Merge tag 'trace-v4.19-rc8-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rosted...
[muen/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_kprobe: " fmt
9
10 #include <linux/module.h>
11 #include <linux/uaccess.h>
12 #include <linux/uprobes.h>
13 #include <linux/namei.h>
14 #include <linux/string.h>
15 #include <linux/rculist.h>
16
17 #include "trace_probe.h"
18
19 #define UPROBE_EVENT_SYSTEM     "uprobes"
20
21 struct uprobe_trace_entry_head {
22         struct trace_entry      ent;
23         unsigned long           vaddr[];
24 };
25
26 #define SIZEOF_TRACE_ENTRY(is_return)                   \
27         (sizeof(struct uprobe_trace_entry_head) +       \
28          sizeof(unsigned long) * (is_return ? 2 : 1))
29
30 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
31         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
32
33 struct trace_uprobe_filter {
34         rwlock_t                rwlock;
35         int                     nr_systemwide;
36         struct list_head        perf_events;
37 };
38
39 /*
40  * uprobe event core functions
41  */
42 struct trace_uprobe {
43         struct list_head                list;
44         struct trace_uprobe_filter      filter;
45         struct uprobe_consumer          consumer;
46         struct path                     path;
47         struct inode                    *inode;
48         char                            *filename;
49         unsigned long                   offset;
50         unsigned long                   nhit;
51         struct trace_probe              tp;
52 };
53
54 #define SIZEOF_TRACE_UPROBE(n)                          \
55         (offsetof(struct trace_uprobe, tp.args) +       \
56         (sizeof(struct probe_arg) * (n)))
57
58 static int register_uprobe_event(struct trace_uprobe *tu);
59 static int unregister_uprobe_event(struct trace_uprobe *tu);
60
61 static DEFINE_MUTEX(uprobe_lock);
62 static LIST_HEAD(uprobe_list);
63
64 struct uprobe_dispatch_data {
65         struct trace_uprobe     *tu;
66         unsigned long           bp_addr;
67 };
68
69 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
70 static int uretprobe_dispatcher(struct uprobe_consumer *con,
71                                 unsigned long func, struct pt_regs *regs);
72
73 #ifdef CONFIG_STACK_GROWSUP
74 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
75 {
76         return addr - (n * sizeof(long));
77 }
78 #else
79 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
80 {
81         return addr + (n * sizeof(long));
82 }
83 #endif
84
85 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
86 {
87         unsigned long ret;
88         unsigned long addr = user_stack_pointer(regs);
89
90         addr = adjust_stack_addr(addr, n);
91
92         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
93                 return 0;
94
95         return ret;
96 }
97
98 /*
99  * Uprobes-specific fetch functions
100  */
101 #define DEFINE_FETCH_stack(type)                                        \
102 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
103                                          void *offset, void *dest)      \
104 {                                                                       \
105         *(type *)dest = (type)get_user_stack_nth(regs,                  \
106                                               ((unsigned long)offset)); \
107 }
108 DEFINE_BASIC_FETCH_FUNCS(stack)
109 /* No string on the stack entry */
110 #define fetch_stack_string      NULL
111 #define fetch_stack_string_size NULL
112
113 #define DEFINE_FETCH_memory(type)                                       \
114 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
115                                           void *addr, void *dest)       \
116 {                                                                       \
117         type retval;                                                    \
118         void __user *vaddr = (void __force __user *) addr;              \
119                                                                         \
120         if (copy_from_user(&retval, vaddr, sizeof(type)))               \
121                 *(type *)dest = 0;                                      \
122         else                                                            \
123                 *(type *) dest = retval;                                \
124 }
125 DEFINE_BASIC_FETCH_FUNCS(memory)
126 /*
127  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
128  * length and relative data location.
