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 | ||
0326f5a9 | 38 | static struct srcu_struct uprobes_srcu; |
2b144498 | 39 | static struct rb_root uprobes_tree = RB_ROOT; |
7b2d81d4 | 40 | |
2b144498 SD |
41 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
42 | ||
43 | #define UPROBES_HASH_SZ 13 | |
7b2d81d4 | 44 | |
2b144498 SD |
45 | /* serialize (un)register */ |
46 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
47 | |
48 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
49 | |
50 | /* serialize uprobe->pending_list */ | |
51 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 52 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 SD |
53 | |
54 | /* | |
7b2d81d4 | 55 | * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe |
2b144498 SD |
56 | * events active at this time. Probably a fine grained per inode count is |
57 | * better? | |
58 | */ | |
59 | static atomic_t uprobe_events = ATOMIC_INIT(0); | |
60 | ||
61 | /* | |
62 | * Maintain a temporary per vma info that can be used to search if a vma | |
63 | * has already been handled. This structure is introduced since extending | |
64 | * vm_area_struct wasnt recommended. | |
65 | */ | |
66 | struct vma_info { | |
7b2d81d4 IM |
67 | struct list_head probe_list; |
68 | struct mm_struct *mm; | |
69 | loff_t vaddr; | |
2b144498 SD |
70 | }; |
71 | ||
3ff54efd SD |
72 | struct uprobe { |
73 | struct rb_node rb_node; /* node in the rb tree */ | |
74 | atomic_t ref; | |
75 | struct rw_semaphore consumer_rwsem; | |
76 | struct list_head pending_list; | |
77 | struct uprobe_consumer *consumers; | |
78 | struct inode *inode; /* Also hold a ref to inode */ | |
79 | loff_t offset; | |
80 | int flags; | |
81 | struct arch_uprobe arch; | |
82 | }; | |
83 | ||
2b144498 SD |
84 | /* |
85 | * valid_vma: Verify if the specified vma is an executable vma | |
86 | * Relax restrictions while unregistering: vm_flags might have | |
87 | * changed after breakpoint was inserted. | |
88 | * - is_register: indicates if we are in register context. | |
89 | * - Return 1 if the specified virtual address is in an | |
90 | * executable vma. | |
91 | */ | |
92 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
93 | { | |
94 | if (!vma->vm_file) | |
95 | return false; | |
96 | ||
97 | if (!is_register) | |
98 | return true; | |
99 | ||
7b2d81d4 | 100 | if ((vma->vm_flags & (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)) == (VM_READ|VM_EXEC)) |
2b144498 SD |
101 | return true; |
102 | ||
103 | return false; | |
104 | } | |
105 | ||
106 | static loff_t vma_address(struct vm_area_struct *vma, loff_t offset) | |
107 | { | |
108 | loff_t vaddr; | |
109 | ||
110 | vaddr = vma->vm_start + offset; | |
111 | vaddr -= vma->vm_pgoff << PAGE_SHIFT; | |
7b2d81d4 | 112 | |
2b144498 SD |
113 | return vaddr; |
114 | } | |
115 | ||
116 | /** | |
117 | * __replace_page - replace page in vma by new page. | |
118 | * based on replace_page in mm/ksm.c | |
119 | * | |
120 | * @vma: vma that holds the pte pointing to page | |
121 | * @page: the cowed page we are replacing by kpage | |
122 | * @kpage: the modified page we replace page by | |
123 | * | |
124 | * Returns 0 on success, -EFAULT on failure. | |
125 | */ | |
7b2d81d4 | 126 | static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage) |
2b144498 SD |
127 | { |
128 | struct mm_struct *mm = vma->vm_mm; | |
129 | pgd_t *pgd; | |
130 | pud_t *pud; | |
131 | pmd_t *pmd; | |
132 | pte_t *ptep; | |
133 | spinlock_t *ptl; | |
134 | unsigned long addr; | |
135 | int err = -EFAULT; | |
136 | ||
137 | addr = page_address_in_vma(page, vma); | |
138 | if (addr == -EFAULT) | |
139 | goto out; | |
140 | ||
141 | pgd = pgd_offset(mm, addr); | |
142 | if (!pgd_present(*pgd)) | |
143 | goto out; | |
144 | ||
145 | pud = pud_offset(pgd, addr); | |
146 | if (!pud_present(*pud)) | |
147 | goto out; | |
148 | ||
149 | pmd = pmd_offset(pud, addr); | |
150 | if (!pmd_present(*pmd)) | |
151 | goto out; | |
152 | ||
153 | ptep = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
154 | if (!ptep) | |
155 | goto out; | |
156 | ||
157 | get_page(kpage); | |
158 | page_add_new_anon_rmap(kpage, vma, addr); | |
159 | ||
160 | flush_cache_page(vma, addr, pte_pfn(*ptep)); | |
161 | ptep_clear_flush(vma, addr, ptep); | |
162 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
163 | ||
164 | page_remove_rmap(page); | |
165 | if (!page_mapped(page)) | |
166 | try_to_free_swap(page); | |
167 | put_page(page); | |
168 | pte_unmap_unlock(ptep, ptl); | |
169 | err = 0; | |
170 | ||
171 | out: | |
172 | return err; | |
173 | } | |
174 | ||
175 | /** | |
5cb4ac3a | 176 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 177 | * @insn: instruction to be checked. |
5cb4ac3a | 178 | * Default implementation of is_swbp_insn |
2b144498 SD |
179 | * Returns true if @insn is a breakpoint instruction. |
180 | */ | |
5cb4ac3a | 181 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 182 | { |
5cb4ac3a | 183 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
184 | } |
185 | ||
186 | /* | |
187 | * NOTE: | |
188 | * Expect the breakpoint instruction to be the smallest size instruction for | |
189 | * the architecture. If an arch has variable length instruction and the | |
190 | * breakpoint instruction is not of the smallest length instruction | |
191 | * supported by that architecture then we need to modify read_opcode / | |
192 | * write_opcode accordingly. This would never be a problem for archs that | |
193 | * have fixed length instructions. | |
194 | */ | |
