]>
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 */ | |
7b2d81d4 | 35 | |
2b144498 SD |
36 | #include <linux/uprobes.h> |
37 | ||
d4b3b638 SD |
38 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
39 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
40 | ||
2b144498 | 41 | static struct rb_root uprobes_tree = RB_ROOT; |
7b2d81d4 | 42 | |
2b144498 SD |
43 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
44 | ||
45 | #define UPROBES_HASH_SZ 13 | |
7b2d81d4 | 46 | |
c5784de2 PZ |
47 | /* |
48 | * We need separate register/unregister and mmap/munmap lock hashes because | |
49 | * of mmap_sem nesting. | |
50 | * | |
51 | * uprobe_register() needs to install probes on (potentially) all processes | |
52 | * and thus needs to acquire multiple mmap_sems (consequtively, not | |
53 | * concurrently), whereas uprobe_mmap() is called while holding mmap_sem | |
54 | * for the particular process doing the mmap. | |
55 | * | |
56 | * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem | |
57 | * because of lock order against i_mmap_mutex. This means there's a hole in | |
58 | * the register vma iteration where a mmap() can happen. | |
59 | * | |
60 | * Thus uprobe_register() can race with uprobe_mmap() and we can try and | |
61 | * install a probe where one is already installed. | |
62 | */ | |
63 | ||
2b144498 SD |
64 | /* serialize (un)register */ |
65 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
66 | |
67 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
68 | |
69 | /* serialize uprobe->pending_list */ | |
70 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 71 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 SD |
72 | |
73 | /* | |
7b2d81d4 | 74 | * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe |
2b144498 SD |
75 | * events active at this time. Probably a fine grained per inode count is |
76 | * better? | |
77 | */ | |
78 | static atomic_t uprobe_events = ATOMIC_INIT(0); | |
79 | ||
3ff54efd SD |
80 | struct uprobe { |
81 | struct rb_node rb_node; /* node in the rb tree */ | |
82 | atomic_t ref; | |
83 | struct rw_semaphore consumer_rwsem; | |
84 | struct list_head pending_list; | |
85 | struct uprobe_consumer *consumers; | |
86 | struct inode *inode; /* Also hold a ref to inode */ | |
87 | loff_t offset; | |
88 | int flags; | |
89 | struct arch_uprobe arch; | |
90 | }; | |
91 | ||
2b144498 SD |
92 | /* |
93 | * valid_vma: Verify if the specified vma is an executable vma | |
94 | * Relax restrictions while unregistering: vm_flags might have | |
95 | * changed after breakpoint was inserted. | |
96 | * - is_register: indicates if we are in register context. | |
97 | * - Return 1 if the specified virtual address is in an | |
98 | * executable vma. | |
99 | */ | |
100 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
101 | { | |
102 | if (!vma->vm_file) | |
103 | return false; | |
104 | ||
105 | if (!is_register) | |
106 | return true; | |
107 | ||
ea131377 ON |
108 | if ((vma->vm_flags & (VM_HUGETLB|VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)) |
109 | == (VM_READ|VM_EXEC)) | |
2b144498 SD |
110 | return true; |
111 | ||
112 | return false; | |
113 | } | |
114 | ||
115 | static loff_t vma_address(struct vm_area_struct *vma, loff_t offset) | |
116 | { | |
117 | loff_t vaddr; | |
118 | ||
119 | vaddr = vma->vm_start + offset; | |
120 | vaddr -= vma->vm_pgoff << PAGE_SHIFT; | |
7b2d81d4 | 121 | |
2b144498 SD |
122 | return vaddr; |
123 | } | |
124 | ||
125 | /** | |
126 | * __replace_page - replace page in vma by new page. | |
127 | * based on replace_page in mm/ksm.c | |
128 | * | |
129 | * @vma: vma that holds the pte pointing to page | |
130 | * @page: the cowed page we are replacing by kpage | |
131 | * @kpage: the modified page we replace page by | |
132 | * | |
133 | * Returns 0 on success, -EFAULT on failure. | |
134 | */ | |
7b2d81d4 | 135 | static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage) |
2b144498 SD |
136 | { |
137 | struct mm_struct *mm = vma->vm_mm; | |
2b144498 | 138 | unsigned long addr; |
5323ce71 ON |
139 | spinlock_t *ptl; |
140 | pte_t *ptep; | |
2b144498 SD |
141 | |
142 | addr = page_address_in_vma(page, vma); | |
143 | if (addr == -EFAULT) | |
5323ce71 | 144 | return -EFAULT; |
2b144498 | 145 | |
5323ce71 | 146 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 147 | if (!ptep) |
5323ce71 | 148 | return -EAGAIN; |
2b144498 SD |
149 | |
150 | get_page(kpage); | |
151 | page_add_new_anon_rmap(kpage, vma, addr); | |
152 | ||
7396fa81 SD |
153 | if (!PageAnon(page)) { |
154 | dec_mm_counter(mm, MM_FILEPAGES); | |
155 | inc_mm_counter(mm, MM_ANONPAGES); | |
156 | } | |
157 | ||
2b144498 SD |
158 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
159 | ptep_clear_flush(vma, addr, ptep); | |
160 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
161 | ||
162 | page_remove_rmap(page); | |
163 | if (!page_mapped(page)) | |
164 | try_to_free_swap(page); | |
165 | put_page(page); | |
166 | pte_unmap_unlock(ptep, ptl); | |
2b144498 | 167 | |
5323ce71 | 168 | return 0; |
2b144498 SD |
169 | } |
170 | ||
171 | /** | |
5cb4ac3a | 172 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 173 | * @insn: instruction to be checked. |
5cb4ac3a | 174 | * Default implementation of is_swbp_insn |
2b144498 SD |
175 | * Returns true if @insn is a breakpoint instruction. |
176 | */ | |
5cb4ac3a | 177 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 178 | { |
5cb4ac3a | 179 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
180 | } |
181 | ||
182 | /* | |
183 | * NOTE: | |
184 | * Expect the breakpoint instruction to be the smallest size instruction for | |
185 | * the architecture. If an arch has variable length instruction and the | |
186 | * breakpoint instruction is not of the smallest length instruction | |
187 | * supported by that architecture then we need to modify read_opcode / | |
188 | * write_opcode accordingly. This would never be a problem for archs that | |
189 | * have fixed length instructions. | |
190 | */ | |
191 | ||
192 | /* | |
193 | * write_opcode - write the opcode at a given virtual address. | |
e3343e6a | 194 | * @auprobe: arch breakpointing information. |
2b144498 | 195 | * @mm: the probed process address space. |
2b144498 SD |
196 | * @vaddr: the virtual address to store the opcode. |
197 | * @opcode: opcode to be written at @vaddr. | |
198 | * | |
199 | * Called with mm->mmap_sem held (for read and with a reference to | |
200 | * mm). | |
201 | * | |
202 | * For mm @mm, write the opcode at @vaddr. | |
203 | * Return 0 (success) or a negative errno. | |
204 | */ | |
e3343e6a | 205 | static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, |
2b144498 SD |
206 | unsigned long vaddr, uprobe_opcode_t opcode) |
207 | { | |
208 | struct page *old_page, *new_page; | |
209 | struct address_space *mapping; | |
210 | void *vaddr_old, *vaddr_new; | |
211 | struct vm_area_struct *vma; | |
3ff54efd | 212 | struct uprobe *uprobe; |
