Commit | Line | Data |
---|---|---|
2b144498 | 1 | /* |
7b2d81d4 | 2 | * User-space Probes (UProbes) |
2b144498 SD |
3 | * |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
35aa621b | 18 | * Copyright (C) IBM Corporation, 2008-2012 |
2b144498 SD |
19 | * Authors: |
20 | * Srikar Dronamraju | |
21 | * Jim Keniston | |
35aa621b | 22 | * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> |
2b144498 SD |
23 | */ |
24 | ||
25 | #include <linux/kernel.h> | |
26 | #include <linux/highmem.h> | |
27 | #include <linux/pagemap.h> /* read_mapping_page */ | |
28 | #include <linux/slab.h> | |
29 | #include <linux/sched.h> | |
30 | #include <linux/rmap.h> /* anon_vma_prepare */ | |
31 | #include <linux/mmu_notifier.h> /* set_pte_at_notify */ | |
32 | #include <linux/swap.h> /* try_to_free_swap */ | |
0326f5a9 SD |
33 | #include <linux/ptrace.h> /* user_enable_single_step */ |
34 | #include <linux/kdebug.h> /* notifier mechanism */ | |
194f8dcb | 35 | #include "../../mm/internal.h" /* munlock_vma_page */ |
7b2d81d4 | 36 | |
2b144498 SD |
37 | #include <linux/uprobes.h> |
38 | ||
d4b3b638 SD |
39 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
40 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
41 | ||
2b144498 | 42 | static struct rb_root uprobes_tree = RB_ROOT; |
7b2d81d4 | 43 | |
2b144498 SD |
44 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
45 | ||
46 | #define UPROBES_HASH_SZ 13 | |
7b2d81d4 | 47 | |
c5784de2 PZ |
48 | /* |
49 | * We need separate register/unregister and mmap/munmap lock hashes because | |
50 | * of mmap_sem nesting. | |
51 | * | |
52 | * uprobe_register() needs to install probes on (potentially) all processes | |
53 | * and thus needs to acquire multiple mmap_sems (consequtively, not | |
54 | * concurrently), whereas uprobe_mmap() is called while holding mmap_sem | |
55 | * for the particular process doing the mmap. | |
56 | * | |
57 | * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem | |
58 | * because of lock order against i_mmap_mutex. This means there's a hole in | |
59 | * the register vma iteration where a mmap() can happen. | |
60 | * | |
61 | * Thus uprobe_register() can race with uprobe_mmap() and we can try and | |
62 | * install a probe where one is already installed. | |
63 | */ | |
64 | ||
2b144498 SD |
65 | /* serialize (un)register */ |
66 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
67 | |
68 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
69 | |
70 | /* serialize uprobe->pending_list */ | |
71 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 72 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 SD |
73 | |
74 | /* | |
7b2d81d4 | 75 | * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe |
2b144498 SD |
76 | * events active at this time. Probably a fine grained per inode count is |
77 | * better? | |
78 | */ | |
79 | static atomic_t uprobe_events = ATOMIC_INIT(0); | |
80 | ||
3ff54efd SD |
81 | struct uprobe { |
82 | struct rb_node rb_node; /* node in the rb tree */ | |
83 | atomic_t ref; | |
84 | struct rw_semaphore consumer_rwsem; | |
85 | struct list_head pending_list; | |
86 | struct uprobe_consumer *consumers; | |
87 | struct inode *inode; /* Also hold a ref to inode */ | |
88 | loff_t offset; | |
89 | int flags; | |
90 | struct arch_uprobe arch; | |
91 | }; | |
92 | ||
2b144498 SD |
93 | /* |
94 | * valid_vma: Verify if the specified vma is an executable vma | |
95 | * Relax restrictions while unregistering: vm_flags might have | |
96 | * changed after breakpoint was inserted. | |
97 | * - is_register: indicates if we are in register context. | |
98 | * - Return 1 if the specified virtual address is in an | |
99 | * executable vma. | |
100 | */ | |
101 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
102 | { | |
e40cfce6 | 103 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED; |
2b144498 | 104 | |
e40cfce6 ON |
105 | if (is_register) |
106 | flags |= VM_WRITE; | |
2b144498 | 107 | |
e40cfce6 | 108 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
109 | } |
110 | ||
57683f72 | 111 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 112 | { |
57683f72 | 113 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
114 | } |
115 | ||
cb113b47 ON |
116 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
117 | { | |
118 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
119 | } | |
120 | ||
2b144498 SD |
121 | /** |
122 | * __replace_page - replace page in vma by new page. | |
123 | * based on replace_page in mm/ksm.c | |
124 | * | |
125 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 126 | * @addr: address the old @page is mapped at |
2b144498 SD |
127 | * @page: the cowed page we are replacing by kpage |
128 | * @kpage: the modified page we replace page by | |
129 | * | |
130 | * Returns 0 on success, -EFAULT on failure. | |
131 | */ | |
c517ee74 ON |
132 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
133 | struct page *page, struct page *kpage) | |
2b144498 SD |
134 | { |
135 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
136 | spinlock_t *ptl; |
137 | pte_t *ptep; | |
9f92448c | 138 | int err; |
2b144498 | 139 | |
194f8dcb | 140 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
141 | lock_page(page); |
142 | ||
143 | err = -EAGAIN; | |
5323ce71 | 144 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 145 | if (!ptep) |
9f92448c | 146 | goto unlock; |
2b144498 SD |
147 | |
148 | get_page(kpage); | |
149 | page_add_new_anon_rmap(kpage, vma, addr); | |
150 | ||
7396fa81 SD |
151 | if (!PageAnon(page)) { |
152 | dec_mm_counter(mm, MM_FILEPAGES); | |
153 | inc_mm_counter(mm, MM_ANONPAGES); | |
154 | } | |
155 | ||
2b144498 SD |
156 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
157 | ptep_clear_flush(vma, addr, ptep); | |
158 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
159 | ||
160 | page_remove_rmap(page); | |
161 | if (!page_mapped(page)) | |
162 | try_to_free_swap(page); | |
2b144498 | 163 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 164 | |
194f8dcb ON |
165 | if (vma->vm_flags & VM_LOCKED) |
166 | munlock_vma_page(page); | |
167 | put_page(page); | |
168 | ||
9f92448c ON |
169 | err = 0; |
170 | unlock: | |
171 | unlock_page(page); | |
172 | return err; | |
2b144498 SD |
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 | ||
cceb55aa ON |
186 | static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode) |
187 | { | |
188 | void *kaddr = kmap_atomic(page); | |
189 | memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE); | |
190 | kunmap_atomic(kaddr); | |
191 | } | |
192 | ||
ed6f6a50 ON |
193 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
194 | { | |
195 | uprobe_opcode_t old_opcode; | |
196 | bool is_swbp; | |
197 | ||
198 | copy_opcode(page, vaddr, &old_opcode); | |