195 | ||
196 | /* | |
197 | * write_opcode - write the opcode at a given virtual address. | |
e3343e6a | 198 | * @auprobe: arch breakpointing information. |
2b144498 | 199 | * @mm: the probed process address space. |
2b144498 SD |
200 | * @vaddr: the virtual address to store the opcode. |
201 | * @opcode: opcode to be written at @vaddr. | |
202 | * | |
203 | * Called with mm->mmap_sem held (for read and with a reference to | |
204 | * mm). | |
205 | * | |
206 | * For mm @mm, write the opcode at @vaddr. | |
207 | * Return 0 (success) or a negative errno. | |
208 | */ | |
e3343e6a | 209 | static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, |
2b144498 SD |
210 | unsigned long vaddr, uprobe_opcode_t opcode) |
211 | { | |
212 | struct page *old_page, *new_page; | |
213 | struct address_space *mapping; | |
214 | void *vaddr_old, *vaddr_new; | |
215 | struct vm_area_struct *vma; | |
3ff54efd | 216 | struct uprobe *uprobe; |
2b144498 SD |
217 | loff_t addr; |
218 | int ret; | |
219 | ||
220 | /* Read the page with vaddr into memory */ | |
221 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma); | |
222 | if (ret <= 0) | |
223 | return ret; | |
7b2d81d4 | 224 | |
2b144498 SD |
225 | ret = -EINVAL; |
226 | ||
227 | /* | |
228 | * We are interested in text pages only. Our pages of interest | |
229 | * should be mapped for read and execute only. We desist from | |
230 | * adding probes in write mapped pages since the breakpoints | |
231 | * might end up in the file copy. | |
232 | */ | |
5cb4ac3a | 233 | if (!valid_vma(vma, is_swbp_insn(&opcode))) |
2b144498 SD |
234 | goto put_out; |
235 | ||
3ff54efd | 236 | uprobe = container_of(auprobe, struct uprobe, arch); |
2b144498 SD |
237 | mapping = uprobe->inode->i_mapping; |
238 | if (mapping != vma->vm_file->f_mapping) | |
239 | goto put_out; | |
240 | ||
241 | addr = vma_address(vma, uprobe->offset); | |
242 | if (vaddr != (unsigned long)addr) | |
243 | goto put_out; | |
244 | ||
245 | ret = -ENOMEM; | |
246 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
247 | if (!new_page) | |
248 | goto put_out; | |
249 | ||
250 | __SetPageUptodate(new_page); | |
251 | ||
252 | /* | |
253 | * lock page will serialize against do_wp_page()'s | |
254 | * PageAnon() handling | |
255 | */ | |
256 | lock_page(old_page); | |
257 | /* copy the page now that we've got it stable */ | |
258 | vaddr_old = kmap_atomic(old_page); | |
259 | vaddr_new = kmap_atomic(new_page); | |
260 | ||
261 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
7b2d81d4 | 262 | |
2b144498 SD |
263 | /* poke the new insn in, ASSUMES we don't cross page boundary */ |
264 | vaddr &= ~PAGE_MASK; | |
5cb4ac3a SD |
265 | BUG_ON(vaddr + UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); |
266 | memcpy(vaddr_new + vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); | |
2b144498 SD |
267 | |
268 | kunmap_atomic(vaddr_new); | |
269 | kunmap_atomic(vaddr_old); | |
270 | ||
271 | ret = anon_vma_prepare(vma); | |
272 | if (ret) | |
273 | goto unlock_out; | |
274 | ||
275 | lock_page(new_page); | |
276 | ret = __replace_page(vma, old_page, new_page); | |
277 | unlock_page(new_page); | |
278 | ||
279 | unlock_out: | |
280 | unlock_page(old_page); | |
281 | page_cache_release(new_page); | |
282 | ||
283 | put_out: | |
7b2d81d4 IM |
284 | put_page(old_page); |
285 | ||
2b144498 SD |
286 | return ret; |
287 | } | |
288 | ||
289 | /** | |
290 | * read_opcode - read the opcode at a given virtual address. | |
291 | * @mm: the probed process address space. | |
292 | * @vaddr: the virtual address to read the opcode. | |
293 | * @opcode: location to store the read opcode. | |
294 | * | |
295 | * Called with mm->mmap_sem held (for read and with a reference to | |
296 | * mm. | |
297 | * | |
298 | * For mm @mm, read the opcode at @vaddr and store it in @opcode. | |
299 | * Return 0 (success) or a negative errno. | |
300 | */ | |
7b2d81d4 | 301 | static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode) |
2b144498 SD |
302 | { |
303 | struct page *page; | |
304 | void *vaddr_new; | |
305 | int ret; | |
306 | ||
307 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &page, NULL); | |
308 | if (ret <= 0) | |
309 | return ret; | |
310 | ||
311 | lock_page(page); | |
312 | vaddr_new = kmap_atomic(page); | |
313 | vaddr &= ~PAGE_MASK; | |
5cb4ac3a | 314 | memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
315 | kunmap_atomic(vaddr_new); |
316 | unlock_page(page); | |
7b2d81d4 IM |
317 | |
318 | put_page(page); | |
319 | ||
2b144498 SD |
320 | return 0; |
321 | } | |
322 | ||
5cb4ac3a | 323 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
2b144498 SD |
324 | { |
325 | uprobe_opcode_t opcode; | |
7b2d81d4 | 326 | int result; |
2b144498 | 327 | |
7b2d81d4 | 328 | result = read_opcode(mm, vaddr, &opcode); |
2b144498 SD |
329 | if (result) |
330 | return result; | |
331 | ||
5cb4ac3a | 332 | if (is_swbp_insn(&opcode)) |
2b144498 SD |
333 | return 1; |
334 | ||
335 | return 0; | |
336 | } | |
337 | ||
338 | /** | |
5cb4ac3a | 339 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 340 | * @auprobe: arch specific probepoint information. |
2b144498 | 341 | * @mm: the probed process address space. |
2b144498 SD |
342 | * @vaddr: the virtual address to insert the opcode. |
343 | * | |
344 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
345 | * Return 0 (success) or a negative errno. | |
346 | */ | |
5cb4ac3a | 347 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 348 | { |
7b2d81d4 | 349 | int result; |
