GNU Linux-libre 4.14.251-gnu1
[releases.git] / kernel / trace / trace_uprobe.c
1 /*
2  * uprobes-based tracing events
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
16  *
17  * Copyright (C) IBM Corporation, 2010-2012
18  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
19  */
20 #define pr_fmt(fmt)     "trace_uprobe: " fmt
21
22 #include <linux/module.h>
23 #include <linux/uaccess.h>
24 #include <linux/uprobes.h>
25 #include <linux/namei.h>
26 #include <linux/string.h>
27 #include <linux/rculist.h>
28
29 #include "trace_probe.h"
30
31 #define UPROBE_EVENT_SYSTEM     "uprobes"
32
33 struct uprobe_trace_entry_head {
34         struct trace_entry      ent;
35         unsigned long           vaddr[];
36 };
37
38 #define SIZEOF_TRACE_ENTRY(is_return)                   \
39         (sizeof(struct uprobe_trace_entry_head) +       \
40          sizeof(unsigned long) * (is_return ? 2 : 1))
41
42 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
43         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
44
45 struct trace_uprobe_filter {
46         rwlock_t                rwlock;
47         int                     nr_systemwide;
48         struct list_head        perf_events;
49 };
50
51 /*
52  * uprobe event core functions
53  */
54 struct trace_uprobe {
55         struct list_head                list;
56         struct trace_uprobe_filter      filter;
57         struct uprobe_consumer          consumer;
58         struct path                     path;
59         struct inode                    *inode;
60         char                            *filename;
61         unsigned long                   offset;
62         unsigned long                   nhit;
63         struct trace_probe              tp;
64 };
65
66 #define SIZEOF_TRACE_UPROBE(n)                          \
67         (offsetof(struct trace_uprobe, tp.args) +       \
68         (sizeof(struct probe_arg) * (n)))
69
70 static int register_uprobe_event(struct trace_uprobe *tu);
71 static int unregister_uprobe_event(struct trace_uprobe *tu);
72
73 static DEFINE_MUTEX(uprobe_lock);
74 static LIST_HEAD(uprobe_list);
75
76 struct uprobe_dispatch_data {
77         struct trace_uprobe     *tu;
78         unsigned long           bp_addr;
79 };
80
81 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
82 static int uretprobe_dispatcher(struct uprobe_consumer *con,
83                                 unsigned long func, struct pt_regs *regs);
84
85 #ifdef CONFIG_STACK_GROWSUP
86 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
87 {
88         return addr - (n * sizeof(long));
89 }
90 #else
91 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
92 {
93         return addr + (n * sizeof(long));
94 }
95 #endif
96
97 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
98 {
99         unsigned long ret;
100         unsigned long addr = user_stack_pointer(regs);
101
102         addr = adjust_stack_addr(addr, n);
103
104         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
105                 return 0;
106
107         return ret;
108 }
109
110 /*
111  * Uprobes-specific fetch functions
112  */
113 #define DEFINE_FETCH_stack(type)                                        \
114 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
115                                          void *offset, void *dest)      \
116 {                                                                       \
117         *(type *)dest = (type)get_user_stack_nth(regs,                  \
118                                               ((unsigned long)offset)); \
119 }
120 DEFINE_BASIC_FETCH_FUNCS(stack)
121 /* No string on the stack entry */
122 #define fetch_stack_string      NULL
123 #define fetch_stack_string_size NULL
124
125 #define DEFINE_FETCH_memory(type)                                       \
126 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
127                                           void *addr, void *dest)       \
128 {                                                                       \
129         type retval;                                                    \
130         void __user *vaddr = (void __force __user *) addr;              \
131                                                                         \
132         if (copy_from_user(&retval, vaddr, sizeof(type)))               \
133                 *(type *)dest = 0;                                      \
134         else                                                            \
135                 *(type *) dest = retval;                                \
136 }
137 DEFINE_BASIC_FETCH_FUNCS(memory)
138 /*
139  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
140  * length and relative data location.
141  */
142 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
143                                             void *addr, void *dest)
144 {
145         long ret;
146         u32 rloc = *(u32 *)dest;
147         int maxlen  = get_rloc_len(rloc);
148         u8 *dst = get_rloc_data(dest);
149         void __user *src = (void __force __user *) addr;
150
151         if (!maxlen)
152                 return;
153
154         ret = strncpy_from_user(dst, src, maxlen);
155         if (ret == maxlen)
156                 dst[ret - 1] = '\0';
157         else if (ret >= 0)
158                 /*
159                  * Include the terminating null byte. In this case it
160                  * was copied by strncpy_from_user but not accounted
161                  * for in ret.
