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 | ||
0326f5a9 | 41 | static struct srcu_struct uprobes_srcu; |
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 | |
2b144498 SD |
48 | /* serialize (un)register */ |
49 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
50 | |
51 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
52 | |
53 | /* serialize uprobe->pending_list */ | |
54 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 55 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 SD |
56 | |
57 | /* | |
7b2d81d4 | 58 | * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe |
2b144498 SD |
59 | * events active at this time. Probably a fine grained per inode count is |
60 | * better? | |
61 | */ | |
62 | static atomic_t uprobe_events = ATOMIC_INIT(0); | |
63 | ||
64 | /* | |
65 | * Maintain a temporary per vma info that can be used to search if a vma | |
66 | * has already been handled. This structure is introduced since extending | |
67 | * vm_area_struct wasnt recommended. | |
68 | */ | |
69 | struct vma_info { | |
7b2d81d4 IM |
70 | struct list_head probe_list; |
71 | struct mm_struct *mm; | |
72 | loff_t vaddr; | |
2b144498 SD |
73 | }; |
74 | ||
3ff54efd SD |
75 | struct uprobe { |
76 | struct rb_node rb_node; /* node in the rb tree */ | |
77 | atomic_t ref; | |
78 | struct rw_semaphore consumer_rwsem; | |
79 | struct list_head pending_list; | |
80 | struct uprobe_consumer *consumers; | |
81 | struct inode *inode; /* Also hold a ref to inode */ | |
82 | loff_t offset; | |
83 | int flags; | |
84 | struct arch_uprobe arch; | |
85 | }; | |
86 | ||
2b144498 SD |
87 | /* |
88 | * valid_vma: Verify if the specified vma is an executable vma | |
89 | * Relax restrictions while unregistering: vm_flags might have | |
90 | * changed after breakpoint was inserted. | |
91 | * - is_register: indicates if we are in register context. | |
92 | * - Return 1 if the specified virtual address is in an | |
93 | * executable vma. | |
94 | */ | |
95 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
96 | { | |
97 | if (!vma->vm_file) | |
98 | return false; | |
99 | ||
100 | if (!is_register) | |
101 | return true; | |
102 | ||
7b2d81d4 | 103 | if ((vma->vm_flags & (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)) == (VM_READ|VM_EXEC)) |
2b144498 SD |
104 | return true; |
105 | ||
106 | return false; | |
107 | } | |
108 | ||
109 | static loff_t vma_address(struct vm_area_struct *vma, loff_t offset) | |
110 | { | |
111 | loff_t vaddr; | |
112 | ||
113 | vaddr = vma->vm_start + offset; | |
114 | vaddr -= vma->vm_pgoff << PAGE_SHIFT; | |
7b2d81d4 | 115 | |
2b144498 SD |
116 | return vaddr; |
117 | } | |
118 | ||
119 | /** | |
120 | * __replace_page - replace page in vma by new page. | |
121 | * based on replace_page in mm/ksm.c | |
122 | * | |
123 | * @vma: vma that holds the pte pointing to page | |
124 | * @page: the cowed page we are replacing by kpage | |
125 | * @kpage: the modified page we replace page by | |
126 | * | |
127 | * Returns 0 on success, -EFAULT on failure. | |
128 | */ | |
7b2d81d4 | 129 | static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage) |
2b144498 SD |
130 | { |
131 | struct mm_struct *mm = vma->vm_mm; | |
132 | pgd_t *pgd; | |
133 | pud_t *pud; | |
134 | pmd_t *pmd; | |
135 | pte_t *ptep; | |
136 | spinlock_t *ptl; | |
137 | unsigned long addr; | |
138 | int err = -EFAULT; | |
139 | ||
140 | addr = page_address_in_vma(page, vma); | |
141 | if (addr == -EFAULT) | |
142 | goto out; | |
143 | ||
144 | pgd = pgd_offset(mm, addr); | |
145 | if (!pgd_present(*pgd)) | |
146 | goto out; | |
147 | ||
148 | pud = pud_offset(pgd, addr); | |
149 | if (!pud_present(*pud)) | |
150 | goto out; | |
151 | ||
152 | pmd = pmd_offset(pud, addr); | |
153 | if (!pmd_present(*pmd)) | |
154 | goto out; | |
155 | ||
156 | ptep = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
157 | if (!ptep) | |
158 | goto out; | |
159 | ||
160 | get_page(kpage); | |
161 | page_add_new_anon_rmap(kpage, vma, addr); | |
162 | ||
7396fa81 SD |
163 | if (!PageAnon(page)) { |
164 | dec_mm_counter(mm, MM_FILEPAGES); | |
165 | inc_mm_counter(mm, MM_ANONPAGES); | |
166 | } | |
167 | ||
2b144498 SD |
168 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
169 | ptep_clear_flush(vma, addr, ptep); | |
170 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
171 | ||
172 | page_remove_rmap(page); | |
173 | if (!page_mapped(page)) | |
174 | try_to_free_swap(page); | |
175 | put_page(page); | |
176 | pte_unmap_unlock(ptep, ptl); | |
177 | err = 0; | |
178 | ||
179 | out: | |
180 | return err; | |
181 | } | |
182 | ||
183 | /** | |
5cb4ac3a | 184 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 185 | * @insn: instruction to be checked. |
5cb4ac3a | 186 | * Default implementation of is_swbp_insn |
2b144498 SD |
187 | * Returns true if @insn is a breakpoint instruction. |
188 | */ | |
5cb4ac3a | 189 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 190 | { |
5cb4ac3a | 191 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
192 | } |
193 | ||
194 | /* | |
195 | * NOTE: | |
196 | * Expect the breakpoint instruction to be the smallest size instruction for | |
197 | * the architecture. If an arch has variable length instruction and the | |
198 | * breakpoint instruction is not of the smallest length instruction | |
199 | * supported by that architecture then we need to modify read_opcode / | |
200 | * write_opcode accordingly. This would never be a problem for archs that | |
201 | * have fixed length instructions. | |
202 | */ | |
203 | ||
204 | /* | |
205 | * write_opcode - write the opcode at a given virtual address. | |
e3343e6a | 206 | * @auprobe: arch breakpointing information. |
2b144498 | 207 | * @mm: the probed process address space. |
2b144498 SD |
208 | * @vaddr: the virtual address to store the opcode. |
209 | * @opcode: opcode to be written at @vaddr. | |
210 | * | |
211 | * Called with mm->mmap_sem held (for read and with a reference to | |
212 | * mm). | |
213 | * | |
214 | * For mm @mm, write the opcode at @vaddr. | |
215 | * Return 0 (success) or a negative errno. | |
216 | */ | |
e3343e6a | 217 | static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, |
2b144498 SD |
218 | unsigned long vaddr, uprobe_opcode_t opcode) |
219 | { | |
220 | struct page *old_page, *new_page; | |
221 | struct address_space *mapping; | |
222 | void *vaddr_old, *vaddr_new; | |
223 | struct vm_area_struct *vma; | |
3ff54efd | 224 | struct uprobe *uprobe; |
2b144498 SD |
225 | loff_t addr; |
226 | int ret; | |
227 | ||
228 | /* Read the page with vaddr into memory */ | |
229 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma); | |
230 | if (ret <= 0) | |
231 | return ret; | |
7b2d81d4 | 232 | |
2b144498 SD |
233 | ret = -EINVAL; |
234 | ||
235 | /* | |
236 | * We are interested in text pages only. Our pages of interest | |
237 | * should be mapped for read and execute only. We desist from | |
