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 */ |
32cdba1e | 36 | #include <linux/percpu-rwsem.h> |
7b2d81d4 | 37 | |
2b144498 SD |
38 | #include <linux/uprobes.h> |
39 | ||
d4b3b638 SD |
40 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
41 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
42 | ||
2b144498 | 43 | static struct rb_root uprobes_tree = RB_ROOT; |
441f1eb7 ON |
44 | /* |
45 | * allows us to skip the uprobe_mmap if there are no uprobe events active | |
46 | * at this time. Probably a fine grained per inode count is better? | |
47 | */ | |
48 | #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree) | |
7b2d81d4 | 49 | |
2b144498 SD |
50 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
51 | ||
52 | #define UPROBES_HASH_SZ 13 | |
2b144498 SD |
53 | /* serialize uprobe->pending_list */ |
54 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 55 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 | 56 | |
32cdba1e ON |
57 | static struct percpu_rw_semaphore dup_mmap_sem; |
58 | ||
cb9a19fe | 59 | /* Have a copy of original instruction */ |
71434f2f | 60 | #define UPROBE_COPY_INSN 0 |
cb9a19fe | 61 | /* Can skip singlestep */ |
bb929284 | 62 | #define UPROBE_SKIP_SSTEP 1 |
cb9a19fe | 63 | |
3ff54efd SD |
64 | struct uprobe { |
65 | struct rb_node rb_node; /* node in the rb tree */ | |
66 | atomic_t ref; | |
e591c8d7 | 67 | struct rw_semaphore register_rwsem; |
3ff54efd SD |
68 | struct rw_semaphore consumer_rwsem; |
69 | struct list_head pending_list; | |
70 | struct uprobe_consumer *consumers; | |
71 | struct inode *inode; /* Also hold a ref to inode */ | |
72 | loff_t offset; | |
71434f2f | 73 | unsigned long flags; |
3ff54efd SD |
74 | struct arch_uprobe arch; |
75 | }; | |
76 | ||
2b144498 SD |
77 | /* |
78 | * valid_vma: Verify if the specified vma is an executable vma | |
79 | * Relax restrictions while unregistering: vm_flags might have | |
80 | * changed after breakpoint was inserted. | |
81 | * - is_register: indicates if we are in register context. | |
82 | * - Return 1 if the specified virtual address is in an | |
83 | * executable vma. | |
84 | */ | |
85 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
86 | { | |
e40cfce6 | 87 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED; |
2b144498 | 88 | |
e40cfce6 ON |
89 | if (is_register) |
90 | flags |= VM_WRITE; | |
2b144498 | 91 | |
e40cfce6 | 92 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
93 | } |
94 | ||
57683f72 | 95 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 96 | { |
57683f72 | 97 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
98 | } |
99 | ||
cb113b47 ON |
100 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
101 | { | |
102 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
103 | } | |
104 | ||
2b144498 SD |
105 | /** |
106 | * __replace_page - replace page in vma by new page. | |
107 | * based on replace_page in mm/ksm.c | |
108 | * | |
109 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 110 | * @addr: address the old @page is mapped at |
2b144498 SD |
111 | * @page: the cowed page we are replacing by kpage |
112 | * @kpage: the modified page we replace page by | |
113 | * | |
114 | * Returns 0 on success, -EFAULT on failure. | |
115 | */ | |
c517ee74 ON |
116 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
117 | struct page *page, struct page *kpage) | |
2b144498 SD |
118 | { |
119 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
120 | spinlock_t *ptl; |
121 | pte_t *ptep; | |
9f92448c | 122 | int err; |
6bdb913f HE |
123 | /* For mmu_notifiers */ |
124 | const unsigned long mmun_start = addr; | |
125 | const unsigned long mmun_end = addr + PAGE_SIZE; | |
2b144498 | 126 | |
194f8dcb | 127 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
128 | lock_page(page); |
129 | ||
6bdb913f | 130 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
9f92448c | 131 | err = -EAGAIN; |
5323ce71 | 132 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 133 | if (!ptep) |
9f92448c | 134 | goto unlock; |
2b144498 SD |
135 | |
136 | get_page(kpage); | |
137 | page_add_new_anon_rmap(kpage, vma, addr); | |
138 | ||
7396fa81 SD |
139 | if (!PageAnon(page)) { |
140 | dec_mm_counter(mm, MM_FILEPAGES); | |
141 | inc_mm_counter(mm, MM_ANONPAGES); | |
142 | } | |
143 | ||
2b144498 SD |
144 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
145 | ptep_clear_flush(vma, addr, ptep); | |
146 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
147 | ||
148 | page_remove_rmap(page); | |
149 | if (!page_mapped(page)) | |
150 | try_to_free_swap(page); | |
2b144498 | 151 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 152 | |
194f8dcb ON |
153 | if (vma->vm_flags & VM_LOCKED) |
154 | munlock_vma_page(page); | |
155 | put_page(page); | |
156 | ||
9f92448c ON |
157 | err = 0; |
158 | unlock: | |
6bdb913f | 159 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
9f92448c ON |
160 | unlock_page(page); |
161 | return err; | |
2b144498 SD |
162 | } |
163 | ||
164 | /** | |
5cb4ac3a | 165 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 166 | * @insn: instruction to be checked. |
5cb4ac3a | 167 | * Default implementation of is_swbp_insn |
2b144498 SD |
168 | * Returns true if @insn is a breakpoint instruction. |
169 | */ | |
5cb4ac3a | 170 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 171 | { |
5cb4ac3a | 172 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
173 | } |
174 | ||
cceb55aa ON |
175 | static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode) |
176 | { | |
177 | void *kaddr = kmap_atomic(page); | |
178 | memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE); | |
179 | kunmap_atomic(kaddr); | |
180 | } | |
181 | ||
ed6f6a50 ON |
182 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
183 | { | |
184 | uprobe_opcode_t old_opcode; | |
185 | bool is_swbp; | |
186 | ||
187 | copy_opcode(page, vaddr, &old_opcode); | |
188 | is_swbp = is_swbp_insn(&old_opcode); | |
189 | ||
190 | if (is_swbp_insn(new_opcode)) { | |
191 | if (is_swbp) /* register: already installed? */ | |
192 | return 0; | |
193 | } else { | |
194 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 195 | return 0; |
ed6f6a50 ON |
196 | } |
197 | ||
198 | return 1; | |
199 | } | |
200 | ||
2b144498 SD |
201 | /* |
202 | * NOTE: | |
203 | * Expect the breakpoint instruction to be the smallest size instruction for | |
204 | * the architecture. If an arch has variable length instruction and the | |
205 | * breakpoint instruction is not of the smallest length instruction | |
cceb55aa | 206 | * supported by that architecture then we need to modify is_swbp_at_addr and |
2b144498 SD |
207 | * write_opcode accordingly. This would never be a problem for archs that |
208 | * have fixed length instructions. | |
209 | */ | |
210 | ||
211 | /* | |
212 | * write_opcode - write the opcode at a given virtual address. | |
213 | * @mm: the probed process address space. | |
2b144498 SD |