162                  */
163                 ret++;
164
165         if (ret < 0) {  /* Failed to fetch string */
166                 ((u8 *)get_rloc_data(dest))[0] = '\0';
167                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
168         } else {
169                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
170         }
171 }
172
173 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
174                                                  void *addr, void *dest)
175 {
176         int len;
177         void __user *vaddr = (void __force __user *) addr;
178
179         len = strnlen_user(vaddr, MAX_STRING_SIZE);
180
181         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
182                 *(u32 *)dest = 0;
183         else
184                 *(u32 *)dest = len;
185 }
186
187 static unsigned long translate_user_vaddr(void *file_offset)
188 {
189         unsigned long base_addr;
190         struct uprobe_dispatch_data *udd;
191
192         udd = (void *) current->utask->vaddr;
193
194         base_addr = udd->bp_addr - udd->tu->offset;
195         return base_addr + (unsigned long)file_offset;
196 }
197
198 #define DEFINE_FETCH_file_offset(type)                                  \
199 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,    \
200                                                void *offset, void *dest)\
201 {                                                                       \
202         void *vaddr = (void *)translate_user_vaddr(offset);             \
203                                                                         \
204         FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);               \
205 }
206 DEFINE_BASIC_FETCH_FUNCS(file_offset)
207 DEFINE_FETCH_file_offset(string)
208 DEFINE_FETCH_file_offset(string_size)
209
210 /* Fetch type information table */
211 static const struct fetch_type uprobes_fetch_type_table[] = {
212         /* Special types */
213         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
214                                         sizeof(u32), 1, "__data_loc char[]"),
215         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
216                                         string_size, sizeof(u32), 0, "u32"),
217         /* Basic types */
218         ASSIGN_FETCH_TYPE(u8,  u8,  0),
219         ASSIGN_FETCH_TYPE(u16, u16, 0),
220         ASSIGN_FETCH_TYPE(u32, u32, 0),
221         ASSIGN_FETCH_TYPE(u64, u64, 0),
222         ASSIGN_FETCH_TYPE(s8,  u8,  1),
223         ASSIGN_FETCH_TYPE(s16, u16, 1),
224         ASSIGN_FETCH_TYPE(s32, u32, 1),
225         ASSIGN_FETCH_TYPE(s64, u64, 1),
226         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
227         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
228         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
229         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
230
231         ASSIGN_FETCH_TYPE_END
232 };
233
234 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
235 {
236         rwlock_init(&filter->rwlock);
237         filter->nr_systemwide = 0;
238         INIT_LIST_HEAD(&filter->perf_events);
239 }
240
241 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
242 {
243         return !filter->nr_systemwide && list_empty(&filter->perf_events);
244 }
245
246 static inline bool is_ret_probe(struct trace_uprobe *tu)
247 {
248         return tu->consumer.ret_handler != NULL;
249 }
250
251 /*
252  * Allocate new trace_uprobe and initialize it (including uprobes).
253  */
254 static struct trace_uprobe *
255 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
256 {
257         struct trace_uprobe *tu;
258
259         if (!event || !is_good_name(event))
260                 return ERR_PTR(-EINVAL);
261
262         if (!group || !is_good_name(group))
263                 return ERR_PTR(-EINVAL);
264
265         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
266         if (!tu)
267                 return ERR_PTR(-ENOMEM);
268
269         tu->tp.call.class = &tu->tp.class;
270         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
271         if (!tu->tp.call.name)
272                 goto error;
273
274         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
275         if (!tu->tp.class.system)
276                 goto error;
277
278         INIT_LIST_HEAD(&tu->list);
279         INIT_LIST_HEAD(&tu->tp.files);
280         tu->consumer.handler = uprobe_dispatcher;
281         if (is_ret)
282                 tu->consumer.ret_handler = uretprobe_dispatcher;
283         init_trace_uprobe_filter(&tu->filter);
284         return tu;
285
286 error:
287         kfree(tu->tp.call.name);
288         kfree(tu);
289
290         return ERR_PTR(-ENOMEM);
291 }
292
293 static void free_trace_uprobe(struct trace_uprobe *tu)
294 {
295         int i;
296
297         for (i = 0; i < tu->tp.nr_args; i++)
298                 traceprobe_free_probe_arg(&tu->tp.args[i]);
299
300         path_put(&tu->path);
301         kfree(tu->tp.call.class->system);
302         kfree(tu->tp.call.name);
303         kfree(tu->filename);
304         kfree(tu);
305 }
306
307 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
308 {
309         struct trace_uprobe *tu;
310
311         list_for_each_entry(tu, &uprobe_list, list)
312                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
313                     strcmp(tu->tp.call.class->system, group) == 0)
314                         return tu;
315
316         return NULL;
317 }
318
319 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
320 static int unregister_trace_uprobe(struct trace_uprobe *tu)
321 {
322         int ret;
323
324         ret = unregister_uprobe_event(tu);
325         if (ret)
326                 return ret;
327
328         list_del(&tu->list);
329         free_trace_uprobe(tu);
330         return 0;
331 }
332
333 /* Register a trace_uprobe and probe_event */
334 static int register_trace_uprobe(struct trace_uprobe *tu)
335 {
336         struct trace_uprobe *old_tu;
337         int ret;
338
339         mutex_lock(&uprobe_lock);
340