199 | is_swbp = is_swbp_insn(&old_opcode); | |
200 | ||
201 | if (is_swbp_insn(new_opcode)) { | |
202 | if (is_swbp) /* register: already installed? */ | |
203 | return 0; | |
204 | } else { | |
205 | if (!is_swbp) /* unregister: was it changed by us? */ | |
206 | return -EINVAL; | |
207 | } | |
208 | ||
209 | return 1; | |
210 | } | |
211 | ||
2b144498 SD |
212 | /* |
213 | * NOTE: | |
214 | * Expect the breakpoint instruction to be the smallest size instruction for | |
215 | * the architecture. If an arch has variable length instruction and the | |
216 | * breakpoint instruction is not of the smallest length instruction | |
cceb55aa | 217 | * supported by that architecture then we need to modify is_swbp_at_addr and |
2b144498 SD |
218 | * write_opcode accordingly. This would never be a problem for archs that |
219 | * have fixed length instructions. | |
220 | */ | |
221 | ||
222 | /* | |
223 | * write_opcode - write the opcode at a given virtual address. | |
224 | * @mm: the probed process address space. | |
2b144498 SD |
225 | * @vaddr: the virtual address to store the opcode. |
226 | * @opcode: opcode to be written at @vaddr. | |
227 | * | |
228 | * Called with mm->mmap_sem held (for read and with a reference to | |
229 | * mm). | |
230 | * | |
231 | * For mm @mm, write the opcode at @vaddr. | |
232 | * Return 0 (success) or a negative errno. | |
233 | */ | |
cceb55aa ON |
234 | static int write_opcode(struct mm_struct *mm, unsigned long vaddr, |
235 | uprobe_opcode_t opcode) | |
2b144498 SD |
236 | { |
237 | struct page *old_page, *new_page; | |
2b144498 SD |
238 | void *vaddr_old, *vaddr_new; |
239 | struct vm_area_struct *vma; | |
2b144498 | 240 | int ret; |
f403072c | 241 | |
5323ce71 | 242 | retry: |
2b144498 | 243 | /* Read the page with vaddr into memory */ |
75ed82ea | 244 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
245 | if (ret <= 0) |
246 | return ret; | |
7b2d81d4 | 247 | |
ed6f6a50 ON |
248 | ret = verify_opcode(old_page, vaddr, &opcode); |
249 | if (ret <= 0) | |
250 | goto put_old; | |
251 | ||
2b144498 SD |
252 | ret = -ENOMEM; |
253 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
254 | if (!new_page) | |
9f92448c | 255 | goto put_old; |
2b144498 SD |
256 | |
257 | __SetPageUptodate(new_page); | |
258 | ||
2b144498 SD |
259 | /* copy the page now that we've got it stable */ |
260 | vaddr_old = kmap_atomic(old_page); | |
261 | vaddr_new = kmap_atomic(new_page); | |
262 | ||
263 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 264 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
265 | |
266 | kunmap_atomic(vaddr_new); | |
267 | kunmap_atomic(vaddr_old); | |
268 | ||
269 | ret = anon_vma_prepare(vma); | |
270 | if (ret) | |
9f92448c | 271 | goto put_new; |
2b144498 | 272 | |
c517ee74 | 273 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 274 | |
9f92448c | 275 | put_new: |
2b144498 | 276 | page_cache_release(new_page); |
9f92448c | 277 | put_old: |
7b2d81d4 IM |
278 | put_page(old_page); |
279 | ||
5323ce71 ON |
280 | if (unlikely(ret == -EAGAIN)) |
281 | goto retry; | |
2b144498 SD |
282 | return ret; |
283 | } | |
284 | ||
5cb4ac3a | 285 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 286 | { |
cceb55aa | 287 | struct page *page; |
2b144498 | 288 | uprobe_opcode_t opcode; |
7b2d81d4 | 289 | int result; |
2b144498 | 290 | |
c00b2750 ON |
291 | if (current->mm == mm) { |
292 | pagefault_disable(); | |
293 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
294 | sizeof(opcode)); | |
295 | pagefault_enable(); | |
296 | ||
297 | if (likely(result == 0)) | |
298 | goto out; | |
299 | } | |
300 | ||
cceb55aa ON |
301 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); |
302 | if (result < 0) | |
2b144498 | 303 | return result; |
2b144498 | 304 | |
cceb55aa ON |
305 | copy_opcode(page, vaddr, &opcode); |
306 | put_page(page); | |
307 | out: | |
308 | return is_swbp_insn(&opcode); | |
2b144498 SD |
309 | } |
310 | ||
311 | /** | |
5cb4ac3a | 312 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 313 | * @auprobe: arch specific probepoint information. |
2b144498 | 314 | * @mm: the probed process address space. |
2b144498 SD |
315 | * @vaddr: the virtual address to insert the opcode. |
316 | * | |
317 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
318 | * Return 0 (success) or a negative errno. | |
319 | */ | |
5cb4ac3a | 320 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 321 | { |
cceb55aa | 322 | return write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
323 | } |
324 | ||
325 | /** | |
326 | * set_orig_insn - Restore the original instruction. | |
327 | * @mm: the probed process address space. | |
e3343e6a | 328 | * @auprobe: arch specific probepoint information. |
2b144498 | 329 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
330 | * |
331 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
332 | * Return 0 (success) or a negative errno. | |
333 | */ | |
7b2d81d4 | 334 | int __weak |
ded86e7c | 335 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 336 | { |
cceb55aa | 337 | return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
338 | } |
339 | ||
340 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
341 | { | |
342 | if (l->inode < r->inode) | |
343 | return -1; | |
7b2d81d4 | 344 | |
2b144498 SD |
345 | if (l->inode > r->inode) |
346 | return 1; | |
2b144498 | 347 | |
7b2d81d4 IM |
348 | if (l->offset < r->offset) |
349 | return -1; | |
350 | ||
351 | if (l->offset > r->offset) | |
352 | return 1; | |
2b144498 SD |
353 | |
354 | return 0; | |
355 | } | |
356 | ||
357 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
358 | { | |
359 | struct uprobe u = { .inode = inode, .offset = offset }; | |
360 | struct rb_node *n = uprobes_tree.rb_node; | |
361 | struct uprobe *uprobe; | |
362 | int match; | |
363 | ||
364 | while (n) { | |
365 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
366 | match = match_uprobe(&u, uprobe); | |
367 | if (!match) { | |
368 | atomic_inc(&uprobe->ref); | |
369 | return uprobe; | |
370 | } | |
7b2d81d4 | 371 | |
2b144498 SD |
372 | if (match < 0) |
373 | n = n->rb_left; | |
374 | else | |
375 | n = n->rb_right; | |
376 | } | |
377 | return NULL; | |
378 | } | |
379 | ||
380 | /* | |
381 | * Find a uprobe corresponding to a given inode:offset | |
382 | * Acquires uprobes_treelock | |
383 | */ | |
384 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
385 | { | |
386 | struct uprobe *uprobe; | |
2b144498 | 387 | |
6f47caa0 | 388 | spin_lock(&uprobes_treelock); |
2b144498 | 389 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 390 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 391 | |
2b144498 SD |
392 | return uprobe; |
393 | } | |
394 | ||
395 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
396 | { | |