5323ce71 | 213 | unsigned long pgoff; |
2b144498 SD |
214 | loff_t addr; |
215 | int ret; | |
5323ce71 | 216 | retry: |
2b144498 SD |
217 | /* Read the page with vaddr into memory */ |
218 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma); | |
219 | if (ret <= 0) | |
220 | return ret; | |
7b2d81d4 | 221 | |
2b144498 SD |
222 | ret = -EINVAL; |
223 | ||
224 | /* | |
225 | * We are interested in text pages only. Our pages of interest | |
226 | * should be mapped for read and execute only. We desist from | |
227 | * adding probes in write mapped pages since the breakpoints | |
228 | * might end up in the file copy. | |
229 | */ | |
5cb4ac3a | 230 | if (!valid_vma(vma, is_swbp_insn(&opcode))) |
2b144498 SD |
231 | goto put_out; |
232 | ||
3ff54efd | 233 | uprobe = container_of(auprobe, struct uprobe, arch); |
2b144498 SD |
234 | mapping = uprobe->inode->i_mapping; |
235 | if (mapping != vma->vm_file->f_mapping) | |
236 | goto put_out; | |
237 | ||
238 | addr = vma_address(vma, uprobe->offset); | |
239 | if (vaddr != (unsigned long)addr) | |
240 | goto put_out; | |
241 | ||
242 | ret = -ENOMEM; | |
243 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
244 | if (!new_page) | |
245 | goto put_out; | |
246 | ||
247 | __SetPageUptodate(new_page); | |
248 | ||
249 | /* | |
250 | * lock page will serialize against do_wp_page()'s | |
251 | * PageAnon() handling | |
252 | */ | |
253 | lock_page(old_page); | |
254 | /* copy the page now that we've got it stable */ | |
255 | vaddr_old = kmap_atomic(old_page); | |
256 | vaddr_new = kmap_atomic(new_page); | |
257 | ||
258 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
7b2d81d4 | 259 | |
2b144498 | 260 | /* poke the new insn in, ASSUMES we don't cross page boundary */ |
5323ce71 ON |
261 | pgoff = (vaddr & ~PAGE_MASK); |
262 | BUG_ON(pgoff + UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
263 | memcpy(vaddr_new + pgoff, &opcode, UPROBE_SWBP_INSN_SIZE); | |
2b144498 SD |
264 | |
265 | kunmap_atomic(vaddr_new); | |
266 | kunmap_atomic(vaddr_old); | |
267 | ||
268 | ret = anon_vma_prepare(vma); | |
269 | if (ret) | |
270 | goto unlock_out; | |
271 | ||
272 | lock_page(new_page); | |
273 | ret = __replace_page(vma, old_page, new_page); | |
274 | unlock_page(new_page); | |
275 | ||
276 | unlock_out: | |
277 | unlock_page(old_page); | |
278 | page_cache_release(new_page); | |
279 | ||
280 | put_out: | |
7b2d81d4 IM |
281 | put_page(old_page); |
282 | ||
5323ce71 ON |
283 | if (unlikely(ret == -EAGAIN)) |
284 | goto retry; | |
2b144498 SD |
285 | return ret; |
286 | } | |
287 | ||
288 | /** | |
289 | * read_opcode - read the opcode at a given virtual address. | |
290 | * @mm: the probed process address space. | |
291 | * @vaddr: the virtual address to read the opcode. | |
292 | * @opcode: location to store the read opcode. | |
293 | * | |
294 | * Called with mm->mmap_sem held (for read and with a reference to | |
295 | * mm. | |
296 | * | |
297 | * For mm @mm, read the opcode at @vaddr and store it in @opcode. | |
298 | * Return 0 (success) or a negative errno. | |
299 | */ | |
7b2d81d4 | 300 | static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode) |
2b144498 SD |
301 | { |
302 | struct page *page; | |
303 | void *vaddr_new; | |
304 | int ret; | |
305 | ||
a3d7bb47 | 306 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); |
2b144498 SD |
307 | if (ret <= 0) |
308 | return ret; | |
309 | ||
310 | lock_page(page); | |
311 | vaddr_new = kmap_atomic(page); | |
312 | vaddr &= ~PAGE_MASK; | |
5cb4ac3a | 313 | memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
314 | kunmap_atomic(vaddr_new); |
315 | unlock_page(page); | |
7b2d81d4 IM |
316 | |
317 | put_page(page); | |
318 | ||
2b144498 SD |
319 | return 0; |
320 | } | |
321 | ||
5cb4ac3a | 322 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
2b144498 SD |
323 | { |
324 | uprobe_opcode_t opcode; | |
7b2d81d4 | 325 | int result; |
2b144498 | 326 | |
c00b2750 ON |
327 | if (current->mm == mm) { |
328 | pagefault_disable(); | |
329 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
330 | sizeof(opcode)); | |
331 | pagefault_enable(); | |
332 | ||
333 | if (likely(result == 0)) | |
334 | goto out; | |
335 | } | |
336 | ||
7b2d81d4 | 337 | result = read_opcode(mm, vaddr, &opcode); |
2b144498 SD |
338 | if (result) |
339 | return result; | |
c00b2750 | 340 | out: |
5cb4ac3a | 341 | if (is_swbp_insn(&opcode)) |
2b144498 SD |
342 | return 1; |
343 | ||
344 | return 0; | |
345 | } | |
346 | ||
347 | /** | |
5cb4ac3a | 348 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 349 | * @auprobe: arch specific probepoint information. |
2b144498 | 350 | * @mm: the probed process address space. |
2b144498 SD |
351 | * @vaddr: the virtual address to insert the opcode. |
352 | * | |
353 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
354 | * Return 0 (success) or a negative errno. | |
355 | */ | |
5cb4ac3a | 356 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 357 | { |
7b2d81d4 | 358 | int result; |
c5784de2 PZ |
359 | /* |
360 | * See the comment near uprobes_hash(). | |
361 | */ | |
5cb4ac3a | 362 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
363 | if (result == 1) |
364 | return -EEXIST; | |
365 | ||
366 | if (result) | |
367 | return result; | |
368 | ||
5cb4ac3a | 369 | return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
370 | } |
371 | ||
372 | /** | |
373 | * set_orig_insn - Restore the original instruction. | |
374 | * @mm: the probed process address space. | |
e3343e6a | 375 | * @auprobe: arch specific probepoint information. |
2b144498 SD |
376 | * @vaddr: the virtual address to insert the opcode. |
377 | * @verify: if true, verify existance of breakpoint instruction. | |
378 | * | |
379 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
380 | * Return 0 (success) or a negative errno. | |
381 | */ | |
7b2d81d4 | 382 | int __weak |
e3343e6a | 383 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify) |
2b144498 SD |
384 | { |
385 | if (verify) { | |
7b2d81d4 | 386 | int result; |
2b144498 | 387 | |
5cb4ac3a | 388 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
389 | if (!result) |
390 | return -EINVAL; | |
391 | ||
392 | if (result != 1) | |
393 | return result; | |
394 | } | |
e3343e6a | 395 | return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
396 | } |
397 | ||
398 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
399 | { | |
400 | if (l->inode < r->inode) | |
401 | return -1; | |
7b2d81d4 | 402 | |
2b144498 SD |
403 | if (l->inode > r->inode) |
404 | return 1; | |
2b144498 | 405 | |
7b2d81d4 IM |
406 | if (l->offset < r->offset) |
407 | return -1; | |
408 | ||
409 | if (l->offset > r->offset) | |
410 | return 1; | |
2b144498 SD |
411 | |
412 | return 0; | |
413 | } | |
414 | ||
415 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
416 | { | |
417 | struct uprobe u = { .inode = inode, .offset = offset }; | |
418 | struct rb_node *n = uprobes_tree.rb_node; | |
419 | struct uprobe *uprobe; | |
