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 | |
7b2d81d4 | 53 | |
c5784de2 PZ |
54 | /* |
55 | * We need separate register/unregister and mmap/munmap lock hashes because | |
56 | * of mmap_sem nesting. | |
57 | * | |
58 | * uprobe_register() needs to install probes on (potentially) all processes | |
59 | * and thus needs to acquire multiple mmap_sems (consequtively, not | |
60 | * concurrently), whereas uprobe_mmap() is called while holding mmap_sem | |
61 | * for the particular process doing the mmap. | |
62 | * | |
63 | * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem | |
64 | * because of lock order against i_mmap_mutex. This means there's a hole in | |
65 | * the register vma iteration where a mmap() can happen. | |
66 | * | |
67 | * Thus uprobe_register() can race with uprobe_mmap() and we can try and | |
68 | * install a probe where one is already installed. | |
69 | */ | |
70 | ||
2b144498 SD |
71 | /* serialize (un)register */ |
72 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
73 | |
74 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
75 | |
76 | /* serialize uprobe->pending_list */ | |
77 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 78 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 | 79 | |
32cdba1e ON |
80 | static struct percpu_rw_semaphore dup_mmap_sem; |
81 | ||
cb9a19fe | 82 | /* Have a copy of original instruction */ |
71434f2f | 83 | #define UPROBE_COPY_INSN 0 |
cb9a19fe | 84 | /* Can skip singlestep */ |
bb929284 | 85 | #define UPROBE_SKIP_SSTEP 1 |
cb9a19fe | 86 | |
3ff54efd SD |
87 | struct uprobe { |
88 | struct rb_node rb_node; /* node in the rb tree */ | |
89 | atomic_t ref; | |
e591c8d7 | 90 | struct rw_semaphore register_rwsem; |
3ff54efd SD |
91 | struct rw_semaphore consumer_rwsem; |
92 | struct list_head pending_list; | |
93 | struct uprobe_consumer *consumers; | |
94 | struct inode *inode; /* Also hold a ref to inode */ | |
95 | loff_t offset; | |
71434f2f | 96 | unsigned long flags; |
3ff54efd SD |
97 | struct arch_uprobe arch; |
98 | }; | |
99 | ||
2b144498 SD |
100 | /* |
101 | * valid_vma: Verify if the specified vma is an executable vma | |
102 | * Relax restrictions while unregistering: vm_flags might have | |
103 | * changed after breakpoint was inserted. | |
104 | * - is_register: indicates if we are in register context. | |
105 | * - Return 1 if the specified virtual address is in an | |
106 | * executable vma. | |
107 | */ | |
108 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
109 | { | |
e40cfce6 | 110 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED; |
2b144498 | 111 | |
e40cfce6 ON |
112 | if (is_register) |
113 | flags |= VM_WRITE; | |
2b144498 | 114 | |
e40cfce6 | 115 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
116 | } |
117 | ||
57683f72 | 118 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 119 | { |
57683f72 | 120 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
121 | } |
122 | ||
cb113b47 ON |
123 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
124 | { | |
125 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
126 | } | |
127 | ||
2b144498 SD |
128 | /** |
129 | * __replace_page - replace page in vma by new page. | |
130 | * based on replace_page in mm/ksm.c | |
131 | * | |
132 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 133 | * @addr: address the old @page is mapped at |
2b144498 SD |
134 | * @page: the cowed page we are replacing by kpage |
135 | * @kpage: the modified page we replace page by | |
136 | * | |
137 | * Returns 0 on success, -EFAULT on failure. | |
138 | */ | |
c517ee74 ON |
139 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
140 | struct page *page, struct page *kpage) | |
2b144498 SD |
141 | { |
142 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
143 | spinlock_t *ptl; |
144 | pte_t *ptep; | |
9f92448c | 145 | int err; |
6bdb913f HE |
146 | /* For mmu_notifiers */ |
147 | const unsigned long mmun_start = addr; | |
148 | const unsigned long mmun_end = addr + PAGE_SIZE; | |
2b144498 | 149 | |
194f8dcb | 150 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
151 | lock_page(page); |
152 | ||
6bdb913f | 153 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
9f92448c | 154 | err = -EAGAIN; |
5323ce71 | 155 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 156 | if (!ptep) |
9f92448c | 157 | goto unlock; |
2b144498 SD |
158 | |
159 | get_page(kpage); | |
160 | page_add_new_anon_rmap(kpage, vma, addr); | |
161 | ||
7396fa81 SD |
162 | if (!PageAnon(page)) { |
163 | dec_mm_counter(mm, MM_FILEPAGES); | |
164 | inc_mm_counter(mm, MM_ANONPAGES); | |
165 | } | |
166 | ||
2b144498 SD |
167 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
168 | ptep_clear_flush(vma, addr, ptep); | |
169 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
170 | ||
171 | page_remove_rmap(page); | |
172 | if (!page_mapped(page)) | |
173 | try_to_free_swap(page); | |
2b144498 | 174 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 175 | |
194f8dcb ON |
176 | if (vma->vm_flags & VM_LOCKED) |
177 | munlock_vma_page(page); | |
178 | put_page(page); | |
179 | ||
9f92448c ON |
180 | err = 0; |
181 | unlock: | |
6bdb913f | 182 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
9f92448c ON |
183 | unlock_page(page); |
184 | return err; | |
2b144498 SD |
185 | } |
186 | ||
187 | /** | |
5cb4ac3a | 188 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 189 | * @insn: instruction to be checked. |
5cb4ac3a | 190 | * Default implementation of is_swbp_insn |
2b144498 SD |
191 | * Returns true if @insn is a breakpoint instruction. |
192 | */ | |
5cb4ac3a | 193 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 194 | { |
5cb4ac3a | 195 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
196 | } |
197 | ||
cceb55aa ON |
198 | static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode) |
199 | { | |
200 | void *kaddr = kmap_atomic(page); | |
201 | memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE); | |
202 | kunmap_atomic(kaddr); | |
203 | } | |
204 | ||
ed6f6a50 ON |
205 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
206 | { | |
207 | uprobe_opcode_t old_opcode; | |
208 | bool is_swbp; | |
209 | ||
210 | copy_opcode(page, vaddr, &old_opcode); | |
211 | is_swbp = is_swbp_insn(&old_opcode); | |
212 | ||
213 | if (is_swbp_insn(new_opcode)) { | |
214 | if (is_swbp) /* register: already installed? */ | |
215 | return 0; | |
216 | } else { | |
217 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 218 | return 0; |
ed6f6a50 ON |
219 | } |
220 | ||
221 | return 1; | |
222 | } | |
223 | ||
2b144498 SD |
224 | /* |
225 | * NOTE: | |
226 | * Expect the breakpoint instruction to be the smallest size instruction for | |
227 | * the architecture. If an arch has variable length instruction and the | |
