Commit | Line | Data |
---|---|---|
2b144498 | 1 | /* |
7b2d81d4 | 2 | * User-space Probes (UProbes) |
2b144498 SD |
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 as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
35aa621b | 18 | * Copyright (C) IBM Corporation, 2008-2012 |
2b144498 SD |
19 | * Authors: |
20 | * Srikar Dronamraju | |
21 | * Jim Keniston | |
35aa621b | 22 | * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> |
2b144498 SD |
23 | */ |
24 | ||
25 | #include <linux/kernel.h> | |
26 | #include <linux/highmem.h> | |
27 | #include <linux/pagemap.h> /* read_mapping_page */ | |
28 | #include <linux/slab.h> | |
29 | #include <linux/sched.h> | |
30 | #include <linux/rmap.h> /* anon_vma_prepare */ | |
31 | #include <linux/mmu_notifier.h> /* set_pte_at_notify */ | |
32 | #include <linux/swap.h> /* try_to_free_swap */ | |
0326f5a9 SD |
33 | #include <linux/ptrace.h> /* user_enable_single_step */ |
34 | #include <linux/kdebug.h> /* notifier mechanism */ | |
194f8dcb | 35 | #include "../../mm/internal.h" /* munlock_vma_page */ |
7b2d81d4 | 36 | |
2b144498 SD |
37 | #include <linux/uprobes.h> |
38 | ||
d4b3b638 SD |
39 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
40 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
41 | ||
2b144498 | 42 | static struct rb_root uprobes_tree = RB_ROOT; |
7b2d81d4 | 43 | |
2b144498 SD |
44 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
45 | ||
46 | #define UPROBES_HASH_SZ 13 | |
7b2d81d4 | 47 | |
c5784de2 PZ |
48 | /* |
49 | * We need separate register/unregister and mmap/munmap lock hashes because | |
50 | * of mmap_sem nesting. | |
51 | * | |
52 | * uprobe_register() needs to install probes on (potentially) all processes | |
53 | * and thus needs to acquire multiple mmap_sems (consequtively, not | |
54 | * concurrently), whereas uprobe_mmap() is called while holding mmap_sem | |
55 | * for the particular process doing the mmap. | |
56 | * | |
57 | * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem | |
58 | * because of lock order against i_mmap_mutex. This means there's a hole in | |
59 | * the register vma iteration where a mmap() can happen. | |
60 | * | |
61 | * Thus uprobe_register() can race with uprobe_mmap() and we can try and | |
62 | * install a probe where one is already installed. | |
63 | */ | |
64 | ||
2b144498 SD |
65 | /* serialize (un)register */ |
66 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
67 | |
68 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
69 | |
70 | /* serialize uprobe->pending_list */ | |
71 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 72 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 SD |
73 | |
74 | /* | |
7b2d81d4 | 75 | * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe |
2b144498 SD |
76 | * events active at this time. Probably a fine grained per inode count is |
77 | * better? | |
78 | */ | |
79 | static atomic_t uprobe_events = ATOMIC_INIT(0); | |
80 | ||
3ff54efd SD |
81 | struct uprobe { |
82 | struct rb_node rb_node; /* node in the rb tree */ | |
83 | atomic_t ref; | |
84 | struct rw_semaphore consumer_rwsem; | |
85 | struct list_head pending_list; | |
86 | struct uprobe_consumer *consumers; | |
87 | struct inode *inode; /* Also hold a ref to inode */ | |
88 | loff_t offset; | |
89 | int flags; | |
90 | struct arch_uprobe arch; | |
91 | }; | |
92 | ||
2b144498 SD |
93 | /* |
94 | * valid_vma: Verify if the specified vma is an executable vma | |
95 | * Relax restrictions while unregistering: vm_flags might have | |
96 | * changed after breakpoint was inserted. | |
97 | * - is_register: indicates if we are in register context. | |
98 | * - Return 1 if the specified virtual address is in an | |
99 | * executable vma. | |
100 | */ | |
101 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
102 | { | |
103 | if (!vma->vm_file) | |
104 | return false; | |
105 | ||
106 | if (!is_register) | |
107 | return true; | |
108 | ||
ea131377 ON |
109 | if ((vma->vm_flags & (VM_HUGETLB|VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)) |
110 | == (VM_READ|VM_EXEC)) | |
2b144498 SD |
111 | return true; |
112 | ||
113 | return false; | |
114 | } | |
115 | ||
57683f72 | 116 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 117 | { |
57683f72 | 118 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
119 | } |
120 | ||
cb113b47 ON |
121 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
122 | { | |
123 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
124 | } | |
125 | ||
2b144498 SD |
126 | /** |
127 | * __replace_page - replace page in vma by new page. | |
128 | * based on replace_page in mm/ksm.c | |
129 | * | |
130 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 131 | * @addr: address the old @page is mapped at |
2b144498 SD |
132 | * @page: the cowed page we are replacing by kpage |
133 | * @kpage: the modified page we replace page by | |
134 | * | |
135 | * Returns 0 on success, -EFAULT on failure. | |
136 | */ | |
c517ee74 ON |
137 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
138 | struct page *page, struct page *kpage) | |
2b144498 SD |
139 | { |
140 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
141 | spinlock_t *ptl; |
142 | pte_t *ptep; | |
9f92448c | 143 | int err; |
2b144498 | 144 | |
194f8dcb | 145 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
146 | lock_page(page); |
147 | ||
148 | err = -EAGAIN; | |
5323ce71 | 149 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 150 | if (!ptep) |
9f92448c | 151 | goto unlock; |
2b144498 SD |
152 | |
153 | get_page(kpage); | |
154 | page_add_new_anon_rmap(kpage, vma, addr); | |
155 | ||
7396fa81 SD |
156 | if (!PageAnon(page)) { |
157 | dec_mm_counter(mm, MM_FILEPAGES); | |
158 | inc_mm_counter(mm, MM_ANONPAGES); | |
159 | } | |
160 | ||
2b144498 SD |
161 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
162 | ptep_clear_flush(vma, addr, ptep); | |
163 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
164 | ||
165 | page_remove_rmap(page); | |
166 | if (!page_mapped(page)) | |
167 | try_to_free_swap(page); | |
2b144498 | 168 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 169 | |
194f8dcb ON |
170 | if (vma->vm_flags & VM_LOCKED) |
171 | munlock_vma_page(page); | |
172 | put_page(page); | |
173 | ||
9f92448c ON |
174 | err = 0; |
175 | unlock: | |
176 | unlock_page(page); | |
177 | return err; | |
2b144498 SD |
178 | } |
179 | ||
180 | /** | |
5cb4ac3a | 181 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 182 | * @insn: instruction to be checked. |
5cb4ac3a | 183 | * Default implementation of is_swbp_insn |
2b144498 SD |
184 | * Returns true if @insn is a breakpoint instruction. |
185 | */ | |
5cb4ac3a | 186 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 187 | { |
5cb4ac3a | 188 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
189 | } |
190 | ||
191 | /* | |
192 | * NOTE: | |
193 | * Expect the breakpoint instruction to be the smallest size instruction for | |