129  */
130 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
131                                             void *addr, void *dest)
132 {
133         long ret;
134         u32 rloc = *(u32 *)dest;
135         int maxlen  = get_rloc_len(rloc);
136         u8 *dst = get_rloc_data(dest);
137         void __user *src = (void __force __user *) addr;
138
139         if (!maxlen)
140                 return;
141
142         ret = strncpy_from_user(dst, src, maxlen);
143         if (ret == maxlen)
144                 dst[--ret] = '\0';
145
146         if (ret < 0) {  /* Failed to fetch string */
147                 ((u8 *)get_rloc_data(dest))[0] = '\0';
148                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
149         } else {
150                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
151         }
152 }
153
154 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
155                                                  void *addr, void *dest)
156 {
157         int len;
158         void __user *vaddr = (void __force __user *) addr;
159
160         len = strnlen_user(vaddr, MAX_STRING_SIZE);
161
162         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
163                 *(u32 *)dest = 0;
164         else
165                 *(u32 *)dest = len;
166 }
167
168 static unsigned long translate_user_vaddr(void *file_offset)
169 {
170         unsigned long base_addr;
171         struct uprobe_dispatch_data *udd;
172
173         udd = (void *) current->utask->vaddr;
174
175         base_addr = udd->bp_addr - udd->tu->offset;
176         return base_addr + (unsigned long)file_offset;
177 }
178
179 #define DEFINE_FETCH_file_offset(type)                                  \
180 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,    \
181                                                void *offset, void *dest)\
182 {                                                                       \
183         void *vaddr = (void *)translate_user_vaddr(offset);             \
184                                                                         \
185         FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);               \
186 }
187 DEFINE_BASIC_FETCH_FUNCS(file_offset)
188 DEFINE_FETCH_file_offset(string)
189 DEFINE_FETCH_file_offset(string_size)
190
191 /* Fetch type information table */
192 static const struct fetch_type uprobes_fetch_type_table[] = {
193         /* Special types */
194         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
195                                         sizeof(u32), 1, "__data_loc char[]"),
196         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
197                                         string_size, sizeof(u32), 0, "u32"),
198         /* Basic types */
199         ASSIGN_FETCH_TYPE(u8,  u8,  0),
200         ASSIGN_FETCH_TYPE(u16, u16, 0),
201         ASSIGN_FETCH_TYPE(u32, u32, 0),
202         ASSIGN_FETCH_TYPE(u64, u64, 0),
203         ASSIGN_FETCH_TYPE(s8,  u8,  1),
204         ASSIGN_FETCH_TYPE(s16, u16, 1),
205         ASSIGN_FETCH_TYPE(s32, u32, 1),
206         ASSIGN_FETCH_TYPE(s64, u64, 1),
207         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
208         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
209         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
210         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
211
212         ASSIGN_FETCH_TYPE_END
213 };
214
215 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
216 {
217         rwlock_init(&filter->rwlock);
218         filter->nr_systemwide = 0;
219         INIT_LIST_HEAD(&filter->perf_events);
220 }
221
222 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
223 {
224         return !filter->nr_systemwide && list_empty(&filter->perf_events);
225 }
226
227 static inline bool is_ret_probe(struct trace_uprobe *tu)
228 {
229         return tu->consumer.ret_handler != NULL;
230 }
231
232 /*
233  * Allocate new trace_uprobe and initialize it (including uprobes).
234  */
235 static struct trace_uprobe *
236 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
237 {
238         struct trace_uprobe *tu;
239
240         if (!event || !is_good_name(event))
241                 return ERR_PTR(-EINVAL);
242
243         if (!group || !is_good_name(group))
244                 return ERR_PTR(-EINVAL);
245
246         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
247         if (!tu)
248                 return ERR_PTR(-ENOMEM);
249
250         tu->tp.call.class = &tu->tp.class;
251         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
252         if (!tu->tp.call.name)
253                 goto error;
254
255         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
256         if (!tu->tp.class.system)
257                 goto error;
258
259         INIT_LIST_HEAD(&tu->list);
260         INIT_LIST_HEAD(&tu->tp.files);
261         tu->consumer.handler = uprobe_dispatcher;
262         if (is_ret)
263                 tu->consumer.ret_handler = uretprobe_dispatcher;
264         init_trace_uprobe_filter(&tu->filter);
265         return tu;
266
267 error:
268         kfree(tu->tp.call.name);
269         kfree(tu);
270
271         return ERR_PTR(-ENOMEM);
272 }
273
274 static void free_trace_uprobe(struct trace_uprobe *tu)
275 {
276         int i;
277
278         for (i = 0; i < tu->tp.nr_args; i++)
279                 traceprobe_free_probe_arg(&tu->tp.args[i]);
280
281         path_put(&tu->path);
282         kfree(tu->tp.call.class->system);
283         kfree(tu->tp.call.name);
284         kfree(tu->filename);
285         kfree(tu);
286 }
287
288 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
289 {
290         struct trace_uprobe *tu;
291
292         list_for_each_entry(tu, &uprobe_list, list)
293                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
294                     strcmp(tu->tp.call.class->system, group) == 0)
295                         return tu;
296
297         return NULL;
298 }
299
300 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
301 static int unregister_trace_uprobe(struct trace_uprobe *tu)
302 {
303         int ret;
304
305         ret = unregister_uprobe_event(tu);
306         if (ret)
307                 return ret;
308
309         list_del(&tu->list);
310         free_trace_uprobe(tu);
311         return 0;
312 }
313
314 /* Register a trace_uprobe and probe_event */
315 static int register_trace_uprobe(struct trace_uprobe *tu)
316 {
317         struct trace_uprobe *old_tu;
318         int ret;
319
320         mutex_lock(&uprobe_lock);
321
322         /* register as an event */
323         old_tu = find_probe_event(trace_event_name(&tu->tp.call),
324                         tu->tp.call.class->system);
325         if (old_tu) {
326                 /* delete old event */