2b144498 | 350 | |
5cb4ac3a | 351 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
352 | if (result == 1) |
353 | return -EEXIST; | |
354 | ||
355 | if (result) | |
356 | return result; | |
357 | ||
5cb4ac3a | 358 | return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
359 | } |
360 | ||
361 | /** | |
362 | * set_orig_insn - Restore the original instruction. | |
363 | * @mm: the probed process address space. | |
e3343e6a | 364 | * @auprobe: arch specific probepoint information. |
2b144498 SD |
365 | * @vaddr: the virtual address to insert the opcode. |
366 | * @verify: if true, verify existance of breakpoint instruction. | |
367 | * | |
368 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
369 | * Return 0 (success) or a negative errno. | |
370 | */ | |
7b2d81d4 | 371 | int __weak |
e3343e6a | 372 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify) |
2b144498 SD |
373 | { |
374 | if (verify) { | |
7b2d81d4 | 375 | int result; |
2b144498 | 376 | |
5cb4ac3a | 377 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
378 | if (!result) |
379 | return -EINVAL; | |
380 | ||
381 | if (result != 1) | |
382 | return result; | |
383 | } | |
e3343e6a | 384 | return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
385 | } |
386 | ||
387 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
388 | { | |
389 | if (l->inode < r->inode) | |
390 | return -1; | |
7b2d81d4 | 391 | |
2b144498 SD |
392 | if (l->inode > r->inode) |
393 | return 1; | |
2b144498 | 394 | |
7b2d81d4 IM |
395 | if (l->offset < r->offset) |
396 | return -1; | |
397 | ||
398 | if (l->offset > r->offset) | |
399 | return 1; | |
2b144498 SD |
400 | |
401 | return 0; | |
402 | } | |
403 | ||
404 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
405 | { | |
406 | struct uprobe u = { .inode = inode, .offset = offset }; | |
407 | struct rb_node *n = uprobes_tree.rb_node; | |
408 | struct uprobe *uprobe; | |
409 | int match; | |
410 | ||
411 | while (n) { | |
412 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
413 | match = match_uprobe(&u, uprobe); | |
414 | if (!match) { | |
415 | atomic_inc(&uprobe->ref); | |
416 | return uprobe; | |
417 | } | |
7b2d81d4 | 418 | |
2b144498 SD |
419 | if (match < 0) |
420 | n = n->rb_left; | |
421 | else | |
422 | n = n->rb_right; | |
423 | } | |
424 | return NULL; | |
425 | } | |
426 | ||
427 | /* | |
428 | * Find a uprobe corresponding to a given inode:offset | |
429 | * Acquires uprobes_treelock | |
430 | */ | |
431 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
432 | { | |
433 | struct uprobe *uprobe; | |
434 | unsigned long flags; | |
435 | ||
436 | spin_lock_irqsave(&uprobes_treelock, flags); | |
437 | uprobe = __find_uprobe(inode, offset); | |
438 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 439 | |
2b144498 SD |
440 | return uprobe; |
441 | } | |
442 | ||
443 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
444 | { | |
445 | struct rb_node **p = &uprobes_tree.rb_node; | |
446 | struct rb_node *parent = NULL; | |
447 | struct uprobe *u; | |
448 | int match; | |
449 | ||
450 | while (*p) { | |
451 | parent = *p; | |
452 | u = rb_entry(parent, struct uprobe, rb_node); | |
453 | match = match_uprobe(uprobe, u); | |
454 | if (!match) { | |
455 | atomic_inc(&u->ref); | |
456 | return u; | |
457 | } | |
458 | ||
459 | if (match < 0) | |
460 | p = &parent->rb_left; | |
461 | else | |
462 | p = &parent->rb_right; | |
463 | ||
464 | } | |
7b2d81d4 | 465 | |
2b144498 SD |
466 | u = NULL; |
467 | rb_link_node(&uprobe->rb_node, parent, p); | |
468 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
469 | /* get access + creation ref */ | |
470 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 471 | |
2b144498 SD |
472 | return u; |
473 | } | |
474 | ||
475 | /* | |
7b2d81d4 | 476 | * Acquire uprobes_treelock. |
2b144498 SD |
477 | * Matching uprobe already exists in rbtree; |
478 | * increment (access refcount) and return the matching uprobe. | |
479 | * | |
480 | * No matching uprobe; insert the uprobe in rb_tree; | |
481 | * get a double refcount (access + creation) and return NULL. | |
482 | */ | |
483 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
484 | { | |
485 | unsigned long flags; | |
486 | struct uprobe *u; | |
487 | ||
488 | spin_lock_irqsave(&uprobes_treelock, flags); | |
489 | u = __insert_uprobe(uprobe); | |
490 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 491 | |
0326f5a9 SD |
492 | /* For now assume that the instruction need not be single-stepped */ |
493 | uprobe->flags |= UPROBE_SKIP_SSTEP; | |
494 | ||
2b144498 SD |
495 | return u; |
496 | } | |
497 | ||
498 | static void put_uprobe(struct uprobe *uprobe) | |
499 | { | |
500 | if (atomic_dec_and_test(&uprobe->ref)) | |
501 | kfree(uprobe); | |
502 | } | |
503 | ||
504 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
505 | { | |
506 | struct uprobe *uprobe, *cur_uprobe; | |
507 | ||
508 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
509 | if (!uprobe) | |
510 | return NULL; | |
511 | ||
512 | uprobe->inode = igrab(inode); | |
513 | uprobe->offset = offset; | |
514 | init_rwsem(&uprobe->consumer_rwsem); | |
515 | INIT_LIST_HEAD(&uprobe->pending_list); | |
516 | ||
517 | /* add to uprobes_tree, sorted on inode:offset */ | |
518 | cur_uprobe = insert_uprobe(uprobe); | |
519 | ||
520 | /* a uprobe exists for this inode:offset combination */ | |
521 | if (cur_uprobe) { | |
522 | kfree(uprobe); | |
523 | uprobe = cur_uprobe; | |
524 | iput(inode); | |
7b2d81d4 | 525 | } else { |
2b144498 | 526 | atomic_inc(&uprobe_events); |