214 | * @vaddr: the virtual address to store the opcode. |
215 | * @opcode: opcode to be written at @vaddr. | |
216 | * | |
217 | * Called with mm->mmap_sem held (for read and with a reference to | |
218 | * mm). | |
219 | * | |
220 | * For mm @mm, write the opcode at @vaddr. | |
221 | * Return 0 (success) or a negative errno. | |
222 | */ | |
cceb55aa ON |
223 | static int write_opcode(struct mm_struct *mm, unsigned long vaddr, |
224 | uprobe_opcode_t opcode) | |
2b144498 SD |
225 | { |
226 | struct page *old_page, *new_page; | |
2b144498 SD |
227 | void *vaddr_old, *vaddr_new; |
228 | struct vm_area_struct *vma; | |
2b144498 | 229 | int ret; |
f403072c | 230 | |
5323ce71 | 231 | retry: |
2b144498 | 232 | /* Read the page with vaddr into memory */ |
75ed82ea | 233 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
234 | if (ret <= 0) |
235 | return ret; | |
7b2d81d4 | 236 | |
ed6f6a50 ON |
237 | ret = verify_opcode(old_page, vaddr, &opcode); |
238 | if (ret <= 0) | |
239 | goto put_old; | |
240 | ||
2b144498 SD |
241 | ret = -ENOMEM; |
242 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
243 | if (!new_page) | |
9f92448c | 244 | goto put_old; |
2b144498 SD |
245 | |
246 | __SetPageUptodate(new_page); | |
247 | ||
2b144498 SD |
248 | /* copy the page now that we've got it stable */ |
249 | vaddr_old = kmap_atomic(old_page); | |
250 | vaddr_new = kmap_atomic(new_page); | |
251 | ||
252 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 253 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
254 | |
255 | kunmap_atomic(vaddr_new); | |
256 | kunmap_atomic(vaddr_old); | |
257 | ||
258 | ret = anon_vma_prepare(vma); | |
259 | if (ret) | |
9f92448c | 260 | goto put_new; |
2b144498 | 261 | |
c517ee74 | 262 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 263 | |
9f92448c | 264 | put_new: |
2b144498 | 265 | page_cache_release(new_page); |
9f92448c | 266 | put_old: |
7b2d81d4 IM |
267 | put_page(old_page); |
268 | ||
5323ce71 ON |
269 | if (unlikely(ret == -EAGAIN)) |
270 | goto retry; | |
2b144498 SD |
271 | return ret; |
272 | } | |
273 | ||
2b144498 | 274 | /** |
5cb4ac3a | 275 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 276 | * @auprobe: arch specific probepoint information. |
2b144498 | 277 | * @mm: the probed process address space. |
2b144498 SD |
278 | * @vaddr: the virtual address to insert the opcode. |
279 | * | |
280 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
281 | * Return 0 (success) or a negative errno. | |
282 | */ | |
5cb4ac3a | 283 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 284 | { |
cceb55aa | 285 | return write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
286 | } |
287 | ||
288 | /** | |
289 | * set_orig_insn - Restore the original instruction. | |
290 | * @mm: the probed process address space. | |
e3343e6a | 291 | * @auprobe: arch specific probepoint information. |
2b144498 | 292 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
293 | * |
294 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
295 | * Return 0 (success) or a negative errno. | |
296 | */ | |
7b2d81d4 | 297 | int __weak |
ded86e7c | 298 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 299 | { |
cceb55aa | 300 | return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
301 | } |
302 | ||
303 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
304 | { | |
305 | if (l->inode < r->inode) | |
306 | return -1; | |
7b2d81d4 | 307 | |
2b144498 SD |
308 | if (l->inode > r->inode) |
309 | return 1; | |
2b144498 | 310 | |
7b2d81d4 IM |
311 | if (l->offset < r->offset) |
312 | return -1; | |
313 | ||
314 | if (l->offset > r->offset) | |
315 | return 1; | |
2b144498 SD |
316 | |
317 | return 0; | |
318 | } | |
319 | ||
320 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
321 | { | |
322 | struct uprobe u = { .inode = inode, .offset = offset }; | |
323 | struct rb_node *n = uprobes_tree.rb_node; | |
324 | struct uprobe *uprobe; | |
325 | int match; | |
326 | ||
327 | while (n) { | |
328 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
329 | match = match_uprobe(&u, uprobe); | |
330 | if (!match) { | |
331 | atomic_inc(&uprobe->ref); | |
332 | return uprobe; | |
333 | } | |
7b2d81d4 | 334 | |
2b144498 SD |
335 | if (match < 0) |
336 | n = n->rb_left; | |
337 | else | |
338 | n = n->rb_right; | |
339 | } | |
340 | return NULL; | |
341 | } | |
342 | ||
343 | /* | |
344 | * Find a uprobe corresponding to a given inode:offset | |
345 | * Acquires uprobes_treelock | |
346 | */ | |
347 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
348 | { | |
349 | struct uprobe *uprobe; | |
2b144498 | 350 | |
6f47caa0 | 351 | spin_lock(&uprobes_treelock); |
2b144498 | 352 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 353 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 354 | |
2b144498 SD |
355 | return uprobe; |
356 | } | |
357 | ||
358 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
359 | { | |
360 | struct rb_node **p = &uprobes_tree.rb_node; | |
361 | struct rb_node *parent = NULL; | |
362 | struct uprobe *u; | |
363 | int match; | |
364 | ||
365 | while (*p) { | |
366 | parent = *p; | |
367 | u = rb_entry(parent, struct uprobe, rb_node); | |
368 | match = match_uprobe(uprobe, u); | |
369 | if (!match) { | |
370 | atomic_inc(&u->ref); | |
371 | return u; | |
372 | } | |
373 | ||
374 | if (match < 0) | |
375 | p = &parent->rb_left; | |
376 | else | |
377 | p = &parent->rb_right; | |
378 | ||
379 | } | |
7b2d81d4 | 380 | |
2b144498 SD |
381 | u = NULL; |
382 | rb_link_node(&uprobe->rb_node, parent, p); | |
383 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
384 | /* get access + creation ref */ | |
385 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 386 | |
2b144498 SD |
387 | return u; |
388 | } | |
389 | ||
390 | /* | |
7b2d81d4 | 391 | * Acquire uprobes_treelock. |
2b144498 SD |
392 | * Matching uprobe already exists in rbtree; |
393 | * increment (access refcount) and return the matching uprobe. | |
394 | * | |
395 | * No matching uprobe; insert the uprobe in rb_tree; | |
396 | * get a double refcount (access + creation) and return NULL. | |
397 | */ | |
398 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
399 | { | |
2b144498 SD |
400 | struct uprobe *u; |
401 | ||
6f47caa0 | 402 | spin_lock(&uprobes_treelock); |
2b144498 | 403 | u = __insert_uprobe(uprobe); |
6f47caa0 | 404 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 405 | |
2b144498 SD |
406 | return u; |
407 | } | |
408 | ||
409 | static void put_uprobe(struct uprobe *uprobe) | |
410 | { | |
411 | if (atomic_dec_and_test(&uprobe->ref)) | |
412 | kfree(uprobe); | |
413 | } | |
414 | ||
415 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
416 | { | |
417 | struct uprobe *uprobe, *cur_uprobe; | |
418 | ||
419 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
420 | if (!uprobe) | |
421 | return NULL; | |
422 | ||
423 | uprobe->inode = igrab(inode); | |
424 | uprobe->offset = offset; | |
