Commit | Line | Data |
---|---|---|
2b144498 | 1 | /* |
7b2d81d4 | 2 | * User-space Probes (UProbes) |
2b144498 SD |
3 | * |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
35aa621b | 18 | * Copyright (C) IBM Corporation, 2008-2012 |
2b144498 SD |
19 | * Authors: |
20 | * Srikar Dronamraju | |
21 | * Jim Keniston | |
35aa621b | 22 | * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> |
2b144498 SD |
23 | */ |
24 | ||
25 | #include <linux/kernel.h> | |
26 | #include <linux/highmem.h> | |
27 | #include <linux/pagemap.h> /* read_mapping_page */ | |
28 | #include <linux/slab.h> | |
29 | #include <linux/sched.h> | |
30 | #include <linux/rmap.h> /* anon_vma_prepare */ | |
31 | #include <linux/mmu_notifier.h> /* set_pte_at_notify */ | |
32 | #include <linux/swap.h> /* try_to_free_swap */ | |
0326f5a9 SD |
33 | #include <linux/ptrace.h> /* user_enable_single_step */ |
34 | #include <linux/kdebug.h> /* notifier mechanism */ | |
7b2d81d4 | 35 | |
2b144498 SD |
36 | #include <linux/uprobes.h> |
37 | ||
d4b3b638 SD |
38 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
39 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
40 | ||
2b144498 | 41 | static struct rb_root uprobes_tree = RB_ROOT; |
7b2d81d4 | 42 | |
2b144498 SD |
43 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
44 | ||
45 | #define UPROBES_HASH_SZ 13 | |
7b2d81d4 | 46 | |
c5784de2 PZ |
47 | /* |
48 | * We need separate register/unregister and mmap/munmap lock hashes because | |
49 | * of mmap_sem nesting. | |
50 | * | |
51 | * uprobe_register() needs to install probes on (potentially) all processes | |
52 | * and thus needs to acquire multiple mmap_sems (consequtively, not | |
53 | * concurrently), whereas uprobe_mmap() is called while holding mmap_sem | |
54 | * for the particular process doing the mmap. | |
55 | * | |
56 | * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem | |
57 | * because of lock order against i_mmap_mutex. This means there's a hole in | |
58 | * the register vma iteration where a mmap() can happen. | |
59 | * | |
60 | * Thus uprobe_register() can race with uprobe_mmap() and we can try and | |
61 | * install a probe where one is already installed. | |
62 | */ | |
63 | ||
2b144498 SD |
64 | /* serialize (un)register */ |
65 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
66 | |
67 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
68 | |
69 | /* serialize uprobe->pending_list */ | |
70 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 71 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 SD |
72 | |
73 | /* | |
7b2d81d4 | 74 | * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe |
2b144498 SD |
75 | * events active at this time. Probably a fine grained per inode count is |
76 | * better? | |
77 | */ | |
78 | static atomic_t uprobe_events = ATOMIC_INIT(0); | |
79 | ||
3ff54efd SD |
80 | struct uprobe { |
81 | struct rb_node rb_node; /* node in the rb tree */ | |
82 | atomic_t ref; | |
83 | struct rw_semaphore consumer_rwsem; | |
84 | struct list_head pending_list; | |
85 | struct uprobe_consumer *consumers; | |
86 | struct inode *inode; /* Also hold a ref to inode */ | |
87 | loff_t offset; | |
88 | int flags; | |
89 | struct arch_uprobe arch; | |
90 | }; | |
91 | ||
2b144498 SD |
92 | /* |
93 | * valid_vma: Verify if the specified vma is an executable vma | |
94 | * Relax restrictions while unregistering: vm_flags might have | |
95 | * changed after breakpoint was inserted. | |
96 | * - is_register: indicates if we are in register context. | |
97 | * - Return 1 if the specified virtual address is in an | |
98 | * executable vma. | |
99 | */ | |
100 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
101 | { | |
102 | if (!vma->vm_file) | |
103 | return false; | |
104 | ||
105 | if (!is_register) | |
106 | return true; | |
107 | ||
ea131377 ON |
108 | if ((vma->vm_flags & (VM_HUGETLB|VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)) |
109 | == (VM_READ|VM_EXEC)) | |
2b144498 SD |
110 | return true; |
111 | ||
112 | return false; | |
113 | } | |
114 | ||
57683f72 | 115 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 116 | { |
57683f72 | 117 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
118 | } |
119 | ||
cb113b47 ON |
120 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
121 | { | |
122 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
123 | } | |
124 | ||
2b144498 SD |
125 | /** |
126 | * __replace_page - replace page in vma by new page. | |
127 | * based on replace_page in mm/ksm.c | |
128 | * | |
129 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 130 | * @addr: address the old @page is mapped at |
2b144498 SD |
131 | * @page: the cowed page we are replacing by kpage |
132 | * @kpage: the modified page we replace page by | |
133 | * | |
134 | * Returns 0 on success, -EFAULT on failure. | |
135 | */ | |
c517ee74 ON |
136 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
137 | struct page *page, struct page *kpage) | |
2b144498 SD |
138 | { |
139 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
140 | spinlock_t *ptl; |
141 | pte_t *ptep; | |
9f92448c | 142 | int err; |
2b144498 | 143 | |
9f92448c ON |
144 | /* freeze PageSwapCache() for try_to_free_swap() below */ |
145 | lock_page(page); | |
146 | ||
147 | err = -EAGAIN; | |
5323ce71 | 148 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 149 | if (!ptep) |
9f92448c | 150 | goto unlock; |
2b144498 SD |
151 | |
152 | get_page(kpage); | |
153 | page_add_new_anon_rmap(kpage, vma, addr); | |
154 | ||
7396fa81 SD |
155 | if (!PageAnon(page)) { |
156 | dec_mm_counter(mm, MM_FILEPAGES); | |
157 | inc_mm_counter(mm, MM_ANONPAGES); | |
158 | } | |
159 | ||
2b144498 SD |
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); | |
2b144498 | 169 | |
9f92448c ON |
170 | err = 0; |
171 | unlock: | |
172 | unlock_page(page); | |
173 | return err; | |
2b144498 SD |
174 | } |
175 | ||
176 | /** | |
5cb4ac3a | 177 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 178 | * @insn: instruction to be checked. |
5cb4ac3a | 179 | * Default implementation of is_swbp_insn |
2b144498 SD |
180 | * Returns true if @insn is a breakpoint instruction. |
181 | */ | |
5cb4ac3a | 182 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 183 | { |
5cb4ac3a | 184 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
185 | } |
186 | ||
187 | /* | |
188 | * NOTE: | |
189 | * Expect the breakpoint instruction to be the smallest size instruction for | |
190 | * the architecture. If an arch has variable length instruction and the | |
191 | * breakpoint instruction is not of the smallest length instruction | |
192 | * supported by that architecture then we need to modify read_opcode / | |
193 | * write_opcode accordingly. This would never be a problem for archs that | |
194 | * have fixed length instructions. | |
195 | */ | |
196 | ||
197 | /* | |
198 | * write_opcode - write the opcode at a given virtual address. | |