194 | * the architecture. If an arch has variable length instruction and the | |
195 | * breakpoint instruction is not of the smallest length instruction | |
196 | * supported by that architecture then we need to modify read_opcode / | |
197 | * write_opcode accordingly. This would never be a problem for archs that | |
198 | * have fixed length instructions. | |
199 | */ | |
200 | ||
201 | /* | |
202 | * write_opcode - write the opcode at a given virtual address. | |
e3343e6a | 203 | * @auprobe: arch breakpointing information. |
2b144498 | 204 | * @mm: the probed process address space. |
2b144498 SD |
205 | * @vaddr: the virtual address to store the opcode. |
206 | * @opcode: opcode to be written at @vaddr. | |
207 | * | |
208 | * Called with mm->mmap_sem held (for read and with a reference to | |
209 | * mm). | |
210 | * | |
211 | * For mm @mm, write the opcode at @vaddr. | |
212 | * Return 0 (success) or a negative errno. | |
213 | */ | |
e3343e6a | 214 | static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, |
2b144498 SD |
215 | unsigned long vaddr, uprobe_opcode_t opcode) |
216 | { | |
217 | struct page *old_page, *new_page; | |
2b144498 SD |
218 | void *vaddr_old, *vaddr_new; |
219 | struct vm_area_struct *vma; | |
2b144498 | 220 | int ret; |
f403072c | 221 | |
5323ce71 | 222 | retry: |
2b144498 SD |
223 | /* Read the page with vaddr into memory */ |
224 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma); | |
225 | if (ret <= 0) | |
226 | return ret; | |
7b2d81d4 | 227 | |
2b144498 SD |
228 | ret = -ENOMEM; |
229 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
230 | if (!new_page) | |
9f92448c | 231 | goto put_old; |
2b144498 SD |
232 | |
233 | __SetPageUptodate(new_page); | |
234 | ||
2b144498 SD |
235 | /* copy the page now that we've got it stable */ |
236 | vaddr_old = kmap_atomic(old_page); | |
237 | vaddr_new = kmap_atomic(new_page); | |
238 | ||
239 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 240 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
241 | |
242 | kunmap_atomic(vaddr_new); | |
243 | kunmap_atomic(vaddr_old); | |
244 | ||
245 | ret = anon_vma_prepare(vma); | |
246 | if (ret) | |
9f92448c | 247 | goto put_new; |
2b144498 | 248 | |
c517ee74 | 249 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 250 | |
9f92448c | 251 | put_new: |
2b144498 | 252 | page_cache_release(new_page); |
9f92448c | 253 | put_old: |
7b2d81d4 IM |
254 | put_page(old_page); |
255 | ||
5323ce71 ON |
256 | if (unlikely(ret == -EAGAIN)) |
257 | goto retry; | |
2b144498 SD |
258 | return ret; |
259 | } | |
260 | ||
261 | /** | |
262 | * read_opcode - read the opcode at a given virtual address. | |
263 | * @mm: the probed process address space. | |
264 | * @vaddr: the virtual address to read the opcode. | |
265 | * @opcode: location to store the read opcode. | |
266 | * | |
267 | * Called with mm->mmap_sem held (for read and with a reference to | |
268 | * mm. | |
269 | * | |
270 | * For mm @mm, read the opcode at @vaddr and store it in @opcode. | |
271 | * Return 0 (success) or a negative errno. | |
272 | */ | |
7b2d81d4 | 273 | static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode) |
2b144498 SD |
274 | { |
275 | struct page *page; | |
276 | void *vaddr_new; | |
277 | int ret; | |
278 | ||
a3d7bb47 | 279 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); |
2b144498 SD |
280 | if (ret <= 0) |
281 | return ret; | |
282 | ||
2b144498 SD |
283 | vaddr_new = kmap_atomic(page); |
284 | vaddr &= ~PAGE_MASK; | |
5cb4ac3a | 285 | memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE); |
2b144498 | 286 | kunmap_atomic(vaddr_new); |
7b2d81d4 IM |
287 | |
288 | put_page(page); | |
289 | ||
2b144498 SD |
290 | return 0; |
291 | } | |
292 | ||
5cb4ac3a | 293 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
2b144498 SD |
294 | { |
295 | uprobe_opcode_t opcode; | |
7b2d81d4 | 296 | int result; |
2b144498 | 297 | |
c00b2750 ON |
298 | if (current->mm == mm) { |
299 | pagefault_disable(); | |
300 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
301 | sizeof(opcode)); | |
302 | pagefault_enable(); | |
303 | ||
304 | if (likely(result == 0)) | |
305 | goto out; | |
306 | } | |
307 | ||
7b2d81d4 | 308 | result = read_opcode(mm, vaddr, &opcode); |
2b144498 SD |
309 | if (result) |
310 | return result; | |
c00b2750 | 311 | out: |
5cb4ac3a | 312 | if (is_swbp_insn(&opcode)) |
2b144498 SD |
313 | return 1; |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
318 | /** | |
5cb4ac3a | 319 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 320 | * @auprobe: arch specific probepoint information. |
2b144498 | 321 | * @mm: the probed process address space. |
2b144498 SD |
322 | * @vaddr: the virtual address to insert the opcode. |
323 | * | |
324 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
325 | * Return 0 (success) or a negative errno. | |
326 | */ | |
5cb4ac3a | 327 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 328 | { |
7b2d81d4 | 329 | int result; |
c5784de2 PZ |
330 | /* |
331 | * See the comment near uprobes_hash(). | |
332 | */ | |
5cb4ac3a | 333 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 | 334 | if (result == 1) |
78f74116 | 335 | return 0; |
2b144498 SD |
336 | |
337 | if (result) | |
338 | return result; | |
339 | ||
5cb4ac3a | 340 | return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
341 | } |
342 | ||
343 | /** | |
344 | * set_orig_insn - Restore the original instruction. | |
345 | * @mm: the probed process address space. | |
e3343e6a | 346 | * @auprobe: arch specific probepoint information. |
2b144498 | 347 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
348 | * |
349 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
350 | * Return 0 (success) or a negative errno. | |
351 | */ | |
7b2d81d4 | 352 | int __weak |
ded86e7c | 353 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 354 | { |
ded86e7c ON |
355 | int result; |
356 | ||
357 | result = is_swbp_at_addr(mm, vaddr); | |
358 | if (!result) | |
359 | return -EINVAL; | |
2b144498 | 360 | |
ded86e7c ON |
361 | if (result != 1) |
362 | return result; | |
2b144498 | 363 | |
e3343e6a | 364 | return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
365 | } |
366 | ||
367 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
368 | { | |
369 | if (l->inode < r->inode) | |
370 | return -1; | |
7b2d81d4 | 371 | |
2b144498 SD |
372 | if (l->inode > r->inode) |
373 | return 1; | |
2b144498 | 374 | |
7b2d81d4 IM |
375 | if (l->offset < r->offset) |
376 | return -1; | |
377 | ||
378 | if (l->offset > r->offset) | |
379 | return 1; | |
2b144498 SD |
380 | |
381 | return 0; | |
382 | } | |
383 | ||
384 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
385 | { | |
386 | struct uprobe u = { .inode = inode, .offset = offset }; | |
387 | struct rb_node *n = uprobes_tree.rb_node; | |
388 | struct uprobe *uprobe; | |
389 | int match; | |
390 | ||
391 | while (n) { | |