e591c8d7 | 425 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 426 | init_rwsem(&uprobe->consumer_rwsem); |
bbc33d05 ON |
427 | /* For now assume that the instruction need not be single-stepped */ |
428 | __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); | |
2b144498 SD |
429 | |
430 | /* add to uprobes_tree, sorted on inode:offset */ | |
431 | cur_uprobe = insert_uprobe(uprobe); | |
432 | ||
433 | /* a uprobe exists for this inode:offset combination */ | |
434 | if (cur_uprobe) { | |
435 | kfree(uprobe); | |
436 | uprobe = cur_uprobe; | |
437 | iput(inode); | |
7b2d81d4 IM |
438 | } |
439 | ||
2b144498 SD |
440 | return uprobe; |
441 | } | |
442 | ||
9a98e03c | 443 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
444 | { |
445 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
446 | uc->next = uprobe->consumers; |
447 | uprobe->consumers = uc; | |
2b144498 | 448 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
449 | } |
450 | ||
451 | /* | |
e3343e6a SD |
452 | * For uprobe @uprobe, delete the consumer @uc. |
453 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
454 | * or return false. |
455 | */ | |
e3343e6a | 456 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
457 | { |
458 | struct uprobe_consumer **con; | |
459 | bool ret = false; | |
460 | ||
461 | down_write(&uprobe->consumer_rwsem); | |
462 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
463 | if (*con == uc) { |
464 | *con = uc->next; | |
2b144498 SD |
465 | ret = true; |
466 | break; | |
467 | } | |
468 | } | |
469 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 470 | |
2b144498 SD |
471 | return ret; |
472 | } | |
473 | ||
e3343e6a | 474 | static int |
d436615e | 475 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 476 | unsigned long nbytes, loff_t offset) |
2b144498 | 477 | { |
2b144498 SD |
478 | struct page *page; |
479 | void *vaddr; | |
593609a5 ON |
480 | unsigned long off; |
481 | pgoff_t idx; | |
2b144498 SD |
482 | |
483 | if (!filp) | |
484 | return -EINVAL; | |
485 | ||
cc359d18 ON |
486 | if (!mapping->a_ops->readpage) |
487 | return -EIO; | |
488 | ||
593609a5 ON |
489 | idx = offset >> PAGE_CACHE_SHIFT; |
490 | off = offset & ~PAGE_MASK; | |
2b144498 SD |
491 | |
492 | /* | |
493 | * Ensure that the page that has the original instruction is | |
494 | * populated and in page-cache. | |
495 | */ | |
496 | page = read_mapping_page(mapping, idx, filp); | |
497 | if (IS_ERR(page)) | |
498 | return PTR_ERR(page); | |
499 | ||
500 | vaddr = kmap_atomic(page); | |
593609a5 | 501 | memcpy(insn, vaddr + off, nbytes); |
2b144498 SD |
502 | kunmap_atomic(vaddr); |
503 | page_cache_release(page); | |
7b2d81d4 | 504 | |
2b144498 SD |
505 | return 0; |
506 | } | |
507 | ||
d436615e | 508 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
509 | { |
510 | struct address_space *mapping; | |
2b144498 | 511 | unsigned long nbytes; |
7b2d81d4 | 512 | int bytes; |
2b144498 | 513 | |
d436615e | 514 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
515 | mapping = uprobe->inode->i_mapping; |
516 | ||
517 | /* Instruction at end of binary; copy only available bytes */ | |
518 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
519 | bytes = uprobe->inode->i_size - uprobe->offset; | |
520 | else | |
521 | bytes = MAX_UINSN_BYTES; | |
522 | ||
523 | /* Instruction at the page-boundary; copy bytes in second page */ | |
524 | if (nbytes < bytes) { | |
fc36f595 ON |
525 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
526 | bytes - nbytes, uprobe->offset + nbytes); | |
527 | if (err) | |
528 | return err; | |
2b144498 SD |
529 | bytes = nbytes; |
530 | } | |
d436615e | 531 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
532 | } |
533 | ||
cb9a19fe ON |
534 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
535 | struct mm_struct *mm, unsigned long vaddr) | |
536 | { | |
537 | int ret = 0; | |
538 | ||
71434f2f | 539 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
540 | return ret; |
541 | ||
d4d3ccc6 ON |
542 | /* TODO: move this into _register, until then we abuse this sem. */ |
543 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 544 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
545 | goto out; |
546 | ||
cb9a19fe ON |
547 | ret = copy_insn(uprobe, file); |
548 | if (ret) | |
549 | goto out; | |
550 | ||
551 | ret = -ENOTSUPP; | |
552 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) | |
553 | goto out; | |
554 | ||
555 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
556 | if (ret) | |
557 | goto out; | |
558 | ||
559 | /* write_opcode() assumes we don't cross page boundary */ | |
560 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
561 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
562 | ||
563 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 564 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
565 | |
566 | out: | |
d4d3ccc6 | 567 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 568 | |
cb9a19fe ON |
569 | return ret; |
570 | } | |
571 | ||
8a7f2fa0 ON |
572 | static inline bool consumer_filter(struct uprobe_consumer *uc, |
573 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
806a98bd | 574 | { |
8a7f2fa0 | 575 | return !uc->filter || uc->filter(uc, ctx, mm); |
806a98bd ON |
576 | } |
577 | ||
8a7f2fa0 ON |
578 | static bool filter_chain(struct uprobe *uprobe, |
579 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
63633cbf | 580 | { |
1ff6fee5 ON |
581 | struct uprobe_consumer *uc; |
582 | bool ret = false; | |
583 | ||
584 | down_read(&uprobe->consumer_rwsem); | |
585 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
8a7f2fa0 | 586 | ret = consumer_filter(uc, ctx, mm); |
1ff6fee5 ON |
587 | if (ret) |
588 | break; | |
589 | } | |
590 | up_read(&uprobe->consumer_rwsem); | |
591 | ||
592 | return ret; | |
63633cbf ON |
593 | } |
594 | ||
e3343e6a SD |
595 | static int |
596 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 597 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 598 | { |
f8ac4ec9 | 599 | bool first_uprobe; |
2b144498 SD |
600 | int ret; |
601 | ||
cb9a19fe ON |
602 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
603 | if (ret) | |
604 | return ret; | |
682968e0 | 605 | |
f8ac4ec9 ON |
606 | /* |
607 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
608 | * the task can hit this breakpoint right after __replace_page(). | |
609 | */ | |
610 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
611 | if (first_uprobe) | |
612 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
613 | ||
816c03fb | 614 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
615 | if (!ret) |
616 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
617 | else if (first_uprobe) | |
f8ac4ec9 | 618 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
619 | |
620 | return ret; | |
621 | } | |
622 | ||
076a365b | 623 | static int |
816c03fb | 624 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 625 | { |