238 | * adding probes in write mapped pages since the breakpoints | |
239 | * might end up in the file copy. | |
240 | */ | |
5cb4ac3a | 241 | if (!valid_vma(vma, is_swbp_insn(&opcode))) |
2b144498 SD |
242 | goto put_out; |
243 | ||
3ff54efd | 244 | uprobe = container_of(auprobe, struct uprobe, arch); |
2b144498 SD |
245 | mapping = uprobe->inode->i_mapping; |
246 | if (mapping != vma->vm_file->f_mapping) | |
247 | goto put_out; | |
248 | ||
249 | addr = vma_address(vma, uprobe->offset); | |
250 | if (vaddr != (unsigned long)addr) | |
251 | goto put_out; | |
252 | ||
253 | ret = -ENOMEM; | |
254 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
255 | if (!new_page) | |
256 | goto put_out; | |
257 | ||
258 | __SetPageUptodate(new_page); | |
259 | ||
260 | /* | |
261 | * lock page will serialize against do_wp_page()'s | |
262 | * PageAnon() handling | |
263 | */ | |
264 | lock_page(old_page); | |
265 | /* copy the page now that we've got it stable */ | |
266 | vaddr_old = kmap_atomic(old_page); | |
267 | vaddr_new = kmap_atomic(new_page); | |
268 | ||
269 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
7b2d81d4 | 270 | |
2b144498 SD |
271 | /* poke the new insn in, ASSUMES we don't cross page boundary */ |
272 | vaddr &= ~PAGE_MASK; | |
5cb4ac3a SD |
273 | BUG_ON(vaddr + UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); |
274 | memcpy(vaddr_new + vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); | |
2b144498 SD |
275 | |
276 | kunmap_atomic(vaddr_new); | |
277 | kunmap_atomic(vaddr_old); | |
278 | ||
279 | ret = anon_vma_prepare(vma); | |
280 | if (ret) | |
281 | goto unlock_out; | |
282 | ||
283 | lock_page(new_page); | |
284 | ret = __replace_page(vma, old_page, new_page); | |
285 | unlock_page(new_page); | |
286 | ||
287 | unlock_out: | |
288 | unlock_page(old_page); | |
289 | page_cache_release(new_page); | |
290 | ||
291 | put_out: | |
7b2d81d4 IM |
292 | put_page(old_page); |
293 | ||
2b144498 SD |
294 | return ret; |
295 | } | |
296 | ||
297 | /** | |
298 | * read_opcode - read the opcode at a given virtual address. | |
299 | * @mm: the probed process address space. | |
300 | * @vaddr: the virtual address to read the opcode. | |
301 | * @opcode: location to store the read opcode. | |
302 | * | |
303 | * Called with mm->mmap_sem held (for read and with a reference to | |
304 | * mm. | |
305 | * | |
306 | * For mm @mm, read the opcode at @vaddr and store it in @opcode. | |
307 | * Return 0 (success) or a negative errno. | |
308 | */ | |
7b2d81d4 | 309 | static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode) |
2b144498 SD |
310 | { |
311 | struct page *page; | |
312 | void *vaddr_new; | |
313 | int ret; | |
314 | ||
a3d7bb47 | 315 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); |
2b144498 SD |
316 | if (ret <= 0) |
317 | return ret; | |
318 | ||
319 | lock_page(page); | |
320 | vaddr_new = kmap_atomic(page); | |
321 | vaddr &= ~PAGE_MASK; | |
5cb4ac3a | 322 | memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
323 | kunmap_atomic(vaddr_new); |
324 | unlock_page(page); | |
7b2d81d4 IM |
325 | |
326 | put_page(page); | |
327 | ||
2b144498 SD |
328 | return 0; |
329 | } | |
330 | ||
5cb4ac3a | 331 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
2b144498 SD |
332 | { |
333 | uprobe_opcode_t opcode; | |
7b2d81d4 | 334 | int result; |
2b144498 | 335 | |
c00b2750 ON |
336 | if (current->mm == mm) { |
337 | pagefault_disable(); | |
338 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
339 | sizeof(opcode)); | |
340 | pagefault_enable(); | |
341 | ||
342 | if (likely(result == 0)) | |
343 | goto out; | |
344 | } | |
345 | ||
7b2d81d4 | 346 | result = read_opcode(mm, vaddr, &opcode); |
2b144498 SD |
347 | if (result) |
348 | return result; | |
c00b2750 | 349 | out: |
5cb4ac3a | 350 | if (is_swbp_insn(&opcode)) |
2b144498 SD |
351 | return 1; |
352 | ||
353 | return 0; | |
354 | } | |
355 | ||
356 | /** | |
5cb4ac3a | 357 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 358 | * @auprobe: arch specific probepoint information. |
2b144498 | 359 | * @mm: the probed process address space. |
2b144498 SD |
360 | * @vaddr: the virtual address to insert the opcode. |
361 | * | |
362 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
363 | * Return 0 (success) or a negative errno. | |
364 | */ | |
5cb4ac3a | 365 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 366 | { |
7b2d81d4 | 367 | int result; |
2b144498 | 368 | |
5cb4ac3a | 369 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
370 | if (result == 1) |
371 | return -EEXIST; | |
372 | ||
373 | if (result) | |
374 | return result; | |
375 | ||
5cb4ac3a | 376 | return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
377 | } |
378 | ||
379 | /** | |
380 | * set_orig_insn - Restore the original instruction. | |
381 | * @mm: the probed process address space. | |
e3343e6a | 382 | * @auprobe: arch specific probepoint information. |
2b144498 SD |
383 | * @vaddr: the virtual address to insert the opcode. |
384 | * @verify: if true, verify existance of breakpoint instruction. | |
385 | * | |
386 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
387 | * Return 0 (success) or a negative errno. | |
388 | */ | |
7b2d81d4 | 389 | int __weak |
e3343e6a | 390 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify) |
2b144498 SD |
391 | { |
392 | if (verify) { | |
7b2d81d4 | 393 | int result; |
2b144498 | 394 | |
5cb4ac3a | 395 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
396 | if (!result) |
397 | return -EINVAL; | |
398 | ||
399 | if (result != 1) | |
400 | return result; | |
401 | } | |
e3343e6a | 402 | return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
403 | } |
404 | ||
405 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
406 | { | |
407 | if (l->inode < r->inode) | |
408 | return -1; | |
7b2d81d4 | 409 | |
2b144498 SD |
410 | if (l->inode > r->inode) |
411 | return 1; | |
2b144498 | 412 | |
7b2d81d4 IM |
413 | if (l->offset < r->offset) |
414 | return -1; | |
415 | ||
416 | if (l->offset > r->offset) | |
417 | return 1; | |
2b144498 SD |
418 | |
419 | return 0; | |
420 | } | |
421 | ||
422 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
423 | { | |
424 | struct uprobe u = { .inode = inode, .offset = offset }; | |
425 | struct rb_node *n = uprobes_tree.rb_node; | |
426 | struct uprobe *uprobe; | |
427 | int match; | |
428 | ||
429 | while (n) { | |
430 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
431 | match = match_uprobe(&u, uprobe); | |
432 | if (!match) { | |
433 | atomic_inc(&uprobe->ref); | |
434 | return uprobe; | |
435 | } | |
7b2d81d4 | 436 | |
2b144498 SD |
437 | if (match < 0) |
438 | n = n->rb_left; | |
439 | else | |
440 | n = n->rb_right; | |
441 | } | |
442 | return NULL; | |
443 | } | |
444 | ||
445 | /* | |
446 | * Find a uprobe corresponding to a given inode:offset | |
447 | * Acquires uprobes_treelock | |
448 | */ | |
449 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
450 | { | |