341         /* register as an event */
342         old_tu = find_probe_event(trace_event_name(&tu->tp.call),
343                         tu->tp.call.class->system);
344         if (old_tu) {
345                 /* delete old event */
346                 ret = unregister_trace_uprobe(old_tu);
347                 if (ret)
348                         goto end;
349         }
350
351         ret = register_uprobe_event(tu);
352         if (ret) {
353                 pr_warn("Failed to register probe event(%d)\n", ret);
354                 goto end;
355         }
356
357         list_add_tail(&tu->list, &uprobe_list);
358
359 end:
360         mutex_unlock(&uprobe_lock);
361
362         return ret;
363 }
364
365 /*
366  * Argument syntax:
367  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
368  *
369  *  - Remove uprobe: -:[GRP/]EVENT
370  */
371 static int create_trace_uprobe(int argc, char **argv)
372 {
373         struct trace_uprobe *tu;
374         char *arg, *event, *group, *filename;
375         char buf[MAX_EVENT_NAME_LEN];
376         struct path path;
377         unsigned long offset;
378         bool is_delete, is_return;
379         int i, ret;
380
381         ret = 0;
382         is_delete = false;
383         is_return = false;
384         event = NULL;
385         group = NULL;
386
387         /* argc must be >= 1 */
388         if (argv[0][0] == '-')
389                 is_delete = true;
390         else if (argv[0][0] == 'r')
391                 is_return = true;
392         else if (argv[0][0] != 'p') {
393                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
394                 return -EINVAL;
395         }
396
397         if (argv[0][1] == ':') {
398                 event = &argv[0][2];
399                 arg = strchr(event, '/');
400
401                 if (arg) {
402                         group = event;
403                         event = arg + 1;
404                         event[-1] = '\0';
405
406                         if (strlen(group) == 0) {
407                                 pr_info("Group name is not specified\n");
408                                 return -EINVAL;
409                         }
410                 }
411                 if (strlen(event) == 0) {
412                         pr_info("Event name is not specified\n");
413                         return -EINVAL;
414                 }
415         }
416         if (!group)
417                 group = UPROBE_EVENT_SYSTEM;
418
419         if (is_delete) {
420                 int ret;
421
422                 if (!event) {
423                         pr_info("Delete command needs an event name.\n");
424                         return -EINVAL;
425                 }
426                 mutex_lock(&uprobe_lock);
427                 tu = find_probe_event(event, group);
428
429                 if (!tu) {
430                         mutex_unlock(&uprobe_lock);
431                         pr_info("Event %s/%s doesn't exist.\n", group, event);
432                         return -ENOENT;
433                 }
434                 /* delete an event */
435                 ret = unregister_trace_uprobe(tu);
436                 mutex_unlock(&uprobe_lock);
437                 return ret;
438         }
439
440         if (argc < 2) {
441                 pr_info("Probe point is not specified.\n");
442                 return -EINVAL;
443         }
444         /* Find the last occurrence, in case the path contains ':' too. */
445         arg = strrchr(argv[1], ':');
446         if (!arg)
447                 return -EINVAL;
448
449         *arg++ = '\0';
450         filename = argv[1];
451         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
452         if (ret)
453                 return ret;
454
455         if (!d_is_reg(path.dentry)) {
456                 ret = -EINVAL;
457                 goto fail_address_parse;
458         }
459
460         ret = kstrtoul(arg, 0, &offset);
461         if (ret)
462                 goto fail_address_parse;
463
464         argc -= 2;
465         argv += 2;
466
467         /* setup a probe */
468         if (!event) {
469                 char *tail;
470                 char *ptr;
471
472                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
473                 if (!tail) {
474                         ret = -ENOMEM;
475                         goto fail_address_parse;
476                 }
477
478                 ptr = strpbrk(tail, ".-_");
479                 if (ptr)
480                         *ptr = '\0';
481
482                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
483                 event = buf;
484                 kfree(tail);
485         }
486
487         tu = alloc_trace_uprobe(group, event, argc, is_return);
488         if (IS_ERR(tu)) {
489                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
490                 ret = PTR_ERR(tu);
491                 goto fail_address_parse;
492         }
493         tu->offset = offset;
494         tu->path = path;
495         tu->filename = kstrdup(filename, GFP_KERNEL);
496
497         if (!tu->filename) {
498                 pr_info("Failed to allocate filename.\n");
499                 ret = -ENOMEM;
500                 goto error;
501         }
502
503         /* parse arguments */
504         ret = 0;
505         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
506                 struct probe_arg *parg = &tu->tp.args[i];
507
508                 /* Increment count for freeing args in error case */
509                 tu->tp.nr_args++;
510
511                 /* Parse argument name */
512                 arg = strchr(argv[i], '=');
513                 if (arg) {
514                         *arg++ = '\0';
515                         parg->name = kstrdup(argv[i], GFP_KERNEL);
516                 } else {
517                         arg = argv[i];
518                         /* If argument name is omitted, set "argN" */
519                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
520                         parg->name = kstrdup(buf, GFP_KERNEL);
521                 }
522