9f68f672 | 626 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 627 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
628 | } |
629 | ||
06b7bcd8 ON |
630 | static inline bool uprobe_is_active(struct uprobe *uprobe) |
631 | { | |
632 | return !RB_EMPTY_NODE(&uprobe->rb_node); | |
633 | } | |
0326f5a9 | 634 | /* |
778b032d ON |
635 | * There could be threads that have already hit the breakpoint. They |
636 | * will recheck the current insn and restart if find_uprobe() fails. | |
637 | * See find_active_uprobe(). | |
0326f5a9 | 638 | */ |
2b144498 SD |
639 | static void delete_uprobe(struct uprobe *uprobe) |
640 | { | |
06b7bcd8 ON |
641 | if (WARN_ON(!uprobe_is_active(uprobe))) |
642 | return; | |
643 | ||
6f47caa0 | 644 | spin_lock(&uprobes_treelock); |
2b144498 | 645 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 646 | spin_unlock(&uprobes_treelock); |
06b7bcd8 | 647 | RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */ |
2b144498 SD |
648 | iput(uprobe->inode); |
649 | put_uprobe(uprobe); | |
2b144498 SD |
650 | } |
651 | ||
26872090 ON |
652 | struct map_info { |
653 | struct map_info *next; | |
654 | struct mm_struct *mm; | |
816c03fb | 655 | unsigned long vaddr; |
26872090 ON |
656 | }; |
657 | ||
658 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 659 | { |
26872090 ON |
660 | struct map_info *next = info->next; |
661 | kfree(info); | |
662 | return next; | |
663 | } | |
664 | ||
665 | static struct map_info * | |
666 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
667 | { | |
668 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 669 | struct vm_area_struct *vma; |
26872090 ON |
670 | struct map_info *curr = NULL; |
671 | struct map_info *prev = NULL; | |
672 | struct map_info *info; | |
673 | int more = 0; | |
2b144498 | 674 | |
26872090 ON |
675 | again: |
676 | mutex_lock(&mapping->i_mmap_mutex); | |
6b2dbba8 | 677 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
678 | if (!valid_vma(vma, is_register)) |
679 | continue; | |
680 | ||
7a5bfb66 ON |
681 | if (!prev && !more) { |
682 | /* | |
683 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
684 | * reclaim. This is optimistic, no harm done if it fails. | |
685 | */ | |
686 | prev = kmalloc(sizeof(struct map_info), | |
687 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
688 | if (prev) | |
689 | prev->next = NULL; | |
690 | } | |
26872090 ON |
691 | if (!prev) { |
692 | more++; | |
693 | continue; | |
2b144498 | 694 | } |
2b144498 | 695 | |
26872090 ON |
696 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
697 | continue; | |
7b2d81d4 | 698 | |
26872090 ON |
699 | info = prev; |
700 | prev = prev->next; | |
701 | info->next = curr; | |
702 | curr = info; | |
2b144498 | 703 | |
26872090 | 704 | info->mm = vma->vm_mm; |
57683f72 | 705 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 706 | } |
2b144498 SD |
707 | mutex_unlock(&mapping->i_mmap_mutex); |
708 | ||
26872090 ON |
709 | if (!more) |
710 | goto out; | |
711 | ||
712 | prev = curr; | |
713 | while (curr) { | |
714 | mmput(curr->mm); | |
715 | curr = curr->next; | |
716 | } | |
7b2d81d4 | 717 | |
26872090 ON |
718 | do { |
719 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
720 | if (!info) { | |
721 | curr = ERR_PTR(-ENOMEM); | |
722 | goto out; | |
723 | } | |
724 | info->next = prev; | |
725 | prev = info; | |
726 | } while (--more); | |
727 | ||
728 | goto again; | |
729 | out: | |
730 | while (prev) | |
731 | prev = free_map_info(prev); | |
732 | return curr; | |
2b144498 SD |
733 | } |
734 | ||
735 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
736 | { | |
26872090 ON |
737 | struct map_info *info; |
738 | int err = 0; | |
2b144498 | 739 | |
32cdba1e | 740 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
741 | info = build_map_info(uprobe->inode->i_mapping, |
742 | uprobe->offset, is_register); | |
32cdba1e ON |
743 | if (IS_ERR(info)) { |
744 | err = PTR_ERR(info); | |
745 | goto out; | |
746 | } | |
7b2d81d4 | 747 | |
26872090 ON |
748 | while (info) { |
749 | struct mm_struct *mm = info->mm; | |
750 | struct vm_area_struct *vma; | |
7b2d81d4 | 751 | |
076a365b | 752 | if (err && is_register) |
26872090 | 753 | goto free; |
7b2d81d4 | 754 | |
77fc4af1 | 755 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
756 | vma = find_vma(mm, info->vaddr); |
757 | if (!vma || !valid_vma(vma, is_register) || | |
758 | vma->vm_file->f_mapping->host != uprobe->inode) | |
26872090 ON |
759 | goto unlock; |
760 | ||
f4d6dfe5 ON |
761 | if (vma->vm_start > info->vaddr || |
762 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 763 | goto unlock; |
2b144498 | 764 | |
806a98bd ON |
765 | if (is_register) { |
766 | /* consult only the "caller", new consumer. */ | |
8a7f2fa0 ON |
767 | if (consumer_filter(uprobe->consumers, |
768 | UPROBE_FILTER_REGISTER, mm)) | |
806a98bd ON |
769 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
770 | } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) { | |
8a7f2fa0 ON |
771 | if (!filter_chain(uprobe, |
772 | UPROBE_FILTER_UNREGISTER, mm)) | |
806a98bd ON |
773 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
774 | } | |
78f74116 | 775 | |
26872090 ON |
776 | unlock: |
777 | up_write(&mm->mmap_sem); | |
778 | free: | |
779 | mmput(mm); | |
780 | info = free_map_info(info); | |
2b144498 | 781 | } |
32cdba1e ON |
782 | out: |
783 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 784 | return err; |
2b144498 SD |
785 | } |
786 | ||
9a98e03c | 787 | static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 788 | { |
9a98e03c | 789 | consumer_add(uprobe, uc); |
bb929284 | 790 | return register_for_each_vma(uprobe, true); |
2b144498 SD |
791 | } |
792 | ||
04aab9b2 | 793 | static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 794 | { |
04aab9b2 ON |
795 | int err; |
796 | ||
797 | if (!consumer_del(uprobe, uc)) /* WARN? */ | |
798 | return; | |
2b144498 | 799 | |
04aab9b2 | 800 | err = register_for_each_vma(uprobe, false); |
bb929284 ON |
801 | /* TODO : cant unregister? schedule a worker thread */ |
802 | if (!uprobe->consumers && !err) | |
803 | delete_uprobe(uprobe); | |
2b144498 SD |
804 | } |
805 | ||
806 | /* | |
7b2d81d4 | 807 | * uprobe_register - register a probe |
2b144498 SD |
808 | * @inode: the file in which the probe has to be placed. |
809 | * @offset: offset from the start of the file. | |
e3343e6a | 810 | * @uc: information on howto handle the probe.. |
2b144498 | 811 | * |
7b2d81d4 | 812 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
813 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
814 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 815 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 816 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 817 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
818 | * unregisters. |
819 | * | |
820 | * Return errno if it cannot successully install probes | |
821 | * else return 0 (success) | |
822 | */ | |
e3343e6a | 823 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