e3343e6a | 199 | * @auprobe: arch breakpointing information. |
2b144498 | 200 | * @mm: the probed process address space. |
2b144498 SD |
201 | * @vaddr: the virtual address to store the opcode. |
202 | * @opcode: opcode to be written at @vaddr. | |
203 | * | |
204 | * Called with mm->mmap_sem held (for read and with a reference to | |
205 | * mm). | |
206 | * | |
207 | * For mm @mm, write the opcode at @vaddr. | |
208 | * Return 0 (success) or a negative errno. | |
209 | */ | |
e3343e6a | 210 | static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, |
2b144498 SD |
211 | unsigned long vaddr, uprobe_opcode_t opcode) |
212 | { | |
213 | struct page *old_page, *new_page; | |
2b144498 SD |
214 | void *vaddr_old, *vaddr_new; |
215 | struct vm_area_struct *vma; | |
2b144498 | 216 | int ret; |
f403072c | 217 | |
5323ce71 | 218 | retry: |
2b144498 SD |
219 | /* Read the page with vaddr into memory */ |
220 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma); | |
221 | if (ret <= 0) | |
222 | return ret; | |
7b2d81d4 | 223 | |
2b144498 SD |
224 | ret = -ENOMEM; |
225 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
226 | if (!new_page) | |
9f92448c | 227 | goto put_old; |
2b144498 SD |
228 | |
229 | __SetPageUptodate(new_page); | |
230 | ||
2b144498 SD |
231 | /* copy the page now that we've got it stable */ |
232 | vaddr_old = kmap_atomic(old_page); | |
233 | vaddr_new = kmap_atomic(new_page); | |
234 | ||
235 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 236 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
237 | |
238 | kunmap_atomic(vaddr_new); | |
239 | kunmap_atomic(vaddr_old); | |
240 | ||
241 | ret = anon_vma_prepare(vma); | |
242 | if (ret) | |
9f92448c | 243 | goto put_new; |
2b144498 | 244 | |
c517ee74 | 245 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 246 | |
9f92448c | 247 | put_new: |
2b144498 | 248 | page_cache_release(new_page); |
9f92448c | 249 | put_old: |
7b2d81d4 IM |
250 | put_page(old_page); |
251 | ||
5323ce71 ON |
252 | if (unlikely(ret == -EAGAIN)) |
253 | goto retry; | |
2b144498 SD |
254 | return ret; |
255 | } | |
256 | ||
257 | /** | |
258 | * read_opcode - read the opcode at a given virtual address. | |
259 | * @mm: the probed process address space. | |
260 | * @vaddr: the virtual address to read the opcode. | |
261 | * @opcode: location to store the read opcode. | |
262 | * | |
263 | * Called with mm->mmap_sem held (for read and with a reference to | |
264 | * mm. | |
265 | * | |
266 | * For mm @mm, read the opcode at @vaddr and store it in @opcode. | |
267 | * Return 0 (success) or a negative errno. | |
268 | */ | |
7b2d81d4 | 269 | static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode) |
2b144498 SD |
270 | { |
271 | struct page *page; | |
272 | void *vaddr_new; | |
273 | int ret; | |
274 | ||
a3d7bb47 | 275 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); |
2b144498 SD |
276 | if (ret <= 0) |
277 | return ret; | |
278 | ||
279 | lock_page(page); | |
280 | vaddr_new = kmap_atomic(page); | |
281 | vaddr &= ~PAGE_MASK; | |
5cb4ac3a | 282 | memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
283 | kunmap_atomic(vaddr_new); |
284 | unlock_page(page); | |
7b2d81d4 IM |
285 | |
286 | put_page(page); | |
287 | ||
2b144498 SD |
288 | return 0; |
289 | } | |
290 | ||
5cb4ac3a | 291 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
2b144498 SD |
292 | { |
293 | uprobe_opcode_t opcode; | |
7b2d81d4 | 294 | int result; |
2b144498 | 295 | |
c00b2750 ON |
296 | if (current->mm == mm) { |
297 | pagefault_disable(); | |
298 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
299 | sizeof(opcode)); | |
300 | pagefault_enable(); | |
301 | ||
302 | if (likely(result == 0)) | |
303 | goto out; | |
304 | } | |
305 | ||
7b2d81d4 | 306 | result = read_opcode(mm, vaddr, &opcode); |
2b144498 SD |
307 | if (result) |
308 | return result; | |
c00b2750 | 309 | out: |
5cb4ac3a | 310 | if (is_swbp_insn(&opcode)) |
2b144498 SD |
311 | return 1; |
312 | ||
313 | return 0; | |
314 | } | |
315 | ||
316 | /** | |
5cb4ac3a | 317 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 318 | * @auprobe: arch specific probepoint information. |
2b144498 | 319 | * @mm: the probed process address space. |
2b144498 SD |
320 | * @vaddr: the virtual address to insert the opcode. |
321 | * | |
322 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
323 | * Return 0 (success) or a negative errno. | |
324 | */ | |
5cb4ac3a | 325 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 326 | { |
7b2d81d4 | 327 | int result; |
c5784de2 PZ |
328 | /* |
329 | * See the comment near uprobes_hash(). | |
330 | */ | |
5cb4ac3a | 331 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
332 | if (result == 1) |
333 | return -EEXIST; | |
334 | ||
335 | if (result) | |
336 | return result; | |
337 | ||
5cb4ac3a | 338 | return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
339 | } |
340 | ||
341 | /** | |
342 | * set_orig_insn - Restore the original instruction. | |
343 | * @mm: the probed process address space. | |
e3343e6a | 344 | * @auprobe: arch specific probepoint information. |
2b144498 SD |
345 | * @vaddr: the virtual address to insert the opcode. |
346 | * @verify: if true, verify existance of breakpoint instruction. | |
347 | * | |
348 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
349 | * Return 0 (success) or a negative errno. | |
350 | */ | |
7b2d81d4 | 351 | int __weak |
e3343e6a | 352 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify) |
2b144498 SD |
353 | { |
354 | if (verify) { | |
7b2d81d4 | 355 | int result; |
2b144498 | 356 | |
5cb4ac3a | 357 | result = is_swbp_at_addr(mm, vaddr); |
2b144498 SD |
358 | if (!result) |
359 | return -EINVAL; | |
360 | ||
361 | if (result != 1) | |
362 | return result; | |
363 | } | |
e3343e6a | 364 | return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
365 | } |
366 | ||
367 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
368 | { | |
369 | if (l->inode < r->inode) | |
370 | return -1; | |
7b2d81d4 | 371 | |
2b144498 SD |
372 | if (l->inode > r->inode) |
373 | return 1; | |
2b144498 | 374 | |
7b2d81d4 IM |
375 | if (l->offset < r->offset) |
376 | return -1; | |
377 | ||
378 | if (l->offset > r->offset) | |
379 | return 1; | |
2b144498 SD |
380 | |
381 | return 0; | |
382 | } | |
383 | ||
384 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
385 | { | |
386 | struct uprobe u = { .inode = inode, .offset = offset }; | |
387 | struct rb_node *n = uprobes_tree.rb_node; | |
388 | struct uprobe *uprobe; | |
389 | int match; | |
390 | ||
391 | while (n) { | |
392 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
393 | match = match_uprobe(&u, uprobe); | |
394 | if (!match) { | |
395 | atomic_inc(&uprobe->ref); | |
396 | return uprobe; | |
397 | } | |
7b2d81d4 | 398 | |
2b144498 SD |
399 | if (match < 0) |
400 | n = n->rb_left; | |
401 | else | |
402 | n = n->rb_right; | |
403 | } | |
404 | return NULL; | |
405 | } | |
