]>
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> | |
e8440c14 | 30 | #include <linux/export.h> |
2b144498 SD |
31 | #include <linux/rmap.h> /* anon_vma_prepare */ |
32 | #include <linux/mmu_notifier.h> /* set_pte_at_notify */ | |
33 | #include <linux/swap.h> /* try_to_free_swap */ | |
0326f5a9 SD |
34 | #include <linux/ptrace.h> /* user_enable_single_step */ |
35 | #include <linux/kdebug.h> /* notifier mechanism */ | |
194f8dcb | 36 | #include "../../mm/internal.h" /* munlock_vma_page */ |
32cdba1e | 37 | #include <linux/percpu-rwsem.h> |
7b2d81d4 | 38 | |
2b144498 SD |
39 | #include <linux/uprobes.h> |
40 | ||
d4b3b638 SD |
41 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
42 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
43 | ||
2b144498 | 44 | static struct rb_root uprobes_tree = RB_ROOT; |
441f1eb7 ON |
45 | /* |
46 | * allows us to skip the uprobe_mmap if there are no uprobe events active | |
47 | * at this time. Probably a fine grained per inode count is better? | |
48 | */ | |
49 | #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree) | |
7b2d81d4 | 50 | |
2b144498 SD |
51 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
52 | ||
53 | #define UPROBES_HASH_SZ 13 | |
2b144498 SD |
54 | /* serialize uprobe->pending_list */ |
55 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 56 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 | 57 | |
32cdba1e ON |
58 | static struct percpu_rw_semaphore dup_mmap_sem; |
59 | ||
cb9a19fe | 60 | /* Have a copy of original instruction */ |
71434f2f | 61 | #define UPROBE_COPY_INSN 0 |
cb9a19fe | 62 | /* Can skip singlestep */ |
bb929284 | 63 | #define UPROBE_SKIP_SSTEP 1 |
cb9a19fe | 64 | |
3ff54efd SD |
65 | struct uprobe { |
66 | struct rb_node rb_node; /* node in the rb tree */ | |
67 | atomic_t ref; | |
e591c8d7 | 68 | struct rw_semaphore register_rwsem; |
3ff54efd SD |
69 | struct rw_semaphore consumer_rwsem; |
70 | struct list_head pending_list; | |
71 | struct uprobe_consumer *consumers; | |
72 | struct inode *inode; /* Also hold a ref to inode */ | |
73 | loff_t offset; | |
71434f2f | 74 | unsigned long flags; |
3ff54efd SD |
75 | struct arch_uprobe arch; |
76 | }; | |
77 | ||
2b144498 SD |
78 | /* |
79 | * valid_vma: Verify if the specified vma is an executable vma | |
80 | * Relax restrictions while unregistering: vm_flags might have | |
81 | * changed after breakpoint was inserted. | |
82 | * - is_register: indicates if we are in register context. | |
83 | * - Return 1 if the specified virtual address is in an | |
84 | * executable vma. | |
85 | */ | |
86 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
87 | { | |
e40cfce6 | 88 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED; |
2b144498 | 89 | |
e40cfce6 ON |
90 | if (is_register) |
91 | flags |= VM_WRITE; | |
2b144498 | 92 | |
e40cfce6 | 93 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
94 | } |
95 | ||
57683f72 | 96 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 97 | { |
57683f72 | 98 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
99 | } |
100 | ||
cb113b47 ON |
101 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
102 | { | |
103 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
104 | } | |
105 | ||
2b144498 SD |
106 | /** |
107 | * __replace_page - replace page in vma by new page. | |
108 | * based on replace_page in mm/ksm.c | |
109 | * | |
110 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 111 | * @addr: address the old @page is mapped at |
2b144498 SD |
112 | * @page: the cowed page we are replacing by kpage |
113 | * @kpage: the modified page we replace page by | |
114 | * | |
115 | * Returns 0 on success, -EFAULT on failure. | |
116 | */ | |
c517ee74 ON |
117 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
118 | struct page *page, struct page *kpage) | |
2b144498 SD |
119 | { |
120 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
121 | spinlock_t *ptl; |
122 | pte_t *ptep; | |
9f92448c | 123 | int err; |
6bdb913f HE |
124 | /* For mmu_notifiers */ |
125 | const unsigned long mmun_start = addr; | |
126 | const unsigned long mmun_end = addr + PAGE_SIZE; | |
2b144498 | 127 | |
194f8dcb | 128 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
129 | lock_page(page); |
130 | ||
6bdb913f | 131 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
9f92448c | 132 | err = -EAGAIN; |
5323ce71 | 133 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 134 | if (!ptep) |
9f92448c | 135 | goto unlock; |
2b144498 SD |
136 | |
137 | get_page(kpage); | |
138 | page_add_new_anon_rmap(kpage, vma, addr); | |
139 | ||
7396fa81 SD |
140 | if (!PageAnon(page)) { |
141 | dec_mm_counter(mm, MM_FILEPAGES); | |
142 | inc_mm_counter(mm, MM_ANONPAGES); | |
143 | } | |
144 | ||
2b144498 SD |
145 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
146 | ptep_clear_flush(vma, addr, ptep); | |
147 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
148 | ||
149 | page_remove_rmap(page); | |
150 | if (!page_mapped(page)) | |
151 | try_to_free_swap(page); | |
2b144498 | 152 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 153 | |
194f8dcb ON |
154 | if (vma->vm_flags & VM_LOCKED) |
155 | munlock_vma_page(page); | |
156 | put_page(page); | |
157 | ||
9f92448c ON |
158 | err = 0; |
159 | unlock: | |
6bdb913f | 160 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
9f92448c ON |
161 | unlock_page(page); |
162 | return err; | |
2b144498 SD |
163 | } |
164 | ||
165 | /** | |
5cb4ac3a | 166 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 167 | * @insn: instruction to be checked. |
5cb4ac3a | 168 | * Default implementation of is_swbp_insn |
2b144498 SD |
169 | * Returns true if @insn is a breakpoint instruction. |
170 | */ | |
5cb4ac3a | 171 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 172 | { |
5cb4ac3a | 173 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
174 | } |
175 | ||
cceb55aa ON |
176 | static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode) |
177 | { | |
178 | void *kaddr = kmap_atomic(page); | |
179 | memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE); | |
180 | kunmap_atomic(kaddr); | |
181 | } | |
182 | ||
ed6f6a50 ON |
183 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
184 | { | |
185 | uprobe_opcode_t old_opcode; | |
186 | bool is_swbp; | |
187 | ||
188 | copy_opcode(page, vaddr, &old_opcode); | |
189 | is_swbp = is_swbp_insn(&old_opcode); | |
190 | ||
191 | if (is_swbp_insn(new_opcode)) { | |
192 | if (is_swbp) /* register: already installed? */ | |
193 | return 0; | |
194 | } else { | |
195 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 196 | return 0; |
ed6f6a50 ON |
197 | } |
198 | ||
199 | return 1; | |
200 | } | |
201 | ||
2b144498 SD |
202 | /* |
203 | * NOTE: | |
204 | * Expect the breakpoint instruction to be the smallest size instruction for | |
205 | * the architecture. If an arch has variable length instruction and the | |
206 | * breakpoint instruction is not of the smallest length instruction | |
cceb55aa | 207 | * supported by that architecture then we need to modify is_swbp_at_addr and |
2b144498 SD |
208 | * write_opcode accordingly. This would never be a problem for archs that |
209 | * have fixed length instructions. | |