824 | { |
825 | struct uprobe *uprobe; | |
7b2d81d4 | 826 | int ret; |
2b144498 | 827 | |
f0744af7 | 828 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 829 | if (offset > i_size_read(inode)) |
7b2d81d4 | 830 | return -EINVAL; |
2b144498 | 831 | |
66d06dff | 832 | retry: |
2b144498 | 833 | uprobe = alloc_uprobe(inode, offset); |
66d06dff ON |
834 | if (!uprobe) |
835 | return -ENOMEM; | |
836 | /* | |
837 | * We can race with uprobe_unregister()->delete_uprobe(). | |
838 | * Check uprobe_is_active() and retry if it is false. | |
839 | */ | |
840 | down_write(&uprobe->register_rwsem); | |
841 | ret = -EAGAIN; | |
842 | if (likely(uprobe_is_active(uprobe))) { | |
9a98e03c ON |
843 | ret = __uprobe_register(uprobe, uc); |
844 | if (ret) | |
04aab9b2 | 845 | __uprobe_unregister(uprobe, uc); |
2b144498 | 846 | } |
66d06dff ON |
847 | up_write(&uprobe->register_rwsem); |
848 | put_uprobe(uprobe); | |
2b144498 | 849 | |
66d06dff ON |
850 | if (unlikely(ret == -EAGAIN)) |
851 | goto retry; | |
2b144498 SD |
852 | return ret; |
853 | } | |
854 | ||
855 | /* | |
7b2d81d4 | 856 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
857 | * @inode: the file in which the probe has to be removed. |
858 | * @offset: offset from the start of the file. | |
e3343e6a | 859 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 860 | */ |
e3343e6a | 861 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 862 | { |
7b2d81d4 | 863 | struct uprobe *uprobe; |
2b144498 | 864 | |
2b144498 SD |
865 | uprobe = find_uprobe(inode, offset); |
866 | if (!uprobe) | |
867 | return; | |
868 | ||
e591c8d7 | 869 | down_write(&uprobe->register_rwsem); |
04aab9b2 | 870 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 871 | up_write(&uprobe->register_rwsem); |
c91368c4 | 872 | put_uprobe(uprobe); |
2b144498 SD |
873 | } |
874 | ||
da1816b1 ON |
875 | static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm) |
876 | { | |
877 | struct vm_area_struct *vma; | |
878 | int err = 0; | |
879 | ||
880 | down_read(&mm->mmap_sem); | |
881 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
882 | unsigned long vaddr; | |
883 | loff_t offset; | |
884 | ||
885 | if (!valid_vma(vma, false) || | |
886 | vma->vm_file->f_mapping->host != uprobe->inode) | |
887 | continue; | |
888 | ||
889 | offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; | |
890 | if (uprobe->offset < offset || | |
891 | uprobe->offset >= offset + vma->vm_end - vma->vm_start) | |
892 | continue; | |
893 | ||
894 | vaddr = offset_to_vaddr(vma, uprobe->offset); | |
895 | err |= remove_breakpoint(uprobe, mm, vaddr); | |
896 | } | |
897 | up_read(&mm->mmap_sem); | |
898 | ||
899 | return err; | |
900 | } | |
901 | ||
891c3970 ON |
902 | static struct rb_node * |
903 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 904 | { |
2b144498 | 905 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
906 | |
907 | while (n) { | |
891c3970 | 908 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 909 | |
891c3970 | 910 | if (inode < u->inode) { |
2b144498 | 911 | n = n->rb_left; |
891c3970 | 912 | } else if (inode > u->inode) { |
2b144498 | 913 | n = n->rb_right; |
891c3970 ON |
914 | } else { |
915 | if (max < u->offset) | |
916 | n = n->rb_left; | |
917 | else if (min > u->offset) | |
918 | n = n->rb_right; | |
919 | else | |
920 | break; | |
921 | } | |
2b144498 | 922 | } |
7b2d81d4 | 923 | |
891c3970 | 924 | return n; |
2b144498 SD |
925 | } |
926 | ||
927 | /* | |
891c3970 | 928 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 929 | */ |
891c3970 ON |
930 | static void build_probe_list(struct inode *inode, |
931 | struct vm_area_struct *vma, | |
932 | unsigned long start, unsigned long end, | |
933 | struct list_head *head) | |
2b144498 | 934 | { |
891c3970 | 935 | loff_t min, max; |
891c3970 ON |
936 | struct rb_node *n, *t; |
937 | struct uprobe *u; | |
7b2d81d4 | 938 | |
891c3970 | 939 | INIT_LIST_HEAD(head); |
cb113b47 | 940 | min = vaddr_to_offset(vma, start); |
891c3970 | 941 | max = min + (end - start) - 1; |
2b144498 | 942 | |
6f47caa0 | 943 | spin_lock(&uprobes_treelock); |
891c3970 ON |
944 | n = find_node_in_range(inode, min, max); |
945 | if (n) { | |
946 | for (t = n; t; t = rb_prev(t)) { | |
947 | u = rb_entry(t, struct uprobe, rb_node); | |
948 | if (u->inode != inode || u->offset < min) | |
949 | break; | |
950 | list_add(&u->pending_list, head); | |
951 | atomic_inc(&u->ref); | |
952 | } | |
953 | for (t = n; (t = rb_next(t)); ) { | |
954 | u = rb_entry(t, struct uprobe, rb_node); | |
955 | if (u->inode != inode || u->offset > max) | |
956 | break; | |
957 | list_add(&u->pending_list, head); | |
958 | atomic_inc(&u->ref); | |
959 | } | |
2b144498 | 960 | } |
6f47caa0 | 961 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
962 | } |
963 | ||
964 | /* | |
5e5be71a | 965 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 966 | * |
5e5be71a ON |
967 | * Currently we ignore all errors and always return 0, the callers |
968 | * can't handle the failure anyway. | |
2b144498 | 969 | */ |
7b2d81d4 | 970 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
971 | { |
972 | struct list_head tmp_list; | |
665605a2 | 973 | struct uprobe *uprobe, *u; |
2b144498 | 974 | struct inode *inode; |
2b144498 | 975 | |
441f1eb7 | 976 | if (no_uprobe_events() || !valid_vma(vma, true)) |
7b2d81d4 | 977 | return 0; |
2b144498 SD |
978 | |
979 | inode = vma->vm_file->f_mapping->host; | |
980 | if (!inode) | |
7b2d81d4 | 981 | return 0; |
2b144498 | 982 | |
2b144498 | 983 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 984 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
806a98bd ON |
985 | /* |
986 | * We can race with uprobe_unregister(), this uprobe can be already | |
987 | * removed. But in this case filter_chain() must return false, all | |
988 | * consumers have gone away. | |
989 | */ | |
665605a2 | 990 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
806a98bd | 991 | if (!fatal_signal_pending(current) && |
8a7f2fa0 | 992 | filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) { |
57683f72 | 993 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 994 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
995 | } |
996 | put_uprobe(uprobe); | |
997 | } | |
2b144498 SD |
998 | mutex_unlock(uprobes_mmap_hash(inode)); |
999 | ||
5e5be71a | 1000 | return 0; |
2b144498 SD |
1001 | } |
1002 | ||
9f68f672 ON |
1003 | static bool |
1004 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1005 | { | |
1006 | loff_t min, max; | |
1007 | struct inode *inode; | |
1008 | struct rb_node *n; | |
1009 | ||
1010 | inode = vma->vm_file->f_mapping->host; | |
1011 | ||
1012 | min = vaddr_to_offset(vma, start); | |
1013 | max = min + (end - start) - 1; | |
1014 | ||
1015 | spin_lock(&uprobes_treelock); | |
1016 | n = find_node_in_range(inode, min, max); | |
1017 | spin_unlock(&uprobes_treelock); | |
1018 | ||
1019 | return !!n; | |
1020 | } | |
1021 | ||
682968e0 SD |
1022 | /* |
