = 0;
1255
1256
1257         tu = container_of(con, struct trace_uprobe, consumer);
1258         tu->nhit++;
1259
1260         udd.tu = tu;
1261         udd.bp_addr = instruction_pointer(regs);
1262
1263         current->utask->vaddr = (unsigned long) &udd;
1264
1265         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1266                 return 0;
1267
1268         dsize = __get_data_size(&tu->tp, regs);
1269         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1270
1271         ucb = uprobe_buffer_get();
1272         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1273
1274         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1275                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1276
1277 #ifdef CONFIG_PERF_EVENTS
1278         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1279                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1280 #endif
1281         uprobe_buffer_put(ucb);
1282         return ret;
1283 }
1284
1285 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1286                                 unsigned long func, struct pt_regs *regs)
1287 {
1288         struct trace_uprobe *tu;
1289         struct uprobe_dispatch_data udd;
1290         struct uprobe_cpu_buffer *ucb;
1291         int dsize, esize;
1292
1293         tu = container_of(con, struct trace_uprobe, consumer);
1294
1295         udd.tu = tu;
1296         udd.bp_addr = func;
1297
1298         current->utask->vaddr = (unsigned long) &udd;
1299
1300         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1301                 return 0;
1302
1303         dsize = __get_data_size(&tu->tp, regs);
1304         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1305
1306         ucb = uprobe_buffer_get();
1307         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1308
1309         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1310                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1311
1312 #ifdef CONFIG_PERF_EVENTS
1313         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1314                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1315 #endif
1316         uprobe_buffer_put(ucb);
1317         return 0;
1318 }
1319
1320 static struct trace_event_functions uprobe_funcs = {
1321         .trace          = print_uprobe_event
1322 };
1323
1324 static inline void init_trace_event_call(struct trace_uprobe *tu)
1325 {
1326         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1327
1328         call->event.funcs = &uprobe_funcs;
1329         call->class->define_fields = uprobe_event_define_fields;
1330
1331         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1332         call->class->reg = trace_uprobe_register;
1333         call->data = tu;
1334 }
1335
1336 static int register_uprobe_event(struct trace_uprobe *tu)
1337 {
1338         init_trace_event_call(tu);
1339
1340         return trace_probe_register_event_call(&tu->tp);
1341 }
1342
1343 static int unregister_uprobe_event(struct trace_uprobe *tu)
1344 {
1345         return trace_probe_unregister_event_call(&tu->tp);
1346 }
1347
1348 #ifdef CONFIG_PERF_EVENTS
1349 struct trace_event_call *
1350 create_local_trace_uprobe(char *name, unsigned long offs,
1351                           unsigned long ref_ctr_offset, bool is_return)
1352 {
1353         struct trace_uprobe *tu;
1354         struct path path;
1355         int ret;
1356
1357         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1358         if (ret)
1359                 return ERR_PTR(ret);
1360
1361         if (!d_is_reg(path.dentry)) {
1362                 path_put(&path);
1363                 return ERR_PTR(-EINVAL);
1364         }
1365
1366         /*
1367          * local trace_kprobes are not added to dyn_event, so they are never
1368          * searched in find_trace_kprobe(). Therefore, there is no concern of
1369          * duplicated name "DUMMY_EVENT" here.
1370          */
1371         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1372                                 is_return);
1373
1374         if (IS_ERR(tu)) {
1375                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1376                         (int)PTR_ERR(tu));
1377                 path_put(&path);
1378                 return ERR_CAST(tu);
1379         }
1380
1381         tu->offset = offs;
1382         tu->path = path;
1383         tu->ref_ctr_offset = ref_ctr_offset;
1384         tu->filename = kstrdup(name, GFP_KERNEL);
1385         init_trace_event_call(tu);
1386
1387         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1388                 ret = -ENOMEM;
1389                 goto error;
1390         }
1391
1392         return trace_probe_event_call(&tu->tp);
1393 error:
1394         free_trace_uprobe(tu);
1395         return ERR_PTR(ret);
1396 }
1397
1398 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1399 {
1400         struct trace_uprobe *tu;
1401
1402         tu = container_of(event_call, struct trace_uprobe, tp.call);
1403
1404         free_trace_uprobe(tu);
1405 }
1406 #endif /* CONFIG_PERF_EVENTS */
1407
1408 /* Make a trace interface for controling probe points */
1409 static __init int init_uprobe_trace(void)
1410 {
1411         struct dentry *d_tracer;
1412         int ret;
1413
1414         ret = dyn_event_register(&trace_uprobe_ops);
1415         if (ret)
1416                 return ret;
1417
1418         d_tracer = tracing_init_dentry();
1419         if (IS_ERR(d_tracer))
1420                 return 0;
1421
1422         trace_create_file("uprobe_events", 0644, d_tracer,
1423                                     NULL, &uprobe_events_ops);
1424         /* Profile interface */
1425         trace_create_file("uprobe_profile", 0444, d_tracer,
1426                                     NULL, &uprobe_profile_ops);
1427         return 0;
1428 }
1429
1430 fs_initcall(init_uprobe_trace);