210 | */ | |
211 | ||
212 | /* | |
213 | * write_opcode - write the opcode at a given virtual address. | |
214 | * @mm: the probed process address space. | |
2b144498 SD |
215 | * @vaddr: the virtual address to store the opcode. |
216 | * @opcode: opcode to be written at @vaddr. | |
217 | * | |
218 | * Called with mm->mmap_sem held (for read and with a reference to | |
219 | * mm). | |
220 | * | |
221 | * For mm @mm, write the opcode at @vaddr. | |
222 | * Return 0 (success) or a negative errno. | |
223 | */ | |
cceb55aa ON |
224 | static int write_opcode(struct mm_struct *mm, unsigned long vaddr, |
225 | uprobe_opcode_t opcode) | |
2b144498 SD |
226 | { |
227 | struct page *old_page, *new_page; | |
2b144498 SD |
228 | void *vaddr_old, *vaddr_new; |
229 | struct vm_area_struct *vma; | |
2b144498 | 230 | int ret; |
f403072c | 231 | |
5323ce71 | 232 | retry: |
2b144498 | 233 | /* Read the page with vaddr into memory */ |
75ed82ea | 234 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
235 | if (ret <= 0) |
236 | return ret; | |
7b2d81d4 | 237 | |
ed6f6a50 ON |
238 | ret = verify_opcode(old_page, vaddr, &opcode); |
239 | if (ret <= 0) | |
240 | goto put_old; | |
241 | ||
2b144498 SD |
242 | ret = -ENOMEM; |
243 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
244 | if (!new_page) | |
9f92448c | 245 | goto put_old; |
2b144498 SD |
246 | |
247 | __SetPageUptodate(new_page); | |
248 | ||
2b144498 SD |
249 | /* copy the page now that we've got it stable */ |
250 | vaddr_old = kmap_atomic(old_page); | |
251 | vaddr_new = kmap_atomic(new_page); | |
252 | ||
253 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 254 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
255 | |
256 | kunmap_atomic(vaddr_new); | |
257 | kunmap_atomic(vaddr_old); | |
258 | ||
259 | ret = anon_vma_prepare(vma); | |
260 | if (ret) | |
9f92448c | 261 | goto put_new; |
2b144498 | 262 | |
c517ee74 | 263 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 264 | |
9f92448c | 265 | put_new: |
2b144498 | 266 | page_cache_release(new_page); |
9f92448c | 267 | put_old: |
7b2d81d4 IM |
268 | put_page(old_page); |
269 | ||
5323ce71 ON |
270 | if (unlikely(ret == -EAGAIN)) |
271 | goto retry; | |
2b144498 SD |
272 | return ret; |
273 | } | |
274 | ||
2b144498 | 275 | /** |
5cb4ac3a | 276 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 277 | * @auprobe: arch specific probepoint information. |
2b144498 | 278 | * @mm: the probed process address space. |
2b144498 SD |
279 | * @vaddr: the virtual address to insert the opcode. |
280 | * | |
281 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
282 | * Return 0 (success) or a negative errno. | |
283 | */ | |
5cb4ac3a | 284 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 285 | { |
cceb55aa | 286 | return write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
287 | } |
288 | ||
289 | /** | |
290 | * set_orig_insn - Restore the original instruction. | |
291 | * @mm: the probed process address space. | |
e3343e6a | 292 | * @auprobe: arch specific probepoint information. |
2b144498 | 293 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
294 | * |
295 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
296 | * Return 0 (success) or a negative errno. | |
297 | */ | |
7b2d81d4 | 298 | int __weak |
ded86e7c | 299 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 300 | { |
cceb55aa | 301 | return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
302 | } |
303 | ||
304 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
305 | { | |
306 | if (l->inode < r->inode) | |
307 | return -1; | |
7b2d81d4 | 308 | |
2b144498 SD |
309 | if (l->inode > r->inode) |
310 | return 1; | |
2b144498 | 311 | |
7b2d81d4 IM |
312 | if (l->offset < r->offset) |
313 | return -1; | |
314 | ||
315 | if (l->offset > r->offset) | |
316 | return 1; | |
2b144498 SD |
317 | |
318 | return 0; | |
319 | } | |
320 | ||
321 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
322 | { | |
323 | struct uprobe u = { .inode = inode, .offset = offset }; | |
324 | struct rb_node *n = uprobes_tree.rb_node; | |
325 | struct uprobe *uprobe; | |
326 | int match; | |
327 | ||
328 | while (n) { | |
329 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
330 | match = match_uprobe(&u, uprobe); | |
331 | if (!match) { | |
332 | atomic_inc(&uprobe->ref); | |
333 | return uprobe; | |
334 | } | |
7b2d81d4 | 335 | |
2b144498 SD |
336 | if (match < 0) |
337 | n = n->rb_left; | |
338 | else | |
339 | n = n->rb_right; | |
340 | } | |
341 | return NULL; | |
342 | } | |
343 | ||
344 | /* | |
345 | * Find a uprobe corresponding to a given inode:offset | |
346 | * Acquires uprobes_treelock | |
347 | */ | |
348 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
349 | { | |
350 | struct uprobe *uprobe; | |
2b144498 | 351 | |
6f47caa0 | 352 | spin_lock(&uprobes_treelock); |
2b144498 | 353 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 354 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 355 | |
2b144498 SD |
356 | return uprobe; |
357 | } | |
358 | ||
359 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
360 | { | |
361 | struct rb_node **p = &uprobes_tree.rb_node; | |
362 | struct rb_node *parent = NULL; | |
363 | struct uprobe *u; | |
364 | int match; | |
365 | ||
366 | while (*p) { | |
367 | parent = *p; | |
368 | u = rb_entry(parent, struct uprobe, rb_node); | |
369 | match = match_uprobe(uprobe, u); | |
370 | if (!match) { | |
371 | atomic_inc(&u->ref); | |
372 | return u; | |
373 | } | |
374 | ||
375 | if (match < 0) | |
376 | p = &parent->rb_left; | |
377 | else | |
378 | p = &parent->rb_right; | |
379 | ||
380 | } | |
7b2d81d4 | 381 | |
2b144498 SD |
382 | u = NULL; |
383 | rb_link_node(&uprobe->rb_node, parent, p); | |
384 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
385 | /* get access + creation ref */ | |
386 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 387 | |
2b144498 SD |
388 | return u; |
389 | } | |
390 | ||
391 | /* | |
7b2d81d4 | 392 | * Acquire uprobes_treelock. |
2b144498 SD |
393 | * Matching uprobe already exists in rbtree; |
394 | * increment (access refcount) and return the matching uprobe. | |
395 | * | |
396 | * No matching uprobe; insert the uprobe in rb_tree; | |
397 | * get a double refcount (access + creation) and return NULL. | |
398 | */ | |
399 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
400 | { | |
2b144498 SD |
401 | struct uprobe *u; |
402 | ||
6f47caa0 | 403 | spin_lock(&uprobes_treelock); |
2b144498 | 404 | u = __insert_uprobe(uprobe); |
6f47caa0 | 405 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 406 | |
2b144498 SD |
407 | return u; |
408 | } | |
409 | ||
410 | static void put_uprobe(struct uprobe *uprobe) | |
411 | { | |
412 | if (atomic_dec_and_test(&uprobe->ref)) | |
413 | kfree(uprobe); | |
414 | } | |
415 | ||
416 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
417 | { | |
418 | struct uprobe *uprobe, *cur_uprobe; | |
419 | ||
420 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
421 | if (!uprobe) | |
422 | return NULL; | |
423 | ||
424 | uprobe->inode = igrab(inode); | |
425 | uprobe->offset = offset; | |
e591c8d7 | 426 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 427 | init_rwsem(&uprobe->consumer_rwsem); |