420 | int match; | |
421 | ||
422 | while (n) { | |
423 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
424 | match = match_uprobe(&u, uprobe); | |
425 | if (!match) { | |
426 | atomic_inc(&uprobe->ref); | |
427 | return uprobe; | |
428 | } | |
7b2d81d4 | 429 | |
2b144498 SD |
430 | if (match < 0) |
431 | n = n->rb_left; | |
432 | else | |
433 | n = n->rb_right; | |
434 | } | |
435 | return NULL; | |
436 | } | |
437 | ||
438 | /* | |
439 | * Find a uprobe corresponding to a given inode:offset | |
440 | * Acquires uprobes_treelock | |
441 | */ | |
442 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
443 | { | |
444 | struct uprobe *uprobe; | |
445 | unsigned long flags; | |
446 | ||
447 | spin_lock_irqsave(&uprobes_treelock, flags); | |
448 | uprobe = __find_uprobe(inode, offset); | |
449 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 450 | |
2b144498 SD |
451 | return uprobe; |
452 | } | |
453 | ||
454 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
455 | { | |
456 | struct rb_node **p = &uprobes_tree.rb_node; | |
457 | struct rb_node *parent = NULL; | |
458 | struct uprobe *u; | |
459 | int match; | |
460 | ||
461 | while (*p) { | |
462 | parent = *p; | |
463 | u = rb_entry(parent, struct uprobe, rb_node); | |
464 | match = match_uprobe(uprobe, u); | |
465 | if (!match) { | |
466 | atomic_inc(&u->ref); | |
467 | return u; | |
468 | } | |
469 | ||
470 | if (match < 0) | |
471 | p = &parent->rb_left; | |
472 | else | |
473 | p = &parent->rb_right; | |
474 | ||
475 | } | |
7b2d81d4 | 476 | |
2b144498 SD |
477 | u = NULL; |
478 | rb_link_node(&uprobe->rb_node, parent, p); | |
479 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
480 | /* get access + creation ref */ | |
481 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 482 | |
2b144498 SD |
483 | return u; |
484 | } | |
485 | ||
486 | /* | |
7b2d81d4 | 487 | * Acquire uprobes_treelock. |
2b144498 SD |
488 | * Matching uprobe already exists in rbtree; |
489 | * increment (access refcount) and return the matching uprobe. | |
490 | * | |
491 | * No matching uprobe; insert the uprobe in rb_tree; | |
492 | * get a double refcount (access + creation) and return NULL. | |
493 | */ | |
494 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
495 | { | |
496 | unsigned long flags; | |
497 | struct uprobe *u; | |
498 | ||
499 | spin_lock_irqsave(&uprobes_treelock, flags); | |
500 | u = __insert_uprobe(uprobe); | |
501 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 502 | |
0326f5a9 SD |
503 | /* For now assume that the instruction need not be single-stepped */ |
504 | uprobe->flags |= UPROBE_SKIP_SSTEP; | |
505 | ||
2b144498 SD |
506 | return u; |
507 | } | |
508 | ||
509 | static void put_uprobe(struct uprobe *uprobe) | |
510 | { | |
511 | if (atomic_dec_and_test(&uprobe->ref)) | |
512 | kfree(uprobe); | |
513 | } | |
514 | ||
515 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
516 | { | |
517 | struct uprobe *uprobe, *cur_uprobe; | |
518 | ||
519 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
520 | if (!uprobe) | |
521 | return NULL; | |
522 | ||
523 | uprobe->inode = igrab(inode); | |
524 | uprobe->offset = offset; | |
525 | init_rwsem(&uprobe->consumer_rwsem); | |
526 | INIT_LIST_HEAD(&uprobe->pending_list); | |
527 | ||
528 | /* add to uprobes_tree, sorted on inode:offset */ | |
529 | cur_uprobe = insert_uprobe(uprobe); | |
530 | ||
531 | /* a uprobe exists for this inode:offset combination */ | |
532 | if (cur_uprobe) { | |
533 | kfree(uprobe); | |
534 | uprobe = cur_uprobe; | |
535 | iput(inode); | |
7b2d81d4 | 536 | } else { |
2b144498 | 537 | atomic_inc(&uprobe_events); |
7b2d81d4 IM |
538 | } |
539 | ||
2b144498 SD |
540 | return uprobe; |
541 | } | |
542 | ||
0326f5a9 SD |
543 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
544 | { | |
545 | struct uprobe_consumer *uc; | |
546 | ||
547 | if (!(uprobe->flags & UPROBE_RUN_HANDLER)) | |
548 | return; | |
549 | ||
550 | down_read(&uprobe->consumer_rwsem); | |
551 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
552 | if (!uc->filter || uc->filter(uc, current)) | |
553 | uc->handler(uc, regs); | |
554 | } | |
555 | up_read(&uprobe->consumer_rwsem); | |
556 | } | |
557 | ||
2b144498 | 558 | /* Returns the previous consumer */ |
7b2d81d4 | 559 | static struct uprobe_consumer * |
e3343e6a | 560 | consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
561 | { |
562 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
563 | uc->next = uprobe->consumers; |
564 | uprobe->consumers = uc; | |
2b144498 | 565 | up_write(&uprobe->consumer_rwsem); |
7b2d81d4 | 566 | |
e3343e6a | 567 | return uc->next; |
2b144498 SD |
568 | } |
569 | ||
570 | /* | |
e3343e6a SD |
571 | * For uprobe @uprobe, delete the consumer @uc. |
572 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
573 | * or return false. |
574 | */ | |
e3343e6a | 575 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
576 | { |
577 | struct uprobe_consumer **con; | |
578 | bool ret = false; | |
579 | ||
580 | down_write(&uprobe->consumer_rwsem); | |
581 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
582 | if (*con == uc) { |
583 | *con = uc->next; | |
2b144498 SD |
584 | ret = true; |
585 | break; | |
586 | } | |
587 | } | |
588 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 589 | |
2b144498 SD |
590 | return ret; |
591 | } | |
592 | ||
e3343e6a | 593 | static int |
d436615e | 594 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
2b144498 SD |
595 | unsigned long nbytes, unsigned long offset) |
596 | { | |
2b144498 SD |
597 | struct page *page; |
598 | void *vaddr; | |
599 | unsigned long off1; | |
600 | unsigned long idx; | |
601 | ||
602 | if (!filp) | |
603 | return -EINVAL; | |
604 | ||
cc359d18 ON |
605 | if (!mapping->a_ops->readpage) |
606 | return -EIO; | |
607 | ||
2b144498 SD |
608 | idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT); |
609 | off1 = offset &= ~PAGE_MASK; | |
610 | ||
611 | /* | |
612 | * Ensure that the page that has the original instruction is | |
613 | * populated and in page-cache. | |
614 | */ | |
615 | page = read_mapping_page(mapping, idx, filp); | |
616 | if (IS_ERR(page)) | |
617 | return PTR_ERR(page); | |
618 | ||
619 | vaddr = kmap_atomic(page); | |
620 | memcpy(insn, vaddr + off1, nbytes); | |
621 | kunmap_atomic(vaddr); | |
622 | page_cache_release(page); | |
7b2d81d4 | 623 | |
2b144498 SD |
624 | return 0; |
625 | } | |
626 | ||
d436615e | 627 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
628 | { |
629 | struct address_space *mapping; | |
2b144498 | 630 | unsigned long nbytes; |
7b2d81d4 | 631 | int bytes; |
2b144498 | 632 | |
d436615e | 633 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
634 | mapping = uprobe->inode->i_mapping; |
635 | ||
636 | /* Instruction at end of binary; copy only available bytes */ | |
637 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
638 | bytes = uprobe->inode->i_size - uprobe->offset; | |
639 | else | |