451 | struct uprobe *uprobe; | |
452 | unsigned long flags; | |
453 | ||
454 | spin_lock_irqsave(&uprobes_treelock, flags); | |
455 | uprobe = __find_uprobe(inode, offset); | |
456 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 457 | |
2b144498 SD |
458 | return uprobe; |
459 | } | |
460 | ||
461 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
462 | { | |
463 | struct rb_node **p = &uprobes_tree.rb_node; | |
464 | struct rb_node *parent = NULL; | |
465 | struct uprobe *u; | |
466 | int match; | |
467 | ||
468 | while (*p) { | |
469 | parent = *p; | |
470 | u = rb_entry(parent, struct uprobe, rb_node); | |
471 | match = match_uprobe(uprobe, u); | |
472 | if (!match) { | |
473 | atomic_inc(&u->ref); | |
474 | return u; | |
475 | } | |
476 | ||
477 | if (match < 0) | |
478 | p = &parent->rb_left; | |
479 | else | |
480 | p = &parent->rb_right; | |
481 | ||
482 | } | |
7b2d81d4 | 483 | |
2b144498 SD |
484 | u = NULL; |
485 | rb_link_node(&uprobe->rb_node, parent, p); | |
486 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
487 | /* get access + creation ref */ | |
488 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 489 | |
2b144498 SD |
490 | return u; |
491 | } | |
492 | ||
493 | /* | |
7b2d81d4 | 494 | * Acquire uprobes_treelock. |
2b144498 SD |
495 | * Matching uprobe already exists in rbtree; |
496 | * increment (access refcount) and return the matching uprobe. | |
497 | * | |
498 | * No matching uprobe; insert the uprobe in rb_tree; | |
499 | * get a double refcount (access + creation) and return NULL. | |
500 | */ | |
501 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
502 | { | |
503 | unsigned long flags; | |
504 | struct uprobe *u; | |
505 | ||
506 | spin_lock_irqsave(&uprobes_treelock, flags); | |
507 | u = __insert_uprobe(uprobe); | |
508 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 509 | |
0326f5a9 SD |
510 | /* For now assume that the instruction need not be single-stepped */ |
511 | uprobe->flags |= UPROBE_SKIP_SSTEP; | |
512 | ||
2b144498 SD |
513 | return u; |
514 | } | |
515 | ||
516 | static void put_uprobe(struct uprobe *uprobe) | |
517 | { | |
518 | if (atomic_dec_and_test(&uprobe->ref)) | |
519 | kfree(uprobe); | |
520 | } | |
521 | ||
522 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
523 | { | |
524 | struct uprobe *uprobe, *cur_uprobe; | |
525 | ||
526 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
527 | if (!uprobe) | |
528 | return NULL; | |
529 | ||
530 | uprobe->inode = igrab(inode); | |
531 | uprobe->offset = offset; | |
532 | init_rwsem(&uprobe->consumer_rwsem); | |
533 | INIT_LIST_HEAD(&uprobe->pending_list); | |
534 | ||
535 | /* add to uprobes_tree, sorted on inode:offset */ | |
536 | cur_uprobe = insert_uprobe(uprobe); | |
537 | ||
538 | /* a uprobe exists for this inode:offset combination */ | |
539 | if (cur_uprobe) { | |
540 | kfree(uprobe); | |
541 | uprobe = cur_uprobe; | |
542 | iput(inode); | |
7b2d81d4 | 543 | } else { |
2b144498 | 544 | atomic_inc(&uprobe_events); |
7b2d81d4 IM |
545 | } |
546 | ||
2b144498 SD |
547 | return uprobe; |
548 | } | |
549 | ||
0326f5a9 SD |
550 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
551 | { | |
552 | struct uprobe_consumer *uc; | |
553 | ||
554 | if (!(uprobe->flags & UPROBE_RUN_HANDLER)) | |
555 | return; | |
556 | ||
557 | down_read(&uprobe->consumer_rwsem); | |
558 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
559 | if (!uc->filter || uc->filter(uc, current)) | |
560 | uc->handler(uc, regs); | |
561 | } | |
562 | up_read(&uprobe->consumer_rwsem); | |
563 | } | |
564 | ||
2b144498 | 565 | /* Returns the previous consumer */ |
7b2d81d4 | 566 | static struct uprobe_consumer * |
e3343e6a | 567 | consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
568 | { |
569 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
570 | uc->next = uprobe->consumers; |
571 | uprobe->consumers = uc; | |
2b144498 | 572 | up_write(&uprobe->consumer_rwsem); |
7b2d81d4 | 573 | |
e3343e6a | 574 | return uc->next; |
2b144498 SD |
575 | } |
576 | ||
577 | /* | |
e3343e6a SD |
578 | * For uprobe @uprobe, delete the consumer @uc. |
579 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
580 | * or return false. |
581 | */ | |
e3343e6a | 582 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
583 | { |
584 | struct uprobe_consumer **con; | |
585 | bool ret = false; | |
586 | ||
587 | down_write(&uprobe->consumer_rwsem); | |
588 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
589 | if (*con == uc) { |
590 | *con = uc->next; | |
2b144498 SD |
591 | ret = true; |
592 | break; | |
593 | } | |
594 | } | |
595 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 596 | |
2b144498 SD |
597 | return ret; |
598 | } | |
599 | ||
e3343e6a SD |
600 | static int |
601 | __copy_insn(struct address_space *mapping, struct vm_area_struct *vma, char *insn, | |
2b144498 SD |
602 | unsigned long nbytes, unsigned long offset) |
603 | { | |
604 | struct file *filp = vma->vm_file; | |
605 | struct page *page; | |
606 | void *vaddr; | |
607 | unsigned long off1; | |
608 | unsigned long idx; | |
609 | ||
610 | if (!filp) | |
611 | return -EINVAL; | |
612 | ||
613 | idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT); | |
614 | off1 = offset &= ~PAGE_MASK; | |
615 | ||
616 | /* | |
617 | * Ensure that the page that has the original instruction is | |
618 | * populated and in page-cache. | |
619 | */ | |
620 | page = read_mapping_page(mapping, idx, filp); | |
621 | if (IS_ERR(page)) | |
622 | return PTR_ERR(page); | |
623 | ||
624 | vaddr = kmap_atomic(page); | |
625 | memcpy(insn, vaddr + off1, nbytes); | |
626 | kunmap_atomic(vaddr); | |
627 | page_cache_release(page); | |
7b2d81d4 | 628 | |
2b144498 SD |
629 | return 0; |
630 | } | |
631 | ||
e3343e6a SD |
632 | static int |
633 | copy_insn(struct uprobe *uprobe, struct vm_area_struct *vma, unsigned long addr) | |
2b144498 SD |
634 | { |
635 | struct address_space *mapping; | |
2b144498 | 636 | unsigned long nbytes; |
7b2d81d4 | 637 | int bytes; |
2b144498 SD |
638 | |
639 | addr &= ~PAGE_MASK; | |
640 | nbytes = PAGE_SIZE - addr; | |
641 | mapping = uprobe->inode->i_mapping; | |
642 | ||
643 | /* Instruction at end of binary; copy only available bytes */ | |
644 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
645 | bytes = uprobe->inode->i_size - uprobe->offset; | |
646 | else | |
647 | bytes = MAX_UINSN_BYTES; | |
648 | ||
649 | /* Instruction at the page-boundary; copy bytes in second page */ | |
650 | if (nbytes < bytes) { | |
3ff54efd | 651 | if (__copy_insn(mapping, vma, uprobe->arch.insn + nbytes, |
2b144498 SD |
652 | bytes - nbytes, uprobe->offset + nbytes)) |
653 | return -ENOMEM; | |
654 | ||
655 | bytes = nbytes; | |
656 | } | |
3ff54efd | 657 | return __copy_insn(mapping, vma, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
658 | } |
659 | ||
682968e0 SD |
660 | /* |
661 | * How mm->uprobes_state.count gets updated | |
662 | * uprobe_mmap() increments the count if | |