bbc33d05 ON |
428 | /* For now assume that the instruction need not be single-stepped */ |
429 | __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); | |
2b144498 SD |
430 | |
431 | /* add to uprobes_tree, sorted on inode:offset */ | |
432 | cur_uprobe = insert_uprobe(uprobe); | |
433 | ||
434 | /* a uprobe exists for this inode:offset combination */ | |
435 | if (cur_uprobe) { | |
436 | kfree(uprobe); | |
437 | uprobe = cur_uprobe; | |
438 | iput(inode); | |
7b2d81d4 IM |
439 | } |
440 | ||
2b144498 SD |
441 | return uprobe; |
442 | } | |
443 | ||
9a98e03c | 444 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
445 | { |
446 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
447 | uc->next = uprobe->consumers; |
448 | uprobe->consumers = uc; | |
2b144498 | 449 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
450 | } |
451 | ||
452 | /* | |
e3343e6a SD |
453 | * For uprobe @uprobe, delete the consumer @uc. |
454 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
455 | * or return false. |
456 | */ | |
e3343e6a | 457 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
458 | { |
459 | struct uprobe_consumer **con; | |
460 | bool ret = false; | |
461 | ||
462 | down_write(&uprobe->consumer_rwsem); | |
463 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
464 | if (*con == uc) { |
465 | *con = uc->next; | |
2b144498 SD |
466 | ret = true; |
467 | break; | |
468 | } | |
469 | } | |
470 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 471 | |
2b144498 SD |
472 | return ret; |
473 | } | |
474 | ||
e3343e6a | 475 | static int |
d436615e | 476 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 477 | unsigned long nbytes, loff_t offset) |
2b144498 | 478 | { |
2b144498 SD |
479 | struct page *page; |
480 | void *vaddr; | |
593609a5 ON |
481 | unsigned long off; |
482 | pgoff_t idx; | |
2b144498 SD |
483 | |
484 | if (!filp) | |
485 | return -EINVAL; | |
486 | ||
cc359d18 ON |
487 | if (!mapping->a_ops->readpage) |
488 | return -EIO; | |
489 | ||
593609a5 ON |
490 | idx = offset >> PAGE_CACHE_SHIFT; |
491 | off = offset & ~PAGE_MASK; | |
2b144498 SD |
492 | |
493 | /* | |
494 | * Ensure that the page that has the original instruction is | |
495 | * populated and in page-cache. | |
496 | */ | |
497 | page = read_mapping_page(mapping, idx, filp); | |
498 | if (IS_ERR(page)) | |
499 | return PTR_ERR(page); | |
500 | ||
501 | vaddr = kmap_atomic(page); | |
593609a5 | 502 | memcpy(insn, vaddr + off, nbytes); |
2b144498 SD |
503 | kunmap_atomic(vaddr); |
504 | page_cache_release(page); | |
7b2d81d4 | 505 | |
2b144498 SD |
506 | return 0; |
507 | } | |
508 | ||
d436615e | 509 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
510 | { |
511 | struct address_space *mapping; | |
2b144498 | 512 | unsigned long nbytes; |
7b2d81d4 | 513 | int bytes; |
2b144498 | 514 | |
d436615e | 515 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
516 | mapping = uprobe->inode->i_mapping; |
517 | ||
518 | /* Instruction at end of binary; copy only available bytes */ | |
519 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
520 | bytes = uprobe->inode->i_size - uprobe->offset; | |
521 | else | |
522 | bytes = MAX_UINSN_BYTES; | |
523 | ||
524 | /* Instruction at the page-boundary; copy bytes in second page */ | |
525 | if (nbytes < bytes) { | |
fc36f595 ON |
526 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
527 | bytes - nbytes, uprobe->offset + nbytes); | |
528 | if (err) | |
529 | return err; | |
2b144498 SD |
530 | bytes = nbytes; |
531 | } | |
d436615e | 532 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
533 | } |
534 | ||
cb9a19fe ON |
535 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
536 | struct mm_struct *mm, unsigned long vaddr) | |
537 | { | |
538 | int ret = 0; | |
539 | ||
71434f2f | 540 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
541 | return ret; |
542 | ||
d4d3ccc6 ON |
543 | /* TODO: move this into _register, until then we abuse this sem. */ |
544 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 545 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
546 | goto out; |
547 | ||
cb9a19fe ON |
548 | ret = copy_insn(uprobe, file); |
549 | if (ret) | |
550 | goto out; | |
551 | ||
552 | ret = -ENOTSUPP; | |
553 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) | |
554 | goto out; | |
555 | ||
556 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
557 | if (ret) | |
558 | goto out; | |
559 | ||
560 | /* write_opcode() assumes we don't cross page boundary */ | |
561 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
562 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
563 | ||
564 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 565 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
566 | |
567 | out: | |
d4d3ccc6 | 568 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 569 | |
cb9a19fe ON |
570 | return ret; |
571 | } | |
572 | ||
8a7f2fa0 ON |
573 | static inline bool consumer_filter(struct uprobe_consumer *uc, |
574 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
806a98bd | 575 | { |
8a7f2fa0 | 576 | return !uc->filter || uc->filter(uc, ctx, mm); |
806a98bd ON |
577 | } |
578 | ||
8a7f2fa0 ON |
579 | static bool filter_chain(struct uprobe *uprobe, |
580 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
63633cbf | 581 | { |
1ff6fee5 ON |
582 | struct uprobe_consumer *uc; |
583 | bool ret = false; | |
584 | ||
585 | down_read(&uprobe->consumer_rwsem); | |
586 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
8a7f2fa0 | 587 | ret = consumer_filter(uc, ctx, mm); |
1ff6fee5 ON |
588 | if (ret) |
589 | break; | |
590 | } | |
591 | up_read(&uprobe->consumer_rwsem); | |
592 | ||
593 | return ret; | |
63633cbf ON |
594 | } |
595 | ||
e3343e6a SD |
596 | static int |
597 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 598 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 599 | { |
f8ac4ec9 | 600 | bool first_uprobe; |
2b144498 SD |
601 | int ret; |
602 | ||
cb9a19fe ON |
603 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
604 | if (ret) | |
605 | return ret; | |
682968e0 | 606 | |
f8ac4ec9 ON |
607 | /* |
608 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
609 | * the task can hit this breakpoint right after __replace_page(). | |
610 | */ | |
611 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
612 | if (first_uprobe) | |
613 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
614 | ||
816c03fb | 615 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
616 | if (!ret) |
617 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
618 | else if (first_uprobe) | |
f8ac4ec9 | 619 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
620 | |
621 | return ret; | |
622 | } | |
623 | ||
076a365b | 624 | static int |
816c03fb | 625 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 626 | { |
9f68f672 | 627 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 628 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
629 | } |
630 | ||
06b7bcd8 ON |
631 | static inline bool uprobe_is_active(struct uprobe *uprobe) |
632 | { | |
633 | return !RB_EMPTY_NODE(&uprobe->rb_node); | |
634 | } | |
0326f5a9 | 635 | /* |
778b032d ON |
636 | * There could be threads that have already hit the breakpoint. They |