1023 | * Called in context of a munmap of a vma. | |
1024 | */ | |
cbc91f71 | 1025 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1026 | { |
441f1eb7 | 1027 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1028 | return; |
1029 | ||
2fd611a9 ON |
1030 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1031 | return; | |
1032 | ||
9f68f672 ON |
1033 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1034 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1035 | return; |
1036 | ||
9f68f672 ON |
1037 | if (vma_has_uprobes(vma, start, end)) |
1038 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1039 | } |
1040 | ||
d4b3b638 SD |
1041 | /* Slot allocation for XOL */ |
1042 | static int xol_add_vma(struct xol_area *area) | |
1043 | { | |
1044 | struct mm_struct *mm; | |
1045 | int ret; | |
1046 | ||
1047 | area->page = alloc_page(GFP_HIGHUSER); | |
1048 | if (!area->page) | |
1049 | return -ENOMEM; | |
1050 | ||
1051 | ret = -EALREADY; | |
1052 | mm = current->mm; | |
1053 | ||
1054 | down_write(&mm->mmap_sem); | |
1055 | if (mm->uprobes_state.xol_area) | |
1056 | goto fail; | |
1057 | ||
1058 | ret = -ENOMEM; | |
1059 | ||
1060 | /* Try to map as high as possible, this is only a hint. */ | |
1061 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1062 | if (area->vaddr & ~PAGE_MASK) { | |
1063 | ret = area->vaddr; | |
1064 | goto fail; | |
1065 | } | |
1066 | ||
1067 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1068 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1069 | if (ret) | |
1070 | goto fail; | |
1071 | ||
1072 | smp_wmb(); /* pairs with get_xol_area() */ | |
1073 | mm->uprobes_state.xol_area = area; | |
1074 | ret = 0; | |
1075 | ||
1076 | fail: | |
1077 | up_write(&mm->mmap_sem); | |
1078 | if (ret) | |
1079 | __free_page(area->page); | |
1080 | ||
1081 | return ret; | |
1082 | } | |
1083 | ||
1084 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1085 | { | |
1086 | struct xol_area *area; | |
1087 | ||
1088 | area = mm->uprobes_state.xol_area; | |
1089 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1090 | ||
1091 | return area; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * xol_alloc_area - Allocate process's xol_area. | |
1096 | * This area will be used for storing instructions for execution out of | |
1097 | * line. | |
1098 | * | |
1099 | * Returns the allocated area or NULL. | |
1100 | */ | |
1101 | static struct xol_area *xol_alloc_area(void) | |
1102 | { | |
1103 | struct xol_area *area; | |
1104 | ||
1105 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1106 | if (unlikely(!area)) | |
1107 | return NULL; | |
1108 | ||
1109 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1110 | ||
1111 | if (!area->bitmap) | |
1112 | goto fail; | |
1113 | ||
1114 | init_waitqueue_head(&area->wq); | |
1115 | if (!xol_add_vma(area)) | |
1116 | return area; | |
1117 | ||
1118 | fail: | |
1119 | kfree(area->bitmap); | |
1120 | kfree(area); | |
1121 | ||
1122 | return get_xol_area(current->mm); | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * uprobe_clear_state - Free the area allocated for slots. | |
1127 | */ | |
1128 | void uprobe_clear_state(struct mm_struct *mm) | |
1129 | { | |
1130 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1131 | ||
1132 | if (!area) | |
1133 | return; | |
1134 | ||
1135 | put_page(area->page); | |
1136 | kfree(area->bitmap); | |
1137 | kfree(area); | |
1138 | } | |
1139 | ||
32cdba1e ON |
1140 | void uprobe_start_dup_mmap(void) |
1141 | { | |
1142 | percpu_down_read(&dup_mmap_sem); | |
1143 | } | |
1144 | ||
1145 | void uprobe_end_dup_mmap(void) | |
1146 | { | |
1147 | percpu_up_read(&dup_mmap_sem); | |
1148 | } | |
1149 | ||
f8ac4ec9 ON |
1150 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1151 | { | |
61559a81 ON |
1152 | newmm->uprobes_state.xol_area = NULL; |
1153 | ||
9f68f672 | 1154 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1155 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1156 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1157 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1158 | } | |
f8ac4ec9 ON |
1159 | } |
1160 | ||
d4b3b638 SD |
1161 | /* |
1162 | * - search for a free slot. | |
1163 | */ | |
1164 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1165 | { | |
1166 | unsigned long slot_addr; | |
1167 | int slot_nr; | |
1168 | ||
1169 | do { | |
1170 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1171 | if (slot_nr < UINSNS_PER_PAGE) { | |
1172 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1173 | break; | |
1174 | ||
1175 | slot_nr = UINSNS_PER_PAGE; | |
1176 | continue; | |
1177 | } | |
1178 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1179 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1180 | ||
1181 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1182 | atomic_inc(&area->slot_count); | |
1183 | ||
1184 | return slot_addr; | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * xol_get_insn_slot - If was not allocated a slot, then | |
1189 | * allocate a slot. | |
1190 | * Returns the allocated slot address or 0. | |
1191 | */ | |
1192 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1193 | { | |
1194 | struct xol_area *area; | |
1195 | unsigned long offset; | |
1196 | void *vaddr; | |
1197 | ||
1198 | area = get_xol_area(current->mm); | |
1199 | if (!area) { | |
1200 | area = xol_alloc_area(); | |
1201 | if (!area) | |
1202 | return 0; | |
1203 | } | |
1204 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1205 | ||
1206 | /* | |
1207 | * Initialize the slot if xol_vaddr points to valid | |
1208 | * instruction slot. | |
1209 | */ | |
1210 | if (unlikely(!current->utask->xol_vaddr)) | |
1211 | return 0; | |
1212 | ||
1213 | current->utask->vaddr = slot_addr; | |
1214 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1215 | vaddr = kmap_atomic(area->page); | |
1216 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1217 | kunmap_atomic(vaddr); | |
65b6ecc0 RV |
1218 | /* |
1219 | * We probably need flush_icache_user_range() but it needs vma. | |
1220 | * This should work on supported architectures too. | |
1221 | */ | |
1222 | flush_dcache_page(area->page); | |
d4b3b638 SD |
1223 | |
1224 | return current->utask->xol_vaddr; | |
1225 | } | |
1226 | ||
1227 | /* | |
1228 | * xol_free_insn_slot - If slot was earlier allocated by | |
1229 | * @xol_get_insn_slot(), make the slot available for | |
1230 | * subsequent requests. | |
1231 | */ | |
1232 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1233 | { | |
1234 | struct xol_area *area; | |
1235 | unsigned long vma_end; | |
1236 | unsigned long slot_addr; | |
1237 | ||
1238 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1239 | return; | |
1240 | ||
1241 | slot_addr = tsk->utask->xol_vaddr; | |
1242 | ||
1243 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1244 | return; | |
1245 | ||
1246 | area = tsk->mm->uprobes_state.xol_area; | |
1247 | vma_end = area->vaddr + PAGE_SIZE; | |