406 | ||
407 | /* | |
408 | * Find a uprobe corresponding to a given inode:offset | |
409 | * Acquires uprobes_treelock | |
410 | */ | |
411 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
412 | { | |
413 | struct uprobe *uprobe; | |
414 | unsigned long flags; | |
415 | ||
416 | spin_lock_irqsave(&uprobes_treelock, flags); | |
417 | uprobe = __find_uprobe(inode, offset); | |
418 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 419 | |
2b144498 SD |
420 | return uprobe; |
421 | } | |
422 | ||
423 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
424 | { | |
425 | struct rb_node **p = &uprobes_tree.rb_node; | |
426 | struct rb_node *parent = NULL; | |
427 | struct uprobe *u; | |
428 | int match; | |
429 | ||
430 | while (*p) { | |
431 | parent = *p; | |
432 | u = rb_entry(parent, struct uprobe, rb_node); | |
433 | match = match_uprobe(uprobe, u); | |
434 | if (!match) { | |
435 | atomic_inc(&u->ref); | |
436 | return u; | |
437 | } | |
438 | ||
439 | if (match < 0) | |
440 | p = &parent->rb_left; | |
441 | else | |
442 | p = &parent->rb_right; | |
443 | ||
444 | } | |
7b2d81d4 | 445 | |
2b144498 SD |
446 | u = NULL; |
447 | rb_link_node(&uprobe->rb_node, parent, p); | |
448 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
449 | /* get access + creation ref */ | |
450 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 451 | |
2b144498 SD |
452 | return u; |
453 | } | |
454 | ||
455 | /* | |
7b2d81d4 | 456 | * Acquire uprobes_treelock. |
2b144498 SD |
457 | * Matching uprobe already exists in rbtree; |
458 | * increment (access refcount) and return the matching uprobe. | |
459 | * | |
460 | * No matching uprobe; insert the uprobe in rb_tree; | |
461 | * get a double refcount (access + creation) and return NULL. | |
462 | */ | |
463 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
464 | { | |
465 | unsigned long flags; | |
466 | struct uprobe *u; | |
467 | ||
468 | spin_lock_irqsave(&uprobes_treelock, flags); | |
469 | u = __insert_uprobe(uprobe); | |
470 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
7b2d81d4 | 471 | |
0326f5a9 SD |
472 | /* For now assume that the instruction need not be single-stepped */ |
473 | uprobe->flags |= UPROBE_SKIP_SSTEP; | |
474 | ||
2b144498 SD |
475 | return u; |
476 | } | |
477 | ||
478 | static void put_uprobe(struct uprobe *uprobe) | |
479 | { | |
480 | if (atomic_dec_and_test(&uprobe->ref)) | |
481 | kfree(uprobe); | |
482 | } | |
483 | ||
484 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
485 | { | |
486 | struct uprobe *uprobe, *cur_uprobe; | |
487 | ||
488 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
489 | if (!uprobe) | |
490 | return NULL; | |
491 | ||
492 | uprobe->inode = igrab(inode); | |
493 | uprobe->offset = offset; | |
494 | init_rwsem(&uprobe->consumer_rwsem); | |
2b144498 SD |
495 | |
496 | /* add to uprobes_tree, sorted on inode:offset */ | |
497 | cur_uprobe = insert_uprobe(uprobe); | |
498 | ||
499 | /* a uprobe exists for this inode:offset combination */ | |
500 | if (cur_uprobe) { | |
501 | kfree(uprobe); | |
502 | uprobe = cur_uprobe; | |
503 | iput(inode); | |
7b2d81d4 | 504 | } else { |
2b144498 | 505 | atomic_inc(&uprobe_events); |
7b2d81d4 IM |
506 | } |
507 | ||
2b144498 SD |
508 | return uprobe; |
509 | } | |
510 | ||
0326f5a9 SD |
511 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
512 | { | |
513 | struct uprobe_consumer *uc; | |
514 | ||
515 | if (!(uprobe->flags & UPROBE_RUN_HANDLER)) | |
516 | return; | |
517 | ||
518 | down_read(&uprobe->consumer_rwsem); | |
519 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
520 | if (!uc->filter || uc->filter(uc, current)) | |
521 | uc->handler(uc, regs); | |
522 | } | |
523 | up_read(&uprobe->consumer_rwsem); | |
524 | } | |
525 | ||
2b144498 | 526 | /* Returns the previous consumer */ |
7b2d81d4 | 527 | static struct uprobe_consumer * |
e3343e6a | 528 | consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
529 | { |
530 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
531 | uc->next = uprobe->consumers; |
532 | uprobe->consumers = uc; | |
2b144498 | 533 | up_write(&uprobe->consumer_rwsem); |
7b2d81d4 | 534 | |
e3343e6a | 535 | return uc->next; |
2b144498 SD |
536 | } |
537 | ||
538 | /* | |
e3343e6a SD |
539 | * For uprobe @uprobe, delete the consumer @uc. |
540 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
541 | * or return false. |
542 | */ | |
e3343e6a | 543 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
544 | { |
545 | struct uprobe_consumer **con; | |
546 | bool ret = false; | |
547 | ||
548 | down_write(&uprobe->consumer_rwsem); | |
549 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
550 | if (*con == uc) { |
551 | *con = uc->next; | |
2b144498 SD |
552 | ret = true; |
553 | break; | |
554 | } | |
555 | } | |
556 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 557 | |
2b144498 SD |
558 | return ret; |
559 | } | |
560 | ||
e3343e6a | 561 | static int |
d436615e | 562 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 563 | unsigned long nbytes, loff_t offset) |
2b144498 | 564 | { |
2b144498 SD |
565 | struct page *page; |
566 | void *vaddr; | |
593609a5 ON |
567 | unsigned long off; |
568 | pgoff_t idx; | |
2b144498 SD |
569 | |
570 | if (!filp) | |
571 | return -EINVAL; | |
572 | ||
cc359d18 ON |
573 | if (!mapping->a_ops->readpage) |
574 | return -EIO; | |
575 | ||
593609a5 ON |
576 | idx = offset >> PAGE_CACHE_SHIFT; |
577 | off = offset & ~PAGE_MASK; | |
2b144498 SD |
578 | |
579 | /* | |
580 | * Ensure that the page that has the original instruction is | |
581 | * populated and in page-cache. | |
582 | */ | |
583 | page = read_mapping_page(mapping, idx, filp); | |
584 | if (IS_ERR(page)) | |
585 | return PTR_ERR(page); | |
586 | ||
587 | vaddr = kmap_atomic(page); | |
593609a5 | 588 | memcpy(insn, vaddr + off, nbytes); |
2b144498 SD |
589 | kunmap_atomic(vaddr); |
590 | page_cache_release(page); | |
7b2d81d4 | 591 | |
2b144498 SD |
592 | return 0; |
593 | } | |
594 | ||
d436615e | 595 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
596 | { |
597 | struct address_space *mapping; | |
2b144498 | 598 | unsigned long nbytes; |
7b2d81d4 | 599 | int bytes; |
2b144498 | 600 | |
d436615e | 601 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
602 | mapping = uprobe->inode->i_mapping; |
603 | ||
604 | /* Instruction at end of binary; copy only available bytes */ | |
605 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
606 | bytes = uprobe->inode->i_size - uprobe->offset; | |
607 | else | |
608 | bytes = MAX_UINSN_BYTES; | |
609 | ||
610 | /* Instruction at the page-boundary; copy bytes in second page */ | |
611 | if (nbytes < bytes) { | |
fc36f595 ON |
612 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
613 | bytes - nbytes, uprobe->offset + nbytes); | |
614 | if (err) | |
615 | return err; | |
2b144498 SD |
616 | bytes = nbytes; |
617 | } | |
d436615e | 618 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