7b2d81d4 IM |
527 | } |
528 | ||
2b144498 SD |
529 | return uprobe; |
530 | } | |
531 | ||
0326f5a9 SD |
532 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
533 | { | |
534 | struct uprobe_consumer *uc; | |
535 | ||
536 | if (!(uprobe->flags & UPROBE_RUN_HANDLER)) | |
537 | return; | |
538 | ||
539 | down_read(&uprobe->consumer_rwsem); | |
540 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
541 | if (!uc->filter || uc->filter(uc, current)) | |
542 | uc->handler(uc, regs); | |
543 | } | |
544 | up_read(&uprobe->consumer_rwsem); | |
545 | } | |
546 | ||
2b144498 | 547 | /* Returns the previous consumer */ |
7b2d81d4 | 548 | static struct uprobe_consumer * |
e3343e6a | 549 | consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
550 | { |
551 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
552 | uc->next = uprobe->consumers; |
553 | uprobe->consumers = uc; | |
2b144498 | 554 | up_write(&uprobe->consumer_rwsem); |
7b2d81d4 | 555 | |
e3343e6a | 556 | return uc->next; |
2b144498 SD |
557 | } |
558 | ||
559 | /* | |
e3343e6a SD |
560 | * For uprobe @uprobe, delete the consumer @uc. |
561 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
562 | * or return false. |
563 | */ | |
e3343e6a | 564 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
565 | { |
566 | struct uprobe_consumer **con; | |
567 | bool ret = false; | |
568 | ||
569 | down_write(&uprobe->consumer_rwsem); | |
570 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
571 | if (*con == uc) { |
572 | *con = uc->next; | |
2b144498 SD |
573 | ret = true; |
574 | break; | |
575 | } | |
576 | } | |
577 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 578 | |
2b144498 SD |
579 | return ret; |
580 | } | |
581 | ||
e3343e6a SD |
582 | static int |
583 | __copy_insn(struct address_space *mapping, struct vm_area_struct *vma, char *insn, | |
2b144498 SD |
584 | unsigned long nbytes, unsigned long offset) |
585 | { | |
586 | struct file *filp = vma->vm_file; | |
587 | struct page *page; | |
588 | void *vaddr; | |
589 | unsigned long off1; | |
590 | unsigned long idx; | |
591 | ||
592 | if (!filp) | |
593 | return -EINVAL; | |
594 | ||
595 | idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT); | |
596 | off1 = offset &= ~PAGE_MASK; | |
597 | ||
598 | /* | |
599 | * Ensure that the page that has the original instruction is | |
600 | * populated and in page-cache. | |
601 | */ | |
602 | page = read_mapping_page(mapping, idx, filp); | |
603 | if (IS_ERR(page)) | |
604 | return PTR_ERR(page); | |
605 | ||
606 | vaddr = kmap_atomic(page); | |
607 | memcpy(insn, vaddr + off1, nbytes); | |
608 | kunmap_atomic(vaddr); | |
609 | page_cache_release(page); | |
7b2d81d4 | 610 | |
2b144498 SD |
611 | return 0; |
612 | } | |
613 | ||
e3343e6a SD |
614 | static int |
615 | copy_insn(struct uprobe *uprobe, struct vm_area_struct *vma, unsigned long addr) | |
2b144498 SD |
616 | { |
617 | struct address_space *mapping; | |
2b144498 | 618 | unsigned long nbytes; |
7b2d81d4 | 619 | int bytes; |
2b144498 SD |
620 | |
621 | addr &= ~PAGE_MASK; | |
622 | nbytes = PAGE_SIZE - addr; | |
623 | mapping = uprobe->inode->i_mapping; | |
624 | ||
625 | /* Instruction at end of binary; copy only available bytes */ | |
626 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
627 | bytes = uprobe->inode->i_size - uprobe->offset; | |
628 | else | |
629 | bytes = MAX_UINSN_BYTES; | |
630 | ||
631 | /* Instruction at the page-boundary; copy bytes in second page */ | |
632 | if (nbytes < bytes) { | |
3ff54efd | 633 | if (__copy_insn(mapping, vma, uprobe->arch.insn + nbytes, |
2b144498 SD |
634 | bytes - nbytes, uprobe->offset + nbytes)) |
635 | return -ENOMEM; | |
636 | ||
637 | bytes = nbytes; | |
638 | } | |
3ff54efd | 639 | return __copy_insn(mapping, vma, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
640 | } |
641 | ||
e3343e6a SD |
642 | static int |
643 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
644 | struct vm_area_struct *vma, loff_t vaddr) | |
2b144498 SD |
645 | { |
646 | unsigned long addr; | |
647 | int ret; | |
648 | ||
649 | /* | |
650 | * If probe is being deleted, unregister thread could be done with | |
651 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
652 | * nobody will be able to cleanup. Also we could be from fork or | |
653 | * mremap path, where the probe might have already been inserted. | |
654 | * Hence behave as if probe already existed. | |
655 | */ | |
656 | if (!uprobe->consumers) | |
657 | return -EEXIST; | |
658 | ||
659 | addr = (unsigned long)vaddr; | |
7b2d81d4 | 660 | |
900771a4 | 661 | if (!(uprobe->flags & UPROBE_COPY_INSN)) { |
2b144498 SD |
662 | ret = copy_insn(uprobe, vma, addr); |
663 | if (ret) | |
664 | return ret; | |
665 | ||
5cb4ac3a | 666 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) |
2b144498 SD |
667 | return -EEXIST; |
668 | ||
0326f5a9 | 669 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm); |
2b144498 SD |
670 | if (ret) |
671 | return ret; | |
672 | ||
900771a4 | 673 | uprobe->flags |= UPROBE_COPY_INSN; |
2b144498 | 674 | } |
5cb4ac3a | 675 | ret = set_swbp(&uprobe->arch, mm, addr); |
2b144498 SD |
676 | |
677 | return ret; | |
678 | } | |
679 | ||
e3343e6a SD |
680 | static void |
681 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr) | |
2b144498 | 682 | { |
e3343e6a | 683 | set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true); |
2b144498 SD |
684 | } |
685 | ||
0326f5a9 SD |
686 | /* |
687 | * There could be threads that have hit the breakpoint and are entering the | |