397 | struct rb_node **p = &uprobes_tree.rb_node; | |
398 | struct rb_node *parent = NULL; | |
399 | struct uprobe *u; | |
400 | int match; | |
401 | ||
402 | while (*p) { | |
403 | parent = *p; | |
404 | u = rb_entry(parent, struct uprobe, rb_node); | |
405 | match = match_uprobe(uprobe, u); | |
406 | if (!match) { | |
407 | atomic_inc(&u->ref); | |
408 | return u; | |
409 | } | |
410 | ||
411 | if (match < 0) | |
412 | p = &parent->rb_left; | |
413 | else | |
414 | p = &parent->rb_right; | |
415 | ||
416 | } | |
7b2d81d4 | 417 | |
2b144498 SD |
418 | u = NULL; |
419 | rb_link_node(&uprobe->rb_node, parent, p); | |
420 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
421 | /* get access + creation ref */ | |
422 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 423 | |
2b144498 SD |
424 | return u; |
425 | } | |
426 | ||
427 | /* | |
7b2d81d4 | 428 | * Acquire uprobes_treelock. |
2b144498 SD |
429 | * Matching uprobe already exists in rbtree; |
430 | * increment (access refcount) and return the matching uprobe. | |
431 | * | |
432 | * No matching uprobe; insert the uprobe in rb_tree; | |
433 | * get a double refcount (access + creation) and return NULL. | |
434 | */ | |
435 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
436 | { | |
2b144498 SD |
437 | struct uprobe *u; |
438 | ||
6f47caa0 | 439 | spin_lock(&uprobes_treelock); |
2b144498 | 440 | u = __insert_uprobe(uprobe); |
6f47caa0 | 441 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 442 | |
0326f5a9 SD |
443 | /* For now assume that the instruction need not be single-stepped */ |
444 | uprobe->flags |= UPROBE_SKIP_SSTEP; | |
445 | ||
2b144498 SD |
446 | return u; |
447 | } | |
448 | ||
449 | static void put_uprobe(struct uprobe *uprobe) | |
450 | { | |
451 | if (atomic_dec_and_test(&uprobe->ref)) | |
452 | kfree(uprobe); | |
453 | } | |
454 | ||
455 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
456 | { | |
457 | struct uprobe *uprobe, *cur_uprobe; | |
458 | ||
459 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
460 | if (!uprobe) | |
461 | return NULL; | |
462 | ||
463 | uprobe->inode = igrab(inode); | |
464 | uprobe->offset = offset; | |
465 | init_rwsem(&uprobe->consumer_rwsem); | |
2b144498 SD |
466 | |
467 | /* add to uprobes_tree, sorted on inode:offset */ | |
468 | cur_uprobe = insert_uprobe(uprobe); | |
469 | ||
470 | /* a uprobe exists for this inode:offset combination */ | |
471 | if (cur_uprobe) { | |
472 | kfree(uprobe); | |
473 | uprobe = cur_uprobe; | |
474 | iput(inode); | |
7b2d81d4 | 475 | } else { |
2b144498 | 476 | atomic_inc(&uprobe_events); |
7b2d81d4 IM |
477 | } |
478 | ||
2b144498 SD |
479 | return uprobe; |
480 | } | |
481 | ||
0326f5a9 SD |
482 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
483 | { | |
484 | struct uprobe_consumer *uc; | |
485 | ||
486 | if (!(uprobe->flags & UPROBE_RUN_HANDLER)) | |
487 | return; | |
488 | ||
489 | down_read(&uprobe->consumer_rwsem); | |
490 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
491 | if (!uc->filter || uc->filter(uc, current)) | |
492 | uc->handler(uc, regs); | |
493 | } | |
494 | up_read(&uprobe->consumer_rwsem); | |
495 | } | |
496 | ||
2b144498 | 497 | /* Returns the previous consumer */ |
7b2d81d4 | 498 | static struct uprobe_consumer * |
e3343e6a | 499 | consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
500 | { |
501 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
502 | uc->next = uprobe->consumers; |
503 | uprobe->consumers = uc; | |
2b144498 | 504 | up_write(&uprobe->consumer_rwsem); |
7b2d81d4 | 505 | |
e3343e6a | 506 | return uc->next; |
2b144498 SD |
507 | } |
508 | ||
509 | /* | |
e3343e6a SD |
510 | * For uprobe @uprobe, delete the consumer @uc. |
511 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
512 | * or return false. |
513 | */ | |
e3343e6a | 514 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
515 | { |
516 | struct uprobe_consumer **con; | |
517 | bool ret = false; | |
518 | ||
519 | down_write(&uprobe->consumer_rwsem); | |
520 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
521 | if (*con == uc) { |
522 | *con = uc->next; | |
2b144498 SD |
523 | ret = true; |
524 | break; | |
525 | } | |
526 | } | |
527 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 528 | |
2b144498 SD |
529 | return ret; |
530 | } | |
531 | ||
e3343e6a | 532 | static int |
d436615e | 533 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 534 | unsigned long nbytes, loff_t offset) |
2b144498 | 535 | { |
2b144498 SD |
536 | struct page *page; |
537 | void *vaddr; | |
593609a5 ON |
538 | unsigned long off; |
539 | pgoff_t idx; | |
2b144498 SD |
540 | |
541 | if (!filp) | |
542 | return -EINVAL; | |
543 | ||
cc359d18 ON |
544 | if (!mapping->a_ops->readpage) |
545 | return -EIO; | |
546 | ||
593609a5 ON |
547 | idx = offset >> PAGE_CACHE_SHIFT; |
548 | off = offset & ~PAGE_MASK; | |
2b144498 SD |
549 | |
550 | /* | |
551 | * Ensure that the page that has the original instruction is | |
552 | * populated and in page-cache. | |
553 | */ | |
554 | page = read_mapping_page(mapping, idx, filp); | |
555 | if (IS_ERR(page)) | |
556 | return PTR_ERR(page); | |
557 | ||
558 | vaddr = kmap_atomic(page); | |
593609a5 | 559 | memcpy(insn, vaddr + off, nbytes); |
2b144498 SD |
560 | kunmap_atomic(vaddr); |
561 | page_cache_release(page); | |
7b2d81d4 | 562 | |
2b144498 SD |
563 | return 0; |
564 | } | |
565 | ||
d436615e | 566 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
567 | { |
568 | struct address_space *mapping; | |
2b144498 | 569 | unsigned long nbytes; |
7b2d81d4 | 570 | int bytes; |
2b144498 | 571 | |
d436615e | 572 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
573 | mapping = uprobe->inode->i_mapping; |
574 | ||
575 | /* Instruction at end of binary; copy only available bytes */ | |
576 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
577 | bytes = uprobe->inode->i_size - uprobe->offset; | |
578 | else | |
579 | bytes = MAX_UINSN_BYTES; | |
580 | ||
581 | /* Instruction at the page-boundary; copy bytes in second page */ | |
582 | if (nbytes < bytes) { | |
fc36f595 ON |
583 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
584 | bytes - nbytes, uprobe->offset + nbytes); | |
585 | if (err) | |
586 | return err; | |
2b144498 SD |
587 | bytes = nbytes; |
588 | } | |
d436615e | 589 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
590 | } |
591 | ||
682968e0 SD |
592 | /* |
593 | * How mm->uprobes_state.count gets updated | |
594 | * uprobe_mmap() increments the count if | |
595 | * - it successfully adds a breakpoint. | |
596 | * - it cannot add a breakpoint, but sees that there is a underlying | |
597 | * breakpoint (via a is_swbp_at_addr()). | |
598 | * | |
599 | * uprobe_munmap() decrements the count if | |