640 | bytes = MAX_UINSN_BYTES; | |
641 | ||
642 | /* Instruction at the page-boundary; copy bytes in second page */ | |
643 | if (nbytes < bytes) { | |
d436615e | 644 | if (__copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
2b144498 SD |
645 | bytes - nbytes, uprobe->offset + nbytes)) |
646 | return -ENOMEM; | |
647 | ||
648 | bytes = nbytes; | |
649 | } | |
d436615e | 650 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
651 | } |
652 | ||
682968e0 SD |
653 | /* |
654 | * How mm->uprobes_state.count gets updated | |
655 | * uprobe_mmap() increments the count if | |
656 | * - it successfully adds a breakpoint. | |
657 | * - it cannot add a breakpoint, but sees that there is a underlying | |
658 | * breakpoint (via a is_swbp_at_addr()). | |
659 | * | |
660 | * uprobe_munmap() decrements the count if | |
661 | * - it sees a underlying breakpoint, (via is_swbp_at_addr) | |
662 | * (Subsequent uprobe_unregister wouldnt find the breakpoint | |
663 | * unless a uprobe_mmap kicks in, since the old vma would be | |
664 | * dropped just after uprobe_munmap.) | |
665 | * | |
666 | * uprobe_register increments the count if: | |
667 | * - it successfully adds a breakpoint. | |
668 | * | |
669 | * uprobe_unregister decrements the count if: | |
670 | * - it sees a underlying breakpoint and removes successfully. | |
671 | * (via is_swbp_at_addr) | |
672 | * (Subsequent uprobe_munmap wouldnt find the breakpoint | |
673 | * since there is no underlying breakpoint after the | |
674 | * breakpoint removal.) | |
675 | */ | |
e3343e6a SD |
676 | static int |
677 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
678 | struct vm_area_struct *vma, loff_t vaddr) | |
2b144498 SD |
679 | { |
680 | unsigned long addr; | |
681 | int ret; | |
682 | ||
683 | /* | |
684 | * If probe is being deleted, unregister thread could be done with | |
685 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
686 | * nobody will be able to cleanup. Also we could be from fork or | |
687 | * mremap path, where the probe might have already been inserted. | |
688 | * Hence behave as if probe already existed. | |
689 | */ | |
690 | if (!uprobe->consumers) | |
691 | return -EEXIST; | |
692 | ||
693 | addr = (unsigned long)vaddr; | |
7b2d81d4 | 694 | |
900771a4 | 695 | if (!(uprobe->flags & UPROBE_COPY_INSN)) { |
d436615e | 696 | ret = copy_insn(uprobe, vma->vm_file); |
2b144498 SD |
697 | if (ret) |
698 | return ret; | |
699 | ||
5cb4ac3a | 700 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) |
c1914a09 | 701 | return -ENOTSUPP; |
2b144498 | 702 | |
7eb9ba5e | 703 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, addr); |
2b144498 SD |
704 | if (ret) |
705 | return ret; | |
706 | ||
900771a4 | 707 | uprobe->flags |= UPROBE_COPY_INSN; |
2b144498 | 708 | } |
682968e0 SD |
709 | |
710 | /* | |
711 | * Ideally, should be updating the probe count after the breakpoint | |
712 | * has been successfully inserted. However a thread could hit the | |
713 | * breakpoint we just inserted even before the probe count is | |
714 | * incremented. If this is the first breakpoint placed, breakpoint | |
715 | * notifier might ignore uprobes and pass the trap to the thread. | |
716 | * Hence increment before and decrement on failure. | |
717 | */ | |
718 | atomic_inc(&mm->uprobes_state.count); | |
5cb4ac3a | 719 | ret = set_swbp(&uprobe->arch, mm, addr); |
682968e0 SD |
720 | if (ret) |
721 | atomic_dec(&mm->uprobes_state.count); | |
2b144498 SD |
722 | |
723 | return ret; | |
724 | } | |
725 | ||
e3343e6a SD |
726 | static void |
727 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr) | |
2b144498 | 728 | { |
682968e0 SD |
729 | if (!set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true)) |
730 | atomic_dec(&mm->uprobes_state.count); | |
2b144498 SD |
731 | } |
732 | ||
0326f5a9 | 733 | /* |
778b032d ON |
734 | * There could be threads that have already hit the breakpoint. They |
735 | * will recheck the current insn and restart if find_uprobe() fails. | |
736 | * See find_active_uprobe(). | |
0326f5a9 | 737 | */ |
2b144498 SD |
738 | static void delete_uprobe(struct uprobe *uprobe) |
739 | { | |
740 | unsigned long flags; | |
741 | ||
742 | spin_lock_irqsave(&uprobes_treelock, flags); | |
743 | rb_erase(&uprobe->rb_node, &uprobes_tree); | |
744 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
745 | iput(uprobe->inode); | |
746 | put_uprobe(uprobe); | |
747 | atomic_dec(&uprobe_events); | |
748 | } | |
749 | ||
26872090 ON |
750 | struct map_info { |
751 | struct map_info *next; | |
752 | struct mm_struct *mm; | |
753 | loff_t vaddr; | |
754 | }; | |
755 | ||
756 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 757 | { |
26872090 ON |
758 | struct map_info *next = info->next; |
759 | kfree(info); | |
760 | return next; | |
761 | } | |
762 | ||
763 | static struct map_info * | |
764 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
765 | { | |
766 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 SD |
767 | struct prio_tree_iter iter; |
768 | struct vm_area_struct *vma; | |
26872090 ON |
769 | struct map_info *curr = NULL; |
770 | struct map_info *prev = NULL; | |
771 | struct map_info *info; | |
772 | int more = 0; | |
2b144498 | 773 | |
26872090 ON |
774 | again: |
775 | mutex_lock(&mapping->i_mmap_mutex); | |
2b144498 SD |
776 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
777 | if (!valid_vma(vma, is_register)) | |
778 | continue; | |
779 | ||
7a5bfb66 ON |
780 | if (!prev && !more) { |
781 | /* | |
782 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
783 | * reclaim. This is optimistic, no harm done if it fails. | |
784 | */ | |
785 | prev = kmalloc(sizeof(struct map_info), | |
786 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
787 | if (prev) | |
788 | prev->next = NULL; | |
789 | } | |
26872090 ON |
790 | if (!prev) { |
791 | more++; | |
792 | continue; | |
2b144498 | 793 | } |
2b144498 | 794 | |
26872090 ON |
795 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
796 | continue; | |
7b2d81d4 | 797 | |
26872090 ON |
798 | info = prev; |
799 | prev = prev->next; | |
800 | info->next = curr; | |
801 | curr = info; | |
2b144498 | 802 | |
26872090 ON |
803 | info->mm = vma->vm_mm; |
804 | info->vaddr = vma_address(vma, offset); | |
805 | } | |
2b144498 SD |
806 | mutex_unlock(&mapping->i_mmap_mutex); |
807 | ||
26872090 ON |
808 | if (!more) |
809 | goto out; | |
810 | ||
811 | prev = curr; | |
812 | while (curr) { | |
813 | mmput(curr->mm); | |
814 | curr = curr->next; | |
815 | } | |
7b2d81d4 | 816 | |
26872090 ON |
817 | do { |
818 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
819 | if (!info) { | |
820 | curr = ERR_PTR(-ENOMEM); | |
821 | goto out; | |
822 | } | |
823 | info->next = prev; | |
824 | prev = info; | |
825 | } while (--more); | |
826 | ||
827 | goto again; | |
828 | out: | |
829 | while (prev) | |
830 | prev = free_map_info(prev); | |
831 | return curr; | |
2b144498 SD |
832 | } |
833 | ||