392 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
393 | match = match_uprobe(&u, uprobe); | |
394 | if (!match) { | |
395 | atomic_inc(&uprobe->ref); | |
396 | return uprobe; | |
397 | } | |
7b2d81d4 | 398 | |
2b144498 SD |
399 | if (match < 0) |
400 | n = n->rb_left; | |
401 | else | |
402 | n = n->rb_right; | |
403 | } | |
404 | return NULL; | |
405 | } | |
406 | ||
407 | /* | |
408 | * Find a uprobe corresponding to a given inode:offset | |
409 | * Acquires uprobes_treelock | |
410 | */ | |
411 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
412 | { | |
413 | struct uprobe *uprobe; | |
2b144498 | 414 | |
6f47caa0 | 415 | spin_lock(&uprobes_treelock); |
2b144498 | 416 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 417 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 418 | |
2b144498 SD |
419 | return uprobe; |
420 | } | |
421 | ||
422 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
423 | { | |
424 | struct rb_node **p = &uprobes_tree.rb_node; | |
425 | struct rb_node *parent = NULL; | |
426 | struct uprobe *u; | |
427 | int match; | |
428 | ||
429 | while (*p) { | |
430 | parent = *p; | |
431 | u = rb_entry(parent, struct uprobe, rb_node); | |
432 | match = match_uprobe(uprobe, u); | |
433 | if (!match) { | |
434 | atomic_inc(&u->ref); | |
435 | return u; | |
436 | } | |
437 | ||
438 | if (match < 0) | |
439 | p = &parent->rb_left; | |
440 | else | |
441 | p = &parent->rb_right; | |
442 | ||
443 | } | |
7b2d81d4 | 444 | |
2b144498 SD |
445 | u = NULL; |
446 | rb_link_node(&uprobe->rb_node, parent, p); | |
447 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
448 | /* get access + creation ref */ | |
449 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 450 | |
2b144498 SD |
451 | return u; |
452 | } | |
453 | ||
454 | /* | |
7b2d81d4 | 455 | * Acquire uprobes_treelock. |
2b144498 SD |
456 | * Matching uprobe already exists in rbtree; |
457 | * increment (access refcount) and return the matching uprobe. | |
458 | * | |
459 | * No matching uprobe; insert the uprobe in rb_tree; | |
460 | * get a double refcount (access + creation) and return NULL. | |
461 | */ | |
462 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
463 | { | |
2b144498 SD |
464 | struct uprobe *u; |
465 | ||
6f47caa0 | 466 | spin_lock(&uprobes_treelock); |
2b144498 | 467 | u = __insert_uprobe(uprobe); |
6f47caa0 | 468 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 469 | |
0326f5a9 SD |
470 | /* For now assume that the instruction need not be single-stepped */ |
471 | uprobe->flags |= UPROBE_SKIP_SSTEP; | |
472 | ||
2b144498 SD |
473 | return u; |
474 | } | |
475 | ||
476 | static void put_uprobe(struct uprobe *uprobe) | |
477 | { | |
478 | if (atomic_dec_and_test(&uprobe->ref)) | |
479 | kfree(uprobe); | |
480 | } | |
481 | ||
482 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
483 | { | |
484 | struct uprobe *uprobe, *cur_uprobe; | |
485 | ||
486 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
487 | if (!uprobe) | |
488 | return NULL; | |
489 | ||
490 | uprobe->inode = igrab(inode); | |
491 | uprobe->offset = offset; | |
492 | init_rwsem(&uprobe->consumer_rwsem); | |
2b144498 SD |
493 | |
494 | /* add to uprobes_tree, sorted on inode:offset */ | |
495 | cur_uprobe = insert_uprobe(uprobe); | |
496 | ||
497 | /* a uprobe exists for this inode:offset combination */ | |
498 | if (cur_uprobe) { | |
499 | kfree(uprobe); | |
500 | uprobe = cur_uprobe; | |
501 | iput(inode); | |
7b2d81d4 | 502 | } else { |
2b144498 | 503 | atomic_inc(&uprobe_events); |
7b2d81d4 IM |
504 | } |
505 | ||
2b144498 SD |
506 | return uprobe; |
507 | } | |
508 | ||
0326f5a9 SD |
509 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
510 | { | |
511 | struct uprobe_consumer *uc; | |
512 | ||
513 | if (!(uprobe->flags & UPROBE_RUN_HANDLER)) | |
514 | return; | |
515 | ||
516 | down_read(&uprobe->consumer_rwsem); | |
517 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
518 | if (!uc->filter || uc->filter(uc, current)) | |
519 | uc->handler(uc, regs); | |
520 | } | |
521 | up_read(&uprobe->consumer_rwsem); | |
522 | } | |
523 | ||
2b144498 | 524 | /* Returns the previous consumer */ |
7b2d81d4 | 525 | static struct uprobe_consumer * |
e3343e6a | 526 | consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
527 | { |
528 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
529 | uc->next = uprobe->consumers; |
530 | uprobe->consumers = uc; | |
2b144498 | 531 | up_write(&uprobe->consumer_rwsem); |
7b2d81d4 | 532 | |
e3343e6a | 533 | return uc->next; |
2b144498 SD |
534 | } |
535 | ||
536 | /* | |
e3343e6a SD |
537 | * For uprobe @uprobe, delete the consumer @uc. |
538 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
539 | * or return false. |
540 | */ | |
e3343e6a | 541 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
542 | { |
543 | struct uprobe_consumer **con; | |
544 | bool ret = false; | |
545 | ||
546 | down_write(&uprobe->consumer_rwsem); | |
547 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
548 | if (*con == uc) { |
549 | *con = uc->next; | |
2b144498 SD |
550 | ret = true; |
551 | break; | |
552 | } | |
553 | } | |
554 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 555 | |
2b144498 SD |
556 | return ret; |
557 | } | |
558 | ||
e3343e6a | 559 | static int |
d436615e | 560 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 561 | unsigned long nbytes, loff_t offset) |
2b144498 | 562 | { |
2b144498 SD |
563 | struct page *page; |
564 | void *vaddr; | |
593609a5 ON |
565 | unsigned long off; |
566 | pgoff_t idx; | |
2b144498 SD |
567 | |
568 | if (!filp) | |
569 | return -EINVAL; | |
570 | ||
cc359d18 ON |
571 | if (!mapping->a_ops->readpage) |
572 | return -EIO; | |
573 | ||
593609a5 ON |
574 | idx = offset >> PAGE_CACHE_SHIFT; |
575 | off = offset & ~PAGE_MASK; | |
2b144498 SD |
576 | |
577 | /* | |
578 | * Ensure that the page that has the original instruction is | |
579 | * populated and in page-cache. | |
580 | */ | |
581 | page = read_mapping_page(mapping, idx, filp); | |
582 | if (IS_ERR(page)) | |
583 | return PTR_ERR(page); | |
584 | ||
585 | vaddr = kmap_atomic(page); | |
593609a5 | 586 | memcpy(insn, vaddr + off, nbytes); |
2b144498 SD |
587 | kunmap_atomic(vaddr); |
588 | page_cache_release(page); | |
7b2d81d4 | 589 | |
2b144498 SD |
590 | return 0; |
591 | } | |
592 | ||
d436615e | 593 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
594 | { |
595 | struct address_space *mapping; | |
2b144498 | 596 | unsigned long nbytes; |
7b2d81d4 | 597 | int bytes; |
2b144498 | 598 | |
d436615e | 599 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
600 | mapping = uprobe->inode->i_mapping; |
601 | ||
602 | /* Instruction at end of binary; copy only available bytes */ | |
603 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
604 | bytes = uprobe->inode->i_size - uprobe->offset; | |
605 | else | |
606 | bytes = MAX_UINSN_BYTES; | |
607 | ||
608 | /* Instruction at the page-boundary; copy bytes in second page */ | |