663 | * - it successfully adds a breakpoint. | |
664 | * - it cannot add a breakpoint, but sees that there is a underlying | |
665 | * breakpoint (via a is_swbp_at_addr()). | |
666 | * | |
667 | * uprobe_munmap() decrements the count if | |
668 | * - it sees a underlying breakpoint, (via is_swbp_at_addr) | |
669 | * (Subsequent uprobe_unregister wouldnt find the breakpoint | |
670 | * unless a uprobe_mmap kicks in, since the old vma would be | |
671 | * dropped just after uprobe_munmap.) | |
672 | * | |
673 | * uprobe_register increments the count if: | |
674 | * - it successfully adds a breakpoint. | |
675 | * | |
676 | * uprobe_unregister decrements the count if: | |
677 | * - it sees a underlying breakpoint and removes successfully. | |
678 | * (via is_swbp_at_addr) | |
679 | * (Subsequent uprobe_munmap wouldnt find the breakpoint | |
680 | * since there is no underlying breakpoint after the | |
681 | * breakpoint removal.) | |
682 | */ | |
e3343e6a SD |
683 | static int |
684 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
685 | struct vm_area_struct *vma, loff_t vaddr) | |
2b144498 SD |
686 | { |
687 | unsigned long addr; | |
688 | int ret; | |
689 | ||
690 | /* | |
691 | * If probe is being deleted, unregister thread could be done with | |
692 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
693 | * nobody will be able to cleanup. Also we could be from fork or | |
694 | * mremap path, where the probe might have already been inserted. | |
695 | * Hence behave as if probe already existed. | |
696 | */ | |
697 | if (!uprobe->consumers) | |
698 | return -EEXIST; | |
699 | ||
700 | addr = (unsigned long)vaddr; | |
7b2d81d4 | 701 | |
900771a4 | 702 | if (!(uprobe->flags & UPROBE_COPY_INSN)) { |
2b144498 SD |
703 | ret = copy_insn(uprobe, vma, addr); |
704 | if (ret) | |
705 | return ret; | |
706 | ||
5cb4ac3a | 707 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) |
2b144498 SD |
708 | return -EEXIST; |
709 | ||
0326f5a9 | 710 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm); |
2b144498 SD |
711 | if (ret) |
712 | return ret; | |
713 | ||
900771a4 | 714 | uprobe->flags |= UPROBE_COPY_INSN; |
2b144498 | 715 | } |
682968e0 SD |
716 | |
717 | /* | |
718 | * Ideally, should be updating the probe count after the breakpoint | |
719 | * has been successfully inserted. However a thread could hit the | |
720 | * breakpoint we just inserted even before the probe count is | |
721 | * incremented. If this is the first breakpoint placed, breakpoint | |
722 | * notifier might ignore uprobes and pass the trap to the thread. | |
723 | * Hence increment before and decrement on failure. | |
724 | */ | |
725 | atomic_inc(&mm->uprobes_state.count); | |
5cb4ac3a | 726 | ret = set_swbp(&uprobe->arch, mm, addr); |
682968e0 SD |
727 | if (ret) |
728 | atomic_dec(&mm->uprobes_state.count); | |
2b144498 SD |
729 | |
730 | return ret; | |
731 | } | |
732 | ||
e3343e6a SD |
733 | static void |
734 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr) | |
2b144498 | 735 | { |
682968e0 SD |
736 | if (!set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true)) |
737 | atomic_dec(&mm->uprobes_state.count); | |
2b144498 SD |
738 | } |
739 | ||
0326f5a9 SD |
740 | /* |
741 | * There could be threads that have hit the breakpoint and are entering the | |
742 | * notifier code and trying to acquire the uprobes_treelock. The thread | |
743 | * calling delete_uprobe() that is removing the uprobe from the rb_tree can | |
744 | * race with these threads and might acquire the uprobes_treelock compared | |
745 | * to some of the breakpoint hit threads. In such a case, the breakpoint | |
746 | * hit threads will not find the uprobe. The current unregistering thread | |
747 | * waits till all other threads have hit a breakpoint, to acquire the | |
748 | * uprobes_treelock before the uprobe is removed from the rbtree. | |
749 | */ | |
2b144498 SD |
750 | static void delete_uprobe(struct uprobe *uprobe) |
751 | { | |
752 | unsigned long flags; | |
753 | ||
0326f5a9 | 754 | synchronize_srcu(&uprobes_srcu); |
2b144498 SD |
755 | spin_lock_irqsave(&uprobes_treelock, flags); |
756 | rb_erase(&uprobe->rb_node, &uprobes_tree); | |
757 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
758 | iput(uprobe->inode); | |
759 | put_uprobe(uprobe); | |
760 | atomic_dec(&uprobe_events); | |
761 | } | |
762 | ||
e3343e6a SD |
763 | static struct vma_info * |
764 | __find_next_vma_info(struct address_space *mapping, struct list_head *head, | |
765 | struct vma_info *vi, loff_t offset, bool is_register) | |
2b144498 SD |
766 | { |
767 | struct prio_tree_iter iter; | |
768 | struct vm_area_struct *vma; | |
769 | struct vma_info *tmpvi; | |
7b2d81d4 | 770 | unsigned long pgoff; |
2b144498 | 771 | int existing_vma; |
7b2d81d4 IM |
772 | loff_t vaddr; |
773 | ||
774 | pgoff = offset >> PAGE_SHIFT; | |
2b144498 SD |
775 | |
776 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
777 | if (!valid_vma(vma, is_register)) | |
778 | continue; | |
779 | ||
780 | existing_vma = 0; | |
781 | vaddr = vma_address(vma, offset); | |
7b2d81d4 | 782 | |
2b144498 SD |
783 | list_for_each_entry(tmpvi, head, probe_list) { |
784 | if (tmpvi->mm == vma->vm_mm && tmpvi->vaddr == vaddr) { | |
785 | existing_vma = 1; | |
786 | break; | |
787 | } | |
788 | } | |
789 | ||
790 | /* | |
791 | * Another vma needs a probe to be installed. However skip | |
792 | * installing the probe if the vma is about to be unlinked. | |
793 | */ | |
7b2d81d4 | 794 | if (!existing_vma && atomic_inc_not_zero(&vma->vm_mm->mm_users)) { |
2b144498 SD |
795 | vi->mm = vma->vm_mm; |
796 | vi->vaddr = vaddr; | |
797 | list_add(&vi->probe_list, head); | |
7b2d81d4 | 798 | |
2b144498 SD |
799 | return vi; |
800 | } | |
801 | } | |
7b2d81d4 | 802 | |
2b144498 SD |
803 | return NULL; |
804 | } | |
805 | ||
806 | /* | |
807 | * Iterate in the rmap prio tree and find a vma where a probe has not | |
808 | * yet been inserted. | |
809 | */ | |
7b2d81d4 | 810 | static struct vma_info * |
e3343e6a SD |
811 | find_next_vma_info(struct address_space *mapping, struct list_head *head, |
812 | loff_t offset, bool is_register) | |
2b144498 SD |
813 | { |
814 | struct vma_info *vi, *retvi; | |
7b2d81d4 | 815 | |
2b144498 SD |
816 | vi = kzalloc(sizeof(struct vma_info), GFP_KERNEL); |
817 | if (!vi) | |
818 | return ERR_PTR(-ENOMEM); | |
819 | ||
820 | mutex_lock(&mapping->i_mmap_mutex); | |
e3343e6a | 821 | retvi = __find_next_vma_info(mapping, head, vi, offset, is_register); |
2b144498 SD |
822 | mutex_unlock(&mapping->i_mmap_mutex); |
823 | ||
824 | if (!retvi) | |
825 | kfree(vi); | |
7b2d81d4 | 826 | |
2b144498 SD |
827 | return retvi; |
828 | } | |
829 | ||
830 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
831 | { | |
832 | struct list_head try_list; | |
833 | struct vm_area_struct *vma; | |
834 | struct address_space *mapping; | |
835 | struct vma_info *vi, *tmpvi; | |
836 | struct mm_struct *mm; | |