688 | * notifier code and trying to acquire the uprobes_treelock. The thread | |
689 | * calling delete_uprobe() that is removing the uprobe from the rb_tree can | |
690 | * race with these threads and might acquire the uprobes_treelock compared | |
691 | * to some of the breakpoint hit threads. In such a case, the breakpoint | |
692 | * hit threads will not find the uprobe. The current unregistering thread | |
693 | * waits till all other threads have hit a breakpoint, to acquire the | |
694 | * uprobes_treelock before the uprobe is removed from the rbtree. | |
695 | */ | |
2b144498 SD |
696 | static void delete_uprobe(struct uprobe *uprobe) |
697 | { | |
698 | unsigned long flags; | |
699 | ||
0326f5a9 | 700 | synchronize_srcu(&uprobes_srcu); |
2b144498 SD |
701 | spin_lock_irqsave(&uprobes_treelock, flags); |
702 | rb_erase(&uprobe->rb_node, &uprobes_tree); | |
703 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
704 | iput(uprobe->inode); | |
705 | put_uprobe(uprobe); | |
706 | atomic_dec(&uprobe_events); | |
707 | } | |
708 | ||
e3343e6a SD |
709 | static struct vma_info * |
710 | __find_next_vma_info(struct address_space *mapping, struct list_head *head, | |
711 | struct vma_info *vi, loff_t offset, bool is_register) | |
2b144498 SD |
712 | { |
713 | struct prio_tree_iter iter; | |
714 | struct vm_area_struct *vma; | |
715 | struct vma_info *tmpvi; | |
7b2d81d4 | 716 | unsigned long pgoff; |
2b144498 | 717 | int existing_vma; |
7b2d81d4 IM |
718 | loff_t vaddr; |
719 | ||
720 | pgoff = offset >> PAGE_SHIFT; | |
2b144498 SD |
721 | |
722 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
723 | if (!valid_vma(vma, is_register)) | |
724 | continue; | |
725 | ||
726 | existing_vma = 0; | |
727 | vaddr = vma_address(vma, offset); | |
7b2d81d4 | 728 | |
2b144498 SD |
729 | list_for_each_entry(tmpvi, head, probe_list) { |
730 | if (tmpvi->mm == vma->vm_mm && tmpvi->vaddr == vaddr) { | |
731 | existing_vma = 1; | |
732 | break; | |
733 | } | |
734 | } | |
735 | ||
736 | /* | |
737 | * Another vma needs a probe to be installed. However skip | |
738 | * installing the probe if the vma is about to be unlinked. | |
739 | */ | |
7b2d81d4 | 740 | if (!existing_vma && atomic_inc_not_zero(&vma->vm_mm->mm_users)) { |
2b144498 SD |
741 | vi->mm = vma->vm_mm; |
742 | vi->vaddr = vaddr; | |
743 | list_add(&vi->probe_list, head); | |
7b2d81d4 | 744 | |
2b144498 SD |
745 | return vi; |
746 | } | |
747 | } | |
7b2d81d4 | 748 | |
2b144498 SD |
749 | return NULL; |
750 | } | |
751 | ||
752 | /* | |
753 | * Iterate in the rmap prio tree and find a vma where a probe has not | |
754 | * yet been inserted. | |
755 | */ | |
7b2d81d4 | 756 | static struct vma_info * |
e3343e6a SD |
757 | find_next_vma_info(struct address_space *mapping, struct list_head *head, |
758 | loff_t offset, bool is_register) | |
2b144498 SD |
759 | { |
760 | struct vma_info *vi, *retvi; | |
7b2d81d4 | 761 | |
2b144498 SD |
762 | vi = kzalloc(sizeof(struct vma_info), GFP_KERNEL); |
763 | if (!vi) | |
764 | return ERR_PTR(-ENOMEM); | |
765 | ||
766 | mutex_lock(&mapping->i_mmap_mutex); | |
e3343e6a | 767 | retvi = __find_next_vma_info(mapping, head, vi, offset, is_register); |
2b144498 SD |
768 | mutex_unlock(&mapping->i_mmap_mutex); |
769 | ||
770 | if (!retvi) | |
771 | kfree(vi); | |
7b2d81d4 | 772 | |
2b144498 SD |
773 | return retvi; |
774 | } | |
775 | ||
776 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
777 | { | |
778 | struct list_head try_list; | |
779 | struct vm_area_struct *vma; | |
780 | struct address_space *mapping; | |
781 | struct vma_info *vi, *tmpvi; | |
782 | struct mm_struct *mm; | |
783 | loff_t vaddr; | |
7b2d81d4 | 784 | int ret; |
2b144498 SD |
785 | |
786 | mapping = uprobe->inode->i_mapping; | |
787 | INIT_LIST_HEAD(&try_list); | |
7b2d81d4 IM |
788 | |
789 | ret = 0; | |
790 | ||
791 | for (;;) { | |
e3343e6a | 792 | vi = find_next_vma_info(mapping, &try_list, uprobe->offset, is_register); |
7b2d81d4 IM |
793 | if (!vi) |
794 | break; | |
795 | ||
2b144498 SD |
796 | if (IS_ERR(vi)) { |
797 | ret = PTR_ERR(vi); | |
798 | break; | |
799 | } | |
7b2d81d4 | 800 | |
2b144498 SD |
801 | mm = vi->mm; |
802 | down_read(&mm->mmap_sem); | |
803 | vma = find_vma(mm, (unsigned long)vi->vaddr); | |
804 | if (!vma || !valid_vma(vma, is_register)) { | |
805 | list_del(&vi->probe_list); | |
806 | kfree(vi); | |
807 | up_read(&mm->mmap_sem); | |
808 | mmput(mm); | |
809 | continue; | |
810 | } | |
811 | vaddr = vma_address(vma, uprobe->offset); | |
812 | if (vma->vm_file->f_mapping->host != uprobe->inode || | |
813 | vaddr != vi->vaddr) { | |
814 | list_del(&vi->probe_list); | |
815 | kfree(vi); | |
816 | up_read(&mm->mmap_sem); | |
817 | mmput(mm); | |
818 | continue; | |
819 | } | |
820 | ||
821 | if (is_register) | |
e3343e6a | 822 | ret = install_breakpoint(uprobe, mm, vma, vi->vaddr); |
2b144498 | 823 | else |
e3343e6a | 824 | remove_breakpoint(uprobe, mm, vi->vaddr); |
2b144498 SD |
825 | |
826 | up_read(&mm->mmap_sem); | |
827 | mmput(mm); | |
828 | if (is_register) { | |
829 | if (ret && ret == -EEXIST) | |
830 | ret = 0; | |
831 | if (ret) | |
832 | break; | |
833 | } | |
834 | } | |
7b2d81d4 | 835 | |
2b144498 SD |
836 | list_for_each_entry_safe(vi, tmpvi, &try_list, probe_list) { |
837 | list_del(&vi->probe_list); | |
838 | kfree(vi); | |
839 | } | |
7b2d81d4 | 840 | |
2b144498 SD |
841 | return ret; |
842 | } | |
843 | ||
7b2d81d4 | 844 | static int __uprobe_register(struct uprobe *uprobe) |
2b144498 SD |
845 | { |