523                 if (!parg->name) {
524                         pr_info("Failed to allocate argument[%d] name.\n", i);
525                         ret = -ENOMEM;
526                         goto error;
527                 }
528
529                 if (!is_good_name(parg->name)) {
530                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
531                         ret = -EINVAL;
532                         goto error;
533                 }
534
535                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
536                         pr_info("Argument[%d] name '%s' conflicts with "
537                                 "another field.\n", i, argv[i]);
538                         ret = -EINVAL;
539                         goto error;
540                 }
541
542                 /* Parse fetch argument */
543                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
544                                                  is_return, false,
545                                                  uprobes_fetch_type_table);
546                 if (ret) {
547                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
548                         goto error;
549                 }
550         }
551
552         ret = register_trace_uprobe(tu);
553         if (ret)
554                 goto error;
555         return 0;
556
557 error:
558         free_trace_uprobe(tu);
559         return ret;
560
561 fail_address_parse:
562         path_put(&path);
563
564         pr_info("Failed to parse address or file.\n");
565
566         return ret;
567 }
568
569 static int cleanup_all_probes(void)
570 {
571         struct trace_uprobe *tu;
572         int ret = 0;
573
574         mutex_lock(&uprobe_lock);
575         while (!list_empty(&uprobe_list)) {
576                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
577                 ret = unregister_trace_uprobe(tu);
578                 if (ret)
579                         break;
580         }
581         mutex_unlock(&uprobe_lock);
582         return ret;
583 }
584
585 /* Probes listing interfaces */
586 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
587 {
588         mutex_lock(&uprobe_lock);
589         return seq_list_start(&uprobe_list, *pos);
590 }
591
592 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
593 {
594         return seq_list_next(v, &uprobe_list, pos);
595 }
596
597 static void probes_seq_stop(struct seq_file *m, void *v)
598 {
599         mutex_unlock(&uprobe_lock);
600 }
601
602 static int probes_seq_show(struct seq_file *m, void *v)
603 {
604         struct trace_uprobe *tu = v;
605         char c = is_ret_probe(tu) ? 'r' : 'p';
606         int i;
607
608         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
609                         trace_event_name(&tu->tp.call));
610         seq_printf(m, " %s:", tu->filename);
611
612         /* Don't print "0x  (null)" when offset is 0 */
613         if (tu->offset) {
614                 seq_printf(m, "0x%p", (void *)tu->offset);
615         } else {
616                 switch (sizeof(void *)) {
617                 case 4:
618                         seq_printf(m, "0x00000000");
619                         break;
620                 case 8:
621                 default:
622                         seq_printf(m, "0x0000000000000000");
623                         break;
624                 }
625         }
626
627         for (i = 0; i < tu->tp.nr_args; i++)
628                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
629
630         seq_putc(m, '\n');
631         return 0;
632 }
633
634 static const struct seq_operations probes_seq_op = {
635         .start  = probes_seq_start,
636         .next   = probes_seq_next,
637         .stop   = probes_seq_stop,
638         .show   = probes_seq_show
639 };
640
641 static int probes_open(struct inode *inode, struct file *file)
642 {
643         int ret;
644
645         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
646                 ret = cleanup_all_probes();
647                 if (ret)
648                         return ret;
649         }
650
651         return seq_open(file, &probes_seq_op);
652 }
653
654 static ssize_t probes_write(struct file *file, const char __user *buffer,
655                             size_t count, loff_t *ppos)
656 {
657         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
658 }
659
660 static const struct file_operations uprobe_events_ops = {
661         .owner          = THIS_MODULE,
662         .open           = probes_open,
663         .read           = seq_read,
664         .llseek         = seq_lseek,
665         .release        = seq_release,
666         .write          = probes_write,
667 };
668
669 /* Probes profiling interfaces */
670 static int probes_profile_seq_show(struct seq_file *m, void *v)
671 {
672         struct trace_uprobe *tu = v;
673
674         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
675                         trace_event_name(&tu->tp.call), tu->nhit);
676         return 0;
677 }
678
679 static const struct seq_operations profile_seq_op = {
680         .start  = probes_seq_start,
681         .next   = probes_seq_next,
682         .stop   = probes_seq_stop,
683         .show   = probes_profile_seq_show
684 };
685
686 static int profile_open(struct inode *inode, struct file *file)
687 {
688         return seq_open(file, &profile_seq_op);
689 }
690
691 static const struct file_operations uprobe_profile_ops = {
692         .owner          = THIS_MODULE,
693         .open           = profile_open,
694         .read           = seq_read,
695         .llseek         = seq_lseek,
696         .release        = seq_release,
697 };
698
699 struct uprobe_cpu_buffer {
700         struct mutex mutex;
701         void *buf;
702 };
703 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
704 static int uprobe_buffer_refcnt;
705
706 static int uprobe_buffer_init(void)
707 {
708         int cpu, err_cpu;
709
710         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
711         if (uprobe_cpu_buffer == NULL)