837 | loff_t vaddr; | |
7b2d81d4 | 838 | int ret; |
2b144498 SD |
839 | |
840 | mapping = uprobe->inode->i_mapping; | |
841 | INIT_LIST_HEAD(&try_list); | |
7b2d81d4 IM |
842 | |
843 | ret = 0; | |
844 | ||
845 | for (;;) { | |
e3343e6a | 846 | vi = find_next_vma_info(mapping, &try_list, uprobe->offset, is_register); |
7b2d81d4 IM |
847 | if (!vi) |
848 | break; | |
849 | ||
2b144498 SD |
850 | if (IS_ERR(vi)) { |
851 | ret = PTR_ERR(vi); | |
852 | break; | |
853 | } | |
7b2d81d4 | 854 | |
2b144498 SD |
855 | mm = vi->mm; |
856 | down_read(&mm->mmap_sem); | |
857 | vma = find_vma(mm, (unsigned long)vi->vaddr); | |
858 | if (!vma || !valid_vma(vma, is_register)) { | |
859 | list_del(&vi->probe_list); | |
860 | kfree(vi); | |
861 | up_read(&mm->mmap_sem); | |
862 | mmput(mm); | |
863 | continue; | |
864 | } | |
865 | vaddr = vma_address(vma, uprobe->offset); | |
866 | if (vma->vm_file->f_mapping->host != uprobe->inode || | |
867 | vaddr != vi->vaddr) { | |
868 | list_del(&vi->probe_list); | |
869 | kfree(vi); | |
870 | up_read(&mm->mmap_sem); | |
871 | mmput(mm); | |
872 | continue; | |
873 | } | |
874 | ||
875 | if (is_register) | |
e3343e6a | 876 | ret = install_breakpoint(uprobe, mm, vma, vi->vaddr); |
2b144498 | 877 | else |
e3343e6a | 878 | remove_breakpoint(uprobe, mm, vi->vaddr); |
2b144498 SD |
879 | |
880 | up_read(&mm->mmap_sem); | |
881 | mmput(mm); | |
882 | if (is_register) { | |
883 | if (ret && ret == -EEXIST) | |
884 | ret = 0; | |
885 | if (ret) | |
886 | break; | |
887 | } | |
888 | } | |
7b2d81d4 | 889 | |
2b144498 SD |
890 | list_for_each_entry_safe(vi, tmpvi, &try_list, probe_list) { |
891 | list_del(&vi->probe_list); | |
892 | kfree(vi); | |
893 | } | |
7b2d81d4 | 894 | |
2b144498 SD |
895 | return ret; |
896 | } | |
897 | ||
7b2d81d4 | 898 | static int __uprobe_register(struct uprobe *uprobe) |
2b144498 SD |
899 | { |
900 | return register_for_each_vma(uprobe, true); | |
901 | } | |
902 | ||
7b2d81d4 | 903 | static void __uprobe_unregister(struct uprobe *uprobe) |
2b144498 SD |
904 | { |
905 | if (!register_for_each_vma(uprobe, false)) | |
906 | delete_uprobe(uprobe); | |
907 | ||
908 | /* TODO : cant unregister? schedule a worker thread */ | |
909 | } | |
910 | ||
911 | /* | |
7b2d81d4 | 912 | * uprobe_register - register a probe |
2b144498 SD |
913 | * @inode: the file in which the probe has to be placed. |
914 | * @offset: offset from the start of the file. | |
e3343e6a | 915 | * @uc: information on howto handle the probe.. |
2b144498 | 916 | * |
7b2d81d4 | 917 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
918 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
919 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 920 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 921 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 922 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
923 | * unregisters. |
924 | * | |
925 | * Return errno if it cannot successully install probes | |
926 | * else return 0 (success) | |
927 | */ | |
e3343e6a | 928 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
929 | { |
930 | struct uprobe *uprobe; | |
7b2d81d4 | 931 | int ret; |
2b144498 | 932 | |
e3343e6a | 933 | if (!inode || !uc || uc->next) |
7b2d81d4 | 934 | return -EINVAL; |
2b144498 SD |
935 | |
936 | if (offset > i_size_read(inode)) | |
7b2d81d4 | 937 | return -EINVAL; |
2b144498 SD |
938 | |
939 | ret = 0; | |
940 | mutex_lock(uprobes_hash(inode)); | |
941 | uprobe = alloc_uprobe(inode, offset); | |
7b2d81d4 | 942 | |
e3343e6a | 943 | if (uprobe && !consumer_add(uprobe, uc)) { |
7b2d81d4 | 944 | ret = __uprobe_register(uprobe); |
2b144498 SD |
945 | if (ret) { |
946 | uprobe->consumers = NULL; | |
7b2d81d4 IM |
947 | __uprobe_unregister(uprobe); |
948 | } else { | |
900771a4 | 949 | uprobe->flags |= UPROBE_RUN_HANDLER; |
7b2d81d4 | 950 | } |
2b144498 SD |
951 | } |
952 | ||
953 | mutex_unlock(uprobes_hash(inode)); | |
954 | put_uprobe(uprobe); | |
955 | ||
956 | return ret; | |
957 | } | |
958 | ||
959 | /* | |
7b2d81d4 | 960 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
961 | * @inode: the file in which the probe has to be removed. |
962 | * @offset: offset from the start of the file. | |
e3343e6a | 963 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 964 | */ |
e3343e6a | 965 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 966 | { |
7b2d81d4 | 967 | struct uprobe *uprobe; |
2b144498 | 968 | |
e3343e6a | 969 | if (!inode || !uc) |
2b144498 SD |
970 | return; |
971 | ||
972 | uprobe = find_uprobe(inode, offset); | |
973 | if (!uprobe) | |
974 | return; | |
975 | ||
976 | mutex_lock(uprobes_hash(inode)); | |
2b144498 | 977 | |
e3343e6a | 978 | if (consumer_del(uprobe, uc)) { |
7b2d81d4 IM |
979 | if (!uprobe->consumers) { |
980 | __uprobe_unregister(uprobe); | |
900771a4 | 981 | uprobe->flags &= ~UPROBE_RUN_HANDLER; |
7b2d81d4 | 982 | } |
2b144498 SD |
983 | } |
984 | ||
2b144498 SD |
985 | mutex_unlock(uprobes_hash(inode)); |
986 | if (uprobe) | |
987 | put_uprobe(uprobe); | |
988 | } | |
989 | ||
990 | /* | |
991 | * Of all the nodes that correspond to the given inode, return the node | |
992 | * with the least offset. | |
993 | */ | |
994 | static struct rb_node *find_least_offset_node(struct inode *inode) | |
995 | { | |
996 | struct uprobe u = { .inode = inode, .offset = 0}; | |
997 | struct rb_node *n = uprobes_tree.rb_node; | |
998 | struct rb_node *close_node = NULL; | |
999 | struct uprobe *uprobe; | |
1000 | int match; | |
1001 | ||
1002 | while (n) { | |
1003 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
1004 | match = match_uprobe(&u, uprobe); | |
7b2d81d4 | 1005 | |
2b144498 SD |
1006 | if (uprobe->inode == inode) |
1007 | close_node = n; | |
1008 | ||
1009 | if (!match) | |
1010 | return close_node; | |
1011 | ||
1012 | if (match < 0) | |
1013 | n = n->rb_left; | |
1014 | else | |
1015 | n = n->rb_right; | |
1016 | } | |
7b2d81d4 | 1017 | |
2b144498 SD |
1018 | return close_node; |
1019 | } | |
1020 | ||
1021 | /* | |
1022 | * For a given inode, build a list of probes that need to be inserted. | |
1023 | */ | |
1024 | static void build_probe_list(struct inode *inode, struct list_head *head) | |
1025 | { | |
1026 | struct uprobe *uprobe; | |
2b144498 | 1027 | unsigned long flags; |
7b2d81d4 | 1028 | struct rb_node *n; |
2b144498 SD |
1029 | |
1030 | spin_lock_irqsave(&uprobes_treelock, flags); | |
7b2d81d4 | 1031 | |
2b144498 | 1032 | n = find_least_offset_node(inode); |
7b2d81d4 | 1033 | |
2b144498 SD |
1034 | for (; n; n = rb_next(n)) { |
1035 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
1036 | if (uprobe->inode != inode) | |
1037 | break; | |
1038 | ||
1039 | list_add(&uprobe->pending_list, head); | |
1040 | atomic_inc(&uprobe->ref); | |
1041 | } | |
7b2d81d4 | 1042 | |
2b144498 SD |
1043 | spin_unlock_irqrestore(&uprobes_treelock, flags); |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Called from mmap_region. | |