846 | return register_for_each_vma(uprobe, true); | |
847 | } | |
848 | ||
7b2d81d4 | 849 | static void __uprobe_unregister(struct uprobe *uprobe) |
2b144498 SD |
850 | { |
851 | if (!register_for_each_vma(uprobe, false)) | |
852 | delete_uprobe(uprobe); | |
853 | ||
854 | /* TODO : cant unregister? schedule a worker thread */ | |
855 | } | |
856 | ||
857 | /* | |
7b2d81d4 | 858 | * uprobe_register - register a probe |
2b144498 SD |
859 | * @inode: the file in which the probe has to be placed. |
860 | * @offset: offset from the start of the file. | |
e3343e6a | 861 | * @uc: information on howto handle the probe.. |
2b144498 | 862 | * |
7b2d81d4 | 863 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
864 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
865 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 866 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 867 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 868 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
869 | * unregisters. |
870 | * | |
871 | * Return errno if it cannot successully install probes | |
872 | * else return 0 (success) | |
873 | */ | |
e3343e6a | 874 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
875 | { |
876 | struct uprobe *uprobe; | |
7b2d81d4 | 877 | int ret; |
2b144498 | 878 | |
e3343e6a | 879 | if (!inode || !uc || uc->next) |
7b2d81d4 | 880 | return -EINVAL; |
2b144498 SD |
881 | |
882 | if (offset > i_size_read(inode)) | |
7b2d81d4 | 883 | return -EINVAL; |
2b144498 SD |
884 | |
885 | ret = 0; | |
886 | mutex_lock(uprobes_hash(inode)); | |
887 | uprobe = alloc_uprobe(inode, offset); | |
7b2d81d4 | 888 | |
e3343e6a | 889 | if (uprobe && !consumer_add(uprobe, uc)) { |
7b2d81d4 | 890 | ret = __uprobe_register(uprobe); |
2b144498 SD |
891 | if (ret) { |
892 | uprobe->consumers = NULL; | |
7b2d81d4 IM |
893 | __uprobe_unregister(uprobe); |
894 | } else { | |
900771a4 | 895 | uprobe->flags |= UPROBE_RUN_HANDLER; |
7b2d81d4 | 896 | } |
2b144498 SD |
897 | } |
898 | ||
899 | mutex_unlock(uprobes_hash(inode)); | |
900 | put_uprobe(uprobe); | |
901 | ||
902 | return ret; | |
903 | } | |
904 | ||
905 | /* | |
7b2d81d4 | 906 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
907 | * @inode: the file in which the probe has to be removed. |
908 | * @offset: offset from the start of the file. | |
e3343e6a | 909 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 910 | */ |
e3343e6a | 911 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 912 | { |
7b2d81d4 | 913 | struct uprobe *uprobe; |
2b144498 | 914 | |
e3343e6a | 915 | if (!inode || !uc) |
2b144498 SD |
916 | return; |
917 | ||
918 | uprobe = find_uprobe(inode, offset); | |
919 | if (!uprobe) | |
920 | return; | |
921 | ||
922 | mutex_lock(uprobes_hash(inode)); | |
2b144498 | 923 | |
e3343e6a | 924 | if (consumer_del(uprobe, uc)) { |
7b2d81d4 IM |
925 | if (!uprobe->consumers) { |
926 | __uprobe_unregister(uprobe); | |
900771a4 | 927 | uprobe->flags &= ~UPROBE_RUN_HANDLER; |
7b2d81d4 | 928 | } |
2b144498 SD |
929 | } |
930 | ||
2b144498 SD |
931 | mutex_unlock(uprobes_hash(inode)); |
932 | if (uprobe) | |
933 | put_uprobe(uprobe); | |
934 | } | |
935 | ||
936 | /* | |
937 | * Of all the nodes that correspond to the given inode, return the node | |
938 | * with the least offset. | |
939 | */ | |
940 | static struct rb_node *find_least_offset_node(struct inode *inode) | |
941 | { | |
942 | struct uprobe u = { .inode = inode, .offset = 0}; | |
943 | struct rb_node *n = uprobes_tree.rb_node; | |
944 | struct rb_node *close_node = NULL; | |
945 | struct uprobe *uprobe; | |
946 | int match; | |
947 | ||
948 | while (n) { | |
949 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
950 | match = match_uprobe(&u, uprobe); | |
7b2d81d4 | 951 | |
2b144498 SD |
952 | if (uprobe->inode == inode) |
953 | close_node = n; | |
954 | ||
955 | if (!match) | |
956 | return close_node; | |
957 | ||
958 | if (match < 0) | |
959 | n = n->rb_left; | |
960 | else | |
961 | n = n->rb_right; | |
962 | } | |
7b2d81d4 | 963 | |
2b144498 SD |
964 | return close_node; |
965 | } | |
966 | ||
967 | /* | |
968 | * For a given inode, build a list of probes that need to be inserted. | |
969 | */ | |
970 | static void build_probe_list(struct inode *inode, struct list_head *head) | |
971 | { | |
972 | struct uprobe *uprobe; | |
2b144498 | 973 | unsigned long flags; |
7b2d81d4 | 974 | struct rb_node *n; |
2b144498 SD |
975 | |
976 | spin_lock_irqsave(&uprobes_treelock, flags); | |
7b2d81d4 | 977 | |
2b144498 | 978 | n = find_least_offset_node(inode); |
7b2d81d4 | 979 | |
2b144498 SD |
980 | for (; n; n = rb_next(n)) { |
981 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
982 | if (uprobe->inode != inode) | |
983 | break; | |
984 | ||
985 | list_add(&uprobe->pending_list, head); | |
986 | atomic_inc(&uprobe->ref); | |
987 | } | |
7b2d81d4 | 988 | |
2b144498 SD |
989 | spin_unlock_irqrestore(&uprobes_treelock, flags); |
990 | } | |
991 | ||
992 | /* | |
993 | * Called from mmap_region. | |
994 | * called with mm->mmap_sem acquired. | |
995 | * | |
996 | * Return -ve no if we fail to insert probes and we cannot | |
997 | * bail-out. | |
7b2d81d4 IM |
998 | * Return 0 otherwise. i.e: |
999 | * | |
2b144498 SD |
1000 | * - successful insertion of probes |
1001 | * - (or) no possible probes to be inserted. | |
1002 | * - (or) insertion of probes failed but we can bail-out. | |
1003 | */ | |
7b2d81d4 | 1004 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1005 | { |
1006 | struct list_head tmp_list; | |