600 | * - it sees a underlying breakpoint, (via is_swbp_at_addr) | |
601 | * (Subsequent uprobe_unregister wouldnt find the breakpoint | |
602 | * unless a uprobe_mmap kicks in, since the old vma would be | |
603 | * dropped just after uprobe_munmap.) | |
604 | * | |
605 | * uprobe_register increments the count if: | |
606 | * - it successfully adds a breakpoint. | |
607 | * | |
608 | * uprobe_unregister decrements the count if: | |
609 | * - it sees a underlying breakpoint and removes successfully. | |
610 | * (via is_swbp_at_addr) | |
611 | * (Subsequent uprobe_munmap wouldnt find the breakpoint | |
612 | * since there is no underlying breakpoint after the | |
613 | * breakpoint removal.) | |
614 | */ | |
e3343e6a SD |
615 | static int |
616 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 617 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 618 | { |
f8ac4ec9 | 619 | bool first_uprobe; |
2b144498 SD |
620 | int ret; |
621 | ||
622 | /* | |
623 | * If probe is being deleted, unregister thread could be done with | |
624 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
625 | * nobody will be able to cleanup. Also we could be from fork or | |
626 | * mremap path, where the probe might have already been inserted. | |
627 | * Hence behave as if probe already existed. | |
628 | */ | |
629 | if (!uprobe->consumers) | |
78f74116 | 630 | return 0; |
2b144498 | 631 | |
900771a4 | 632 | if (!(uprobe->flags & UPROBE_COPY_INSN)) { |
d436615e | 633 | ret = copy_insn(uprobe, vma->vm_file); |
2b144498 SD |
634 | if (ret) |
635 | return ret; | |
636 | ||
5cb4ac3a | 637 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) |
c1914a09 | 638 | return -ENOTSUPP; |
2b144498 | 639 | |
816c03fb | 640 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
641 | if (ret) |
642 | return ret; | |
643 | ||
d9c4a30e ON |
644 | /* write_opcode() assumes we don't cross page boundary */ |
645 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
646 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
647 | ||
900771a4 | 648 | uprobe->flags |= UPROBE_COPY_INSN; |
2b144498 | 649 | } |
682968e0 | 650 | |
f8ac4ec9 ON |
651 | /* |
652 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
653 | * the task can hit this breakpoint right after __replace_page(). | |
654 | */ | |
655 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
656 | if (first_uprobe) | |
657 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
658 | ||
816c03fb | 659 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
660 | if (!ret) |
661 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
662 | else if (first_uprobe) | |
f8ac4ec9 | 663 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
664 | |
665 | return ret; | |
666 | } | |
667 | ||
e3343e6a | 668 | static void |
816c03fb | 669 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 670 | { |
9f68f672 ON |
671 | /* can happen if uprobe_register() fails */ |
672 | if (!test_bit(MMF_HAS_UPROBES, &mm->flags)) | |
673 | return; | |
674 | ||
675 | set_bit(MMF_RECALC_UPROBES, &mm->flags); | |
ded86e7c | 676 | set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
677 | } |
678 | ||
0326f5a9 | 679 | /* |
778b032d ON |
680 | * There could be threads that have already hit the breakpoint. They |
681 | * will recheck the current insn and restart if find_uprobe() fails. | |
682 | * See find_active_uprobe(). | |
0326f5a9 | 683 | */ |
2b144498 SD |
684 | static void delete_uprobe(struct uprobe *uprobe) |
685 | { | |
6f47caa0 | 686 | spin_lock(&uprobes_treelock); |
2b144498 | 687 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 688 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
689 | iput(uprobe->inode); |
690 | put_uprobe(uprobe); | |
691 | atomic_dec(&uprobe_events); | |
692 | } | |
693 | ||
26872090 ON |
694 | struct map_info { |
695 | struct map_info *next; | |
696 | struct mm_struct *mm; | |
816c03fb | 697 | unsigned long vaddr; |
26872090 ON |
698 | }; |
699 | ||
700 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 701 | { |
26872090 ON |
702 | struct map_info *next = info->next; |
703 | kfree(info); | |
704 | return next; | |
705 | } | |
706 | ||
707 | static struct map_info * | |
708 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
709 | { | |
710 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 SD |
711 | struct prio_tree_iter iter; |
712 | struct vm_area_struct *vma; | |
26872090 ON |
713 | struct map_info *curr = NULL; |
714 | struct map_info *prev = NULL; | |
715 | struct map_info *info; | |
716 | int more = 0; | |
2b144498 | 717 | |
26872090 ON |
718 | again: |
719 | mutex_lock(&mapping->i_mmap_mutex); | |
2b144498 SD |
720 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
721 | if (!valid_vma(vma, is_register)) | |
722 | continue; | |
723 | ||
7a5bfb66 ON |
724 | if (!prev && !more) { |
725 | /* | |
726 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
727 | * reclaim. This is optimistic, no harm done if it fails. | |
728 | */ | |
729 | prev = kmalloc(sizeof(struct map_info), | |
730 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
731 | if (prev) | |
732 | prev->next = NULL; | |
733 | } | |
26872090 ON |
734 | if (!prev) { |
735 | more++; | |
736 | continue; | |
2b144498 | 737 | } |
2b144498 | 738 | |
26872090 ON |
739 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
740 | continue; | |
7b2d81d4 | 741 | |
26872090 ON |
742 | info = prev; |
743 | prev = prev->next; | |
744 | info->next = curr; | |
745 | curr = info; | |
2b144498 | 746 | |
26872090 | 747 | info->mm = vma->vm_mm; |
57683f72 | 748 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 749 | } |
2b144498 SD |
750 | mutex_unlock(&mapping->i_mmap_mutex); |
751 | ||
26872090 ON |
752 | if (!more) |
753 | goto out; | |
754 | ||
755 | prev = curr; | |
756 | while (curr) { | |
757 | mmput(curr->mm); | |
758 | curr = curr->next; | |
759 | } | |
7b2d81d4 | 760 | |
26872090 ON |
761 | do { |
762 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
763 | if (!info) { | |
764 | curr = ERR_PTR(-ENOMEM); | |
765 | goto out; | |
766 | } | |
767 | info->next = prev; | |
768 | prev = info; | |
769 | } while (--more); | |
770 | ||
771 | goto again; | |
772 | out: | |
773 | while (prev) | |
774 | prev = free_map_info(prev); | |
775 | return curr; | |
2b144498 SD |
776 | } |
777 | ||
778 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
779 | { | |
26872090 ON |
780 | struct map_info *info; |
781 | int err = 0; | |
2b144498 | 782 | |
26872090 ON |
783 | info = build_map_info(uprobe->inode->i_mapping, |
784 | uprobe->offset, is_register); | |
785 | if (IS_ERR(info)) | |
786 | return PTR_ERR(info); | |
7b2d81d4 | 787 | |
26872090 ON |
788 | while (info) { |
789 | struct mm_struct *mm = info->mm; | |
790 | struct vm_area_struct *vma; | |
7b2d81d4 | 791 | |
26872090 ON |
792 | if (err) |