609 | if (nbytes < bytes) { | |
fc36f595 ON |
610 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
611 | bytes - nbytes, uprobe->offset + nbytes); | |
612 | if (err) | |
613 | return err; | |
2b144498 SD |
614 | bytes = nbytes; |
615 | } | |
d436615e | 616 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
617 | } |
618 | ||
682968e0 SD |
619 | /* |
620 | * How mm->uprobes_state.count gets updated | |
621 | * uprobe_mmap() increments the count if | |
622 | * - it successfully adds a breakpoint. | |
623 | * - it cannot add a breakpoint, but sees that there is a underlying | |
624 | * breakpoint (via a is_swbp_at_addr()). | |
625 | * | |
626 | * uprobe_munmap() decrements the count if | |
627 | * - it sees a underlying breakpoint, (via is_swbp_at_addr) | |
628 | * (Subsequent uprobe_unregister wouldnt find the breakpoint | |
629 | * unless a uprobe_mmap kicks in, since the old vma would be | |
630 | * dropped just after uprobe_munmap.) | |
631 | * | |
632 | * uprobe_register increments the count if: | |
633 | * - it successfully adds a breakpoint. | |
634 | * | |
635 | * uprobe_unregister decrements the count if: | |
636 | * - it sees a underlying breakpoint and removes successfully. | |
637 | * (via is_swbp_at_addr) | |
638 | * (Subsequent uprobe_munmap wouldnt find the breakpoint | |
639 | * since there is no underlying breakpoint after the | |
640 | * breakpoint removal.) | |
641 | */ | |
e3343e6a SD |
642 | static int |
643 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 644 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 645 | { |
f8ac4ec9 | 646 | bool first_uprobe; |
2b144498 SD |
647 | int ret; |
648 | ||
649 | /* | |
650 | * If probe is being deleted, unregister thread could be done with | |
651 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
652 | * nobody will be able to cleanup. Also we could be from fork or | |
653 | * mremap path, where the probe might have already been inserted. | |
654 | * Hence behave as if probe already existed. | |
655 | */ | |
656 | if (!uprobe->consumers) | |
78f74116 | 657 | return 0; |
2b144498 | 658 | |
900771a4 | 659 | if (!(uprobe->flags & UPROBE_COPY_INSN)) { |
d436615e | 660 | ret = copy_insn(uprobe, vma->vm_file); |
2b144498 SD |
661 | if (ret) |
662 | return ret; | |
663 | ||
5cb4ac3a | 664 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) |
c1914a09 | 665 | return -ENOTSUPP; |
2b144498 | 666 | |
816c03fb | 667 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
668 | if (ret) |
669 | return ret; | |
670 | ||
d9c4a30e ON |
671 | /* write_opcode() assumes we don't cross page boundary */ |
672 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
673 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
674 | ||
900771a4 | 675 | uprobe->flags |= UPROBE_COPY_INSN; |
2b144498 | 676 | } |
682968e0 | 677 | |
f8ac4ec9 ON |
678 | /* |
679 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
680 | * the task can hit this breakpoint right after __replace_page(). | |
681 | */ | |
682 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
683 | if (first_uprobe) | |
684 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
685 | ||
816c03fb | 686 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
687 | if (!ret) |
688 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
689 | else if (first_uprobe) | |
f8ac4ec9 | 690 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
691 | |
692 | return ret; | |
693 | } | |
694 | ||
e3343e6a | 695 | static void |
816c03fb | 696 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 697 | { |
9f68f672 ON |
698 | /* can happen if uprobe_register() fails */ |
699 | if (!test_bit(MMF_HAS_UPROBES, &mm->flags)) | |
700 | return; | |
701 | ||
702 | set_bit(MMF_RECALC_UPROBES, &mm->flags); | |
ded86e7c | 703 | set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
704 | } |
705 | ||
0326f5a9 | 706 | /* |
778b032d ON |
707 | * There could be threads that have already hit the breakpoint. They |
708 | * will recheck the current insn and restart if find_uprobe() fails. | |
709 | * See find_active_uprobe(). | |
0326f5a9 | 710 | */ |
2b144498 SD |
711 | static void delete_uprobe(struct uprobe *uprobe) |
712 | { | |
6f47caa0 | 713 | spin_lock(&uprobes_treelock); |
2b144498 | 714 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 715 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
716 | iput(uprobe->inode); |
717 | put_uprobe(uprobe); | |
718 | atomic_dec(&uprobe_events); | |
719 | } | |
720 | ||
26872090 ON |
721 | struct map_info { |
722 | struct map_info *next; | |
723 | struct mm_struct *mm; | |
816c03fb | 724 | unsigned long vaddr; |
26872090 ON |
725 | }; |
726 | ||
727 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 728 | { |
26872090 ON |
729 | struct map_info *next = info->next; |
730 | kfree(info); | |
731 | return next; | |
732 | } | |
733 | ||
734 | static struct map_info * | |
735 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
736 | { | |
737 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 738 | struct vm_area_struct *vma; |
26872090 ON |
739 | struct map_info *curr = NULL; |
740 | struct map_info *prev = NULL; | |
741 | struct map_info *info; | |
742 | int more = 0; | |
2b144498 | 743 | |
26872090 ON |
744 | again: |
745 | mutex_lock(&mapping->i_mmap_mutex); | |
6b2dbba8 | 746 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
747 | if (!valid_vma(vma, is_register)) |
748 | continue; | |
749 | ||
7a5bfb66 ON |
750 | if (!prev && !more) { |
751 | /* | |
752 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
753 | * reclaim. This is optimistic, no harm done if it fails. | |
754 | */ | |
755 | prev = kmalloc(sizeof(struct map_info), | |
756 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
757 | if (prev) | |
758 | prev->next = NULL; | |
759 | } | |
26872090 ON |
760 | if (!prev) { |
761 | more++; | |
762 | continue; | |
2b144498 | 763 | } |
2b144498 | 764 | |
26872090 ON |
765 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
766 | continue; | |
7b2d81d4 | 767 | |
26872090 ON |
768 | info = prev; |
769 | prev = prev->next; | |
770 | info->next = curr; | |
771 | curr = info; | |
2b144498 | 772 | |
26872090 | 773 | info->mm = vma->vm_mm; |
57683f72 | 774 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 775 | } |
2b144498 SD |
776 | mutex_unlock(&mapping->i_mmap_mutex); |
777 | ||
26872090 ON |
778 | if (!more) |
779 | goto out; | |
780 | ||
781 | prev = curr; | |
782 | while (curr) { | |
783 | mmput(curr->mm); | |
784 | curr = curr->next; | |
785 | } | |
7b2d81d4 | 786 | |
26872090 ON |
787 | do { |
788 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
789 | if (!info) { | |
790 | curr = ERR_PTR(-ENOMEM); | |
791 | goto out; | |
792 | } | |
793 | info->next = prev; | |
794 | prev = info; | |
795 | } while (--more); | |
796 | ||
797 | goto again; | |
798 | out: | |
799 | while (prev) | |
800 | prev = free_map_info(prev); | |
801 | return curr; | |
2b144498 SD |