1007 | struct uprobe *uprobe, *u; | |
1008 | struct inode *inode; | |
7b2d81d4 | 1009 | int ret; |
2b144498 SD |
1010 | |
1011 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) | |
7b2d81d4 | 1012 | return 0; |
2b144498 SD |
1013 | |
1014 | inode = vma->vm_file->f_mapping->host; | |
1015 | if (!inode) | |
7b2d81d4 | 1016 | return 0; |
2b144498 SD |
1017 | |
1018 | INIT_LIST_HEAD(&tmp_list); | |
1019 | mutex_lock(uprobes_mmap_hash(inode)); | |
1020 | build_probe_list(inode, &tmp_list); | |
7b2d81d4 IM |
1021 | |
1022 | ret = 0; | |
1023 | ||
2b144498 SD |
1024 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
1025 | loff_t vaddr; | |
1026 | ||
1027 | list_del(&uprobe->pending_list); | |
1028 | if (!ret) { | |
1029 | vaddr = vma_address(vma, uprobe->offset); | |
7b2d81d4 | 1030 | if (vaddr >= vma->vm_start && vaddr < vma->vm_end) { |
e3343e6a | 1031 | ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
7b2d81d4 IM |
1032 | /* Ignore double add: */ |
1033 | if (ret == -EEXIST) | |
1034 | ret = 0; | |
2b144498 | 1035 | } |
2b144498 SD |
1036 | } |
1037 | put_uprobe(uprobe); | |
1038 | } | |
1039 | ||
1040 | mutex_unlock(uprobes_mmap_hash(inode)); | |
1041 | ||
1042 | return ret; | |
1043 | } | |
1044 | ||
0326f5a9 SD |
1045 | /** |
1046 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1047 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1048 | * instruction. | |
1049 | * Return the address of the breakpoint instruction. | |
1050 | */ | |
1051 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1052 | { | |
1053 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Called with no locks held. | |
1058 | * Called in context of a exiting or a exec-ing thread. | |
1059 | */ | |
1060 | void uprobe_free_utask(struct task_struct *t) | |
1061 | { | |
1062 | struct uprobe_task *utask = t->utask; | |
1063 | ||
1064 | if (t->uprobe_srcu_id != -1) | |
1065 | srcu_read_unlock_raw(&uprobes_srcu, t->uprobe_srcu_id); | |
1066 | ||
1067 | if (!utask) | |
1068 | return; | |
1069 | ||
1070 | if (utask->active_uprobe) | |
1071 | put_uprobe(utask->active_uprobe); | |
1072 | ||
1073 | kfree(utask); | |
1074 | t->utask = NULL; | |
1075 | } | |
1076 | ||
1077 | /* | |
1078 | * Called in context of a new clone/fork from copy_process. | |
1079 | */ | |
1080 | void uprobe_copy_process(struct task_struct *t) | |
1081 | { | |
1082 | t->utask = NULL; | |
1083 | t->uprobe_srcu_id = -1; | |
1084 | } | |
1085 | ||
1086 | /* | |
1087 | * Allocate a uprobe_task object for the task. | |
1088 | * Called when the thread hits a breakpoint for the first time. | |
1089 | * | |
1090 | * Returns: | |
1091 | * - pointer to new uprobe_task on success | |
1092 | * - NULL otherwise | |
1093 | */ | |
1094 | static struct uprobe_task *add_utask(void) | |
1095 | { | |
1096 | struct uprobe_task *utask; | |
1097 | ||
1098 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1099 | if (unlikely(!utask)) | |
1100 | return NULL; | |
1101 | ||
1102 | utask->active_uprobe = NULL; | |
1103 | current->utask = utask; | |
1104 | return utask; | |
1105 | } | |
1106 | ||
1107 | /* Prepare to single-step probed instruction out of line. */ | |
1108 | static int | |
1109 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1110 | { | |
1111 | return -EFAULT; | |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * If we are singlestepping, then ensure this thread is not connected to | |
1116 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1117 | * triggers the signal, restart the original insn even if the task is | |
1118 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1119 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1120 | * _same_ instruction should be repeated again after return from the signal | |
1121 | * handler, and SSTEP can never finish in this case. | |
1122 | */ | |
1123 | bool uprobe_deny_signal(void) | |
1124 | { | |
1125 | struct task_struct *t = current; | |
1126 | struct uprobe_task *utask = t->utask; | |
1127 | ||
1128 | if (likely(!utask || !utask->active_uprobe)) | |
1129 | return false; | |
1130 | ||
1131 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1132 | ||
1133 | if (signal_pending(t)) { | |
1134 | spin_lock_irq(&t->sighand->siglock); | |
1135 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1136 | spin_unlock_irq(&t->sighand->siglock); | |
1137 | ||
1138 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1139 | utask->state = UTASK_SSTEP_TRAPPED; | |
1140 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1141 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1142 | } | |
1143 | } | |
1144 | ||
1145 | return true; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * Avoid singlestepping the original instruction if the original instruction | |
1150 | * is a NOP or can be emulated. | |
1151 | */ | |
1152 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1153 | { | |
1154 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) | |
1155 | return true; | |
1156 | ||
1157 | uprobe->flags &= ~UPROBE_SKIP_SSTEP; | |
1158 | return false; | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * Run handler and ask thread to singlestep. | |
1163 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1164 | */ | |
1165 | static void handle_swbp(struct pt_regs *regs) | |
1166 | { | |
1167 | struct vm_area_struct *vma; | |
1168 | struct uprobe_task *utask; | |
1169 | struct uprobe *uprobe; | |
1170 | struct mm_struct *mm; | |
1171 | unsigned long bp_vaddr; | |
1172 | ||
1173 | uprobe = NULL; | |