793 | goto free; | |
7b2d81d4 | 794 | |
77fc4af1 | 795 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
796 | vma = find_vma(mm, info->vaddr); |
797 | if (!vma || !valid_vma(vma, is_register) || | |
798 | vma->vm_file->f_mapping->host != uprobe->inode) | |
26872090 ON |
799 | goto unlock; |
800 | ||
f4d6dfe5 ON |
801 | if (vma->vm_start > info->vaddr || |
802 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 803 | goto unlock; |
2b144498 | 804 | |
78f74116 | 805 | if (is_register) |
26872090 | 806 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
78f74116 | 807 | else |
26872090 | 808 | remove_breakpoint(uprobe, mm, info->vaddr); |
78f74116 | 809 | |
26872090 ON |
810 | unlock: |
811 | up_write(&mm->mmap_sem); | |
812 | free: | |
813 | mmput(mm); | |
814 | info = free_map_info(info); | |
2b144498 | 815 | } |
7b2d81d4 | 816 | |
26872090 | 817 | return err; |
2b144498 SD |
818 | } |
819 | ||
7b2d81d4 | 820 | static int __uprobe_register(struct uprobe *uprobe) |
2b144498 SD |
821 | { |
822 | return register_for_each_vma(uprobe, true); | |
823 | } | |
824 | ||
7b2d81d4 | 825 | static void __uprobe_unregister(struct uprobe *uprobe) |
2b144498 SD |
826 | { |
827 | if (!register_for_each_vma(uprobe, false)) | |
828 | delete_uprobe(uprobe); | |
829 | ||
830 | /* TODO : cant unregister? schedule a worker thread */ | |
831 | } | |
832 | ||
833 | /* | |
7b2d81d4 | 834 | * uprobe_register - register a probe |
2b144498 SD |
835 | * @inode: the file in which the probe has to be placed. |
836 | * @offset: offset from the start of the file. | |
e3343e6a | 837 | * @uc: information on howto handle the probe.. |
2b144498 | 838 | * |
7b2d81d4 | 839 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
840 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
841 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 842 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 843 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 844 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
845 | * unregisters. |
846 | * | |
847 | * Return errno if it cannot successully install probes | |
848 | * else return 0 (success) | |
849 | */ | |
e3343e6a | 850 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
851 | { |
852 | struct uprobe *uprobe; | |
7b2d81d4 | 853 | int ret; |
2b144498 | 854 | |
e3343e6a | 855 | if (!inode || !uc || uc->next) |
7b2d81d4 | 856 | return -EINVAL; |
2b144498 SD |
857 | |
858 | if (offset > i_size_read(inode)) | |
7b2d81d4 | 859 | return -EINVAL; |
2b144498 SD |
860 | |
861 | ret = 0; | |
862 | mutex_lock(uprobes_hash(inode)); | |
863 | uprobe = alloc_uprobe(inode, offset); | |
7b2d81d4 | 864 | |
e3343e6a | 865 | if (uprobe && !consumer_add(uprobe, uc)) { |
7b2d81d4 | 866 | ret = __uprobe_register(uprobe); |
2b144498 SD |
867 | if (ret) { |
868 | uprobe->consumers = NULL; | |
7b2d81d4 IM |
869 | __uprobe_unregister(uprobe); |
870 | } else { | |
900771a4 | 871 | uprobe->flags |= UPROBE_RUN_HANDLER; |
7b2d81d4 | 872 | } |
2b144498 SD |
873 | } |
874 | ||
875 | mutex_unlock(uprobes_hash(inode)); | |
6d1d8dfa SAS |
876 | if (uprobe) |
877 | put_uprobe(uprobe); | |
2b144498 SD |
878 | |
879 | return ret; | |
880 | } | |
881 | ||
882 | /* | |
7b2d81d4 | 883 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
884 | * @inode: the file in which the probe has to be removed. |
885 | * @offset: offset from the start of the file. | |
e3343e6a | 886 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 887 | */ |
e3343e6a | 888 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 889 | { |
7b2d81d4 | 890 | struct uprobe *uprobe; |
2b144498 | 891 | |
e3343e6a | 892 | if (!inode || !uc) |
2b144498 SD |
893 | return; |
894 | ||
895 | uprobe = find_uprobe(inode, offset); | |
896 | if (!uprobe) | |
897 | return; | |
898 | ||
899 | mutex_lock(uprobes_hash(inode)); | |
2b144498 | 900 | |
e3343e6a | 901 | if (consumer_del(uprobe, uc)) { |
7b2d81d4 IM |
902 | if (!uprobe->consumers) { |
903 | __uprobe_unregister(uprobe); | |
900771a4 | 904 | uprobe->flags &= ~UPROBE_RUN_HANDLER; |
7b2d81d4 | 905 | } |
2b144498 SD |
906 | } |
907 | ||
2b144498 SD |
908 | mutex_unlock(uprobes_hash(inode)); |
909 | if (uprobe) | |
910 | put_uprobe(uprobe); | |
911 | } | |
912 | ||
891c3970 ON |
913 | static struct rb_node * |
914 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 915 | { |
2b144498 | 916 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
917 | |
918 | while (n) { | |
891c3970 | 919 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 920 | |
891c3970 | 921 | if (inode < u->inode) { |
2b144498 | 922 | n = n->rb_left; |
891c3970 | 923 | } else if (inode > u->inode) { |
2b144498 | 924 | n = n->rb_right; |
891c3970 ON |
925 | } else { |
926 | if (max < u->offset) | |
927 | n = n->rb_left; | |
928 | else if (min > u->offset) | |
929 | n = n->rb_right; | |
930 | else | |
931 | break; | |
932 | } | |
2b144498 | 933 | } |
7b2d81d4 | 934 | |
891c3970 | 935 | return n; |
2b144498 SD |
936 | } |
937 | ||
938 | /* | |
891c3970 | 939 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 940 | */ |
891c3970 ON |
941 | static void build_probe_list(struct inode *inode, |
942 | struct vm_area_struct *vma, | |
943 | unsigned long start, unsigned long end, | |
944 | struct list_head *head) | |
2b144498 | 945 | { |
891c3970 | 946 | loff_t min, max; |
891c3970 ON |
947 | struct rb_node *n, *t; |
948 | struct uprobe *u; | |
7b2d81d4 | 949 | |
891c3970 | 950 | INIT_LIST_HEAD(head); |
cb113b47 | 951 | min = vaddr_to_offset(vma, start); |
891c3970 | 952 | max = min + (end - start) - 1; |
2b144498 | 953 | |
6f47caa0 | 954 | spin_lock(&uprobes_treelock); |
891c3970 ON |
955 | n = find_node_in_range(inode, min, max); |
956 | if (n) { | |
957 | for (t = n; t; t = rb_prev(t)) { | |
958 | u = rb_entry(t, struct uprobe, rb_node); | |
959 | if (u->inode != inode || u->offset < min) | |
960 | break; | |
961 | list_add(&u->pending_list, head); | |
962 | atomic_inc(&u->ref); | |
963 | } | |
964 | for (t = n; (t = rb_next(t)); ) { | |
965 | u = rb_entry(t, struct uprobe, rb_node); | |
966 | if (u->inode != inode || u->offset > max) | |
967 | break; | |
968 | list_add(&u->pending_list, head); | |
969 | atomic_inc(&u->ref); | |
970 | } | |
2b144498 | 971 | } |
6f47caa0 | 972 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
973 | } |
974 | ||
975 | /* | |
5e5be71a | 976 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 977 | * |
5e5be71a ON |
978 | * Currently we ignore all errors and always return 0, the callers |
979 | * can't handle the failure anyway. | |
2b144498 | 980 | */ |
7b2d81d4 | 981 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
982 | { |
983 | struct list_head tmp_list; | |
665605a2 | 984 | struct uprobe *uprobe, *u; |
2b144498 | 985 | struct inode *inode; |