1048 | * called with mm->mmap_sem acquired. | |
1049 | * | |
1050 | * Return -ve no if we fail to insert probes and we cannot | |
1051 | * bail-out. | |
7b2d81d4 IM |
1052 | * Return 0 otherwise. i.e: |
1053 | * | |
2b144498 SD |
1054 | * - successful insertion of probes |
1055 | * - (or) no possible probes to be inserted. | |
1056 | * - (or) insertion of probes failed but we can bail-out. | |
1057 | */ | |
7b2d81d4 | 1058 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1059 | { |
1060 | struct list_head tmp_list; | |
1061 | struct uprobe *uprobe, *u; | |
1062 | struct inode *inode; | |
682968e0 | 1063 | int ret, count; |
2b144498 SD |
1064 | |
1065 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) | |
7b2d81d4 | 1066 | return 0; |
2b144498 SD |
1067 | |
1068 | inode = vma->vm_file->f_mapping->host; | |
1069 | if (!inode) | |
7b2d81d4 | 1070 | return 0; |
2b144498 SD |
1071 | |
1072 | INIT_LIST_HEAD(&tmp_list); | |
1073 | mutex_lock(uprobes_mmap_hash(inode)); | |
1074 | build_probe_list(inode, &tmp_list); | |
7b2d81d4 IM |
1075 | |
1076 | ret = 0; | |
682968e0 | 1077 | count = 0; |
7b2d81d4 | 1078 | |
2b144498 SD |
1079 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
1080 | loff_t vaddr; | |
1081 | ||
1082 | list_del(&uprobe->pending_list); | |
1083 | if (!ret) { | |
1084 | vaddr = vma_address(vma, uprobe->offset); | |
682968e0 SD |
1085 | |
1086 | if (vaddr < vma->vm_start || vaddr >= vma->vm_end) { | |
1087 | put_uprobe(uprobe); | |
1088 | continue; | |
2b144498 | 1089 | } |
682968e0 SD |
1090 | |
1091 | ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); | |
1092 | ||
1093 | /* Ignore double add: */ | |
1094 | if (ret == -EEXIST) { | |
1095 | ret = 0; | |
1096 | ||
1097 | if (!is_swbp_at_addr(vma->vm_mm, vaddr)) | |
1098 | continue; | |
1099 | ||
1100 | /* | |
1101 | * Unable to insert a breakpoint, but | |
1102 | * breakpoint lies underneath. Increment the | |
1103 | * probe count. | |
1104 | */ | |
1105 | atomic_inc(&vma->vm_mm->uprobes_state.count); | |
1106 | } | |
1107 | ||
1108 | if (!ret) | |
1109 | count++; | |
2b144498 SD |
1110 | } |
1111 | put_uprobe(uprobe); | |
1112 | } | |
1113 | ||
1114 | mutex_unlock(uprobes_mmap_hash(inode)); | |
1115 | ||
682968e0 SD |
1116 | if (ret) |
1117 | atomic_sub(count, &vma->vm_mm->uprobes_state.count); | |
1118 | ||
2b144498 SD |
1119 | return ret; |
1120 | } | |
1121 | ||
682968e0 SD |
1122 | /* |
1123 | * Called in context of a munmap of a vma. | |
1124 | */ | |
cbc91f71 | 1125 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 SD |
1126 | { |
1127 | struct list_head tmp_list; | |
1128 | struct uprobe *uprobe, *u; | |
1129 | struct inode *inode; | |
1130 | ||
1131 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) | |
1132 | return; | |
1133 | ||
1134 | if (!atomic_read(&vma->vm_mm->uprobes_state.count)) | |
1135 | return; | |
1136 | ||
1137 | inode = vma->vm_file->f_mapping->host; | |
1138 | if (!inode) | |
1139 | return; | |
1140 | ||
1141 | INIT_LIST_HEAD(&tmp_list); | |
1142 | mutex_lock(uprobes_mmap_hash(inode)); | |
1143 | build_probe_list(inode, &tmp_list); | |
1144 | ||
1145 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { | |
1146 | loff_t vaddr; | |
1147 | ||
1148 | list_del(&uprobe->pending_list); | |
1149 | vaddr = vma_address(vma, uprobe->offset); | |
1150 | ||
cbc91f71 | 1151 | if (vaddr >= start && vaddr < end) { |
682968e0 SD |
1152 | /* |
1153 | * An unregister could have removed the probe before | |
1154 | * unmap. So check before we decrement the count. | |
1155 | */ | |
1156 | if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1) | |
1157 | atomic_dec(&vma->vm_mm->uprobes_state.count); | |
1158 | } | |
1159 | put_uprobe(uprobe); | |
1160 | } | |
1161 | mutex_unlock(uprobes_mmap_hash(inode)); | |
1162 | } | |
1163 | ||
d4b3b638 SD |
1164 | /* Slot allocation for XOL */ |
1165 | static int xol_add_vma(struct xol_area *area) | |
1166 | { | |
1167 | struct mm_struct *mm; | |
1168 | int ret; | |
1169 | ||
1170 | area->page = alloc_page(GFP_HIGHUSER); | |
1171 | if (!area->page) | |
1172 | return -ENOMEM; | |
1173 | ||
1174 | ret = -EALREADY; | |
1175 | mm = current->mm; | |
1176 | ||
1177 | down_write(&mm->mmap_sem); | |
1178 | if (mm->uprobes_state.xol_area) | |
1179 | goto fail; | |
1180 | ||
1181 | ret = -ENOMEM; | |
1182 | ||
1183 | /* Try to map as high as possible, this is only a hint. */ | |
1184 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1185 | if (area->vaddr & ~PAGE_MASK) { | |
1186 | ret = area->vaddr; | |
1187 | goto fail; | |
1188 | } | |
1189 | ||
1190 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1191 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1192 | if (ret) | |
1193 | goto fail; | |
1194 | ||
1195 | smp_wmb(); /* pairs with get_xol_area() */ | |
1196 | mm->uprobes_state.xol_area = area; | |
1197 | ret = 0; | |
1198 | ||
1199 | fail: | |
1200 | up_write(&mm->mmap_sem); | |
1201 | if (ret) | |
1202 | __free_page(area->page); | |
1203 | ||
1204 | return ret; | |
1205 | } | |
1206 | ||
1207 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1208 | { | |
1209 | struct xol_area *area; | |
1210 | ||
1211 | area = mm->uprobes_state.xol_area; | |
1212 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1213 | ||
1214 | return area; | |
1215 | } | |
1216 | ||
1217 | /* | |
1218 | * xol_alloc_area - Allocate process's xol_area. | |
1219 | * This area will be used for storing instructions for execution out of | |
1220 | * line. | |
1221 | * | |
1222 | * Returns the allocated area or NULL. | |
1223 | */ | |
1224 | static struct xol_area *xol_alloc_area(void) | |
1225 | { | |
1226 | struct xol_area *area; | |
1227 | ||
1228 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1229 | if (unlikely(!area)) | |
1230 | return NULL; | |
1231 | ||
1232 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1233 | ||
1234 | if (!area->bitmap) | |
1235 | goto fail; | |
1236 | ||
1237 | init_waitqueue_head(&area->wq); | |
1238 | if (!xol_add_vma(area)) | |
1239 | return area; | |
1240 | ||
1241 | fail: | |
1242 | kfree(area->bitmap); | |
1243 | kfree(area); | |
1244 | ||
1245 | return get_xol_area(current->mm); | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * uprobe_clear_state - Free the area allocated for slots. | |
1250 | */ | |
1251 | void uprobe_clear_state(struct mm_struct *mm) | |
1252 | { | |
1253 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1254 | ||
1255 | if (!area) | |
1256 | return; | |
1257 | ||
1258 | put_page(area->page); | |
1259 | kfree(area->bitmap); | |
1260 | kfree(area); | |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * uprobe_reset_state - Free the area allocated for slots. | |
1265 | */ | |
1266 | void uprobe_reset_state(struct mm_struct *mm) | |
1267 | { | |
1268 | mm->uprobes_state.xol_area = NULL; | |
682968e0 | 1269 | atomic_set(&mm->uprobes_state.count, 0); |
d4b3b638 SD |
1270 | } |
1271 | ||
1272 | /* | |
1273 | * - search for a free slot. | |
1274 | */ | |