802 | } |
803 | ||
804 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
805 | { | |
26872090 ON |
806 | struct map_info *info; |
807 | int err = 0; | |
2b144498 | 808 | |
26872090 ON |
809 | info = build_map_info(uprobe->inode->i_mapping, |
810 | uprobe->offset, is_register); | |
811 | if (IS_ERR(info)) | |
812 | return PTR_ERR(info); | |
7b2d81d4 | 813 | |
26872090 ON |
814 | while (info) { |
815 | struct mm_struct *mm = info->mm; | |
816 | struct vm_area_struct *vma; | |
7b2d81d4 | 817 | |
26872090 ON |
818 | if (err) |
819 | goto free; | |
7b2d81d4 | 820 | |
77fc4af1 | 821 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
822 | vma = find_vma(mm, info->vaddr); |
823 | if (!vma || !valid_vma(vma, is_register) || | |
824 | vma->vm_file->f_mapping->host != uprobe->inode) | |
26872090 ON |
825 | goto unlock; |
826 | ||
f4d6dfe5 ON |
827 | if (vma->vm_start > info->vaddr || |
828 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 829 | goto unlock; |
2b144498 | 830 | |
78f74116 | 831 | if (is_register) |
26872090 | 832 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
78f74116 | 833 | else |
26872090 | 834 | remove_breakpoint(uprobe, mm, info->vaddr); |
78f74116 | 835 | |
26872090 ON |
836 | unlock: |
837 | up_write(&mm->mmap_sem); | |
838 | free: | |
839 | mmput(mm); | |
840 | info = free_map_info(info); | |
2b144498 | 841 | } |
7b2d81d4 | 842 | |
26872090 | 843 | return err; |
2b144498 SD |
844 | } |
845 | ||
7b2d81d4 | 846 | static int __uprobe_register(struct uprobe *uprobe) |
2b144498 SD |
847 | { |
848 | return register_for_each_vma(uprobe, true); | |
849 | } | |
850 | ||
7b2d81d4 | 851 | static void __uprobe_unregister(struct uprobe *uprobe) |
2b144498 SD |
852 | { |
853 | if (!register_for_each_vma(uprobe, false)) | |
854 | delete_uprobe(uprobe); | |
855 | ||
856 | /* TODO : cant unregister? schedule a worker thread */ | |
857 | } | |
858 | ||
859 | /* | |
7b2d81d4 | 860 | * uprobe_register - register a probe |
2b144498 SD |
861 | * @inode: the file in which the probe has to be placed. |
862 | * @offset: offset from the start of the file. | |
e3343e6a | 863 | * @uc: information on howto handle the probe.. |
2b144498 | 864 | * |
7b2d81d4 | 865 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
866 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
867 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 868 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 869 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 870 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
871 | * unregisters. |
872 | * | |
873 | * Return errno if it cannot successully install probes | |
874 | * else return 0 (success) | |
875 | */ | |
e3343e6a | 876 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
877 | { |
878 | struct uprobe *uprobe; | |
7b2d81d4 | 879 | int ret; |
2b144498 | 880 | |
e3343e6a | 881 | if (!inode || !uc || uc->next) |
7b2d81d4 | 882 | return -EINVAL; |
2b144498 SD |
883 | |
884 | if (offset > i_size_read(inode)) | |
7b2d81d4 | 885 | return -EINVAL; |
2b144498 SD |
886 | |
887 | ret = 0; | |
888 | mutex_lock(uprobes_hash(inode)); | |
889 | uprobe = alloc_uprobe(inode, offset); | |
7b2d81d4 | 890 | |
e3343e6a | 891 | if (uprobe && !consumer_add(uprobe, uc)) { |
7b2d81d4 | 892 | ret = __uprobe_register(uprobe); |
2b144498 SD |
893 | if (ret) { |
894 | uprobe->consumers = NULL; | |
7b2d81d4 IM |
895 | __uprobe_unregister(uprobe); |
896 | } else { | |
900771a4 | 897 | uprobe->flags |= UPROBE_RUN_HANDLER; |
7b2d81d4 | 898 | } |
2b144498 SD |
899 | } |
900 | ||
901 | mutex_unlock(uprobes_hash(inode)); | |
6d1d8dfa SAS |
902 | if (uprobe) |
903 | put_uprobe(uprobe); | |
2b144498 SD |
904 | |
905 | return ret; | |
906 | } | |
907 | ||
908 | /* | |
7b2d81d4 | 909 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
910 | * @inode: the file in which the probe has to be removed. |
911 | * @offset: offset from the start of the file. | |
e3343e6a | 912 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 913 | */ |
e3343e6a | 914 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 915 | { |
7b2d81d4 | 916 | struct uprobe *uprobe; |
2b144498 | 917 | |
e3343e6a | 918 | if (!inode || !uc) |
2b144498 SD |
919 | return; |
920 | ||
921 | uprobe = find_uprobe(inode, offset); | |
922 | if (!uprobe) | |
923 | return; | |
924 | ||
925 | mutex_lock(uprobes_hash(inode)); | |
2b144498 | 926 | |
e3343e6a | 927 | if (consumer_del(uprobe, uc)) { |
7b2d81d4 IM |
928 | if (!uprobe->consumers) { |
929 | __uprobe_unregister(uprobe); | |
900771a4 | 930 | uprobe->flags &= ~UPROBE_RUN_HANDLER; |
7b2d81d4 | 931 | } |
2b144498 SD |
932 | } |
933 | ||
2b144498 SD |
934 | mutex_unlock(uprobes_hash(inode)); |
935 | if (uprobe) | |
936 | put_uprobe(uprobe); | |
937 | } | |
938 | ||
891c3970 ON |
939 | static struct rb_node * |
940 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 941 | { |
2b144498 | 942 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
943 | |
944 | while (n) { | |
891c3970 | 945 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 946 | |
891c3970 | 947 | if (inode < u->inode) { |
2b144498 | 948 | n = n->rb_left; |
891c3970 | 949 | } else if (inode > u->inode) { |
2b144498 | 950 | n = n->rb_right; |
891c3970 ON |
951 | } else { |
952 | if (max < u->offset) | |
953 | n = n->rb_left; | |
954 | else if (min > u->offset) | |
955 | n = n->rb_right; | |
956 | else | |
957 | break; | |
958 | } | |
2b144498 | 959 | } |
7b2d81d4 | 960 | |
891c3970 | 961 | return n; |
2b144498 SD |
962 | } |
963 | ||
964 | /* | |
891c3970 | 965 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 966 | */ |
891c3970 ON |
967 | static void build_probe_list(struct inode *inode, |
968 | struct vm_area_struct *vma, | |
969 | unsigned long start, unsigned long end, | |
970 | struct list_head *head) | |
2b144498 | 971 | { |
891c3970 | 972 | loff_t min, max; |
891c3970 ON |
973 | struct rb_node *n, *t; |
974 | struct uprobe *u; | |
7b2d81d4 | 975 | |
891c3970 | 976 | INIT_LIST_HEAD(head); |
cb113b47 | 977 | min = vaddr_to_offset(vma, start); |
891c3970 | 978 | max = min + (end - start) - 1; |
2b144498 | 979 | |
6f47caa0 | 980 | spin_lock(&uprobes_treelock); |
891c3970 ON |
981 | n = find_node_in_range(inode, min, max); |
982 | if (n) { | |
983 | for (t = n; t; t = rb_prev(t)) { | |
984 | u = rb_entry(t, struct uprobe, rb_node); | |
985 | if (u->inode != inode || u->offset < min) | |
986 | break; | |
987 | list_add(&u->pending_list, head); | |
988 | atomic_inc(&u->ref); | |
989 | } | |
990 | for (t = n; (t = rb_next(t)); ) { | |
991 | u = rb_entry(t, struct uprobe, rb_node); | |
992 | if (u->inode != inode || u->offset > max) | |
993 | break; | |
994 | list_add(&u->pending_list, head); | |
995 | atomic_inc(&u->ref); | |
996 | } | |
2b144498 | 997 | } |
6f47caa0 | 998 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
999 | } |
1000 | ||
1001 | /* | |
5e5be71a | 1002 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 1003 | * |