619 | } |
620 | ||
682968e0 SD |
621 | /* |
622 | * How mm->uprobes_state.count gets updated | |
623 | * uprobe_mmap() increments the count if | |
624 | * - it successfully adds a breakpoint. | |
625 | * - it cannot add a breakpoint, but sees that there is a underlying | |
626 | * breakpoint (via a is_swbp_at_addr()). | |
627 | * | |
628 | * uprobe_munmap() decrements the count if | |
629 | * - it sees a underlying breakpoint, (via is_swbp_at_addr) | |
630 | * (Subsequent uprobe_unregister wouldnt find the breakpoint | |
631 | * unless a uprobe_mmap kicks in, since the old vma would be | |
632 | * dropped just after uprobe_munmap.) | |
633 | * | |
634 | * uprobe_register increments the count if: | |
635 | * - it successfully adds a breakpoint. | |
636 | * | |
637 | * uprobe_unregister decrements the count if: | |
638 | * - it sees a underlying breakpoint and removes successfully. | |
639 | * (via is_swbp_at_addr) | |
640 | * (Subsequent uprobe_munmap wouldnt find the breakpoint | |
641 | * since there is no underlying breakpoint after the | |
642 | * breakpoint removal.) | |
643 | */ | |
e3343e6a SD |
644 | static int |
645 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 646 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 647 | { |
2b144498 SD |
648 | int ret; |
649 | ||
650 | /* | |
651 | * If probe is being deleted, unregister thread could be done with | |
652 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
653 | * nobody will be able to cleanup. Also we could be from fork or | |
654 | * mremap path, where the probe might have already been inserted. | |
655 | * Hence behave as if probe already existed. | |
656 | */ | |
657 | if (!uprobe->consumers) | |
658 | return -EEXIST; | |
659 | ||
900771a4 | 660 | if (!(uprobe->flags & UPROBE_COPY_INSN)) { |
d436615e | 661 | ret = copy_insn(uprobe, vma->vm_file); |
2b144498 SD |
662 | if (ret) |
663 | return ret; | |
664 | ||
5cb4ac3a | 665 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) |
c1914a09 | 666 | return -ENOTSUPP; |
2b144498 | 667 | |
816c03fb | 668 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
669 | if (ret) |
670 | return ret; | |
671 | ||
d9c4a30e ON |
672 | /* write_opcode() assumes we don't cross page boundary */ |
673 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
674 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
675 | ||
900771a4 | 676 | uprobe->flags |= UPROBE_COPY_INSN; |
2b144498 | 677 | } |
682968e0 SD |
678 | |
679 | /* | |
680 | * Ideally, should be updating the probe count after the breakpoint | |
681 | * has been successfully inserted. However a thread could hit the | |
682 | * breakpoint we just inserted even before the probe count is | |
683 | * incremented. If this is the first breakpoint placed, breakpoint | |
684 | * notifier might ignore uprobes and pass the trap to the thread. | |
685 | * Hence increment before and decrement on failure. | |
686 | */ | |
687 | atomic_inc(&mm->uprobes_state.count); | |
816c03fb | 688 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
682968e0 SD |
689 | if (ret) |
690 | atomic_dec(&mm->uprobes_state.count); | |
2b144498 SD |
691 | |
692 | return ret; | |
693 | } | |
694 | ||
e3343e6a | 695 | static void |
816c03fb | 696 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 697 | { |
816c03fb | 698 | if (!set_orig_insn(&uprobe->arch, mm, vaddr, true)) |
682968e0 | 699 | atomic_dec(&mm->uprobes_state.count); |
2b144498 SD |
700 | } |
701 | ||
0326f5a9 | 702 | /* |
778b032d ON |
703 | * There could be threads that have already hit the breakpoint. They |
704 | * will recheck the current insn and restart if find_uprobe() fails. | |
705 | * See find_active_uprobe(). | |
0326f5a9 | 706 | */ |
2b144498 SD |
707 | static void delete_uprobe(struct uprobe *uprobe) |
708 | { | |
709 | unsigned long flags; | |
710 | ||
711 | spin_lock_irqsave(&uprobes_treelock, flags); | |
712 | rb_erase(&uprobe->rb_node, &uprobes_tree); | |
713 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
714 | iput(uprobe->inode); | |
715 | put_uprobe(uprobe); | |
716 | atomic_dec(&uprobe_events); | |
717 | } | |
718 | ||
26872090 ON |
719 | struct map_info { |
720 | struct map_info *next; | |
721 | struct mm_struct *mm; | |
816c03fb | 722 | unsigned long vaddr; |
26872090 ON |
723 | }; |
724 | ||
725 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 726 | { |
26872090 ON |
727 | struct map_info *next = info->next; |
728 | kfree(info); | |
729 | return next; | |
730 | } | |
731 | ||
732 | static struct map_info * | |
733 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
734 | { | |
735 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 SD |
736 | struct prio_tree_iter iter; |
737 | struct vm_area_struct *vma; | |
26872090 ON |
738 | struct map_info *curr = NULL; |
739 | struct map_info *prev = NULL; | |
740 | struct map_info *info; | |
741 | int more = 0; | |
2b144498 | 742 | |
26872090 ON |
743 | again: |
744 | mutex_lock(&mapping->i_mmap_mutex); | |
2b144498 SD |
745 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
746 | if (!valid_vma(vma, is_register)) | |
747 | continue; | |
748 | ||
7a5bfb66 ON |
749 | if (!prev && !more) { |
750 | /* | |
751 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
752 | * reclaim. This is optimistic, no harm done if it fails. | |
753 | */ | |
754 | prev = kmalloc(sizeof(struct map_info), | |
755 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
756 | if (prev) | |
757 | prev->next = NULL; | |
758 | } | |
26872090 ON |
759 | if (!prev) { |
760 | more++; | |
761 | continue; | |
2b144498 | 762 | } |
2b144498 | 763 | |
26872090 ON |
764 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
765 | continue; | |
7b2d81d4 | 766 | |
26872090 ON |
767 | info = prev; |
768 | prev = prev->next; | |
769 | info->next = curr; | |
770 | curr = info; | |
2b144498 | 771 | |
26872090 | 772 | info->mm = vma->vm_mm; |
57683f72 | 773 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 774 | } |
2b144498 SD |
775 | mutex_unlock(&mapping->i_mmap_mutex); |
776 | ||
26872090 ON |
777 | if (!more) |
778 | goto out; | |
779 | ||
780 | prev = curr; | |
781 | while (curr) { | |
782 | mmput(curr->mm); | |
783 | curr = curr->next; | |
784 | } | |
7b2d81d4 | 785 | |
26872090 ON |
786 | do { |
787 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
788 | if (!info) { | |
789 | curr = ERR_PTR(-ENOMEM); | |
790 | goto out; | |
791 | } | |
792 | info->next = prev; | |
793 | prev = info; | |
794 | } while (--more); | |
795 | ||
796 | goto again; | |
797 | out: | |
798 | while (prev) | |
799 | prev = free_map_info(prev); | |
800 | return curr; | |
2b144498 SD |
801 | } |
802 | ||
803 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
804 | { | |
26872090 ON |
805 | struct map_info *info; |
806 | int err = 0; | |
2b144498 | 807 | |
26872090 ON |
808 | info = build_map_info(uprobe->inode->i_mapping, |
809 | uprobe->offset, is_register); | |