228 | * breakpoint instruction is not of the smallest length instruction | |
cceb55aa | 229 | * supported by that architecture then we need to modify is_swbp_at_addr and |
2b144498 SD |
230 | * write_opcode accordingly. This would never be a problem for archs that |
231 | * have fixed length instructions. | |
232 | */ | |
233 | ||
234 | /* | |
235 | * write_opcode - write the opcode at a given virtual address. | |
236 | * @mm: the probed process address space. | |
2b144498 SD |
237 | * @vaddr: the virtual address to store the opcode. |
238 | * @opcode: opcode to be written at @vaddr. | |
239 | * | |
240 | * Called with mm->mmap_sem held (for read and with a reference to | |
241 | * mm). | |
242 | * | |
243 | * For mm @mm, write the opcode at @vaddr. | |
244 | * Return 0 (success) or a negative errno. | |
245 | */ | |
cceb55aa ON |
246 | static int write_opcode(struct mm_struct *mm, unsigned long vaddr, |
247 | uprobe_opcode_t opcode) | |
2b144498 SD |
248 | { |
249 | struct page *old_page, *new_page; | |
2b144498 SD |
250 | void *vaddr_old, *vaddr_new; |
251 | struct vm_area_struct *vma; | |
2b144498 | 252 | int ret; |
f403072c | 253 | |
5323ce71 | 254 | retry: |
2b144498 | 255 | /* Read the page with vaddr into memory */ |
75ed82ea | 256 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
257 | if (ret <= 0) |
258 | return ret; | |
7b2d81d4 | 259 | |
ed6f6a50 ON |
260 | ret = verify_opcode(old_page, vaddr, &opcode); |
261 | if (ret <= 0) | |
262 | goto put_old; | |
263 | ||
2b144498 SD |
264 | ret = -ENOMEM; |
265 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
266 | if (!new_page) | |
9f92448c | 267 | goto put_old; |
2b144498 SD |
268 | |
269 | __SetPageUptodate(new_page); | |
270 | ||
2b144498 SD |
271 | /* copy the page now that we've got it stable */ |
272 | vaddr_old = kmap_atomic(old_page); | |
273 | vaddr_new = kmap_atomic(new_page); | |
274 | ||
275 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 276 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
277 | |
278 | kunmap_atomic(vaddr_new); | |
279 | kunmap_atomic(vaddr_old); | |
280 | ||
281 | ret = anon_vma_prepare(vma); | |
282 | if (ret) | |
9f92448c | 283 | goto put_new; |
2b144498 | 284 | |
c517ee74 | 285 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 286 | |
9f92448c | 287 | put_new: |
2b144498 | 288 | page_cache_release(new_page); |
9f92448c | 289 | put_old: |
7b2d81d4 IM |
290 | put_page(old_page); |
291 | ||
5323ce71 ON |
292 | if (unlikely(ret == -EAGAIN)) |
293 | goto retry; | |
2b144498 SD |
294 | return ret; |
295 | } | |
296 | ||
2b144498 | 297 | /** |
5cb4ac3a | 298 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 299 | * @auprobe: arch specific probepoint information. |
2b144498 | 300 | * @mm: the probed process address space. |
2b144498 SD |
301 | * @vaddr: the virtual address to insert the opcode. |
302 | * | |
303 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
304 | * Return 0 (success) or a negative errno. | |
305 | */ | |
5cb4ac3a | 306 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 307 | { |
cceb55aa | 308 | return write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
309 | } |
310 | ||
311 | /** | |
312 | * set_orig_insn - Restore the original instruction. | |
313 | * @mm: the probed process address space. | |
e3343e6a | 314 | * @auprobe: arch specific probepoint information. |
2b144498 | 315 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
316 | * |
317 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
318 | * Return 0 (success) or a negative errno. | |
319 | */ | |
7b2d81d4 | 320 | int __weak |
ded86e7c | 321 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 322 | { |
cceb55aa | 323 | return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
324 | } |
325 | ||
326 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
327 | { | |
328 | if (l->inode < r->inode) | |
329 | return -1; | |
7b2d81d4 | 330 | |
2b144498 SD |
331 | if (l->inode > r->inode) |
332 | return 1; | |
2b144498 | 333 | |
7b2d81d4 IM |
334 | if (l->offset < r->offset) |
335 | return -1; | |
336 | ||
337 | if (l->offset > r->offset) | |
338 | return 1; | |
2b144498 SD |
339 | |
340 | return 0; | |
341 | } | |
342 | ||
343 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
344 | { | |
345 | struct uprobe u = { .inode = inode, .offset = offset }; | |
346 | struct rb_node *n = uprobes_tree.rb_node; | |
347 | struct uprobe *uprobe; | |
348 | int match; | |
349 | ||
350 | while (n) { | |
351 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
352 | match = match_uprobe(&u, uprobe); | |
353 | if (!match) { | |
354 | atomic_inc(&uprobe->ref); | |
355 | return uprobe; | |
356 | } | |
7b2d81d4 | 357 | |
2b144498 SD |
358 | if (match < 0) |
359 | n = n->rb_left; | |
360 | else | |
361 | n = n->rb_right; | |
362 | } | |
363 | return NULL; | |
364 | } | |
365 | ||
366 | /* | |
367 | * Find a uprobe corresponding to a given inode:offset | |
368 | * Acquires uprobes_treelock | |
369 | */ | |
370 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
371 | { | |
372 | struct uprobe *uprobe; | |
2b144498 | 373 | |
6f47caa0 | 374 | spin_lock(&uprobes_treelock); |
2b144498 | 375 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 376 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 377 | |
2b144498 SD |
378 | return uprobe; |
379 | } | |
380 | ||
381 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
382 | { | |
383 | struct rb_node **p = &uprobes_tree.rb_node; | |
384 | struct rb_node *parent = NULL; | |
385 | struct uprobe *u; | |
386 | int match; | |
387 | ||
388 | while (*p) { | |
389 | parent = *p; | |
390 | u = rb_entry(parent, struct uprobe, rb_node); | |
391 | match = match_uprobe(uprobe, u); | |
392 | if (!match) { | |
393 | atomic_inc(&u->ref); | |
394 | return u; | |
395 | } | |
396 | ||
397 | if (match < 0) | |
398 | p = &parent->rb_left; | |
399 | else | |
400 | p = &parent->rb_right; | |
401 | ||
402 | } | |
7b2d81d4 | 403 | |
2b144498 SD |
404 | u = NULL; |
405 | rb_link_node(&uprobe->rb_node, parent, p); | |
406 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
407 | /* get access + creation ref */ | |
408 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 409 | |
2b144498 SD |
410 | return u; |
411 | } | |
412 | ||
413 | /* | |
7b2d81d4 | 414 | * Acquire uprobes_treelock. |
2b144498 SD |
415 | * Matching uprobe already exists in rbtree; |
416 | * increment (access refcount) and return the matching uprobe. | |
417 | * | |
418 | * No matching uprobe; insert the uprobe in rb_tree; | |
419 | * get a double refcount (access + creation) and return NULL. | |
420 | */ | |
421 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
422 | { | |
2b144498 SD |
423 | struct uprobe *u; |
424 | ||