1275 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1276 | { | |
1277 | unsigned long slot_addr; | |
1278 | int slot_nr; | |
1279 | ||
1280 | do { | |
1281 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1282 | if (slot_nr < UINSNS_PER_PAGE) { | |
1283 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1284 | break; | |
1285 | ||
1286 | slot_nr = UINSNS_PER_PAGE; | |
1287 | continue; | |
1288 | } | |
1289 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1290 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1291 | ||
1292 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1293 | atomic_inc(&area->slot_count); | |
1294 | ||
1295 | return slot_addr; | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * xol_get_insn_slot - If was not allocated a slot, then | |
1300 | * allocate a slot. | |
1301 | * Returns the allocated slot address or 0. | |
1302 | */ | |
1303 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1304 | { | |
1305 | struct xol_area *area; | |
1306 | unsigned long offset; | |
1307 | void *vaddr; | |
1308 | ||
1309 | area = get_xol_area(current->mm); | |
1310 | if (!area) { | |
1311 | area = xol_alloc_area(); | |
1312 | if (!area) | |
1313 | return 0; | |
1314 | } | |
1315 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1316 | ||
1317 | /* | |
1318 | * Initialize the slot if xol_vaddr points to valid | |
1319 | * instruction slot. | |
1320 | */ | |
1321 | if (unlikely(!current->utask->xol_vaddr)) | |
1322 | return 0; | |
1323 | ||
1324 | current->utask->vaddr = slot_addr; | |
1325 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1326 | vaddr = kmap_atomic(area->page); | |
1327 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1328 | kunmap_atomic(vaddr); | |
1329 | ||
1330 | return current->utask->xol_vaddr; | |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * xol_free_insn_slot - If slot was earlier allocated by | |
1335 | * @xol_get_insn_slot(), make the slot available for | |
1336 | * subsequent requests. | |
1337 | */ | |
1338 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1339 | { | |
1340 | struct xol_area *area; | |
1341 | unsigned long vma_end; | |
1342 | unsigned long slot_addr; | |
1343 | ||
1344 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1345 | return; | |
1346 | ||
1347 | slot_addr = tsk->utask->xol_vaddr; | |
1348 | ||
1349 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1350 | return; | |
1351 | ||
1352 | area = tsk->mm->uprobes_state.xol_area; | |
1353 | vma_end = area->vaddr + PAGE_SIZE; | |
1354 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1355 | unsigned long offset; | |
1356 | int slot_nr; | |
1357 | ||
1358 | offset = slot_addr - area->vaddr; | |
1359 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1360 | if (slot_nr >= UINSNS_PER_PAGE) | |
1361 | return; | |
1362 | ||
1363 | clear_bit(slot_nr, area->bitmap); | |
1364 | atomic_dec(&area->slot_count); | |
1365 | if (waitqueue_active(&area->wq)) | |
1366 | wake_up(&area->wq); | |
1367 | ||
1368 | tsk->utask->xol_vaddr = 0; | |
1369 | } | |
1370 | } | |
1371 | ||
0326f5a9 SD |
1372 | /** |
1373 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1374 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1375 | * instruction. | |
1376 | * Return the address of the breakpoint instruction. | |
1377 | */ | |
1378 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1379 | { | |
1380 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1381 | } | |
1382 | ||
1383 | /* | |
1384 | * Called with no locks held. | |
1385 | * Called in context of a exiting or a exec-ing thread. | |
1386 | */ | |
1387 | void uprobe_free_utask(struct task_struct *t) | |
1388 | { | |
1389 | struct uprobe_task *utask = t->utask; | |
1390 | ||
1391 | if (t->uprobe_srcu_id != -1) | |
1392 | srcu_read_unlock_raw(&uprobes_srcu, t->uprobe_srcu_id); | |
1393 | ||
1394 | if (!utask) | |
1395 | return; | |
1396 | ||
1397 | if (utask->active_uprobe) | |
1398 | put_uprobe(utask->active_uprobe); | |
1399 | ||
d4b3b638 | 1400 | xol_free_insn_slot(t); |
0326f5a9 SD |
1401 | kfree(utask); |
1402 | t->utask = NULL; | |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * Called in context of a new clone/fork from copy_process. | |
1407 | */ | |
1408 | void uprobe_copy_process(struct task_struct *t) | |
1409 | { | |
1410 | t->utask = NULL; | |
1411 | t->uprobe_srcu_id = -1; | |
1412 | } | |
1413 | ||
1414 | /* | |
1415 | * Allocate a uprobe_task object for the task. | |
1416 | * Called when the thread hits a breakpoint for the first time. | |
1417 | * | |
1418 | * Returns: | |
1419 | * - pointer to new uprobe_task on success | |
1420 | * - NULL otherwise | |
1421 | */ | |
1422 | static struct uprobe_task *add_utask(void) | |
1423 | { | |
1424 | struct uprobe_task *utask; | |
1425 | ||
1426 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1427 | if (unlikely(!utask)) | |
1428 | return NULL; | |
1429 | ||
1430 | utask->active_uprobe = NULL; | |
1431 | current->utask = utask; | |
1432 | return utask; | |
1433 | } | |
1434 | ||
1435 | /* Prepare to single-step probed instruction out of line. */ | |
1436 | static int | |
1437 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1438 | { | |
d4b3b638 SD |
1439 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1440 | return 0; | |
1441 | ||
0326f5a9 SD |
1442 | return -EFAULT; |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * If we are singlestepping, then ensure this thread is not connected to | |
1447 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1448 | * triggers the signal, restart the original insn even if the task is | |
1449 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1450 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1451 | * _same_ instruction should be repeated again after return from the signal | |
1452 | * handler, and SSTEP can never finish in this case. | |
1453 | */ | |
1454 | bool uprobe_deny_signal(void) | |
1455 | { | |
1456 | struct task_struct *t = current; | |
1457 | struct uprobe_task *utask = t->utask; | |
1458 | ||
1459 | if (likely(!utask || !utask->active_uprobe)) | |
1460 | return false; | |
1461 | ||
1462 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1463 | ||
1464 | if (signal_pending(t)) { | |
1465 | spin_lock_irq(&t->sighand->siglock); | |
1466 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1467 | spin_unlock_irq(&t->sighand->siglock); | |
1468 | ||
1469 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1470 | utask->state = UTASK_SSTEP_TRAPPED; | |
1471 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1472 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1473 | } | |
1474 | } | |
1475 | ||
1476 | return true; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Avoid singlestepping the original instruction if the original instruction | |
1481 | * is a NOP or can be emulated. | |
1482 | */ | |
1483 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1484 | { | |
1485 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) | |