1248 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1249 | unsigned long offset; | |
1250 | int slot_nr; | |
1251 | ||
1252 | offset = slot_addr - area->vaddr; | |
1253 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1254 | if (slot_nr >= UINSNS_PER_PAGE) | |
1255 | return; | |
1256 | ||
1257 | clear_bit(slot_nr, area->bitmap); | |
1258 | atomic_dec(&area->slot_count); | |
1259 | if (waitqueue_active(&area->wq)) | |
1260 | wake_up(&area->wq); | |
1261 | ||
1262 | tsk->utask->xol_vaddr = 0; | |
1263 | } | |
1264 | } | |
1265 | ||
0326f5a9 SD |
1266 | /** |
1267 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1268 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1269 | * instruction. | |
1270 | * Return the address of the breakpoint instruction. | |
1271 | */ | |
1272 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1273 | { | |
1274 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Called with no locks held. | |
1279 | * Called in context of a exiting or a exec-ing thread. | |
1280 | */ | |
1281 | void uprobe_free_utask(struct task_struct *t) | |
1282 | { | |
1283 | struct uprobe_task *utask = t->utask; | |
1284 | ||
0326f5a9 SD |
1285 | if (!utask) |
1286 | return; | |
1287 | ||
1288 | if (utask->active_uprobe) | |
1289 | put_uprobe(utask->active_uprobe); | |
1290 | ||
d4b3b638 | 1291 | xol_free_insn_slot(t); |
0326f5a9 SD |
1292 | kfree(utask); |
1293 | t->utask = NULL; | |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * Called in context of a new clone/fork from copy_process. | |
1298 | */ | |
1299 | void uprobe_copy_process(struct task_struct *t) | |
1300 | { | |
1301 | t->utask = NULL; | |
0326f5a9 SD |
1302 | } |
1303 | ||
1304 | /* | |
1305 | * Allocate a uprobe_task object for the task. | |
1306 | * Called when the thread hits a breakpoint for the first time. | |
1307 | * | |
1308 | * Returns: | |
1309 | * - pointer to new uprobe_task on success | |
1310 | * - NULL otherwise | |
1311 | */ | |
1312 | static struct uprobe_task *add_utask(void) | |
1313 | { | |
1314 | struct uprobe_task *utask; | |
1315 | ||
1316 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1317 | if (unlikely(!utask)) | |
1318 | return NULL; | |
1319 | ||
0326f5a9 SD |
1320 | current->utask = utask; |
1321 | return utask; | |
1322 | } | |
1323 | ||
1324 | /* Prepare to single-step probed instruction out of line. */ | |
1325 | static int | |
1326 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1327 | { | |
d4b3b638 SD |
1328 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1329 | return 0; | |
1330 | ||
0326f5a9 SD |
1331 | return -EFAULT; |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * If we are singlestepping, then ensure this thread is not connected to | |
1336 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1337 | * triggers the signal, restart the original insn even if the task is | |
1338 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1339 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1340 | * _same_ instruction should be repeated again after return from the signal | |
1341 | * handler, and SSTEP can never finish in this case. | |
1342 | */ | |
1343 | bool uprobe_deny_signal(void) | |
1344 | { | |
1345 | struct task_struct *t = current; | |
1346 | struct uprobe_task *utask = t->utask; | |
1347 | ||
1348 | if (likely(!utask || !utask->active_uprobe)) | |
1349 | return false; | |
1350 | ||
1351 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1352 | ||
1353 | if (signal_pending(t)) { | |
1354 | spin_lock_irq(&t->sighand->siglock); | |
1355 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1356 | spin_unlock_irq(&t->sighand->siglock); | |
1357 | ||
1358 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1359 | utask->state = UTASK_SSTEP_TRAPPED; | |
1360 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1361 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1362 | } | |
1363 | } | |
1364 | ||
1365 | return true; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * Avoid singlestepping the original instruction if the original instruction | |
1370 | * is a NOP or can be emulated. | |
1371 | */ | |
1372 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1373 | { | |
71434f2f | 1374 | if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) { |
0578a970 ON |
1375 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
1376 | return true; | |
71434f2f | 1377 | clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); |
0578a970 | 1378 | } |
0326f5a9 SD |
1379 | return false; |
1380 | } | |
1381 | ||
499a4f3e ON |
1382 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1383 | { | |
1384 | struct vm_area_struct *vma; | |
1385 | ||
1386 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1387 | if (!valid_vma(vma, false)) | |
1388 | continue; | |
1389 | /* | |
1390 | * This is not strictly accurate, we can race with | |
1391 | * uprobe_unregister() and see the already removed | |
1392 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 1393 | * Or this uprobe can be filtered out. |
499a4f3e ON |
1394 | */ |
1395 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1396 | return; | |
1397 | } | |
1398 | ||
1399 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1400 | } | |
1401 | ||
ec75fba9 ON |
1402 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
1403 | { | |
1404 | struct page *page; | |
1405 | uprobe_opcode_t opcode; | |
1406 | int result; | |
1407 | ||
1408 | pagefault_disable(); | |
1409 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1410 | sizeof(opcode)); | |
1411 | pagefault_enable(); | |
1412 | ||
1413 | if (likely(result == 0)) | |
1414 | goto out; | |
1415 | ||
1416 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1417 | if (result < 0) | |
1418 | return result; | |
1419 | ||
1420 | copy_opcode(page, vaddr, &opcode); | |
1421 | put_page(page); | |
1422 | out: | |
1423 | return is_swbp_insn(&opcode); | |
1424 | } | |
1425 | ||
d790d346 | 1426 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1427 | { |
3a9ea052 ON |
1428 | struct mm_struct *mm = current->mm; |
1429 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1430 | struct vm_area_struct *vma; |
0326f5a9 | 1431 | |
0326f5a9 SD |
1432 | down_read(&mm->mmap_sem); |
1433 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1434 | if (vma && vma->vm_start <= bp_vaddr) { |
1435 | if (valid_vma(vma, false)) { | |
cb113b47 ON |
1436 | struct inode *inode = vma->vm_file->f_mapping->host; |
1437 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); | |
0326f5a9 | 1438 | |
3a9ea052 ON |
1439 | uprobe = find_uprobe(inode, offset); |
1440 | } | |
d790d346 ON |
1441 | |
1442 | if (!uprobe) | |
1443 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1444 | } else { | |
1445 | *is_swbp = -EFAULT; | |
0326f5a9 | 1446 | } |
499a4f3e ON |
1447 | |
1448 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1449 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1450 | up_read(&mm->mmap_sem); |
1451 | ||
3a9ea052 ON |