6f47caa0 | 425 | spin_lock(&uprobes_treelock); |
2b144498 | 426 | u = __insert_uprobe(uprobe); |
6f47caa0 | 427 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 428 | |
2b144498 SD |
429 | return u; |
430 | } | |
431 | ||
432 | static void put_uprobe(struct uprobe *uprobe) | |
433 | { | |
434 | if (atomic_dec_and_test(&uprobe->ref)) | |
435 | kfree(uprobe); | |
436 | } | |
437 | ||
438 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
439 | { | |
440 | struct uprobe *uprobe, *cur_uprobe; | |
441 | ||
442 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
443 | if (!uprobe) | |
444 | return NULL; | |
445 | ||
446 | uprobe->inode = igrab(inode); | |
447 | uprobe->offset = offset; | |
e591c8d7 | 448 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 449 | init_rwsem(&uprobe->consumer_rwsem); |
bbc33d05 ON |
450 | /* For now assume that the instruction need not be single-stepped */ |
451 | __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); | |
2b144498 SD |
452 | |
453 | /* add to uprobes_tree, sorted on inode:offset */ | |
454 | cur_uprobe = insert_uprobe(uprobe); | |
455 | ||
456 | /* a uprobe exists for this inode:offset combination */ | |
457 | if (cur_uprobe) { | |
458 | kfree(uprobe); | |
459 | uprobe = cur_uprobe; | |
460 | iput(inode); | |
7b2d81d4 IM |
461 | } |
462 | ||
2b144498 SD |
463 | return uprobe; |
464 | } | |
465 | ||
0326f5a9 SD |
466 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
467 | { | |
468 | struct uprobe_consumer *uc; | |
469 | ||
e591c8d7 | 470 | down_read(&uprobe->register_rwsem); |
fe20d71f ON |
471 | for (uc = uprobe->consumers; uc; uc = uc->next) |
472 | uc->handler(uc, regs); | |
e591c8d7 | 473 | up_read(&uprobe->register_rwsem); |
0326f5a9 SD |
474 | } |
475 | ||
9a98e03c | 476 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
477 | { |
478 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
479 | uc->next = uprobe->consumers; |
480 | uprobe->consumers = uc; | |
2b144498 | 481 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
482 | } |
483 | ||
484 | /* | |
e3343e6a SD |
485 | * For uprobe @uprobe, delete the consumer @uc. |
486 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
487 | * or return false. |
488 | */ | |
e3343e6a | 489 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
490 | { |
491 | struct uprobe_consumer **con; | |
492 | bool ret = false; | |
493 | ||
494 | down_write(&uprobe->consumer_rwsem); | |
495 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
496 | if (*con == uc) { |
497 | *con = uc->next; | |
2b144498 SD |
498 | ret = true; |
499 | break; | |
500 | } | |
501 | } | |
502 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 503 | |
2b144498 SD |
504 | return ret; |
505 | } | |
506 | ||
e3343e6a | 507 | static int |
d436615e | 508 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 509 | unsigned long nbytes, loff_t offset) |
2b144498 | 510 | { |
2b144498 SD |
511 | struct page *page; |
512 | void *vaddr; | |
593609a5 ON |
513 | unsigned long off; |
514 | pgoff_t idx; | |
2b144498 SD |
515 | |
516 | if (!filp) | |
517 | return -EINVAL; | |
518 | ||
cc359d18 ON |
519 | if (!mapping->a_ops->readpage) |
520 | return -EIO; | |
521 | ||
593609a5 ON |
522 | idx = offset >> PAGE_CACHE_SHIFT; |
523 | off = offset & ~PAGE_MASK; | |
2b144498 SD |
524 | |
525 | /* | |
526 | * Ensure that the page that has the original instruction is | |
527 | * populated and in page-cache. | |
528 | */ | |
529 | page = read_mapping_page(mapping, idx, filp); | |
530 | if (IS_ERR(page)) | |
531 | return PTR_ERR(page); | |
532 | ||
533 | vaddr = kmap_atomic(page); | |
593609a5 | 534 | memcpy(insn, vaddr + off, nbytes); |
2b144498 SD |
535 | kunmap_atomic(vaddr); |
536 | page_cache_release(page); | |
7b2d81d4 | 537 | |
2b144498 SD |
538 | return 0; |
539 | } | |
540 | ||
d436615e | 541 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
542 | { |
543 | struct address_space *mapping; | |
2b144498 | 544 | unsigned long nbytes; |
7b2d81d4 | 545 | int bytes; |
2b144498 | 546 | |
d436615e | 547 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
548 | mapping = uprobe->inode->i_mapping; |
549 | ||
550 | /* Instruction at end of binary; copy only available bytes */ | |
551 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
552 | bytes = uprobe->inode->i_size - uprobe->offset; | |
553 | else | |
554 | bytes = MAX_UINSN_BYTES; | |
555 | ||
556 | /* Instruction at the page-boundary; copy bytes in second page */ | |
557 | if (nbytes < bytes) { | |
fc36f595 ON |
558 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
559 | bytes - nbytes, uprobe->offset + nbytes); | |
560 | if (err) | |
561 | return err; | |
2b144498 SD |
562 | bytes = nbytes; |
563 | } | |
d436615e | 564 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
565 | } |
566 | ||
cb9a19fe ON |
567 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
568 | struct mm_struct *mm, unsigned long vaddr) | |
569 | { | |
570 | int ret = 0; | |
571 | ||
71434f2f | 572 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
573 | return ret; |
574 | ||
d4d3ccc6 ON |
575 | /* TODO: move this into _register, until then we abuse this sem. */ |
576 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 577 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
578 | goto out; |
579 | ||
cb9a19fe ON |
580 | ret = copy_insn(uprobe, file); |
581 | if (ret) | |
582 | goto out; | |
583 | ||
584 | ret = -ENOTSUPP; | |
585 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) | |
586 | goto out; | |
587 | ||
588 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
589 | if (ret) | |
590 | goto out; | |
591 | ||
592 | /* write_opcode() assumes we don't cross page boundary */ | |
593 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
594 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
595 | ||
596 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 597 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
598 | |
599 | out: | |
d4d3ccc6 | 600 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 601 | |
cb9a19fe ON |
602 | return ret; |
603 | } | |
604 | ||
63633cbf ON |
605 | static bool filter_chain(struct uprobe *uprobe) |
606 | { | |
1ff6fee5 ON |
607 | struct uprobe_consumer *uc; |
608 | bool ret = false; | |
609 | ||
610 | down_read(&uprobe->consumer_rwsem); | |
611 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
612 | /* TODO: ret = uc->filter(...) */ | |
613 | ret = true; | |
614 | if (ret) | |
615 | break; | |
616 | } | |
617 | up_read(&uprobe->consumer_rwsem); | |
618 | ||
619 | return ret; | |
63633cbf ON |
620 | } |
621 | ||
e3343e6a SD |
622 | static int |
623 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 624 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 625 | { |
f8ac4ec9 | 626 | bool first_uprobe; |