2b144498 SD |
986 | |
987 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) | |
7b2d81d4 | 988 | return 0; |
2b144498 SD |
989 | |
990 | inode = vma->vm_file->f_mapping->host; | |
991 | if (!inode) | |
7b2d81d4 | 992 | return 0; |
2b144498 | 993 | |
2b144498 | 994 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 995 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
7b2d81d4 | 996 | |
665605a2 | 997 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
5e5be71a | 998 | if (!fatal_signal_pending(current)) { |
57683f72 | 999 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1000 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1001 | } |
1002 | put_uprobe(uprobe); | |
1003 | } | |
2b144498 SD |
1004 | mutex_unlock(uprobes_mmap_hash(inode)); |
1005 | ||
5e5be71a | 1006 | return 0; |
2b144498 SD |
1007 | } |
1008 | ||
9f68f672 ON |
1009 | static bool |
1010 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1011 | { | |
1012 | loff_t min, max; | |
1013 | struct inode *inode; | |
1014 | struct rb_node *n; | |
1015 | ||
1016 | inode = vma->vm_file->f_mapping->host; | |
1017 | ||
1018 | min = vaddr_to_offset(vma, start); | |
1019 | max = min + (end - start) - 1; | |
1020 | ||
1021 | spin_lock(&uprobes_treelock); | |
1022 | n = find_node_in_range(inode, min, max); | |
1023 | spin_unlock(&uprobes_treelock); | |
1024 | ||
1025 | return !!n; | |
1026 | } | |
1027 | ||
682968e0 SD |
1028 | /* |
1029 | * Called in context of a munmap of a vma. | |
1030 | */ | |
cbc91f71 | 1031 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1032 | { |
682968e0 SD |
1033 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) |
1034 | return; | |
1035 | ||
2fd611a9 ON |
1036 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1037 | return; | |
1038 | ||
9f68f672 ON |
1039 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1040 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1041 | return; |
1042 | ||
9f68f672 ON |
1043 | if (vma_has_uprobes(vma, start, end)) |
1044 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1045 | } |
1046 | ||
d4b3b638 SD |
1047 | /* Slot allocation for XOL */ |
1048 | static int xol_add_vma(struct xol_area *area) | |
1049 | { | |
1050 | struct mm_struct *mm; | |
1051 | int ret; | |
1052 | ||
1053 | area->page = alloc_page(GFP_HIGHUSER); | |
1054 | if (!area->page) | |
1055 | return -ENOMEM; | |
1056 | ||
1057 | ret = -EALREADY; | |
1058 | mm = current->mm; | |
1059 | ||
1060 | down_write(&mm->mmap_sem); | |
1061 | if (mm->uprobes_state.xol_area) | |
1062 | goto fail; | |
1063 | ||
1064 | ret = -ENOMEM; | |
1065 | ||
1066 | /* Try to map as high as possible, this is only a hint. */ | |
1067 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1068 | if (area->vaddr & ~PAGE_MASK) { | |
1069 | ret = area->vaddr; | |
1070 | goto fail; | |
1071 | } | |
1072 | ||
1073 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1074 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1075 | if (ret) | |
1076 | goto fail; | |
1077 | ||
1078 | smp_wmb(); /* pairs with get_xol_area() */ | |
1079 | mm->uprobes_state.xol_area = area; | |
1080 | ret = 0; | |
1081 | ||
1082 | fail: | |
1083 | up_write(&mm->mmap_sem); | |
1084 | if (ret) | |
1085 | __free_page(area->page); | |
1086 | ||
1087 | return ret; | |
1088 | } | |
1089 | ||
1090 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1091 | { | |
1092 | struct xol_area *area; | |
1093 | ||
1094 | area = mm->uprobes_state.xol_area; | |
1095 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1096 | ||
1097 | return area; | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * xol_alloc_area - Allocate process's xol_area. | |
1102 | * This area will be used for storing instructions for execution out of | |
1103 | * line. | |
1104 | * | |
1105 | * Returns the allocated area or NULL. | |
1106 | */ | |
1107 | static struct xol_area *xol_alloc_area(void) | |
1108 | { | |
1109 | struct xol_area *area; | |
1110 | ||
1111 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1112 | if (unlikely(!area)) | |
1113 | return NULL; | |
1114 | ||
1115 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1116 | ||
1117 | if (!area->bitmap) | |
1118 | goto fail; | |
1119 | ||
1120 | init_waitqueue_head(&area->wq); | |
1121 | if (!xol_add_vma(area)) | |
1122 | return area; | |
1123 | ||
1124 | fail: | |
1125 | kfree(area->bitmap); | |
1126 | kfree(area); | |
1127 | ||
1128 | return get_xol_area(current->mm); | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * uprobe_clear_state - Free the area allocated for slots. | |
1133 | */ | |
1134 | void uprobe_clear_state(struct mm_struct *mm) | |
1135 | { | |
1136 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1137 | ||
1138 | if (!area) | |
1139 | return; | |
1140 | ||
1141 | put_page(area->page); | |
1142 | kfree(area->bitmap); | |
1143 | kfree(area); | |
1144 | } | |
1145 | ||
f8ac4ec9 ON |
1146 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1147 | { | |
61559a81 ON |
1148 | newmm->uprobes_state.xol_area = NULL; |
1149 | ||
9f68f672 | 1150 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1151 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1152 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1153 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1154 | } | |
f8ac4ec9 ON |
1155 | } |
1156 | ||
d4b3b638 SD |
1157 | /* |
1158 | * - search for a free slot. | |
1159 | */ | |
1160 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1161 | { | |
1162 | unsigned long slot_addr; | |
1163 | int slot_nr; | |
1164 | ||
1165 | do { | |
1166 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1167 | if (slot_nr < UINSNS_PER_PAGE) { | |
1168 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1169 | break; | |
1170 | ||
1171 | slot_nr = UINSNS_PER_PAGE; | |
1172 | continue; | |
1173 | } | |
1174 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1175 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1176 | ||
1177 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1178 | atomic_inc(&area->slot_count); | |
1179 | ||
1180 | return slot_addr; | |
1181 | } | |
1182 | ||
1183 | /* | |
1184 | * xol_get_insn_slot - If was not allocated a slot, then | |
1185 | * allocate a slot. | |
1186 | * Returns the allocated slot address or 0. | |
1187 | */ | |
1188 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1189 | { | |
1190 | struct xol_area *area; | |
1191 | unsigned long offset; | |
1192 | void *vaddr; | |
1193 | ||
1194 | area = get_xol_area(current->mm); | |
1195 | if (!area) { | |
1196 | area = xol_alloc_area(); | |
1197 | if (!area) | |
1198 | return 0; | |
1199 | } | |
1200 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1201 | ||
1202 | /* | |
1203 | * Initialize the slot if xol_vaddr points to valid | |
1204 | * instruction slot. | |