637 | * will recheck the current insn and restart if find_uprobe() fails. | |
638 | * See find_active_uprobe(). | |
0326f5a9 | 639 | */ |
2b144498 SD |
640 | static void delete_uprobe(struct uprobe *uprobe) |
641 | { | |
06b7bcd8 ON |
642 | if (WARN_ON(!uprobe_is_active(uprobe))) |
643 | return; | |
644 | ||
6f47caa0 | 645 | spin_lock(&uprobes_treelock); |
2b144498 | 646 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 647 | spin_unlock(&uprobes_treelock); |
06b7bcd8 | 648 | RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */ |
2b144498 SD |
649 | iput(uprobe->inode); |
650 | put_uprobe(uprobe); | |
2b144498 SD |
651 | } |
652 | ||
26872090 ON |
653 | struct map_info { |
654 | struct map_info *next; | |
655 | struct mm_struct *mm; | |
816c03fb | 656 | unsigned long vaddr; |
26872090 ON |
657 | }; |
658 | ||
659 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 660 | { |
26872090 ON |
661 | struct map_info *next = info->next; |
662 | kfree(info); | |
663 | return next; | |
664 | } | |
665 | ||
666 | static struct map_info * | |
667 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
668 | { | |
669 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 670 | struct vm_area_struct *vma; |
26872090 ON |
671 | struct map_info *curr = NULL; |
672 | struct map_info *prev = NULL; | |
673 | struct map_info *info; | |
674 | int more = 0; | |
2b144498 | 675 | |
26872090 ON |
676 | again: |
677 | mutex_lock(&mapping->i_mmap_mutex); | |
6b2dbba8 | 678 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
679 | if (!valid_vma(vma, is_register)) |
680 | continue; | |
681 | ||
7a5bfb66 ON |
682 | if (!prev && !more) { |
683 | /* | |
684 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
685 | * reclaim. This is optimistic, no harm done if it fails. | |
686 | */ | |
687 | prev = kmalloc(sizeof(struct map_info), | |
688 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
689 | if (prev) | |
690 | prev->next = NULL; | |
691 | } | |
26872090 ON |
692 | if (!prev) { |
693 | more++; | |
694 | continue; | |
2b144498 | 695 | } |
2b144498 | 696 | |
26872090 ON |
697 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
698 | continue; | |
7b2d81d4 | 699 | |
26872090 ON |
700 | info = prev; |
701 | prev = prev->next; | |
702 | info->next = curr; | |
703 | curr = info; | |
2b144498 | 704 | |
26872090 | 705 | info->mm = vma->vm_mm; |
57683f72 | 706 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 707 | } |
2b144498 SD |
708 | mutex_unlock(&mapping->i_mmap_mutex); |
709 | ||
26872090 ON |
710 | if (!more) |
711 | goto out; | |
712 | ||
713 | prev = curr; | |
714 | while (curr) { | |
715 | mmput(curr->mm); | |
716 | curr = curr->next; | |
717 | } | |
7b2d81d4 | 718 | |
26872090 ON |
719 | do { |
720 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
721 | if (!info) { | |
722 | curr = ERR_PTR(-ENOMEM); | |
723 | goto out; | |
724 | } | |
725 | info->next = prev; | |
726 | prev = info; | |
727 | } while (--more); | |
728 | ||
729 | goto again; | |
730 | out: | |
731 | while (prev) | |
732 | prev = free_map_info(prev); | |
733 | return curr; | |
2b144498 SD |
734 | } |
735 | ||
bdf8647c ON |
736 | static int |
737 | register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) | |
2b144498 | 738 | { |
bdf8647c | 739 | bool is_register = !!new; |
26872090 ON |
740 | struct map_info *info; |
741 | int err = 0; | |
2b144498 | 742 | |
32cdba1e | 743 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
744 | info = build_map_info(uprobe->inode->i_mapping, |
745 | uprobe->offset, is_register); | |
32cdba1e ON |
746 | if (IS_ERR(info)) { |
747 | err = PTR_ERR(info); | |
748 | goto out; | |
749 | } | |
7b2d81d4 | 750 | |
26872090 ON |
751 | while (info) { |
752 | struct mm_struct *mm = info->mm; | |
753 | struct vm_area_struct *vma; | |
7b2d81d4 | 754 | |
076a365b | 755 | if (err && is_register) |
26872090 | 756 | goto free; |
7b2d81d4 | 757 | |
77fc4af1 | 758 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
759 | vma = find_vma(mm, info->vaddr); |
760 | if (!vma || !valid_vma(vma, is_register) || | |
761 | vma->vm_file->f_mapping->host != uprobe->inode) | |
26872090 ON |
762 | goto unlock; |
763 | ||
f4d6dfe5 ON |
764 | if (vma->vm_start > info->vaddr || |
765 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 766 | goto unlock; |
2b144498 | 767 | |
806a98bd ON |
768 | if (is_register) { |
769 | /* consult only the "caller", new consumer. */ | |
bdf8647c | 770 | if (consumer_filter(new, |
8a7f2fa0 | 771 | UPROBE_FILTER_REGISTER, mm)) |
806a98bd ON |
772 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
773 | } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) { | |
8a7f2fa0 ON |
774 | if (!filter_chain(uprobe, |
775 | UPROBE_FILTER_UNREGISTER, mm)) | |
806a98bd ON |
776 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
777 | } | |
78f74116 | 778 | |
26872090 ON |
779 | unlock: |
780 | up_write(&mm->mmap_sem); | |
781 | free: | |
782 | mmput(mm); | |
783 | info = free_map_info(info); | |
2b144498 | 784 | } |
32cdba1e ON |
785 | out: |
786 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 787 | return err; |
2b144498 SD |
788 | } |
789 | ||
9a98e03c | 790 | static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 791 | { |
9a98e03c | 792 | consumer_add(uprobe, uc); |
bdf8647c | 793 | return register_for_each_vma(uprobe, uc); |
2b144498 SD |
794 | } |
795 | ||
04aab9b2 | 796 | static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 797 | { |
04aab9b2 ON |
798 | int err; |
799 | ||
800 | if (!consumer_del(uprobe, uc)) /* WARN? */ | |
801 | return; | |
2b144498 | 802 | |
bdf8647c | 803 | err = register_for_each_vma(uprobe, NULL); |
bb929284 ON |
804 | /* TODO : cant unregister? schedule a worker thread */ |
805 | if (!uprobe->consumers && !err) | |
806 | delete_uprobe(uprobe); | |
2b144498 SD |
807 | } |
808 | ||
809 | /* | |
7b2d81d4 | 810 | * uprobe_register - register a probe |
2b144498 SD |
811 | * @inode: the file in which the probe has to be placed. |
812 | * @offset: offset from the start of the file. | |
e3343e6a | 813 | * @uc: information on howto handle the probe.. |
2b144498 | 814 | * |
7b2d81d4 | 815 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
816 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
817 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 818 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 819 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 820 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
821 | * unregisters. |
822 | * | |
823 | * Return errno if it cannot successully install probes | |
824 | * else return 0 (success) | |
825 | */ | |
e3343e6a | 826 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
827 | { |
828 | struct uprobe *uprobe; | |
7b2d81d4 | 829 | int ret; |
2b144498 | 830 | |
f0744af7 | 831 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 832 | if (offset > i_size_read(inode)) |
7b2d81d4 | 833 | return -EINVAL; |
2b144498 | 834 | |
66d06dff | 835 | retry: |
2b144498 | 836 | uprobe = alloc_uprobe(inode, offset); |
66d06dff ON |
837 | if (!uprobe) |
838 | return -ENOMEM; | |
839 | /* | |
840 | * We can race with uprobe_unregister()->delete_uprobe(). | |