834 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
835 | { | |
26872090 ON |
836 | struct map_info *info; |
837 | int err = 0; | |
2b144498 | 838 | |
26872090 ON |
839 | info = build_map_info(uprobe->inode->i_mapping, |
840 | uprobe->offset, is_register); | |
841 | if (IS_ERR(info)) | |
842 | return PTR_ERR(info); | |
7b2d81d4 | 843 | |
26872090 ON |
844 | while (info) { |
845 | struct mm_struct *mm = info->mm; | |
846 | struct vm_area_struct *vma; | |
847 | loff_t vaddr; | |
7b2d81d4 | 848 | |
26872090 ON |
849 | if (err) |
850 | goto free; | |
7b2d81d4 | 851 | |
77fc4af1 | 852 | down_write(&mm->mmap_sem); |
26872090 ON |
853 | vma = find_vma(mm, (unsigned long)info->vaddr); |
854 | if (!vma || !valid_vma(vma, is_register)) | |
855 | goto unlock; | |
856 | ||
2b144498 SD |
857 | vaddr = vma_address(vma, uprobe->offset); |
858 | if (vma->vm_file->f_mapping->host != uprobe->inode || | |
26872090 ON |
859 | vaddr != info->vaddr) |
860 | goto unlock; | |
2b144498 | 861 | |
2b144498 | 862 | if (is_register) { |
26872090 | 863 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
c5784de2 PZ |
864 | /* |
865 | * We can race against uprobe_mmap(), see the | |
866 | * comment near uprobe_hash(). | |
867 | */ | |
26872090 ON |
868 | if (err == -EEXIST) |
869 | err = 0; | |
870 | } else { | |
871 | remove_breakpoint(uprobe, mm, info->vaddr); | |
2b144498 | 872 | } |
26872090 ON |
873 | unlock: |
874 | up_write(&mm->mmap_sem); | |
875 | free: | |
876 | mmput(mm); | |
877 | info = free_map_info(info); | |
2b144498 | 878 | } |
7b2d81d4 | 879 | |
26872090 | 880 | return err; |
2b144498 SD |
881 | } |
882 | ||
7b2d81d4 | 883 | static int __uprobe_register(struct uprobe *uprobe) |
2b144498 SD |
884 | { |
885 | return register_for_each_vma(uprobe, true); | |
886 | } | |
887 | ||
7b2d81d4 | 888 | static void __uprobe_unregister(struct uprobe *uprobe) |
2b144498 SD |
889 | { |
890 | if (!register_for_each_vma(uprobe, false)) | |
891 | delete_uprobe(uprobe); | |
892 | ||
893 | /* TODO : cant unregister? schedule a worker thread */ | |
894 | } | |
895 | ||
896 | /* | |
7b2d81d4 | 897 | * uprobe_register - register a probe |
2b144498 SD |
898 | * @inode: the file in which the probe has to be placed. |
899 | * @offset: offset from the start of the file. | |
e3343e6a | 900 | * @uc: information on howto handle the probe.. |
2b144498 | 901 | * |
7b2d81d4 | 902 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
903 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
904 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 905 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 906 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 907 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
908 | * unregisters. |
909 | * | |
910 | * Return errno if it cannot successully install probes | |
911 | * else return 0 (success) | |
912 | */ | |
e3343e6a | 913 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
914 | { |
915 | struct uprobe *uprobe; | |
7b2d81d4 | 916 | int ret; |
2b144498 | 917 | |
e3343e6a | 918 | if (!inode || !uc || uc->next) |
7b2d81d4 | 919 | return -EINVAL; |
2b144498 SD |
920 | |
921 | if (offset > i_size_read(inode)) | |
7b2d81d4 | 922 | return -EINVAL; |
2b144498 SD |
923 | |
924 | ret = 0; | |
925 | mutex_lock(uprobes_hash(inode)); | |
926 | uprobe = alloc_uprobe(inode, offset); | |
7b2d81d4 | 927 | |
e3343e6a | 928 | if (uprobe && !consumer_add(uprobe, uc)) { |
7b2d81d4 | 929 | ret = __uprobe_register(uprobe); |
2b144498 SD |
930 | if (ret) { |
931 | uprobe->consumers = NULL; | |
7b2d81d4 IM |
932 | __uprobe_unregister(uprobe); |
933 | } else { | |
900771a4 | 934 | uprobe->flags |= UPROBE_RUN_HANDLER; |
7b2d81d4 | 935 | } |
2b144498 SD |
936 | } |
937 | ||
938 | mutex_unlock(uprobes_hash(inode)); | |
939 | put_uprobe(uprobe); | |
940 | ||
941 | return ret; | |
942 | } | |
943 | ||
944 | /* | |
7b2d81d4 | 945 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
946 | * @inode: the file in which the probe has to be removed. |
947 | * @offset: offset from the start of the file. | |
e3343e6a | 948 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 949 | */ |
e3343e6a | 950 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 951 | { |
7b2d81d4 | 952 | struct uprobe *uprobe; |
2b144498 | 953 | |
e3343e6a | 954 | if (!inode || !uc) |
2b144498 SD |
955 | return; |
956 | ||
957 | uprobe = find_uprobe(inode, offset); | |
958 | if (!uprobe) | |
959 | return; | |
960 | ||
961 | mutex_lock(uprobes_hash(inode)); | |
2b144498 | 962 | |
e3343e6a | 963 | if (consumer_del(uprobe, uc)) { |
7b2d81d4 IM |
964 | if (!uprobe->consumers) { |
965 | __uprobe_unregister(uprobe); | |
900771a4 | 966 | uprobe->flags &= ~UPROBE_RUN_HANDLER; |
7b2d81d4 | 967 | } |
2b144498 SD |
968 | } |
969 | ||
2b144498 SD |
970 | mutex_unlock(uprobes_hash(inode)); |
971 | if (uprobe) | |
972 | put_uprobe(uprobe); | |
973 | } | |
974 | ||
975 | /* | |
976 | * Of all the nodes that correspond to the given inode, return the node | |
977 | * with the least offset. | |
978 | */ | |
979 | static struct rb_node *find_least_offset_node(struct inode *inode) | |
980 | { | |
981 | struct uprobe u = { .inode = inode, .offset = 0}; | |
982 | struct rb_node *n = uprobes_tree.rb_node; | |
983 | struct rb_node *close_node = NULL; | |
984 | struct uprobe *uprobe; | |
985 | int match; | |
986 | ||
987 | while (n) { | |
988 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
989 | match = match_uprobe(&u, uprobe); | |
7b2d81d4 | 990 | |
2b144498 SD |
991 | if (uprobe->inode == inode) |
992 | close_node = n; | |
993 | ||
994 | if (!match) | |
995 | return close_node; | |
996 | ||
997 | if (match < 0) | |
998 | n = n->rb_left; | |
999 | else | |
1000 | n = n->rb_right; | |
1001 | } | |
7b2d81d4 | 1002 | |
2b144498 SD |
1003 | return close_node; |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * For a given inode, build a list of probes that need to be inserted. | |
1008 | */ | |
1009 | static void build_probe_list(struct inode *inode, struct list_head *head) | |
1010 | { | |
1011 | struct uprobe *uprobe; | |
2b144498 | 1012 | unsigned long flags; |
7b2d81d4 | 1013 | struct rb_node *n; |
2b144498 SD |
1014 | |
1015 | spin_lock_irqsave(&uprobes_treelock, flags); | |
7b2d81d4 | 1016 | |
2b144498 | 1017 | n = find_least_offset_node(inode); |
7b2d81d4 | 1018 | |
2b144498 SD |
1019 | for (; n; n = rb_next(n)) { |
1020 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
1021 | if (uprobe->inode != inode) | |
1022 | break; | |
1023 | ||
1024 | list_add(&uprobe->pending_list, head); | |
1025 | atomic_inc(&uprobe->ref); | |
1026 | } | |
7b2d81d4 | 1027 | |
2b144498 SD |
1028 | spin_unlock_irqrestore(&uprobes_treelock, flags); |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * Called from mmap_region. | |
1033 | * called with mm->mmap_sem acquired. | |
1034 | * | |
1035 | * Return -ve no if we fail to insert probes and we cannot | |
1036 | * bail-out. | |
7b2d81d4 IM |
1037 | * Return 0 otherwise. i.e: |