712                 return -ENOMEM;
713
714         for_each_possible_cpu(cpu) {
715                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
716                                                   GFP_KERNEL, 0);
717                 if (p == NULL) {
718                         err_cpu = cpu;
719                         goto err;
720                 }
721                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
722                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
723         }
724
725         return 0;
726
727 err:
728         for_each_possible_cpu(cpu) {
729                 if (cpu == err_cpu)
730                         break;
731                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
732         }
733
734         free_percpu(uprobe_cpu_buffer);
735         return -ENOMEM;
736 }
737
738 static int uprobe_buffer_enable(void)
739 {
740         int ret = 0;
741
742         BUG_ON(!mutex_is_locked(&event_mutex));
743
744         if (uprobe_buffer_refcnt++ == 0) {
745                 ret = uprobe_buffer_init();
746                 if (ret < 0)
747                         uprobe_buffer_refcnt--;
748         }
749
750         return ret;
751 }
752
753 static void uprobe_buffer_disable(void)
754 {
755         int cpu;
756
757         BUG_ON(!mutex_is_locked(&event_mutex));
758
759         if (--uprobe_buffer_refcnt == 0) {
760                 for_each_possible_cpu(cpu)
761                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
762                                                              cpu)->buf);
763
764                 free_percpu(uprobe_cpu_buffer);
765                 uprobe_cpu_buffer = NULL;
766         }
767 }
768
769 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
770 {
771         struct uprobe_cpu_buffer *ucb;
772         int cpu;
773
774         cpu = raw_smp_processor_id();
775         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
776
777         /*
778          * Use per-cpu buffers for fastest access, but we might migrate
779          * so the mutex makes sure we have sole access to it.
780          */
781         mutex_lock(&ucb->mutex);
782
783         return ucb;
784 }
785
786 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
787 {
788         mutex_unlock(&ucb->mutex);
789 }
790
791 static void __uprobe_trace_func(struct trace_uprobe *tu,
792                                 unsigned long func, struct pt_regs *regs,
793                                 struct uprobe_cpu_buffer *ucb, int dsize,
794                                 struct trace_event_file *trace_file)
795 {
796         struct uprobe_trace_entry_head *entry;
797         struct ring_buffer_event *event;
798         struct ring_buffer *buffer;
799         void *data;
800         int size, esize;
801         struct trace_event_call *call = &tu->tp.call;
802
803         WARN_ON(call != trace_file->event_call);
804
805         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
806                 return;
807
808         if (trace_trigger_soft_disabled(trace_file))
809                 return;
810
811         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
812         size = esize + tu->tp.size + dsize;
813         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
814                                                 call->event.type, size, 0, 0);
815         if (!event)
816                 return;
817
818         entry = ring_buffer_event_data(event);
819         if (is_ret_probe(tu)) {
820                 entry->vaddr[0] = func;
821                 entry->vaddr[1] = instruction_pointer(regs);
822                 data = DATAOF_TRACE_ENTRY(entry, true);
823         } else {
824                 entry->vaddr[0] = instruction_pointer(regs);
825                 data = DATAOF_TRACE_ENTRY(entry, false);
826         }
827
828         memcpy(data, ucb->buf, tu->tp.size + dsize);
829
830         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
831 }
832
833 /* uprobe handler */
834 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
835                              struct uprobe_cpu_buffer *ucb, int dsize)
836 {
837         struct event_file_link *link;
838
839         if (is_ret_probe(tu))
840                 return 0;
841
842         rcu_read_lock();
843         list_for_each_entry_rcu(link, &tu->tp.files, list)
844                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
845         rcu_read_unlock();
846
847         return 0;
848 }
849
850 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
851                                  struct pt_regs *regs,
852                                  struct uprobe_cpu_buffer *ucb, int dsize)
853 {
854         struct event_file_link *link;
855
856         rcu_read_lock();
857         list_for_each_entry_rcu(link, &tu->tp.files, list)
858                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
859         rcu_read_unlock();
860 }
861
862 /* Event entry printers */
863 static enum print_line_t
864 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
865 {
866         struct uprobe_trace_entry_head *entry;
867         struct trace_seq *s = &iter->seq;
868         struct trace_uprobe *tu;
869         u8 *data;
870         int i;
871
872         entry = (struct uprobe_trace_entry_head *)iter->ent;
873         tu = container_of(event, struct trace_uprobe, tp.call.event);
874
875         if (is_ret_probe(tu)) {
876                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
877                                  trace_event_name(&tu->tp.call),
878                                  entry->vaddr[1], entry->vaddr[0]);
879                 data = DATAOF_TRACE_ENTRY(entry, true);
880         } else {
881                 trace_seq_printf(s, "%s: (0x%lx)",
882                                  trace_event_name(&tu->tp.call),
883                                  entry->vaddr[0]);
884                 data = DATAOF_TRACE_ENTRY(entry, false);
885         }
886
887         for (i = 0; i < tu->tp.nr_args; i++) {
888                 struct probe_arg *parg = &tu->tp.args[i];
889