1486 | return true; | |
1487 | ||
1488 | uprobe->flags &= ~UPROBE_SKIP_SSTEP; | |
1489 | return false; | |
1490 | } | |
1491 | ||
d790d346 | 1492 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1493 | { |
3a9ea052 ON |
1494 | struct mm_struct *mm = current->mm; |
1495 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1496 | struct vm_area_struct *vma; |
0326f5a9 | 1497 | |
0326f5a9 SD |
1498 | down_read(&mm->mmap_sem); |
1499 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1500 | if (vma && vma->vm_start <= bp_vaddr) { |
1501 | if (valid_vma(vma, false)) { | |
1502 | struct inode *inode; | |
1503 | loff_t offset; | |
0326f5a9 | 1504 | |
3a9ea052 ON |
1505 | inode = vma->vm_file->f_mapping->host; |
1506 | offset = bp_vaddr - vma->vm_start; | |
1507 | offset += (vma->vm_pgoff << PAGE_SHIFT); | |
1508 | uprobe = find_uprobe(inode, offset); | |
1509 | } | |
d790d346 ON |
1510 | |
1511 | if (!uprobe) | |
1512 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1513 | } else { | |
1514 | *is_swbp = -EFAULT; | |
0326f5a9 SD |
1515 | } |
1516 | ||
1517 | srcu_read_unlock_raw(&uprobes_srcu, current->uprobe_srcu_id); | |
1518 | current->uprobe_srcu_id = -1; | |
1519 | up_read(&mm->mmap_sem); | |
1520 | ||
3a9ea052 ON |
1521 | return uprobe; |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * Run handler and ask thread to singlestep. | |
1526 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1527 | */ | |
1528 | static void handle_swbp(struct pt_regs *regs) | |
1529 | { | |
1530 | struct uprobe_task *utask; | |
1531 | struct uprobe *uprobe; | |
1532 | unsigned long bp_vaddr; | |
d790d346 | 1533 | int is_swbp; |
3a9ea052 ON |
1534 | |
1535 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1536 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1537 | |
0326f5a9 SD |
1538 | if (!uprobe) { |
1539 | /* No matching uprobe; signal SIGTRAP. */ | |
1540 | send_sig(SIGTRAP, current, 0); | |
1541 | return; | |
1542 | } | |
1543 | ||
1544 | utask = current->utask; | |
1545 | if (!utask) { | |
1546 | utask = add_utask(); | |
1547 | /* Cannot allocate; re-execute the instruction. */ | |
1548 | if (!utask) | |
1549 | goto cleanup_ret; | |
1550 | } | |
1551 | utask->active_uprobe = uprobe; | |
1552 | handler_chain(uprobe, regs); | |
1553 | if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs)) | |
1554 | goto cleanup_ret; | |
1555 | ||
1556 | utask->state = UTASK_SSTEP; | |
1557 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { | |
1558 | user_enable_single_step(current); | |
1559 | return; | |
1560 | } | |
1561 | ||
1562 | cleanup_ret: | |
1563 | if (utask) { | |
1564 | utask->active_uprobe = NULL; | |
1565 | utask->state = UTASK_RUNNING; | |
1566 | } | |
1567 | if (uprobe) { | |
1568 | if (!(uprobe->flags & UPROBE_SKIP_SSTEP)) | |
1569 | ||
1570 | /* | |
1571 | * cannot singlestep; cannot skip instruction; | |
1572 | * re-execute the instruction. | |
1573 | */ | |
1574 | instruction_pointer_set(regs, bp_vaddr); | |
1575 | ||
1576 | put_uprobe(uprobe); | |
1577 | } | |
1578 | } | |
1579 | ||
1580 | /* | |
1581 | * Perform required fix-ups and disable singlestep. | |
1582 | * Allow pending signals to take effect. | |
1583 | */ | |
1584 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1585 | { | |
1586 | struct uprobe *uprobe; | |
1587 | ||
1588 | uprobe = utask->active_uprobe; | |
1589 | if (utask->state == UTASK_SSTEP_ACK) | |
1590 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1591 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1592 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1593 | else | |
1594 | WARN_ON_ONCE(1); | |
1595 | ||
1596 | put_uprobe(uprobe); | |
1597 | utask->active_uprobe = NULL; | |
1598 | utask->state = UTASK_RUNNING; | |
1599 | user_disable_single_step(current); | |
d4b3b638 | 1600 | xol_free_insn_slot(current); |
0326f5a9 SD |
1601 | |
1602 | spin_lock_irq(¤t->sighand->siglock); | |
1603 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1604 | spin_unlock_irq(¤t->sighand->siglock); | |
1605 | } | |
1606 | ||
1607 | /* | |
1608 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on | |
1609 | * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and | |
1610 | * allows the thread to return from interrupt. | |
1611 | * | |
1612 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and | |
1613 | * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from | |
1614 | * interrupt. | |
1615 | * | |
1616 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1617 | * uprobe_notify_resume(). | |
1618 | */ | |
1619 | void uprobe_notify_resume(struct pt_regs *regs) | |
1620 | { | |
1621 | struct uprobe_task *utask; | |
1622 | ||
1623 | utask = current->utask; | |
1624 | if (!utask || utask->state == UTASK_BP_HIT) | |
1625 | handle_swbp(regs); | |
1626 | else | |
1627 | handle_singlestep(utask, regs); | |
1628 | } | |
1629 | ||
1630 | /* | |
1631 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1632 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1633 | */ | |
1634 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1635 | { | |
1636 | struct uprobe_task *utask; | |
1637 | ||
682968e0 SD |
1638 | if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count)) |
1639 | /* task is currently not uprobed */ | |
0326f5a9 SD |
1640 | return 0; |
1641 | ||
1642 | utask = current->utask; | |
1643 | if (utask) | |
1644 | utask->state = UTASK_BP_HIT; | |
1645 | ||
1646 | set_thread_flag(TIF_UPROBE); | |
1647 | current->uprobe_srcu_id = srcu_read_lock_raw(&uprobes_srcu); | |
1648 | ||
1649 | return 1; | |
1650 | } | |
1651 | ||
1652 | /* | |
1653 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1654 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1655 | */ | |
1656 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1657 | { | |
1658 | struct uprobe_task *utask = current->utask; | |
1659 | ||
1660 | if (!current->mm || !utask || !utask->active_uprobe) | |
1661 | /* task is currently not uprobed */ | |
1662 | return 0; | |
1663 | ||
1664 | utask->state = UTASK_SSTEP_ACK; | |
1665 | set_thread_flag(TIF_UPROBE); | |
1666 | return 1; | |
1667 | } | |
1668 | ||
1669 | static struct notifier_block uprobe_exception_nb = { | |
1670 | .notifier_call = arch_uprobe_exception_notify, | |
1671 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1672 | }; | |
1673 | ||
2b144498 SD |
1674 | static int __init init_uprobes(void) |
1675 | { | |
1676 | int i; | |
1677 | ||
1678 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1679 | mutex_init(&uprobes_mutex[i]); | |
1680 | mutex_init(&uprobes_mmap_mutex[i]); | |
1681 | } | |
0326f5a9 SD |
1682 | init_srcu_struct(&uprobes_srcu); |
1683 | ||
1684 | return register_die_notifier(&uprobe_exception_nb); | |
2b144498 | 1685 | } |
0326f5a9 | 1686 | module_init(init_uprobes); |
2b144498 SD |
1687 | |
1688 | static void __exit exit_uprobes(void) | |
1689 | { | |
1690 | } | |
2b144498 | 1691 | module_exit(exit_uprobes); |