841 | * Check uprobe_is_active() and retry if it is false. | |
842 | */ | |
843 | down_write(&uprobe->register_rwsem); | |
844 | ret = -EAGAIN; | |
845 | if (likely(uprobe_is_active(uprobe))) { | |
9a98e03c ON |
846 | ret = __uprobe_register(uprobe, uc); |
847 | if (ret) | |
04aab9b2 | 848 | __uprobe_unregister(uprobe, uc); |
2b144498 | 849 | } |
66d06dff ON |
850 | up_write(&uprobe->register_rwsem); |
851 | put_uprobe(uprobe); | |
2b144498 | 852 | |
66d06dff ON |
853 | if (unlikely(ret == -EAGAIN)) |
854 | goto retry; | |
2b144498 SD |
855 | return ret; |
856 | } | |
e8440c14 | 857 | EXPORT_SYMBOL_GPL(uprobe_register); |
2b144498 | 858 | |
bdf8647c ON |
859 | /* |
860 | * uprobe_apply - unregister a already registered probe. | |
861 | * @inode: the file in which the probe has to be removed. | |
862 | * @offset: offset from the start of the file. | |
863 | * @uc: consumer which wants to add more or remove some breakpoints | |
864 | * @add: add or remove the breakpoints | |
865 | */ | |
866 | int uprobe_apply(struct inode *inode, loff_t offset, | |
867 | struct uprobe_consumer *uc, bool add) | |
868 | { | |
869 | struct uprobe *uprobe; | |
870 | struct uprobe_consumer *con; | |
871 | int ret = -ENOENT; | |
872 | ||
873 | uprobe = find_uprobe(inode, offset); | |
874 | if (!uprobe) | |
875 | return ret; | |
876 | ||
877 | down_write(&uprobe->register_rwsem); | |
878 | for (con = uprobe->consumers; con && con != uc ; con = con->next) | |
879 | ; | |
880 | if (con) | |
881 | ret = register_for_each_vma(uprobe, add ? uc : NULL); | |
882 | up_write(&uprobe->register_rwsem); | |
883 | put_uprobe(uprobe); | |
884 | ||
885 | return ret; | |
886 | } | |
887 | ||
2b144498 | 888 | /* |
7b2d81d4 | 889 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
890 | * @inode: the file in which the probe has to be removed. |
891 | * @offset: offset from the start of the file. | |
e3343e6a | 892 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 893 | */ |
e3343e6a | 894 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 895 | { |
7b2d81d4 | 896 | struct uprobe *uprobe; |
2b144498 | 897 | |
2b144498 SD |
898 | uprobe = find_uprobe(inode, offset); |
899 | if (!uprobe) | |
900 | return; | |
901 | ||
e591c8d7 | 902 | down_write(&uprobe->register_rwsem); |
04aab9b2 | 903 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 904 | up_write(&uprobe->register_rwsem); |
c91368c4 | 905 | put_uprobe(uprobe); |
2b144498 | 906 | } |
e8440c14 | 907 | EXPORT_SYMBOL_GPL(uprobe_unregister); |
2b144498 | 908 | |
da1816b1 ON |
909 | static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm) |
910 | { | |
911 | struct vm_area_struct *vma; | |
912 | int err = 0; | |
913 | ||
914 | down_read(&mm->mmap_sem); | |
915 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
916 | unsigned long vaddr; | |
917 | loff_t offset; | |
918 | ||
919 | if (!valid_vma(vma, false) || | |
920 | vma->vm_file->f_mapping->host != uprobe->inode) | |
921 | continue; | |
922 | ||
923 | offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; | |
924 | if (uprobe->offset < offset || | |
925 | uprobe->offset >= offset + vma->vm_end - vma->vm_start) | |
926 | continue; | |
927 | ||
928 | vaddr = offset_to_vaddr(vma, uprobe->offset); | |
929 | err |= remove_breakpoint(uprobe, mm, vaddr); | |
930 | } | |
931 | up_read(&mm->mmap_sem); | |
932 | ||
933 | return err; | |
934 | } | |
935 | ||
891c3970 ON |
936 | static struct rb_node * |
937 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 938 | { |
2b144498 | 939 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
940 | |
941 | while (n) { | |
891c3970 | 942 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 943 | |
891c3970 | 944 | if (inode < u->inode) { |
2b144498 | 945 | n = n->rb_left; |
891c3970 | 946 | } else if (inode > u->inode) { |
2b144498 | 947 | n = n->rb_right; |
891c3970 ON |
948 | } else { |
949 | if (max < u->offset) | |
950 | n = n->rb_left; | |
951 | else if (min > u->offset) | |
952 | n = n->rb_right; | |
953 | else | |
954 | break; | |
955 | } | |
2b144498 | 956 | } |
7b2d81d4 | 957 | |
891c3970 | 958 | return n; |
2b144498 SD |
959 | } |
960 | ||
961 | /* | |
891c3970 | 962 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 963 | */ |
891c3970 ON |
964 | static void build_probe_list(struct inode *inode, |
965 | struct vm_area_struct *vma, | |
966 | unsigned long start, unsigned long end, | |
967 | struct list_head *head) | |
2b144498 | 968 | { |
891c3970 | 969 | loff_t min, max; |
891c3970 ON |
970 | struct rb_node *n, *t; |
971 | struct uprobe *u; | |
7b2d81d4 | 972 | |
891c3970 | 973 | INIT_LIST_HEAD(head); |
cb113b47 | 974 | min = vaddr_to_offset(vma, start); |
891c3970 | 975 | max = min + (end - start) - 1; |
2b144498 | 976 | |
6f47caa0 | 977 | spin_lock(&uprobes_treelock); |
891c3970 ON |
978 | n = find_node_in_range(inode, min, max); |
979 | if (n) { | |
980 | for (t = n; t; t = rb_prev(t)) { | |
981 | u = rb_entry(t, struct uprobe, rb_node); | |
982 | if (u->inode != inode || u->offset < min) | |
983 | break; | |
984 | list_add(&u->pending_list, head); | |
985 | atomic_inc(&u->ref); | |
986 | } | |
987 | for (t = n; (t = rb_next(t)); ) { | |
988 | u = rb_entry(t, struct uprobe, rb_node); | |
989 | if (u->inode != inode || u->offset > max) | |
990 | break; | |
991 | list_add(&u->pending_list, head); | |
992 | atomic_inc(&u->ref); | |
993 | } | |
2b144498 | 994 | } |
6f47caa0 | 995 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
996 | } |
997 | ||
998 | /* | |
5e5be71a | 999 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 1000 | * |
5e5be71a ON |
1001 | * Currently we ignore all errors and always return 0, the callers |
1002 | * can't handle the failure anyway. | |
2b144498 | 1003 | */ |
7b2d81d4 | 1004 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1005 | { |
1006 | struct list_head tmp_list; | |
665605a2 | 1007 | struct uprobe *uprobe, *u; |
2b144498 | 1008 | struct inode *inode; |
2b144498 | 1009 | |
441f1eb7 | 1010 | if (no_uprobe_events() || !valid_vma(vma, true)) |
7b2d81d4 | 1011 | return 0; |
2b144498 SD |
1012 | |
1013 | inode = vma->vm_file->f_mapping->host; | |
1014 | if (!inode) | |
7b2d81d4 | 1015 | return 0; |
2b144498 | 1016 | |
2b144498 | 1017 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1018 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
806a98bd ON |
1019 | /* |
1020 | * We can race with uprobe_unregister(), this uprobe can be already | |
1021 | * removed. But in this case filter_chain() must return false, all | |
1022 | * consumers have gone away. | |
1023 | */ | |
665605a2 | 1024 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
806a98bd | 1025 | if (!fatal_signal_pending(current) && |
8a7f2fa0 | 1026 | filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) { |
57683f72 | 1027 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1028 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1029 | } |
1030 | put_uprobe(uprobe); | |
1031 | } | |
2b144498 SD |
1032 | mutex_unlock(uprobes_mmap_hash(inode)); |
1033 | ||
5e5be71a | 1034 | return 0; |
2b144498 SD |
1035 | } |
1036 | ||
9f68f672 ON |
1037 | static bool |
1038 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1039 | { | |
1040 | loff_t min, max; | |