1038 | * | |
2b144498 SD |
1039 | * - successful insertion of probes |
1040 | * - (or) no possible probes to be inserted. | |
1041 | * - (or) insertion of probes failed but we can bail-out. | |
1042 | */ | |
7b2d81d4 | 1043 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1044 | { |
1045 | struct list_head tmp_list; | |
1046 | struct uprobe *uprobe, *u; | |
1047 | struct inode *inode; | |
682968e0 | 1048 | int ret, count; |
2b144498 SD |
1049 | |
1050 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) | |
7b2d81d4 | 1051 | return 0; |
2b144498 SD |
1052 | |
1053 | inode = vma->vm_file->f_mapping->host; | |
1054 | if (!inode) | |
7b2d81d4 | 1055 | return 0; |
2b144498 SD |
1056 | |
1057 | INIT_LIST_HEAD(&tmp_list); | |
1058 | mutex_lock(uprobes_mmap_hash(inode)); | |
1059 | build_probe_list(inode, &tmp_list); | |
7b2d81d4 IM |
1060 | |
1061 | ret = 0; | |
682968e0 | 1062 | count = 0; |
7b2d81d4 | 1063 | |
2b144498 SD |
1064 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
1065 | loff_t vaddr; | |
1066 | ||
1067 | list_del(&uprobe->pending_list); | |
1068 | if (!ret) { | |
1069 | vaddr = vma_address(vma, uprobe->offset); | |
682968e0 SD |
1070 | |
1071 | if (vaddr < vma->vm_start || vaddr >= vma->vm_end) { | |
1072 | put_uprobe(uprobe); | |
1073 | continue; | |
2b144498 | 1074 | } |
682968e0 SD |
1075 | |
1076 | ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); | |
c5784de2 PZ |
1077 | /* |
1078 | * We can race against uprobe_register(), see the | |
1079 | * comment near uprobe_hash(). | |
1080 | */ | |
682968e0 SD |
1081 | if (ret == -EEXIST) { |
1082 | ret = 0; | |
1083 | ||
1084 | if (!is_swbp_at_addr(vma->vm_mm, vaddr)) | |
1085 | continue; | |
1086 | ||
1087 | /* | |
1088 | * Unable to insert a breakpoint, but | |
1089 | * breakpoint lies underneath. Increment the | |
1090 | * probe count. | |
1091 | */ | |
1092 | atomic_inc(&vma->vm_mm->uprobes_state.count); | |
1093 | } | |
1094 | ||
1095 | if (!ret) | |
1096 | count++; | |
2b144498 SD |
1097 | } |
1098 | put_uprobe(uprobe); | |
1099 | } | |
1100 | ||
1101 | mutex_unlock(uprobes_mmap_hash(inode)); | |
1102 | ||
682968e0 SD |
1103 | if (ret) |
1104 | atomic_sub(count, &vma->vm_mm->uprobes_state.count); | |
1105 | ||
2b144498 SD |
1106 | return ret; |
1107 | } | |
1108 | ||
682968e0 SD |
1109 | /* |
1110 | * Called in context of a munmap of a vma. | |
1111 | */ | |
cbc91f71 | 1112 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 SD |
1113 | { |
1114 | struct list_head tmp_list; | |
1115 | struct uprobe *uprobe, *u; | |
1116 | struct inode *inode; | |
1117 | ||
1118 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) | |
1119 | return; | |
1120 | ||
1121 | if (!atomic_read(&vma->vm_mm->uprobes_state.count)) | |
1122 | return; | |
1123 | ||
1124 | inode = vma->vm_file->f_mapping->host; | |
1125 | if (!inode) | |
1126 | return; | |
1127 | ||
1128 | INIT_LIST_HEAD(&tmp_list); | |
1129 | mutex_lock(uprobes_mmap_hash(inode)); | |
1130 | build_probe_list(inode, &tmp_list); | |
1131 | ||
1132 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { | |
1133 | loff_t vaddr; | |
1134 | ||
1135 | list_del(&uprobe->pending_list); | |
1136 | vaddr = vma_address(vma, uprobe->offset); | |
1137 | ||
cbc91f71 | 1138 | if (vaddr >= start && vaddr < end) { |
682968e0 SD |
1139 | /* |
1140 | * An unregister could have removed the probe before | |
1141 | * unmap. So check before we decrement the count. | |
1142 | */ | |
1143 | if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1) | |
1144 | atomic_dec(&vma->vm_mm->uprobes_state.count); | |
1145 | } | |
1146 | put_uprobe(uprobe); | |
1147 | } | |
1148 | mutex_unlock(uprobes_mmap_hash(inode)); | |
1149 | } | |
1150 | ||
d4b3b638 SD |
1151 | /* Slot allocation for XOL */ |
1152 | static int xol_add_vma(struct xol_area *area) | |
1153 | { | |
1154 | struct mm_struct *mm; | |
1155 | int ret; | |
1156 | ||
1157 | area->page = alloc_page(GFP_HIGHUSER); | |
1158 | if (!area->page) | |
1159 | return -ENOMEM; | |
1160 | ||
1161 | ret = -EALREADY; | |
1162 | mm = current->mm; | |
1163 | ||
1164 | down_write(&mm->mmap_sem); | |
1165 | if (mm->uprobes_state.xol_area) | |
1166 | goto fail; | |
1167 | ||
1168 | ret = -ENOMEM; | |
1169 | ||
1170 | /* Try to map as high as possible, this is only a hint. */ | |
1171 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1172 | if (area->vaddr & ~PAGE_MASK) { | |
1173 | ret = area->vaddr; | |
1174 | goto fail; | |
1175 | } | |
1176 | ||
1177 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1178 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1179 | if (ret) | |
1180 | goto fail; | |
1181 | ||
1182 | smp_wmb(); /* pairs with get_xol_area() */ | |
1183 | mm->uprobes_state.xol_area = area; | |
1184 | ret = 0; | |
1185 | ||
1186 | fail: | |
1187 | up_write(&mm->mmap_sem); | |
1188 | if (ret) | |
1189 | __free_page(area->page); | |
1190 | ||
1191 | return ret; | |
1192 | } | |
1193 | ||
1194 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1195 | { | |
1196 | struct xol_area *area; | |
1197 | ||
1198 | area = mm->uprobes_state.xol_area; | |
1199 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1200 | ||
1201 | return area; | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * xol_alloc_area - Allocate process's xol_area. | |
1206 | * This area will be used for storing instructions for execution out of | |
1207 | * line. | |
1208 | * | |
1209 | * Returns the allocated area or NULL. | |
1210 | */ | |
1211 | static struct xol_area *xol_alloc_area(void) | |
1212 | { | |
1213 | struct xol_area *area; | |
1214 | ||
1215 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1216 | if (unlikely(!area)) | |
1217 | return NULL; | |
1218 | ||
1219 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1220 | ||
1221 | if (!area->bitmap) | |
1222 | goto fail; | |
1223 | ||
1224 | init_waitqueue_head(&area->wq); | |
1225 | if (!xol_add_vma(area)) | |
1226 | return area; | |
1227 | ||
1228 | fail: | |
1229 | kfree(area->bitmap); | |
1230 | kfree(area); | |
1231 | ||
1232 | return get_xol_area(current->mm); | |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * uprobe_clear_state - Free the area allocated for slots. | |
1237 | */ | |
1238 | void uprobe_clear_state(struct mm_struct *mm) | |
1239 | { | |
1240 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1241 | ||
1242 | if (!area) | |
1243 | return; | |
1244 | ||
1245 | put_page(area->page); | |
1246 | kfree(area->bitmap); | |
1247 | kfree(area); | |
1248 | } | |
1249 | ||
1250 | /* | |
1251 | * uprobe_reset_state - Free the area allocated for slots. | |
1252 | */ | |
1253 | void uprobe_reset_state(struct mm_struct *mm) | |
1254 | { | |
1255 | mm->uprobes_state.xol_area = NULL; | |
682968e0 | 1256 | atomic_set(&mm->uprobes_state.count, 0); |
d4b3b638 SD |
1257 | } |
1258 | ||
1259 | /* | |
1260 | * - search for a free slot. | |
1261 | */ | |
1262 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1263 | { | |
1264 | unsigned long slot_addr; | |