327                 ret = unregister_trace_uprobe(old_tu);
328                 if (ret)
329                         goto end;
330         }
331
332         ret = register_uprobe_event(tu);
333         if (ret) {
334                 pr_warn("Failed to register probe event(%d)\n", ret);
335                 goto end;
336         }
337
338         list_add_tail(&tu->list, &uprobe_list);
339
340 end:
341         mutex_unlock(&uprobe_lock);
342
343         return ret;
344 }
345
346 /*
347  * Argument syntax:
348  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
349  *
350  *  - Remove uprobe: -:[GRP/]EVENT
351  */
352 static int create_trace_uprobe(int argc, char **argv)
353 {
354         struct trace_uprobe *tu;
355         char *arg, *event, *group, *filename;
356         char buf[MAX_EVENT_NAME_LEN];
357         struct path path;
358         unsigned long offset;
359         bool is_delete, is_return;
360         int i, ret;
361
362         ret = 0;
363         is_delete = false;
364         is_return = false;
365         event = NULL;
366         group = NULL;
367
368         /* argc must be >= 1 */
369         if (argv[0][0] == '-')
370                 is_delete = true;
371         else if (argv[0][0] == 'r')
372                 is_return = true;
373         else if (argv[0][0] != 'p') {
374                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
375                 return -EINVAL;
376         }
377
378         if (argv[0][1] == ':') {
379                 event = &argv[0][2];
380                 arg = strchr(event, '/');
381
382                 if (arg) {
383                         group = event;
384                         event = arg + 1;
385                         event[-1] = '\0';
386
387                         if (strlen(group) == 0) {
388                                 pr_info("Group name is not specified\n");
389                                 return -EINVAL;
390                         }
391                 }
392                 if (strlen(event) == 0) {
393                         pr_info("Event name is not specified\n");
394                         return -EINVAL;
395                 }
396         }
397         if (!group)
398                 group = UPROBE_EVENT_SYSTEM;
399
400         if (is_delete) {
401                 int ret;
402
403                 if (!event) {
404                         pr_info("Delete command needs an event name.\n");
405                         return -EINVAL;
406                 }
407                 mutex_lock(&uprobe_lock);
408                 tu = find_probe_event(event, group);
409
410                 if (!tu) {
411                         mutex_unlock(&uprobe_lock);
412                         pr_info("Event %s/%s doesn't exist.\n", group, event);
413                         return -ENOENT;
414                 }
415                 /* delete an event */
416                 ret = unregister_trace_uprobe(tu);
417                 mutex_unlock(&uprobe_lock);
418                 return ret;
419         }
420
421         if (argc < 2) {
422                 pr_info("Probe point is not specified.\n");
423                 return -EINVAL;
424         }
425         /* Find the last occurrence, in case the path contains ':' too. */
426         arg = strrchr(argv[1], ':');
427         if (!arg)
428                 return -EINVAL;
429
430         *arg++ = '\0';
431         filename = argv[1];
432         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
433         if (ret)
434                 return ret;
435
436         if (!d_is_reg(path.dentry)) {
437                 ret = -EINVAL;
438                 goto fail_address_parse;
439         }
440
441         ret = kstrtoul(arg, 0, &offset);
442         if (ret)
443                 goto fail_address_parse;
444
445         argc -= 2;
446         argv += 2;
447
448         /* setup a probe */
449         if (!event) {
450                 char *tail;
451                 char *ptr;
452
453                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
454                 if (!tail) {
455                         ret = -ENOMEM;
456                         goto fail_address_parse;
457                 }
458
459                 ptr = strpbrk(tail, ".-_");
460                 if (ptr)
461                         *ptr = '\0';
462
463                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
464                 event = buf;
465                 kfree(tail);
466         }
467
468         tu = alloc_trace_uprobe(group, event, argc, is_return);
469         if (IS_ERR(tu)) {
470                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
471                 ret = PTR_ERR(tu);
472                 goto fail_address_parse;
473         }
474         tu->offset = offset;
475         tu->path = path;
476         tu->filename = kstrdup(filename, GFP_KERNEL);
477
478         if (!tu->filename) {
479                 pr_info("Failed to allocate filename.\n");
480                 ret = -ENOMEM;
481                 goto error;
482         }
483
484         /* parse arguments */
485         ret = 0;
486         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
487                 struct probe_arg *parg = &tu->tp.args[i];
488
489                 /* Increment count for freeing args in error case */
490                 tu->tp.nr_args++;
491
492                 /* Parse argument name */
493                 arg = strchr(argv[i], '=');
494                 if (arg) {
495                         *arg++ = '\0';
496                         parg->name = kstrdup(argv[i], GFP_KERNEL);
497                 } else {
498                         arg = argv[i];
499                         /* If argument name is omitted, set "argN" */
500                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
501                         parg->name = kstrdup(buf, GFP_KERNEL);
502                 }
503
504                 if (!parg->name) {
505                         pr_info("Failed to allocate argument[%d] name.\n", i);
506                         ret = -ENOMEM;
507                         goto error;
508                 }
509
510                 if (!is_good_name(parg->name)) {
511                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
512                         ret = -EINVAL;
513                         goto error;
514                 }
515
516                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
517                         pr_info("Argument[%d] name '%s' conflicts with "
518                                 "another field.\n", i, argv[i]);
519                         ret = -EINVAL;
520                         goto error;
521                 }
522
523                 /* Parse fetch argument */
524                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