890                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
891                         goto out;
892         }
893
894         trace_seq_putc(s, '\n');
895
896  out:
897         return trace_handle_return(s);
898 }
899
900 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
901                                 enum uprobe_filter_ctx ctx,
902                                 struct mm_struct *mm);
903
904 static int
905 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
906                    filter_func_t filter)
907 {
908         bool enabled = trace_probe_is_enabled(&tu->tp);
909         struct event_file_link *link = NULL;
910         int ret;
911
912         if (file) {
913                 if (tu->tp.flags & TP_FLAG_PROFILE)
914                         return -EINTR;
915
916                 link = kmalloc(sizeof(*link), GFP_KERNEL);
917                 if (!link)
918                         return -ENOMEM;
919
920                 link->file = file;
921                 list_add_tail_rcu(&link->list, &tu->tp.files);
922
923                 tu->tp.flags |= TP_FLAG_TRACE;
924         } else {
925                 if (tu->tp.flags & TP_FLAG_TRACE)
926                         return -EINTR;
927
928                 tu->tp.flags |= TP_FLAG_PROFILE;
929         }
930
931         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
932
933         if (enabled)
934                 return 0;
935
936         ret = uprobe_buffer_enable();
937         if (ret)
938                 goto err_flags;
939
940         tu->consumer.filter = filter;
941         tu->inode = d_real_inode(tu->path.dentry);
942         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
943         if (ret)
944                 goto err_buffer;
945
946         return 0;
947
948  err_buffer:
949         uprobe_buffer_disable();
950
951  err_flags:
952         if (file) {
953                 list_del(&link->list);
954                 kfree(link);
955                 tu->tp.flags &= ~TP_FLAG_TRACE;
956         } else {
957                 tu->tp.flags &= ~TP_FLAG_PROFILE;
958         }
959         return ret;
960 }
961
962 static void
963 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
964 {
965         if (!trace_probe_is_enabled(&tu->tp))
966                 return;
967
968         if (file) {
969                 struct event_file_link *link;
970
971                 link = find_event_file_link(&tu->tp, file);
972                 if (!link)
973                         return;
974
975                 list_del_rcu(&link->list);
976                 /* synchronize with u{,ret}probe_trace_func */
977                 synchronize_rcu();
978                 kfree(link);
979
980                 if (!list_empty(&tu->tp.files))
981                         return;
982         }
983
984         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
985
986         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
987         tu->inode = NULL;
988         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
989
990         uprobe_buffer_disable();
991 }
992
993 static int uprobe_event_define_fields(struct trace_event_call *event_call)
994 {
995         int ret, i, size;
996         struct uprobe_trace_entry_head field;
997         struct trace_uprobe *tu = event_call->data;
998
999         if (is_ret_probe(tu)) {
1000                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1001                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1002                 size = SIZEOF_TRACE_ENTRY(true);
1003         } else {
1004                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1005                 size = SIZEOF_TRACE_ENTRY(false);
1006         }
1007         /* Set argument names as fields */
1008         for (i = 0; i < tu->tp.nr_args; i++) {
1009                 struct probe_arg *parg = &tu->tp.args[i];
1010
1011                 ret = trace_define_field(event_call, parg->type->fmttype,
1012                                          parg->name, size + parg->offset,
1013                                          parg->type->size, parg->type->is_signed,
1014                                          FILTER_OTHER);
1015
1016                 if (ret)
1017                         return ret;
1018         }
1019         return 0;
1020 }
1021
1022 #ifdef CONFIG_PERF_EVENTS
1023 static bool
1024 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1025 {
1026         struct perf_event *event;
1027
1028         if (filter->nr_systemwide)
1029                 return true;
1030
1031         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1032                 if (event->hw.target->mm == mm)
1033                         return true;
1034         }
1035
1036         return false;
1037 }
1038
1039 static inline bool
1040 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1041 {
1042         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1043 }
1044
1045 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1046 {
1047         bool done;
1048
1049         write_lock(&tu->filter.rwlock);
1050         if (event->hw.target) {
1051                 list_del(&event->hw.tp_list);
1052                 done = tu->filter.nr_systemwide ||
1053                         (event->hw.target->flags & PF_EXITING) ||
1054                         uprobe_filter_event(tu, event);
1055         } else {
1056                 tu->filter.nr_systemwide--;
1057                 done = tu->filter.nr_systemwide;
1058         }
1059         write_unlock(&tu->filter.rwlock);
1060
1061         if (!done)
1062                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1063
1064         return 0;
1065 }
1066
1067 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1068 {
1069         bool done;
1070         int err;
1071
1072         write_lock(&tu->filter.rwlock);
1073         if (event->hw.target) {
1074                 /*
1075                  * event->parent != NULL means copy_process(), we can avoid
1076                  * uprobe_apply(). current->mm must be probed and we can rely
1077                  * on dup_mmap() which preserves the already installed bp's.