1205 | */ | |
1206 | if (unlikely(!current->utask->xol_vaddr)) | |
1207 | return 0; | |
1208 | ||
1209 | current->utask->vaddr = slot_addr; | |
1210 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1211 | vaddr = kmap_atomic(area->page); | |
1212 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1213 | kunmap_atomic(vaddr); | |
1214 | ||
1215 | return current->utask->xol_vaddr; | |
1216 | } | |
1217 | ||
1218 | /* | |
1219 | * xol_free_insn_slot - If slot was earlier allocated by | |
1220 | * @xol_get_insn_slot(), make the slot available for | |
1221 | * subsequent requests. | |
1222 | */ | |
1223 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1224 | { | |
1225 | struct xol_area *area; | |
1226 | unsigned long vma_end; | |
1227 | unsigned long slot_addr; | |
1228 | ||
1229 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1230 | return; | |
1231 | ||
1232 | slot_addr = tsk->utask->xol_vaddr; | |
1233 | ||
1234 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1235 | return; | |
1236 | ||
1237 | area = tsk->mm->uprobes_state.xol_area; | |
1238 | vma_end = area->vaddr + PAGE_SIZE; | |
1239 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1240 | unsigned long offset; | |
1241 | int slot_nr; | |
1242 | ||
1243 | offset = slot_addr - area->vaddr; | |
1244 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1245 | if (slot_nr >= UINSNS_PER_PAGE) | |
1246 | return; | |
1247 | ||
1248 | clear_bit(slot_nr, area->bitmap); | |
1249 | atomic_dec(&area->slot_count); | |
1250 | if (waitqueue_active(&area->wq)) | |
1251 | wake_up(&area->wq); | |
1252 | ||
1253 | tsk->utask->xol_vaddr = 0; | |
1254 | } | |
1255 | } | |
1256 | ||
0326f5a9 SD |
1257 | /** |
1258 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1259 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1260 | * instruction. | |
1261 | * Return the address of the breakpoint instruction. | |
1262 | */ | |
1263 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1264 | { | |
1265 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * Called with no locks held. | |
1270 | * Called in context of a exiting or a exec-ing thread. | |
1271 | */ | |
1272 | void uprobe_free_utask(struct task_struct *t) | |
1273 | { | |
1274 | struct uprobe_task *utask = t->utask; | |
1275 | ||
0326f5a9 SD |
1276 | if (!utask) |
1277 | return; | |
1278 | ||
1279 | if (utask->active_uprobe) | |
1280 | put_uprobe(utask->active_uprobe); | |
1281 | ||
d4b3b638 | 1282 | xol_free_insn_slot(t); |
0326f5a9 SD |
1283 | kfree(utask); |
1284 | t->utask = NULL; | |
1285 | } | |
1286 | ||
1287 | /* | |
1288 | * Called in context of a new clone/fork from copy_process. | |
1289 | */ | |
1290 | void uprobe_copy_process(struct task_struct *t) | |
1291 | { | |
1292 | t->utask = NULL; | |
0326f5a9 SD |
1293 | } |
1294 | ||
1295 | /* | |
1296 | * Allocate a uprobe_task object for the task. | |
1297 | * Called when the thread hits a breakpoint for the first time. | |
1298 | * | |
1299 | * Returns: | |
1300 | * - pointer to new uprobe_task on success | |
1301 | * - NULL otherwise | |
1302 | */ | |
1303 | static struct uprobe_task *add_utask(void) | |
1304 | { | |
1305 | struct uprobe_task *utask; | |
1306 | ||
1307 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1308 | if (unlikely(!utask)) | |
1309 | return NULL; | |
1310 | ||
0326f5a9 SD |
1311 | current->utask = utask; |
1312 | return utask; | |
1313 | } | |
1314 | ||
1315 | /* Prepare to single-step probed instruction out of line. */ | |
1316 | static int | |
1317 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1318 | { | |
d4b3b638 SD |
1319 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1320 | return 0; | |
1321 | ||
0326f5a9 SD |
1322 | return -EFAULT; |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * If we are singlestepping, then ensure this thread is not connected to | |
1327 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1328 | * triggers the signal, restart the original insn even if the task is | |
1329 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1330 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1331 | * _same_ instruction should be repeated again after return from the signal | |
1332 | * handler, and SSTEP can never finish in this case. | |
1333 | */ | |
1334 | bool uprobe_deny_signal(void) | |
1335 | { | |
1336 | struct task_struct *t = current; | |
1337 | struct uprobe_task *utask = t->utask; | |
1338 | ||
1339 | if (likely(!utask || !utask->active_uprobe)) | |
1340 | return false; | |
1341 | ||
1342 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1343 | ||
1344 | if (signal_pending(t)) { | |
1345 | spin_lock_irq(&t->sighand->siglock); | |
1346 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1347 | spin_unlock_irq(&t->sighand->siglock); | |
1348 | ||
1349 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1350 | utask->state = UTASK_SSTEP_TRAPPED; | |
1351 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1352 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | return true; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * Avoid singlestepping the original instruction if the original instruction | |
1361 | * is a NOP or can be emulated. | |
1362 | */ | |
1363 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1364 | { | |
0578a970 ON |
1365 | if (uprobe->flags & UPROBE_SKIP_SSTEP) { |
1366 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) | |
1367 | return true; | |
1368 | uprobe->flags &= ~UPROBE_SKIP_SSTEP; | |
1369 | } | |
0326f5a9 SD |
1370 | return false; |
1371 | } | |
1372 | ||
499a4f3e ON |
1373 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1374 | { | |
1375 | struct vm_area_struct *vma; | |
1376 | ||
1377 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1378 | if (!valid_vma(vma, false)) | |
1379 | continue; | |
1380 | /* | |
1381 | * This is not strictly accurate, we can race with | |
1382 | * uprobe_unregister() and see the already removed | |
1383 | * uprobe if delete_uprobe() was not yet called. | |
1384 | */ | |
1385 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1386 | return; | |
1387 | } | |
1388 | ||
1389 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1390 | } | |
1391 | ||
d790d346 | 1392 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1393 | { |
3a9ea052 ON |
1394 | struct mm_struct *mm = current->mm; |
1395 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1396 | struct vm_area_struct *vma; |
0326f5a9 | 1397 | |
0326f5a9 SD |
1398 | down_read(&mm->mmap_sem); |
1399 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1400 | if (vma && vma->vm_start <= bp_vaddr) { |
1401 | if (valid_vma(vma, false)) { | |
cb113b47 ON |
1402 | struct inode *inode = vma->vm_file->f_mapping->host; |
1403 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); | |