2b144498 SD |
627 | int ret; |
628 | ||
629 | /* | |
630 | * If probe is being deleted, unregister thread could be done with | |
631 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
63633cbf ON |
632 | * nobody will be able to cleanup. But in this case filter_chain() |
633 | * must return false, all consumers have gone away. | |
2b144498 | 634 | */ |
63633cbf | 635 | if (!filter_chain(uprobe)) |
78f74116 | 636 | return 0; |
2b144498 | 637 | |
cb9a19fe ON |
638 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
639 | if (ret) | |
640 | return ret; | |
682968e0 | 641 | |
f8ac4ec9 ON |
642 | /* |
643 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
644 | * the task can hit this breakpoint right after __replace_page(). | |
645 | */ | |
646 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
647 | if (first_uprobe) | |
648 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
649 | ||
816c03fb | 650 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
651 | if (!ret) |
652 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
653 | else if (first_uprobe) | |
f8ac4ec9 | 654 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
655 | |
656 | return ret; | |
657 | } | |
658 | ||
076a365b | 659 | static int |
816c03fb | 660 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 661 | { |
9f68f672 | 662 | if (!test_bit(MMF_HAS_UPROBES, &mm->flags)) |
076a365b | 663 | return 0; |
9f68f672 | 664 | |
63633cbf ON |
665 | if (filter_chain(uprobe)) |
666 | return 0; | |
667 | ||
9f68f672 | 668 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 669 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
670 | } |
671 | ||
0326f5a9 | 672 | /* |
778b032d ON |
673 | * There could be threads that have already hit the breakpoint. They |
674 | * will recheck the current insn and restart if find_uprobe() fails. | |
675 | * See find_active_uprobe(). | |
0326f5a9 | 676 | */ |
2b144498 SD |
677 | static void delete_uprobe(struct uprobe *uprobe) |
678 | { | |
6f47caa0 | 679 | spin_lock(&uprobes_treelock); |
2b144498 | 680 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 681 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
682 | iput(uprobe->inode); |
683 | put_uprobe(uprobe); | |
2b144498 SD |
684 | } |
685 | ||
26872090 ON |
686 | struct map_info { |
687 | struct map_info *next; | |
688 | struct mm_struct *mm; | |
816c03fb | 689 | unsigned long vaddr; |
26872090 ON |
690 | }; |
691 | ||
692 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 693 | { |
26872090 ON |
694 | struct map_info *next = info->next; |
695 | kfree(info); | |
696 | return next; | |
697 | } | |
698 | ||
699 | static struct map_info * | |
700 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
701 | { | |
702 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 703 | struct vm_area_struct *vma; |
26872090 ON |
704 | struct map_info *curr = NULL; |
705 | struct map_info *prev = NULL; | |
706 | struct map_info *info; | |
707 | int more = 0; | |
2b144498 | 708 | |
26872090 ON |
709 | again: |
710 | mutex_lock(&mapping->i_mmap_mutex); | |
6b2dbba8 | 711 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
712 | if (!valid_vma(vma, is_register)) |
713 | continue; | |
714 | ||
7a5bfb66 ON |
715 | if (!prev && !more) { |
716 | /* | |
717 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
718 | * reclaim. This is optimistic, no harm done if it fails. | |
719 | */ | |
720 | prev = kmalloc(sizeof(struct map_info), | |
721 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
722 | if (prev) | |
723 | prev->next = NULL; | |
724 | } | |
26872090 ON |
725 | if (!prev) { |
726 | more++; | |
727 | continue; | |
2b144498 | 728 | } |
2b144498 | 729 | |
26872090 ON |
730 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
731 | continue; | |
7b2d81d4 | 732 | |
26872090 ON |
733 | info = prev; |
734 | prev = prev->next; | |
735 | info->next = curr; | |
736 | curr = info; | |
2b144498 | 737 | |
26872090 | 738 | info->mm = vma->vm_mm; |
57683f72 | 739 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 740 | } |
2b144498 SD |
741 | mutex_unlock(&mapping->i_mmap_mutex); |
742 | ||
26872090 ON |
743 | if (!more) |
744 | goto out; | |
745 | ||
746 | prev = curr; | |
747 | while (curr) { | |
748 | mmput(curr->mm); | |
749 | curr = curr->next; | |
750 | } | |
7b2d81d4 | 751 | |
26872090 ON |
752 | do { |
753 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
754 | if (!info) { | |
755 | curr = ERR_PTR(-ENOMEM); | |
756 | goto out; | |
757 | } | |
758 | info->next = prev; | |
759 | prev = info; | |
760 | } while (--more); | |
761 | ||
762 | goto again; | |
763 | out: | |
764 | while (prev) | |
765 | prev = free_map_info(prev); | |
766 | return curr; | |
2b144498 SD |
767 | } |
768 | ||
769 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
770 | { | |
26872090 ON |
771 | struct map_info *info; |
772 | int err = 0; | |
2b144498 | 773 | |
32cdba1e | 774 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
775 | info = build_map_info(uprobe->inode->i_mapping, |
776 | uprobe->offset, is_register); | |
32cdba1e ON |
777 | if (IS_ERR(info)) { |
778 | err = PTR_ERR(info); | |
779 | goto out; | |
780 | } | |
7b2d81d4 | 781 | |
26872090 ON |
782 | while (info) { |
783 | struct mm_struct *mm = info->mm; | |
784 | struct vm_area_struct *vma; | |
7b2d81d4 | 785 | |
076a365b | 786 | if (err && is_register) |
26872090 | 787 | goto free; |
7b2d81d4 | 788 | |
77fc4af1 | 789 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
790 | vma = find_vma(mm, info->vaddr); |
791 | if (!vma || !valid_vma(vma, is_register) || | |
792 | vma->vm_file->f_mapping->host != uprobe->inode) | |
26872090 ON |
793 | goto unlock; |
794 | ||
f4d6dfe5 ON |
795 | if (vma->vm_start > info->vaddr || |
796 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 797 | goto unlock; |
2b144498 | 798 | |
78f74116 | 799 | if (is_register) |
26872090 | 800 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
78f74116 | 801 | else |
076a365b | 802 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
78f74116 | 803 | |
26872090 ON |
804 | unlock: |
805 | up_write(&mm->mmap_sem); | |
806 | free: | |
807 | mmput(mm); | |
808 | info = free_map_info(info); | |
2b144498 | 809 | } |
32cdba1e ON |
810 | out: |
811 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 812 | return err; |
2b144498 SD |
813 | } |
814 | ||
9a98e03c | 815 | static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 816 | { |
9a98e03c | 817 | consumer_add(uprobe, uc); |
bb929284 | 818 | return register_for_each_vma(uprobe, true); |
2b144498 SD |
819 | } |
820 | ||
04aab9b2 | 821 | static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 822 | { |
04aab9b2 ON |
823 | int err; |
824 | ||
825 | if (!consumer_del(uprobe, uc)) /* WARN? */ | |
826 | return; | |