1078                  *
1079                  * attr.enable_on_exec means that exec/mmap will install the
1080                  * breakpoints we need.
1081                  */
1082                 done = tu->filter.nr_systemwide ||
1083                         event->parent || event->attr.enable_on_exec ||
1084                         uprobe_filter_event(tu, event);
1085                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1086         } else {
1087                 done = tu->filter.nr_systemwide;
1088                 tu->filter.nr_systemwide++;
1089         }
1090         write_unlock(&tu->filter.rwlock);
1091
1092         err = 0;
1093         if (!done) {
1094                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1095                 if (err)
1096                         uprobe_perf_close(tu, event);
1097         }
1098         return err;
1099 }
1100
1101 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1102                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1103 {
1104         struct trace_uprobe *tu;
1105         int ret;
1106
1107         tu = container_of(uc, struct trace_uprobe, consumer);
1108         read_lock(&tu->filter.rwlock);
1109         ret = __uprobe_perf_filter(&tu->filter, mm);
1110         read_unlock(&tu->filter.rwlock);
1111
1112         return ret;
1113 }
1114
1115 static void __uprobe_perf_func(struct trace_uprobe *tu,
1116                                unsigned long func, struct pt_regs *regs,
1117                                struct uprobe_cpu_buffer *ucb, int dsize)
1118 {
1119         struct trace_event_call *call = &tu->tp.call;
1120         struct uprobe_trace_entry_head *entry;
1121         struct bpf_prog *prog = call->prog;
1122         struct hlist_head *head;
1123         void *data;
1124         int size, esize;
1125         int rctx;
1126
1127         if (prog && !trace_call_bpf(prog, regs))
1128                 return;
1129
1130         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1131
1132         size = esize + tu->tp.size + dsize;
1133         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1134         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1135                 return;
1136
1137         preempt_disable();
1138         head = this_cpu_ptr(call->perf_events);
1139         if (hlist_empty(head))
1140                 goto out;
1141
1142         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1143         if (!entry)
1144                 goto out;
1145
1146         if (is_ret_probe(tu)) {
1147                 entry->vaddr[0] = func;
1148                 entry->vaddr[1] = instruction_pointer(regs);
1149                 data = DATAOF_TRACE_ENTRY(entry, true);
1150         } else {
1151                 entry->vaddr[0] = instruction_pointer(regs);
1152                 data = DATAOF_TRACE_ENTRY(entry, false);
1153         }
1154
1155         memcpy(data, ucb->buf, tu->tp.size + dsize);
1156
1157         if (size - esize > tu->tp.size + dsize) {
1158                 int len = tu->tp.size + dsize;
1159
1160                 memset(data + len, 0, size - esize - len);
1161         }
1162
1163         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1164                               head, NULL, NULL);
1165  out:
1166         preempt_enable();
1167 }
1168
1169 /* uprobe profile handler */
1170 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1171                             struct uprobe_cpu_buffer *ucb, int dsize)
1172 {
1173         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1174                 return UPROBE_HANDLER_REMOVE;
1175
1176         if (!is_ret_probe(tu))
1177                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1178         return 0;
1179 }
1180
1181 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1182                                 struct pt_regs *regs,
1183                                 struct uprobe_cpu_buffer *ucb, int dsize)
1184 {
1185         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1186 }
1187 #endif  /* CONFIG_PERF_EVENTS */
1188
1189 static int
1190 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1191                       void *data)
1192 {
1193         struct trace_uprobe *tu = event->data;
1194         struct trace_event_file *file = data;
1195
1196         switch (type) {
1197         case TRACE_REG_REGISTER:
1198                 return probe_event_enable(tu, file, NULL);
1199
1200         case TRACE_REG_UNREGISTER:
1201                 probe_event_disable(tu, file);
1202                 return 0;
1203
1204 #ifdef CONFIG_PERF_EVENTS
1205         case TRACE_REG_PERF_REGISTER:
1206                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1207
1208         case TRACE_REG_PERF_UNREGISTER:
1209                 probe_event_disable(tu, NULL);
1210                 return 0;
1211
1212         case TRACE_REG_PERF_OPEN:
1213                 return uprobe_perf_open(tu, data);
1214