810 | if (IS_ERR(info)) | |
811 | return PTR_ERR(info); | |
7b2d81d4 | 812 | |
26872090 ON |
813 | while (info) { |
814 | struct mm_struct *mm = info->mm; | |
815 | struct vm_area_struct *vma; | |
7b2d81d4 | 816 | |
26872090 ON |
817 | if (err) |
818 | goto free; | |
7b2d81d4 | 819 | |
77fc4af1 | 820 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
821 | vma = find_vma(mm, info->vaddr); |
822 | if (!vma || !valid_vma(vma, is_register) || | |
823 | vma->vm_file->f_mapping->host != uprobe->inode) | |
26872090 ON |
824 | goto unlock; |
825 | ||
f4d6dfe5 ON |
826 | if (vma->vm_start > info->vaddr || |
827 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 828 | goto unlock; |
2b144498 | 829 | |
2b144498 | 830 | if (is_register) { |
26872090 | 831 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
c5784de2 PZ |
832 | /* |
833 | * We can race against uprobe_mmap(), see the | |
834 | * comment near uprobe_hash(). | |
835 | */ | |
26872090 ON |
836 | if (err == -EEXIST) |
837 | err = 0; | |
838 | } else { | |
839 | remove_breakpoint(uprobe, mm, info->vaddr); | |
2b144498 | 840 | } |
26872090 ON |
841 | unlock: |
842 | up_write(&mm->mmap_sem); | |
843 | free: | |
844 | mmput(mm); | |
845 | info = free_map_info(info); | |
2b144498 | 846 | } |
7b2d81d4 | 847 | |
26872090 | 848 | return err; |
2b144498 SD |
849 | } |
850 | ||
7b2d81d4 | 851 | static int __uprobe_register(struct uprobe *uprobe) |
2b144498 SD |
852 | { |
853 | return register_for_each_vma(uprobe, true); | |
854 | } | |
855 | ||
7b2d81d4 | 856 | static void __uprobe_unregister(struct uprobe *uprobe) |
2b144498 SD |
857 | { |
858 | if (!register_for_each_vma(uprobe, false)) | |
859 | delete_uprobe(uprobe); | |
860 | ||
861 | /* TODO : cant unregister? schedule a worker thread */ | |
862 | } | |
863 | ||
864 | /* | |
7b2d81d4 | 865 | * uprobe_register - register a probe |
2b144498 SD |
866 | * @inode: the file in which the probe has to be placed. |
867 | * @offset: offset from the start of the file. | |
e3343e6a | 868 | * @uc: information on howto handle the probe.. |
2b144498 | 869 | * |
7b2d81d4 | 870 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
871 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
872 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 873 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 874 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 875 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
876 | * unregisters. |
877 | * | |
878 | * Return errno if it cannot successully install probes | |
879 | * else return 0 (success) | |
880 | */ | |
e3343e6a | 881 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
882 | { |
883 | struct uprobe *uprobe; | |
7b2d81d4 | 884 | int ret; |
2b144498 | 885 | |
e3343e6a | 886 | if (!inode || !uc || uc->next) |
7b2d81d4 | 887 | return -EINVAL; |
2b144498 SD |
888 | |
889 | if (offset > i_size_read(inode)) | |
7b2d81d4 | 890 | return -EINVAL; |
2b144498 SD |
891 | |
892 | ret = 0; | |
893 | mutex_lock(uprobes_hash(inode)); | |
894 | uprobe = alloc_uprobe(inode, offset); | |
7b2d81d4 | 895 | |
e3343e6a | 896 | if (uprobe && !consumer_add(uprobe, uc)) { |
7b2d81d4 | 897 | ret = __uprobe_register(uprobe); |
2b144498 SD |
898 | if (ret) { |
899 | uprobe->consumers = NULL; | |
7b2d81d4 IM |
900 | __uprobe_unregister(uprobe); |
901 | } else { | |
900771a4 | 902 | uprobe->flags |= UPROBE_RUN_HANDLER; |
7b2d81d4 | 903 | } |
2b144498 SD |
904 | } |
905 | ||
906 | mutex_unlock(uprobes_hash(inode)); | |
907 | put_uprobe(uprobe); | |
908 | ||
909 | return ret; | |
910 | } | |
911 | ||
912 | /* | |
7b2d81d4 | 913 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
914 | * @inode: the file in which the probe has to be removed. |
915 | * @offset: offset from the start of the file. | |
e3343e6a | 916 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 917 | */ |
e3343e6a | 918 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 919 | { |
7b2d81d4 | 920 | struct uprobe *uprobe; |
2b144498 | 921 | |
e3343e6a | 922 | if (!inode || !uc) |
2b144498 SD |
923 | return; |
924 | ||
925 | uprobe = find_uprobe(inode, offset); | |
926 | if (!uprobe) | |
927 | return; | |
928 | ||
929 | mutex_lock(uprobes_hash(inode)); | |
2b144498 | 930 | |
e3343e6a | 931 | if (consumer_del(uprobe, uc)) { |
7b2d81d4 IM |
932 | if (!uprobe->consumers) { |
933 | __uprobe_unregister(uprobe); | |
900771a4 | 934 | uprobe->flags &= ~UPROBE_RUN_HANDLER; |
7b2d81d4 | 935 | } |
2b144498 SD |
936 | } |
937 | ||
2b144498 SD |
938 | mutex_unlock(uprobes_hash(inode)); |
939 | if (uprobe) | |
940 | put_uprobe(uprobe); | |
941 | } | |
942 | ||
891c3970 ON |
943 | static struct rb_node * |
944 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 945 | { |
2b144498 | 946 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
947 | |
948 | while (n) { | |
891c3970 | 949 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 950 | |
891c3970 | 951 | if (inode < u->inode) { |
2b144498 | 952 | n = n->rb_left; |
891c3970 | 953 | } else if (inode > u->inode) { |
2b144498 | 954 | n = n->rb_right; |
891c3970 ON |
955 | } else { |
956 | if (max < u->offset) | |
957 | n = n->rb_left; | |
958 | else if (min > u->offset) | |
959 | n = n->rb_right; | |
960 | else | |
961 | break; | |
962 | } | |
2b144498 | 963 | } |
7b2d81d4 | 964 | |
891c3970 | 965 | return n; |
2b144498 SD |
966 | } |
967 | ||
968 | /* | |
891c3970 | 969 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 970 | */ |
891c3970 ON |
971 | static void build_probe_list(struct inode *inode, |
972 | struct vm_area_struct *vma, | |
973 | unsigned long start, unsigned long end, | |
974 | struct list_head *head) | |
2b144498 | 975 | { |
891c3970 | 976 | loff_t min, max; |
2b144498 | 977 | unsigned long flags; |
891c3970 ON |
978 | struct rb_node *n, *t; |
979 | struct uprobe *u; | |
7b2d81d4 | 980 | |
891c3970 | 981 | INIT_LIST_HEAD(head); |
cb113b47 | 982 | min = vaddr_to_offset(vma, start); |
891c3970 | 983 | max = min + (end - start) - 1; |
2b144498 | 984 | |
891c3970 ON |
985 | spin_lock_irqsave(&uprobes_treelock, flags); |
986 | n = find_node_in_range(inode, min, max); | |
987 | if (n) { | |
988 | for (t = n; t; t = rb_prev(t)) { | |
989 | u = rb_entry(t, struct uprobe, rb_node); | |
990 | if (u->inode != inode || u->offset < min) | |
991 | break; | |
992 | list_add(&u->pending_list, head); | |
993 | atomic_inc(&u->ref); | |
994 | } | |
995 | for (t = n; (t = rb_next(t)); ) { | |
996 | u = rb_entry(t, struct uprobe, rb_node); | |
997 | if (u->inode != inode || u->offset > max) | |
998 | break; | |
999 | list_add(&u->pending_list, head); | |
1000 | atomic_inc(&u->ref); | |
1001 | } | |
2b144498 SD |
1002 | } |
1003 | spin_unlock_irqrestore(&uprobes_treelock, flags); | |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * Called from mmap_region. | |
1008 | * called with mm->mmap_sem acquired. | |
1009 | * | |
1010 | * Return -ve no if we fail to insert probes and we cannot | |