525                                                  is_return, false,
526                                                  uprobes_fetch_type_table);
527                 if (ret) {
528                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
529                         goto error;
530                 }
531         }
532
533         ret = register_trace_uprobe(tu);
534         if (ret)
535                 goto error;
536         return 0;
537
538 error:
539         free_trace_uprobe(tu);
540         return ret;
541
542 fail_address_parse:
543         path_put(&path);
544
545         pr_info("Failed to parse address or file.\n");
546
547         return ret;
548 }
549
550 static int cleanup_all_probes(void)
551 {
552         struct trace_uprobe *tu;
553         int ret = 0;
554
555         mutex_lock(&uprobe_lock);
556         while (!list_empty(&uprobe_list)) {
557                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
558                 ret = unregister_trace_uprobe(tu);
559                 if (ret)
560                         break;
561         }
562         mutex_unlock(&uprobe_lock);
563         return ret;
564 }
565
566 /* Probes listing interfaces */
567 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
568 {
569         mutex_lock(&uprobe_lock);
570         return seq_list_start(&uprobe_list, *pos);
571 }
572
573 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
574 {
575         return seq_list_next(v, &uprobe_list, pos);
576 }
577
578 static void probes_seq_stop(struct seq_file *m, void *v)
579 {
580         mutex_unlock(&uprobe_lock);
581 }
582
583 static int probes_seq_show(struct seq_file *m, void *v)
584 {
585         struct trace_uprobe *tu = v;
586         char c = is_ret_probe(tu) ? 'r' : 'p';
587         int i;
588
589         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, tu->tp.call.class->system,
590                         trace_event_name(&tu->tp.call), tu->filename,
591                         (int)(sizeof(void *) * 2), tu->offset);
592
593         for (i = 0; i < tu->tp.nr_args; i++)
594                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
595
596         seq_putc(m, '\n');
597         return 0;
598 }
599
600 static const struct seq_operations probes_seq_op = {
601         .start  = probes_seq_start,
602         .next   = probes_seq_next,
603         .stop   = probes_seq_stop,
604         .show   = probes_seq_show
605 };
606
607 static int probes_open(struct inode *inode, struct file *file)
608 {
609         int ret;
610
611         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
612                 ret = cleanup_all_probes();
613                 if (ret)
614                         return ret;
615         }
616
617         return seq_open(file, &probes_seq_op);
618 }
619
620 static ssize_t probes_write(struct file *file, const char __user *buffer,
621                             size_t count, loff_t *ppos)
622 {
623         return trace_parse_run_command(file, buffer, count, ppos, create_trace_uprobe);
624 }
625
626 static const struct file_operations uprobe_events_ops = {
627         .owner          = THIS_MODULE,
628         .open           = probes_open,
629         .read           = seq_read,
630         .llseek         = seq_lseek,
631         .release        = seq_release,
632         .write          = probes_write,
633 };
634
635 /* Probes profiling interfaces */
636 static int probes_profile_seq_show(struct seq_file *m, void *v)
637 {
638         struct trace_uprobe *tu = v;
639
640         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
641                         trace_event_name(&tu->tp.call), tu->nhit);
642         return 0;
643 }
644
645 static const struct seq_operations profile_seq_op = {
646         .start  = probes_seq_start,
647         .next   = probes_seq_next,
648         .stop   = probes_seq_stop,
649         .show   = probes_profile_seq_show
650 };
651
652 static int profile_open(struct inode *inode, struct file *file)
653 {
654         return seq_open(file, &profile_seq_op);
655 }
656
657 static const struct file_operations uprobe_profile_ops = {
658         .owner          = THIS_MODULE,
659         .open           = profile_open,
660         .read           = seq_read,
661         .llseek         = seq_lseek,
662         .release        = seq_release,
663 };
664
665 struct uprobe_cpu_buffer {
666         struct mutex mutex;
667         void *buf;
668 };
669 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
670 static int uprobe_buffer_refcnt;
671
672 static int uprobe_buffer_init(void)
673 {
674         int cpu, err_cpu;
675
676         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
677         if (uprobe_cpu_buffer == NULL)
678                 return -ENOMEM;
679
680         for_each_possible_cpu(cpu) {
681                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
682                                                   GFP_KERNEL, 0);
683                 if (p == NULL) {
684                         err_cpu = cpu;
685                         goto err;
686                 }
687                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
688                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
689         }
690
691         return 0;
692
693 err:
694         for_each_possible_cpu(cpu) {
695                 if (cpu == err_cpu)
696                         break;
697                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
698         }
699
700         free_percpu(uprobe_cpu_buffer);
701         return -ENOMEM;
702 }
703
704 static int uprobe_buffer_enable(void)
705 {
706         int ret = 0;
707
708         BUG_ON(!mutex_is_locked(&event_mutex));
709
710         if (uprobe_buffer_refcnt++ == 0) {
711                 ret = uprobe_buffer_init();
712                 if (ret < 0)
713                         uprobe_buffer_refcnt--;
714         }
715
716         return ret;
717 }
718
719 static void uprobe_buffer_disable(void)
720 {
721         int cpu;
722
723         BUG_ON(!mutex_is_locked(&event_mutex));
724
725         if (--uprobe_buffer_refcnt == 0) {
726                 for_each_possible_cpu(cpu)
727                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
728                                                              cpu)->buf);
729
730                 free_percpu(uprobe_cpu_buffer);
731                 uprobe_cpu_buffer = NULL;
732         }
733 }
734
735 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
736 {
737         struct uprobe_cpu_buffer *ucb;
738         int cpu;
739
740         cpu = raw_smp_processor_id();
741         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
742
743         /*
744          * Use per-cpu buffers for fastest access, but we might migrate
745          * so the mutex makes sure we have sole access to it.