2b144498 | 827 | |
04aab9b2 | 828 | err = register_for_each_vma(uprobe, false); |
bb929284 ON |
829 | /* TODO : cant unregister? schedule a worker thread */ |
830 | if (!uprobe->consumers && !err) | |
831 | delete_uprobe(uprobe); | |
2b144498 SD |
832 | } |
833 | ||
834 | /* | |
7b2d81d4 | 835 | * uprobe_register - register a probe |
2b144498 SD |
836 | * @inode: the file in which the probe has to be placed. |
837 | * @offset: offset from the start of the file. | |
e3343e6a | 838 | * @uc: information on howto handle the probe.. |
2b144498 | 839 | * |
7b2d81d4 | 840 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
841 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
842 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 843 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 844 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 845 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
846 | * unregisters. |
847 | * | |
848 | * Return errno if it cannot successully install probes | |
849 | * else return 0 (success) | |
850 | */ | |
e3343e6a | 851 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
852 | { |
853 | struct uprobe *uprobe; | |
7b2d81d4 | 854 | int ret; |
2b144498 | 855 | |
f0744af7 | 856 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 857 | if (offset > i_size_read(inode)) |
7b2d81d4 | 858 | return -EINVAL; |
2b144498 | 859 | |
9a98e03c | 860 | ret = -ENOMEM; |
2b144498 SD |
861 | mutex_lock(uprobes_hash(inode)); |
862 | uprobe = alloc_uprobe(inode, offset); | |
9a98e03c | 863 | if (uprobe) { |
e591c8d7 | 864 | down_write(&uprobe->register_rwsem); |
9a98e03c ON |
865 | ret = __uprobe_register(uprobe, uc); |
866 | if (ret) | |
04aab9b2 | 867 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 868 | up_write(&uprobe->register_rwsem); |
2b144498 | 869 | } |
2b144498 | 870 | mutex_unlock(uprobes_hash(inode)); |
6d1d8dfa SAS |
871 | if (uprobe) |
872 | put_uprobe(uprobe); | |
2b144498 SD |
873 | |
874 | return ret; | |
875 | } | |
876 | ||
877 | /* | |
7b2d81d4 | 878 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
879 | * @inode: the file in which the probe has to be removed. |
880 | * @offset: offset from the start of the file. | |
e3343e6a | 881 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 882 | */ |
e3343e6a | 883 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 884 | { |
7b2d81d4 | 885 | struct uprobe *uprobe; |
2b144498 | 886 | |
2b144498 SD |
887 | uprobe = find_uprobe(inode, offset); |
888 | if (!uprobe) | |
889 | return; | |
890 | ||
891 | mutex_lock(uprobes_hash(inode)); | |
e591c8d7 | 892 | down_write(&uprobe->register_rwsem); |
04aab9b2 | 893 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 894 | up_write(&uprobe->register_rwsem); |
2b144498 | 895 | mutex_unlock(uprobes_hash(inode)); |
c91368c4 | 896 | put_uprobe(uprobe); |
2b144498 SD |
897 | } |
898 | ||
891c3970 ON |
899 | static struct rb_node * |
900 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 901 | { |
2b144498 | 902 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
903 | |
904 | while (n) { | |
891c3970 | 905 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 906 | |
891c3970 | 907 | if (inode < u->inode) { |
2b144498 | 908 | n = n->rb_left; |
891c3970 | 909 | } else if (inode > u->inode) { |
2b144498 | 910 | n = n->rb_right; |
891c3970 ON |
911 | } else { |
912 | if (max < u->offset) | |
913 | n = n->rb_left; | |
914 | else if (min > u->offset) | |
915 | n = n->rb_right; | |
916 | else | |
917 | break; | |
918 | } | |
2b144498 | 919 | } |
7b2d81d4 | 920 | |
891c3970 | 921 | return n; |
2b144498 SD |
922 | } |
923 | ||
924 | /* | |
891c3970 | 925 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 926 | */ |
891c3970 ON |
927 | static void build_probe_list(struct inode *inode, |
928 | struct vm_area_struct *vma, | |
929 | unsigned long start, unsigned long end, | |
930 | struct list_head *head) | |
2b144498 | 931 | { |
891c3970 | 932 | loff_t min, max; |
891c3970 ON |
933 | struct rb_node *n, *t; |
934 | struct uprobe *u; | |
7b2d81d4 | 935 | |
891c3970 | 936 | INIT_LIST_HEAD(head); |
cb113b47 | 937 | min = vaddr_to_offset(vma, start); |
891c3970 | 938 | max = min + (end - start) - 1; |
2b144498 | 939 | |
6f47caa0 | 940 | spin_lock(&uprobes_treelock); |
891c3970 ON |
941 | n = find_node_in_range(inode, min, max); |
942 | if (n) { | |
943 | for (t = n; t; t = rb_prev(t)) { | |
944 | u = rb_entry(t, struct uprobe, rb_node); | |
945 | if (u->inode != inode || u->offset < min) | |
946 | break; | |
947 | list_add(&u->pending_list, head); | |
948 | atomic_inc(&u->ref); | |
949 | } | |
950 | for (t = n; (t = rb_next(t)); ) { | |
951 | u = rb_entry(t, struct uprobe, rb_node); | |
952 | if (u->inode != inode || u->offset > max) | |
953 | break; | |
954 | list_add(&u->pending_list, head); | |
955 | atomic_inc(&u->ref); | |
956 | } | |
2b144498 | 957 | } |
6f47caa0 | 958 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
959 | } |
960 | ||
961 | /* | |
5e5be71a | 962 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 963 | * |
5e5be71a ON |
964 | * Currently we ignore all errors and always return 0, the callers |
965 | * can't handle the failure anyway. | |
2b144498 | 966 | */ |
7b2d81d4 | 967 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
968 | { |
969 | struct list_head tmp_list; | |
665605a2 | 970 | struct uprobe *uprobe, *u; |
2b144498 | 971 | struct inode *inode; |
2b144498 | 972 | |
441f1eb7 | 973 | if (no_uprobe_events() || !valid_vma(vma, true)) |
7b2d81d4 | 974 | return 0; |
2b144498 SD |
975 | |
976 | inode = vma->vm_file->f_mapping->host; | |
977 | if (!inode) | |
7b2d81d4 | 978 | return 0; |
2b144498 | 979 | |
2b144498 | 980 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 981 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
7b2d81d4 | 982 | |
665605a2 | 983 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
5e5be71a | 984 | if (!fatal_signal_pending(current)) { |
57683f72 | 985 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 986 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
987 | } |
988 | put_uprobe(uprobe); | |
989 | } | |
2b144498 SD |
990 | mutex_unlock(uprobes_mmap_hash(inode)); |
991 | ||
5e5be71a | 992 | return 0; |
2b144498 SD |
993 | } |
994 | ||
9f68f672 ON |
995 | static bool |
996 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
997 | { | |
998 | loff_t min, max; | |
999 | struct inode *inode; | |
1000 | struct rb_node *n; | |
1001 | ||
1002 | inode = vma->vm_file->f_mapping->host; | |
1003 | ||
1004 | min = vaddr_to_offset(vma, start); | |
1005 | max = min + (end - start) - 1; | |
1006 | ||
1007 | spin_lock(&uprobes_treelock); | |
1008 | n = find_node_in_range(inode, min, max); | |
1009 | spin_unlock(&uprobes_treelock); | |
1010 | ||