1011 | * bail-out. | |
7b2d81d4 IM |
1012 | * Return 0 otherwise. i.e: |
1013 | * | |
2b144498 SD |
1014 | * - successful insertion of probes |
1015 | * - (or) no possible probes to be inserted. | |
1016 | * - (or) insertion of probes failed but we can bail-out. | |
1017 | */ | |
7b2d81d4 | 1018 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1019 | { |
1020 | struct list_head tmp_list; | |
665605a2 | 1021 | struct uprobe *uprobe, *u; |
2b144498 | 1022 | struct inode *inode; |
682968e0 | 1023 | int ret, count; |
2b144498 SD |
1024 | |
1025 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) | |
7b2d81d4 | 1026 | return 0; |
2b144498 SD |
1027 | |
1028 | inode = vma->vm_file->f_mapping->host; | |
1029 | if (!inode) | |
7b2d81d4 | 1030 | return 0; |
2b144498 | 1031 | |
2b144498 | 1032 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1033 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
7b2d81d4 IM |
1034 | |
1035 | ret = 0; | |
682968e0 | 1036 | count = 0; |
7b2d81d4 | 1037 | |
665605a2 | 1038 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
2b144498 | 1039 | if (!ret) { |
57683f72 | 1040 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
682968e0 | 1041 | |
682968e0 | 1042 | ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
c5784de2 PZ |
1043 | /* |
1044 | * We can race against uprobe_register(), see the | |
1045 | * comment near uprobe_hash(). | |
1046 | */ | |
682968e0 SD |
1047 | if (ret == -EEXIST) { |
1048 | ret = 0; | |
1049 | ||
1050 | if (!is_swbp_at_addr(vma->vm_mm, vaddr)) | |
1051 | continue; | |
1052 | ||
1053 | /* | |
1054 | * Unable to insert a breakpoint, but | |
1055 | * breakpoint lies underneath. Increment the | |
1056 | * probe count. | |
1057 | */ | |
1058 | atomic_inc(&vma->vm_mm->uprobes_state.count); | |
1059 | } | |
1060 | ||
1061 | if (!ret) | |
1062 | count++; | |
2b144498 SD |
1063 | } |
1064 | put_uprobe(uprobe); | |
1065 | } | |
1066 | ||
1067 | mutex_unlock(uprobes_mmap_hash(inode)); | |
1068 | ||
682968e0 SD |
1069 | if (ret) |
1070 | atomic_sub(count, &vma->vm_mm->uprobes_state.count); | |
1071 | ||
2b144498 SD |
1072 | return ret; |
1073 | } | |
1074 | ||
682968e0 SD |
1075 | /* |
1076 | * Called in context of a munmap of a vma. | |
1077 | */ | |
cbc91f71 | 1078 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 SD |
1079 | { |
1080 | struct list_head tmp_list; | |
665605a2 | 1081 | struct uprobe *uprobe, *u; |
682968e0 SD |
1082 | struct inode *inode; |
1083 | ||
1084 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) | |
1085 | return; | |
1086 | ||
2fd611a9 ON |
1087 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1088 | return; | |
1089 | ||
682968e0 SD |
1090 | if (!atomic_read(&vma->vm_mm->uprobes_state.count)) |
1091 | return; | |
1092 | ||
1093 | inode = vma->vm_file->f_mapping->host; | |
1094 | if (!inode) | |
1095 | return; | |
1096 | ||
682968e0 | 1097 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1098 | build_probe_list(inode, vma, start, end, &tmp_list); |
682968e0 | 1099 | |
665605a2 | 1100 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
57683f72 | 1101 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
891c3970 ON |
1102 | /* |
1103 | * An unregister could have removed the probe before | |
1104 | * unmap. So check before we decrement the count. | |
1105 | */ | |
1106 | if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1) | |
1107 | atomic_dec(&vma->vm_mm->uprobes_state.count); | |
682968e0 SD |
1108 | put_uprobe(uprobe); |
1109 | } | |
1110 | mutex_unlock(uprobes_mmap_hash(inode)); | |
1111 | } | |
1112 | ||
d4b3b638 SD |
1113 | /* Slot allocation for XOL */ |
1114 | static int xol_add_vma(struct xol_area *area) | |
1115 | { | |
1116 | struct mm_struct *mm; | |
1117 | int ret; | |
1118 | ||
1119 | area->page = alloc_page(GFP_HIGHUSER); | |
1120 | if (!area->page) | |
1121 | return -ENOMEM; | |
1122 | ||
1123 | ret = -EALREADY; | |
1124 | mm = current->mm; | |
1125 | ||
1126 | down_write(&mm->mmap_sem); | |
1127 | if (mm->uprobes_state.xol_area) | |
1128 | goto fail; | |
1129 | ||
1130 | ret = -ENOMEM; | |
1131 | ||
1132 | /* Try to map as high as possible, this is only a hint. */ | |
1133 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1134 | if (area->vaddr & ~PAGE_MASK) { | |
1135 | ret = area->vaddr; | |
1136 | goto fail; | |
1137 | } | |
1138 | ||
1139 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1140 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1141 | if (ret) | |
1142 | goto fail; | |
1143 | ||
1144 | smp_wmb(); /* pairs with get_xol_area() */ | |
1145 | mm->uprobes_state.xol_area = area; | |
1146 | ret = 0; | |
1147 | ||
1148 | fail: | |
1149 | up_write(&mm->mmap_sem); | |
1150 | if (ret) | |
1151 | __free_page(area->page); | |
1152 | ||
1153 | return ret; | |
1154 | } | |
1155 | ||
1156 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1157 | { | |
1158 | struct xol_area *area; | |
1159 | ||
1160 | area = mm->uprobes_state.xol_area; | |
1161 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1162 | ||
1163 | return area; | |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * xol_alloc_area - Allocate process's xol_area. | |
1168 | * This area will be used for storing instructions for execution out of | |
1169 | * line. | |
1170 | * | |
1171 | * Returns the allocated area or NULL. | |
1172 | */ | |
1173 | static struct xol_area *xol_alloc_area(void) | |
1174 | { | |
1175 | struct xol_area *area; | |
1176 | ||
1177 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1178 | if (unlikely(!area)) | |
1179 | return NULL; | |
1180 | ||
1181 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1182 | ||
1183 | if (!area->bitmap) | |
1184 | goto fail; | |
1185 | ||
1186 | init_waitqueue_head(&area->wq); | |
1187 | if (!xol_add_vma(area)) | |
1188 | return area; | |
1189 | ||
1190 | fail: | |
1191 | kfree(area->bitmap); | |
1192 | kfree(area); | |
1193 | ||
1194 | return get_xol_area(current->mm); | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * uprobe_clear_state - Free the area allocated for slots. | |
1199 | */ | |
1200 | void uprobe_clear_state(struct mm_struct *mm) | |
1201 | { | |
1202 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1203 | ||
1204 | if (!area) | |
1205 | return; | |
1206 | ||
1207 | put_page(area->page); | |
1208 | kfree(area->bitmap); | |
1209 | kfree(area); | |
1210 | } | |
1211 | ||
1212 | /* | |
1213 | * uprobe_reset_state - Free the area allocated for slots. | |
1214 | */ | |
1215 | void uprobe_reset_state(struct mm_struct *mm) | |
1216 | { | |
1217 | mm->uprobes_state.xol_area = NULL; | |
682968e0 | 1218 | atomic_set(&mm->uprobes_state.count, 0); |
d4b3b638 SD |
1219 | } |
1220 | ||
1221 | /* | |
1222 | * - search for a free slot. | |
1223 | */ | |
1224 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1225 | { | |
1226 | unsigned long slot_addr; | |
1227 | int slot_nr; | |
1228 | ||
1229 | do { | |