1265 | int slot_nr; | |
1266 | ||
1267 | do { | |
1268 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1269 | if (slot_nr < UINSNS_PER_PAGE) { | |
1270 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1271 | break; | |
1272 | ||
1273 | slot_nr = UINSNS_PER_PAGE; | |
1274 | continue; | |
1275 | } | |
1276 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1277 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1278 | ||
1279 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1280 | atomic_inc(&area->slot_count); | |
1281 | ||
1282 | return slot_addr; | |
1283 | } | |
1284 | ||
1285 | /* | |
1286 | * xol_get_insn_slot - If was not allocated a slot, then | |
1287 | * allocate a slot. | |
1288 | * Returns the allocated slot address or 0. | |
1289 | */ | |
1290 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1291 | { | |
1292 | struct xol_area *area; | |
1293 | unsigned long offset; | |
1294 | void *vaddr; | |
1295 | ||
1296 | area = get_xol_area(current->mm); | |
1297 | if (!area) { | |
1298 | area = xol_alloc_area(); | |
1299 | if (!area) | |
1300 | return 0; | |
1301 | } | |
1302 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1303 | ||
1304 | /* | |
1305 | * Initialize the slot if xol_vaddr points to valid | |
1306 | * instruction slot. | |
1307 | */ | |
1308 | if (unlikely(!current->utask->xol_vaddr)) | |
1309 | return 0; | |
1310 | ||
1311 | current->utask->vaddr = slot_addr; | |
1312 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1313 | vaddr = kmap_atomic(area->page); | |
1314 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1315 | kunmap_atomic(vaddr); | |
1316 | ||
1317 | return current->utask->xol_vaddr; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * xol_free_insn_slot - If slot was earlier allocated by | |
1322 | * @xol_get_insn_slot(), make the slot available for | |
1323 | * subsequent requests. | |
1324 | */ | |
1325 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1326 | { | |
1327 | struct xol_area *area; | |
1328 | unsigned long vma_end; | |
1329 | unsigned long slot_addr; | |
1330 | ||
1331 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1332 | return; | |
1333 | ||
1334 | slot_addr = tsk->utask->xol_vaddr; | |
1335 | ||
1336 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1337 | return; | |
1338 | ||
1339 | area = tsk->mm->uprobes_state.xol_area; | |
1340 | vma_end = area->vaddr + PAGE_SIZE; | |
1341 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1342 | unsigned long offset; | |
1343 | int slot_nr; | |
1344 | ||
1345 | offset = slot_addr - area->vaddr; | |
1346 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1347 | if (slot_nr >= UINSNS_PER_PAGE) | |
1348 | return; | |
1349 | ||
1350 | clear_bit(slot_nr, area->bitmap); | |
1351 | atomic_dec(&area->slot_count); | |
1352 | if (waitqueue_active(&area->wq)) | |
1353 | wake_up(&area->wq); | |
1354 | ||
1355 | tsk->utask->xol_vaddr = 0; | |
1356 | } | |
1357 | } | |
1358 | ||
0326f5a9 SD |
1359 | /** |
1360 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1361 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1362 | * instruction. | |
1363 | * Return the address of the breakpoint instruction. | |
1364 | */ | |
1365 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1366 | { | |
1367 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Called with no locks held. | |
1372 | * Called in context of a exiting or a exec-ing thread. | |
1373 | */ | |
1374 | void uprobe_free_utask(struct task_struct *t) | |
1375 | { | |
1376 | struct uprobe_task *utask = t->utask; | |
1377 | ||
0326f5a9 SD |
1378 | if (!utask) |
1379 | return; | |
1380 | ||
1381 | if (utask->active_uprobe) | |
1382 | put_uprobe(utask->active_uprobe); | |
1383 | ||
d4b3b638 | 1384 | xol_free_insn_slot(t); |
0326f5a9 SD |
1385 | kfree(utask); |
1386 | t->utask = NULL; | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * Called in context of a new clone/fork from copy_process. | |
1391 | */ | |
1392 | void uprobe_copy_process(struct task_struct *t) | |
1393 | { | |
1394 | t->utask = NULL; | |
0326f5a9 SD |
1395 | } |
1396 | ||
1397 | /* | |
1398 | * Allocate a uprobe_task object for the task. | |
1399 | * Called when the thread hits a breakpoint for the first time. | |
1400 | * | |
1401 | * Returns: | |
1402 | * - pointer to new uprobe_task on success | |
1403 | * - NULL otherwise | |
1404 | */ | |
1405 | static struct uprobe_task *add_utask(void) | |
1406 | { | |
1407 | struct uprobe_task *utask; | |
1408 | ||
1409 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1410 | if (unlikely(!utask)) | |
1411 | return NULL; | |
1412 | ||
1413 | utask->active_uprobe = NULL; | |
1414 | current->utask = utask; | |
1415 | return utask; | |
1416 | } | |
1417 | ||
1418 | /* Prepare to single-step probed instruction out of line. */ | |
1419 | static int | |
1420 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1421 | { | |
d4b3b638 SD |
1422 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1423 | return 0; | |
1424 | ||
0326f5a9 SD |
1425 | return -EFAULT; |
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * If we are singlestepping, then ensure this thread is not connected to | |
1430 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1431 | * triggers the signal, restart the original insn even if the task is | |
1432 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1433 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1434 | * _same_ instruction should be repeated again after return from the signal | |
1435 | * handler, and SSTEP can never finish in this case. | |
1436 | */ | |
1437 | bool uprobe_deny_signal(void) | |
1438 | { | |
1439 | struct task_struct *t = current; | |
1440 | struct uprobe_task *utask = t->utask; | |
1441 | ||
1442 | if (likely(!utask || !utask->active_uprobe)) | |
1443 | return false; | |
1444 | ||
1445 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1446 | ||
1447 | if (signal_pending(t)) { | |
1448 | spin_lock_irq(&t->sighand->siglock); | |
1449 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1450 | spin_unlock_irq(&t->sighand->siglock); | |
1451 | ||
1452 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1453 | utask->state = UTASK_SSTEP_TRAPPED; | |
1454 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1455 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1456 | } | |
1457 | } | |
1458 | ||
1459 | return true; | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * Avoid singlestepping the original instruction if the original instruction | |
1464 | * is a NOP or can be emulated. | |
1465 | */ | |
1466 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1467 | { | |
1468 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) | |
1469 | return true; | |
1470 | ||
1471 | uprobe->flags &= ~UPROBE_SKIP_SSTEP; | |
1472 | return false; | |
1473 | } | |
1474 | ||
d790d346 | 1475 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1476 | { |
3a9ea052 ON |
1477 | struct mm_struct *mm = current->mm; |