746          */
747         mutex_lock(&ucb->mutex);
748
749         return ucb;
750 }
751
752 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
753 {
754         mutex_unlock(&ucb->mutex);
755 }
756
757 static void __uprobe_trace_func(struct trace_uprobe *tu,
758                                 unsigned long func, struct pt_regs *regs,
759                                 struct uprobe_cpu_buffer *ucb, int dsize,
760                                 struct trace_event_file *trace_file)
761 {
762         struct uprobe_trace_entry_head *entry;
763         struct ring_buffer_event *event;
764         struct ring_buffer *buffer;
765         void *data;
766         int size, esize;
767         struct trace_event_call *call = &tu->tp.call;
768
769         WARN_ON(call != trace_file->event_call);
770
771         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
772                 return;
773
774         if (trace_trigger_soft_disabled(trace_file))
775                 return;
776
777         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
778         size = esize + tu->tp.size + dsize;
779         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
780                                                 call->event.type, size, 0, 0);
781         if (!event)
782                 return;
783
784         entry = ring_buffer_event_data(event);
785         if (is_ret_probe(tu)) {
786                 entry->vaddr[0] = func;
787                 entry->vaddr[1] = instruction_pointer(regs);
788                 data = DATAOF_TRACE_ENTRY(entry, true);
789         } else {
790                 entry->vaddr[0] = instruction_pointer(regs);
791                 data = DATAOF_TRACE_ENTRY(entry, false);
792         }
793
794         memcpy(data, ucb->buf, tu->tp.size + dsize);
795
796         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
797 }
798
799 /* uprobe handler */
800 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
801                              struct uprobe_cpu_buffer *ucb, int dsize)
802 {
803         struct event_file_link *link;
804
805         if (is_ret_probe(tu))
806                 return 0;
807
808         rcu_read_lock();
809         list_for_each_entry_rcu(link, &tu->tp.files, list)
810                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
811         rcu_read_unlock();
812
813         return 0;
814 }
815
816 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
817                                  struct pt_regs *regs,
818                                  struct uprobe_cpu_buffer *ucb, int dsize)
819 {
820         struct event_file_link *link;
821
822         rcu_read_lock();
823         list_for_each_entry_rcu(link, &tu->tp.files, list)
824                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
825         rcu_read_unlock();
826 }
827
828 /* Event entry printers */
829 static enum print_line_t
830 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
831 {
832         struct uprobe_trace_entry_head *entry;
833         struct trace_seq *s = &iter->seq;
834         struct trace_uprobe *tu;
835         u8 *data;
836         int i;
837
838         entry = (struct uprobe_trace_entry_head *)iter->ent;
839         tu = container_of(event, struct trace_uprobe, tp.call.event);
840
841         if (is_ret_probe(tu)) {
842                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
843                                  trace_event_name(&tu->tp.call),
844                                  entry->vaddr[1], entry->vaddr[0]);
845                 data = DATAOF_TRACE_ENTRY(entry, true);
846         } else {
847                 trace_seq_printf(s, "%s: (0x%lx)",
848                                  trace_event_name(&tu->tp.call),
849                                  entry->vaddr[0]);
850                 data = DATAOF_TRACE_ENTRY(entry, false);
851         }
852
853         for (i = 0; i < tu->tp.nr_args; i++) {
854                 struct probe_arg *parg = &tu->tp.args[i];
855
856                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
857                         goto out;
858         }
859
860         trace_seq_putc(s, '\n');
861
862  out:
863         return trace_handle_return(s);
864 }
865
866 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
867                                 enum uprobe_filter_ctx ctx,
868                                 struct mm_struct *mm);
869
870 static int
871 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
872                    filter_func_t filter)
873 {
874         bool enabled = trace_probe_is_enabled(&tu->tp);
875         struct event_file_link *link = NULL;
876         int ret;
877
878         if (file) {
879                 if (tu->tp.flags & TP_FLAG_PROFILE)
880                         return -EINTR;
881
882                 link = kmalloc(sizeof(*link), GFP_KERNEL);
883                 if (!link)
884                         return -ENOMEM;
885
886                 link->file = file;
887                 list_add_tail_rcu(&link->list, &tu->tp.files);
888
889                 tu->tp.flags |= TP_FLAG_TRACE;
890         } else {