1230 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1231 | if (slot_nr < UINSNS_PER_PAGE) { | |
1232 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1233 | break; | |
1234 | ||
1235 | slot_nr = UINSNS_PER_PAGE; | |
1236 | continue; | |
1237 | } | |
1238 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1239 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1240 | ||
1241 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1242 | atomic_inc(&area->slot_count); | |
1243 | ||
1244 | return slot_addr; | |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * xol_get_insn_slot - If was not allocated a slot, then | |
1249 | * allocate a slot. | |
1250 | * Returns the allocated slot address or 0. | |
1251 | */ | |
1252 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1253 | { | |
1254 | struct xol_area *area; | |
1255 | unsigned long offset; | |
1256 | void *vaddr; | |
1257 | ||
1258 | area = get_xol_area(current->mm); | |
1259 | if (!area) { | |
1260 | area = xol_alloc_area(); | |
1261 | if (!area) | |
1262 | return 0; | |
1263 | } | |
1264 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1265 | ||
1266 | /* | |
1267 | * Initialize the slot if xol_vaddr points to valid | |
1268 | * instruction slot. | |
1269 | */ | |
1270 | if (unlikely(!current->utask->xol_vaddr)) | |
1271 | return 0; | |
1272 | ||
1273 | current->utask->vaddr = slot_addr; | |
1274 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1275 | vaddr = kmap_atomic(area->page); | |
1276 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1277 | kunmap_atomic(vaddr); | |
1278 | ||
1279 | return current->utask->xol_vaddr; | |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * xol_free_insn_slot - If slot was earlier allocated by | |
1284 | * @xol_get_insn_slot(), make the slot available for | |
1285 | * subsequent requests. | |
1286 | */ | |
1287 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1288 | { | |
1289 | struct xol_area *area; | |
1290 | unsigned long vma_end; | |
1291 | unsigned long slot_addr; | |
1292 | ||
1293 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1294 | return; | |
1295 | ||
1296 | slot_addr = tsk->utask->xol_vaddr; | |
1297 | ||
1298 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1299 | return; | |
1300 | ||
1301 | area = tsk->mm->uprobes_state.xol_area; | |
1302 | vma_end = area->vaddr + PAGE_SIZE; | |
1303 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1304 | unsigned long offset; | |
1305 | int slot_nr; | |
1306 | ||
1307 | offset = slot_addr - area->vaddr; | |
1308 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1309 | if (slot_nr >= UINSNS_PER_PAGE) | |
1310 | return; | |
1311 | ||
1312 | clear_bit(slot_nr, area->bitmap); | |
1313 | atomic_dec(&area->slot_count); | |
1314 | if (waitqueue_active(&area->wq)) | |
1315 | wake_up(&area->wq); | |
1316 | ||
1317 | tsk->utask->xol_vaddr = 0; | |
1318 | } | |
1319 | } | |
1320 | ||
0326f5a9 SD |
1321 | /** |
1322 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1323 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1324 | * instruction. | |
1325 | * Return the address of the breakpoint instruction. | |
1326 | */ | |
1327 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1328 | { | |
1329 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1330 | } | |
1331 | ||
1332 | /* | |
1333 | * Called with no locks held. | |
1334 | * Called in context of a exiting or a exec-ing thread. | |
1335 | */ | |
1336 | void uprobe_free_utask(struct task_struct *t) | |
1337 | { | |
1338 | struct uprobe_task *utask = t->utask; | |
1339 | ||
0326f5a9 SD |
1340 | if (!utask) |
1341 | return; | |
1342 | ||
1343 | if (utask->active_uprobe) | |
1344 | put_uprobe(utask->active_uprobe); | |
1345 | ||
d4b3b638 | 1346 | xol_free_insn_slot(t); |
0326f5a9 SD |
1347 | kfree(utask); |
1348 | t->utask = NULL; | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * Called in context of a new clone/fork from copy_process. | |
1353 | */ | |
1354 | void uprobe_copy_process(struct task_struct *t) | |
1355 | { | |
1356 | t->utask = NULL; | |
0326f5a9 SD |
1357 | } |
1358 | ||
1359 | /* | |
1360 | * Allocate a uprobe_task object for the task. | |
1361 | * Called when the thread hits a breakpoint for the first time. | |
1362 | * | |
1363 | * Returns: | |
1364 | * - pointer to new uprobe_task on success | |
1365 | * - NULL otherwise | |
1366 | */ | |
1367 | static struct uprobe_task *add_utask(void) | |
1368 | { | |
1369 | struct uprobe_task *utask; | |
1370 | ||
1371 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1372 | if (unlikely(!utask)) | |
1373 | return NULL; | |
1374 | ||
0326f5a9 SD |
1375 | current->utask = utask; |
1376 | return utask; | |
1377 | } | |
1378 | ||
1379 | /* Prepare to single-step probed instruction out of line. */ | |
1380 | static int | |
1381 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1382 | { | |
d4b3b638 SD |
1383 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1384 | return 0; | |
1385 | ||
0326f5a9 SD |
1386 | return -EFAULT; |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * If we are singlestepping, then ensure this thread is not connected to | |
1391 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1392 | * triggers the signal, restart the original insn even if the task is | |
1393 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1394 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1395 | * _same_ instruction should be repeated again after return from the signal | |
1396 | * handler, and SSTEP can never finish in this case. | |
1397 | */ | |
1398 | bool uprobe_deny_signal(void) | |
1399 | { | |
1400 | struct task_struct *t = current; | |
1401 | struct uprobe_task *utask = t->utask; | |
1402 | ||
1403 | if (likely(!utask || !utask->active_uprobe)) | |
1404 | return false; | |
1405 | ||
1406 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1407 | ||
1408 | if (signal_pending(t)) { | |
1409 | spin_lock_irq(&t->sighand->siglock); | |
1410 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1411 | spin_unlock_irq(&t->sighand->siglock); | |
1412 | ||
1413 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1414 | utask->state = UTASK_SSTEP_TRAPPED; | |
1415 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1416 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1417 | } | |
1418 | } | |
1419 | ||
1420 | return true; | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * Avoid singlestepping the original instruction if the original instruction | |
1425 | * is a NOP or can be emulated. | |
1426 | */ | |
1427 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1428 | { | |
1429 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) | |
1430 | return true; | |
1431 | ||
1432 | uprobe->flags &= ~UPROBE_SKIP_SSTEP; | |
1433 | return false; | |
1434 | } | |
1435 | ||
d790d346 | 1436 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1437 | { |
3a9ea052 ON |
1438 | struct mm_struct *mm = current->mm; |