1041 | struct inode *inode; | |
1042 | struct rb_node *n; | |
1043 | ||
1044 | inode = vma->vm_file->f_mapping->host; | |
1045 | ||
1046 | min = vaddr_to_offset(vma, start); | |
1047 | max = min + (end - start) - 1; | |
1048 | ||
1049 | spin_lock(&uprobes_treelock); | |
1050 | n = find_node_in_range(inode, min, max); | |
1051 | spin_unlock(&uprobes_treelock); | |
1052 | ||
1053 | return !!n; | |
1054 | } | |
1055 | ||
682968e0 SD |
1056 | /* |
1057 | * Called in context of a munmap of a vma. | |
1058 | */ | |
cbc91f71 | 1059 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1060 | { |
441f1eb7 | 1061 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1062 | return; |
1063 | ||
2fd611a9 ON |
1064 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1065 | return; | |
1066 | ||
9f68f672 ON |
1067 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1068 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1069 | return; |
1070 | ||
9f68f672 ON |
1071 | if (vma_has_uprobes(vma, start, end)) |
1072 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1073 | } |
1074 | ||
d4b3b638 SD |
1075 | /* Slot allocation for XOL */ |
1076 | static int xol_add_vma(struct xol_area *area) | |
1077 | { | |
c8a82538 ON |
1078 | struct mm_struct *mm = current->mm; |
1079 | int ret = -EALREADY; | |
d4b3b638 SD |
1080 | |
1081 | down_write(&mm->mmap_sem); | |
1082 | if (mm->uprobes_state.xol_area) | |
1083 | goto fail; | |
1084 | ||
1085 | ret = -ENOMEM; | |
d4b3b638 SD |
1086 | /* Try to map as high as possible, this is only a hint. */ |
1087 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1088 | if (area->vaddr & ~PAGE_MASK) { | |
1089 | ret = area->vaddr; | |
1090 | goto fail; | |
1091 | } | |
1092 | ||
1093 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1094 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1095 | if (ret) | |
1096 | goto fail; | |
1097 | ||
1098 | smp_wmb(); /* pairs with get_xol_area() */ | |
1099 | mm->uprobes_state.xol_area = area; | |
1100 | ret = 0; | |
c8a82538 | 1101 | fail: |
d4b3b638 | 1102 | up_write(&mm->mmap_sem); |
d4b3b638 SD |
1103 | |
1104 | return ret; | |
1105 | } | |
1106 | ||
d4b3b638 | 1107 | /* |
9b545df8 ON |
1108 | * get_xol_area - Allocate process's xol_area if necessary. |
1109 | * This area will be used for storing instructions for execution out of line. | |
d4b3b638 SD |
1110 | * |
1111 | * Returns the allocated area or NULL. | |
1112 | */ | |
9b545df8 | 1113 | static struct xol_area *get_xol_area(void) |
d4b3b638 | 1114 | { |
9b545df8 | 1115 | struct mm_struct *mm = current->mm; |
d4b3b638 SD |
1116 | struct xol_area *area; |
1117 | ||
9b545df8 ON |
1118 | area = mm->uprobes_state.xol_area; |
1119 | if (area) | |
1120 | goto ret; | |
1121 | ||
d4b3b638 SD |
1122 | area = kzalloc(sizeof(*area), GFP_KERNEL); |
1123 | if (unlikely(!area)) | |
c8a82538 | 1124 | goto out; |
d4b3b638 SD |
1125 | |
1126 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
d4b3b638 | 1127 | if (!area->bitmap) |
c8a82538 ON |
1128 | goto free_area; |
1129 | ||
1130 | area->page = alloc_page(GFP_HIGHUSER); | |
1131 | if (!area->page) | |
1132 | goto free_bitmap; | |
d4b3b638 SD |
1133 | |
1134 | init_waitqueue_head(&area->wq); | |
1135 | if (!xol_add_vma(area)) | |
1136 | return area; | |
1137 | ||
c8a82538 ON |
1138 | __free_page(area->page); |
1139 | free_bitmap: | |
d4b3b638 | 1140 | kfree(area->bitmap); |
c8a82538 | 1141 | free_area: |
d4b3b638 | 1142 | kfree(area); |
c8a82538 | 1143 | out: |
9b545df8 ON |
1144 | area = mm->uprobes_state.xol_area; |
1145 | ret: | |
1146 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1147 | return area; | |
d4b3b638 SD |
1148 | } |
1149 | ||
1150 | /* | |
1151 | * uprobe_clear_state - Free the area allocated for slots. | |
1152 | */ | |
1153 | void uprobe_clear_state(struct mm_struct *mm) | |
1154 | { | |
1155 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1156 | ||
1157 | if (!area) | |
1158 | return; | |
1159 | ||
1160 | put_page(area->page); | |
1161 | kfree(area->bitmap); | |
1162 | kfree(area); | |
1163 | } | |
1164 | ||
32cdba1e ON |
1165 | void uprobe_start_dup_mmap(void) |
1166 | { | |
1167 | percpu_down_read(&dup_mmap_sem); | |
1168 | } | |
1169 | ||
1170 | void uprobe_end_dup_mmap(void) | |
1171 | { | |
1172 | percpu_up_read(&dup_mmap_sem); | |
1173 | } | |
1174 | ||
f8ac4ec9 ON |
1175 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1176 | { | |
61559a81 ON |
1177 | newmm->uprobes_state.xol_area = NULL; |
1178 | ||
9f68f672 | 1179 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1180 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1181 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1182 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1183 | } | |
f8ac4ec9 ON |
1184 | } |
1185 | ||
d4b3b638 SD |
1186 | /* |
1187 | * - search for a free slot. | |
1188 | */ | |
1189 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1190 | { | |
1191 | unsigned long slot_addr; | |
1192 | int slot_nr; | |
1193 | ||
1194 | do { | |
1195 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1196 | if (slot_nr < UINSNS_PER_PAGE) { | |
1197 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1198 | break; | |
1199 | ||
1200 | slot_nr = UINSNS_PER_PAGE; | |
1201 | continue; | |
1202 | } | |
1203 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1204 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1205 | ||
1206 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1207 | atomic_inc(&area->slot_count); | |
1208 | ||
1209 | return slot_addr; | |
1210 | } | |
1211 | ||
1212 | /* | |
a6cb3f6d | 1213 | * xol_get_insn_slot - allocate a slot for xol. |
d4b3b638 SD |
1214 | * Returns the allocated slot address or 0. |
1215 | */ | |
a6cb3f6d | 1216 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe) |
d4b3b638 SD |
1217 | { |
1218 | struct xol_area *area; | |
1219 | unsigned long offset; | |
a6cb3f6d | 1220 | unsigned long xol_vaddr; |
d4b3b638 SD |
1221 | void *vaddr; |
1222 | ||
9b545df8 ON |
1223 | area = get_xol_area(); |
1224 | if (!area) | |
1225 | return 0; | |
d4b3b638 | 1226 | |
a6cb3f6d ON |
1227 | xol_vaddr = xol_take_insn_slot(area); |
1228 | if (unlikely(!xol_vaddr)) | |
d4b3b638 SD |
1229 | return 0; |
1230 | ||
a6cb3f6d ON |
1231 | /* Initialize the slot */ |
1232 | offset = xol_vaddr & ~PAGE_MASK; | |
d4b3b638 SD |
1233 | vaddr = kmap_atomic(area->page); |
1234 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1235 | kunmap_atomic(vaddr); | |
65b6ecc0 RV |
1236 | /* |
1237 | * We probably need flush_icache_user_range() but it needs vma. | |
1238 | * This should work on supported architectures too. | |
1239 | */ | |
1240 | flush_dcache_page(area->page); | |
d4b3b638 | 1241 | |
a6cb3f6d | 1242 | return xol_vaddr; |
d4b3b638 SD |
1243 | } |
1244 | ||
1245 | /* | |
1246 | * xol_free_insn_slot - If slot was earlier allocated by | |
1247 | * @xol_get_insn_slot(), make the slot available for | |
1248 | * subsequent requests. | |
1249 | */ | |
1250 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1251 | { | |
1252 | struct xol_area *area; | |
1253 | unsigned long vma_end; | |
1254 | unsigned long slot_addr; | |
1255 | ||
1256 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1257 | return; | |