891                 if (tu->tp.flags & TP_FLAG_TRACE)
892                         return -EINTR;
893
894                 tu->tp.flags |= TP_FLAG_PROFILE;
895         }
896
897         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
898
899         if (enabled)
900                 return 0;
901
902         ret = uprobe_buffer_enable();
903         if (ret)
904                 goto err_flags;
905
906         tu->consumer.filter = filter;
907         tu->inode = d_real_inode(tu->path.dentry);
908         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
909         if (ret)
910                 goto err_buffer;
911
912         return 0;
913
914  err_buffer:
915         uprobe_buffer_disable();
916
917  err_flags:
918         if (file) {
919                 list_del(&link->list);
920                 kfree(link);
921                 tu->tp.flags &= ~TP_FLAG_TRACE;
922         } else {
923                 tu->tp.flags &= ~TP_FLAG_PROFILE;
924         }
925         return ret;
926 }
927
928 static void
929 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
930 {
931         if (!trace_probe_is_enabled(&tu->tp))
932                 return;
933
934         if (file) {
935                 struct event_file_link *link;
936
937                 link = find_event_file_link(&tu->tp, file);
938                 if (!link)
939                         return;
940
941                 list_del_rcu(&link->list);
942                 /* synchronize with u{,ret}probe_trace_func */
943                 synchronize_rcu();
944                 kfree(link);
945
946                 if (!list_empty(&tu->tp.files))
947                         return;
948         }
949
950         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
951
952         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
953         tu->inode = NULL;
954         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
955
956         uprobe_buffer_disable();
957 }
958
959 static int uprobe_event_define_fields(struct trace_event_call *event_call)
960 {
961         int ret, i, size;
962         struct uprobe_trace_entry_head field;
963         struct trace_uprobe *tu = event_call->data;
964
965         if (is_ret_probe(tu)) {
966                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
967                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
968                 size = SIZEOF_TRACE_ENTRY(true);
969         } else {
970                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
971                 size = SIZEOF_TRACE_ENTRY(false);
972         }
973         /* Set argument names as fields */
974         for (i = 0; i < tu->tp.nr_args; i++) {
975                 struct probe_arg *parg = &tu->tp.args[i];
976
977                 ret = trace_define_field(event_call, parg->type->fmttype,
978                                          parg->name, size + parg->offset,
979                                          parg->type->size, parg->type->is_signed,
980                                          FILTER_OTHER);
981
982                 if (ret)
983                         return ret;
984         }
985         return 0;
986 }
987
988 #ifdef CONFIG_PERF_EVENTS
989 static bool
990 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
991 {
992         struct perf_event *event;
993
994         if (filter->nr_systemwide)
995                 return true;
996
997         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
998                 if (event->hw.target->mm == mm)
999                         return true;
1000         }
1001
1002         return false;
1003 }
1004
1005 static inline bool
1006 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1007 {
1008         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1009 }
1010
1011 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1012 {
1013         bool done;
1014
1015         write_lock(&tu->filter.rwlock);
1016         if (event->hw.target) {
1017                 list_del(&event->hw.tp_list);
1018                 done = tu->filter.nr_systemwide ||
1019                         (event->hw.target->flags & PF_EXITING) ||
1020                         uprobe_filter_event(tu, event);
1021         } else {
1022                 tu->filter.nr_systemwide--;
1023                 done = tu->filter.nr_systemwide;
1024         }
1025         write_unlock(&tu->filter.rwlock);
1026
1027         if (!done)
1028                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1029
1030         return 0;
1031 }
1032
1033 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1034 {
1035         bool done;
1036         int err;
1037
1038         write_lock(&tu->filter.rwlock);
1039         if (event->hw.target) {
1040                 /*
1041                  * event->parent != NULL means copy_process(), we can avoid
1042                  * uprobe_apply(). current->mm must be probed and we can rely
1043                  * on dup_mmap() which preserves the already installed bp's.
1044                  *
1045                  * attr.enable_on_exec means that exec/mmap will install the
1046                  * breakpoints we need.