1439 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1440 | struct vm_area_struct *vma; |
0326f5a9 | 1441 | |
0326f5a9 SD |
1442 | down_read(&mm->mmap_sem); |
1443 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1444 | if (vma && vma->vm_start <= bp_vaddr) { |
1445 | if (valid_vma(vma, false)) { | |
cb113b47 ON |
1446 | struct inode *inode = vma->vm_file->f_mapping->host; |
1447 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); | |
0326f5a9 | 1448 | |
3a9ea052 ON |
1449 | uprobe = find_uprobe(inode, offset); |
1450 | } | |
d790d346 ON |
1451 | |
1452 | if (!uprobe) | |
1453 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1454 | } else { | |
1455 | *is_swbp = -EFAULT; | |
0326f5a9 | 1456 | } |
0326f5a9 SD |
1457 | up_read(&mm->mmap_sem); |
1458 | ||
3a9ea052 ON |
1459 | return uprobe; |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * Run handler and ask thread to singlestep. | |
1464 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1465 | */ | |
1466 | static void handle_swbp(struct pt_regs *regs) | |
1467 | { | |
1468 | struct uprobe_task *utask; | |
1469 | struct uprobe *uprobe; | |
1470 | unsigned long bp_vaddr; | |
56bb4cf6 | 1471 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1472 | |
1473 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1474 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1475 | |
0326f5a9 | 1476 | if (!uprobe) { |
56bb4cf6 ON |
1477 | if (is_swbp > 0) { |
1478 | /* No matching uprobe; signal SIGTRAP. */ | |
1479 | send_sig(SIGTRAP, current, 0); | |
1480 | } else { | |
1481 | /* | |
1482 | * Either we raced with uprobe_unregister() or we can't | |
1483 | * access this memory. The latter is only possible if | |
1484 | * another thread plays with our ->mm. In both cases | |
1485 | * we can simply restart. If this vma was unmapped we | |
1486 | * can pretend this insn was not executed yet and get | |
1487 | * the (correct) SIGSEGV after restart. | |
1488 | */ | |
1489 | instruction_pointer_set(regs, bp_vaddr); | |
1490 | } | |
0326f5a9 SD |
1491 | return; |
1492 | } | |
1493 | ||
1494 | utask = current->utask; | |
1495 | if (!utask) { | |
1496 | utask = add_utask(); | |
1497 | /* Cannot allocate; re-execute the instruction. */ | |
1498 | if (!utask) | |
1499 | goto cleanup_ret; | |
1500 | } | |
1501 | utask->active_uprobe = uprobe; | |
1502 | handler_chain(uprobe, regs); | |
1503 | if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs)) | |
1504 | goto cleanup_ret; | |
1505 | ||
1506 | utask->state = UTASK_SSTEP; | |
1507 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { | |
1508 | user_enable_single_step(current); | |
1509 | return; | |
1510 | } | |
1511 | ||
1512 | cleanup_ret: | |
1513 | if (utask) { | |
1514 | utask->active_uprobe = NULL; | |
1515 | utask->state = UTASK_RUNNING; | |
1516 | } | |
1517 | if (uprobe) { | |
1518 | if (!(uprobe->flags & UPROBE_SKIP_SSTEP)) | |
1519 | ||
1520 | /* | |
1521 | * cannot singlestep; cannot skip instruction; | |
1522 | * re-execute the instruction. | |
1523 | */ | |
1524 | instruction_pointer_set(regs, bp_vaddr); | |
1525 | ||
1526 | put_uprobe(uprobe); | |
1527 | } | |
1528 | } | |
1529 | ||
1530 | /* | |
1531 | * Perform required fix-ups and disable singlestep. | |
1532 | * Allow pending signals to take effect. | |
1533 | */ | |
1534 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1535 | { | |
1536 | struct uprobe *uprobe; | |
1537 | ||
1538 | uprobe = utask->active_uprobe; | |
1539 | if (utask->state == UTASK_SSTEP_ACK) | |
1540 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1541 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1542 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1543 | else | |
1544 | WARN_ON_ONCE(1); | |
1545 | ||
1546 | put_uprobe(uprobe); | |
1547 | utask->active_uprobe = NULL; | |
1548 | utask->state = UTASK_RUNNING; | |
1549 | user_disable_single_step(current); | |
d4b3b638 | 1550 | xol_free_insn_slot(current); |
0326f5a9 SD |
1551 | |
1552 | spin_lock_irq(¤t->sighand->siglock); | |
1553 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1554 | spin_unlock_irq(¤t->sighand->siglock); | |
1555 | } | |
1556 | ||
1557 | /* | |
1558 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on | |
1559 | * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and | |
1560 | * allows the thread to return from interrupt. | |
1561 | * | |
1562 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and | |
1563 | * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from | |
1564 | * interrupt. | |
1565 | * | |
1566 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1567 | * uprobe_notify_resume(). | |
1568 | */ | |
1569 | void uprobe_notify_resume(struct pt_regs *regs) | |
1570 | { | |
1571 | struct uprobe_task *utask; | |
1572 | ||
1573 | utask = current->utask; | |
1574 | if (!utask || utask->state == UTASK_BP_HIT) | |
1575 | handle_swbp(regs); | |
1576 | else | |
1577 | handle_singlestep(utask, regs); | |
1578 | } | |
1579 | ||
1580 | /* | |
1581 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1582 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1583 | */ | |
1584 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1585 | { | |
1586 | struct uprobe_task *utask; | |
1587 | ||
682968e0 SD |
1588 | if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count)) |
1589 | /* task is currently not uprobed */ | |
0326f5a9 SD |
1590 | return 0; |
1591 | ||
1592 | utask = current->utask; | |
1593 | if (utask) | |
1594 | utask->state = UTASK_BP_HIT; | |
1595 | ||
1596 | set_thread_flag(TIF_UPROBE); | |
0326f5a9 SD |
1597 | |
1598 | return 1; | |
1599 | } | |
1600 | ||
1601 | /* | |
1602 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1603 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1604 | */ | |
1605 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1606 | { | |
1607 | struct uprobe_task *utask = current->utask; | |
1608 | ||
1609 | if (!current->mm || !utask || !utask->active_uprobe) | |
1610 | /* task is currently not uprobed */ | |
1611 | return 0; | |
1612 | ||
1613 | utask->state = UTASK_SSTEP_ACK; | |
1614 | set_thread_flag(TIF_UPROBE); | |
1615 | return 1; | |
1616 | } | |
1617 | ||
1618 | static struct notifier_block uprobe_exception_nb = { | |
1619 | .notifier_call = arch_uprobe_exception_notify, | |
1620 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1621 | }; | |
1622 | ||
2b144498 SD |
1623 | static int __init init_uprobes(void) |
1624 | { | |
1625 | int i; | |
1626 | ||
1627 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1628 | mutex_init(&uprobes_mutex[i]); | |
1629 | mutex_init(&uprobes_mmap_mutex[i]); | |
1630 | } | |
0326f5a9 SD |
1631 | |
1632 | return register_die_notifier(&uprobe_exception_nb); | |
2b144498 | 1633 | } |
0326f5a9 | 1634 | module_init(init_uprobes); |
2b144498 SD |
1635 | |
1636 | static void __exit exit_uprobes(void) | |
1637 | { | |
1638 | } | |
2b144498 | 1639 | module_exit(exit_uprobes); |