1258 | ||
1259 | slot_addr = tsk->utask->xol_vaddr; | |
af4355e9 | 1260 | if (unlikely(!slot_addr)) |
d4b3b638 SD |
1261 | return; |
1262 | ||
1263 | area = tsk->mm->uprobes_state.xol_area; | |
1264 | vma_end = area->vaddr + PAGE_SIZE; | |
1265 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1266 | unsigned long offset; | |
1267 | int slot_nr; | |
1268 | ||
1269 | offset = slot_addr - area->vaddr; | |
1270 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1271 | if (slot_nr >= UINSNS_PER_PAGE) | |
1272 | return; | |
1273 | ||
1274 | clear_bit(slot_nr, area->bitmap); | |
1275 | atomic_dec(&area->slot_count); | |
1276 | if (waitqueue_active(&area->wq)) | |
1277 | wake_up(&area->wq); | |
1278 | ||
1279 | tsk->utask->xol_vaddr = 0; | |
1280 | } | |
1281 | } | |
1282 | ||
0326f5a9 SD |
1283 | /** |
1284 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1285 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1286 | * instruction. | |
1287 | * Return the address of the breakpoint instruction. | |
1288 | */ | |
1289 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1290 | { | |
1291 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Called with no locks held. | |
1296 | * Called in context of a exiting or a exec-ing thread. | |
1297 | */ | |
1298 | void uprobe_free_utask(struct task_struct *t) | |
1299 | { | |
1300 | struct uprobe_task *utask = t->utask; | |
1301 | ||
0326f5a9 SD |
1302 | if (!utask) |
1303 | return; | |
1304 | ||
1305 | if (utask->active_uprobe) | |
1306 | put_uprobe(utask->active_uprobe); | |
1307 | ||
d4b3b638 | 1308 | xol_free_insn_slot(t); |
0326f5a9 SD |
1309 | kfree(utask); |
1310 | t->utask = NULL; | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * Called in context of a new clone/fork from copy_process. | |
1315 | */ | |
1316 | void uprobe_copy_process(struct task_struct *t) | |
1317 | { | |
1318 | t->utask = NULL; | |
0326f5a9 SD |
1319 | } |
1320 | ||
1321 | /* | |
5a2df662 ON |
1322 | * Allocate a uprobe_task object for the task if if necessary. |
1323 | * Called when the thread hits a breakpoint. | |
0326f5a9 SD |
1324 | * |
1325 | * Returns: | |
1326 | * - pointer to new uprobe_task on success | |
1327 | * - NULL otherwise | |
1328 | */ | |
5a2df662 | 1329 | static struct uprobe_task *get_utask(void) |
0326f5a9 | 1330 | { |
5a2df662 ON |
1331 | if (!current->utask) |
1332 | current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1333 | return current->utask; | |
0326f5a9 SD |
1334 | } |
1335 | ||
1336 | /* Prepare to single-step probed instruction out of line. */ | |
1337 | static int | |
a6cb3f6d | 1338 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr) |
0326f5a9 | 1339 | { |
a6cb3f6d ON |
1340 | struct uprobe_task *utask; |
1341 | unsigned long xol_vaddr; | |
aba51024 | 1342 | int err; |
a6cb3f6d | 1343 | |
608e7427 ON |
1344 | utask = get_utask(); |
1345 | if (!utask) | |
1346 | return -ENOMEM; | |
a6cb3f6d ON |
1347 | |
1348 | xol_vaddr = xol_get_insn_slot(uprobe); | |
1349 | if (!xol_vaddr) | |
1350 | return -ENOMEM; | |
1351 | ||
1352 | utask->xol_vaddr = xol_vaddr; | |
1353 | utask->vaddr = bp_vaddr; | |
d4b3b638 | 1354 | |
aba51024 ON |
1355 | err = arch_uprobe_pre_xol(&uprobe->arch, regs); |
1356 | if (unlikely(err)) { | |
1357 | xol_free_insn_slot(current); | |
1358 | return err; | |
1359 | } | |
1360 | ||
608e7427 ON |
1361 | utask->active_uprobe = uprobe; |
1362 | utask->state = UTASK_SSTEP; | |
aba51024 | 1363 | return 0; |
0326f5a9 SD |
1364 | } |
1365 | ||
1366 | /* | |
1367 | * If we are singlestepping, then ensure this thread is not connected to | |
1368 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1369 | * triggers the signal, restart the original insn even if the task is | |
1370 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1371 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1372 | * _same_ instruction should be repeated again after return from the signal | |
1373 | * handler, and SSTEP can never finish in this case. | |
1374 | */ | |
1375 | bool uprobe_deny_signal(void) | |
1376 | { | |
1377 | struct task_struct *t = current; | |
1378 | struct uprobe_task *utask = t->utask; | |
1379 | ||
1380 | if (likely(!utask || !utask->active_uprobe)) | |
1381 | return false; | |
1382 | ||
1383 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1384 | ||
1385 | if (signal_pending(t)) { | |
1386 | spin_lock_irq(&t->sighand->siglock); | |
1387 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1388 | spin_unlock_irq(&t->sighand->siglock); | |
1389 | ||
1390 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1391 | utask->state = UTASK_SSTEP_TRAPPED; | |
1392 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1393 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1394 | } | |
1395 | } | |
1396 | ||
1397 | return true; | |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * Avoid singlestepping the original instruction if the original instruction | |
1402 | * is a NOP or can be emulated. | |
1403 | */ | |
1404 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1405 | { | |
71434f2f | 1406 | if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) { |
0578a970 ON |
1407 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
1408 | return true; | |
71434f2f | 1409 | clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); |
0578a970 | 1410 | } |
0326f5a9 SD |
1411 | return false; |
1412 | } | |
1413 | ||
499a4f3e ON |
1414 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1415 | { | |
1416 | struct vm_area_struct *vma; | |
1417 | ||
1418 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1419 | if (!valid_vma(vma, false)) | |
1420 | continue; | |
1421 | /* | |
1422 | * This is not strictly accurate, we can race with | |
1423 | * uprobe_unregister() and see the already removed | |
1424 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 1425 | * Or this uprobe can be filtered out. |
499a4f3e ON |
1426 | */ |
1427 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1428 | return; | |
1429 | } | |
1430 | ||
1431 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1432 | } | |
1433 | ||
ec75fba9 ON |
1434 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
1435 | { | |
1436 | struct page *page; | |
1437 | uprobe_opcode_t opcode; | |
1438 | int result; | |
1439 | ||
1440 | pagefault_disable(); | |
1441 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1442 | sizeof(opcode)); | |
1443 | pagefault_enable(); | |
1444 | ||
1445 | if (likely(result == 0)) | |
1446 | goto out; | |
1447 | ||
1448 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1449 | if (result < 0) | |
1450 | return result; | |
1451 | ||
1452 | copy_opcode(page, vaddr, &opcode); | |
1453 | put_page(page); | |
1454 | out: | |
1455 | return is_swbp_insn(&opcode); | |
1456 | } | |
1457 | ||
d790d346 | 1458 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1459 | { |
3a9ea052 ON |
1460 | struct mm_struct *mm = current->mm; |
1461 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1462 | struct vm_area_struct *vma; |
0326f5a9 | 1463 | |
0326f5a9 SD |
1464 | down_read(&mm->mmap_sem); |
1465 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1466 | if (vma && vma->vm_start <= bp_vaddr) { |
1467 | if (valid_vma(vma, false)) { | |