1047                  */
1048                 done = tu->filter.nr_systemwide ||
1049                         event->parent || event->attr.enable_on_exec ||
1050                         uprobe_filter_event(tu, event);
1051                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1052         } else {
1053                 done = tu->filter.nr_systemwide;
1054                 tu->filter.nr_systemwide++;
1055         }
1056         write_unlock(&tu->filter.rwlock);
1057
1058         err = 0;
1059         if (!done) {
1060                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1061                 if (err)
1062                         uprobe_perf_close(tu, event);
1063         }
1064         return err;
1065 }
1066
1067 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1068                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1069 {
1070         struct trace_uprobe *tu;
1071         int ret;
1072
1073         tu = container_of(uc, struct trace_uprobe, consumer);
1074         read_lock(&tu->filter.rwlock);
1075         ret = __uprobe_perf_filter(&tu->filter, mm);
1076         read_unlock(&tu->filter.rwlock);
1077
1078         return ret;
1079 }
1080
1081 static void __uprobe_perf_func(struct trace_uprobe *tu,
1082                                unsigned long func, struct pt_regs *regs,
1083                                struct uprobe_cpu_buffer *ucb, int dsize)
1084 {
1085         struct trace_event_call *call = &tu->tp.call;
1086         struct uprobe_trace_entry_head *entry;
1087         struct hlist_head *head;
1088         void *data;
1089         int size, esize;
1090         int rctx;
1091
1092         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1093                 return;
1094
1095         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1096
1097         size = esize + tu->tp.size + dsize;
1098         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1099         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1100                 return;
1101
1102         preempt_disable();
1103         head = this_cpu_ptr(call->perf_events);
1104         if (hlist_empty(head))
1105                 goto out;
1106
1107         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1108         if (!entry)
1109                 goto out;
1110
1111         if (is_ret_probe(tu)) {
1112                 entry->vaddr[0] = func;
1113                 entry->vaddr[1] = instruction_pointer(regs);
1114                 data = DATAOF_TRACE_ENTRY(entry, true);
1115         } else {
1116                 entry->vaddr[0] = instruction_pointer(regs);
1117                 data = DATAOF_TRACE_ENTRY(entry, false);
1118         }
1119
1120         memcpy(data, ucb->buf, tu->tp.size + dsize);
1121
1122         if (size - esize > tu->tp.size + dsize) {
1123                 int len = tu->tp.size + dsize;
1124
1125                 memset(data + len, 0, size - esize - len);
1126         }
1127
1128         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1129                               head, NULL);
1130  out:
1131         preempt_enable();
1132 }
1133
1134 /* uprobe profile handler */
1135 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1136                             struct uprobe_cpu_buffer *ucb, int dsize)
1137 {
1138         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1139                 return UPROBE_HANDLER_REMOVE;
1140
1141         if (!is_ret_probe(tu))
1142                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1143         return 0;
1144 }
1145
1146 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1147                                 struct pt_regs *regs,
1148                                 struct uprobe_cpu_buffer *ucb, int dsize)
1149 {
1150         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1151 }
1152
1153 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1154                         const char **filename, u64 *probe_offset,
1155                         bool perf_type_tracepoint)
1156 {
1157         const char *pevent = trace_event_name(event->tp_event);
1158         const char *group = event->tp_event->class->system;
1159         struct trace_uprobe *tu;
1160
1161         if (perf_type_tracepoint)
1162                 tu = find_probe_event(pevent, group);
1163         else
1164                 tu = event->tp_event->data;
1165         if (!tu)
1166                 return -EINVAL;
1167
1168         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1169                                     : BPF_FD_TYPE_UPROBE;
1170         *filename = tu->filename;
1171         *probe_offset = tu->offset;
1172         return 0;
1173 }
1174 #endif  /* CONFIG_PERF_EVENTS */
1175
1176 static int
1177 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1178                       void *data)
1179 {
1180         struct trace_uprobe *tu = event->data;
1181         struct trace_event_file *file = data;
1182
1183         switch (type) {
1184         case TRACE_REG_REGISTER:
1185                 return probe_event_enable(tu, file, NULL);
1186
1187         case TRACE_REG_UNREGISTER:
1188                 probe_event_disable(tu, file);
1189                 return 0;
1190
1191 #ifdef CONFIG_PERF_EVENTS
1192         case TRACE_REG_PERF_REGISTER:
1193                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1194
1195         case TRACE_REG_PERF_UNREGISTER:
1196                 probe_event_disable(tu, NULL);
1197                 return 0;
1198
1199         case TRACE_REG_PERF_OPEN:
1200                 return uprobe_perf_open(tu, data);
1201
1202         case TRACE_REG_PERF_CLOSE:
1203                 return uprobe_perf_close(tu, data);
1204
1205 #endif
1206         default:
1207                 return 0;
1208         }
1209         return 0;
1210 }
1211
1212 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1213 {
1214         struct trace_uprobe *tu;
1215         struct uprobe_dispatch_data udd;
1216         struct uprobe_cpu_buffer *ucb;
1217         int dsize, esize;
1218         int ret = 0;
1219
1220
1221         tu = container_of(con, struct trace_uprobe, consumer);
1222         tu->nhit++;
1223
1224         udd.tu = tu;
1225         udd.bp_addr = instruction_pointer(regs);
1226
1227         current->utask->vaddr = (unsigned long) &udd;
1228