1011 | return !!n; | |
1012 | } | |
1013 | ||
682968e0 SD |
1014 | /* |
1015 | * Called in context of a munmap of a vma. | |
1016 | */ | |
cbc91f71 | 1017 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1018 | { |
441f1eb7 | 1019 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1020 | return; |
1021 | ||
2fd611a9 ON |
1022 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1023 | return; | |
1024 | ||
9f68f672 ON |
1025 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1026 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1027 | return; |
1028 | ||
9f68f672 ON |
1029 | if (vma_has_uprobes(vma, start, end)) |
1030 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1031 | } |
1032 | ||
d4b3b638 SD |
1033 | /* Slot allocation for XOL */ |
1034 | static int xol_add_vma(struct xol_area *area) | |
1035 | { | |
1036 | struct mm_struct *mm; | |
1037 | int ret; | |
1038 | ||
1039 | area->page = alloc_page(GFP_HIGHUSER); | |
1040 | if (!area->page) | |
1041 | return -ENOMEM; | |
1042 | ||
1043 | ret = -EALREADY; | |
1044 | mm = current->mm; | |
1045 | ||
1046 | down_write(&mm->mmap_sem); | |
1047 | if (mm->uprobes_state.xol_area) | |
1048 | goto fail; | |
1049 | ||
1050 | ret = -ENOMEM; | |
1051 | ||
1052 | /* Try to map as high as possible, this is only a hint. */ | |
1053 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1054 | if (area->vaddr & ~PAGE_MASK) { | |
1055 | ret = area->vaddr; | |
1056 | goto fail; | |
1057 | } | |
1058 | ||
1059 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1060 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1061 | if (ret) | |
1062 | goto fail; | |
1063 | ||
1064 | smp_wmb(); /* pairs with get_xol_area() */ | |
1065 | mm->uprobes_state.xol_area = area; | |
1066 | ret = 0; | |
1067 | ||
1068 | fail: | |
1069 | up_write(&mm->mmap_sem); | |
1070 | if (ret) | |
1071 | __free_page(area->page); | |
1072 | ||
1073 | return ret; | |
1074 | } | |
1075 | ||
1076 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1077 | { | |
1078 | struct xol_area *area; | |
1079 | ||
1080 | area = mm->uprobes_state.xol_area; | |
1081 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1082 | ||
1083 | return area; | |
1084 | } | |
1085 | ||
1086 | /* | |
1087 | * xol_alloc_area - Allocate process's xol_area. | |
1088 | * This area will be used for storing instructions for execution out of | |
1089 | * line. | |
1090 | * | |
1091 | * Returns the allocated area or NULL. | |
1092 | */ | |
1093 | static struct xol_area *xol_alloc_area(void) | |
1094 | { | |
1095 | struct xol_area *area; | |
1096 | ||
1097 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1098 | if (unlikely(!area)) | |
1099 | return NULL; | |
1100 | ||
1101 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1102 | ||
1103 | if (!area->bitmap) | |
1104 | goto fail; | |
1105 | ||
1106 | init_waitqueue_head(&area->wq); | |
1107 | if (!xol_add_vma(area)) | |
1108 | return area; | |
1109 | ||
1110 | fail: | |
1111 | kfree(area->bitmap); | |
1112 | kfree(area); | |
1113 | ||
1114 | return get_xol_area(current->mm); | |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * uprobe_clear_state - Free the area allocated for slots. | |
1119 | */ | |
1120 | void uprobe_clear_state(struct mm_struct *mm) | |
1121 | { | |
1122 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1123 | ||
1124 | if (!area) | |
1125 | return; | |
1126 | ||
1127 | put_page(area->page); | |
1128 | kfree(area->bitmap); | |
1129 | kfree(area); | |
1130 | } | |
1131 | ||
32cdba1e ON |
1132 | void uprobe_start_dup_mmap(void) |
1133 | { | |
1134 | percpu_down_read(&dup_mmap_sem); | |
1135 | } | |
1136 | ||
1137 | void uprobe_end_dup_mmap(void) | |
1138 | { | |
1139 | percpu_up_read(&dup_mmap_sem); | |
1140 | } | |
1141 | ||
f8ac4ec9 ON |
1142 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1143 | { | |
61559a81 ON |
1144 | newmm->uprobes_state.xol_area = NULL; |
1145 | ||
9f68f672 | 1146 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1147 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1148 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1149 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1150 | } | |
f8ac4ec9 ON |
1151 | } |
1152 | ||
d4b3b638 SD |
1153 | /* |
1154 | * - search for a free slot. | |
1155 | */ | |
1156 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1157 | { | |
1158 | unsigned long slot_addr; | |
1159 | int slot_nr; | |
1160 | ||
1161 | do { | |
1162 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1163 | if (slot_nr < UINSNS_PER_PAGE) { | |
1164 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1165 | break; | |
1166 | ||
1167 | slot_nr = UINSNS_PER_PAGE; | |
1168 | continue; | |
1169 | } | |
1170 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1171 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1172 | ||
1173 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1174 | atomic_inc(&area->slot_count); | |
1175 | ||
1176 | return slot_addr; | |
1177 | } | |
1178 | ||
1179 | /* | |
1180 | * xol_get_insn_slot - If was not allocated a slot, then | |
1181 | * allocate a slot. | |
1182 | * Returns the allocated slot address or 0. | |
1183 | */ | |
1184 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1185 | { | |
1186 | struct xol_area *area; | |
1187 | unsigned long offset; | |
1188 | void *vaddr; | |
1189 | ||
1190 | area = get_xol_area(current->mm); | |
1191 | if (!area) { | |
1192 | area = xol_alloc_area(); | |
1193 | if (!area) | |
1194 | return 0; | |
1195 | } | |
1196 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1197 | ||
1198 | /* | |
1199 | * Initialize the slot if xol_vaddr points to valid | |
1200 | * instruction slot. | |
1201 | */ | |
1202 | if (unlikely(!current->utask->xol_vaddr)) | |
1203 | return 0; | |
1204 | ||
1205 | current->utask->vaddr = slot_addr; | |
1206 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1207 | vaddr = kmap_atomic(area->page); | |
1208 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1209 | kunmap_atomic(vaddr); | |
65b6ecc0 RV |
1210 | /* |
1211 | * We probably need flush_icache_user_range() but it needs vma. | |
1212 | * This should work on supported architectures too. | |
1213 | */ | |
1214 | flush_dcache_page(area->page); | |
d4b3b638 SD |
1215 | |
1216 | return current->utask->xol_vaddr; | |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * xol_free_insn_slot - If slot was earlier allocated by | |
1221 | * @xol_get_insn_slot(), make the slot available for | |
1222 | * subsequent requests. | |
1223 | */ | |
1224 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1225 | { | |
1226 | struct xol_area *area; | |
1227 | unsigned long vma_end; | |
1228 | unsigned long slot_addr; | |
1229 | ||
1230 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1231 | return; | |