1452 | return uprobe; |
1453 | } | |
1454 | ||
da1816b1 ON |
1455 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
1456 | { | |
1457 | struct uprobe_consumer *uc; | |
1458 | int remove = UPROBE_HANDLER_REMOVE; | |
1459 | ||
1460 | down_read(&uprobe->register_rwsem); | |
1461 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
1462 | int rc = uc->handler(uc, regs); | |
1463 | ||
1464 | WARN(rc & ~UPROBE_HANDLER_MASK, | |
1465 | "bad rc=0x%x from %pf()\n", rc, uc->handler); | |
1466 | remove &= rc; | |
1467 | } | |
1468 | ||
1469 | if (remove && uprobe->consumers) { | |
1470 | WARN_ON(!uprobe_is_active(uprobe)); | |
1471 | unapply_uprobe(uprobe, current->mm); | |
1472 | } | |
1473 | up_read(&uprobe->register_rwsem); | |
1474 | } | |
1475 | ||
3a9ea052 ON |
1476 | /* |
1477 | * Run handler and ask thread to singlestep. | |
1478 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1479 | */ | |
1480 | static void handle_swbp(struct pt_regs *regs) | |
1481 | { | |
1482 | struct uprobe_task *utask; | |
1483 | struct uprobe *uprobe; | |
1484 | unsigned long bp_vaddr; | |
56bb4cf6 | 1485 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1486 | |
1487 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1488 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1489 | |
0326f5a9 | 1490 | if (!uprobe) { |
56bb4cf6 ON |
1491 | if (is_swbp > 0) { |
1492 | /* No matching uprobe; signal SIGTRAP. */ | |
1493 | send_sig(SIGTRAP, current, 0); | |
1494 | } else { | |
1495 | /* | |
1496 | * Either we raced with uprobe_unregister() or we can't | |
1497 | * access this memory. The latter is only possible if | |
1498 | * another thread plays with our ->mm. In both cases | |
1499 | * we can simply restart. If this vma was unmapped we | |
1500 | * can pretend this insn was not executed yet and get | |
1501 | * the (correct) SIGSEGV after restart. | |
1502 | */ | |
1503 | instruction_pointer_set(regs, bp_vaddr); | |
1504 | } | |
0326f5a9 SD |
1505 | return; |
1506 | } | |
142b18dd ON |
1507 | /* |
1508 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1509 | * After we hit the bp, _unregister + _register can install the | |
1510 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1511 | */ | |
1512 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1513 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
142b18dd | 1514 | goto restart; |
0326f5a9 SD |
1515 | |
1516 | utask = current->utask; | |
1517 | if (!utask) { | |
1518 | utask = add_utask(); | |
1519 | /* Cannot allocate; re-execute the instruction. */ | |
1520 | if (!utask) | |
0578a970 | 1521 | goto restart; |
0326f5a9 | 1522 | } |
746a9e6b | 1523 | |
0326f5a9 | 1524 | handler_chain(uprobe, regs); |
0578a970 ON |
1525 | if (can_skip_sstep(uprobe, regs)) |
1526 | goto out; | |
0326f5a9 | 1527 | |
0326f5a9 | 1528 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { |
746a9e6b ON |
1529 | utask->active_uprobe = uprobe; |
1530 | utask->state = UTASK_SSTEP; | |
0326f5a9 SD |
1531 | return; |
1532 | } | |
1533 | ||
0578a970 ON |
1534 | restart: |
1535 | /* | |
1536 | * cannot singlestep; cannot skip instruction; | |
1537 | * re-execute the instruction. | |
1538 | */ | |
1539 | instruction_pointer_set(regs, bp_vaddr); | |
1540 | out: | |
8bd87445 | 1541 | put_uprobe(uprobe); |
0326f5a9 SD |
1542 | } |
1543 | ||
1544 | /* | |
1545 | * Perform required fix-ups and disable singlestep. | |
1546 | * Allow pending signals to take effect. | |
1547 | */ | |
1548 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1549 | { | |
1550 | struct uprobe *uprobe; | |
1551 | ||
1552 | uprobe = utask->active_uprobe; | |
1553 | if (utask->state == UTASK_SSTEP_ACK) | |
1554 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1555 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1556 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1557 | else | |
1558 | WARN_ON_ONCE(1); | |
1559 | ||
1560 | put_uprobe(uprobe); | |
1561 | utask->active_uprobe = NULL; | |
1562 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1563 | xol_free_insn_slot(current); |
0326f5a9 SD |
1564 | |
1565 | spin_lock_irq(¤t->sighand->siglock); | |
1566 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1567 | spin_unlock_irq(¤t->sighand->siglock); | |
1568 | } | |
1569 | ||
1570 | /* | |
1b08e907 ON |
1571 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1572 | * allows the thread to return from interrupt. After that handle_swbp() | |
1573 | * sets utask->active_uprobe. | |
0326f5a9 | 1574 | * |
1b08e907 ON |
1575 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1576 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1577 | * |
1578 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1579 | * uprobe_notify_resume(). | |
1580 | */ | |
1581 | void uprobe_notify_resume(struct pt_regs *regs) | |
1582 | { | |
1583 | struct uprobe_task *utask; | |
1584 | ||
db023ea5 ON |
1585 | clear_thread_flag(TIF_UPROBE); |
1586 | ||
0326f5a9 | 1587 | utask = current->utask; |
1b08e907 | 1588 | if (utask && utask->active_uprobe) |
0326f5a9 | 1589 | handle_singlestep(utask, regs); |
1b08e907 ON |
1590 | else |
1591 | handle_swbp(regs); | |
0326f5a9 SD |
1592 | } |
1593 | ||
1594 | /* | |
1595 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1596 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1597 | */ | |
1598 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1599 | { | |
f8ac4ec9 | 1600 | if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) |
0326f5a9 SD |
1601 | return 0; |
1602 | ||
0326f5a9 | 1603 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1604 | return 1; |
1605 | } | |
1606 | ||
1607 | /* | |
1608 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1609 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1610 | */ | |
1611 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1612 | { | |
1613 | struct uprobe_task *utask = current->utask; | |
1614 | ||
1615 | if (!current->mm || !utask || !utask->active_uprobe) | |
1616 | /* task is currently not uprobed */ | |
1617 | return 0; | |
1618 | ||
1619 | utask->state = UTASK_SSTEP_ACK; | |
1620 | set_thread_flag(TIF_UPROBE); | |
1621 | return 1; | |
1622 | } | |
1623 | ||
1624 | static struct notifier_block uprobe_exception_nb = { | |
1625 | .notifier_call = arch_uprobe_exception_notify, | |
1626 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1627 | }; | |
1628 | ||
2b144498 SD |
1629 | static int __init init_uprobes(void) |
1630 | { | |
1631 | int i; | |
1632 | ||
66d06dff | 1633 | for (i = 0; i < UPROBES_HASH_SZ; i++) |
2b144498 | 1634 | mutex_init(&uprobes_mmap_mutex[i]); |
0326f5a9 | 1635 | |
32cdba1e ON |
1636 | if (percpu_init_rwsem(&dup_mmap_sem)) |
1637 | return -ENOMEM; | |
1638 | ||
0326f5a9 | 1639 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 1640 | } |
0326f5a9 | 1641 | module_init(init_uprobes); |
2b144498 SD |
1642 | |
1643 | static void __exit exit_uprobes(void) | |
1644 | { | |
1645 | } | |
2b144498 | 1646 | module_exit(exit_uprobes); |