cb113b47 ON |
1468 | struct inode *inode = vma->vm_file->f_mapping->host; |
1469 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); | |
0326f5a9 | 1470 | |
3a9ea052 ON |
1471 | uprobe = find_uprobe(inode, offset); |
1472 | } | |
d790d346 ON |
1473 | |
1474 | if (!uprobe) | |
1475 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1476 | } else { | |
1477 | *is_swbp = -EFAULT; | |
0326f5a9 | 1478 | } |
499a4f3e ON |
1479 | |
1480 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1481 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1482 | up_read(&mm->mmap_sem); |
1483 | ||
3a9ea052 ON |
1484 | return uprobe; |
1485 | } | |
1486 | ||
da1816b1 ON |
1487 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
1488 | { | |
1489 | struct uprobe_consumer *uc; | |
1490 | int remove = UPROBE_HANDLER_REMOVE; | |
1491 | ||
1492 | down_read(&uprobe->register_rwsem); | |
1493 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
1494 | int rc = uc->handler(uc, regs); | |
1495 | ||
1496 | WARN(rc & ~UPROBE_HANDLER_MASK, | |
1497 | "bad rc=0x%x from %pf()\n", rc, uc->handler); | |
1498 | remove &= rc; | |
1499 | } | |
1500 | ||
1501 | if (remove && uprobe->consumers) { | |
1502 | WARN_ON(!uprobe_is_active(uprobe)); | |
1503 | unapply_uprobe(uprobe, current->mm); | |
1504 | } | |
1505 | up_read(&uprobe->register_rwsem); | |
1506 | } | |
1507 | ||
3a9ea052 ON |
1508 | /* |
1509 | * Run handler and ask thread to singlestep. | |
1510 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1511 | */ | |
1512 | static void handle_swbp(struct pt_regs *regs) | |
1513 | { | |
3a9ea052 ON |
1514 | struct uprobe *uprobe; |
1515 | unsigned long bp_vaddr; | |
56bb4cf6 | 1516 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1517 | |
1518 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1519 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1520 | |
0326f5a9 | 1521 | if (!uprobe) { |
56bb4cf6 ON |
1522 | if (is_swbp > 0) { |
1523 | /* No matching uprobe; signal SIGTRAP. */ | |
1524 | send_sig(SIGTRAP, current, 0); | |
1525 | } else { | |
1526 | /* | |
1527 | * Either we raced with uprobe_unregister() or we can't | |
1528 | * access this memory. The latter is only possible if | |
1529 | * another thread plays with our ->mm. In both cases | |
1530 | * we can simply restart. If this vma was unmapped we | |
1531 | * can pretend this insn was not executed yet and get | |
1532 | * the (correct) SIGSEGV after restart. | |
1533 | */ | |
1534 | instruction_pointer_set(regs, bp_vaddr); | |
1535 | } | |
0326f5a9 SD |
1536 | return; |
1537 | } | |
74e59dfc ON |
1538 | |
1539 | /* change it in advance for ->handler() and restart */ | |
1540 | instruction_pointer_set(regs, bp_vaddr); | |
1541 | ||
142b18dd ON |
1542 | /* |
1543 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1544 | * After we hit the bp, _unregister + _register can install the | |
1545 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1546 | */ | |
1547 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1548 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
74e59dfc | 1549 | goto out; |
0326f5a9 | 1550 | |
0326f5a9 | 1551 | handler_chain(uprobe, regs); |
0578a970 ON |
1552 | if (can_skip_sstep(uprobe, regs)) |
1553 | goto out; | |
0326f5a9 | 1554 | |
608e7427 | 1555 | if (!pre_ssout(uprobe, regs, bp_vaddr)) |
0326f5a9 | 1556 | return; |
0326f5a9 | 1557 | |
74e59dfc | 1558 | /* can_skip_sstep() succeeded, or restart if can't singlestep */ |
0578a970 | 1559 | out: |
8bd87445 | 1560 | put_uprobe(uprobe); |
0326f5a9 SD |
1561 | } |
1562 | ||
1563 | /* | |
1564 | * Perform required fix-ups and disable singlestep. | |
1565 | * Allow pending signals to take effect. | |
1566 | */ | |
1567 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1568 | { | |
1569 | struct uprobe *uprobe; | |
1570 | ||
1571 | uprobe = utask->active_uprobe; | |
1572 | if (utask->state == UTASK_SSTEP_ACK) | |
1573 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1574 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1575 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1576 | else | |
1577 | WARN_ON_ONCE(1); | |
1578 | ||
1579 | put_uprobe(uprobe); | |
1580 | utask->active_uprobe = NULL; | |
1581 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1582 | xol_free_insn_slot(current); |
0326f5a9 SD |
1583 | |
1584 | spin_lock_irq(¤t->sighand->siglock); | |
1585 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1586 | spin_unlock_irq(¤t->sighand->siglock); | |
1587 | } | |
1588 | ||
1589 | /* | |
1b08e907 ON |
1590 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1591 | * allows the thread to return from interrupt. After that handle_swbp() | |
1592 | * sets utask->active_uprobe. | |
0326f5a9 | 1593 | * |
1b08e907 ON |
1594 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1595 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1596 | * |
1597 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1598 | * uprobe_notify_resume(). | |
1599 | */ | |
1600 | void uprobe_notify_resume(struct pt_regs *regs) | |
1601 | { | |
1602 | struct uprobe_task *utask; | |
1603 | ||
db023ea5 ON |
1604 | clear_thread_flag(TIF_UPROBE); |
1605 | ||
0326f5a9 | 1606 | utask = current->utask; |
1b08e907 | 1607 | if (utask && utask->active_uprobe) |
0326f5a9 | 1608 | handle_singlestep(utask, regs); |
1b08e907 ON |
1609 | else |
1610 | handle_swbp(regs); | |
0326f5a9 SD |
1611 | } |
1612 | ||
1613 | /* | |
1614 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1615 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1616 | */ | |
1617 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1618 | { | |
f8ac4ec9 | 1619 | if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) |
0326f5a9 SD |
1620 | return 0; |
1621 | ||
0326f5a9 | 1622 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1623 | return 1; |
1624 | } | |
1625 | ||
1626 | /* | |
1627 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1628 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1629 | */ | |
1630 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1631 | { | |
1632 | struct uprobe_task *utask = current->utask; | |
1633 | ||
1634 | if (!current->mm || !utask || !utask->active_uprobe) | |
1635 | /* task is currently not uprobed */ | |
1636 | return 0; | |
1637 | ||
1638 | utask->state = UTASK_SSTEP_ACK; | |
1639 | set_thread_flag(TIF_UPROBE); | |
1640 | return 1; | |
1641 | } | |
1642 | ||
1643 | static struct notifier_block uprobe_exception_nb = { | |
1644 | .notifier_call = arch_uprobe_exception_notify, | |
1645 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1646 | }; | |
1647 | ||
2b144498 SD |
1648 | static int __init init_uprobes(void) |
1649 | { | |
1650 | int i; | |
1651 | ||
66d06dff | 1652 | for (i = 0; i < UPROBES_HASH_SZ; i++) |
2b144498 | 1653 | mutex_init(&uprobes_mmap_mutex[i]); |
0326f5a9 | 1654 | |
32cdba1e ON |
1655 | if (percpu_init_rwsem(&dup_mmap_sem)) |
1656 | return -ENOMEM; | |
1657 | ||
0326f5a9 | 1658 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 1659 | } |
0326f5a9 | 1660 | module_init(init_uprobes); |
2b144498 SD |
1661 | |
1662 | static void __exit exit_uprobes(void) | |
1663 | { | |
1664 | } | |
2b144498 | 1665 | module_exit(exit_uprobes); |