0326f5a9 | 1404 | |
3a9ea052 ON |
1405 | uprobe = find_uprobe(inode, offset); |
1406 | } | |
d790d346 ON |
1407 | |
1408 | if (!uprobe) | |
1409 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1410 | } else { | |
1411 | *is_swbp = -EFAULT; | |
0326f5a9 | 1412 | } |
499a4f3e ON |
1413 | |
1414 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1415 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1416 | up_read(&mm->mmap_sem); |
1417 | ||
3a9ea052 ON |
1418 | return uprobe; |
1419 | } | |
1420 | ||
9d778782 SAS |
1421 | void __weak arch_uprobe_enable_step(struct arch_uprobe *arch) |
1422 | { | |
1423 | user_enable_single_step(current); | |
1424 | } | |
1425 | ||
1426 | void __weak arch_uprobe_disable_step(struct arch_uprobe *arch) | |
1427 | { | |
1428 | user_disable_single_step(current); | |
1429 | } | |
1430 | ||
3a9ea052 ON |
1431 | /* |
1432 | * Run handler and ask thread to singlestep. | |
1433 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1434 | */ | |
1435 | static void handle_swbp(struct pt_regs *regs) | |
1436 | { | |
1437 | struct uprobe_task *utask; | |
1438 | struct uprobe *uprobe; | |
1439 | unsigned long bp_vaddr; | |
56bb4cf6 | 1440 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1441 | |
1442 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1443 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1444 | |
0326f5a9 | 1445 | if (!uprobe) { |
56bb4cf6 ON |
1446 | if (is_swbp > 0) { |
1447 | /* No matching uprobe; signal SIGTRAP. */ | |
1448 | send_sig(SIGTRAP, current, 0); | |
1449 | } else { | |
1450 | /* | |
1451 | * Either we raced with uprobe_unregister() or we can't | |
1452 | * access this memory. The latter is only possible if | |
1453 | * another thread plays with our ->mm. In both cases | |
1454 | * we can simply restart. If this vma was unmapped we | |
1455 | * can pretend this insn was not executed yet and get | |
1456 | * the (correct) SIGSEGV after restart. | |
1457 | */ | |
1458 | instruction_pointer_set(regs, bp_vaddr); | |
1459 | } | |
0326f5a9 SD |
1460 | return; |
1461 | } | |
1462 | ||
1b08e907 | 1463 | utask = current->utask; |
0326f5a9 SD |
1464 | if (!utask) { |
1465 | utask = add_utask(); | |
1466 | /* Cannot allocate; re-execute the instruction. */ | |
1467 | if (!utask) | |
0578a970 | 1468 | goto restart; |
0326f5a9 | 1469 | } |
746a9e6b | 1470 | |
0326f5a9 | 1471 | handler_chain(uprobe, regs); |
0578a970 ON |
1472 | if (can_skip_sstep(uprobe, regs)) |
1473 | goto out; | |
0326f5a9 | 1474 | |
0326f5a9 | 1475 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { |
9d778782 | 1476 | arch_uprobe_enable_step(&uprobe->arch); |
746a9e6b ON |
1477 | utask->active_uprobe = uprobe; |
1478 | utask->state = UTASK_SSTEP; | |
0326f5a9 SD |
1479 | return; |
1480 | } | |
1481 | ||
0578a970 ON |
1482 | restart: |
1483 | /* | |
1484 | * cannot singlestep; cannot skip instruction; | |
1485 | * re-execute the instruction. | |
1486 | */ | |
1487 | instruction_pointer_set(regs, bp_vaddr); | |
1488 | out: | |
8bd87445 | 1489 | put_uprobe(uprobe); |
0326f5a9 SD |
1490 | } |
1491 | ||
1492 | /* | |
1493 | * Perform required fix-ups and disable singlestep. | |
1494 | * Allow pending signals to take effect. | |
1495 | */ | |
1496 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1497 | { | |
1498 | struct uprobe *uprobe; | |
1499 | ||
1500 | uprobe = utask->active_uprobe; | |
1501 | if (utask->state == UTASK_SSTEP_ACK) | |
1502 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1503 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1504 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1505 | else | |
1506 | WARN_ON_ONCE(1); | |
1507 | ||
9d778782 | 1508 | arch_uprobe_disable_step(&uprobe->arch); |
0326f5a9 SD |
1509 | put_uprobe(uprobe); |
1510 | utask->active_uprobe = NULL; | |
1511 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1512 | xol_free_insn_slot(current); |
0326f5a9 SD |
1513 | |
1514 | spin_lock_irq(¤t->sighand->siglock); | |
1515 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1516 | spin_unlock_irq(¤t->sighand->siglock); | |
1517 | } | |
1518 | ||
1519 | /* | |
1b08e907 ON |
1520 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1521 | * allows the thread to return from interrupt. After that handle_swbp() | |
1522 | * sets utask->active_uprobe. | |
0326f5a9 | 1523 | * |
1b08e907 ON |
1524 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1525 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1526 | * |
1527 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1528 | * uprobe_notify_resume(). | |
1529 | */ | |
1530 | void uprobe_notify_resume(struct pt_regs *regs) | |
1531 | { | |
1532 | struct uprobe_task *utask; | |
1533 | ||
db023ea5 ON |
1534 | clear_thread_flag(TIF_UPROBE); |
1535 | ||
0326f5a9 | 1536 | utask = current->utask; |
1b08e907 | 1537 | if (utask && utask->active_uprobe) |
0326f5a9 | 1538 | handle_singlestep(utask, regs); |
1b08e907 ON |
1539 | else |
1540 | handle_swbp(regs); | |
0326f5a9 SD |
1541 | } |
1542 | ||
1543 | /* | |
1544 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1545 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1546 | */ | |
1547 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1548 | { | |
f8ac4ec9 | 1549 | if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) |
0326f5a9 SD |
1550 | return 0; |
1551 | ||
0326f5a9 | 1552 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1553 | return 1; |
1554 | } | |
1555 | ||
1556 | /* | |
1557 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1558 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1559 | */ | |
1560 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1561 | { | |
1562 | struct uprobe_task *utask = current->utask; | |
1563 | ||
1564 | if (!current->mm || !utask || !utask->active_uprobe) | |
1565 | /* task is currently not uprobed */ | |
1566 | return 0; | |
1567 | ||
1568 | utask->state = UTASK_SSTEP_ACK; | |
1569 | set_thread_flag(TIF_UPROBE); | |
1570 | return 1; | |
1571 | } | |
1572 | ||
1573 | static struct notifier_block uprobe_exception_nb = { | |
1574 | .notifier_call = arch_uprobe_exception_notify, | |
1575 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1576 | }; | |
1577 | ||
2b144498 SD |
1578 | static int __init init_uprobes(void) |
1579 | { | |
1580 | int i; | |
1581 | ||
1582 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1583 | mutex_init(&uprobes_mutex[i]); | |
1584 | mutex_init(&uprobes_mmap_mutex[i]); | |
1585 | } | |
0326f5a9 SD |
1586 | |
1587 | return register_die_notifier(&uprobe_exception_nb); | |
2b144498 | 1588 | } |
0326f5a9 | 1589 | module_init(init_uprobes); |
2b144498 SD |
1590 | |
1591 | static void __exit exit_uprobes(void) | |
1592 | { | |
1593 | } | |
2b144498 | 1594 | module_exit(exit_uprobes); |