5e5be71a ON |
1004 | * Currently we ignore all errors and always return 0, the callers |
1005 | * can't handle the failure anyway. | |
2b144498 | 1006 | */ |
7b2d81d4 | 1007 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1008 | { |
1009 | struct list_head tmp_list; | |
665605a2 | 1010 | struct uprobe *uprobe, *u; |
2b144498 | 1011 | struct inode *inode; |
2b144498 SD |
1012 | |
1013 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) | |
7b2d81d4 | 1014 | return 0; |
2b144498 SD |
1015 | |
1016 | inode = vma->vm_file->f_mapping->host; | |
1017 | if (!inode) | |
7b2d81d4 | 1018 | return 0; |
2b144498 | 1019 | |
2b144498 | 1020 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1021 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
7b2d81d4 | 1022 | |
665605a2 | 1023 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
5e5be71a | 1024 | if (!fatal_signal_pending(current)) { |
57683f72 | 1025 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1026 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1027 | } |
1028 | put_uprobe(uprobe); | |
1029 | } | |
2b144498 SD |
1030 | mutex_unlock(uprobes_mmap_hash(inode)); |
1031 | ||
5e5be71a | 1032 | return 0; |
2b144498 SD |
1033 | } |
1034 | ||
9f68f672 ON |
1035 | static bool |
1036 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1037 | { | |
1038 | loff_t min, max; | |
1039 | struct inode *inode; | |
1040 | struct rb_node *n; | |
1041 | ||
1042 | inode = vma->vm_file->f_mapping->host; | |
1043 | ||
1044 | min = vaddr_to_offset(vma, start); | |
1045 | max = min + (end - start) - 1; | |
1046 | ||
1047 | spin_lock(&uprobes_treelock); | |
1048 | n = find_node_in_range(inode, min, max); | |
1049 | spin_unlock(&uprobes_treelock); | |
1050 | ||
1051 | return !!n; | |
1052 | } | |
1053 | ||
682968e0 SD |
1054 | /* |
1055 | * Called in context of a munmap of a vma. | |
1056 | */ | |
cbc91f71 | 1057 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1058 | { |
682968e0 SD |
1059 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) |
1060 | return; | |
1061 | ||
2fd611a9 ON |
1062 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1063 | return; | |
1064 | ||
9f68f672 ON |
1065 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1066 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1067 | return; |
1068 | ||
9f68f672 ON |
1069 | if (vma_has_uprobes(vma, start, end)) |
1070 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1071 | } |
1072 | ||
d4b3b638 SD |
1073 | /* Slot allocation for XOL */ |
1074 | static int xol_add_vma(struct xol_area *area) | |
1075 | { | |
1076 | struct mm_struct *mm; | |
1077 | int ret; | |
1078 | ||
1079 | area->page = alloc_page(GFP_HIGHUSER); | |
1080 | if (!area->page) | |
1081 | return -ENOMEM; | |
1082 | ||
1083 | ret = -EALREADY; | |
1084 | mm = current->mm; | |
1085 | ||
1086 | down_write(&mm->mmap_sem); | |
1087 | if (mm->uprobes_state.xol_area) | |
1088 | goto fail; | |
1089 | ||
1090 | ret = -ENOMEM; | |
1091 | ||
1092 | /* Try to map as high as possible, this is only a hint. */ | |
1093 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1094 | if (area->vaddr & ~PAGE_MASK) { | |
1095 | ret = area->vaddr; | |
1096 | goto fail; | |
1097 | } | |
1098 | ||
1099 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1100 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1101 | if (ret) | |
1102 | goto fail; | |
1103 | ||
1104 | smp_wmb(); /* pairs with get_xol_area() */ | |
1105 | mm->uprobes_state.xol_area = area; | |
1106 | ret = 0; | |
1107 | ||
1108 | fail: | |
1109 | up_write(&mm->mmap_sem); | |
1110 | if (ret) | |
1111 | __free_page(area->page); | |
1112 | ||
1113 | return ret; | |
1114 | } | |
1115 | ||
1116 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1117 | { | |
1118 | struct xol_area *area; | |
1119 | ||
1120 | area = mm->uprobes_state.xol_area; | |
1121 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1122 | ||
1123 | return area; | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * xol_alloc_area - Allocate process's xol_area. | |
1128 | * This area will be used for storing instructions for execution out of | |
1129 | * line. | |
1130 | * | |
1131 | * Returns the allocated area or NULL. | |
1132 | */ | |
1133 | static struct xol_area *xol_alloc_area(void) | |
1134 | { | |
1135 | struct xol_area *area; | |
1136 | ||
1137 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1138 | if (unlikely(!area)) | |
1139 | return NULL; | |
1140 | ||
1141 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1142 | ||
1143 | if (!area->bitmap) | |
1144 | goto fail; | |
1145 | ||
1146 | init_waitqueue_head(&area->wq); | |
1147 | if (!xol_add_vma(area)) | |
1148 | return area; | |
1149 | ||
1150 | fail: | |
1151 | kfree(area->bitmap); | |
1152 | kfree(area); | |
1153 | ||
1154 | return get_xol_area(current->mm); | |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * uprobe_clear_state - Free the area allocated for slots. | |
1159 | */ | |
1160 | void uprobe_clear_state(struct mm_struct *mm) | |
1161 | { | |
1162 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1163 | ||
1164 | if (!area) | |
1165 | return; | |
1166 | ||
1167 | put_page(area->page); | |
1168 | kfree(area->bitmap); | |
1169 | kfree(area); | |
1170 | } | |
1171 | ||
f8ac4ec9 ON |
1172 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1173 | { | |
61559a81 ON |
1174 | newmm->uprobes_state.xol_area = NULL; |
1175 | ||
9f68f672 | 1176 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1177 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1178 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1179 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1180 | } | |
f8ac4ec9 ON |
1181 | } |
1182 | ||
d4b3b638 SD |
1183 | /* |
1184 | * - search for a free slot. | |
1185 | */ | |
1186 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1187 | { | |
1188 | unsigned long slot_addr; | |
1189 | int slot_nr; | |
1190 | ||
1191 | do { | |
1192 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1193 | if (slot_nr < UINSNS_PER_PAGE) { | |
1194 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1195 | break; | |
1196 | ||
1197 | slot_nr = UINSNS_PER_PAGE; | |
1198 | continue; | |
1199 | } | |
1200 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1201 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1202 | ||
1203 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1204 | atomic_inc(&area->slot_count); | |
1205 | ||
1206 | return slot_addr; | |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * xol_get_insn_slot - If was not allocated a slot, then | |
1211 | * allocate a slot. | |
1212 | * Returns the allocated slot address or 0. | |
1213 | */ | |
1214 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1215 | { | |
1216 | struct xol_area *area; | |
1217 | unsigned long offset; | |
1218 | void *vaddr; | |
1219 | ||
1220 | area = get_xol_area(current->mm); | |
1221 | if (!area) { | |
1222 | area = xol_alloc_area(); | |