1232 | ||
1233 | slot_addr = tsk->utask->xol_vaddr; | |
1234 | ||
1235 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1236 | return; | |
1237 | ||
1238 | area = tsk->mm->uprobes_state.xol_area; | |
1239 | vma_end = area->vaddr + PAGE_SIZE; | |
1240 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1241 | unsigned long offset; | |
1242 | int slot_nr; | |
1243 | ||
1244 | offset = slot_addr - area->vaddr; | |
1245 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1246 | if (slot_nr >= UINSNS_PER_PAGE) | |
1247 | return; | |
1248 | ||
1249 | clear_bit(slot_nr, area->bitmap); | |
1250 | atomic_dec(&area->slot_count); | |
1251 | if (waitqueue_active(&area->wq)) | |
1252 | wake_up(&area->wq); | |
1253 | ||
1254 | tsk->utask->xol_vaddr = 0; | |
1255 | } | |
1256 | } | |
1257 | ||
0326f5a9 SD |
1258 | /** |
1259 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1260 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1261 | * instruction. | |
1262 | * Return the address of the breakpoint instruction. | |
1263 | */ | |
1264 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1265 | { | |
1266 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1267 | } | |
1268 | ||
1269 | /* | |
1270 | * Called with no locks held. | |
1271 | * Called in context of a exiting or a exec-ing thread. | |
1272 | */ | |
1273 | void uprobe_free_utask(struct task_struct *t) | |
1274 | { | |
1275 | struct uprobe_task *utask = t->utask; | |
1276 | ||
0326f5a9 SD |
1277 | if (!utask) |
1278 | return; | |
1279 | ||
1280 | if (utask->active_uprobe) | |
1281 | put_uprobe(utask->active_uprobe); | |
1282 | ||
d4b3b638 | 1283 | xol_free_insn_slot(t); |
0326f5a9 SD |
1284 | kfree(utask); |
1285 | t->utask = NULL; | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * Called in context of a new clone/fork from copy_process. | |
1290 | */ | |
1291 | void uprobe_copy_process(struct task_struct *t) | |
1292 | { | |
1293 | t->utask = NULL; | |
0326f5a9 SD |
1294 | } |
1295 | ||
1296 | /* | |
1297 | * Allocate a uprobe_task object for the task. | |
1298 | * Called when the thread hits a breakpoint for the first time. | |
1299 | * | |
1300 | * Returns: | |
1301 | * - pointer to new uprobe_task on success | |
1302 | * - NULL otherwise | |
1303 | */ | |
1304 | static struct uprobe_task *add_utask(void) | |
1305 | { | |
1306 | struct uprobe_task *utask; | |
1307 | ||
1308 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1309 | if (unlikely(!utask)) | |
1310 | return NULL; | |
1311 | ||
0326f5a9 SD |
1312 | current->utask = utask; |
1313 | return utask; | |
1314 | } | |
1315 | ||
1316 | /* Prepare to single-step probed instruction out of line. */ | |
1317 | static int | |
1318 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1319 | { | |
d4b3b638 SD |
1320 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1321 | return 0; | |
1322 | ||
0326f5a9 SD |
1323 | return -EFAULT; |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * If we are singlestepping, then ensure this thread is not connected to | |
1328 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1329 | * triggers the signal, restart the original insn even if the task is | |
1330 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1331 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1332 | * _same_ instruction should be repeated again after return from the signal | |
1333 | * handler, and SSTEP can never finish in this case. | |
1334 | */ | |
1335 | bool uprobe_deny_signal(void) | |
1336 | { | |
1337 | struct task_struct *t = current; | |
1338 | struct uprobe_task *utask = t->utask; | |
1339 | ||
1340 | if (likely(!utask || !utask->active_uprobe)) | |
1341 | return false; | |
1342 | ||
1343 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1344 | ||
1345 | if (signal_pending(t)) { | |
1346 | spin_lock_irq(&t->sighand->siglock); | |
1347 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1348 | spin_unlock_irq(&t->sighand->siglock); | |
1349 | ||
1350 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1351 | utask->state = UTASK_SSTEP_TRAPPED; | |
1352 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1353 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1354 | } | |
1355 | } | |
1356 | ||
1357 | return true; | |
1358 | } | |
1359 | ||
1360 | /* | |
1361 | * Avoid singlestepping the original instruction if the original instruction | |
1362 | * is a NOP or can be emulated. | |
1363 | */ | |
1364 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1365 | { | |
71434f2f | 1366 | if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) { |
0578a970 ON |
1367 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
1368 | return true; | |
71434f2f | 1369 | clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); |
0578a970 | 1370 | } |
0326f5a9 SD |
1371 | return false; |
1372 | } | |
1373 | ||
499a4f3e ON |
1374 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1375 | { | |
1376 | struct vm_area_struct *vma; | |
1377 | ||
1378 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1379 | if (!valid_vma(vma, false)) | |
1380 | continue; | |
1381 | /* | |
1382 | * This is not strictly accurate, we can race with | |
1383 | * uprobe_unregister() and see the already removed | |
1384 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 1385 | * Or this uprobe can be filtered out. |
499a4f3e ON |
1386 | */ |
1387 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1388 | return; | |
1389 | } | |
1390 | ||
1391 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1392 | } | |
1393 | ||
ec75fba9 ON |
1394 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
1395 | { | |
1396 | struct page *page; | |
1397 | uprobe_opcode_t opcode; | |
1398 | int result; | |
1399 | ||
1400 | pagefault_disable(); | |
1401 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1402 | sizeof(opcode)); | |
1403 | pagefault_enable(); | |
1404 | ||
1405 | if (likely(result == 0)) | |
1406 | goto out; | |
1407 | ||
1408 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1409 | if (result < 0) | |
1410 | return result; | |
1411 | ||
1412 | copy_opcode(page, vaddr, &opcode); | |
1413 | put_page(page); | |
1414 | out: | |
1415 | return is_swbp_insn(&opcode); | |
1416 | } | |
1417 | ||
d790d346 | 1418 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1419 | { |
3a9ea052 ON |
1420 | struct mm_struct *mm = current->mm; |
1421 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1422 | struct vm_area_struct *vma; |
0326f5a9 | 1423 | |
0326f5a9 SD |
1424 | down_read(&mm->mmap_sem); |
1425 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1426 | if (vma && vma->vm_start <= bp_vaddr) { |
1427 | if (valid_vma(vma, false)) { | |
cb113b47 ON |
1428 | struct inode *inode = vma->vm_file->f_mapping->host; |