1478 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1479 | struct vm_area_struct *vma; |
0326f5a9 | 1480 | |
0326f5a9 SD |
1481 | down_read(&mm->mmap_sem); |
1482 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1483 | if (vma && vma->vm_start <= bp_vaddr) { |
1484 | if (valid_vma(vma, false)) { | |
1485 | struct inode *inode; | |
1486 | loff_t offset; | |
0326f5a9 | 1487 | |
3a9ea052 ON |
1488 | inode = vma->vm_file->f_mapping->host; |
1489 | offset = bp_vaddr - vma->vm_start; | |
1490 | offset += (vma->vm_pgoff << PAGE_SHIFT); | |
1491 | uprobe = find_uprobe(inode, offset); | |
1492 | } | |
d790d346 ON |
1493 | |
1494 | if (!uprobe) | |
1495 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1496 | } else { | |
1497 | *is_swbp = -EFAULT; | |
0326f5a9 | 1498 | } |
0326f5a9 SD |
1499 | up_read(&mm->mmap_sem); |
1500 | ||
3a9ea052 ON |
1501 | return uprobe; |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * Run handler and ask thread to singlestep. | |
1506 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1507 | */ | |
1508 | static void handle_swbp(struct pt_regs *regs) | |
1509 | { | |
1510 | struct uprobe_task *utask; | |
1511 | struct uprobe *uprobe; | |
1512 | unsigned long bp_vaddr; | |
56bb4cf6 | 1513 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1514 | |
1515 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1516 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1517 | |
0326f5a9 | 1518 | if (!uprobe) { |
56bb4cf6 ON |
1519 | if (is_swbp > 0) { |
1520 | /* No matching uprobe; signal SIGTRAP. */ | |
1521 | send_sig(SIGTRAP, current, 0); | |
1522 | } else { | |
1523 | /* | |
1524 | * Either we raced with uprobe_unregister() or we can't | |
1525 | * access this memory. The latter is only possible if | |
1526 | * another thread plays with our ->mm. In both cases | |
1527 | * we can simply restart. If this vma was unmapped we | |
1528 | * can pretend this insn was not executed yet and get | |
1529 | * the (correct) SIGSEGV after restart. | |
1530 | */ | |
1531 | instruction_pointer_set(regs, bp_vaddr); | |
1532 | } | |
0326f5a9 SD |
1533 | return; |
1534 | } | |
1535 | ||
1536 | utask = current->utask; | |
1537 | if (!utask) { | |
1538 | utask = add_utask(); | |
1539 | /* Cannot allocate; re-execute the instruction. */ | |
1540 | if (!utask) | |
1541 | goto cleanup_ret; | |
1542 | } | |
1543 | utask->active_uprobe = uprobe; | |
1544 | handler_chain(uprobe, regs); | |
1545 | if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs)) | |
1546 | goto cleanup_ret; | |
1547 | ||
1548 | utask->state = UTASK_SSTEP; | |
1549 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { | |
1550 | user_enable_single_step(current); | |
1551 | return; | |
1552 | } | |
1553 | ||
1554 | cleanup_ret: | |
1555 | if (utask) { | |
1556 | utask->active_uprobe = NULL; | |
1557 | utask->state = UTASK_RUNNING; | |
1558 | } | |
1559 | if (uprobe) { | |
1560 | if (!(uprobe->flags & UPROBE_SKIP_SSTEP)) | |
1561 | ||
1562 | /* | |
1563 | * cannot singlestep; cannot skip instruction; | |
1564 | * re-execute the instruction. | |
1565 | */ | |
1566 | instruction_pointer_set(regs, bp_vaddr); | |
1567 | ||
1568 | put_uprobe(uprobe); | |
1569 | } | |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * Perform required fix-ups and disable singlestep. | |
1574 | * Allow pending signals to take effect. | |
1575 | */ | |
1576 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1577 | { | |
1578 | struct uprobe *uprobe; | |
1579 | ||
1580 | uprobe = utask->active_uprobe; | |
1581 | if (utask->state == UTASK_SSTEP_ACK) | |
1582 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1583 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1584 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1585 | else | |
1586 | WARN_ON_ONCE(1); | |
1587 | ||
1588 | put_uprobe(uprobe); | |
1589 | utask->active_uprobe = NULL; | |
1590 | utask->state = UTASK_RUNNING; | |
1591 | user_disable_single_step(current); | |
d4b3b638 | 1592 | xol_free_insn_slot(current); |
0326f5a9 SD |
1593 | |
1594 | spin_lock_irq(¤t->sighand->siglock); | |
1595 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1596 | spin_unlock_irq(¤t->sighand->siglock); | |
1597 | } | |
1598 | ||
1599 | /* | |
1600 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on | |
1601 | * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and | |
1602 | * allows the thread to return from interrupt. | |
1603 | * | |
1604 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and | |
1605 | * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from | |
1606 | * interrupt. | |
1607 | * | |
1608 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1609 | * uprobe_notify_resume(). | |
1610 | */ | |
1611 | void uprobe_notify_resume(struct pt_regs *regs) | |
1612 | { | |
1613 | struct uprobe_task *utask; | |
1614 | ||
1615 | utask = current->utask; | |
1616 | if (!utask || utask->state == UTASK_BP_HIT) | |
1617 | handle_swbp(regs); | |
1618 | else | |
1619 | handle_singlestep(utask, regs); | |
1620 | } | |
1621 | ||
1622 | /* | |
1623 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1624 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1625 | */ | |
1626 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1627 | { | |
1628 | struct uprobe_task *utask; | |
1629 | ||
682968e0 SD |
1630 | if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count)) |
1631 | /* task is currently not uprobed */ | |
0326f5a9 SD |
1632 | return 0; |
1633 | ||
1634 | utask = current->utask; | |
1635 | if (utask) | |
1636 | utask->state = UTASK_BP_HIT; | |
1637 | ||
1638 | set_thread_flag(TIF_UPROBE); | |
0326f5a9 SD |
1639 | |
1640 | return 1; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1645 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1646 | */ | |
1647 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1648 | { | |
1649 | struct uprobe_task *utask = current->utask; | |
1650 | ||
1651 | if (!current->mm || !utask || !utask->active_uprobe) | |
1652 | /* task is currently not uprobed */ | |
1653 | return 0; | |
1654 | ||
1655 | utask->state = UTASK_SSTEP_ACK; | |
1656 | set_thread_flag(TIF_UPROBE); | |
1657 | return 1; | |
1658 | } | |
1659 | ||
1660 | static struct notifier_block uprobe_exception_nb = { | |
1661 | .notifier_call = arch_uprobe_exception_notify, | |
1662 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1663 | }; | |
1664 | ||
2b144498 SD |
1665 | static int __init init_uprobes(void) |
1666 | { | |
1667 | int i; | |
1668 | ||
1669 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1670 | mutex_init(&uprobes_mutex[i]); | |
1671 | mutex_init(&uprobes_mmap_mutex[i]); | |
1672 | } | |
0326f5a9 SD |
1673 | |
1674 | return register_die_notifier(&uprobe_exception_nb); | |
2b144498 | 1675 | } |
0326f5a9 | 1676 | module_init(init_uprobes); |
2b144498 SD |
1677 | |
1678 | static void __exit exit_uprobes(void) | |
1679 | { | |
1680 | } | |
2b144498 | 1681 | module_exit(exit_uprobes); |