1223 | if (!area) | |
1224 | return 0; | |
1225 | } | |
1226 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1227 | ||
1228 | /* | |
1229 | * Initialize the slot if xol_vaddr points to valid | |
1230 | * instruction slot. | |
1231 | */ | |
1232 | if (unlikely(!current->utask->xol_vaddr)) | |
1233 | return 0; | |
1234 | ||
1235 | current->utask->vaddr = slot_addr; | |
1236 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1237 | vaddr = kmap_atomic(area->page); | |
1238 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1239 | kunmap_atomic(vaddr); | |
1240 | ||
1241 | return current->utask->xol_vaddr; | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * xol_free_insn_slot - If slot was earlier allocated by | |
1246 | * @xol_get_insn_slot(), make the slot available for | |
1247 | * subsequent requests. | |
1248 | */ | |
1249 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1250 | { | |
1251 | struct xol_area *area; | |
1252 | unsigned long vma_end; | |
1253 | unsigned long slot_addr; | |
1254 | ||
1255 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1256 | return; | |
1257 | ||
1258 | slot_addr = tsk->utask->xol_vaddr; | |
1259 | ||
1260 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1261 | return; | |
1262 | ||
1263 | area = tsk->mm->uprobes_state.xol_area; | |
1264 | vma_end = area->vaddr + PAGE_SIZE; | |
1265 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1266 | unsigned long offset; | |
1267 | int slot_nr; | |
1268 | ||
1269 | offset = slot_addr - area->vaddr; | |
1270 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1271 | if (slot_nr >= UINSNS_PER_PAGE) | |
1272 | return; | |
1273 | ||
1274 | clear_bit(slot_nr, area->bitmap); | |
1275 | atomic_dec(&area->slot_count); | |
1276 | if (waitqueue_active(&area->wq)) | |
1277 | wake_up(&area->wq); | |
1278 | ||
1279 | tsk->utask->xol_vaddr = 0; | |
1280 | } | |
1281 | } | |
1282 | ||
0326f5a9 SD |
1283 | /** |
1284 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1285 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1286 | * instruction. | |
1287 | * Return the address of the breakpoint instruction. | |
1288 | */ | |
1289 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1290 | { | |
1291 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Called with no locks held. | |
1296 | * Called in context of a exiting or a exec-ing thread. | |
1297 | */ | |
1298 | void uprobe_free_utask(struct task_struct *t) | |
1299 | { | |
1300 | struct uprobe_task *utask = t->utask; | |
1301 | ||
0326f5a9 SD |
1302 | if (!utask) |
1303 | return; | |
1304 | ||
1305 | if (utask->active_uprobe) | |
1306 | put_uprobe(utask->active_uprobe); | |
1307 | ||
d4b3b638 | 1308 | xol_free_insn_slot(t); |
0326f5a9 SD |
1309 | kfree(utask); |
1310 | t->utask = NULL; | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * Called in context of a new clone/fork from copy_process. | |
1315 | */ | |
1316 | void uprobe_copy_process(struct task_struct *t) | |
1317 | { | |
1318 | t->utask = NULL; | |
0326f5a9 SD |
1319 | } |
1320 | ||
1321 | /* | |
1322 | * Allocate a uprobe_task object for the task. | |
1323 | * Called when the thread hits a breakpoint for the first time. | |
1324 | * | |
1325 | * Returns: | |
1326 | * - pointer to new uprobe_task on success | |
1327 | * - NULL otherwise | |
1328 | */ | |
1329 | static struct uprobe_task *add_utask(void) | |
1330 | { | |
1331 | struct uprobe_task *utask; | |
1332 | ||
1333 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1334 | if (unlikely(!utask)) | |
1335 | return NULL; | |
1336 | ||
0326f5a9 SD |
1337 | current->utask = utask; |
1338 | return utask; | |
1339 | } | |
1340 | ||
1341 | /* Prepare to single-step probed instruction out of line. */ | |
1342 | static int | |
1343 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1344 | { | |
d4b3b638 SD |
1345 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1346 | return 0; | |
1347 | ||
0326f5a9 SD |
1348 | return -EFAULT; |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * If we are singlestepping, then ensure this thread is not connected to | |
1353 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1354 | * triggers the signal, restart the original insn even if the task is | |
1355 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1356 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1357 | * _same_ instruction should be repeated again after return from the signal | |
1358 | * handler, and SSTEP can never finish in this case. | |
1359 | */ | |
1360 | bool uprobe_deny_signal(void) | |
1361 | { | |
1362 | struct task_struct *t = current; | |
1363 | struct uprobe_task *utask = t->utask; | |
1364 | ||
1365 | if (likely(!utask || !utask->active_uprobe)) | |
1366 | return false; | |
1367 | ||
1368 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1369 | ||
1370 | if (signal_pending(t)) { | |
1371 | spin_lock_irq(&t->sighand->siglock); | |
1372 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1373 | spin_unlock_irq(&t->sighand->siglock); | |
1374 | ||
1375 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1376 | utask->state = UTASK_SSTEP_TRAPPED; | |
1377 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1378 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1379 | } | |
1380 | } | |
1381 | ||
1382 | return true; | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * Avoid singlestepping the original instruction if the original instruction | |
1387 | * is a NOP or can be emulated. | |
1388 | */ | |
1389 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1390 | { | |
1391 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) | |
1392 | return true; | |
1393 | ||
1394 | uprobe->flags &= ~UPROBE_SKIP_SSTEP; | |
1395 | return false; | |
1396 | } | |
1397 | ||
499a4f3e ON |
1398 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1399 | { | |
1400 | struct vm_area_struct *vma; | |
1401 | ||
1402 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1403 | if (!valid_vma(vma, false)) | |
1404 | continue; | |
1405 | /* | |
1406 | * This is not strictly accurate, we can race with | |
1407 | * uprobe_unregister() and see the already removed | |
1408 | * uprobe if delete_uprobe() was not yet called. | |
1409 | */ | |
1410 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1411 | return; | |
1412 | } | |
1413 | ||
1414 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1415 | } | |
1416 | ||
d790d346 | 1417 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1418 | { |
3a9ea052 ON |
1419 | struct mm_struct *mm = current->mm; |
1420 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1421 | struct vm_area_struct *vma; |
0326f5a9 | 1422 | |
0326f5a9 SD |
1423 | down_read(&mm->mmap_sem); |
1424 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1425 | if (vma && vma->vm_start <= bp_vaddr) { |
1426 | if (valid_vma(vma, false)) { | |
cb113b47 ON |
1427 | struct inode *inode = vma->vm_file->f_mapping->host; |
1428 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); | |