1429 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); | |
0326f5a9 | 1430 | |
3a9ea052 ON |
1431 | uprobe = find_uprobe(inode, offset); |
1432 | } | |
d790d346 ON |
1433 | |
1434 | if (!uprobe) | |
1435 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1436 | } else { | |
1437 | *is_swbp = -EFAULT; | |
0326f5a9 | 1438 | } |
499a4f3e ON |
1439 | |
1440 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1441 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1442 | up_read(&mm->mmap_sem); |
1443 | ||
3a9ea052 ON |
1444 | return uprobe; |
1445 | } | |
1446 | ||
1447 | /* | |
1448 | * Run handler and ask thread to singlestep. | |
1449 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1450 | */ | |
1451 | static void handle_swbp(struct pt_regs *regs) | |
1452 | { | |
1453 | struct uprobe_task *utask; | |
1454 | struct uprobe *uprobe; | |
1455 | unsigned long bp_vaddr; | |
56bb4cf6 | 1456 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1457 | |
1458 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1459 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1460 | |
0326f5a9 | 1461 | if (!uprobe) { |
56bb4cf6 ON |
1462 | if (is_swbp > 0) { |
1463 | /* No matching uprobe; signal SIGTRAP. */ | |
1464 | send_sig(SIGTRAP, current, 0); | |
1465 | } else { | |
1466 | /* | |
1467 | * Either we raced with uprobe_unregister() or we can't | |
1468 | * access this memory. The latter is only possible if | |
1469 | * another thread plays with our ->mm. In both cases | |
1470 | * we can simply restart. If this vma was unmapped we | |
1471 | * can pretend this insn was not executed yet and get | |
1472 | * the (correct) SIGSEGV after restart. | |
1473 | */ | |
1474 | instruction_pointer_set(regs, bp_vaddr); | |
1475 | } | |
0326f5a9 SD |
1476 | return; |
1477 | } | |
142b18dd ON |
1478 | /* |
1479 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1480 | * After we hit the bp, _unregister + _register can install the | |
1481 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1482 | */ | |
1483 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1484 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
142b18dd | 1485 | goto restart; |
0326f5a9 SD |
1486 | |
1487 | utask = current->utask; | |
1488 | if (!utask) { | |
1489 | utask = add_utask(); | |
1490 | /* Cannot allocate; re-execute the instruction. */ | |
1491 | if (!utask) | |
0578a970 | 1492 | goto restart; |
0326f5a9 | 1493 | } |
746a9e6b | 1494 | |
0326f5a9 | 1495 | handler_chain(uprobe, regs); |
0578a970 ON |
1496 | if (can_skip_sstep(uprobe, regs)) |
1497 | goto out; | |
0326f5a9 | 1498 | |
0326f5a9 | 1499 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { |
746a9e6b ON |
1500 | utask->active_uprobe = uprobe; |
1501 | utask->state = UTASK_SSTEP; | |
0326f5a9 SD |
1502 | return; |
1503 | } | |
1504 | ||
0578a970 ON |
1505 | restart: |
1506 | /* | |
1507 | * cannot singlestep; cannot skip instruction; | |
1508 | * re-execute the instruction. | |
1509 | */ | |
1510 | instruction_pointer_set(regs, bp_vaddr); | |
1511 | out: | |
8bd87445 | 1512 | put_uprobe(uprobe); |
0326f5a9 SD |
1513 | } |
1514 | ||
1515 | /* | |
1516 | * Perform required fix-ups and disable singlestep. | |
1517 | * Allow pending signals to take effect. | |
1518 | */ | |
1519 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1520 | { | |
1521 | struct uprobe *uprobe; | |
1522 | ||
1523 | uprobe = utask->active_uprobe; | |
1524 | if (utask->state == UTASK_SSTEP_ACK) | |
1525 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1526 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1527 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1528 | else | |
1529 | WARN_ON_ONCE(1); | |
1530 | ||
1531 | put_uprobe(uprobe); | |
1532 | utask->active_uprobe = NULL; | |
1533 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1534 | xol_free_insn_slot(current); |
0326f5a9 SD |
1535 | |
1536 | spin_lock_irq(¤t->sighand->siglock); | |
1537 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1538 | spin_unlock_irq(¤t->sighand->siglock); | |
1539 | } | |
1540 | ||
1541 | /* | |
1b08e907 ON |
1542 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1543 | * allows the thread to return from interrupt. After that handle_swbp() | |
1544 | * sets utask->active_uprobe. | |
0326f5a9 | 1545 | * |
1b08e907 ON |
1546 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1547 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1548 | * |
1549 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1550 | * uprobe_notify_resume(). | |
1551 | */ | |
1552 | void uprobe_notify_resume(struct pt_regs *regs) | |
1553 | { | |
1554 | struct uprobe_task *utask; | |
1555 | ||
db023ea5 ON |
1556 | clear_thread_flag(TIF_UPROBE); |
1557 | ||
0326f5a9 | 1558 | utask = current->utask; |
1b08e907 | 1559 | if (utask && utask->active_uprobe) |
0326f5a9 | 1560 | handle_singlestep(utask, regs); |
1b08e907 ON |
1561 | else |
1562 | handle_swbp(regs); | |
0326f5a9 SD |
1563 | } |
1564 | ||
1565 | /* | |
1566 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1567 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1568 | */ | |
1569 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1570 | { | |
f8ac4ec9 | 1571 | if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) |
0326f5a9 SD |
1572 | return 0; |
1573 | ||
0326f5a9 | 1574 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1575 | return 1; |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1580 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1581 | */ | |
1582 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1583 | { | |
1584 | struct uprobe_task *utask = current->utask; | |
1585 | ||
1586 | if (!current->mm || !utask || !utask->active_uprobe) | |
1587 | /* task is currently not uprobed */ | |
1588 | return 0; | |
1589 | ||
1590 | utask->state = UTASK_SSTEP_ACK; | |
1591 | set_thread_flag(TIF_UPROBE); | |
1592 | return 1; | |
1593 | } | |
1594 | ||
1595 | static struct notifier_block uprobe_exception_nb = { | |
1596 | .notifier_call = arch_uprobe_exception_notify, | |
1597 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1598 | }; | |
1599 | ||
2b144498 SD |
1600 | static int __init init_uprobes(void) |
1601 | { | |
1602 | int i; | |
1603 | ||
1604 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1605 | mutex_init(&uprobes_mutex[i]); | |
1606 | mutex_init(&uprobes_mmap_mutex[i]); | |
1607 | } | |
0326f5a9 | 1608 | |
32cdba1e ON |
1609 | if (percpu_init_rwsem(&dup_mmap_sem)) |
1610 | return -ENOMEM; | |
1611 | ||
0326f5a9 | 1612 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 1613 | } |
0326f5a9 | 1614 | module_init(init_uprobes); |
2b144498 SD |
1615 | |
1616 | static void __exit exit_uprobes(void) | |
1617 | { | |
1618 | } | |
2b144498 | 1619 | module_exit(exit_uprobes); |