1215         case TRACE_REG_PERF_CLOSE:
1216                 return uprobe_perf_close(tu, data);
1217
1218 #endif
1219         default:
1220                 return 0;
1221         }
1222         return 0;
1223 }
1224
1225 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1226 {
1227         struct trace_uprobe *tu;
1228         struct uprobe_dispatch_data udd;
1229         struct uprobe_cpu_buffer *ucb;
1230         int dsize, esize;
1231         int ret = 0;
1232
1233
1234         tu = container_of(con, struct trace_uprobe, consumer);
1235         tu->nhit++;
1236
1237         udd.tu = tu;
1238         udd.bp_addr = instruction_pointer(regs);
1239
1240         current->utask->vaddr = (unsigned long) &udd;
1241
1242         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1243                 return 0;
1244
1245         dsize = __get_data_size(&tu->tp, regs);
1246         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1247
1248         ucb = uprobe_buffer_get();
1249         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1250
1251         if (tu->tp.flags & TP_FLAG_TRACE)
1252                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1253
1254 #ifdef CONFIG_PERF_EVENTS
1255         if (tu->tp.flags & TP_FLAG_PROFILE)
1256                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1257 #endif
1258         uprobe_buffer_put(ucb);
1259         return ret;
1260 }
1261
1262 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1263                                 unsigned long func, struct pt_regs *regs)
1264 {
1265         struct trace_uprobe *tu;
1266         struct uprobe_dispatch_data udd;
1267         struct uprobe_cpu_buffer *ucb;
1268         int dsize, esize;
1269
1270         tu = container_of(con, struct trace_uprobe, consumer);
1271
1272         udd.tu = tu;
1273         udd.bp_addr = func;
1274
1275         current->utask->vaddr = (unsigned long) &udd;
1276
1277         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1278                 return 0;
1279
1280         dsize = __get_data_size(&tu->tp, regs);
1281         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1282
1283         ucb = uprobe_buffer_get();
1284         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1285
1286         if (tu->tp.flags & TP_FLAG_TRACE)
1287                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1288
1289 #ifdef CONFIG_PERF_EVENTS
1290         if (tu->tp.flags & TP_FLAG_PROFILE)
1291                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1292 #endif
1293         uprobe_buffer_put(ucb);
1294         return 0;
1295 }
1296
1297 static struct trace_event_functions uprobe_funcs = {
1298         .trace          = print_uprobe_event
1299 };
1300
1301 static int register_uprobe_event(struct trace_uprobe *tu)
1302 {
1303         struct trace_event_call *call = &tu->tp.call;
1304         int ret;
1305
1306         /* Initialize trace_event_call */
1307         INIT_LIST_HEAD(&call->class->fields);
1308         call->event.funcs = &uprobe_funcs;
1309         call->class->define_fields = uprobe_event_define_fields;
1310
1311         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1312                 return -ENOMEM;
1313
1314         ret = register_trace_event(&call->event);
1315         if (!ret) {
1316                 kfree(call->print_fmt);
1317                 return -ENODEV;
1318         }
1319
1320         call->flags = TRACE_EVENT_FL_UPROBE;
1321         call->class->reg = trace_uprobe_register;
1322         call->data = tu;
1323         ret = trace_add_event_call(call);
1324
1325         if (ret) {
1326                 pr_info("Failed to register uprobe event: %s\n",
1327                         trace_event_name(call));
1328                 kfree(call->print_fmt);
1329                 unregister_trace_event(&call->event);
1330         }
1331
1332         return ret;
1333 }
1334
1335 static int unregister_uprobe_event(struct trace_uprobe *tu)
1336 {
1337         int ret;
1338
1339         /* tu->event is unregistered in trace_remove_event_call() */
1340         ret = trace_remove_event_call(&tu->tp.call);
1341         if (ret)
1342                 return ret;
1343         kfree(tu->tp.call.print_fmt);
1344         tu->tp.call.print_fmt = NULL;
1345         return 0;
1346 }
1347
1348 /* Make a trace interface for controling probe points */
1349 static __init int init_uprobe_trace(void)
1350 {
1351         struct dentry *d_tracer;
1352
1353         d_tracer = tracing_init_dentry();
1354         if (IS_ERR(d_tracer))
1355                 return 0;
1356
1357         trace_create_file("uprobe_events", 0644, d_tracer,
1358                                     NULL, &uprobe_events_ops);
1359         /* Profile interface */
1360         trace_create_file("uprobe_profile", 0444, d_tracer,
1361                                     NULL, &uprobe_profile_ops);
1362         return 0;
1363 }
1364
1365 fs_initcall(init_uprobe_trace);