0326f5a9 | 1429 | |
3a9ea052 ON |
1430 | uprobe = find_uprobe(inode, offset); |
1431 | } | |
d790d346 ON |
1432 | |
1433 | if (!uprobe) | |
1434 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1435 | } else { | |
1436 | *is_swbp = -EFAULT; | |
0326f5a9 | 1437 | } |
499a4f3e ON |
1438 | |
1439 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1440 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1441 | up_read(&mm->mmap_sem); |
1442 | ||
3a9ea052 ON |
1443 | return uprobe; |
1444 | } | |
1445 | ||
9d778782 SAS |
1446 | void __weak arch_uprobe_enable_step(struct arch_uprobe *arch) |
1447 | { | |
1448 | user_enable_single_step(current); | |
1449 | } | |
1450 | ||
1451 | void __weak arch_uprobe_disable_step(struct arch_uprobe *arch) | |
1452 | { | |
1453 | user_disable_single_step(current); | |
1454 | } | |
1455 | ||
3a9ea052 ON |
1456 | /* |
1457 | * Run handler and ask thread to singlestep. | |
1458 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1459 | */ | |
1460 | static void handle_swbp(struct pt_regs *regs) | |
1461 | { | |
1462 | struct uprobe_task *utask; | |
1463 | struct uprobe *uprobe; | |
1464 | unsigned long bp_vaddr; | |
56bb4cf6 | 1465 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1466 | |
1467 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1468 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1469 | |
0326f5a9 | 1470 | if (!uprobe) { |
56bb4cf6 ON |
1471 | if (is_swbp > 0) { |
1472 | /* No matching uprobe; signal SIGTRAP. */ | |
1473 | send_sig(SIGTRAP, current, 0); | |
1474 | } else { | |
1475 | /* | |
1476 | * Either we raced with uprobe_unregister() or we can't | |
1477 | * access this memory. The latter is only possible if | |
1478 | * another thread plays with our ->mm. In both cases | |
1479 | * we can simply restart. If this vma was unmapped we | |
1480 | * can pretend this insn was not executed yet and get | |
1481 | * the (correct) SIGSEGV after restart. | |
1482 | */ | |
1483 | instruction_pointer_set(regs, bp_vaddr); | |
1484 | } | |
0326f5a9 SD |
1485 | return; |
1486 | } | |
1487 | ||
1488 | utask = current->utask; | |
1489 | if (!utask) { | |
1490 | utask = add_utask(); | |
1491 | /* Cannot allocate; re-execute the instruction. */ | |
1492 | if (!utask) | |
1493 | goto cleanup_ret; | |
1494 | } | |
1495 | utask->active_uprobe = uprobe; | |
1496 | handler_chain(uprobe, regs); | |
1497 | if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs)) | |
1498 | goto cleanup_ret; | |
1499 | ||
1500 | utask->state = UTASK_SSTEP; | |
1501 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { | |
9d778782 | 1502 | arch_uprobe_enable_step(&uprobe->arch); |
0326f5a9 SD |
1503 | return; |
1504 | } | |
1505 | ||
1506 | cleanup_ret: | |
1507 | if (utask) { | |
1508 | utask->active_uprobe = NULL; | |
1509 | utask->state = UTASK_RUNNING; | |
1510 | } | |
8bd87445 | 1511 | if (!(uprobe->flags & UPROBE_SKIP_SSTEP)) |
0326f5a9 | 1512 | |
8bd87445 SAS |
1513 | /* |
1514 | * cannot singlestep; cannot skip instruction; | |
1515 | * re-execute the instruction. | |
1516 | */ | |
1517 | instruction_pointer_set(regs, bp_vaddr); | |
0326f5a9 | 1518 | |
8bd87445 | 1519 | put_uprobe(uprobe); |
0326f5a9 SD |
1520 | } |
1521 | ||
1522 | /* | |
1523 | * Perform required fix-ups and disable singlestep. | |
1524 | * Allow pending signals to take effect. | |
1525 | */ | |
1526 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1527 | { | |
1528 | struct uprobe *uprobe; | |
1529 | ||
1530 | uprobe = utask->active_uprobe; | |
1531 | if (utask->state == UTASK_SSTEP_ACK) | |
1532 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1533 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1534 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1535 | else | |
1536 | WARN_ON_ONCE(1); | |
1537 | ||
9d778782 | 1538 | arch_uprobe_disable_step(&uprobe->arch); |
0326f5a9 SD |
1539 | put_uprobe(uprobe); |
1540 | utask->active_uprobe = NULL; | |
1541 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1542 | xol_free_insn_slot(current); |
0326f5a9 SD |
1543 | |
1544 | spin_lock_irq(¤t->sighand->siglock); | |
1545 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1546 | spin_unlock_irq(¤t->sighand->siglock); | |
1547 | } | |
1548 | ||
1549 | /* | |
1550 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on | |
1551 | * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and | |
1552 | * allows the thread to return from interrupt. | |
1553 | * | |
1554 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and | |
1555 | * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from | |
1556 | * interrupt. | |
1557 | * | |
1558 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1559 | * uprobe_notify_resume(). | |
1560 | */ | |
1561 | void uprobe_notify_resume(struct pt_regs *regs) | |
1562 | { | |
1563 | struct uprobe_task *utask; | |
1564 | ||
1565 | utask = current->utask; | |
1566 | if (!utask || utask->state == UTASK_BP_HIT) | |
1567 | handle_swbp(regs); | |
1568 | else | |
1569 | handle_singlestep(utask, regs); | |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1574 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1575 | */ | |
1576 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1577 | { | |
1578 | struct uprobe_task *utask; | |
1579 | ||
f8ac4ec9 | 1580 | if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) |
0326f5a9 SD |
1581 | return 0; |
1582 | ||
1583 | utask = current->utask; | |
1584 | if (utask) | |
1585 | utask->state = UTASK_BP_HIT; | |
1586 | ||
1587 | set_thread_flag(TIF_UPROBE); | |
0326f5a9 SD |
1588 | |
1589 | return 1; | |
1590 | } | |
1591 | ||
1592 | /* | |
1593 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1594 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1595 | */ | |
1596 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1597 | { | |
1598 | struct uprobe_task *utask = current->utask; | |
1599 | ||
1600 | if (!current->mm || !utask || !utask->active_uprobe) | |
1601 | /* task is currently not uprobed */ | |
1602 | return 0; | |
1603 | ||
1604 | utask->state = UTASK_SSTEP_ACK; | |
1605 | set_thread_flag(TIF_UPROBE); | |
1606 | return 1; | |
1607 | } | |
1608 | ||
1609 | static struct notifier_block uprobe_exception_nb = { | |
1610 | .notifier_call = arch_uprobe_exception_notify, | |
1611 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1612 | }; | |
1613 | ||
2b144498 SD |
1614 | static int __init init_uprobes(void) |
1615 | { | |
1616 | int i; | |
1617 | ||
1618 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1619 | mutex_init(&uprobes_mutex[i]); | |
1620 | mutex_init(&uprobes_mmap_mutex[i]); | |
1621 | } | |
0326f5a9 SD |
1622 | |
1623 | return register_die_notifier(&uprobe_exception_nb); | |
2b144498 | 1624 | } |
0326f5a9 | 1625 | module_init(init_uprobes); |
2b144498 SD |
1626 | |
1627 | static void __exit exit_uprobes(void) | |
1628 | { | |
1629 | } | |
2b144498 | 1630 | module_exit(exit_uprobes); |