1229         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1230                 return 0;
1231
1232         dsize = __get_data_size(&tu->tp, regs);
1233         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1234
1235         ucb = uprobe_buffer_get();
1236         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1237
1238         if (tu->tp.flags & TP_FLAG_TRACE)
1239                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1240
1241 #ifdef CONFIG_PERF_EVENTS
1242         if (tu->tp.flags & TP_FLAG_PROFILE)
1243                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1244 #endif
1245         uprobe_buffer_put(ucb);
1246         return ret;
1247 }
1248
1249 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1250                                 unsigned long func, struct pt_regs *regs)
1251 {
1252         struct trace_uprobe *tu;
1253         struct uprobe_dispatch_data udd;
1254         struct uprobe_cpu_buffer *ucb;
1255         int dsize, esize;
1256
1257         tu = container_of(con, struct trace_uprobe, consumer);
1258
1259         udd.tu = tu;
1260         udd.bp_addr = func;
1261
1262         current->utask->vaddr = (unsigned long) &udd;
1263
1264         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1265                 return 0;
1266
1267         dsize = __get_data_size(&tu->tp, regs);
1268         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1269
1270         ucb = uprobe_buffer_get();
1271         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1272
1273         if (tu->tp.flags & TP_FLAG_TRACE)
1274                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1275
1276 #ifdef CONFIG_PERF_EVENTS
1277         if (tu->tp.flags & TP_FLAG_PROFILE)
1278                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1279 #endif
1280         uprobe_buffer_put(ucb);
1281         return 0;
1282 }
1283
1284 static struct trace_event_functions uprobe_funcs = {
1285         .trace          = print_uprobe_event
1286 };
1287
1288 static inline void init_trace_event_call(struct trace_uprobe *tu,
1289                                          struct trace_event_call *call)
1290 {
1291         INIT_LIST_HEAD(&call->class->fields);
1292         call->event.funcs = &uprobe_funcs;
1293         call->class->define_fields = uprobe_event_define_fields;
1294
1295         call->flags = TRACE_EVENT_FL_UPROBE;
1296         call->class->reg = trace_uprobe_register;
1297         call->data = tu;
1298 }
1299
1300 static int register_uprobe_event(struct trace_uprobe *tu)
1301 {
1302         struct trace_event_call *call = &tu->tp.call;
1303         int ret = 0;
1304
1305         init_trace_event_call(tu, call);
1306
1307         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1308                 return -ENOMEM;
1309
1310         ret = register_trace_event(&call->event);
1311         if (!ret) {
1312                 kfree(call->print_fmt);
1313                 return -ENODEV;
1314         }
1315
1316         ret = trace_add_event_call(call);
1317
1318         if (ret) {
1319                 pr_info("Failed to register uprobe event: %s\n",
1320                         trace_event_name(call));
1321                 kfree(call->print_fmt);
1322                 unregister_trace_event(&call->event);
1323         }
1324
1325         return ret;
1326 }
1327
1328 static int unregister_uprobe_event(struct trace_uprobe *tu)
1329 {
1330         int ret;
1331
1332         /* tu->event is unregistered in trace_remove_event_call() */
1333         ret = trace_remove_event_call(&tu->tp.call);
1334         if (ret)
1335                 return ret;
1336         kfree(tu->tp.call.print_fmt);
1337         tu->tp.call.print_fmt = NULL;
1338         return 0;
1339 }
1340
1341 #ifdef CONFIG_PERF_EVENTS
1342 struct trace_event_call *
1343 create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
1344 {
1345         struct trace_uprobe *tu;
1346         struct path path;
1347         int ret;
1348
1349         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1350         if (ret)
1351                 return ERR_PTR(ret);
1352
1353         if (!d_is_reg(path.dentry)) {
1354                 path_put(&path);
1355                 return ERR_PTR(-EINVAL);
1356         }
1357
1358         /*
1359          * local trace_kprobes are not added to probe_list, so they are never
1360          * searched in find_trace_kprobe(). Therefore, there is no concern of
1361          * duplicated name "DUMMY_EVENT" here.
1362          */
1363         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1364                                 is_return);
1365
1366         if (IS_ERR(tu)) {
1367                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1368                         (int)PTR_ERR(tu));
1369                 path_put(&path);
1370                 return ERR_CAST(tu);
1371         }
1372
1373         tu->offset = offs;
1374         tu->path = path;
1375         tu->filename = kstrdup(name, GFP_KERNEL);
1376         init_trace_event_call(tu, &tu->tp.call);
1377
1378         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1379                 ret = -ENOMEM;
1380                 goto error;
1381         }
1382
1383         return &tu->tp.call;
1384 error:
1385         free_trace_uprobe(tu);
1386         return ERR_PTR(ret);
1387 }
1388
1389 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1390 {
1391         struct trace_uprobe *tu;
1392
1393         tu = container_of(event_call, struct trace_uprobe, tp.call);
1394
1395         kfree(tu->tp.call.print_fmt);
1396         tu->tp.call.print_fmt = NULL;
1397
1398         free_trace_uprobe(tu);
1399 }
1400 #endif /* CONFIG_PERF_EVENTS */
1401
1402 /* Make a trace interface for controling probe points */
1403 static __init int init_uprobe_trace(void)
1404 {
1405         struct dentry *d_tracer;
1406
1407         d_tracer = tracing_init_dentry();
1408         if (IS_ERR(d_tracer))
1409                 return 0;
1410
1411         trace_create_file("uprobe_events", 0644, d_tracer,
1412                                     NULL, &uprobe_events_ops);
1413         /* Profile interface */
1414         trace_create_file("uprobe_profile", 0444, d_tracer,
1415                                     NULL, &uprobe_profile_ops);
1416         return 0;
1417 }
1418
1419 fs_initcall(init_uprobe_trace);