1174 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
1175 | mm = current->mm; | |
1176 | down_read(&mm->mmap_sem); | |
1177 | vma = find_vma(mm, bp_vaddr); | |
1178 | ||
1179 | if (vma && vma->vm_start <= bp_vaddr && valid_vma(vma, false)) { | |
1180 | struct inode *inode; | |
1181 | loff_t offset; | |
1182 | ||
1183 | inode = vma->vm_file->f_mapping->host; | |
1184 | offset = bp_vaddr - vma->vm_start; | |
1185 | offset += (vma->vm_pgoff << PAGE_SHIFT); | |
1186 | uprobe = find_uprobe(inode, offset); | |
1187 | } | |
1188 | ||
1189 | srcu_read_unlock_raw(&uprobes_srcu, current->uprobe_srcu_id); | |
1190 | current->uprobe_srcu_id = -1; | |
1191 | up_read(&mm->mmap_sem); | |
1192 | ||
1193 | if (!uprobe) { | |
1194 | /* No matching uprobe; signal SIGTRAP. */ | |
1195 | send_sig(SIGTRAP, current, 0); | |
1196 | return; | |
1197 | } | |
1198 | ||
1199 | utask = current->utask; | |
1200 | if (!utask) { | |
1201 | utask = add_utask(); | |
1202 | /* Cannot allocate; re-execute the instruction. */ | |
1203 | if (!utask) | |
1204 | goto cleanup_ret; | |
1205 | } | |
1206 | utask->active_uprobe = uprobe; | |
1207 | handler_chain(uprobe, regs); | |
1208 | if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs)) | |
1209 | goto cleanup_ret; | |
1210 | ||
1211 | utask->state = UTASK_SSTEP; | |
1212 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { | |
1213 | user_enable_single_step(current); | |
1214 | return; | |
1215 | } | |
1216 | ||
1217 | cleanup_ret: | |
1218 | if (utask) { | |
1219 | utask->active_uprobe = NULL; | |
1220 | utask->state = UTASK_RUNNING; | |
1221 | } | |
1222 | if (uprobe) { | |
1223 | if (!(uprobe->flags & UPROBE_SKIP_SSTEP)) | |
1224 | ||
1225 | /* | |
1226 | * cannot singlestep; cannot skip instruction; | |
1227 | * re-execute the instruction. | |
1228 | */ | |
1229 | instruction_pointer_set(regs, bp_vaddr); | |
1230 | ||
1231 | put_uprobe(uprobe); | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * Perform required fix-ups and disable singlestep. | |
1237 | * Allow pending signals to take effect. | |
1238 | */ | |
1239 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1240 | { | |
1241 | struct uprobe *uprobe; | |
1242 | ||
1243 | uprobe = utask->active_uprobe; | |
1244 | if (utask->state == UTASK_SSTEP_ACK) | |
1245 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1246 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1247 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1248 | else | |
1249 | WARN_ON_ONCE(1); | |
1250 | ||
1251 | put_uprobe(uprobe); | |
1252 | utask->active_uprobe = NULL; | |
1253 | utask->state = UTASK_RUNNING; | |
1254 | user_disable_single_step(current); | |
1255 | ||
1256 | spin_lock_irq(¤t->sighand->siglock); | |
1257 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1258 | spin_unlock_irq(¤t->sighand->siglock); | |
1259 | } | |
1260 | ||
1261 | /* | |
1262 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on | |
1263 | * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and | |
1264 | * allows the thread to return from interrupt. | |
1265 | * | |
1266 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and | |
1267 | * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from | |
1268 | * interrupt. | |
1269 | * | |
1270 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1271 | * uprobe_notify_resume(). | |
1272 | */ | |
1273 | void uprobe_notify_resume(struct pt_regs *regs) | |
1274 | { | |
1275 | struct uprobe_task *utask; | |
1276 | ||
1277 | utask = current->utask; | |
1278 | if (!utask || utask->state == UTASK_BP_HIT) | |
1279 | handle_swbp(regs); | |
1280 | else | |
1281 | handle_singlestep(utask, regs); | |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1286 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1287 | */ | |
1288 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1289 | { | |
1290 | struct uprobe_task *utask; | |
1291 | ||
1292 | if (!current->mm) | |
1293 | return 0; | |
1294 | ||
1295 | utask = current->utask; | |
1296 | if (utask) | |
1297 | utask->state = UTASK_BP_HIT; | |
1298 | ||
1299 | set_thread_flag(TIF_UPROBE); | |
1300 | current->uprobe_srcu_id = srcu_read_lock_raw(&uprobes_srcu); | |
1301 | ||
1302 | return 1; | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1307 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1308 | */ | |
1309 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1310 | { | |
1311 | struct uprobe_task *utask = current->utask; | |
1312 | ||
1313 | if (!current->mm || !utask || !utask->active_uprobe) | |
1314 | /* task is currently not uprobed */ | |
1315 | return 0; | |
1316 | ||
1317 | utask->state = UTASK_SSTEP_ACK; | |
1318 | set_thread_flag(TIF_UPROBE); | |
1319 | return 1; | |
1320 | } | |
1321 | ||
1322 | static struct notifier_block uprobe_exception_nb = { | |
1323 | .notifier_call = arch_uprobe_exception_notify, | |
1324 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1325 | }; | |
1326 | ||
2b144498 SD |
1327 | static int __init init_uprobes(void) |
1328 | { | |
1329 | int i; | |
1330 | ||
1331 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1332 | mutex_init(&uprobes_mutex[i]); | |
1333 | mutex_init(&uprobes_mmap_mutex[i]); | |
1334 | } | |
0326f5a9 SD |
1335 | init_srcu_struct(&uprobes_srcu); |
1336 | ||
1337 | return register_die_notifier(&uprobe_exception_nb); | |
2b144498 | 1338 | } |
0326f5a9 | 1339 | module_init(init_uprobes); |
2b144498 SD |
1340 | |
1341 | static void __exit exit_uprobes(void) | |
1342 | { | |
1343 | } | |
2b144498 | 1344 | module_exit(exit_uprobes); |